[PATCH] D30225: [LIR] re-enable generation of memmove with runtime checks

[Eli Friedman via Phabricator via llvm-commits]

efriedma added a comment.

I guess we can just generate memmove for the first iteration of this; we'll see if any testcases come up where it would help to generate memcpy instead.

In terms of where to run this... not sure there's really any existing pass that's appropriate.  You want to run after inlining, but before the loop vectorizer.  Maybe a new pass just before the vectorizer; call it loop-versioning-loop-idiom or something like that.


Repository:
  rL LLVM

https://reviews.llvm.org/D30225