Fix a bug in the code that checks if we can vectorize loops while using dynamic
authorNadav Rotem <nrotem@apple.com>
Fri, 21 Dec 2012 00:07:35 +0000 (00:07 +0000)
committerNadav Rotem <nrotem@apple.com>
Fri, 21 Dec 2012 00:07:35 +0000 (00:07 +0000)
commit55306bdea5d2d53be39f3ac59fadf5220ee6b5d0
treef8f8d9f0290bd9e8131d1669ca290f86bac18509
parent71a9c2137b590467fabfcde1ef9f87bcb8059c92
Fix a bug in the code that checks if we can vectorize loops while using dynamic
memory bound checks.  Before the fix we were able to vectorize this loop from
the Livermore Loops benchmark:

for ( k=1 ; k<n ; k++ )
  x[k] = x[k-1] + y[k];

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@170811 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Transforms/Vectorize/LoopVectorize.cpp
test/Transforms/LoopVectorize/same-base-access.ll [new file with mode: 0644]
test/Transforms/LoopVectorize/scalar-store.ll [deleted file]