projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 4581dae) | patch
Second attempt at fixing the performance regressions introduced
authorOwen Anderson <resistor@mac.com>
Sat, 27 Nov 2010 08:15:55 +0000 (08:15 +0000)
committerOwen Anderson <resistor@mac.com>
Sat, 27 Nov 2010 08:15:55 +0000 (08:15 +0000)
commit35bf4d6d8018160557a92b86181acbcef76f86eb
treefc85169448d2343bf3c6e1931eda1227994d1949tree | snapshot
parent4581dae9ea6d21da0a584aad9f1143343bb7c32ecommit | diff
Second attempt at fixing the performance regressions introduced
by my recent GVN improvement.  Looking through a single layer of
PHI nodes when attempting to sink GEPs, we need to iteratively
look through arbitrary PHI nests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@120202 91177308-0d34-0410-b5e6-96231b3b80d8
include/llvm/Transforms/Utils/AddrModeMatcher.h diff | blob | history
lib/Transforms/Scalar/CodeGenPrepare.cpp diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors