[all-commits] [llvm/llvm-project] 813cf3: [X86] Fold XOR of VGF2P8AFFINEQB and its source (#...

[Walter via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 813cf3358e5eefc4a9f7393ea373f50c943fd893
      https://github.com/llvm/llvm-project/commit/813cf3358e5eefc4a9f7393ea373f50c943fd893
  Author: Walter <90877047+WalterKruger at users.noreply.github.com>
  Date:   2026-06-01 (Mon, 01 Jun 2026)

  Changed paths:
    M llvm/lib/Target/X86/X86ISelLowering.cpp
    M llvm/test/CodeGen/X86/gfni-operand-and-fold.ll
    M llvm/test/CodeGen/X86/gfni-xor-fold-avx512.ll
    M llvm/test/CodeGen/X86/gfni-xor-fold.ll

  Log Message:
  -----------
  [X86] Fold XOR of VGF2P8AFFINEQB and its source (#198448)

Adds a optimization to fold XORs between `vgf2p8affineqb` and its
source. That XOR can done through the existing affine transformation by
XORing its matrix with the identity matrix. This patch:

- Folds XOR between `vgf2p8affineqb` and its source.
- Only occurs when the matrix is constant, ensuring that it can't
increase the dependency chain.
- Doesn't occur if the affine is multi use, preventing an increase in
code size.
- Includes test coverage for both positive and negative cases.

Fixes #184061



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