[PATCH] D99118: [SROA] Speculate around PHIs: allow load to be in a unique-successor unique-predecessor block

Mon Mar 22 15:14:34 PDT 2021

nikic added a comment.

If there is only one successor, and the one successor has only one predecessor, wouldn't these blocks be merged by SimplifyCFG? I'm not sure 2009-02-20-InstCombine-SROA.ll is really testing something meaningful because the specific `-instcombine -sroa` pass combination shouldn't occur in the wild. The primary SROA passes run before InstCombine, and the secondary post-unroll pass has so many other passes scheduled in between that reasoning about this in terms of just `-instcombine -sroa` is not very helpful.

I'm not really opposed to this change, but I'm also not sure it's necessary. Maybe in this case simply deleting the test case is the right thing to do?

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D99118/new/

https://reviews.llvm.org/D99118