cpu: x64 brgemm: fix for destination data check #4497
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This fixes issue that brgemm will return unsupported when a post op results in a u8 or s8 data type for float16 and bfloat16 brgemm calculations.
github-actions
bot
added
the
platform:cpu-x64
Contributor
Author
|
make test |
dzarukin
reviewed
Jan 6, 2026
Contributor
dzarukin
left a comment
dzarukin
left a comment
There was a problem hiding this comment.
Please add test cases to brgemm driver in benchdnn to verify it's passing.
ankalinin
approved these changes
Jan 6, 2026
|
@georgen117 Could you help backport to rls-v3.10? OpenVINO CPU's target is rls-v3.10. Thank you very much. |
Contributor
Author
done: #4504 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This fixes issue that brgemm will return unsupported when a post op results in a u8 or s8 data type for float16 and bfloat16 brgemm calculations.
This issue was reported in responce to earlyer fix that was published to catch unsupported data types.
The added check did not account for the fact that post ops could result in the destination data type being u8 or s8.
u8 and s8 has been added as a valid destination data type for float16 and bfloat16 brgemm operations. Only bf16 models are known to be affected by this issue.
this was validatated using the sde and following benchdnn tests.
benchdnn --brgemm --dt=bf16:bf16:u8 --attr-scales=dst:common:0.05 32x32:32x32
benchdnn --brgemm --dt=bf16:bf16:s8 --attr-scales=dst:common:0.05 32x32:32x32
benchdnn --brgemm --dt=f16:f16:u8 --attr-scales=dst:common:0.05 32x32:32x32
benchdnn --brgemm --dt=f16:f16:s8 --attr-scales=dst:common:0.05 32x32:32x32
this should resolve an issue reported by the OpenVINO team related to post-step affecting some known models.
see PR#3887 to see original fix for MFDNN-13836