Skip to content

xe: gemm: reintroduce kernels to spotfix regressions#4766

Open
Simonsays095 wants to merge 1 commit intouxlfoundation:mainfrom
Simonsays095:xe2_regression_fix
Open

xe: gemm: reintroduce kernels to spotfix regressions#4766
Simonsays095 wants to merge 1 commit intouxlfoundation:mainfrom
Simonsays095:xe2_regression_fix

Conversation

@Simonsays095
Copy link
Contributor

Addresses MFDNN-14483. Reintroduces one kernel that was removed. Since the performance model sometimes dispatches to this kernel when the other kernels would be better, this kernel is strictly limited to a few cases present in currently-used models. This is why one kernel is duplicated 3x - each entry covers another shape.

@Simonsays095
Copy link
Contributor Author

make test perf-gpu
set primitive=gpu:gemm

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

platform:gpu-intel Codeowner: @oneapi-src/onednn-gpu-intel

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants