[PATCH] D26066: [PowerPC] Improvements for BUILD_VECTOR Vol. 4

[Nemanja Ivanovic via llvm-commits]

nemanjai updated this revision to Diff 78676.
nemanjai added a comment.

Updated the code in the PeepHole optimization to ensure that an attempt to remove the same instruction twice cannot occur.
Looking a little deeper into the relationship between FRSP and XVCVDPSP, I realized that it is safe to remove an FRSP that feeds an XVCVDPSP even if the FRSP only produced one half of the input vector. This is because the two operations are equivalent (other than one being scalar and the other one a vector operation) and they do not modify a single precision input.


Repository:
  rL LLVM

https://reviews.llvm.org/D26066

Files:
  lib/Target/PowerPC/PPCMIPeephole.cpp
  test/CodeGen/PowerPC/build-vector-tests.ll
  test/CodeGen/PowerPC/power9-moves-and-splats.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D26066.78676.patch
Type: text/x-patch
Size: 190852 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20161121/d9156406/attachment-0001.bin>