perf(linalg): run SiLU on a dedicated AVX-512 kernel - #2589
Draft
cverrier wants to merge 2 commits into
Draft
Conversation
AVX-512 hosts ran silu_f32 and silu_f16 through a kernel composed over the zmm sigmoid, which pays a scratch copy and an extra traversal that the fused FMA kernel avoids. They now use the fused kernel at both widths. The composed kernel has no remaining users, and is removed.
AVX-512 hosts ran SiLU through the 256-bit FMA kernel for both f32 and f16. A 16-wide kernel is faster on both, worth ~8% on f32 buffers and ~29% on f16 — the latter only once the f16 wrapper hands it a slice long enough to amortize the fixed cost every f32 kernel call pays regardless of its length.
cverrier
force-pushed
the
perf/run-silu-on-dedicated-avx512-kernel
branch
from
August 13, 2026 07:48
bc41a8d to
31a36ba
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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
AVX-512 hosts ran SiLU through the 256-bit FMA kernel for both f32 and f16. A 16-wide kernel is faster on both, worth ~8% on f32 buffers and ~29% on f16 — the latter only once the f16 wrapper hands it a slice long enough to amortize the fixed cost every f32 kernel call pays regardless of its length.
Ready for review once #2567 is merged.