[PATCH] D58574: [X86][SSE] Generalize X86ISD::BLENDI support to more value types with fix for revert from r354713

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added a comment.

In D58574#1408061 <https://reviews.llvm.org/D58574#1408061>, @craig.topper wrote:

> I will fix the commuting code too.
>
> But doing only that means we’ll turn v4i32 blends into movss/movsd sometimes on pre-avx2 targets. For v2i64 it can happen on avx2 targets as well.


Good point - I imagine even for OptSize we typically want to avoid the domain swap.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D58574/new/

https://reviews.llvm.org/D58574