[RISCV] Promote fixed-length bf16 arith vector ops with zvfbfmin (#112393)

The aim is to have the same set of promotions on fixed-length bf16
vectors as on fixed-length f16 vectors, and then deduplicate them
similarly to what was done for scalable vectors.

It looks like fneg/fabs/fcopysign end up getting expanded because fsub
is now legal, and the default operation action must be expand.

Author: lukel97

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -1379,7 +1379,14 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
           setOperationAction(
               {ISD::VP_MERGE, ISD::VP_SELECT, ISD::VSELECT, ISD::SELECT}, VT,
               Custom);
-          // TODO: Promote to fp32.
+          MVT F32VecVT = MVT::getVectorVT(MVT::f32, VT.getVectorElementCount());
+          // Don't promote f16 vector operations to f32 if f32 vector type is
+          // not legal.
+          // TODO: could split the f16 vector into two vectors and do promotion.
+          if (!isTypeLegal(F32VecVT))
+            continue;
+          setOperationPromotedToType(ZvfhminZvfbfminPromoteOps, VT, F32VecVT);
+          // TODO: Promote VP ops to fp32.
           continue;
         }