[PATCH] D64128: [CodeGen] Generate llvm.ptrmask instead of inttoptr(and(ptrtoint, C)) if possible.

Wed Jul 3 16:47:03 PDT 2019

hfinkel added a comment.

In D64128#1568857 <https://reviews.llvm.org/D64128#1568857>, @efriedma wrote:

> I don't think this transform is valid, for the same reasons we don't do it in IR optimizations.

I believe that in the problematic cases we previously discussed (e.g., from https://reviews.llvm.org/D59065#1449682), they all depend on some control dependency being introduced somewhere in between the initial pointer casts and the other operations. If they're all syntactically together like this, maybe that's safe?

Does this actually catch the ABI code that motivated this in the first place? Isn't that in lib/CodeGen/TargetInfo.cpp (e.g., in emitRoundPointerUpToAlignment)?

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D64128/new/

https://reviews.llvm.org/D64128