[PATCH] D49499: [X86] Prefer unpckhpd over movhlps in isel for fake unary cases

[Craig Topper via Phabricator via llvm-commits]

craig.topper added inline comments.


================
Comment at: lib/Target/X86/X86InstrSSE.td:831
                       Sched<[SchedWriteFShuffle.XMM]>;
   let isCommutable = 1 in
   def MOVHLPSrr : PSI<0x12, MRMSrcReg, (outs VR128:$dst),
----------------
RKSimon wrote:
> Shouldn't it only be commutable if it has SSE2? Looking at X86InstrInfo::commuteInstructionImpl it asserts that SSE2 is available, it doesn't early out....
It's checked in findCommutedOpIndices which should be called before commuteInstructionImpl.


https://reviews.llvm.org/D49499