tiny226 commited on
Commit
4f3da59
·
verified ·
1 Parent(s): c6d1ff4

Upload folder using huggingface_hub (part 6)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +18 -0
  2. triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.cubin +0 -0
  3. triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.json +1 -0
  4. triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.llir +633 -0
  5. triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ptx +642 -0
  6. triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.source +172 -0
  7. triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttgir +135 -0
  8. triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttir +128 -0
  9. triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/__grp__l2norm_fwd_kernel.json +1 -0
  10. triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.cubin +0 -0
  11. triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.json +1 -0
  12. triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.llir +630 -0
  13. triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ptx +638 -0
  14. triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.source +172 -0
  15. triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttgir +135 -0
  16. triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttir +128 -0
  17. triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/__grp__l2norm_fwd_kernel.json +1 -0
  18. triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.cubin +3 -0
  19. triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.json +1 -0
  20. triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.llir +0 -0
  21. triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ptx +0 -0
  22. triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.source +172 -0
  23. triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttgir +135 -0
  24. triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttir +125 -0
  25. triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/__grp__l2norm_fwd_kernel.json +1 -0
  26. triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.cubin +0 -0
  27. triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.json +1 -0
  28. triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.llir +1143 -0
  29. triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ptx +966 -0
  30. triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.source +172 -0
  31. triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttgir +135 -0
  32. triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttir +128 -0
  33. triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/__grp__l2norm_bwd_kernel.json +1 -0
  34. triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.cubin +0 -0
  35. triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.json +1 -0
  36. triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.llir +1269 -0
  37. triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ptx +1322 -0
  38. triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.source +210 -0
  39. triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttgir +146 -0
  40. triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttir +144 -0
  41. triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/__grp__l2norm_bwd_kernel.json +1 -0
  42. triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.cubin +3 -0
  43. triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.json +1 -0
  44. triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.llir +0 -0
  45. triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ptx +0 -0
  46. triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.source +210 -0
  47. triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttgir +154 -0
  48. triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttir +144 -0
  49. triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/__grp__l2norm_fwd_kernel.json +1 -0
  50. triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.cubin +0 -0
.gitattributes CHANGED
@@ -67,3 +67,21 @@ triton/gpu0/A7X6T6PKJCRMT3VM6R2EQWTJL4JVXK2EY6S4UGCLO42GUJM54JWQ/chunk_bwd_kerne
67
  triton/gpu0/AEQGKJYL6DLKXOG3TU4HKZV7AVJHMVX2MVTUSYTRLB5UQBUBVLEA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
68
  triton/gpu0/AKSFD27IP2GUNEHHZEPXU4XCHBVIYLE7I6AGS43DEHS44EMEKUNA/recompute_w_u_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
69
  triton/gpu0/AOEYMJ4TTFDXRPXMBW52VLMJ7OBEVHHGC7Z5TZAFKQFJAGELKGPA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
67
  triton/gpu0/AEQGKJYL6DLKXOG3TU4HKZV7AVJHMVX2MVTUSYTRLB5UQBUBVLEA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
68
  triton/gpu0/AKSFD27IP2GUNEHHZEPXU4XCHBVIYLE7I6AGS43DEHS44EMEKUNA/recompute_w_u_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
69
  triton/gpu0/AOEYMJ4TTFDXRPXMBW52VLMJ7OBEVHHGC7Z5TZAFKQFJAGELKGPA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
70
+ triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
71
+ triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
72
+ triton/gpu0/AYUU53VTNNZ4LFIJYMXEZRWXU5JY3NM7NVTEQ247SL47KZXLXBPQ/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin filter=lfs diff=lfs merge=lfs -text
73
+ triton/gpu0/AZ2POZCGJMUQFHK2T3FRUE454XEXY5VLJNMHENMV74CHSJIGNLKA/chunk_gated_delta_rule_bwd_kernel_dhu_blockdim64.cubin filter=lfs diff=lfs merge=lfs -text
74
+ triton/gpu0/B3KG5EI256WXP2F453JJA5EZ23LGPQX6JMCCQOBMDMXZ7L3O74IQ/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
75
+ triton/gpu0/BAM5EFAHFGOPXBL6ESYNHVXXIUC6CEF6BE7RENTTQGFKFM2SL4FA/recompute_w_u_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
76
+ triton/gpu0/BDSEFEY6V4ZNS7CUIRSXLJG5JS3IBFXLJBNWJ3G3MGRPRBXZTBDQ/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
77
+ triton/gpu0/BVUXRJFXLWMPTXZYYXZZDKTF7DL2TDROBXMZSECMK7XWW56KU4XA/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
78
+ triton/gpu0/BXR5KI5TYDWMFJZJNP2HRCVZPXHZ4G26FQNU664UID3Y7EBYWWSA/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
79
+ triton/gpu0/C3BBG2WA46OLFJWCPGPIHHKU45J4OYISJP4HOWWEO7G3S2AUFDJA/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
80
+ triton/gpu0/C4JTGE6RONYBE2TIA3VQNFH5ET3IPXF5BENOYW3UC6DBIRHSX7AQ/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
81
+ triton/gpu0/C4ZXA7K33QWBGGKKAVHEWMPIFY3L7QKRVHELLC7C7FYSTLXUXK6Q/chunk_bwd_kernel_dv_local.cubin filter=lfs diff=lfs merge=lfs -text
82
+ triton/gpu0/CFFDWJFZ47MY2YWOXFQIWADZNBHOOK5GE3QPM3NRGAJ7LC4T2JYA/chunk_gated_delta_rule_fwd_kkt_solve_kernel.cubin filter=lfs diff=lfs merge=lfs -text
83
+ triton/gpu0/CHNWL4ABYPDOXW6CHFJAGTHSJZ3DYKK772V6UXBJ66ARZIICL44A/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
84
+ triton/gpu0/CHXV3TIVRYKYD2ETLGKNLFVZBCZX6KIIFHQFBRRQFOZQQXYXG56A/l2norm_bwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
85
+ triton/gpu0/CMD3VKOAXCGTVA3ESN2NTA2JXDOINGPYLUNE3XL7W6I7ZR3B65AA/l2norm_fwd_kernel.cubin filter=lfs diff=lfs merge=lfs -text
86
+ triton/gpu0/CMHJC57VN4MELHWQTXVJHXBMJELLE5JKMINJOJUUCE2VNETPUN7Q/chunk_gated_delta_rule_fwd_kernel_h_blockdim64.cubin filter=lfs diff=lfs merge=lfs -text
87
+ triton/gpu0/COEH6RZWJ4F4NLI6BWD6G2VBWJCNYUFXAFKSMEBEK2VRZ3DVOIQA/chunk_fwd_kernel_o.cubin filter=lfs diff=lfs merge=lfs -text
triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.cubin ADDED
Binary file (28.6 kB). View file
 
triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}
triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.llir ADDED
@@ -0,0 +1,633 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ; ModuleID = 'LLVMDialectModule'
2
+ source_filename = "LLVMDialectModule"
3
+ 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"
4
+
5
+ @global_smem = external addrspace(3) global [0 x i8], align 16
6
+ @.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1
7
+ @.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1
8
+
9
+ ; Function Attrs: nounwind
10
+ 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 {
11
+ %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13
12
+ %9 = zext nneg i32 %8 to i64, !dbg !14
13
+ %10 = shl nuw nsw i64 %9, 3, !dbg !15
14
+ %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16
15
+ %12 = and i32 %11, 16, !dbg !16
16
+ %.lobit = lshr exact i32 %12, 4, !dbg !16
17
+ %13 = or disjoint i32 %.lobit, 2, !dbg !16
18
+ %14 = or disjoint i32 %.lobit, 4, !dbg !16
19
+ %15 = or disjoint i32 %.lobit, 6, !dbg !16
20
+ %16 = zext nneg i32 %.lobit to i64, !dbg !17
21
+ %17 = zext nneg i32 %13 to i64, !dbg !17
22
+ %18 = zext nneg i32 %14 to i64, !dbg !17
23
+ %19 = zext nneg i32 %15 to i64, !dbg !17
24
+ %20 = or disjoint i64 %10, %16, !dbg !17
25
+ %21 = or disjoint i64 %10, %17, !dbg !17
26
+ %22 = or disjoint i64 %10, %18, !dbg !17
27
+ %23 = or disjoint i64 %10, %19, !dbg !17
28
+ %24 = sext i32 %4 to i64, !dbg !18
29
+ %25 = icmp slt i64 %20, %24, !dbg !18
30
+ %26 = icmp slt i64 %21, %24, !dbg !18
31
+ %27 = icmp slt i64 %22, %24, !dbg !18
32
+ %28 = icmp slt i64 %23, %24, !dbg !18
33
+ %29 = shl nuw nsw i32 %11, 3, !dbg !19
34
+ %30 = and i32 %29, 120, !dbg !19
35
+ %31 = shl nuw nsw i64 %20, 7, !dbg !20
36
+ %32 = shl nuw nsw i64 %21, 7, !dbg !20
37
+ %33 = shl nuw nsw i64 %22, 7, !dbg !20
38
+ %34 = shl nuw nsw i64 %23, 7, !dbg !20
39
+ %35 = getelementptr bfloat, ptr addrspace(1) %0, i64 %31, !dbg !21
40
+ %36 = getelementptr bfloat, ptr addrspace(1) %0, i64 %32, !dbg !21
41
+ %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !21
42
+ %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !21
43
+ %39 = zext nneg i32 %30 to i64, !dbg !22
44
+ %40 = getelementptr bfloat, ptr addrspace(1) %35, i64 %39, !dbg !22
45
+ %41 = getelementptr bfloat, ptr addrspace(1) %36, i64 %39, !dbg !22
46
+ %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %39, !dbg !22
47
+ %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %39, !dbg !22
48
+ %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
49
+ %45 = extractvalue { i32, i32, i32, i32 } %44, 0, !dbg !23
50
+ %46 = bitcast i32 %45 to <2 x bfloat>, !dbg !23
51
+ %47 = extractvalue { i32, i32, i32, i32 } %44, 1, !dbg !23
52
+ %48 = bitcast i32 %47 to <2 x bfloat>, !dbg !23
53
+ %49 = extractvalue { i32, i32, i32, i32 } %44, 2, !dbg !23
54
+ %50 = bitcast i32 %49 to <2 x bfloat>, !dbg !23
55
+ %51 = extractvalue { i32, i32, i32, i32 } %44, 3, !dbg !23
56
+ %52 = bitcast i32 %51 to <2 x bfloat>, !dbg !23
57
+ %53 = extractelement <2 x bfloat> %46, i64 0, !dbg !23
58
+ %54 = extractelement <2 x bfloat> %46, i64 1, !dbg !23
59
+ %55 = extractelement <2 x bfloat> %48, i64 0, !dbg !23
60
+ %56 = extractelement <2 x bfloat> %48, i64 1, !dbg !23
61
+ %57 = extractelement <2 x bfloat> %50, i64 0, !dbg !23
62
+ %58 = extractelement <2 x bfloat> %50, i64 1, !dbg !23
63
+ %59 = extractelement <2 x bfloat> %52, i64 0, !dbg !23
64
+ %60 = extractelement <2 x bfloat> %52, i64 1, !dbg !23
65
+ %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
66
+ %62 = extractvalue { i32, i32, i32, i32 } %61, 0, !dbg !23
67
+ %63 = bitcast i32 %62 to <2 x bfloat>, !dbg !23
68
+ %64 = extractvalue { i32, i32, i32, i32 } %61, 1, !dbg !23
69
+ %65 = bitcast i32 %64 to <2 x bfloat>, !dbg !23
70
+ %66 = extractvalue { i32, i32, i32, i32 } %61, 2, !dbg !23
71
+ %67 = bitcast i32 %66 to <2 x bfloat>, !dbg !23
72
+ %68 = extractvalue { i32, i32, i32, i32 } %61, 3, !dbg !23
73
+ %69 = bitcast i32 %68 to <2 x bfloat>, !dbg !23
74
+ %70 = extractelement <2 x bfloat> %63, i64 0, !dbg !23
75
+ %71 = extractelement <2 x bfloat> %63, i64 1, !dbg !23
76
+ %72 = extractelement <2 x bfloat> %65, i64 0, !dbg !23
77
+ %73 = extractelement <2 x bfloat> %65, i64 1, !dbg !23
78
+ %74 = extractelement <2 x bfloat> %67, i64 0, !dbg !23
79
+ %75 = extractelement <2 x bfloat> %67, i64 1, !dbg !23
80
+ %76 = extractelement <2 x bfloat> %69, i64 0, !dbg !23
81
+ %77 = extractelement <2 x bfloat> %69, i64 1, !dbg !23
82
+ %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
83
+ %79 = extractvalue { i32, i32, i32, i32 } %78, 0, !dbg !23
84
+ %80 = bitcast i32 %79 to <2 x bfloat>, !dbg !23
85
+ %81 = extractvalue { i32, i32, i32, i32 } %78, 1, !dbg !23
86
+ %82 = bitcast i32 %81 to <2 x bfloat>, !dbg !23
87
+ %83 = extractvalue { i32, i32, i32, i32 } %78, 2, !dbg !23
88
+ %84 = bitcast i32 %83 to <2 x bfloat>, !dbg !23
89
+ %85 = extractvalue { i32, i32, i32, i32 } %78, 3, !dbg !23
90
+ %86 = bitcast i32 %85 to <2 x bfloat>, !dbg !23
91
+ %87 = extractelement <2 x bfloat> %80, i64 0, !dbg !23
92
+ %88 = extractelement <2 x bfloat> %80, i64 1, !dbg !23
93
+ %89 = extractelement <2 x bfloat> %82, i64 0, !dbg !23
94
+ %90 = extractelement <2 x bfloat> %82, i64 1, !dbg !23
95
+ %91 = extractelement <2 x bfloat> %84, i64 0, !dbg !23
96
+ %92 = extractelement <2 x bfloat> %84, i64 1, !dbg !23
97
+ %93 = extractelement <2 x bfloat> %86, i64 0, !dbg !23
98
+ %94 = extractelement <2 x bfloat> %86, i64 1, !dbg !23
99
+ %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
100
+ %96 = extractvalue { i32, i32, i32, i32 } %95, 0, !dbg !23
101
+ %97 = bitcast i32 %96 to <2 x bfloat>, !dbg !23
102
+ %98 = extractvalue { i32, i32, i32, i32 } %95, 1, !dbg !23
103
+ %99 = bitcast i32 %98 to <2 x bfloat>, !dbg !23
104
+ %100 = extractvalue { i32, i32, i32, i32 } %95, 2, !dbg !23
105
+ %101 = bitcast i32 %100 to <2 x bfloat>, !dbg !23
106
+ %102 = extractvalue { i32, i32, i32, i32 } %95, 3, !dbg !23
107
+ %103 = bitcast i32 %102 to <2 x bfloat>, !dbg !23
108
+ %104 = extractelement <2 x bfloat> %97, i64 0, !dbg !23
109
+ %105 = extractelement <2 x bfloat> %97, i64 1, !dbg !23
110
+ %106 = extractelement <2 x bfloat> %99, i64 0, !dbg !23
111
+ %107 = extractelement <2 x bfloat> %99, i64 1, !dbg !23
112
+ %108 = extractelement <2 x bfloat> %101, i64 0, !dbg !23
113
+ %109 = extractelement <2 x bfloat> %101, i64 1, !dbg !23
114
+ %110 = extractelement <2 x bfloat> %103, i64 0, !dbg !23
115
+ %111 = extractelement <2 x bfloat> %103, i64 1, !dbg !23
116
+ %112 = fpext bfloat %53 to float, !dbg !24
117
+ %113 = fpext bfloat %54 to float, !dbg !24
118
+ %114 = fpext bfloat %55 to float, !dbg !24
119
+ %115 = fpext bfloat %56 to float, !dbg !24
120
+ %116 = fpext bfloat %57 to float, !dbg !24
121
+ %117 = fpext bfloat %58 to float, !dbg !24
122
+ %118 = fpext bfloat %59 to float, !dbg !24
123
+ %119 = fpext bfloat %60 to float, !dbg !24
124
+ %120 = fpext bfloat %70 to float, !dbg !24
125
+ %121 = fpext bfloat %71 to float, !dbg !24
126
+ %122 = fpext bfloat %72 to float, !dbg !24
127
+ %123 = fpext bfloat %73 to float, !dbg !24
128
+ %124 = fpext bfloat %74 to float, !dbg !24
129
+ %125 = fpext bfloat %75 to float, !dbg !24
130
+ %126 = fpext bfloat %76 to float, !dbg !24
131
+ %127 = fpext bfloat %77 to float, !dbg !24
132
+ %128 = fpext bfloat %87 to float, !dbg !24
133
+ %129 = fpext bfloat %88 to float, !dbg !24
134
+ %130 = fpext bfloat %89 to float, !dbg !24
135
+ %131 = fpext bfloat %90 to float, !dbg !24
136
+ %132 = fpext bfloat %91 to float, !dbg !24
137
+ %133 = fpext bfloat %92 to float, !dbg !24
138
+ %134 = fpext bfloat %93 to float, !dbg !24
139
+ %135 = fpext bfloat %94 to float, !dbg !24
140
+ %136 = fpext bfloat %104 to float, !dbg !24
141
+ %137 = fpext bfloat %105 to float, !dbg !24
142
+ %138 = fpext bfloat %106 to float, !dbg !24
143
+ %139 = fpext bfloat %107 to float, !dbg !24
144
+ %140 = fpext bfloat %108 to float, !dbg !24
145
+ %141 = fpext bfloat %109 to float, !dbg !24
146
+ %142 = fpext bfloat %110 to float, !dbg !24
147
+ %143 = fpext bfloat %111 to float, !dbg !24
148
+ %144 = fmul float %112, %112, !dbg !25
149
+ %145 = fmul float %113, %113, !dbg !25
150
+ %146 = fmul float %114, %114, !dbg !25
151
+ %147 = fmul float %115, %115, !dbg !25
152
+ %148 = fmul float %116, %116, !dbg !25
153
+ %149 = fmul float %117, %117, !dbg !25
154
+ %150 = fmul float %118, %118, !dbg !25
155
+ %151 = fmul float %119, %119, !dbg !25
156
+ %152 = fmul float %120, %120, !dbg !25
157
+ %153 = fmul float %121, %121, !dbg !25
158
+ %154 = fmul float %122, %122, !dbg !25
159
+ %155 = fmul float %123, %123, !dbg !25
160
+ %156 = fmul float %124, %124, !dbg !25
161
+ %157 = fmul float %125, %125, !dbg !25
162
+ %158 = fmul float %126, %126, !dbg !25
163
+ %159 = fmul float %127, %127, !dbg !25
164
+ %160 = fmul float %128, %128, !dbg !25
165
+ %161 = fmul float %129, %129, !dbg !25
166
+ %162 = fmul float %130, %130, !dbg !25
167
+ %163 = fmul float %131, %131, !dbg !25
168
+ %164 = fmul float %132, %132, !dbg !25
169
+ %165 = fmul float %133, %133, !dbg !25
170
+ %166 = fmul float %134, %134, !dbg !25
171
+ %167 = fmul float %135, %135, !dbg !25
172
+ %168 = fmul float %136, %136, !dbg !25
173
+ %169 = fmul float %137, %137, !dbg !25
174
+ %170 = fmul float %138, %138, !dbg !25
175
+ %171 = fmul float %139, %139, !dbg !25
176
+ %172 = fmul float %140, %140, !dbg !25
177
+ %173 = fmul float %141, %141, !dbg !25
178
+ %174 = fmul float %142, %142, !dbg !25
179
+ %175 = fmul float %143, %143, !dbg !25
180
+ %176 = fadd float %144, %145, !dbg !26
181
+ %177 = fadd float %146, %176, !dbg !26
182
+ %178 = fadd float %147, %177, !dbg !26
183
+ %179 = fadd float %148, %178, !dbg !26
184
+ %180 = fadd float %149, %179, !dbg !26
185
+ %181 = fadd float %150, %180, !dbg !26
186
+ %182 = fadd float %151, %181, !dbg !26
187
+ %183 = fadd float %152, %153, !dbg !26
188
+ %184 = fadd float %154, %183, !dbg !26
189
+ %185 = fadd float %155, %184, !dbg !26
190
+ %186 = fadd float %156, %185, !dbg !26
191
+ %187 = fadd float %157, %186, !dbg !26
192
+ %188 = fadd float %158, %187, !dbg !26
193
+ %189 = fadd float %159, %188, !dbg !26
194
+ %190 = fadd float %160, %161, !dbg !26
195
+ %191 = fadd float %162, %190, !dbg !26
196
+ %192 = fadd float %163, %191, !dbg !26
197
+ %193 = fadd float %164, %192, !dbg !26
198
+ %194 = fadd float %165, %193, !dbg !26
199
+ %195 = fadd float %166, %194, !dbg !26
200
+ %196 = fadd float %167, %195, !dbg !26
201
+ %197 = fadd float %168, %169, !dbg !26
202
+ %198 = fadd float %170, %197, !dbg !26
203
+ %199 = fadd float %171, %198, !dbg !26
204
+ %200 = fadd float %172, %199, !dbg !26
205
+ %201 = fadd float %173, %200, !dbg !26
206
+ %202 = fadd float %174, %201, !dbg !26
207
+ %203 = fadd float %175, %202, !dbg !26
208
+ %204 = bitcast float %182 to i32, !dbg !29
209
+ %205 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %204, i32 8, i32 31), !dbg !29
210
+ %206 = bitcast i32 %205 to float, !dbg !29
211
+ %207 = fadd float %182, %206, !dbg !26
212
+ %208 = bitcast float %207 to i32, !dbg !29
213
+ %209 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %208, i32 4, i32 31), !dbg !29
214
+ %210 = bitcast i32 %209 to float, !dbg !29
215
+ %211 = fadd float %207, %210, !dbg !26
216
+ %212 = bitcast float %211 to i32, !dbg !29
217
+ %213 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %212, i32 2, i32 31), !dbg !29
218
+ %214 = bitcast i32 %213 to float, !dbg !29
219
+ %215 = fadd float %211, %214, !dbg !26
220
+ %216 = bitcast float %215 to i32, !dbg !29
221
+ %217 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %216, i32 1, i32 31), !dbg !29
222
+ %218 = bitcast i32 %217 to float, !dbg !29
223
+ %219 = fadd float %215, %218, !dbg !26
224
+ %220 = bitcast float %189 to i32, !dbg !29
225
+ %221 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %220, i32 8, i32 31), !dbg !29
226
+ %222 = bitcast i32 %221 to float, !dbg !29
227
+ %223 = fadd float %189, %222, !dbg !26
228
+ %224 = bitcast float %223 to i32, !dbg !29
229
+ %225 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %224, i32 4, i32 31), !dbg !29
230
+ %226 = bitcast i32 %225 to float, !dbg !29
231
+ %227 = fadd float %223, %226, !dbg !26
232
+ %228 = bitcast float %227 to i32, !dbg !29
233
+ %229 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %228, i32 2, i32 31), !dbg !29
234
+ %230 = bitcast i32 %229 to float, !dbg !29
235
+ %231 = fadd float %227, %230, !dbg !26
236
+ %232 = bitcast float %231 to i32, !dbg !29
237
+ %233 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %232, i32 1, i32 31), !dbg !29
238
+ %234 = bitcast i32 %233 to float, !dbg !29
239
+ %235 = fadd float %231, %234, !dbg !26
240
+ %236 = bitcast float %196 to i32, !dbg !29
241
+ %237 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %236, i32 8, i32 31), !dbg !29
242
+ %238 = bitcast i32 %237 to float, !dbg !29
243
+ %239 = fadd float %196, %238, !dbg !26
244
+ %240 = bitcast float %239 to i32, !dbg !29
245
+ %241 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %240, i32 4, i32 31), !dbg !29
246
+ %242 = bitcast i32 %241 to float, !dbg !29
247
+ %243 = fadd float %239, %242, !dbg !26
248
+ %244 = bitcast float %243 to i32, !dbg !29
249
+ %245 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %244, i32 2, i32 31), !dbg !29
250
+ %246 = bitcast i32 %245 to float, !dbg !29
251
+ %247 = fadd float %243, %246, !dbg !26
252
+ %248 = bitcast float %247 to i32, !dbg !29
253
+ %249 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %248, i32 1, i32 31), !dbg !29
254
+ %250 = bitcast i32 %249 to float, !dbg !29
255
+ %251 = fadd float %247, %250, !dbg !26
256
+ %252 = bitcast float %203 to i32, !dbg !29
257
+ %253 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %252, i32 8, i32 31), !dbg !29
258
+ %254 = bitcast i32 %253 to float, !dbg !29
259
+ %255 = fadd float %203, %254, !dbg !26
260
+ %256 = bitcast float %255 to i32, !dbg !29
261
+ %257 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %256, i32 4, i32 31), !dbg !29
262
+ %258 = bitcast i32 %257 to float, !dbg !29
263
+ %259 = fadd float %255, %258, !dbg !26
264
+ %260 = bitcast float %259 to i32, !dbg !29
265
+ %261 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %260, i32 2, i32 31), !dbg !29
266
+ %262 = bitcast i32 %261 to float, !dbg !29
267
+ %263 = fadd float %259, %262, !dbg !26
268
+ %264 = bitcast float %263 to i32, !dbg !29
269
+ %265 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %264, i32 1, i32 31), !dbg !29
270
+ %266 = bitcast i32 %265 to float, !dbg !29
271
+ %267 = fadd float %263, %266, !dbg !26
272
+ %268 = fadd float %3, %219, !dbg !32
273
+ %269 = fadd float %3, %235, !dbg !32
274
+ %270 = fadd float %3, %251, !dbg !32
275
+ %271 = fadd float %3, %267, !dbg !32
276
+ %272 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
277
+ %.not.i = icmp eq i32 %272, 0, !dbg !33
278
+ %273 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
279
+ %.not1.i = icmp eq i32 %273, 0, !dbg !33
280
+ br i1 %.not.i, label %279, label %274, !dbg !33
281
+
282
+ 274: ; preds = %7
283
+ br i1 %.not1.i, label %277, label %275, !dbg !33
284
+
285
+ 275: ; preds = %274
286
+ %276 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %268) #6, !dbg !33
287
+ br label %__nv_sqrtf.exit, !dbg !33
288
+
289
+ 277: ; preds = %274
290
+ %278 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %268) #6, !dbg !33
291
+ br label %__nv_sqrtf.exit, !dbg !33
292
+
293
+ 279: ; preds = %7
294
+ br i1 %.not1.i, label %282, label %280, !dbg !33
295
+
296
+ 280: ; preds = %279
297
+ %281 = tail call float @llvm.nvvm.sqrt.rn.f(float %268) #6, !dbg !33
298
+ br label %__nv_sqrtf.exit, !dbg !33
299
+
300
+ 282: ; preds = %279
301
+ %283 = tail call float @llvm.nvvm.sqrt.approx.f(float %268) #6, !dbg !33
302
+ br label %__nv_sqrtf.exit, !dbg !33
303
+
304
+ __nv_sqrtf.exit: ; preds = %275, %277, %280, %282
305
+ %.0.i = phi float [ %276, %275 ], [ %278, %277 ], [ %281, %280 ], [ %283, %282 ], !dbg !33
306
+ %284 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
307
+ %.not.i1 = icmp eq i32 %284, 0, !dbg !33
308
+ %285 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
309
+ %.not1.i4 = icmp eq i32 %285, 0, !dbg !33
310
+ br i1 %.not.i1, label %291, label %286, !dbg !33
311
+
312
+ 286: ; preds = %__nv_sqrtf.exit
313
+ br i1 %.not1.i4, label %289, label %287, !dbg !33
314
+
315
+ 287: ; preds = %286
316
+ %288 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %269) #6, !dbg !33
317
+ br label %__nv_sqrtf.exit5, !dbg !33
318
+
319
+ 289: ; preds = %286
320
+ %290 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %269) #6, !dbg !33
321
+ br label %__nv_sqrtf.exit5, !dbg !33
322
+
323
+ 291: ; preds = %__nv_sqrtf.exit
324
+ br i1 %.not1.i4, label %294, label %292, !dbg !33
325
+
326
+ 292: ; preds = %291
327
+ %293 = tail call float @llvm.nvvm.sqrt.rn.f(float %269) #6, !dbg !33
328
+ br label %__nv_sqrtf.exit5, !dbg !33
329
+
330
+ 294: ; preds = %291
331
+ %295 = tail call float @llvm.nvvm.sqrt.approx.f(float %269) #6, !dbg !33
332
+ br label %__nv_sqrtf.exit5, !dbg !33
333
+
334
+ __nv_sqrtf.exit5: ; preds = %287, %289, %292, %294
335
+ %.0.i3 = phi float [ %288, %287 ], [ %290, %289 ], [ %293, %292 ], [ %295, %294 ], !dbg !33
336
+ %296 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
337
+ %.not.i6 = icmp eq i32 %296, 0, !dbg !33
338
+ %297 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
339
+ %.not1.i9 = icmp eq i32 %297, 0, !dbg !33
340
+ br i1 %.not.i6, label %303, label %298, !dbg !33
341
+
342
+ 298: ; preds = %__nv_sqrtf.exit5
343
+ br i1 %.not1.i9, label %301, label %299, !dbg !33
344
+
345
+ 299: ; preds = %298
346
+ %300 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %270) #6, !dbg !33
347
+ br label %__nv_sqrtf.exit10, !dbg !33
348
+
349
+ 301: ; preds = %298
350
+ %302 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %270) #6, !dbg !33
351
+ br label %__nv_sqrtf.exit10, !dbg !33
352
+
353
+ 303: ; preds = %__nv_sqrtf.exit5
354
+ br i1 %.not1.i9, label %306, label %304, !dbg !33
355
+
356
+ 304: ; preds = %303
357
+ %305 = tail call float @llvm.nvvm.sqrt.rn.f(float %270) #6, !dbg !33
358
+ br label %__nv_sqrtf.exit10, !dbg !33
359
+
360
+ 306: ; preds = %303
361
+ %307 = tail call float @llvm.nvvm.sqrt.approx.f(float %270) #6, !dbg !33
362
+ br label %__nv_sqrtf.exit10, !dbg !33
363
+
364
+ __nv_sqrtf.exit10: ; preds = %299, %301, %304, %306
365
+ %.0.i8 = phi float [ %300, %299 ], [ %302, %301 ], [ %305, %304 ], [ %307, %306 ], !dbg !33
366
+ %308 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
367
+ %.not.i11 = icmp eq i32 %308, 0, !dbg !33
368
+ %309 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
369
+ %.not1.i14 = icmp eq i32 %309, 0, !dbg !33
370
+ br i1 %.not.i11, label %315, label %310, !dbg !33
371
+
372
+ 310: ; preds = %__nv_sqrtf.exit10
373
+ br i1 %.not1.i14, label %313, label %311, !dbg !33
374
+
375
+ 311: ; preds = %310
376
+ %312 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %271) #6, !dbg !33
377
+ br label %__nv_sqrtf.exit15, !dbg !33
378
+
379
+ 313: ; preds = %310
380
+ %314 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %271) #6, !dbg !33
381
+ br label %__nv_sqrtf.exit15, !dbg !33
382
+
383
+ 315: ; preds = %__nv_sqrtf.exit10
384
+ br i1 %.not1.i14, label %318, label %316, !dbg !33
385
+
386
+ 316: ; preds = %315
387
+ %317 = tail call float @llvm.nvvm.sqrt.rn.f(float %271) #6, !dbg !33
388
+ br label %__nv_sqrtf.exit15, !dbg !33
389
+
390
+ 318: ; preds = %315
391
+ %319 = tail call float @llvm.nvvm.sqrt.approx.f(float %271) #6, !dbg !33
392
+ br label %__nv_sqrtf.exit15, !dbg !33
393
+
394
+ __nv_sqrtf.exit15: ; preds = %311, %313, %316, %318
395
+ %.0.i13 = phi float [ %312, %311 ], [ %314, %313 ], [ %317, %316 ], [ %319, %318 ], !dbg !33
396
+ %320 = and i32 %11, 7, !dbg !16
397
+ %321 = zext nneg i32 %320 to i64, !dbg !17
398
+ %322 = or disjoint i64 %10, %321, !dbg !17
399
+ %323 = getelementptr float, ptr addrspace(1) %2, i64 %322, !dbg !34
400
+ %324 = getelementptr bfloat, ptr addrspace(1) %1, i64 %34, !dbg !35
401
+ %325 = getelementptr bfloat, ptr addrspace(1) %324, i64 %39, !dbg !36
402
+ %326 = getelementptr bfloat, ptr addrspace(1) %1, i64 %33, !dbg !35
403
+ %327 = getelementptr bfloat, ptr addrspace(1) %326, i64 %39, !dbg !36
404
+ %328 = getelementptr bfloat, ptr addrspace(1) %1, i64 %32, !dbg !35
405
+ %329 = getelementptr bfloat, ptr addrspace(1) %328, i64 %39, !dbg !36
406
+ %330 = getelementptr bfloat, ptr addrspace(1) %1, i64 %31, !dbg !35
407
+ %331 = getelementptr bfloat, ptr addrspace(1) %330, i64 %39, !dbg !36
408
+ %332 = icmp slt i64 %322, %24, !dbg !18
409
+ %333 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37
410
+ %334 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37
411
+ %335 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37
412
+ %336 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37
413
+ %337 = fmul float %333, %112, !dbg !38
414
+ %338 = fmul float %333, %113, !dbg !38
415
+ %339 = fmul float %333, %114, !dbg !38
416
+ %340 = fmul float %333, %115, !dbg !38
417
+ %341 = fmul float %333, %116, !dbg !38
418
+ %342 = fmul float %333, %117, !dbg !38
419
+ %343 = fmul float %333, %118, !dbg !38
420
+ %344 = fmul float %333, %119, !dbg !38
421
+ %345 = fmul float %334, %120, !dbg !38
422
+ %346 = fmul float %334, %121, !dbg !38
423
+ %347 = fmul float %334, %122, !dbg !38
424
+ %348 = fmul float %334, %123, !dbg !38
425
+ %349 = fmul float %334, %124, !dbg !38
426
+ %350 = fmul float %334, %125, !dbg !38
427
+ %351 = fmul float %334, %126, !dbg !38
428
+ %352 = fmul float %334, %127, !dbg !38
429
+ %353 = fmul float %335, %128, !dbg !38
430
+ %354 = fmul float %335, %129, !dbg !38
431
+ %355 = fmul float %335, %130, !dbg !38
432
+ %356 = fmul float %335, %131, !dbg !38
433
+ %357 = fmul float %335, %132, !dbg !38
434
+ %358 = fmul float %335, %133, !dbg !38
435
+ %359 = fmul float %335, %134, !dbg !38
436
+ %360 = fmul float %335, %135, !dbg !38
437
+ %361 = fmul float %336, %136, !dbg !38
438
+ %362 = fmul float %336, %137, !dbg !38
439
+ %363 = fmul float %336, %138, !dbg !38
440
+ %364 = fmul float %336, %139, !dbg !38
441
+ %365 = fmul float %336, %140, !dbg !38
442
+ %366 = fmul float %336, %141, !dbg !38
443
+ %367 = fmul float %336, %142, !dbg !38
444
+ %368 = fmul float %336, %143, !dbg !38
445
+ %369 = fptrunc float %337 to bfloat, !dbg !39
446
+ %370 = fptrunc float %338 to bfloat, !dbg !39
447
+ %371 = fptrunc float %339 to bfloat, !dbg !39
448
+ %372 = fptrunc float %340 to bfloat, !dbg !39
449
+ %373 = fptrunc float %341 to bfloat, !dbg !39
450
+ %374 = fptrunc float %342 to bfloat, !dbg !39
451
+ %375 = fptrunc float %343 to bfloat, !dbg !39
452
+ %376 = fptrunc float %344 to bfloat, !dbg !39
453
+ %377 = fptrunc float %345 to bfloat, !dbg !39
454
+ %378 = fptrunc float %346 to bfloat, !dbg !39
455
+ %379 = fptrunc float %347 to bfloat, !dbg !39
456
+ %380 = fptrunc float %348 to bfloat, !dbg !39
457
+ %381 = fptrunc float %349 to bfloat, !dbg !39
458
+ %382 = fptrunc float %350 to bfloat, !dbg !39
459
+ %383 = fptrunc float %351 to bfloat, !dbg !39
460
+ %384 = fptrunc float %352 to bfloat, !dbg !39
461
+ %385 = fptrunc float %353 to bfloat, !dbg !39
462
+ %386 = fptrunc float %354 to bfloat, !dbg !39
463
+ %387 = fptrunc float %355 to bfloat, !dbg !39
464
+ %388 = fptrunc float %356 to bfloat, !dbg !39
465
+ %389 = fptrunc float %357 to bfloat, !dbg !39
466
+ %390 = fptrunc float %358 to bfloat, !dbg !39
467
+ %391 = fptrunc float %359 to bfloat, !dbg !39
468
+ %392 = fptrunc float %360 to bfloat, !dbg !39
469
+ %393 = fptrunc float %361 to bfloat, !dbg !39
470
+ %394 = fptrunc float %362 to bfloat, !dbg !39
471
+ %395 = fptrunc float %363 to bfloat, !dbg !39
472
+ %396 = fptrunc float %364 to bfloat, !dbg !39
473
+ %397 = fptrunc float %365 to bfloat, !dbg !39
474
+ %398 = fptrunc float %366 to bfloat, !dbg !39
475
+ %399 = fptrunc float %367 to bfloat, !dbg !39
476
+ %400 = fptrunc float %368 to bfloat, !dbg !39
477
+ %401 = insertelement <2 x bfloat> poison, bfloat %369, i64 0, !dbg !40
478
+ %402 = insertelement <2 x bfloat> %401, bfloat %370, i64 1, !dbg !40
479
+ %403 = bitcast <2 x bfloat> %402 to i32, !dbg !40
480
+ %404 = insertelement <2 x bfloat> poison, bfloat %371, i64 0, !dbg !40
481
+ %405 = insertelement <2 x bfloat> %404, bfloat %372, i64 1, !dbg !40
482
+ %406 = bitcast <2 x bfloat> %405 to i32, !dbg !40
483
+ %407 = insertelement <2 x bfloat> poison, bfloat %373, i64 0, !dbg !40
484
+ %408 = insertelement <2 x bfloat> %407, bfloat %374, i64 1, !dbg !40
485
+ %409 = bitcast <2 x bfloat> %408 to i32, !dbg !40
486
+ %410 = insertelement <2 x bfloat> poison, bfloat %375, i64 0, !dbg !40
487
+ %411 = insertelement <2 x bfloat> %410, bfloat %376, i64 1, !dbg !40
488
+ %412 = bitcast <2 x bfloat> %411 to i32, !dbg !40
489
+ 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
490
+ %413 = insertelement <2 x bfloat> poison, bfloat %377, i64 0, !dbg !40
491
+ %414 = insertelement <2 x bfloat> %413, bfloat %378, i64 1, !dbg !40
492
+ %415 = bitcast <2 x bfloat> %414 to i32, !dbg !40
493
+ %416 = insertelement <2 x bfloat> poison, bfloat %379, i64 0, !dbg !40
494
+ %417 = insertelement <2 x bfloat> %416, bfloat %380, i64 1, !dbg !40
495
+ %418 = bitcast <2 x bfloat> %417 to i32, !dbg !40
496
+ %419 = insertelement <2 x bfloat> poison, bfloat %381, i64 0, !dbg !40
497
+ %420 = insertelement <2 x bfloat> %419, bfloat %382, i64 1, !dbg !40
498
+ %421 = bitcast <2 x bfloat> %420 to i32, !dbg !40
499
+ %422 = insertelement <2 x bfloat> poison, bfloat %383, i64 0, !dbg !40
500
+ %423 = insertelement <2 x bfloat> %422, bfloat %384, i64 1, !dbg !40
501
+ %424 = bitcast <2 x bfloat> %423 to i32, !dbg !40
502
+ 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
503
+ %425 = insertelement <2 x bfloat> poison, bfloat %385, i64 0, !dbg !40
504
+ %426 = insertelement <2 x bfloat> %425, bfloat %386, i64 1, !dbg !40
505
+ %427 = bitcast <2 x bfloat> %426 to i32, !dbg !40
506
+ %428 = insertelement <2 x bfloat> poison, bfloat %387, i64 0, !dbg !40
507
+ %429 = insertelement <2 x bfloat> %428, bfloat %388, i64 1, !dbg !40
508
+ %430 = bitcast <2 x bfloat> %429 to i32, !dbg !40
509
+ %431 = insertelement <2 x bfloat> poison, bfloat %389, i64 0, !dbg !40
510
+ %432 = insertelement <2 x bfloat> %431, bfloat %390, i64 1, !dbg !40
511
+ %433 = bitcast <2 x bfloat> %432 to i32, !dbg !40
512
+ %434 = insertelement <2 x bfloat> poison, bfloat %391, i64 0, !dbg !40
513
+ %435 = insertelement <2 x bfloat> %434, bfloat %392, i64 1, !dbg !40
514
+ %436 = bitcast <2 x bfloat> %435 to i32, !dbg !40
515
+ 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
516
+ %437 = insertelement <2 x bfloat> poison, bfloat %393, i64 0, !dbg !40
517
+ %438 = insertelement <2 x bfloat> %437, bfloat %394, i64 1, !dbg !40
518
+ %439 = bitcast <2 x bfloat> %438 to i32, !dbg !40
519
+ %440 = insertelement <2 x bfloat> poison, bfloat %395, i64 0, !dbg !40
520
+ %441 = insertelement <2 x bfloat> %440, bfloat %396, i64 1, !dbg !40
521
+ %442 = bitcast <2 x bfloat> %441 to i32, !dbg !40
522
+ %443 = insertelement <2 x bfloat> poison, bfloat %397, i64 0, !dbg !40
523
+ %444 = insertelement <2 x bfloat> %443, bfloat %398, i64 1, !dbg !40
524
+ %445 = bitcast <2 x bfloat> %444 to i32, !dbg !40
525
+ %446 = insertelement <2 x bfloat> poison, bfloat %399, i64 0, !dbg !40
526
+ %447 = insertelement <2 x bfloat> %446, bfloat %400, i64 1, !dbg !40
527
+ %448 = bitcast <2 x bfloat> %447 to i32, !dbg !40
528
+ 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
529
+ %449 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %12, !dbg !41
530
+ %450 = bitcast float %333 to i32, !dbg !41
531
+ %451 = bitcast float %334 to i32, !dbg !41
532
+ %452 = bitcast float %335 to i32, !dbg !41
533
+ %453 = bitcast float %336 to i32, !dbg !41
534
+ 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
535
+ tail call void @llvm.nvvm.bar.warp.sync(i32 -1), !dbg !41
536
+ %454 = shl nuw nsw i32 %11, 1, !dbg !41
537
+ %455 = and i32 %454, 12, !dbg !41
538
+ %456 = shl nuw nsw i32 %11, 4, !dbg !41
539
+ %457 = and i32 %456, 16, !dbg !41
540
+ %458 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %457, !dbg !41
541
+ %459 = getelementptr inbounds nuw i8, ptr addrspace(3) %458, i32 %455, !dbg !41
542
+ %460 = load i32, ptr addrspace(3) %459, align 4, !dbg !41
543
+ %461 = and i32 %11, 24, !dbg !41
544
+ %462 = icmp eq i32 %461, 0, !dbg !41
545
+ %463 = and i1 %462, %332, !dbg !41
546
+ 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
547
+ ret void, !dbg !42
548
+ }
549
+
550
+ ; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
551
+ declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1
552
+
553
+ ; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
554
+ declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1
555
+
556
+ ; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite)
557
+ declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2
558
+
559
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
560
+ declare float @llvm.nvvm.div.full(float, float) #3
561
+
562
+ ; Function Attrs: convergent nocallback nounwind
563
+ declare void @llvm.nvvm.bar.warp.sync(i32) #4
564
+
565
+ declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5
566
+
567
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
568
+ declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3
569
+
570
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
571
+ declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3
572
+
573
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
574
+ declare float @llvm.nvvm.sqrt.rn.f(float) #3
575
+
576
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
577
+ declare float @llvm.nvvm.sqrt.approx.f(float) #3
578
+
579
+ attributes #0 = { nounwind "nvvm.reqntid"="32" }
580
+ attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) }
581
+ attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) }
582
+ attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) }
583
+ attributes #4 = { convergent nocallback nounwind }
584
+ 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" }
585
+ attributes #6 = { nounwind }
586
+
587
+ !llvm.dbg.cu = !{!0}
588
+ !llvm.module.flags = !{!2, !3}
589
+ !llvm.ident = !{!4}
590
+
591
+ !0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly)
592
+ !1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm")
593
+ !2 = !{i32 2, !"Debug Info Version", i32 3}
594
+ !3 = !{i32 4, !"nvvm-reflect-ftz", i32 1}
595
+ !4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"}
596
+ !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)
597
+ !6 = !DISubroutineType(cc: DW_CC_normal, types: !7)
598
+ !7 = !{null, !8, !8, !10, !11, !12, !8, !8}
599
+ !8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1)
600
+ !9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed)
601
+ !10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1)
602
+ !11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float)
603
+ !12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
604
+ !13 = !DILocation(line: 70, column: 24, scope: !5)
605
+ !14 = !DILocation(line: 70, column: 30, scope: !5)
606
+ !15 = !DILocation(line: 71, column: 16, scope: !5)
607
+ !16 = !DILocation(line: 71, column: 34, scope: !5)
608
+ !17 = !DILocation(line: 71, column: 21, scope: !5)
609
+ !18 = !DILocation(line: 73, column: 16, scope: !5)
610
+ !19 = !DILocation(line: 74, column: 30, scope: !5)
611
+ !20 = !DILocation(line: 75, column: 29, scope: !5)
612
+ !21 = !DILocation(line: 75, column: 14, scope: !5)
613
+ !22 = !DILocation(line: 75, column: 33, scope: !5)
614
+ !23 = !DILocation(line: 79, column: 18, scope: !5)
615
+ !24 = !DILocation(line: 79, column: 47, scope: !5)
616
+ !25 = !DILocation(line: 80, column: 38, scope: !5)
617
+ !26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29)
618
+ !27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0)
619
+ !28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language")
620
+ !29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30)
621
+ !30 = !DILocation(line: 80, column: 43, scope: !31)
622
+ !31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0)
623
+ !32 = !DILocation(line: 80, column: 48, scope: !5)
624
+ !33 = !DILocation(line: 80, column: 25, scope: !5)
625
+ !34 = !DILocation(line: 77, column: 20, scope: !5)
626
+ !35 = !DILocation(line: 76, column: 14, scope: !5)
627
+ !36 = !DILocation(line: 76, column: 33, scope: !5)
628
+ !37 = !DILocation(line: 80, column: 17, scope: !5)
629
+ !38 = !DILocation(line: 81, column: 16, scope: !5)
630
+ !39 = !DILocation(line: 83, column: 25, scope: !5)
631
+ !40 = !DILocation(line: 83, column: 18, scope: !5)
632
+ !41 = !DILocation(line: 84, column: 21, scope: !5)
633
+ !42 = !DILocation(line: 84, column: 4, scope: !5)
triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ptx ADDED
@@ -0,0 +1,642 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ //
2
+ // Generated by LLVM NVPTX Back-End
3
+ //
4
+
5
+ .version 9.1
6
+ .target sm_100a
7
+ .address_size 64
8
+
9
+ // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel
10
+ .extern .shared .align 16 .b8 global_smem[];
11
+ .global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90};
12
+ .global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84};
13
+ // @l2norm_fwd_kernel
14
+ .visible .entry l2norm_fwd_kernel(
15
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0,
16
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1,
17
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2,
18
+ .param .f32 l2norm_fwd_kernel_param_3,
19
+ .param .u32 l2norm_fwd_kernel_param_4,
20
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5,
21
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6
22
+ )
23
+ .reqntid 32
24
+ {
25
+ .reg .pred %p<8>;
26
+ .reg .b16 %rs<33>;
27
+ .reg .b32 %r<196>;
28
+ .reg .b64 %rd<39>;
29
+ .loc 1 69 0 // l2norm.py:69:0
30
+ $L__func_begin0:
31
+ .loc 1 69 0 // l2norm.py:69:0
32
+
33
+ // %bb.0: // %__nv_sqrtf.exit
34
+ ld.param.b64 %rd10, [l2norm_fwd_kernel_param_0];
35
+ ld.param.b64 %rd11, [l2norm_fwd_kernel_param_1];
36
+ $L__tmp0:
37
+ .loc 1 70 24 // l2norm.py:70:24
38
+ mov.u32 %r40, %ctaid.x;
39
+ ld.param.b64 %rd12, [l2norm_fwd_kernel_param_2];
40
+ .loc 1 71 16 // l2norm.py:71:16
41
+ mul.wide.u32 %rd13, %r40, 8;
42
+ ld.param.b32 %r41, [l2norm_fwd_kernel_param_3];
43
+ ld.param.s32 %rd14, [l2norm_fwd_kernel_param_4];
44
+ .loc 1 71 34 // l2norm.py:71:34
45
+ mov.u32 %r42, %tid.x;
46
+ and.b32 %r43, %r42, 16;
47
+ bfe.u32 %r44, %r42, 4, 1;
48
+ or.b32 %r45, %r44, 2;
49
+ or.b32 %r46, %r44, 4;
50
+ or.b32 %r47, %r44, 6;
51
+ .loc 1 71 21 // l2norm.py:71:21
52
+ cvt.u64.u32 %rd15, %r44;
53
+ cvt.u64.u32 %rd16, %r45;
54
+ cvt.u64.u32 %rd17, %r46;
55
+ cvt.u64.u32 %rd18, %r47;
56
+ or.b64 %rd19, %rd13, %rd15;
57
+ or.b64 %rd20, %rd13, %rd16;
58
+ or.b64 %rd21, %rd13, %rd17;
59
+ or.b64 %rd22, %rd13, %rd18;
60
+ .loc 1 73 16 // l2norm.py:73:16
61
+ setp.lt.s64 %p1, %rd19, %rd14;
62
+ setp.lt.s64 %p2, %rd20, %rd14;
63
+ setp.lt.s64 %p3, %rd21, %rd14;
64
+ setp.lt.s64 %p4, %rd22, %rd14;
65
+ .loc 1 74 30 // l2norm.py:74:30
66
+ shl.b32 %r48, %r42, 3;
67
+ and.b32 %r49, %r48, 120;
68
+ .loc 1 75 14 // l2norm.py:75:14
69
+ shl.b64 %rd23, %rd19, 8;
70
+ add.s64 %rd24, %rd10, %rd23;
71
+ shl.b64 %rd25, %rd20, 8;
72
+ add.s64 %rd26, %rd10, %rd25;
73
+ shl.b64 %rd27, %rd21, 8;
74
+ add.s64 %rd28, %rd10, %rd27;
75
+ shl.b64 %rd29, %rd22, 8;
76
+ add.s64 %rd30, %rd10, %rd29;
77
+ .loc 1 75 33 // l2norm.py:75:33
78
+ mul.wide.u32 %rd31, %r49, 2;
79
+ add.s64 %rd1, %rd24, %rd31;
80
+ add.s64 %rd2, %rd26, %rd31;
81
+ add.s64 %rd3, %rd28, %rd31;
82
+ add.s64 %rd4, %rd30, %rd31;
83
+ mov.b32 %r5, 0;
84
+ .loc 1 79 18 // l2norm.py:79:18
85
+ // begin inline asm
86
+ mov.u32 %r1, %r5;
87
+ mov.u32 %r2, %r5;
88
+ mov.u32 %r3, %r5;
89
+ mov.u32 %r4, %r5;
90
+ @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ];
91
+ // end inline asm
92
+ mov.b32 {%rs1, %rs2}, %r1;
93
+ mov.b32 {%rs3, %rs4}, %r2;
94
+ mov.b32 {%rs5, %rs6}, %r3;
95
+ mov.b32 {%rs7, %rs8}, %r4;
96
+ // begin inline asm
97
+ mov.u32 %r6, %r5;
98
+ mov.u32 %r7, %r5;
99
+ mov.u32 %r8, %r5;
100
+ mov.u32 %r9, %r5;
101
+ @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ];
102
+ // end inline asm
103
+ mov.b32 {%rs9, %rs10}, %r6;
104
+ mov.b32 {%rs11, %rs12}, %r7;
105
+ mov.b32 {%rs13, %rs14}, %r8;
106
+ mov.b32 {%rs15, %rs16}, %r9;
107
+ // begin inline asm
108
+ mov.u32 %r10, %r5;
109
+ mov.u32 %r11, %r5;
110
+ mov.u32 %r12, %r5;
111
+ mov.u32 %r13, %r5;
112
+ @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ];
113
+ // end inline asm
114
+ mov.b32 {%rs17, %rs18}, %r10;
115
+ mov.b32 {%rs19, %rs20}, %r11;
116
+ mov.b32 {%rs21, %rs22}, %r12;
117
+ mov.b32 {%rs23, %rs24}, %r13;
118
+ // begin inline asm
119
+ mov.u32 %r14, %r5;
120
+ mov.u32 %r15, %r5;
121
+ mov.u32 %r16, %r5;
122
+ mov.u32 %r17, %r5;
123
+ @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ];
124
+ // end inline asm
125
+ mov.b32 {%rs25, %rs26}, %r14;
126
+ mov.b32 {%rs27, %rs28}, %r15;
127
+ mov.b32 {%rs29, %rs30}, %r16;
128
+ mov.b32 {%rs31, %rs32}, %r17;
129
+ .loc 1 79 47 // l2norm.py:79:47
130
+ cvt.f32.bf16 %r50, %rs1;
131
+ cvt.f32.bf16 %r51, %rs2;
132
+ cvt.f32.bf16 %r52, %rs3;
133
+ cvt.f32.bf16 %r53, %rs4;
134
+ cvt.f32.bf16 %r54, %rs5;
135
+ cvt.f32.bf16 %r55, %rs6;
136
+ cvt.f32.bf16 %r56, %rs7;
137
+ cvt.f32.bf16 %r57, %rs8;
138
+ cvt.f32.bf16 %r58, %rs9;
139
+ cvt.f32.bf16 %r59, %rs10;
140
+ cvt.f32.bf16 %r60, %rs11;
141
+ cvt.f32.bf16 %r61, %rs12;
142
+ cvt.f32.bf16 %r62, %rs13;
143
+ cvt.f32.bf16 %r63, %rs14;
144
+ cvt.f32.bf16 %r64, %rs15;
145
+ cvt.f32.bf16 %r65, %rs16;
146
+ cvt.f32.bf16 %r66, %rs17;
147
+ cvt.f32.bf16 %r67, %rs18;
148
+ cvt.f32.bf16 %r68, %rs19;
149
+ cvt.f32.bf16 %r69, %rs20;
150
+ cvt.f32.bf16 %r70, %rs21;
151
+ cvt.f32.bf16 %r71, %rs22;
152
+ cvt.f32.bf16 %r72, %rs23;
153
+ cvt.f32.bf16 %r73, %rs24;
154
+ cvt.f32.bf16 %r74, %rs25;
155
+ cvt.f32.bf16 %r75, %rs26;
156
+ cvt.f32.bf16 %r76, %rs27;
157
+ cvt.f32.bf16 %r77, %rs28;
158
+ cvt.f32.bf16 %r78, %rs29;
159
+ cvt.f32.bf16 %r79, %rs30;
160
+ cvt.f32.bf16 %r80, %rs31;
161
+ cvt.f32.bf16 %r81, %rs32;
162
+ .loc 1 80 38 // l2norm.py:80:38
163
+ mul.f32 %r82, %r51, %r51;
164
+ mul.f32 %r83, %r59, %r59;
165
+ mul.f32 %r84, %r67, %r67;
166
+ mul.f32 %r85, %r75, %r75;
167
+ $L__tmp1:
168
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
169
+ fma.rn.f32.bf16 %r86, %rs1, %rs1, %r82;
170
+ fma.rn.f32.bf16 %r87, %rs3, %rs3, %r86;
171
+ fma.rn.f32.bf16 %r88, %rs4, %rs4, %r87;
172
+ fma.rn.f32.bf16 %r89, %rs5, %rs5, %r88;
173
+ fma.rn.f32.bf16 %r90, %rs6, %rs6, %r89;
174
+ fma.rn.f32.bf16 %r91, %rs7, %rs7, %r90;
175
+ fma.rn.f32.bf16 %r92, %rs8, %rs8, %r91;
176
+ fma.rn.f32.bf16 %r93, %rs9, %rs9, %r83;
177
+ fma.rn.f32.bf16 %r94, %rs11, %rs11, %r93;
178
+ fma.rn.f32.bf16 %r95, %rs12, %rs12, %r94;
179
+ fma.rn.f32.bf16 %r96, %rs13, %rs13, %r95;
180
+ fma.rn.f32.bf16 %r97, %rs14, %rs14, %r96;
181
+ fma.rn.f32.bf16 %r98, %rs15, %rs15, %r97;
182
+ fma.rn.f32.bf16 %r99, %rs16, %rs16, %r98;
183
+ fma.rn.f32.bf16 %r100, %rs17, %rs17, %r84;
184
+ fma.rn.f32.bf16 %r101, %rs19, %rs19, %r100;
185
+ fma.rn.f32.bf16 %r102, %rs20, %rs20, %r101;
186
+ fma.rn.f32.bf16 %r103, %rs21, %rs21, %r102;
187
+ fma.rn.f32.bf16 %r104, %rs22, %rs22, %r103;
188
+ fma.rn.f32.bf16 %r105, %rs23, %rs23, %r104;
189
+ fma.rn.f32.bf16 %r106, %rs24, %rs24, %r105;
190
+ fma.rn.f32.bf16 %r107, %rs25, %rs25, %r85;
191
+ fma.rn.f32.bf16 %r108, %rs27, %rs27, %r107;
192
+ fma.rn.f32.bf16 %r109, %rs28, %rs28, %r108;
193
+ fma.rn.f32.bf16 %r110, %rs29, %rs29, %r109;
194
+ fma.rn.f32.bf16 %r111, %rs30, %rs30, %r110;
195
+ fma.rn.f32.bf16 %r112, %rs31, %rs31, %r111;
196
+ fma.rn.f32.bf16 %r113, %rs32, %rs32, %r112;
197
+ $L__tmp2:
198
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
199
+ shfl.sync.bfly.b32 %r114, %r92, 8, 31, -1;
200
+ $L__tmp3:
201
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
202
+ add.f32 %r115, %r92, %r114;
203
+ $L__tmp4:
204
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
205
+ shfl.sync.bfly.b32 %r116, %r115, 4, 31, -1;
206
+ $L__tmp5:
207
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
208
+ add.f32 %r117, %r115, %r116;
209
+ $L__tmp6:
210
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
211
+ shfl.sync.bfly.b32 %r118, %r117, 2, 31, -1;
212
+ $L__tmp7:
213
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
214
+ add.f32 %r119, %r117, %r118;
215
+ $L__tmp8:
216
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
217
+ shfl.sync.bfly.b32 %r120, %r119, 1, 31, -1;
218
+ $L__tmp9:
219
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
220
+ add.f32 %r121, %r119, %r120;
221
+ $L__tmp10:
222
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
223
+ shfl.sync.bfly.b32 %r122, %r99, 8, 31, -1;
224
+ $L__tmp11:
225
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
226
+ add.f32 %r123, %r99, %r122;
227
+ $L__tmp12:
228
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
229
+ shfl.sync.bfly.b32 %r124, %r123, 4, 31, -1;
230
+ $L__tmp13:
231
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
232
+ add.f32 %r125, %r123, %r124;
233
+ $L__tmp14:
234
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
235
+ shfl.sync.bfly.b32 %r126, %r125, 2, 31, -1;
236
+ $L__tmp15:
237
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
238
+ add.f32 %r127, %r125, %r126;
239
+ $L__tmp16:
240
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
241
+ shfl.sync.bfly.b32 %r128, %r127, 1, 31, -1;
242
+ $L__tmp17:
243
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
244
+ add.f32 %r129, %r127, %r128;
245
+ $L__tmp18:
246
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
247
+ shfl.sync.bfly.b32 %r130, %r106, 8, 31, -1;
248
+ $L__tmp19:
249
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
250
+ add.f32 %r131, %r106, %r130;
251
+ $L__tmp20:
252
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
253
+ shfl.sync.bfly.b32 %r132, %r131, 4, 31, -1;
254
+ $L__tmp21:
255
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
256
+ add.f32 %r133, %r131, %r132;
257
+ $L__tmp22:
258
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
259
+ shfl.sync.bfly.b32 %r134, %r133, 2, 31, -1;
260
+ $L__tmp23:
261
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
262
+ add.f32 %r135, %r133, %r134;
263
+ $L__tmp24:
264
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
265
+ shfl.sync.bfly.b32 %r136, %r135, 1, 31, -1;
266
+ $L__tmp25:
267
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
268
+ add.f32 %r137, %r135, %r136;
269
+ $L__tmp26:
270
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
271
+ shfl.sync.bfly.b32 %r138, %r113, 8, 31, -1;
272
+ $L__tmp27:
273
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
274
+ add.f32 %r139, %r113, %r138;
275
+ $L__tmp28:
276
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
277
+ shfl.sync.bfly.b32 %r140, %r139, 4, 31, -1;
278
+ $L__tmp29:
279
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
280
+ add.f32 %r141, %r139, %r140;
281
+ $L__tmp30:
282
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
283
+ shfl.sync.bfly.b32 %r142, %r141, 2, 31, -1;
284
+ $L__tmp31:
285
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
286
+ add.f32 %r143, %r141, %r142;
287
+ $L__tmp32:
288
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
289
+ shfl.sync.bfly.b32 %r144, %r143, 1, 31, -1;
290
+ $L__tmp33:
291
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
292
+ add.f32 %r145, %r143, %r144;
293
+ $L__tmp34:
294
+ .loc 1 80 48 // l2norm.py:80:48
295
+ add.f32 %r146, %r41, %r121;
296
+ add.f32 %r147, %r41, %r129;
297
+ add.f32 %r148, %r41, %r137;
298
+ add.f32 %r149, %r41, %r145;
299
+ .loc 1 80 25 // l2norm.py:80:25
300
+ sqrt.approx.ftz.f32 %r150, %r146;
301
+ sqrt.approx.ftz.f32 %r151, %r147;
302
+ sqrt.approx.ftz.f32 %r152, %r148;
303
+ sqrt.approx.ftz.f32 %r153, %r149;
304
+ .loc 1 71 34 // l2norm.py:71:34
305
+ and.b32 %r154, %r42, 7;
306
+ .loc 1 71 21 // l2norm.py:71:21
307
+ cvt.u64.u32 %rd32, %r154;
308
+ or.b64 %rd33, %rd13, %rd32;
309
+ .loc 1 77 20 // l2norm.py:77:20
310
+ shl.b64 %rd34, %rd33, 2;
311
+ add.s64 %rd9, %rd12, %rd34;
312
+ .loc 1 76 14 // l2norm.py:76:14
313
+ add.s64 %rd35, %rd11, %rd29;
314
+ .loc 1 76 33 // l2norm.py:76:33
315
+ add.s64 %rd8, %rd35, %rd31;
316
+ .loc 1 76 14 // l2norm.py:76:14
317
+ add.s64 %rd36, %rd11, %rd27;
318
+ .loc 1 76 33 // l2norm.py:76:33
319
+ add.s64 %rd7, %rd36, %rd31;
320
+ .loc 1 76 14 // l2norm.py:76:14
321
+ add.s64 %rd37, %rd11, %rd25;
322
+ .loc 1 76 33 // l2norm.py:76:33
323
+ add.s64 %rd6, %rd37, %rd31;
324
+ .loc 1 76 14 // l2norm.py:76:14
325
+ add.s64 %rd38, %rd11, %rd23;
326
+ .loc 1 76 33 // l2norm.py:76:33
327
+ add.s64 %rd5, %rd38, %rd31;
328
+ .loc 1 73 16 // l2norm.py:73:16
329
+ setp.lt.s64 %p6, %rd33, %rd14;
330
+ mov.b32 %r155, 0f3F800000;
331
+ .loc 1 80 17 // l2norm.py:80:17
332
+ div.full.f32 %r35, %r155, %r150;
333
+ div.full.f32 %r36, %r155, %r151;
334
+ div.full.f32 %r37, %r155, %r152;
335
+ div.full.f32 %r38, %r155, %r153;
336
+ .loc 1 81 16 // l2norm.py:81:16
337
+ mul.f32 %r156, %r35, %r50;
338
+ mul.f32 %r157, %r35, %r51;
339
+ mul.f32 %r158, %r35, %r52;
340
+ mul.f32 %r159, %r35, %r53;
341
+ mul.f32 %r160, %r35, %r54;
342
+ mul.f32 %r161, %r35, %r55;
343
+ mul.f32 %r162, %r35, %r56;
344
+ mul.f32 %r163, %r35, %r57;
345
+ mul.f32 %r164, %r36, %r58;
346
+ mul.f32 %r165, %r36, %r59;
347
+ mul.f32 %r166, %r36, %r60;
348
+ mul.f32 %r167, %r36, %r61;
349
+ mul.f32 %r168, %r36, %r62;
350
+ mul.f32 %r169, %r36, %r63;
351
+ mul.f32 %r170, %r36, %r64;
352
+ mul.f32 %r171, %r36, %r65;
353
+ mul.f32 %r172, %r37, %r66;
354
+ mul.f32 %r173, %r37, %r67;
355
+ mul.f32 %r174, %r37, %r68;
356
+ mul.f32 %r175, %r37, %r69;
357
+ mul.f32 %r176, %r37, %r70;
358
+ mul.f32 %r177, %r37, %r71;
359
+ mul.f32 %r178, %r37, %r72;
360
+ mul.f32 %r179, %r37, %r73;
361
+ mul.f32 %r180, %r38, %r74;
362
+ mul.f32 %r181, %r38, %r75;
363
+ mul.f32 %r182, %r38, %r76;
364
+ mul.f32 %r183, %r38, %r77;
365
+ mul.f32 %r184, %r38, %r78;
366
+ mul.f32 %r185, %r38, %r79;
367
+ mul.f32 %r186, %r38, %r80;
368
+ mul.f32 %r187, %r38, %r81;
369
+ .loc 1 83 18 // l2norm.py:83:18
370
+ cvt.rn.bf16x2.f32 %r18, %r157, %r156;
371
+ cvt.rn.bf16x2.f32 %r19, %r159, %r158;
372
+ cvt.rn.bf16x2.f32 %r20, %r161, %r160;
373
+ cvt.rn.bf16x2.f32 %r21, %r163, %r162;
374
+ // begin inline asm
375
+ @%p1 st.global.v4.b32 [ %rd5 + 0 ], { %r18, %r19, %r20, %r21 };
376
+ // end inline asm
377
+ cvt.rn.bf16x2.f32 %r22, %r165, %r164;
378
+ cvt.rn.bf16x2.f32 %r23, %r167, %r166;
379
+ cvt.rn.bf16x2.f32 %r24, %r169, %r168;
380
+ cvt.rn.bf16x2.f32 %r25, %r171, %r170;
381
+ // begin inline asm
382
+ @%p2 st.global.v4.b32 [ %rd6 + 0 ], { %r22, %r23, %r24, %r25 };
383
+ // end inline asm
384
+ cvt.rn.bf16x2.f32 %r26, %r173, %r172;
385
+ cvt.rn.bf16x2.f32 %r27, %r175, %r174;
386
+ cvt.rn.bf16x2.f32 %r28, %r177, %r176;
387
+ cvt.rn.bf16x2.f32 %r29, %r179, %r178;
388
+ // begin inline asm
389
+ @%p3 st.global.v4.b32 [ %rd7 + 0 ], { %r26, %r27, %r28, %r29 };
390
+ // end inline asm
391
+ cvt.rn.bf16x2.f32 %r30, %r181, %r180;
392
+ cvt.rn.bf16x2.f32 %r31, %r183, %r182;
393
+ cvt.rn.bf16x2.f32 %r32, %r185, %r184;
394
+ cvt.rn.bf16x2.f32 %r33, %r187, %r186;
395
+ // begin inline asm
396
+ @%p4 st.global.v4.b32 [ %rd8 + 0 ], { %r30, %r31, %r32, %r33 };
397
+ // end inline asm
398
+ .loc 1 84 21 // l2norm.py:84:21
399
+ mov.b32 %r188, global_smem;
400
+ add.s32 %r34, %r188, %r43;
401
+ // begin inline asm
402
+ st.shared.v4.b32 [ %r34 + 0 ], { %r35, %r36, %r37, %r38 };
403
+ // end inline asm
404
+ bar.warp.sync -1;
405
+ shl.b32 %r189, %r42, 1;
406
+ and.b32 %r190, %r189, 12;
407
+ shl.b32 %r191, %r42, 4;
408
+ and.b32 %r192, %r191, 16;
409
+ add.s32 %r193, %r188, %r192;
410
+ add.s32 %r194, %r193, %r190;
411
+ ld.shared.b32 %r39, [%r194];
412
+ and.b32 %r195, %r42, 24;
413
+ setp.eq.b32 %p7, %r195, 0;
414
+ and.pred %p5, %p7, %p6;
415
+ // begin inline asm
416
+ @%p5 st.global.b32 [ %rd9 + 0 ], { %r39 };
417
+ // end inline asm
418
+ .loc 1 84 4 // l2norm.py:84:4
419
+ ret;
420
+ $L__tmp35:
421
+ $L__func_end0:
422
+ // -- End function
423
+ }
424
+ .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py"
425
+ .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py"
426
+ .section .debug_abbrev
427
+ {
428
+ .b8 1 // Abbreviation Code
429
+ .b8 17 // DW_TAG_compile_unit
430
+ .b8 1 // DW_CHILDREN_yes
431
+ .b8 37 // DW_AT_producer
432
+ .b8 8 // DW_FORM_string
433
+ .b8 19 // DW_AT_language
434
+ .b8 5 // DW_FORM_data2
435
+ .b8 3 // DW_AT_name
436
+ .b8 8 // DW_FORM_string
437
+ .b8 16 // DW_AT_stmt_list
438
+ .b8 6 // DW_FORM_data4
439
+ .b8 27 // DW_AT_comp_dir
440
+ .b8 8 // DW_FORM_string
441
+ .b8 0 // EOM(1)
442
+ .b8 0 // EOM(2)
443
+ .b8 2 // Abbreviation Code
444
+ .b8 46 // DW_TAG_subprogram
445
+ .b8 0 // DW_CHILDREN_no
446
+ .b8 3 // DW_AT_name
447
+ .b8 8 // DW_FORM_string
448
+ .b8 32 // DW_AT_inline
449
+ .b8 11 // DW_FORM_data1
450
+ .b8 0 // EOM(1)
451
+ .b8 0 // EOM(2)
452
+ .b8 3 // Abbreviation Code
453
+ .b8 46 // DW_TAG_subprogram
454
+ .b8 1 // DW_CHILDREN_yes
455
+ .b8 17 // DW_AT_low_pc
456
+ .b8 1 // DW_FORM_addr
457
+ .b8 18 // DW_AT_high_pc
458
+ .b8 1 // DW_FORM_addr
459
+ .b8 49 // DW_AT_abstract_origin
460
+ .b8 19 // DW_FORM_ref4
461
+ .b8 0 // EOM(1)
462
+ .b8 0 // EOM(2)
463
+ .b8 4 // Abbreviation Code
464
+ .b8 29 // DW_TAG_inlined_subroutine
465
+ .b8 1 // DW_CHILDREN_yes
466
+ .b8 49 // DW_AT_abstract_origin
467
+ .b8 19 // DW_FORM_ref4
468
+ .b8 17 // DW_AT_low_pc
469
+ .b8 1 // DW_FORM_addr
470
+ .b8 18 // DW_AT_high_pc
471
+ .b8 1 // DW_FORM_addr
472
+ .b8 88 // DW_AT_call_file
473
+ .b8 11 // DW_FORM_data1
474
+ .b8 89 // DW_AT_call_line
475
+ .b8 11 // DW_FORM_data1
476
+ .b8 87 // DW_AT_call_column
477
+ .b8 11 // DW_FORM_data1
478
+ .b8 0 // EOM(1)
479
+ .b8 0 // EOM(2)
480
+ .b8 5 // Abbreviation Code
481
+ .b8 29 // DW_TAG_inlined_subroutine
482
+ .b8 0 // DW_CHILDREN_no
483
+ .b8 49 // DW_AT_abstract_origin
484
+ .b8 19 // DW_FORM_ref4
485
+ .b8 17 // DW_AT_low_pc
486
+ .b8 1 // DW_FORM_addr
487
+ .b8 18 // DW_AT_high_pc
488
+ .b8 1 // DW_FORM_addr
489
+ .b8 88 // DW_AT_call_file
490
+ .b8 11 // DW_FORM_data1
491
+ .b8 89 // DW_AT_call_line
492
+ .b8 5 // DW_FORM_data2
493
+ .b8 87 // DW_AT_call_column
494
+ .b8 11 // DW_FORM_data1
495
+ .b8 0 // EOM(1)
496
+ .b8 0 // EOM(2)
497
+ .b8 0 // EOM(3)
498
+ }
499
+ .section .debug_info
500
+ {
501
+ .b32 196 // Length of Unit
502
+ .b8 2 // DWARF version number
503
+ .b8 0
504
+ .b32 .debug_abbrev // Offset Into Abbrev. Section
505
+ .b8 8 // Address Size (in bytes)
506
+ .b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit
507
+ .b8 116 // DW_AT_producer
508
+ .b8 114
509
+ .b8 105
510
+ .b8 116
511
+ .b8 111
512
+ .b8 110
513
+ .b8 0
514
+ .b8 2 // DW_AT_language
515
+ .b8 0
516
+ .b8 108 // DW_AT_name
517
+ .b8 50
518
+ .b8 110
519
+ .b8 111
520
+ .b8 114
521
+ .b8 109
522
+ .b8 46
523
+ .b8 112
524
+ .b8 121
525
+ .b8 0
526
+ .b32 .debug_line // DW_AT_stmt_list
527
+ .b8 47 // DW_AT_comp_dir
528
+ .b8 104
529
+ .b8 111
530
+ .b8 109
531
+ .b8 101
532
+ .b8 47
533
+ .b8 115
534
+ .b8 105
535
+ .b8 108
536
+ .b8 118
537
+ .b8 101
538
+ .b8 114
539
+ .b8 47
540
+ .b8 97
541
+ .b8 102
542
+ .b8 102
543
+ .b8 105
544
+ .b8 110
545
+ .b8 101
546
+ .b8 47
547
+ .b8 46
548
+ .b8 118
549
+ .b8 101
550
+ .b8 110
551
+ .b8 118
552
+ .b8 47
553
+ .b8 108
554
+ .b8 105
555
+ .b8 98
556
+ .b8 47
557
+ .b8 112
558
+ .b8 121
559
+ .b8 116
560
+ .b8 104
561
+ .b8 111
562
+ .b8 110
563
+ .b8 51
564
+ .b8 46
565
+ .b8 49
566
+ .b8 50
567
+ .b8 47
568
+ .b8 115
569
+ .b8 105
570
+ .b8 116
571
+ .b8 101
572
+ .b8 45
573
+ .b8 112
574
+ .b8 97
575
+ .b8 99
576
+ .b8 107
577
+ .b8 97
578
+ .b8 103
579
+ .b8 101
580
+ .b8 115
581
+ .b8 47
582
+ .b8 102
583
+ .b8 108
584
+ .b8 97
585
+ .b8 47
586
+ .b8 109
587
+ .b8 111
588
+ .b8 100
589
+ .b8 117
590
+ .b8 108
591
+ .b8 101
592
+ .b8 115
593
+ .b8 47
594
+ .b8 110
595
+ .b8 111
596
+ .b8 114
597
+ .b8 109
598
+ .b8 0
599
+ .b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram
600
+ .b8 108 // DW_AT_name
601
+ .b8 50
602
+ .b8 110
603
+ .b8 111
604
+ .b8 114
605
+ .b8 109
606
+ .b8 95
607
+ .b8 102
608
+ .b8 119
609
+ .b8 100
610
+ .b8 95
611
+ .b8 107
612
+ .b8 101
613
+ .b8 114
614
+ .b8 110
615
+ .b8 101
616
+ .b8 108
617
+ .b8 0
618
+ .b8 1 // DW_AT_inline
619
+ .b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram
620
+ .b64 $L__func_begin0 // DW_AT_low_pc
621
+ .b64 $L__func_end0 // DW_AT_high_pc
622
+ .b32 107 // DW_AT_abstract_origin
623
+ .b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine
624
+ .b32 107 // DW_AT_abstract_origin
625
+ .b64 $L__tmp1 // DW_AT_low_pc
626
+ .b64 $L__tmp34 // DW_AT_high_pc
627
+ .b8 1 // DW_AT_call_file
628
+ .b8 80 // DW_AT_call_line
629
+ .b8 43 // DW_AT_call_column
630
+ .b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine
631
+ .b32 107 // DW_AT_abstract_origin
632
+ .b64 $L__tmp1 // DW_AT_low_pc
633
+ .b64 $L__tmp34 // DW_AT_high_pc
634
+ .b8 2 // DW_AT_call_file
635
+ .b8 37 // DW_AT_call_line
636
+ .b8 1
637
+ .b8 36 // DW_AT_call_column
638
+ .b8 0 // End Of Children Mark
639
+ .b8 0 // End Of Children Mark
640
+ .b8 0 // End Of Children Mark
641
+ }
642
+ .section .debug_macinfo { }
triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.source ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
2
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0)
3
+ #loc38 = loc(unknown)
4
+ #loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0)
5
+ #loc45 = loc("x"(#loc))
6
+ #loc46 = loc("y"(#loc))
7
+ #loc47 = loc("rstd"(#loc))
8
+ #loc48 = loc("eps"(#loc))
9
+ #loc49 = loc("T"(#loc))
10
+ #loc81 = loc("input"(#loc36))
11
+ #loc82 = loc("a"(#loc41))
12
+ #loc83 = loc("b"(#loc41))
13
+ module {
14
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %i_t = tt.get_program_id x : i32 loc(#loc50)
16
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51)
17
+ %o_t = arith.constant 8 : i32 loc(#loc52)
18
+ %o_t_1 = arith.constant 8 : i64 loc(#loc52)
19
+ %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52)
20
+ %o_t_3 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc53)
21
+ %o_t_4 = arith.extsi %o_t_3 : tensor<8xi32> to tensor<8xi64> loc(#loc54)
22
+ %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<8xi64> loc(#loc54)
23
+ %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<8xi64> loc(#loc54)
24
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55)
25
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc56)
26
+ %m_t_7 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc56)
27
+ %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<8xi64> loc(#loc56)
