[PATCH] D17899: [x86, AVX] optimize masked loads with constant masks

[Sanjay Patel via llvm-commits]

spatel created this revision.
spatel added reviewers: RKSimon, delena, ab.
spatel added a subscriber: llvm-commits.
Herald added a subscriber: mcrosier.

Instead of a variable-blend instruction, form a blend with immediate because those are always cheaper.

Note the FIXME for AVX512: I saw that masked loads were followed by blends after this change but there is currently no blend at all. I assume that's because the blend is part of the AVX512 masked load itself? I don't know enough about AVX512 to be sure how to solve it, so I've just enabled this for AVX1/AVX2 for now.

http://reviews.llvm.org/D17899

Files:
  lib/Target/X86/X86ISelLowering.cpp
  test/CodeGen/X86/masked_memop.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D17899.49859.patch
Type: text/x-patch
Size: 7579 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160304/ec44ef1d/attachment-0001.bin>