projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 82a827d) | patch
[x86] don't use a shuffle when a vselect will do; NFCI
authorSanjay Patel <spatel@rotateright.com>
Thu, 10 Mar 2016 22:35:33 +0000 (22:35 +0000)
committerSanjay Patel <spatel@rotateright.com>
Thu, 10 Mar 2016 22:35:33 +0000 (22:35 +0000)
commit0181943b89e156f85b77e7e97a2a467ebf0fb736
tree8a9515318cd2e943f999628340516f1c5af3d9cbtree | snapshot
parent82a827de3ba5a907d72508758269eb075f533322commit | diff
[x86] don't use a shuffle when a vselect will do; NFCI

Looking at the IR definition of a masked load made me realize
there was no reason to use a shuffle here, so we don't need
to convert the format of the mask at all.

llvm-svn: 263167
llvm/lib/Target/X86/X86ISelLowering.cpp diff | blob | history
Domain: System / Toolchain;