X86ISelLowering.cpp

    unsigned EltAlign = 0;
    getMaxByValAlign(ATy->getElementType(), EltAlign);
    if (EltAlign > MaxAlign)
      MaxAlign = EltAlign;
  } else if (const StructType *STy = dyn_cast<StructType>(Ty)) {
    for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
      unsigned EltAlign = 0;
      getMaxByValAlign(STy->getElementType(i), EltAlign);
      if (EltAlign > MaxAlign)
        MaxAlign = EltAlign;
      if (MaxAlign == 16)
        break;
    }
  }
  return;
}

/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
/// function arguments in the caller parameter area. For X86, aggregates
/// that contain SSE vectors are placed at 16-byte boundaries while the rest
/// are at 4-byte boundaries.
unsigned X86TargetLowering::getByValTypeAlignment(const Type *Ty) const {
  if (Subtarget->is64Bit()) {
    // Max of 8 and alignment of type.
    unsigned TyAlign = TD->getABITypeAlignment(Ty);
    if (TyAlign > 8)
      return TyAlign;
    return 8;
  }

  unsigned Align = 4;
  if (Subtarget->hasSSE1())
    getMaxByValAlign(Ty, Align);
  return Align;
}

/// getOptimalMemOpType - Returns the target specific optimal type for load
/// and store operations as a result of memset, memcpy, and memmove
/// lowering. It returns MVT::iAny if SelectionDAG should be responsible for
/// determining it.
EVT
X86TargetLowering::getOptimalMemOpType(uint64_t Size, unsigned Align,
                                       bool isSrcConst, bool isSrcStr,
                                       SelectionDAG &DAG) const {
  // FIXME: This turns off use of xmm stores for memset/memcpy on targets like
  // linux.  This is because the stack realignment code can't handle certain
  // cases like PR2962.  This should be removed when PR2962 is fixed.
  const Function *F = DAG.getMachineFunction().getFunction();
  bool NoImplicitFloatOps = F->hasFnAttr(Attribute::NoImplicitFloat);
  if (!NoImplicitFloatOps && Subtarget->getStackAlignment() >= 16) {
    if ((isSrcConst || isSrcStr) && Subtarget->hasSSE2() && Size >= 16)
      return MVT::v4i32;
    if ((isSrcConst || isSrcStr) && Subtarget->hasSSE1() && Size >= 16)
      return MVT::v4f32;
  }
  if (Subtarget->is64Bit() && Size >= 8)
    return MVT::i64;
  return MVT::i32;
}

/// getPICJumpTableRelocaBase - Returns relocation base for the given PIC
/// jumptable.
SDValue X86TargetLowering::getPICJumpTableRelocBase(SDValue Table,
                                                      SelectionDAG &DAG) const {
  if (usesGlobalOffsetTable())
    return DAG.getGLOBAL_OFFSET_TABLE(getPointerTy());
  if (!Subtarget->is64Bit())
    // This doesn't have DebugLoc associated with it, but is not really the
    // same as a Register.
    return DAG.getNode(X86ISD::GlobalBaseReg, DebugLoc::getUnknownLoc(),
                       getPointerTy());
  return Table;
}

/// getFunctionAlignment - Return the Log2 alignment of this function.
unsigned X86TargetLowering::getFunctionAlignment(const Function *F) const {
  return F->hasFnAttr(Attribute::OptimizeForSize) ? 0 : 4;
}

//===----------------------------------------------------------------------===//
//               Return Value Calling Convention Implementation
//===----------------------------------------------------------------------===//

#include "X86GenCallingConv.inc"

SDValue
X86TargetLowering::LowerReturn(SDValue Chain,
                               CallingConv::ID CallConv, bool isVarArg,
                               const SmallVectorImpl<ISD::OutputArg> &Outs,
                               DebugLoc dl, SelectionDAG &DAG) {

  SmallVector<CCValAssign, 16> RVLocs;
  CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
                 RVLocs, *DAG.getContext());
  CCInfo.AnalyzeReturn(Outs, RetCC_X86);

  // If this is the first return lowered for this function, add the regs to the
  // liveout set for the function.
  if (DAG.getMachineFunction().getRegInfo().liveout_empty()) {
    for (unsigned i = 0; i != RVLocs.size(); ++i)
      if (RVLocs[i].isRegLoc())
        DAG.getMachineFunction().getRegInfo().addLiveOut(RVLocs[i].getLocReg());
  }

  SDValue Flag;

  SmallVector<SDValue, 6> RetOps;
  RetOps.push_back(Chain); // Operand #0 = Chain (updated below)
  // Operand #1 = Bytes To Pop
  RetOps.push_back(DAG.getTargetConstant(getBytesToPopOnReturn(), MVT::i16));

  // Copy the result values into the output registers.
  for (unsigned i = 0; i != RVLocs.size(); ++i) {
    CCValAssign &VA = RVLocs[i];
    assert(VA.isRegLoc() && "Can only return in registers!");
    SDValue ValToCopy = Outs[i].Val;

    // Returns in ST0/ST1 are handled specially: these are pushed as operands to
    // the RET instruction and handled by the FP Stackifier.
    if (VA.getLocReg() == X86::ST0 ||
        VA.getLocReg() == X86::ST1) {
      // If this is a copy from an xmm register to ST(0), use an FPExtend to
      // change the value to the FP stack register class.
      if (isScalarFPTypeInSSEReg(VA.getValVT()))
        ValToCopy = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f80, ValToCopy);
      RetOps.push_back(ValToCopy);
      // Don't emit a copytoreg.
      continue;
    }

