brgemm: support arbitrary K on AMX #2319
Open
Conversation
This file contains 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
github-actions
bot
added
platform:cpu-x64
|
make test |
densamoilov
approved these changes
Dec 27, 2024
dzarukin
approved these changes
Dec 27, 2024
ankalinin
force-pushed
the
akalinin/brgemm_k_tail_pr
branch
from
527c2a0 to
4bd35fa
Compare
github-actions
bot
added
platform:cpu-aarch64
ankalinin
force-pushed
the
akalinin/brgemm_k_tail_pr
branch
from
4bd35fa to
280499f
Compare
|
I restarted the AArch64 CI because the encountered failure is a known sporadic bug on the c7g. |
|
cc: @Radu2k |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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Outdated suggestions cannot be applied.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.
There are two problems regarding brgemm K value on AMX:
brgemm doesn't support K not divisible by vnni granularity
for K not divisible by tile width (32 or 64) the blocking by K dimension may be not optimal.
To get around this limitation brgemm primitives are forced to either transform the matrix A or call many small brgemm kernels with different tile configurations. Both ways leads to performance lost.
This PR implements support of arbitrary K in brgemm on AMX and implements corresponding updates in 1x1 convolutions and in matmul to use this new ability.
Performance update for convolutions:
Performance update for matmul
