Repository navigation
perf(gemm): native float32 GEMM on loong64 - #81
Draft
tannevaled wants to merge 1 commit into
Draft
tannevaled wants to merge 1 commit into
tannevaled wants to merge 1 commit into
Conversation
gemmMicro8x16LASX is the float64 8x8 LASX kernel at twice the lanes: 16 eight-lane accumulators, the B row in two registers, each A value broadcast with xvldrepl.w (XVMOVQ .W8), and xvfmadd.s through go-asmgen v0.23.0's new XVFMADDF encoder. Gated on useLASX; without it, float32 products widen to the float64 GEMM. Validated under qemu la464 in CI (NDARRAY_REQUIRE_LASX): the Loongson host, cfarm401, refuses our key, so it is not measured on hardware yet. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
tannevaled
force-pushed
the
perf/gemm32-loong64
branch
from
October 31, 2026 20:36
3c76426 to
ef2785b
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Native float32 GEMM on loong64:
gemmMicro8x16LASX(LASX 8×16), using go-asmgen v0.23.0's newXVFMADDFencoder (go-asmgen/asmgen#37). It is gated onuseLASX.la464,NDARRAY_REQUIRE_LASX).Still a draft: not measured on hardware. The Loongson host (cfarm401) refuses our key, and qemu timings mean nothing. On a 3C5000L an earlier LASX kernel (
blocks8) was 26–47% slower than scalar code, so this one merges only after a hardware A/B shows the gain.🤖 Generated with Claude Code