This is an archive of the discontinued LLVM Phabricator instance.

Differential D157666

[mlir][sparse][gpu] refine sparse gpu round-trip and lowering test
ClosedPublic

Authored by aartbik on Aug 10 2023, 5:08 PM.

Details

Reviewers
Peiming
ThomasRaoux
K-Wu
nicolasvasilache
herhut
Commits
rG6c4cd7a13e1c: [mlir][sparse][gpu] refine sparse gpu round-trip and lowering test
Summary

Tests had become inconsistent, and contained a few slip ups
(e.g. non-async versions did not lower)

Diff Detail

Event Timeline

aartbik created this revision.Aug 10 2023, 5:08 PM
Herald added a reviewer: ThomasRaoux. · View Herald TranscriptAug 10 2023, 5:08 PM
Herald added a reviewer: K-Wu. · View Herald Transcript
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: K-Wu, bviyer, Moerafaat and 23 others. · View Herald Transcript
aartbik requested review of this revision.Aug 10 2023, 5:08 PM
Herald added a reviewer: nicolasvasilache. · View Herald TranscriptAug 10 2023, 5:08 PM
Herald added a reviewer: herhut. · View Herald Transcript
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: stephenneuendorffer, nicolasvasilache. · View Herald Transcript
K-Wu accepted this revision.Aug 10 2023, 5:12 PM

LGTM. Approval

This revision is now accepted and ready to land.Aug 10 2023, 5:12 PM
aartbik updated this revision to Diff 549207.Aug 10 2023, 5:12 PM

rebased with main

This revision was landed with ongoing or failed builds.Aug 10 2023, 5:19 PM
Closed by commit rG6c4cd7a13e1c: [mlir][sparse][gpu] refine sparse gpu round-trip and lowering test (authored by aartbik). · Explain Why
This revision was automatically updated to reflect the committed changes.
aartbik added a commit: rG6c4cd7a13e1c: [mlir][sparse][gpu] refine sparse gpu round-trip and lowering test.
Harbormaster completed remote builds in B251813: Diff 549207.Aug 10 2023, 10:15 PM

