[Mlir-commits] [mlir] [mlir] Add sm_90a GEMM test 128x128x128 (F32 += F16 * F16) (PR #69913)
llvmlistbot at llvm.org
llvmlistbot at llvm.org
Mon Oct 23 03:36:39 PDT 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-mlir-gpu
Author: Guray Ozen (grypp)
<details>
<summary>Changes</summary>
This PR adds a test that performs GEMM 128x128x128 (F32 += F16 * F16). It uses `sm_90a` features in NVGPU dialect.
Simplified algorithm is as follows:
**Prologue**
```
mgroup = mbarriers.init x 2
tma.load ... shmem_buffer_lhs<0 x 128 x 64>
tma.load ... shmem_buffer_rhs<0 x 64 x 64>
tma.load ... shmem_buffer_rhs<0 x 64 x 64>
mbarrier.expect_tx 32768
tma.load ... shmem_buffer_lhs<1 x 128 x 64>
tma.load ... shmem_buffer_rhs<1 x 64 x 64>
tma.load ... shmem_buffer_rhs<1 x 64 x 64>
mbarrier.expect_tx 32768
```
**Mainloop**
```
matrixD =
for(i = 0;...2) {
mbarrier.try_wait [i]
lhs = shmem_buffer_lhs<pipe x 128 x 64>
rhs = shmem_buffer_rhs<pipe x 64 x 128>
yield nvgpu.warpgroup.mma (lhs, rhs)
// Expanded : nvgpu.warpgroup.mma [128][128]+=[128][64]*[64][128]
// wgmma.m64n128k16(A[0:64][0:16] * B[0:16][0:128])
// wgmma.m64n128k16(A[0:64][16:32] * B[16:32][0:128])
// wgmma.m64n128k16(A[0:64][32:48] * B[32:48][0:128])
// wgmma.m64n128k16(A[0:64][48:64] * B[48:64][0:128])
// wgmma.m64n128k16(A[64:128][0:16] * B[0:16][0:128])
// wgmma.m64n128k16(A[64:128][16:32] * B[16:32][0:128])
// wgmma.m64n128k16(A[64:128][32:48] * B[32:48][0:128])
// wgmma.m64n128k16(A[64:128][48:64] * B[48:64][0:128])
```
**Epilogue**
```
//reg->shmem
warpgroup.mma.store matrixD, shmem
//shmem->glbmem
parallel-for(i=0;...128)
parallel-for(j=0;...128)
store shmem, globalmem
```
---
Full diff: https://github.com/llvm/llvm-project/pull/69913.diff
1 Files Affected:
- (added) mlir/test/Integration/GPU/CUDA/sm90/gemm_f32_f16_f16_128x128x128.mlir (+277)
``````````diff
diff --git a/mlir/test/Integration/GPU/CUDA/sm90/gemm_f32_f16_f16_128x128x128.mlir b/mlir/test/Integration/GPU/CUDA/sm90/gemm_f32_f16_f16_128x128x128.mlir
new file mode 100644
index 000000000000000..fbf804ca559c78c
--- /dev/null
+++ b/mlir/test/Integration/GPU/CUDA/sm90/gemm_f32_f16_f16_128x128x128.mlir
@@ -0,0 +1,277 @@
+// RUN: mlir-opt %s \
+// RUN: -test-lower-to-nvvm="cubin-chip=sm_90a cubin-features=+ptx80 opt-level=3" \
+// RUN: | mlir-cpu-runner \
+// RUN: --shared-libs=%mlir_cuda_runtime \
+// RUN: --shared-libs=%mlir_runner_utils \
+// RUN: --entry-point-result=void \
+// RUN: | FileCheck %s
+
+// CHECK: Correct Results : 16384
+// CHECK: Incorrect Results : 0
+
+// This program performs 128x128x128 GEMM (F32 += F16 * F16)
+//
+// ## Sequential
+// for(128)
+// for(128)
+// for(128)
+// D += A * B
+//
+// ## Parallel 1 CTA with 1 Warpgroup with 2 pipelining stage
+//
+// cuda kernel() {
+// mbarriers.init[2]
+// for(i = 0;...2) {
+// tma.load shmem_buffer<i x...>
+// mbarrier.expect_tx group[i]
+// }
+// result =
+// for(i = 0;...2) {
+// pipe = i % 2
+// mbarrier.wait [pipe]
+// lhs = shmem_buffer_lhs<pipe x 128 x 64>
+// rhs = shmem_buffer_rhs<pipe x 64 x 128>
+// yield nvgpu.warpgroup.mma (lhs, rhs)
+// ---------------------------------------------------------------------
+// Expanded : nvgpu.warpgroup.mma [128][128]+=[128][64]*[64][128]
+// wgmma.m64n128k16(A[0:64][0:16] * B[0:16][0:128])
+// wgmma.m64n128k16(A[0:64][16:32] * B[16:32][0:128])
+// wgmma.m64n128k16(A[0:64][32:48] * B[32:48][0:128])
+// wgmma.m64n128k16(A[0:64][48:64] * B[48:64][0:128])
+// wgmma.m64n128k16(A[64:128][0:16] * B[0:16][0:128])
+// wgmma.m64n128k16(A[64:128][16:32] * B[16:32][0:128])
+// wgmma.m64n128k16(A[64:128][32:48] * B[32:48][0:128])
+// wgmma.m64n128k16(A[64:128][48:64] * B[48:64][0:128])
+// ---------------------------------------------------------------------
+// }
+// nvgpu.store result -> shmem_buffer_result
+
+
+!barrierType = !nvgpu.mbarrier.group<memorySpace = #gpu.address_space<workgroup>, num_barriers = 2>
+!lhsTensorMap = !nvgpu.tensormap.descriptor<tensor = memref<128x64xf16, 3>, swizzle = swizzle_128b, l2promo=none, oob=zero, interleave=none>
+!rhsTensorMap = !nvgpu.tensormap.descriptor<tensor = memref<64x128xf16, 3>, swizzle = swizzle_128b, l2promo=none, oob=zero, interleave=none>
+
+func.func private @printMemrefF32(memref<*xf32>)
+llvm.func @printf(!llvm.ptr<i8>, ...) -> i32
+
+memref.global "private" @dynamicShmem : memref<0xf16, 3> {alignment = 16 : i64}
+memref.global "private" @accShmem : memref<0xf32, 3> {alignment = 16 : i64}
+
+func.func @main() {
+ %c214016_i32 = arith.constant 214016 : i32
+ %hc1 = arith.constant 1 : index
+ %hc4096 = arith.constant 4096 : index
+ %hc0 = arith.constant 0 : index
+ %hc64 = arith.constant 64 : index
+ %hc16 = arith.constant 16 : index
+ %hc8 = arith.constant 8 : index
+ %hc128 = arith.constant 128 : index
+ %hc32 = arith.constant 32 : index
+ %hc256 = arith.constant 256 : index
+ %f0 = arith.constant 0.0 : f32
+
+ // Step 1. Allocate and Initilize LHS and RHS Matrices
+ %matrixAHost = memref.alloc() : memref<128x128xf16>
+ %matrixBHost = memref.alloc() : memref<128x128xf16>
+ %matrixDHost = memref.alloc() : memref<128x128xf32>
+ %matrixRefHost = memref.alloc() : memref<128x128xf32>
+ scf.for %i = %hc0 to %hc128 step %hc1 {
+ scf.for %j = %hc0 to %hc128 step %hc1 {
+ %v0 = arith.muli %i, %hc128 : index // i * 128
+ %v00 = arith.addi %v0, %j : index // i * 128 + j
+ %v01 = arith.divui %v00, %hc8 : index // (i * 128 + j) / 8
+ %v02 = arith.remui %v01, %hc16 : index // <<<<< mod 128
+ %v2 = arith.index_cast %v02 : index to i32
+ %vR = arith.sitofp %v2 : i32 to f16
+ memref.store %vR, %matrixBHost[%i, %j] : memref<128x128xf16>
+ %b0 = arith.muli %j, %hc64 : index
+ %b00 = arith.addi %b0, %i : index
+ %b01 = arith.divui %b00, %hc8 : index
+ %b02 = arith.remui %b01, %hc16 : index // <<<<< mod 128
+ %v1 = arith.index_cast %b02 : index to i32
+ %vL = arith.sitofp %v1 : i32 to f16
+ memref.store %vL, %matrixAHost[%j, %i] : memref<128x128xf16>
+ memref.store %f0, %matrixDHost[%i, %j] : memref<128x128xf32>
+ memref.store %f0, %matrixRefHost[%i, %j] : memref<128x128xf32>
+ }
+ }
+
+ // Step 2. Allocate Device Memory for LHS and RHS Matrices and Copy H2D
+ %token = gpu.wait async
+ %matrixA:2 = gpu.alloc async [%token] () : memref<128x128xf16>
+ %matrixB:2 = gpu.alloc async [%token] () : memref<128x128xf16>
+ %matrixD:2 = gpu.alloc async [%token] () : memref<128x128xf32>
+ %1 = gpu.memcpy async [%token] %matrixA, %matrixAHost : memref<128x128xf16>, memref<128x128xf16>
+ %2 = gpu.memcpy async [%token] %matrixB, %matrixBHost : memref<128x128xf16>, memref<128x128xf16>
+ %castA = memref.cast %matrixA : memref<128x128xf16> to memref<*xf16>
+ %castB = memref.cast %matrixB : memref<128x128xf16> to memref<*xf16>
+
+ // Step 3. Create TMA Descriptor
+ %descA = nvgpu.tma.create.descriptor %castA box[%hc128, %hc64] : memref<*xf16> -> !lhsTensorMap
+ %descB = nvgpu.tma.create.descriptor %castB box[%hc64, %hc64] : memref<*xf16> -> !rhsTensorMap
+
+ // Step 4. Launch GPU Kernel
+ gpu.launch blocks(%arg0, %arg1, %arg2) in (%arg6 = %hc1, %arg7 = %hc1, %arg8 = %hc1)
+ threads(%arg3, %arg4, %arg5) in (%arg9 = %hc128, %arg10 = %hc1, %arg11 = %hc1)
+ dynamic_shared_memory_size %c214016_i32
+ {
+ memref.assume_alignment %matrixD, 16 : memref<128x128xf32>
+
+ %c256 = arith.constant 256 : index
+ %c10000000 = arith.constant 10000000 : index
+ %c32768 = arith.constant 32768 : index
+ %c320 = arith.constant 320 : index
+ %c192 = arith.constant 192 : index
+ %c6 = arith.constant 6 : index
+ %c5 = arith.constant 5 : index
+ %c4 = arith.constant 4 : index
+ %c3 = arith.constant 3 : index
+ %c7 = arith.constant 7 : index
+ %c64 = arith.constant 64 : index
+ %c1 = arith.constant 1 : index
+ %c2 = arith.constant 2 : index
+ %c0 = arith.constant 0 : index
+ %c128 = arith.constant 128 : index
+ %c32 = arith.constant 32 : index
+ %c16 = arith.constant 16 : index
+ %c4096 = arith.constant 4096 : index
+ %c8 = arith.constant 8 : index
+ %txcount = arith.constant 32768 : index
+
+ %tidx = gpu.thread_id x
+ %dynamicMem = memref.get_global @dynamicShmem : memref<0xf16, 3>
+ %lhsShmem = memref.reinterpret_cast %dynamicMem to offset: [0], sizes: [7, 128, 64], strides: [8192, 64, 1] : memref<0xf16, 3> to memref<7x128x64xf16, 3>
