[PATCH] D120337: [InstCombine] Remove one-use limitation from X-Y==0 fold

[Nikita Popov via Phabricator via llvm-commits]

nikic added a comment.

I just looked into the code-size regression for tramp3d-v4, and what happens is that we're performing additional (desirable) jump-threading in many places, and in particular in @_ZN6EngineILi3Ed10MultiPatchI7GridTag6RemoteI5BrickEEED2Ev a redundant branch is eliminated entirely and the function shrinks by just enough to be eligible for inlining at hot call sites -- of which there are a lot, or at least LLVM thinks so.

As such, at least for the tramp3d-v4 case, it looks like this change is having a positive effect and just ends up crossing an inlining threshold.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D120337/new/

https://reviews.llvm.org/D120337