[PATCH] D155843: [Analysis] Analysis of storing unchanged loaded value

[Nikita Popov via Phabricator via llvm-commits]

nikic added a comment.

> InstCombine is responsible for these optimization.

Not really. The pass responsible for this is GVN, via MemDepAnalysis and to a lesser degree EarlyCSE via MSSA. InstCombine implements a very weak version of load CSE optimization for phase ordering reasons. I'm skeptical about making InstCombine handle patterns that GVN cannot handle, because the optimization in InstCombine is fundamentally limited to a tiny instruction window.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D155843/new/

https://reviews.llvm.org/D155843