28
+ %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc57)
29
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58)
30
+ %m_x_10 = arith.constant 128 : i32 loc(#loc59)
31
+ %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59)
32
+ %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59)
33
+ %m_x_13 = tt.broadcast %m_x : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc60)
34
+ %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc60)
35
+ %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<8x128xi1> loc(#loc60)
36
+ %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc61)
37
+ %p_x_16 = arith.constant 128 : i32 loc(#loc62)
38
+ %p_x_17 = arith.constant 128 : i64 loc(#loc62)
39
+ %p_x_18 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc62)
40
+ %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<8x1xi64> loc(#loc62)
41
+ %p_x_20 = tt.splat %x : !tt.ptr<bf16> -> tensor<8x1x!tt.ptr<bf16>> loc(#loc63)
42
+ %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<8x1x!tt.ptr<bf16>>, tensor<8x1xi64> loc(#loc63)
43
+ %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64)
44
+ %p_x_23 = tt.broadcast %p_x_21 : tensor<8x1x!tt.ptr<bf16>> -> tensor<8x128x!tt.ptr<bf16>> loc(#loc65)
45
+ %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc65)
46
+ %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<8x128x!tt.ptr<bf16>>, tensor<8x128xi32> loc(#loc65)
47
+ %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc66)
48
+ %p_y_26 = arith.constant 128 : i32 loc(#loc67)
49
+ %p_y_27 = arith.constant 128 : i64 loc(#loc67)
50
+ %p_y_28 = arith.constant dense<128> : tensor<8x1xi64> loc(#loc67)
51
+ %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<8x1xi64> loc(#loc67)
52
+ %p_y_30 = tt.splat %y : !tt.ptr<bf16> -> tensor<8x1x!tt.ptr<bf16>> loc(#loc68)
53
+ %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<8x1x!tt.ptr<bf16>>, tensor<8x1xi64> loc(#loc68)
54
+ %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69)
55
+ %p_y_33 = tt.broadcast %p_y_31 : tensor<8x1x!tt.ptr<bf16>> -> tensor<8x128x!tt.ptr<bf16>> loc(#loc70)
56
+ %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc70)
57
+ %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<8x128x!tt.ptr<bf16>>, tensor<8x128xi32> loc(#loc70)
58
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<8x!tt.ptr<f32>> loc(#loc71)
59
+ %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<8x!tt.ptr<f32>>, tensor<8xi64> loc(#loc71)
60
+ %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72)
61
+ %b_x_37 = arith.constant dense<0.000000e+00> : tensor<8x128xf32> loc(#loc72)
62
+ %b_x_38 = arith.truncf %b_x_37 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc72)
63
+ %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<8x128x!tt.ptr<bf16>> loc(#loc72)
64
+ %b_x_40 = arith.extf %b_x_39 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc73)
65
+ %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<8x128xf32> loc(#loc74)
66
+ %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%b_rstd) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc75)
67
+ %b_rstd_42 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc76)
68
+ %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<8xf32> loc(#loc76)
69
+ %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<8xf32> loc(#loc77)
70
+ %b_rstd_45 = arith.constant 1 : i32 loc(#loc78)
71
+ %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78)
72
+ %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc78)
73
+ %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<8xf32> loc(#loc78)
74
+ %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc79)
75
+ %b_y_49 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc80)
76
+ %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<8x128xf32> loc(#loc80)
77
+ %0 = arith.truncf %b_y_50 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc32)
78
+ tt.store %p_y_35, %0, %m_x_15 : tensor<8x128x!tt.ptr<bf16>> loc(#loc33)
79
+ tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<8x!tt.ptr<f32>> loc(#loc34)
80
+ tt.return loc(#loc35)
81
+ } loc(#loc)
82
+ tt.func private @triton.language.standard.sum__fp32S8_128S_c1_cFalse_cNone(%input: tensor<8x128xf32> loc("input"(#loc36))) -> tensor<8xf32> attributes {noinline = false} {
83
+ %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({
84
+ ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)):
85
+ %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37)
86
+ tt.reduce.return %2 : f32 loc(#loc37)
87
+ }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc37)
88
+ tt.return %0 : tensor<8xf32> loc(#loc39)
89
+ ^bb1: // no predecessors
90
+ %1 = ub.poison : tensor<8xf32> loc(#loc40)
91
+ tt.return %1 : tensor<8xf32> loc(#loc40)
92
+ } loc(#loc36)
93
+ tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} {
94
+ %0 = arith.addf %a, %b : f32 loc(#loc42)
95
+ tt.return %0 : f32 loc(#loc43)
96
+ ^bb1: // no predecessors
97
+ %1 = ub.poison : f32 loc(#loc44)
98
+ tt.return %1 : f32 loc(#loc44)
99
+ } loc(#loc41)
100
+ } loc(#loc)
101
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
102
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
103
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
104
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
105
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
106
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23)
107
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
108
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
109
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
110
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
111
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
112
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
113
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
114
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
115
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37)
116
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
117
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18)
118
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29)
119
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
120
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37)
121
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
122
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
123
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
124
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
125
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
126
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
127
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
128
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
129
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
130
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
131
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
132
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
133
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
134
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
135
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
136
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
137
+ #loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11)
138
+ #loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4)
139
+ #loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
140
+ #loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11)
141
+ #loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4)
142
+ #loc50 = loc("i_t"(#loc1))
143
+ #loc51 = loc("i_t"(#loc2))
144
+ #loc52 = loc("o_t"(#loc3))
145
+ #loc53 = loc("o_t"(#loc4))
146
+ #loc54 = loc("o_t"(#loc5))
147
+ #loc55 = loc("o_d"(#loc6))
148
+ #loc56 = loc("m_t"(#loc7))
149
+ #loc57 = loc("m_x"(#loc8))
150
+ #loc58 = loc("m_x"(#loc9))
151
+ #loc59 = loc("m_x"(#loc10))
152
+ #loc60 = loc("m_x"(#loc11))
153
+ #loc61 = loc("p_x"(#loc12))
154
+ #loc62 = loc("p_x"(#loc13))
155
+ #loc63 = loc("p_x"(#loc14))
156
+ #loc64 = loc("p_x"(#loc15))
157
+ #loc65 = loc("p_x"(#loc16))
158
+ #loc66 = loc("p_y"(#loc17))
159
+ #loc67 = loc("p_y"(#loc18))
160
+ #loc68 = loc("p_y"(#loc19))
161
+ #loc69 = loc("p_y"(#loc20))
162
+ #loc70 = loc("p_y"(#loc21))
163
+ #loc71 = loc("p_rstd"(#loc22))
164
+ #loc72 = loc("b_x"(#loc23))
165
+ #loc73 = loc("b_x"(#loc24))
166
+ #loc74 = loc("b_rstd"(#loc25))
167
+ #loc75 = loc("b_rstd"(#loc26))
168
+ #loc76 = loc("b_rstd"(#loc27))
169
+ #loc77 = loc("b_rstd"(#loc28))
170
+ #loc78 = loc("b_rstd"(#loc29))
171
+ #loc79 = loc("b_y"(#loc30))
172
+ #loc80 = loc("b_y"(#loc31))
triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttgir ADDED
@@ -0,0 +1,135 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [1, 1], order = [1, 0]}>
2
+ #blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [1], order = [0]}>
3
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
4
+ #loc1 = loc(unknown)
5
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
6
+ #loc34 = loc("x"(#loc))
7
+ #loc35 = loc("y"(#loc))
8
+ #loc36 = loc("rstd"(#loc))
9
+ #loc37 = loc("eps"(#loc))
10
+ #loc38 = loc("T"(#loc))
11
+ #loc59 = loc("b_rstd"(#loc23))
12
+ #loc66 = loc(callsite(#loc1 at #loc59))
13
+ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 1 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} {
14
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %cst = arith.constant dense<1.000000e+00> : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1)
16
+ %cst_0 = arith.constant dense<0.000000e+00> : tensor<8x128xbf16, #blocked> loc(#loc1)
17
+ %c8_i64 = arith.constant 8 : i64 loc(#loc1)
18
+ %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1)
19
+ %cst_2 = arith.constant dense<128> : tensor<8x1xi64, #blocked> loc(#loc1)
20
+ %i_t = tt.get_program_id x : i32 loc(#loc39)
21
+ %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40)
22
+ %o_t = arith.muli %i_t_3, %c8_i64 : i64 loc(#loc41)
23
+ %o_t_4 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42)
24
+ %o_t_5 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32, #blocked1> loc(#loc42)
25
+ %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)
26
+ %o_t_7 = arith.extsi %o_t_5 : tensor<8xi32, #blocked1> to tensor<8xi64, #blocked1> loc(#loc43)
27
+ %o_t_8 = tt.splat %o_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43)
28
+ %o_t_9 = tt.splat %o_t : i64 -> tensor<8xi64, #blocked1> loc(#loc43)
29
+ %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43)
30
+ %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<8xi64, #blocked1> loc(#loc43)
31
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc44)
32
+ %m_t_12 = tt.splat %m_t : i64 -> tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44)
33
+ %m_t_13 = tt.splat %m_t : i64 -> tensor<8xi64, #blocked1> loc(#loc44)
34
+ %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44)
35
+ %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<8xi64, #blocked1> loc(#loc44)
36
+ %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<8xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi1, #blocked> loc(#loc45)
37
+ %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46)
38
+ %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)
39
+ %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47)
40
+ %m_x_19 = tt.broadcast %m_x : tensor<8x1xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48)
41
+ %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<8x128xi1, #blocked> loc(#loc48)
42
+ %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<8x128xi1, #blocked> loc(#loc48)
43
+ %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<8xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xi64, #blocked> loc(#loc49)
44
+ %p_x_22 = arith.muli %p_x, %cst_2 : tensor<8x1xi64, #blocked> loc(#loc50)
45
+ %p_x_23 = tt.splat %x : !tt.ptr<bf16> -> tensor<8x1x!tt.ptr<bf16>, #blocked> loc(#loc51)
46
+ %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<8x1x!tt.ptr<bf16>, #blocked>, tensor<8x1xi64, #blocked> loc(#loc51)
47
+ %p_x_25 = tt.broadcast %p_x_24 : tensor<8x1x!tt.ptr<bf16>, #blocked> -> tensor<8x128x!tt.ptr<bf16>, #blocked> loc(#loc52)
48
+ %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<8x128xi32, #blocked> loc(#loc52)
49
+ %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<8x128x!tt.ptr<bf16>, #blocked>, tensor<8x128xi32, #blocked> loc(#loc52)
50
+ %p_y = tt.splat %y : !tt.ptr<bf16> -> tensor<8x1x!tt.ptr<bf16>, #blocked> loc(#loc53)
51
+ %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<8x1x!tt.ptr<bf16>, #blocked>, tensor<8x1xi64, #blocked> loc(#loc53)
52
+ %p_y_29 = tt.broadcast %p_y_28 : tensor<8x1x!tt.ptr<bf16>, #blocked> -> tensor<8x128x!tt.ptr<bf16>, #blocked> loc(#loc54)
53
+ %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<8x128x!tt.ptr<bf16>, #blocked>, tensor<8x128xi32, #blocked> loc(#loc54)
54
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<8x!tt.ptr<f32>, #blocked1> loc(#loc55)
55
+ %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<8x!tt.ptr<f32>, #blocked1>, tensor<8xi64, #blocked1> loc(#loc55)
56
+ %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<8x128x!tt.ptr<bf16>, #blocked> loc(#loc56)
57
+ %b_x_32 = arith.extf %b_x : tensor<8x128xbf16, #blocked> to tensor<8x128xf32, #blocked> loc(#loc57)
58
+ %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<8x128xf32, #blocked> loc(#loc58)
59
+ %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({
60
+ ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))):
61
+ %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67)
62
+ tt.reduce.return %b_rstd_42 : f32 loc(#loc65)
63
+ }) : (tensor<8x128xf32, #blocked>) -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65)
64
+ %b_rstd_34 = tt.splat %eps : f32 -> tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60)
65
+ %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60)
66
+ %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61)
67
+ %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62)
68
+ %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8x1xf32, #blocked> loc(#loc63)
69
+ %b_y_38 = tt.broadcast %b_y : tensor<8x1xf32, #blocked> -> tensor<8x128xf32, #blocked> loc(#loc64)
70
+ %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<8x128xf32, #blocked> loc(#loc64)
71
+ %0 = arith.truncf %b_y_39 : tensor<8x128xf32, #blocked> to tensor<8x128xbf16, #blocked> loc(#loc30)
72
+ tt.store %p_y_30, %0, %m_x_21 : tensor<8x128x!tt.ptr<bf16>, #blocked> loc(#loc31)
73
+ %1 = ttg.convert_layout %b_rstd_37 : tensor<8xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<8xf32, #blocked1> loc(#loc32)
74
+ tt.store %p_rstd_31, %1, %m_t_15 : tensor<8x!tt.ptr<f32>, #blocked1> loc(#loc32)
75
+ tt.return loc(#loc33)
76
+ } loc(#loc)
77
+ } loc(#loc)
78
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
79
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
80
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
81
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
82
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
83
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
84
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
85
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
86
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
87
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
88
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
89
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
90
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
91
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
92
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
93
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
94
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
95
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
96
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
97
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
98
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
99
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
100
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
101
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
102
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
103
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
104
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
105
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
106
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
107
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
108
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
109
+ #loc39 = loc("i_t"(#loc2))
110
+ #loc40 = loc("i_t"(#loc3))
111
+ #loc41 = loc("o_t"(#loc4))
112
+ #loc42 = loc("o_t"(#loc5))
113
+ #loc43 = loc("o_t"(#loc6))
114
+ #loc44 = loc("m_t"(#loc7))
115
+ #loc45 = loc("m_x"(#loc8))
116
+ #loc46 = loc("m_x"(#loc9))
117
+ #loc47 = loc("m_x"(#loc10))
118
+ #loc48 = loc("m_x"(#loc11))
119
+ #loc49 = loc("p_x"(#loc12))
120
+ #loc50 = loc("p_x"(#loc13))
121
+ #loc51 = loc("p_x"(#loc14))
122
+ #loc52 = loc("p_x"(#loc15))
123
+ #loc53 = loc("p_y"(#loc16))
124
+ #loc54 = loc("p_y"(#loc17))
125
+ #loc55 = loc("p_rstd"(#loc18))
126
+ #loc56 = loc("b_x"(#loc19))
127
+ #loc57 = loc("b_x"(#loc20))
128
+ #loc58 = loc("b_rstd"(#loc21))
129
+ #loc60 = loc("b_rstd"(#loc25))
130
+ #loc61 = loc("b_rstd"(#loc26))
131
+ #loc62 = loc("b_rstd"(#loc27))
132
+ #loc63 = loc("b_y"(#loc28))
133
+ #loc64 = loc("b_y"(#loc29))
134
+ #loc65 = loc(callsite(#loc22 at #loc59))
135
+ #loc67 = loc(callsite(#loc24 at #loc65))
triton/gpu0/APMLY4LQ6EOVNTSV25VI5PL76QMSITQBTW3UL6ARWPY5RKPDBM3Q/l2norm_fwd_kernel.ttir ADDED
@@ -0,0 +1,128 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
2
+ #loc3 = loc(unknown)
3
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
4
+ #loc35 = loc("x"(#loc))
5
+ #loc36 = loc("y"(#loc))
6
+ #loc37 = loc("rstd"(#loc))
7
+ #loc38 = loc("eps"(#loc))
8
+ #loc39 = loc("T"(#loc))
9
+ #loc62 = loc("b_rstd"(#loc25))
10
+ #loc68 = loc(callsite(#loc3 at #loc62))
11
+ module {
12
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
13
+ %b_x = arith.constant dense<0.000000e+00> : tensor<8x128xbf16> loc(#loc40)
14
+ %b_rstd = arith.constant dense<1.000000e+00> : tensor<8xf32> loc(#loc41)
15
+ %cst = arith.constant dense<128> : tensor<8x1xi64> loc(#loc3)
16
+ %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42)
17
+ %o_t = arith.constant 8 : i64 loc(#loc43)
18
+ %i_t = tt.get_program_id x : i32 loc(#loc44)
19
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45)
20
+ %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43)
21
+ %o_t_2 = tt.make_range {end = 8 : i32, start = 0 : i32} : tensor<8xi32> loc(#loc46)
22
+ %o_t_3 = arith.extsi %o_t_2 : tensor<8xi32> to tensor<8xi64> loc(#loc47)
23
+ %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<8xi64> loc(#loc47)
24
+ %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<8xi64> loc(#loc47)
25
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48)
26
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc49)
27
+ %m_t_6 = tt.splat %m_t : i64 -> tensor<8xi64> loc(#loc49)
28
+ %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<8xi64> loc(#loc49)
29
+ %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<8xi1> -> tensor<8x1xi1> loc(#loc50)
30
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51)
31
+ %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42)
32
+ %m_x_11 = tt.broadcast %m_x_8 : tensor<8x1xi1> -> tensor<8x128xi1> loc(#loc52)
33
+ %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<8x128xi1> loc(#loc52)
34
+ %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<8x128xi1> loc(#loc52)
35
+ %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<8xi64> -> tensor<8x1xi64> loc(#loc53)
36
+ %p_x_14 = arith.muli %p_x, %cst : tensor<8x1xi64> loc(#loc54)
37
+ %p_x_15 = tt.splat %x : !tt.ptr<bf16> -> tensor<8x1x!tt.ptr<bf16>> loc(#loc55)
38
+ %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<8x1x!tt.ptr<bf16>>, tensor<8x1xi64> loc(#loc55)
39
+ %p_x_17 = tt.broadcast %p_x_16 : tensor<8x1x!tt.ptr<bf16>> -> tensor<8x128x!tt.ptr<bf16>> loc(#loc56)
40
+ %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<8x128xi32> loc(#loc56)
41
+ %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<8x128x!tt.ptr<bf16>>, tensor<8x128xi32> loc(#loc56)
42
+ %p_y = tt.splat %y : !tt.ptr<bf16> -> tensor<8x1x!tt.ptr<bf16>> loc(#loc57)
43
+ %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<8x1x!tt.ptr<bf16>>, tensor<8x1xi64> loc(#loc57)
44
+ %p_y_21 = tt.broadcast %p_y_20 : tensor<8x1x!tt.ptr<bf16>> -> tensor<8x128x!tt.ptr<bf16>> loc(#loc58)
45
+ %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<8x128x!tt.ptr<bf16>>, tensor<8x128xi32> loc(#loc58)
46
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<8x!tt.ptr<f32>> loc(#loc59)
47
+ %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<8x!tt.ptr<f32>>, tensor<8xi64> loc(#loc59)
48
+ %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<8x128x!tt.ptr<bf16>> loc(#loc40)
49
+ %b_x_25 = arith.extf %b_x_24 : tensor<8x128xbf16> to tensor<8x128xf32> loc(#loc60)
50
+ %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<8x128xf32> loc(#loc61)
51
+ %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({
52
+ ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))):
53
+ %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69)
54
+ tt.reduce.return %b_rstd_36 : f32 loc(#loc67)
55
+ }) : (tensor<8x128xf32>) -> tensor<8xf32> loc(#loc67)
56
+ %b_rstd_28 = tt.splat %eps : f32 -> tensor<8xf32> loc(#loc63)
57
+ %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<8xf32> loc(#loc63)
58
+ %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<8xf32> loc(#loc64)
59
+ %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<8xf32> loc(#loc41)
60
+ %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<8xf32> -> tensor<8x1xf32> loc(#loc65)
61
+ %b_y_32 = tt.broadcast %b_y : tensor<8x1xf32> -> tensor<8x128xf32> loc(#loc66)
62
+ %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<8x128xf32> loc(#loc66)
63
+ %0 = arith.truncf %b_y_33 : tensor<8x128xf32> to tensor<8x128xbf16> loc(#loc31)
64
+ tt.store %p_y_22, %0, %m_x_13 : tensor<8x128x!tt.ptr<bf16>> loc(#loc32)
65
+ tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<8x!tt.ptr<f32>> loc(#loc33)
66
+ tt.return loc(#loc34)
67
+ } loc(#loc)
68
+ } loc(#loc)
69
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
70
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
71
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
72
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
73
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
74
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
75
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
76
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
77
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23)
78
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
79
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
80
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
81
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
82
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
83
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
84
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
85
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
86
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
87
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
88
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
89
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
90
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
91
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
92
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
93
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
94
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
95
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
96
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
97
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
98
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
99
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
100
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
101
+ #loc40 = loc("b_x"(#loc1))
102
+ #loc41 = loc("b_rstd"(#loc2))
103
+ #loc42 = loc("m_x"(#loc4))
104
+ #loc43 = loc("o_t"(#loc5))
105
+ #loc44 = loc("i_t"(#loc6))
106
+ #loc45 = loc("i_t"(#loc7))
107
+ #loc46 = loc("o_t"(#loc8))
108
+ #loc47 = loc("o_t"(#loc9))
109
+ #loc48 = loc("o_d"(#loc10))
110
+ #loc49 = loc("m_t"(#loc11))
111
+ #loc50 = loc("m_x"(#loc12))
112
+ #loc51 = loc("m_x"(#loc13))
113
+ #loc52 = loc("m_x"(#loc14))
114
+ #loc53 = loc("p_x"(#loc15))
115
+ #loc54 = loc("p_x"(#loc16))
116
+ #loc55 = loc("p_x"(#loc17))
117
+ #loc56 = loc("p_x"(#loc18))
118
+ #loc57 = loc("p_y"(#loc19))
119
+ #loc58 = loc("p_y"(#loc20))
120
+ #loc59 = loc("p_rstd"(#loc21))
121
+ #loc60 = loc("b_x"(#loc22))
122
+ #loc61 = loc("b_rstd"(#loc23))
123
+ #loc63 = loc("b_rstd"(#loc27))
124
+ #loc64 = loc("b_rstd"(#loc28))
125
+ #loc65 = loc("b_y"(#loc29))
126
+ #loc66 = loc("b_y"(#loc30))
127
+ #loc67 = loc(callsite(#loc24 at #loc62))
128
+ #loc69 = loc(callsite(#loc26 at #loc67))
triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/__grp__l2norm_fwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}}
triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.cubin ADDED
Binary file (28.8 kB). View file
 
triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}
triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.llir ADDED
@@ -0,0 +1,630 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ; ModuleID = 'LLVMDialectModule'
2
+ source_filename = "LLVMDialectModule"
3
+ 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"
4
+
5
+ @global_smem = external addrspace(3) global [0 x i8], align 16
6
+ @.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1
7
+ @.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1
8
+
9
+ ; Function Attrs: nounwind
10
+ 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 {
11
+ %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13
12
+ %9 = zext nneg i32 %8 to i64, !dbg !14
13
+ %10 = shl nuw nsw i64 %9, 4, !dbg !15
14
+ %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16
15
+ %12 = and i32 %11, 48, !dbg !16
16
+ %13 = lshr exact i32 %12, 4, !dbg !16
17
+ %14 = or disjoint i32 %13, 4, !dbg !16
18
+ %15 = or disjoint i32 %13, 8, !dbg !16
19
+ %16 = or disjoint i32 %13, 12, !dbg !16
20
+ %17 = and i32 %11, 15, !dbg !16
21
+ %18 = zext nneg i32 %13 to i64, !dbg !17
22
+ %19 = zext nneg i32 %14 to i64, !dbg !17
23
+ %20 = zext nneg i32 %15 to i64, !dbg !17
24
+ %21 = zext nneg i32 %16 to i64, !dbg !17
25
+ %22 = or disjoint i64 %10, %18, !dbg !17
26
+ %23 = or disjoint i64 %10, %19, !dbg !17
27
+ %24 = or disjoint i64 %10, %20, !dbg !17
28
+ %25 = or disjoint i64 %10, %21, !dbg !17
29
+ %26 = sext i32 %4 to i64, !dbg !18
30
+ %27 = icmp slt i64 %22, %26, !dbg !18
31
+ %28 = icmp slt i64 %23, %26, !dbg !18
32
+ %29 = icmp slt i64 %24, %26, !dbg !18
33
+ %30 = icmp slt i64 %25, %26, !dbg !18
34
+ %31 = shl nuw nsw i32 %17, 3, !dbg !19
35
+ %32 = shl nuw nsw i64 %22, 7, !dbg !20
36
+ %33 = shl nuw nsw i64 %23, 7, !dbg !20
37
+ %34 = shl nuw nsw i64 %24, 7, !dbg !20
38
+ %35 = shl nuw nsw i64 %25, 7, !dbg !20
39
+ %36 = getelementptr bfloat, ptr addrspace(1) %0, i64 %32, !dbg !21
40
+ %37 = getelementptr bfloat, ptr addrspace(1) %0, i64 %33, !dbg !21
41
+ %38 = getelementptr bfloat, ptr addrspace(1) %0, i64 %34, !dbg !21
42
+ %39 = getelementptr bfloat, ptr addrspace(1) %0, i64 %35, !dbg !21
43
+ %40 = zext nneg i32 %31 to i64, !dbg !22
44
+ %41 = getelementptr bfloat, ptr addrspace(1) %36, i64 %40, !dbg !22
45
+ %42 = getelementptr bfloat, ptr addrspace(1) %37, i64 %40, !dbg !22
46
+ %43 = getelementptr bfloat, ptr addrspace(1) %38, i64 %40, !dbg !22
47
+ %44 = getelementptr bfloat, ptr addrspace(1) %39, i64 %40, !dbg !22
48
+ %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
49
+ %46 = extractvalue { i32, i32, i32, i32 } %45, 0, !dbg !23
50
+ %47 = bitcast i32 %46 to <2 x bfloat>, !dbg !23
51
+ %48 = extractvalue { i32, i32, i32, i32 } %45, 1, !dbg !23
52
+ %49 = bitcast i32 %48 to <2 x bfloat>, !dbg !23
53
+ %50 = extractvalue { i32, i32, i32, i32 } %45, 2, !dbg !23
54
+ %51 = bitcast i32 %50 to <2 x bfloat>, !dbg !23
55
+ %52 = extractvalue { i32, i32, i32, i32 } %45, 3, !dbg !23
56
+ %53 = bitcast i32 %52 to <2 x bfloat>, !dbg !23
57
+ %54 = extractelement <2 x bfloat> %47, i64 0, !dbg !23
58
+ %55 = extractelement <2 x bfloat> %47, i64 1, !dbg !23
59
+ %56 = extractelement <2 x bfloat> %49, i64 0, !dbg !23
60
+ %57 = extractelement <2 x bfloat> %49, i64 1, !dbg !23
61
+ %58 = extractelement <2 x bfloat> %51, i64 0, !dbg !23
62
+ %59 = extractelement <2 x bfloat> %51, i64 1, !dbg !23
63
+ %60 = extractelement <2 x bfloat> %53, i64 0, !dbg !23
64
+ %61 = extractelement <2 x bfloat> %53, i64 1, !dbg !23
65
+ %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
66
+ %63 = extractvalue { i32, i32, i32, i32 } %62, 0, !dbg !23
67
+ %64 = bitcast i32 %63 to <2 x bfloat>, !dbg !23
68
+ %65 = extractvalue { i32, i32, i32, i32 } %62, 1, !dbg !23
69
+ %66 = bitcast i32 %65 to <2 x bfloat>, !dbg !23
70
+ %67 = extractvalue { i32, i32, i32, i32 } %62, 2, !dbg !23
71
+ %68 = bitcast i32 %67 to <2 x bfloat>, !dbg !23
72
+ %69 = extractvalue { i32, i32, i32, i32 } %62, 3, !dbg !23
73
+ %70 = bitcast i32 %69 to <2 x bfloat>, !dbg !23
74
+ %71 = extractelement <2 x bfloat> %64, i64 0, !dbg !23
75
+ %72 = extractelement <2 x bfloat> %64, i64 1, !dbg !23
76
+ %73 = extractelement <2 x bfloat> %66, i64 0, !dbg !23
77
+ %74 = extractelement <2 x bfloat> %66, i64 1, !dbg !23
78
+ %75 = extractelement <2 x bfloat> %68, i64 0, !dbg !23
79
+ %76 = extractelement <2 x bfloat> %68, i64 1, !dbg !23
80
+ %77 = extractelement <2 x bfloat> %70, i64 0, !dbg !23
81
+ %78 = extractelement <2 x bfloat> %70, i64 1, !dbg !23
82
+ %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
83
+ %80 = extractvalue { i32, i32, i32, i32 } %79, 0, !dbg !23
84
+ %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23
85
+ %82 = extractvalue { i32, i32, i32, i32 } %79, 1, !dbg !23
86
+ %83 = bitcast i32 %82 to <2 x bfloat>, !dbg !23
87
+ %84 = extractvalue { i32, i32, i32, i32 } %79, 2, !dbg !23
88
+ %85 = bitcast i32 %84 to <2 x bfloat>, !dbg !23
89
+ %86 = extractvalue { i32, i32, i32, i32 } %79, 3, !dbg !23
90
+ %87 = bitcast i32 %86 to <2 x bfloat>, !dbg !23
91
+ %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23
92
+ %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23
93
+ %90 = extractelement <2 x bfloat> %83, i64 0, !dbg !23
94
+ %91 = extractelement <2 x bfloat> %83, i64 1, !dbg !23
95
+ %92 = extractelement <2 x bfloat> %85, i64 0, !dbg !23
96
+ %93 = extractelement <2 x bfloat> %85, i64 1, !dbg !23
97
+ %94 = extractelement <2 x bfloat> %87, i64 0, !dbg !23
98
+ %95 = extractelement <2 x bfloat> %87, i64 1, !dbg !23
99
+ %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
100
+ %97 = extractvalue { i32, i32, i32, i32 } %96, 0, !dbg !23
101
+ %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23
102
+ %99 = extractvalue { i32, i32, i32, i32 } %96, 1, !dbg !23
103
+ %100 = bitcast i32 %99 to <2 x bfloat>, !dbg !23
104
+ %101 = extractvalue { i32, i32, i32, i32 } %96, 2, !dbg !23
105
+ %102 = bitcast i32 %101 to <2 x bfloat>, !dbg !23
106
+ %103 = extractvalue { i32, i32, i32, i32 } %96, 3, !dbg !23
107
+ %104 = bitcast i32 %103 to <2 x bfloat>, !dbg !23
108
+ %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23
109
+ %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23
110
+ %107 = extractelement <2 x bfloat> %100, i64 0, !dbg !23
111
+ %108 = extractelement <2 x bfloat> %100, i64 1, !dbg !23
112
+ %109 = extractelement <2 x bfloat> %102, i64 0, !dbg !23
113
+ %110 = extractelement <2 x bfloat> %102, i64 1, !dbg !23
114
+ %111 = extractelement <2 x bfloat> %104, i64 0, !dbg !23
115
+ %112 = extractelement <2 x bfloat> %104, i64 1, !dbg !23
116
+ %113 = fpext bfloat %54 to float, !dbg !24
117
+ %114 = fpext bfloat %55 to float, !dbg !24
118
+ %115 = fpext bfloat %56 to float, !dbg !24
119
+ %116 = fpext bfloat %57 to float, !dbg !24
120
+ %117 = fpext bfloat %58 to float, !dbg !24
121
+ %118 = fpext bfloat %59 to float, !dbg !24
122
+ %119 = fpext bfloat %60 to float, !dbg !24
123
+ %120 = fpext bfloat %61 to float, !dbg !24
124
+ %121 = fpext bfloat %71 to float, !dbg !24
125
+ %122 = fpext bfloat %72 to float, !dbg !24
126
+ %123 = fpext bfloat %73 to float, !dbg !24
127
+ %124 = fpext bfloat %74 to float, !dbg !24
128
+ %125 = fpext bfloat %75 to float, !dbg !24
129
+ %126 = fpext bfloat %76 to float, !dbg !24
130
+ %127 = fpext bfloat %77 to float, !dbg !24
131
+ %128 = fpext bfloat %78 to float, !dbg !24
132
+ %129 = fpext bfloat %88 to float, !dbg !24
133
+ %130 = fpext bfloat %89 to float, !dbg !24
134
+ %131 = fpext bfloat %90 to float, !dbg !24
135
+ %132 = fpext bfloat %91 to float, !dbg !24
136
+ %133 = fpext bfloat %92 to float, !dbg !24
137
+ %134 = fpext bfloat %93 to float, !dbg !24
138
+ %135 = fpext bfloat %94 to float, !dbg !24
139
+ %136 = fpext bfloat %95 to float, !dbg !24
140
+ %137 = fpext bfloat %105 to float, !dbg !24
141
+ %138 = fpext bfloat %106 to float, !dbg !24
142
+ %139 = fpext bfloat %107 to float, !dbg !24
143
+ %140 = fpext bfloat %108 to float, !dbg !24
144
+ %141 = fpext bfloat %109 to float, !dbg !24
145
+ %142 = fpext bfloat %110 to float, !dbg !24
146
+ %143 = fpext bfloat %111 to float, !dbg !24
147
+ %144 = fpext bfloat %112 to float, !dbg !24
148
+ %145 = fmul float %113, %113, !dbg !25
149
+ %146 = fmul float %114, %114, !dbg !25
150
+ %147 = fmul float %115, %115, !dbg !25
151
+ %148 = fmul float %116, %116, !dbg !25
152
+ %149 = fmul float %117, %117, !dbg !25
153
+ %150 = fmul float %118, %118, !dbg !25
154
+ %151 = fmul float %119, %119, !dbg !25
155
+ %152 = fmul float %120, %120, !dbg !25
156
+ %153 = fmul float %121, %121, !dbg !25
157
+ %154 = fmul float %122, %122, !dbg !25
158
+ %155 = fmul float %123, %123, !dbg !25
159
+ %156 = fmul float %124, %124, !dbg !25
160
+ %157 = fmul float %125, %125, !dbg !25
161
+ %158 = fmul float %126, %126, !dbg !25
162
+ %159 = fmul float %127, %127, !dbg !25
163
+ %160 = fmul float %128, %128, !dbg !25
164
+ %161 = fmul float %129, %129, !dbg !25
165
+ %162 = fmul float %130, %130, !dbg !25
166
+ %163 = fmul float %131, %131, !dbg !25
167
+ %164 = fmul float %132, %132, !dbg !25
168
+ %165 = fmul float %133, %133, !dbg !25
169
+ %166 = fmul float %134, %134, !dbg !25
170
+ %167 = fmul float %135, %135, !dbg !25
171
+ %168 = fmul float %136, %136, !dbg !25
172
+ %169 = fmul float %137, %137, !dbg !25
173
+ %170 = fmul float %138, %138, !dbg !25
174
+ %171 = fmul float %139, %139, !dbg !25
175
+ %172 = fmul float %140, %140, !dbg !25
176
+ %173 = fmul float %141, %141, !dbg !25
177
+ %174 = fmul float %142, %142, !dbg !25
178
+ %175 = fmul float %143, %143, !dbg !25
179
+ %176 = fmul float %144, %144, !dbg !25
180
+ %177 = fadd float %145, %146, !dbg !26
181
+ %178 = fadd float %147, %177, !dbg !26
182
+ %179 = fadd float %148, %178, !dbg !26
183
+ %180 = fadd float %149, %179, !dbg !26
184
+ %181 = fadd float %150, %180, !dbg !26
185
+ %182 = fadd float %151, %181, !dbg !26
186
+ %183 = fadd float %152, %182, !dbg !26
187
+ %184 = fadd float %153, %154, !dbg !26
188
+ %185 = fadd float %155, %184, !dbg !26
189
+ %186 = fadd float %156, %185, !dbg !26
190
+ %187 = fadd float %157, %186, !dbg !26
191
+ %188 = fadd float %158, %187, !dbg !26
192
+ %189 = fadd float %159, %188, !dbg !26
193
+ %190 = fadd float %160, %189, !dbg !26
194
+ %191 = fadd float %161, %162, !dbg !26
195
+ %192 = fadd float %163, %191, !dbg !26
196
+ %193 = fadd float %164, %192, !dbg !26
197
+ %194 = fadd float %165, %193, !dbg !26
198
+ %195 = fadd float %166, %194, !dbg !26
199
+ %196 = fadd float %167, %195, !dbg !26
200
+ %197 = fadd float %168, %196, !dbg !26
201
+ %198 = fadd float %169, %170, !dbg !26
202
+ %199 = fadd float %171, %198, !dbg !26
203
+ %200 = fadd float %172, %199, !dbg !26
204
+ %201 = fadd float %173, %200, !dbg !26
205
+ %202 = fadd float %174, %201, !dbg !26
206
+ %203 = fadd float %175, %202, !dbg !26
207
+ %204 = fadd float %176, %203, !dbg !26
208
+ %205 = bitcast float %183 to i32, !dbg !29
209
+ %206 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %205, i32 8, i32 31), !dbg !29
210
+ %207 = bitcast i32 %206 to float, !dbg !29
211
+ %208 = fadd float %183, %207, !dbg !26
212
+ %209 = bitcast float %208 to i32, !dbg !29
213
+ %210 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %209, i32 4, i32 31), !dbg !29
214
+ %211 = bitcast i32 %210 to float, !dbg !29
215
+ %212 = fadd float %208, %211, !dbg !26
216
+ %213 = bitcast float %212 to i32, !dbg !29
217
+ %214 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %213, i32 2, i32 31), !dbg !29
218
+ %215 = bitcast i32 %214 to float, !dbg !29
219
+ %216 = fadd float %212, %215, !dbg !26
220
+ %217 = bitcast float %216 to i32, !dbg !29
221
+ %218 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %217, i32 1, i32 31), !dbg !29
222
+ %219 = bitcast i32 %218 to float, !dbg !29
223
+ %220 = fadd float %216, %219, !dbg !26
224
+ %221 = bitcast float %190 to i32, !dbg !29
225
+ %222 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %221, i32 8, i32 31), !dbg !29
226
+ %223 = bitcast i32 %222 to float, !dbg !29
227
+ %224 = fadd float %190, %223, !dbg !26
228
+ %225 = bitcast float %224 to i32, !dbg !29
229
+ %226 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %225, i32 4, i32 31), !dbg !29
230
+ %227 = bitcast i32 %226 to float, !dbg !29
231
+ %228 = fadd float %224, %227, !dbg !26
232
+ %229 = bitcast float %228 to i32, !dbg !29
233
+ %230 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %229, i32 2, i32 31), !dbg !29
234
+ %231 = bitcast i32 %230 to float, !dbg !29
235
+ %232 = fadd float %228, %231, !dbg !26
236
+ %233 = bitcast float %232 to i32, !dbg !29
237
+ %234 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %233, i32 1, i32 31), !dbg !29
238
+ %235 = bitcast i32 %234 to float, !dbg !29
239
+ %236 = fadd float %232, %235, !dbg !26
240
+ %237 = bitcast float %197 to i32, !dbg !29
241
+ %238 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %237, i32 8, i32 31), !dbg !29
242
+ %239 = bitcast i32 %238 to float, !dbg !29
243
+ %240 = fadd float %197, %239, !dbg !26
244
+ %241 = bitcast float %240 to i32, !dbg !29
245
+ %242 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %241, i32 4, i32 31), !dbg !29
246
+ %243 = bitcast i32 %242 to float, !dbg !29
247
+ %244 = fadd float %240, %243, !dbg !26
248
+ %245 = bitcast float %244 to i32, !dbg !29
249
+ %246 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %245, i32 2, i32 31), !dbg !29
250
+ %247 = bitcast i32 %246 to float, !dbg !29
251
+ %248 = fadd float %244, %247, !dbg !26
252
+ %249 = bitcast float %248 to i32, !dbg !29
253
+ %250 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %249, i32 1, i32 31), !dbg !29
254
+ %251 = bitcast i32 %250 to float, !dbg !29
255
+ %252 = fadd float %248, %251, !dbg !26
256
+ %253 = bitcast float %204 to i32, !dbg !29
257
+ %254 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %253, i32 8, i32 31), !dbg !29
258
+ %255 = bitcast i32 %254 to float, !dbg !29
259
+ %256 = fadd float %204, %255, !dbg !26
260
+ %257 = bitcast float %256 to i32, !dbg !29
261
+ %258 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %257, i32 4, i32 31), !dbg !29
262
+ %259 = bitcast i32 %258 to float, !dbg !29
263
+ %260 = fadd float %256, %259, !dbg !26
264
+ %261 = bitcast float %260 to i32, !dbg !29
265
+ %262 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %261, i32 2, i32 31), !dbg !29
266
+ %263 = bitcast i32 %262 to float, !dbg !29
267
+ %264 = fadd float %260, %263, !dbg !26
268
+ %265 = bitcast float %264 to i32, !dbg !29
269
+ %266 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %265, i32 1, i32 31), !dbg !29
270
+ %267 = bitcast i32 %266 to float, !dbg !29
271
+ %268 = fadd float %264, %267, !dbg !26
272
+ %269 = fadd float %3, %220, !dbg !32
273
+ %270 = fadd float %3, %236, !dbg !32
274
+ %271 = fadd float %3, %252, !dbg !32
275
+ %272 = fadd float %3, %268, !dbg !32
276
+ %273 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
277
+ %.not.i = icmp eq i32 %273, 0, !dbg !33
278
+ %274 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
279
+ %.not1.i = icmp eq i32 %274, 0, !dbg !33
280
+ br i1 %.not.i, label %280, label %275, !dbg !33
281
+
282
+ 275: ; preds = %7
283
+ br i1 %.not1.i, label %278, label %276, !dbg !33
284
+
285
+ 276: ; preds = %275
286
+ %277 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %269) #6, !dbg !33
287
+ br label %__nv_sqrtf.exit, !dbg !33
288
+
289
+ 278: ; preds = %275
290
+ %279 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %269) #6, !dbg !33
291
+ br label %__nv_sqrtf.exit, !dbg !33
292
+
293
+ 280: ; preds = %7
294
+ br i1 %.not1.i, label %283, label %281, !dbg !33
295
+
296
+ 281: ; preds = %280
297
+ %282 = tail call float @llvm.nvvm.sqrt.rn.f(float %269) #6, !dbg !33
298
+ br label %__nv_sqrtf.exit, !dbg !33
299
+
300
+ 283: ; preds = %280
301
+ %284 = tail call float @llvm.nvvm.sqrt.approx.f(float %269) #6, !dbg !33
302
+ br label %__nv_sqrtf.exit, !dbg !33
303
+
304
+ __nv_sqrtf.exit: ; preds = %276, %278, %281, %283
305
+ %.0.i = phi float [ %277, %276 ], [ %279, %278 ], [ %282, %281 ], [ %284, %283 ], !dbg !33
306
+ %285 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
307
+ %.not.i1 = icmp eq i32 %285, 0, !dbg !33
308
+ %286 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
309
+ %.not1.i4 = icmp eq i32 %286, 0, !dbg !33
310
+ br i1 %.not.i1, label %292, label %287, !dbg !33
311
+
312
+ 287: ; preds = %__nv_sqrtf.exit
313
+ br i1 %.not1.i4, label %290, label %288, !dbg !33
314
+
315
+ 288: ; preds = %287
316
+ %289 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %270) #6, !dbg !33
317
+ br label %__nv_sqrtf.exit5, !dbg !33
318
+
319
+ 290: ; preds = %287
320
+ %291 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %270) #6, !dbg !33
321
+ br label %__nv_sqrtf.exit5, !dbg !33
322
+
323
+ 292: ; preds = %__nv_sqrtf.exit
324
+ br i1 %.not1.i4, label %295, label %293, !dbg !33
325
+
326
+ 293: ; preds = %292
327
+ %294 = tail call float @llvm.nvvm.sqrt.rn.f(float %270) #6, !dbg !33
328
+ br label %__nv_sqrtf.exit5, !dbg !33
329
+
330
+ 295: ; preds = %292
331
+ %296 = tail call float @llvm.nvvm.sqrt.approx.f(float %270) #6, !dbg !33
332
+ br label %__nv_sqrtf.exit5, !dbg !33
333
+
334
+ __nv_sqrtf.exit5: ; preds = %288, %290, %293, %295
335
+ %.0.i3 = phi float [ %289, %288 ], [ %291, %290 ], [ %294, %293 ], [ %296, %295 ], !dbg !33
336
+ %297 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
337
+ %.not.i6 = icmp eq i32 %297, 0, !dbg !33
338
+ %298 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
339
+ %.not1.i9 = icmp eq i32 %298, 0, !dbg !33
340
+ br i1 %.not.i6, label %304, label %299, !dbg !33
341
+
342
+ 299: ; preds = %__nv_sqrtf.exit5
343
+ br i1 %.not1.i9, label %302, label %300, !dbg !33
344
+
345
+ 300: ; preds = %299
346
+ %301 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %271) #6, !dbg !33
347
+ br label %__nv_sqrtf.exit10, !dbg !33
348
+
349
+ 302: ; preds = %299
350
+ %303 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %271) #6, !dbg !33
351
+ br label %__nv_sqrtf.exit10, !dbg !33
352
+
353
+ 304: ; preds = %__nv_sqrtf.exit5
354
+ br i1 %.not1.i9, label %307, label %305, !dbg !33
355
+
356
+ 305: ; preds = %304
357
+ %306 = tail call float @llvm.nvvm.sqrt.rn.f(float %271) #6, !dbg !33
358
+ br label %__nv_sqrtf.exit10, !dbg !33
359
+
360
+ 307: ; preds = %304
361
+ %308 = tail call float @llvm.nvvm.sqrt.approx.f(float %271) #6, !dbg !33
362
+ br label %__nv_sqrtf.exit10, !dbg !33
363
+
364
+ __nv_sqrtf.exit10: ; preds = %300, %302, %305, %307
365
+ %.0.i8 = phi float [ %301, %300 ], [ %303, %302 ], [ %306, %305 ], [ %308, %307 ], !dbg !33
366
+ %309 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
367
+ %.not.i11 = icmp eq i32 %309, 0, !dbg !33
368
+ %310 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
369
+ %.not1.i14 = icmp eq i32 %310, 0, !dbg !33
370
+ br i1 %.not.i11, label %316, label %311, !dbg !33
371
+
372
+ 311: ; preds = %__nv_sqrtf.exit10
373
+ br i1 %.not1.i14, label %314, label %312, !dbg !33
374
+
375
+ 312: ; preds = %311
376
+ %313 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %272) #6, !dbg !33
377
+ br label %__nv_sqrtf.exit15, !dbg !33
378
+
379
+ 314: ; preds = %311
380
+ %315 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %272) #6, !dbg !33
381
+ br label %__nv_sqrtf.exit15, !dbg !33
382
+
383
+ 316: ; preds = %__nv_sqrtf.exit10
384
+ br i1 %.not1.i14, label %319, label %317, !dbg !33
385
+
386
+ 317: ; preds = %316
387
+ %318 = tail call float @llvm.nvvm.sqrt.rn.f(float %272) #6, !dbg !33
388
+ br label %__nv_sqrtf.exit15, !dbg !33
389
+
390
+ 319: ; preds = %316
391
+ %320 = tail call float @llvm.nvvm.sqrt.approx.f(float %272) #6, !dbg !33
392
+ br label %__nv_sqrtf.exit15, !dbg !33
393
+
394
+ __nv_sqrtf.exit15: ; preds = %312, %314, %317, %319
395
+ %.0.i13 = phi float [ %313, %312 ], [ %315, %314 ], [ %318, %317 ], [ %320, %319 ], !dbg !33
396
+ %321 = zext nneg i32 %17 to i64, !dbg !17
397
+ %322 = or disjoint i64 %10, %321, !dbg !17
398
+ %323 = getelementptr float, ptr addrspace(1) %2, i64 %322, !dbg !34
399
+ %324 = getelementptr bfloat, ptr addrspace(1) %1, i64 %35, !dbg !35
400
+ %325 = getelementptr bfloat, ptr addrspace(1) %324, i64 %40, !dbg !36
401
+ %326 = getelementptr bfloat, ptr addrspace(1) %1, i64 %34, !dbg !35
402
+ %327 = getelementptr bfloat, ptr addrspace(1) %326, i64 %40, !dbg !36
403
+ %328 = getelementptr bfloat, ptr addrspace(1) %1, i64 %33, !dbg !35
404
+ %329 = getelementptr bfloat, ptr addrspace(1) %328, i64 %40, !dbg !36
405
+ %330 = getelementptr bfloat, ptr addrspace(1) %1, i64 %32, !dbg !35
406
+ %331 = getelementptr bfloat, ptr addrspace(1) %330, i64 %40, !dbg !36
407
+ %332 = icmp slt i64 %322, %26, !dbg !18
408
+ %333 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37
409
+ %334 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37
410
+ %335 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37
411
+ %336 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37
412
+ %337 = fmul float %333, %113, !dbg !38
413
+ %338 = fmul float %333, %114, !dbg !38
414
+ %339 = fmul float %333, %115, !dbg !38
415
+ %340 = fmul float %333, %116, !dbg !38
416
+ %341 = fmul float %333, %117, !dbg !38
417
+ %342 = fmul float %333, %118, !dbg !38
418
+ %343 = fmul float %333, %119, !dbg !38
419
+ %344 = fmul float %333, %120, !dbg !38
420
+ %345 = fmul float %334, %121, !dbg !38
421
+ %346 = fmul float %334, %122, !dbg !38
422
+ %347 = fmul float %334, %123, !dbg !38
423
+ %348 = fmul float %334, %124, !dbg !38
424
+ %349 = fmul float %334, %125, !dbg !38
425
+ %350 = fmul float %334, %126, !dbg !38
426
+ %351 = fmul float %334, %127, !dbg !38
427
+ %352 = fmul float %334, %128, !dbg !38
428
+ %353 = fmul float %335, %129, !dbg !38
429
+ %354 = fmul float %335, %130, !dbg !38
430
+ %355 = fmul float %335, %131, !dbg !38
431
+ %356 = fmul float %335, %132, !dbg !38
432
+ %357 = fmul float %335, %133, !dbg !38
433
+ %358 = fmul float %335, %134, !dbg !38
434
+ %359 = fmul float %335, %135, !dbg !38
435
+ %360 = fmul float %335, %136, !dbg !38
436
+ %361 = fmul float %336, %137, !dbg !38
437
+ %362 = fmul float %336, %138, !dbg !38
438
+ %363 = fmul float %336, %139, !dbg !38
439
+ %364 = fmul float %336, %140, !dbg !38
440
+ %365 = fmul float %336, %141, !dbg !38
441
+ %366 = fmul float %336, %142, !dbg !38
442
+ %367 = fmul float %336, %143, !dbg !38
443
+ %368 = fmul float %336, %144, !dbg !38
444
+ %369 = fptrunc float %337 to bfloat, !dbg !39
445
+ %370 = fptrunc float %338 to bfloat, !dbg !39
446
+ %371 = fptrunc float %339 to bfloat, !dbg !39
447
+ %372 = fptrunc float %340 to bfloat, !dbg !39
448
+ %373 = fptrunc float %341 to bfloat, !dbg !39
449
+ %374 = fptrunc float %342 to bfloat, !dbg !39
450
+ %375 = fptrunc float %343 to bfloat, !dbg !39
451
+ %376 = fptrunc float %344 to bfloat, !dbg !39
452
+ %377 = fptrunc float %345 to bfloat, !dbg !39
453
+ %378 = fptrunc float %346 to bfloat, !dbg !39
454
+ %379 = fptrunc float %347 to bfloat, !dbg !39
455
+ %380 = fptrunc float %348 to bfloat, !dbg !39
456
+ %381 = fptrunc float %349 to bfloat, !dbg !39
457
+ %382 = fptrunc float %350 to bfloat, !dbg !39
458
+ %383 = fptrunc float %351 to bfloat, !dbg !39
459
+ %384 = fptrunc float %352 to bfloat, !dbg !39
460
+ %385 = fptrunc float %353 to bfloat, !dbg !39
461
+ %386 = fptrunc float %354 to bfloat, !dbg !39
462
+ %387 = fptrunc float %355 to bfloat, !dbg !39
463
+ %388 = fptrunc float %356 to bfloat, !dbg !39
464
+ %389 = fptrunc float %357 to bfloat, !dbg !39
465
+ %390 = fptrunc float %358 to bfloat, !dbg !39
466
+ %391 = fptrunc float %359 to bfloat, !dbg !39
467
+ %392 = fptrunc float %360 to bfloat, !dbg !39
468
+ %393 = fptrunc float %361 to bfloat, !dbg !39
469
+ %394 = fptrunc float %362 to bfloat, !dbg !39
470
+ %395 = fptrunc float %363 to bfloat, !dbg !39
471
+ %396 = fptrunc float %364 to bfloat, !dbg !39
472
+ %397 = fptrunc float %365 to bfloat, !dbg !39
473
+ %398 = fptrunc float %366 to bfloat, !dbg !39
474
+ %399 = fptrunc float %367 to bfloat, !dbg !39
475
+ %400 = fptrunc float %368 to bfloat, !dbg !39
476
+ %401 = insertelement <2 x bfloat> poison, bfloat %369, i64 0, !dbg !40
477
+ %402 = insertelement <2 x bfloat> %401, bfloat %370, i64 1, !dbg !40
478
+ %403 = bitcast <2 x bfloat> %402 to i32, !dbg !40
479
+ %404 = insertelement <2 x bfloat> poison, bfloat %371, i64 0, !dbg !40
480
+ %405 = insertelement <2 x bfloat> %404, bfloat %372, i64 1, !dbg !40
481
+ %406 = bitcast <2 x bfloat> %405 to i32, !dbg !40
482
+ %407 = insertelement <2 x bfloat> poison, bfloat %373, i64 0, !dbg !40
483
+ %408 = insertelement <2 x bfloat> %407, bfloat %374, i64 1, !dbg !40
484
+ %409 = bitcast <2 x bfloat> %408 to i32, !dbg !40
485
+ %410 = insertelement <2 x bfloat> poison, bfloat %375, i64 0, !dbg !40
486
+ %411 = insertelement <2 x bfloat> %410, bfloat %376, i64 1, !dbg !40
487
+ %412 = bitcast <2 x bfloat> %411 to i32, !dbg !40
488
+ 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
489
+ %413 = insertelement <2 x bfloat> poison, bfloat %377, i64 0, !dbg !40
490
+ %414 = insertelement <2 x bfloat> %413, bfloat %378, i64 1, !dbg !40
491
+ %415 = bitcast <2 x bfloat> %414 to i32, !dbg !40
492
+ %416 = insertelement <2 x bfloat> poison, bfloat %379, i64 0, !dbg !40
493
+ %417 = insertelement <2 x bfloat> %416, bfloat %380, i64 1, !dbg !40
494
+ %418 = bitcast <2 x bfloat> %417 to i32, !dbg !40
495
+ %419 = insertelement <2 x bfloat> poison, bfloat %381, i64 0, !dbg !40
496
+ %420 = insertelement <2 x bfloat> %419, bfloat %382, i64 1, !dbg !40
497
+ %421 = bitcast <2 x bfloat> %420 to i32, !dbg !40
498
+ %422 = insertelement <2 x bfloat> poison, bfloat %383, i64 0, !dbg !40
499
+ %423 = insertelement <2 x bfloat> %422, bfloat %384, i64 1, !dbg !40
500
+ %424 = bitcast <2 x bfloat> %423 to i32, !dbg !40
501
+ 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
502
+ %425 = insertelement <2 x bfloat> poison, bfloat %385, i64 0, !dbg !40
503
+ %426 = insertelement <2 x bfloat> %425, bfloat %386, i64 1, !dbg !40
504
+ %427 = bitcast <2 x bfloat> %426 to i32, !dbg !40
505
+ %428 = insertelement <2 x bfloat> poison, bfloat %387, i64 0, !dbg !40
506
+ %429 = insertelement <2 x bfloat> %428, bfloat %388, i64 1, !dbg !40
507
+ %430 = bitcast <2 x bfloat> %429 to i32, !dbg !40
508
+ %431 = insertelement <2 x bfloat> poison, bfloat %389, i64 0, !dbg !40
509
+ %432 = insertelement <2 x bfloat> %431, bfloat %390, i64 1, !dbg !40
510
+ %433 = bitcast <2 x bfloat> %432 to i32, !dbg !40
511
+ %434 = insertelement <2 x bfloat> poison, bfloat %391, i64 0, !dbg !40
512
+ %435 = insertelement <2 x bfloat> %434, bfloat %392, i64 1, !dbg !40
513
+ %436 = bitcast <2 x bfloat> %435 to i32, !dbg !40
514
+ 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
515
+ %437 = insertelement <2 x bfloat> poison, bfloat %393, i64 0, !dbg !40
516
+ %438 = insertelement <2 x bfloat> %437, bfloat %394, i64 1, !dbg !40
517
+ %439 = bitcast <2 x bfloat> %438 to i32, !dbg !40
518
+ %440 = insertelement <2 x bfloat> poison, bfloat %395, i64 0, !dbg !40
519
+ %441 = insertelement <2 x bfloat> %440, bfloat %396, i64 1, !dbg !40
520
+ %442 = bitcast <2 x bfloat> %441 to i32, !dbg !40
521
+ %443 = insertelement <2 x bfloat> poison, bfloat %397, i64 0, !dbg !40
522
+ %444 = insertelement <2 x bfloat> %443, bfloat %398, i64 1, !dbg !40
523
+ %445 = bitcast <2 x bfloat> %444 to i32, !dbg !40
524
+ %446 = insertelement <2 x bfloat> poison, bfloat %399, i64 0, !dbg !40
525
+ %447 = insertelement <2 x bfloat> %446, bfloat %400, i64 1, !dbg !40
526
+ %448 = bitcast <2 x bfloat> %447 to i32, !dbg !40
527
+ 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
528
+ %449 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %12, !dbg !41
529
+ %450 = bitcast float %333 to i32, !dbg !41
530
+ %451 = bitcast float %334 to i32, !dbg !41
531
+ %452 = bitcast float %335 to i32, !dbg !41
532
+ %453 = bitcast float %336 to i32, !dbg !41
533
+ 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
534
+ tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41
535
+ %454 = shl nuw nsw i32 %11, 4, !dbg !41
536
+ %455 = and i32 %454, 48, !dbg !41
537
+ %456 = and i32 %11, 12, !dbg !41
538
+ %457 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %455, !dbg !41
539
+ %458 = getelementptr inbounds nuw i8, ptr addrspace(3) %457, i32 %456, !dbg !41
540
+ %459 = load i32, ptr addrspace(3) %458, align 4, !dbg !41
541
+ %460 = icmp eq i32 %12, 0, !dbg !41
542
+ %461 = and i1 %460, %332, !dbg !41
543
+ 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
544
+ ret void, !dbg !42
545
+ }
546
+
547
+ ; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
548
+ declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1
549
+
550
+ ; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
551
+ declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1
552
+
553
+ ; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite)
554
+ declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2
555
+
556
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
557
+ declare float @llvm.nvvm.div.full(float, float) #3
558
+
559
+ ; Function Attrs: convergent nocallback nounwind
560
+ declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4
561
+
562
+ declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5
563
+
564
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
565
+ declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3
566
+
567
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
568
+ declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3
569
+
570
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
571
+ declare float @llvm.nvvm.sqrt.rn.f(float) #3
572
+
573
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
574
+ declare float @llvm.nvvm.sqrt.approx.f(float) #3
575
+
576
+ attributes #0 = { nounwind "nvvm.reqntid"="64" }
577
+ attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) }
578
+ attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) }
579
+ attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) }
580
+ attributes #4 = { convergent nocallback nounwind }
581
+ 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" }
582
+ attributes #6 = { nounwind }
583
+
584
+ !llvm.dbg.cu = !{!0}
585
+ !llvm.module.flags = !{!2, !3}
586
+ !llvm.ident = !{!4}
587
+
588
+ !0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly)
589
+ !1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm")
590
+ !2 = !{i32 2, !"Debug Info Version", i32 3}
591
+ !3 = !{i32 4, !"nvvm-reflect-ftz", i32 1}
592
+ !4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"}
593
+ !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)
594
+ !6 = !DISubroutineType(cc: DW_CC_normal, types: !7)
595
+ !7 = !{null, !8, !8, !10, !11, !12, !8, !8}
596
+ !8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1)
597
+ !9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed)
598
+ !10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1)
599
+ !11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float)
600
+ !12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
601
+ !13 = !DILocation(line: 70, column: 24, scope: !5)
602
+ !14 = !DILocation(line: 70, column: 30, scope: !5)
603
+ !15 = !DILocation(line: 71, column: 16, scope: !5)
604
+ !16 = !DILocation(line: 71, column: 34, scope: !5)
605
+ !17 = !DILocation(line: 71, column: 21, scope: !5)
606
+ !18 = !DILocation(line: 73, column: 16, scope: !5)
607
+ !19 = !DILocation(line: 74, column: 30, scope: !5)
608
+ !20 = !DILocation(line: 75, column: 29, scope: !5)
609
+ !21 = !DILocation(line: 75, column: 14, scope: !5)
610
+ !22 = !DILocation(line: 75, column: 33, scope: !5)
611
+ !23 = !DILocation(line: 79, column: 18, scope: !5)
612
+ !24 = !DILocation(line: 79, column: 47, scope: !5)
613
+ !25 = !DILocation(line: 80, column: 38, scope: !5)
614
+ !26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29)
615
+ !27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0)
616
+ !28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language")
617
+ !29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30)
618
+ !30 = !DILocation(line: 80, column: 43, scope: !31)
619
+ !31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0)
620
+ !32 = !DILocation(line: 80, column: 48, scope: !5)
621
+ !33 = !DILocation(line: 80, column: 25, scope: !5)
622
+ !34 = !DILocation(line: 77, column: 20, scope: !5)
623
+ !35 = !DILocation(line: 76, column: 14, scope: !5)
624
+ !36 = !DILocation(line: 76, column: 33, scope: !5)
625
+ !37 = !DILocation(line: 80, column: 17, scope: !5)
626
+ !38 = !DILocation(line: 81, column: 16, scope: !5)
627
+ !39 = !DILocation(line: 83, column: 25, scope: !5)
628
+ !40 = !DILocation(line: 83, column: 18, scope: !5)
629
+ !41 = !DILocation(line: 84, column: 21, scope: !5)
630
+ !42 = !DILocation(line: 84, column: 4, scope: !5)
triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ptx ADDED
@@ -0,0 +1,638 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ //
2
+ // Generated by LLVM NVPTX Back-End
3
+ //
4
+
5
+ .version 9.1
6
+ .target sm_100a
7
+ .address_size 64
8
+
9
+ // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel
10
+ .extern .shared .align 16 .b8 global_smem[];
11
+ .global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90};
12
+ .global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84};
13
+ // @l2norm_fwd_kernel
14
+ .visible .entry l2norm_fwd_kernel(
15
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0,
16
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1,
17
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2,
18
+ .param .f32 l2norm_fwd_kernel_param_3,
19
+ .param .u32 l2norm_fwd_kernel_param_4,
20
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5,
21
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6
22
+ )
23
+ .reqntid 64
24
+ {
25
+ .reg .pred %p<8>;
26
+ .reg .b16 %rs<33>;
27
+ .reg .b32 %r<193>;
28
+ .reg .b64 %rd<39>;
29
+ .loc 1 69 0 // l2norm.py:69:0
30
+ $L__func_begin0:
31
+ .loc 1 69 0 // l2norm.py:69:0
32
+
33
+ // %bb.0: // %__nv_sqrtf.exit
34
+ ld.param.b64 %rd10, [l2norm_fwd_kernel_param_0];
35
+ ld.param.b64 %rd11, [l2norm_fwd_kernel_param_1];
36
+ $L__tmp0:
37
+ .loc 1 70 24 // l2norm.py:70:24
38
+ mov.u32 %r40, %ctaid.x;
39
+ ld.param.b64 %rd12, [l2norm_fwd_kernel_param_2];
40
+ .loc 1 71 16 // l2norm.py:71:16
41
+ mul.wide.u32 %rd13, %r40, 16;
42
+ ld.param.b32 %r41, [l2norm_fwd_kernel_param_3];
43
+ ld.param.s32 %rd14, [l2norm_fwd_kernel_param_4];
44
+ .loc 1 71 34 // l2norm.py:71:34
45
+ mov.u32 %r42, %tid.x;
46
+ and.b32 %r43, %r42, 48;
47
+ bfe.u32 %r44, %r42, 4, 2;
48
+ or.b32 %r45, %r44, 4;
49
+ or.b32 %r46, %r44, 8;
50
+ or.b32 %r47, %r44, 12;
51
+ and.b32 %r48, %r42, 15;
52
+ .loc 1 71 21 // l2norm.py:71:21
53
+ cvt.u64.u32 %rd15, %r44;
54
+ cvt.u64.u32 %rd16, %r45;
55
+ cvt.u64.u32 %rd17, %r46;
56
+ cvt.u64.u32 %rd18, %r47;
57
+ or.b64 %rd19, %rd13, %rd15;
58
+ or.b64 %rd20, %rd13, %rd16;
59
+ or.b64 %rd21, %rd13, %rd17;
60
+ or.b64 %rd22, %rd13, %rd18;
61
+ .loc 1 73 16 // l2norm.py:73:16
62
+ setp.lt.s64 %p1, %rd19, %rd14;
63
+ setp.lt.s64 %p2, %rd20, %rd14;
64
+ setp.lt.s64 %p3, %rd21, %rd14;
65
+ setp.lt.s64 %p4, %rd22, %rd14;
66
+ .loc 1 74 30 // l2norm.py:74:30
67
+ shl.b32 %r49, %r48, 3;
68
+ .loc 1 75 14 // l2norm.py:75:14
69
+ shl.b64 %rd23, %rd19, 8;
70
+ add.s64 %rd24, %rd10, %rd23;
71
+ shl.b64 %rd25, %rd20, 8;
72
+ add.s64 %rd26, %rd10, %rd25;
73
+ shl.b64 %rd27, %rd21, 8;
74
+ add.s64 %rd28, %rd10, %rd27;
75
+ shl.b64 %rd29, %rd22, 8;
76
+ add.s64 %rd30, %rd10, %rd29;
77
+ .loc 1 75 33 // l2norm.py:75:33
78
+ mul.wide.u32 %rd31, %r49, 2;
79
+ add.s64 %rd1, %rd24, %rd31;
80
+ add.s64 %rd2, %rd26, %rd31;
81
+ add.s64 %rd3, %rd28, %rd31;
82
+ add.s64 %rd4, %rd30, %rd31;
83
+ mov.b32 %r5, 0;
84
+ .loc 1 79 18 // l2norm.py:79:18
85
+ // begin inline asm
86
+ mov.u32 %r1, %r5;
87
+ mov.u32 %r2, %r5;
88
+ mov.u32 %r3, %r5;
89
+ mov.u32 %r4, %r5;
90
+ @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ];
91
+ // end inline asm
92
+ mov.b32 {%rs1, %rs2}, %r1;
93
+ mov.b32 {%rs3, %rs4}, %r2;
94
+ mov.b32 {%rs5, %rs6}, %r3;
95
+ mov.b32 {%rs7, %rs8}, %r4;
96
+ // begin inline asm
97
+ mov.u32 %r6, %r5;
98
+ mov.u32 %r7, %r5;
99
+ mov.u32 %r8, %r5;
100
+ mov.u32 %r9, %r5;
101
+ @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ];
102
+ // end inline asm
103
+ mov.b32 {%rs9, %rs10}, %r6;
104
+ mov.b32 {%rs11, %rs12}, %r7;
105
+ mov.b32 {%rs13, %rs14}, %r8;
106
+ mov.b32 {%rs15, %rs16}, %r9;
107
+ // begin inline asm
108
+ mov.u32 %r10, %r5;
109
+ mov.u32 %r11, %r5;
110
+ mov.u32 %r12, %r5;
111
+ mov.u32 %r13, %r5;
112
+ @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ];
113
+ // end inline asm
114
+ mov.b32 {%rs17, %rs18}, %r10;
115
+ mov.b32 {%rs19, %rs20}, %r11;
116
+ mov.b32 {%rs21, %rs22}, %r12;
117
+ mov.b32 {%rs23, %rs24}, %r13;
118
+ // begin inline asm
119
+ mov.u32 %r14, %r5;
120
