[all-commits] [llvm/llvm-project] df170b: [X86] X86FixupInstTuning - fold VPERM2x128 -> VINS...

[Simon Pilgrim via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: df170b20e9e0bbec29f69a51c0ba6e68ecabb160
      https://github.com/llvm/llvm-project/commit/df170b20e9e0bbec29f69a51c0ba6e68ecabb160
  Author: Simon Pilgrim <llvm-dev at redking.me.uk>
  Date:   2026-06-05 (Fri, 05 Jun 2026)

  Changed paths:
    M llvm/lib/Target/X86/X86FixupInstTuning.cpp
    M llvm/test/CodeGen/X86/slow-pmulld.ll
    M llvm/test/CodeGen/X86/vector-interleave.ll
    M llvm/test/CodeGen/X86/vector-interleaved-load-i16-stride-6.ll
    M llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-5.ll
    M llvm/test/CodeGen/X86/vector-interleaved-load-i32-stride-6.ll
    M llvm/test/CodeGen/X86/vector-interleaved-load-i64-stride-4.ll
    M llvm/test/CodeGen/X86/vector-interleaved-load-i8-stride-6.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i16-stride-2.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i32-stride-2.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i64-stride-4.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i64-stride-6.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i64-stride-7.ll
    M llvm/test/CodeGen/X86/vector-interleaved-store-i8-stride-2.ll
    M llvm/test/CodeGen/X86/vector-shuffle-256-v32.ll
    M llvm/test/CodeGen/X86/vector-shuffle-combining-avx.ll
    M llvm/test/CodeGen/X86/x86-interleaved-access.ll

  Log Message:
  -----------
  [X86] X86FixupInstTuning - fold VPERM2x128 -> VINSERTx128 when shuffling lower xmm half ymm sources (#201618)

VINSERTx128 is never slower than VPERM2x128 and notably quicker on some
targets (btver2, znver1, e-cores, etc.).

Shuffle lowering avoids some VINSERT patterns for AVX targets as it can
affect folding/commutation - but by the time we get to the fixup passes,
these are all done and we can safely convert to VINSERTF128/I128.

There's more variants of the VPERM2 immediate mask that could be folded,
but its incredibly difficult to hit them as its easily commutable.

I hit this while working on #199445.



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