    // 64-bit vector (MMX) values are returned in XMM0 / XMM1 except for v1i64
    // which is returned in RAX / RDX.
    if (Subtarget->is64Bit()) {
      EVT ValVT = ValToCopy.getValueType();
      if (ValVT.isVector() && ValVT.getSizeInBits() == 64) {
        ValToCopy = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, ValToCopy);
        if (VA.getLocReg() == X86::XMM0 || VA.getLocReg() == X86::XMM1)
          ValToCopy = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, ValToCopy);
      }
    }

    Chain = DAG.getCopyToReg(Chain, dl, VA.getLocReg(), ValToCopy, Flag);
    Flag = Chain.getValue(1);
  }

  // The x86-64 ABI for returning structs by value requires that we copy
  // the sret argument into %rax for the return. We saved the argument into
  // a virtual register in the entry block, so now we copy the value out
  // and into %rax.
  if (Subtarget->is64Bit() &&
      DAG.getMachineFunction().getFunction()->hasStructRetAttr()) {
    MachineFunction &MF = DAG.getMachineFunction();
    X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
    unsigned Reg = FuncInfo->getSRetReturnReg();
    if (!Reg) {
      Reg = MF.getRegInfo().createVirtualRegister(getRegClassFor(MVT::i64));
      FuncInfo->setSRetReturnReg(Reg);
    }
    SDValue Val = DAG.getCopyFromReg(Chain, dl, Reg, getPointerTy());

    Chain = DAG.getCopyToReg(Chain, dl, X86::RAX, Val, Flag);
    Flag = Chain.getValue(1);
  }

  RetOps[0] = Chain;  // Update chain.

  // Add the flag if we have it.
  if (Flag.getNode())
    RetOps.push_back(Flag);

  return DAG.getNode(X86ISD::RET_FLAG, dl,
                     MVT::Other, &RetOps[0], RetOps.size());
}

/// LowerCallResult - Lower the result values of a call into the
/// appropriate copies out of appropriate physical registers.
///
SDValue
X86TargetLowering::LowerCallResult(SDValue Chain, SDValue InFlag,
                                   CallingConv::ID CallConv, bool isVarArg,
                                   const SmallVectorImpl<ISD::InputArg> &Ins,
                                   DebugLoc dl, SelectionDAG &DAG,
                                   SmallVectorImpl<SDValue> &InVals) {

  // Assign locations to each value returned by this call.
  SmallVector<CCValAssign, 16> RVLocs;
  bool Is64Bit = Subtarget->is64Bit();
  CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
                 RVLocs, *DAG.getContext());
  CCInfo.AnalyzeCallResult(Ins, RetCC_X86);

  // Copy all of the result registers out of their specified physreg.
  for (unsigned i = 0; i != RVLocs.size(); ++i) {
    CCValAssign &VA = RVLocs[i];
    EVT CopyVT = VA.getValVT();

    // If this is x86-64, and we disabled SSE, we can't return FP values
    if ((CopyVT == MVT::f32 || CopyVT == MVT::f64) &&
        ((Is64Bit || Ins[i].Flags.isInReg()) && !Subtarget->hasSSE1())) {
      llvm_report_error("SSE register return with SSE disabled");
    }

    // If this is a call to a function that returns an fp value on the floating
    // point stack, but where we prefer to use the value in xmm registers, copy
    // it out as F80 and use a truncate to move it from fp stack reg to xmm reg.
    if ((VA.getLocReg() == X86::ST0 ||
         VA.getLocReg() == X86::ST1) &&
        isScalarFPTypeInSSEReg(VA.getValVT())) {
      CopyVT = MVT::f80;
    }

    SDValue Val;
    if (Is64Bit && CopyVT.isVector() && CopyVT.getSizeInBits() == 64) {
      // For x86-64, MMX values are returned in XMM0 / XMM1 except for v1i64.
      if (VA.getLocReg() == X86::XMM0 || VA.getLocReg() == X86::XMM1) {
        Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(),
                                   MVT::v2i64, InFlag).getValue(1);
        Val = Chain.getValue(0);
        Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64,
                          Val, DAG.getConstant(0, MVT::i64));
      } else {
        Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(),
                                   MVT::i64, InFlag).getValue(1);
        Val = Chain.getValue(0);
      }
      Val = DAG.getNode(ISD::BIT_CONVERT, dl, CopyVT, Val);
    } else {
      Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(),
                                 CopyVT, InFlag).getValue(1);
      Val = Chain.getValue(0);
    }
    InFlag = Chain.getValue(2);

    if (CopyVT != VA.getValVT()) {
      // Round the F80 the right size, which also moves to the appropriate xmm
      // register.
      Val = DAG.getNode(ISD::FP_ROUND, dl, VA.getValVT(), Val,
                        // This truncation won't change the value.
                        DAG.getIntPtrConstant(1));
    }

    InVals.push_back(Val);
  }

  return Chain;
}


//===----------------------------------------------------------------------===//
//                C & StdCall & Fast Calling Convention implementation
//===----------------------------------------------------------------------===//
//  StdCall calling convention seems to be standard for many Windows' API
//  routines and around. It differs from C calling convention just a little:
//  callee should clean up the stack, not caller. Symbols should be also
//  decorated in some fancy way :) It doesn't support any vector arguments.
//  For info on fast calling convention see Fast Calling Convention (tail call)
//  implementation LowerX86_32FastCCCallTo.

/// CallIsStructReturn - Determines whether a call uses struct return
/// semantics.
static bool CallIsStructReturn(const SmallVectorImpl<ISD::OutputArg> &Outs) {
  if (Outs.empty())
    return false;

  return Outs[0].Flags.isSRet();
}

/// ArgsAreStructReturn - Determines whether a function uses struct
/// return semantics.
static bool
ArgsAreStructReturn(const SmallVectorImpl<ISD::InputArg> &Ins) {
  if (Ins.empty())
    return false;

  return Ins[0].Flags.isSRet();
}

/// IsCalleePop - Determines whether the callee is required to pop its
/// own arguments. Callee pop is necessary to support tail calls.
bool X86TargetLowering::IsCalleePop(bool IsVarArg, CallingConv::ID CallingConv){
  if (IsVarArg)
    return false;

  switch (CallingConv) {
  default:
    return false;
  case CallingConv::X86_StdCall:
    return !Subtarget->is64Bit();
  case CallingConv::X86_FastCall:
    return !Subtarget->is64Bit();
  case CallingConv::Fast:
    return PerformTailCallOpt;
  }
}

/// CCAssignFnForNode - Selects the correct CCAssignFn for a the
/// given CallingConvention value.
CCAssignFn *X86TargetLowering::CCAssignFnForNode(CallingConv::ID CC) const {
  if (Subtarget->is64Bit()) {
    if (Subtarget->isTargetWin64())
      return CC_X86_Win64_C;
    else
      return CC_X86_64_C;
  }

  if (CC == CallingConv::X86_FastCall)
    return CC_X86_32_FastCall;
  else if (CC == CallingConv::Fast)
    return CC_X86_32_FastCC;
  else
    return CC_X86_32_C;
}

/// NameDecorationForCallConv - Selects the appropriate decoration to
/// apply to a MachineFunction containing a given calling convention.
NameDecorationStyle
X86TargetLowering::NameDecorationForCallConv(CallingConv::ID CallConv) {
  if (CallConv == CallingConv::X86_FastCall)
    return FastCall;
  else if (CallConv == CallingConv::X86_StdCall)
    return StdCall;
  return None;
}


/// CreateCopyOfByValArgument - Make a copy of an aggregate at address specified
/// by "Src" to address "Dst" with size and alignment information specified by
/// the specific parameter attribute. The copy will be passed as a byval
/// function parameter.
static SDValue
CreateCopyOfByValArgument(SDValue Src, SDValue Dst, SDValue Chain,
                          ISD::ArgFlagsTy Flags, SelectionDAG &DAG,
                          DebugLoc dl) {
  SDValue SizeNode     = DAG.getConstant(Flags.getByValSize(), MVT::i32);
  return DAG.getMemcpy(Chain, dl, Dst, Src, SizeNode, Flags.getByValAlign(),
                       /*AlwaysInline=*/true, NULL, 0, NULL, 0);
}

SDValue
X86TargetLowering::LowerMemArgument(SDValue Chain,
                                    CallingConv::ID CallConv,
                                    const SmallVectorImpl<ISD::InputArg> &Ins,
                                    DebugLoc dl, SelectionDAG &DAG,
                                    const CCValAssign &VA,
                                    MachineFrameInfo *MFI,
                                    unsigned i) {

  // Create the nodes corresponding to a load from this parameter slot.
  ISD::ArgFlagsTy Flags = Ins[i].Flags;
  bool AlwaysUseMutable = (CallConv==CallingConv::Fast) && PerformTailCallOpt;
  bool isImmutable = !AlwaysUseMutable && !Flags.isByVal();
  EVT ValVT;

  // If value is passed by pointer we have address passed instead of the value
  // itself.
  if (VA.getLocInfo() == CCValAssign::Indirect)
    ValVT = VA.getLocVT();
  else
    ValVT = VA.getValVT();

  // FIXME: For now, all byval parameter objects are marked mutable. This can be
  // changed with more analysis.
  // In case of tail call optimization mark all arguments mutable. Since they
  // could be overwritten by lowering of arguments in case of a tail call.
  int FI = MFI->CreateFixedObject(ValVT.getSizeInBits()/8,
                                  VA.getLocMemOffset(), isImmutable);
  SDValue FIN = DAG.getFrameIndex(FI, getPointerTy());
  if (Flags.isByVal())
    return FIN;
  return DAG.getLoad(ValVT, dl, Chain, FIN,
                     PseudoSourceValue::getFixedStack(FI), 0);
}

SDValue
X86TargetLowering::LowerFormalArguments(SDValue Chain,
                                        CallingConv::ID CallConv,
                                        bool isVarArg,
                                      const SmallVectorImpl<ISD::InputArg> &Ins,
                                        DebugLoc dl,
                                        SelectionDAG &DAG,
                                        SmallVectorImpl<SDValue> &InVals) {

  MachineFunction &MF = DAG.getMachineFunction();
  X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();

  const Function* Fn = MF.getFunction();
  if (Fn->hasExternalLinkage() &&
      Subtarget->isTargetCygMing() &&
      Fn->getName() == "main")
    FuncInfo->setForceFramePointer(true);

  // Decorate the function name.
  FuncInfo->setDecorationStyle(NameDecorationForCallConv(CallConv));

