[llvm] [SROA] Unfold gep of index phi (round 2) (PR #83494)

[Artem Belevich via llvm-commits]

Artem-B wrote:

> It doesn't really matter where the GEP is, later optimizations will place them wherever optimal.

It does matter in practice. Later optimization pass may *eventually* place them wherever optimal, but that assumes there will be enough of those "later" passes. If we happened to be in SROA pass at the tail end of the compilation, that may not be the case. 

Anyways, I'm OK with the keep-it-simple approach.

https://github.com/llvm/llvm-project/pull/83494