[PATCH] D20789: Consecutive memory access in Loop Vectorizer - fixed and simplified

Adam Nemet via llvm-commits llvm-commits at lists.llvm.org
Mon Jun 13 06:45:50 PDT 2016

anemet added a comment.

In http://reviews.llvm.org/D20789#456052, @delena wrote:

> In http://reviews.llvm.org/D20789#456016, @anemet wrote:
> > In http://reviews.llvm.org/D20789#455936, @delena wrote:
> >
> > > I changed interface of getPtrStride(). Instead of one "Assume" parameter which means "additional run-time assumptions", I put 2 parameters - "Assume memory access versioning in run-time" and "Assume no-wrap in address calculation".
> > >  isConsecutivePtr() sets to "true" only the first assumption.
> >
> >
> > Why?  What's wrong with calling getPtrStirde with Assume=true?
> 2 tests were failing. "Assume-versioning"=true is ok for isConsecutivePtr(), but Assume-no-wrap should be "false".

Why were the tests failing?  You need to provide some explanation rather than just communicate the conclusion of your investigation.

As far as I can see both of these just add run-time checks from things that cannot be proven at compile time.



More information about the llvm-commits mailing list