LoopVectorizer: When we vectorizer and widen loops we process many elements at once...
authorNadav Rotem <nrotem@apple.com>
Mon, 7 Jan 2013 21:54:51 +0000 (21:54 +0000)
committerNadav Rotem <nrotem@apple.com>
Mon, 7 Jan 2013 21:54:51 +0000 (21:54 +0000)
commit9a6c6a373629fb5a3cc5afd08aafa51339df95df
tree1cbd326c4bbe4297f2e8f43f828f7e6acad6ee45
parent6c1d4972cf1cd6b6072e31c05f97abb1ed7a8497
LoopVectorizer: When we vectorizer and widen loops we process many elements at once. This is a good thing, except for
small loops. On small loops post-loop that handles scalars (and runs slower) can take more time to execute than the
rest of the loop. This patch disables widening of loops with a small static trip count.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171798 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Transforms/Vectorize/LoopVectorize.cpp
test/Transforms/LoopVectorize/X86/unroll-small-loops.ll [new file with mode: 0644]