[PATCH] D106450: [InstCombine] Fold (gep (oneuse(gep Ptr, Idx0)), Idx1) -> (gep Ptr, (add Idx0, Idx1)) (PR51069)

[Whitney Tsang via Phabricator via llvm-commits]

Whitney added a comment.

In D106450#2916523 <https://reviews.llvm.org/D106450#2916523>, @RKSimon wrote:

> @Whitney we need the end-to-end test case to have a better understanding of where your original IR is appearing and why cse isn't solving this - running it through -O3  seems to be fine: https://c.godbolt.org/z/jWYcra73x (vs just -instcombine) which suggests a phase ordering issue is at play

We are working on another reduced LLVM IR test case that shows the problem with -O3. Unfortunately the important benchmark is huge and we are not allowed to share.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D106450/new/

https://reviews.llvm.org/D106450