Revision Contents

PathSize
mlir/
test/
Conversion/
GPUCommon/
113 lines
Dialect/
GPU/
44 lines

Diff 549208

mlir/test/Conversion/GPUCommon/lower-sparse-to-gpu-runtime-calls.mlir

Show First 20 Lines • Show All 48 Lines • ▼ Show 20 Lines func.func @matmul(%arg0: index) {
%bufferSz, %token6 = gpu.spmm_buffer_size async [%token5] %spmat, %dnmat, %dnmat : index into f64 %bufferSz, %token6 = gpu.spmm_buffer_size async [%token5] %spmat, %dnmat, %dnmat : index into f64
%token7 = gpu.spmm async [%token6] %spmat, %dnmat, %dnmat, %mem2 : memref<?xf64> into f64 %token7 = gpu.spmm async [%token6] %spmat, %dnmat, %dnmat, %mem2 : memref<?xf64> into f64
%token8 = gpu.destroy_sp_mat async [%token7] %spmat %token8 = gpu.destroy_sp_mat async [%token7] %spmat
%token9 = gpu.destroy_dn_tensor async [%token8] %dnmat %token9 = gpu.destroy_dn_tensor async [%token8] %dnmat
gpu.wait [%token9] gpu.wait [%token9]
return return
} }
// CHECK-LABEL: func @sddmm
// CHECK: llvm.call @mgpuStreamCreate
// CHECK: llvm.call @mgpuMemAlloc
// CHECK: llvm.call @mgpuMemAlloc
// CHECK: llvm.call @mgpuCreateCsr
// CHECK: llvm.call @mgpuCreateDnMat
// CHECK: llvm.call @mgpuSDDMMBufferSize
// CHECK: llvm.call @mgpuSDDMM
// CHECK: llvm.call @mgpuDestroySpMat
// CHECK: llvm.call @mgpuDestroyDnMat
// CHECK: llvm.call @mgpuStreamSynchronize
// CHECK: llvm.call @mgpuStreamDestroy
func.func @sddmm(%arg0: index) {
%token0 = gpu.wait async
%mem1, %token1 = gpu.alloc async [%token0] (%arg0) : memref<?xindex>
%mem2, %token2 = gpu.alloc async [%token1] (%arg0) : memref<?xf64>
%spmat, %token4 = gpu.create_csr async [%token2] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf64>
%dnmat, %token5 = gpu.create_dn_tensor async [%token4] %mem2, %arg0, %arg0 : index, index into memref<?xf64>
%bufferSz, %token6 = gpu.sddmm_buffer_size async [%token5] %dnmat, %dnmat, %spmat into f64
%token7 = gpu.sddmm async [%token6] %dnmat, %dnmat, %spmat, %mem2 : memref<?xf64> into f64
%token8 = gpu.destroy_sp_mat async [%token7] %spmat
%token9 = gpu.destroy_dn_tensor async [%token8] %dnmat
gpu.wait [%token9]
return
}
// CHECK-LABEL: func @spgemm // CHECK-LABEL: func @spgemm
// CHECK: llvm.call @mgpuStreamCreate // CHECK: llvm.call @mgpuStreamCreate
// CHECK: llvm.call @mgpuMemAlloc // CHECK: llvm.call @mgpuMemAlloc
// CHECK: llvm.call @mgpuMemAlloc // CHECK: llvm.call @mgpuMemAlloc
// CHECK: llvm.call @mgpuCreateCsr // CHECK: llvm.call @mgpuCreateCsr
// CHECK: llvm.call @mgpuCreateCsr // CHECK: llvm.call @mgpuCreateCsr
// CHECK: llvm.call @mgpuCreateCsr // CHECK: llvm.call @mgpuCreateCsr
// CHECK: llvm.call @mgpuSpGEMMCreateDescr // CHECK: llvm.call @mgpuSpGEMMCreateDescr
// CHECK: llvm.call @malloc
// CHECK: llvm.call @mgpuSpGEMMWorkEstimation // CHECK: llvm.call @mgpuSpGEMMWorkEstimation
// CHECK: llvm.call @mgpuMemAlloc
// CHECK: llvm.call @mgpuSpGEMMWorkEstimation
// CHECK: llvm.call @mgpuMemAlloc
// CHECK: llvm.call @mgpuSpGEMMCompute // CHECK: llvm.call @mgpuSpGEMMCompute
// CHECK: llvm.call @mgpuMemAlloc // CHECK: llvm.call @mgpuSpGEMMGetSize
// CHECK: llvm.call @mgpuMemAlloc // CHECK: llvm.call @mgpuSetCsrPointers
// CHECK: llvm.call @mgpuStreamSynchronize
// CHECK: llvm.call @mgpuStreamDestroy
// CHECK: llvm.call @mgpuStreamCreate
// CHECK: llvm.call @mgpuSpGEMMCopy // CHECK: llvm.call @mgpuSpGEMMCopy
// CHECK: llvm.call @mgpuSpGEMMDestroyDescr
// CHECK: llvm.call @mgpuDestroySpMat // CHECK: llvm.call @mgpuDestroySpMat
// CHECK: llvm.call @mgpuDestroySpMat // CHECK: llvm.call @mgpuDestroySpMat
// CHECK: llvm.call @mgpuDestroySpMat // CHECK: llvm.call @mgpuDestroySpMat
// CHECK: llvm.call @mgpuStreamSynchronize // CHECK: llvm.call @mgpuStreamSynchronize
// CHECK: llvm.call @mgpuStreamDestroy // CHECK: llvm.call @mgpuStreamDestroy
func.func @spgemm(%arg0: index) { func.func @spgemm(%arg0: index) {
%token0 = gpu.wait async %token0 = gpu.wait async
%mem1, %token1 = gpu.alloc async [%token0] (%arg0) : memref<?xindex> %mem1, %token1 = gpu.alloc async [%token0] (%arg0) : memref<?xindex>
%mem2, %token2 = gpu.alloc async [%token1] (%arg0) : memref<?xf64> %mem2, %token2 = gpu.alloc async [%token1] (%arg0) : memref<?xf32>
%spmatA, %token3 = gpu.create_csr async [%token2] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf64> %spmatA, %token3 = gpu.create_csr async [%token2] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32>
%spmatB, %token4 = gpu.create_csr async [%token3] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf64> %spmatB, %token4 = gpu.create_csr async [%token3] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32>
%spmatC, %token5 = gpu.create_csr async [%token4] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf64> %spmatC, %token5 = gpu.create_csr async [%token4] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32>
%spgemmDesc, %token6 = gpu.spgemm_create_descr async [%token5] %spgemmDesc, %token6 = gpu.spgemm_create_descr async [%token5]
// Used as nullptr %alloc = memref.alloc() : memref<0xi8> // nullptr
%alloc = memref.alloc() : memref<0xi8>
%c0 = arith.constant 0 : index %c0 = arith.constant 0 : index
%bufferSz1, %token7 = gpu.spgemm_work_estimation_or_compute async %bufferSz1, %token7 = gpu.spgemm_work_estimation_or_compute async
[%token6]{WORK_ESTIMATION} [%token6]{WORK_ESTIMATION}
%spmatA{NON_TRANSPOSE}, %spmatB{NON_TRANSPOSE}, %spmatA, %spmatB, %spmatC,
%spmatC, %spgemmDesc, %c0, %spgemmDesc, %c0, %alloc: f32 into memref<0xi8>
%alloc: f32 into memref<0xi8> %bufferSz2, %token8 = gpu.spgemm_work_estimation_or_compute async
%buf1, %token8 = gpu.alloc async [%token7] (%bufferSz1) : memref<?xi8> [%token7]{COMPUTE}