+ mov.u32 %r15, %r5;
121
+ mov.u32 %r16, %r5;
122
+ mov.u32 %r17, %r5;
123
+ @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ];
124
+ // end inline asm
125
+ mov.b32 {%rs25, %rs26}, %r14;
126
+ mov.b32 {%rs27, %rs28}, %r15;
127
+ mov.b32 {%rs29, %rs30}, %r16;
128
+ mov.b32 {%rs31, %rs32}, %r17;
129
+ .loc 1 79 47 // l2norm.py:79:47
130
+ cvt.f32.bf16 %r50, %rs1;
131
+ cvt.f32.bf16 %r51, %rs2;
132
+ cvt.f32.bf16 %r52, %rs3;
133
+ cvt.f32.bf16 %r53, %rs4;
134
+ cvt.f32.bf16 %r54, %rs5;
135
+ cvt.f32.bf16 %r55, %rs6;
136
+ cvt.f32.bf16 %r56, %rs7;
137
+ cvt.f32.bf16 %r57, %rs8;
138
+ cvt.f32.bf16 %r58, %rs9;
139
+ cvt.f32.bf16 %r59, %rs10;
140
+ cvt.f32.bf16 %r60, %rs11;
141
+ cvt.f32.bf16 %r61, %rs12;
142
+ cvt.f32.bf16 %r62, %rs13;
143
+ cvt.f32.bf16 %r63, %rs14;
144
+ cvt.f32.bf16 %r64, %rs15;
145
+ cvt.f32.bf16 %r65, %rs16;
146
+ cvt.f32.bf16 %r66, %rs17;
147
+ cvt.f32.bf16 %r67, %rs18;
148
+ cvt.f32.bf16 %r68, %rs19;
149
+ cvt.f32.bf16 %r69, %rs20;
150
+ cvt.f32.bf16 %r70, %rs21;
151
+ cvt.f32.bf16 %r71, %rs22;
152
+ cvt.f32.bf16 %r72, %rs23;
153
+ cvt.f32.bf16 %r73, %rs24;
154
+ cvt.f32.bf16 %r74, %rs25;
155
+ cvt.f32.bf16 %r75, %rs26;
156
+ cvt.f32.bf16 %r76, %rs27;
157
+ cvt.f32.bf16 %r77, %rs28;
158
+ cvt.f32.bf16 %r78, %rs29;
159
+ cvt.f32.bf16 %r79, %rs30;
160
+ cvt.f32.bf16 %r80, %rs31;
161
+ cvt.f32.bf16 %r81, %rs32;
162
+ .loc 1 80 38 // l2norm.py:80:38
163
+ mul.f32 %r82, %r51, %r51;
164
+ mul.f32 %r83, %r59, %r59;
165
+ mul.f32 %r84, %r67, %r67;
166
+ mul.f32 %r85, %r75, %r75;
167
+ $L__tmp1:
168
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
169
+ fma.rn.f32.bf16 %r86, %rs1, %rs1, %r82;
170
+ fma.rn.f32.bf16 %r87, %rs3, %rs3, %r86;
171
+ fma.rn.f32.bf16 %r88, %rs4, %rs4, %r87;
172
+ fma.rn.f32.bf16 %r89, %rs5, %rs5, %r88;
173
+ fma.rn.f32.bf16 %r90, %rs6, %rs6, %r89;
174
+ fma.rn.f32.bf16 %r91, %rs7, %rs7, %r90;
175
+ fma.rn.f32.bf16 %r92, %rs8, %rs8, %r91;
176
+ fma.rn.f32.bf16 %r93, %rs9, %rs9, %r83;
177
+ fma.rn.f32.bf16 %r94, %rs11, %rs11, %r93;
178
+ fma.rn.f32.bf16 %r95, %rs12, %rs12, %r94;
179
+ fma.rn.f32.bf16 %r96, %rs13, %rs13, %r95;
180
+ fma.rn.f32.bf16 %r97, %rs14, %rs14, %r96;
181
+ fma.rn.f32.bf16 %r98, %rs15, %rs15, %r97;
182
+ fma.rn.f32.bf16 %r99, %rs16, %rs16, %r98;
183
+ fma.rn.f32.bf16 %r100, %rs17, %rs17, %r84;
184
+ fma.rn.f32.bf16 %r101, %rs19, %rs19, %r100;
185
+ fma.rn.f32.bf16 %r102, %rs20, %rs20, %r101;
186
+ fma.rn.f32.bf16 %r103, %rs21, %rs21, %r102;
187
+ fma.rn.f32.bf16 %r104, %rs22, %rs22, %r103;
188
+ fma.rn.f32.bf16 %r105, %rs23, %rs23, %r104;
189
+ fma.rn.f32.bf16 %r106, %rs24, %rs24, %r105;
190
+ fma.rn.f32.bf16 %r107, %rs25, %rs25, %r85;
191
+ fma.rn.f32.bf16 %r108, %rs27, %rs27, %r107;
192
+ fma.rn.f32.bf16 %r109, %rs28, %rs28, %r108;
193
+ fma.rn.f32.bf16 %r110, %rs29, %rs29, %r109;
194
+ fma.rn.f32.bf16 %r111, %rs30, %rs30, %r110;
195
+ fma.rn.f32.bf16 %r112, %rs31, %rs31, %r111;
196
+ fma.rn.f32.bf16 %r113, %rs32, %rs32, %r112;
197
+ $L__tmp2:
198
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
199
+ shfl.sync.bfly.b32 %r114, %r92, 8, 31, -1;
200
+ $L__tmp3:
201
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
202
+ add.f32 %r115, %r92, %r114;
203
+ $L__tmp4:
204
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
205
+ shfl.sync.bfly.b32 %r116, %r115, 4, 31, -1;
206
+ $L__tmp5:
207
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
208
+ add.f32 %r117, %r115, %r116;
209
+ $L__tmp6:
210
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
211
+ shfl.sync.bfly.b32 %r118, %r117, 2, 31, -1;
212
+ $L__tmp7:
213
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
214
+ add.f32 %r119, %r117, %r118;
215
+ $L__tmp8:
216
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
217
+ shfl.sync.bfly.b32 %r120, %r119, 1, 31, -1;
218
+ $L__tmp9:
219
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
220
+ add.f32 %r121, %r119, %r120;
221
+ $L__tmp10:
222
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
223
+ shfl.sync.bfly.b32 %r122, %r99, 8, 31, -1;
224
+ $L__tmp11:
225
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
226
+ add.f32 %r123, %r99, %r122;
227
+ $L__tmp12:
228
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
229
+ shfl.sync.bfly.b32 %r124, %r123, 4, 31, -1;
230
+ $L__tmp13:
231
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
232
+ add.f32 %r125, %r123, %r124;
233
+ $L__tmp14:
234
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
235
+ shfl.sync.bfly.b32 %r126, %r125, 2, 31, -1;
236
+ $L__tmp15:
237
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
238
+ add.f32 %r127, %r125, %r126;
239
+ $L__tmp16:
240
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
241
+ shfl.sync.bfly.b32 %r128, %r127, 1, 31, -1;
242
+ $L__tmp17:
243
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
244
+ add.f32 %r129, %r127, %r128;
245
+ $L__tmp18:
246
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
247
+ shfl.sync.bfly.b32 %r130, %r106, 8, 31, -1;
248
+ $L__tmp19:
249
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
250
+ add.f32 %r131, %r106, %r130;
251
+ $L__tmp20:
252
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
253
+ shfl.sync.bfly.b32 %r132, %r131, 4, 31, -1;
254
+ $L__tmp21:
255
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
256
+ add.f32 %r133, %r131, %r132;
257
+ $L__tmp22:
258
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
259
+ shfl.sync.bfly.b32 %r134, %r133, 2, 31, -1;
260
+ $L__tmp23:
261
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
262
+ add.f32 %r135, %r133, %r134;
263
+ $L__tmp24:
264
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
265
+ shfl.sync.bfly.b32 %r136, %r135, 1, 31, -1;
266
+ $L__tmp25:
267
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
268
+ add.f32 %r137, %r135, %r136;
269
+ $L__tmp26:
270
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
271
+ shfl.sync.bfly.b32 %r138, %r113, 8, 31, -1;
272
+ $L__tmp27:
273
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
274
+ add.f32 %r139, %r113, %r138;
275
+ $L__tmp28:
276
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
277
+ shfl.sync.bfly.b32 %r140, %r139, 4, 31, -1;
278
+ $L__tmp29:
279
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
280
+ add.f32 %r141, %r139, %r140;
281
+ $L__tmp30:
282
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
283
+ shfl.sync.bfly.b32 %r142, %r141, 2, 31, -1;
284
+ $L__tmp31:
285
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
286
+ add.f32 %r143, %r141, %r142;
287
+ $L__tmp32:
288
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
289
+ shfl.sync.bfly.b32 %r144, %r143, 1, 31, -1;
290
+ $L__tmp33:
291
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
292
+ add.f32 %r145, %r143, %r144;
293
+ $L__tmp34:
294
+ .loc 1 80 48 // l2norm.py:80:48
295
+ add.f32 %r146, %r41, %r121;
296
+ add.f32 %r147, %r41, %r129;
297
+ add.f32 %r148, %r41, %r137;
298
+ add.f32 %r149, %r41, %r145;
299
+ .loc 1 80 25 // l2norm.py:80:25
300
+ sqrt.approx.ftz.f32 %r150, %r146;
301
+ sqrt.approx.ftz.f32 %r151, %r147;
302
+ sqrt.approx.ftz.f32 %r152, %r148;
303
+ sqrt.approx.ftz.f32 %r153, %r149;
304
+ .loc 1 71 21 // l2norm.py:71:21
305
+ cvt.u64.u32 %rd32, %r48;
306
+ or.b64 %rd33, %rd13, %rd32;
307
+ .loc 1 77 20 // l2norm.py:77:20
308
+ shl.b64 %rd34, %rd33, 2;
309
+ add.s64 %rd9, %rd12, %rd34;
310
+ .loc 1 76 14 // l2norm.py:76:14
311
+ add.s64 %rd35, %rd11, %rd29;
312
+ .loc 1 76 33 // l2norm.py:76:33
313
+ add.s64 %rd8, %rd35, %rd31;
314
+ .loc 1 76 14 // l2norm.py:76:14
315
+ add.s64 %rd36, %rd11, %rd27;
316
+ .loc 1 76 33 // l2norm.py:76:33
317
+ add.s64 %rd7, %rd36, %rd31;
318
+ .loc 1 76 14 // l2norm.py:76:14
319
+ add.s64 %rd37, %rd11, %rd25;
320
+ .loc 1 76 33 // l2norm.py:76:33
321
+ add.s64 %rd6, %rd37, %rd31;
322
+ .loc 1 76 14 // l2norm.py:76:14
323
+ add.s64 %rd38, %rd11, %rd23;
324
+ .loc 1 76 33 // l2norm.py:76:33
325
+ add.s64 %rd5, %rd38, %rd31;
326
+ .loc 1 73 16 // l2norm.py:73:16
327
+ setp.lt.s64 %p6, %rd33, %rd14;
328
+ mov.b32 %r154, 0f3F800000;
329
+ .loc 1 80 17 // l2norm.py:80:17
330
+ div.full.f32 %r35, %r154, %r150;
331
+ div.full.f32 %r36, %r154, %r151;
332
+ div.full.f32 %r37, %r154, %r152;
333
+ div.full.f32 %r38, %r154, %r153;
334
+ .loc 1 81 16 // l2norm.py:81:16
335
+ mul.f32 %r155, %r35, %r50;
336
+ mul.f32 %r156, %r35, %r51;
337
+ mul.f32 %r157, %r35, %r52;
338
+ mul.f32 %r158, %r35, %r53;
339
+ mul.f32 %r159, %r35, %r54;
340
+ mul.f32 %r160, %r35, %r55;
341
+ mul.f32 %r161, %r35, %r56;
342
+ mul.f32 %r162, %r35, %r57;
343
+ mul.f32 %r163, %r36, %r58;
344
+ mul.f32 %r164, %r36, %r59;
345
+ mul.f32 %r165, %r36, %r60;
346
+ mul.f32 %r166, %r36, %r61;
347
+ mul.f32 %r167, %r36, %r62;
348
+ mul.f32 %r168, %r36, %r63;
349
+ mul.f32 %r169, %r36, %r64;
350
+ mul.f32 %r170, %r36, %r65;
351
+ mul.f32 %r171, %r37, %r66;
352
+ mul.f32 %r172, %r37, %r67;
353
+ mul.f32 %r173, %r37, %r68;
354
+ mul.f32 %r174, %r37, %r69;
355
+ mul.f32 %r175, %r37, %r70;
356
+ mul.f32 %r176, %r37, %r71;
357
+ mul.f32 %r177, %r37, %r72;
358
+ mul.f32 %r178, %r37, %r73;
359
+ mul.f32 %r179, %r38, %r74;
360
+ mul.f32 %r180, %r38, %r75;
361
+ mul.f32 %r181, %r38, %r76;
362
+ mul.f32 %r182, %r38, %r77;
363
+ mul.f32 %r183, %r38, %r78;
364
+ mul.f32 %r184, %r38, %r79;
365
+ mul.f32 %r185, %r38, %r80;
366
+ mul.f32 %r186, %r38, %r81;
367
+ .loc 1 83 18 // l2norm.py:83:18
368
+ cvt.rn.bf16x2.f32 %r18, %r156, %r155;
369
+ cvt.rn.bf16x2.f32 %r19, %r158, %r157;
370
+ cvt.rn.bf16x2.f32 %r20, %r160, %r159;
371
+ cvt.rn.bf16x2.f32 %r21, %r162, %r161;
372
+ // begin inline asm
373
+ @%p1 st.global.v4.b32 [ %rd5 + 0 ], { %r18, %r19, %r20, %r21 };
374
+ // end inline asm
375
+ cvt.rn.bf16x2.f32 %r22, %r164, %r163;
376
+ cvt.rn.bf16x2.f32 %r23, %r166, %r165;
377
+ cvt.rn.bf16x2.f32 %r24, %r168, %r167;
378
+ cvt.rn.bf16x2.f32 %r25, %r170, %r169;
379
+ // begin inline asm
380
+ @%p2 st.global.v4.b32 [ %rd6 + 0 ], { %r22, %r23, %r24, %r25 };
381
+ // end inline asm
382
+ cvt.rn.bf16x2.f32 %r26, %r172, %r171;
383
+ cvt.rn.bf16x2.f32 %r27, %r174, %r173;
384
+ cvt.rn.bf16x2.f32 %r28, %r176, %r175;
385
+ cvt.rn.bf16x2.f32 %r29, %r178, %r177;
386
+ // begin inline asm
387
+ @%p3 st.global.v4.b32 [ %rd7 + 0 ], { %r26, %r27, %r28, %r29 };
388
+ // end inline asm
389
+ cvt.rn.bf16x2.f32 %r30, %r180, %r179;
390
+ cvt.rn.bf16x2.f32 %r31, %r182, %r181;
391
+ cvt.rn.bf16x2.f32 %r32, %r184, %r183;
392
+ cvt.rn.bf16x2.f32 %r33, %r186, %r185;
393
+ // begin inline asm
394
+ @%p4 st.global.v4.b32 [ %rd8 + 0 ], { %r30, %r31, %r32, %r33 };
395
+ // end inline asm
396
+ .loc 1 84 21 // l2norm.py:84:21
397
+ mov.b32 %r187, global_smem;
398
+ add.s32 %r34, %r187, %r43;
399
+ // begin inline asm
400
+ st.shared.v4.b32 [ %r34 + 0 ], { %r35, %r36, %r37, %r38 };
401
+ // end inline asm
402
+ bar.sync 0;
403
+ shl.b32 %r188, %r42, 4;
404
+ and.b32 %r189, %r188, 48;
405
+ and.b32 %r190, %r42, 12;
406
+ add.s32 %r191, %r187, %r189;
407
+ add.s32 %r192, %r191, %r190;
408
+ ld.shared.b32 %r39, [%r192];
409
+ setp.eq.b32 %p7, %r43, 0;
410
+ and.pred %p5, %p7, %p6;
411
+ // begin inline asm
412
+ @%p5 st.global.b32 [ %rd9 + 0 ], { %r39 };
413
+ // end inline asm
414
+ .loc 1 84 4 // l2norm.py:84:4
415
+ ret;
416
+ $L__tmp35:
417
+ $L__func_end0:
418
+ // -- End function
419
+ }
420
+ .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py"
421
+ .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py"
422
+ .section .debug_abbrev
423
+ {
424
+ .b8 1 // Abbreviation Code
425
+ .b8 17 // DW_TAG_compile_unit
426
+ .b8 1 // DW_CHILDREN_yes
427
+ .b8 37 // DW_AT_producer
428
+ .b8 8 // DW_FORM_string
429
+ .b8 19 // DW_AT_language
430
+ .b8 5 // DW_FORM_data2
431
+ .b8 3 // DW_AT_name
432
+ .b8 8 // DW_FORM_string
433
+ .b8 16 // DW_AT_stmt_list
434
+ .b8 6 // DW_FORM_data4
435
+ .b8 27 // DW_AT_comp_dir
436
+ .b8 8 // DW_FORM_string
437
+ .b8 0 // EOM(1)
438
+ .b8 0 // EOM(2)
439
+ .b8 2 // Abbreviation Code
440
+ .b8 46 // DW_TAG_subprogram
441
+ .b8 0 // DW_CHILDREN_no
442
+ .b8 3 // DW_AT_name
443
+ .b8 8 // DW_FORM_string
444
+ .b8 32 // DW_AT_inline
445
+ .b8 11 // DW_FORM_data1
446
+ .b8 0 // EOM(1)
447
+ .b8 0 // EOM(2)
448
+ .b8 3 // Abbreviation Code
449
+ .b8 46 // DW_TAG_subprogram
450
+ .b8 1 // DW_CHILDREN_yes
451
+ .b8 17 // DW_AT_low_pc
452
+ .b8 1 // DW_FORM_addr
453
+ .b8 18 // DW_AT_high_pc
454
+ .b8 1 // DW_FORM_addr
455
+ .b8 49 // DW_AT_abstract_origin
456
+ .b8 19 // DW_FORM_ref4
457
+ .b8 0 // EOM(1)
458
+ .b8 0 // EOM(2)
459
+ .b8 4 // Abbreviation Code
460
+ .b8 29 // DW_TAG_inlined_subroutine
461
+ .b8 1 // DW_CHILDREN_yes
462
+ .b8 49 // DW_AT_abstract_origin
463
+ .b8 19 // DW_FORM_ref4
464
+ .b8 17 // DW_AT_low_pc
465
+ .b8 1 // DW_FORM_addr
466
+ .b8 18 // DW_AT_high_pc
467
+ .b8 1 // DW_FORM_addr
468
+ .b8 88 // DW_AT_call_file
469
+ .b8 11 // DW_FORM_data1
470
+ .b8 89 // DW_AT_call_line
471
+ .b8 11 // DW_FORM_data1
472
+ .b8 87 // DW_AT_call_column
473
+ .b8 11 // DW_FORM_data1
474
+ .b8 0 // EOM(1)
475
+ .b8 0 // EOM(2)
476
+ .b8 5 // Abbreviation Code
477
+ .b8 29 // DW_TAG_inlined_subroutine
478
+ .b8 0 // DW_CHILDREN_no
479
+ .b8 49 // DW_AT_abstract_origin
480
+ .b8 19 // DW_FORM_ref4
481
+ .b8 17 // DW_AT_low_pc
482
+ .b8 1 // DW_FORM_addr
483
+ .b8 18 // DW_AT_high_pc
484
+ .b8 1 // DW_FORM_addr
485
+ .b8 88 // DW_AT_call_file
486
+ .b8 11 // DW_FORM_data1
487
+ .b8 89 // DW_AT_call_line
488
+ .b8 5 // DW_FORM_data2
489
+ .b8 87 // DW_AT_call_column
490
+ .b8 11 // DW_FORM_data1
491
+ .b8 0 // EOM(1)
492
+ .b8 0 // EOM(2)
493
+ .b8 0 // EOM(3)
494
+ }
495
+ .section .debug_info
496
+ {
497
+ .b32 196 // Length of Unit
498
+ .b8 2 // DWARF version number
499
+ .b8 0
500
+ .b32 .debug_abbrev // Offset Into Abbrev. Section
501
+ .b8 8 // Address Size (in bytes)
502
+ .b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit
503
+ .b8 116 // DW_AT_producer
504
+ .b8 114
505
+ .b8 105
506
+ .b8 116
507
+ .b8 111
508
+ .b8 110
509
+ .b8 0
510
+ .b8 2 // DW_AT_language
511
+ .b8 0
512
+ .b8 108 // DW_AT_name
513
+ .b8 50
514
+ .b8 110
515
+ .b8 111
516
+ .b8 114
517
+ .b8 109
518
+ .b8 46
519
+ .b8 112
520
+ .b8 121
521
+ .b8 0
522
+ .b32 .debug_line // DW_AT_stmt_list
523
+ .b8 47 // DW_AT_comp_dir
524
+ .b8 104
525
+ .b8 111
526
+ .b8 109
527
+ .b8 101
528
+ .b8 47
529
+ .b8 115
530
+ .b8 105
531
+ .b8 108
532
+ .b8 118
533
+ .b8 101
534
+ .b8 114
535
+ .b8 47
536
+ .b8 97
537
+ .b8 102
538
+ .b8 102
539
+ .b8 105
540
+ .b8 110
541
+ .b8 101
542
+ .b8 47
543
+ .b8 46
544
+ .b8 118
545
+ .b8 101
546
+ .b8 110
547
+ .b8 118
548
+ .b8 47
549
+ .b8 108
550
+ .b8 105
551
+ .b8 98
552
+ .b8 47
553
+ .b8 112
554
+ .b8 121
555
+ .b8 116
556
+ .b8 104
557
+ .b8 111
558
+ .b8 110
559
+ .b8 51
560
+ .b8 46
561
+ .b8 49
562
+ .b8 50
563
+ .b8 47
564
+ .b8 115
565
+ .b8 105
566
+ .b8 116
567
+ .b8 101
568
+ .b8 45
569
+ .b8 112
570
+ .b8 97
571
+ .b8 99
572
+ .b8 107
573
+ .b8 97
574
+ .b8 103
575
+ .b8 101
576
+ .b8 115
577
+ .b8 47
578
+ .b8 102
579
+ .b8 108
580
+ .b8 97
581
+ .b8 47
582
+ .b8 109
583
+ .b8 111
584
+ .b8 100
585
+ .b8 117
586
+ .b8 108
587
+ .b8 101
588
+ .b8 115
589
+ .b8 47
590
+ .b8 110
591
+ .b8 111
592
+ .b8 114
593
+ .b8 109
594
+ .b8 0
595
+ .b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram
596
+ .b8 108 // DW_AT_name
597
+ .b8 50
598
+ .b8 110
599
+ .b8 111
600
+ .b8 114
601
+ .b8 109
602
+ .b8 95
603
+ .b8 102
604
+ .b8 119
605
+ .b8 100
606
+ .b8 95
607
+ .b8 107
608
+ .b8 101
609
+ .b8 114
610
+ .b8 110
611
+ .b8 101
612
+ .b8 108
613
+ .b8 0
614
+ .b8 1 // DW_AT_inline
615
+ .b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram
616
+ .b64 $L__func_begin0 // DW_AT_low_pc
617
+ .b64 $L__func_end0 // DW_AT_high_pc
618
+ .b32 107 // DW_AT_abstract_origin
619
+ .b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine
620
+ .b32 107 // DW_AT_abstract_origin
621
+ .b64 $L__tmp1 // DW_AT_low_pc
622
+ .b64 $L__tmp34 // DW_AT_high_pc
623
+ .b8 1 // DW_AT_call_file
624
+ .b8 80 // DW_AT_call_line
625
+ .b8 43 // DW_AT_call_column
626
+ .b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine
627
+ .b32 107 // DW_AT_abstract_origin
628
+ .b64 $L__tmp1 // DW_AT_low_pc
629
+ .b64 $L__tmp34 // DW_AT_high_pc
630
+ .b8 2 // DW_AT_call_file
631
+ .b8 37 // DW_AT_call_line
632
+ .b8 1
633
+ .b8 36 // DW_AT_call_column
634
+ .b8 0 // End Of Children Mark
635
+ .b8 0 // End Of Children Mark
636
+ .b8 0 // End Of Children Mark
637
+ }
638
+ .section .debug_macinfo { }
triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.source ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
2
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0)
3
+ #loc38 = loc(unknown)
4
+ #loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0)
5
+ #loc45 = loc("x"(#loc))
6
+ #loc46 = loc("y"(#loc))
7
+ #loc47 = loc("rstd"(#loc))
8
+ #loc48 = loc("eps"(#loc))
9
+ #loc49 = loc("T"(#loc))
10
+ #loc81 = loc("input"(#loc36))
11
+ #loc82 = loc("a"(#loc41))
12
+ #loc83 = loc("b"(#loc41))
13
+ module {
14
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %i_t = tt.get_program_id x : i32 loc(#loc50)
16
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51)
17
+ %o_t = arith.constant 16 : i32 loc(#loc52)
18
+ %o_t_1 = arith.constant 16 : i64 loc(#loc52)
19
+ %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52)
20
+ %o_t_3 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc53)
21
+ %o_t_4 = arith.extsi %o_t_3 : tensor<16xi32> to tensor<16xi64> loc(#loc54)
22
+ %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<16xi64> loc(#loc54)
23
+ %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<16xi64> loc(#loc54)
24
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55)
25
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc56)
26
+ %m_t_7 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc56)
27
+ %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<16xi64> loc(#loc56)
28
+ %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc57)
29
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58)
30
+ %m_x_10 = arith.constant 128 : i32 loc(#loc59)
31
+ %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59)
32
+ %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59)
33
+ %m_x_13 = tt.broadcast %m_x : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc60)
34
+ %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc60)
35
+ %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<16x128xi1> loc(#loc60)
36
+ %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc61)
37
+ %p_x_16 = arith.constant 128 : i32 loc(#loc62)
38
+ %p_x_17 = arith.constant 128 : i64 loc(#loc62)
39
+ %p_x_18 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc62)
40
+ %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<16x1xi64> loc(#loc62)
41
+ %p_x_20 = tt.splat %x : !tt.ptr<bf16> -> tensor<16x1x!tt.ptr<bf16>> loc(#loc63)
42
+ %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<16x1x!tt.ptr<bf16>>, tensor<16x1xi64> loc(#loc63)
43
+ %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64)
44
+ %p_x_23 = tt.broadcast %p_x_21 : tensor<16x1x!tt.ptr<bf16>> -> tensor<16x128x!tt.ptr<bf16>> loc(#loc65)
45
+ %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc65)
46
+ %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<16x128x!tt.ptr<bf16>>, tensor<16x128xi32> loc(#loc65)
47
+ %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc66)
48
+ %p_y_26 = arith.constant 128 : i32 loc(#loc67)
49
+ %p_y_27 = arith.constant 128 : i64 loc(#loc67)
50
+ %p_y_28 = arith.constant dense<128> : tensor<16x1xi64> loc(#loc67)
51
+ %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<16x1xi64> loc(#loc67)
52
+ %p_y_30 = tt.splat %y : !tt.ptr<bf16> -> tensor<16x1x!tt.ptr<bf16>> loc(#loc68)
53
+ %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<16x1x!tt.ptr<bf16>>, tensor<16x1xi64> loc(#loc68)
54
+ %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69)
55
+ %p_y_33 = tt.broadcast %p_y_31 : tensor<16x1x!tt.ptr<bf16>> -> tensor<16x128x!tt.ptr<bf16>> loc(#loc70)
56
+ %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc70)
57
+ %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<16x128x!tt.ptr<bf16>>, tensor<16x128xi32> loc(#loc70)
58
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<16x!tt.ptr<f32>> loc(#loc71)
59
+ %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<16x!tt.ptr<f32>>, tensor<16xi64> loc(#loc71)
60
+ %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72)
61
+ %b_x_37 = arith.constant dense<0.000000e+00> : tensor<16x128xf32> loc(#loc72)
62
+ %b_x_38 = arith.truncf %b_x_37 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc72)
63
+ %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<16x128x!tt.ptr<bf16>> loc(#loc72)
64
+ %b_x_40 = arith.extf %b_x_39 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc73)
65
+ %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<16x128xf32> loc(#loc74)
66
+ %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%b_rstd) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc75)
67
+ %b_rstd_42 = tt.splat %eps : f32 -> tensor<16xf32> loc(#loc76)
68
+ %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<16xf32> loc(#loc76)
69
+ %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<16xf32> loc(#loc77)
70
+ %b_rstd_45 = arith.constant 1 : i32 loc(#loc78)
71
+ %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78)
72
+ %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<16xf32> loc(#loc78)
73
+ %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<16xf32> loc(#loc78)
74
+ %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc79)
75
+ %b_y_49 = tt.broadcast %b_y : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc80)
76
+ %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<16x128xf32> loc(#loc80)
77
+ %0 = arith.truncf %b_y_50 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc32)
78
+ tt.store %p_y_35, %0, %m_x_15 : tensor<16x128x!tt.ptr<bf16>> loc(#loc33)
79
+ tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<16x!tt.ptr<f32>> loc(#loc34)
80
+ tt.return loc(#loc35)
81
+ } loc(#loc)
82
+ tt.func private @triton.language.standard.sum__fp32S16_128S_c1_cFalse_cNone(%input: tensor<16x128xf32> loc("input"(#loc36))) -> tensor<16xf32> attributes {noinline = false} {
83
+ %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({
84
+ ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)):
85
+ %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37)
86
+ tt.reduce.return %2 : f32 loc(#loc37)
87
+ }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc37)
88
+ tt.return %0 : tensor<16xf32> loc(#loc39)
89
+ ^bb1: // no predecessors
90
+ %1 = ub.poison : tensor<16xf32> loc(#loc40)
91
+ tt.return %1 : tensor<16xf32> loc(#loc40)
92
+ } loc(#loc36)
93
+ tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} {
94
+ %0 = arith.addf %a, %b : f32 loc(#loc42)
95
+ tt.return %0 : f32 loc(#loc43)
96
+ ^bb1: // no predecessors
97
+ %1 = ub.poison : f32 loc(#loc44)
98
+ tt.return %1 : f32 loc(#loc44)
99
+ } loc(#loc41)
100
+ } loc(#loc)
101
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
102
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
103
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
104
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
105
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
106
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23)
107
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
108
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
109
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
110
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
111
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
112
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
113
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
114
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
115
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37)
116
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
117
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18)
118
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29)
119
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
120
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37)
121
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
122
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
123
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
124
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
125
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
126
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
127
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
128
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
129
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
130
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
131
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
132
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
133
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
134
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
135
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
136
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
137
+ #loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11)
138
+ #loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4)
139
+ #loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
140
+ #loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11)
141
+ #loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4)
142
+ #loc50 = loc("i_t"(#loc1))
143
+ #loc51 = loc("i_t"(#loc2))
144
+ #loc52 = loc("o_t"(#loc3))
145
+ #loc53 = loc("o_t"(#loc4))
146
+ #loc54 = loc("o_t"(#loc5))
147
+ #loc55 = loc("o_d"(#loc6))
148
+ #loc56 = loc("m_t"(#loc7))
149
+ #loc57 = loc("m_x"(#loc8))
150
+ #loc58 = loc("m_x"(#loc9))
151
+ #loc59 = loc("m_x"(#loc10))
152
+ #loc60 = loc("m_x"(#loc11))
153
+ #loc61 = loc("p_x"(#loc12))
154
+ #loc62 = loc("p_x"(#loc13))
155
+ #loc63 = loc("p_x"(#loc14))
156
+ #loc64 = loc("p_x"(#loc15))
157
+ #loc65 = loc("p_x"(#loc16))
158
+ #loc66 = loc("p_y"(#loc17))
159
+ #loc67 = loc("p_y"(#loc18))
160
+ #loc68 = loc("p_y"(#loc19))
161
+ #loc69 = loc("p_y"(#loc20))
162
+ #loc70 = loc("p_y"(#loc21))
163
+ #loc71 = loc("p_rstd"(#loc22))
164
+ #loc72 = loc("b_x"(#loc23))
165
+ #loc73 = loc("b_x"(#loc24))
166
+ #loc74 = loc("b_rstd"(#loc25))
167
+ #loc75 = loc("b_rstd"(#loc26))
168
+ #loc76 = loc("b_rstd"(#loc27))
169
+ #loc77 = loc("b_rstd"(#loc28))
170
+ #loc78 = loc("b_rstd"(#loc29))
171
+ #loc79 = loc("b_y"(#loc30))
172
+ #loc80 = loc("b_y"(#loc31))
triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttgir ADDED
@@ -0,0 +1,135 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}>
2
+ #blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}>
3
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
4
+ #loc1 = loc(unknown)
5
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
6
+ #loc34 = loc("x"(#loc))
7
+ #loc35 = loc("y"(#loc))
8
+ #loc36 = loc("rstd"(#loc))
9
+ #loc37 = loc("eps"(#loc))
10
+ #loc38 = loc("T"(#loc))
11
+ #loc59 = loc("b_rstd"(#loc23))
12
+ #loc66 = loc(callsite(#loc1 at #loc59))
13
+ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} {
14
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %cst = arith.constant dense<1.000000e+00> : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1)
16
+ %cst_0 = arith.constant dense<0.000000e+00> : tensor<16x128xbf16, #blocked> loc(#loc1)
17
+ %c16_i64 = arith.constant 16 : i64 loc(#loc1)
18
+ %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1)
19
+ %cst_2 = arith.constant dense<128> : tensor<16x1xi64, #blocked> loc(#loc1)
20
+ %i_t = tt.get_program_id x : i32 loc(#loc39)
21
+ %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40)
22
+ %o_t = arith.muli %i_t_3, %c16_i64 : i64 loc(#loc41)
23
+ %o_t_4 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42)
24
+ %o_t_5 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32, #blocked1> loc(#loc42)
25
+ %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)
26
+ %o_t_7 = arith.extsi %o_t_5 : tensor<16xi32, #blocked1> to tensor<16xi64, #blocked1> loc(#loc43)
27
+ %o_t_8 = tt.splat %o_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43)
28
+ %o_t_9 = tt.splat %o_t : i64 -> tensor<16xi64, #blocked1> loc(#loc43)
29
+ %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43)
30
+ %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<16xi64, #blocked1> loc(#loc43)
31
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc44)
32
+ %m_t_12 = tt.splat %m_t : i64 -> tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44)
33
+ %m_t_13 = tt.splat %m_t : i64 -> tensor<16xi64, #blocked1> loc(#loc44)
34
+ %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44)
35
+ %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<16xi64, #blocked1> loc(#loc44)
36
+ %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<16xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi1, #blocked> loc(#loc45)
37
+ %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46)
38
+ %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)
39
+ %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47)
40
+ %m_x_19 = tt.broadcast %m_x : tensor<16x1xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc48)
41
+ %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<16x128xi1, #blocked> loc(#loc48)
42
+ %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<16x128xi1, #blocked> loc(#loc48)
43
+ %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<16xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xi64, #blocked> loc(#loc49)
44
+ %p_x_22 = arith.muli %p_x, %cst_2 : tensor<16x1xi64, #blocked> loc(#loc50)
45
+ %p_x_23 = tt.splat %x : !tt.ptr<bf16> -> tensor<16x1x!tt.ptr<bf16>, #blocked> loc(#loc51)
46
+ %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<16x1x!tt.ptr<bf16>, #blocked>, tensor<16x1xi64, #blocked> loc(#loc51)
47
+ %p_x_25 = tt.broadcast %p_x_24 : tensor<16x1x!tt.ptr<bf16>, #blocked> -> tensor<16x128x!tt.ptr<bf16>, #blocked> loc(#loc52)
48
+ %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<16x128xi32, #blocked> loc(#loc52)
49
+ %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<16x128x!tt.ptr<bf16>, #blocked>, tensor<16x128xi32, #blocked> loc(#loc52)
50
+ %p_y = tt.splat %y : !tt.ptr<bf16> -> tensor<16x1x!tt.ptr<bf16>, #blocked> loc(#loc53)
51
+ %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<16x1x!tt.ptr<bf16>, #blocked>, tensor<16x1xi64, #blocked> loc(#loc53)
52
+ %p_y_29 = tt.broadcast %p_y_28 : tensor<16x1x!tt.ptr<bf16>, #blocked> -> tensor<16x128x!tt.ptr<bf16>, #blocked> loc(#loc54)
53
+ %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<16x128x!tt.ptr<bf16>, #blocked>, tensor<16x128xi32, #blocked> loc(#loc54)
54
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<16x!tt.ptr<f32>, #blocked1> loc(#loc55)
55
+ %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<16x!tt.ptr<f32>, #blocked1>, tensor<16xi64, #blocked1> loc(#loc55)
56
+ %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<16x128x!tt.ptr<bf16>, #blocked> loc(#loc56)
57
+ %b_x_32 = arith.extf %b_x : tensor<16x128xbf16, #blocked> to tensor<16x128xf32, #blocked> loc(#loc57)
58
+ %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<16x128xf32, #blocked> loc(#loc58)
59
+ %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({
60
+ ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))):
61
+ %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67)
62
+ tt.reduce.return %b_rstd_42 : f32 loc(#loc65)
63
+ }) : (tensor<16x128xf32, #blocked>) -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65)
64
+ %b_rstd_34 = tt.splat %eps : f32 -> tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60)
65
+ %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60)
66
+ %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61)
67
+ %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62)
68
+ %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16x1xf32, #blocked> loc(#loc63)
69
+ %b_y_38 = tt.broadcast %b_y : tensor<16x1xf32, #blocked> -> tensor<16x128xf32, #blocked> loc(#loc64)
70
+ %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<16x128xf32, #blocked> loc(#loc64)
71
+ %0 = arith.truncf %b_y_39 : tensor<16x128xf32, #blocked> to tensor<16x128xbf16, #blocked> loc(#loc30)
72
+ tt.store %p_y_30, %0, %m_x_21 : tensor<16x128x!tt.ptr<bf16>, #blocked> loc(#loc31)
73
+ %1 = ttg.convert_layout %b_rstd_37 : tensor<16xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<16xf32, #blocked1> loc(#loc32)
74
+ tt.store %p_rstd_31, %1, %m_t_15 : tensor<16x!tt.ptr<f32>, #blocked1> loc(#loc32)
75
+ tt.return loc(#loc33)
76
+ } loc(#loc)
77
+ } loc(#loc)
78
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
79
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
80
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
81
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
82
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
83
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
84
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
85
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
86
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
87
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
88
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
89
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
90
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
91
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
92
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
93
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
94
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
95
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
96
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
97
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
98
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
99
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
100
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
101
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
102
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
103
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
104
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
105
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
106
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
107
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
108
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
109
+ #loc39 = loc("i_t"(#loc2))
110
+ #loc40 = loc("i_t"(#loc3))
111
+ #loc41 = loc("o_t"(#loc4))
112
+ #loc42 = loc("o_t"(#loc5))
113
+ #loc43 = loc("o_t"(#loc6))
114
+ #loc44 = loc("m_t"(#loc7))
115
+ #loc45 = loc("m_x"(#loc8))
116
+ #loc46 = loc("m_x"(#loc9))
117
+ #loc47 = loc("m_x"(#loc10))
118
+ #loc48 = loc("m_x"(#loc11))
119
+ #loc49 = loc("p_x"(#loc12))
120
+ #loc50 = loc("p_x"(#loc13))
121
+ #loc51 = loc("p_x"(#loc14))
122
+ #loc52 = loc("p_x"(#loc15))
123
+ #loc53 = loc("p_y"(#loc16))
124
+ #loc54 = loc("p_y"(#loc17))
125
+ #loc55 = loc("p_rstd"(#loc18))
126
+ #loc56 = loc("b_x"(#loc19))
127
+ #loc57 = loc("b_x"(#loc20))
128
+ #loc58 = loc("b_rstd"(#loc21))
129
+ #loc60 = loc("b_rstd"(#loc25))
130
+ #loc61 = loc("b_rstd"(#loc26))
131
+ #loc62 = loc("b_rstd"(#loc27))
132
+ #loc63 = loc("b_y"(#loc28))
133
+ #loc64 = loc("b_y"(#loc29))
134
+ #loc65 = loc(callsite(#loc22 at #loc59))
135
+ #loc67 = loc(callsite(#loc24 at #loc65))
triton/gpu0/ARLKORWLIGGFADQNU2FOPQUU3BDQ5BG3L7NN5UWZVKSN2XYPXO4A/l2norm_fwd_kernel.ttir ADDED
@@ -0,0 +1,128 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
2
+ #loc3 = loc(unknown)
3
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
4
+ #loc35 = loc("x"(#loc))
5
+ #loc36 = loc("y"(#loc))
6
+ #loc37 = loc("rstd"(#loc))
7
+ #loc38 = loc("eps"(#loc))
8
+ #loc39 = loc("T"(#loc))
9
+ #loc62 = loc("b_rstd"(#loc25))
10
+ #loc68 = loc(callsite(#loc3 at #loc62))
11
+ module {
12
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
13
+ %b_x = arith.constant dense<0.000000e+00> : tensor<16x128xbf16> loc(#loc40)
14
+ %b_rstd = arith.constant dense<1.000000e+00> : tensor<16xf32> loc(#loc41)
15
+ %cst = arith.constant dense<128> : tensor<16x1xi64> loc(#loc3)
16
+ %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42)
17
+ %o_t = arith.constant 16 : i64 loc(#loc43)
18
+ %i_t = tt.get_program_id x : i32 loc(#loc44)
19
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45)
20
+ %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43)
21
+ %o_t_2 = tt.make_range {end = 16 : i32, start = 0 : i32} : tensor<16xi32> loc(#loc46)
22
+ %o_t_3 = arith.extsi %o_t_2 : tensor<16xi32> to tensor<16xi64> loc(#loc47)
23
+ %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<16xi64> loc(#loc47)
24
+ %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<16xi64> loc(#loc47)
25
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48)
26
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc49)
27
+ %m_t_6 = tt.splat %m_t : i64 -> tensor<16xi64> loc(#loc49)
28
+ %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<16xi64> loc(#loc49)
29
+ %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<16xi1> -> tensor<16x1xi1> loc(#loc50)
30
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51)
31
+ %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42)
32
+ %m_x_11 = tt.broadcast %m_x_8 : tensor<16x1xi1> -> tensor<16x128xi1> loc(#loc52)
33
+ %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<16x128xi1> loc(#loc52)
34
+ %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<16x128xi1> loc(#loc52)
35
+ %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<16xi64> -> tensor<16x1xi64> loc(#loc53)
36
+ %p_x_14 = arith.muli %p_x, %cst : tensor<16x1xi64> loc(#loc54)
37
+ %p_x_15 = tt.splat %x : !tt.ptr<bf16> -> tensor<16x1x!tt.ptr<bf16>> loc(#loc55)
38
+ %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<16x1x!tt.ptr<bf16>>, tensor<16x1xi64> loc(#loc55)
39
+ %p_x_17 = tt.broadcast %p_x_16 : tensor<16x1x!tt.ptr<bf16>> -> tensor<16x128x!tt.ptr<bf16>> loc(#loc56)
40
+ %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<16x128xi32> loc(#loc56)
41
+ %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<16x128x!tt.ptr<bf16>>, tensor<16x128xi32> loc(#loc56)
42
+ %p_y = tt.splat %y : !tt.ptr<bf16> -> tensor<16x1x!tt.ptr<bf16>> loc(#loc57)
43
+ %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<16x1x!tt.ptr<bf16>>, tensor<16x1xi64> loc(#loc57)
44
+ %p_y_21 = tt.broadcast %p_y_20 : tensor<16x1x!tt.ptr<bf16>> -> tensor<16x128x!tt.ptr<bf16>> loc(#loc58)
45
+ %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<16x128x!tt.ptr<bf16>>, tensor<16x128xi32> loc(#loc58)
46
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<16x!tt.ptr<f32>> loc(#loc59)
47
+ %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<16x!tt.ptr<f32>>, tensor<16xi64> loc(#loc59)
48
+ %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<16x128x!tt.ptr<bf16>> loc(#loc40)
49
+ %b_x_25 = arith.extf %b_x_24 : tensor<16x128xbf16> to tensor<16x128xf32> loc(#loc60)
50
+ %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<16x128xf32> loc(#loc61)
51
+ %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({
52
+ ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))):
53
+ %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69)
54
+ tt.reduce.return %b_rstd_36 : f32 loc(#loc67)
55
+ }) : (tensor<16x128xf32>) -> tensor<16xf32> loc(#loc67)
56
+ %b_rstd_28 = tt.splat %eps : f32 -> tensor<16xf32> loc(#loc63)
57
+ %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<16xf32> loc(#loc63)
58
+ %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<16xf32> loc(#loc64)
59
+ %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<16xf32> loc(#loc41)
60
+ %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<16xf32> -> tensor<16x1xf32> loc(#loc65)
61
+ %b_y_32 = tt.broadcast %b_y : tensor<16x1xf32> -> tensor<16x128xf32> loc(#loc66)
62
+ %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<16x128xf32> loc(#loc66)
63
+ %0 = arith.truncf %b_y_33 : tensor<16x128xf32> to tensor<16x128xbf16> loc(#loc31)
64
+ tt.store %p_y_22, %0, %m_x_13 : tensor<16x128x!tt.ptr<bf16>> loc(#loc32)
65
+ tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<16x!tt.ptr<f32>> loc(#loc33)
66
+ tt.return loc(#loc34)
67
+ } loc(#loc)
68
+ } loc(#loc)
69
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
70
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
71
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
72
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
73
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
74
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
75
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
76
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
77
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23)
78
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
79
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
80
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
81
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
82
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
83
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
84
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
85
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
86
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
87
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
88
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
89
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
90
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
91
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
92
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
93
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
94
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
95
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
96
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
97
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
98
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
99
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
100
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
101
+ #loc40 = loc("b_x"(#loc1))
102
+ #loc41 = loc("b_rstd"(#loc2))
103
+ #loc42 = loc("m_x"(#loc4))
104
+ #loc43 = loc("o_t"(#loc5))
105
+ #loc44 = loc("i_t"(#loc6))
106
+ #loc45 = loc("i_t"(#loc7))
107
+ #loc46 = loc("o_t"(#loc8))
108
+ #loc47 = loc("o_t"(#loc9))
109
+ #loc48 = loc("o_d"(#loc10))
110
+ #loc49 = loc("m_t"(#loc11))
111
+ #loc50 = loc("m_x"(#loc12))
112
+ #loc51 = loc("m_x"(#loc13))
113
+ #loc52 = loc("m_x"(#loc14))
114
+ #loc53 = loc("p_x"(#loc15))
115
+ #loc54 = loc("p_x"(#loc16))
116
+ #loc55 = loc("p_x"(#loc17))
117
+ #loc56 = loc("p_x"(#loc18))
118
+ #loc57 = loc("p_y"(#loc19))
119
+ #loc58 = loc("p_y"(#loc20))
120
+ #loc59 = loc("p_rstd"(#loc21))
121
+ #loc60 = loc("b_x"(#loc22))
122
+ #loc61 = loc("b_rstd"(#loc23))
123
+ #loc63 = loc("b_rstd"(#loc27))
124
+ #loc64 = loc("b_rstd"(#loc28))
125
+ #loc65 = loc("b_y"(#loc29))
126
+ #loc66 = loc("b_y"(#loc30))
127
+ #loc67 = loc(callsite(#loc24 at #loc62))
128
+ #loc69 = loc(callsite(#loc26 at #loc67))
triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/__grp__l2norm_fwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}}
triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.cubin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:01f06397f0b30f222dafe7d0ff24bfb233bf0b57b9c1dc7883c733a1b6747267
3
+ size 160768
triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}
triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.llir ADDED
The diff for this file is too large to render. See raw diff
 
triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ptx ADDED
The diff for this file is too large to render. See raw diff
 
triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.source ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
2
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0)
3
+ #loc38 = loc(unknown)
4
+ #loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0)
5
+ #loc45 = loc("x"(#loc))
6
+ #loc46 = loc("y"(#loc))
7
+ #loc47 = loc("rstd"(#loc))
8
+ #loc48 = loc("eps"(#loc))
9
+ #loc49 = loc("T"(#loc))
10
+ #loc81 = loc("input"(#loc36))
11
+ #loc82 = loc("a"(#loc41))
12
+ #loc83 = loc("b"(#loc41))
13
+ module {
14
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %i_t = tt.get_program_id x : i32 loc(#loc50)
16
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51)
17
+ %o_t = arith.constant 128 : i32 loc(#loc52)
18
+ %o_t_1 = arith.constant 128 : i64 loc(#loc52)
19
+ %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52)
20
+ %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc53)
21
+ %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc54)
22
+ %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc54)
23
+ %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc54)
24
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55)
25
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc56)
26
+ %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc56)
27
+ %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc56)
28
+ %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc57)
29
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58)
30
+ %m_x_10 = arith.constant 128 : i32 loc(#loc59)
31
+ %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59)
32
+ %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59)
33
+ %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc60)
34
+ %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc60)
35
+ %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc60)
36
+ %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc61)
37
+ %p_x_16 = arith.constant 128 : i32 loc(#loc62)
38
+ %p_x_17 = arith.constant 128 : i64 loc(#loc62)
39
+ %p_x_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc62)
40
+ %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<128x1xi64> loc(#loc62)
41
+ %p_x_20 = tt.splat %x : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc63)
42
+ %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc63)
43
+ %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64)
44
+ %p_x_23 = tt.broadcast %p_x_21 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc65)
45
+ %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc65)
46
+ %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc65)
47
+ %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc66)
48
+ %p_y_26 = arith.constant 128 : i32 loc(#loc67)
49
+ %p_y_27 = arith.constant 128 : i64 loc(#loc67)
50
+ %p_y_28 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc67)
51
+ %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<128x1xi64> loc(#loc67)
52
+ %p_y_30 = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc68)
53
+ %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc68)
54
+ %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69)
55
+ %p_y_33 = tt.broadcast %p_y_31 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc70)
56
+ %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc70)
57
+ %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc70)
58
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>> loc(#loc71)
59
+ %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr<f32>>, tensor<128xi64> loc(#loc71)
60
+ %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72)
61
+ %b_x_37 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc72)
62
+ %b_x_38 = arith.truncf %b_x_37 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc72)
63
+ %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<128x128x!tt.ptr<bf16>> loc(#loc72)
64
+ %b_x_40 = arith.extf %b_x_39 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc73)
65
+ %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<128x128xf32> loc(#loc74)
66
+ %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_rstd) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc75)
67
+ %b_rstd_42 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc76)
68
+ %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<128xf32> loc(#loc76)
69
+ %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<128xf32> loc(#loc77)
70
+ %b_rstd_45 = arith.constant 1 : i32 loc(#loc78)
71
+ %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78)
72
+ %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc78)
73
+ %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<128xf32> loc(#loc78)
74
+ %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc79)
75
+ %b_y_49 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc80)
76
+ %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<128x128xf32> loc(#loc80)
77
+ %0 = arith.truncf %b_y_50 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc32)
78
+ tt.store %p_y_35, %0, %m_x_15 : tensor<128x128x!tt.ptr<bf16>> loc(#loc33)
79
+ tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<128x!tt.ptr<f32>> loc(#loc34)
80
+ tt.return loc(#loc35)
81
+ } loc(#loc)
82
+ tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc36))) -> tensor<128xf32> attributes {noinline = false} {
83
+ %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({
84
+ ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)):
85
+ %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37)
86
+ tt.reduce.return %2 : f32 loc(#loc37)
87
+ }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc37)
88
+ tt.return %0 : tensor<128xf32> loc(#loc39)
89
+ ^bb1: // no predecessors
90
+ %1 = ub.poison : tensor<128xf32> loc(#loc40)
91
+ tt.return %1 : tensor<128xf32> loc(#loc40)
92
+ } loc(#loc36)
93
+ tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} {
94
+ %0 = arith.addf %a, %b : f32 loc(#loc42)
95
+ tt.return %0 : f32 loc(#loc43)
96
+ ^bb1: // no predecessors
97
+ %1 = ub.poison : f32 loc(#loc44)
98
+ tt.return %1 : f32 loc(#loc44)
99
+ } loc(#loc41)
100
+ } loc(#loc)
101
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
102
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
103
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
104
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
105
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
106
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23)
107
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
108
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
109
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
110
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
111
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
112
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
113
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
114
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
115
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37)
116
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
117
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18)
118
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29)
119
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
120
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37)
121
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
122
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
123
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
124
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
125
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
126
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
127
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
128
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
129
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
130
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
131
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
132
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
133
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
134
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
135
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
136
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
137
+ #loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11)
138
+ #loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4)
139
+ #loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
140
+ #loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11)
141
+ #loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4)
142
+ #loc50 = loc("i_t"(#loc1))
143
+ #loc51 = loc("i_t"(#loc2))
144
+ #loc52 = loc("o_t"(#loc3))
145
+ #loc53 = loc("o_t"(#loc4))
146
+ #loc54 = loc("o_t"(#loc5))
147
+ #loc55 = loc("o_d"(#loc6))
148
+ #loc56 = loc("m_t"(#loc7))
149
+ #loc57 = loc("m_x"(#loc8))
150
+ #loc58 = loc("m_x"(#loc9))
151
+ #loc59 = loc("m_x"(#loc10))
152
+ #loc60 = loc("m_x"(#loc11))
153
+ #loc61 = loc("p_x"(#loc12))
154
+ #loc62 = loc("p_x"(#loc13))
155
+ #loc63 = loc("p_x"(#loc14))
156
+ #loc64 = loc("p_x"(#loc15))
157
+ #loc65 = loc("p_x"(#loc16))
158
+ #loc66 = loc("p_y"(#loc17))
159
+ #loc67 = loc("p_y"(#loc18))
160
+ #loc68 = loc("p_y"(#loc19))
161
+ #loc69 = loc("p_y"(#loc20))
162
+ #loc70 = loc("p_y"(#loc21))
163
+ #loc71 = loc("p_rstd"(#loc22))
164
+ #loc72 = loc("b_x"(#loc23))
165
+ #loc73 = loc("b_x"(#loc24))
166
+ #loc74 = loc("b_rstd"(#loc25))
167
+ #loc75 = loc("b_rstd"(#loc26))
168
+ #loc76 = loc("b_rstd"(#loc27))
169
+ #loc77 = loc("b_rstd"(#loc28))
170
+ #loc78 = loc("b_rstd"(#loc29))
171
+ #loc79 = loc("b_y"(#loc30))
172
+ #loc80 = loc("b_y"(#loc31))
triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttgir ADDED
@@ -0,0 +1,135 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}>
2
+ #blocked1 = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}>
3
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
4
+ #loc1 = loc(unknown)
5
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
6
+ #loc34 = loc("x"(#loc))
7
+ #loc35 = loc("y"(#loc))
8
+ #loc36 = loc("rstd"(#loc))
9
+ #loc37 = loc("eps"(#loc))
10
+ #loc38 = loc("T"(#loc))
11
+ #loc59 = loc("b_rstd"(#loc23))
12
+ #loc66 = loc(callsite(#loc1 at #loc59))
13
+ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} {
14
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %cst = arith.constant dense<1.000000e+00> : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1)
16
+ %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked> loc(#loc1)
17
+ %c128_i64 = arith.constant 128 : i64 loc(#loc1)
18
+ %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1)
19
+ %cst_2 = arith.constant dense<128> : tensor<128x1xi64, #blocked> loc(#loc1)
20
+ %i_t = tt.get_program_id x : i32 loc(#loc39)
21
+ %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40)
22
+ %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc41)
23
+ %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42)
24
+ %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked1> loc(#loc42)
25
+ %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)
26
+ %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked1> to tensor<128xi64, #blocked1> loc(#loc43)
27
+ %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43)
28
+ %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked1> loc(#loc43)
29
+ %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43)
30
+ %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked1> loc(#loc43)
31
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc44)
32
+ %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44)
33
+ %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked1> loc(#loc44)
34
+ %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44)
35
+ %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked1> loc(#loc44)
36
+ %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc45)
37
+ %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46)
38
+ %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)
39
+ %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47)
40
+ %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48)
41
+ %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc48)
42
+ %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked> loc(#loc48)
43
+ %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi64, #blocked> loc(#loc49)
44
+ %p_x_22 = arith.muli %p_x, %cst_2 : tensor<128x1xi64, #blocked> loc(#loc50)
45
+ %p_x_23 = tt.splat %x : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>, #blocked> loc(#loc51)
46
+ %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<128x1x!tt.ptr<bf16>, #blocked>, tensor<128x1xi64, #blocked> loc(#loc51)
47
+ %p_x_25 = tt.broadcast %p_x_24 : tensor<128x1x!tt.ptr<bf16>, #blocked> -> tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc52)
48
+ %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<128x128xi32, #blocked> loc(#loc52)
49
+ %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<128x128x!tt.ptr<bf16>, #blocked>, tensor<128x128xi32, #blocked> loc(#loc52)
50
+ %p_y = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>, #blocked> loc(#loc53)
51
+ %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<128x1x!tt.ptr<bf16>, #blocked>, tensor<128x1xi64, #blocked> loc(#loc53)
52
+ %p_y_29 = tt.broadcast %p_y_28 : tensor<128x1x!tt.ptr<bf16>, #blocked> -> tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc54)
53
+ %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<128x128x!tt.ptr<bf16>, #blocked>, tensor<128x128xi32, #blocked> loc(#loc54)
54
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>, #blocked1> loc(#loc55)
55
+ %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr<f32>, #blocked1>, tensor<128xi64, #blocked1> loc(#loc55)
56
+ %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc56)
57
+ %b_x_32 = arith.extf %b_x : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc57)
58
+ %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<128x128xf32, #blocked> loc(#loc58)
59
+ %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({
60
+ ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))):
61
+ %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67)
62
+ tt.reduce.return %b_rstd_42 : f32 loc(#loc65)
63
+ }) : (tensor<128x128xf32, #blocked>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65)
64
+ %b_rstd_34 = tt.splat %eps : f32 -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60)
65
+ %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60)
66
+ %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61)
67
+ %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62)
68
+ %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc63)
69
+ %b_y_38 = tt.broadcast %b_y : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc64)
70
+ %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<128x128xf32, #blocked> loc(#loc64)
71
+ %0 = arith.truncf %b_y_39 : tensor<128x128xf32, #blocked> to tensor<128x128xbf16, #blocked> loc(#loc30)
72
+ tt.store %p_y_30, %0, %m_x_21 : tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc31)
73
+ %1 = ttg.convert_layout %b_rstd_37 : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128xf32, #blocked1> loc(#loc32)
74
+ tt.store %p_rstd_31, %1, %m_t_15 : tensor<128x!tt.ptr<f32>, #blocked1> loc(#loc32)
75
+ tt.return loc(#loc33)
76
+ } loc(#loc)
77
+ } loc(#loc)
78
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
79
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
80
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
81
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
82
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
83
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
84
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
85
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
86
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
87
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
88
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
89
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
90
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
91
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
92
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
93
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
94
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
95
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
96
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
97
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
98
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
99
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
100
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
101
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
102
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
103
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
104
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
105
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
106
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
107
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
108
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
109
+ #loc39 = loc("i_t"(#loc2))
110
+ #loc40 = loc("i_t"(#loc3))
111
+ #loc41 = loc("o_t"(#loc4))
112
+ #loc42 = loc("o_t"(#loc5))
113
+ #loc43 = loc("o_t"(#loc6))
114
+ #loc44 = loc("m_t"(#loc7))
115
+ #loc45 = loc("m_x"(#loc8))
116
+ #loc46 = loc("m_x"(#loc9))
117
+ #loc47 = loc("m_x"(#loc10))
118
+ #loc48 = loc("m_x"(#loc11))
119
+ #loc49 = loc("p_x"(#loc12))
120
+ #loc50 = loc("p_x"(#loc13))
121
+ #loc51 = loc("p_x"(#loc14))
122
+ #loc52 = loc("p_x"(#loc15))
123
+ #loc53 = loc("p_y"(#loc16))
124
+ #loc54 = loc("p_y"(#loc17))
125
+ #loc55 = loc("p_rstd"(#loc18))
126
+ #loc56 = loc("b_x"(#loc19))
127
+ #loc57 = loc("b_x"(#loc20))
128
+ #loc58 = loc("b_rstd"(#loc21))
129
+ #loc60 = loc("b_rstd"(#loc25))
130
+ #loc61 = loc("b_rstd"(#loc26))
131
+ #loc62 = loc("b_rstd"(#loc27))
132
+ #loc63 = loc("b_y"(#loc28))
133
+ #loc64 = loc("b_y"(#loc29))
134
+ #loc65 = loc(callsite(#loc22 at #loc59))
135
+ #loc67 = loc(callsite(#loc24 at #loc65))
triton/gpu0/AUMQSD5G6VQCBUJBTICJWDFBC2JNON7RBB2M4ZACHZDVVKWFQNLQ/l2norm_fwd_kernel.ttir ADDED
@@ -0,0 +1,125 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
2
+ #loc3 = loc(unknown)
3
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
4
+ #loc34 = loc("x"(#loc))
5
+ #loc35 = loc("y"(#loc))
6
+ #loc36 = loc("rstd"(#loc))
7
+ #loc37 = loc("eps"(#loc))
8
+ #loc38 = loc("T"(#loc))
9
+ #loc60 = loc("b_rstd"(#loc24))
10
+ #loc66 = loc(callsite(#loc3 at #loc60))
11
+ module {
12
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
13
+ %b_x = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc39)
14
+ %b_rstd = arith.constant dense<1.000000e+00> : tensor<128xf32> loc(#loc40)
15
