Skip to content

LATX, AVX: use direct widening multiply-add sequences - #456

Closed
luzeng87 wants to merge 1 commit into
lat-opensource:masterfrom
luzeng87:avx-pmadd-lowering
Closed

LATX, AVX: use direct widening multiply-add sequences#456
luzeng87 wants to merge 1 commit into
lat-opensource:masterfrom
luzeng87:avx-pmadd-lowering

Conversation

@luzeng87

@luzeng87 luzeng87 commented Sep 2, 2026

Copy link
Copy Markdown
Contributor

Summary

  • lower VPMADDWD to widening even/odd multiplies plus addition without clearing and copying a temporary vector
  • lower VPMADDUBSW with the mixed unsigned-byte/signed-byte widening multiply instructions
  • preserve source/destination overlap by selecting a temporary sequence only when required
  • apply the same lowering to LSX lane translation and native LASX translation

Validation

  • clean 3A6000 build: passed
  • VPMADDWD and VPMADDUBSW over 100,000 deterministic input sets: passed
  • signed saturation and source-value checks: passed
  • JIT, cold AOT, and hot AOT: passed with a non-empty AOT file
  • git diff --check: passed

Signed-off-by: Lu Zeng <luzeng87@gmail.com>
@luzeng87
luzeng87 marked this pull request as draft September 3, 2026 03:07
@luzeng87

luzeng87 commented Sep 3, 2026

Copy link
Copy Markdown
Contributor Author

Superseded by grouped AVX lowering PR #448. The original commit remains separate inside that PR for review.

@luzeng87 luzeng87 closed this Sep 3, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant