[all-commits] [llvm/llvm-project] fb054e: [X86] splitAndLowerShuffle - split a v8f32 bitcast...

[Simon Pilgrim via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: fb054e6cf2b138835301966abc7ac94af92261c3
      https://github.com/llvm/llvm-project/commit/fb054e6cf2b138835301966abc7ac94af92261c3
  Author: Simon Pilgrim <llvm-dev at redking.me.uk>
  Date:   2025-06-10 (Tue, 10 Jun 2025)

  Changed paths:
    M llvm/lib/Target/X86/X86ISelLowering.cpp
    M llvm/test/CodeGen/X86/any_extend_vector_inreg_of_broadcast_from_memory.ll
    M llvm/test/CodeGen/X86/oddshuffles.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-3.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-4.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-5.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-6.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-7.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-8.ll
    M llvm/test/CodeGen/X86/vector-shuffle-256-v8.ll
    M llvm/test/CodeGen/X86/zero_extend_vector_inreg_of_broadcast_from_memory.ll

  Log Message:
  -----------
  [X86] splitAndLowerShuffle - split a v8f32 bitcast from v8i32 operands as 2 v4i32 shuffles (#143493)

AVX1 performs v8i32 shuffles as bitcast v8f32, but if we split these back to v4f32 instead of peeking through the bitcasts, we can lose track of the original domain.

Fixes an issue I noticed working on #142972 where we were using v4f32 blends instead of v8i16 resulting in a lot of domain crossing.

Its also helps avoid unnecessary use of VINSERTPS nodes which can be tricky to commute or concatenate back to 256-bit vectors.



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