pass element size to delinearization
authorSebastian Pop <spop@codeaurora.org>
Tue, 27 May 2014 22:42:09 +0000 (22:42 +0000)
committerSebastian Pop <spop@codeaurora.org>
Tue, 27 May 2014 22:42:09 +0000 (22:42 +0000)
commitbc9009a20a7b1ffbf86c8a9f880efd068024e59a
treef972709389e0ce8595dba1e9c81be92396beb7a9
parentf93ef123300cf943980fef09cee723cf49c96017
pass element size to delinearization

Instead of relying on the delinearization to infer the size of an element,
compute the element size from the base address type. This is a much more precise
way of computing the element size than before, as we would have mixed together
the size of an element with the strides of the innermost dimension.

llvm-svn: 209695
polly/include/polly/ScopDetection.h
polly/lib/Analysis/ScopDetection.cpp
polly/lib/Analysis/TempScopInfo.cpp