[PATCH] D134606: [LAA] Change to function analysis for new PM.

[Arthur Eubanks via Phabricator via llvm-commits]

aeubanks added a comment.

I looked at the repro in #50940 to see why analyses weren't getting invalidated. I think something like https://github.com/llvm/llvm-project/commit/f4c0810a79776a63fd2ba57be0199a31e35d5ad7 is more appropriate. We should be invalidating all loop analyses after a loop adaptor is finished, if any loop pass made a change. That fixes https://github.com/llvm/llvm-project/issues/50940. Thoughts on that?

Not saying that making LAA a function analysis isn't useful, maybe we want to do that anyway like you said.

And https://reviews.llvm.org/D134611 is a different issue of not updating analyses within a pass, which could be solved either with your patch, manually calling the analysis manager to invalidate, or keeping analyses up to date within the pass when making a change (maybe this isn't possible with LAA).


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D134606/new/

https://reviews.llvm.org/D134606