[Diffusion] rG5419b671375c: [SimplifyCFG] Update FoldTwoEntryPHINode to handle and/or of select and binop…

[Nikita Popov via Phabricator via llvm-commits]

nikic added a comment.

Well, the InstCombine canonicalization is there so that other code doesn't have to care about it. It just needs to not explicitly break the canonical pattern. Of course, there may be some phase ordering consideration that makes it worthwhile to consider the non-canonical form (SimplifyCFG would be a prime place where this may be true), but it should not be a general problem.

As the test cases from rG7ef6f019090f3979fa345105b9ac95ac589c6cf9 <https://reviews.llvm.org/rG7ef6f019090f3979fa345105b9ac95ac589c6cf9> are in non-canonical form and it's not clear where this form is coming from, I don't see anything actionable here.


BRANCHES
  llvm_jit_comment, main, ml-spread-test

Users:
  aqjune (Author)

https://reviews.llvm.org/rG5419b671375c