+ %cst = arith.constant dense<128> : tensor<128x1xi64> loc(#loc3)
16
+ %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc41)
17
+ %c128_i64 = arith.constant 128 : i64 loc(#loc3)
18
+ %i_t = tt.get_program_id x : i32 loc(#loc42)
19
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc43)
20
+ %o_t = arith.muli %i_t_0, %c128_i64 : i64 loc(#loc44)
21
+ %o_t_1 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc45)
22
+ %o_t_2 = arith.extsi %o_t_1 : tensor<128xi32> to tensor<128xi64> loc(#loc46)
23
+ %o_t_3 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc46)
24
+ %o_t_4 = arith.addi %o_t_3, %o_t_2 : tensor<128xi64> loc(#loc46)
25
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc47)
26
+ %m_t_5 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc47)
27
+ %m_t_6 = arith.cmpi slt, %o_t_4, %m_t_5 : tensor<128xi64> loc(#loc47)
28
+ %m_x_7 = tt.expand_dims %m_t_6 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc48)
29
+ %m_x_8 = tt.expand_dims %o_t_1 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc49)
30
+ %m_x_9 = arith.cmpi slt, %m_x_8, %m_x : tensor<1x128xi32> loc(#loc41)
31
+ %m_x_10 = tt.broadcast %m_x_7 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc50)
32
+ %m_x_11 = tt.broadcast %m_x_9 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc50)
33
+ %m_x_12 = arith.andi %m_x_10, %m_x_11 : tensor<128x128xi1> loc(#loc50)
34
+ %p_x = tt.expand_dims %o_t_4 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc51)
35
+ %p_x_13 = arith.muli %p_x, %cst : tensor<128x1xi64> loc(#loc52)
36
+ %p_x_14 = tt.splat %x : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc53)
37
+ %p_x_15 = tt.addptr %p_x_14, %p_x_13 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc53)
38
+ %p_x_16 = tt.broadcast %p_x_15 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc54)
39
+ %p_x_17 = tt.broadcast %m_x_8 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc54)
40
+ %p_x_18 = tt.addptr %p_x_16, %p_x_17 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc54)
41
+ %p_y = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc55)
42
+ %p_y_19 = tt.addptr %p_y, %p_x_13 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc55)
43
+ %p_y_20 = tt.broadcast %p_y_19 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc56)
44
+ %p_y_21 = tt.addptr %p_y_20, %p_x_17 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc56)
45
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>> loc(#loc57)
46
+ %p_rstd_22 = tt.addptr %p_rstd, %o_t_4 : tensor<128x!tt.ptr<f32>>, tensor<128xi64> loc(#loc57)
47
+ %b_x_23 = tt.load %p_x_18, %m_x_12, %b_x : tensor<128x128x!tt.ptr<bf16>> loc(#loc39)
48
+ %b_x_24 = arith.extf %b_x_23 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc58)
49
+ %b_rstd_25 = arith.mulf %b_x_24, %b_x_24 : tensor<128x128xf32> loc(#loc59)
50
+ %b_rstd_26 = "tt.reduce"(%b_rstd_25) <{axis = 1 : i32}> ({
51
+ ^bb0(%b_rstd_33: f32 loc(callsite(#loc3 at #loc60)), %b_rstd_34: f32 loc(callsite(#loc3 at #loc60))):
52
+ %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : f32 loc(#loc67)
53
+ tt.reduce.return %b_rstd_35 : f32 loc(#loc65)
54
+ }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc65)
55
+ %b_rstd_27 = tt.splat %eps : f32 -> tensor<128xf32> loc(#loc61)
56
+ %b_rstd_28 = arith.addf %b_rstd_26, %b_rstd_27 : tensor<128xf32> loc(#loc61)
57
+ %b_rstd_29 = math.sqrt %b_rstd_28 : tensor<128xf32> loc(#loc62)
58
+ %b_rstd_30 = arith.divf %b_rstd, %b_rstd_29 : tensor<128xf32> loc(#loc40)
59
+ %b_y = tt.expand_dims %b_rstd_30 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc63)
60
+ %b_y_31 = tt.broadcast %b_y : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc64)
61
+ %b_y_32 = arith.mulf %b_x_24, %b_y_31 : tensor<128x128xf32> loc(#loc64)
62
+ %0 = arith.truncf %b_y_32 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc30)
63
+ tt.store %p_y_21, %0, %m_x_12 : tensor<128x128x!tt.ptr<bf16>> loc(#loc31)
64
+ tt.store %p_rstd_22, %b_rstd_30, %m_t_6 : tensor<128x!tt.ptr<f32>> loc(#loc32)
65
+ tt.return loc(#loc33)
66
+ } loc(#loc)
67
+ } loc(#loc)
68
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
69
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
70
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
71
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
72
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
73
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
74
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
75
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
76
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
77
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
78
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
79
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
80
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
81
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
82
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
83
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
84
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
85
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
86
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
87
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
88
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
89
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
90
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
91
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
92
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
93
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
94
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
95
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
96
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
97
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
98
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
99
+ #loc39 = loc("b_x"(#loc1))
100
+ #loc40 = loc("b_rstd"(#loc2))
101
+ #loc41 = loc("m_x"(#loc4))
102
+ #loc42 = loc("i_t"(#loc5))
103
+ #loc43 = loc("i_t"(#loc6))
104
+ #loc44 = loc("o_t"(#loc7))
105
+ #loc45 = loc("o_t"(#loc8))
106
+ #loc46 = loc("o_t"(#loc9))
107
+ #loc47 = loc("m_t"(#loc10))
108
+ #loc48 = loc("m_x"(#loc11))
109
+ #loc49 = loc("m_x"(#loc12))
110
+ #loc50 = loc("m_x"(#loc13))
111
+ #loc51 = loc("p_x"(#loc14))
112
+ #loc52 = loc("p_x"(#loc15))
113
+ #loc53 = loc("p_x"(#loc16))
114
+ #loc54 = loc("p_x"(#loc17))
115
+ #loc55 = loc("p_y"(#loc18))
116
+ #loc56 = loc("p_y"(#loc19))
117
+ #loc57 = loc("p_rstd"(#loc20))
118
+ #loc58 = loc("b_x"(#loc21))
119
+ #loc59 = loc("b_rstd"(#loc22))
120
+ #loc61 = loc("b_rstd"(#loc26))
121
+ #loc62 = loc("b_rstd"(#loc27))
122
+ #loc63 = loc("b_y"(#loc28))
123
+ #loc64 = loc("b_y"(#loc29))
124
+ #loc65 = loc(callsite(#loc23 at #loc60))
125
+ #loc67 = loc(callsite(#loc25 at #loc65))
triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/__grp__l2norm_fwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}}
triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.cubin ADDED
Binary file (46.2 kB). View file
 
triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}
triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.llir ADDED
@@ -0,0 +1,1143 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ; ModuleID = 'LLVMDialectModule'
2
+ source_filename = "LLVMDialectModule"
3
+ 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"
4
+
5
+ @global_smem = external addrspace(3) global [0 x i8], align 16
6
+ @.str = private unnamed_addr constant [11 x i8] c"__CUDA_FTZ\00", align 1
7
+ @.str.2 = private unnamed_addr constant [17 x i8] c"__CUDA_PREC_SQRT\00", align 1
8
+
9
+ ; Function Attrs: nounwind
10
+ 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 {
11
+ %8 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !13
12
+ %9 = zext nneg i32 %8 to i64, !dbg !14
13
+ %10 = shl nuw nsw i64 %9, 5, !dbg !15
14
+ %11 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !16
15
+ %12 = and i32 %11, 48, !dbg !16
16
+ %13 = lshr exact i32 %12, 4, !dbg !16
17
+ %14 = or disjoint i32 %13, 4, !dbg !16
18
+ %15 = or disjoint i32 %13, 8, !dbg !16
19
+ %16 = or disjoint i32 %13, 12, !dbg !16
20
+ %17 = or disjoint i32 %13, 16, !dbg !16
21
+ %18 = or disjoint i32 %13, 20, !dbg !16
22
+ %19 = or disjoint i32 %13, 24, !dbg !16
23
+ %20 = or disjoint i32 %13, 28, !dbg !16
24
+ %21 = zext nneg i32 %13 to i64, !dbg !17
25
+ %22 = zext nneg i32 %14 to i64, !dbg !17
26
+ %23 = zext nneg i32 %15 to i64, !dbg !17
27
+ %24 = zext nneg i32 %16 to i64, !dbg !17
28
+ %25 = zext nneg i32 %17 to i64, !dbg !17
29
+ %26 = zext nneg i32 %18 to i64, !dbg !17
30
+ %27 = zext nneg i32 %19 to i64, !dbg !17
31
+ %28 = zext nneg i32 %20 to i64, !dbg !17
32
+ %29 = or disjoint i64 %10, %21, !dbg !17
33
+ %30 = or disjoint i64 %10, %22, !dbg !17
34
+ %31 = or disjoint i64 %10, %23, !dbg !17
35
+ %32 = or disjoint i64 %10, %24, !dbg !17
36
+ %33 = or disjoint i64 %10, %25, !dbg !17
37
+ %34 = or disjoint i64 %10, %26, !dbg !17
38
+ %35 = or disjoint i64 %10, %27, !dbg !17
39
+ %36 = or disjoint i64 %10, %28, !dbg !17
40
+ %37 = sext i32 %4 to i64, !dbg !18
41
+ %38 = icmp slt i64 %29, %37, !dbg !18
42
+ %39 = icmp slt i64 %30, %37, !dbg !18
43
+ %40 = icmp slt i64 %31, %37, !dbg !18
44
+ %41 = icmp slt i64 %32, %37, !dbg !18
45
+ %42 = icmp slt i64 %33, %37, !dbg !18
46
+ %43 = icmp slt i64 %34, %37, !dbg !18
47
+ %44 = icmp slt i64 %35, %37, !dbg !18
48
+ %45 = icmp slt i64 %36, %37, !dbg !18
49
+ %46 = shl nuw nsw i32 %11, 3, !dbg !19
50
+ %47 = and i32 %46, 120, !dbg !19
51
+ %48 = shl nuw nsw i64 %29, 7, !dbg !20
52
+ %49 = shl nuw nsw i64 %30, 7, !dbg !20
53
+ %50 = shl nuw nsw i64 %31, 7, !dbg !20
54
+ %51 = shl nuw nsw i64 %32, 7, !dbg !20
55
+ %52 = shl nuw nsw i64 %33, 7, !dbg !20
56
+ %53 = shl nuw nsw i64 %34, 7, !dbg !20
57
+ %54 = shl nuw nsw i64 %35, 7, !dbg !20
58
+ %55 = shl nuw nsw i64 %36, 7, !dbg !20
59
+ %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !21
60
+ %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %49, !dbg !21
61
+ %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !21
62
+ %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %51, !dbg !21
63
+ %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !21
64
+ %61 = getelementptr bfloat, ptr addrspace(1) %0, i64 %53, !dbg !21
65
+ %62 = getelementptr bfloat, ptr addrspace(1) %0, i64 %54, !dbg !21
66
+ %63 = getelementptr bfloat, ptr addrspace(1) %0, i64 %55, !dbg !21
67
+ %64 = zext nneg i32 %47 to i64, !dbg !22
68
+ %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %64, !dbg !22
69
+ %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %64, !dbg !22
70
+ %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %64, !dbg !22
71
+ %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %64, !dbg !22
72
+ %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %64, !dbg !22
73
+ %70 = getelementptr bfloat, ptr addrspace(1) %61, i64 %64, !dbg !22
74
+ %71 = getelementptr bfloat, ptr addrspace(1) %62, i64 %64, !dbg !22
75
+ %72 = getelementptr bfloat, ptr addrspace(1) %63, i64 %64, !dbg !22
76
+ %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
77
+ %74 = extractvalue { i32, i32, i32, i32 } %73, 0, !dbg !23
78
+ %75 = bitcast i32 %74 to <2 x bfloat>, !dbg !23
79
+ %76 = extractvalue { i32, i32, i32, i32 } %73, 1, !dbg !23
80
+ %77 = bitcast i32 %76 to <2 x bfloat>, !dbg !23
81
+ %78 = extractvalue { i32, i32, i32, i32 } %73, 2, !dbg !23
82
+ %79 = bitcast i32 %78 to <2 x bfloat>, !dbg !23
83
+ %80 = extractvalue { i32, i32, i32, i32 } %73, 3, !dbg !23
84
+ %81 = bitcast i32 %80 to <2 x bfloat>, !dbg !23
85
+ %82 = extractelement <2 x bfloat> %75, i64 0, !dbg !23
86
+ %83 = extractelement <2 x bfloat> %75, i64 1, !dbg !23
87
+ %84 = extractelement <2 x bfloat> %77, i64 0, !dbg !23
88
+ %85 = extractelement <2 x bfloat> %77, i64 1, !dbg !23
89
+ %86 = extractelement <2 x bfloat> %79, i64 0, !dbg !23
90
+ %87 = extractelement <2 x bfloat> %79, i64 1, !dbg !23
91
+ %88 = extractelement <2 x bfloat> %81, i64 0, !dbg !23
92
+ %89 = extractelement <2 x bfloat> %81, i64 1, !dbg !23
93
+ %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
94
+ %91 = extractvalue { i32, i32, i32, i32 } %90, 0, !dbg !23
95
+ %92 = bitcast i32 %91 to <2 x bfloat>, !dbg !23
96
+ %93 = extractvalue { i32, i32, i32, i32 } %90, 1, !dbg !23
97
+ %94 = bitcast i32 %93 to <2 x bfloat>, !dbg !23
98
+ %95 = extractvalue { i32, i32, i32, i32 } %90, 2, !dbg !23
99
+ %96 = bitcast i32 %95 to <2 x bfloat>, !dbg !23
100
+ %97 = extractvalue { i32, i32, i32, i32 } %90, 3, !dbg !23
101
+ %98 = bitcast i32 %97 to <2 x bfloat>, !dbg !23
102
+ %99 = extractelement <2 x bfloat> %92, i64 0, !dbg !23
103
+ %100 = extractelement <2 x bfloat> %92, i64 1, !dbg !23
104
+ %101 = extractelement <2 x bfloat> %94, i64 0, !dbg !23
105
+ %102 = extractelement <2 x bfloat> %94, i64 1, !dbg !23
106
+ %103 = extractelement <2 x bfloat> %96, i64 0, !dbg !23
107
+ %104 = extractelement <2 x bfloat> %96, i64 1, !dbg !23
108
+ %105 = extractelement <2 x bfloat> %98, i64 0, !dbg !23
109
+ %106 = extractelement <2 x bfloat> %98, i64 1, !dbg !23
110
+ %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
111
+ %108 = extractvalue { i32, i32, i32, i32 } %107, 0, !dbg !23
112
+ %109 = bitcast i32 %108 to <2 x bfloat>, !dbg !23
113
+ %110 = extractvalue { i32, i32, i32, i32 } %107, 1, !dbg !23
114
+ %111 = bitcast i32 %110 to <2 x bfloat>, !dbg !23
115
+ %112 = extractvalue { i32, i32, i32, i32 } %107, 2, !dbg !23
116
+ %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !23
117
+ %114 = extractvalue { i32, i32, i32, i32 } %107, 3, !dbg !23
118
+ %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !23
119
+ %116 = extractelement <2 x bfloat> %109, i64 0, !dbg !23
120
+ %117 = extractelement <2 x bfloat> %109, i64 1, !dbg !23
121
+ %118 = extractelement <2 x bfloat> %111, i64 0, !dbg !23
122
+ %119 = extractelement <2 x bfloat> %111, i64 1, !dbg !23
123
+ %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !23
124
+ %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !23
125
+ %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !23
126
+ %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !23
127
+ %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
128
+ %125 = extractvalue { i32, i32, i32, i32 } %124, 0, !dbg !23
129
+ %126 = bitcast i32 %125 to <2 x bfloat>, !dbg !23
130
+ %127 = extractvalue { i32, i32, i32, i32 } %124, 1, !dbg !23
131
+ %128 = bitcast i32 %127 to <2 x bfloat>, !dbg !23
132
+ %129 = extractvalue { i32, i32, i32, i32 } %124, 2, !dbg !23
133
+ %130 = bitcast i32 %129 to <2 x bfloat>, !dbg !23
134
+ %131 = extractvalue { i32, i32, i32, i32 } %124, 3, !dbg !23
135
+ %132 = bitcast i32 %131 to <2 x bfloat>, !dbg !23
136
+ %133 = extractelement <2 x bfloat> %126, i64 0, !dbg !23
137
+ %134 = extractelement <2 x bfloat> %126, i64 1, !dbg !23
138
+ %135 = extractelement <2 x bfloat> %128, i64 0, !dbg !23
139
+ %136 = extractelement <2 x bfloat> %128, i64 1, !dbg !23
140
+ %137 = extractelement <2 x bfloat> %130, i64 0, !dbg !23
141
+ %138 = extractelement <2 x bfloat> %130, i64 1, !dbg !23
142
+ %139 = extractelement <2 x bfloat> %132, i64 0, !dbg !23
143
+ %140 = extractelement <2 x bfloat> %132, i64 1, !dbg !23
144
+ %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
145
+ %142 = extractvalue { i32, i32, i32, i32 } %141, 0, !dbg !23
146
+ %143 = bitcast i32 %142 to <2 x bfloat>, !dbg !23
147
+ %144 = extractvalue { i32, i32, i32, i32 } %141, 1, !dbg !23
148
+ %145 = bitcast i32 %144 to <2 x bfloat>, !dbg !23
149
+ %146 = extractvalue { i32, i32, i32, i32 } %141, 2, !dbg !23
150
+ %147 = bitcast i32 %146 to <2 x bfloat>, !dbg !23
151
+ %148 = extractvalue { i32, i32, i32, i32 } %141, 3, !dbg !23
152
+ %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !23
153
+ %150 = extractelement <2 x bfloat> %143, i64 0, !dbg !23
154
+ %151 = extractelement <2 x bfloat> %143, i64 1, !dbg !23
155
+ %152 = extractelement <2 x bfloat> %145, i64 0, !dbg !23
156
+ %153 = extractelement <2 x bfloat> %145, i64 1, !dbg !23
157
+ %154 = extractelement <2 x bfloat> %147, i64 0, !dbg !23
158
+ %155 = extractelement <2 x bfloat> %147, i64 1, !dbg !23
159
+ %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !23
160
+ %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !23
161
+ %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
162
+ %159 = extractvalue { i32, i32, i32, i32 } %158, 0, !dbg !23
163
+ %160 = bitcast i32 %159 to <2 x bfloat>, !dbg !23
164
+ %161 = extractvalue { i32, i32, i32, i32 } %158, 1, !dbg !23
165
+ %162 = bitcast i32 %161 to <2 x bfloat>, !dbg !23
166
+ %163 = extractvalue { i32, i32, i32, i32 } %158, 2, !dbg !23
167
+ %164 = bitcast i32 %163 to <2 x bfloat>, !dbg !23
168
+ %165 = extractvalue { i32, i32, i32, i32 } %158, 3, !dbg !23
169
+ %166 = bitcast i32 %165 to <2 x bfloat>, !dbg !23
170
+ %167 = extractelement <2 x bfloat> %160, i64 0, !dbg !23
171
+ %168 = extractelement <2 x bfloat> %160, i64 1, !dbg !23
172
+ %169 = extractelement <2 x bfloat> %162, i64 0, !dbg !23
173
+ %170 = extractelement <2 x bfloat> %162, i64 1, !dbg !23
174
+ %171 = extractelement <2 x bfloat> %164, i64 0, !dbg !23
175
+ %172 = extractelement <2 x bfloat> %164, i64 1, !dbg !23
176
+ %173 = extractelement <2 x bfloat> %166, i64 0, !dbg !23
177
+ %174 = extractelement <2 x bfloat> %166, i64 1, !dbg !23
178
+ %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
179
+ %176 = extractvalue { i32, i32, i32, i32 } %175, 0, !dbg !23
180
+ %177 = bitcast i32 %176 to <2 x bfloat>, !dbg !23
181
+ %178 = extractvalue { i32, i32, i32, i32 } %175, 1, !dbg !23
182
+ %179 = bitcast i32 %178 to <2 x bfloat>, !dbg !23
183
+ %180 = extractvalue { i32, i32, i32, i32 } %175, 2, !dbg !23
184
+ %181 = bitcast i32 %180 to <2 x bfloat>, !dbg !23
185
+ %182 = extractvalue { i32, i32, i32, i32 } %175, 3, !dbg !23
186
+ %183 = bitcast i32 %182 to <2 x bfloat>, !dbg !23
187
+ %184 = extractelement <2 x bfloat> %177, i64 0, !dbg !23
188
+ %185 = extractelement <2 x bfloat> %177, i64 1, !dbg !23
189
+ %186 = extractelement <2 x bfloat> %179, i64 0, !dbg !23
190
+ %187 = extractelement <2 x bfloat> %179, i64 1, !dbg !23
191
+ %188 = extractelement <2 x bfloat> %181, i64 0, !dbg !23
192
+ %189 = extractelement <2 x bfloat> %181, i64 1, !dbg !23
193
+ %190 = extractelement <2 x bfloat> %183, i64 0, !dbg !23
194
+ %191 = extractelement <2 x bfloat> %183, i64 1, !dbg !23
195
+ %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
196
+ %193 = extractvalue { i32, i32, i32, i32 } %192, 0, !dbg !23
197
+ %194 = bitcast i32 %193 to <2 x bfloat>, !dbg !23
198
+ %195 = extractvalue { i32, i32, i32, i32 } %192, 1, !dbg !23
199
+ %196 = bitcast i32 %195 to <2 x bfloat>, !dbg !23
200
+ %197 = extractvalue { i32, i32, i32, i32 } %192, 2, !dbg !23
201
+ %198 = bitcast i32 %197 to <2 x bfloat>, !dbg !23
202
+ %199 = extractvalue { i32, i32, i32, i32 } %192, 3, !dbg !23
203
+ %200 = bitcast i32 %199 to <2 x bfloat>, !dbg !23
204
+ %201 = extractelement <2 x bfloat> %194, i64 0, !dbg !23
205
+ %202 = extractelement <2 x bfloat> %194, i64 1, !dbg !23
206
+ %203 = extractelement <2 x bfloat> %196, i64 0, !dbg !23
207
+ %204 = extractelement <2 x bfloat> %196, i64 1, !dbg !23
208
+ %205 = extractelement <2 x bfloat> %198, i64 0, !dbg !23
209
+ %206 = extractelement <2 x bfloat> %198, i64 1, !dbg !23
210
+ %207 = extractelement <2 x bfloat> %200, i64 0, !dbg !23
211
+ %208 = extractelement <2 x bfloat> %200, i64 1, !dbg !23
212
+ %209 = fpext bfloat %82 to float, !dbg !24
213
+ %210 = fpext bfloat %83 to float, !dbg !24
214
+ %211 = fpext bfloat %84 to float, !dbg !24
215
+ %212 = fpext bfloat %85 to float, !dbg !24
216
+ %213 = fpext bfloat %86 to float, !dbg !24
217
+ %214 = fpext bfloat %87 to float, !dbg !24
218
+ %215 = fpext bfloat %88 to float, !dbg !24
219
+ %216 = fpext bfloat %89 to float, !dbg !24
220
+ %217 = fpext bfloat %99 to float, !dbg !24
221
+ %218 = fpext bfloat %100 to float, !dbg !24
222
+ %219 = fpext bfloat %101 to float, !dbg !24
223
+ %220 = fpext bfloat %102 to float, !dbg !24
224
+ %221 = fpext bfloat %103 to float, !dbg !24
225
+ %222 = fpext bfloat %104 to float, !dbg !24
226
+ %223 = fpext bfloat %105 to float, !dbg !24
227
+ %224 = fpext bfloat %106 to float, !dbg !24
228
+ %225 = fpext bfloat %116 to float, !dbg !24
229
+ %226 = fpext bfloat %117 to float, !dbg !24
230
+ %227 = fpext bfloat %118 to float, !dbg !24
231
+ %228 = fpext bfloat %119 to float, !dbg !24
232
+ %229 = fpext bfloat %120 to float, !dbg !24
233
+ %230 = fpext bfloat %121 to float, !dbg !24
234
+ %231 = fpext bfloat %122 to float, !dbg !24
235
+ %232 = fpext bfloat %123 to float, !dbg !24
236
+ %233 = fpext bfloat %133 to float, !dbg !24
237
+ %234 = fpext bfloat %134 to float, !dbg !24
238
+ %235 = fpext bfloat %135 to float, !dbg !24
239
+ %236 = fpext bfloat %136 to float, !dbg !24
240
+ %237 = fpext bfloat %137 to float, !dbg !24
241
+ %238 = fpext bfloat %138 to float, !dbg !24
242
+ %239 = fpext bfloat %139 to float, !dbg !24
243
+ %240 = fpext bfloat %140 to float, !dbg !24
244
+ %241 = fpext bfloat %150 to float, !dbg !24
245
+ %242 = fpext bfloat %151 to float, !dbg !24
246
+ %243 = fpext bfloat %152 to float, !dbg !24
247
+ %244 = fpext bfloat %153 to float, !dbg !24
248
+ %245 = fpext bfloat %154 to float, !dbg !24
249
+ %246 = fpext bfloat %155 to float, !dbg !24
250
+ %247 = fpext bfloat %156 to float, !dbg !24
251
+ %248 = fpext bfloat %157 to float, !dbg !24
252
+ %249 = fpext bfloat %167 to float, !dbg !24
253
+ %250 = fpext bfloat %168 to float, !dbg !24
254
+ %251 = fpext bfloat %169 to float, !dbg !24
255
+ %252 = fpext bfloat %170 to float, !dbg !24
256
+ %253 = fpext bfloat %171 to float, !dbg !24
257
+ %254 = fpext bfloat %172 to float, !dbg !24
258
+ %255 = fpext bfloat %173 to float, !dbg !24
259
+ %256 = fpext bfloat %174 to float, !dbg !24
260
+ %257 = fpext bfloat %184 to float, !dbg !24
261
+ %258 = fpext bfloat %185 to float, !dbg !24
262
+ %259 = fpext bfloat %186 to float, !dbg !24
263
+ %260 = fpext bfloat %187 to float, !dbg !24
264
+ %261 = fpext bfloat %188 to float, !dbg !24
265
+ %262 = fpext bfloat %189 to float, !dbg !24
266
+ %263 = fpext bfloat %190 to float, !dbg !24
267
+ %264 = fpext bfloat %191 to float, !dbg !24
268
+ %265 = fpext bfloat %201 to float, !dbg !24
269
+ %266 = fpext bfloat %202 to float, !dbg !24
270
+ %267 = fpext bfloat %203 to float, !dbg !24
271
+ %268 = fpext bfloat %204 to float, !dbg !24
272
+ %269 = fpext bfloat %205 to float, !dbg !24
273
+ %270 = fpext bfloat %206 to float, !dbg !24
274
+ %271 = fpext bfloat %207 to float, !dbg !24
275
+ %272 = fpext bfloat %208 to float, !dbg !24
276
+ %273 = fmul float %209, %209, !dbg !25
277
+ %274 = fmul float %210, %210, !dbg !25
278
+ %275 = fmul float %211, %211, !dbg !25
279
+ %276 = fmul float %212, %212, !dbg !25
280
+ %277 = fmul float %213, %213, !dbg !25
281
+ %278 = fmul float %214, %214, !dbg !25
282
+ %279 = fmul float %215, %215, !dbg !25
283
+ %280 = fmul float %216, %216, !dbg !25
284
+ %281 = fmul float %217, %217, !dbg !25
285
+ %282 = fmul float %218, %218, !dbg !25
286
+ %283 = fmul float %219, %219, !dbg !25
287
+ %284 = fmul float %220, %220, !dbg !25
288
+ %285 = fmul float %221, %221, !dbg !25
289
+ %286 = fmul float %222, %222, !dbg !25
290
+ %287 = fmul float %223, %223, !dbg !25
291
+ %288 = fmul float %224, %224, !dbg !25
292
+ %289 = fmul float %225, %225, !dbg !25
293
+ %290 = fmul float %226, %226, !dbg !25
294
+ %291 = fmul float %227, %227, !dbg !25
295
+ %292 = fmul float %228, %228, !dbg !25
296
+ %293 = fmul float %229, %229, !dbg !25
297
+ %294 = fmul float %230, %230, !dbg !25
298
+ %295 = fmul float %231, %231, !dbg !25
299
+ %296 = fmul float %232, %232, !dbg !25
300
+ %297 = fmul float %233, %233, !dbg !25
301
+ %298 = fmul float %234, %234, !dbg !25
302
+ %299 = fmul float %235, %235, !dbg !25
303
+ %300 = fmul float %236, %236, !dbg !25
304
+ %301 = fmul float %237, %237, !dbg !25
305
+ %302 = fmul float %238, %238, !dbg !25
306
+ %303 = fmul float %239, %239, !dbg !25
307
+ %304 = fmul float %240, %240, !dbg !25
308
+ %305 = fmul float %241, %241, !dbg !25
309
+ %306 = fmul float %242, %242, !dbg !25
310
+ %307 = fmul float %243, %243, !dbg !25
311
+ %308 = fmul float %244, %244, !dbg !25
312
+ %309 = fmul float %245, %245, !dbg !25
313
+ %310 = fmul float %246, %246, !dbg !25
314
+ %311 = fmul float %247, %247, !dbg !25
315
+ %312 = fmul float %248, %248, !dbg !25
316
+ %313 = fmul float %249, %249, !dbg !25
317
+ %314 = fmul float %250, %250, !dbg !25
318
+ %315 = fmul float %251, %251, !dbg !25
319
+ %316 = fmul float %252, %252, !dbg !25
320
+ %317 = fmul float %253, %253, !dbg !25
321
+ %318 = fmul float %254, %254, !dbg !25
322
+ %319 = fmul float %255, %255, !dbg !25
323
+ %320 = fmul float %256, %256, !dbg !25
324
+ %321 = fmul float %257, %257, !dbg !25
325
+ %322 = fmul float %258, %258, !dbg !25
326
+ %323 = fmul float %259, %259, !dbg !25
327
+ %324 = fmul float %260, %260, !dbg !25
328
+ %325 = fmul float %261, %261, !dbg !25
329
+ %326 = fmul float %262, %262, !dbg !25
330
+ %327 = fmul float %263, %263, !dbg !25
331
+ %328 = fmul float %264, %264, !dbg !25
332
+ %329 = fmul float %265, %265, !dbg !25
333
+ %330 = fmul float %266, %266, !dbg !25
334
+ %331 = fmul float %267, %267, !dbg !25
335
+ %332 = fmul float %268, %268, !dbg !25
336
+ %333 = fmul float %269, %269, !dbg !25
337
+ %334 = fmul float %270, %270, !dbg !25
338
+ %335 = fmul float %271, %271, !dbg !25
339
+ %336 = fmul float %272, %272, !dbg !25
340
+ %337 = fadd float %273, %274, !dbg !26
341
+ %338 = fadd float %275, %337, !dbg !26
342
+ %339 = fadd float %276, %338, !dbg !26
343
+ %340 = fadd float %277, %339, !dbg !26
344
+ %341 = fadd float %278, %340, !dbg !26
345
+ %342 = fadd float %279, %341, !dbg !26
346
+ %343 = fadd float %280, %342, !dbg !26
347
+ %344 = fadd float %281, %282, !dbg !26
348
+ %345 = fadd float %283, %344, !dbg !26
349
+ %346 = fadd float %284, %345, !dbg !26
350
+ %347 = fadd float %285, %346, !dbg !26
351
+ %348 = fadd float %286, %347, !dbg !26
352
+ %349 = fadd float %287, %348, !dbg !26
353
+ %350 = fadd float %288, %349, !dbg !26
354
+ %351 = fadd float %289, %290, !dbg !26
355
+ %352 = fadd float %291, %351, !dbg !26
356
+ %353 = fadd float %292, %352, !dbg !26
357
+ %354 = fadd float %293, %353, !dbg !26
358
+ %355 = fadd float %294, %354, !dbg !26
359
+ %356 = fadd float %295, %355, !dbg !26
360
+ %357 = fadd float %296, %356, !dbg !26
361
+ %358 = fadd float %297, %298, !dbg !26
362
+ %359 = fadd float %299, %358, !dbg !26
363
+ %360 = fadd float %300, %359, !dbg !26
364
+ %361 = fadd float %301, %360, !dbg !26
365
+ %362 = fadd float %302, %361, !dbg !26
366
+ %363 = fadd float %303, %362, !dbg !26
367
+ %364 = fadd float %304, %363, !dbg !26
368
+ %365 = fadd float %305, %306, !dbg !26
369
+ %366 = fadd float %307, %365, !dbg !26
370
+ %367 = fadd float %308, %366, !dbg !26
371
+ %368 = fadd float %309, %367, !dbg !26
372
+ %369 = fadd float %310, %368, !dbg !26
373
+ %370 = fadd float %311, %369, !dbg !26
374
+ %371 = fadd float %312, %370, !dbg !26
375
+ %372 = fadd float %313, %314, !dbg !26
376
+ %373 = fadd float %315, %372, !dbg !26
377
+ %374 = fadd float %316, %373, !dbg !26
378
+ %375 = fadd float %317, %374, !dbg !26
379
+ %376 = fadd float %318, %375, !dbg !26
380
+ %377 = fadd float %319, %376, !dbg !26
381
+ %378 = fadd float %320, %377, !dbg !26
382
+ %379 = fadd float %321, %322, !dbg !26
383
+ %380 = fadd float %323, %379, !dbg !26
384
+ %381 = fadd float %324, %380, !dbg !26
385
+ %382 = fadd float %325, %381, !dbg !26
386
+ %383 = fadd float %326, %382, !dbg !26
387
+ %384 = fadd float %327, %383, !dbg !26
388
+ %385 = fadd float %328, %384, !dbg !26
389
+ %386 = fadd float %329, %330, !dbg !26
390
+ %387 = fadd float %331, %386, !dbg !26
391
+ %388 = fadd float %332, %387, !dbg !26
392
+ %389 = fadd float %333, %388, !dbg !26
393
+ %390 = fadd float %334, %389, !dbg !26
394
+ %391 = fadd float %335, %390, !dbg !26
395
+ %392 = fadd float %336, %391, !dbg !26
396
+ %393 = bitcast float %343 to i32, !dbg !29
397
+ %394 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %393, i32 8, i32 31), !dbg !29
398
+ %395 = bitcast i32 %394 to float, !dbg !29
399
+ %396 = fadd float %343, %395, !dbg !26
400
+ %397 = bitcast float %396 to i32, !dbg !29
401
+ %398 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %397, i32 4, i32 31), !dbg !29
402
+ %399 = bitcast i32 %398 to float, !dbg !29
403
+ %400 = fadd float %396, %399, !dbg !26
404
+ %401 = bitcast float %400 to i32, !dbg !29
405
+ %402 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %401, i32 2, i32 31), !dbg !29
406
+ %403 = bitcast i32 %402 to float, !dbg !29
407
+ %404 = fadd float %400, %403, !dbg !26
408
+ %405 = bitcast float %404 to i32, !dbg !29
409
+ %406 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %405, i32 1, i32 31), !dbg !29
410
+ %407 = bitcast i32 %406 to float, !dbg !29
411
+ %408 = fadd float %404, %407, !dbg !26
412
+ %409 = bitcast float %350 to i32, !dbg !29
413
+ %410 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %409, i32 8, i32 31), !dbg !29
414
+ %411 = bitcast i32 %410 to float, !dbg !29
415
+ %412 = fadd float %350, %411, !dbg !26
416
+ %413 = bitcast float %412 to i32, !dbg !29
417
+ %414 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %413, i32 4, i32 31), !dbg !29
418
+ %415 = bitcast i32 %414 to float, !dbg !29
419
+ %416 = fadd float %412, %415, !dbg !26
420
+ %417 = bitcast float %416 to i32, !dbg !29
421
+ %418 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %417, i32 2, i32 31), !dbg !29
422
+ %419 = bitcast i32 %418 to float, !dbg !29
423
+ %420 = fadd float %416, %419, !dbg !26
424
+ %421 = bitcast float %420 to i32, !dbg !29
425
+ %422 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %421, i32 1, i32 31), !dbg !29
426
+ %423 = bitcast i32 %422 to float, !dbg !29
427
+ %424 = fadd float %420, %423, !dbg !26
428
+ %425 = bitcast float %357 to i32, !dbg !29
429
+ %426 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %425, i32 8, i32 31), !dbg !29
430
+ %427 = bitcast i32 %426 to float, !dbg !29
431
+ %428 = fadd float %357, %427, !dbg !26
432
+ %429 = bitcast float %428 to i32, !dbg !29
433
+ %430 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %429, i32 4, i32 31), !dbg !29
434
+ %431 = bitcast i32 %430 to float, !dbg !29
435
+ %432 = fadd float %428, %431, !dbg !26
436
+ %433 = bitcast float %432 to i32, !dbg !29
437
+ %434 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %433, i32 2, i32 31), !dbg !29
438
+ %435 = bitcast i32 %434 to float, !dbg !29
439
+ %436 = fadd float %432, %435, !dbg !26
440
+ %437 = bitcast float %436 to i32, !dbg !29
441
+ %438 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %437, i32 1, i32 31), !dbg !29
442
+ %439 = bitcast i32 %438 to float, !dbg !29
443
+ %440 = fadd float %436, %439, !dbg !26
444
+ %441 = bitcast float %364 to i32, !dbg !29
445
+ %442 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %441, i32 8, i32 31), !dbg !29
446
+ %443 = bitcast i32 %442 to float, !dbg !29
447
+ %444 = fadd float %364, %443, !dbg !26
448
+ %445 = bitcast float %444 to i32, !dbg !29
449
+ %446 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %445, i32 4, i32 31), !dbg !29
450
+ %447 = bitcast i32 %446 to float, !dbg !29
451
+ %448 = fadd float %444, %447, !dbg !26
452
+ %449 = bitcast float %448 to i32, !dbg !29
453
+ %450 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %449, i32 2, i32 31), !dbg !29
454
+ %451 = bitcast i32 %450 to float, !dbg !29
455
+ %452 = fadd float %448, %451, !dbg !26
456
+ %453 = bitcast float %452 to i32, !dbg !29
457
+ %454 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %453, i32 1, i32 31), !dbg !29
458
+ %455 = bitcast i32 %454 to float, !dbg !29
459
+ %456 = fadd float %452, %455, !dbg !26
460
+ %457 = bitcast float %371 to i32, !dbg !29
461
+ %458 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %457, i32 8, i32 31), !dbg !29
462
+ %459 = bitcast i32 %458 to float, !dbg !29
463
+ %460 = fadd float %371, %459, !dbg !26
464
+ %461 = bitcast float %460 to i32, !dbg !29
465
+ %462 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %461, i32 4, i32 31), !dbg !29
466
+ %463 = bitcast i32 %462 to float, !dbg !29
467
+ %464 = fadd float %460, %463, !dbg !26
468
+ %465 = bitcast float %464 to i32, !dbg !29
469
+ %466 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %465, i32 2, i32 31), !dbg !29
470
+ %467 = bitcast i32 %466 to float, !dbg !29
471
+ %468 = fadd float %464, %467, !dbg !26
472
+ %469 = bitcast float %468 to i32, !dbg !29
473
+ %470 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %469, i32 1, i32 31), !dbg !29
474
+ %471 = bitcast i32 %470 to float, !dbg !29
475
+ %472 = fadd float %468, %471, !dbg !26
476
+ %473 = bitcast float %378 to i32, !dbg !29
477
+ %474 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %473, i32 8, i32 31), !dbg !29
478
+ %475 = bitcast i32 %474 to float, !dbg !29
479
+ %476 = fadd float %378, %475, !dbg !26
480
+ %477 = bitcast float %476 to i32, !dbg !29
481
+ %478 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %477, i32 4, i32 31), !dbg !29
482
+ %479 = bitcast i32 %478 to float, !dbg !29
483
+ %480 = fadd float %476, %479, !dbg !26
484
+ %481 = bitcast float %480 to i32, !dbg !29
485
+ %482 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %481, i32 2, i32 31), !dbg !29
486
+ %483 = bitcast i32 %482 to float, !dbg !29
487
+ %484 = fadd float %480, %483, !dbg !26
488
+ %485 = bitcast float %484 to i32, !dbg !29
489
+ %486 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %485, i32 1, i32 31), !dbg !29
490
+ %487 = bitcast i32 %486 to float, !dbg !29
491
+ %488 = fadd float %484, %487, !dbg !26
492
+ %489 = bitcast float %385 to i32, !dbg !29
493
+ %490 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %489, i32 8, i32 31), !dbg !29
494
+ %491 = bitcast i32 %490 to float, !dbg !29
495
+ %492 = fadd float %385, %491, !dbg !26
496
+ %493 = bitcast float %492 to i32, !dbg !29
497
+ %494 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %493, i32 4, i32 31), !dbg !29
498
+ %495 = bitcast i32 %494 to float, !dbg !29
499
+ %496 = fadd float %492, %495, !dbg !26
500
+ %497 = bitcast float %496 to i32, !dbg !29
501
+ %498 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %497, i32 2, i32 31), !dbg !29
502
+ %499 = bitcast i32 %498 to float, !dbg !29
503
+ %500 = fadd float %496, %499, !dbg !26
504
+ %501 = bitcast float %500 to i32, !dbg !29
505
+ %502 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %501, i32 1, i32 31), !dbg !29
506
+ %503 = bitcast i32 %502 to float, !dbg !29
507
+ %504 = fadd float %500, %503, !dbg !26
508
+ %505 = bitcast float %392 to i32, !dbg !29
509
+ %506 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %505, i32 8, i32 31), !dbg !29
510
+ %507 = bitcast i32 %506 to float, !dbg !29
511
+ %508 = fadd float %392, %507, !dbg !26
512
+ %509 = bitcast float %508 to i32, !dbg !29
513
+ %510 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %509, i32 4, i32 31), !dbg !29
514
+ %511 = bitcast i32 %510 to float, !dbg !29
515
+ %512 = fadd float %508, %511, !dbg !26
516
+ %513 = bitcast float %512 to i32, !dbg !29
517
+ %514 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %513, i32 2, i32 31), !dbg !29
518
+ %515 = bitcast i32 %514 to float, !dbg !29
519
+ %516 = fadd float %512, %515, !dbg !26
520
+ %517 = bitcast float %516 to i32, !dbg !29
521
+ %518 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %517, i32 1, i32 31), !dbg !29
522
+ %519 = bitcast i32 %518 to float, !dbg !29
523
+ %520 = fadd float %516, %519, !dbg !26
524
+ %521 = fadd float %3, %408, !dbg !32
525
+ %522 = fadd float %3, %424, !dbg !32
526
+ %523 = fadd float %3, %440, !dbg !32
527
+ %524 = fadd float %3, %456, !dbg !32
528
+ %525 = fadd float %3, %472, !dbg !32
529
+ %526 = fadd float %3, %488, !dbg !32
530
+ %527 = fadd float %3, %504, !dbg !32
531
+ %528 = fadd float %3, %520, !dbg !32
532
+ %529 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
533
+ %.not.i = icmp eq i32 %529, 0, !dbg !33
534
+ %530 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
535
+ %.not1.i = icmp eq i32 %530, 0, !dbg !33
536
+ br i1 %.not.i, label %536, label %531, !dbg !33
537
+
538
+ 531: ; preds = %7
539
+ br i1 %.not1.i, label %534, label %532, !dbg !33
540
+
541
+ 532: ; preds = %531
542
+ %533 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %521) #6, !dbg !33
543
+ br label %__nv_sqrtf.exit, !dbg !33
544
+
545
+ 534: ; preds = %531
546
+ %535 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %521) #6, !dbg !33
547
+ br label %__nv_sqrtf.exit, !dbg !33
548
+
549
+ 536: ; preds = %7
550
+ br i1 %.not1.i, label %539, label %537, !dbg !33
551
+
552
+ 537: ; preds = %536
553
+ %538 = tail call float @llvm.nvvm.sqrt.rn.f(float %521) #6, !dbg !33
554
+ br label %__nv_sqrtf.exit, !dbg !33
555
+
556
+ 539: ; preds = %536
557
+ %540 = tail call float @llvm.nvvm.sqrt.approx.f(float %521) #6, !dbg !33
558
+ br label %__nv_sqrtf.exit, !dbg !33
559
+
560
+ __nv_sqrtf.exit: ; preds = %532, %534, %537, %539
561
+ %.0.i = phi float [ %533, %532 ], [ %535, %534 ], [ %538, %537 ], [ %540, %539 ], !dbg !33
562
+ %541 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
563
+ %.not.i1 = icmp eq i32 %541, 0, !dbg !33
564
+ %542 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
565
+ %.not1.i4 = icmp eq i32 %542, 0, !dbg !33
566
+ br i1 %.not.i1, label %548, label %543, !dbg !33
567
+
568
+ 543: ; preds = %__nv_sqrtf.exit
569
+ br i1 %.not1.i4, label %546, label %544, !dbg !33
570
+
571
+ 544: ; preds = %543
572
+ %545 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %522) #6, !dbg !33
573
+ br label %__nv_sqrtf.exit5, !dbg !33
574
+
575
+ 546: ; preds = %543
576
+ %547 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %522) #6, !dbg !33
577
+ br label %__nv_sqrtf.exit5, !dbg !33
578
+
579
+ 548: ; preds = %__nv_sqrtf.exit
580
+ br i1 %.not1.i4, label %551, label %549, !dbg !33
581
+
582
+ 549: ; preds = %548
583
+ %550 = tail call float @llvm.nvvm.sqrt.rn.f(float %522) #6, !dbg !33
584
+ br label %__nv_sqrtf.exit5, !dbg !33
585
+
586
+ 551: ; preds = %548
587
+ %552 = tail call float @llvm.nvvm.sqrt.approx.f(float %522) #6, !dbg !33
588
+ br label %__nv_sqrtf.exit5, !dbg !33
589
+
590
+ __nv_sqrtf.exit5: ; preds = %544, %546, %549, %551
591
+ %.0.i3 = phi float [ %545, %544 ], [ %547, %546 ], [ %550, %549 ], [ %552, %551 ], !dbg !33
592
+ %553 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
593
+ %.not.i6 = icmp eq i32 %553, 0, !dbg !33
594
+ %554 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
595
+ %.not1.i9 = icmp eq i32 %554, 0, !dbg !33
596
+ br i1 %.not.i6, label %560, label %555, !dbg !33
597
+
598
+ 555: ; preds = %__nv_sqrtf.exit5
599
+ br i1 %.not1.i9, label %558, label %556, !dbg !33
600
+
601
+ 556: ; preds = %555
602
+ %557 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %523) #6, !dbg !33
603
+ br label %__nv_sqrtf.exit10, !dbg !33
604
+
605
+ 558: ; preds = %555
606
+ %559 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %523) #6, !dbg !33
607
+ br label %__nv_sqrtf.exit10, !dbg !33
608
+
609
+ 560: ; preds = %__nv_sqrtf.exit5
610
+ br i1 %.not1.i9, label %563, label %561, !dbg !33
611
+
612
+ 561: ; preds = %560
613
+ %562 = tail call float @llvm.nvvm.sqrt.rn.f(float %523) #6, !dbg !33
614
+ br label %__nv_sqrtf.exit10, !dbg !33
615
+
616
+ 563: ; preds = %560
617
+ %564 = tail call float @llvm.nvvm.sqrt.approx.f(float %523) #6, !dbg !33
618
+ br label %__nv_sqrtf.exit10, !dbg !33
619
+
620
+ __nv_sqrtf.exit10: ; preds = %556, %558, %561, %563
621
+ %.0.i8 = phi float [ %557, %556 ], [ %559, %558 ], [ %562, %561 ], [ %564, %563 ], !dbg !33
622
+ %565 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
623
+ %.not.i11 = icmp eq i32 %565, 0, !dbg !33
624
+ %566 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
625
+ %.not1.i14 = icmp eq i32 %566, 0, !dbg !33
626
+ br i1 %.not.i11, label %572, label %567, !dbg !33
627
+
628
+ 567: ; preds = %__nv_sqrtf.exit10
629
+ br i1 %.not1.i14, label %570, label %568, !dbg !33
630
+
631
+ 568: ; preds = %567
632
+ %569 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %524) #6, !dbg !33
633
+ br label %__nv_sqrtf.exit15, !dbg !33
634
+
635
+ 570: ; preds = %567
636
+ %571 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %524) #6, !dbg !33
637
+ br label %__nv_sqrtf.exit15, !dbg !33
638
+
639
+ 572: ; preds = %__nv_sqrtf.exit10
640
+ br i1 %.not1.i14, label %575, label %573, !dbg !33
641
+
642
+ 573: ; preds = %572
643
+ %574 = tail call float @llvm.nvvm.sqrt.rn.f(float %524) #6, !dbg !33
644
+ br label %__nv_sqrtf.exit15, !dbg !33
645
+
646
+ 575: ; preds = %572
647
+ %576 = tail call float @llvm.nvvm.sqrt.approx.f(float %524) #6, !dbg !33
648
+ br label %__nv_sqrtf.exit15, !dbg !33
649
+
650
+ __nv_sqrtf.exit15: ; preds = %568, %570, %573, %575
651
+ %.0.i13 = phi float [ %569, %568 ], [ %571, %570 ], [ %574, %573 ], [ %576, %575 ], !dbg !33
652
+ %577 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
653
+ %.not.i16 = icmp eq i32 %577, 0, !dbg !33
654
+ %578 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
655
+ %.not1.i19 = icmp eq i32 %578, 0, !dbg !33
656
+ br i1 %.not.i16, label %584, label %579, !dbg !33
657
+
658
+ 579: ; preds = %__nv_sqrtf.exit15
659
+ br i1 %.not1.i19, label %582, label %580, !dbg !33
660
+
661
+ 580: ; preds = %579
662
+ %581 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %525) #6, !dbg !33
663
+ br label %__nv_sqrtf.exit20, !dbg !33
664
+
665
+ 582: ; preds = %579
666
+ %583 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %525) #6, !dbg !33
667
+ br label %__nv_sqrtf.exit20, !dbg !33
668
+
669
+ 584: ; preds = %__nv_sqrtf.exit15
670
+ br i1 %.not1.i19, label %587, label %585, !dbg !33
671
+
672
+ 585: ; preds = %584
673
+ %586 = tail call float @llvm.nvvm.sqrt.rn.f(float %525) #6, !dbg !33
674
+ br label %__nv_sqrtf.exit20, !dbg !33
675
+
676
+ 587: ; preds = %584
677
+ %588 = tail call float @llvm.nvvm.sqrt.approx.f(float %525) #6, !dbg !33
678
+ br label %__nv_sqrtf.exit20, !dbg !33
679
+
680
+ __nv_sqrtf.exit20: ; preds = %580, %582, %585, %587
681
+ %.0.i18 = phi float [ %581, %580 ], [ %583, %582 ], [ %586, %585 ], [ %588, %587 ], !dbg !33
682
+ %589 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
683
+ %.not.i21 = icmp eq i32 %589, 0, !dbg !33
684
+ %590 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
685
+ %.not1.i24 = icmp eq i32 %590, 0, !dbg !33
686
+ br i1 %.not.i21, label %596, label %591, !dbg !33
687
+
688
+ 591: ; preds = %__nv_sqrtf.exit20
689
+ br i1 %.not1.i24, label %594, label %592, !dbg !33
690
+
691
+ 592: ; preds = %591
692
+ %593 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %526) #6, !dbg !33
693
+ br label %__nv_sqrtf.exit25, !dbg !33
694
+
695
+ 594: ; preds = %591
696
+ %595 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %526) #6, !dbg !33
697
+ br label %__nv_sqrtf.exit25, !dbg !33
698
+
699
+ 596: ; preds = %__nv_sqrtf.exit20
700
+ br i1 %.not1.i24, label %599, label %597, !dbg !33
701
+
702
+ 597: ; preds = %596
703
+ %598 = tail call float @llvm.nvvm.sqrt.rn.f(float %526) #6, !dbg !33
704
+ br label %__nv_sqrtf.exit25, !dbg !33
705
+
706
+ 599: ; preds = %596
707
+ %600 = tail call float @llvm.nvvm.sqrt.approx.f(float %526) #6, !dbg !33
708
+ br label %__nv_sqrtf.exit25, !dbg !33
709
+
710
+ __nv_sqrtf.exit25: ; preds = %592, %594, %597, %599
711
+ %.0.i23 = phi float [ %593, %592 ], [ %595, %594 ], [ %598, %597 ], [ %600, %599 ], !dbg !33
712
+ %601 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
713
+ %.not.i26 = icmp eq i32 %601, 0, !dbg !33
714
+ %602 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
715
+ %.not1.i29 = icmp eq i32 %602, 0, !dbg !33
716
+ br i1 %.not.i26, label %608, label %603, !dbg !33
717
+
718
+ 603: ; preds = %__nv_sqrtf.exit25
719
+ br i1 %.not1.i29, label %606, label %604, !dbg !33
720
+
721
+ 604: ; preds = %603
722
+ %605 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %527) #6, !dbg !33
723
+ br label %__nv_sqrtf.exit30, !dbg !33
724
+
725
+ 606: ; preds = %603
726
+ %607 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %527) #6, !dbg !33
727
+ br label %__nv_sqrtf.exit30, !dbg !33
728
+
729
+ 608: ; preds = %__nv_sqrtf.exit25
730
+ br i1 %.not1.i29, label %611, label %609, !dbg !33
731
+
732
+ 609: ; preds = %608
733
+ %610 = tail call float @llvm.nvvm.sqrt.rn.f(float %527) #6, !dbg !33
734
+ br label %__nv_sqrtf.exit30, !dbg !33
735
+
736
+ 611: ; preds = %608
737
+ %612 = tail call float @llvm.nvvm.sqrt.approx.f(float %527) #6, !dbg !33
738
+ br label %__nv_sqrtf.exit30, !dbg !33
739
+
740
+ __nv_sqrtf.exit30: ; preds = %604, %606, %609, %611
741
+ %.0.i28 = phi float [ %605, %604 ], [ %607, %606 ], [ %610, %609 ], [ %612, %611 ], !dbg !33
742
+ %613 = tail call i32 @__nvvm_reflect(ptr nonnull @.str) #6, !dbg !33
743
+ %.not.i31 = icmp eq i32 %613, 0, !dbg !33
744
+ %614 = tail call i32 @__nvvm_reflect(ptr nonnull @.str.2) #6, !dbg !33
745
+ %.not1.i34 = icmp eq i32 %614, 0, !dbg !33
746
+ br i1 %.not.i31, label %620, label %615, !dbg !33
747
+
748
+ 615: ; preds = %__nv_sqrtf.exit30
749
+ br i1 %.not1.i34, label %618, label %616, !dbg !33
750
+
751
+ 616: ; preds = %615
752
+ %617 = tail call float @llvm.nvvm.sqrt.rn.ftz.f(float %528) #6, !dbg !33
753
+ br label %__nv_sqrtf.exit35, !dbg !33
754
+
755
+ 618: ; preds = %615
756
+ %619 = tail call float @llvm.nvvm.sqrt.approx.ftz.f(float %528) #6, !dbg !33
757
+ br label %__nv_sqrtf.exit35, !dbg !33
758
+
759
+ 620: ; preds = %__nv_sqrtf.exit30
760
+ br i1 %.not1.i34, label %623, label %621, !dbg !33
761
+
762
+ 621: ; preds = %620
763
+ %622 = tail call float @llvm.nvvm.sqrt.rn.f(float %528) #6, !dbg !33
764
+ br label %__nv_sqrtf.exit35, !dbg !33
765
+
766
+ 623: ; preds = %620
767
+ %624 = tail call float @llvm.nvvm.sqrt.approx.f(float %528) #6, !dbg !33
768
+ br label %__nv_sqrtf.exit35, !dbg !33
769
+
770
+ __nv_sqrtf.exit35: ; preds = %616, %618, %621, %623
771
+ %.0.i33 = phi float [ %617, %616 ], [ %619, %618 ], [ %622, %621 ], [ %624, %623 ], !dbg !33
772
+ %625 = and i32 %11, 31, !dbg !16
773
+ %626 = zext nneg i32 %625 to i64, !dbg !17
774
+ %627 = or disjoint i64 %10, %626, !dbg !17
775
+ %628 = getelementptr float, ptr addrspace(1) %2, i64 %627, !dbg !34
776
+ %629 = getelementptr bfloat, ptr addrspace(1) %1, i64 %55, !dbg !35
777
+ %630 = getelementptr bfloat, ptr addrspace(1) %629, i64 %64, !dbg !36
778
+ %631 = getelementptr bfloat, ptr addrspace(1) %1, i64 %54, !dbg !35
779
+ %632 = getelementptr bfloat, ptr addrspace(1) %631, i64 %64, !dbg !36
780
+ %633 = getelementptr bfloat, ptr addrspace(1) %1, i64 %53, !dbg !35
781
+ %634 = getelementptr bfloat, ptr addrspace(1) %633, i64 %64, !dbg !36
782
+ %635 = getelementptr bfloat, ptr addrspace(1) %1, i64 %52, !dbg !35
783
+ %636 = getelementptr bfloat, ptr addrspace(1) %635, i64 %64, !dbg !36
784
+ %637 = getelementptr bfloat, ptr addrspace(1) %1, i64 %51, !dbg !35
785
+ %638 = getelementptr bfloat, ptr addrspace(1) %637, i64 %64, !dbg !36
786
+ %639 = getelementptr bfloat, ptr addrspace(1) %1, i64 %50, !dbg !35
787
+ %640 = getelementptr bfloat, ptr addrspace(1) %639, i64 %64, !dbg !36
788
+ %641 = getelementptr bfloat, ptr addrspace(1) %1, i64 %49, !dbg !35
789
+ %642 = getelementptr bfloat, ptr addrspace(1) %641, i64 %64, !dbg !36
790
+ %643 = getelementptr bfloat, ptr addrspace(1) %1, i64 %48, !dbg !35
791
+ %644 = getelementptr bfloat, ptr addrspace(1) %643, i64 %64, !dbg !36
792
+ %645 = icmp slt i64 %627, %37, !dbg !18
793
+ %646 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i), !dbg !37
794
+ %647 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i3), !dbg !37
795
+ %648 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i8), !dbg !37
796
+ %649 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i13), !dbg !37
797
+ %650 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i18), !dbg !37
798
+ %651 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i23), !dbg !37
799
+ %652 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i28), !dbg !37
800
+ %653 = tail call float @llvm.nvvm.div.full(float 1.000000e+00, float %.0.i33), !dbg !37
801
+ %654 = fmul float %646, %209, !dbg !38
802
+ %655 = fmul float %646, %210, !dbg !38
803
+ %656 = fmul float %646, %211, !dbg !38
804
+ %657 = fmul float %646, %212, !dbg !38
805
+ %658 = fmul float %646, %213, !dbg !38
806
+ %659 = fmul float %646, %214, !dbg !38
807
+ %660 = fmul float %646, %215, !dbg !38
808
+ %661 = fmul float %646, %216, !dbg !38
809
+ %662 = fmul float %647, %217, !dbg !38
810
+ %663 = fmul float %647, %218, !dbg !38
811
+ %664 = fmul float %647, %219, !dbg !38
812
+ %665 = fmul float %647, %220, !dbg !38
813
+ %666 = fmul float %647, %221, !dbg !38
814
+ %667 = fmul float %647, %222, !dbg !38
815
+ %668 = fmul float %647, %223, !dbg !38
816
+ %669 = fmul float %647, %224, !dbg !38
817
+ %670 = fmul float %648, %225, !dbg !38
818
+ %671 = fmul float %648, %226, !dbg !38
819
+ %672 = fmul float %648, %227, !dbg !38
820
+ %673 = fmul float %648, %228, !dbg !38
821
+ %674 = fmul float %648, %229, !dbg !38
822
+ %675 = fmul float %648, %230, !dbg !38
823
+ %676 = fmul float %648, %231, !dbg !38
824
+ %677 = fmul float %648, %232, !dbg !38
825
+ %678 = fmul float %649, %233, !dbg !38
826
+ %679 = fmul float %649, %234, !dbg !38
827
+ %680 = fmul float %649, %235, !dbg !38
828
+ %681 = fmul float %649, %236, !dbg !38
829
+ %682 = fmul float %649, %237, !dbg !38
830
+ %683 = fmul float %649, %238, !dbg !38
831
+ %684 = fmul float %649, %239, !dbg !38
832
+ %685 = fmul float %649, %240, !dbg !38
833
+ %686 = fmul float %650, %241, !dbg !38
834
+ %687 = fmul float %650, %242, !dbg !38
835
+ %688 = fmul float %650, %243, !dbg !38
836
+ %689 = fmul float %650, %244, !dbg !38
837
+ %690 = fmul float %650, %245, !dbg !38
838
+ %691 = fmul float %650, %246, !dbg !38
839
+ %692 = fmul float %650, %247, !dbg !38
840
+ %693 = fmul float %650, %248, !dbg !38
841
+ %694 = fmul float %651, %249, !dbg !38
842
+ %695 = fmul float %651, %250, !dbg !38
843
+ %696 = fmul float %651, %251, !dbg !38
844
+ %697 = fmul float %651, %252, !dbg !38
845
+ %698 = fmul float %651, %253, !dbg !38
846
+ %699 = fmul float %651, %254, !dbg !38
847
+ %700 = fmul float %651, %255, !dbg !38
848
+ %701 = fmul float %651, %256, !dbg !38
849
+ %702 = fmul float %652, %257, !dbg !38
850
+ %703 = fmul float %652, %258, !dbg !38
851
+ %704 = fmul float %652, %259, !dbg !38
852
+ %705 = fmul float %652, %260, !dbg !38
853
+ %706 = fmul float %652, %261, !dbg !38
854
+ %707 = fmul float %652, %262, !dbg !38
855
+ %708 = fmul float %652, %263, !dbg !38
856
+ %709 = fmul float %652, %264, !dbg !38
857
+ %710 = fmul float %653, %265, !dbg !38
858
+ %711 = fmul float %653, %266, !dbg !38
859
+ %712 = fmul float %653, %267, !dbg !38
860
+ %713 = fmul float %653, %268, !dbg !38
861
+ %714 = fmul float %653, %269, !dbg !38
862
+ %715 = fmul float %653, %270, !dbg !38
863
+ %716 = fmul float %653, %271, !dbg !38
864
+ %717 = fmul float %653, %272, !dbg !38
865
+ %718 = fptrunc float %654 to bfloat, !dbg !39
866
+ %719 = fptrunc float %655 to bfloat, !dbg !39
867
+ %720 = fptrunc float %656 to bfloat, !dbg !39
868
+ %721 = fptrunc float %657 to bfloat, !dbg !39
869
+ %722 = fptrunc float %658 to bfloat, !dbg !39
870
+ %723 = fptrunc float %659 to bfloat, !dbg !39
871
+ %724 = fptrunc float %660 to bfloat, !dbg !39
872
+ %725 = fptrunc float %661 to bfloat, !dbg !39
873
+ %726 = fptrunc float %662 to bfloat, !dbg !39
874
+ %727 = fptrunc float %663 to bfloat, !dbg !39
875
+ %728 = fptrunc float %664 to bfloat, !dbg !39
876
+ %729 = fptrunc float %665 to bfloat, !dbg !39
877
+ %730 = fptrunc float %666 to bfloat, !dbg !39
878
+ %731 = fptrunc float %667 to bfloat, !dbg !39
879
+ %732 = fptrunc float %668 to bfloat, !dbg !39
880
+ %733 = fptrunc float %669 to bfloat, !dbg !39
881
+ %734 = fptrunc float %670 to bfloat, !dbg !39
882
+ %735 = fptrunc float %671 to bfloat, !dbg !39
883
+ %736 = fptrunc float %672 to bfloat, !dbg !39
884
+ %737 = fptrunc float %673 to bfloat, !dbg !39
885
+ %738 = fptrunc float %674 to bfloat, !dbg !39
886
+ %739 = fptrunc float %675 to bfloat, !dbg !39
887
+ %740 = fptrunc float %676 to bfloat, !dbg !39
888
+ %741 = fptrunc float %677 to bfloat, !dbg !39
889
+ %742 = fptrunc float %678 to bfloat, !dbg !39
890
+ %743 = fptrunc float %679 to bfloat, !dbg !39
891
+ %744 = fptrunc float %680 to bfloat, !dbg !39
892
+ %745 = fptrunc float %681 to bfloat, !dbg !39
893
+ %746 = fptrunc float %682 to bfloat, !dbg !39
894
+ %747 = fptrunc float %683 to bfloat, !dbg !39
895
+ %748 = fptrunc float %684 to bfloat, !dbg !39
896
+ %749 = fptrunc float %685 to bfloat, !dbg !39
897
+ %750 = fptrunc float %686 to bfloat, !dbg !39
898
+ %751 = fptrunc float %687 to bfloat, !dbg !39
899
+ %752 = fptrunc float %688 to bfloat, !dbg !39
900
+ %753 = fptrunc float %689 to bfloat, !dbg !39
901
+ %754 = fptrunc float %690 to bfloat, !dbg !39
902
+ %755 = fptrunc float %691 to bfloat, !dbg !39
903
+ %756 = fptrunc float %692 to bfloat, !dbg !39
904
+ %757 = fptrunc float %693 to bfloat, !dbg !39
905
+ %758 = fptrunc float %694 to bfloat, !dbg !39
906
+ %759 = fptrunc float %695 to bfloat, !dbg !39
907
+ %760 = fptrunc float %696 to bfloat, !dbg !39
908
+ %761 = fptrunc float %697 to bfloat, !dbg !39
909
+ %762 = fptrunc float %698 to bfloat, !dbg !39
910
+ %763 = fptrunc float %699 to bfloat, !dbg !39
911
+ %764 = fptrunc float %700 to bfloat, !dbg !39
912
+ %765 = fptrunc float %701 to bfloat, !dbg !39
913
+ %766 = fptrunc float %702 to bfloat, !dbg !39
914
+ %767 = fptrunc float %703 to bfloat, !dbg !39
915
+ %768 = fptrunc float %704 to bfloat, !dbg !39
916
+ %769 = fptrunc float %705 to bfloat, !dbg !39
917
+ %770 = fptrunc float %706 to bfloat, !dbg !39
918
+ %771 = fptrunc float %707 to bfloat, !dbg !39
919
+ %772 = fptrunc float %708 to bfloat, !dbg !39
920
+ %773 = fptrunc float %709 to bfloat, !dbg !39
921
+ %774 = fptrunc float %710 to bfloat, !dbg !39
922
+ %775 = fptrunc float %711 to bfloat, !dbg !39
923
+ %776 = fptrunc float %712 to bfloat, !dbg !39
924
+ %777 = fptrunc float %713 to bfloat, !dbg !39
925
+ %778 = fptrunc float %714 to bfloat, !dbg !39
926
+ %779 = fptrunc float %715 to bfloat, !dbg !39
927
+ %780 = fptrunc float %716 to bfloat, !dbg !39
928
+ %781 = fptrunc float %717 to bfloat, !dbg !39
929
+ %782 = insertelement <2 x bfloat> poison, bfloat %718, i64 0, !dbg !40
930
+ %783 = insertelement <2 x bfloat> %782, bfloat %719, i64 1, !dbg !40
931
+ %784 = bitcast <2 x bfloat> %783 to i32, !dbg !40
932
+ %785 = insertelement <2 x bfloat> poison, bfloat %720, i64 0, !dbg !40
933
+ %786 = insertelement <2 x bfloat> %785, bfloat %721, i64 1, !dbg !40
934
+ %787 = bitcast <2 x bfloat> %786 to i32, !dbg !40
935
+ %788 = insertelement <2 x bfloat> poison, bfloat %722, i64 0, !dbg !40
936
+ %789 = insertelement <2 x bfloat> %788, bfloat %723, i64 1, !dbg !40
937
+ %790 = bitcast <2 x bfloat> %789 to i32, !dbg !40
938
+ %791 = insertelement <2 x bfloat> poison, bfloat %724, i64 0, !dbg !40
939
+ %792 = insertelement <2 x bfloat> %791, bfloat %725, i64 1, !dbg !40
940
+ %793 = bitcast <2 x bfloat> %792 to i32, !dbg !40
941
+ 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
942
+ %794 = insertelement <2 x bfloat> poison, bfloat %726, i64 0, !dbg !40
943
+ %795 = insertelement <2 x bfloat> %794, bfloat %727, i64 1, !dbg !40
944
+ %796 = bitcast <2 x bfloat> %795 to i32, !dbg !40
945
+ %797 = insertelement <2 x bfloat> poison, bfloat %728, i64 0, !dbg !40
946
+ %798 = insertelement <2 x bfloat> %797, bfloat %729, i64 1, !dbg !40
947
+ %799 = bitcast <2 x bfloat> %798 to i32, !dbg !40
948
+ %800 = insertelement <2 x bfloat> poison, bfloat %730, i64 0, !dbg !40
949
+ %801 = insertelement <2 x bfloat> %800, bfloat %731, i64 1, !dbg !40
950
+ %802 = bitcast <2 x bfloat> %801 to i32, !dbg !40
951
+ %803 = insertelement <2 x bfloat> poison, bfloat %732, i64 0, !dbg !40
952
+ %804 = insertelement <2 x bfloat> %803, bfloat %733, i64 1, !dbg !40
953
+ %805 = bitcast <2 x bfloat> %804 to i32, !dbg !40
954
+ 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
955
+ %806 = insertelement <2 x bfloat> poison, bfloat %734, i64 0, !dbg !40
956
+ %807 = insertelement <2 x bfloat> %806, bfloat %735, i64 1, !dbg !40
957
+ %808 = bitcast <2 x bfloat> %807 to i32, !dbg !40
958
+ %809 = insertelement <2 x bfloat> poison, bfloat %736, i64 0, !dbg !40
959
+ %810 = insertelement <2 x bfloat> %809, bfloat %737, i64 1, !dbg !40
960
+ %811 = bitcast <2 x bfloat> %810 to i32, !dbg !40
961
+ %812 = insertelement <2 x bfloat> poison, bfloat %738, i64 0, !dbg !40
962
+ %813 = insertelement <2 x bfloat> %812, bfloat %739, i64 1, !dbg !40
963
+ %814 = bitcast <2 x bfloat> %813 to i32, !dbg !40
964
+ %815 = insertelement <2 x bfloat> poison, bfloat %740, i64 0, !dbg !40
965
+ %816 = insertelement <2 x bfloat> %815, bfloat %741, i64 1, !dbg !40
966
+ %817 = bitcast <2 x bfloat> %816 to i32, !dbg !40
967
+ 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
968
+ %818 = insertelement <2 x bfloat> poison, bfloat %742, i64 0, !dbg !40
969
+ %819 = insertelement <2 x bfloat> %818, bfloat %743, i64 1, !dbg !40
970
+ %820 = bitcast <2 x bfloat> %819 to i32, !dbg !40
971
+ %821 = insertelement <2 x bfloat> poison, bfloat %744, i64 0, !dbg !40
972
+ %822 = insertelement <2 x bfloat> %821, bfloat %745, i64 1, !dbg !40
973
+ %823 = bitcast <2 x bfloat> %822 to i32, !dbg !40
974
+ %824 = insertelement <2 x bfloat> poison, bfloat %746, i64 0, !dbg !40
975
+ %825 = insertelement <2 x bfloat> %824, bfloat %747, i64 1, !dbg !40
976
+ %826 = bitcast <2 x bfloat> %825 to i32, !dbg !40
977
+ %827 = insertelement <2 x bfloat> poison, bfloat %748, i64 0, !dbg !40
978
+ %828 = insertelement <2 x bfloat> %827, bfloat %749, i64 1, !dbg !40
979
+ %829 = bitcast <2 x bfloat> %828 to i32, !dbg !40
980
+ 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
981
+ %830 = insertelement <2 x bfloat> poison, bfloat %750, i64 0, !dbg !40
982
+ %831 = insertelement <2 x bfloat> %830, bfloat %751, i64 1, !dbg !40
983
+ %832 = bitcast <2 x bfloat> %831 to i32, !dbg !40
984
+ %833 = insertelement <2 x bfloat> poison, bfloat %752, i64 0, !dbg !40
985
+ %834 = insertelement <2 x bfloat> %833, bfloat %753, i64 1, !dbg !40
986
+ %835 = bitcast <2 x bfloat> %834 to i32, !dbg !40
987
+ %836 = insertelement <2 x bfloat> poison, bfloat %754, i64 0, !dbg !40
988
+ %837 = insertelement <2 x bfloat> %836, bfloat %755, i64 1, !dbg !40
989
+ %838 = bitcast <2 x bfloat> %837 to i32, !dbg !40
990
+ %839 = insertelement <2 x bfloat> poison, bfloat %756, i64 0, !dbg !40
991
+ %840 = insertelement <2 x bfloat> %839, bfloat %757, i64 1, !dbg !40
992
+ %841 = bitcast <2 x bfloat> %840 to i32, !dbg !40
993
+ 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
994
+ %842 = insertelement <2 x bfloat> poison, bfloat %758, i64 0, !dbg !40
995
+ %843 = insertelement <2 x bfloat> %842, bfloat %759, i64 1, !dbg !40
996
+ %844 = bitcast <2 x bfloat> %843 to i32, !dbg !40
997
+ %845 = insertelement <2 x bfloat> poison, bfloat %760, i64 0, !dbg !40
998
+ %846 = insertelement <2 x bfloat> %845, bfloat %761, i64 1, !dbg !40
999
+ %847 = bitcast <2 x bfloat> %846 to i32, !dbg !40
1000
+ %848 = insertelement <2 x bfloat> poison, bfloat %762, i64 0, !dbg !40
1001
+ %849 = insertelement <2 x bfloat> %848, bfloat %763, i64 1, !dbg !40
1002
+ %850 = bitcast <2 x bfloat> %849 to i32, !dbg !40
1003
+ %851 = insertelement <2 x bfloat> poison, bfloat %764, i64 0, !dbg !40
1004
+ %852 = insertelement <2 x bfloat> %851, bfloat %765, i64 1, !dbg !40
1005
+ %853 = bitcast <2 x bfloat> %852 to i32, !dbg !40
1006
+ 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
1007
+ %854 = insertelement <2 x bfloat> poison, bfloat %766, i64 0, !dbg !40
1008
+ %855 = insertelement <2 x bfloat> %854, bfloat %767, i64 1, !dbg !40
1009
+ %856 = bitcast <2 x bfloat> %855 to i32, !dbg !40
1010
+ %857 = insertelement <2 x bfloat> poison, bfloat %768, i64 0, !dbg !40
1011
+ %858 = insertelement <2 x bfloat> %857, bfloat %769, i64 1, !dbg !40
1012
+ %859 = bitcast <2 x bfloat> %858 to i32, !dbg !40
1013
+ %860 = insertelement <2 x bfloat> poison, bfloat %770, i64 0, !dbg !40
1014
+ %861 = insertelement <2 x bfloat> %860, bfloat %771, i64 1, !dbg !40
1015
+ %862 = bitcast <2 x bfloat> %861 to i32, !dbg !40
1016
+ %863 = insertelement <2 x bfloat> poison, bfloat %772, i64 0, !dbg !40
1017
+ %864 = insertelement <2 x bfloat> %863, bfloat %773, i64 1, !dbg !40
1018
+ %865 = bitcast <2 x bfloat> %864 to i32, !dbg !40
1019
+ 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
1020
+ %866 = insertelement <2 x bfloat> poison, bfloat %774, i64 0, !dbg !40
1021
+ %867 = insertelement <2 x bfloat> %866, bfloat %775, i64 1, !dbg !40
1022
+ %868 = bitcast <2 x bfloat> %867 to i32, !dbg !40
1023
+ %869 = insertelement <2 x bfloat> poison, bfloat %776, i64 0, !dbg !40
1024
+ %870 = insertelement <2 x bfloat> %869, bfloat %777, i64 1, !dbg !40
1025
+ %871 = bitcast <2 x bfloat> %870 to i32, !dbg !40
1026
+ %872 = insertelement <2 x bfloat> poison, bfloat %778, i64 0, !dbg !40
1027
+ %873 = insertelement <2 x bfloat> %872, bfloat %779, i64 1, !dbg !40
1028
+ %874 = bitcast <2 x bfloat> %873 to i32, !dbg !40
1029
+ %875 = insertelement <2 x bfloat> poison, bfloat %780, i64 0, !dbg !40
1030
+ %876 = insertelement <2 x bfloat> %875, bfloat %781, i64 1, !dbg !40
1031
+ %877 = bitcast <2 x bfloat> %876 to i32, !dbg !40
1032
+ 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
1033
+ %878 = shl nuw nsw i32 %12, 1, !dbg !41
1034
+ %879 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %878, !dbg !41
1035
+ %880 = bitcast float %646 to i32, !dbg !41
1036
+ %881 = bitcast float %647 to i32, !dbg !41
1037
+ %882 = bitcast float %648 to i32, !dbg !41
1038
+ %883 = bitcast float %649 to i32, !dbg !41
1039
+ 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
1040
+ %884 = getelementptr inbounds nuw i8, ptr addrspace(3) %879, i32 16, !dbg !41
1041
+ %885 = bitcast float %650 to i32, !dbg !41
1042
+ %886 = bitcast float %651 to i32, !dbg !41
