[PATCH] D45945: [LoopRotate] Fix incorrect SCEV invalidation in loop rotation

[Chandler Carruth via Phabricator via llvm-commits]

chandlerc added a comment.

In https://reviews.llvm.org/D45945#1075073, @mkazantsev wrote:

> We've fixed a bunch of bugs of this sort, so I think it's fine to go with.


So generally, as a matter of politeness, good to avoid the review if you're gonna go with post-submit.

However I also have a question here -- are we going to have any SCEV left?

We don't re-run SCEV mid-way through the loop pass manager. If we invalidate all outer scevs when we loop rotate, won't that completely defeat SCEVs analysis on outer loops? Or will we lazily rebuild it?


Repository:
  rL LLVM

https://reviews.llvm.org/D45945