[PATCH] D25222: [AVX-512] Use AVX512 feature instead of VLX to determine whether to use extended 128/256-bit register classes for addRegisterClass.

[Craig Topper via llvm-commits]

craig.topper updated this revision to Diff 73880.
craig.topper added a comment.

Here's a new version with a test case that shows using the extended register with vpmovdb.

While doing this I discovered that the current code for getLargestLegalSuperClass has a bug in it that was responsible for xmm16 being used and spilled without VLX enabled. I considered fixing that bug separately from this, but figured that was unnecessary churn.


https://reviews.llvm.org/D25222

Files:
  lib/Target/X86/X86ISelLowering.cpp
  lib/Target/X86/X86RegisterInfo.cpp
  test/CodeGen/X86/vector-half-conversions.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D25222.73880.patch
Type: text/x-patch
Size: 12302 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20161007/0a663db0/attachment.bin>