[PATCH] D85759: [SLPVectorizer] Fix regression from cost model refactoring

[Sam Parker via Phabricator via llvm-commits]

samparker added a subscriber: sanwou01.
samparker added a comment.

There were at least two main paths, through different APIs, and then the logic diverged depending on whether arguments, or just their types were passed - and I have no good idea why. But looking at my original patch again, I also don't know why I wouldn't have just done what is proposed here... I made almost the same change in getVectorCallCosts too, so it might be worth looking there too. I know what @sanwou01 saw an important uplift from this change though, so it would be nice if we could maintain that (not sure if he committed a test) and fix this regression.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D85759/new/

https://reviews.llvm.org/D85759