projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 7336f7f) | patch
On Sandybridge loading unaligned 256bits using two XMM loads (vmovups and vinsertf128...
authorNadav Rotem <nrotem@apple.com>
Fri, 18 Jan 2013 23:10:30 +0000 (23:10 +0000)
committerNadav Rotem <nrotem@apple.com>
Fri, 18 Jan 2013 23:10:30 +0000 (23:10 +0000)
commit48177ac90fb940833b9deea1a6716092348cfe82
tree5252f0617e256f0dd1f8b26082f05a088d7232b9tree | snapshot
parent7336f7febb5170b374a4cbffee273ad82ff8a1a3commit | diff
On Sandybridge loading unaligned 256bits using two XMM loads (vmovups and vinsertf128) is faster than using a single vmovups instruction.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@172868 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/X86/X86ISelLowering.cpp diff | blob | history
test/CodeGen/X86/sandybridge-loads.ll [new file with mode: 0644] blob
test/CodeGen/X86/v8i1-masks.ll diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors