[llvm] X86: make VBMI2 funnel shifts use VSHLD/VSHRD for const splats (PR #169401)

[Simon Pilgrim via llvm-commits]

================
@@ -57622,6 +57626,49 @@ static SDValue combineFP_TO_xINT_SAT(SDNode *N, SelectionDAG &DAG,
   return SDValue();
 }
 
+// Combiner: turn uniform-constant splat funnel shifts into VSHLD/VSHRD
----------------
RKSimon wrote:

we can't currently easily match constant splats inside td patterns - it'd be a nice to have someday, but beyond the scope of this patch - plus we'd hit the problem that the X86ISD::VSHRD node has an implicit operand swap that is going to be messy to remove.

https://github.com/llvm/llvm-project/pull/169401