cpu: aarch64: prefer brgemm over jit for 1x1 convolutions with sve_256 #3411
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
github-actions
bot
added
platform:cpu-aarch64
Ryo-not-rio
requested changes
Jun 10, 2025
| @@ -44,6 +44,7 @@ using namespace data_type; | |||
|
|
|||
| template <cpu_isa_t isa> | |||
| status_t brgemm_1x1_convolution_fwd_t<isa>::pd_t::init(engine_t *engine) { | |||
|
|
|||
Comment on lines
84
to
85
|
|
||
|
|
Comment on lines
208
to
209
|
|
||
|
|
Anallear
force-pushed
the
fixreg
branch
2 times, most recently
from
d5ee1d9 to
d94617f
Compare
Moving brgemm_1x1_convolution_fwd_t<sve_256> before the jit_sve_1x1_convolution_fwd implementation based on benchmark results showing better performance.
Ryo-not-rio
approved these changes
Jun 13, 2025
Radu2k
approved these changes
Jun 13, 2025
vpirogov
approved these changes
Jun 17, 2025
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Outdated suggestions cannot be applied.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.
Description
Moving brgemm_1x1_convolution_fwd_t<sve_256> before the jit_sve_1x1_convolution_fwd implementation based on benchmark results showing better performance.
Performance improvements
mb: mini batch size , ic: input channels , S: Stride
Brgconv,
S=1, mb=64, ic=384 ,time= 21.25 ms
S=2, mb=64, ic=384 ,time= 5.48 ms
S=3, mb=64, ic=384 ,time= 2.57 ms
JIT_1X1,
S=1, mb=64, ic=384 ,time= 21.62 ms
S=2, mb=64, ic=384 ,time= 6.16 ms
S=3, mb=64, ic=384 ,time= 5.87 ms
Bug fixes