Skip to content

matmul bug fixes. #547

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

Merged
luitjens merged 1 commit into from
Jan 5, 2024
+6 −1
Merged

matmul bug fixes. #547

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
7 changes: 6 additions & 1 deletion include/matx/transforms/matmul.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1107,7 +1107,8 @@ __MATX_INLINE__ auto getCublasSupportedTensor( const Op &in, cudaStream_t stream
(in.Stride(RANK-1) != (index_t)1 && in.Stride(RANK-2) != (index_t)1) ||
// cublas allows 0 strides, but verify that the corresponding size is 1
(in.Stride(RANK-1) == (index_t)0 && in.Size(RANK-1) != (index_t)1) ||
(in.Stride(RANK-2) == (index_t)0 && in.Size(RANK-2) != (index_t)1)
(in.Stride(RANK-2) == (index_t)0 && in.Size(RANK-2) != (index_t)1) ||
in.Stride(RANK-2) == 0 // WAR for CUBLAS bug
) {
supported = false;
}
Expand Down Expand Up @@ -1192,6 +1193,10 @@ void matmul_impl(TensorTypeC C, const TensorTypeA A,
if(!b.isSameView(B_)) {
(b = B_).run(stream);
}

if(beta != 0 && !c.isSameView(C)) {
(c = C).run(stream);
}

#if MATX_ENABLE_CUTLASS != 1
// cublasLt does not allow transpose modes on C. Thus we need to make sure that the right most dimension has a stride of 1.
Expand Down