[all-commits] [llvm/llvm-project] 869d8b: [SLP][X86][NFC] Add pre-commit test for width-3 sh...

[Milin Bhade via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 869d8b74ff2b1b823005e9b7ac57c2bf78904503
      https://github.com/llvm/llvm-project/commit/869d8b74ff2b1b823005e9b7ac57c2bf78904503
  Author: Milin Bhade <mbhade at amd.com>
  Date:   2026-09-16 (Wed, 16 Sep 2026)

  Changed paths:
    A llvm/test/Transforms/SLPVectorizer/X86/nonpow2-shared-weight-reduction.ll

  Log Message:
  -----------
  [SLP][X86][NFC] Add pre-commit test for width-3 shared-weight reduction (#222243)

Pre-commit test for three independent reductions over contiguous bytes
sharing a single broadcast weight. It captures current codegen: even
with -slp-vectorize-non-power-of-2=true the width-3 tree is discarded to
scalar because the non-power-of-2 <3 x i8> load is over-priced (the
folded vpinsr*(mem) is double-counted). A follow-up X86 cost fix flips
the NPOT run to the width-3 vector form.

cc @alexey-bataev

---------

Co-authored-by: Cursor <cursoragent at cursor.com>



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications