[all-commits] [llvm/llvm-project] 87c971: [SLP]Emit loop-carried horizontal reductions as lo...

[Alexey Bataev via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 87c971d36345c0ff4f7a7ae75080bd55e83e4fca
      https://github.com/llvm/llvm-project/commit/87c971d36345c0ff4f7a7ae75080bd55e83e4fca
  Author: Alexey Bataev <a.bataev at outlook.com>
  Date:   2026-09-21 (Mon, 21 Sep 2026)

  Changed paths:
    M llvm/lib/Transforms/Vectorize/SLPVectorizer.cpp
    M llvm/test/Transforms/SLPVectorizer/AArch64/gather-root.ll
    M llvm/test/Transforms/SLPVectorizer/AArch64/getelementptr.ll
    M llvm/test/Transforms/SLPVectorizer/AArch64/horizontal.ll
    M llvm/test/Transforms/SLPVectorizer/AArch64/loop-accumulator-reduction.ll
    M llvm/test/Transforms/SLPVectorizer/X86/loop-accumulator-reduction.ll
    M llvm/test/Transforms/SLPVectorizer/X86/slp-fma-loss.ll

  Log Message:
  -----------
  [SLP]Emit loop-carried horizontal reductions as loop vector accumulator

A reduction whose root is the backedge value of a loop accumulator phi
is emitted as a lane-wise vector accumulator in the loop plus a single
reduction in the exit blocks, instead of a horizontal reduction on every
iteration. The loop-carried dependency chain becomes lane-wise and the
reduction is paid once per loop execution.

Note: It is partial vectorization within the loop, LV cannot handle it.
SLP already does something similar with multi-register reductions,
this adds another level - phi-based loop accumulator reductions.

Reviewers: RKSimon, bababuck

Pull Request: https://github.com/llvm/llvm-project/pull/221598



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