%bufferSz1_1, %token9 = gpu.spgemm_work_estimation_or_compute async %spmatA, %spmatB, %spmatC,
[%token8]{WORK_ESTIMATION} %spmatA, %spmatB, %spgemmDesc, %c0, %alloc: f32 into memref<0xi8>
%spmatC, %spgemmDesc, %bufferSz1, %rows, %cols, %nnz, %token9 = gpu.spgemm_get_size async [%token8] %spmatC
%buf1: f32 into memref<?xi8> %token10 = gpu.set_csr_pointers async [%token8] %spmatC, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32>
%buf2, %token13 = gpu.alloc async [%token9] (%bufferSz1_1) : memref<?xi8> %token11 = gpu.spgemm_copy async [%token10] %spmatA, %spmatB, %spmatC, %spgemmDesc: f32
%bufferSz2_2, %token14 = gpu.spgemm_work_estimation_or_compute async %token12 = gpu.spgemm_destroy_descr async [%token11] %spgemmDesc
[%token13]{COMPUTE} %spmatA, %spmatB, %spmatC, %token13 = gpu.destroy_sp_mat async [%token12] %spmatA
%spgemmDesc, %bufferSz1_1, %token14 = gpu.destroy_sp_mat async [%token13] %spmatB
%buf2: f32 into memref<?xi8> %token15 = gpu.destroy_sp_mat async [%token14] %spmatC
%rows, %cols, %nnz, %token15 = gpu.spgemm_get_size async [%token14] %spmatC gpu.wait [%token15]
%mem_columns, %token16 = gpu.alloc async [%token15] (%cols) : memref<?xi32> return
%mem_values, %token17 = gpu.alloc async [%token16] (%nnz) : memref<?xf32> }
gpu.wait [%token17]
%token18 = gpu.wait async // CHECK-LABEL: func @sddmm
%token19 = gpu.spgemm_copy async [%token18] %spmatA, %spmatB, %spmatC, %spgemmDesc: f32 // CHECK: llvm.call @mgpuStreamCreate
%token20 = gpu.destroy_sp_mat async [%token19] %spmatA // CHECK: llvm.call @mgpuMemAlloc
%token21 = gpu.destroy_sp_mat async [%token20] %spmatB // CHECK: llvm.call @mgpuMemAlloc
%token22 = gpu.destroy_sp_mat async [%token21] %spmatC // CHECK: llvm.call @mgpuCreateCsr
gpu.wait [%token22] // CHECK: llvm.call @mgpuCreateDnMat
// CHECK: llvm.call @mgpuSDDMMBufferSize
// CHECK: llvm.call @mgpuSDDMM
// CHECK: llvm.call @mgpuDestroySpMat
// CHECK: llvm.call @mgpuDestroyDnMat
// CHECK: llvm.call @mgpuStreamSynchronize
// CHECK: llvm.call @mgpuStreamDestroy
func.func @sddmm(%arg0: index) {
%token0 = gpu.wait async
%mem1, %token1 = gpu.alloc async [%token0] (%arg0) : memref<?xindex>
%mem2, %token2 = gpu.alloc async [%token1] (%arg0) : memref<?xf64>
%spmat, %token4 = gpu.create_csr async [%token2] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf64>
%dnmat, %token5 = gpu.create_dn_tensor async [%token4] %mem2, %arg0, %arg0 : index, index into memref<?xf64>
%bufferSz, %token6 = gpu.sddmm_buffer_size async [%token5] %dnmat, %dnmat, %spmat into f64
%token7 = gpu.sddmm async [%token6] %dnmat, %dnmat, %spmat, %mem2 : memref<?xf64> into f64
%token8 = gpu.destroy_sp_mat async [%token7] %spmat
%token9 = gpu.destroy_dn_tensor async [%token8] %dnmat
gpu.wait [%token9]
return return
} }
} }

