Add qd8-qb4w-bf16 scalar kernels#9960
Open
GregoryComer wants to merge 2 commits into
Open
Conversation
27 tasks
dsharlet
approved these changes
Apr 14, 2026
597b894 to
2b8ee89
Compare
Contributor
Author
|
I regenerated the kernel build sources - CI should pass now. Verified on x86 Linux and ARM Mac locally. |
Contributor
Author
|
@dsharlet Could you retrigger CI when you have time? Not urgent. |
2b8ee89 to
032143b
Compare
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Add scalar qd8-qb4w-bf16 kernels. This is mainly just an output conversion, so kernels are very similar to the existing qd8-qb4w-f16/f32. Performance is with ~5% of scalar fp16. The scalar rounding logic is strictly correct for NaNs/INFs and has correct round to even support, so it's a little expensive. The M4 Max f16 kernels beat bf16 slightly because it uses native scalar fp16 arithmetic in the default build config.
Benchmarks
M4 Max
Cortex-X1 (Pixel 7 Pro)
Broadwell