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.

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

xe: sdpa: Fix KQ gemm alignment for the K tensor #2283

Merged
merged 2 commits into from
Dec 19, 2024

Conversation

umar456
Copy link
Contributor

@umar456 umar456 commented Dec 18, 2024

Description

This PR fixes an issue with the alignment on the KQ matrix multiplication. This fixes several accuracy issues for smaller tensor sizes.

@umar456 umar456 added bug A confirmed library bug platform:gpu-intel Codeowner: @oneapi-src/onednn-gpu-intel labels Dec 18, 2024
@umar456 umar456 requested a review from a team as a code owner December 18, 2024 01:48
@umar456
Copy link
Contributor Author

umar456 commented Dec 18, 2024

make test
disable device_cpu
disable benchdnn_all
enable benchdnn_nightly
enable benchdnn_graph

@umar456 umar456 force-pushed the uarshad/sdpa_alignment_fix branch from 02fd3ad to b3bcf85 Compare December 18, 2024 17:24
@umar456
Copy link
Contributor Author

umar456 commented Dec 18, 2024

make test
disable device_cpu
disable benchdnn_all
enable benchdnn_nightly
enable benchdnn_graph

@umar456 umar456 force-pushed the uarshad/sdpa_alignment_fix branch from b3bcf85 to cd1fd9c Compare December 19, 2024 05:21
@umar456
Copy link
Contributor Author

umar456 commented Dec 19, 2024

make test
disable device_cpu
disable benchdnn_all
enable benchdnn_nightly
enable benchdnn_graph

@umar456 umar456 merged commit c3b5d23 into main Dec 19, 2024
8 of 10 checks passed
@umar456 umar456 deleted the uarshad/sdpa_alignment_fix branch December 19, 2024 15:46
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug A confirmed library bug platform:gpu-intel Codeowner: @oneapi-src/onednn-gpu-intel
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants