Exemple #1
0
// Write ByVal Arg to arg registers and stack.
static void
WriteByValArg(SDValue& ByValChain, SDValue Chain, SDLoc DL,
              SmallVector<std::pair<unsigned, SDValue>, 16>& RegsToPass,
              SmallVector<SDValue, 8>& MemOpChains, int& LastFI,
              MachineFrameInfo *MFI, SelectionDAG &DAG, SDValue Arg,
              const CCValAssign &VA, const ISD::ArgFlagsTy& Flags,
              MVT PtrType, bool isLittle) {
  unsigned LocMemOffset = VA.getLocMemOffset();
  unsigned Offset = 0;
  uint32_t RemainingSize = Flags.getByValSize();
  unsigned ByValAlign = Flags.getByValAlign();

  if (RemainingSize == 0)
    return;

  // Create a fixed object on stack at offset LocMemOffset and copy
  // remaining part of byval arg to it using memcpy.
  SDValue Src = DAG.getNode(ISD::ADD, DL, MVT::i32, Arg,
                            DAG.getConstant(Offset, MVT::i32));
  LastFI = MFI->CreateFixedObject(RemainingSize, LocMemOffset, true);
  SDValue Dst = DAG.getFrameIndex(LastFI, PtrType);
  ByValChain = DAG.getMemcpy(ByValChain, DL, Dst, Src,
                             DAG.getConstant(RemainingSize, MVT::i32),
                             std::min(ByValAlign, (unsigned)4),
                             /*isVolatile=*/false, /*AlwaysInline=*/false,
                             MachinePointerInfo(0), MachinePointerInfo(0));
}
// HandleByVal - Allocate space on the stack large enough to pass an argument
// by value. The size and alignment information of the argument is encoded in
// its parameter attribute.
void CCState::HandleByVal(unsigned ValNo, MVT ValVT,
                          MVT LocVT, CCValAssign::LocInfo LocInfo,
                          int MinSize, int MinAlign,
                          ISD::ArgFlagsTy ArgFlags) {
  unsigned Align = ArgFlags.getByValAlign();
  unsigned Size  = ArgFlags.getByValSize();
  if (MinSize > (int)Size)
    Size = MinSize;
  if (MinAlign > (int)Align)
    Align = MinAlign;
  MF.getFrameInfo()->ensureMaxAlignment(Align);
  TM.getTargetLowering()->HandleByVal(this, Size, Align);
  unsigned Offset = AllocateStack(Size, Align);
  addLoc(CCValAssign::getMem(ValNo, ValVT, Offset, LocVT, LocInfo));
}
// HandleByVal - Allocate a stack slot large enough to pass an argument by
// value. The size and alignment information of the argument is encoded in its
// parameter attribute.
void Hexagon_CCState::HandleByVal(unsigned ValNo, EVT ValVT,
                                EVT LocVT, CCValAssign::LocInfo LocInfo,
                                int MinSize, int MinAlign,
                                ISD::ArgFlagsTy ArgFlags) {
  unsigned Align = ArgFlags.getByValAlign();
  unsigned Size  = ArgFlags.getByValSize();
  if (MinSize > (int)Size)
    Size = MinSize;
  if (MinAlign > (int)Align)
    Align = MinAlign;
  unsigned Offset = AllocateStack(Size, Align);

  addLoc(CCValAssign::getMem(ValNo, ValVT.getSimpleVT(), Offset,
                             LocVT.getSimpleVT(), LocInfo));
}
Exemple #4
0
/// LowerCCCCallTo - functions arguments are copied from virtual regs to
/// (physical regs)/(stack frame), CALLSEQ_START and CALLSEQ_END are emitted.
// TODO: sret.
SDValue MSP430TargetLowering::LowerCCCCallTo(
    SDValue Chain, SDValue Callee, CallingConv::ID CallConv, bool isVarArg,
    bool isTailCall, const SmallVectorImpl<ISD::OutputArg> &Outs,
    const SmallVectorImpl<SDValue> &OutVals,
    const SmallVectorImpl<ISD::InputArg> &Ins, const SDLoc &dl,
    SelectionDAG &DAG, SmallVectorImpl<SDValue> &InVals) const {
  // Analyze operands of the call, assigning locations to each operand.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                 *DAG.getContext());
  AnalyzeArguments(CCInfo, ArgLocs, Outs);

  // Get a count of how many bytes are to be pushed on the stack.
  unsigned NumBytes = CCInfo.getNextStackOffset();
  auto PtrVT = getPointerTy(DAG.getDataLayout());

  Chain = DAG.getCALLSEQ_START(Chain,
                               DAG.getConstant(NumBytes, dl, PtrVT, true), dl);

  SmallVector<std::pair<unsigned, SDValue>, 4> RegsToPass;
  SmallVector<SDValue, 12> MemOpChains;
  SDValue StackPtr;

  // Walk the register/memloc assignments, inserting copies/loads.
  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    CCValAssign &VA = ArgLocs[i];

    SDValue Arg = OutVals[i];

    // 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, VA.getLocVT(), Arg);
        break;
      case CCValAssign::ZExt:
        Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
        break;
      case CCValAssign::AExt:
        Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
        break;
    }

    // Arguments that can be passed on register must be kept at RegsToPass
    // vector
    if (VA.isRegLoc()) {
      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
    } else {
      assert(VA.isMemLoc());

      if (!StackPtr.getNode())
        StackPtr = DAG.getCopyFromReg(Chain, dl, MSP430::SP, PtrVT);

      SDValue PtrOff =
          DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr,
                      DAG.getIntPtrConstant(VA.getLocMemOffset(), dl));

      SDValue MemOp;
      ISD::ArgFlagsTy Flags = Outs[i].Flags;

      if (Flags.isByVal()) {
        SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), dl, MVT::i16);
        MemOp = DAG.getMemcpy(Chain, dl, PtrOff, Arg, SizeNode,
                              Flags.getByValAlign(),
                              /*isVolatile*/false,
                              /*AlwaysInline=*/true,
                              /*isTailCall=*/false,
                              MachinePointerInfo(),
                              MachinePointerInfo());
      } else {
        MemOp = DAG.getStore(Chain, dl, Arg, PtrOff, MachinePointerInfo(),
                             false, false, 0);
      }

      MemOpChains.push_back(MemOp);
    }
  }

  // Transform all store nodes into one single node because all store nodes are
  // independent of each other.
  if (!MemOpChains.empty())
    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);

  // Build a sequence of copy-to-reg nodes chained together with token chain and
  // flag operands which copy the outgoing args into registers.  The InFlag in
  // necessary since all emitted instructions must be stuck together.
  SDValue InFlag;
  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 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.
  // Likewise ExternalSymbol -> TargetExternalSymbol.
  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
    Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, MVT::i16);
  else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee))
    Callee = DAG.getTargetExternalSymbol(E->getSymbol(), MVT::i16);

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

  // 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()));

  if (InFlag.getNode())
    Ops.push_back(InFlag);

  Chain = DAG.getNode(MSP430ISD::CALL, dl, NodeTys, Ops);
  InFlag = Chain.getValue(1);

  // Create the CALLSEQ_END node.
  Chain = DAG.getCALLSEQ_END(Chain, DAG.getConstant(NumBytes, dl, PtrVT, true),
                             DAG.getConstant(0, dl, PtrVT, true), InFlag, dl);
  InFlag = Chain.getValue(1);

  // Handle result values, copying them out of physregs into vregs that we
  // return.
  return LowerCallResult(Chain, InFlag, CallConv, isVarArg, Ins, dl,
                         DAG, InVals);
}
// LowerCCCCallTo - functions arguments are copied from virtual regs to
// (physical regs)/(stack frame), CALLSEQ_START and CALLSEQ_END are emitted.
SDValue LanaiTargetLowering::LowerCCCCallTo(
    SDValue Chain, SDValue Callee, CallingConv::ID CallConv, bool IsVarArg,
    bool IsTailCall, const SmallVectorImpl<ISD::OutputArg> &Outs,
    const SmallVectorImpl<SDValue> &OutVals,
    const SmallVectorImpl<ISD::InputArg> &Ins, SDLoc DL, SelectionDAG &DAG,
    SmallVectorImpl<SDValue> &InVals) const {
  // Analyze operands of the call, assigning locations to each operand.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, IsVarArg, DAG.getMachineFunction(), ArgLocs,
                 *DAG.getContext());
  GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee);
  MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();

  NumFixedArgs = 0;
  if (IsVarArg && G) {
    const Function *CalleeFn = dyn_cast<Function>(G->getGlobal());
    if (CalleeFn)
      NumFixedArgs = CalleeFn->getFunctionType()->getNumParams();
  }
  if (NumFixedArgs)
    CCInfo.AnalyzeCallOperands(Outs, CC_Lanai32_VarArg);
  else {
    if (CallConv == CallingConv::Fast)
      CCInfo.AnalyzeCallOperands(Outs, CC_Lanai32_Fast);
    else
      CCInfo.AnalyzeCallOperands(Outs, CC_Lanai32);
  }

  // Get a count of how many bytes are to be pushed on the stack.
  unsigned NumBytes = CCInfo.getNextStackOffset();

  // Create local copies for byval args.
  SmallVector<SDValue, 8> ByValArgs;
  for (unsigned I = 0, E = Outs.size(); I != E; ++I) {
    ISD::ArgFlagsTy Flags = Outs[I].Flags;
    if (!Flags.isByVal())
      continue;

    SDValue Arg = OutVals[I];
    unsigned Size = Flags.getByValSize();
    unsigned Align = Flags.getByValAlign();

    int FI = MFI->CreateStackObject(Size, Align, false);
    SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy(DAG.getDataLayout()));
    SDValue SizeNode = DAG.getConstant(Size, DL, MVT::i32);

    Chain = DAG.getMemcpy(Chain, DL, FIPtr, Arg, SizeNode, Align,
                          /*IsVolatile=*/false,
                          /*AlwaysInline=*/false,
                          /*IsTailCall=*/false, MachinePointerInfo(),
                          MachinePointerInfo());
    ByValArgs.push_back(FIPtr);
  }

  Chain = DAG.getCALLSEQ_START(
      Chain,
      DAG.getConstant(NumBytes, DL, getPointerTy(DAG.getDataLayout()), true),
      DL);

  SmallVector<std::pair<unsigned, SDValue>, 4> RegsToPass;
  SmallVector<SDValue, 12> MemOpChains;
  SDValue StackPtr;

  // Walk the register/memloc assignments, inserting copies/loads.
  for (unsigned I = 0, J = 0, E = ArgLocs.size(); I != E; ++I) {
    CCValAssign &VA = ArgLocs[I];
    SDValue Arg = OutVals[I];
    ISD::ArgFlagsTy Flags = Outs[I].Flags;

    // Promote the value if needed.
    switch (VA.getLocInfo()) {
    case CCValAssign::Full:
      break;
    case CCValAssign::SExt:
      Arg = DAG.getNode(ISD::SIGN_EXTEND, DL, VA.getLocVT(), Arg);
      break;
    case CCValAssign::ZExt:
      Arg = DAG.getNode(ISD::ZERO_EXTEND, DL, VA.getLocVT(), Arg);
      break;
    case CCValAssign::AExt:
      Arg = DAG.getNode(ISD::ANY_EXTEND, DL, VA.getLocVT(), Arg);
      break;
    default:
      llvm_unreachable("Unknown loc info!");
    }

    // Use local copy if it is a byval arg.
    if (Flags.isByVal())
      Arg = ByValArgs[J++];

    // Arguments that can be passed on register must be kept at RegsToPass
    // vector
    if (VA.isRegLoc()) {
      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
    } else {
      assert(VA.isMemLoc());

      if (StackPtr.getNode() == 0)
        StackPtr = DAG.getCopyFromReg(Chain, DL, Lanai::SP,
                                      getPointerTy(DAG.getDataLayout()));

      SDValue PtrOff =
          DAG.getNode(ISD::ADD, DL, getPointerTy(DAG.getDataLayout()), StackPtr,
                      DAG.getIntPtrConstant(VA.getLocMemOffset(), DL));

      MemOpChains.push_back(DAG.getStore(
          Chain, DL, Arg, PtrOff, MachinePointerInfo(), false, false, 0));
    }
  }

  // Transform all store nodes into one single node because all store nodes are
  // independent of each other.
  if (!MemOpChains.empty())
    Chain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
                        ArrayRef<SDValue>(&MemOpChains[0], MemOpChains.size()));

  SDValue InFlag;

  // Build a sequence of copy-to-reg nodes chained together with token chain and
  // flag operands which copy the outgoing args into registers.  The InFlag in
  // necessary since all emitted instructions must be stuck together.
  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 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.
  // Likewise ExternalSymbol -> TargetExternalSymbol.
  uint8_t OpFlag = LanaiII::MO_NO_FLAG;
  if (G) {
    Callee = DAG.getTargetGlobalAddress(
        G->getGlobal(), DL, getPointerTy(DAG.getDataLayout()), 0, OpFlag);
  } else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee)) {
    Callee = DAG.getTargetExternalSymbol(
        E->getSymbol(), getPointerTy(DAG.getDataLayout()), OpFlag);
  }

