[PATCH] D156702: Add support for GEP of GEP instruction combining when types do not exactly match

[Nikita Popov via Phabricator via llvm-commits]

nikic added a comment.

Can you please explain what the larger context/motivation for this is?

Longer term at least, I think we would be better off going in the opposite direction and making sure that each GEP has at most one variable index. For example, if you have two GEPs with variable indices and one of them is loop invariant, LICM will be able to move it out of the loop (possibly after reassociating the two GEPs). This is not possible if both indices are in the same GEP.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D156702/new/

https://reviews.llvm.org/D156702