[llvm] [RISCV] custom scmp(x, 0) and scmp(0, x) lowering for RVV (PR #151753)

[Olaf Bernstein via llvm-commits]

================
@@ -8223,6 +8225,40 @@ SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
   case ISD::SADDSAT:
   case ISD::SSUBSAT:
     return lowerToScalableOp(Op, DAG);
+  case ISD::SCMP: {
+    SDLoc DL(Op);
+    EVT VT = Op->getValueType(0);
+    EVT CCVT = getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT);
+
+    SDValue LHS = DAG.getFreeze(Op->getOperand(0));
+    SDValue RHS = DAG.getFreeze(Op->getOperand(1));
+    unsigned SEW = VT.getScalarSizeInBits();
+
+    SDValue Shift = DAG.getConstant(SEW-1, DL, VT);
+    SDValue Zero = DAG.getConstant(0, DL, VT);
+    SDValue One = DAG.getConstant(1, DL, VT);
+    SDValue MinusOne = DAG.getAllOnesConstant(DL, VT);
+
+    if (ISD::isConstantSplatVectorAllZeros(RHS.getNode())) {
+      SDValue Sra = DAG.getNode(ISD::SRA, DL, VT, LHS, Shift);
+      if (SEW <= 32) {
+        // scmp(lhs, 0) -> vor.vv(vsra.vi(lhs,SEW-1), vmin.vx(lhs,1))
----------------
camel-cdr wrote:

OK, I made the SEW=64 case also use the vor+vsra+vmin lowering. I originally thought that using two GPRs may be worse than a v0 clobber, if it causes a GPR spill, but I didn't consider that you can use a single GPR and load-immediates in the loop if you are starved of GPRs.

https://github.com/llvm/llvm-project/pull/151753