Fix an issue where fast math flags were dropped during scalarization.
authorOwen Anderson <resistor@mac.com>
Tue, 1 Mar 2016 19:35:52 +0000 (19:35 +0000)
committerOwen Anderson <resistor@mac.com>
Tue, 1 Mar 2016 19:35:52 +0000 (19:35 +0000)
commit7ea02fc7879ccfa9aba47c0dec7d57e8fbeb3b9e
treeb70ac649bf2480bdf65e8c49174eed1bd09fae7c
parentb3a7935d549c1d73729a4666ec6133e80658ce34
Fix an issue where fast math flags were dropped during scalarization.

Most portions of InstCombine properly propagate fast math flags, but
apparently the vector scalarization section was overlooked.

llvm-svn: 262376
llvm/include/llvm/IR/InstrTypes.h
llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
llvm/test/Transforms/InstCombine/fast-math-scalarization.ll [new file with mode: 0644]