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

  // Add a register mask operand representing the call-preserved registers.
  // TODO: Should return-twice functions be handled?
  const uint32_t *Mask =
      TRI->getCallPreservedMask(DAG.getMachineFunction(), CallConv);
  assert(Mask && "Missing call preserved mask for calling convention");
  Ops.push_back(DAG.getRegisterMask(Mask));

  // 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()));

  if (InFlag.getNode())
    Ops.push_back(InFlag);

  Chain = DAG.getNode(LanaiISD::CALL, DL, NodeTys,
                      ArrayRef<SDValue>(&Ops[0], Ops.size()));
  InFlag = Chain.getValue(1);

  // Create the CALLSEQ_END node.
  Chain = DAG.getCALLSEQ_END(
      Chain,
      DAG.getConstant(NumBytes, DL, getPointerTy(DAG.getDataLayout()), true),
      DAG.getConstant(0, DL, getPointerTy(DAG.getDataLayout()), true), InFlag,
      DL);
  InFlag = Chain.getValue(1);

  // Handle result values, copying them out of physregs into vregs that we
  // return.
  return LowerCallResult(Chain, InFlag, CallConv, IsVarArg, Ins, DL, DAG,
                         InVals);
}
Exemple #6
0
SDValue Y86TargetLowering::LowerCall(CallLoweringInfo &CLI,
                                     SmallVectorImpl<SDValue> &InVals) const {
  bool &isTailCall = CLI.IsTailCall;

  if (isTailCall) {
    // Here we would check whether we can actually eliminate the tail
    // call. But since we don't support it yet, we reject the opportunity
    // and also throw an error in the case there's an assumed expectation.
    if (CLI.CS && CLI.CS->isMustTailCall())
      report_fatal_error("tail call elimination not yet supported");
    isTailCall = false;
  }

  bool isVarArg = CLI.IsVarArg;
  SelectionDAG &DAG = CLI.DAG;
  SmallVectorImpl<ISD::OutputArg> &Outs = CLI.Outs;
  CallingConv::ID CallConv = CLI.CallConv;
  MachineFunction &MFunction = DAG.getMachineFunction();

  if (isVarArg)
    report_fatal_error("varargs not yet supported");

  // Gather info about call operands.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, isVarArg, MFunction, ArgLocs, *DAG.getContext());
  CCInfo.AnalyzeCallOperands(Outs, CC_Y86);

  // Get a count of how many bytes are to be pushed on the stack.
  unsigned NumBytes = CCInfo.getNextStackOffset();

  // Start the calling sequence.
  SDValue Chain = CLI.Chain;
  SDLoc &dl = CLI.DL;
  Chain = DAG.getCALLSEQ_START(Chain,
                               DAG.getIntPtrConstant(NumBytes, true), dl);

  // Walk the register/memloc assignments and treat them for argument
  // passing (intermediate nodes maybe be used for chaining).
  const Y86RegisterInfo *RegInfo =
    static_cast<const Y86RegisterInfo*>(DAG.getSubtarget().getRegisterInfo());
  SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
  SmallVector<SDValue, 8> MemOpChains;
  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    // Skip inalloca arguments, they have already been written.
    ISD::ArgFlagsTy Flags = Outs[i].Flags;
    if (Flags.isInAlloca())
      continue;

    SmallVectorImpl<SDValue> &OutVals = CLI.OutVals;
    CCValAssign &VA = ArgLocs[i];
    SDValue Arg = OutVals[i];
    if (VA.isRegLoc()) {
      // Registers and corresponding arguments are added to a separate
      // list so they are chained all together later on.
      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
    } else {
      assert(VA.isMemLoc());
      if (Flags.isByVal() || !isTailCall) {
        // First create a node to copy the arguments from a register (they
        // are accessed as a displacement from ESP).
        SDValue StackRegister =
            DAG.getCopyFromReg(Chain, dl, RegInfo->getStackRegister(),
                               getPointerTy());

        // The add node is used to adjust the offset to the exact location.
        unsigned MemLocOffset = VA.getLocMemOffset();
        SDValue MemLoc = DAG.getIntPtrConstant(MemLocOffset);
        MemLoc = DAG.getNode(ISD::ADD, dl, getPointerTy(),
                             StackRegister, MemLoc);

