[PATCH] Enable unrolling of multi-exit loops

[Andrew Trick]

Sorry for the delay reviewing--I had to be careful with this one, and the diff is unclear. It does look like great improvement and it's about time to move on to full-fledge support for multi-exit loops.

With this change, ComputeExitLimit design loses some information. We cannot as precisely determine the trip count of a particular exit given knowledge of NSW. But I see now that information was probably not very useful. The idea of conditionally considering NW/NSW based on the loop structure makes the code simpler and the API safer. As a result, the loop unroller can actually be more aggressive without adding any complexity. So I think this is a great change.

One thing I'm concerned about though is that I think you're using SCEVDivision for the first time in the standard LLVM pipeline (outside of the delinearizer). SCEVDivision looks like it recurses over the expression operands. Every time someone does this, we eventually have to track down some case that results in pathological compile time. Can you prove that this recursion will never visit the same expression twice? If not, I think you need to add a visited set as we do elsewhere, or find another way to check for exact division.

http://reviews.llvm.org/D5550