[PATCH] D105390: [X86] Lower insertions into upper half of an 256-bit vector as broadcast+blend (PR50971)

[Roman Lebedev via Phabricator via llvm-commits]

lebedev.ri updated this revision to Diff 362063.
lebedev.ri added a comment.

@RKSimon thank you for taking a look!
Thinking about it more, i am not okay with following https://reviews.llvm.org/D105390#inline-1016171
`Maybe generalize the IdxVal >= NumEltsIn128 limit to insert with broadcast if the scalar is already used in a (a) another insertelement/psinrw/pinsrb, (b) scalar_to_vector or (c) broadcast.` suggestion.
While that clearly results in improvements mostly, not all additional changes are wins per-se.
This is counter-productive. The relaxation is clearly separable from the existing diff.
I see no reason why it must be done at once.

I'm interested in looking into that later, or not proceeding with this patch at all.
Thanks.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D105390/new/

https://reviews.llvm.org/D105390

Files:
  llvm/lib/Target/X86/X86ISelLowering.cpp
  llvm/test/CodeGen/X86/avx-insertelt.ll
  llvm/test/CodeGen/X86/avx2-masked-gather.ll
  llvm/test/CodeGen/X86/avx512-insert-extract.ll
  llvm/test/CodeGen/X86/avx512-masked-memop-64-32.ll
  llvm/test/CodeGen/X86/insertelement-shuffle.ll
  llvm/test/CodeGen/X86/masked_expandload.ll
  llvm/test/CodeGen/X86/masked_gather.ll
  llvm/test/CodeGen/X86/masked_gather_scatter.ll
  llvm/test/CodeGen/X86/masked_load.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D105390.362063.patch
Type: text/x-patch
Size: 102631 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20210727/b434dbb4/attachment.bin>