projects / oota-llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 68f32cb) | patch
Don't do the X * 0.0 -> 0.0 transformation in instcombine, because
authorDan Gohman <gohman@apple.com>
Thu, 4 Jun 2009 17:12:12 +0000 (17:12 +0000)
committerDan Gohman <gohman@apple.com>
Thu, 4 Jun 2009 17:12:12 +0000 (17:12 +0000)
commit77b81fe487de709d41dfcc97bbb2941ae1bb80af
tree9730d59d8f6ae8de964c38fd56884553c77db692tree | snapshot
parent68f32cbb1f9e62d5e6047b048c0d7d217b8717e1commit | diff
Don't do the X * 0.0 -> 0.0 transformation in instcombine, because
instcombine doesn't know when it's safe. To partially compensate
for this, introduce new code to do this transformation in
dagcombine, which can use UnsafeFPMath.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@72872 91177308-0d34-0410-b5e6-96231b3b80d8
lib/CodeGen/SelectionDAG/DAGCombiner.cpp diff | blob | history
lib/CodeGen/SelectionDAG/SelectionDAG.cpp diff | blob | history
lib/Transforms/Scalar/InstructionCombining.cpp diff | blob | history
C/C++ LLVM-based compilers that forbids OOTA behaviors