  MachineFrameInfo *MFI = MF.getFrameInfo();
  bool Is64Bit = Subtarget->is64Bit();
  bool IsWin64 = Subtarget->isTargetWin64();

  assert(!(isVarArg && CallConv == CallingConv::Fast) &&
         "Var args not supported with calling convention fastcc");

  // Assign locations to all of the incoming arguments.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
                 ArgLocs, *DAG.getContext());
  CCInfo.AnalyzeFormalArguments(Ins, CCAssignFnForNode(CallConv));

  unsigned LastVal = ~0U;
  SDValue ArgValue;
  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    CCValAssign &VA = ArgLocs[i];
    // TODO: If an arg is passed in two places (e.g. reg and stack), skip later
    // places.
    assert(VA.getValNo() != LastVal &&
           "Don't support value assigned to multiple locs yet");
    LastVal = VA.getValNo();

    if (VA.isRegLoc()) {
      EVT RegVT = VA.getLocVT();
      TargetRegisterClass *RC = NULL;
      if (RegVT == MVT::i32)
        RC = X86::GR32RegisterClass;
      else if (Is64Bit && RegVT == MVT::i64)
        RC = X86::GR64RegisterClass;
      else if (RegVT == MVT::f32)
        RC = X86::FR32RegisterClass;
      else if (RegVT == MVT::f64)
        RC = X86::FR64RegisterClass;
      else if (RegVT.isVector() && RegVT.getSizeInBits() == 128)
        RC = X86::VR128RegisterClass;
      else if (RegVT.isVector() && RegVT.getSizeInBits() == 64)
        RC = X86::VR64RegisterClass;
      else
        llvm_unreachable("Unknown argument type!");

      unsigned Reg = MF.addLiveIn(VA.getLocReg(), RC);
      ArgValue = DAG.getCopyFromReg(Chain, dl, Reg, RegVT);

      // If this is an 8 or 16-bit value, it is really passed promoted to 32
      // bits.  Insert an assert[sz]ext to capture this, then truncate to the
      // right size.
      if (VA.getLocInfo() == CCValAssign::SExt)
        ArgValue = DAG.getNode(ISD::AssertSext, dl, RegVT, ArgValue,
                               DAG.getValueType(VA.getValVT()));
      else if (VA.getLocInfo() == CCValAssign::ZExt)
        ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
                               DAG.getValueType(VA.getValVT()));
      else if (VA.getLocInfo() == CCValAssign::BCvt)
        ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getValVT(), ArgValue);

      if (VA.isExtInLoc()) {
        // Handle MMX values passed in XMM regs.
        if (RegVT.isVector()) {
          ArgValue = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64,
                                 ArgValue, DAG.getConstant(0, MVT::i64));
          ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getValVT(), ArgValue);
        } else
          ArgValue = DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), ArgValue);
      }
    } else {
      assert(VA.isMemLoc());
      ArgValue = LowerMemArgument(Chain, CallConv, Ins, dl, DAG, VA, MFI, i);
    }

    // If value is passed via pointer - do a load.
    if (VA.getLocInfo() == CCValAssign::Indirect)
      ArgValue = DAG.getLoad(VA.getValVT(), dl, Chain, ArgValue, NULL, 0);

    InVals.push_back(ArgValue);
  }

  // The x86-64 ABI for returning structs by value requires that we copy
  // the sret argument into %rax for the return. Save the argument into
  // a virtual register so that we can access it from the return points.
  if (Is64Bit && MF.getFunction()->hasStructRetAttr()) {
    X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
    unsigned Reg = FuncInfo->getSRetReturnReg();
    if (!Reg) {
      Reg = MF.getRegInfo().createVirtualRegister(getRegClassFor(MVT::i64));
      FuncInfo->setSRetReturnReg(Reg);
    }
    SDValue Copy = DAG.getCopyToReg(DAG.getEntryNode(), dl, Reg, InVals[0]);
    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Copy, Chain);
  }

  unsigned StackSize = CCInfo.getNextStackOffset();
  // align stack specially for tail calls
  if (PerformTailCallOpt && CallConv == CallingConv::Fast)
    StackSize = GetAlignedArgumentStackSize(StackSize, DAG);

  // If the function takes variable number of arguments, make a frame index for
  // the start of the first vararg value... for expansion of llvm.va_start.
  if (isVarArg) {
    if (Is64Bit || CallConv != CallingConv::X86_FastCall) {
      VarArgsFrameIndex = MFI->CreateFixedObject(1, StackSize);
    }
    if (Is64Bit) {
      unsigned TotalNumIntRegs = 0, TotalNumXMMRegs = 0;

      // FIXME: We should really autogenerate these arrays
      static const unsigned GPR64ArgRegsWin64[] = {
        X86::RCX, X86::RDX, X86::R8,  X86::R9
      };
      static const unsigned XMMArgRegsWin64[] = {
        X86::XMM0, X86::XMM1, X86::XMM2, X86::XMM3
      };
      static const unsigned GPR64ArgRegs64Bit[] = {
        X86::RDI, X86::RSI, X86::RDX, X86::RCX, X86::R8, X86::R9
      };
      static const unsigned XMMArgRegs64Bit[] = {
        X86::XMM0, X86::XMM1, X86::XMM2, X86::XMM3,
        X86::XMM4, X86::XMM5, X86::XMM6, X86::XMM7
      };
      const unsigned *GPR64ArgRegs, *XMMArgRegs;

      if (IsWin64) {
        TotalNumIntRegs = 4; TotalNumXMMRegs = 4;
        GPR64ArgRegs = GPR64ArgRegsWin64;
        XMMArgRegs = XMMArgRegsWin64;
      } else {
        TotalNumIntRegs = 6; TotalNumXMMRegs = 8;
        GPR64ArgRegs = GPR64ArgRegs64Bit;
        XMMArgRegs = XMMArgRegs64Bit;
      }
      unsigned NumIntRegs = CCInfo.getFirstUnallocated(GPR64ArgRegs,
                                                       TotalNumIntRegs);
      unsigned NumXMMRegs = CCInfo.getFirstUnallocated(XMMArgRegs,
                                                       TotalNumXMMRegs);

      bool NoImplicitFloatOps = Fn->hasFnAttr(Attribute::NoImplicitFloat);
      assert(!(NumXMMRegs && !Subtarget->hasSSE1()) &&
             "SSE register cannot be used when SSE is disabled!");
      assert(!(NumXMMRegs && UseSoftFloat && NoImplicitFloatOps) &&
             "SSE register cannot be used when SSE is disabled!");
      if (UseSoftFloat || NoImplicitFloatOps || !Subtarget->hasSSE1())
        // Kernel mode asks for SSE to be disabled, so don't push them
        // on the stack.
        TotalNumXMMRegs = 0;

      // For X86-64, if there are vararg parameters that are passed via
      // registers, then we must store them to their spots on the stack so they
      // may be loaded by deferencing the result of va_next.
      VarArgsGPOffset = NumIntRegs * 8;
      VarArgsFPOffset = TotalNumIntRegs * 8 + NumXMMRegs * 16;
      RegSaveFrameIndex = MFI->CreateStackObject(TotalNumIntRegs * 8 +
                                                 TotalNumXMMRegs * 16, 16);

      // Store the integer parameter registers.
      SmallVector<SDValue, 8> MemOps;
      SDValue RSFIN = DAG.getFrameIndex(RegSaveFrameIndex, getPointerTy());
      unsigned Offset = VarArgsGPOffset;
      for (; NumIntRegs != TotalNumIntRegs; ++NumIntRegs) {
        SDValue FIN = DAG.getNode(ISD::ADD, dl, getPointerTy(), RSFIN,
                                  DAG.getIntPtrConstant(Offset));
        unsigned VReg = MF.addLiveIn(GPR64ArgRegs[NumIntRegs],
                                     X86::GR64RegisterClass);
        SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::i64);
        SDValue Store =
          DAG.getStore(Val.getValue(1), dl, Val, FIN,
                       PseudoSourceValue::getFixedStack(RegSaveFrameIndex),
                       Offset);
        MemOps.push_back(Store);
        Offset += 8;
      }

      if (TotalNumXMMRegs != 0 && NumXMMRegs != TotalNumXMMRegs) {
        // Now store the XMM (fp + vector) parameter registers.
        SmallVector<SDValue, 11> SaveXMMOps;
        SaveXMMOps.push_back(Chain);

        unsigned AL = MF.addLiveIn(X86::AL, X86::GR8RegisterClass);
        SDValue ALVal = DAG.getCopyFromReg(DAG.getEntryNode(), dl, AL, MVT::i8);
        SaveXMMOps.push_back(ALVal);

        SaveXMMOps.push_back(DAG.getIntPtrConstant(RegSaveFrameIndex));
        SaveXMMOps.push_back(DAG.getIntPtrConstant(VarArgsFPOffset));

        for (; NumXMMRegs != TotalNumXMMRegs; ++NumXMMRegs) {
          unsigned VReg = MF.addLiveIn(XMMArgRegs[NumXMMRegs],
                                       X86::VR128RegisterClass);
          SDValue Val = DAG.getCopyFromReg(Chain, dl, VReg, MVT::v4f32);
          SaveXMMOps.push_back(Val);
        }
        MemOps.push_back(DAG.getNode(X86ISD::VASTART_SAVE_XMM_REGS, dl,
                                     MVT::Other,
                                     &SaveXMMOps[0], SaveXMMOps.size()));
      }

      if (!MemOps.empty())
        Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
                            &MemOps[0], MemOps.size());
    }
  }

  // Some CCs need callee pop.
  if (IsCalleePop(isVarArg, CallConv)) {
    BytesToPopOnReturn  = StackSize; // Callee pops everything.
    BytesCallerReserves = 0;
  } else {
    BytesToPopOnReturn  = 0; // Callee pops nothing.
    // If this is an sret function, the return should pop the hidden pointer.
    if (!Is64Bit && CallConv != CallingConv::Fast && ArgsAreStructReturn(Ins))
      BytesToPopOnReturn = 4;
    BytesCallerReserves = StackSize;
  }

  if (!Is64Bit) {
    RegSaveFrameIndex = 0xAAAAAAA;   // RegSaveFrameIndex is X86-64 only.
    if (CallConv == CallingConv::X86_FastCall)
      VarArgsFrameIndex = 0xAAAAAAA;   // fastcc functions can't have varargs.
  }

