[all-commits] [llvm/llvm-project] 97a1f4: DAG: Avoid breaking legal vector_shuffle with mult...

[Matt Arsenault via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 97a1f494a6778df24cfca817a30b73b1f0a7228a
      https://github.com/llvm/llvm-project/commit/97a1f494a6778df24cfca817a30b73b1f0a7228a
  Author: Matt Arsenault <Matthew.Arsenault at amd.com>
  Date:   2025-01-30 (Thu, 30 Jan 2025)

  Changed paths:
    M llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4f32.v2f32.ll
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4f32.v3f32.ll
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4f32.v4f32.ll
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4i32.v2i32.ll
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4i32.v3i32.ll
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4i32.v4i32.ll
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4p3.v2p3.ll
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4p3.v3p3.ll
    M llvm/test/CodeGen/AMDGPU/shufflevector.v4p3.v4p3.ll

  Log Message:
  -----------
  DAG: Avoid breaking legal vector_shuffle with multiple uses (#123712)

Previously this combine would undo AMDGPU's new custom legalization of
wide vector shuffles into 2 element pieces. The comment also
states that this combine is only done before legalization,
but the case with a build_vector source was unconditional.

We probably don't want to do this if the multiple uses are full
scalarization of the vector, but this seems to work well enough.
Scalarizing extracts should have folded out pre-legalize.



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