[PATCH] D25652: [AVX-512] Add support for commuting VPERMT2(B/W/D/Q/PS/PD) to/from VPERMI2(B/W/D/Q/PS/PD).

[Craig Topper via llvm-commits]

craig.topper updated this revision to Diff 75391.
craig.topper added a comment.

Turns out we weren't properly commuting masked instructions due to findCommutedOpIndices miscalculating for these instructions.

I've also added an additional new test(avx512-vpermv3-commute.ll) to cover commuting these instructions explicitly. Version without commuting with extra moves was commited in r284808. So this patch just shows the moves being removed due to commuting.


https://reviews.llvm.org/D25652

Files:
  lib/Target/X86/X86InstrAVX512.td
  lib/Target/X86/X86InstrInfo.cpp
  test/CodeGen/X86/avx512-insert-extract.ll
  test/CodeGen/X86/avx512-mask-op.ll
  test/CodeGen/X86/avx512-vpermv3-commute.ll
  test/CodeGen/X86/avx512vbmi-intrinsics.ll
  test/CodeGen/X86/avx512vbmivl-intrinsics.ll
  test/CodeGen/X86/merge-consecutive-loads-512.ll
  test/CodeGen/X86/vector-shuffle-512-v8.ll
  test/CodeGen/X86/vector-shuffle-combining-avx512bw.ll
  test/CodeGen/X86/vector-shuffle-v1.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D25652.75391.patch
Type: text/x-patch
Size: 55784 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20161021/368d7242/attachment.bin>