  FuncInfo->setBytesToPopOnReturn(BytesToPopOnReturn);

  return Chain;
}

SDValue
X86TargetLowering::LowerMemOpCallTo(SDValue Chain,
                                    SDValue StackPtr, SDValue Arg,
                                    DebugLoc dl, SelectionDAG &DAG,
                                    const CCValAssign &VA,
                                    ISD::ArgFlagsTy Flags) {
  const unsigned FirstStackArgOffset = (Subtarget->isTargetWin64() ? 32 : 0);
  unsigned LocMemOffset = FirstStackArgOffset + VA.getLocMemOffset();
  SDValue PtrOff = DAG.getIntPtrConstant(LocMemOffset);
  PtrOff = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, PtrOff);
  if (Flags.isByVal()) {
    return CreateCopyOfByValArgument(Arg, PtrOff, Chain, Flags, DAG, dl);
  }
  return DAG.getStore(Chain, dl, Arg, PtrOff,
                      PseudoSourceValue::getStack(), LocMemOffset);
}

/// EmitTailCallLoadRetAddr - Emit a load of return address if tail call
/// optimization is performed and it is required.
SDValue
X86TargetLowering::EmitTailCallLoadRetAddr(SelectionDAG &DAG,
                                           SDValue &OutRetAddr,
                                           SDValue Chain,
                                           bool IsTailCall,
                                           bool Is64Bit,
                                           int FPDiff,
                                           DebugLoc dl) {
  if (!IsTailCall || FPDiff==0) return Chain;

  // Adjust the Return address stack slot.
  EVT VT = getPointerTy();
  OutRetAddr = getReturnAddressFrameIndex(DAG);

  // Load the "old" Return address.
  OutRetAddr = DAG.getLoad(VT, dl, Chain, OutRetAddr, NULL, 0);
  return SDValue(OutRetAddr.getNode(), 1);
}

/// EmitTailCallStoreRetAddr - Emit a store of the return adress if tail call
/// optimization is performed and it is required (FPDiff!=0).
static SDValue
EmitTailCallStoreRetAddr(SelectionDAG & DAG, MachineFunction &MF,
                         SDValue Chain, SDValue RetAddrFrIdx,
                         bool Is64Bit, int FPDiff, DebugLoc dl) {
  // Store the return address to the appropriate stack slot.
  if (!FPDiff) return Chain;
  // Calculate the new stack slot for the return address.
  int SlotSize = Is64Bit ? 8 : 4;
  int NewReturnAddrFI =
    MF.getFrameInfo()->CreateFixedObject(SlotSize, FPDiff-SlotSize);
  EVT VT = Is64Bit ? MVT::i64 : MVT::i32;
  SDValue NewRetAddrFrIdx = DAG.getFrameIndex(NewReturnAddrFI, VT);
  Chain = DAG.getStore(Chain, dl, RetAddrFrIdx, NewRetAddrFrIdx,
                       PseudoSourceValue::getFixedStack(NewReturnAddrFI), 0);
  return Chain;
}

SDValue
X86TargetLowering::LowerCall(SDValue Chain, SDValue Callee,
                             CallingConv::ID CallConv, bool isVarArg,
                             bool isTailCall,
                             const SmallVectorImpl<ISD::OutputArg> &Outs,
                             const SmallVectorImpl<ISD::InputArg> &Ins,
                             DebugLoc dl, SelectionDAG &DAG,
                             SmallVectorImpl<SDValue> &InVals) {

  MachineFunction &MF = DAG.getMachineFunction();
  bool Is64Bit        = Subtarget->is64Bit();
  bool IsStructRet    = CallIsStructReturn(Outs);

  assert((!isTailCall ||
          (CallConv == CallingConv::Fast && PerformTailCallOpt)) &&
         "IsEligibleForTailCallOptimization missed a case!");
  assert(!(isVarArg && CallConv == CallingConv::Fast) &&
         "Var args not supported with calling convention fastcc");

  // Analyze operands of the call, assigning locations to each operand.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
                 ArgLocs, *DAG.getContext());
  CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CallConv));

  // Get a count of how many bytes are to be pushed on the stack.
  unsigned NumBytes = CCInfo.getNextStackOffset();
  if (PerformTailCallOpt && CallConv == CallingConv::Fast)
    NumBytes = GetAlignedArgumentStackSize(NumBytes, DAG);

  int FPDiff = 0;
  if (isTailCall) {
    // Lower arguments at fp - stackoffset + fpdiff.
    unsigned NumBytesCallerPushed =
      MF.getInfo<X86MachineFunctionInfo>()->getBytesToPopOnReturn();
    FPDiff = NumBytesCallerPushed - NumBytes;

    // Set the delta of movement of the returnaddr stackslot.
    // But only set if delta is greater than previous delta.
    if (FPDiff < (MF.getInfo<X86MachineFunctionInfo>()->getTCReturnAddrDelta()))
      MF.getInfo<X86MachineFunctionInfo>()->setTCReturnAddrDelta(FPDiff);
  }

  Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(NumBytes, true));

  SDValue RetAddrFrIdx;
  // Load return adress for tail calls.
  Chain = EmitTailCallLoadRetAddr(DAG, RetAddrFrIdx, Chain, isTailCall, Is64Bit,
                                  FPDiff, dl);

  SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
  SmallVector<SDValue, 8> MemOpChains;
  SDValue StackPtr;

  // Walk the register/memloc assignments, inserting copies/loads.  In the case
  // of tail call optimization arguments are handle later.
  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    CCValAssign &VA = ArgLocs[i];
    EVT RegVT = VA.getLocVT();
    SDValue Arg = Outs[i].Val;
    ISD::ArgFlagsTy Flags = Outs[i].Flags;
    bool isByVal = Flags.isByVal();

