[PATCH] D39340: Modifying reassociate for improved CSE

[Florian Hahn via Phabricator via llvm-commits]

fhahn added a reviewer: fhahn.
fhahn added a comment.

In https://reviews.llvm.org/D39340#909660, @escha wrote:

> Yeah, actually, I don't quite understand what N-ary reassociate *is* compared to this. One thing I do know about it is that it uses SCEV, which completely rules it out for floating point, which is the main thing I care about. Does it handle the integer case here?


It looks like NaryReassoicate tries to do the same thing as this patch, although using a different approach. I think ideally we would not have code in 2 passes that try to do the same thing.

If this patch handles all cases handled by NaryReassoicate AND floating point operations, it might make sense to get rid of NaryReassociate. Could you try and check if this patch handles all the NaryReassoicate test cases?


Repository:
  rL LLVM

https://reviews.llvm.org/D39340