projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: d04ec0c) | patch
Fix a regression from r147481.
authorChad Rosier <mcrosier@apple.com>
Fri, 9 Mar 2012 02:00:48 +0000 (02:00 +0000)
committerChad Rosier <mcrosier@apple.com>
Fri, 9 Mar 2012 02:00:48 +0000 (02:00 +0000)
commitabd6674166d99a8da98269051fbd869d9d8c3ba1
tree4b68f0ac12b9221bc2931e98600677d9a545ca4atree | snapshot
parentd04ec0c855176ebddd459c044bdd24f49938fae4commit | diff
Fix a regression from r147481.

Original commit message from r147481:
DAGCombine for transforming 128->256 casts into a vmovaps, rather
then a vxorps + vinsertf128 pair if the original vector came from a load.

Fix:
Unaligned loads need to generate a vmovups.
rdar://10974078

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152366 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/X86/X86InstrFragmentsSIMD.td diff | blob | history
lib/Target/X86/X86InstrSSE.td diff | blob | history
test/CodeGen/X86/avx-shuffle.ll diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors