[PATCH] D46946: [X86] Lowering shift intrinsics to native IR

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

We already do a lot in InstCombineCalls to convert x86 vector shift calls to generic shifts (mainly for constants, valuetracking might allow us to do a bit more?), and as others have mentioned, splitting IR like this is likely to cause us a lot of problems.

It could be useful to allow the backend to remove out of bounds shift masks where the instruction supports implicit zeroing (e.g. to avoid sanitizer warnings: https://gcc.godbolt.org/z/Abcq8n)


Repository:
  rL LLVM

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D46946/new/

https://reviews.llvm.org/D46946