    // Promote the value if needed.
    switch (VA.getLocInfo()) {
    default: llvm_unreachable("Unknown loc info!");
    case CCValAssign::Full: break;
    case CCValAssign::SExt:
      Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, RegVT, Arg);
      break;
    case CCValAssign::ZExt:
      Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, RegVT, Arg);
      break;
    case CCValAssign::AExt:
      if (RegVT.isVector() && RegVT.getSizeInBits() == 128) {
        // Special case: passing MMX values in XMM registers.
        Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, Arg);
        Arg = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, Arg);
        Arg = getMOVL(DAG, dl, MVT::v2i64, DAG.getUNDEF(MVT::v2i64), Arg);
      } else
        Arg = DAG.getNode(ISD::ANY_EXTEND, dl, RegVT, Arg);
      break;
    case CCValAssign::BCvt:
      Arg = DAG.getNode(ISD::BIT_CONVERT, dl, RegVT, Arg);
      break;
    case CCValAssign::Indirect: {
      // Store the argument.
      SDValue SpillSlot = DAG.CreateStackTemporary(VA.getValVT());
      int FI = cast<FrameIndexSDNode>(SpillSlot)->getIndex();
      Chain = DAG.getStore(Chain, dl, Arg, SpillSlot,
                           PseudoSourceValue::getFixedStack(FI), 0);
      Arg = SpillSlot;
      break;
    }
    }

    if (VA.isRegLoc()) {
      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
    } else {
      if (!isTailCall || (isTailCall && isByVal)) {
        assert(VA.isMemLoc());
        if (StackPtr.getNode() == 0)
          StackPtr = DAG.getCopyFromReg(Chain, dl, X86StackPtr, getPointerTy());

        MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Arg,
                                               dl, DAG, VA, Flags));
      }
    }
  }

  if (!MemOpChains.empty())
    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
                        &MemOpChains[0], MemOpChains.size());

  // Build a sequence of copy-to-reg nodes chained together with token chain
  // and flag operands which copy the outgoing args into registers.
  SDValue InFlag;
  // Tail call byval lowering might overwrite argument registers so in case of
  // tail call optimization the copies to registers are lowered later.
  if (!isTailCall)
    for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
      Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
                               RegsToPass[i].second, InFlag);
      InFlag = Chain.getValue(1);
    }


  if (Subtarget->isPICStyleGOT()) {
    // ELF / PIC requires GOT in the EBX register before function calls via PLT
    // GOT pointer.
    if (!isTailCall) {
      Chain = DAG.getCopyToReg(Chain, dl, X86::EBX,
                               DAG.getNode(X86ISD::GlobalBaseReg,
                                           DebugLoc::getUnknownLoc(),
                                           getPointerTy()),
                               InFlag);
      InFlag = Chain.getValue(1);
    } else {
      // If we are tail calling and generating PIC/GOT style code load the
      // address of the callee into ECX. The value in ecx is used as target of
      // the tail jump. This is done to circumvent the ebx/callee-saved problem
      // for tail calls on PIC/GOT architectures. Normally we would just put the
      // address of GOT into ebx and then call target@PLT. But for tail calls
      // ebx would be restored (since ebx is callee saved) before jumping to the
      // target@PLT.

      // Note: The actual moving to ECX is done further down.
      GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee);
      if (G && !G->getGlobal()->hasHiddenVisibility() &&
          !G->getGlobal()->hasProtectedVisibility())
        Callee = LowerGlobalAddress(Callee, DAG);
      else if (isa<ExternalSymbolSDNode>(Callee))
        Callee = LowerExternalSymbol(Callee, DAG);
    }
  }

  if (Is64Bit && isVarArg) {
    // From AMD64 ABI document:
    // For calls that may call functions that use varargs or stdargs
    // (prototype-less calls or calls to functions containing ellipsis (...) in
    // the declaration) %al is used as hidden argument to specify the number
    // of SSE registers used. The contents of %al do not need to match exactly
    // the number of registers, but must be an ubound on the number of SSE
    // registers used and is in the range 0 - 8 inclusive.

    // FIXME: Verify this on Win64
    // Count the number of XMM registers allocated.
    static const unsigned XMMArgRegs[] = {
      X86::XMM0, X86::XMM1, X86::XMM2, X86::XMM3,
      X86::XMM4, X86::XMM5, X86::XMM6, X86::XMM7
    };
    unsigned NumXMMRegs = CCInfo.getFirstUnallocated(XMMArgRegs, 8);
    assert((Subtarget->hasSSE1() || !NumXMMRegs)
           && "SSE registers cannot be used when SSE is disabled");

    Chain = DAG.getCopyToReg(Chain, dl, X86::AL,
                             DAG.getConstant(NumXMMRegs, MVT::i8), InFlag);
    InFlag = Chain.getValue(1);
  }


  // For tail calls lower the arguments to the 'real' stack slot.
  if (isTailCall) {
    // Force all the incoming stack arguments to be loaded from the stack
    // before any new outgoing arguments are stored to the stack, because the
    // outgoing stack slots may alias the incoming argument stack slots, and
    // the alias isn't otherwise explicit. This is slightly more conservative
    // than necessary, because it means that each store effectively depends
    // on every argument instead of just those arguments it would clobber.
    SDValue ArgChain = DAG.getStackArgumentTokenFactor(Chain);

