Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

【Hackathon 5th No.110】为 Paddle 增强 sparse.matmul API #59890

Merged
merged 27 commits into from
Jan 12, 2024
Merged

【Hackathon 5th No.110】为 Paddle 增强 sparse.matmul API #59890

Show file tree
Hide file tree
Changes from 9 commits
Commits
Show all changes
27 commits
Select commit Hold shift + click to select a range
4641b7f
add cusparseSpGEMM
MayYouBeProsperous Nov 5, 2023
9c82903
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into sm
MayYouBeProsperous Nov 10, 2023
782f92d
refine paddle.sparse.matmul
MayYouBeProsperous Nov 10, 2023
6c5dbb7
test
MayYouBeProsperous Nov 27, 2023
96c9cd5
fix
MayYouBeProsperous Dec 11, 2023
044ba38
coo*coo
MayYouBeProsperous Dec 12, 2023
ace8c96
add grad op
MayYouBeProsperous Dec 13, 2023
ff40558
fix
MayYouBeProsperous Dec 14, 2023
03a08cc
fix
MayYouBeProsperous Dec 15, 2023
14deb87
fix 3d
MayYouBeProsperous Dec 19, 2023
a2019d1
codestyle
MayYouBeProsperous Dec 19, 2023
67bb445
fix
MayYouBeProsperous Dec 19, 2023
0ee3545
fix
MayYouBeProsperous Dec 19, 2023
c8ff3d5
use CastCsrKernel
MayYouBeProsperous Dec 21, 2023
253ca38
optimize int32 index
MayYouBeProsperous Dec 30, 2023
0851dd3
Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into sm
MayYouBeProsperous Dec 30, 2023
8dfe895
fix
MayYouBeProsperous Dec 30, 2023
9c80efe
fix
MayYouBeProsperous Dec 31, 2023
477dc5c
fix cuda<11.8
MayYouBeProsperous Jan 2, 2024
ea71c64
ci
MayYouBeProsperous Jan 3, 2024
4397dbb
fix batched computation
MayYouBeProsperous Jan 9, 2024
28449d4
fix
MayYouBeProsperous Jan 9, 2024
76e5ed9
fix
MayYouBeProsperous Jan 9, 2024
ee0fee5
fix cuda version check
MayYouBeProsperous Jan 10, 2024
0d23935
fix
MayYouBeProsperous Jan 10, 2024
e01251b
fix bugs
MayYouBeProsperous Jan 11, 2024
1f34f1c
fix
MayYouBeProsperous Jan 11, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
48 changes: 28 additions & 20 deletions paddle/fluid/platform/dynload/cusparse.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,26 +30,34 @@ namespace dynload {

#if defined(PADDLE_WITH_CUDA)
#if CUDA_VERSION >= 11000
#define CUSPARSE_ROUTINE_EACH(__macro) \
__macro(cusparseCreate); \
__macro(cusparseSetStream); \
__macro(cusparseCreateMatDescr); \
__macro(cusparseDestroy); \
__macro(cusparseSnnz); \
__macro(cusparseDnnz); \
__macro(cusparseSetMatType); \
__macro(cusparseSetMatIndexBase); \
__macro(cusparseCreateCsr); \
__macro(cusparseCreateCoo); \
__macro(cusparseCreateDnMat); \
__macro(cusparseCreateDnVec); \
__macro(cusparseSpMM_bufferSize); \
__macro(cusparseSpMM); \
__macro(cusparseDestroySpMat); \
__macro(cusparseDestroyDnMat); \
__macro(cusparseDestroyDnVec); \
__macro(cusparseSpMV_bufferSize); \
__macro(cusparseSpMV);
#define CUSPARSE_ROUTINE_EACH(__macro) \
__macro(cusparseCreate); \
__macro(cusparseSetStream); \
__macro(cusparseCreateMatDescr); \
__macro(cusparseDestroy); \
__macro(cusparseSnnz); \
__macro(cusparseDnnz); \
__macro(cusparseSetMatType); \
__macro(cusparseSetMatIndexBase); \
__macro(cusparseCreateCsr); \
__macro(cusparseCreateCoo); \
__macro(cusparseCreateDnMat); \
__macro(cusparseCreateDnVec); \
__macro(cusparseSpMM_bufferSize); \
__macro(cusparseSpMM); \
__macro(cusparseDestroySpMat); \
__macro(cusparseDestroyDnMat); \
__macro(cusparseDestroyDnVec); \
__macro(cusparseSpMV_bufferSize); \
__macro(cusparseSpMV); \
__macro(cusparseSpMatGetSize); \
__macro(cusparseSpGEMM_createDescr); \
__macro(cusparseSpGEMM_compute); \
__macro(cusparseSpGEMM_workEstimation); \
__macro(cusparseSpGEMM_copy); \
__macro(cusparseSpGEMM_destroyDescr); \
__macro(cusparseCsr2cscEx2_bufferSize); \
__macro(cusparseCsr2cscEx2);

