projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 2c693dc) | patch
Synthesize x86 max/min instructions also for vectors (i.e. produce
authorDuncan Sands <baldrick@free.fr>
Sat, 17 Sep 2011 16:49:39 +0000 (16:49 +0000)
committerDuncan Sands <baldrick@free.fr>
Sat, 17 Sep 2011 16:49:39 +0000 (16:49 +0000)
commit6bcd2196e5febb97efe188d0d6de403c577d96fc
treeb147a99b1f7f3b65365ba8535f8466ab53f5d766tree | snapshot
parent2c693dc1261d8b8d92b5c96cf00ec873dde86bb2commit | diff
Synthesize x86 max/min instructions also for vectors (i.e. produce
maxps and maxpd).  This broke the sse41-blend.ll testcase by causing
maxpd to be produced rather than a cmp+blend pair, which is the reason
I tweaked it.  Gives a small speedup on doduc with dragonegg when the
GCC vectorizer is used.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139986 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/X86/X86ISelLowering.cpp diff | blob | history
test/CodeGen/X86/sse-minmax.ll diff | blob | history
test/CodeGen/X86/sse41-blend.ll diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors