[PowerPC] Add custom lowering for ssubo (#111748)

This patch is to improve the codegen for ssubo node for i32 in 64-bit
mode by custom lowering.

Author: maryammo

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

@@ -198,6 +198,11 @@ PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM,
 
   setOperationAction(ISD::UADDO, RegVT, Custom);
 
+  // On P10, the default lowering generates better code using the
+  // setbc instruction.
+  if (!Subtarget.hasP10Vector() && isPPC64)
+    setOperationAction(ISD::SSUBO, MVT::i32, Custom);
+
   // Match BITREVERSE to customized fast code sequence in the td file.
   setOperationAction(ISD::BITREVERSE, MVT::i32, Legal);
   setOperationAction(ISD::BITREVERSE, MVT::i64, Legal);
@@ -12041,6 +12046,27 @@ SDValue PPCTargetLowering::LowerUaddo(SDValue Op, SelectionDAG &DAG) const {
   return Res;
 }
 
+SDValue PPCTargetLowering::LowerSSUBO(SDValue Op, SelectionDAG &DAG) const {
+
+  SDLoc dl(Op);
+  SDValue LHS = Op.getOperand(0);
+  SDValue RHS = Op.getOperand(1);
+
+  SDValue Sub = DAG.getNode(ISD::SUB, dl, MVT::i32, LHS, RHS);
+
+  SDValue Xor1 = DAG.getNode(ISD::XOR, dl, MVT::i32, RHS, LHS);
+  SDValue Xor2 = DAG.getNode(ISD::XOR, dl, MVT::i32, Sub, LHS);
+
+  SDValue And = DAG.getNode(ISD::AND, dl, MVT::i32, Xor1, Xor2);
+
+  SDValue Overflow = DAG.getNode(ISD::SRL, dl, MVT::i32, And,
+                                 DAG.getConstant(31, dl, MVT::i32));
+  SDValue OverflowTrunc =
+      DAG.getNode(ISD::TRUNCATE, dl, Op.getNode()->getValueType(1), Overflow);
+
+  return DAG.getMergeValues({Sub, OverflowTrunc}, dl);
+}
+
 /// LowerOperation - Provide custom lowering hooks for some operations.
 ///
 SDValue PPCTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
@@ -12063,6 +12089,8 @@ SDValue PPCTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::SETCC:              return LowerSETCC(Op, DAG);
   case ISD::INIT_TRAMPOLINE:    return LowerINIT_TRAMPOLINE(Op, DAG);
   case ISD::ADJUST_TRAMPOLINE:  return LowerADJUST_TRAMPOLINE(Op, DAG);
+  case ISD::SSUBO:
+    return LowerSSUBO(Op, DAG);
 
   case ISD::INLINEASM:
   case ISD::INLINEASM_BR:       return LowerINLINEASM(Op, DAG);

llvm/lib/Target/PowerPC/PPCISelLowering.h

@@ -1285,6 +1285,7 @@ namespace llvm {
     SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerUaddo(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerSSUBO(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerINIT_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerADJUST_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerINLINEASM(SDValue Op, SelectionDAG &DAG) const;

llvm/test/CodeGen/PowerPC/saddo-ssubo.ll

@@ -129,12 +129,11 @@ entry:
 define i1 @test_ssubo_i32(i32 %a, i32 %b) nounwind {
 ; CHECK-LABEL: test_ssubo_i32:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    sub 5, 3, 4
-; CHECK-NEXT:    cmpwi 1, 4, 0
-; CHECK-NEXT:    cmpw 5, 3
-; CHECK-NEXT:    li 3, 1
-; CHECK-NEXT:    creqv 20, 5, 0
-; CHECK-NEXT:    isel 3, 0, 3, 20
+; CHECK-NEXT:    xor 5, 4, 3
+; CHECK-NEXT:    sub 4, 3, 4
+; CHECK-NEXT:    xor 3, 4, 3
+; CHECK-NEXT:    and 3, 5, 3
+; CHECK-NEXT:    srwi 3, 3, 31
 ; CHECK-NEXT:    blr
 entry:
   %res = call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %a, i32 %b) nounwind