[X86] Remove x86ISD::INC/DEC. Just select them from X86ISD::ADD/SUB at isel time

[craig.topper]

Summary

INC/DEC are pretty much the same as ADD/SUB except that they don't update the C flag.

This patch removes the special nodes and just pattern matches from ADD/SUB during isel if the C flag isn't being used.

I had to avoid selecting DEC is the result isn't used. This will become a SUB immediate which will turned into a CMP later by optimizeCompareInstr. This lead to the one test change where we use a CMP instead of a DEC for an overflow intrinsic since we only checked the flag.

This also exposed a hole in our RMW flag matching use of hasNoCarryFlagUses. Our root node for the match is a store and there's no guarantee that all the flag users have been selected yet. So hasNoCarryFlagUses needs to check copyToReg and machine opcodes, but it also needs to check for the pre-match SETCC, SETCC_CARRY, BRCOND, and CMOV opcodes.

[craig.topper]

Remove the atomic INC/DEC ISD opcodes too.

[RKSimon]

lib/Target/X86/X86ISelDAGToDAG.cpp

2311 ↗

(On Diff #179242)

(style) return true here for ISD::CopyToReg special case and remove else + indentation below?

[craig.topper]

lib/Target/X86/X86ISelDAGToDAG.cpp

2311 ↗

(On Diff #179242)

We’re inside the outer for loop over users of the flags. We may have a mix of selected and unselected users. We can’t return true until we’ve visited them all.

[RKSimon]

lib/Target/X86/X86ISelDAGToDAG.cpp

2311 ↗

(On Diff #179242)

Sorry, I'm miscounting braces.....

[craig.topper]

lib/Target/X86/X86ISelDAGToDAG.cpp

2311 ↗

(On Diff #179242)

I'm going to add a continue there instead. And I'll move the duplicate CC switch into a helper.

[craig.topper]

Reduce code duplication and indendation in hasNoCarryFlagUses.

[craig.topper]

Merge some duplicate code in EmitTest. Remove now unused SDNodes from X86InstrInfo.td

[craig.topper]

Ping

[craig.topper]

Rebase

[RKSimon]

I'm happy with this - @spatel @andreadb any comments?

[spatel]

LGTM

lib/Target/X86/X86ISelLowering.cpp

18660 ↗

(On Diff #179810)

instructions selection -> instruction selection ?
This is an awkward/long sentence, so it might be better to just edit:

// Transform to an x86-specific ALU node with flags if there is a chance of
// folding a memory operand or only the flags are used. Otherwise, leave 
// the node alone and emit a 'test' instruction.

llvm/trunk/lib/Target/X86/X86FastISel.cpp

case Intrinsic::umul_with_overflow: {
     const Value *LHS = II->getArgOperand(0);
     const Value *RHS = II->getArgOperand(1);
 
     // Canonicalize immediate to the RHS.
     if (isa<ConstantInt>(LHS) && !isa<ConstantInt>(RHS) &&
         isCommutativeIntrinsic(II))
       std::swap(LHS, RHS);
 
-    bool UseIncDec = false;
-    if (isa<ConstantInt>(RHS) && cast<ConstantInt>(RHS)->isOne())
-      UseIncDec = true;
-
     unsigned BaseOpc, CondOpc;
     switch (II->getIntrinsicID()) {
     default: llvm_unreachable("Unexpected intrinsic!");
     case Intrinsic::sadd_with_overflow:
-      BaseOpc = UseIncDec ? unsigned(X86ISD::INC) : unsigned(ISD::ADD);
-      CondOpc = X86::SETOr;
-      break;
+      BaseOpc = ISD::ADD; CondOpc = X86::SETOr; break;
     case Intrinsic::uadd_with_overflow:
       BaseOpc = ISD::ADD; CondOpc = X86::SETBr; break;
     case Intrinsic::ssub_with_overflow:
-      BaseOpc = UseIncDec ? unsigned(X86ISD::DEC) : unsigned(ISD::SUB);
-      CondOpc = X86::SETOr;
-      break;
+      BaseOpc = ISD::SUB; CondOpc = X86::SETOr; break;
     case Intrinsic::usub_with_overflow:
       BaseOpc = ISD::SUB; CondOpc = X86::SETBr; break;
     case Intrinsic::smul_with_overflow:
       BaseOpc = X86ISD::SMUL; CondOpc = X86::SETOr; break;
     case Intrinsic::umul_with_overflow:
       BaseOpc = X86ISD::UMUL; CondOpc = X86::SETOr; break;
     }
 
     unsigned LHSReg = getRegForValue(LHS);
     if (LHSReg == 0)
       return false;
     bool LHSIsKill = hasTrivialKill(LHS);
 
     unsigned ResultReg = 0;
     // Check if we have an immediate version.
     if (const auto *CI = dyn_cast<ConstantInt>(RHS)) {
       static const uint16_t Opc[2][4] = {
         { X86::INC8r, X86::INC16r, X86::INC32r, X86::INC64r },
         { X86::DEC8r, X86::DEC16r, X86::DEC32r, X86::DEC64r }
       };
 
-      if (BaseOpc == X86ISD::INC || BaseOpc == X86ISD::DEC) {
+      if (CI->isOne() && (BaseOpc == ISD::ADD || BaseOpc == ISD::SUB) &&
+          CondOpc == X86::SETOr) {
+        // We can use INC/DEC.
         ResultReg = createResultReg(TLI.getRegClassFor(VT));
-        bool IsDec = BaseOpc == X86ISD::DEC;
+        bool IsDec = BaseOpc == ISD::SUB;
         BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
                 TII.get(Opc[IsDec][VT.SimpleTy-MVT::i8]), ResultReg)
           .addReg(LHSReg, getKillRegState(LHSIsKill));
       } else
         ResultReg = fastEmit_ri(VT, VT, BaseOpc, LHSReg, LHSIsKill,
                                 CI->getZExtValue());
     }
 

llvm/trunk/lib/Target/X86/X86ISelDAGToDAG.cpp

for (SDNode::use_iterator FlagUI = UI->use_begin(),
       default:
         return false;
       }
     }
   }
   return true;
 }
 
+static bool mayUseCarryFlag(X86::CondCode CC) {
+  switch (CC) {
+  // Comparisons which don't examine the CF flag.
+  case X86::COND_O: case X86::COND_NO:
+  case X86::COND_E: case X86::COND_NE:
+  case X86::COND_S: case X86::COND_NS:
+  case X86::COND_P: case X86::COND_NP:
+  case X86::COND_L: case X86::COND_GE:
+  case X86::COND_G: case X86::COND_LE:
+    return false;
+  // Anything else: assume conservatively.
+  default:
+    return true;
+  }
+}
+
 /// Test whether the given node which sets flags has any uses which require the
 /// CF flag to be accurate.
  bool X86DAGToDAGISel::hasNoCarryFlagUses(SDValue Flags) const {
   // Examine each user of the node.
   for (SDNode::use_iterator UI = Flags->use_begin(), UE = Flags->use_end();
          UI != UE; ++UI) {
     // Only check things that use the flags.
     if (UI.getUse().getResNo() != Flags.getResNo())
       continue;
+
+    unsigned UIOpc = UI->getOpcode();
+
+    if (UIOpc == ISD::CopyToReg) {
       // Only examine CopyToReg uses that copy to EFLAGS.
-    if (UI->getOpcode() != ISD::CopyToReg ||
-        cast<RegisterSDNode>(UI->getOperand(1))->getReg() != X86::EFLAGS)
+      if (cast<RegisterSDNode>(UI->getOperand(1))->getReg() != X86::EFLAGS)
         return false;
       // Examine each user of the CopyToReg use.
       for (SDNode::use_iterator FlagUI = UI->use_begin(), FlagUE = UI->use_end();
            FlagUI != FlagUE; ++FlagUI) {
         // Only examine the Flag result.
         if (FlagUI.getUse().getResNo() != 1)
           continue;
         // Anything unusual: assume conservatively.
         if (!FlagUI->isMachineOpcode())
           return false;
         // Examine the condition code of the user.
         X86::CondCode CC = getCondFromOpc(FlagUI->getMachineOpcode());
 
-      switch (CC) {
-      // Comparisons which don't examine the CF flag.
-      case X86::COND_O: case X86::COND_NO:
-      case X86::COND_E: case X86::COND_NE:
-      case X86::COND_S: case X86::COND_NS:
-      case X86::COND_P: case X86::COND_NP:
-      case X86::COND_L: case X86::COND_GE:
-      case X86::COND_G: case X86::COND_LE:
-        continue;
-      // Anything else: assume conservatively.
-      default:
-        return false;
-      }
-    }
+        if (mayUseCarryFlag(CC))
+          return false;
+      }
+
+      // This CopyToReg is ok. Move on to the next user.
+    }
+
+    // This might be an unselected node. So look for the pre-isel opcodes that
+    // use flags.
+    unsigned CCOpNo;
+    switch (UIOpc) {
+      // Something unusual. Be conservative.
+    case X86ISD::SETCC:       CCOpNo = 0; break;
+    case X86ISD::SETCC_CARRY: CCOpNo = 0; break;
+    case X86ISD::CMOV:        CCOpNo = 2; break;
+    case X86ISD::BRCOND:      CCOpNo = 2; break;
+
+    X86::CondCode CC = (X86::CondCode)UI->getConstantOperandVal(CCOpNo);
+    if (mayUseCarryFlag(CC))
+      return false;
   }
   return true;
 }
 
 /// Check whether or not the chain ending in StoreNode is suitable for doing
 /// the {load; op; store} to modify transformation.
 static bool isFusableLoadOpStorePattern(StoreSDNode *StoreNode,
                                         SDValue StoredVal, SelectionDAG *CurDAG,
bool X86DAGToDAGISel::foldLoadStoreIntoMemOperand(SDNode *Node) {
   if (MemVT != MVT::i64 && MemVT != MVT::i32 && MemVT != MVT::i16 &&
       MemVT != MVT::i8)
     return false;
 
   bool IsCommutable = false;
   switch (Opc) {
   default:
     return false;
-  case X86ISD::INC:
-  case X86ISD::DEC:
   case X86ISD::SUB:
   case X86ISD::SBB:
     break;
   case X86ISD::ADD:
   case X86ISD::ADC:
   case X86ISD::AND:
   case X86ISD::OR:
   case X86ISD::XOR:
case MVT::i8:
       return Opc8;
     default:
       llvm_unreachable("Invalid size!");
     }
   };
 
   MachineSDNode *Result;
   switch (Opc) {
-  case X86ISD::INC:
-  case X86ISD::DEC: {
-    unsigned NewOpc =
-        Opc == X86ISD::INC
+  case X86ISD::ADD:
+  case X86ISD::SUB:
+    // Try to match inc/dec.
+    if (!Subtarget->slowIncDec() ||
+        CurDAG->getMachineFunction().getFunction().optForSize()) {
+      bool IsOne = isOneConstant(StoredVal.getOperand(1));
+      bool IsNegOne = isAllOnesConstant(StoredVal.getOperand(1));
+      // ADD/SUB with 1/-1 and carry flag isn't used can use inc/dec.
+      if ((IsOne || IsNegOne) && hasNoCarryFlagUses(StoredVal.getValue(1))) {
+        unsigned NewOpc = 
+          ((Opc == X86ISD::ADD) == IsOne)
               ? SelectOpcode(X86::INC64m, X86::INC32m, X86::INC16m, X86::INC8m)
               : SelectOpcode(X86::DEC64m, X86::DEC32m, X86::DEC16m, X86::DEC8m);
         const SDValue Ops[] = {Base, Scale, Index, Disp, Segment, InputChain};
-    Result =
-        CurDAG->getMachineNode(NewOpc, SDLoc(Node), MVT::i32, MVT::Other, Ops);
+        Result = CurDAG->getMachineNode(NewOpc, SDLoc(Node), MVT::i32,
+                                        MVT::Other, Ops);
         break;
       }
-  case X86ISD::ADD:
+    }
+    LLVM_FALLTHROUGH;
   case X86ISD::ADC:
-  case X86ISD::SUB:
   case X86ISD::SBB:
   case X86ISD::AND:
   case X86ISD::OR:
   case X86ISD::XOR: {
     auto SelectRegOpcode = [SelectOpcode](unsigned Opc) {
       switch (Opc) {
       case X86ISD::ADD:
         return SelectOpcode(X86::ADD64mr, X86::ADD32mr, X86::ADD16mr,

llvm/trunk/lib/Target/X86/X86ISelLowering.h

enum NodeType : unsigned {
       /// Vector comparison generating mask bits for fp and
       /// integer signed and unsigned data types.
       CMPM,
       // Vector comparison with rounding mode for FP values
       CMPM_RND,
 
       // Arithmetic operations with FLAGS results.
       ADD, SUB, ADC, SBB, SMUL, UMUL,
-      INC, DEC, OR, XOR, AND,
+      OR, XOR, AND,
 
       // Bit field extract.
       BEXTR,
 
       // Zero High Bits Starting with Specified Bit Position.
       BZHI,
 
       // X86-specific multiply by immediate.
enum NodeType : unsigned {
       LCMPXCHG_DAG = ISD::FIRST_TARGET_MEMORY_OPCODE,
       LCMPXCHG8_DAG,
       LCMPXCHG16_DAG,
       LCMPXCHG8_SAVE_EBX_DAG,
       LCMPXCHG16_SAVE_RBX_DAG,
 
       /// LOCK-prefixed arithmetic read-modify-write instructions.
       /// EFLAGS, OUTCHAIN = LADD(INCHAIN, PTR, RHS)
-      LADD, LSUB, LOR, LXOR, LAND, LINC, LDEC,
+      LADD, LSUB, LOR, LXOR, LAND,
 
       // Load, scalar_to_vector, and zero extend.
       VZEXT_LOAD,
 
       // Store FP control world into i16 memory.
       FNSTCW16m,
 
       /// This instruction implements FP_TO_SINT with the

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

static SDValue EmitTest(SDValue Op, unsigned X86CC, const SDLoc &dl,
   unsigned NumOperands = 0;
 
   SDValue ArithOp = Op;
 
   // NOTICE: In the code below we use ArithOp to hold the arithmetic operation
   // which may be the result of a CAST.  We use the variable 'Op', which is the
   // non-casted variable when we check for possible users.
   switch (ArithOp.getOpcode()) {
-  case ISD::ADD:
-    // We only want to rewrite this as a target-specific node with attached
-    // flags if there is a reasonable chance of either using that to do custom
-    // instructions selection that can fold some of the memory operands, or if
-    // only the flags are used. If there are other uses, leave the node alone
-    // and emit a test instruction.
-    for (SDNode::use_iterator UI = Op.getNode()->use_begin(),
-         UE = Op.getNode()->use_end(); UI != UE; ++UI)
-      if (UI->getOpcode() != ISD::CopyToReg &&
-          UI->getOpcode() != ISD::SETCC &&
-          UI->getOpcode() != ISD::STORE)
-        goto default_case;
-
-    if (auto *C = dyn_cast<ConstantSDNode>(ArithOp.getOperand(1))) {
-      // An add of one will be selected as an INC.
-      if (C->isOne() &&
-          (!Subtarget.slowIncDec() ||
-           DAG.getMachineFunction().getFunction().optForSize())) {
-        Opcode = X86ISD::INC;
-        NumOperands = 1;
-        break;
-      }
-
-      // An add of negative one (subtract of one) will be selected as a DEC.
-      if (C->isAllOnesValue() &&
-          (!Subtarget.slowIncDec() ||
-           DAG.getMachineFunction().getFunction().optForSize())) {
-        Opcode = X86ISD::DEC;
-        NumOperands = 1;
-        break;
-      }
-    }
-
-    // Otherwise use a regular EFLAGS-setting add.
-    Opcode = X86ISD::ADD;
-    NumOperands = 2;
-    break;
-
   case ISD::AND:
     // If the primary 'and' result isn't used, don't bother using X86ISD::AND,
     // because a TEST instruction will be better.
     if (!hasNonFlagsUse(Op))
       break;
 
     LLVM_FALLTHROUGH;
+  case ISD::ADD:
   case ISD::SUB:
   case ISD::OR:
   case ISD::XOR:
-    // Similar to ISD::ADD above, check if the uses will preclude useful
-    // lowering of the target-specific node.
+    // Transform to an x86-specific ALU node with flags if there is a chance of
+    // using an RMW op or only the flags are used. Otherwise, leave
+    // the node alone and emit a 'test' instruction.
     for (SDNode::use_iterator UI = Op.getNode()->use_begin(),
            UE = Op.getNode()->use_end(); UI != UE; ++UI)
       if (UI->getOpcode() != ISD::CopyToReg &&
           UI->getOpcode() != ISD::SETCC &&
           UI->getOpcode() != ISD::STORE)
         goto default_case;
 
     // Otherwise use a regular EFLAGS-setting instruction.
     switch (ArithOp.getOpcode()) {
     default: llvm_unreachable("unexpected operator!");
+    case ISD::ADD: Opcode = X86ISD::ADD; break;
     case ISD::SUB: Opcode = X86ISD::SUB; break;
     case ISD::XOR: Opcode = X86ISD::XOR; break;
     case ISD::AND: Opcode = X86ISD::AND; break;
     case ISD::OR:  Opcode = X86ISD::OR;  break;
     }
 
     NumOperands = 2;
     break;
   case X86ISD::ADD:
   case X86ISD::SUB:
-  case X86ISD::INC:
-  case X86ISD::DEC:
   case X86ISD::OR:
   case X86ISD::XOR:
   case X86ISD::AND:
     return SDValue(Op.getNode(), 1);
   default:
   default_case:
     break;
   }
getX86XALUOOp(X86::CondCode &Cond, SDValue Op, SelectionDAG &DAG) {
   SDValue Value, Overflow;
   SDValue LHS = Op.getOperand(0);
   SDValue RHS = Op.getOperand(1);
   unsigned BaseOp = 0;
   SDLoc DL(Op);
   switch (Op.getOpcode()) {
   default: llvm_unreachable("Unknown ovf instruction!");
   case ISD::SADDO:
-    // A subtract of one will be selected as a INC. Note that INC doesn't
-    // set CF, so we can't do this for UADDO.
-    if (isOneConstant(RHS)) {
-      BaseOp = X86ISD::INC;
-      Cond = X86::COND_O;
-      break;
-    }
     BaseOp = X86ISD::ADD;
     Cond = X86::COND_O;
     break;
   case ISD::UADDO:
     BaseOp = X86ISD::ADD;
     Cond = X86::COND_B;
     break;
   case ISD::SSUBO:
-    // A subtract of one will be selected as a DEC. Note that DEC doesn't
-    // set CF, so we can't do this for USUBO.
-    if (isOneConstant(RHS)) {
-      BaseOp = X86ISD::DEC;
-      Cond = X86::COND_O;
-      break;
-    }
     BaseOp = X86ISD::SUB;
     Cond = X86::COND_O;
     break;
   case ISD::USUBO:
     BaseOp = X86ISD::SUB;
     Cond = X86::COND_B;
     break;
   case ISD::SMULO:

 static bool isX86LogicalCmp(SDValue Op) {
   unsigned Opc = Op.getOpcode();
   if (Opc == X86ISD::CMP || Opc == X86ISD::COMI || Opc == X86ISD::UCOMI ||
       Opc == X86ISD::SAHF)
     return true;
   if (Op.getResNo() == 1 &&
       (Opc == X86ISD::ADD || Opc == X86ISD::SUB || Opc == X86ISD::ADC ||
        Opc == X86ISD::SBB || Opc == X86ISD::SMUL || Opc == X86ISD::UMUL ||
-       Opc == X86ISD::INC || Opc == X86ISD::DEC || Opc == X86ISD::OR ||
-       Opc == X86ISD::XOR || Opc == X86ISD::AND))
+       Opc == X86ISD::OR || Opc == X86ISD::XOR || Opc == X86ISD::AND))
     return true;
 
   return false;
 }
 
 static bool isTruncWithZeroHighBitsInput(SDValue V, SelectionDAG &DAG) {
   if (V.getOpcode() != ISD::TRUNCATE)
     return false;
static SDValue LowerBITREVERSE(SDValue Op, const X86Subtarget &Subtarget,
   SDValue LoMask = DAG.getBuildVector(VT, DL, LoMaskElts);
   SDValue HiMask = DAG.getBuildVector(VT, DL, HiMaskElts);
   Lo = DAG.getNode(X86ISD::PSHUFB, DL, VT, LoMask, Lo);
   Hi = DAG.getNode(X86ISD::PSHUFB, DL, VT, HiMask, Hi);
   return DAG.getNode(ISD::OR, DL, VT, Lo, Hi);
 }
 
 static SDValue lowerAtomicArithWithLOCK(SDValue N, SelectionDAG &DAG,
-                                        const X86Subtarget &Subtarget,
-                                        bool AllowIncDec = true) {
+                                        const X86Subtarget &Subtarget) {
   unsigned NewOpc = 0;
   switch (N->getOpcode()) {
   case ISD::ATOMIC_LOAD_ADD:
     NewOpc = X86ISD::LADD;
     break;
   case ISD::ATOMIC_LOAD_SUB:
     NewOpc = X86ISD::LSUB;
     break;
   case ISD::ATOMIC_LOAD_OR:
     NewOpc = X86ISD::LOR;
     break;
   case ISD::ATOMIC_LOAD_XOR:
     NewOpc = X86ISD::LXOR;
     break;
   case ISD::ATOMIC_LOAD_AND:
     NewOpc = X86ISD::LAND;
     break;
   default:
     llvm_unreachable("Unknown ATOMIC_LOAD_ opcode");
   }
 
   MachineMemOperand *MMO = cast<MemSDNode>(N)->getMemOperand();
 
-  if (auto *C = dyn_cast<ConstantSDNode>(N->getOperand(2))) {
-    // Convert to inc/dec if they aren't slow or we are optimizing for size.
-    if (AllowIncDec && (!Subtarget.slowIncDec() ||
-                        DAG.getMachineFunction().getFunction().optForSize())) {
-      if ((NewOpc == X86ISD::LADD && C->isOne()) ||
-          (NewOpc == X86ISD::LSUB && C->isAllOnesValue()))
-        return DAG.getMemIntrinsicNode(X86ISD::LINC, SDLoc(N),
-                                       DAG.getVTList(MVT::i32, MVT::Other),
-                                       {N->getOperand(0), N->getOperand(1)},
-                                       /*MemVT=*/N->getSimpleValueType(0), MMO);
-      if ((NewOpc == X86ISD::LSUB && C->isOne()) ||
-          (NewOpc == X86ISD::LADD && C->isAllOnesValue()))
-        return DAG.getMemIntrinsicNode(X86ISD::LDEC, SDLoc(N),
-                                       DAG.getVTList(MVT::i32, MVT::Other),
-                                       {N->getOperand(0), N->getOperand(1)},
-                                       /*MemVT=*/N->getSimpleValueType(0), MMO);
-    }
-  }
-
   return DAG.getMemIntrinsicNode(
       NewOpc, SDLoc(N), DAG.getVTList(MVT::i32, MVT::Other),
       {N->getOperand(0), N->getOperand(1), N->getOperand(2)},
       /*MemVT=*/N->getSimpleValueType(0), MMO);
 }
 
 /// Lower atomic_load_ops into LOCK-prefixed operations.
 static SDValue lowerAtomicArith(SDValue N, SelectionDAG &DAG,
case X86ISD::LCMPXCHG8_SAVE_EBX_DAG:
     return "X86ISD::LCMPXCHG8_SAVE_EBX_DAG";
   case X86ISD::LCMPXCHG16_SAVE_RBX_DAG:
     return "X86ISD::LCMPXCHG16_SAVE_RBX_DAG";
   case X86ISD::LADD:               return "X86ISD::LADD";
   case X86ISD::LSUB:               return "X86ISD::LSUB";
   case X86ISD::LOR:                return "X86ISD::LOR";
   case X86ISD::LXOR:               return "X86ISD::LXOR";
   case X86ISD::LAND:               return "X86ISD::LAND";
-  case X86ISD::LINC:               return "X86ISD::LINC";
-  case X86ISD::LDEC:               return "X86ISD::LDEC";
   case X86ISD::VZEXT_MOVL:         return "X86ISD::VZEXT_MOVL";
   case X86ISD::VZEXT_LOAD:         return "X86ISD::VZEXT_LOAD";
   case X86ISD::VTRUNC:             return "X86ISD::VTRUNC";
   case X86ISD::VTRUNCS:            return "X86ISD::VTRUNCS";
   case X86ISD::VTRUNCUS:           return "X86ISD::VTRUNCUS";
   case X86ISD::VTRUNCSTORES:       return "X86ISD::VTRUNCSTORES";
   case X86ISD::VTRUNCSTOREUS:      return "X86ISD::VTRUNCSTOREUS";
   case X86ISD::VMTRUNCSTORES:      return "X86ISD::VMTRUNCSTORES";
const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
   case X86ISD::PCMPGT:             return "X86ISD::PCMPGT";
   case X86ISD::PHMINPOS:           return "X86ISD::PHMINPOS";
   case X86ISD::ADD:                return "X86ISD::ADD";
   case X86ISD::SUB:                return "X86ISD::SUB";
   case X86ISD::ADC:                return "X86ISD::ADC";
   case X86ISD::SBB:                return "X86ISD::SBB";
   case X86ISD::SMUL:               return "X86ISD::SMUL";
   case X86ISD::UMUL:               return "X86ISD::UMUL";
-  case X86ISD::INC:                return "X86ISD::INC";
-  case X86ISD::DEC:                return "X86ISD::DEC";
   case X86ISD::OR:                 return "X86ISD::OR";
   case X86ISD::XOR:                return "X86ISD::XOR";
   case X86ISD::AND:                return "X86ISD::AND";
   case X86ISD::BEXTR:              return "X86ISD::BEXTR";
   case X86ISD::BZHI:               return "X86ISD::BZHI";
   case X86ISD::MUL_IMM:            return "X86ISD::MUL_IMM";
   case X86ISD::MOVMSK:             return "X86ISD::MOVMSK";
   case X86ISD::PTEST:              return "X86ISD::PTEST";
if (Comparison == -Addend) {
     // The CC is fine, but we need to rewrite the LHS of the comparison as an
     // atomic sub.
     auto *AN = cast<AtomicSDNode>(CmpLHS.getNode());
     auto AtomicSub = DAG.getAtomic(
         ISD::ATOMIC_LOAD_SUB, SDLoc(CmpLHS), CmpLHS.getValueType(),
         /*Chain*/ CmpLHS.getOperand(0), /*LHS*/ CmpLHS.getOperand(1),
         /*RHS*/ DAG.getConstant(-Addend, SDLoc(CmpRHS), CmpRHS.getValueType()),
         AN->getMemOperand());
-    // If the comparision uses the CF flag we can't use INC/DEC instructions.
-    bool NeedCF = false;
-    switch (CC) {
-    default: break;
-    case X86::COND_A: case X86::COND_AE:
-    case X86::COND_B: case X86::COND_BE:
-      NeedCF = true;
-      break;
-    }
-    auto LockOp = lowerAtomicArithWithLOCK(AtomicSub, DAG, Subtarget, !NeedCF);
+    auto LockOp = lowerAtomicArithWithLOCK(AtomicSub, DAG, Subtarget);
     DAG.ReplaceAllUsesOfValueWith(CmpLHS.getValue(0),
                                   DAG.getUNDEF(CmpLHS.getValueType()));
     DAG.ReplaceAllUsesOfValueWith(CmpLHS.getValue(1), LockOp.getValue(1));
     return LockOp;
   }
 
   // We can handle comparisons with zero in a number of cases by manipulating
   // the CC used.

llvm/trunk/lib/Target/X86/X86InstrArithmetic.td

def NOT32m : I<0xF7, MRM2m, (outs), (ins i32mem:$dst),
                [(store (not (loadi32 addr:$dst)), addr:$dst)]>,
                OpSize32;
 def NOT64m : RI<0xF7, MRM2m, (outs), (ins i64mem:$dst), "not{q}\t$dst",
                 [(store (not (loadi64 addr:$dst)), addr:$dst)]>,
                 Requires<[In64BitMode]>;
 } // SchedRW
 } // CodeSize
 
+def X86add_flag_nocf : PatFrag<(ops node:$lhs, node:$rhs),
+                               (X86add_flag node:$lhs, node:$rhs), [{
+  return hasNoCarryFlagUses(SDValue(N, 1));
+}]>;
+
+def X86sub_flag_nocf : PatFrag<(ops node:$lhs, node:$rhs),
+                               (X86sub_flag node:$lhs, node:$rhs), [{
+  // Only use DEC if the result is used.
+  return !SDValue(N, 0).use_empty() && hasNoCarryFlagUses(SDValue(N, 1));
+}]>;
+
 // TODO: inc/dec is slow for P4, but fast for Pentium-M.
 let Defs = [EFLAGS] in {
 let Constraints = "$src1 = $dst", SchedRW = [WriteALU] in {
 let CodeSize = 2 in
 def INC8r  : I<0xFE, MRM0r, (outs GR8 :$dst), (ins GR8 :$src1),
                "inc{b}\t$dst",
-               [(set GR8:$dst, EFLAGS, (X86inc_flag GR8:$src1))]>;
+               [(set GR8:$dst, EFLAGS, (X86add_flag_nocf GR8:$src1, 1))]>;
 let isConvertibleToThreeAddress = 1, CodeSize = 2 in { // Can xform into LEA.
 def INC16r : I<0xFF, MRM0r, (outs GR16:$dst), (ins GR16:$src1),
                "inc{w}\t$dst",
-               [(set GR16:$dst, EFLAGS, (X86inc_flag GR16:$src1))]>, OpSize16;
+               [(set GR16:$dst, EFLAGS, (X86add_flag_nocf GR16:$src1, 1))]>,
+               OpSize16;
 def INC32r : I<0xFF, MRM0r, (outs GR32:$dst), (ins GR32:$src1),
                "inc{l}\t$dst",
-               [(set GR32:$dst, EFLAGS, (X86inc_flag GR32:$src1))]>, OpSize32;
+               [(set GR32:$dst, EFLAGS, (X86add_flag_nocf GR32:$src1, 1))]>,
+               OpSize32;
 def INC64r : RI<0xFF, MRM0r, (outs GR64:$dst), (ins GR64:$src1), "inc{q}\t$dst",
-                [(set GR64:$dst, EFLAGS, (X86inc_flag GR64:$src1))]>;
+                [(set GR64:$dst, EFLAGS, (X86add_flag_nocf GR64:$src1, 1))]>;
 } // isConvertibleToThreeAddress = 1, CodeSize = 2
 