        // Complete the actual memory access.
        SDValue MemAccess;
        if (Flags.isByVal()) {
          SDValue SizeNode = DAG.getConstant(Flags.getByValSize(), MVT::i32);
          MemAccess = DAG.getMemcpy(Chain, dl, MemLoc, Arg, SizeNode,
                                    Flags.getByValAlign(), false, true,
                                    MachinePointerInfo(), MachinePointerInfo());
        } else {
          MemAccess = DAG.getStore(Chain, dl, Arg, MemLoc,
                                   MachinePointerInfo::getStack(MemLocOffset),
                                   false, false, 0);
        }
        MemOpChains.push_back(MemAccess);
      }
    }
  }

  if (!MemOpChains.empty())
    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);

  // 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;
  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
    Chain = DAG.getCopyToReg(Chain, dl,
                             RegsToPass[i].first,   // Reg. assignment.
                             RegsToPass[i].second,  // Argument.
                             InFlag);
    InFlag = Chain.getValue(1);
  }

  SDValue Callee = CLI.Callee;
  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
    // 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.
    const GlobalValue *GV = G->getGlobal();
    Callee = DAG.getTargetGlobalAddress(GV, dl, getPointerTy(),
                                        G->getOffset());
  } else {
    llvm_unreachable("External symbol calls not yet implemented");
  }

  // Fill-up call operands, beginning with the chain and callee.
  SmallVector<SDValue, 8> Ops;
  Ops.push_back(Chain);
  Ops.push_back(Callee);

  // 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 a register mask operand representing the call-preserved registers.
  const uint32_t *Mask = RegInfo->getCallPreservedMask(CallConv);
  assert(Mask && "Missing call preserved mask for calling convention");
  Ops.push_back(DAG.getRegisterMask(Mask));

  if (InFlag.getNode())
    Ops.push_back(InFlag);

  SDVTList VTs = DAG.getVTList(MVT::Other, MVT::Glue);
  Chain = DAG.getNode(Y86ISD::CALL, dl, VTs, Ops);
  InFlag = Chain.getValue(1);

  // Finish the calling sequence. First operand is the total number of bytes
  // to pop. The second operand is the fraction of those bytes to be popped
  // by the callee.
  unsigned NumBytesForCalleeToPop = 0;
  if (Y86::isCalleePop(CallConv, isVarArg, isTailCall))
    NumBytesForCalleeToPop = NumBytes;
  Chain = DAG.getCALLSEQ_END(Chain,
                             DAG.getIntPtrConstant(NumBytes, true),
                             DAG.getIntPtrConstant(NumBytesForCalleeToPop, true),
                             InFlag, dl);
  InFlag = Chain.getValue(1);

  SmallVectorImpl<ISD::InputArg> &Ins = CLI.Ins;
  return LowerCallResult(Chain, InFlag, CallConv, isVarArg,
                         Ins, dl, DAG, InVals);

}
SDValue
SparcTargetLowering::LowerCall(SDValue Chain, SDValue Callee,
                               CallingConv::ID CallConv, bool isVarArg,
                               bool &isTailCall,
                               const SmallVectorImpl<ISD::OutputArg> &Outs,
                               const SmallVectorImpl<SDValue> &OutVals,
                               const SmallVectorImpl<ISD::InputArg> &Ins,
                               DebugLoc dl, SelectionDAG &DAG,
                               SmallVectorImpl<SDValue> &InVals) const {
  // Sparc target does not yet support tail call optimization.
  isTailCall = false;

  // Analyze operands of the call, assigning locations to each operand.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CallConv, isVarArg, DAG.getTarget(), ArgLocs,
                 *DAG.getContext());
  CCInfo.AnalyzeCallOperands(Outs, CC_Sparc32);

  // Get the size of the outgoing arguments stack space requirement.
  unsigned ArgsSize = CCInfo.getNextStackOffset();

  // Keep stack frames 8-byte aligned.
  ArgsSize = (ArgsSize+7) & ~7;

  MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();

  //Create local copies for byval args.
  SmallVector<SDValue, 8> ByValArgs;
  for (unsigned i = 0,  e = Outs.size(); i != e; ++i) {
    ISD::ArgFlagsTy Flags = Outs[i].Flags;
    if (!Flags.isByVal())
      continue;

    SDValue Arg = OutVals[i];
    unsigned Size = Flags.getByValSize();
    unsigned Align = Flags.getByValAlign();

    int FI = MFI->CreateStackObject(Size, Align, false);
    SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy());
    SDValue SizeNode = DAG.getConstant(Size, MVT::i32);

