[PATCH] D58470: [X86] Reduce some patterns by using FP instructions for integer types even when AVX2 is available and execution domain fixing will do the right thing

[Craig Topper via Phabricator via llvm-commits]

craig.topper added a comment.

I’ve never understood if the peephole pass is as powerful at folding loads as SelectionDAG. I added the extra patterns to make sure AVX1 and 2 were being handled in a consistent way.

I do know that the execution domain fixing pass is going to make a domain decision for these instructions regardless of what instruction we started with. Though I guess it doesn’t run in O0 so maybe we should pick the best domain in isel?


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D58470/new/

https://reviews.llvm.org/D58470