+ %rhsShmem2 = memref.reinterpret_cast %dynamicMem to offset: [0], sizes: [14, 64, 128], strides: [8192,128,1] : memref<0xf16, 3> to memref<14x64x128xf16,3>
+ %rhsShmem = memref.subview %rhsShmem2[7, 0, 0][7, 64, 128][1, 1, 1] : memref<14x64x128xf16,3> to memref<7x64x128xf16, strided<[8192, 128, 1], offset: 57344>, 3>
+
+ // Step 1. [GPU] Create Async Transactional Barriers (mbarriers)
+ %barrier = nvgpu.mbarrier.create -> !barrierType
+ %cnd = arith.cmpi eq, %tidx, %c0 : index
+
+ // Step 2. [GPU] Initialize mbarriers
+ nvgpu.mbarrier.init %barrier[%c0], %c1 : !barrierType
+ nvgpu.mbarrier.init %barrier[%c1], %c1 : !barrierType
+
+ // Step 3. [GPU] Prefetch TMA Descriptors to L1 Cache
+ nvgpu.tma.prefetch.descriptor %descA : !lhsTensorMap
+ nvgpu.tma.prefetch.descriptor %descB : !rhsTensorMap
+
+ // Step 4. [GPU] TMA Load Pipeline 1
+ scf.if %cnd {
+ %pipe = arith.constant 0 : index
+ %lhsSlice = memref.subview %lhsShmem [0, 0, 0][1, 64, 128][1, 1, 1] : memref<7x128x64xf16,3> to memref<1x64x128xf16, strided<[8192, 64, 1]>, 3>
+ %rhsSlice = memref.subview %rhsShmem [0, 0, 0][1, 128, 64][1, 1, 1] : memref<7x64x128xf16, strided<[8192, 128, 1], offset: 57344>, 3> to memref<1x128x64xf16, strided<[8192, 128, 1], offset: 57344>, 3>
+ %rhsSlice2 = memref.subview %rhsSlice[0, 32, 0][1, 128, 64][1,1,1] : memref<1x128x64xf16, strided<[8192, 128, 1], offset: 57344>, 3> to memref<1x128x64xf16, strided<[8192, 128, 1], offset: 61440>, 3>
+ nvgpu.mbarrier.arrive.expect_tx %barrier[%pipe], %txcount : !barrierType
+ %dim = arith.muli %pipe, %c64 : index
+ nvgpu.tma.async.load %descA[%dim, %c0], %barrier[%pipe] to %lhsSlice : !lhsTensorMap, !barrierType -> memref<1x64x128xf16, strided<[8192, 64, 1]>, 3>
+ nvgpu.tma.async.load %descB[%c0, %dim], %barrier[%pipe] to %rhsSlice : !rhsTensorMap, !barrierType -> memref<1x128x64xf16, strided<[8192, 128, 1], offset: 57344>, 3>
+ nvgpu.tma.async.load %descB[%c64, %dim], %barrier[%pipe] to %rhsSlice2 : !rhsTensorMap, !barrierType -> memref<1x128x64xf16, strided<[8192, 128, 1], offset: 61440>, 3>
+ }
+ // Step 4. [GPU] TMA Load Pipeline 2
+ scf.if %cnd {
+ %pipe = arith.constant 1 : index
+ %lhsSlice = memref.subview %lhsShmem [1, 0, 0][1, 64, 128][1, 1, 1] : memref<7x128x64xf16,3> to memref<1x64x128xf16, strided<[8192, 64, 1], offset: 8192>, 3>
+ %rhsSlice = memref.subview %rhsShmem [1, 0, 0][1, 128, 64][1, 1, 1] : memref<7x64x128xf16, strided<[8192, 128, 1], offset: 57344>, 3> to memref<1x128x64xf16, strided<[8192, 128, 1], offset: 65536>, 3>
+ %rhsSlice2 = memref.subview %rhsSlice[0, 32, 0][1, 128, 64][1,1,1] : memref<1x128x64xf16, strided<[8192, 128, 1], offset: 65536>, 3> to memref<1x128x64xf16, strided<[8192, 128, 1], offset: 69632>, 3>