 // Short forms only valid in 32-bit mode. Selected during MCInst lowering.
 let CodeSize = 1, hasSideEffects = 0 in {
 def INC16r_alt : I<0x40, AddRegFrm, (outs GR16:$dst), (ins GR16:$src1),
                    "inc{w}\t$dst", []>,
                  OpSize16, Requires<[Not64BitMode]>;
 def INC32r_alt : I<0x40, AddRegFrm, (outs GR32:$dst), (ins GR32:$src1),
def INC64m : RI<0xFF, MRM0m, (outs), (ins i64mem:$dst), "inc{q}\t$dst",
                    (implicit EFLAGS)]>;
 } // Predicates
 } // CodeSize = 2, SchedRW
 
 let Constraints = "$src1 = $dst", SchedRW = [WriteALU] in {
 let CodeSize = 2 in
 def DEC8r  : I<0xFE, MRM1r, (outs GR8 :$dst), (ins GR8 :$src1),
                "dec{b}\t$dst",
-               [(set GR8:$dst, EFLAGS, (X86dec_flag GR8:$src1))]>;
+               [(set GR8:$dst, EFLAGS, (X86sub_flag_nocf GR8:$src1, 1))]>;
 let isConvertibleToThreeAddress = 1, CodeSize = 2 in { // Can xform into LEA.
 def DEC16r : I<0xFF, MRM1r, (outs GR16:$dst), (ins GR16:$src1),
                "dec{w}\t$dst",
-               [(set GR16:$dst, EFLAGS, (X86dec_flag GR16:$src1))]>, OpSize16;
+               [(set GR16:$dst, EFLAGS, (X86sub_flag_nocf GR16:$src1, 1))]>,
+               OpSize16;
 def DEC32r : I<0xFF, MRM1r, (outs GR32:$dst), (ins GR32:$src1),
                "dec{l}\t$dst",
-               [(set GR32:$dst, EFLAGS, (X86dec_flag GR32:$src1))]>, OpSize32;
+               [(set GR32:$dst, EFLAGS, (X86sub_flag_nocf GR32:$src1, 1))]>,
+               OpSize32;
 def DEC64r : RI<0xFF, MRM1r, (outs GR64:$dst), (ins GR64:$src1), "dec{q}\t$dst",
-                [(set GR64:$dst, EFLAGS, (X86dec_flag GR64:$src1))]>;
+                [(set GR64:$dst, EFLAGS, (X86sub_flag_nocf GR64:$src1, 1))]>;
 } // isConvertibleToThreeAddress = 1, CodeSize = 2
 
 // Short forms only valid in 32-bit mode. Selected during MCInst lowering.
 let CodeSize = 1, hasSideEffects = 0 in {
 def DEC16r_alt : I<0x48, AddRegFrm, (outs GR16:$dst), (ins GR16:$src1),
                    "dec{w}\t$dst", []>,
                  OpSize16, Requires<[Not64BitMode]>;
 def DEC32r_alt : I<0x48, AddRegFrm, (outs GR32:$dst), (ins GR32:$src1),

llvm/trunk/lib/Target/X86/X86InstrCompiler.td

 }
 
 defm LOCK_ADD : LOCK_ArithBinOp<0x00, 0x80, 0x83, MRM0m, X86lock_add, "add">;
 defm LOCK_SUB : LOCK_ArithBinOp<0x28, 0x80, 0x83, MRM5m, X86lock_sub, "sub">;
 defm LOCK_OR  : LOCK_ArithBinOp<0x08, 0x80, 0x83, MRM1m, X86lock_or , "or">;
 defm LOCK_AND : LOCK_ArithBinOp<0x20, 0x80, 0x83, MRM4m, X86lock_and, "and">;
 defm LOCK_XOR : LOCK_ArithBinOp<0x30, 0x80, 0x83, MRM6m, X86lock_xor, "xor">;
 
-multiclass LOCK_ArithUnOp<bits<8> Opc8, bits<8> Opc, Format Form,
-                          string frag, string mnemonic> {
+def X86lock_add_nocf : PatFrag<(ops node:$lhs, node:$rhs),
+                               (X86lock_add node:$lhs, node:$rhs), [{
+  return hasNoCarryFlagUses(SDValue(N, 0));
+}]>;
+
+def X86lock_sub_nocf : PatFrag<(ops node:$lhs, node:$rhs),
+                               (X86lock_sub node:$lhs, node:$rhs), [{
+  return hasNoCarryFlagUses(SDValue(N, 0));
+}]>;
+
+let Predicates = [UseIncDec] in {
   let Defs = [EFLAGS], mayLoad = 1, mayStore = 1, isCodeGenOnly = 1,
-    SchedRW = [WriteALURMW] in {
-def NAME#8m  : I<Opc8, Form, (outs), (ins i8mem :$dst),
-                 !strconcat(mnemonic, "{b}\t$dst"),
-                 [(set EFLAGS, (!cast<PatFrag>(frag # "_8") addr:$dst))]>,
+      SchedRW = [WriteALURMW]  in {
+    def LOCK_INC8m  : I<0xFE, MRM0m, (outs), (ins i8mem :$dst),
+                        "inc{b}\t$dst",
+                        [(set EFLAGS, (X86lock_add_nocf addr:$dst, (i8 1)))]>,
                         LOCK;
-def NAME#16m : I<Opc, Form, (outs), (ins i16mem:$dst),
-                 !strconcat(mnemonic, "{w}\t$dst"),
-                 [(set EFLAGS, (!cast<PatFrag>(frag # "_16") addr:$dst))]>,
+    def LOCK_INC16m : I<0xFF, MRM0m, (outs), (ins i16mem:$dst),
+                        "inc{w}\t$dst",
+                        [(set EFLAGS, (X86lock_add_nocf addr:$dst, (i16 1)))]>,
                         OpSize16, LOCK;
-def NAME#32m : I<Opc, Form, (outs), (ins i32mem:$dst),
-                 !strconcat(mnemonic, "{l}\t$dst"),
-                 [(set EFLAGS, (!cast<PatFrag>(frag # "_32") addr:$dst))]>,
+    def LOCK_INC32m : I<0xFF, MRM0m, (outs), (ins i32mem:$dst),
+                        "inc{l}\t$dst",
+                        [(set EFLAGS, (X86lock_add_nocf addr:$dst, (i32 1)))]>,
                         OpSize32, LOCK;
-def NAME#64m : RI<Opc, Form, (outs), (ins i64mem:$dst),
-                  !strconcat(mnemonic, "{q}\t$dst"),
-                  [(set EFLAGS, (!cast<PatFrag>(frag # "_64") addr:$dst))]>,
+    def LOCK_INC64m : RI<0xFF, MRM0m, (outs), (ins i64mem:$dst),
+                         "inc{q}\t$dst",
+                         [(set EFLAGS, (X86lock_add_nocf addr:$dst, (i64 1)))]>,
                          LOCK;
-}
-}
 
-multiclass unary_atomic_intrin<SDNode atomic_op> {
-  def _8 : PatFrag<(ops node:$ptr),
-                   (atomic_op  node:$ptr), [{
-    return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i8;
-  }]>;
-  def _16 : PatFrag<(ops node:$ptr),
-                    (atomic_op node:$ptr), [{
-    return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i16;
-  }]>;
-  def _32 : PatFrag<(ops node:$ptr),
-                    (atomic_op node:$ptr), [{
-    return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i32;
-  }]>;
-  def _64 : PatFrag<(ops node:$ptr),
-                    (atomic_op node:$ptr), [{
-    return cast<MemIntrinsicSDNode>(N)->getMemoryVT() == MVT::i64;
-  }]>;
+    def LOCK_DEC8m  : I<0xFE, MRM1m, (outs), (ins i8mem :$dst),
+                        "dec{b}\t$dst",
+                        [(set EFLAGS, (X86lock_sub_nocf addr:$dst, (i8 1)))]>,
+                        LOCK;
+    def LOCK_DEC16m : I<0xFF, MRM1m, (outs), (ins i16mem:$dst),
+                        "dec{w}\t$dst",
+                        [(set EFLAGS, (X86lock_sub_nocf addr:$dst, (i16 1)))]>,
+                        OpSize16, LOCK;
+    def LOCK_DEC32m : I<0xFF, MRM1m, (outs), (ins i32mem:$dst),
+                        "dec{l}\t$dst",
+                        [(set EFLAGS, (X86lock_sub_nocf addr:$dst, (i32 1)))]>,
+                        OpSize32, LOCK;
+    def LOCK_DEC64m : RI<0xFF, MRM1m, (outs), (ins i64mem:$dst),
+                         "dec{q}\t$dst",
+                         [(set EFLAGS, (X86lock_sub_nocf addr:$dst, (i64 1)))]>,
+                         LOCK;
   }
 
-defm X86lock_inc : unary_atomic_intrin<X86lock_inc>;
-defm X86lock_dec : unary_atomic_intrin<X86lock_dec>;
-
-defm LOCK_INC    : LOCK_ArithUnOp<0xFE, 0xFF, MRM0m, "X86lock_inc", "inc">;
-defm LOCK_DEC    : LOCK_ArithUnOp<0xFE, 0xFF, MRM1m, "X86lock_dec", "dec">;
+  // Additional patterns for -1 constant.
+  def : Pat<(X86lock_add addr:$dst, (i8  -1)), (LOCK_DEC8m  addr:$dst)>;
+  def : Pat<(X86lock_add addr:$dst, (i16 -1)), (LOCK_DEC16m addr:$dst)>;
+  def : Pat<(X86lock_add addr:$dst, (i32 -1)), (LOCK_DEC32m addr:$dst)>;
+  def : Pat<(X86lock_add addr:$dst, (i64 -1)), (LOCK_DEC64m addr:$dst)>;
+  def : Pat<(X86lock_sub addr:$dst, (i8  -1)), (LOCK_INC8m  addr:$dst)>;
+  def : Pat<(X86lock_sub addr:$dst, (i16 -1)), (LOCK_INC16m addr:$dst)>;
+  def : Pat<(X86lock_sub addr:$dst, (i32 -1)), (LOCK_INC32m addr:$dst)>;
+  def : Pat<(X86lock_sub addr:$dst, (i64 -1)), (LOCK_INC64m addr:$dst)>;
+}
 
 // Atomic compare and swap.
 multiclass LCMPXCHG_UnOp<bits<8> Opc, Format Form, string mnemonic,
                          SDPatternOperator frag, X86MemOperand x86memop> {
 let isCodeGenOnly = 1, usesCustomInserter = 1 in {
   def NAME : I<Opc, Form, (outs), (ins x86memop:$ptr),
                !strconcat(mnemonic, "\t$ptr"),
                [(frag addr:$ptr)]>, TB, LOCK;
let Predicates = [UseIncDec] in {
   def : Pat<(add GR8:$src, 1),   (INC8r GR8:$src)>;
   def : Pat<(add GR16:$src, 1),  (INC16r GR16:$src)>;
   def : Pat<(add GR32:$src, 1),  (INC32r GR32:$src)>;
   def : Pat<(add GR64:$src, 1),  (INC64r GR64:$src)>;
   def : Pat<(add GR8:$src, -1),  (DEC8r GR8:$src)>;
   def : Pat<(add GR16:$src, -1), (DEC16r GR16:$src)>;
   def : Pat<(add GR32:$src, -1), (DEC32r GR32:$src)>;
   def : Pat<(add GR64:$src, -1), (DEC64r GR64:$src)>;
+
+  def : Pat<(X86add_flag_nocf GR8:$src, -1),  (DEC8r GR8:$src)>;
+  def : Pat<(X86add_flag_nocf GR16:$src, -1), (DEC16r GR16:$src)>;
+  def : Pat<(X86add_flag_nocf GR32:$src, -1), (DEC32r GR32:$src)>;
+  def : Pat<(X86add_flag_nocf GR64:$src, -1), (DEC64r GR64:$src)>;
+  def : Pat<(X86sub_flag_nocf GR8:$src, -1),  (INC8r GR8:$src)>;
+  def : Pat<(X86sub_flag_nocf GR16:$src, -1), (INC16r GR16:$src)>;
+  def : Pat<(X86sub_flag_nocf GR32:$src, -1), (INC32r GR32:$src)>;
+  def : Pat<(X86sub_flag_nocf GR64:$src, -1), (INC64r GR64:$src)>;
 }
 
 // or reg/reg.
 def : Pat<(or GR8 :$src1, GR8 :$src2), (OR8rr  GR8 :$src1, GR8 :$src2)>;
 def : Pat<(or GR16:$src1, GR16:$src2), (OR16rr GR16:$src1, GR16:$src2)>;
 def : Pat<(or GR32:$src1, GR32:$src2), (OR32rr GR32:$src1, GR32:$src2)>;
 def : Pat<(or GR64:$src1, GR64:$src2), (OR64rr GR64:$src1, GR64:$src2)>;
 

llvm/trunk/lib/Target/X86/X86InstrInfo.td

 def X86sub_flag  : SDNode<"X86ISD::SUB",  SDTBinaryArithWithFlags>;
 def X86smul_flag : SDNode<"X86ISD::SMUL", SDTBinaryArithWithFlags,
                           [SDNPCommutative]>;
 def X86umul_flag : SDNode<"X86ISD::UMUL", SDT2ResultBinaryArithWithFlags,
                           [SDNPCommutative]>;
 def X86adc_flag  : SDNode<"X86ISD::ADC",  SDTBinaryArithWithFlagsInOut>;
 def X86sbb_flag  : SDNode<"X86ISD::SBB",  SDTBinaryArithWithFlagsInOut>;
 
-def X86inc_flag  : SDNode<"X86ISD::INC",  SDTUnaryArithWithFlags>;
-def X86dec_flag  : SDNode<"X86ISD::DEC",  SDTUnaryArithWithFlags>;
 def X86or_flag   : SDNode<"X86ISD::OR",   SDTBinaryArithWithFlags,
                           [SDNPCommutative]>;
 def X86xor_flag  : SDNode<"X86ISD::XOR",  SDTBinaryArithWithFlags,
                           [SDNPCommutative]>;
 def X86and_flag  : SDNode<"X86ISD::AND",  SDTBinaryArithWithFlags,
                           [SDNPCommutative]>;
 
 def X86lock_add  : SDNode<"X86ISD::LADD",  SDTLockBinaryArithWithFlags,
                           [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
                            SDNPMemOperand]>;
 def X86lock_sub  : SDNode<"X86ISD::LSUB",  SDTLockBinaryArithWithFlags,
                           [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
                            SDNPMemOperand]>;
 def X86lock_or  : SDNode<"X86ISD::LOR",  SDTLockBinaryArithWithFlags,
                          [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
                           SDNPMemOperand]>;
 def X86lock_xor  : SDNode<"X86ISD::LXOR",  SDTLockBinaryArithWithFlags,
                           [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
                            SDNPMemOperand]>;
 def X86lock_and  : SDNode<"X86ISD::LAND",  SDTLockBinaryArithWithFlags,
                           [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
                            SDNPMemOperand]>;
 
-def X86lock_inc  : SDNode<"X86ISD::LINC",  SDTLockUnaryArithWithFlags,
-                          [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
-                           SDNPMemOperand]>;
-def X86lock_dec  : SDNode<"X86ISD::LDEC",  SDTLockUnaryArithWithFlags,
-                          [SDNPHasChain, SDNPMayStore, SDNPMayLoad,
-                           SDNPMemOperand]>;
-
 def X86bextr  : SDNode<"X86ISD::BEXTR",  SDTIntBinOp>;
 
 def X86bzhi   : SDNode<"X86ISD::BZHI",   SDTIntBinOp>;
 
 def X86mul_imm : SDNode<"X86ISD::MUL_IMM", SDTIntBinOp>;
 
 def X86WinAlloca : SDNode<"X86ISD::WIN_ALLOCA", SDT_X86WIN_ALLOCA,
                           [SDNPHasChain, SDNPOutGlue]>;

llvm/trunk/test/CodeGen/X86/sub-with-overflow.ll

 
 declare i32 @printf(i8*, ...) nounwind
 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32, i32)
 declare {i32, i1} @llvm.usub.with.overflow.i32(i32, i32)
 
 define i1 @func3(i32 %x) nounwind {
 ; CHECK-LABEL: func3:
 ; CHECK:       # %bb.0: # %entry
-; CHECK-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; CHECK-NEXT:    decl %eax
+; CHECK-NEXT:    cmpl $1, {{[0-9]+}}(%esp)
 ; CHECK-NEXT:    seto %al
 ; CHECK-NEXT:    retl
 entry:
   %t = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %x, i32 1)
   %obit = extractvalue {i32, i1} %t, 1
   ret i1 %obit
 
 }