File size: 5,722 Bytes
9425aed | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 | // ════════════════════════════════════════════════════════════════
// SOV_PIPELINE.MLIR — Polyhedral fusion spec
// MLIR affine/linalg pipeline: U * rho * U† → SINGLE KERNEL
// Targets: ARM64 SVE2 | x86_64 AVX-512 | NVIDIA PTX
//
// Compile pipeline:
// mlir-opt --affine-loop-fusion --linalg-tile="tile-sizes=16,16"
// --vectorize --gpu-kernel-outlining
// --convert-linalg-to-loops --convert-vector-to-scf
// --convert-scf-to-llvm
// sov_pipeline.mlir -o sov_llvm.mlir
// ════════════════════════════════════════════════════════════════
module @sov_gemm_fused {
// ── Types ──────────────────────────────────────────────────────
// complex<f64> represented as !sov.complex = vector<2xf64>
// Density matrix: memref<?x?xvector<2xf64>>
// ── exp(H) computation (scaling & squaring) ───────────────────
func.func private @sov_exp_ham(
%H : memref<?x?xvector<2xf64>>,
%dt : f64
) -> memref<?x?xvector<2xf64>>
// ── U† = conj(transpose(U)) ───────────────────────────────────
func.func private @sov_adjoint(
%U : memref<?x?xvector<2xf64>>
) -> memref<?x?xvector<2xf64>>
// ── Plasma verify (inlined into kernel) ───────────────────────
func.func private @sov_plasma_verify_mlir(
%rho : memref<?x?xvector<2xf64>>
) -> i1
// ── Blake3 hash ───────────────────────────────────────────────
func.func private @sov_blake3_hash(
%rho : memref<?x?xvector<2xf64>>,
%hash : memref<32xi8>
)
// ── Ed25519 sign ──────────────────────────────────────────────
func.func private @sov_ed25519_sign(
%hash : memref<32xi8>,
%sk : memref<32xi8>,
%sig : memref<64xi8>
)
// ── MAIN FUSED KERNEL ─────────────────────────────────────────
// Computes: out_rho = exp(-i*dt*H) * rho * exp(+i*dt*H)
// Then: plasma verify + blake3 hash + ed25519 sign
// MLIR linalg.matmul fusion merges tmp=U*rho and out=tmp*U†
// into ONE kernel (zero intermediate materialisation on GPU).
func.func @sov_fused_uru(
%H : memref<?x?xvector<2xf64>>,
%rho : memref<?x?xvector<2xf64>>,
%dt : f64,
%sk : memref<32xi8>,
%pk : memref<32xi8>,
%out_rho : memref<?x?xvector<2xf64>>,
%out_hash: memref<32xi8>,
%out_sig : memref<64xi8>
) {
%n = memref.dim %H, %c0 : memref<?x?xvector<2xf64>>
// 1. Compute U = exp(-i * dt * H)
%U = func.call @sov_exp_ham(%H, %dt) : (memref<?x?xvector<2xf64>>, f64) -> memref<?x?xvector<2xf64>>
%Udag = func.call @sov_adjoint(%U) : (memref<?x?xvector<2xf64>>) -> memref<?x?xvector<2xf64>>
// 2. tmp = U * rho (linalg: affine_map column-major, tile 16x16)
%tmp = memref.alloc(%n, %n) : memref<?x?xvector<2xf64>>
linalg.matmul
ins(%U, %rho : memref<?x?xvector<2xf64>>, memref<?x?xvector<2xf64>>)
outs(%tmp : memref<?x?xvector<2xf64>>)
// 3. out_rho = tmp * U† (fused with step 2 by --affine-loop-fusion)
linalg.matmul
ins(%tmp, %Udag : memref<?x?xvector<2xf64>>, memref<?x?xvector<2xf64>>)
outs(%out_rho : memref<?x?xvector<2xf64>>)
// 4. Plasma verify (constant-time, inlined)
%ok = func.call @sov_plasma_verify_mlir(%out_rho) : (memref<?x?xvector<2xf64>>) -> i1
cf.assert %ok, "PLASMA: out_rho failed density check"
// 5. Blake3 hash of out_rho
func.call @sov_blake3_hash(%out_rho, %out_hash)
: (memref<?x?xvector<2xf64>>, memref<32xi8>) -> ()
// 6. Ed25519 sign
func.call @sov_ed25519_sign(%out_hash, %sk, %out_sig)
: (memref<32xi8>, memref<32xi8>, memref<64xi8>) -> ()
memref.dealloc %tmp : memref<?x?xvector<2xf64>>
return
}
// ── GPU VARIANT (outlined by --gpu-kernel-outlining) ──────────
// After outlining, linalg.matmul loops become gpu.launch regions
// targeting PTX (nvptx64) or SPIR-V (AMD/Intel) depending on llc target.
func.func @sov_fused_uru_gpu(
%H : memref<?x?xvector<2xf64>>,
%rho : memref<?x?xvector<2xf64>>,
%dt : f64,
%sk : memref<32xi8>,
%pk : memref<32xi8>,
%out_rho : memref<?x?xvector<2xf64>>,
%out_hash: memref<32xi8>,
%out_sig : memref<64xi8>
) {
// Host: compute U, U†, allocate device buffers
// Device: fused matmul kernel (PTX/SPIR-V)
// Host: plasma verify + hash + sign (constant-time, CPU)
func.call @sov_fused_uru(%H, %rho, %dt, %sk, %pk, %out_rho, %out_hash, %out_sig)
: (memref<?x?xvector<2xf64>>, memref<?x?xvector<2xf64>>, f64,
memref<32xi8>, memref<32xi8>, memref<?x?xvector<2xf64>>,
memref<32xi8>, memref<64xi8>) -> ()
return
}
// Constants
arith.constant %c0 = 0 : index
} // module @sov_gemm_fused
|