[PATCH] D63336: [CodeGen] Check for HardwareLoop Latch ExitBlock

[Dave Green via Phabricator via llvm-commits]

dmgreen added a comment.

Am I correct in saying that this loop will find the first exiting block that dominates all the latches in the loop (with some other conditions like a loop invariant non zero exit count). Plus is now also a latch?

But the loop may have other exits and other latches?

The "other exits" sounds like it should be OK for arm low overhead loops. At least I can't think of a reason right now why they are not OK. For vector tail predicated loops I imagine that you would need to ensure that a "LCTP" is placed on the other exits, to ensure the tail predicate doesn't pollute extra instructions after the loop.

I'm not sure about the "other latches" part. Is that something that's OK? I imagine it doesn't come up very often.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D63336/new/

https://reviews.llvm.org/D63336