[PATCH] D145008: [ControlHeightReduction] Don't combine a "poison" branch

Wed Mar 1 14:13:09 PST 2023

nikic added a comment.

In D145008#4162887 <https://reviews.llvm.org/D145008#4162887>, @kazu wrote:

> OK.  Now, what if `Cond` is guaranteed to be a `poison`?  Should we still transform the code?

I'd say "Yes", in the sense that we should not go out of our way to handle this special case. We should only make sure that the transform is correct.

A branch on poison should really not reach this code and get optimized away to unreachable earlier -- the reason it isn't is that this would take some annoying test updates, see D140892 <https://reviews.llvm.org/D140892>.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D145008/new/

https://reviews.llvm.org/D145008