CUSPARSE_ROUTINE_EACH(PLATFORM_DECLARE_DYNAMIC_LOAD_CUSPARSE_WRAP)
#endif
Expand Down
48 changes: 28 additions & 20 deletions paddle/phi/backends/dynload/cusparse.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -42,26 +42,34 @@ extern void *cusparse_dso_handle;

#if defined(PADDLE_WITH_CUDA)
#if CUDA_VERSION >= 11000
#define CUSPARSE_ROUTINE_EACH(__macro) \
__macro(cusparseCreate); \
__macro(cusparseSetStream); \
__macro(cusparseCreateMatDescr); \
__macro(cusparseDestroy); \
__macro(cusparseSnnz); \
__macro(cusparseDnnz); \
__macro(cusparseSetMatType); \
__macro(cusparseSetMatIndexBase); \
__macro(cusparseCreateCsr); \
__macro(cusparseCreateCoo); \
__macro(cusparseCreateDnMat); \
__macro(cusparseCreateDnVec); \
__macro(cusparseSpMM_bufferSize); \
__macro(cusparseSpMM); \
__macro(cusparseDestroySpMat); \
__macro(cusparseDestroyDnMat); \
__macro(cusparseDestroyDnVec); \
__macro(cusparseSpMV_bufferSize); \
__macro(cusparseSpMV);
#define CUSPARSE_ROUTINE_EACH(__macro) \
__macro(cusparseCreate); \
__macro(cusparseSetStream); \
__macro(cusparseCreateMatDescr); \
__macro(cusparseDestroy); \
__macro(cusparseSnnz); \
__macro(cusparseDnnz); \
__macro(cusparseSetMatType); \
__macro(cusparseSetMatIndexBase); \
__macro(cusparseCreateCsr); \
__macro(cusparseCreateCoo); \
__macro(cusparseCreateDnMat); \
__macro(cusparseCreateDnVec); \
__macro(cusparseSpMM_bufferSize); \
__macro(cusparseSpMM); \
__macro(cusparseDestroySpMat); \
__macro(cusparseDestroyDnMat); \
__macro(cusparseDestroyDnVec); \
__macro(cusparseSpMV_bufferSize); \
__macro(cusparseSpMV); \
__macro(cusparseSpMatGetSize); \
__macro(cusparseSpGEMM_createDescr); \
__macro(cusparseSpGEMM_compute); \
__macro(cusparseSpGEMM_workEstimation); \
__macro(cusparseSpGEMM_copy); \
__macro(cusparseSpGEMM_destroyDescr); \
__macro(cusparseCsr2cscEx2_bufferSize); \
__macro(cusparseCsr2cscEx2);

CUSPARSE_ROUTINE_EACH(DECLARE_DYNAMIC_LOAD_CUSPARSE_WRAP)
#endif
Expand Down
17 changes: 17 additions & 0 deletions paddle/phi/kernels/funcs/sparse/sparse_blas.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,9 @@
namespace phi {
namespace funcs {
namespace sparse {
template <typename T, typename IntT>
SparseCsrTensor CSRTanspose(const phi::GPUContext& dev_ctx,
const phi::SparseCsrTensor& x);

template <typename DeviceContext>
class SparseBlas {
Expand Down Expand Up @@ -54,6 +57,15 @@ class SparseBlas {
T beta,
TensorType* mat_out) const;

template <typename T>
void SPGEMM(bool transa,
bool transb,
T alpha,
const SparseCsrTensor& mat_a,
const SparseCsrTensor& mat_b,
T beta,
SparseCsrTensor* mat_out) const;

private:
const DeviceContext& dev_ctx_;
};
Expand All @@ -78,6 +90,11 @@ class SparseBlasT : private SparseBlas<DeviceContext> {
Base()->template SDDMM<T>(args...);
}

template <typename... ARGS>
void SPGEMM(ARGS... args) const {
Base()->template SPGEMM<T>(args...);
}

private:
const SparseBlas<DeviceContext>* Base() const {
return static_cast<const SparseBlas<DeviceContext>*>(this);
Expand Down
Loading