+ nvgpu.mbarrier.arrive.expect_tx %barrier[%pipe], %txcount : !barrierType
+ %dim = arith.muli %pipe, %c64 : index
+ nvgpu.tma.async.load %descA[%dim, %c0], %barrier[%pipe] to %lhsSlice : !lhsTensorMap, !barrierType -> memref<1x64x128xf16, strided<[8192, 64, 1], offset: 8192>, 3>
+ nvgpu.tma.async.load %descB[%c0, %dim], %barrier[%pipe] to %rhsSlice : !rhsTensorMap, !barrierType -> memref<1x128x64xf16, strided<[8192, 128, 1], offset: 65536>, 3>
+ nvgpu.tma.async.load %descB[%c64, %dim], %barrier[%pipe] to %rhsSlice2 : !rhsTensorMap, !barrierType -> memref<1x128x64xf16, strided<[8192, 128, 1], offset: 69632>, 3>
+ }
+
+ // Step 5. [GPU] Initiliaze accumulator matrix
+ %14 = nvgpu.warpgroup.mma.init.accumulator -> <fragmented = vector<128x128xf32>>
+
+ // Step 6. [GPU] Main Loop Starts
+ %15 = scf.for %i = %c0 to %c2 step %c1 iter_args(%mc = %14)
+ -> (!nvgpu.warpgroup.accumulator<fragmented = vector<128x128xf32>>)
+ {
+ %ticks = arith.constant 10000000 : index
+ // TMA wait
+ nvgpu.mbarrier.try_wait.parity %barrier[%i], %c0, %ticks : !barrierType
+ %lhsSlice = memref.subview %lhsShmem [%i, 0, 0][1, 64, 128][1, 1, 1] : memref<7x128x64xf16,3> to memref<1x64x128xf16, strided<[8192, 64, 1], offset: ?>, 3>
+ %rhsSlice = memref.subview %rhsShmem [%i, 0, 0][1, 128, 64][1, 1, 1] : memref<7x64x128xf16, strided<[8192, 128, 1], offset: 57344>, 3> to memref<1x128x64xf16, strided<[8192, 128, 1], offset: ?>, 3>
+ // Descriptor WGMMA
+ %dA = nvgpu.warpgroup.generate.descriptor %lhsSlice, %descA : memref<1x64x128xf16, strided<[8192, 64, 1], offset: ?>, 3>, !lhsTensorMap -> !nvgpu.warpgroup.descriptor<tensor=memref<128x64xf16, 3>>
+ %dB = nvgpu.warpgroup.generate.descriptor %rhsSlice, %descB : memref<1x128x64xf16, strided<[8192, 128, 1], offset: ?>, 3>, !rhsTensorMap -> !nvgpu.warpgroup.descriptor<tensor=memref<64x128xf16, 3>>
+ // Perform WGMMA 128x128x64
+ %md = nvgpu.warpgroup.mma %dA, %dB, %mc {transposeB} : <tensor = memref<128x64xf16,3>>, <tensor = memref<64x128xf16,3>>, <fragmented = vector<128x128xf32>> -> <fragmented = vector<128x128xf32>>
+ scf.yield %md : !nvgpu.warpgroup.accumulator<fragmented = vector<128x128xf32>>
+ }
+
+ // Step 10. Wait all to finish mma
+ nvvm.wgmma.wait.group.sync.aligned 0
+
+ // Step 11. [GPU] Epilogue, store fragmented register to shared memory
+ %accShmem = memref.get_global @accShmem : memref<0xf32, 3>
+ %accShmemPtr = memref.reinterpret_cast %accShmem to offset: [0], sizes: [128, 128], strides: [128, 1] : memref<0xf32, 3> to memref<128x128xf32, 3>
+ nvgpu.warpgroup.mma.store %15, %accShmemPtr : <fragmented = vector<128x128xf32>> to memref<128x128xf32, 3>
+
+ %17 = arith.divui %tidx, %c32 : index
+ %18 = arith.remui %tidx, %c32 : index