    SmallVector<SDValue, 8> MemOpChains2;
    SDValue FIN;
    int FI = 0;
    // Do not flag preceeding copytoreg stuff together with the following stuff.
    InFlag = SDValue();
    for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
      CCValAssign &VA = ArgLocs[i];
      if (!VA.isRegLoc()) {
        assert(VA.isMemLoc());
        SDValue Arg = Outs[i].Val;
        ISD::ArgFlagsTy Flags = Outs[i].Flags;
        // Create frame index.
        int32_t Offset = VA.getLocMemOffset()+FPDiff;
        uint32_t OpSize = (VA.getLocVT().getSizeInBits()+7)/8;
        FI = MF.getFrameInfo()->CreateFixedObject(OpSize, Offset);
        FIN = DAG.getFrameIndex(FI, getPointerTy());

        if (Flags.isByVal()) {
          // Copy relative to framepointer.
          SDValue Source = DAG.getIntPtrConstant(VA.getLocMemOffset());
          if (StackPtr.getNode() == 0)
            StackPtr = DAG.getCopyFromReg(Chain, dl, X86StackPtr,
                                          getPointerTy());
          Source = DAG.getNode(ISD::ADD, dl, getPointerTy(), StackPtr, Source);

          MemOpChains2.push_back(CreateCopyOfByValArgument(Source, FIN,
                                                           ArgChain,
                                                           Flags, DAG, dl));
        } else {
          // Store relative to framepointer.
          MemOpChains2.push_back(
            DAG.getStore(ArgChain, dl, Arg, FIN,
                         PseudoSourceValue::getFixedStack(FI), 0));
        }
      }
    }

    if (!MemOpChains2.empty())
      Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
                          &MemOpChains2[0], MemOpChains2.size());

    // Copy arguments to their registers.
    for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
      Chain = DAG.getCopyToReg(Chain, dl, RegsToPass[i].first,
                               RegsToPass[i].second, InFlag);
      InFlag = Chain.getValue(1);
    }
    InFlag =SDValue();

    // Store the return address to the appropriate stack slot.
    Chain = EmitTailCallStoreRetAddr(DAG, MF, Chain, RetAddrFrIdx, Is64Bit,
                                     FPDiff, dl);
  }

  // If the callee is a GlobalAddress node (quite common, every direct call is)
  // turn it into a TargetGlobalAddress node so that legalize doesn't hack it.
  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
    // We should use extra load for direct calls to dllimported functions in
    // non-JIT mode.
    GlobalValue *GV = G->getGlobal();
    if (!GV->hasDLLImportLinkage()) {
      unsigned char OpFlags = 0;

      // On ELF targets, in both X86-64 and X86-32 mode, direct calls to
      // external symbols most go through the PLT in PIC mode.  If the symbol
      // has hidden or protected visibility, or if it is static or local, then
      // we don't need to use the PLT - we can directly call it.
      if (Subtarget->isTargetELF() &&
          getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
          GV->hasDefaultVisibility() && !GV->hasLocalLinkage()) {
        OpFlags = X86II::MO_PLT;
      } else if (Subtarget->isPICStyleStubAny() &&
               (GV->isDeclaration() || GV->isWeakForLinker()) &&
               Subtarget->getDarwinVers() < 9) {
        // PC-relative references to external symbols should go through $stub,
        // unless we're building with the leopard linker or later, which
        // automatically synthesizes these stubs.
        OpFlags = X86II::MO_DARWIN_STUB;
      }

      Callee = DAG.getTargetGlobalAddress(GV, getPointerTy(),
                                          G->getOffset(), OpFlags);
    }
  } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
    unsigned char OpFlags = 0;

    // On ELF targets, in either X86-64 or X86-32 mode, direct calls to external
    // symbols should go through the PLT.
    if (Subtarget->isTargetELF() &&
        getTargetMachine().getRelocationModel() == Reloc::PIC_) {
      OpFlags = X86II::MO_PLT;
    } else if (Subtarget->isPICStyleStubAny() &&
             Subtarget->getDarwinVers() < 9) {
      // PC-relative references to external symbols should go through $stub,
      // unless we're building with the leopard linker or later, which
      // automatically synthesizes these stubs.
      OpFlags = X86II::MO_DARWIN_STUB;
    }

    Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy(),
                                         OpFlags);
  } else if (isTailCall) {
    unsigned Opc = Is64Bit ? X86::R11 : X86::EAX;

    Chain = DAG.getCopyToReg(Chain,  dl,
                             DAG.getRegister(Opc, getPointerTy()),
                             Callee,InFlag);
    Callee = DAG.getRegister(Opc, getPointerTy());
    // Add register as live out.
    MF.getRegInfo().addLiveOut(Opc);
  }

  // Returns a chain & a flag for retval copy to use.
  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
  SmallVector<SDValue, 8> Ops;

  if (isTailCall) {
    Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(NumBytes, true),
                           DAG.getIntPtrConstant(0, true), InFlag);
    InFlag = Chain.getValue(1);
  }

  Ops.push_back(Chain);
  Ops.push_back(Callee);

  if (isTailCall)
    Ops.push_back(DAG.getConstant(FPDiff, MVT::i32));

  // Add argument registers to the end of the list so that they are known live
  // into the call.
  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
    Ops.push_back(DAG.getRegister(RegsToPass[i].first,
                                  RegsToPass[i].second.getValueType()));

  // Add an implicit use GOT pointer in EBX.