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

[rls-v3.8-pc] Fixed int4 weights alignment check in matmul #2945

Merged
merged 2 commits into from
Mar 27, 2025

Conversation

vpirogov
Copy link
Contributor

@vpirogov vpirogov commented Mar 25, 2025

Partial backport of #2917.

@vpirogov vpirogov requested a review from a team as a code owner March 25, 2025 00:01
@github-actions github-actions bot added backport component:tests Codeowner: @oneapi-src/onednn-arch labels Mar 25, 2025
@vpirogov
Copy link
Contributor Author

make test
disable device_cpu
enable device_gpu
disable benchdnn_all
enable benchdnn_matmul

IMPLICATION(src_strides[d] > 1, src_strides[d] % 2 == 0),
VERBOSE_BAD_DIM, "src", d);
}
}
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Seems like we should factor out the check, but fine for this backport:

bool valid_subbyte_strides(const memory_desc &md, int *bad_dim) { ... }

/* ... */

VCHECK_MATMUL(valid_subbyte_strides(weights_desc, &bad_dim), VERBOSE_BAD_DIM, "weights", bad_dim);
VCHECK_MATMUL(valid_subbyte_strides(src_desc, &bad_dim), VERBOSE_BAD_DIM, "src", bad_dim);

@vpirogov vpirogov changed the title Fixed int4 weights alignment check in matmul [rls-v3.8-pc] Fixed int4 weights alignment check in matmul Mar 27, 2025
@vpirogov vpirogov merged commit d97791a into rls-v3.8-pc Mar 27, 2025
28 of 32 checks passed
@vpirogov vpirogov deleted the vpirogov/v3.8-pc-int4 branch March 27, 2025 18:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
backport component:tests Codeowner: @oneapi-src/onednn-arch
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants