[llvm] [X86] Use GFNI for vXi8 shifts/rotates (PR #89115)

[Simon Pilgrim via llvm-commits]

================
@@ -29175,6 +29198,14 @@ static SDValue LowerShiftByScalarImmediate(SDValue Op, SelectionDAG &DAG,
     if (VT == MVT::v16i8 && Subtarget.hasXOP())
       return SDValue();
 
+    if (Subtarget.hasGFNI()) {
----------------
RKSimon wrote:

No - 512-bit GFNI don't require BWI (although it will fail to use the mask predicates - see [avx512-gfni-intrinsics.ll](https://github.com/llvm/llvm-project/blob/main/llvm/test/CodeGen/X86/avx512-gfni-intrinsics.ll)).

I've added test coverage to ensure we track this

https://github.com/llvm/llvm-project/pull/89115