1043
+ %887 = bitcast float %652 to i32, !dbg !41
1044
+ %888 = bitcast float %653 to i32, !dbg !41
1045
+ 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
1046
+ tail call void @llvm.nvvm.barrier.cta.sync.aligned.all(i32 0), !dbg !41
1047
+ %889 = shl nuw nsw i32 %11, 5, !dbg !41
1048
+ %890 = and i32 %889, 96, !dbg !41
1049
+ %891 = and i32 %11, 28, !dbg !41
1050
+ %892 = getelementptr inbounds nuw i8, ptr addrspace(3) @global_smem, i32 %890, !dbg !41
1051
+ %893 = getelementptr inbounds nuw i8, ptr addrspace(3) %892, i32 %891, !dbg !41
1052
+ %894 = load i32, ptr addrspace(3) %893, align 4, !dbg !41
1053
+ %895 = and i32 %11, 32, !dbg !41
1054
+ %896 = icmp eq i32 %895, 0, !dbg !41
1055
+ %897 = and i1 %896, %645, !dbg !41
1056
+ 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
1057
+ ret void, !dbg !42
1058
+ }
1059
+
1060
+ ; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
1061
+ declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1
1062
+
1063
+ ; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
1064
+ declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1
1065
+
1066
+ ; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite)
1067
+ declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2
1068
+
1069
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
1070
+ declare float @llvm.nvvm.div.full(float, float) #3
1071
+
1072
+ ; Function Attrs: convergent nocallback nounwind
1073
+ declare void @llvm.nvvm.barrier.cta.sync.aligned.all(i32) #4
1074
+
1075
+ declare i32 @__nvvm_reflect(ptr) local_unnamed_addr #5
1076
+
1077
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
1078
+ declare float @llvm.nvvm.sqrt.rn.ftz.f(float) #3
1079
+
1080
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
1081
+ declare float @llvm.nvvm.sqrt.approx.ftz.f(float) #3
1082
+
1083
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
1084
+ declare float @llvm.nvvm.sqrt.rn.f(float) #3
1085
+
1086
+ ; Function Attrs: mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none)
1087
+ declare float @llvm.nvvm.sqrt.approx.f(float) #3
1088
+
1089
+ attributes #0 = { nounwind "nvvm.reqntid"="64" }
1090
+ attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) }
1091
+ attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) }
1092
+ attributes #3 = { mustprogress nocallback nocreateundeforpoison nofree nosync nounwind willreturn memory(none) }
1093
+ attributes #4 = { convergent nocallback nounwind }
1094
+ 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" }
1095
+ attributes #6 = { nounwind }
1096
+
1097
+ !llvm.dbg.cu = !{!0}
1098
+ !llvm.module.flags = !{!2, !3}
1099
+ !llvm.ident = !{!4}
1100
+
1101
+ !0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly)
1102
+ !1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm")
1103
+ !2 = !{i32 2, !"Debug Info Version", i32 3}
1104
+ !3 = !{i32 4, !"nvvm-reflect-ftz", i32 1}
1105
+ !4 = !{!"clang version 3.8.0 (tags/RELEASE_380/final)"}
1106
+ !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)
1107
+ !6 = !DISubroutineType(cc: DW_CC_normal, types: !7)
1108
+ !7 = !{null, !8, !8, !10, !11, !12, !8, !8}
1109
+ !8 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !9, size: 64, dwarfAddressSpace: 1)
1110
+ !9 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed)
1111
+ !10 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !11, size: 64, dwarfAddressSpace: 1)
1112
+ !11 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float)
1113
+ !12 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
1114
+ !13 = !DILocation(line: 70, column: 24, scope: !5)
1115
+ !14 = !DILocation(line: 70, column: 30, scope: !5)
1116
+ !15 = !DILocation(line: 71, column: 16, scope: !5)
1117
+ !16 = !DILocation(line: 71, column: 34, scope: !5)
1118
+ !17 = !DILocation(line: 71, column: 21, scope: !5)
1119
+ !18 = !DILocation(line: 73, column: 16, scope: !5)
1120
+ !19 = !DILocation(line: 74, column: 30, scope: !5)
1121
+ !20 = !DILocation(line: 75, column: 29, scope: !5)
1122
+ !21 = !DILocation(line: 75, column: 14, scope: !5)
1123
+ !22 = !DILocation(line: 75, column: 33, scope: !5)
1124
+ !23 = !DILocation(line: 79, column: 18, scope: !5)
1125
+ !24 = !DILocation(line: 79, column: 47, scope: !5)
1126
+ !25 = !DILocation(line: 80, column: 38, scope: !5)
1127
+ !26 = !DILocation(line: 263, column: 15, scope: !27, inlinedAt: !29)
1128
+ !27 = distinct !DILexicalBlockFile(scope: !5, file: !28, discriminator: 0)
1129
+ !28 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language")
1130
+ !29 = !DILocation(line: 293, column: 36, scope: !27, inlinedAt: !30)
1131
+ !30 = !DILocation(line: 80, column: 43, scope: !31)
1132
+ !31 = distinct !DILexicalBlockFile(scope: !5, file: !1, discriminator: 0)
1133
+ !32 = !DILocation(line: 80, column: 48, scope: !5)
1134
+ !33 = !DILocation(line: 80, column: 25, scope: !5)
1135
+ !34 = !DILocation(line: 77, column: 20, scope: !5)
1136
+ !35 = !DILocation(line: 76, column: 14, scope: !5)
1137
+ !36 = !DILocation(line: 76, column: 33, scope: !5)
1138
+ !37 = !DILocation(line: 80, column: 17, scope: !5)
1139
+ !38 = !DILocation(line: 81, column: 16, scope: !5)
1140
+ !39 = !DILocation(line: 83, column: 25, scope: !5)
1141
+ !40 = !DILocation(line: 83, column: 18, scope: !5)
1142
+ !41 = !DILocation(line: 84, column: 21, scope: !5)
1143
+ !42 = !DILocation(line: 84, column: 4, scope: !5)
triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ptx ADDED
@@ -0,0 +1,966 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ //
2
+ // Generated by LLVM NVPTX Back-End
3
+ //
4
+
5
+ .version 9.1
6
+ .target sm_100a
7
+ .address_size 64
8
+
9
+ // .globl l2norm_fwd_kernel // -- Begin function l2norm_fwd_kernel
10
+ .extern .shared .align 16 .b8 global_smem[];
11
+ .global .align 1 .b8 _$_str[11] = {95, 95, 67, 85, 68, 65, 95, 70, 84, 90};
12
+ .global .align 1 .b8 _$_str_$_2[17] = {95, 95, 67, 85, 68, 65, 95, 80, 82, 69, 67, 95, 83, 81, 82, 84};
13
+ // @l2norm_fwd_kernel
14
+ .visible .entry l2norm_fwd_kernel(
15
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_0,
16
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_1,
17
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_2,
18
+ .param .f32 l2norm_fwd_kernel_param_3,
19
+ .param .u32 l2norm_fwd_kernel_param_4,
20
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_5,
21
+ .param .u64 .ptr .global .align 1 l2norm_fwd_kernel_param_6
22
+ )
23
+ .reqntid 64
24
+ {
25
+ .reg .pred %p<12>;
26
+ .reg .b16 %rs<65>;
27
+ .reg .b32 %r<373>;
28
+ .reg .b64 %rd<67>;
29
+ .loc 1 69 0 // l2norm.py:69:0
30
+ $L__func_begin0:
31
+ .loc 1 69 0 // l2norm.py:69:0
32
+
33
+ // %bb.0: // %__nv_sqrtf.exit
34
+ ld.param.b64 %rd18, [l2norm_fwd_kernel_param_0];
35
+ ld.param.b64 %rd19, [l2norm_fwd_kernel_param_1];
36
+ $L__tmp0:
37
+ .loc 1 70 24 // l2norm.py:70:24
38
+ mov.u32 %r77, %ctaid.x;
39
+ ld.param.b64 %rd20, [l2norm_fwd_kernel_param_2];
40
+ .loc 1 71 16 // l2norm.py:71:16
41
+ mul.wide.u32 %rd21, %r77, 32;
42
+ ld.param.b32 %r78, [l2norm_fwd_kernel_param_3];
43
+ ld.param.s32 %rd22, [l2norm_fwd_kernel_param_4];
44
+ .loc 1 71 34 // l2norm.py:71:34
45
+ mov.u32 %r79, %tid.x;
46
+ and.b32 %r80, %r79, 48;
47
+ bfe.u32 %r81, %r79, 4, 2;
48
+ or.b32 %r82, %r81, 4;
49
+ or.b32 %r83, %r81, 8;
50
+ or.b32 %r84, %r81, 12;
51
+ or.b32 %r85, %r81, 16;
52
+ or.b32 %r86, %r81, 20;
53
+ or.b32 %r87, %r81, 24;
54
+ or.b32 %r88, %r81, 28;
55
+ .loc 1 71 21 // l2norm.py:71:21
56
+ cvt.u64.u32 %rd23, %r81;
57
+ cvt.u64.u32 %rd24, %r82;
58
+ cvt.u64.u32 %rd25, %r83;
59
+ cvt.u64.u32 %rd26, %r84;
60
+ cvt.u64.u32 %rd27, %r85;
61
+ cvt.u64.u32 %rd28, %r86;
62
+ cvt.u64.u32 %rd29, %r87;
63
+ cvt.u64.u32 %rd30, %r88;
64
+ or.b64 %rd31, %rd21, %rd23;
65
+ or.b64 %rd32, %rd21, %rd24;
66
+ or.b64 %rd33, %rd21, %rd25;
67
+ or.b64 %rd34, %rd21, %rd26;
68
+ or.b64 %rd35, %rd21, %rd27;
69
+ or.b64 %rd36, %rd21, %rd28;
70
+ or.b64 %rd37, %rd21, %rd29;
71
+ or.b64 %rd38, %rd21, %rd30;
72
+ .loc 1 73 16 // l2norm.py:73:16
73
+ setp.lt.s64 %p1, %rd31, %rd22;
74
+ setp.lt.s64 %p2, %rd32, %rd22;
75
+ setp.lt.s64 %p3, %rd33, %rd22;
76
+ setp.lt.s64 %p4, %rd34, %rd22;
77
+ setp.lt.s64 %p5, %rd35, %rd22;
78
+ setp.lt.s64 %p6, %rd36, %rd22;
79
+ setp.lt.s64 %p7, %rd37, %rd22;
80
+ setp.lt.s64 %p8, %rd38, %rd22;
81
+ .loc 1 74 30 // l2norm.py:74:30
82
+ shl.b32 %r89, %r79, 3;
83
+ and.b32 %r90, %r89, 120;
84
+ .loc 1 75 14 // l2norm.py:75:14
85
+ shl.b64 %rd39, %rd31, 8;
86
+ add.s64 %rd40, %rd18, %rd39;
87
+ shl.b64 %rd41, %rd32, 8;
88
+ add.s64 %rd42, %rd18, %rd41;
89
+ shl.b64 %rd43, %rd33, 8;
90
+ add.s64 %rd44, %rd18, %rd43;
91
+ shl.b64 %rd45, %rd34, 8;
92
+ add.s64 %rd46, %rd18, %rd45;
93
+ shl.b64 %rd47, %rd35, 8;
94
+ add.s64 %rd48, %rd18, %rd47;
95
+ shl.b64 %rd49, %rd36, 8;
96
+ add.s64 %rd50, %rd18, %rd49;
97
+ shl.b64 %rd51, %rd37, 8;
98
+ add.s64 %rd52, %rd18, %rd51;
99
+ shl.b64 %rd53, %rd38, 8;
100
+ add.s64 %rd54, %rd18, %rd53;
101
+ .loc 1 75 33 // l2norm.py:75:33
102
+ mul.wide.u32 %rd55, %r90, 2;
103
+ add.s64 %rd1, %rd40, %rd55;
104
+ add.s64 %rd2, %rd42, %rd55;
105
+ add.s64 %rd3, %rd44, %rd55;
106
+ add.s64 %rd4, %rd46, %rd55;
107
+ add.s64 %rd5, %rd48, %rd55;
108
+ add.s64 %rd6, %rd50, %rd55;
109
+ add.s64 %rd7, %rd52, %rd55;
110
+ add.s64 %rd8, %rd54, %rd55;
111
+ mov.b32 %r5, 0;
112
+ .loc 1 79 18 // l2norm.py:79:18
113
+ // begin inline asm
114
+ mov.u32 %r1, %r5;
115
+ mov.u32 %r2, %r5;
116
+ mov.u32 %r3, %r5;
117
+ mov.u32 %r4, %r5;
118
+ @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ];
119
+ // end inline asm
120
+ mov.b32 {%rs1, %rs2}, %r1;
121
+ mov.b32 {%rs3, %rs4}, %r2;
122
+ mov.b32 {%rs5, %rs6}, %r3;
123
+ mov.b32 {%rs7, %rs8}, %r4;
124
+ // begin inline asm
125
+ mov.u32 %r6, %r5;
126
+ mov.u32 %r7, %r5;
127
+ mov.u32 %r8, %r5;
128
+ mov.u32 %r9, %r5;
129
+ @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ];
130
+ // end inline asm
131
+ mov.b32 {%rs9, %rs10}, %r6;
132
+ mov.b32 {%rs11, %rs12}, %r7;
133
+ mov.b32 {%rs13, %rs14}, %r8;
134
+ mov.b32 {%rs15, %rs16}, %r9;
135
+ // begin inline asm
136
+ mov.u32 %r10, %r5;
137
+ mov.u32 %r11, %r5;
138
+ mov.u32 %r12, %r5;
139
+ mov.u32 %r13, %r5;
140
+ @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ];
141
+ // end inline asm
142
+ mov.b32 {%rs17, %rs18}, %r10;
143
+ mov.b32 {%rs19, %rs20}, %r11;
144
+ mov.b32 {%rs21, %rs22}, %r12;
145
+ mov.b32 {%rs23, %rs24}, %r13;
146
+ // begin inline asm
147
+ mov.u32 %r14, %r5;
148
+ mov.u32 %r15, %r5;
149
+ mov.u32 %r16, %r5;
150
+ mov.u32 %r17, %r5;
151
+ @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ];
152
+ // end inline asm
153
+ mov.b32 {%rs25, %rs26}, %r14;
154
+ mov.b32 {%rs27, %rs28}, %r15;
155
+ mov.b32 {%rs29, %rs30}, %r16;
156
+ mov.b32 {%rs31, %rs32}, %r17;
157
+ // begin inline asm
158
+ mov.u32 %r18, %r5;
159
+ mov.u32 %r19, %r5;
160
+ mov.u32 %r20, %r5;
161
+ mov.u32 %r21, %r5;
162
+ @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ];
163
+ // end inline asm
164
+ mov.b32 {%rs33, %rs34}, %r18;
165
+ mov.b32 {%rs35, %rs36}, %r19;
166
+ mov.b32 {%rs37, %rs38}, %r20;
167
+ mov.b32 {%rs39, %rs40}, %r21;
168
+ // begin inline asm
169
+ mov.u32 %r22, %r5;
170
+ mov.u32 %r23, %r5;
171
+ mov.u32 %r24, %r5;
172
+ mov.u32 %r25, %r5;
173
+ @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ];
174
+ // end inline asm
175
+ mov.b32 {%rs41, %rs42}, %r22;
176
+ mov.b32 {%rs43, %rs44}, %r23;
177
+ mov.b32 {%rs45, %rs46}, %r24;
178
+ mov.b32 {%rs47, %rs48}, %r25;
179
+ // begin inline asm
180
+ mov.u32 %r26, %r5;
181
+ mov.u32 %r27, %r5;
182
+ mov.u32 %r28, %r5;
183
+ mov.u32 %r29, %r5;
184
+ @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ];
185
+ // end inline asm
186
+ mov.b32 {%rs49, %rs50}, %r26;
187
+ mov.b32 {%rs51, %rs52}, %r27;
188
+ mov.b32 {%rs53, %rs54}, %r28;
189
+ mov.b32 {%rs55, %rs56}, %r29;
190
+ // begin inline asm
191
+ mov.u32 %r30, %r5;
192
+ mov.u32 %r31, %r5;
193
+ mov.u32 %r32, %r5;
194
+ mov.u32 %r33, %r5;
195
+ @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ];
196
+ // end inline asm
197
+ mov.b32 {%rs57, %rs58}, %r30;
198
+ mov.b32 {%rs59, %rs60}, %r31;
199
+ mov.b32 {%rs61, %rs62}, %r32;
200
+ mov.b32 {%rs63, %rs64}, %r33;
201
+ .loc 1 79 47 // l2norm.py:79:47
202
+ cvt.f32.bf16 %r91, %rs1;
203
+ cvt.f32.bf16 %r92, %rs2;
204
+ cvt.f32.bf16 %r93, %rs3;
205
+ cvt.f32.bf16 %r94, %rs4;
206
+ cvt.f32.bf16 %r95, %rs5;
207
+ cvt.f32.bf16 %r96, %rs6;
208
+ cvt.f32.bf16 %r97, %rs7;
209
+ cvt.f32.bf16 %r98, %rs8;
210
+ cvt.f32.bf16 %r99, %rs9;
211
+ cvt.f32.bf16 %r100, %rs10;
212
+ cvt.f32.bf16 %r101, %rs11;
213
+ cvt.f32.bf16 %r102, %rs12;
214
+ cvt.f32.bf16 %r103, %rs13;
215
+ cvt.f32.bf16 %r104, %rs14;
216
+ cvt.f32.bf16 %r105, %rs15;
217
+ cvt.f32.bf16 %r106, %rs16;
218
+ cvt.f32.bf16 %r107, %rs17;
219
+ cvt.f32.bf16 %r108, %rs18;
220
+ cvt.f32.bf16 %r109, %rs19;
221
+ cvt.f32.bf16 %r110, %rs20;
222
+ cvt.f32.bf16 %r111, %rs21;
223
+ cvt.f32.bf16 %r112, %rs22;
224
+ cvt.f32.bf16 %r113, %rs23;
225
+ cvt.f32.bf16 %r114, %rs24;
226
+ cvt.f32.bf16 %r115, %rs25;
227
+ cvt.f32.bf16 %r116, %rs26;
228
+ cvt.f32.bf16 %r117, %rs27;
229
+ cvt.f32.bf16 %r118, %rs28;
230
+ cvt.f32.bf16 %r119, %rs29;
231
+ cvt.f32.bf16 %r120, %rs30;
232
+ cvt.f32.bf16 %r121, %rs31;
233
+ cvt.f32.bf16 %r122, %rs32;
234
+ cvt.f32.bf16 %r123, %rs33;
235
+ cvt.f32.bf16 %r124, %rs34;
236
+ cvt.f32.bf16 %r125, %rs35;
237
+ cvt.f32.bf16 %r126, %rs36;
238
+ cvt.f32.bf16 %r127, %rs37;
239
+ cvt.f32.bf16 %r128, %rs38;
240
+ cvt.f32.bf16 %r129, %rs39;
241
+ cvt.f32.bf16 %r130, %rs40;
242
+ cvt.f32.bf16 %r131, %rs41;
243
+ cvt.f32.bf16 %r132, %rs42;
244
+ cvt.f32.bf16 %r133, %rs43;
245
+ cvt.f32.bf16 %r134, %rs44;
246
+ cvt.f32.bf16 %r135, %rs45;
247
+ cvt.f32.bf16 %r136, %rs46;
248
+ cvt.f32.bf16 %r137, %rs47;
249
+ cvt.f32.bf16 %r138, %rs48;
250
+ cvt.f32.bf16 %r139, %rs49;
251
+ cvt.f32.bf16 %r140, %rs50;
252
+ cvt.f32.bf16 %r141, %rs51;
253
+ cvt.f32.bf16 %r142, %rs52;
254
+ cvt.f32.bf16 %r143, %rs53;
255
+ cvt.f32.bf16 %r144, %rs54;
256
+ cvt.f32.bf16 %r145, %rs55;
257
+ cvt.f32.bf16 %r146, %rs56;
258
+ cvt.f32.bf16 %r147, %rs57;
259
+ cvt.f32.bf16 %r148, %rs58;
260
+ cvt.f32.bf16 %r149, %rs59;
261
+ cvt.f32.bf16 %r150, %rs60;
262
+ cvt.f32.bf16 %r151, %rs61;
263
+ cvt.f32.bf16 %r152, %rs62;
264
+ cvt.f32.bf16 %r153, %rs63;
265
+ cvt.f32.bf16 %r154, %rs64;
266
+ .loc 1 80 38 // l2norm.py:80:38
267
+ mul.f32 %r155, %r92, %r92;
268
+ mul.f32 %r156, %r100, %r100;
269
+ mul.f32 %r157, %r108, %r108;
270
+ mul.f32 %r158, %r116, %r116;
271
+ mul.f32 %r159, %r124, %r124;
272
+ mul.f32 %r160, %r132, %r132;
273
+ mul.f32 %r161, %r140, %r140;
274
+ mul.f32 %r162, %r148, %r148;
275
+ $L__tmp1:
276
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
277
+ fma.rn.f32.bf16 %r163, %rs1, %rs1, %r155;
278
+ fma.rn.f32.bf16 %r164, %rs3, %rs3, %r163;
279
+ fma.rn.f32.bf16 %r165, %rs4, %rs4, %r164;
280
+ fma.rn.f32.bf16 %r166, %rs5, %rs5, %r165;
281
+ fma.rn.f32.bf16 %r167, %rs6, %rs6, %r166;
282
+ fma.rn.f32.bf16 %r168, %rs7, %rs7, %r167;
283
+ fma.rn.f32.bf16 %r169, %rs8, %rs8, %r168;
284
+ fma.rn.f32.bf16 %r170, %rs9, %rs9, %r156;
285
+ fma.rn.f32.bf16 %r171, %rs11, %rs11, %r170;
286
+ fma.rn.f32.bf16 %r172, %rs12, %rs12, %r171;
287
+ fma.rn.f32.bf16 %r173, %rs13, %rs13, %r172;
288
+ fma.rn.f32.bf16 %r174, %rs14, %rs14, %r173;
289
+ fma.rn.f32.bf16 %r175, %rs15, %rs15, %r174;
290
+ fma.rn.f32.bf16 %r176, %rs16, %rs16, %r175;
291
+ fma.rn.f32.bf16 %r177, %rs17, %rs17, %r157;
292
+ fma.rn.f32.bf16 %r178, %rs19, %rs19, %r177;
293
+ fma.rn.f32.bf16 %r179, %rs20, %rs20, %r178;
294
+ fma.rn.f32.bf16 %r180, %rs21, %rs21, %r179;
295
+ fma.rn.f32.bf16 %r181, %rs22, %rs22, %r180;
296
+ fma.rn.f32.bf16 %r182, %rs23, %rs23, %r181;
297
+ fma.rn.f32.bf16 %r183, %rs24, %rs24, %r182;
298
+ fma.rn.f32.bf16 %r184, %rs25, %rs25, %r158;
299
+ fma.rn.f32.bf16 %r185, %rs27, %rs27, %r184;
300
+ fma.rn.f32.bf16 %r186, %rs28, %rs28, %r185;
301
+ fma.rn.f32.bf16 %r187, %rs29, %rs29, %r186;
302
+ fma.rn.f32.bf16 %r188, %rs30, %rs30, %r187;
303
+ fma.rn.f32.bf16 %r189, %rs31, %rs31, %r188;
304
+ fma.rn.f32.bf16 %r190, %rs32, %rs32, %r189;
305
+ fma.rn.f32.bf16 %r191, %rs33, %rs33, %r159;
306
+ fma.rn.f32.bf16 %r192, %rs35, %rs35, %r191;
307
+ fma.rn.f32.bf16 %r193, %rs36, %rs36, %r192;
308
+ fma.rn.f32.bf16 %r194, %rs37, %rs37, %r193;
309
+ fma.rn.f32.bf16 %r195, %rs38, %rs38, %r194;
310
+ fma.rn.f32.bf16 %r196, %rs39, %rs39, %r195;
311
+ fma.rn.f32.bf16 %r197, %rs40, %rs40, %r196;
312
+ fma.rn.f32.bf16 %r198, %rs41, %rs41, %r160;
313
+ fma.rn.f32.bf16 %r199, %rs43, %rs43, %r198;
314
+ fma.rn.f32.bf16 %r200, %rs44, %rs44, %r199;
315
+ fma.rn.f32.bf16 %r201, %rs45, %rs45, %r200;
316
+ fma.rn.f32.bf16 %r202, %rs46, %rs46, %r201;
317
+ fma.rn.f32.bf16 %r203, %rs47, %rs47, %r202;
318
+ fma.rn.f32.bf16 %r204, %rs48, %rs48, %r203;
319
+ fma.rn.f32.bf16 %r205, %rs49, %rs49, %r161;
320
+ fma.rn.f32.bf16 %r206, %rs51, %rs51, %r205;
321
+ fma.rn.f32.bf16 %r207, %rs52, %rs52, %r206;
322
+ fma.rn.f32.bf16 %r208, %rs53, %rs53, %r207;
323
+ fma.rn.f32.bf16 %r209, %rs54, %rs54, %r208;
324
+ fma.rn.f32.bf16 %r210, %rs55, %rs55, %r209;
325
+ fma.rn.f32.bf16 %r211, %rs56, %rs56, %r210;
326
+ fma.rn.f32.bf16 %r212, %rs57, %rs57, %r162;
327
+ fma.rn.f32.bf16 %r213, %rs59, %rs59, %r212;
328
+ fma.rn.f32.bf16 %r214, %rs60, %rs60, %r213;
329
+ fma.rn.f32.bf16 %r215, %rs61, %rs61, %r214;
330
+ fma.rn.f32.bf16 %r216, %rs62, %rs62, %r215;
331
+ fma.rn.f32.bf16 %r217, %rs63, %rs63, %r216;
332
+ fma.rn.f32.bf16 %r218, %rs64, %rs64, %r217;
333
+ $L__tmp2:
334
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
335
+ shfl.sync.bfly.b32 %r219, %r169, 8, 31, -1;
336
+ $L__tmp3:
337
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
338
+ add.f32 %r220, %r169, %r219;
339
+ $L__tmp4:
340
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
341
+ shfl.sync.bfly.b32 %r221, %r220, 4, 31, -1;
342
+ $L__tmp5:
343
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
344
+ add.f32 %r222, %r220, %r221;
345
+ $L__tmp6:
346
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
347
+ shfl.sync.bfly.b32 %r223, %r222, 2, 31, -1;
348
+ $L__tmp7:
349
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
350
+ add.f32 %r224, %r222, %r223;
351
+ $L__tmp8:
352
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
353
+ shfl.sync.bfly.b32 %r225, %r224, 1, 31, -1;
354
+ $L__tmp9:
355
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
356
+ add.f32 %r226, %r224, %r225;
357
+ $L__tmp10:
358
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
359
+ shfl.sync.bfly.b32 %r227, %r176, 8, 31, -1;
360
+ $L__tmp11:
361
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
362
+ add.f32 %r228, %r176, %r227;
363
+ $L__tmp12:
364
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
365
+ shfl.sync.bfly.b32 %r229, %r228, 4, 31, -1;
366
+ $L__tmp13:
367
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
368
+ add.f32 %r230, %r228, %r229;
369
+ $L__tmp14:
370
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
371
+ shfl.sync.bfly.b32 %r231, %r230, 2, 31, -1;
372
+ $L__tmp15:
373
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
374
+ add.f32 %r232, %r230, %r231;
375
+ $L__tmp16:
376
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
377
+ shfl.sync.bfly.b32 %r233, %r232, 1, 31, -1;
378
+ $L__tmp17:
379
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
380
+ add.f32 %r234, %r232, %r233;
381
+ $L__tmp18:
382
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
383
+ shfl.sync.bfly.b32 %r235, %r183, 8, 31, -1;
384
+ $L__tmp19:
385
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
386
+ add.f32 %r236, %r183, %r235;
387
+ $L__tmp20:
388
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
389
+ shfl.sync.bfly.b32 %r237, %r236, 4, 31, -1;
390
+ $L__tmp21:
391
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
392
+ add.f32 %r238, %r236, %r237;
393
+ $L__tmp22:
394
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
395
+ shfl.sync.bfly.b32 %r239, %r238, 2, 31, -1;
396
+ $L__tmp23:
397
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
398
+ add.f32 %r240, %r238, %r239;
399
+ $L__tmp24:
400
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
401
+ shfl.sync.bfly.b32 %r241, %r240, 1, 31, -1;
402
+ $L__tmp25:
403
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
404
+ add.f32 %r242, %r240, %r241;
405
+ $L__tmp26:
406
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
407
+ shfl.sync.bfly.b32 %r243, %r190, 8, 31, -1;
408
+ $L__tmp27:
409
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
410
+ add.f32 %r244, %r190, %r243;
411
+ $L__tmp28:
412
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
413
+ shfl.sync.bfly.b32 %r245, %r244, 4, 31, -1;
414
+ $L__tmp29:
415
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
416
+ add.f32 %r246, %r244, %r245;
417
+ $L__tmp30:
418
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
419
+ shfl.sync.bfly.b32 %r247, %r246, 2, 31, -1;
420
+ $L__tmp31:
421
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
422
+ add.f32 %r248, %r246, %r247;
423
+ $L__tmp32:
424
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
425
+ shfl.sync.bfly.b32 %r249, %r248, 1, 31, -1;
426
+ $L__tmp33:
427
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
428
+ add.f32 %r250, %r248, %r249;
429
+ $L__tmp34:
430
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
431
+ shfl.sync.bfly.b32 %r251, %r197, 8, 31, -1;
432
+ $L__tmp35:
433
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
434
+ add.f32 %r252, %r197, %r251;
435
+ $L__tmp36:
436
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
437
+ shfl.sync.bfly.b32 %r253, %r252, 4, 31, -1;
438
+ $L__tmp37:
439
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
440
+ add.f32 %r254, %r252, %r253;
441
+ $L__tmp38:
442
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
443
+ shfl.sync.bfly.b32 %r255, %r254, 2, 31, -1;
444
+ $L__tmp39:
445
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
446
+ add.f32 %r256, %r254, %r255;
447
+ $L__tmp40:
448
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
449
+ shfl.sync.bfly.b32 %r257, %r256, 1, 31, -1;
450
+ $L__tmp41:
451
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
452
+ add.f32 %r258, %r256, %r257;
453
+ $L__tmp42:
454
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
455
+ shfl.sync.bfly.b32 %r259, %r204, 8, 31, -1;
456
+ $L__tmp43:
457
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
458
+ add.f32 %r260, %r204, %r259;
459
+ $L__tmp44:
460
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
461
+ shfl.sync.bfly.b32 %r261, %r260, 4, 31, -1;
462
+ $L__tmp45:
463
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
464
+ add.f32 %r262, %r260, %r261;
465
+ $L__tmp46:
466
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
467
+ shfl.sync.bfly.b32 %r263, %r262, 2, 31, -1;
468
+ $L__tmp47:
469
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
470
+ add.f32 %r264, %r262, %r263;
471
+ $L__tmp48:
472
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
473
+ shfl.sync.bfly.b32 %r265, %r264, 1, 31, -1;
474
+ $L__tmp49:
475
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
476
+ add.f32 %r266, %r264, %r265;
477
+ $L__tmp50:
478
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
479
+ shfl.sync.bfly.b32 %r267, %r211, 8, 31, -1;
480
+ $L__tmp51:
481
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
482
+ add.f32 %r268, %r211, %r267;
483
+ $L__tmp52:
484
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
485
+ shfl.sync.bfly.b32 %r269, %r268, 4, 31, -1;
486
+ $L__tmp53:
487
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
488
+ add.f32 %r270, %r268, %r269;
489
+ $L__tmp54:
490
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
491
+ shfl.sync.bfly.b32 %r271, %r270, 2, 31, -1;
492
+ $L__tmp55:
493
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
494
+ add.f32 %r272, %r270, %r271;
495
+ $L__tmp56:
496
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
497
+ shfl.sync.bfly.b32 %r273, %r272, 1, 31, -1;
498
+ $L__tmp57:
499
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
500
+ add.f32 %r274, %r272, %r273;
501
+ $L__tmp58:
502
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
503
+ shfl.sync.bfly.b32 %r275, %r218, 8, 31, -1;
504
+ $L__tmp59:
505
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
506
+ add.f32 %r276, %r218, %r275;
507
+ $L__tmp60:
508
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
509
+ shfl.sync.bfly.b32 %r277, %r276, 4, 31, -1;
510
+ $L__tmp61:
511
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
512
+ add.f32 %r278, %r276, %r277;
513
+ $L__tmp62:
514
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
515
+ shfl.sync.bfly.b32 %r279, %r278, 2, 31, -1;
516
+ $L__tmp63:
517
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
518
+ add.f32 %r280, %r278, %r279;
519
+ $L__tmp64:
520
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:80:43 ]
521
+ shfl.sync.bfly.b32 %r281, %r280, 1, 31, -1;
522
+ $L__tmp65:
523
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:80:43 ] ]
524
+ add.f32 %r282, %r280, %r281;
525
+ $L__tmp66:
526
+ .loc 1 80 48 // l2norm.py:80:48
527
+ add.f32 %r283, %r78, %r226;
528
+ add.f32 %r284, %r78, %r234;
529
+ add.f32 %r285, %r78, %r242;
530
+ add.f32 %r286, %r78, %r250;
531
+ add.f32 %r287, %r78, %r258;
532
+ add.f32 %r288, %r78, %r266;
533
+ add.f32 %r289, %r78, %r274;
534
+ add.f32 %r290, %r78, %r282;
535
+ .loc 1 80 25 // l2norm.py:80:25
536
+ sqrt.approx.ftz.f32 %r291, %r283;
537
+ sqrt.approx.ftz.f32 %r292, %r284;
538
+ sqrt.approx.ftz.f32 %r293, %r285;
539
+ sqrt.approx.ftz.f32 %r294, %r286;
540
+ sqrt.approx.ftz.f32 %r295, %r287;
541
+ sqrt.approx.ftz.f32 %r296, %r288;
542
+ sqrt.approx.ftz.f32 %r297, %r289;
543
+ sqrt.approx.ftz.f32 %r298, %r290;
544
+ .loc 1 71 34 // l2norm.py:71:34
545
+ and.b32 %r299, %r79, 31;
546
+ .loc 1 71 21 // l2norm.py:71:21
547
+ cvt.u64.u32 %rd56, %r299;
548
+ or.b64 %rd57, %rd21, %rd56;
549
+ .loc 1 77 20 // l2norm.py:77:20
550
+ shl.b64 %rd58, %rd57, 2;
551
+ add.s64 %rd17, %rd20, %rd58;
552
+ .loc 1 76 14 // l2norm.py:76:14
553
+ add.s64 %rd59, %rd19, %rd53;
554
+ .loc 1 76 33 // l2norm.py:76:33
555
+ add.s64 %rd16, %rd59, %rd55;
556
+ .loc 1 76 14 // l2norm.py:76:14
557
+ add.s64 %rd60, %rd19, %rd51;
558
+ .loc 1 76 33 // l2norm.py:76:33
559
+ add.s64 %rd15, %rd60, %rd55;
560
+ .loc 1 76 14 // l2norm.py:76:14
561
+ add.s64 %rd61, %rd19, %rd49;
562
+ .loc 1 76 33 // l2norm.py:76:33
563
+ add.s64 %rd14, %rd61, %rd55;
564
+ .loc 1 76 14 // l2norm.py:76:14
565
+ add.s64 %rd62, %rd19, %rd47;
566
+ .loc 1 76 33 // l2norm.py:76:33
567
+ add.s64 %rd13, %rd62, %rd55;
568
+ .loc 1 76 14 // l2norm.py:76:14
569
+ add.s64 %rd63, %rd19, %rd45;
570
+ .loc 1 76 33 // l2norm.py:76:33
571
+ add.s64 %rd12, %rd63, %rd55;
572
+ .loc 1 76 14 // l2norm.py:76:14
573
+ add.s64 %rd64, %rd19, %rd43;
574
+ .loc 1 76 33 // l2norm.py:76:33
575
+ add.s64 %rd11, %rd64, %rd55;
576
+ .loc 1 76 14 // l2norm.py:76:14
577
+ add.s64 %rd65, %rd19, %rd41;
578
+ .loc 1 76 33 // l2norm.py:76:33
579
+ add.s64 %rd10, %rd65, %rd55;
580
+ .loc 1 76 14 // l2norm.py:76:14
581
+ add.s64 %rd66, %rd19, %rd39;
582
+ .loc 1 76 33 // l2norm.py:76:33
583
+ add.s64 %rd9, %rd66, %rd55;
584
+ .loc 1 73 16 // l2norm.py:73:16
585
+ setp.lt.s64 %p10, %rd57, %rd22;
586
+ mov.b32 %r300, 0f3F800000;
587
+ .loc 1 80 17 // l2norm.py:80:17
588
+ div.full.f32 %r67, %r300, %r291;
589
+ div.full.f32 %r68, %r300, %r292;
590
+ div.full.f32 %r69, %r300, %r293;
591
+ div.full.f32 %r70, %r300, %r294;
592
+ div.full.f32 %r72, %r300, %r295;
593
+ div.full.f32 %r73, %r300, %r296;
594
+ div.full.f32 %r74, %r300, %r297;
595
+ div.full.f32 %r75, %r300, %r298;
596
+ .loc 1 81 16 // l2norm.py:81:16
597
+ mul.f32 %r301, %r67, %r91;
598
+ mul.f32 %r302, %r67, %r92;
599
+ mul.f32 %r303, %r67, %r93;
600
+ mul.f32 %r304, %r67, %r94;
601
+ mul.f32 %r305, %r67, %r95;
602
+ mul.f32 %r306, %r67, %r96;
603
+ mul.f32 %r307, %r67, %r97;
604
+ mul.f32 %r308, %r67, %r98;
605
+ mul.f32 %r309, %r68, %r99;
606
+ mul.f32 %r310, %r68, %r100;
607
+ mul.f32 %r311, %r68, %r101;
608
+ mul.f32 %r312, %r68, %r102;
609
+ mul.f32 %r313, %r68, %r103;
610
+ mul.f32 %r314, %r68, %r104;
611
+ mul.f32 %r315, %r68, %r105;
612
+ mul.f32 %r316, %r68, %r106;
613
+ mul.f32 %r317, %r69, %r107;
614
+ mul.f32 %r318, %r69, %r108;
615
+ mul.f32 %r319, %r69, %r109;
616
+ mul.f32 %r320, %r69, %r110;
617
+ mul.f32 %r321, %r69, %r111;
618
+ mul.f32 %r322, %r69, %r112;
619
+ mul.f32 %r323, %r69, %r113;
620
+ mul.f32 %r324, %r69, %r114;
621
+ mul.f32 %r325, %r70, %r115;
622
+ mul.f32 %r326, %r70, %r116;
623
+ mul.f32 %r327, %r70, %r117;
624
+ mul.f32 %r328, %r70, %r118;
625
+ mul.f32 %r329, %r70, %r119;
626
+ mul.f32 %r330, %r70, %r120;
627
+ mul.f32 %r331, %r70, %r121;
628
+ mul.f32 %r332, %r70, %r122;
629
+ mul.f32 %r333, %r72, %r123;
630
+ mul.f32 %r334, %r72, %r124;
631
+ mul.f32 %r335, %r72, %r125;
632
+ mul.f32 %r336, %r72, %r126;
633
+ mul.f32 %r337, %r72, %r127;
634
+ mul.f32 %r338, %r72, %r128;
635
+ mul.f32 %r339, %r72, %r129;
636
+ mul.f32 %r340, %r72, %r130;
637
+ mul.f32 %r341, %r73, %r131;
638
+ mul.f32 %r342, %r73, %r132;
639
+ mul.f32 %r343, %r73, %r133;
640
+ mul.f32 %r344, %r73, %r134;
641
+ mul.f32 %r345, %r73, %r135;
642
+ mul.f32 %r346, %r73, %r136;
643
+ mul.f32 %r347, %r73, %r137;
644
+ mul.f32 %r348, %r73, %r138;
645
+ mul.f32 %r349, %r74, %r139;
646
+ mul.f32 %r350, %r74, %r140;
647
+ mul.f32 %r351, %r74, %r141;
648
+ mul.f32 %r352, %r74, %r142;
649
+ mul.f32 %r353, %r74, %r143;
650
+ mul.f32 %r354, %r74, %r144;
651
+ mul.f32 %r355, %r74, %r145;
652
+ mul.f32 %r356, %r74, %r146;
653
+ mul.f32 %r357, %r75, %r147;
654
+ mul.f32 %r358, %r75, %r148;
655
+ mul.f32 %r359, %r75, %r149;
656
+ mul.f32 %r360, %r75, %r150;
657
+ mul.f32 %r361, %r75, %r151;
658
+ mul.f32 %r362, %r75, %r152;
659
+ mul.f32 %r363, %r75, %r153;
660
+ mul.f32 %r364, %r75, %r154;
661
+ .loc 1 83 18 // l2norm.py:83:18
662
+ cvt.rn.bf16x2.f32 %r34, %r302, %r301;
663
+ cvt.rn.bf16x2.f32 %r35, %r304, %r303;
664
+ cvt.rn.bf16x2.f32 %r36, %r306, %r305;
665
+ cvt.rn.bf16x2.f32 %r37, %r308, %r307;
666
+ // begin inline asm
667
+ @%p1 st.global.v4.b32 [ %rd9 + 0 ], { %r34, %r35, %r36, %r37 };
668
+ // end inline asm
669
+ cvt.rn.bf16x2.f32 %r38, %r310, %r309;
670
+ cvt.rn.bf16x2.f32 %r39, %r312, %r311;
671
+ cvt.rn.bf16x2.f32 %r40, %r314, %r313;
672
+ cvt.rn.bf16x2.f32 %r41, %r316, %r315;
673
+ // begin inline asm
674
+ @%p2 st.global.v4.b32 [ %rd10 + 0 ], { %r38, %r39, %r40, %r41 };
675
+ // end inline asm
676
+ cvt.rn.bf16x2.f32 %r42, %r318, %r317;
677
+ cvt.rn.bf16x2.f32 %r43, %r320, %r319;
678
+ cvt.rn.bf16x2.f32 %r44, %r322, %r321;
679
+ cvt.rn.bf16x2.f32 %r45, %r324, %r323;
680
+ // begin inline asm
681
+ @%p3 st.global.v4.b32 [ %rd11 + 0 ], { %r42, %r43, %r44, %r45 };
682
+ // end inline asm
683
+ cvt.rn.bf16x2.f32 %r46, %r326, %r325;
684
+ cvt.rn.bf16x2.f32 %r47, %r328, %r327;
685
+ cvt.rn.bf16x2.f32 %r48, %r330, %r329;
686
+ cvt.rn.bf16x2.f32 %r49, %r332, %r331;
687
+ // begin inline asm
688
+ @%p4 st.global.v4.b32 [ %rd12 + 0 ], { %r46, %r47, %r48, %r49 };
689
+ // end inline asm
690
+ cvt.rn.bf16x2.f32 %r50, %r334, %r333;
691
+ cvt.rn.bf16x2.f32 %r51, %r336, %r335;
692
+ cvt.rn.bf16x2.f32 %r52, %r338, %r337;
693
+ cvt.rn.bf16x2.f32 %r53, %r340, %r339;
694
+ // begin inline asm
695
+ @%p5 st.global.v4.b32 [ %rd13 + 0 ], { %r50, %r51, %r52, %r53 };
696
+ // end inline asm
697
+ cvt.rn.bf16x2.f32 %r54, %r342, %r341;
698
+ cvt.rn.bf16x2.f32 %r55, %r344, %r343;
699
+ cvt.rn.bf16x2.f32 %r56, %r346, %r345;
700
+ cvt.rn.bf16x2.f32 %r57, %r348, %r347;
701
+ // begin inline asm
702
+ @%p6 st.global.v4.b32 [ %rd14 + 0 ], { %r54, %r55, %r56, %r57 };
703
+ // end inline asm
704
+ cvt.rn.bf16x2.f32 %r58, %r350, %r349;
705
+ cvt.rn.bf16x2.f32 %r59, %r352, %r351;
706
+ cvt.rn.bf16x2.f32 %r60, %r354, %r353;
707
+ cvt.rn.bf16x2.f32 %r61, %r356, %r355;
708
+ // begin inline asm
709
+ @%p7 st.global.v4.b32 [ %rd15 + 0 ], { %r58, %r59, %r60, %r61 };
710
+ // end inline asm
711
+ cvt.rn.bf16x2.f32 %r62, %r358, %r357;
712
+ cvt.rn.bf16x2.f32 %r63, %r360, %r359;
713
+ cvt.rn.bf16x2.f32 %r64, %r362, %r361;
714
+ cvt.rn.bf16x2.f32 %r65, %r364, %r363;
715
+ // begin inline asm
716
+ @%p8 st.global.v4.b32 [ %rd16 + 0 ], { %r62, %r63, %r64, %r65 };
717
+ // end inline asm
718
+ .loc 1 84 21 // l2norm.py:84:21
719
+ shl.b32 %r365, %r80, 1;
720
+ mov.b32 %r366, global_smem;
721
+ add.s32 %r66, %r366, %r365;
722
+ // begin inline asm
723
+ st.shared.v4.b32 [ %r66 + 0 ], { %r67, %r68, %r69, %r70 };
724
+ // end inline asm
725
+ add.s32 %r71, %r66, 16;
726
+ // begin inline asm
727
+ st.shared.v4.b32 [ %r71 + 0 ], { %r72, %r73, %r74, %r75 };
728
+ // end inline asm
729
+ bar.sync 0;
730
+ shl.b32 %r367, %r79, 5;
731
+ and.b32 %r368, %r367, 96;
732
+ and.b32 %r369, %r79, 28;
733
+ add.s32 %r370, %r366, %r368;
734
+ add.s32 %r371, %r370, %r369;
735
+ ld.shared.b32 %r76, [%r371];
736
+ and.b32 %r372, %r79, 32;
737
+ setp.eq.b32 %p11, %r372, 0;
738
+ and.pred %p9, %p11, %p10;
739
+ // begin inline asm
740
+ @%p9 st.global.b32 [ %rd17 + 0 ], { %r76 };
741
+ // end inline asm
742
+ .loc 1 84 4 // l2norm.py:84:4
743
+ ret;
744
+ $L__tmp67:
745
+ $L__func_end0:
746
+ // -- End function
747
+ }
748
+ .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py"
749
+ .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py"
750
+ .section .debug_abbrev
751
+ {
752
+ .b8 1 // Abbreviation Code
753
+ .b8 17 // DW_TAG_compile_unit
754
+ .b8 1 // DW_CHILDREN_yes
755
+ .b8 37 // DW_AT_producer
756
+ .b8 8 // DW_FORM_string
757
+ .b8 19 // DW_AT_language
758
+ .b8 5 // DW_FORM_data2
759
+ .b8 3 // DW_AT_name
760
+ .b8 8 // DW_FORM_string
761
+ .b8 16 // DW_AT_stmt_list
762
+ .b8 6 // DW_FORM_data4
763
+ .b8 27 // DW_AT_comp_dir
764
+ .b8 8 // DW_FORM_string
765
+ .b8 0 // EOM(1)
766
+ .b8 0 // EOM(2)
767
+ .b8 2 // Abbreviation Code
768
+ .b8 46 // DW_TAG_subprogram
769
+ .b8 0 // DW_CHILDREN_no
770
+ .b8 3 // DW_AT_name
771
+ .b8 8 // DW_FORM_string
772
+ .b8 32 // DW_AT_inline
773
+ .b8 11 // DW_FORM_data1
774
+ .b8 0 // EOM(1)
775
+ .b8 0 // EOM(2)
776
+ .b8 3 // Abbreviation Code
777
+ .b8 46 // DW_TAG_subprogram
778
+ .b8 1 // DW_CHILDREN_yes
779
+ .b8 17 // DW_AT_low_pc
780
+ .b8 1 // DW_FORM_addr
781
+ .b8 18 // DW_AT_high_pc
782
+ .b8 1 // DW_FORM_addr
783
+ .b8 49 // DW_AT_abstract_origin
784
+ .b8 19 // DW_FORM_ref4
785
+ .b8 0 // EOM(1)
786
+ .b8 0 // EOM(2)
787
+ .b8 4 // Abbreviation Code
788
+ .b8 29 // DW_TAG_inlined_subroutine
789
+ .b8 1 // DW_CHILDREN_yes
790
+ .b8 49 // DW_AT_abstract_origin
791
+ .b8 19 // DW_FORM_ref4
792
+ .b8 17 // DW_AT_low_pc
793
+ .b8 1 // DW_FORM_addr
794
+ .b8 18 // DW_AT_high_pc
795
+ .b8 1 // DW_FORM_addr
796
+ .b8 88 // DW_AT_call_file
797
+ .b8 11 // DW_FORM_data1
798
+ .b8 89 // DW_AT_call_line
799
+ .b8 11 // DW_FORM_data1
800
+ .b8 87 // DW_AT_call_column
801
+ .b8 11 // DW_FORM_data1
802
+ .b8 0 // EOM(1)
803
+ .b8 0 // EOM(2)
804
+ .b8 5 // Abbreviation Code
805
+ .b8 29 // DW_TAG_inlined_subroutine
806
+ .b8 0 // DW_CHILDREN_no
807
+ .b8 49 // DW_AT_abstract_origin
808
+ .b8 19 // DW_FORM_ref4
809
+ .b8 17 // DW_AT_low_pc
810
+ .b8 1 // DW_FORM_addr
811
+ .b8 18 // DW_AT_high_pc
812
+ .b8 1 // DW_FORM_addr
813
+ .b8 88 // DW_AT_call_file
814
+ .b8 11 // DW_FORM_data1
815
+ .b8 89 // DW_AT_call_line
816
+ .b8 5 // DW_FORM_data2
817
+ .b8 87 // DW_AT_call_column
818
+ .b8 11 // DW_FORM_data1
819
+ .b8 0 // EOM(1)
820
+ .b8 0 // EOM(2)
821
+ .b8 0 // EOM(3)
822
+ }
823
+ .section .debug_info
824
+ {
825
+ .b32 196 // Length of Unit
826
+ .b8 2 // DWARF version number
827
+ .b8 0
828
+ .b32 .debug_abbrev // Offset Into Abbrev. Section
829
+ .b8 8 // Address Size (in bytes)
830
+ .b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit
831
+ .b8 116 // DW_AT_producer
832
+ .b8 114
833
+ .b8 105
834
+ .b8 116
835
+ .b8 111
836
+ .b8 110
837
+ .b8 0
838
+ .b8 2 // DW_AT_language
839
+ .b8 0
840
+ .b8 108 // DW_AT_name
841
+ .b8 50
842
+ .b8 110
843
+ .b8 111
844
+ .b8 114
845
+ .b8 109
846
+ .b8 46
847
+ .b8 112
848
+ .b8 121
849
+ .b8 0
850
+ .b32 .debug_line // DW_AT_stmt_list
851
+ .b8 47 // DW_AT_comp_dir
852
+ .b8 104
853
+ .b8 111
854
+ .b8 109
855
+ .b8 101
856
+ .b8 47
857
+ .b8 115
858
+ .b8 105
859
+ .b8 108
860
+ .b8 118
861
+ .b8 101
862
+ .b8 114
863
+ .b8 47
864
+ .b8 97
865
+ .b8 102
866
+ .b8 102
867
+ .b8 105
868
+ .b8 110
869
+ .b8 101
870
+ .b8 47
871
+ .b8 46
872
+ .b8 118
873
+ .b8 101
874
+ .b8 110
875
+ .b8 118
876
+ .b8 47
877
+ .b8 108
878
+ .b8 105
879
+ .b8 98
880
+ .b8 47
881
+ .b8 112
882
+ .b8 121
883
+ .b8 116
884
+ .b8 104
885
+ .b8 111
886
+ .b8 110
887
+ .b8 51
888
+ .b8 46
889
+ .b8 49
890
+ .b8 50
891
+ .b8 47
892
+ .b8 115
893
+ .b8 105
894
+ .b8 116
895
+ .b8 101
896
+ .b8 45
897
+ .b8 112
898
+ .b8 97
899
+ .b8 99
900
+ .b8 107
901
+ .b8 97
902
+ .b8 103
903
+ .b8 101
904
+ .b8 115
905
+ .b8 47
906
+ .b8 102
907
+ .b8 108
908
+ .b8 97
909
+ .b8 47
910
+ .b8 109
911
+ .b8 111
912
+ .b8 100
913
+ .b8 117
914
+ .b8 108
915
+ .b8 101
916
+ .b8 115
917
+ .b8 47
918
+ .b8 110
919
+ .b8 111
920
+ .b8 114
921
+ .b8 109
922
+ .b8 0
923
+ .b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram
924
+ .b8 108 // DW_AT_name
925
+ .b8 50
926
+ .b8 110
927
+ .b8 111
928
+ .b8 114
929
+ .b8 109
930
+ .b8 95
931
+ .b8 102
932
+ .b8 119
933
+ .b8 100
934
+ .b8 95
935
+ .b8 107
936
+ .b8 101
937
+ .b8 114
938
+ .b8 110
939
+ .b8 101
940
+ .b8 108
941
+ .b8 0
942
+ .b8 1 // DW_AT_inline
943
+ .b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram
944
+ .b64 $L__func_begin0 // DW_AT_low_pc
945
+ .b64 $L__func_end0 // DW_AT_high_pc
946
+ .b32 107 // DW_AT_abstract_origin
947
+ .b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine
948
+ .b32 107 // DW_AT_abstract_origin
949
+ .b64 $L__tmp1 // DW_AT_low_pc
950
+ .b64 $L__tmp66 // DW_AT_high_pc
951
+ .b8 1 // DW_AT_call_file
952
+ .b8 80 // DW_AT_call_line
953
+ .b8 43 // DW_AT_call_column
954
+ .b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine
955
+ .b32 107 // DW_AT_abstract_origin
956
+ .b64 $L__tmp1 // DW_AT_low_pc
957
+ .b64 $L__tmp66 // DW_AT_high_pc
958
+ .b8 2 // DW_AT_call_file
959
+ .b8 37 // DW_AT_call_line
960
+ .b8 1
961
+ .b8 36 // DW_AT_call_column
962
+ .b8 0 // End Of Children Mark
963
+ .b8 0 // End Of Children Mark
964
+ .b8 0 // End Of Children Mark
965
+ }
966
+ .section .debug_macinfo { }
triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.source ADDED
@@ -0,0 +1,172 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
2
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0)
3
+ #loc38 = loc(unknown)
4
+ #loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0)
5
+ #loc45 = loc("x"(#loc))
6
+ #loc46 = loc("y"(#loc))
7
+ #loc47 = loc("rstd"(#loc))
8
+ #loc48 = loc("eps"(#loc))
9
+ #loc49 = loc("T"(#loc))
10
+ #loc81 = loc("input"(#loc36))
11
+ #loc82 = loc("a"(#loc41))
12
+ #loc83 = loc("b"(#loc41))
13
+ module {
14
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %i_t = tt.get_program_id x : i32 loc(#loc50)
16
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc51)
17
+ %o_t = arith.constant 32 : i32 loc(#loc52)
18
+ %o_t_1 = arith.constant 32 : i64 loc(#loc52)
19
+ %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc52)
20
+ %o_t_3 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc53)
21
+ %o_t_4 = arith.extsi %o_t_3 : tensor<32xi32> to tensor<32xi64> loc(#loc54)
22
+ %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<32xi64> loc(#loc54)
23
+ %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<32xi64> loc(#loc54)
24
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc55)
25
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc56)
26
+ %m_t_7 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc56)
27
+ %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<32xi64> loc(#loc56)
28
+ %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc57)
29
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc58)
30
+ %m_x_10 = arith.constant 128 : i32 loc(#loc59)
31
+ %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc59)
32
+ %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc59)
33
+ %m_x_13 = tt.broadcast %m_x : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc60)
34
+ %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc60)
35
+ %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<32x128xi1> loc(#loc60)
36
+ %p_x = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc61)
37
+ %p_x_16 = arith.constant 128 : i32 loc(#loc62)
38
+ %p_x_17 = arith.constant 128 : i64 loc(#loc62)
39
+ %p_x_18 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc62)
40
+ %p_x_19 = arith.muli %p_x, %p_x_18 : tensor<32x1xi64> loc(#loc62)
41
+ %p_x_20 = tt.splat %x : !tt.ptr<bf16> -> tensor<32x1x!tt.ptr<bf16>> loc(#loc63)
42
+ %p_x_21 = tt.addptr %p_x_20, %p_x_19 : tensor<32x1x!tt.ptr<bf16>>, tensor<32x1xi64> loc(#loc63)
43
+ %p_x_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc64)
44
+ %p_x_23 = tt.broadcast %p_x_21 : tensor<32x1x!tt.ptr<bf16>> -> tensor<32x128x!tt.ptr<bf16>> loc(#loc65)
45
+ %p_x_24 = tt.broadcast %p_x_22 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc65)
46
+ %p_x_25 = tt.addptr %p_x_23, %p_x_24 : tensor<32x128x!tt.ptr<bf16>>, tensor<32x128xi32> loc(#loc65)
47
+ %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc66)
48
+ %p_y_26 = arith.constant 128 : i32 loc(#loc67)
49
+ %p_y_27 = arith.constant 128 : i64 loc(#loc67)
50
+ %p_y_28 = arith.constant dense<128> : tensor<32x1xi64> loc(#loc67)
51
+ %p_y_29 = arith.muli %p_y, %p_y_28 : tensor<32x1xi64> loc(#loc67)
52
+ %p_y_30 = tt.splat %y : !tt.ptr<bf16> -> tensor<32x1x!tt.ptr<bf16>> loc(#loc68)
53
+ %p_y_31 = tt.addptr %p_y_30, %p_y_29 : tensor<32x1x!tt.ptr<bf16>>, tensor<32x1xi64> loc(#loc68)
54
+ %p_y_32 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc69)
55
+ %p_y_33 = tt.broadcast %p_y_31 : tensor<32x1x!tt.ptr<bf16>> -> tensor<32x128x!tt.ptr<bf16>> loc(#loc70)
56
+ %p_y_34 = tt.broadcast %p_y_32 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc70)
57
+ %p_y_35 = tt.addptr %p_y_33, %p_y_34 : tensor<32x128x!tt.ptr<bf16>>, tensor<32x128xi32> loc(#loc70)
58
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<32x!tt.ptr<f32>> loc(#loc71)
59
+ %p_rstd_36 = tt.addptr %p_rstd, %o_t_6 : tensor<32x!tt.ptr<f32>>, tensor<32xi64> loc(#loc71)
60
+ %b_x = arith.constant 0.000000e+00 : f32 loc(#loc72)
61
+ %b_x_37 = arith.constant dense<0.000000e+00> : tensor<32x128xf32> loc(#loc72)
62
+ %b_x_38 = arith.truncf %b_x_37 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc72)
63
+ %b_x_39 = tt.load %p_x_25, %m_x_15, %b_x_38 : tensor<32x128x!tt.ptr<bf16>> loc(#loc72)
64
+ %b_x_40 = arith.extf %b_x_39 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc73)
65
+ %b_rstd = arith.mulf %b_x_40, %b_x_40 : tensor<32x128xf32> loc(#loc74)
66
+ %b_rstd_41 = tt.call @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%b_rstd) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc75)
67
+ %b_rstd_42 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc76)
68
+ %b_rstd_43 = arith.addf %b_rstd_41, %b_rstd_42 : tensor<32xf32> loc(#loc76)
69
+ %b_rstd_44 = math.sqrt %b_rstd_43 : tensor<32xf32> loc(#loc77)
70
+ %b_rstd_45 = arith.constant 1 : i32 loc(#loc78)
71
+ %b_rstd_46 = arith.constant 1.000000e+00 : f32 loc(#loc78)
72
+ %b_rstd_47 = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc78)
73
+ %b_rstd_48 = arith.divf %b_rstd_47, %b_rstd_44 : tensor<32xf32> loc(#loc78)
74
+ %b_y = tt.expand_dims %b_rstd_48 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc79)
75
+ %b_y_49 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc80)
76
+ %b_y_50 = arith.mulf %b_x_40, %b_y_49 : tensor<32x128xf32> loc(#loc80)
77
+ %0 = arith.truncf %b_y_50 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc32)
78
+ tt.store %p_y_35, %0, %m_x_15 : tensor<32x128x!tt.ptr<bf16>> loc(#loc33)
79
+ tt.store %p_rstd_36, %b_rstd_48, %m_t_8 : tensor<32x!tt.ptr<f32>> loc(#loc34)
80
+ tt.return loc(#loc35)
81
+ } loc(#loc)
82
+ tt.func private @triton.language.standard.sum__fp32S32_128S_c1_cFalse_cNone(%input: tensor<32x128xf32> loc("input"(#loc36))) -> tensor<32xf32> attributes {noinline = false} {
83
+ %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({
84
+ ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)):
85
+ %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc37)
86
+ tt.reduce.return %2 : f32 loc(#loc37)
87
+ }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc37)
88
+ tt.return %0 : tensor<32xf32> loc(#loc39)
89
+ ^bb1: // no predecessors
90
+ %1 = ub.poison : tensor<32xf32> loc(#loc40)
91
+ tt.return %1 : tensor<32xf32> loc(#loc40)
92
+ } loc(#loc36)
93
+ tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc41)), %b: f32 loc("b"(#loc41))) -> f32 attributes {noinline = false} {
94
+ %0 = arith.addf %a, %b : f32 loc(#loc42)
95
+ tt.return %0 : f32 loc(#loc43)
96
+ ^bb1: // no predecessors
97
+ %1 = ub.poison : f32 loc(#loc44)
98
+ tt.return %1 : f32 loc(#loc44)
99
+ } loc(#loc41)
100
+ } loc(#loc)
101
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
102
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
103
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
104
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
105
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
106
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23)
107
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
108
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
109
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
110
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
111
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
112
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
113
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
114
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
115
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:37)
116
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
117
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:18)
118
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:29)
119
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
120
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:37)
121
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
122
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
123
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
124
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
125
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
126
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
127
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
128
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
129
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
130
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
131
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
132
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
133
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
134
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
135
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
136
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
137
+ #loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11)
138
+ #loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4)
139
+ #loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
140
+ #loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11)
141
+ #loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4)
142
+ #loc50 = loc("i_t"(#loc1))
143
+ #loc51 = loc("i_t"(#loc2))
144
+ #loc52 = loc("o_t"(#loc3))
145
+ #loc53 = loc("o_t"(#loc4))
146
+ #loc54 = loc("o_t"(#loc5))
147
+ #loc55 = loc("o_d"(#loc6))
148
+ #loc56 = loc("m_t"(#loc7))
149
+ #loc57 = loc("m_x"(#loc8))
150
+ #loc58 = loc("m_x"(#loc9))
151
+ #loc59 = loc("m_x"(#loc10))
152
+ #loc60 = loc("m_x"(#loc11))
153
+ #loc61 = loc("p_x"(#loc12))
154
+ #loc62 = loc("p_x"(#loc13))
155
+ #loc63 = loc("p_x"(#loc14))
156
+ #loc64 = loc("p_x"(#loc15))
157
+ #loc65 = loc("p_x"(#loc16))
158
+ #loc66 = loc("p_y"(#loc17))
159
+ #loc67 = loc("p_y"(#loc18))
160
+ #loc68 = loc("p_y"(#loc19))
161
+ #loc69 = loc("p_y"(#loc20))
162
+ #loc70 = loc("p_y"(#loc21))
163
+ #loc71 = loc("p_rstd"(#loc22))
164
+ #loc72 = loc("b_x"(#loc23))
165
+ #loc73 = loc("b_x"(#loc24))
166
+ #loc74 = loc("b_rstd"(#loc25))
167
+ #loc75 = loc("b_rstd"(#loc26))
168
+ #loc76 = loc("b_rstd"(#loc27))
169
+ #loc77 = loc("b_rstd"(#loc28))
170
+ #loc78 = loc("b_rstd"(#loc29))
171
+ #loc79 = loc("b_y"(#loc30))
172
+ #loc80 = loc("b_y"(#loc31))
triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttgir ADDED
@@ -0,0 +1,135 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}>
2
+ #blocked1 = #ttg.blocked<{sizePerThread = [1], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}>
3
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
4
+ #loc1 = loc(unknown)
5
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
6
+ #loc34 = loc("x"(#loc))
7
+ #loc35 = loc("y"(#loc))
8
+ #loc36 = loc("rstd"(#loc))
9
+ #loc37 = loc("eps"(#loc))
10
+ #loc38 = loc("T"(#loc))
11
+ #loc59 = loc("b_rstd"(#loc23))
12
+ #loc66 = loc(callsite(#loc1 at #loc59))
13
+ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} {
14
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %cst = arith.constant dense<1.000000e+00> : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1)
16
+ %cst_0 = arith.constant dense<0.000000e+00> : tensor<32x128xbf16, #blocked> loc(#loc1)
17
+ %c32_i64 = arith.constant 32 : i64 loc(#loc1)
18
+ %cst_1 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1)
19
+ %cst_2 = arith.constant dense<128> : tensor<32x1xi64, #blocked> loc(#loc1)
20
+ %i_t = tt.get_program_id x : i32 loc(#loc39)
21
+ %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc40)
22
+ %o_t = arith.muli %i_t_3, %c32_i64 : i64 loc(#loc41)
23
+ %o_t_4 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc42)
24
+ %o_t_5 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32, #blocked1> loc(#loc42)
25
+ %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)
26
+ %o_t_7 = arith.extsi %o_t_5 : tensor<32xi32, #blocked1> to tensor<32xi64, #blocked1> loc(#loc43)
27
+ %o_t_8 = tt.splat %o_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43)
28
+ %o_t_9 = tt.splat %o_t : i64 -> tensor<32xi64, #blocked1> loc(#loc43)
29
+ %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc43)
30
+ %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<32xi64, #blocked1> loc(#loc43)
31
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc44)
32
+ %m_t_12 = tt.splat %m_t : i64 -> tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44)
33
+ %m_t_13 = tt.splat %m_t : i64 -> tensor<32xi64, #blocked1> loc(#loc44)
34
+ %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc44)
35
+ %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<32xi64, #blocked1> loc(#loc44)
36
+ %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<32xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi1, #blocked> loc(#loc45)
37
+ %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc46)
38
+ %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)
39
+ %m_x_18 = arith.cmpi slt, %m_x_17, %cst_1 : tensor<1x128xi32, #blocked> loc(#loc47)
40
+ %m_x_19 = tt.broadcast %m_x : tensor<32x1xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48)
41
+ %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked> -> tensor<32x128xi1, #blocked> loc(#loc48)
42
+ %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<32x128xi1, #blocked> loc(#loc48)
43
+ %p_x = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<32xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xi64, #blocked> loc(#loc49)
44
+ %p_x_22 = arith.muli %p_x, %cst_2 : tensor<32x1xi64, #blocked> loc(#loc50)
45
+ %p_x_23 = tt.splat %x : !tt.ptr<bf16> -> tensor<32x1x!tt.ptr<bf16>, #blocked> loc(#loc51)
46
+ %p_x_24 = tt.addptr %p_x_23, %p_x_22 : tensor<32x1x!tt.ptr<bf16>, #blocked>, tensor<32x1xi64, #blocked> loc(#loc51)
47
+ %p_x_25 = tt.broadcast %p_x_24 : tensor<32x1x!tt.ptr<bf16>, #blocked> -> tensor<32x128x!tt.ptr<bf16>, #blocked> loc(#loc52)
48
+ %p_x_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked> -> tensor<32x128xi32, #blocked> loc(#loc52)
49
+ %p_x_27 = tt.addptr %p_x_25, %p_x_26 : tensor<32x128x!tt.ptr<bf16>, #blocked>, tensor<32x128xi32, #blocked> loc(#loc52)
50
+ %p_y = tt.splat %y : !tt.ptr<bf16> -> tensor<32x1x!tt.ptr<bf16>, #blocked> loc(#loc53)
51
+ %p_y_28 = tt.addptr %p_y, %p_x_22 : tensor<32x1x!tt.ptr<bf16>, #blocked>, tensor<32x1xi64, #blocked> loc(#loc53)
52
+ %p_y_29 = tt.broadcast %p_y_28 : tensor<32x1x!tt.ptr<bf16>, #blocked> -> tensor<32x128x!tt.ptr<bf16>, #blocked> loc(#loc54)
53
+ %p_y_30 = tt.addptr %p_y_29, %p_x_26 : tensor<32x128x!tt.ptr<bf16>, #blocked>, tensor<32x128xi32, #blocked> loc(#loc54)
54
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<32x!tt.ptr<f32>, #blocked1> loc(#loc55)
55
+ %p_rstd_31 = tt.addptr %p_rstd, %o_t_11 : tensor<32x!tt.ptr<f32>, #blocked1>, tensor<32xi64, #blocked1> loc(#loc55)
56
+ %b_x = tt.load %p_x_27, %m_x_21, %cst_0 : tensor<32x128x!tt.ptr<bf16>, #blocked> loc(#loc56)
57
+ %b_x_32 = arith.extf %b_x : tensor<32x128xbf16, #blocked> to tensor<32x128xf32, #blocked> loc(#loc57)
58
+ %b_rstd = arith.mulf %b_x_32, %b_x_32 : tensor<32x128xf32, #blocked> loc(#loc58)
59
+ %b_rstd_33 = "tt.reduce"(%b_rstd) <{axis = 1 : i32}> ({
60
+ ^bb0(%b_rstd_40: f32 loc(callsite(#loc1 at #loc59)), %b_rstd_41: f32 loc(callsite(#loc1 at #loc59))):
61
+ %b_rstd_42 = arith.addf %b_rstd_40, %b_rstd_41 : f32 loc(#loc67)
62
+ tt.reduce.return %b_rstd_42 : f32 loc(#loc65)
63
+ }) : (tensor<32x128xf32, #blocked>) -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc65)
64
+ %b_rstd_34 = tt.splat %eps : f32 -> tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60)
65
+ %b_rstd_35 = arith.addf %b_rstd_33, %b_rstd_34 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc60)
66
+ %b_rstd_36 = math.sqrt %b_rstd_35 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc61)
67
+ %b_rstd_37 = arith.divf %cst, %b_rstd_36 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc62)
68
+ %b_y = tt.expand_dims %b_rstd_37 {axis = 1 : i32} : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32x1xf32, #blocked> loc(#loc63)
69
+ %b_y_38 = tt.broadcast %b_y : tensor<32x1xf32, #blocked> -> tensor<32x128xf32, #blocked> loc(#loc64)
70
+ %b_y_39 = arith.mulf %b_x_32, %b_y_38 : tensor<32x128xf32, #blocked> loc(#loc64)
71
+ %0 = arith.truncf %b_y_39 : tensor<32x128xf32, #blocked> to tensor<32x128xbf16, #blocked> loc(#loc30)
72
+ tt.store %p_y_30, %0, %m_x_21 : tensor<32x128x!tt.ptr<bf16>, #blocked> loc(#loc31)
73
+ %1 = ttg.convert_layout %b_rstd_37 : tensor<32xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<32xf32, #blocked1> loc(#loc32)
74
+ tt.store %p_rstd_31, %1, %m_t_15 : tensor<32x!tt.ptr<f32>, #blocked1> loc(#loc32)
75
+ tt.return loc(#loc33)
76
+ } loc(#loc)
77
+ } loc(#loc)
78
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
79
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
80
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
81
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
82
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
83
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
84
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
85
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
86
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
87
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
88
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
89
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
90
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
91
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
92
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
93
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
94
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
95
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
96
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
97
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
98
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
99
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
100
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
101
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
102
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
103
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
104
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
105
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
106
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
107
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
108
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
109
+ #loc39 = loc("i_t"(#loc2))
110
+ #loc40 = loc("i_t"(#loc3))
111
+ #loc41 = loc("o_t"(#loc4))
112
+ #loc42 = loc("o_t"(#loc5))
113
+ #loc43 = loc("o_t"(#loc6))
114
+ #loc44 = loc("m_t"(#loc7))
115
+ #loc45 = loc("m_x"(#loc8))
116
+ #loc46 = loc("m_x"(#loc9))
117
+ #loc47 = loc("m_x"(#loc10))
118
+ #loc48 = loc("m_x"(#loc11))
119
+ #loc49 = loc("p_x"(#loc12))
120
+ #loc50 = loc("p_x"(#loc13))
121
+ #loc51 = loc("p_x"(#loc14))
122
+ #loc52 = loc("p_x"(#loc15))
123
+ #loc53 = loc("p_y"(#loc16))
124
+ #loc54 = loc("p_y"(#loc17))
125
+ #loc55 = loc("p_rstd"(#loc18))
126
+ #loc56 = loc("b_x"(#loc19))
127
+ #loc57 = loc("b_x"(#loc20))
128
+ #loc58 = loc("b_rstd"(#loc21))
129
+ #loc60 = loc("b_rstd"(#loc25))
130
+ #loc61 = loc("b_rstd"(#loc26))
131
+ #loc62 = loc("b_rstd"(#loc27))
132
+ #loc63 = loc("b_y"(#loc28))
133
+ #loc64 = loc("b_y"(#loc29))
134
+ #loc65 = loc(callsite(#loc22 at #loc59))
135
+ #loc67 = loc(callsite(#loc24 at #loc65))
triton/gpu0/AUPUMNPJXH6A35AOPQUO34OMYXZDFS4GMEAKGLYW2ALFIJQXM7XA/l2norm_fwd_kernel.ttir ADDED
@@ -0,0 +1,128 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":69:0)
2
+ #loc3 = loc(unknown)
3
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:43)
4
+ #loc35 = loc("x"(#loc))
5
+ #loc36 = loc("y"(#loc))
6
+ #loc37 = loc("rstd"(#loc))
7
+ #loc38 = loc("eps"(#loc))
8
+ #loc39 = loc("T"(#loc))
9
+ #loc62 = loc("b_rstd"(#loc25))
10
+ #loc68 = loc(callsite(#loc3 at #loc62))
11
+ module {
12
+ tt.func public @l2norm_fwd_kernel(%x: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("x"(#loc)), %y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
13
+ %b_x = arith.constant dense<0.000000e+00> : tensor<32x128xbf16> loc(#loc40)
14
+ %b_rstd = arith.constant dense<1.000000e+00> : tensor<32xf32> loc(#loc41)
15
+ %cst = arith.constant dense<128> : tensor<32x1xi64> loc(#loc3)
16
+ %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc42)
17
+ %o_t = arith.constant 32 : i64 loc(#loc43)
18
+ %i_t = tt.get_program_id x : i32 loc(#loc44)
19
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc45)
20
+ %o_t_1 = arith.muli %i_t_0, %o_t : i64 loc(#loc43)
21
+ %o_t_2 = tt.make_range {end = 32 : i32, start = 0 : i32} : tensor<32xi32> loc(#loc46)
22
+ %o_t_3 = arith.extsi %o_t_2 : tensor<32xi32> to tensor<32xi64> loc(#loc47)
23
+ %o_t_4 = tt.splat %o_t_1 : i64 -> tensor<32xi64> loc(#loc47)
24
+ %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<32xi64> loc(#loc47)
25
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc48)
26
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc49)
27
+ %m_t_6 = tt.splat %m_t : i64 -> tensor<32xi64> loc(#loc49)
28
+ %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<32xi64> loc(#loc49)
29
+ %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<32xi1> -> tensor<32x1xi1> loc(#loc50)
30
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc51)
31
+ %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc42)
32
+ %m_x_11 = tt.broadcast %m_x_8 : tensor<32x1xi1> -> tensor<32x128xi1> loc(#loc52)
33
+ %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<32x128xi1> loc(#loc52)
34
+ %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<32x128xi1> loc(#loc52)
35
+ %p_x = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<32xi64> -> tensor<32x1xi64> loc(#loc53)
36
+ %p_x_14 = arith.muli %p_x, %cst : tensor<32x1xi64> loc(#loc54)
37
+ %p_x_15 = tt.splat %x : !tt.ptr<bf16> -> tensor<32x1x!tt.ptr<bf16>> loc(#loc55)
38
+ %p_x_16 = tt.addptr %p_x_15, %p_x_14 : tensor<32x1x!tt.ptr<bf16>>, tensor<32x1xi64> loc(#loc55)
39
+ %p_x_17 = tt.broadcast %p_x_16 : tensor<32x1x!tt.ptr<bf16>> -> tensor<32x128x!tt.ptr<bf16>> loc(#loc56)
40
+ %p_x_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<32x128xi32> loc(#loc56)
41
+ %p_x_19 = tt.addptr %p_x_17, %p_x_18 : tensor<32x128x!tt.ptr<bf16>>, tensor<32x128xi32> loc(#loc56)
42
+ %p_y = tt.splat %y : !tt.ptr<bf16> -> tensor<32x1x!tt.ptr<bf16>> loc(#loc57)
43
+ %p_y_20 = tt.addptr %p_y, %p_x_14 : tensor<32x1x!tt.ptr<bf16>>, tensor<32x1xi64> loc(#loc57)
44
+ %p_y_21 = tt.broadcast %p_y_20 : tensor<32x1x!tt.ptr<bf16>> -> tensor<32x128x!tt.ptr<bf16>> loc(#loc58)
45
+ %p_y_22 = tt.addptr %p_y_21, %p_x_18 : tensor<32x128x!tt.ptr<bf16>>, tensor<32x128xi32> loc(#loc58)
46
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<32x!tt.ptr<f32>> loc(#loc59)
47
+ %p_rstd_23 = tt.addptr %p_rstd, %o_t_5 : tensor<32x!tt.ptr<f32>>, tensor<32xi64> loc(#loc59)
48
+ %b_x_24 = tt.load %p_x_19, %m_x_13, %b_x : tensor<32x128x!tt.ptr<bf16>> loc(#loc40)
49
+ %b_x_25 = arith.extf %b_x_24 : tensor<32x128xbf16> to tensor<32x128xf32> loc(#loc60)
50
+ %b_rstd_26 = arith.mulf %b_x_25, %b_x_25 : tensor<32x128xf32> loc(#loc61)
51
+ %b_rstd_27 = "tt.reduce"(%b_rstd_26) <{axis = 1 : i32}> ({
52
+ ^bb0(%b_rstd_34: f32 loc(callsite(#loc3 at #loc62)), %b_rstd_35: f32 loc(callsite(#loc3 at #loc62))):
53
+ %b_rstd_36 = arith.addf %b_rstd_34, %b_rstd_35 : f32 loc(#loc69)
54
+ tt.reduce.return %b_rstd_36 : f32 loc(#loc67)
55
+ }) : (tensor<32x128xf32>) -> tensor<32xf32> loc(#loc67)
56
+ %b_rstd_28 = tt.splat %eps : f32 -> tensor<32xf32> loc(#loc63)
57
+ %b_rstd_29 = arith.addf %b_rstd_27, %b_rstd_28 : tensor<32xf32> loc(#loc63)
58
+ %b_rstd_30 = math.sqrt %b_rstd_29 : tensor<32xf32> loc(#loc64)
59
+ %b_rstd_31 = arith.divf %b_rstd, %b_rstd_30 : tensor<32xf32> loc(#loc41)
60
+ %b_y = tt.expand_dims %b_rstd_31 {axis = 1 : i32} : tensor<32xf32> -> tensor<32x1xf32> loc(#loc65)
61
+ %b_y_32 = tt.broadcast %b_y : tensor<32x1xf32> -> tensor<32x128xf32> loc(#loc66)
62
+ %b_y_33 = arith.mulf %b_x_25, %b_y_32 : tensor<32x128xf32> loc(#loc66)
63
