squall/cutlass
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Merge pull request #214 from Peter9606/separate-stream-error

Browse Source 
Bugfix: memsetAsync uses wrong default stream
This commit is contained in:
Haicheng Wu 2021-03-24 12:09:01 -04:00 committed by GitHub
commit 8a3e4b8d02
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
7 changed files with 15 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
include/cutlass/conv/device/implicit_gemm_convolution.h
View File

@ -244,7 +244,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);

6
include/cutlass/gemm/device/gemm.h
View File

@ -482,7 +482,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);
@ -673,7 +673,7 @@ public:
/// Initializes GEMM state from arguments. /// Initializes GEMM state from arguments.
Status initialize(Arguments const &args, void *workspace = nullptr, cudaStream_t stream = nullptr) { Status initialize(Arguments const &args, void *workspace = nullptr, cudaStream_t stream = nullptr) {
return underlying_operator_.initialize(to_underlying_arguments(args), workspace); return underlying_operator_.initialize(to_underlying_arguments(args), workspace, stream);
} }
/// Lightweight update given a subset of arguments /// Lightweight update given a subset of arguments
@ -699,7 +699,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);

6
include/cutlass/gemm/device/gemm_array.h
View File

@ -473,7 +473,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);
@ -700,7 +700,7 @@ public:
/// Initializes GEMM state from arguments. /// Initializes GEMM state from arguments.
Status initialize(Arguments const &args, void *workspace = nullptr, cudaStream_t stream = nullptr) { Status initialize(Arguments const &args, void *workspace = nullptr, cudaStream_t stream = nullptr) {
return underlying_operator_.initialize(to_underlying_arguments(args), workspace); return underlying_operator_.initialize(to_underlying_arguments(args), workspace, stream);
} }
/// Lightweight update given a subset of arguments /// Lightweight update given a subset of arguments
@ -726,7 +726,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);

6
include/cutlass/gemm/device/gemm_batched.h
View File

@ -451,7 +451,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);
@ -666,7 +666,7 @@ public:
/// Initializes GEMM state from arguments. /// Initializes GEMM state from arguments.
Status initialize(Arguments const &args, void *workspace = nullptr, cudaStream_t stream = nullptr) { Status initialize(Arguments const &args, void *workspace = nullptr, cudaStream_t stream = nullptr) {
return underlying_operator_.initialize(to_underlying_arguments(args), workspace); return underlying_operator_.initialize(to_underlying_arguments(args), workspace, stream);
} }
/// Lightweight update given a subset of arguments /// Lightweight update given a subset of arguments
@ -692,7 +692,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);

6
include/cutlass/gemm/device/gemm_complex.h
View File

@ -465,7 +465,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);
@ -674,7 +674,7 @@ public:
/// Initializes GEMM state from arguments. /// Initializes GEMM state from arguments.
Status initialize(Arguments const &args, void *workspace = nullptr, cudaStream_t stream = nullptr) { Status initialize(Arguments const &args, void *workspace = nullptr, cudaStream_t stream = nullptr) {
return underlying_operator_.initialize(to_underlying_arguments(args), workspace); return underlying_operator_.initialize(to_underlying_arguments(args), workspace, stream);
} }
/// Lightweight update given a subset of arguments /// Lightweight update given a subset of arguments
@ -700,7 +700,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);

2
include/cutlass/gemm/device/gemm_sparse.h
View File

@ -498,7 +498,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);

2
include/cutlass/reduction/device/reduce_split_k.h
View File

@ -197,7 +197,7 @@ public:
void *workspace = nullptr, void *workspace = nullptr,
cudaStream_t stream = nullptr) { cudaStream_t stream = nullptr) {
Status status = initialize(args, workspace); Status status = initialize(args, workspace, stream);
if (status == Status::kSuccess) { if (status == Status::kSuccess) {
status = run(stream); status = run(stream);