[PATCH] D32953: [X86][SSE] Improve combineLogicBlendIntoPBLENDV to use general masks.

[Simon Pilgrim via Phabricator via llvm-commits]

RKSimon added inline comments.


================
Comment at: lib/Target/X86/X86ISelLowering.cpp:31566
 
   if (!((VT == MVT::v2i64) || (VT == MVT::v4i64 && Subtarget.hasInt256())))
     return SDValue();
----------------
delena wrote:
> Why VT should be v2i64 of v4i64? Isn't the same transformation profitable for other integer types?
It's due to us promoting to v2i64/v4i64 for X86ISD::ANDNP - it's not actually necessary but seems to have been used as an early out. 

I'll remove it - it won't make any difference to the codegen now but will make it easier to perform more thorough ANDNP matching mentioned in the TODO.


Repository:
  rL LLVM

https://reviews.llvm.org/D32953