mlir/test/Dialect/GPU/sparse-roundtrip.mlir

Show First 20 Lines • Show All 48 Lines • ▼ Show 20 Lines func.func @matmul(%arg0: index) {
%bufferSz, %token6 = gpu.spmm_buffer_size async [%token5] %spmat, %dnmat, %dnmat : index into f64 %bufferSz, %token6 = gpu.spmm_buffer_size async [%token5] %spmat, %dnmat, %dnmat : index into f64
%token7 = gpu.spmm async [%token6] %spmat, %dnmat, %dnmat, %mem2 : memref<?xf64> into f64 %token7 = gpu.spmm async [%token6] %spmat, %dnmat, %dnmat, %mem2 : memref<?xf64> into f64
%token8 = gpu.destroy_sp_mat async [%token7] %spmat %token8 = gpu.destroy_sp_mat async [%token7] %spmat
%token9 = gpu.destroy_dn_tensor async [%token8] %dnmat %token9 = gpu.destroy_dn_tensor async [%token8] %dnmat
gpu.wait [%token9] gpu.wait [%token9]
return return
} }
// CHECK-LABEL: func @spgemm // CHECK-LABEL: func @spgemm
// CHECK: %{{.*}} = gpu.wait async // CHECK: %{{.*}} = gpu.wait async
// CHECK: %{{.*}}, %{{.*}} = gpu.alloc async [%{{.*}}] (%{{.*}}) : memref<?xindex> // CHECK: %{{.*}}, %{{.*}} = gpu.alloc async [%{{.*}}] (%{{.*}}) : memref<?xindex>
// CHECK: %{{.*}}, %{{.*}} = gpu.alloc async [%{{.*}}] (%{{.*}}) : memref<?xf32> // CHECK: %{{.*}}, %{{.*}} = gpu.alloc async [%{{.*}}] (%{{.*}}) : memref<?xf32>
// CHECK: %{{.*}}, %{{.*}} = gpu.create_csr async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf32> // CHECK: %{{.*}}, %{{.*}} = gpu.create_csr async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf32>
// CHECK: %{{.*}}, %{{.*}} = gpu.create_csr async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf32> // CHECK: %{{.*}}, %{{.*}} = gpu.create_csr async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf32>
// CHECK: %{{.*}}, %{{.*}} = gpu.create_csr async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf32> // CHECK: %{{.*}}, %{{.*}} = gpu.create_csr async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf32>
// CHECK: %{{.*}}, %{{.*}} = gpu.spgemm_create_descr async [%{{.*}}] // CHECK: %{{.*}}, %{{.*}} = gpu.spgemm_create_descr async [%{{.*}}]
// CHECK: %{{.*}}, %{{.*}} = gpu.spgemm_work_estimation_or_compute async [%{{.*}}]{ WORK_ESTIMATION} %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : f32 into memref<0xi8> // CHECK: %{{.*}}, %{{.*}} = gpu.spgemm_work_estimation_or_compute async [%{{.*}}]{ WORK_ESTIMATION} %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : f32 into memref<0xi8>
// CHECK: %{{.*}}, %{{.*}} = gpu.spgemm_work_estimation_or_compute async [%{{.*}}]{ COMPUTE} %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : f32 into memref<0xi8> // CHECK: %{{.*}}, %{{.*}} = gpu.spgemm_work_estimation_or_compute async [%{{.*}}]{ COMPUTE} %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : f32 into memref<0xi8>
// CHECK: %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} = gpu.spgemm_get_size async [%{{.*}}] %{{.*}} // CHECK: %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} = gpu.spgemm_get_size async [%{{.*}}] %{{.*}}
// CHECK %{{.*}} = gpu.set_csr_pointers async [%{{.*}}] %{{.*}}, {{.*}}, {{.*}}, {{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf32> // CHECK: %{{.*}} = gpu.set_csr_pointers async [%{{.*}}] %{{.*}}, {{.*}}, {{.*}}, {{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf32>
// CHECK: %{{.*}} = gpu.spgemm_copy async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : f32 // CHECK: %{{.*}} = gpu.spgemm_copy async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : f32
// CHECK: %{{.*}} = gpu.spgemm_destroy_descr async [%{{.*}}] %{{.*}} // CHECK: %{{.*}} = gpu.spgemm_destroy_descr async [%{{.*}}] %{{.*}}
// CHECK: gpu.destroy_sp_mat %{{.*}} // CHECK: %{{.*}} = gpu.destroy_sp_mat async [%{{.*}}] %{{.*}}
// CHECK: gpu.destroy_sp_mat %{{.*}} // CHECK: %{{.*}} = gpu.destroy_sp_mat async [%{{.*}}] %{{.*}}
// CHECK: gpu.destroy_sp_mat %{{.*}} // CHECK: %{{.*}} = gpu.destroy_sp_mat async [%{{.*}}] %{{.*}}