+ %0 = arith.truncf %b_y_33 : tensor<32x128xf32> to tensor<32x128xbf16> loc(#loc31)
64
+ tt.store %p_y_22, %0, %m_x_13 : tensor<32x128x!tt.ptr<bf16>> loc(#loc32)
65
+ tt.store %p_rstd_23, %b_rstd_31, %m_t_7 : tensor<32x!tt.ptr<f32>> loc(#loc33)
66
+ tt.return loc(#loc34)
67
+ } loc(#loc)
68
+ } loc(#loc)
69
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:18)
70
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:17)
71
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:41)
72
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:16)
73
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:24)
74
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":70:30)
75
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:34)
76
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":71:21)
77
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":72:23)
78
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":73:16)
79
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:14)
80
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:30)
81
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":74:26)
82
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:18)
83
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:29)
84
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:14)
85
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":75:33)
86
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:14)
87
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":76:33)
88
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":77:20)
89
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":79:47)
90
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:38)
91
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
92
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
93
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:48)
94
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":80:25)
95
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:23)
96
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":81:16)
97
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:25)
98
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":83:18)
99
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:21)
100
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":84:4)
101
+ #loc40 = loc("b_x"(#loc1))
102
+ #loc41 = loc("b_rstd"(#loc2))
103
+ #loc42 = loc("m_x"(#loc4))
104
+ #loc43 = loc("o_t"(#loc5))
105
+ #loc44 = loc("i_t"(#loc6))
106
+ #loc45 = loc("i_t"(#loc7))
107
+ #loc46 = loc("o_t"(#loc8))
108
+ #loc47 = loc("o_t"(#loc9))
109
+ #loc48 = loc("o_d"(#loc10))
110
+ #loc49 = loc("m_t"(#loc11))
111
+ #loc50 = loc("m_x"(#loc12))
112
+ #loc51 = loc("m_x"(#loc13))
113
+ #loc52 = loc("m_x"(#loc14))
114
+ #loc53 = loc("p_x"(#loc15))
115
+ #loc54 = loc("p_x"(#loc16))
116
+ #loc55 = loc("p_x"(#loc17))
117
+ #loc56 = loc("p_x"(#loc18))
118
+ #loc57 = loc("p_y"(#loc19))
119
+ #loc58 = loc("p_y"(#loc20))
120
+ #loc59 = loc("p_rstd"(#loc21))
121
+ #loc60 = loc("b_x"(#loc22))
122
+ #loc61 = loc("b_rstd"(#loc23))
123
+ #loc63 = loc("b_rstd"(#loc27))
124
+ #loc64 = loc("b_rstd"(#loc28))
125
+ #loc65 = loc("b_y"(#loc29))
126
+ #loc66 = loc("b_y"(#loc30))
127
+ #loc67 = loc(callsite(#loc24 at #loc62))
128
+ #loc69 = loc(callsite(#loc26 at #loc67))
triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/__grp__l2norm_bwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}}
triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.cubin ADDED
Binary file (62.5 kB). View file
 
triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}
triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.llir ADDED
@@ -0,0 +1,1269 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ; ModuleID = 'LLVMDialectModule'
2
+ source_filename = "LLVMDialectModule"
3
+ 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"
4
+
5
+ ; Function Attrs: nounwind
6
+ 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 {
7
+ %9 = tail call i32 @llvm.nvvm.read.ptx.sreg.ctaid.x(), !dbg !12
8
+ %10 = zext nneg i32 %9 to i64, !dbg !13
9
+ %11 = shl nuw nsw i64 %10, 7, !dbg !14
10
+ %12 = tail call i32 @llvm.nvvm.read.ptx.sreg.tid.x(), !dbg !15
11
+ %13 = lshr i32 %12, 4, !dbg !15
12
+ %14 = and i32 %13, 15, !dbg !15
13
+ %15 = or disjoint i32 %14, 16, !dbg !15
14
+ %16 = insertelement <2 x i32> poison, i32 %13, i64 0, !dbg !15
15
+ %17 = insertelement <2 x i32> %16, i32 %14, i64 1, !dbg !15
16
+ %18 = or <2 x i32> %17, <i32 48, i32 32>, !dbg !15
17
+ %19 = insertelement <4 x i32> poison, i32 %13, i64 0, !dbg !15
18
+ %20 = insertelement <4 x i32> %19, i32 %14, i64 1, !dbg !15
19
+ %21 = shufflevector <2 x i32> %18, <2 x i32> poison, <8 x i32> <i32 poison, i32 poison, i32 poison, i32 poison, i32 0, i32 1, i32 poison, i32 poison>, !dbg !16
20
+ %22 = insertelement <8 x i32> %21, i32 %15, i64 6, !dbg !16
21
+ %23 = insertelement <8 x i32> %22, i32 %14, i64 7, !dbg !16
22
+ %24 = shufflevector <4 x i32> %20, <4 x i32> poison, <8 x i32> <i32 0, i32 1, i32 1, i32 1, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !16
23
+ %25 = or <8 x i32> %24, <i32 112, i32 96, i32 80, i32 64, i32 poison, i32 poison, i32 poison, i32 poison>, !dbg !16
24
+ %26 = shufflevector <8 x i32> %25, <8 x i32> %23, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 12, i32 13, i32 14, i32 15>, !dbg !16
25
+ %27 = zext <8 x i32> %26 to <8 x i64>, !dbg !16
26
+ %28 = insertelement <8 x i64> poison, i64 %11, i64 0, !dbg !16
27
+ %29 = shufflevector <8 x i64> %28, <8 x i64> poison, <8 x i32> zeroinitializer, !dbg !16
28
+ %30 = or disjoint <8 x i64> %29, %27, !dbg !16
29
+ %31 = insertelement <8 x i32> poison, i32 %5, i64 0, !dbg !17
30
+ %32 = shufflevector <8 x i32> %31, <8 x i32> poison, <8 x i32> zeroinitializer, !dbg !17
31
+ %33 = sext <8 x i32> %32 to <8 x i64>, !dbg !17
32
+ %34 = icmp slt <8 x i64> %30, %33, !dbg !17
33
+ %35 = shl nuw nsw i32 %12, 3, !dbg !18
34
+ %36 = and i32 %35, 120, !dbg !18
35
+ %37 = extractelement <8 x i64> %30, i64 7, !dbg !19
36
+ %38 = shl nuw nsw i64 %37, 7, !dbg !19
37
+ %39 = extractelement <8 x i64> %30, i64 6, !dbg !19
38
+ %40 = shl nuw nsw i64 %39, 7, !dbg !19
39
+ %41 = extractelement <8 x i64> %30, i64 5, !dbg !19
40
+ %42 = shl nuw nsw i64 %41, 7, !dbg !19
41
+ %43 = extractelement <8 x i64> %30, i64 4, !dbg !19
42
+ %44 = shl nuw nsw i64 %43, 7, !dbg !19
43
+ %45 = extractelement <8 x i64> %30, i64 3, !dbg !19
44
+ %46 = shl nuw nsw i64 %45, 7, !dbg !19
45
+ %47 = extractelement <8 x i64> %30, i64 2, !dbg !19
46
+ %48 = shl nuw nsw i64 %47, 7, !dbg !19
47
+ %49 = extractelement <8 x i64> %30, i64 1, !dbg !19
48
+ %50 = shl nuw nsw i64 %49, 7, !dbg !19
49
+ %51 = extractelement <8 x i64> %30, i64 0, !dbg !19
50
+ %52 = shl nuw nsw i64 %51, 7, !dbg !19
51
+ %53 = getelementptr bfloat, ptr addrspace(1) %0, i64 %38, !dbg !20
52
+ %54 = getelementptr bfloat, ptr addrspace(1) %0, i64 %40, !dbg !20
53
+ %55 = getelementptr bfloat, ptr addrspace(1) %0, i64 %42, !dbg !20
54
+ %56 = getelementptr bfloat, ptr addrspace(1) %0, i64 %44, !dbg !20
55
+ %57 = getelementptr bfloat, ptr addrspace(1) %0, i64 %46, !dbg !20
56
+ %58 = getelementptr bfloat, ptr addrspace(1) %0, i64 %48, !dbg !20
57
+ %59 = getelementptr bfloat, ptr addrspace(1) %0, i64 %50, !dbg !20
58
+ %60 = getelementptr bfloat, ptr addrspace(1) %0, i64 %52, !dbg !20
59
+ %61 = zext nneg i32 %36 to i64, !dbg !21
60
+ %62 = getelementptr bfloat, ptr addrspace(1) %53, i64 %61, !dbg !21
61
+ %63 = getelementptr bfloat, ptr addrspace(1) %54, i64 %61, !dbg !21
62
+ %64 = getelementptr bfloat, ptr addrspace(1) %55, i64 %61, !dbg !21
63
+ %65 = getelementptr bfloat, ptr addrspace(1) %56, i64 %61, !dbg !21
64
+ %66 = getelementptr bfloat, ptr addrspace(1) %57, i64 %61, !dbg !21
65
+ %67 = getelementptr bfloat, ptr addrspace(1) %58, i64 %61, !dbg !21
66
+ %68 = getelementptr bfloat, ptr addrspace(1) %59, i64 %61, !dbg !21
67
+ %69 = getelementptr bfloat, ptr addrspace(1) %60, i64 %61, !dbg !21
68
+ %70 = getelementptr float, ptr addrspace(1) %1, i64 %37, !dbg !22
69
+ %71 = getelementptr float, ptr addrspace(1) %1, i64 %39, !dbg !22
70
+ %72 = getelementptr float, ptr addrspace(1) %1, i64 %41, !dbg !22
71
+ %73 = getelementptr float, ptr addrspace(1) %1, i64 %43, !dbg !22
72
+ %74 = getelementptr float, ptr addrspace(1) %1, i64 %45, !dbg !22
73
+ %75 = getelementptr float, ptr addrspace(1) %1, i64 %47, !dbg !22
74
+ %76 = getelementptr float, ptr addrspace(1) %1, i64 %49, !dbg !22
75
+ %77 = getelementptr float, ptr addrspace(1) %1, i64 %51, !dbg !22
76
+ %78 = getelementptr bfloat, ptr addrspace(1) %2, i64 %38, !dbg !23
77
+ %79 = getelementptr bfloat, ptr addrspace(1) %2, i64 %40, !dbg !23
78
+ %80 = getelementptr bfloat, ptr addrspace(1) %2, i64 %42, !dbg !23
79
+ %81 = getelementptr bfloat, ptr addrspace(1) %2, i64 %44, !dbg !23
80
+ %82 = getelementptr bfloat, ptr addrspace(1) %2, i64 %46, !dbg !23
81
+ %83 = getelementptr bfloat, ptr addrspace(1) %2, i64 %48, !dbg !23
82
+ %84 = getelementptr bfloat, ptr addrspace(1) %2, i64 %50, !dbg !23
83
+ %85 = getelementptr bfloat, ptr addrspace(1) %2, i64 %52, !dbg !23
84
+ %86 = getelementptr bfloat, ptr addrspace(1) %78, i64 %61, !dbg !24
85
+ %87 = getelementptr bfloat, ptr addrspace(1) %79, i64 %61, !dbg !24
86
+ %88 = getelementptr bfloat, ptr addrspace(1) %80, i64 %61, !dbg !24
87
+ %89 = getelementptr bfloat, ptr addrspace(1) %81, i64 %61, !dbg !24
88
+ %90 = getelementptr bfloat, ptr addrspace(1) %82, i64 %61, !dbg !24
89
+ %91 = getelementptr bfloat, ptr addrspace(1) %83, i64 %61, !dbg !24
90
+ %92 = getelementptr bfloat, ptr addrspace(1) %84, i64 %61, !dbg !24
91
+ %93 = getelementptr bfloat, ptr addrspace(1) %85, i64 %61, !dbg !24
92
+ %94 = getelementptr bfloat, ptr addrspace(1) %3, i64 %38, !dbg !25
93
+ %95 = getelementptr bfloat, ptr addrspace(1) %3, i64 %40, !dbg !25
94
+ %96 = getelementptr bfloat, ptr addrspace(1) %3, i64 %42, !dbg !25
95
+ %97 = getelementptr bfloat, ptr addrspace(1) %3, i64 %44, !dbg !25
96
+ %98 = getelementptr bfloat, ptr addrspace(1) %3, i64 %46, !dbg !25
97
+ %99 = getelementptr bfloat, ptr addrspace(1) %3, i64 %48, !dbg !25
98
+ %100 = getelementptr bfloat, ptr addrspace(1) %3, i64 %50, !dbg !25
99
+ %101 = getelementptr bfloat, ptr addrspace(1) %3, i64 %52, !dbg !25
100
+ %102 = getelementptr bfloat, ptr addrspace(1) %94, i64 %61, !dbg !26
101
+ %103 = getelementptr bfloat, ptr addrspace(1) %95, i64 %61, !dbg !26
102
+ %104 = getelementptr bfloat, ptr addrspace(1) %96, i64 %61, !dbg !26
103
+ %105 = getelementptr bfloat, ptr addrspace(1) %97, i64 %61, !dbg !26
104
+ %106 = getelementptr bfloat, ptr addrspace(1) %98, i64 %61, !dbg !26
105
+ %107 = getelementptr bfloat, ptr addrspace(1) %99, i64 %61, !dbg !26
106
+ %108 = getelementptr bfloat, ptr addrspace(1) %100, i64 %61, !dbg !26
107
+ %109 = getelementptr bfloat, ptr addrspace(1) %101, i64 %61, !dbg !26
108
+ %110 = extractelement <8 x i1> %34, i64 7, !dbg !27
109
+ %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
110
+ %112 = extractvalue { i32, i32, i32, i32 } %111, 0, !dbg !27
111
+ %113 = bitcast i32 %112 to <2 x bfloat>, !dbg !27
112
+ %114 = extractvalue { i32, i32, i32, i32 } %111, 1, !dbg !27
113
+ %115 = bitcast i32 %114 to <2 x bfloat>, !dbg !27
114
+ %116 = extractvalue { i32, i32, i32, i32 } %111, 2, !dbg !27
115
+ %117 = bitcast i32 %116 to <2 x bfloat>, !dbg !27
116
+ %118 = extractvalue { i32, i32, i32, i32 } %111, 3, !dbg !27
117
+ %119 = bitcast i32 %118 to <2 x bfloat>, !dbg !27
118
+ %120 = extractelement <2 x bfloat> %113, i64 0, !dbg !27
119
+ %121 = extractelement <2 x bfloat> %113, i64 1, !dbg !27
120
+ %122 = extractelement <2 x bfloat> %115, i64 0, !dbg !27
121
+ %123 = extractelement <2 x bfloat> %115, i64 1, !dbg !27
122
+ %124 = extractelement <2 x bfloat> %117, i64 0, !dbg !27
123
+ %125 = extractelement <2 x bfloat> %117, i64 1, !dbg !27
124
+ %126 = extractelement <2 x bfloat> %119, i64 0, !dbg !27
125
+ %127 = extractelement <2 x bfloat> %119, i64 1, !dbg !27
126
+ %128 = extractelement <8 x i1> %34, i64 6, !dbg !27
127
+ %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
128
+ %130 = extractvalue { i32, i32, i32, i32 } %129, 0, !dbg !27
129
+ %131 = bitcast i32 %130 to <2 x bfloat>, !dbg !27
130
+ %132 = extractvalue { i32, i32, i32, i32 } %129, 1, !dbg !27
131
+ %133 = bitcast i32 %132 to <2 x bfloat>, !dbg !27
132
+ %134 = extractvalue { i32, i32, i32, i32 } %129, 2, !dbg !27
133
+ %135 = bitcast i32 %134 to <2 x bfloat>, !dbg !27
134
+ %136 = extractvalue { i32, i32, i32, i32 } %129, 3, !dbg !27
135
+ %137 = bitcast i32 %136 to <2 x bfloat>, !dbg !27
136
+ %138 = extractelement <2 x bfloat> %131, i64 0, !dbg !27
137
+ %139 = extractelement <2 x bfloat> %131, i64 1, !dbg !27
138
+ %140 = extractelement <2 x bfloat> %133, i64 0, !dbg !27
139
+ %141 = extractelement <2 x bfloat> %133, i64 1, !dbg !27
140
+ %142 = extractelement <2 x bfloat> %135, i64 0, !dbg !27
141
+ %143 = extractelement <2 x bfloat> %135, i64 1, !dbg !27
142
+ %144 = extractelement <2 x bfloat> %137, i64 0, !dbg !27
143
+ %145 = extractelement <2 x bfloat> %137, i64 1, !dbg !27
144
+ %146 = extractelement <8 x i1> %34, i64 5, !dbg !27
145
+ %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
146
+ %148 = extractvalue { i32, i32, i32, i32 } %147, 0, !dbg !27
147
+ %149 = bitcast i32 %148 to <2 x bfloat>, !dbg !27
148
+ %150 = extractvalue { i32, i32, i32, i32 } %147, 1, !dbg !27
149
+ %151 = bitcast i32 %150 to <2 x bfloat>, !dbg !27
150
+ %152 = extractvalue { i32, i32, i32, i32 } %147, 2, !dbg !27
151
+ %153 = bitcast i32 %152 to <2 x bfloat>, !dbg !27
152
+ %154 = extractvalue { i32, i32, i32, i32 } %147, 3, !dbg !27
153
+ %155 = bitcast i32 %154 to <2 x bfloat>, !dbg !27
154
+ %156 = extractelement <2 x bfloat> %149, i64 0, !dbg !27
155
+ %157 = extractelement <2 x bfloat> %149, i64 1, !dbg !27
156
+ %158 = extractelement <2 x bfloat> %151, i64 0, !dbg !27
157
+ %159 = extractelement <2 x bfloat> %151, i64 1, !dbg !27
158
+ %160 = extractelement <2 x bfloat> %153, i64 0, !dbg !27
159
+ %161 = extractelement <2 x bfloat> %153, i64 1, !dbg !27
160
+ %162 = extractelement <2 x bfloat> %155, i64 0, !dbg !27
161
+ %163 = extractelement <2 x bfloat> %155, i64 1, !dbg !27
162
+ %164 = extractelement <8 x i1> %34, i64 4, !dbg !27
163
+ %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
164
+ %166 = extractvalue { i32, i32, i32, i32 } %165, 0, !dbg !27
165
+ %167 = bitcast i32 %166 to <2 x bfloat>, !dbg !27
166
+ %168 = extractvalue { i32, i32, i32, i32 } %165, 1, !dbg !27
167
+ %169 = bitcast i32 %168 to <2 x bfloat>, !dbg !27
168
+ %170 = extractvalue { i32, i32, i32, i32 } %165, 2, !dbg !27
169
+ %171 = bitcast i32 %170 to <2 x bfloat>, !dbg !27
170
+ %172 = extractvalue { i32, i32, i32, i32 } %165, 3, !dbg !27
171
+ %173 = bitcast i32 %172 to <2 x bfloat>, !dbg !27
172
+ %174 = extractelement <2 x bfloat> %167, i64 0, !dbg !27
173
+ %175 = extractelement <2 x bfloat> %167, i64 1, !dbg !27
174
+ %176 = extractelement <2 x bfloat> %169, i64 0, !dbg !27
175
+ %177 = extractelement <2 x bfloat> %169, i64 1, !dbg !27
176
+ %178 = extractelement <2 x bfloat> %171, i64 0, !dbg !27
177
+ %179 = extractelement <2 x bfloat> %171, i64 1, !dbg !27
178
+ %180 = extractelement <2 x bfloat> %173, i64 0, !dbg !27
179
+ %181 = extractelement <2 x bfloat> %173, i64 1, !dbg !27
180
+ %182 = extractelement <8 x i1> %34, i64 3, !dbg !27
181
+ %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
182
+ %184 = extractvalue { i32, i32, i32, i32 } %183, 0, !dbg !27
183
+ %185 = bitcast i32 %184 to <2 x bfloat>, !dbg !27
184
+ %186 = extractvalue { i32, i32, i32, i32 } %183, 1, !dbg !27
185
+ %187 = bitcast i32 %186 to <2 x bfloat>, !dbg !27
186
+ %188 = extractvalue { i32, i32, i32, i32 } %183, 2, !dbg !27
187
+ %189 = bitcast i32 %188 to <2 x bfloat>, !dbg !27
188
+ %190 = extractvalue { i32, i32, i32, i32 } %183, 3, !dbg !27
189
+ %191 = bitcast i32 %190 to <2 x bfloat>, !dbg !27
190
+ %192 = extractelement <2 x bfloat> %185, i64 0, !dbg !27
191
+ %193 = extractelement <2 x bfloat> %185, i64 1, !dbg !27
192
+ %194 = extractelement <2 x bfloat> %187, i64 0, !dbg !27
193
+ %195 = extractelement <2 x bfloat> %187, i64 1, !dbg !27
194
+ %196 = extractelement <2 x bfloat> %189, i64 0, !dbg !27
195
+ %197 = extractelement <2 x bfloat> %189, i64 1, !dbg !27
196
+ %198 = extractelement <2 x bfloat> %191, i64 0, !dbg !27
197
+ %199 = extractelement <2 x bfloat> %191, i64 1, !dbg !27
198
+ %200 = extractelement <8 x i1> %34, i64 2, !dbg !27
199
+ %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
200
+ %202 = extractvalue { i32, i32, i32, i32 } %201, 0, !dbg !27
201
+ %203 = bitcast i32 %202 to <2 x bfloat>, !dbg !27
202
+ %204 = extractvalue { i32, i32, i32, i32 } %201, 1, !dbg !27
203
+ %205 = bitcast i32 %204 to <2 x bfloat>, !dbg !27
204
+ %206 = extractvalue { i32, i32, i32, i32 } %201, 2, !dbg !27
205
+ %207 = bitcast i32 %206 to <2 x bfloat>, !dbg !27
206
+ %208 = extractvalue { i32, i32, i32, i32 } %201, 3, !dbg !27
207
+ %209 = bitcast i32 %208 to <2 x bfloat>, !dbg !27
208
+ %210 = extractelement <2 x bfloat> %203, i64 0, !dbg !27
209
+ %211 = extractelement <2 x bfloat> %203, i64 1, !dbg !27
210
+ %212 = extractelement <2 x bfloat> %205, i64 0, !dbg !27
211
+ %213 = extractelement <2 x bfloat> %205, i64 1, !dbg !27
212
+ %214 = extractelement <2 x bfloat> %207, i64 0, !dbg !27
213
+ %215 = extractelement <2 x bfloat> %207, i64 1, !dbg !27
214
+ %216 = extractelement <2 x bfloat> %209, i64 0, !dbg !27
215
+ %217 = extractelement <2 x bfloat> %209, i64 1, !dbg !27
216
+ %218 = extractelement <8 x i1> %34, i64 1, !dbg !27
217
+ %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
218
+ %220 = extractvalue { i32, i32, i32, i32 } %219, 0, !dbg !27
219
+ %221 = bitcast i32 %220 to <2 x bfloat>, !dbg !27
220
+ %222 = extractvalue { i32, i32, i32, i32 } %219, 1, !dbg !27
221
+ %223 = bitcast i32 %222 to <2 x bfloat>, !dbg !27
222
+ %224 = extractvalue { i32, i32, i32, i32 } %219, 2, !dbg !27
223
+ %225 = bitcast i32 %224 to <2 x bfloat>, !dbg !27
224
+ %226 = extractvalue { i32, i32, i32, i32 } %219, 3, !dbg !27
225
+ %227 = bitcast i32 %226 to <2 x bfloat>, !dbg !27
226
+ %228 = extractelement <2 x bfloat> %221, i64 0, !dbg !27
227
+ %229 = extractelement <2 x bfloat> %221, i64 1, !dbg !27
228
+ %230 = extractelement <2 x bfloat> %223, i64 0, !dbg !27
229
+ %231 = extractelement <2 x bfloat> %223, i64 1, !dbg !27
230
+ %232 = extractelement <2 x bfloat> %225, i64 0, !dbg !27
231
+ %233 = extractelement <2 x bfloat> %225, i64 1, !dbg !27
232
+ %234 = extractelement <2 x bfloat> %227, i64 0, !dbg !27
233
+ %235 = extractelement <2 x bfloat> %227, i64 1, !dbg !27
234
+ %236 = extractelement <8 x i1> %34, i64 0, !dbg !27
235
+ %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
236
+ %238 = extractvalue { i32, i32, i32, i32 } %237, 0, !dbg !27
237
+ %239 = bitcast i32 %238 to <2 x bfloat>, !dbg !27
238
+ %240 = extractvalue { i32, i32, i32, i32 } %237, 1, !dbg !27
239
+ %241 = bitcast i32 %240 to <2 x bfloat>, !dbg !27
240
+ %242 = extractvalue { i32, i32, i32, i32 } %237, 2, !dbg !27
241
+ %243 = bitcast i32 %242 to <2 x bfloat>, !dbg !27
242
+ %244 = extractvalue { i32, i32, i32, i32 } %237, 3, !dbg !27
243
+ %245 = bitcast i32 %244 to <2 x bfloat>, !dbg !27
244
+ %246 = extractelement <2 x bfloat> %239, i64 0, !dbg !27
245
+ %247 = extractelement <2 x bfloat> %239, i64 1, !dbg !27
246
+ %248 = extractelement <2 x bfloat> %241, i64 0, !dbg !27
247
+ %249 = extractelement <2 x bfloat> %241, i64 1, !dbg !27
248
+ %250 = extractelement <2 x bfloat> %243, i64 0, !dbg !27
249
+ %251 = extractelement <2 x bfloat> %243, i64 1, !dbg !27
250
+ %252 = extractelement <2 x bfloat> %245, i64 0, !dbg !27
251
+ %253 = extractelement <2 x bfloat> %245, i64 1, !dbg !27
252
+ %254 = fpext bfloat %120 to float, !dbg !28
253
+ %255 = fpext bfloat %121 to float, !dbg !28
254
+ %256 = fpext bfloat %122 to float, !dbg !28
255
+ %257 = fpext bfloat %123 to float, !dbg !28
256
+ %258 = fpext bfloat %124 to float, !dbg !28
257
+ %259 = fpext bfloat %125 to float, !dbg !28
258
+ %260 = fpext bfloat %126 to float, !dbg !28
259
+ %261 = fpext bfloat %127 to float, !dbg !28
260
+ %262 = fpext bfloat %138 to float, !dbg !28
261
+ %263 = fpext bfloat %139 to float, !dbg !28
262
+ %264 = fpext bfloat %140 to float, !dbg !28
263
+ %265 = fpext bfloat %141 to float, !dbg !28
264
+ %266 = fpext bfloat %142 to float, !dbg !28
265
+ %267 = fpext bfloat %143 to float, !dbg !28
266
+ %268 = fpext bfloat %144 to float, !dbg !28
267
+ %269 = fpext bfloat %145 to float, !dbg !28
268
+ %270 = fpext bfloat %156 to float, !dbg !28
269
+ %271 = fpext bfloat %157 to float, !dbg !28
270
+ %272 = fpext bfloat %158 to float, !dbg !28
271
+ %273 = fpext bfloat %159 to float, !dbg !28
272
+ %274 = fpext bfloat %160 to float, !dbg !28
273
+ %275 = fpext bfloat %161 to float, !dbg !28
274
+ %276 = fpext bfloat %162 to float, !dbg !28
275
+ %277 = fpext bfloat %163 to float, !dbg !28
276
+ %278 = fpext bfloat %174 to float, !dbg !28
277
+ %279 = fpext bfloat %175 to float, !dbg !28
278
+ %280 = fpext bfloat %176 to float, !dbg !28
279
+ %281 = fpext bfloat %177 to float, !dbg !28
280
+ %282 = fpext bfloat %178 to float, !dbg !28
281
+ %283 = fpext bfloat %179 to float, !dbg !28
282
+ %284 = fpext bfloat %180 to float, !dbg !28
283
+ %285 = fpext bfloat %181 to float, !dbg !28
284
+ %286 = fpext bfloat %192 to float, !dbg !28
285
+ %287 = fpext bfloat %193 to float, !dbg !28
286
+ %288 = fpext bfloat %194 to float, !dbg !28
287
+ %289 = fpext bfloat %195 to float, !dbg !28
288
+ %290 = fpext bfloat %196 to float, !dbg !28
289
+ %291 = fpext bfloat %197 to float, !dbg !28
290
+ %292 = fpext bfloat %198 to float, !dbg !28
291
+ %293 = fpext bfloat %199 to float, !dbg !28
292
+ %294 = fpext bfloat %210 to float, !dbg !28
293
+ %295 = fpext bfloat %211 to float, !dbg !28
294
+ %296 = fpext bfloat %212 to float, !dbg !28
295
+ %297 = fpext bfloat %213 to float, !dbg !28
296
+ %298 = fpext bfloat %214 to float, !dbg !28
297
+ %299 = fpext bfloat %215 to float, !dbg !28
298
+ %300 = fpext bfloat %216 to float, !dbg !28
299
+ %301 = fpext bfloat %217 to float, !dbg !28
300
+ %302 = fpext bfloat %228 to float, !dbg !28
301
+ %303 = fpext bfloat %229 to float, !dbg !28
302
+ %304 = fpext bfloat %230 to float, !dbg !28
303
+ %305 = fpext bfloat %231 to float, !dbg !28
304
+ %306 = fpext bfloat %232 to float, !dbg !28
305
+ %307 = fpext bfloat %233 to float, !dbg !28
306
+ %308 = fpext bfloat %234 to float, !dbg !28
307
+ %309 = fpext bfloat %235 to float, !dbg !28
308
+ %310 = fpext bfloat %246 to float, !dbg !28
309
+ %311 = fpext bfloat %247 to float, !dbg !28
310
+ %312 = fpext bfloat %248 to float, !dbg !28
311
+ %313 = fpext bfloat %249 to float, !dbg !28
312
+ %314 = fpext bfloat %250 to float, !dbg !28
313
+ %315 = fpext bfloat %251 to float, !dbg !28
314
+ %316 = fpext bfloat %252 to float, !dbg !28
315
+ %317 = fpext bfloat %253 to float, !dbg !28
316
+ %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
317
+ %319 = bitcast i32 %318 to float, !dbg !29
318
+ %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
319
+ %321 = bitcast i32 %320 to float, !dbg !29
320
+ %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
321
+ %323 = bitcast i32 %322 to float, !dbg !29
322
+ %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
323
+ %325 = bitcast i32 %324 to float, !dbg !29
324
+ %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
325
+ %327 = bitcast i32 %326 to float, !dbg !29
326
+ %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
327
+ %329 = bitcast i32 %328 to float, !dbg !29
328
+ %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
329
+ %331 = bitcast i32 %330 to float, !dbg !29
330
+ %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
331
+ %333 = bitcast i32 %332 to float, !dbg !29
332
+ %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
333
+ %335 = extractvalue { i32, i32, i32, i32 } %334, 0, !dbg !30
334
+ %336 = bitcast i32 %335 to <2 x bfloat>, !dbg !30
335
+ %337 = extractvalue { i32, i32, i32, i32 } %334, 1, !dbg !30
336
+ %338 = bitcast i32 %337 to <2 x bfloat>, !dbg !30
337
+ %339 = extractvalue { i32, i32, i32, i32 } %334, 2, !dbg !30
338
+ %340 = bitcast i32 %339 to <2 x bfloat>, !dbg !30
339
+ %341 = extractvalue { i32, i32, i32, i32 } %334, 3, !dbg !30
340
+ %342 = bitcast i32 %341 to <2 x bfloat>, !dbg !30
341
+ %343 = extractelement <2 x bfloat> %336, i64 0, !dbg !30
342
+ %344 = extractelement <2 x bfloat> %336, i64 1, !dbg !30
343
+ %345 = extractelement <2 x bfloat> %338, i64 0, !dbg !30
344
+ %346 = extractelement <2 x bfloat> %338, i64 1, !dbg !30
345
+ %347 = extractelement <2 x bfloat> %340, i64 0, !dbg !30
346
+ %348 = extractelement <2 x bfloat> %340, i64 1, !dbg !30
347
+ %349 = extractelement <2 x bfloat> %342, i64 0, !dbg !30
348
+ %350 = extractelement <2 x bfloat> %342, i64 1, !dbg !30
349
+ %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
350
+ %352 = extractvalue { i32, i32, i32, i32 } %351, 0, !dbg !30
351
+ %353 = bitcast i32 %352 to <2 x bfloat>, !dbg !30
352
+ %354 = extractvalue { i32, i32, i32, i32 } %351, 1, !dbg !30
353
+ %355 = bitcast i32 %354 to <2 x bfloat>, !dbg !30
354
+ %356 = extractvalue { i32, i32, i32, i32 } %351, 2, !dbg !30
355
+ %357 = bitcast i32 %356 to <2 x bfloat>, !dbg !30
356
+ %358 = extractvalue { i32, i32, i32, i32 } %351, 3, !dbg !30
357
+ %359 = bitcast i32 %358 to <2 x bfloat>, !dbg !30
358
+ %360 = extractelement <2 x bfloat> %353, i64 0, !dbg !30
359
+ %361 = extractelement <2 x bfloat> %353, i64 1, !dbg !30
360
+ %362 = extractelement <2 x bfloat> %355, i64 0, !dbg !30
361
+ %363 = extractelement <2 x bfloat> %355, i64 1, !dbg !30
362
+ %364 = extractelement <2 x bfloat> %357, i64 0, !dbg !30
363
+ %365 = extractelement <2 x bfloat> %357, i64 1, !dbg !30
364
+ %366 = extractelement <2 x bfloat> %359, i64 0, !dbg !30
365
+ %367 = extractelement <2 x bfloat> %359, i64 1, !dbg !30
366
+ %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
367
+ %369 = extractvalue { i32, i32, i32, i32 } %368, 0, !dbg !30
368
+ %370 = bitcast i32 %369 to <2 x bfloat>, !dbg !30
369
+ %371 = extractvalue { i32, i32, i32, i32 } %368, 1, !dbg !30
370
+ %372 = bitcast i32 %371 to <2 x bfloat>, !dbg !30
371
+ %373 = extractvalue { i32, i32, i32, i32 } %368, 2, !dbg !30
372
+ %374 = bitcast i32 %373 to <2 x bfloat>, !dbg !30
373
+ %375 = extractvalue { i32, i32, i32, i32 } %368, 3, !dbg !30
374
+ %376 = bitcast i32 %375 to <2 x bfloat>, !dbg !30
375
+ %377 = extractelement <2 x bfloat> %370, i64 0, !dbg !30
376
+ %378 = extractelement <2 x bfloat> %370, i64 1, !dbg !30
377
+ %379 = extractelement <2 x bfloat> %372, i64 0, !dbg !30
378
+ %380 = extractelement <2 x bfloat> %372, i64 1, !dbg !30
379
+ %381 = extractelement <2 x bfloat> %374, i64 0, !dbg !30
380
+ %382 = extractelement <2 x bfloat> %374, i64 1, !dbg !30
381
+ %383 = extractelement <2 x bfloat> %376, i64 0, !dbg !30
382
+ %384 = extractelement <2 x bfloat> %376, i64 1, !dbg !30
383
+ %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
384
+ %386 = extractvalue { i32, i32, i32, i32 } %385, 0, !dbg !30
385
+ %387 = bitcast i32 %386 to <2 x bfloat>, !dbg !30
386
+ %388 = extractvalue { i32, i32, i32, i32 } %385, 1, !dbg !30
387
+ %389 = bitcast i32 %388 to <2 x bfloat>, !dbg !30
388
+ %390 = extractvalue { i32, i32, i32, i32 } %385, 2, !dbg !30
389
+ %391 = bitcast i32 %390 to <2 x bfloat>, !dbg !30
390
+ %392 = extractvalue { i32, i32, i32, i32 } %385, 3, !dbg !30
391
+ %393 = bitcast i32 %392 to <2 x bfloat>, !dbg !30
392
+ %394 = extractelement <2 x bfloat> %387, i64 0, !dbg !30
393
+ %395 = extractelement <2 x bfloat> %387, i64 1, !dbg !30
394
+ %396 = extractelement <2 x bfloat> %389, i64 0, !dbg !30
395
+ %397 = extractelement <2 x bfloat> %389, i64 1, !dbg !30
396
+ %398 = extractelement <2 x bfloat> %391, i64 0, !dbg !30
397
+ %399 = extractelement <2 x bfloat> %391, i64 1, !dbg !30
398
+ %400 = extractelement <2 x bfloat> %393, i64 0, !dbg !30
399
+ %401 = extractelement <2 x bfloat> %393, i64 1, !dbg !30
400
+ %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
401
+ %403 = extractvalue { i32, i32, i32, i32 } %402, 0, !dbg !30
402
+ %404 = bitcast i32 %403 to <2 x bfloat>, !dbg !30
403
+ %405 = extractvalue { i32, i32, i32, i32 } %402, 1, !dbg !30
404
+ %406 = bitcast i32 %405 to <2 x bfloat>, !dbg !30
405
+ %407 = extractvalue { i32, i32, i32, i32 } %402, 2, !dbg !30
406
+ %408 = bitcast i32 %407 to <2 x bfloat>, !dbg !30
407
+ %409 = extractvalue { i32, i32, i32, i32 } %402, 3, !dbg !30
408
+ %410 = bitcast i32 %409 to <2 x bfloat>, !dbg !30
409
+ %411 = extractelement <2 x bfloat> %404, i64 0, !dbg !30
410
+ %412 = extractelement <2 x bfloat> %404, i64 1, !dbg !30
411
+ %413 = extractelement <2 x bfloat> %406, i64 0, !dbg !30
412
+ %414 = extractelement <2 x bfloat> %406, i64 1, !dbg !30
413
+ %415 = extractelement <2 x bfloat> %408, i64 0, !dbg !30
414
+ %416 = extractelement <2 x bfloat> %408, i64 1, !dbg !30
415
+ %417 = extractelement <2 x bfloat> %410, i64 0, !dbg !30
416
+ %418 = extractelement <2 x bfloat> %410, i64 1, !dbg !30
417
+ %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
418
+ %420 = extractvalue { i32, i32, i32, i32 } %419, 0, !dbg !30
419
+ %421 = bitcast i32 %420 to <2 x bfloat>, !dbg !30
420
+ %422 = extractvalue { i32, i32, i32, i32 } %419, 1, !dbg !30
421
+ %423 = bitcast i32 %422 to <2 x bfloat>, !dbg !30
422
+ %424 = extractvalue { i32, i32, i32, i32 } %419, 2, !dbg !30
423
+ %425 = bitcast i32 %424 to <2 x bfloat>, !dbg !30
424
+ %426 = extractvalue { i32, i32, i32, i32 } %419, 3, !dbg !30
425
+ %427 = bitcast i32 %426 to <2 x bfloat>, !dbg !30
426
+ %428 = extractelement <2 x bfloat> %421, i64 0, !dbg !30
427
+ %429 = extractelement <2 x bfloat> %421, i64 1, !dbg !30
428
+ %430 = extractelement <2 x bfloat> %423, i64 0, !dbg !30
429
+ %431 = extractelement <2 x bfloat> %423, i64 1, !dbg !30
430
+ %432 = extractelement <2 x bfloat> %425, i64 0, !dbg !30
431
+ %433 = extractelement <2 x bfloat> %425, i64 1, !dbg !30
432
+ %434 = extractelement <2 x bfloat> %427, i64 0, !dbg !30
433
+ %435 = extractelement <2 x bfloat> %427, i64 1, !dbg !30
434
+ %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
435
+ %437 = extractvalue { i32, i32, i32, i32 } %436, 0, !dbg !30
436
+ %438 = bitcast i32 %437 to <2 x bfloat>, !dbg !30
437
+ %439 = extractvalue { i32, i32, i32, i32 } %436, 1, !dbg !30
438
+ %440 = bitcast i32 %439 to <2 x bfloat>, !dbg !30
439
+ %441 = extractvalue { i32, i32, i32, i32 } %436, 2, !dbg !30
440
+ %442 = bitcast i32 %441 to <2 x bfloat>, !dbg !30
441
+ %443 = extractvalue { i32, i32, i32, i32 } %436, 3, !dbg !30
442
+ %444 = bitcast i32 %443 to <2 x bfloat>, !dbg !30
443
+ %445 = extractelement <2 x bfloat> %438, i64 0, !dbg !30
444
+ %446 = extractelement <2 x bfloat> %438, i64 1, !dbg !30
445
+ %447 = extractelement <2 x bfloat> %440, i64 0, !dbg !30
446
+ %448 = extractelement <2 x bfloat> %440, i64 1, !dbg !30
447
+ %449 = extractelement <2 x bfloat> %442, i64 0, !dbg !30
448
+ %450 = extractelement <2 x bfloat> %442, i64 1, !dbg !30
449
+ %451 = extractelement <2 x bfloat> %444, i64 0, !dbg !30
450
+ %452 = extractelement <2 x bfloat> %444, i64 1, !dbg !30
451
+ %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
452
+ %454 = extractvalue { i32, i32, i32, i32 } %453, 0, !dbg !30
453
+ %455 = bitcast i32 %454 to <2 x bfloat>, !dbg !30
454
+ %456 = extractvalue { i32, i32, i32, i32 } %453, 1, !dbg !30
455
+ %457 = bitcast i32 %456 to <2 x bfloat>, !dbg !30
456
+ %458 = extractvalue { i32, i32, i32, i32 } %453, 2, !dbg !30
457
+ %459 = bitcast i32 %458 to <2 x bfloat>, !dbg !30
458
+ %460 = extractvalue { i32, i32, i32, i32 } %453, 3, !dbg !30
459
+ %461 = bitcast i32 %460 to <2 x bfloat>, !dbg !30
460
+ %462 = extractelement <2 x bfloat> %455, i64 0, !dbg !30
461
+ %463 = extractelement <2 x bfloat> %455, i64 1, !dbg !30
462
+ %464 = extractelement <2 x bfloat> %457, i64 0, !dbg !30
463
+ %465 = extractelement <2 x bfloat> %457, i64 1, !dbg !30
464
+ %466 = extractelement <2 x bfloat> %459, i64 0, !dbg !30
465
+ %467 = extractelement <2 x bfloat> %459, i64 1, !dbg !30
466
+ %468 = extractelement <2 x bfloat> %461, i64 0, !dbg !30
467
+ %469 = extractelement <2 x bfloat> %461, i64 1, !dbg !30
468
+ %470 = fpext bfloat %343 to float, !dbg !31
469
+ %471 = fpext bfloat %344 to float, !dbg !31
470
+ %472 = fpext bfloat %345 to float, !dbg !31
471
+ %473 = fpext bfloat %346 to float, !dbg !31
472
+ %474 = fpext bfloat %347 to float, !dbg !31
473
+ %475 = fpext bfloat %348 to float, !dbg !31
474
+ %476 = fpext bfloat %349 to float, !dbg !31
475
+ %477 = fpext bfloat %350 to float, !dbg !31
476
+ %478 = fpext bfloat %360 to float, !dbg !31
477
+ %479 = fpext bfloat %361 to float, !dbg !31
478
+ %480 = fpext bfloat %362 to float, !dbg !31
479
+ %481 = fpext bfloat %363 to float, !dbg !31
480
+ %482 = fpext bfloat %364 to float, !dbg !31
481
+ %483 = fpext bfloat %365 to float, !dbg !31
482
+ %484 = fpext bfloat %366 to float, !dbg !31
483
+ %485 = fpext bfloat %367 to float, !dbg !31
484
+ %486 = fpext bfloat %377 to float, !dbg !31
485
+ %487 = fpext bfloat %378 to float, !dbg !31
486
+ %488 = fpext bfloat %379 to float, !dbg !31
487
+ %489 = fpext bfloat %380 to float, !dbg !31
488
+ %490 = fpext bfloat %381 to float, !dbg !31
489
+ %491 = fpext bfloat %382 to float, !dbg !31
490
+ %492 = fpext bfloat %383 to float, !dbg !31
491
+ %493 = fpext bfloat %384 to float, !dbg !31
492
+ %494 = fpext bfloat %394 to float, !dbg !31
493
+ %495 = fpext bfloat %395 to float, !dbg !31
494
+ %496 = fpext bfloat %396 to float, !dbg !31
495
+ %497 = fpext bfloat %397 to float, !dbg !31
496
+ %498 = fpext bfloat %398 to float, !dbg !31
497
+ %499 = fpext bfloat %399 to float, !dbg !31
498
+ %500 = fpext bfloat %400 to float, !dbg !31
499
+ %501 = fpext bfloat %401 to float, !dbg !31
500
+ %502 = fpext bfloat %411 to float, !dbg !31
501
+ %503 = fpext bfloat %412 to float, !dbg !31
502
+ %504 = fpext bfloat %413 to float, !dbg !31
503
+ %505 = fpext bfloat %414 to float, !dbg !31
504
+ %506 = fpext bfloat %415 to float, !dbg !31
505
+ %507 = fpext bfloat %416 to float, !dbg !31
506
+ %508 = fpext bfloat %417 to float, !dbg !31
507
+ %509 = fpext bfloat %418 to float, !dbg !31
508
+ %510 = fpext bfloat %428 to float, !dbg !31
509
+ %511 = fpext bfloat %429 to float, !dbg !31
510
+ %512 = fpext bfloat %430 to float, !dbg !31
511
+ %513 = fpext bfloat %431 to float, !dbg !31
512
+ %514 = fpext bfloat %432 to float, !dbg !31
513
+ %515 = fpext bfloat %433 to float, !dbg !31
514
+ %516 = fpext bfloat %434 to float, !dbg !31
515
+ %517 = fpext bfloat %435 to float, !dbg !31
516
+ %518 = fpext bfloat %445 to float, !dbg !31
517
+ %519 = fpext bfloat %446 to float, !dbg !31
518
+ %520 = fpext bfloat %447 to float, !dbg !31
519
+ %521 = fpext bfloat %448 to float, !dbg !31
520
+ %522 = fpext bfloat %449 to float, !dbg !31
521
+ %523 = fpext bfloat %450 to float, !dbg !31
522
+ %524 = fpext bfloat %451 to float, !dbg !31
523
+ %525 = fpext bfloat %452 to float, !dbg !31
524
+ %526 = fpext bfloat %462 to float, !dbg !31
525
+ %527 = fpext bfloat %463 to float, !dbg !31
526
+ %528 = fpext bfloat %464 to float, !dbg !31
527
+ %529 = fpext bfloat %465 to float, !dbg !31
528
+ %530 = fpext bfloat %466 to float, !dbg !31
529
+ %531 = fpext bfloat %467 to float, !dbg !31
530
+ %532 = fpext bfloat %468 to float, !dbg !31
531
+ %533 = fpext bfloat %469 to float, !dbg !31
532
+ %534 = fmul float %319, %470, !dbg !32
533
+ %535 = fmul float %319, %471, !dbg !32
534
+ %536 = fmul float %319, %472, !dbg !32
535
+ %537 = fmul float %319, %473, !dbg !32
536
+ %538 = fmul float %319, %474, !dbg !32
537
+ %539 = fmul float %319, %475, !dbg !32
538
+ %540 = fmul float %319, %476, !dbg !32
539
+ %541 = fmul float %319, %477, !dbg !32
540
+ %542 = fmul float %321, %478, !dbg !32
541
+ %543 = fmul float %321, %479, !dbg !32
542
+ %544 = fmul float %321, %480, !dbg !32
543
+ %545 = fmul float %321, %481, !dbg !32
544
+ %546 = fmul float %321, %482, !dbg !32
545
+ %547 = fmul float %321, %483, !dbg !32
546
+ %548 = fmul float %321, %484, !dbg !32
547
+ %549 = fmul float %321, %485, !dbg !32
548
+ %550 = fmul float %323, %486, !dbg !32
549
+ %551 = fmul float %323, %487, !dbg !32
550
+ %552 = fmul float %323, %488, !dbg !32
551
+ %553 = fmul float %323, %489, !dbg !32
552
+ %554 = fmul float %323, %490, !dbg !32
553
+ %555 = fmul float %323, %491, !dbg !32
554
+ %556 = fmul float %323, %492, !dbg !32
555
+ %557 = fmul float %323, %493, !dbg !32
556
+ %558 = fmul float %325, %494, !dbg !32
557
+ %559 = fmul float %325, %495, !dbg !32
558
+ %560 = fmul float %325, %496, !dbg !32
559
+ %561 = fmul float %325, %497, !dbg !32
560
+ %562 = fmul float %325, %498, !dbg !32
561
+ %563 = fmul float %325, %499, !dbg !32
562
+ %564 = fmul float %325, %500, !dbg !32
563
+ %565 = fmul float %325, %501, !dbg !32
564
+ %566 = fmul float %327, %502, !dbg !32
565
+ %567 = fmul float %327, %503, !dbg !32
566
+ %568 = fmul float %327, %504, !dbg !32
567
+ %569 = fmul float %327, %505, !dbg !32
568
+ %570 = fmul float %327, %506, !dbg !32
569
+ %571 = fmul float %327, %507, !dbg !32
570
+ %572 = fmul float %327, %508, !dbg !32
571
+ %573 = fmul float %327, %509, !dbg !32
572
+ %574 = fmul float %329, %510, !dbg !32
573
+ %575 = fmul float %329, %511, !dbg !32
574
+ %576 = fmul float %329, %512, !dbg !32
575
+ %577 = fmul float %329, %513, !dbg !32
576
+ %578 = fmul float %329, %514, !dbg !32
577
+ %579 = fmul float %329, %515, !dbg !32
578
+ %580 = fmul float %329, %516, !dbg !32
579
+ %581 = fmul float %329, %517, !dbg !32
580
+ %582 = fmul float %331, %518, !dbg !32
581
+ %583 = fmul float %331, %519, !dbg !32
582
+ %584 = fmul float %331, %520, !dbg !32
583
+ %585 = fmul float %331, %521, !dbg !32
584
+ %586 = fmul float %331, %522, !dbg !32
585
+ %587 = fmul float %331, %523, !dbg !32
586
+ %588 = fmul float %331, %524, !dbg !32
587
+ %589 = fmul float %331, %525, !dbg !32
588
+ %590 = fmul float %333, %526, !dbg !32
589
+ %591 = fmul float %333, %527, !dbg !32
590
+ %592 = fmul float %333, %528, !dbg !32
591
+ %593 = fmul float %333, %529, !dbg !32
592
+ %594 = fmul float %333, %530, !dbg !32
593
+ %595 = fmul float %333, %531, !dbg !32
594
+ %596 = fmul float %333, %532, !dbg !32
595
+ %597 = fmul float %333, %533, !dbg !32
596
+ %598 = fmul float %254, %470, !dbg !33
597
+ %599 = fmul float %255, %471, !dbg !33
598
+ %600 = fmul float %256, %472, !dbg !33
599
+ %601 = fmul float %257, %473, !dbg !33
600
+ %602 = fmul float %258, %474, !dbg !33
601
+ %603 = fmul float %259, %475, !dbg !33
602
+ %604 = fmul float %260, %476, !dbg !33
603
+ %605 = fmul float %261, %477, !dbg !33
604
+ %606 = fmul float %262, %478, !dbg !33
605
+ %607 = fmul float %263, %479, !dbg !33
606
+ %608 = fmul float %264, %480, !dbg !33
607
+ %609 = fmul float %265, %481, !dbg !33
608
+ %610 = fmul float %266, %482, !dbg !33
609
+ %611 = fmul float %267, %483, !dbg !33
610
+ %612 = fmul float %268, %484, !dbg !33
611
+ %613 = fmul float %269, %485, !dbg !33
612
+ %614 = fmul float %270, %486, !dbg !33
613
+ %615 = fmul float %271, %487, !dbg !33
614
+ %616 = fmul float %272, %488, !dbg !33
615
+ %617 = fmul float %273, %489, !dbg !33
616
+ %618 = fmul float %274, %490, !dbg !33
617
+ %619 = fmul float %275, %491, !dbg !33
618
+ %620 = fmul float %276, %492, !dbg !33
619
+ %621 = fmul float %277, %493, !dbg !33
620
+ %622 = fmul float %278, %494, !dbg !33
621
+ %623 = fmul float %279, %495, !dbg !33
622
+ %624 = fmul float %280, %496, !dbg !33
623
+ %625 = fmul float %281, %497, !dbg !33
624
+ %626 = fmul float %282, %498, !dbg !33
625
+ %627 = fmul float %283, %499, !dbg !33
626
+ %628 = fmul float %284, %500, !dbg !33
627
+ %629 = fmul float %285, %501, !dbg !33
628
+ %630 = fmul float %286, %502, !dbg !33
629
+ %631 = fmul float %287, %503, !dbg !33
630
+ %632 = fmul float %288, %504, !dbg !33
631
+ %633 = fmul float %289, %505, !dbg !33
632
+ %634 = fmul float %290, %506, !dbg !33
633
+ %635 = fmul float %291, %507, !dbg !33
634
+ %636 = fmul float %292, %508, !dbg !33
635
+ %637 = fmul float %293, %509, !dbg !33
636
+ %638 = fmul float %294, %510, !dbg !33
637
+ %639 = fmul float %295, %511, !dbg !33
638
+ %640 = fmul float %296, %512, !dbg !33
639
+ %641 = fmul float %297, %513, !dbg !33
640
+ %642 = fmul float %298, %514, !dbg !33
641
+ %643 = fmul float %299, %515, !dbg !33
642
+ %644 = fmul float %300, %516, !dbg !33
643
+ %645 = fmul float %301, %517, !dbg !33
644
+ %646 = fmul float %302, %518, !dbg !33
645
+ %647 = fmul float %303, %519, !dbg !33
646
+ %648 = fmul float %304, %520, !dbg !33
647
+ %649 = fmul float %305, %521, !dbg !33
648
+ %650 = fmul float %306, %522, !dbg !33
649
+ %651 = fmul float %307, %523, !dbg !33
650
+ %652 = fmul float %308, %524, !dbg !33
651
+ %653 = fmul float %309, %525, !dbg !33
652
+ %654 = fmul float %310, %526, !dbg !33
653
+ %655 = fmul float %311, %527, !dbg !33
654
+ %656 = fmul float %312, %528, !dbg !33
655
+ %657 = fmul float %313, %529, !dbg !33
656
+ %658 = fmul float %314, %530, !dbg !33
657
+ %659 = fmul float %315, %531, !dbg !33
658
+ %660 = fmul float %316, %532, !dbg !33
659
+ %661 = fmul float %317, %533, !dbg !33
660
+ %662 = fadd float %598, %599, !dbg !34
661
+ %663 = fadd float %600, %662, !dbg !34
662
+ %664 = fadd float %601, %663, !dbg !34
663
+ %665 = fadd float %602, %664, !dbg !34
664
+ %666 = fadd float %603, %665, !dbg !34
665
+ %667 = fadd float %604, %666, !dbg !34
666
+ %668 = fadd float %605, %667, !dbg !34
667
+ %669 = fadd float %606, %607, !dbg !34
668
+ %670 = fadd float %608, %669, !dbg !34
669
+ %671 = fadd float %609, %670, !dbg !34
670
+ %672 = fadd float %610, %671, !dbg !34
671
+ %673 = fadd float %611, %672, !dbg !34
672
+ %674 = fadd float %612, %673, !dbg !34
673
+ %675 = fadd float %613, %674, !dbg !34
674
+ %676 = fadd float %614, %615, !dbg !34
675
+ %677 = fadd float %616, %676, !dbg !34
676
+ %678 = fadd float %617, %677, !dbg !34
677
+ %679 = fadd float %618, %678, !dbg !34
678
+ %680 = fadd float %619, %679, !dbg !34
679
+ %681 = fadd float %620, %680, !dbg !34
680
+ %682 = fadd float %621, %681, !dbg !34
681
+ %683 = fadd float %622, %623, !dbg !34
682
+ %684 = fadd float %624, %683, !dbg !34
683
+ %685 = fadd float %625, %684, !dbg !34
684
+ %686 = fadd float %626, %685, !dbg !34
685
+ %687 = fadd float %627, %686, !dbg !34
686
+ %688 = fadd float %628, %687, !dbg !34
687
+ %689 = fadd float %629, %688, !dbg !34
688
+ %690 = fadd float %630, %631, !dbg !34
689
+ %691 = fadd float %632, %690, !dbg !34
690
+ %692 = fadd float %633, %691, !dbg !34
691
+ %693 = fadd float %634, %692, !dbg !34
692
+ %694 = fadd float %635, %693, !dbg !34
693
+ %695 = fadd float %636, %694, !dbg !34
694
+ %696 = fadd float %637, %695, !dbg !34
695
+ %697 = fadd float %638, %639, !dbg !34
696
+ %698 = fadd float %640, %697, !dbg !34
697
+ %699 = fadd float %641, %698, !dbg !34
698
+ %700 = fadd float %642, %699, !dbg !34
699
+ %701 = fadd float %643, %700, !dbg !34
700
+ %702 = fadd float %644, %701, !dbg !34
701
+ %703 = fadd float %645, %702, !dbg !34
702
+ %704 = fadd float %646, %647, !dbg !34
703
+ %705 = fadd float %648, %704, !dbg !34
704
+ %706 = fadd float %649, %705, !dbg !34
705
+ %707 = fadd float %650, %706, !dbg !34
706
+ %708 = fadd float %651, %707, !dbg !34
707
+ %709 = fadd float %652, %708, !dbg !34
708
+ %710 = fadd float %653, %709, !dbg !34
709
+ %711 = fadd float %654, %655, !dbg !34
710
+ %712 = fadd float %656, %711, !dbg !34
711
+ %713 = fadd float %657, %712, !dbg !34
712
+ %714 = fadd float %658, %713, !dbg !34
713
+ %715 = fadd float %659, %714, !dbg !34
714
+ %716 = fadd float %660, %715, !dbg !34
715
+ %717 = fadd float %661, %716, !dbg !34
716
+ %718 = bitcast float %668 to i32, !dbg !37
717
+ %719 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %718, i32 8, i32 31), !dbg !37
718
+ %720 = bitcast i32 %719 to float, !dbg !37
719
+ %721 = fadd float %668, %720, !dbg !34
720
+ %722 = bitcast float %721 to i32, !dbg !37
721
+ %723 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %722, i32 4, i32 31), !dbg !37
722
+ %724 = bitcast i32 %723 to float, !dbg !37
723
+ %725 = fadd float %721, %724, !dbg !34
724
+ %726 = bitcast float %725 to i32, !dbg !37
725
+ %727 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %726, i32 2, i32 31), !dbg !37
726
+ %728 = bitcast i32 %727 to float, !dbg !37
727
+ %729 = fadd float %725, %728, !dbg !34
728
+ %730 = bitcast float %729 to i32, !dbg !37
729
+ %731 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %730, i32 1, i32 31), !dbg !37
730
+ %732 = bitcast i32 %731 to float, !dbg !37
731
+ %733 = fadd float %729, %732, !dbg !34
732
+ %734 = bitcast float %675 to i32, !dbg !37
733
+ %735 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %734, i32 8, i32 31), !dbg !37
734
+ %736 = bitcast i32 %735 to float, !dbg !37
735
+ %737 = fadd float %675, %736, !dbg !34
736
+ %738 = bitcast float %737 to i32, !dbg !37
737
+ %739 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %738, i32 4, i32 31), !dbg !37
738
+ %740 = bitcast i32 %739 to float, !dbg !37
739
+ %741 = fadd float %737, %740, !dbg !34
740
+ %742 = bitcast float %741 to i32, !dbg !37
741
+ %743 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %742, i32 2, i32 31), !dbg !37
742
+ %744 = bitcast i32 %743 to float, !dbg !37
743
+ %745 = fadd float %741, %744, !dbg !34
744
+ %746 = bitcast float %745 to i32, !dbg !37
745
+ %747 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %746, i32 1, i32 31), !dbg !37
746
+ %748 = bitcast i32 %747 to float, !dbg !37
747
+ %749 = fadd float %745, %748, !dbg !34
748
+ %750 = bitcast float %682 to i32, !dbg !37
749
+ %751 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %750, i32 8, i32 31), !dbg !37
750
+ %752 = bitcast i32 %751 to float, !dbg !37
751
+ %753 = fadd float %682, %752, !dbg !34
752
+ %754 = bitcast float %753 to i32, !dbg !37
753
+ %755 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %754, i32 4, i32 31), !dbg !37
754
+ %756 = bitcast i32 %755 to float, !dbg !37
755
+ %757 = fadd float %753, %756, !dbg !34
756
+ %758 = bitcast float %757 to i32, !dbg !37
757
+ %759 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %758, i32 2, i32 31), !dbg !37
758
+ %760 = bitcast i32 %759 to float, !dbg !37
759
+ %761 = fadd float %757, %760, !dbg !34
760
+ %762 = bitcast float %761 to i32, !dbg !37
761
+ %763 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %762, i32 1, i32 31), !dbg !37
762
+ %764 = bitcast i32 %763 to float, !dbg !37
763
+ %765 = fadd float %761, %764, !dbg !34
764
+ %766 = bitcast float %689 to i32, !dbg !37
765
+ %767 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %766, i32 8, i32 31), !dbg !37
766
+ %768 = bitcast i32 %767 to float, !dbg !37
767
+ %769 = fadd float %689, %768, !dbg !34
768
+ %770 = bitcast float %769 to i32, !dbg !37
769
+ %771 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %770, i32 4, i32 31), !dbg !37
770
+ %772 = bitcast i32 %771 to float, !dbg !37
771
+ %773 = fadd float %769, %772, !dbg !34
772
+ %774 = bitcast float %773 to i32, !dbg !37
773
+ %775 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %774, i32 2, i32 31), !dbg !37
774
+ %776 = bitcast i32 %775 to float, !dbg !37
775
+ %777 = fadd float %773, %776, !dbg !34
776
+ %778 = bitcast float %777 to i32, !dbg !37
777
+ %779 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %778, i32 1, i32 31), !dbg !37
778
+ %780 = bitcast i32 %779 to float, !dbg !37
779
+ %781 = fadd float %777, %780, !dbg !34
780
+ %782 = bitcast float %696 to i32, !dbg !37
781
+ %783 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %782, i32 8, i32 31), !dbg !37
782
+ %784 = bitcast i32 %783 to float, !dbg !37
783
+ %785 = fadd float %696, %784, !dbg !34
784
+ %786 = bitcast float %785 to i32, !dbg !37
785
+ %787 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %786, i32 4, i32 31), !dbg !37
786
+ %788 = bitcast i32 %787 to float, !dbg !37
787
+ %789 = fadd float %785, %788, !dbg !34
788
+ %790 = bitcast float %789 to i32, !dbg !37
789
+ %791 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %790, i32 2, i32 31), !dbg !37
790
+ %792 = bitcast i32 %791 to float, !dbg !37
791
+ %793 = fadd float %789, %792, !dbg !34
792
+ %794 = bitcast float %793 to i32, !dbg !37
793
+ %795 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %794, i32 1, i32 31), !dbg !37
794
+ %796 = bitcast i32 %795 to float, !dbg !37
795
+ %797 = fadd float %793, %796, !dbg !34
796
+ %798 = bitcast float %703 to i32, !dbg !37
797
+ %799 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %798, i32 8, i32 31), !dbg !37
798
+ %800 = bitcast i32 %799 to float, !dbg !37
799
+ %801 = fadd float %703, %800, !dbg !34
800
+ %802 = bitcast float %801 to i32, !dbg !37
801
+ %803 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %802, i32 4, i32 31), !dbg !37
802
+ %804 = bitcast i32 %803 to float, !dbg !37
803
+ %805 = fadd float %801, %804, !dbg !34
804
+ %806 = bitcast float %805 to i32, !dbg !37
805
+ %807 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %806, i32 2, i32 31), !dbg !37
806
+ %808 = bitcast i32 %807 to float, !dbg !37
807
+ %809 = fadd float %805, %808, !dbg !34
808
+ %810 = bitcast float %809 to i32, !dbg !37
809
+ %811 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %810, i32 1, i32 31), !dbg !37
810
+ %812 = bitcast i32 %811 to float, !dbg !37
811
+ %813 = fadd float %809, %812, !dbg !34
812
+ %814 = bitcast float %710 to i32, !dbg !37
813
+ %815 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %814, i32 8, i32 31), !dbg !37
814
+ %816 = bitcast i32 %815 to float, !dbg !37
815
+ %817 = fadd float %710, %816, !dbg !34
816
+ %818 = bitcast float %817 to i32, !dbg !37
817
+ %819 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %818, i32 4, i32 31), !dbg !37
818
+ %820 = bitcast i32 %819 to float, !dbg !37
819
+ %821 = fadd float %817, %820, !dbg !34
820
+ %822 = bitcast float %821 to i32, !dbg !37
821
+ %823 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %822, i32 2, i32 31), !dbg !37
822
+ %824 = bitcast i32 %823 to float, !dbg !37
823
+ %825 = fadd float %821, %824, !dbg !34
824
+ %826 = bitcast float %825 to i32, !dbg !37
825
+ %827 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %826, i32 1, i32 31), !dbg !37
826
+ %828 = bitcast i32 %827 to float, !dbg !37
827
+ %829 = fadd float %825, %828, !dbg !34
828
+ %830 = bitcast float %717 to i32, !dbg !37
829
+ %831 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %830, i32 8, i32 31), !dbg !37
830
+ %832 = bitcast i32 %831 to float, !dbg !37
831
+ %833 = fadd float %717, %832, !dbg !34
832
+ %834 = bitcast float %833 to i32, !dbg !37
833
+ %835 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %834, i32 4, i32 31), !dbg !37
834
+ %836 = bitcast i32 %835 to float, !dbg !37
835
+ %837 = fadd float %833, %836, !dbg !34
836
+ %838 = bitcast float %837 to i32, !dbg !37
837
+ %839 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %838, i32 2, i32 31), !dbg !37
838
+ %840 = bitcast i32 %839 to float, !dbg !37
839
+ %841 = fadd float %837, %840, !dbg !34
840
+ %842 = bitcast float %841 to i32, !dbg !37
841
+ %843 = tail call i32 @llvm.nvvm.shfl.sync.bfly.i32(i32 -1, i32 %842, i32 1, i32 31), !dbg !37
842
+ %844 = bitcast i32 %843 to float, !dbg !37
843
+ %845 = fadd float %841, %844, !dbg !34
844
+ %846 = fmul float %733, %254, !dbg !40
845
+ %847 = fmul float %733, %255, !dbg !40
846
+ %848 = fmul float %733, %256, !dbg !40
847
+ %849 = fmul float %733, %257, !dbg !40
848
+ %850 = fmul float %733, %258, !dbg !40
849
+ %851 = fmul float %733, %259, !dbg !40
850
+ %852 = fmul float %733, %260, !dbg !40
851
+ %853 = fmul float %733, %261, !dbg !40
852
+ %854 = fmul float %749, %262, !dbg !40
853
+ %855 = fmul float %749, %263, !dbg !40
854
+ %856 = fmul float %749, %264, !dbg !40
855
+ %857 = fmul float %749, %265, !dbg !40
856
+ %858 = fmul float %749, %266, !dbg !40
857
+ %859 = fmul float %749, %267, !dbg !40
858
+ %860 = fmul float %749, %268, !dbg !40
859
+ %861 = fmul float %749, %269, !dbg !40
860
+ %862 = fmul float %765, %270, !dbg !40
861
+ %863 = fmul float %765, %271, !dbg !40
862
+ %864 = fmul float %765, %272, !dbg !40
863
+ %865 = fmul float %765, %273, !dbg !40
864
+ %866 = fmul float %765, %274, !dbg !40
865
+ %867 = fmul float %765, %275, !dbg !40
866
+ %868 = fmul float %765, %276, !dbg !40
867
+ %869 = fmul float %765, %277, !dbg !40
868
+ %870 = fmul float %781, %278, !dbg !40
869
+ %871 = fmul float %781, %279, !dbg !40
870
+ %872 = fmul float %781, %280, !dbg !40
871
+ %873 = fmul float %781, %281, !dbg !40
872
+ %874 = fmul float %781, %282, !dbg !40
873
+ %875 = fmul float %781, %283, !dbg !40
874
+ %876 = fmul float %781, %284, !dbg !40
875
+ %877 = fmul float %781, %285, !dbg !40
876
+ %878 = fmul float %797, %286, !dbg !40
877
+ %879 = fmul float %797, %287, !dbg !40
878
+ %880 = fmul float %797, %288, !dbg !40
879
+ %881 = fmul float %797, %289, !dbg !40
880
+ %882 = fmul float %797, %290, !dbg !40
881
+ %883 = fmul float %797, %291, !dbg !40
882
+ %884 = fmul float %797, %292, !dbg !40
883
+ %885 = fmul float %797, %293, !dbg !40
884
+ %886 = fmul float %813, %294, !dbg !40
885
+ %887 = fmul float %813, %295, !dbg !40
886
+ %888 = fmul float %813, %296, !dbg !40
887
+ %889 = fmul float %813, %297, !dbg !40
888
+ %890 = fmul float %813, %298, !dbg !40
889
+ %891 = fmul float %813, %299, !dbg !40
890
+ %892 = fmul float %813, %300, !dbg !40
891
+ %893 = fmul float %813, %301, !dbg !40
892
+ %894 = fmul float %829, %302, !dbg !40
893
+ %895 = fmul float %829, %303, !dbg !40
894
+ %896 = fmul float %829, %304, !dbg !40
895
+ %897 = fmul float %829, %305, !dbg !40
896
+ %898 = fmul float %829, %306, !dbg !40
897
+ %899 = fmul float %829, %307, !dbg !40
898
+ %900 = fmul float %829, %308, !dbg !40
899
+ %901 = fmul float %829, %309, !dbg !40
900
+ %902 = fmul float %845, %310, !dbg !40
901
+ %903 = fmul float %845, %311, !dbg !40
902
+ %904 = fmul float %845, %312, !dbg !40
903
+ %905 = fmul float %845, %313, !dbg !40
904
+ %906 = fmul float %845, %314, !dbg !40
905
+ %907 = fmul float %845, %315, !dbg !40
906
+ %908 = fmul float %845, %316, !dbg !40
907
+ %909 = fmul float %845, %317, !dbg !40
908
+ %910 = fmul float %846, %319, !dbg !41
909
+ %911 = fmul float %847, %319, !dbg !41
910
+ %912 = fmul float %848, %319, !dbg !41
911
+ %913 = fmul float %849, %319, !dbg !41
912
+ %914 = fmul float %850, %319, !dbg !41
913
+ %915 = fmul float %851, %319, !dbg !41
914
+ %916 = fmul float %852, %319, !dbg !41
915
+ %917 = fmul float %853, %319, !dbg !41
916
+ %918 = fmul float %854, %321, !dbg !41
917
+ %919 = fmul float %855, %321, !dbg !41
918
+ %920 = fmul float %856, %321, !dbg !41
919
+ %921 = fmul float %857, %321, !dbg !41
920
+ %922 = fmul float %858, %321, !dbg !41
921
+ %923 = fmul float %859, %321, !dbg !41
922
+ %924 = fmul float %860, %321, !dbg !41
923
+ %925 = fmul float %861, %321, !dbg !41
924
+ %926 = fmul float %862, %323, !dbg !41
925
+ %927 = fmul float %863, %323, !dbg !41
926
+ %928 = fmul float %864, %323, !dbg !41
927
+ %929 = fmul float %865, %323, !dbg !41
928
+ %930 = fmul float %866, %323, !dbg !41
929
+ %931 = fmul float %867, %323, !dbg !41
930
+ %932 = fmul float %868, %323, !dbg !41
931
+ %933 = fmul float %869, %323, !dbg !41
932
+ %934 = fmul float %870, %325, !dbg !41
933
+ %935 = fmul float %871, %325, !dbg !41
934
+ %936 = fmul float %872, %325, !dbg !41
935
+ %937 = fmul float %873, %325, !dbg !41
936
+ %938 = fmul float %874, %325, !dbg !41
937
+ %939 = fmul float %875, %325, !dbg !41
938
+ %940 = fmul float %876, %325, !dbg !41
939
+ %941 = fmul float %877, %325, !dbg !41
940
+ %942 = fmul float %878, %327, !dbg !41
941
+ %943 = fmul float %879, %327, !dbg !41
942
+ %944 = fmul float %880, %327, !dbg !41
943
+ %945 = fmul float %881, %327, !dbg !41
944
+ %946 = fmul float %882, %327, !dbg !41
945
+ %947 = fmul float %883, %327, !dbg !41
946
+ %948 = fmul float %884, %327, !dbg !41
947
+ %949 = fmul float %885, %327, !dbg !41
948
+ %950 = fmul float %886, %329, !dbg !41
949
+ %951 = fmul float %887, %329, !dbg !41
950
+ %952 = fmul float %888, %329, !dbg !41
951
+ %953 = fmul float %889, %329, !dbg !41
952
+ %954 = fmul float %890, %329, !dbg !41
953
+ %955 = fmul float %891, %329, !dbg !41
954
+ %956 = fmul float %892, %329, !dbg !41
955
+ %957 = fmul float %893, %329, !dbg !41
956
+ %958 = fmul float %894, %331, !dbg !41
957
+ %959 = fmul float %895, %331, !dbg !41
958
+ %960 = fmul float %896, %331, !dbg !41
959
+ %961 = fmul float %897, %331, !dbg !41
960
+ %962 = fmul float %898, %331, !dbg !41
961
+ %963 = fmul float %899, %331, !dbg !41
962
+ %964 = fmul float %900, %331, !dbg !41
963
+ %965 = fmul float %901, %331, !dbg !41
964
+ %966 = fmul float %902, %333, !dbg !41
965
+ %967 = fmul float %903, %333, !dbg !41
966
+ %968 = fmul float %904, %333, !dbg !41
967
+ %969 = fmul float %905, %333, !dbg !41
968
+ %970 = fmul float %906, %333, !dbg !41
969
+ %971 = fmul float %907, %333, !dbg !41
970
+ %972 = fmul float %908, %333, !dbg !41
971
+ %973 = fmul float %909, %333, !dbg !41
972
+ %974 = fsub float %534, %910, !dbg !42
973
+ %975 = fsub float %535, %911, !dbg !42
974
+ %976 = fsub float %536, %912, !dbg !42
975
+ %977 = fsub float %537, %913, !dbg !42
976
+ %978 = fsub float %538, %914, !dbg !42
977
+ %979 = fsub float %539, %915, !dbg !42
978
+ %980 = fsub float %540, %916, !dbg !42
979
+ %981 = fsub float %541, %917, !dbg !42
980
+ %982 = fsub float %542, %918, !dbg !42
981
+ %983 = fsub float %543, %919, !dbg !42
982
+ %984 = fsub float %544, %920, !dbg !42
983
+ %985 = fsub float %545, %921, !dbg !42
984
+ %986 = fsub float %546, %922, !dbg !42
985
+ %987 = fsub float %547, %923, !dbg !42
986
+ %988 = fsub float %548, %924, !dbg !42
987
+ %989 = fsub float %549, %925, !dbg !42
988
+ %990 = fsub float %550, %926, !dbg !42
989
+ %991 = fsub float %551, %927, !dbg !42
990
+ %992 = fsub float %552, %928, !dbg !42
991
+ %993 = fsub float %553, %929, !dbg !42
992
+ %994 = fsub float %554, %930, !dbg !42
993
+ %995 = fsub float %555, %931, !dbg !42
994
+ %996 = fsub float %556, %932, !dbg !42
995
+ %997 = fsub float %557, %933, !dbg !42
996
+ %998 = fsub float %558, %934, !dbg !42
997
+ %999 = fsub float %559, %935, !dbg !42
998
+ %1000 = fsub float %560, %936, !dbg !42
999
+ %1001 = fsub float %561, %937, !dbg !42
1000
+ %1002 = fsub float %562, %938, !dbg !42
1001
+ %1003 = fsub float %563, %939, !dbg !42
1002
+ %1004 = fsub float %564, %940, !dbg !42
1003
+ %1005 = fsub float %565, %941, !dbg !42
1004
+ %1006 = fsub float %566, %942, !dbg !42
1005
+ %1007 = fsub float %567, %943, !dbg !42
1006
+ %1008 = fsub float %568, %944, !dbg !42
1007
+ %1009 = fsub float %569, %945, !dbg !42
1008
+ %1010 = fsub float %570, %946, !dbg !42
1009
+ %1011 = fsub float %571, %947, !dbg !42
1010
+ %1012 = fsub float %572, %948, !dbg !42
1011
+ %1013 = fsub float %573, %949, !dbg !42
1012
+ %1014 = fsub float %574, %950, !dbg !42
1013
+ %1015 = fsub float %575, %951, !dbg !42
1014
+ %1016 = fsub float %576, %952, !dbg !42
1015
+ %1017 = fsub float %577, %953, !dbg !42
1016
+ %1018 = fsub float %578, %954, !dbg !42
1017
+ %1019 = fsub float %579, %955, !dbg !42
1018
+ %1020 = fsub float %580, %956, !dbg !42
1019
+ %1021 = fsub float %581, %957, !dbg !42
1020
+ %1022 = insertelement <2 x float> poison, float %582, i64 0, !dbg !42
1021
+ %1023 = insertelement <2 x float> %1022, float %583, i64 1, !dbg !42
1022
+ %1024 = insertelement <2 x float> poison, float %958, i64 0, !dbg !42
1023
+ %1025 = insertelement <2 x float> %1024, float %959, i64 1, !dbg !42
1024
+ %1026 = fsub <2 x float> %1023, %1025, !dbg !42
1025
+ %1027 = insertelement <2 x float> poison, float %584, i64 0, !dbg !42
1026
+ %1028 = insertelement <2 x float> %1027, float %585, i64 1, !dbg !42
1027
+ %1029 = insertelement <2 x float> poison, float %960, i64 0, !dbg !42
1028
+ %1030 = insertelement <2 x float> %1029, float %961, i64 1, !dbg !42
1029
+ %1031 = fsub <2 x float> %1028, %1030, !dbg !42
1030
+ %1032 = insertelement <2 x float> poison, float %586, i64 0, !dbg !42
1031
+ %1033 = insertelement <2 x float> %1032, float %587, i64 1, !dbg !42
1032
+ %1034 = insertelement <2 x float> poison, float %962, i64 0, !dbg !42
1033
+ %1035 = insertelement <2 x float> %1034, float %963, i64 1, !dbg !42
1034
+ %1036 = fsub <2 x float> %1033, %1035, !dbg !42
1035
+ %1037 = insertelement <2 x float> poison, float %588, i64 0, !dbg !42
1036
+ %1038 = insertelement <2 x float> %1037, float %589, i64 1, !dbg !42
1037
+ %1039 = insertelement <2 x float> poison, float %964, i64 0, !dbg !42
1038
+ %1040 = insertelement <2 x float> %1039, float %965, i64 1, !dbg !42
1039
+ %1041 = fsub <2 x float> %1038, %1040, !dbg !42
1040
+ %1042 = insertelement <2 x float> poison, float %590, i64 0, !dbg !42
1041
+ %1043 = insertelement <2 x float> %1042, float %591, i64 1, !dbg !42
1042
+ %1044 = insertelement <2 x float> poison, float %966, i64 0, !dbg !42
1043
+ %1045 = insertelement <2 x float> %1044, float %967, i64 1, !dbg !42
1044
+ %1046 = fsub <2 x float> %1043, %1045, !dbg !42
1045
+ %1047 = insertelement <2 x float> poison, float %592, i64 0, !dbg !42
1046
+ %1048 = insertelement <2 x float> %1047, float %593, i64 1, !dbg !42
1047
+ %1049 = insertelement <2 x float> poison, float %968, i64 0, !dbg !42
1048
+ %1050 = insertelement <2 x float> %1049, float %969, i64 1, !dbg !42
1049
+ %1051 = fsub <2 x float> %1048, %1050, !dbg !42
1050
+ %1052 = insertelement <2 x float> poison, float %594, i64 0, !dbg !42
1051
+ %1053 = insertelement <2 x float> %1052, float %595, i64 1, !dbg !42
1052
+ %1054 = insertelement <2 x float> poison, float %970, i64 0, !dbg !42
1053
+ %1055 = insertelement <2 x float> %1054, float %971, i64 1, !dbg !42
1054
+ %1056 = fsub <2 x float> %1053, %1055, !dbg !42
1055
+ %1057 = insertelement <2 x float> poison, float %596, i64 0, !dbg !42
1056
+ %1058 = insertelement <2 x float> %1057, float %597, i64 1, !dbg !42
1057
+ %1059 = insertelement <2 x float> poison, float %972, i64 0, !dbg !42
1058
