[PATCH] D48809: [FPEnv] Split double width StrictFP vector operations as needed

[Cameron McInally via Phabricator via llvm-commits]

cameron.mcinally created this revision.
cameron.mcinally added reviewers: andrew.w.kaylor, craig.topper, uweigand, hfinkel.
Herald added a subscriber: llvm-commits.

Here's a little patch to split illegal StrictFP vector operations. It's very much like the code already committed to expand  StrictFP vector operations.

Some notes:

1. There's one line that I'm not too happy about. I'll add a comment inline to elucidate it.

2. The double width vector math lib calls and rounding calls are not split on X86, but rather expanded. The newly added tests for those may be overkill. Thoughts on this?

3. It may make sense to break out the StrictFP FMA operations into their own test file. We're testing some of these non-FMA operations twice for NO-FMA and HAS-FMA. That said, the generated assembly is slightly different when FMAs are enabled, so there may be some value in keeping them as-is.


Repository:
  rL LLVM

https://reviews.llvm.org/D48809

Files:
  lib/CodeGen/SelectionDAG/LegalizeTypes.h
  lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
  test/CodeGen/X86/vector-constrained-fp-intrinsics.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D48809.153650.patch
Type: text/x-patch
Size: 60709 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20180701/51ecece7/attachment.bin>