[PATCH] D62531: SpeculateAroundPHIs: Respect convergent

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In D62531#1519636 <https://reviews.llvm.org/D62531#1519636>, @jdoerfert wrote:

> I don't think we need the AMDGPU test case specialization here. A generic test with a convergent and one with a non-duplicatable call would make more sense to me.


It concludes an arbitrary call is too expensive, so the problem doesn't manifest. I didn't see any control to override the cost


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D62531/new/

https://reviews.llvm.org/D62531