LegalizeDAG: Try to use Overflow operations when expanding ADD/SUB

[jvesely]

ping

[resistor]

LGTM.

llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp

if (hasCarry) {
     } else {
       Lo = DAG.getNode(ISD::SUBC, dl, VTList, LoOps);
       HiOps[2] = Lo.getValue(1);
       Hi = DAG.getNode(ISD::SUBE, dl, VTList, HiOps);
     }
     return;
   }
 
+  bool hasOVF =
+    TLI.isOperationLegalOrCustom(N->getOpcode() == ISD::ADD ?
+                                   ISD::UADDO : ISD::USUBO,
+                                 TLI.getTypeToExpandTo(*DAG.getContext(), NVT));
+  if (hasOVF) {
+    SDVTList VTList = DAG.getVTList(NVT, NVT);
+    TargetLoweringBase::BooleanContent BoolType = TLI.getBooleanContents(NVT);
+    int RevOpc;
+    if (N->getOpcode() == ISD::ADD) {
+      RevOpc = ISD::SUB;
+      Lo = DAG.getNode(ISD::UADDO, dl, VTList, LoOps);
+      Hi = DAG.getNode(ISD::ADD, dl, NVT, makeArrayRef(HiOps, 2));
+    } else {
+      RevOpc = ISD::ADD;
+      Lo = DAG.getNode(ISD::USUBO, dl, VTList, LoOps);
+      Hi = DAG.getNode(ISD::SUB, dl, NVT, makeArrayRef(HiOps, 2));
+    }
+    SDValue OVF = Lo.getValue(1);
+
+    switch (BoolType) {
+    default:
+    case TargetLoweringBase::UndefinedBooleanContent:
+      OVF = DAG.getNode(ISD::AND, dl, NVT, DAG.getConstant(1, NVT), OVF);
+      // Fallthrough
+    case TargetLoweringBase::ZeroOrOneBooleanContent:
+      Hi = DAG.getNode(N->getOpcode(), dl, NVT, Hi, OVF);
+      break;
+    case TargetLoweringBase::ZeroOrNegativeOneBooleanContent:
+      Hi = DAG.getNode(RevOpc, dl, NVT, Hi, OVF);
+    }
+    return;
+  }
+
   if (N->getOpcode() == ISD::ADD) {
     Lo = DAG.getNode(ISD::ADD, dl, NVT, LoOps);
     Hi = DAG.getNode(ISD::ADD, dl, NVT, makeArrayRef(HiOps, 2));
     SDValue Cmp1 = DAG.getSetCC(dl, getSetCCResultType(NVT), Lo, LoOps[0],
                                 ISD::SETULT);
     SDValue Carry1 = DAG.getSelect(dl, NVT, Cmp1,
                                    DAG.getConstant(1, NVT),
                                    DAG.getConstant(0, NVT));

llvm/trunk/test/CodeGen/Hexagon/adde.ll

 ; RUN: llc -march=hexagon -hexagon-expand-condsets=0 < %s | FileCheck %s
 
-; CHECK: r{{[0-9]+:[0-9]+}} = #0
-; CHECK: r{{[0-9]+:[0-9]+}} = #1
+; CHECK: r{{[0-9]+:[0-9]+}} = add(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
 ; CHECK: r{{[0-9]+:[0-9]+}} = add(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
 ; CHECK: p{{[0-9]+}} = cmp.gtu(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
-; CHECK: p{{[0-9]+}} = cmp.gtu(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
-; CHECK: r{{[0-9]+}} = mux(p{{[0-9]+}}, r{{[0-9]+}}, r{{[0-9]+}})
-; CHECK: r{{[0-9]+}} = mux(p{{[0-9]+}}, r{{[0-9]+}}, r{{[0-9]+}})
-; CHECK: r{{[0-9]+:[0-9]+}} = combine(r{{[0-9]+}}, r{{[0-9]+}})
-; CHECK: r{{[0-9]+}} = mux(p{{[0-9]+}}, r{{[0-9]+}}, r{{[0-9]+}})
-; CHECK: r{{[0-9]+}} = mux(p{{[0-9]+}}, r{{[0-9]+}}, r{{[0-9]+}})
-; CHECK: r{{[0-9]+:[0-9]+}} = combine(r{{[0-9]+}}, r{{[0-9]+}})
+; CHECK: r{{[0-9]+}} = mux(p{{[0-9]+}}, #1, #0)
 ; CHECK: r{{[0-9]+:[0-9]+}} = add(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
 
 
 define void @check_adde_addc (i64 %AL, i64 %AH, i64 %BL, i64 %BH, i64* %RL, i64* %RH) {
 entry:
         %tmp1 = zext i64 %AL to i128
         %tmp23 = zext i64 %AH to i128
         %tmp4 = shl i128 %tmp23, 64

llvm/trunk/test/CodeGen/Hexagon/sube.ll

 ; RUN: llc -march=hexagon -hexagon-expand-condsets=0 < %s | FileCheck %s
 
-; CHECK: r{{[0-9]+:[0-9]+}} = #0
-; CHECK: r{{[0-9]+:[0-9]+}} = #1
-; CHECK: p{{[0-9]+}} = cmp.gtu(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
-; CHECK: r{{[0-9]+}} = mux(p{{[0-9]+}}, r{{[0-9]+}}, r{{[0-9]+}})
-; CHECK: r{{[0-9]+}} = mux(p{{[0-9]+}}, r{{[0-9]+}}, r{{[0-9]+}})
 ; CHECK: r{{[0-9]+:[0-9]+}} = sub(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
 ; CHECK: r{{[0-9]+:[0-9]+}} = sub(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
-; CHECK: r{{[0-9]+:[0-9]+}} = combine(r{{[0-9]+}}, r{{[0-9]+}})
+; CHECK: p{{[0-9]+}} = cmp.gtu(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
+; CHECK: r{{[0-9]+}} = mux(p{{[0-9]+}}, #1, #0)
+; CHECK: r{{[0-9]+:[0-9]+}} = sub(r{{[0-9]+:[0-9]+}}, r{{[0-9]+:[0-9]+}})
 
 define void @check_sube_subc(i64 %AL, i64 %AH, i64 %BL, i64 %BH, i64* %RL, i64* %RH) {
 entry:
         %tmp1 = zext i64 %AL to i128
         %tmp23 = zext i64 %AH to i128
         %tmp4 = shl i128 %tmp23, 64
         %tmp5 = or i128 %tmp4, %tmp1
         %tmp67 = zext i64 %BL to i128

llvm/trunk/test/CodeGen/NVPTX/add-128bit.ll

 ; RUN: llc < %s -march=nvptx -mcpu=sm_20 | FileCheck %s
 
 target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v16:16:16-v32:32:32-v64:64:64-v128:128:128-n16:32:64"
 
 
 
 define void @foo(i64 %a, i64 %add, i128* %retptr) {
 ; CHECK:        add.s64
 ; CHECK:        setp.lt.u64
-; CHECK:        setp.lt.u64
-; CHECK:        selp.b64
 ; CHECK:        selp.b64
-; CHECK:        add.s64
+; CHECK:        sub.s64
   %t1 = sext i64 %a to i128
   %add2 = zext i64 %add to i128
   %val = add i128 %t1, %add2
   store i128 %val, i128* %retptr
   ret void
 }

llvm/trunk/test/CodeGen/R600/sub.ll

define void @test_sub_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
   store <4 x i32> %result, <4 x i32> addrspace(1)* %out
   ret void
 }
 
 ; FUNC-LABEL: {{^}}s_sub_i64:
 ; SI: s_sub_u32
 ; SI: s_subb_u32
 
-; EG-DAG: SETGE_UINT
-; EG-DAG: CNDE_INT
-; EG-DAG: SUB_INT
 ; EG-DAG: SUB_INT
+; EG-DAG: SETGT_UINT
 ; EG-DAG: SUB_INT
+; EG-DAG: ADD_INT
 define void @s_sub_i64(i64 addrspace(1)* noalias %out, i64 %a, i64 %b) nounwind {
   %result = sub i64 %a, %b
   store i64 %result, i64 addrspace(1)* %out, align 8
   ret void
 }
 
 ; FUNC-LABEL: {{^}}v_sub_i64:
 ; SI: v_sub_i32_e32
 ; SI: v_subb_u32_e32
 
-; EG-DAG: SETGE_UINT
-; EG-DAG: CNDE_INT
-; EG-DAG: SUB_INT
 ; EG-DAG: SUB_INT
+; EG-DAG: SETGT_UINT
 ; EG-DAG: SUB_INT
+; EG-DAG: ADD_INT
 define void @v_sub_i64(i64 addrspace(1)* noalias %out, i64 addrspace(1)* noalias %inA, i64 addrspace(1)* noalias %inB) nounwind {
   %tid = call i32 @llvm.r600.read.tidig.x() readnone
   %a_ptr = getelementptr i64, i64 addrspace(1)* %inA, i32 %tid
   %b_ptr = getelementptr i64, i64 addrspace(1)* %inB, i32 %tid
   %a = load i64, i64 addrspace(1)* %a_ptr
   %b = load i64, i64 addrspace(1)* %b_ptr
   %result = sub i64 %a, %b
   store i64 %result, i64 addrspace(1)* %out, align 8