[llvm] [InstCombine] Push freeze through non-recurrence PHIs (PR #157678)

[Nikita Popov via llvm-commits]

nikic wrote:

I think the problem here is similar to what I ran into with the original change in https://github.com/llvm/llvm-project/pull/154336, just for phi nodes. For cyclic phis, we only push the freeze to that phi, and then it will get pushed through in a separate transform. This kind of incrementally pushing freeze (instead of freezing the whole graph) ends up being slow in some cases, though I can't say I fully understand what causes it in this specific case.

Probably if we want to push through freeze we need to integrate the recurrence and non-recurrence cases...

https://github.com/llvm/llvm-project/pull/157678