+ %1060 = insertelement <2 x float> %1059, float %973, i64 1, !dbg !42
1059
+ %1061 = fsub <2 x float> %1058, %1060, !dbg !42
1060
+ %1062 = fptrunc float %974 to bfloat, !dbg !43
1061
+ %1063 = fptrunc float %975 to bfloat, !dbg !43
1062
+ %1064 = fptrunc float %976 to bfloat, !dbg !43
1063
+ %1065 = fptrunc float %977 to bfloat, !dbg !43
1064
+ %1066 = fptrunc float %978 to bfloat, !dbg !43
1065
+ %1067 = fptrunc float %979 to bfloat, !dbg !43
1066
+ %1068 = fptrunc float %980 to bfloat, !dbg !43
1067
+ %1069 = fptrunc float %981 to bfloat, !dbg !43
1068
+ %1070 = fptrunc float %982 to bfloat, !dbg !43
1069
+ %1071 = fptrunc float %983 to bfloat, !dbg !43
1070
+ %1072 = fptrunc float %984 to bfloat, !dbg !43
1071
+ %1073 = fptrunc float %985 to bfloat, !dbg !43
1072
+ %1074 = fptrunc float %986 to bfloat, !dbg !43
1073
+ %1075 = fptrunc float %987 to bfloat, !dbg !43
1074
+ %1076 = fptrunc float %988 to bfloat, !dbg !43
1075
+ %1077 = fptrunc float %989 to bfloat, !dbg !43
1076
+ %1078 = fptrunc float %990 to bfloat, !dbg !43
1077
+ %1079 = fptrunc float %991 to bfloat, !dbg !43
1078
+ %1080 = fptrunc float %992 to bfloat, !dbg !43
1079
+ %1081 = fptrunc float %993 to bfloat, !dbg !43
1080
+ %1082 = fptrunc float %994 to bfloat, !dbg !43
1081
+ %1083 = fptrunc float %995 to bfloat, !dbg !43
1082
+ %1084 = fptrunc float %996 to bfloat, !dbg !43
1083
+ %1085 = fptrunc float %997 to bfloat, !dbg !43
1084
+ %1086 = fptrunc float %998 to bfloat, !dbg !43
1085
+ %1087 = fptrunc float %999 to bfloat, !dbg !43
1086
+ %1088 = fptrunc float %1000 to bfloat, !dbg !43
1087
+ %1089 = fptrunc float %1001 to bfloat, !dbg !43
1088
+ %1090 = fptrunc float %1002 to bfloat, !dbg !43
1089
+ %1091 = fptrunc float %1003 to bfloat, !dbg !43
1090
+ %1092 = fptrunc float %1004 to bfloat, !dbg !43
1091
+ %1093 = fptrunc float %1005 to bfloat, !dbg !43
1092
+ %1094 = fptrunc float %1006 to bfloat, !dbg !43
1093
+ %1095 = fptrunc float %1007 to bfloat, !dbg !43
1094
+ %1096 = fptrunc float %1008 to bfloat, !dbg !43
1095
+ %1097 = fptrunc float %1009 to bfloat, !dbg !43
1096
+ %1098 = fptrunc float %1010 to bfloat, !dbg !43
1097
+ %1099 = fptrunc float %1011 to bfloat, !dbg !43
1098
+ %1100 = fptrunc float %1012 to bfloat, !dbg !43
1099
+ %1101 = fptrunc float %1013 to bfloat, !dbg !43
1100
+ %1102 = fptrunc float %1014 to bfloat, !dbg !43
1101
+ %1103 = fptrunc float %1015 to bfloat, !dbg !43
1102
+ %1104 = fptrunc float %1016 to bfloat, !dbg !43
1103
+ %1105 = fptrunc float %1017 to bfloat, !dbg !43
1104
+ %1106 = fptrunc float %1018 to bfloat, !dbg !43
1105
+ %1107 = fptrunc float %1019 to bfloat, !dbg !43
1106
+ %1108 = fptrunc float %1020 to bfloat, !dbg !43
1107
+ %1109 = fptrunc float %1021 to bfloat, !dbg !43
1108
+ %1110 = fptrunc <2 x float> %1026 to <2 x bfloat>, !dbg !43
1109
+ %1111 = fptrunc <2 x float> %1031 to <2 x bfloat>, !dbg !43
1110
+ %1112 = fptrunc <2 x float> %1036 to <2 x bfloat>, !dbg !43
1111
+ %1113 = fptrunc <2 x float> %1041 to <2 x bfloat>, !dbg !43
1112
+ %1114 = fptrunc <2 x float> %1046 to <2 x bfloat>, !dbg !43
1113
+ %1115 = fptrunc <2 x float> %1051 to <2 x bfloat>, !dbg !43
1114
+ %1116 = fptrunc <2 x float> %1056 to <2 x bfloat>, !dbg !43
1115
+ %1117 = fptrunc <2 x float> %1061 to <2 x bfloat>, !dbg !43
1116
+ %1118 = insertelement <2 x bfloat> poison, bfloat %1062, i64 0, !dbg !44
1117
+ %1119 = insertelement <2 x bfloat> %1118, bfloat %1063, i64 1, !dbg !44
1118
+ %1120 = bitcast <2 x bfloat> %1119 to i32, !dbg !44
1119
+ %1121 = insertelement <2 x bfloat> poison, bfloat %1064, i64 0, !dbg !44
1120
+ %1122 = insertelement <2 x bfloat> %1121, bfloat %1065, i64 1, !dbg !44
1121
+ %1123 = bitcast <2 x bfloat> %1122 to i32, !dbg !44
1122
+ %1124 = insertelement <2 x bfloat> poison, bfloat %1066, i64 0, !dbg !44
1123
+ %1125 = insertelement <2 x bfloat> %1124, bfloat %1067, i64 1, !dbg !44
1124
+ %1126 = bitcast <2 x bfloat> %1125 to i32, !dbg !44
1125
+ %1127 = insertelement <2 x bfloat> poison, bfloat %1068, i64 0, !dbg !44
1126
+ %1128 = insertelement <2 x bfloat> %1127, bfloat %1069, i64 1, !dbg !44
1127
+ %1129 = bitcast <2 x bfloat> %1128 to i32, !dbg !44
1128
+ 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
1129
+ %1130 = insertelement <2 x bfloat> poison, bfloat %1070, i64 0, !dbg !44
1130
+ %1131 = insertelement <2 x bfloat> %1130, bfloat %1071, i64 1, !dbg !44
1131
+ %1132 = bitcast <2 x bfloat> %1131 to i32, !dbg !44
1132
+ %1133 = insertelement <2 x bfloat> poison, bfloat %1072, i64 0, !dbg !44
1133
+ %1134 = insertelement <2 x bfloat> %1133, bfloat %1073, i64 1, !dbg !44
1134
+ %1135 = bitcast <2 x bfloat> %1134 to i32, !dbg !44
1135
+ %1136 = insertelement <2 x bfloat> poison, bfloat %1074, i64 0, !dbg !44
1136
+ %1137 = insertelement <2 x bfloat> %1136, bfloat %1075, i64 1, !dbg !44
1137
+ %1138 = bitcast <2 x bfloat> %1137 to i32, !dbg !44
1138
+ %1139 = insertelement <2 x bfloat> poison, bfloat %1076, i64 0, !dbg !44
1139
+ %1140 = insertelement <2 x bfloat> %1139, bfloat %1077, i64 1, !dbg !44
1140
+ %1141 = bitcast <2 x bfloat> %1140 to i32, !dbg !44
1141
+ 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
+ %1142 = insertelement <2 x bfloat> poison, bfloat %1078, i64 0, !dbg !44
1143
+ %1143 = insertelement <2 x bfloat> %1142, bfloat %1079, i64 1, !dbg !44
1144
+ %1144 = bitcast <2 x bfloat> %1143 to i32, !dbg !44
1145
+ %1145 = insertelement <2 x bfloat> poison, bfloat %1080, i64 0, !dbg !44
1146
+ %1146 = insertelement <2 x bfloat> %1145, bfloat %1081, i64 1, !dbg !44
1147
+ %1147 = bitcast <2 x bfloat> %1146 to i32, !dbg !44
1148
+ %1148 = insertelement <2 x bfloat> poison, bfloat %1082, i64 0, !dbg !44
1149
+ %1149 = insertelement <2 x bfloat> %1148, bfloat %1083, i64 1, !dbg !44
1150
+ %1150 = bitcast <2 x bfloat> %1149 to i32, !dbg !44
1151
+ %1151 = insertelement <2 x bfloat> poison, bfloat %1084, i64 0, !dbg !44
1152
+ %1152 = insertelement <2 x bfloat> %1151, bfloat %1085, i64 1, !dbg !44
1153
+ %1153 = bitcast <2 x bfloat> %1152 to i32, !dbg !44
1154
+ 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
1155
+ %1154 = insertelement <2 x bfloat> poison, bfloat %1086, i64 0, !dbg !44
1156
+ %1155 = insertelement <2 x bfloat> %1154, bfloat %1087, i64 1, !dbg !44
1157
+ %1156 = bitcast <2 x bfloat> %1155 to i32, !dbg !44
1158
+ %1157 = insertelement <2 x bfloat> poison, bfloat %1088, i64 0, !dbg !44
1159
+ %1158 = insertelement <2 x bfloat> %1157, bfloat %1089, i64 1, !dbg !44
1160
+ %1159 = bitcast <2 x bfloat> %1158 to i32, !dbg !44
1161
+ %1160 = insertelement <2 x bfloat> poison, bfloat %1090, i64 0, !dbg !44
1162
+ %1161 = insertelement <2 x bfloat> %1160, bfloat %1091, i64 1, !dbg !44
1163
+ %1162 = bitcast <2 x bfloat> %1161 to i32, !dbg !44
1164
+ %1163 = insertelement <2 x bfloat> poison, bfloat %1092, i64 0, !dbg !44
1165
+ %1164 = insertelement <2 x bfloat> %1163, bfloat %1093, i64 1, !dbg !44
1166
+ %1165 = bitcast <2 x bfloat> %1164 to i32, !dbg !44
1167
+ 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
1168
+ %1166 = insertelement <2 x bfloat> poison, bfloat %1094, i64 0, !dbg !44
1169
+ %1167 = insertelement <2 x bfloat> %1166, bfloat %1095, i64 1, !dbg !44
1170
+ %1168 = bitcast <2 x bfloat> %1167 to i32, !dbg !44
1171
+ %1169 = insertelement <2 x bfloat> poison, bfloat %1096, i64 0, !dbg !44
1172
+ %1170 = insertelement <2 x bfloat> %1169, bfloat %1097, i64 1, !dbg !44
1173
+ %1171 = bitcast <2 x bfloat> %1170 to i32, !dbg !44
1174
+ %1172 = insertelement <2 x bfloat> poison, bfloat %1098, i64 0, !dbg !44
1175
+ %1173 = insertelement <2 x bfloat> %1172, bfloat %1099, i64 1, !dbg !44
1176
+ %1174 = bitcast <2 x bfloat> %1173 to i32, !dbg !44
1177
+ %1175 = insertelement <2 x bfloat> poison, bfloat %1100, i64 0, !dbg !44
1178
+ %1176 = insertelement <2 x bfloat> %1175, bfloat %1101, i64 1, !dbg !44
1179
+ %1177 = bitcast <2 x bfloat> %1176 to i32, !dbg !44
1180
+ 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
1181
+ %1178 = insertelement <2 x bfloat> poison, bfloat %1102, i64 0, !dbg !44
1182
+ %1179 = insertelement <2 x bfloat> %1178, bfloat %1103, i64 1, !dbg !44
1183
+ %1180 = bitcast <2 x bfloat> %1179 to i32, !dbg !44
1184
+ %1181 = insertelement <2 x bfloat> poison, bfloat %1104, i64 0, !dbg !44
1185
+ %1182 = insertelement <2 x bfloat> %1181, bfloat %1105, i64 1, !dbg !44
1186
+ %1183 = bitcast <2 x bfloat> %1182 to i32, !dbg !44
1187
+ %1184 = insertelement <2 x bfloat> poison, bfloat %1106, i64 0, !dbg !44
1188
+ %1185 = insertelement <2 x bfloat> %1184, bfloat %1107, i64 1, !dbg !44
1189
+ %1186 = bitcast <2 x bfloat> %1185 to i32, !dbg !44
1190
+ %1187 = insertelement <2 x bfloat> poison, bfloat %1108, i64 0, !dbg !44
1191
+ %1188 = insertelement <2 x bfloat> %1187, bfloat %1109, i64 1, !dbg !44
1192
+ %1189 = bitcast <2 x bfloat> %1188 to i32, !dbg !44
1193
+ 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
1194
+ %1190 = bitcast <2 x bfloat> %1110 to i32, !dbg !44
1195
+ %1191 = bitcast <2 x bfloat> %1111 to i32, !dbg !44
1196
+ %1192 = bitcast <2 x bfloat> %1112 to i32, !dbg !44
1197
+ %1193 = bitcast <2 x bfloat> %1113 to i32, !dbg !44
1198
+ 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
1199
+ %1194 = bitcast <2 x bfloat> %1114 to i32, !dbg !44
1200
+ %1195 = bitcast <2 x bfloat> %1115 to i32, !dbg !44
1201
+ %1196 = bitcast <2 x bfloat> %1116 to i32, !dbg !44
1202
+ %1197 = bitcast <2 x bfloat> %1117 to i32, !dbg !44
1203
+ 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
1204
+ ret void, !dbg !45
1205
+ }
1206
+
1207
+ ; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
1208
+ declare noundef range(i32 0, 2147483647) i32 @llvm.nvvm.read.ptx.sreg.ctaid.x() #1
1209
+
1210
+ ; Function Attrs: mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none)
1211
+ declare noundef range(i32 0, 1024) i32 @llvm.nvvm.read.ptx.sreg.tid.x() #1
1212
+
1213
+ ; Function Attrs: convergent nocallback nounwind memory(inaccessiblemem: readwrite)
1214
+ declare i32 @llvm.nvvm.shfl.sync.bfly.i32(i32, i32, i32, i32) #2
1215
+
1216
+ attributes #0 = { nounwind "nvvm.reqntid"="256" }
1217
+ attributes #1 = { mustprogress nocallback nofree nosync nounwind speculatable willreturn memory(none) }
1218
+ attributes #2 = { convergent nocallback nounwind memory(inaccessiblemem: readwrite) }
1219
+ attributes #3 = { nounwind }
1220
+
1221
+ !llvm.dbg.cu = !{!0}
1222
+ !llvm.module.flags = !{!2, !3}
1223
+
1224
+ !0 = distinct !DICompileUnit(language: DW_LANG_C, file: !1, producer: "triton", isOptimized: true, runtimeVersion: 0, emissionKind: LineTablesOnly)
1225
+ !1 = !DIFile(filename: "l2norm.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm")
1226
+ !2 = !{i32 2, !"Debug Info Version", i32 3}
1227
+ !3 = !{i32 4, !"nvvm-reflect-ftz", i32 1}
1228
+ !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)
1229
+ !5 = !DISubroutineType(cc: DW_CC_normal, types: !6)
1230
+ !6 = !{null, !7, !9, !7, !7, !10, !11, !7, !7}
1231
+ !7 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !8, size: 64, dwarfAddressSpace: 1)
1232
+ !8 = !DIBasicType(name: "unknown_type", encoding: DW_ATE_signed)
1233
+ !9 = !DIDerivedType(tag: DW_TAG_pointer_type, name: "pointer", baseType: !10, size: 64, dwarfAddressSpace: 1)
1234
+ !10 = !DIBasicType(name: "float", size: 32, encoding: DW_ATE_float)
1235
+ !11 = !DIBasicType(name: "int", size: 32, encoding: DW_ATE_signed)
1236
+ !12 = !DILocation(line: 94, column: 24, scope: !4)
1237
+ !13 = !DILocation(line: 94, column: 30, scope: !4)
1238
+ !14 = !DILocation(line: 95, column: 16, scope: !4)
1239
+ !15 = !DILocation(line: 95, column: 34, scope: !4)
1240
+ !16 = !DILocation(line: 95, column: 21, scope: !4)
1241
+ !17 = !DILocation(line: 97, column: 16, scope: !4)
1242
+ !18 = !DILocation(line: 98, column: 30, scope: !4)
1243
+ !19 = !DILocation(line: 99, column: 29, scope: !4)
1244
+ !20 = !DILocation(line: 99, column: 14, scope: !4)
1245
+ !21 = !DILocation(line: 99, column: 33, scope: !4)
1246
+ !22 = !DILocation(line: 100, column: 20, scope: !4)
1247
+ !23 = !DILocation(line: 101, column: 16, scope: !4)
1248
+ !24 = !DILocation(line: 101, column: 35, scope: !4)
1249
+ !25 = !DILocation(line: 102, column: 16, scope: !4)
1250
+ !26 = !DILocation(line: 102, column: 35, scope: !4)
1251
+ !27 = !DILocation(line: 104, column: 18, scope: !4)
1252
+ !28 = !DILocation(line: 104, column: 47, scope: !4)
1253
+ !29 = !DILocation(line: 105, column: 21, scope: !4)
1254
+ !30 = !DILocation(line: 106, column: 19, scope: !4)
1255
+ !31 = !DILocation(line: 106, column: 49, scope: !4)
1256
+ !32 = !DILocation(line: 107, column: 18, scope: !4)
1257
+ !33 = !DILocation(line: 107, column: 50, scope: !4)
1258
+ !34 = !DILocation(line: 263, column: 15, scope: !35, inlinedAt: !37)
1259
+ !35 = distinct !DILexicalBlockFile(scope: !4, file: !36, discriminator: 0)
1260
+ !36 = !DIFile(filename: "standard.py", directory: "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language")
1261
+ !37 = !DILocation(line: 293, column: 36, scope: !35, inlinedAt: !38)
1262
+ !38 = !DILocation(line: 107, column: 55, scope: !39)
1263
+ !39 = distinct !DILexicalBlockFile(scope: !4, file: !1, discriminator: 0)
1264
+ !40 = !DILocation(line: 107, column: 69, scope: !4)
1265
+ !41 = !DILocation(line: 107, column: 75, scope: !4)
1266
+ !42 = !DILocation(line: 107, column: 36, scope: !4)
1267
+ !43 = !DILocation(line: 108, column: 27, scope: !4)
1268
+ !44 = !DILocation(line: 108, column: 19, scope: !4)
1269
+ !45 = !DILocation(line: 108, column: 4, scope: !4)
triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ptx ADDED
@@ -0,0 +1,1322 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ //
2
+ // Generated by LLVM NVPTX Back-End
3
+ //
4
+
5
+ .version 9.1
6
+ .target sm_100a
7
+ .address_size 64
8
+
9
+ // .globl l2norm_bwd_kernel // -- Begin function l2norm_bwd_kernel
10
+ // @l2norm_bwd_kernel
11
+ .visible .entry l2norm_bwd_kernel(
12
+ .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_0,
13
+ .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_1,
14
+ .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_2,
15
+ .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_3,
16
+ .param .f32 l2norm_bwd_kernel_param_4,
17
+ .param .u32 l2norm_bwd_kernel_param_5,
18
+ .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_6,
19
+ .param .u64 .ptr .global .align 1 l2norm_bwd_kernel_param_7
20
+ )
21
+ .reqntid 256
22
+ {
23
+ .reg .pred %p<9>;
24
+ .reg .b16 %rs<129>;
25
+ .reg .b32 %r<631>;
26
+ .reg .b64 %rd<120>;
27
+ .loc 1 93 0 // l2norm.py:93:0
28
+ $L__func_begin0:
29
+ .loc 1 93 0 // l2norm.py:93:0
30
+
31
+ // %bb.0:
32
+ ld.param.b64 %rd33, [l2norm_bwd_kernel_param_0];
33
+ ld.param.b64 %rd34, [l2norm_bwd_kernel_param_1];
34
+ $L__tmp0:
35
+ .loc 1 94 24 // l2norm.py:94:24
36
+ mov.u32 %r106, %ctaid.x;
37
+ ld.param.b64 %rd35, [l2norm_bwd_kernel_param_2];
38
+ .loc 1 95 16 // l2norm.py:95:16
39
+ mul.wide.u32 %rd36, %r106, 128;
40
+ ld.param.b64 %rd37, [l2norm_bwd_kernel_param_3];
41
+ .loc 1 95 34 // l2norm.py:95:34
42
+ mov.u32 %r107, %tid.x;
43
+ shr.u32 %r108, %r107, 4;
44
+ ld.param.s32 %rd38, [l2norm_bwd_kernel_param_5];
45
+ bfe.u32 %r109, %r107, 4, 4;
46
+ or.b32 %r110, %r109, 16;
47
+ or.b32 %r111, %r109, 32;
48
+ or.b32 %r112, %r108, 48;
49
+ .loc 1 95 21 // l2norm.py:95:21
50
+ or.b32 %r113, %r109, 64;
51
+ or.b32 %r114, %r109, 80;
52
+ or.b32 %r115, %r109, 96;
53
+ or.b32 %r116, %r108, 112;
54
+ cvt.u64.u32 %rd39, %r116;
55
+ cvt.u64.u32 %rd40, %r115;
56
+ cvt.u64.u32 %rd41, %r114;
57
+ cvt.u64.u32 %rd42, %r113;
58
+ cvt.u64.u32 %rd43, %r112;
59
+ cvt.u64.u32 %rd44, %r111;
60
+ cvt.u64.u32 %rd45, %r110;
61
+ cvt.u64.u32 %rd46, %r109;
62
+ or.b64 %rd47, %rd36, %rd46;
63
+ or.b64 %rd48, %rd36, %rd45;
64
+ or.b64 %rd49, %rd36, %rd44;
65
+ or.b64 %rd50, %rd36, %rd43;
66
+ or.b64 %rd51, %rd36, %rd42;
67
+ or.b64 %rd52, %rd36, %rd41;
68
+ or.b64 %rd53, %rd36, %rd40;
69
+ or.b64 %rd54, %rd36, %rd39;
70
+ .loc 1 97 16 // l2norm.py:97:16
71
+ setp.lt.s64 %p8, %rd54, %rd38;
72
+ setp.lt.s64 %p7, %rd53, %rd38;
73
+ setp.lt.s64 %p6, %rd52, %rd38;
74
+ setp.lt.s64 %p5, %rd51, %rd38;
75
+ setp.lt.s64 %p4, %rd50, %rd38;
76
+ setp.lt.s64 %p3, %rd49, %rd38;
77
+ setp.lt.s64 %p2, %rd48, %rd38;
78
+ setp.lt.s64 %p1, %rd47, %rd38;
79
+ .loc 1 98 30 // l2norm.py:98:30
80
+ shl.b32 %r117, %r107, 3;
81
+ and.b32 %r118, %r117, 120;
82
+ .loc 1 99 14 // l2norm.py:99:14
83
+ shl.b64 %rd55, %rd47, 8;
84
+ add.s64 %rd56, %rd33, %rd55;
85
+ shl.b64 %rd57, %rd48, 8;
86
+ add.s64 %rd58, %rd33, %rd57;
87
+ shl.b64 %rd59, %rd49, 8;
88
+ add.s64 %rd60, %rd33, %rd59;
89
+ shl.b64 %rd61, %rd50, 8;
90
+ add.s64 %rd62, %rd33, %rd61;
91
+ shl.b64 %rd63, %rd51, 8;
92
+ add.s64 %rd64, %rd33, %rd63;
93
+ shl.b64 %rd65, %rd52, 8;
94
+ add.s64 %rd66, %rd33, %rd65;
95
+ shl.b64 %rd67, %rd53, 8;
96
+ add.s64 %rd68, %rd33, %rd67;
97
+ shl.b64 %rd69, %rd54, 8;
98
+ add.s64 %rd70, %rd33, %rd69;
99
+ .loc 1 99 33 // l2norm.py:99:33
100
+ mul.wide.u32 %rd71, %r118, 2;
101
+ add.s64 %rd1, %rd56, %rd71;
102
+ add.s64 %rd2, %rd58, %rd71;
103
+ add.s64 %rd3, %rd60, %rd71;
104
+ add.s64 %rd4, %rd62, %rd71;
105
+ add.s64 %rd5, %rd64, %rd71;
106
+ add.s64 %rd6, %rd66, %rd71;
107
+ add.s64 %rd7, %rd68, %rd71;
108
+ add.s64 %rd8, %rd70, %rd71;
109
+ .loc 1 100 20 // l2norm.py:100:20
110
+ shl.b64 %rd72, %rd47, 2;
111
+ add.s64 %rd9, %rd34, %rd72;
112
+ shl.b64 %rd73, %rd48, 2;
113
+ add.s64 %rd10, %rd34, %rd73;
114
+ shl.b64 %rd74, %rd49, 2;
115
+ add.s64 %rd11, %rd34, %rd74;
116
+ shl.b64 %rd75, %rd50, 2;
117
+ add.s64 %rd12, %rd34, %rd75;
118
+ shl.b64 %rd76, %rd51, 2;
119
+ add.s64 %rd13, %rd34, %rd76;
120
+ shl.b64 %rd77, %rd52, 2;
121
+ add.s64 %rd14, %rd34, %rd77;
122
+ shl.b64 %rd78, %rd53, 2;
123
+ add.s64 %rd15, %rd34, %rd78;
124
+ shl.b64 %rd79, %rd54, 2;
125
+ add.s64 %rd16, %rd34, %rd79;
126
+ .loc 1 101 16 // l2norm.py:101:16
127
+ add.s64 %rd80, %rd35, %rd55;
128
+ add.s64 %rd81, %rd35, %rd57;
129
+ add.s64 %rd82, %rd35, %rd59;
130
+ add.s64 %rd83, %rd35, %rd61;
131
+ add.s64 %rd84, %rd35, %rd63;
132
+ add.s64 %rd85, %rd35, %rd65;
133
+ add.s64 %rd86, %rd35, %rd67;
134
+ add.s64 %rd87, %rd35, %rd69;
135
+ .loc 1 101 35 // l2norm.py:101:35
136
+ add.s64 %rd17, %rd80, %rd71;
137
+ add.s64 %rd18, %rd81, %rd71;
138
+ add.s64 %rd19, %rd82, %rd71;
139
+ add.s64 %rd20, %rd83, %rd71;
140
+ add.s64 %rd21, %rd84, %rd71;
141
+ add.s64 %rd22, %rd85, %rd71;
142
+ add.s64 %rd23, %rd86, %rd71;
143
+ add.s64 %rd24, %rd87, %rd71;
144
+ .loc 1 102 16 // l2norm.py:102:16
145
+ add.s64 %rd88, %rd37, %rd55;
146
+ add.s64 %rd89, %rd37, %rd57;
147
+ add.s64 %rd90, %rd37, %rd59;
148
+ add.s64 %rd91, %rd37, %rd61;
149
+ add.s64 %rd92, %rd37, %rd63;
150
+ add.s64 %rd93, %rd37, %rd65;
151
+ add.s64 %rd94, %rd37, %rd67;
152
+ add.s64 %rd95, %rd37, %rd69;
153
+ .loc 1 102 35 // l2norm.py:102:35
154
+ add.s64 %rd25, %rd88, %rd71;
155
+ add.s64 %rd26, %rd89, %rd71;
156
+ add.s64 %rd27, %rd90, %rd71;
157
+ add.s64 %rd28, %rd91, %rd71;
158
+ add.s64 %rd29, %rd92, %rd71;
159
+ add.s64 %rd30, %rd93, %rd71;
160
+ add.s64 %rd31, %rd94, %rd71;
161
+ add.s64 %rd32, %rd95, %rd71;
162
+ mov.b32 %r5, 0;
163
+ .loc 1 104 18 // l2norm.py:104:18
164
+ // begin inline asm
165
+ mov.u32 %r1, %r5;
166
+ mov.u32 %r2, %r5;
167
+ mov.u32 %r3, %r5;
168
+ mov.u32 %r4, %r5;
169
+ @%p1 ld.global.v4.b32 { %r1, %r2, %r3, %r4 }, [ %rd1 + 0 ];
170
+ // end inline asm
171
+ mov.b32 {%rs1, %rs2}, %r1;
172
+ mov.b32 {%rs3, %rs4}, %r2;
173
+ mov.b32 {%rs5, %rs6}, %r3;
174
+ mov.b32 {%rs7, %rs8}, %r4;
175
+ // begin inline asm
176
+ mov.u32 %r6, %r5;
177
+ mov.u32 %r7, %r5;
178
+ mov.u32 %r8, %r5;
179
+ mov.u32 %r9, %r5;
180
+ @%p2 ld.global.v4.b32 { %r6, %r7, %r8, %r9 }, [ %rd2 + 0 ];
181
+ // end inline asm
182
+ mov.b32 {%rs9, %rs10}, %r6;
183
+ mov.b32 {%rs11, %rs12}, %r7;
184
+ mov.b32 {%rs13, %rs14}, %r8;
185
+ mov.b32 {%rs15, %rs16}, %r9;
186
+ // begin inline asm
187
+ mov.u32 %r10, %r5;
188
+ mov.u32 %r11, %r5;
189
+ mov.u32 %r12, %r5;
190
+ mov.u32 %r13, %r5;
191
+ @%p3 ld.global.v4.b32 { %r10, %r11, %r12, %r13 }, [ %rd3 + 0 ];
192
+ // end inline asm
193
+ mov.b32 {%rs17, %rs18}, %r10;
194
+ mov.b32 {%rs19, %rs20}, %r11;
195
+ mov.b32 {%rs21, %rs22}, %r12;
196
+ mov.b32 {%rs23, %rs24}, %r13;
197
+ // begin inline asm
198
+ mov.u32 %r14, %r5;
199
+ mov.u32 %r15, %r5;
200
+ mov.u32 %r16, %r5;
201
+ mov.u32 %r17, %r5;
202
+ @%p4 ld.global.v4.b32 { %r14, %r15, %r16, %r17 }, [ %rd4 + 0 ];
203
+ // end inline asm
204
+ mov.b32 {%rs25, %rs26}, %r14;
205
+ mov.b32 {%rs27, %rs28}, %r15;
206
+ mov.b32 {%rs29, %rs30}, %r16;
207
+ mov.b32 {%rs31, %rs32}, %r17;
208
+ // begin inline asm
209
+ mov.u32 %r18, %r5;
210
+ mov.u32 %r19, %r5;
211
+ mov.u32 %r20, %r5;
212
+ mov.u32 %r21, %r5;
213
+ @%p5 ld.global.v4.b32 { %r18, %r19, %r20, %r21 }, [ %rd5 + 0 ];
214
+ // end inline asm
215
+ mov.b32 {%rs33, %rs34}, %r18;
216
+ mov.b32 {%rs35, %rs36}, %r19;
217
+ mov.b32 {%rs37, %rs38}, %r20;
218
+ mov.b32 {%rs39, %rs40}, %r21;
219
+ // begin inline asm
220
+ mov.u32 %r22, %r5;
221
+ mov.u32 %r23, %r5;
222
+ mov.u32 %r24, %r5;
223
+ mov.u32 %r25, %r5;
224
+ @%p6 ld.global.v4.b32 { %r22, %r23, %r24, %r25 }, [ %rd6 + 0 ];
225
+ // end inline asm
226
+ mov.b32 {%rs41, %rs42}, %r22;
227
+ mov.b32 {%rs43, %rs44}, %r23;
228
+ mov.b32 {%rs45, %rs46}, %r24;
229
+ mov.b32 {%rs47, %rs48}, %r25;
230
+ // begin inline asm
231
+ mov.u32 %r26, %r5;
232
+ mov.u32 %r27, %r5;
233
+ mov.u32 %r28, %r5;
234
+ mov.u32 %r29, %r5;
235
+ @%p7 ld.global.v4.b32 { %r26, %r27, %r28, %r29 }, [ %rd7 + 0 ];
236
+ // end inline asm
237
+ mov.b32 {%rs49, %rs50}, %r26;
238
+ mov.b32 {%rs51, %rs52}, %r27;
239
+ mov.b32 {%rs53, %rs54}, %r28;
240
+ mov.b32 {%rs55, %rs56}, %r29;
241
+ // begin inline asm
242
+ mov.u32 %r30, %r5;
243
+ mov.u32 %r31, %r5;
244
+ mov.u32 %r32, %r5;
245
+ mov.u32 %r33, %r5;
246
+ @%p8 ld.global.v4.b32 { %r30, %r31, %r32, %r33 }, [ %rd8 + 0 ];
247
+ // end inline asm
248
+ mov.b32 {%rs57, %rs58}, %r30;
249
+ mov.b32 {%rs59, %rs60}, %r31;
250
+ mov.b32 {%rs61, %rs62}, %r32;
251
+ mov.b32 {%rs63, %rs64}, %r33;
252
+ .loc 1 104 47 // l2norm.py:104:47
253
+ cvt.f32.bf16 %r119, %rs1;
254
+ cvt.f32.bf16 %r120, %rs2;
255
+ cvt.f32.bf16 %r121, %rs3;
256
+ cvt.f32.bf16 %r122, %rs4;
257
+ cvt.f32.bf16 %r123, %rs5;
258
+ cvt.f32.bf16 %r124, %rs6;
259
+ cvt.f32.bf16 %r125, %rs7;
260
+ cvt.f32.bf16 %r126, %rs8;
261
+ cvt.f32.bf16 %r127, %rs9;
262
+ cvt.f32.bf16 %r128, %rs10;
263
+ cvt.f32.bf16 %r129, %rs11;
264
+ cvt.f32.bf16 %r130, %rs12;
265
+ cvt.f32.bf16 %r131, %rs13;
266
+ cvt.f32.bf16 %r132, %rs14;
267
+ cvt.f32.bf16 %r133, %rs15;
268
+ cvt.f32.bf16 %r134, %rs16;
269
+ cvt.f32.bf16 %r135, %rs17;
270
+ cvt.f32.bf16 %r136, %rs18;
271
+ cvt.f32.bf16 %r137, %rs19;
272
+ cvt.f32.bf16 %r138, %rs20;
273
+ cvt.f32.bf16 %r139, %rs21;
274
+ cvt.f32.bf16 %r140, %rs22;
275
+ cvt.f32.bf16 %r141, %rs23;
276
+ cvt.f32.bf16 %r142, %rs24;
277
+ cvt.f32.bf16 %r143, %rs25;
278
+ cvt.f32.bf16 %r144, %rs26;
279
+ cvt.f32.bf16 %r145, %rs27;
280
+ cvt.f32.bf16 %r146, %rs28;
281
+ cvt.f32.bf16 %r147, %rs29;
282
+ cvt.f32.bf16 %r148, %rs30;
283
+ cvt.f32.bf16 %r149, %rs31;
284
+ cvt.f32.bf16 %r150, %rs32;
285
+ cvt.f32.bf16 %r151, %rs33;
286
+ cvt.f32.bf16 %r152, %rs34;
287
+ cvt.f32.bf16 %r153, %rs35;
288
+ cvt.f32.bf16 %r154, %rs36;
289
+ cvt.f32.bf16 %r155, %rs37;
290
+ cvt.f32.bf16 %r156, %rs38;
291
+ cvt.f32.bf16 %r157, %rs39;
292
+ cvt.f32.bf16 %r158, %rs40;
293
+ cvt.f32.bf16 %r159, %rs41;
294
+ cvt.f32.bf16 %r160, %rs42;
295
+ cvt.f32.bf16 %r161, %rs43;
296
+ cvt.f32.bf16 %r162, %rs44;
297
+ cvt.f32.bf16 %r163, %rs45;
298
+ cvt.f32.bf16 %r164, %rs46;
299
+ cvt.f32.bf16 %r165, %rs47;
300
+ cvt.f32.bf16 %r166, %rs48;
301
+ cvt.f32.bf16 %r167, %rs49;
302
+ cvt.f32.bf16 %r168, %rs50;
303
+ cvt.f32.bf16 %r169, %rs51;
304
+ cvt.f32.bf16 %r170, %rs52;
305
+ cvt.f32.bf16 %r171, %rs53;
306
+ cvt.f32.bf16 %r172, %rs54;
307
+ cvt.f32.bf16 %r173, %rs55;
308
+ cvt.f32.bf16 %r174, %rs56;
309
+ cvt.f32.bf16 %r175, %rs57;
310
+ cvt.f32.bf16 %r176, %rs58;
311
+ cvt.f32.bf16 %r177, %rs59;
312
+ cvt.f32.bf16 %r178, %rs60;
313
+ cvt.f32.bf16 %r179, %rs61;
314
+ cvt.f32.bf16 %r180, %rs62;
315
+ cvt.f32.bf16 %r181, %rs63;
316
+ cvt.f32.bf16 %r182, %rs64;
317
+ .loc 1 105 21 // l2norm.py:105:21
318
+ // begin inline asm
319
+ mov.u32 %r34, %r5;
320
+ @%p1 ld.global.b32 { %r34 }, [ %rd9 + 0 ];
321
+ // end inline asm
322
+ // begin inline asm
323
+ mov.u32 %r35, %r5;
324
+ @%p2 ld.global.b32 { %r35 }, [ %rd10 + 0 ];
325
+ // end inline asm
326
+ // begin inline asm
327
+ mov.u32 %r36, %r5;
328
+ @%p3 ld.global.b32 { %r36 }, [ %rd11 + 0 ];
329
+ // end inline asm
330
+ // begin inline asm
331
+ mov.u32 %r37, %r5;
332
+ @%p4 ld.global.b32 { %r37 }, [ %rd12 + 0 ];
333
+ // end inline asm
334
+ // begin inline asm
335
+ mov.u32 %r38, %r5;
336
+ @%p5 ld.global.b32 { %r38 }, [ %rd13 + 0 ];
337
+ // end inline asm
338
+ // begin inline asm
339
+ mov.u32 %r39, %r5;
340
+ @%p6 ld.global.b32 { %r39 }, [ %rd14 + 0 ];
341
+ // end inline asm
342
+ // begin inline asm
343
+ mov.u32 %r40, %r5;
344
+ @%p7 ld.global.b32 { %r40 }, [ %rd15 + 0 ];
345
+ // end inline asm
346
+ // begin inline asm
347
+ mov.u32 %r41, %r5;
348
+ @%p8 ld.global.b32 { %r41 }, [ %rd16 + 0 ];
349
+ // end inline asm
350
+ .loc 1 106 19 // l2norm.py:106:19
351
+ // begin inline asm
352
+ mov.u32 %r42, %r5;
353
+ mov.u32 %r43, %r5;
354
+ mov.u32 %r44, %r5;
355
+ mov.u32 %r45, %r5;
356
+ @%p1 ld.global.v4.b32 { %r42, %r43, %r44, %r45 }, [ %rd17 + 0 ];
357
+ // end inline asm
358
+ mov.b32 {%rs65, %rs66}, %r42;
359
+ mov.b32 {%rs67, %rs68}, %r43;
360
+ mov.b32 {%rs69, %rs70}, %r44;
361
+ mov.b32 {%rs71, %rs72}, %r45;
362
+ // begin inline asm
363
+ mov.u32 %r46, %r5;
364
+ mov.u32 %r47, %r5;
365
+ mov.u32 %r48, %r5;
366
+ mov.u32 %r49, %r5;
367
+ @%p2 ld.global.v4.b32 { %r46, %r47, %r48, %r49 }, [ %rd18 + 0 ];
368
+ // end inline asm
369
+ mov.b32 {%rs73, %rs74}, %r46;
370
+ mov.b32 {%rs75, %rs76}, %r47;
371
+ mov.b32 {%rs77, %rs78}, %r48;
372
+ mov.b32 {%rs79, %rs80}, %r49;
373
+ // begin inline asm
374
+ mov.u32 %r50, %r5;
375
+ mov.u32 %r51, %r5;
376
+ mov.u32 %r52, %r5;
377
+ mov.u32 %r53, %r5;
378
+ @%p3 ld.global.v4.b32 { %r50, %r51, %r52, %r53 }, [ %rd19 + 0 ];
379
+ // end inline asm
380
+ mov.b32 {%rs81, %rs82}, %r50;
381
+ mov.b32 {%rs83, %rs84}, %r51;
382
+ mov.b32 {%rs85, %rs86}, %r52;
383
+ mov.b32 {%rs87, %rs88}, %r53;
384
+ // begin inline asm
385
+ mov.u32 %r54, %r5;
386
+ mov.u32 %r55, %r5;
387
+ mov.u32 %r56, %r5;
388
+ mov.u32 %r57, %r5;
389
+ @%p4 ld.global.v4.b32 { %r54, %r55, %r56, %r57 }, [ %rd20 + 0 ];
390
+ // end inline asm
391
+ mov.b32 {%rs89, %rs90}, %r54;
392
+ mov.b32 {%rs91, %rs92}, %r55;
393
+ mov.b32 {%rs93, %rs94}, %r56;
394
+ mov.b32 {%rs95, %rs96}, %r57;
395
+ // begin inline asm
396
+ mov.u32 %r58, %r5;
397
+ mov.u32 %r59, %r5;
398
+ mov.u32 %r60, %r5;
399
+ mov.u32 %r61, %r5;
400
+ @%p5 ld.global.v4.b32 { %r58, %r59, %r60, %r61 }, [ %rd21 + 0 ];
401
+ // end inline asm
402
+ mov.b32 {%rs97, %rs98}, %r58;
403
+ mov.b32 {%rs99, %rs100}, %r59;
404
+ mov.b32 {%rs101, %rs102}, %r60;
405
+ mov.b32 {%rs103, %rs104}, %r61;
406
+ // begin inline asm
407
+ mov.u32 %r62, %r5;
408
+ mov.u32 %r63, %r5;
409
+ mov.u32 %r64, %r5;
410
+ mov.u32 %r65, %r5;
411
+ @%p6 ld.global.v4.b32 { %r62, %r63, %r64, %r65 }, [ %rd22 + 0 ];
412
+ // end inline asm
413
+ mov.b32 {%rs105, %rs106}, %r62;
414
+ mov.b32 {%rs107, %rs108}, %r63;
415
+ mov.b32 {%rs109, %rs110}, %r64;
416
+ mov.b32 {%rs111, %rs112}, %r65;
417
+ // begin inline asm
418
+ mov.u32 %r66, %r5;
419
+ mov.u32 %r67, %r5;
420
+ mov.u32 %r68, %r5;
421
+ mov.u32 %r69, %r5;
422
+ @%p7 ld.global.v4.b32 { %r66, %r67, %r68, %r69 }, [ %rd23 + 0 ];
423
+ // end inline asm
424
+ mov.b32 {%rs113, %rs114}, %r66;
425
+ mov.b32 {%rs115, %rs116}, %r67;
426
+ mov.b32 {%rs117, %rs118}, %r68;
427
+ mov.b32 {%rs119, %rs120}, %r69;
428
+ // begin inline asm
429
+ mov.u32 %r70, %r5;
430
+ mov.u32 %r71, %r5;
431
+ mov.u32 %r72, %r5;
432
+ mov.u32 %r73, %r5;
433
+ @%p8 ld.global.v4.b32 { %r70, %r71, %r72, %r73 }, [ %rd24 + 0 ];
434
+ // end inline asm
435
+ mov.b32 {%rs121, %rs122}, %r70;
436
+ mov.b32 {%rs123, %rs124}, %r71;
437
+ mov.b32 {%rs125, %rs126}, %r72;
438
+ mov.b32 {%rs127, %rs128}, %r73;
439
+ .loc 1 106 49 // l2norm.py:106:49
440
+ cvt.f32.bf16 %r183, %rs65;
441
+ cvt.f32.bf16 %r184, %rs66;
442
+ cvt.f32.bf16 %r185, %rs67;
443
+ cvt.f32.bf16 %r186, %rs68;
444
+ cvt.f32.bf16 %r187, %rs69;
445
+ cvt.f32.bf16 %r188, %rs70;
446
+ cvt.f32.bf16 %r189, %rs71;
447
+ cvt.f32.bf16 %r190, %rs72;
448
+ cvt.f32.bf16 %r191, %rs73;
449
+ cvt.f32.bf16 %r192, %rs74;
450
+ cvt.f32.bf16 %r193, %rs75;
451
+ cvt.f32.bf16 %r194, %rs76;
452
+ cvt.f32.bf16 %r195, %rs77;
453
+ cvt.f32.bf16 %r196, %rs78;
454
+ cvt.f32.bf16 %r197, %rs79;
455
+ cvt.f32.bf16 %r198, %rs80;
456
+ cvt.f32.bf16 %r199, %rs81;
457
+ cvt.f32.bf16 %r200, %rs82;
458
+ cvt.f32.bf16 %r201, %rs83;
459
+ cvt.f32.bf16 %r202, %rs84;
460
+ cvt.f32.bf16 %r203, %rs85;
461
+ cvt.f32.bf16 %r204, %rs86;
462
+ cvt.f32.bf16 %r205, %rs87;
463
+ cvt.f32.bf16 %r206, %rs88;
464
+ cvt.f32.bf16 %r207, %rs89;
465
+ cvt.f32.bf16 %r208, %rs90;
466
+ cvt.f32.bf16 %r209, %rs91;
467
+ cvt.f32.bf16 %r210, %rs92;
468
+ cvt.f32.bf16 %r211, %rs93;
469
+ cvt.f32.bf16 %r212, %rs94;
470
+ cvt.f32.bf16 %r213, %rs95;
471
+ cvt.f32.bf16 %r214, %rs96;
472
+ cvt.f32.bf16 %r215, %rs97;
473
+ cvt.f32.bf16 %r216, %rs98;
474
+ cvt.f32.bf16 %r217, %rs99;
475
+ cvt.f32.bf16 %r218, %rs100;
476
+ cvt.f32.bf16 %r219, %rs101;
477
+ cvt.f32.bf16 %r220, %rs102;
478
+ cvt.f32.bf16 %r221, %rs103;
479
+ cvt.f32.bf16 %r222, %rs104;
480
+ cvt.f32.bf16 %r223, %rs105;
481
+ cvt.f32.bf16 %r224, %rs106;
482
+ cvt.f32.bf16 %r225, %rs107;
483
+ cvt.f32.bf16 %r226, %rs108;
484
+ cvt.f32.bf16 %r227, %rs109;
485
+ cvt.f32.bf16 %r228, %rs110;
486
+ cvt.f32.bf16 %r229, %rs111;
487
+ cvt.f32.bf16 %r230, %rs112;
488
+ cvt.f32.bf16 %r231, %rs113;
489
+ cvt.f32.bf16 %r232, %rs114;
490
+ cvt.f32.bf16 %r233, %rs115;
491
+ cvt.f32.bf16 %r234, %rs116;
492
+ cvt.f32.bf16 %r235, %rs117;
493
+ cvt.f32.bf16 %r236, %rs118;
494
+ cvt.f32.bf16 %r237, %rs119;
495
+ cvt.f32.bf16 %r238, %rs120;
496
+ cvt.f32.bf16 %r239, %rs121;
497
+ cvt.f32.bf16 %r240, %rs122;
498
+ cvt.f32.bf16 %r241, %rs123;
499
+ cvt.f32.bf16 %r242, %rs124;
500
+ cvt.f32.bf16 %r243, %rs125;
501
+ cvt.f32.bf16 %r244, %rs126;
502
+ cvt.f32.bf16 %r245, %rs127;
503
+ cvt.f32.bf16 %r246, %rs128;
504
+ .loc 1 107 18 // l2norm.py:107:18
505
+ mul.f32 %r247, %r40, %r231;
506
+ mul.f32 %r248, %r40, %r232;
507
+ mul.f32 %r249, %r40, %r233;
508
+ mul.f32 %r250, %r40, %r234;
509
+ mul.f32 %r251, %r40, %r235;
510
+ mul.f32 %r252, %r40, %r236;
511
+ mul.f32 %r253, %r40, %r237;
512
+ mul.f32 %r254, %r40, %r238;
513
+ mul.f32 %r255, %r41, %r239;
514
+ mul.f32 %r256, %r41, %r240;
515
+ mul.f32 %r257, %r41, %r241;
516
+ mul.f32 %r258, %r41, %r242;
517
+ mul.f32 %r259, %r41, %r243;
518
+ mul.f32 %r260, %r41, %r244;
519
+ mul.f32 %r261, %r41, %r245;
520
+ mul.f32 %r262, %r41, %r246;
521
+ .loc 1 107 50 // l2norm.py:107:50
522
+ mul.f32 %r263, %r120, %r184;
523
+ mul.f32 %r264, %r128, %r192;
524
+ mul.f32 %r265, %r136, %r200;
525
+ mul.f32 %r266, %r144, %r208;
526
+ mul.f32 %r267, %r152, %r216;
527
+ mul.f32 %r268, %r160, %r224;
528
+ mul.f32 %r269, %r168, %r232;
529
+ mul.f32 %r270, %r176, %r240;
530
+ $L__tmp1:
531
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
532
+ fma.rn.f32.bf16 %r271, %rs1, %rs65, %r263;
533
+ fma.rn.f32.bf16 %r272, %rs3, %rs67, %r271;
534
+ fma.rn.f32.bf16 %r273, %rs4, %rs68, %r272;
535
+ fma.rn.f32.bf16 %r274, %rs5, %rs69, %r273;
536
+ fma.rn.f32.bf16 %r275, %rs6, %rs70, %r274;
537
+ fma.rn.f32.bf16 %r276, %rs7, %rs71, %r275;
538
+ fma.rn.f32.bf16 %r277, %rs8, %rs72, %r276;
539
+ fma.rn.f32.bf16 %r278, %rs9, %rs73, %r264;
540
+ fma.rn.f32.bf16 %r279, %rs11, %rs75, %r278;
541
+ fma.rn.f32.bf16 %r280, %rs12, %rs76, %r279;
542
+ fma.rn.f32.bf16 %r281, %rs13, %rs77, %r280;
543
+ fma.rn.f32.bf16 %r282, %rs14, %rs78, %r281;
544
+ fma.rn.f32.bf16 %r283, %rs15, %rs79, %r282;
545
+ fma.rn.f32.bf16 %r284, %rs16, %rs80, %r283;
546
+ fma.rn.f32.bf16 %r285, %rs17, %rs81, %r265;
547
+ fma.rn.f32.bf16 %r286, %rs19, %rs83, %r285;
548
+ fma.rn.f32.bf16 %r287, %rs20, %rs84, %r286;
549
+ fma.rn.f32.bf16 %r288, %rs21, %rs85, %r287;
550
+ fma.rn.f32.bf16 %r289, %rs22, %rs86, %r288;
551
+ fma.rn.f32.bf16 %r290, %rs23, %rs87, %r289;
552
+ fma.rn.f32.bf16 %r291, %rs24, %rs88, %r290;
553
+ fma.rn.f32.bf16 %r292, %rs25, %rs89, %r266;
554
+ fma.rn.f32.bf16 %r293, %rs27, %rs91, %r292;
555
+ fma.rn.f32.bf16 %r294, %rs28, %rs92, %r293;
556
+ fma.rn.f32.bf16 %r295, %rs29, %rs93, %r294;
557
+ fma.rn.f32.bf16 %r296, %rs30, %rs94, %r295;
558
+ fma.rn.f32.bf16 %r297, %rs31, %rs95, %r296;
559
+ fma.rn.f32.bf16 %r298, %rs32, %rs96, %r297;
560
+ fma.rn.f32.bf16 %r299, %rs33, %rs97, %r267;
561
+ fma.rn.f32.bf16 %r300, %rs35, %rs99, %r299;
562
+ fma.rn.f32.bf16 %r301, %rs36, %rs100, %r300;
563
+ fma.rn.f32.bf16 %r302, %rs37, %rs101, %r301;
564
+ fma.rn.f32.bf16 %r303, %rs38, %rs102, %r302;
565
+ fma.rn.f32.bf16 %r304, %rs39, %rs103, %r303;
566
+ fma.rn.f32.bf16 %r305, %rs40, %rs104, %r304;
567
+ fma.rn.f32.bf16 %r306, %rs41, %rs105, %r268;
568
+ fma.rn.f32.bf16 %r307, %rs43, %rs107, %r306;
569
+ fma.rn.f32.bf16 %r308, %rs44, %rs108, %r307;
570
+ fma.rn.f32.bf16 %r309, %rs45, %rs109, %r308;
571
+ fma.rn.f32.bf16 %r310, %rs46, %rs110, %r309;
572
+ fma.rn.f32.bf16 %r311, %rs47, %rs111, %r310;
573
+ fma.rn.f32.bf16 %r312, %rs48, %rs112, %r311;
574
+ fma.rn.f32.bf16 %r313, %rs49, %rs113, %r269;
575
+ fma.rn.f32.bf16 %r314, %rs51, %rs115, %r313;
576
+ fma.rn.f32.bf16 %r315, %rs52, %rs116, %r314;
577
+ fma.rn.f32.bf16 %r316, %rs53, %rs117, %r315;
578
+ fma.rn.f32.bf16 %r317, %rs54, %rs118, %r316;
579
+ fma.rn.f32.bf16 %r318, %rs55, %rs119, %r317;
580
+ fma.rn.f32.bf16 %r319, %rs56, %rs120, %r318;
581
+ fma.rn.f32.bf16 %r320, %rs57, %rs121, %r270;
582
+ fma.rn.f32.bf16 %r321, %rs59, %rs123, %r320;
583
+ fma.rn.f32.bf16 %r322, %rs60, %rs124, %r321;
584
+ fma.rn.f32.bf16 %r323, %rs61, %rs125, %r322;
585
+ fma.rn.f32.bf16 %r324, %rs62, %rs126, %r323;
586
+ fma.rn.f32.bf16 %r325, %rs63, %rs127, %r324;
587
+ fma.rn.f32.bf16 %r326, %rs64, %rs128, %r325;
588
+ $L__tmp2:
589
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
590
+ shfl.sync.bfly.b32 %r327, %r277, 8, 31, -1;
591
+ $L__tmp3:
592
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
593
+ add.f32 %r328, %r277, %r327;
594
+ $L__tmp4:
595
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
596
+ shfl.sync.bfly.b32 %r329, %r328, 4, 31, -1;
597
+ $L__tmp5:
598
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
599
+ add.f32 %r330, %r328, %r329;
600
+ $L__tmp6:
601
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
602
+ shfl.sync.bfly.b32 %r331, %r330, 2, 31, -1;
603
+ $L__tmp7:
604
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
605
+ add.f32 %r332, %r330, %r331;
606
+ $L__tmp8:
607
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
608
+ shfl.sync.bfly.b32 %r333, %r332, 1, 31, -1;
609
+ $L__tmp9:
610
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
611
+ add.f32 %r334, %r332, %r333;
612
+ $L__tmp10:
613
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
614
+ shfl.sync.bfly.b32 %r335, %r284, 8, 31, -1;
615
+ $L__tmp11:
616
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
617
+ add.f32 %r336, %r284, %r335;
618
+ $L__tmp12:
619
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
620
+ shfl.sync.bfly.b32 %r337, %r336, 4, 31, -1;
621
+ $L__tmp13:
622
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
623
+ add.f32 %r338, %r336, %r337;
624
+ $L__tmp14:
625
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
626
+ shfl.sync.bfly.b32 %r339, %r338, 2, 31, -1;
627
+ $L__tmp15:
628
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
629
+ add.f32 %r340, %r338, %r339;
630
+ $L__tmp16:
631
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
632
+ shfl.sync.bfly.b32 %r341, %r340, 1, 31, -1;
633
+ $L__tmp17:
634
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
635
+ add.f32 %r342, %r340, %r341;
636
+ $L__tmp18:
637
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
638
+ shfl.sync.bfly.b32 %r343, %r291, 8, 31, -1;
639
+ $L__tmp19:
640
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
641
+ add.f32 %r344, %r291, %r343;
642
+ $L__tmp20:
643
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
644
+ shfl.sync.bfly.b32 %r345, %r344, 4, 31, -1;
645
+ $L__tmp21:
646
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
647
+ add.f32 %r346, %r344, %r345;
648
+ $L__tmp22:
649
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
650
+ shfl.sync.bfly.b32 %r347, %r346, 2, 31, -1;
651
+ $L__tmp23:
652
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
653
+ add.f32 %r348, %r346, %r347;
654
+ $L__tmp24:
655
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
656
+ shfl.sync.bfly.b32 %r349, %r348, 1, 31, -1;
657
+ $L__tmp25:
658
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
659
+ add.f32 %r350, %r348, %r349;
660
+ $L__tmp26:
661
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
662
+ shfl.sync.bfly.b32 %r351, %r298, 8, 31, -1;
663
+ $L__tmp27:
664
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
665
+ add.f32 %r352, %r298, %r351;
666
+ $L__tmp28:
667
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
668
+ shfl.sync.bfly.b32 %r353, %r352, 4, 31, -1;
669
+ $L__tmp29:
670
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
671
+ add.f32 %r354, %r352, %r353;
672
+ $L__tmp30:
673
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
674
+ shfl.sync.bfly.b32 %r355, %r354, 2, 31, -1;
675
+ $L__tmp31:
676
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
677
+ add.f32 %r356, %r354, %r355;
678
+ $L__tmp32:
679
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
680
+ shfl.sync.bfly.b32 %r357, %r356, 1, 31, -1;
681
+ $L__tmp33:
682
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
683
+ add.f32 %r358, %r356, %r357;
684
+ $L__tmp34:
685
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
686
+ shfl.sync.bfly.b32 %r359, %r305, 8, 31, -1;
687
+ $L__tmp35:
688
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
689
+ add.f32 %r360, %r305, %r359;
690
+ $L__tmp36:
691
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
692
+ shfl.sync.bfly.b32 %r361, %r360, 4, 31, -1;
693
+ $L__tmp37:
694
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
695
+ add.f32 %r362, %r360, %r361;
696
+ $L__tmp38:
697
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
698
+ shfl.sync.bfly.b32 %r363, %r362, 2, 31, -1;
699
+ $L__tmp39:
700
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
701
+ add.f32 %r364, %r362, %r363;
702
+ $L__tmp40:
703
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
704
+ shfl.sync.bfly.b32 %r365, %r364, 1, 31, -1;
705
+ $L__tmp41:
706
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
707
+ add.f32 %r366, %r364, %r365;
708
+ $L__tmp42:
709
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
710
+ shfl.sync.bfly.b32 %r367, %r312, 8, 31, -1;
711
+ $L__tmp43:
712
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
713
+ add.f32 %r368, %r312, %r367;
714
+ $L__tmp44:
715
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
716
+ shfl.sync.bfly.b32 %r369, %r368, 4, 31, -1;
717
+ $L__tmp45:
718
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
719
+ add.f32 %r370, %r368, %r369;
720
+ $L__tmp46:
721
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
722
+ shfl.sync.bfly.b32 %r371, %r370, 2, 31, -1;
723
+ $L__tmp47:
724
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
725
+ add.f32 %r372, %r370, %r371;
726
+ $L__tmp48:
727
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
728
+ shfl.sync.bfly.b32 %r373, %r372, 1, 31, -1;
729
+ $L__tmp49:
730
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
731
+ add.f32 %r374, %r372, %r373;
732
+ $L__tmp50:
733
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
734
+ shfl.sync.bfly.b32 %r375, %r319, 8, 31, -1;
735
+ $L__tmp51:
736
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
737
+ add.f32 %r376, %r319, %r375;
738
+ $L__tmp52:
739
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
740
+ shfl.sync.bfly.b32 %r377, %r376, 4, 31, -1;
741
+ $L__tmp53:
742
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
743
+ add.f32 %r378, %r376, %r377;
744
+ $L__tmp54:
745
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
746
+ shfl.sync.bfly.b32 %r379, %r378, 2, 31, -1;
747
+ $L__tmp55:
748
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
749
+ add.f32 %r380, %r378, %r379;
750
+ $L__tmp56:
751
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
752
+ shfl.sync.bfly.b32 %r381, %r380, 1, 31, -1;
753
+ $L__tmp57:
754
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
755
+ add.f32 %r382, %r380, %r381;
756
+ $L__tmp58:
757
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
758
+ shfl.sync.bfly.b32 %r383, %r326, 8, 31, -1;
759
+ $L__tmp59:
760
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
761
+ add.f32 %r384, %r326, %r383;
762
+ $L__tmp60:
763
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
764
+ shfl.sync.bfly.b32 %r385, %r384, 4, 31, -1;
765
+ $L__tmp61:
766
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
767
+ add.f32 %r386, %r384, %r385;
768
+ $L__tmp62:
769
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
770
+ shfl.sync.bfly.b32 %r387, %r386, 2, 31, -1;
771
+ $L__tmp63:
772
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
773
+ add.f32 %r388, %r386, %r387;
774
+ $L__tmp64:
775
+ .loc 2 293 36 // standard.py:293:36 @[ l2norm.py:107:55 ]
776
+ shfl.sync.bfly.b32 %r389, %r388, 1, 31, -1;
777
+ $L__tmp65:
778
+ .loc 2 263 15 // standard.py:263:15 @[ standard.py:293:36 @[ l2norm.py:107:55 ] ]
779
+ add.f32 %r390, %r388, %r389;
780
+ $L__tmp66:
781
+ .loc 1 107 69 // l2norm.py:107:69
782
+ mul.f32 %r391, %r334, %r119;
783
+ mul.f32 %r392, %r334, %r120;
784
+ mul.f32 %r393, %r334, %r121;
785
+ mul.f32 %r394, %r334, %r122;
786
+ mul.f32 %r395, %r334, %r123;
787
+ mul.f32 %r396, %r334, %r124;
788
+ mul.f32 %r397, %r334, %r125;
789
+ mul.f32 %r398, %r334, %r126;
790
+ mul.f32 %r399, %r342, %r127;
791
+ mul.f32 %r400, %r342, %r128;
792
+ mul.f32 %r401, %r342, %r129;
793
+ mul.f32 %r402, %r342, %r130;
794
+ mul.f32 %r403, %r342, %r131;
795
+ mul.f32 %r404, %r342, %r132;
796
+ mul.f32 %r405, %r342, %r133;
797
+ mul.f32 %r406, %r342, %r134;
798
+ mul.f32 %r407, %r350, %r135;
799
+ mul.f32 %r408, %r350, %r136;
800
+ mul.f32 %r409, %r350, %r137;
801
+ mul.f32 %r410, %r350, %r138;
802
+ mul.f32 %r411, %r350, %r139;
803
+ mul.f32 %r412, %r350, %r140;
804
+ mul.f32 %r413, %r350, %r141;
805
+ mul.f32 %r414, %r350, %r142;
806
+ mul.f32 %r415, %r358, %r143;
807
+ mul.f32 %r416, %r358, %r144;
808
+ mul.f32 %r417, %r358, %r145;
809
+ mul.f32 %r418, %r358, %r146;
810
+ mul.f32 %r419, %r358, %r147;
811
+ mul.f32 %r420, %r358, %r148;
812
+ mul.f32 %r421, %r358, %r149;
813
+ mul.f32 %r422, %r358, %r150;
814
+ mul.f32 %r423, %r366, %r151;
815
+ mul.f32 %r424, %r366, %r152;
816
+ mul.f32 %r425, %r366, %r153;
817
+ mul.f32 %r426, %r366, %r154;
818
+ mul.f32 %r427, %r366, %r155;
819
+ mul.f32 %r428, %r366, %r156;
820
+ mul.f32 %r429, %r366, %r157;
821
+ mul.f32 %r430, %r366, %r158;
822
+ mul.f32 %r431, %r374, %r159;
823
+ mul.f32 %r432, %r374, %r160;
824
+ mul.f32 %r433, %r374, %r161;
825
+ mul.f32 %r434, %r374, %r162;
826
+ mul.f32 %r435, %r374, %r163;
827
+ mul.f32 %r436, %r374, %r164;
828
+ mul.f32 %r437, %r374, %r165;
829
+ mul.f32 %r438, %r374, %r166;
830
+ mul.f32 %r439, %r382, %r167;
831
+ mul.f32 %r440, %r382, %r168;
832
+ mul.f32 %r441, %r382, %r169;
833
+ mul.f32 %r442, %r382, %r170;
834
+ mul.f32 %r443, %r382, %r171;
835
+ mul.f32 %r444, %r382, %r172;
836
+ mul.f32 %r445, %r382, %r173;
837
+ mul.f32 %r446, %r382, %r174;
838
+ mul.f32 %r447, %r390, %r175;
839
+ mul.f32 %r448, %r390, %r176;
840
+ mul.f32 %r449, %r390, %r177;
841
+ mul.f32 %r450, %r390, %r178;
842
+ mul.f32 %r451, %r390, %r179;
843
+ mul.f32 %r452, %r390, %r180;
844
+ mul.f32 %r453, %r390, %r181;
845
+ mul.f32 %r454, %r390, %r182;
846
+ .loc 1 107 75 // l2norm.py:107:75
847
+ mul.f32 %r455, %r391, %r34;
848
+ mul.f32 %r456, %r392, %r34;
849
+ mul.f32 %r457, %r393, %r34;
850
+ mul.f32 %r458, %r394, %r34;
851
+ mul.f32 %r459, %r395, %r34;
852
+ mul.f32 %r460, %r396, %r34;
853
+ mul.f32 %r461, %r397, %r34;
854
+ mul.f32 %r462, %r398, %r34;
855
+ mul.f32 %r463, %r399, %r35;
856
+ mul.f32 %r464, %r400, %r35;
857
+ mul.f32 %r465, %r401, %r35;
858
+ mul.f32 %r466, %r402, %r35;
859
+ mul.f32 %r467, %r403, %r35;
860
+ mul.f32 %r468, %r404, %r35;
861
+ mul.f32 %r469, %r405, %r35;
862
+ mul.f32 %r470, %r406, %r35;
863
+ mul.f32 %r471, %r407, %r36;
864
+ mul.f32 %r472, %r408, %r36;
865
+ mul.f32 %r473, %r409, %r36;
866
+ mul.f32 %r474, %r410, %r36;
867
+ mul.f32 %r475, %r411, %r36;
868
+ mul.f32 %r476, %r412, %r36;
869
+ mul.f32 %r477, %r413, %r36;
870
+ mul.f32 %r478, %r414, %r36;
871
+ mul.f32 %r479, %r415, %r37;
872
+ mul.f32 %r480, %r416, %r37;
873
+ mul.f32 %r481, %r417, %r37;
874
+ mul.f32 %r482, %r418, %r37;
875
+ mul.f32 %r483, %r419, %r37;
876
+ mul.f32 %r484, %r420, %r37;
877
+ mul.f32 %r485, %r421, %r37;
878
+ mul.f32 %r486, %r422, %r37;
879
+ mul.f32 %r487, %r423, %r38;
880
+ mul.f32 %r488, %r424, %r38;
881
+ mul.f32 %r489, %r425, %r38;
882
+ mul.f32 %r490, %r426, %r38;
883
+ mul.f32 %r491, %r427, %r38;
884
+ mul.f32 %r492, %r428, %r38;
885
+ mul.f32 %r493, %r429, %r38;
886
+ mul.f32 %r494, %r430, %r38;
887
+ mul.f32 %r495, %r431, %r39;
888
+ mul.f32 %r496, %r432, %r39;
889
+ mul.f32 %r497, %r433, %r39;
890
+ mul.f32 %r498, %r434, %r39;
891
+ mul.f32 %r499, %r435, %r39;
892
+ mul.f32 %r500, %r436, %r39;
893
+ mul.f32 %r501, %r437, %r39;
894
+ mul.f32 %r502, %r438, %r39;
895
+ mul.f32 %r503, %r439, %r40;
896
+ mul.f32 %r504, %r440, %r40;
897
+ mul.f32 %r505, %r441, %r40;
898
+ mul.f32 %r506, %r442, %r40;
899
+ mul.f32 %r507, %r443, %r40;
900
+ mul.f32 %r508, %r444, %r40;
901
+ mul.f32 %r509, %r445, %r40;
902
+ mul.f32 %r510, %r446, %r40;
903
+ mul.f32 %r511, %r447, %r41;
904
+ mul.f32 %r512, %r448, %r41;
905
+ mul.f32 %r513, %r449, %r41;
906
+ mul.f32 %r514, %r450, %r41;
907
+ mul.f32 %r515, %r451, %r41;
908
+ mul.f32 %r516, %r452, %r41;
909
+ mul.f32 %r517, %r453, %r41;
910
+ mul.f32 %r518, %r454, %r41;
911
+ .loc 1 107 36 // l2norm.py:107:36
912
+ neg.f32 %r519, %r455;
913
+ fma.rn.f32 %r520, %r34, %r183, %r519;
914
+ neg.f32 %r521, %r456;
915
+ fma.rn.f32 %r522, %r34, %r184, %r521;
916
+ neg.f32 %r523, %r457;
917
+ fma.rn.f32 %r524, %r34, %r185, %r523;
918
+ neg.f32 %r525, %r458;
919
+ fma.rn.f32 %r526, %r34, %r186, %r525;
920
+ neg.f32 %r527, %r459;
921
+ fma.rn.f32 %r528, %r34, %r187, %r527;
922
+ neg.f32 %r529, %r460;
923
+ fma.rn.f32 %r530, %r34, %r188, %r529;
924
+ neg.f32 %r531, %r461;
925
+ fma.rn.f32 %r532, %r34, %r189, %r531;
926
+ neg.f32 %r533, %r462;
927
+ fma.rn.f32 %r534, %r34, %r190, %r533;
928
+ neg.f32 %r535, %r463;
929
+ fma.rn.f32 %r536, %r35, %r191, %r535;
930
+ neg.f32 %r537, %r464;
931
+ fma.rn.f32 %r538, %r35, %r192, %r537;
932
+ neg.f32 %r539, %r465;
933
+ fma.rn.f32 %r540, %r35, %r193, %r539;
934
+ neg.f32 %r541, %r466;
935
+ fma.rn.f32 %r542, %r35, %r194, %r541;
936
+ neg.f32 %r543, %r467;
937
+ fma.rn.f32 %r544, %r35, %r195, %r543;
938
+ neg.f32 %r545, %r468;
939
+ fma.rn.f32 %r546, %r35, %r196, %r545;
940
+ neg.f32 %r547, %r469;
941
+ fma.rn.f32 %r548, %r35, %r197, %r547;
942
+ neg.f32 %r549, %r470;
943
+ fma.rn.f32 %r550, %r35, %r198, %r549;
944
+ neg.f32 %r551, %r471;
945
+ fma.rn.f32 %r552, %r36, %r199, %r551;
946
+ neg.f32 %r553, %r472;
947
+ fma.rn.f32 %r554, %r36, %r200, %r553;
948
+ neg.f32 %r555, %r473;
949
+ fma.rn.f32 %r556, %r36, %r201, %r555;
950
+ neg.f32 %r557, %r474;
951
+ fma.rn.f32 %r558, %r36, %r202, %r557;
952
+ neg.f32 %r559, %r475;
953
+ fma.rn.f32 %r560, %r36, %r203, %r559;
954
+ neg.f32 %r561, %r476;
955
+ fma.rn.f32 %r562, %r36, %r204, %r561;
956
+ neg.f32 %r563, %r477;
957
+ fma.rn.f32 %r564, %r36, %r205, %r563;
958
+ neg.f32 %r565, %r478;
959
+ fma.rn.f32 %r566, %r36, %r206, %r565;
960
+ neg.f32 %r567, %r479;
961
+ fma.rn.f32 %r568, %r37, %r207, %r567;
962
+ neg.f32 %r569, %r480;
963
+ fma.rn.f32 %r570, %r37, %r208, %r569;
964
+ neg.f32 %r571, %r481;
965
+ fma.rn.f32 %r572, %r37, %r209, %r571;
966
+ neg.f32 %r573, %r482;
967
+ fma.rn.f32 %r574, %r37, %r210, %r573;
968
+ neg.f32 %r575, %r483;
969
+ fma.rn.f32 %r576, %r37, %r211, %r575;
970
+ neg.f32 %r577, %r484;
971
+ fma.rn.f32 %r578, %r37, %r212, %r577;
972
+ neg.f32 %r579, %r485;
973
+ fma.rn.f32 %r580, %r37, %r213, %r579;
974
+ neg.f32 %r581, %r486;
975
+ fma.rn.f32 %r582, %r37, %r214, %r581;
976
+ neg.f32 %r583, %r487;
977
+ fma.rn.f32 %r584, %r38, %r215, %r583;
978
+ neg.f32 %r585, %r488;
979
+ fma.rn.f32 %r586, %r38, %r216, %r585;
980
+ neg.f32 %r587, %r489;
981
+ fma.rn.f32 %r588, %r38, %r217, %r587;
982
+ neg.f32 %r589, %r490;
983
+ fma.rn.f32 %r590, %r38, %r218, %r589;
984
+ neg.f32 %r591, %r491;
985
+ fma.rn.f32 %r592, %r38, %r219, %r591;
986
+ neg.f32 %r593, %r492;
987
+ fma.rn.f32 %r594, %r38, %r220, %r593;
988
+ neg.f32 %r595, %r493;
989
+ fma.rn.f32 %r596, %r38, %r221, %r595;
990
+ neg.f32 %r597, %r494;
991
+ fma.rn.f32 %r598, %r38, %r222, %r597;
992
+ neg.f32 %r599, %r495;
993
+ fma.rn.f32 %r600, %r39, %r223, %r599;
994
+ neg.f32 %r601, %r496;
995
+ fma.rn.f32 %r602, %r39, %r224, %r601;
996
+ neg.f32 %r603, %r497;
997
+ fma.rn.f32 %r604, %r39, %r225, %r603;
998
+ neg.f32 %r605, %r498;
999
+ fma.rn.f32 %r606, %r39, %r226, %r605;
1000
+ neg.f32 %r607, %r499;
1001
+ fma.rn.f32 %r608, %r39, %r227, %r607;
1002
+ neg.f32 %r609, %r500;
1003
+ fma.rn.f32 %r610, %r39, %r228, %r609;
1004
+ neg.f32 %r611, %r501;
1005
+ fma.rn.f32 %r612, %r39, %r229, %r611;
1006
+ neg.f32 %r613, %r502;
1007
+ fma.rn.f32 %r614, %r39, %r230, %r613;
1008
+ mov.b64 %rd96, {%r247, %r248};
1009
+ mov.b64 %rd97, {%r503, %r504};
1010
+ sub.f32x2 %rd98, %rd96, %rd97;
1011
+ mov.b64 %rd99, {%r249, %r250};
1012
+ mov.b64 %rd100, {%r505, %r506};
1013
+ sub.f32x2 %rd101, %rd99, %rd100;
1014
+ mov.b64 %rd102, {%r251, %r252};
1015
+ mov.b64 %rd103, {%r507, %r508};
1016
+ sub.f32x2 %rd104, %rd102, %rd103;
1017
+ mov.b64 %rd105, {%r253, %r254};
1018
+ mov.b64 %rd106, {%r509, %r510};
1019
+ sub.f32x2 %rd107, %rd105, %rd106;
1020
+ mov.b64 %rd108, {%r255, %r256};
1021
+ mov.b64 %rd109, {%r511, %r512};
1022
+ sub.f32x2 %rd110, %rd108, %rd109;
1023
+ mov.b64 %rd111, {%r257, %r258};
1024
+ mov.b64 %rd112, {%r513, %r514};
1025
+ sub.f32x2 %rd113, %rd111, %rd112;
1026
+ mov.b64 %rd114, {%r259, %r260};
1027
+ mov.b64 %rd115, {%r515, %r516};
1028
+ sub.f32x2 %rd116, %rd114, %rd115;
1029
+ mov.b64 %rd117, {%r261, %r262};
1030
+ mov.b64 %rd118, {%r517, %r518};
1031
+ sub.f32x2 %rd119, %rd117, %rd118;
1032
+ .loc 1 108 27 // l2norm.py:108:27
1033
+ mov.b64 {%r615, %r616}, %rd98;
1034
+ cvt.rn.bf16x2.f32 %r98, %r616, %r615;
1035
+ mov.b64 {%r617, %r618}, %rd101;
1036
+ cvt.rn.bf16x2.f32 %r99, %r618, %r617;
1037
+ mov.b64 {%r619, %r620}, %rd104;
1038
+ cvt.rn.bf16x2.f32 %r100, %r620, %r619;
1039
+ mov.b64 {%r621, %r622}, %rd107;
1040
+ cvt.rn.bf16x2.f32 %r101, %r622, %r621;
1041
+ mov.b64 {%r623, %r624}, %rd110;
1042
+ cvt.rn.bf16x2.f32 %r102, %r624, %r623;
1043
+ mov.b64 {%r625, %r626}, %rd113;
1044
+ cvt.rn.bf16x2.f32 %r103, %r626, %r625;
1045
+ mov.b64 {%r627, %r628}, %rd116;
1046
+ cvt.rn.bf16x2.f32 %r104, %r628, %r627;
1047
+ mov.b64 {%r629, %r630}, %rd119;
1048
+ cvt.rn.bf16x2.f32 %r105, %r630, %r629;
1049
+ .loc 1 108 19 // l2norm.py:108:19
1050
+ cvt.rn.bf16x2.f32 %r74, %r522, %r520;
1051
+ cvt.rn.bf16x2.f32 %r75, %r526, %r524;
1052
+ cvt.rn.bf16x2.f32 %r76, %r530, %r528;
1053
+ cvt.rn.bf16x2.f32 %r77, %r534, %r532;
1054
+ // begin inline asm
1055
+ @%p1 st.global.v4.b32 [ %rd25 + 0 ], { %r74, %r75, %r76, %r77 };
1056
+ // end inline asm
1057
+ cvt.rn.bf16x2.f32 %r78, %r538, %r536;
1058
+ cvt.rn.bf16x2.f32 %r79, %r542, %r540;
1059
+ cvt.rn.bf16x2.f32 %r80, %r546, %r544;
1060
+ cvt.rn.bf16x2.f32 %r81, %r550, %r548;
1061
+ // begin inline asm
1062
+ @%p2 st.global.v4.b32 [ %rd26 + 0 ], { %r78, %r79, %r80, %r81 };
1063
+ // end inline asm
1064
+ cvt.rn.bf16x2.f32 %r82, %r554, %r552;
1065
+ cvt.rn.bf16x2.f32 %r83, %r558, %r556;
1066
+ cvt.rn.bf16x2.f32 %r84, %r562, %r560;
1067
+ cvt.rn.bf16x2.f32 %r85, %r566, %r564;
1068
+ // begin inline asm
1069
+ @%p3 st.global.v4.b32 [ %rd27 + 0 ], { %r82, %r83, %r84, %r85 };
1070
+ // end inline asm
1071
+ cvt.rn.bf16x2.f32 %r86, %r570, %r568;
1072
+ cvt.rn.bf16x2.f32 %r87, %r574, %r572;
1073
+ cvt.rn.bf16x2.f32 %r88, %r578, %r576;
1074
+ cvt.rn.bf16x2.f32 %r89, %r582, %r580;
1075
+ // begin inline asm
1076
+ @%p4 st.global.v4.b32 [ %rd28 + 0 ], { %r86, %r87, %r88, %r89 };
1077
+ // end inline asm
1078
+ cvt.rn.bf16x2.f32 %r90, %r586, %r584;
1079
+ cvt.rn.bf16x2.f32 %r91, %r590, %r588;
1080
+ cvt.rn.bf16x2.f32 %r92, %r594, %r592;
1081
+ cvt.rn.bf16x2.f32 %r93, %r598, %r596;
1082
+ // begin inline asm
1083
+ @%p5 st.global.v4.b32 [ %rd29 + 0 ], { %r90, %r91, %r92, %r93 };
1084
+ // end inline asm
1085
+ cvt.rn.bf16x2.f32 %r94, %r602, %r600;
1086
+ cvt.rn.bf16x2.f32 %r95, %r606, %r604;
1087
+ cvt.rn.bf16x2.f32 %r96, %r610, %r608;
1088
+ cvt.rn.bf16x2.f32 %r97, %r614, %r612;
1089
+ // begin inline asm
1090
+ @%p6 st.global.v4.b32 [ %rd30 + 0 ], { %r94, %r95, %r96, %r97 };
1091
+ // end inline asm
1092
+ // begin inline asm
1093
+ @%p7 st.global.v4.b32 [ %rd31 + 0 ], { %r98, %r99, %r100, %r101 };
1094
+ // end inline asm
1095
+ // begin inline asm
1096
+ @%p8 st.global.v4.b32 [ %rd32 + 0 ], { %r102, %r103, %r104, %r105 };
1097
+ // end inline asm
1098
+ .loc 1 108 4 // l2norm.py:108:4
1099
+ ret;
1100
+ $L__tmp67:
1101
+ $L__func_end0:
1102
+ // -- End function
1103
+ }
1104
+ .file 1 "/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py"
1105
+ .file 2 "/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py"
1106
+ .section .debug_abbrev
1107
+ {
1108
+ .b8 1 // Abbreviation Code
1109
+ .b8 17 // DW_TAG_compile_unit
1110
+ .b8 1 // DW_CHILDREN_yes
1111
+ .b8 37 // DW_AT_producer
1112
+ .b8 8 // DW_FORM_string
1113
+ .b8 19 // DW_AT_language
1114
+ .b8 5 // DW_FORM_data2
1115
+ .b8 3 // DW_AT_name
1116
+ .b8 8 // DW_FORM_string
1117
+ .b8 16 // DW_AT_stmt_list
1118
+ .b8 6 // DW_FORM_data4
1119
+ .b8 27 // DW_AT_comp_dir
1120
+ .b8 8 // DW_FORM_string
1121
+ .b8 0 // EOM(1)
1122
+ .b8 0 // EOM(2)
1123
+ .b8 2 // Abbreviation Code
1124
+ .b8 46 // DW_TAG_subprogram
1125
+ .b8 0 // DW_CHILDREN_no
1126
+ .b8 3 // DW_AT_name
1127
+ .b8 8 // DW_FORM_string
1128
+ .b8 32 // DW_AT_inline
1129
+ .b8 11 // DW_FORM_data1
1130
+ .b8 0 // EOM(1)
1131
+ .b8 0 // EOM(2)
1132
+ .b8 3 // Abbreviation Code
1133
+ .b8 46 // DW_TAG_subprogram
1134
+ .b8 1 // DW_CHILDREN_yes
1135
+ .b8 17 // DW_AT_low_pc
1136
+ .b8 1 // DW_FORM_addr
1137
+ .b8 18 // DW_AT_high_pc
1138
+ .b8 1 // DW_FORM_addr
1139
+ .b8 49 // DW_AT_abstract_origin
1140
+ .b8 19 // DW_FORM_ref4
1141
+ .b8 0 // EOM(1)
1142
+ .b8 0 // EOM(2)
1143
+ .b8 4 // Abbreviation Code
1144
+ .b8 29 // DW_TAG_inlined_subroutine
1145
+ .b8 1 // DW_CHILDREN_yes
1146
+ .b8 49 // DW_AT_abstract_origin
1147
+ .b8 19 // DW_FORM_ref4
1148
+ .b8 17 // DW_AT_low_pc
1149
+ .b8 1 // DW_FORM_addr
1150
+ .b8 18 // DW_AT_high_pc
1151
+ .b8 1 // DW_FORM_addr
1152
+ .b8 88 // DW_AT_call_file
1153
+ .b8 11 // DW_FORM_data1
1154
+ .b8 89 // DW_AT_call_line
1155
+ .b8 11 // DW_FORM_data1
1156
+ .b8 87 // DW_AT_call_column
1157
+ .b8 11 // DW_FORM_data1
1158
+ .b8 0 // EOM(1)
1159
+ .b8 0 // EOM(2)
1160
+ .b8 5 // Abbreviation Code
1161
+ .b8 29 // DW_TAG_inlined_subroutine
1162
+ .b8 0 // DW_CHILDREN_no
1163
+ .b8 49 // DW_AT_abstract_origin
1164
+ .b8 19 // DW_FORM_ref4
1165
+ .b8 17 // DW_AT_low_pc
1166
+ .b8 1 // DW_FORM_addr
1167
+ .b8 18 // DW_AT_high_pc
1168
+ .b8 1 // DW_FORM_addr
1169
+ .b8 88 // DW_AT_call_file
1170
+ .b8 11 // DW_FORM_data1
1171
+ .b8 89 // DW_AT_call_line
1172
+ .b8 5 // DW_FORM_data2
1173
+ .b8 87 // DW_AT_call_column
1174
+ .b8 11 // DW_FORM_data1
1175
+ .b8 0 // EOM(1)
1176
+ .b8 0 // EOM(2)
1177
+ .b8 0 // EOM(3)
1178
+ }
1179
+ .section .debug_info
1180
+ {
1181
+ .b32 196 // Length of Unit
1182
+ .b8 2 // DWARF version number
1183
+ .b8 0
1184
+ .b32 .debug_abbrev // Offset Into Abbrev. Section
1185
+ .b8 8 // Address Size (in bytes)
1186
+ .b8 1 // Abbrev [1] 0xb:0xbd DW_TAG_compile_unit
1187
+ .b8 116 // DW_AT_producer
1188
+ .b8 114
1189
+ .b8 105
1190
+ .b8 116
1191
+ .b8 111
1192
+ .b8 110
1193
+ .b8 0
1194
+ .b8 2 // DW_AT_language
1195
+ .b8 0
1196
+ .b8 108 // DW_AT_name
1197
+ .b8 50
1198
+ .b8 110
1199
+ .b8 111
1200
+ .b8 114
1201
+ .b8 109
1202
+ .b8 46
1203
+ .b8 112
1204
+ .b8 121
1205
+ .b8 0
1206
+ .b32 .debug_line // DW_AT_stmt_list
1207
+ .b8 47 // DW_AT_comp_dir
1208
+ .b8 104
1209
+ .b8 111
1210
+ .b8 109
1211
+ .b8 101
1212
+ .b8 47
1213
+ .b8 115
1214
+ .b8 105
1215
+ .b8 108
1216
+ .b8 118
1217
+ .b8 101
1218
+ .b8 114
1219
+ .b8 47
1220
+ .b8 97
1221
+ .b8 102
1222
+ .b8 102
1223
+ .b8 105
1224
+ .b8 110
1225
+ .b8 101
1226
+ .b8 47
1227
+ .b8 46
1228
+ .b8 118
1229
+ .b8 101
1230
+ .b8 110
1231
+ .b8 118
1232
+ .b8 47
1233
+ .b8 108
1234
+ .b8 105
1235
+ .b8 98
1236
+ .b8 47
1237
+ .b8 112
1238
+ .b8 121
1239
+ .b8 116
1240
+ .b8 104
1241
+ .b8 111
1242
+ .b8 110
1243
+ .b8 51
1244
+ .b8 46
1245
+ .b8 49
1246
+ .b8 50
1247
+ .b8 47
1248
+ .b8 115
1249
+ .b8 105
1250
+ .b8 116
1251
+ .b8 101
1252
+ .b8 45
1253
+ .b8 112
1254
+ .b8 97
1255
+ .b8 99
1256
+ .b8 107
1257
+ .b8 97
1258
+ .b8 103
1259
+ .b8 101
1260
+ .b8 115
1261
+ .b8 47
1262
+ .b8 102
1263
+ .b8 108
1264
+ .b8 97
1265
+ .b8 47
1266
+ .b8 109
1267
+ .b8 111
1268
+ .b8 100
1269
+ .b8 117
1270
+ .b8 108
1271
+ .b8 101
1272
+ .b8 115
1273
+ .b8 47
1274
+ .b8 110
1275
+ .b8 111
1276
+ .b8 114
1277
+ .b8 109
1278
+ .b8 0
1279
+ .b8 2 // Abbrev [2] 0x6b:0x14 DW_TAG_subprogram
1280
+ .b8 108 // DW_AT_name
1281
+ .b8 50
1282
+ .b8 110
1283
+ .b8 111
1284
+ .b8 114
1285
+ .b8 109
1286
+ .b8 95
1287
+ .b8 98
1288
+ .b8 119
1289
+ .b8 100
1290
+ .b8 95
1291
+ .b8 107
1292
+ .b8 101
1293
+ .b8 114
1294
+ .b8 110
1295
+ .b8 101
1296
+ .b8 108
1297
+ .b8 0
1298
+ .b8 1 // DW_AT_inline
1299
+ .b8 3 // Abbrev [3] 0x7f:0x48 DW_TAG_subprogram
1300
+ .b64 $L__func_begin0 // DW_AT_low_pc
1301
+ .b64 $L__func_end0 // DW_AT_high_pc
1302
+ .b32 107 // DW_AT_abstract_origin
1303
+ .b8 4 // Abbrev [4] 0x94:0x32 DW_TAG_inlined_subroutine
1304
+ .b32 107 // DW_AT_abstract_origin
1305
+ .b64 $L__tmp1 // DW_AT_low_pc
1306
+ .b64 $L__tmp66 // DW_AT_high_pc
1307
+ .b8 1 // DW_AT_call_file
1308
+ .b8 107 // DW_AT_call_line
1309
+ .b8 55 // DW_AT_call_column
1310
+ .b8 5 // Abbrev [5] 0xac:0x19 DW_TAG_inlined_subroutine
1311
+ .b32 107 // DW_AT_abstract_origin
1312
+ .b64 $L__tmp1 // DW_AT_low_pc
1313
+ .b64 $L__tmp66 // DW_AT_high_pc
1314
+ .b8 2 // DW_AT_call_file
1315
+ .b8 37 // DW_AT_call_line
1316
+ .b8 1
1317
+ .b8 36 // DW_AT_call_column
1318
+ .b8 0 // End Of Children Mark
1319
+ .b8 0 // End Of Children Mark
1320
+ .b8 0 // End Of Children Mark
1321
+ }
1322
+ .section .debug_macinfo { }
triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.source ADDED
@@ -0,0 +1,210 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0)
2
+ #loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0)
3
+ #loc47 = loc(unknown)
4
+ #loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0)
5
+ #loc54 = loc("y"(#loc))
6
+ #loc55 = loc("rstd"(#loc))
7
+ #loc56 = loc("dy"(#loc))
8
+ #loc57 = loc("dx"(#loc))
9
+ #loc58 = loc("eps"(#loc))
10
+ #loc59 = loc("T"(#loc))
11
+ #loc101 = loc("input"(#loc45))
12
+ #loc102 = loc("a"(#loc50))
13
+ #loc103 = loc("b"(#loc50))
14
+ module {
15
+ tt.func public @l2norm_bwd_kernel(%y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
16
+ %i_t = tt.get_program_id x : i32 loc(#loc60)
17
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61)
18
+ %o_t = arith.constant 128 : i32 loc(#loc62)
19
+ %o_t_1 = arith.constant 128 : i64 loc(#loc62)
20
+ %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62)
21
+ %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc63)
22
+ %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc64)
23
+ %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc64)
24
+ %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc64)
25
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65)
26
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc66)
27
+ %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc66)
28
+ %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc66)
29
+ %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc67)
30
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68)
31
+ %m_x_10 = arith.constant 128 : i32 loc(#loc69)
32
+ %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69)
33
+ %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69)
34
+ %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc70)
35
+ %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc70)
36
+ %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc70)
37
+ %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc71)
38
+ %p_y_16 = arith.constant 128 : i32 loc(#loc72)
39
+ %p_y_17 = arith.constant 128 : i64 loc(#loc72)
40
+ %p_y_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc72)
41
+ %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<128x1xi64> loc(#loc72)
42
+ %p_y_20 = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc73)
43
+ %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc73)
44
+ %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74)
45
+ %p_y_23 = tt.broadcast %p_y_21 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc75)
46
+ %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc75)
47
+ %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc75)
48
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>> loc(#loc76)
49
+ %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr<f32>>, tensor<128xi64> loc(#loc76)
50
+ %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc77)
51
+ %p_dy_27 = arith.constant 128 : i32 loc(#loc78)
52
+ %p_dy_28 = arith.constant 128 : i64 loc(#loc78)
53
+ %p_dy_29 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc78)
54
+ %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<128x1xi64> loc(#loc78)
55
