cpu: aarch64: Re-Enable JIT Depthwise Convolution for BF16 #3441
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
dzarukin
approved these changes
Jun 18, 2025
renato-arantes
commented
Jun 19, 2025
| if (jcp.dst_dt == data_type::f32) { | ||
| ld1w(ZRegS(0), P_ALL_ONE, ptr(reg_tmp_addr)); | ||
| fadd(zregs_acc, zregs_acc, ZRegS(0)); | ||
| } else if (jcp.dst_dt == data_type::bf16) { |
There was a problem hiding this comment.
Just curious: which change addressed the aforementioned failures?
This is the change that deals with the failed tests. The post-op sum had not been addressed previously.
renato-arantes
force-pushed
the
jit_bf16_depth_again
branch
from
7432276 to
7a71365
Compare
jondea
approved these changes
Jun 19, 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Outdated suggestions cannot be applied.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
This PR re-enables the PR 3308, which was reverted due to failing nightly tests. It enables JIT depthwise convolution for
bf16, preventing it from going torefand significantly improving its performance, from ~4% for one thread to ~2.5x for 32 threads, when compared to thef32JIT operation. These performance numbers were extracted from a benchmark executed on an AWSc7g.16xlargeinstance.General
make testandmake test_benchdnn_*) pass locally for each commit?Performance improvements
Below are small test case logs that demonstrate performance numbers before and after.
OMP_NUM_THREADS=16 ./tests/benchdnn/benchdnn --conv --dt=bf16 --mode=p --alg=convolution_direct g32mb64_ic32oc32_ih112oh112kh3sh1dh0ph1_iw112ow112kw3sw1dw0pw1bf16: total perf: min(ms):0.815918 avg(ms):0.826171f32: total perf: min(ms):1.02075 avg(ms):1.03415ref: total perf: min(ms):225.645 avg(ms):225.808