Allow overflow of indices with constant dim-sizes.
authorMichael Kruse <llvm@meinersbur.de>
Mon, 11 Apr 2016 14:34:08 +0000 (14:34 +0000)
committerMichael Kruse <llvm@meinersbur.de>
Mon, 11 Apr 2016 14:34:08 +0000 (14:34 +0000)
commit3b425ff232d69ab7326c8d0d43d5485a15e7911b
treeb1f8ed3b9f5525ae5d7442331909d47ebcafa4a5
parent2d5487cf446f9a58f6d2f76df33102423c83f285
Allow overflow of indices with constant dim-sizes.

Allow overflow of indices into the next higher dimension if it has
constant size. E.g.

    float A[32][2];
    ((float*)A)[5];

is effectively the same as

    A[2][1];

This can happen since r265379 as a side effect if ScopDetection
recognizes an access as affine, but ScopInfo rejects the GetElementPtr.

Differential Revision: http://reviews.llvm.org/D18878

llvm-svn: 265942
polly/include/polly/ScopInfo.h
polly/lib/Analysis/ScopInfo.cpp
polly/test/ScopInfo/multidim_fixedsize_multi_offset.ll
polly/test/ScopInfo/process_added_dimensions.ll