[all-commits] [llvm/llvm-project] ca085a: [SelectionDAG] Pre-commit tests for dagcombine imp...

[Krzysztof Drewniak via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: ca085afb41ab167597e76daf732b4f5a6eb36298
      https://github.com/llvm/llvm-project/commit/ca085afb41ab167597e76daf732b4f5a6eb36298
  Author: Krzysztof Drewniak <Krzysztof.Drewniak at amd.com>
  Date:   2026-06-04 (Thu, 04 Jun 2026)

  Changed paths:
    A llvm/test/CodeGen/X86/freeze-fp.ll
    M llvm/test/CodeGen/X86/freeze-vector.ll
    M llvm/test/CodeGen/X86/madd.ll
    M llvm/test/CodeGen/X86/vector-shuffle-combining-avx512bwvl.ll

  Log Message:
  -----------
  [SelectionDAG] Pre-commit tests for dagcombine improvements (#201270)

I've got a stack of dagcombine improvements that together make an
infinite cycle relating to freeze insertion in vector-manipulation IR.
Here we have

- Handling freeze(undef) in demanded-elts for shufflevector
- Improvements to noundef checks for bitcast, concat, and select
- Improvements to extract(concat), extract(extract), and extract(insert)
handling

Even though the regression I'm fixing is an AMDGPU one, these tests are
mainly X86 because the AMDGPU calling convention makes it hard to
demonstrate the folds I'm adding.

AI note: I got an LLM to find most of these tests, especially some of
the fiddly ones that needed control flow.



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