[PATCH] D41096: [X86] Initial support for prefer-vector-width function attribute

[Craig Topper via Phabricator via llvm-commits]

craig.topper added a comment.

Yeah we should be able to add another subtarget feature and process it in getSubtargetImpl as well.

I don't think we'll want to enable it by default on any Intel CPUs though since the issue there isn't as severe. So hopefully 256 is still a win in many workloads. And I don't know if I can sign up to do the work to implement the X86 isel legalization support that it would require to fully enforce it.


https://reviews.llvm.org/D41096