BBVectorize: Cache fixed-order pairs instead of recomputing pointer info.
authorHal Finkel <hfinkel@anl.gov>
Tue, 30 Oct 2012 20:17:37 +0000 (20:17 +0000)
committerHal Finkel <hfinkel@anl.gov>
Tue, 30 Oct 2012 20:17:37 +0000 (20:17 +0000)
commita9779bfbc9ab0cf3f157453fd0afd110b04a9fdc
treeb25f11b8fbacb85aa56a28ebbd3452b71b982dbe
parent2f34d754d00fbe2e4a98762d71d0fae5f4b0cf45
BBVectorize: Cache fixed-order pairs instead of recomputing pointer info.

Instead of recomputing relative pointer information just prior to fusing,
cache this information (which also needs to be computed during the
candidate-pair selection process). This cuts down on the total number of
SE queries made, and also is a necessary intermediate step on the road toward
including shuffle costs in the pair selection procedure.

No functionality change is intended.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@167049 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Transforms/Vectorize/BBVectorize.cpp