+ %p_dy_31 = tt.splat %dy : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc79)
56
+ %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc79)
57
+ %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80)
58
+ %p_dy_34 = tt.broadcast %p_dy_32 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc81)
59
+ %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc81)
60
+ %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc81)
61
+ %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc82)
62
+ %p_dx_37 = arith.constant 128 : i32 loc(#loc83)
63
+ %p_dx_38 = arith.constant 128 : i64 loc(#loc83)
64
+ %p_dx_39 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc83)
65
+ %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<128x1xi64> loc(#loc83)
66
+ %p_dx_41 = tt.splat %dx : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc84)
67
+ %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc84)
68
+ %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85)
69
+ %p_dx_44 = tt.broadcast %p_dx_42 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc86)
70
+ %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc86)
71
+ %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc86)
72
+ %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87)
73
+ %b_y_47 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc87)
74
+ %b_y_48 = arith.truncf %b_y_47 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc87)
75
+ %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<128x128x!tt.ptr<bf16>> loc(#loc87)
76
+ %b_y_50 = arith.extf %b_y_49 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc88)
77
+ %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89)
78
+ %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc89)
79
+ %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<128x!tt.ptr<f32>> loc(#loc89)
80
+ %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90)
81
+ %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc90)
82
+ %b_dy_54 = arith.truncf %b_dy_53 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc90)
83
+ %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<128x128x!tt.ptr<bf16>> loc(#loc90)
84
+ %b_dy_56 = arith.extf %b_dy_55 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc91)
85
+ %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc92)
86
+ %b_dx_57 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc93)
87
+ %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<128x128xf32> loc(#loc93)
88
+ %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<128x128xf32> loc(#loc94)
89
+ %b_dx_60 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc95)
90
+ %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc96)
91
+ %b_dx_62 = tt.broadcast %b_dx_61 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc97)
92
+ %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<128x128xf32> loc(#loc97)
93
+ %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc98)
94
+ %b_dx_65 = tt.broadcast %b_dx_64 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc99)
95
+ %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<128x128xf32> loc(#loc99)
96
+ %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<128x128xf32> loc(#loc100)
97
+ %0 = arith.truncf %b_dx_67 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc42)
98
+ tt.store %p_dx_46, %0, %m_x_15 : tensor<128x128x!tt.ptr<bf16>> loc(#loc43)
99
+ tt.return loc(#loc44)
100
+ } loc(#loc)
101
+ tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc45))) -> tensor<128xf32> attributes {noinline = false} {
102
+ %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({
103
+ ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)):
104
+ %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46)
105
+ tt.reduce.return %2 : f32 loc(#loc46)
106
+ }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc46)
107
+ tt.return %0 : tensor<128xf32> loc(#loc48)
108
+ ^bb1: // no predecessors
109
+ %1 = ub.poison : tensor<128xf32> loc(#loc49)
110
+ tt.return %1 : tensor<128xf32> loc(#loc49)
111
+ } loc(#loc45)
112
+ tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} {
113
+ %0 = arith.addf %a, %b : f32 loc(#loc51)
114
+ tt.return %0 : f32 loc(#loc52)
115
+ ^bb1: // no predecessors
116
+ %1 = ub.poison : f32 loc(#loc53)
117
+ tt.return %1 : f32 loc(#loc53)
118
+ } loc(#loc50)
119
+ } loc(#loc)
120
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24)
121
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30)
122
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16)
123
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34)
124
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21)
125
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23)
126
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16)
127
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14)
128
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30)
129
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41)
130
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26)
131
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18)
132
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29)
133
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14)
134
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37)
135
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33)
136
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20)
137
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20)
138
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31)
139
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16)
140
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39)
141
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35)
142
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20)
143
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31)
144
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16)
145
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39)
146
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35)
147
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18)
148
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47)
149
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21)
150
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19)
151
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49)
152
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25)
153
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18)
154
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50)
155
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55)
156
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58)
157
+ #loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69)
158
+ #loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82)
159
+ #loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75)
160
+ #loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36)
161
+ #loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27)
162
+ #loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19)
163
+ #loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4)
164
+ #loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
165
+ #loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11)
166
+ #loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4)
167
+ #loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
168
+ #loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11)
169
+ #loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4)
170
+ #loc60 = loc("i_t"(#loc1))
171
+ #loc61 = loc("i_t"(#loc2))
172
+ #loc62 = loc("o_t"(#loc3))
173
+ #loc63 = loc("o_t"(#loc4))
174
+ #loc64 = loc("o_t"(#loc5))
175
+ #loc65 = loc("o_d"(#loc6))
176
+ #loc66 = loc("m_t"(#loc7))
177
+ #loc67 = loc("m_x"(#loc8))
178
+ #loc68 = loc("m_x"(#loc9))
179
+ #loc69 = loc("m_x"(#loc10))
180
+ #loc70 = loc("m_x"(#loc11))
181
+ #loc71 = loc("p_y"(#loc12))
182
+ #loc72 = loc("p_y"(#loc13))
183
+ #loc73 = loc("p_y"(#loc14))
184
+ #loc74 = loc("p_y"(#loc15))
185
+ #loc75 = loc("p_y"(#loc16))
186
+ #loc76 = loc("p_rstd"(#loc17))
187
+ #loc77 = loc("p_dy"(#loc18))
188
+ #loc78 = loc("p_dy"(#loc19))
189
+ #loc79 = loc("p_dy"(#loc20))
190
+ #loc80 = loc("p_dy"(#loc21))
191
+ #loc81 = loc("p_dy"(#loc22))
192
+ #loc82 = loc("p_dx"(#loc23))
193
+ #loc83 = loc("p_dx"(#loc24))
194
+ #loc84 = loc("p_dx"(#loc25))
195
+ #loc85 = loc("p_dx"(#loc26))
196
+ #loc86 = loc("p_dx"(#loc27))
197
+ #loc87 = loc("b_y"(#loc28))
198
+ #loc88 = loc("b_y"(#loc29))
199
+ #loc89 = loc("b_rstd"(#loc30))
200
+ #loc90 = loc("b_dy"(#loc31))
201
+ #loc91 = loc("b_dy"(#loc32))
202
+ #loc92 = loc("b_dx"(#loc33))
203
+ #loc93 = loc("b_dx"(#loc34))
204
+ #loc94 = loc("b_dx"(#loc35))
205
+ #loc95 = loc("b_dx"(#loc36))
206
+ #loc96 = loc("b_dx"(#loc37))
207
+ #loc97 = loc("b_dx"(#loc38))
208
+ #loc98 = loc("b_dx"(#loc39))
209
+ #loc99 = loc("b_dx"(#loc40))
210
+ #loc100 = loc("b_dx"(#loc41))
triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttgir ADDED
@@ -0,0 +1,146 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #blocked = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [8, 1], order = [1, 0]}>
2
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0)
3
+ #loc1 = loc(unknown)
4
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55)
5
+ #loc39 = loc("y"(#loc))
6
+ #loc40 = loc("rstd"(#loc))
7
+ #loc41 = loc("dy"(#loc))
8
+ #loc42 = loc("dx"(#loc))
9
+ #loc43 = loc("eps"(#loc))
10
+ #loc44 = loc("T"(#loc))
11
+ #loc72 = loc("b_dx"(#loc30))
12
+ #loc78 = loc(callsite(#loc1 at #loc72))
13
+ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 8 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} {
14
+ tt.func public @l2norm_bwd_kernel(%y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
15
+ %c128_i64 = arith.constant 128 : i64 loc(#loc1)
16
+ %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked> loc(#loc1)
17
+ %cst_0 = arith.constant dense<0.000000e+00> : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc1)
18
+ %cst_1 = arith.constant dense<128> : tensor<128x1xi64, #blocked> loc(#loc1)
19
+ %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked> loc(#loc1)
20
+ %i_t = tt.get_program_id x : i32 loc(#loc45)
21
+ %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46)
22
+ %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc47)
23
+ %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc48)
24
+ %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)
25
+ %o_t_6 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49)
26
+ %o_t_7 = arith.addi %o_t_6, %o_t_5 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc49)
27
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc50)
28
+ %m_t_8 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50)
29
+ %m_t_9 = arith.cmpi slt, %o_t_7, %m_t_8 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc50)
30
+ %m_x = tt.expand_dims %m_t_9 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi1, #blocked> loc(#loc51)
31
+ %m_x_10 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked}>> loc(#loc52)
32
+ %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)
33
+ %m_x_12 = arith.cmpi slt, %m_x_11, %cst_2 : tensor<1x128xi32, #blocked> loc(#loc53)
34
+ %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc54)
35
+ %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1, #blocked> -> tensor<128x128xi1, #blocked> loc(#loc54)
36
+ %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1, #blocked> loc(#loc54)
37
+ %p_y = tt.expand_dims %o_t_7 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xi64, #blocked> loc(#loc55)
38
+ %p_y_16 = arith.muli %p_y, %cst_1 : tensor<128x1xi64, #blocked> loc(#loc56)
39
+ %p_y_17 = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>, #blocked> loc(#loc57)
40
+ %p_y_18 = tt.addptr %p_y_17, %p_y_16 : tensor<128x1x!tt.ptr<bf16>, #blocked>, tensor<128x1xi64, #blocked> loc(#loc57)
41
+ %p_y_19 = tt.broadcast %p_y_18 : tensor<128x1x!tt.ptr<bf16>, #blocked> -> tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc58)
42
+ %p_y_20 = tt.broadcast %m_x_11 : tensor<1x128xi32, #blocked> -> tensor<128x128xi32, #blocked> loc(#loc58)
43
+ %p_y_21 = tt.addptr %p_y_19, %p_y_20 : tensor<128x128x!tt.ptr<bf16>, #blocked>, tensor<128x128xi32, #blocked> loc(#loc58)
44
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59)
45
+ %p_rstd_22 = tt.addptr %p_rstd, %o_t_7 : tensor<128x!tt.ptr<f32>, #ttg.slice<{dim = 1, parent = #blocked}>>, tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc59)
46
+ %p_dy = tt.splat %dy : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>, #blocked> loc(#loc60)
47
+ %p_dy_23 = tt.addptr %p_dy, %p_y_16 : tensor<128x1x!tt.ptr<bf16>, #blocked>, tensor<128x1xi64, #blocked> loc(#loc60)
48
+ %p_dy_24 = tt.broadcast %p_dy_23 : tensor<128x1x!tt.ptr<bf16>, #blocked> -> tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc61)
49
+ %p_dy_25 = tt.addptr %p_dy_24, %p_y_20 : tensor<128x128x!tt.ptr<bf16>, #blocked>, tensor<128x128xi32, #blocked> loc(#loc61)
50
+ %p_dx = tt.splat %dx : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>, #blocked> loc(#loc62)
51
+ %p_dx_26 = tt.addptr %p_dx, %p_y_16 : tensor<128x1x!tt.ptr<bf16>, #blocked>, tensor<128x1xi64, #blocked> loc(#loc62)
52
+ %p_dx_27 = tt.broadcast %p_dx_26 : tensor<128x1x!tt.ptr<bf16>, #blocked> -> tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc63)
53
+ %p_dx_28 = tt.addptr %p_dx_27, %p_y_20 : tensor<128x128x!tt.ptr<bf16>, #blocked>, tensor<128x128xi32, #blocked> loc(#loc63)
54
+ %b_y = tt.load %p_y_21, %m_x_15, %cst : tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc64)
55
+ %b_y_29 = arith.extf %b_y : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc65)
56
+ %b_rstd = tt.load %p_rstd_22, %m_t_9, %cst_0 : tensor<128x!tt.ptr<f32>, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc66)
57
+ %b_dy = tt.load %p_dy_25, %m_x_15, %cst : tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc67)
58
+ %b_dy_30 = arith.extf %b_dy : tensor<128x128xbf16, #blocked> to tensor<128x128xf32, #blocked> loc(#loc68)
59
+ %b_dx = tt.expand_dims %b_rstd {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> -> tensor<128x1xf32, #blocked> loc(#loc69)
60
+ %b_dx_31 = tt.broadcast %b_dx : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc70)
61
+ %b_dx_32 = arith.mulf %b_dy_30, %b_dx_31 : tensor<128x128xf32, #blocked> loc(#loc70)
62
+ %b_dx_33 = arith.mulf %b_dy_30, %b_y_29 : tensor<128x128xf32, #blocked> loc(#loc71)
63
+ %b_dx_34 = "tt.reduce"(%b_dx_33) <{axis = 1 : i32}> ({
64
+ ^bb0(%b_dx_40: f32 loc(callsite(#loc1 at #loc72)), %b_dx_41: f32 loc(callsite(#loc1 at #loc72))):
65
+ %b_dx_42 = arith.addf %b_dx_40, %b_dx_41 : f32 loc(#loc79)
66
+ tt.reduce.return %b_dx_42 : f32 loc(#loc77)
67
+ }) : (tensor<128x128xf32, #blocked>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked}>> loc(#loc77)
68
+ %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)
69
+ %b_dx_36 = tt.broadcast %b_dx_35 : tensor<128x1xf32, #blocked> -> tensor<128x128xf32, #blocked> loc(#loc74)
70
+ %b_dx_37 = arith.mulf %b_dx_36, %b_y_29 : tensor<128x128xf32, #blocked> loc(#loc74)
71
+ %b_dx_38 = arith.mulf %b_dx_37, %b_dx_31 : tensor<128x128xf32, #blocked> loc(#loc75)
72
+ %b_dx_39 = arith.subf %b_dx_32, %b_dx_38 : tensor<128x128xf32, #blocked> loc(#loc76)
73
+ %0 = arith.truncf %b_dx_39 : tensor<128x128xf32, #blocked> to tensor<128x128xbf16, #blocked> loc(#loc36)
74
+ tt.store %p_dx_28, %0, %m_x_15 : tensor<128x128x!tt.ptr<bf16>, #blocked> loc(#loc37)
75
+ tt.return loc(#loc38)
76
+ } loc(#loc)
77
+ } loc(#loc)
78
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24)
79
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30)
80
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16)
81
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34)
82
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21)
83
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16)
84
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14)
85
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30)
86
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41)
87
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26)
88
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18)
89
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29)
90
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14)
91
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33)
92
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20)
93
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16)
94
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35)
95
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16)
96
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35)
97
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18)
98
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47)
99
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21)
100
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19)
101
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49)
102
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25)
103
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18)
104
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50)
105
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
106
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
107
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58)
108
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69)
109
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75)
110
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36)
111
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27)
112
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19)
113
+ #loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4)
114
+ #loc45 = loc("i_t"(#loc2))
115
+ #loc46 = loc("i_t"(#loc3))
116
+ #loc47 = loc("o_t"(#loc4))
117
+ #loc48 = loc("o_t"(#loc5))
118
+ #loc49 = loc("o_t"(#loc6))
119
+ #loc50 = loc("m_t"(#loc7))
120
+ #loc51 = loc("m_x"(#loc8))
121
+ #loc52 = loc("m_x"(#loc9))
122
+ #loc53 = loc("m_x"(#loc10))
123
+ #loc54 = loc("m_x"(#loc11))
124
+ #loc55 = loc("p_y"(#loc12))
125
+ #loc56 = loc("p_y"(#loc13))
126
+ #loc57 = loc("p_y"(#loc14))
127
+ #loc58 = loc("p_y"(#loc15))
128
+ #loc59 = loc("p_rstd"(#loc16))
129
+ #loc60 = loc("p_dy"(#loc17))
130
+ #loc61 = loc("p_dy"(#loc18))
131
+ #loc62 = loc("p_dx"(#loc19))
132
+ #loc63 = loc("p_dx"(#loc20))
133
+ #loc64 = loc("b_y"(#loc21))
134
+ #loc65 = loc("b_y"(#loc22))
135
+ #loc66 = loc("b_rstd"(#loc23))
136
+ #loc67 = loc("b_dy"(#loc24))
137
+ #loc68 = loc("b_dy"(#loc25))
138
+ #loc69 = loc("b_dx"(#loc26))
139
+ #loc70 = loc("b_dx"(#loc27))
140
+ #loc71 = loc("b_dx"(#loc28))
141
+ #loc73 = loc("b_dx"(#loc32))
142
+ #loc74 = loc("b_dx"(#loc33))
143
+ #loc75 = loc("b_dx"(#loc34))
144
+ #loc76 = loc("b_dx"(#loc35))
145
+ #loc77 = loc(callsite(#loc29 at #loc72))
146
+ #loc79 = loc(callsite(#loc31 at #loc77))
triton/gpu0/AVZNMTOEEYM7LL6RAVSMNAUEOCOHYRYMQVA7AUWCEJA7E7L55K3Q/l2norm_bwd_kernel.ttir ADDED
@@ -0,0 +1,144 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0)
2
+ #loc1 = loc(unknown)
3
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55)
4
+ #loc39 = loc("y"(#loc))
5
+ #loc40 = loc("rstd"(#loc))
6
+ #loc41 = loc("dy"(#loc))
7
+ #loc42 = loc("dx"(#loc))
8
+ #loc43 = loc("eps"(#loc))
9
+ #loc44 = loc("T"(#loc))
10
+ #loc72 = loc("b_dx"(#loc30))
11
+ #loc78 = loc(callsite(#loc1 at #loc72))
12
+ module {
13
+ tt.func public @l2norm_bwd_kernel(%y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
14
+ %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc1)
15
+ %b_rstd = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc45)
16
+ %cst_0 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc1)
17
+ %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc46)
18
+ %c128_i64 = arith.constant 128 : i64 loc(#loc1)
19
+ %i_t = tt.get_program_id x : i32 loc(#loc47)
20
+ %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc48)
21
+ %o_t = arith.muli %i_t_1, %c128_i64 : i64 loc(#loc49)
22
+ %o_t_2 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc50)
23
+ %o_t_3 = arith.extsi %o_t_2 : tensor<128xi32> to tensor<128xi64> loc(#loc51)
24
+ %o_t_4 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc51)
25
+ %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<128xi64> loc(#loc51)
26
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc52)
27
+ %m_t_6 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc52)
28
+ %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<128xi64> loc(#loc52)
29
+ %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc53)
30
+ %m_x_9 = tt.expand_dims %o_t_2 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc54)
31
+ %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc46)
32
+ %m_x_11 = tt.broadcast %m_x_8 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc55)
33
+ %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc55)
34
+ %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<128x128xi1> loc(#loc55)
35
+ %p_y = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc56)
36
+ %p_y_14 = arith.muli %p_y, %cst_0 : tensor<128x1xi64> loc(#loc57)
37
+ %p_y_15 = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc58)
38
+ %p_y_16 = tt.addptr %p_y_15, %p_y_14 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc58)
39
+ %p_y_17 = tt.broadcast %p_y_16 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc59)
40
+ %p_y_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc59)
41
+ %p_y_19 = tt.addptr %p_y_17, %p_y_18 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc59)
42
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>> loc(#loc60)
43
+ %p_rstd_20 = tt.addptr %p_rstd, %o_t_5 : tensor<128x!tt.ptr<f32>>, tensor<128xi64> loc(#loc60)
44
+ %p_dy = tt.splat %dy : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc61)
45
+ %p_dy_21 = tt.addptr %p_dy, %p_y_14 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc61)
46
+ %p_dy_22 = tt.broadcast %p_dy_21 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc62)
47
+ %p_dy_23 = tt.addptr %p_dy_22, %p_y_18 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc62)
48
+ %p_dx = tt.splat %dx : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc63)
49
+ %p_dx_24 = tt.addptr %p_dx, %p_y_14 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc63)
50
+ %p_dx_25 = tt.broadcast %p_dx_24 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc64)
51
+ %p_dx_26 = tt.addptr %p_dx_25, %p_y_18 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc64)
52
+ %b_y = tt.load %p_y_19, %m_x_13, %cst : tensor<128x128x!tt.ptr<bf16>> loc(#loc65)
53
+ %b_y_27 = arith.extf %b_y : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc66)
54
+ %b_rstd_28 = tt.load %p_rstd_20, %m_t_7, %b_rstd : tensor<128x!tt.ptr<f32>> loc(#loc45)
55
+ %b_dy = tt.load %p_dy_23, %m_x_13, %cst : tensor<128x128x!tt.ptr<bf16>> loc(#loc67)
56
+ %b_dy_29 = arith.extf %b_dy : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc68)
57
+ %b_dx = tt.expand_dims %b_rstd_28 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc69)
58
+ %b_dx_30 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc70)
59
+ %b_dx_31 = arith.mulf %b_dy_29, %b_dx_30 : tensor<128x128xf32> loc(#loc70)
60
+ %b_dx_32 = arith.mulf %b_dy_29, %b_y_27 : tensor<128x128xf32> loc(#loc71)
61
+ %b_dx_33 = "tt.reduce"(%b_dx_32) <{axis = 1 : i32}> ({
62
+ ^bb0(%b_dx_39: f32 loc(callsite(#loc1 at #loc72)), %b_dx_40: f32 loc(callsite(#loc1 at #loc72))):
63
+ %b_dx_41 = arith.addf %b_dx_39, %b_dx_40 : f32 loc(#loc79)
64
+ tt.reduce.return %b_dx_41 : f32 loc(#loc77)
65
+ }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc77)
66
+ %b_dx_34 = tt.expand_dims %b_dx_33 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc73)
67
+ %b_dx_35 = tt.broadcast %b_dx_34 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc74)
68
+ %b_dx_36 = arith.mulf %b_dx_35, %b_y_27 : tensor<128x128xf32> loc(#loc74)
69
+ %b_dx_37 = arith.mulf %b_dx_36, %b_dx_30 : tensor<128x128xf32> loc(#loc75)
70
+ %b_dx_38 = arith.subf %b_dx_31, %b_dx_37 : tensor<128x128xf32> loc(#loc76)
71
+ %0 = arith.truncf %b_dx_38 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc36)
72
+ tt.store %p_dx_26, %0, %m_x_13 : tensor<128x128x!tt.ptr<bf16>> loc(#loc37)
73
+ tt.return loc(#loc38)
74
+ } loc(#loc)
75
+ } loc(#loc)
76
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21)
77
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41)
78
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24)
79
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30)
80
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16)
81
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34)
82
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21)
83
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16)
84
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14)
85
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30)
86
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26)
87
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18)
88
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29)
89
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14)
90
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33)
91
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20)
92
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16)
93
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35)
94
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16)
95
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35)
96
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18)
97
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47)
98
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19)
99
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49)
100
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25)
101
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18)
102
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50)
103
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
104
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
105
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58)
106
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69)
107
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75)
108
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36)
109
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27)
110
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19)
111
+ #loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4)
112
+ #loc45 = loc("b_rstd"(#loc2))
113
+ #loc46 = loc("m_x"(#loc3))
114
+ #loc47 = loc("i_t"(#loc4))
115
+ #loc48 = loc("i_t"(#loc5))
116
+ #loc49 = loc("o_t"(#loc6))
117
+ #loc50 = loc("o_t"(#loc7))
118
+ #loc51 = loc("o_t"(#loc8))
119
+ #loc52 = loc("m_t"(#loc9))
120
+ #loc53 = loc("m_x"(#loc10))
121
+ #loc54 = loc("m_x"(#loc11))
122
+ #loc55 = loc("m_x"(#loc12))
123
+ #loc56 = loc("p_y"(#loc13))
124
+ #loc57 = loc("p_y"(#loc14))
125
+ #loc58 = loc("p_y"(#loc15))
126
+ #loc59 = loc("p_y"(#loc16))
127
+ #loc60 = loc("p_rstd"(#loc17))
128
+ #loc61 = loc("p_dy"(#loc18))
129
+ #loc62 = loc("p_dy"(#loc19))
130
+ #loc63 = loc("p_dx"(#loc20))
131
+ #loc64 = loc("p_dx"(#loc21))
132
+ #loc65 = loc("b_y"(#loc22))
133
+ #loc66 = loc("b_y"(#loc23))
134
+ #loc67 = loc("b_dy"(#loc24))
135
+ #loc68 = loc("b_dy"(#loc25))
136
+ #loc69 = loc("b_dx"(#loc26))
137
+ #loc70 = loc("b_dx"(#loc27))
138
+ #loc71 = loc("b_dx"(#loc28))
139
+ #loc73 = loc("b_dx"(#loc32))
140
+ #loc74 = loc("b_dx"(#loc33))
141
+ #loc75 = loc("b_dx"(#loc34))
142
+ #loc76 = loc("b_dx"(#loc35))
143
+ #loc77 = loc(callsite(#loc29 at #loc72))
144
+ #loc79 = loc(callsite(#loc31 at #loc77))
triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/__grp__l2norm_bwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}}
triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.cubin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:891a986d506009060f5cc64e4ab10f2b7bbcaa1fbdb0463144c7091e004da2f5
3
+ size 241800
triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}
triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.llir ADDED
The diff for this file is too large to render. See raw diff
 
triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ptx ADDED
The diff for this file is too large to render. See raw diff
 
triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.source ADDED
@@ -0,0 +1,210 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0)
2
+ #loc45 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":287:0)
3
+ #loc47 = loc(unknown)
4
+ #loc50 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":262:0)
5
+ #loc54 = loc("y"(#loc))
6
+ #loc55 = loc("rstd"(#loc))
7
+ #loc56 = loc("dy"(#loc))
8
+ #loc57 = loc("dx"(#loc))
9
+ #loc58 = loc("eps"(#loc))
10
+ #loc59 = loc("T"(#loc))
11
+ #loc101 = loc("input"(#loc45))
12
+ #loc102 = loc("a"(#loc50))
13
+ #loc103 = loc("b"(#loc50))
14
+ module {
15
+ tt.func public @l2norm_bwd_kernel(%y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
16
+ %i_t = tt.get_program_id x : i32 loc(#loc60)
17
+ %i_t_0 = arith.extsi %i_t : i32 to i64 loc(#loc61)
18
+ %o_t = arith.constant 128 : i32 loc(#loc62)
19
+ %o_t_1 = arith.constant 128 : i64 loc(#loc62)
20
+ %o_t_2 = arith.muli %i_t_0, %o_t_1 : i64 loc(#loc62)
21
+ %o_t_3 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc63)
22
+ %o_t_4 = arith.extsi %o_t_3 : tensor<128xi32> to tensor<128xi64> loc(#loc64)
23
+ %o_t_5 = tt.splat %o_t_2 : i64 -> tensor<128xi64> loc(#loc64)
24
+ %o_t_6 = arith.addi %o_t_5, %o_t_4 : tensor<128xi64> loc(#loc64)
25
+ %o_d = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc65)
26
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc66)
27
+ %m_t_7 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc66)
28
+ %m_t_8 = arith.cmpi slt, %o_t_6, %m_t_7 : tensor<128xi64> loc(#loc66)
29
+ %m_x = tt.expand_dims %m_t_8 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc67)
30
+ %m_x_9 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc68)
31
+ %m_x_10 = arith.constant 128 : i32 loc(#loc69)
32
+ %m_x_11 = arith.constant dense<128> : tensor<1x128xi32> loc(#loc69)
33
+ %m_x_12 = arith.cmpi slt, %m_x_9, %m_x_11 : tensor<1x128xi32> loc(#loc69)
34
+ %m_x_13 = tt.broadcast %m_x : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc70)
35
+ %m_x_14 = tt.broadcast %m_x_12 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc70)
36
+ %m_x_15 = arith.andi %m_x_13, %m_x_14 : tensor<128x128xi1> loc(#loc70)
37
+ %p_y = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc71)
38
+ %p_y_16 = arith.constant 128 : i32 loc(#loc72)
39
+ %p_y_17 = arith.constant 128 : i64 loc(#loc72)
40
+ %p_y_18 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc72)
41
+ %p_y_19 = arith.muli %p_y, %p_y_18 : tensor<128x1xi64> loc(#loc72)
42
+ %p_y_20 = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc73)
43
+ %p_y_21 = tt.addptr %p_y_20, %p_y_19 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc73)
44
+ %p_y_22 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc74)
45
+ %p_y_23 = tt.broadcast %p_y_21 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc75)
46
+ %p_y_24 = tt.broadcast %p_y_22 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc75)
47
+ %p_y_25 = tt.addptr %p_y_23, %p_y_24 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc75)
48
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>> loc(#loc76)
49
+ %p_rstd_26 = tt.addptr %p_rstd, %o_t_6 : tensor<128x!tt.ptr<f32>>, tensor<128xi64> loc(#loc76)
50
+ %p_dy = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc77)
51
+ %p_dy_27 = arith.constant 128 : i32 loc(#loc78)
52
+ %p_dy_28 = arith.constant 128 : i64 loc(#loc78)
53
+ %p_dy_29 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc78)
54
+ %p_dy_30 = arith.muli %p_dy, %p_dy_29 : tensor<128x1xi64> loc(#loc78)
55
+ %p_dy_31 = tt.splat %dy : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc79)
56
+ %p_dy_32 = tt.addptr %p_dy_31, %p_dy_30 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc79)
57
+ %p_dy_33 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc80)
58
+ %p_dy_34 = tt.broadcast %p_dy_32 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc81)
59
+ %p_dy_35 = tt.broadcast %p_dy_33 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc81)
60
+ %p_dy_36 = tt.addptr %p_dy_34, %p_dy_35 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc81)
61
+ %p_dx = tt.expand_dims %o_t_6 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc82)
62
+ %p_dx_37 = arith.constant 128 : i32 loc(#loc83)
63
+ %p_dx_38 = arith.constant 128 : i64 loc(#loc83)
64
+ %p_dx_39 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc83)
65
+ %p_dx_40 = arith.muli %p_dx, %p_dx_39 : tensor<128x1xi64> loc(#loc83)
66
+ %p_dx_41 = tt.splat %dx : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc84)
67
+ %p_dx_42 = tt.addptr %p_dx_41, %p_dx_40 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc84)
68
+ %p_dx_43 = tt.expand_dims %o_d {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc85)
69
+ %p_dx_44 = tt.broadcast %p_dx_42 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc86)
70
+ %p_dx_45 = tt.broadcast %p_dx_43 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc86)
71
+ %p_dx_46 = tt.addptr %p_dx_44, %p_dx_45 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc86)
72
+ %b_y = arith.constant 0.000000e+00 : f32 loc(#loc87)
73
+ %b_y_47 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc87)
74
+ %b_y_48 = arith.truncf %b_y_47 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc87)
75
+ %b_y_49 = tt.load %p_y_25, %m_x_15, %b_y_48 : tensor<128x128x!tt.ptr<bf16>> loc(#loc87)
76
+ %b_y_50 = arith.extf %b_y_49 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc88)
77
+ %b_rstd = arith.constant 0.000000e+00 : f32 loc(#loc89)
78
+ %b_rstd_51 = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc89)
79
+ %b_rstd_52 = tt.load %p_rstd_26, %m_t_8, %b_rstd_51 : tensor<128x!tt.ptr<f32>> loc(#loc89)
80
+ %b_dy = arith.constant 0.000000e+00 : f32 loc(#loc90)
81
+ %b_dy_53 = arith.constant dense<0.000000e+00> : tensor<128x128xf32> loc(#loc90)
82
+ %b_dy_54 = arith.truncf %b_dy_53 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc90)
83
+ %b_dy_55 = tt.load %p_dy_36, %m_x_15, %b_dy_54 : tensor<128x128x!tt.ptr<bf16>> loc(#loc90)
84
+ %b_dy_56 = arith.extf %b_dy_55 : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc91)
85
+ %b_dx = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc92)
86
+ %b_dx_57 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc93)
87
+ %b_dx_58 = arith.mulf %b_dy_56, %b_dx_57 : tensor<128x128xf32> loc(#loc93)
88
+ %b_dx_59 = arith.mulf %b_dy_56, %b_y_50 : tensor<128x128xf32> loc(#loc94)
89
+ %b_dx_60 = tt.call @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%b_dx_59) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc95)
90
+ %b_dx_61 = tt.expand_dims %b_dx_60 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc96)
91
+ %b_dx_62 = tt.broadcast %b_dx_61 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc97)
92
+ %b_dx_63 = arith.mulf %b_dx_62, %b_y_50 : tensor<128x128xf32> loc(#loc97)
93
+ %b_dx_64 = tt.expand_dims %b_rstd_52 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc98)
94
+ %b_dx_65 = tt.broadcast %b_dx_64 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc99)
95
+ %b_dx_66 = arith.mulf %b_dx_63, %b_dx_65 : tensor<128x128xf32> loc(#loc99)
96
+ %b_dx_67 = arith.subf %b_dx_58, %b_dx_66 : tensor<128x128xf32> loc(#loc100)
97
+ %0 = arith.truncf %b_dx_67 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc42)
98
+ tt.store %p_dx_46, %0, %m_x_15 : tensor<128x128x!tt.ptr<bf16>> loc(#loc43)
99
+ tt.return loc(#loc44)
100
+ } loc(#loc)
101
+ tt.func private @triton.language.standard.sum__fp32S128_128S_c1_cFalse_cNone(%input: tensor<128x128xf32> loc("input"(#loc45))) -> tensor<128xf32> attributes {noinline = false} {
102
+ %0 = "tt.reduce"(%input) <{axis = 1 : i32}> ({
103
+ ^bb0(%arg1: f32 loc(unknown), %arg2: f32 loc(unknown)):
104
+ %2 = tt.call @triton.language.standard._sum_combine__fp32_fp32(%arg1, %arg2) : (f32, f32) -> f32 loc(#loc46)
105
+ tt.reduce.return %2 : f32 loc(#loc46)
106
+ }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc46)
107
+ tt.return %0 : tensor<128xf32> loc(#loc48)
108
+ ^bb1: // no predecessors
109
+ %1 = ub.poison : tensor<128xf32> loc(#loc49)
110
+ tt.return %1 : tensor<128xf32> loc(#loc49)
111
+ } loc(#loc45)
112
+ tt.func private @triton.language.standard._sum_combine__fp32_fp32(%a: f32 loc("a"(#loc50)), %b: f32 loc("b"(#loc50))) -> f32 attributes {noinline = false} {
113
+ %0 = arith.addf %a, %b : f32 loc(#loc51)
114
+ tt.return %0 : f32 loc(#loc52)
115
+ ^bb1: // no predecessors
116
+ %1 = ub.poison : f32 loc(#loc53)
117
+ tt.return %1 : f32 loc(#loc53)
118
+ } loc(#loc50)
119
+ } loc(#loc)
120
+ #loc1 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24)
121
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30)
122
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16)
123
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34)
124
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21)
125
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":96:23)
126
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16)
127
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14)
128
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30)
129
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41)
130
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26)
131
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18)
132
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29)
133
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14)
134
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:37)
135
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33)
136
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20)
137
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:20)
138
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:31)
139
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16)
140
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:39)
141
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35)
142
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:20)
143
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:31)
144
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16)
145
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:39)
146
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35)
147
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18)
148
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47)
149
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21)
150
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19)
151
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49)
152
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25)
153
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18)
154
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50)
155
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55)
156
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58)
157
+ #loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69)
158
+ #loc39 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:82)
159
+ #loc40 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75)
160
+ #loc41 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36)
161
+ #loc42 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27)
162
+ #loc43 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19)
163
+ #loc44 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4)
164
+ #loc46 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
165
+ #loc48 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:11)
166
+ #loc49 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:4)
167
+ #loc51 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
168
+ #loc52 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:11)
169
+ #loc53 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:4)
170
+ #loc60 = loc("i_t"(#loc1))
171
+ #loc61 = loc("i_t"(#loc2))
172
+ #loc62 = loc("o_t"(#loc3))
173
+ #loc63 = loc("o_t"(#loc4))
174
+ #loc64 = loc("o_t"(#loc5))
175
+ #loc65 = loc("o_d"(#loc6))
176
+ #loc66 = loc("m_t"(#loc7))
177
+ #loc67 = loc("m_x"(#loc8))
178
+ #loc68 = loc("m_x"(#loc9))
179
+ #loc69 = loc("m_x"(#loc10))
180
+ #loc70 = loc("m_x"(#loc11))
181
+ #loc71 = loc("p_y"(#loc12))
182
+ #loc72 = loc("p_y"(#loc13))
183
+ #loc73 = loc("p_y"(#loc14))
184
+ #loc74 = loc("p_y"(#loc15))
185
+ #loc75 = loc("p_y"(#loc16))
186
+ #loc76 = loc("p_rstd"(#loc17))
187
+ #loc77 = loc("p_dy"(#loc18))
188
+ #loc78 = loc("p_dy"(#loc19))
189
+ #loc79 = loc("p_dy"(#loc20))
190
+ #loc80 = loc("p_dy"(#loc21))
191
+ #loc81 = loc("p_dy"(#loc22))
192
+ #loc82 = loc("p_dx"(#loc23))
193
+ #loc83 = loc("p_dx"(#loc24))
194
+ #loc84 = loc("p_dx"(#loc25))
195
+ #loc85 = loc("p_dx"(#loc26))
196
+ #loc86 = loc("p_dx"(#loc27))
197
+ #loc87 = loc("b_y"(#loc28))
198
+ #loc88 = loc("b_y"(#loc29))
199
+ #loc89 = loc("b_rstd"(#loc30))
200
+ #loc90 = loc("b_dy"(#loc31))
201
+ #loc91 = loc("b_dy"(#loc32))
202
+ #loc92 = loc("b_dx"(#loc33))
203
+ #loc93 = loc("b_dx"(#loc34))
204
+ #loc94 = loc("b_dx"(#loc35))
205
+ #loc95 = loc("b_dx"(#loc36))
206
+ #loc96 = loc("b_dx"(#loc37))
207
+ #loc97 = loc("b_dx"(#loc38))
208
+ #loc98 = loc("b_dx"(#loc39))
209
+ #loc99 = loc("b_dx"(#loc40))
210
+ #loc100 = loc("b_dx"(#loc41))
triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttgir ADDED
@@ -0,0 +1,154 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #blocked = #ttg.blocked<{sizePerThread = [2], threadsPerWarp = [32], warpsPerCTA = [2], order = [0]}>
2
+ #blocked1 = #ttg.blocked<{sizePerThread = [1, 8], threadsPerWarp = [2, 16], warpsPerCTA = [2, 1], order = [1, 0]}>
3
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0)
4
+ #loc1 = loc(unknown)
5
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55)
6
+ #loc39 = loc("y"(#loc))
7
+ #loc40 = loc("rstd"(#loc))
8
+ #loc41 = loc("dy"(#loc))
9
+ #loc42 = loc("dx"(#loc))
10
+ #loc43 = loc("eps"(#loc))
11
+ #loc44 = loc("T"(#loc))
12
+ #loc72 = loc("b_dx"(#loc30))
13
+ #loc78 = loc(callsite(#loc1 at #loc72))
14
+ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 2 : i32, ttg.target = "cuda:100", "ttg.threads-per-warp" = 32 : i32} {
15
+ tt.func public @l2norm_bwd_kernel(%y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
16
+ %c128_i64 = arith.constant 128 : i64 loc(#loc1)
17
+ %cst = arith.constant dense<0.000000e+00> : tensor<128xf32, #blocked> loc(#loc1)
18
+ %cst_0 = arith.constant dense<0.000000e+00> : tensor<128x128xbf16, #blocked1> loc(#loc1)
19
+ %cst_1 = arith.constant dense<128> : tensor<128x1xi64, #blocked1> loc(#loc1)
20
+ %cst_2 = arith.constant dense<128> : tensor<1x128xi32, #blocked1> loc(#loc1)
21
+ %i_t = tt.get_program_id x : i32 loc(#loc45)
22
+ %i_t_3 = arith.extsi %i_t : i32 to i64 loc(#loc46)
23
+ %o_t = arith.muli %i_t_3, %c128_i64 : i64 loc(#loc47)
24
+ %o_t_4 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc48)
25
+ %o_t_5 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #blocked> loc(#loc48)
26
+ %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)
27
+ %o_t_7 = arith.extsi %o_t_5 : tensor<128xi32, #blocked> to tensor<128xi64, #blocked> loc(#loc49)
28
+ %o_t_8 = tt.splat %o_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49)
29
+ %o_t_9 = tt.splat %o_t : i64 -> tensor<128xi64, #blocked> loc(#loc49)
30
+ %o_t_10 = arith.addi %o_t_8, %o_t_6 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc49)
31
+ %o_t_11 = arith.addi %o_t_9, %o_t_7 : tensor<128xi64, #blocked> loc(#loc49)
32
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc50)
33
+ %m_t_12 = tt.splat %m_t : i64 -> tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50)
34
+ %m_t_13 = tt.splat %m_t : i64 -> tensor<128xi64, #blocked> loc(#loc50)
35
+ %m_t_14 = arith.cmpi slt, %o_t_10, %m_t_12 : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc50)
36
+ %m_t_15 = arith.cmpi slt, %o_t_11, %m_t_13 : tensor<128xi64, #blocked> loc(#loc50)
37
+ %m_x = tt.expand_dims %m_t_14 {axis = 1 : i32} : tensor<128xi1, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi1, #blocked1> loc(#loc51)
38
+ %m_x_16 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32, #ttg.slice<{dim = 0, parent = #blocked1}>> loc(#loc52)
39
+ %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)
40
+ %m_x_18 = arith.cmpi slt, %m_x_17, %cst_2 : tensor<1x128xi32, #blocked1> loc(#loc53)
41
+ %m_x_19 = tt.broadcast %m_x : tensor<128x1xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54)
42
+ %m_x_20 = tt.broadcast %m_x_18 : tensor<1x128xi1, #blocked1> -> tensor<128x128xi1, #blocked1> loc(#loc54)
43
+ %m_x_21 = arith.andi %m_x_19, %m_x_20 : tensor<128x128xi1, #blocked1> loc(#loc54)
44
+ %p_y = tt.expand_dims %o_t_10 {axis = 1 : i32} : tensor<128xi64, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xi64, #blocked1> loc(#loc55)
45
+ %p_y_22 = arith.muli %p_y, %cst_1 : tensor<128x1xi64, #blocked1> loc(#loc56)
46
+ %p_y_23 = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>, #blocked1> loc(#loc57)
47
+ %p_y_24 = tt.addptr %p_y_23, %p_y_22 : tensor<128x1x!tt.ptr<bf16>, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc57)
48
+ %p_y_25 = tt.broadcast %p_y_24 : tensor<128x1x!tt.ptr<bf16>, #blocked1> -> tensor<128x128x!tt.ptr<bf16>, #blocked1> loc(#loc58)
49
+ %p_y_26 = tt.broadcast %m_x_17 : tensor<1x128xi32, #blocked1> -> tensor<128x128xi32, #blocked1> loc(#loc58)
50
+ %p_y_27 = tt.addptr %p_y_25, %p_y_26 : tensor<128x128x!tt.ptr<bf16>, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc58)
51
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>, #blocked> loc(#loc59)
52
+ %p_rstd_28 = tt.addptr %p_rstd, %o_t_11 : tensor<128x!tt.ptr<f32>, #blocked>, tensor<128xi64, #blocked> loc(#loc59)
53
+ %p_dy = tt.splat %dy : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>, #blocked1> loc(#loc60)
54
+ %p_dy_29 = tt.addptr %p_dy, %p_y_22 : tensor<128x1x!tt.ptr<bf16>, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc60)
55
+ %p_dy_30 = tt.broadcast %p_dy_29 : tensor<128x1x!tt.ptr<bf16>, #blocked1> -> tensor<128x128x!tt.ptr<bf16>, #blocked1> loc(#loc61)
56
+ %p_dy_31 = tt.addptr %p_dy_30, %p_y_26 : tensor<128x128x!tt.ptr<bf16>, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc61)
57
+ %p_dx = tt.splat %dx : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>, #blocked1> loc(#loc62)
58
+ %p_dx_32 = tt.addptr %p_dx, %p_y_22 : tensor<128x1x!tt.ptr<bf16>, #blocked1>, tensor<128x1xi64, #blocked1> loc(#loc62)
59
+ %p_dx_33 = tt.broadcast %p_dx_32 : tensor<128x1x!tt.ptr<bf16>, #blocked1> -> tensor<128x128x!tt.ptr<bf16>, #blocked1> loc(#loc63)
60
+ %p_dx_34 = tt.addptr %p_dx_33, %p_y_26 : tensor<128x128x!tt.ptr<bf16>, #blocked1>, tensor<128x128xi32, #blocked1> loc(#loc63)
61
+ %b_y = tt.load %p_y_27, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr<bf16>, #blocked1> loc(#loc64)
62
+ %b_y_35 = arith.extf %b_y : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc65)
63
+ %b_rstd = tt.load %p_rstd_28, %m_t_15, %cst : tensor<128x!tt.ptr<f32>, #blocked> loc(#loc66)
64
+ %b_dy = tt.load %p_dy_31, %m_x_21, %cst_0 : tensor<128x128x!tt.ptr<bf16>, #blocked1> loc(#loc67)
65
+ %b_dy_36 = arith.extf %b_dy : tensor<128x128xbf16, #blocked1> to tensor<128x128xf32, #blocked1> loc(#loc68)
66
+ %b_dx = ttg.convert_layout %b_rstd : tensor<128xf32, #blocked> -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc69)
67
+ %b_dx_37 = tt.expand_dims %b_dx {axis = 1 : i32} : tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> -> tensor<128x1xf32, #blocked1> loc(#loc69)
68
+ %b_dx_38 = tt.broadcast %b_dx_37 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc70)
69
+ %b_dx_39 = arith.mulf %b_dy_36, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc70)
70
+ %b_dx_40 = arith.mulf %b_dy_36, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc71)
71
+ %b_dx_41 = "tt.reduce"(%b_dx_40) <{axis = 1 : i32}> ({
72
+ ^bb0(%b_dx_47: f32 loc(callsite(#loc1 at #loc72)), %b_dx_48: f32 loc(callsite(#loc1 at #loc72))):
73
+ %b_dx_49 = arith.addf %b_dx_47, %b_dx_48 : f32 loc(#loc79)
74
+ tt.reduce.return %b_dx_49 : f32 loc(#loc77)
75
+ }) : (tensor<128x128xf32, #blocked1>) -> tensor<128xf32, #ttg.slice<{dim = 1, parent = #blocked1}>> loc(#loc77)
76
+ %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)
77
+ %b_dx_43 = tt.broadcast %b_dx_42 : tensor<128x1xf32, #blocked1> -> tensor<128x128xf32, #blocked1> loc(#loc74)
78
+ %b_dx_44 = arith.mulf %b_dx_43, %b_y_35 : tensor<128x128xf32, #blocked1> loc(#loc74)
79
+ %b_dx_45 = arith.mulf %b_dx_44, %b_dx_38 : tensor<128x128xf32, #blocked1> loc(#loc75)
80
+ %b_dx_46 = arith.subf %b_dx_39, %b_dx_45 : tensor<128x128xf32, #blocked1> loc(#loc76)
81
+ %0 = arith.truncf %b_dx_46 : tensor<128x128xf32, #blocked1> to tensor<128x128xbf16, #blocked1> loc(#loc36)
82
+ tt.store %p_dx_34, %0, %m_x_21 : tensor<128x128x!tt.ptr<bf16>, #blocked1> loc(#loc37)
83
+ tt.return loc(#loc38)
84
+ } loc(#loc)
85
+ } loc(#loc)
86
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24)
87
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30)
88
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16)
89
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34)
90
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21)
91
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16)
92
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14)
93
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30)
94
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41)
95
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26)
96
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18)
97
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29)
98
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14)
99
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33)
100
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20)
101
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16)
102
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35)
103
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16)
104
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35)
105
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18)
106
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47)
107
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21)
108
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19)
109
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49)
110
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25)
111
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18)
112
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50)
113
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
114
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
115
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58)
116
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69)
117
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75)
118
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36)
119
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27)
120
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19)
121
+ #loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4)
122
+ #loc45 = loc("i_t"(#loc2))
123
+ #loc46 = loc("i_t"(#loc3))
124
+ #loc47 = loc("o_t"(#loc4))
125
+ #loc48 = loc("o_t"(#loc5))
126
+ #loc49 = loc("o_t"(#loc6))
127
+ #loc50 = loc("m_t"(#loc7))
128
+ #loc51 = loc("m_x"(#loc8))
129
+ #loc52 = loc("m_x"(#loc9))
130
+ #loc53 = loc("m_x"(#loc10))
131
+ #loc54 = loc("m_x"(#loc11))
132
+ #loc55 = loc("p_y"(#loc12))
133
+ #loc56 = loc("p_y"(#loc13))
134
+ #loc57 = loc("p_y"(#loc14))
135
+ #loc58 = loc("p_y"(#loc15))
136
+ #loc59 = loc("p_rstd"(#loc16))
137
+ #loc60 = loc("p_dy"(#loc17))
138
+ #loc61 = loc("p_dy"(#loc18))
139
+ #loc62 = loc("p_dx"(#loc19))
140
+ #loc63 = loc("p_dx"(#loc20))
141
+ #loc64 = loc("b_y"(#loc21))
142
+ #loc65 = loc("b_y"(#loc22))
143
+ #loc66 = loc("b_rstd"(#loc23))
144
+ #loc67 = loc("b_dy"(#loc24))
145
+ #loc68 = loc("b_dy"(#loc25))
146
+ #loc69 = loc("b_dx"(#loc26))
147
+ #loc70 = loc("b_dx"(#loc27))
148
+ #loc71 = loc("b_dx"(#loc28))
149
+ #loc73 = loc("b_dx"(#loc32))
150
+ #loc74 = loc("b_dx"(#loc33))
151
+ #loc75 = loc("b_dx"(#loc34))
152
+ #loc76 = loc("b_dx"(#loc35))
153
+ #loc77 = loc(callsite(#loc29 at #loc72))
154
+ #loc79 = loc(callsite(#loc31 at #loc77))
triton/gpu0/AXWDTT3IKG4XFV2WZ2GR2YK7L7GXEUZ4UEPVIJDYBIHYWKM62LOA/l2norm_bwd_kernel.ttir ADDED
@@ -0,0 +1,144 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #loc = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":93:0)
2
+ #loc1 = loc(unknown)
3
+ #loc30 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:55)
4
+ #loc39 = loc("y"(#loc))
5
+ #loc40 = loc("rstd"(#loc))
6
+ #loc41 = loc("dy"(#loc))
7
+ #loc42 = loc("dx"(#loc))
8
+ #loc43 = loc("eps"(#loc))
9
+ #loc44 = loc("T"(#loc))
10
+ #loc72 = loc("b_dx"(#loc30))
11
+ #loc78 = loc(callsite(#loc1 at #loc72))
12
+ module {
13
+ tt.func public @l2norm_bwd_kernel(%y: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("y"(#loc)), %rstd: !tt.ptr<f32> {tt.divisibility = 16 : i32} loc("rstd"(#loc)), %dy: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dy"(#loc)), %dx: !tt.ptr<bf16> {tt.divisibility = 16 : i32} loc("dx"(#loc)), %eps: f32 loc("eps"(#loc)), %T: i32 loc("T"(#loc))) attributes {noinline = false} {
14
+ %cst = arith.constant dense<0.000000e+00> : tensor<128x128xbf16> loc(#loc1)
15
+ %b_rstd = arith.constant dense<0.000000e+00> : tensor<128xf32> loc(#loc45)
16
+ %cst_0 = arith.constant dense<128> : tensor<128x1xi64> loc(#loc1)
17
+ %m_x = arith.constant dense<128> : tensor<1x128xi32> loc(#loc46)
18
+ %c128_i64 = arith.constant 128 : i64 loc(#loc1)
19
+ %i_t = tt.get_program_id x : i32 loc(#loc47)
20
+ %i_t_1 = arith.extsi %i_t : i32 to i64 loc(#loc48)
21
+ %o_t = arith.muli %i_t_1, %c128_i64 : i64 loc(#loc49)
22
+ %o_t_2 = tt.make_range {end = 128 : i32, start = 0 : i32} : tensor<128xi32> loc(#loc50)
23
+ %o_t_3 = arith.extsi %o_t_2 : tensor<128xi32> to tensor<128xi64> loc(#loc51)
24
+ %o_t_4 = tt.splat %o_t : i64 -> tensor<128xi64> loc(#loc51)
25
+ %o_t_5 = arith.addi %o_t_4, %o_t_3 : tensor<128xi64> loc(#loc51)
26
+ %m_t = arith.extsi %T : i32 to i64 loc(#loc52)
27
+ %m_t_6 = tt.splat %m_t : i64 -> tensor<128xi64> loc(#loc52)
28
+ %m_t_7 = arith.cmpi slt, %o_t_5, %m_t_6 : tensor<128xi64> loc(#loc52)
29
+ %m_x_8 = tt.expand_dims %m_t_7 {axis = 1 : i32} : tensor<128xi1> -> tensor<128x1xi1> loc(#loc53)
30
+ %m_x_9 = tt.expand_dims %o_t_2 {axis = 0 : i32} : tensor<128xi32> -> tensor<1x128xi32> loc(#loc54)
31
+ %m_x_10 = arith.cmpi slt, %m_x_9, %m_x : tensor<1x128xi32> loc(#loc46)
32
+ %m_x_11 = tt.broadcast %m_x_8 : tensor<128x1xi1> -> tensor<128x128xi1> loc(#loc55)
33
+ %m_x_12 = tt.broadcast %m_x_10 : tensor<1x128xi1> -> tensor<128x128xi1> loc(#loc55)
34
+ %m_x_13 = arith.andi %m_x_11, %m_x_12 : tensor<128x128xi1> loc(#loc55)
35
+ %p_y = tt.expand_dims %o_t_5 {axis = 1 : i32} : tensor<128xi64> -> tensor<128x1xi64> loc(#loc56)
36
+ %p_y_14 = arith.muli %p_y, %cst_0 : tensor<128x1xi64> loc(#loc57)
37
+ %p_y_15 = tt.splat %y : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc58)
38
+ %p_y_16 = tt.addptr %p_y_15, %p_y_14 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc58)
39
+ %p_y_17 = tt.broadcast %p_y_16 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc59)
40
+ %p_y_18 = tt.broadcast %m_x_9 : tensor<1x128xi32> -> tensor<128x128xi32> loc(#loc59)
41
+ %p_y_19 = tt.addptr %p_y_17, %p_y_18 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc59)
42
+ %p_rstd = tt.splat %rstd : !tt.ptr<f32> -> tensor<128x!tt.ptr<f32>> loc(#loc60)
43
+ %p_rstd_20 = tt.addptr %p_rstd, %o_t_5 : tensor<128x!tt.ptr<f32>>, tensor<128xi64> loc(#loc60)
44
+ %p_dy = tt.splat %dy : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc61)
45
+ %p_dy_21 = tt.addptr %p_dy, %p_y_14 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc61)
46
+ %p_dy_22 = tt.broadcast %p_dy_21 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc62)
47
+ %p_dy_23 = tt.addptr %p_dy_22, %p_y_18 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc62)
48
+ %p_dx = tt.splat %dx : !tt.ptr<bf16> -> tensor<128x1x!tt.ptr<bf16>> loc(#loc63)
49
+ %p_dx_24 = tt.addptr %p_dx, %p_y_14 : tensor<128x1x!tt.ptr<bf16>>, tensor<128x1xi64> loc(#loc63)
50
+ %p_dx_25 = tt.broadcast %p_dx_24 : tensor<128x1x!tt.ptr<bf16>> -> tensor<128x128x!tt.ptr<bf16>> loc(#loc64)
51
+ %p_dx_26 = tt.addptr %p_dx_25, %p_y_18 : tensor<128x128x!tt.ptr<bf16>>, tensor<128x128xi32> loc(#loc64)
52
+ %b_y = tt.load %p_y_19, %m_x_13, %cst : tensor<128x128x!tt.ptr<bf16>> loc(#loc65)
53
+ %b_y_27 = arith.extf %b_y : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc66)
54
+ %b_rstd_28 = tt.load %p_rstd_20, %m_t_7, %b_rstd : tensor<128x!tt.ptr<f32>> loc(#loc45)
55
+ %b_dy = tt.load %p_dy_23, %m_x_13, %cst : tensor<128x128x!tt.ptr<bf16>> loc(#loc67)
56
+ %b_dy_29 = arith.extf %b_dy : tensor<128x128xbf16> to tensor<128x128xf32> loc(#loc68)
57
+ %b_dx = tt.expand_dims %b_rstd_28 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc69)
58
+ %b_dx_30 = tt.broadcast %b_dx : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc70)
59
+ %b_dx_31 = arith.mulf %b_dy_29, %b_dx_30 : tensor<128x128xf32> loc(#loc70)
60
+ %b_dx_32 = arith.mulf %b_dy_29, %b_y_27 : tensor<128x128xf32> loc(#loc71)
61
+ %b_dx_33 = "tt.reduce"(%b_dx_32) <{axis = 1 : i32}> ({
62
+ ^bb0(%b_dx_39: f32 loc(callsite(#loc1 at #loc72)), %b_dx_40: f32 loc(callsite(#loc1 at #loc72))):
63
+ %b_dx_41 = arith.addf %b_dx_39, %b_dx_40 : f32 loc(#loc79)
64
+ tt.reduce.return %b_dx_41 : f32 loc(#loc77)
65
+ }) : (tensor<128x128xf32>) -> tensor<128xf32> loc(#loc77)
66
+ %b_dx_34 = tt.expand_dims %b_dx_33 {axis = 1 : i32} : tensor<128xf32> -> tensor<128x1xf32> loc(#loc73)
67
+ %b_dx_35 = tt.broadcast %b_dx_34 : tensor<128x1xf32> -> tensor<128x128xf32> loc(#loc74)
68
+ %b_dx_36 = arith.mulf %b_dx_35, %b_y_27 : tensor<128x128xf32> loc(#loc74)
69
+ %b_dx_37 = arith.mulf %b_dx_36, %b_dx_30 : tensor<128x128xf32> loc(#loc75)
70
+ %b_dx_38 = arith.subf %b_dx_31, %b_dx_37 : tensor<128x128xf32> loc(#loc76)
71
+ %0 = arith.truncf %b_dx_38 : tensor<128x128xf32> to tensor<128x128xbf16> loc(#loc36)
72
+ tt.store %p_dx_26, %0, %m_x_13 : tensor<128x128x!tt.ptr<bf16>> loc(#loc37)
73
+ tt.return loc(#loc38)
74
+ } loc(#loc)
75
+ } loc(#loc)
76
+ #loc2 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":105:21)
77
+ #loc3 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:41)
78
+ #loc4 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:24)
79
+ #loc5 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":94:30)
80
+ #loc6 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:16)
81
+ #loc7 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:34)
82
+ #loc8 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":95:21)
83
+ #loc9 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":97:16)
84
+ #loc10 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:14)
85
+ #loc11 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:30)
86
+ #loc12 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":98:26)
87
+ #loc13 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:18)
88
+ #loc14 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:29)
89
+ #loc15 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:14)
90
+ #loc16 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":99:33)
91
+ #loc17 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":100:20)
92
+ #loc18 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:16)
93
+ #loc19 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":101:35)
94
+ #loc20 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:16)
95
+ #loc21 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":102:35)
96
+ #loc22 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:18)
97
+ #loc23 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":104:47)
98
+ #loc24 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:19)
99
+ #loc25 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":106:49)
100
+ #loc26 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:25)
101
+ #loc27 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:18)
102
+ #loc28 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:50)
103
+ #loc29 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":293:36)
104
+ #loc31 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/triton/language/standard.py":263:15)
105
+ #loc32 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:58)
106
+ #loc33 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:69)
107
+ #loc34 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:75)
108
+ #loc35 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":107:36)
109
+ #loc36 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:27)
110
+ #loc37 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:19)
111
+ #loc38 = loc("/home/silver/affine/.venv/lib/python3.12/site-packages/fla/modules/norm/l2norm.py":108:4)
112
+ #loc45 = loc("b_rstd"(#loc2))
113
+ #loc46 = loc("m_x"(#loc3))
114
+ #loc47 = loc("i_t"(#loc4))
115
+ #loc48 = loc("i_t"(#loc5))
116
+ #loc49 = loc("o_t"(#loc6))
117
+ #loc50 = loc("o_t"(#loc7))
118
+ #loc51 = loc("o_t"(#loc8))
119
+ #loc52 = loc("m_t"(#loc9))
120
+ #loc53 = loc("m_x"(#loc10))
121
+ #loc54 = loc("m_x"(#loc11))
122
+ #loc55 = loc("m_x"(#loc12))
123
+ #loc56 = loc("p_y"(#loc13))
124
+ #loc57 = loc("p_y"(#loc14))
125
+ #loc58 = loc("p_y"(#loc15))
126
+ #loc59 = loc("p_y"(#loc16))
127
+ #loc60 = loc("p_rstd"(#loc17))
128
+ #loc61 = loc("p_dy"(#loc18))
129
+ #loc62 = loc("p_dy"(#loc19))
130
+ #loc63 = loc("p_dx"(#loc20))
131
+ #loc64 = loc("p_dx"(#loc21))
132
+ #loc65 = loc("b_y"(#loc22))
133
+ #loc66 = loc("b_y"(#loc23))
134
+ #loc67 = loc("b_dy"(#loc24))
135
+ #loc68 = loc("b_dy"(#loc25))
136
+ #loc69 = loc("b_dx"(#loc26))
137
+ #loc70 = loc("b_dx"(#loc27))
138
+ #loc71 = loc("b_dx"(#loc28))
139
+ #loc73 = loc("b_dx"(#loc32))
140
+ #loc74 = loc("b_dx"(#loc33))
141
+ #loc75 = loc("b_dx"(#loc34))
142
+ #loc76 = loc("b_dx"(#loc35))
143
+ #loc77 = loc(callsite(#loc29 at #loc72))
144
+ #loc79 = loc(callsite(#loc31 at #loc77))
triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/__grp__l2norm_fwd_kernel.json ADDED
@@ -0,0 +1 @@
 
 
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"}}
triton/gpu0/AYNNBEP2C4BWUBPD2FUJPNMVGLRIMC6DGANTGH3CMSV37SGVUQJA/l2norm_fwd_kernel.cubin ADDED
Binary file (15.2 kB). View file