+ scf.for %arg12 = %17 to %c128 step %c4 {
+ %19 = arith.muli %18, %c4 : index
+ %20 = vector.load %accShmemPtr[%arg12, %19] : memref<128x128xf32, 3>, vector<4xf32>
+ vector.store %20, %matrixD[%arg12, %19] : memref<128x128xf32>, vector<4xf32>
+ }
+ gpu.terminator
+ }
+
+ // Step 13. Copy D2H
+ %5 = gpu.memcpy async [%token] %matrixDHost, %matrixD : memref<128x128xf32>, memref<128x128xf32>
+ gpu.wait [%token]
+
+ // Step 14. Compute on host
+ linalg.matmul ins(%matrixAHost, %matrixBHost : memref<128x128xf16>, memref<128x128xf16>) outs(%matrixRefHost : memref<128x128xf32>)
+
+ // Step 15. Verify
+ %ic1 = arith.constant 1 : i32
+ %ic0 = arith.constant 0 : i32
+ %tolerance = arith.constant 0.00000001 : f32
+ %errorCount, %correctCount =
+ scf.for %i = %hc0 to %hc128 step %hc1 iter_args(%ec1 = %ic0, %cc1 = %ic0) -> (i32,i32) {
+ %ec2, %cc2 =
+ scf.for %j = %hc0 to %hc128 step %hc1 iter_args(%ec2 = %ec1, %cc2 = %cc1) -> (i32,i32){
+ %v1 = memref.load %matrixRefHost[%i,%j] : memref<128x128xf32>
+ %v2 = memref.load %matrixDHost[%i,%j] : memref<128x128xf32>
+ %g1 = arith.subf %v1,%v2 : f32
+ %g2 = math.absf %g1: f32
+ %g3 = arith.cmpf ult, %tolerance, %g2 : f32
+ %ec3, %cc3 = scf.if %g3 -> (i32, i32) {
+ %coor = arith.constant dense<-1> : vector<2xi32>
+ %i32 = arith.index_cast %i : index to i32
+ %j32 = arith.index_cast %j : index to i32
+ %coord1 = vector.insert %i32, %coor[0] : i32 into vector<2xi32>
+ %coord2 = vector.insert %j32, %coord1[1] : i32 into vector<2xi32>
+ // vector.print %coord2 : vector<2xi32>
+ %ec3 = arith.addi %ec2, %ic1 : i32
+ scf.yield %ec3, %cc2 : i32, i32
+ } else {
+ %cc3 = arith.addi %cc2, %ic1 : i32
+ scf.yield %ec2, %cc3 : i32, i32
+ }
+ scf.yield %ec3, %cc3 : i32,i32
+ }
+ scf.yield %ec2,%cc2 : i32,i32
+ }
+
+ %s0 = llvm.mlir.addressof @str_correct : !llvm.ptr<array<18 x i8>>
+ %s1 = llvm.mlir.constant(0 : index) : i64
+ %s2 = llvm.getelementptr %s0[%s1, %s1]
+ : (!llvm.ptr<array<18 x i8>>, i64, i64) -> !llvm.ptr<i8>
+ func.call @printCString(%s2) : (!llvm.ptr<i8>) -> ()
+ vector.print %correctCount : i32
+ %s3 = llvm.mlir.addressof @str_incorrect : !llvm.ptr<array<20 x i8>>
+ %s4 = llvm.getelementptr %s3[%s1, %s1]
+ : (!llvm.ptr<array<20 x i8>>, i64, i64) -> !llvm.ptr<i8>
+ func.call @printCString(%s4) : (!llvm.ptr<i8>) -> ()
+ vector.print %errorCount : i32
+
+ return
+}
+llvm.mlir.global internal constant @str_correct("Correct Results : ") {addr_space = 0 : i32}
+llvm.mlir.global internal constant @str_incorrect("Incorrect Results : ") {addr_space = 0 : i32}
+func.func private @printCString(!llvm.ptr<i8>)
+
``````````
</details>
https://github.com/llvm/llvm-project/pull/69913
More information about the Mlir-commits
mailing list