projects
/
platform
/
core
/
ml
/
nntrainer.git
/ commitdiff
commit
grep
author
committer
pickaxe
?
search:
re
summary
|
shortlog
|
log
|
commit
| commitdiff |
tree
raw
|
patch
| inline |
side by side
(parent:
a8993a2
)
[ blas/bugfix ] Fix irrelevant function call
author
skykongkong8
<ss.kong@samsung.com>
Mon, 26 Aug 2024 02:17:19 +0000
(11:17 +0900)
committer
MyungJoo Ham
<myungjoo.ham@samsung.com>
Mon, 26 Aug 2024 08:29:34 +0000
(17:29 +0900)
- Since current function implementations are not using CBLAS params, should directly call function from cblas.h
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: skykongkong8 <ss.kong@samsung.com>
nntrainer/tensor/blas_interface.cpp
patch
|
blob
|
history
diff --git
a/nntrainer/tensor/blas_interface.cpp
b/nntrainer/tensor/blas_interface.cpp
index 938d14c49bbd776e4b210468d37acff34e8a52cf..91187c50ccb0e7cf019374da62009bc546201038 100644
(file)
--- a/
nntrainer/tensor/blas_interface.cpp
+++ b/
nntrainer/tensor/blas_interface.cpp
@@
-343,7
+343,8
@@
static void sgemm_FP16(const unsigned int TStorageOrder, bool TransA,
scopy(M * K, A, 1, A_, 1);
scopy(N * K, B, 1, B_, 1);
scopy(M * N, C, 1, C_, 1);
- sgemm(order, transA, transB, M, N, K, alpha, A_, lda, B_, ldb, beta, C_, ldc);
+ cblas_sgemm(order, transA, transB, M, N, K, alpha, A_, lda, B_, ldb, beta, C_,
+ ldc);
scopy(M * N, C_, 1, C, 1);
delete[] A_;