// CHECK: gpu.wait [%{{.*}}]
// CHECK: return // CHECK: return
func.func @spgemm(%arg0: index) { func.func @spgemm(%arg0: index) {
%token0 = gpu.wait async %token0 = gpu.wait async
%mem1, %token1 = gpu.alloc async [%token0] (%arg0) : memref<?xindex> %mem1, %token1 = gpu.alloc async [%token0] (%arg0) : memref<?xindex>
%mem2, %token2 = gpu.alloc async [%token1] (%arg0) : memref<?xf32> %mem2, %token2 = gpu.alloc async [%token1] (%arg0) : memref<?xf32>
%spmatA, %token3 = gpu.create_csr async [%token2] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32> %spmatA, %token3 = gpu.create_csr async [%token2] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32>
%spmatB, %token4 = gpu.create_csr async [%token3] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32> %spmatB, %token4 = gpu.create_csr async [%token3] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32>
%spmatC, %token5 = gpu.create_csr async [%token4] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32> %spmatC, %token5 = gpu.create_csr async [%token4] %arg0, %arg0, %arg0, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32>
%spgemmDesc, %token6 = gpu.spgemm_create_descr async [%token5] %spgemmDesc, %token6 = gpu.spgemm_create_descr async [%token5]
%alloc = memref.alloc() : memref<0xi8> // nullptr %alloc = memref.alloc() : memref<0xi8> // nullptr
%c0 = arith.constant 0 : index %c0 = arith.constant 0 : index
%bufferSz1, %token7 = gpu.spgemm_work_estimation_or_compute async %bufferSz1, %token7 = gpu.spgemm_work_estimation_or_compute async
[%token6]{WORK_ESTIMATION} [%token6]{WORK_ESTIMATION}
%spmatA, %spmatB, %spmatC, %spmatA, %spmatB, %spmatC,
%spgemmDesc, %c0, %alloc: f32 into memref<0xi8> %spgemmDesc, %c0, %alloc: f32 into memref<0xi8>
%bufferSz2, %token8 = gpu.spgemm_work_estimation_or_compute async %bufferSz2, %token8 = gpu.spgemm_work_estimation_or_compute async
[%token7]{COMPUTE} [%token7]{COMPUTE}
%spmatA, %spmatB, %spmatC, %spmatA, %spmatB, %spmatC,
%spgemmDesc, %c0, %alloc: f32 into memref<0xi8> %spgemmDesc, %c0, %alloc: f32 into memref<0xi8>
%rows, %cols, %nnz, %token9 = gpu.spgemm_get_size async [%token8] %spmatC %rows, %cols, %nnz, %token9 = gpu.spgemm_get_size async [%token8] %spmatC
%token10 = gpu.set_csr_pointers async [%token8] %spmatC, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32> %token10 = gpu.set_csr_pointers async [%token8] %spmatC, %mem1, %mem1, %mem2 : memref<?xindex>, memref<?xindex>, memref<?xf32>
%token11 = gpu.spgemm_copy async [%token10] %spmatA, %spmatB, %spmatC, %spgemmDesc: f32 %token11 = gpu.spgemm_copy async [%token10] %spmatA, %spmatB, %spmatC, %spgemmDesc: f32
%token12 = gpu.spgemm_destroy_descr async [%token11] %spgemmDesc %token12 = gpu.spgemm_destroy_descr async [%token11] %spgemmDesc
gpu.destroy_sp_mat %spmatA %token13 = gpu.destroy_sp_mat async [%token12] %spmatA
gpu.destroy_sp_mat %spmatB %token14 = gpu.destroy_sp_mat async [%token13] %spmatB
gpu.destroy_sp_mat %spmatC %token15 = gpu.destroy_sp_mat async [%token14] %spmatC
gpu.wait [%token15]
return return
} }
// CHECK-LABEL: func @sddmm // CHECK-LABEL: func @sddmm
// CHECK: %{{.*}} = gpu.wait async // CHECK: %{{.*}} = gpu.wait async
// CHECK: %{{.*}}, %{{.*}} = gpu.alloc async [%{{.*}}] (%{{.*}}) : memref<?xindex> // CHECK: %{{.*}}, %{{.*}} = gpu.alloc async [%{{.*}}] (%{{.*}}) : memref<?xindex>
// CHECK: %{{.*}}, %{{.*}} = gpu.alloc async [%{{.*}}] (%{{.*}}) : memref<?xf64> // CHECK: %{{.*}}, %{{.*}} = gpu.alloc async [%{{.*}}] (%{{.*}}) : memref<?xf64>
// CHECK: %{{.*}}, %{{.*}} = gpu.create_csr async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf64> // CHECK: %{{.*}}, %{{.*}} = gpu.create_csr async [%{{.*}}] %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}} : memref<?xindex>, memref<?xindex>, memref<?xf64>
Show All 22 Lines