    Chain = DAG.getMemcpy(Chain, dl, FIPtr, Arg, SizeNode, Align,
                          false,        //isVolatile,
                          (Size <= 32), //AlwaysInline if size <= 32
                          MachinePointerInfo(), MachinePointerInfo());
    ByValArgs.push_back(FIPtr);
  }

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

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

  const unsigned StackOffset = 92;
  bool hasStructRetAttr = false;
  // Walk the register/memloc assignments, inserting copies/loads.
  for (unsigned i = 0, realArgIdx = 0, byvalArgIdx = 0, e = ArgLocs.size();
       i != e;
       ++i, ++realArgIdx) {
    CCValAssign &VA = ArgLocs[i];
    SDValue Arg = OutVals[realArgIdx];

    ISD::ArgFlagsTy Flags = Outs[realArgIdx].Flags;

    //Use local copy if it is a byval arg.
    if (Flags.isByVal())
      Arg = ByValArgs[byvalArgIdx++];

    // 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, VA.getLocVT(), Arg);
      break;
    case CCValAssign::ZExt:
      Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
      break;
    case CCValAssign::AExt:
      Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
      break;
    case CCValAssign::BCvt:
      Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
      break;
    }

    if (Flags.isSRet()) {
      assert(VA.needsCustom());
      // store SRet argument in %sp+64
      SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
      SDValue PtrOff = DAG.getIntPtrConstant(64);
      PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
      MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
                                         MachinePointerInfo(),
                                         false, false, 0));
      hasStructRetAttr = true;
      continue;
    }

    if (VA.needsCustom()) {
      assert(VA.getLocVT() == MVT::f64);

      if (VA.isMemLoc()) {
        unsigned Offset = VA.getLocMemOffset() + StackOffset;
        //if it is double-word aligned, just store.
        if (Offset % 8 == 0) {
          SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
          SDValue PtrOff = DAG.getIntPtrConstant(Offset);
          PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
          MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
                                             MachinePointerInfo(),
                                             false, false, 0));
          continue;
        }
      }

      SDValue StackPtr = DAG.CreateStackTemporary(MVT::f64, MVT::i32);
      SDValue Store = DAG.getStore(DAG.getEntryNode(), dl,
                                   Arg, StackPtr, MachinePointerInfo(),
                                   false, false, 0);
      // Sparc is big-endian, so the high part comes first.
      SDValue Hi = DAG.getLoad(MVT::i32, dl, Store, StackPtr,
                               MachinePointerInfo(), false, false, 0);
      // Increment the pointer to the other half.
      StackPtr = DAG.getNode(ISD::ADD, dl, StackPtr.getValueType(), StackPtr,
                             DAG.getIntPtrConstant(4));
      // Load the low part.
      SDValue Lo = DAG.getLoad(MVT::i32, dl, Store, StackPtr,
                               MachinePointerInfo(), false, false, 0);

      if (VA.isRegLoc()) {
        RegsToPass.push_back(std::make_pair(VA.getLocReg(), Hi));
        assert(i+1 != e);
        CCValAssign &NextVA = ArgLocs[++i];
        if (NextVA.isRegLoc()) {
          RegsToPass.push_back(std::make_pair(NextVA.getLocReg(), Lo));
        } else {
          //Store the low part in stack.
          unsigned Offset = NextVA.getLocMemOffset() + StackOffset;
          SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
          SDValue PtrOff = DAG.getIntPtrConstant(Offset);
          PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
          MemOpChains.push_back(DAG.getStore(Chain, dl, Lo, PtrOff,
                                             MachinePointerInfo(),
                                             false, false, 0));
        }
      } else {
        unsigned Offset = VA.getLocMemOffset() + StackOffset;
        // Store the high part.
        SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
        SDValue PtrOff = DAG.getIntPtrConstant(Offset);
        PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
        MemOpChains.push_back(DAG.getStore(Chain, dl, Hi, PtrOff,
                                           MachinePointerInfo(),
                                           false, false, 0));
        // Store the low part.
        PtrOff = DAG.getIntPtrConstant(Offset+4);
        PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
        MemOpChains.push_back(DAG.getStore(Chain, dl, Lo, PtrOff,
                                           MachinePointerInfo(),
                                           false, false, 0));
      }
      continue;
    }

    // Arguments that can be passed on register must be kept at
    // RegsToPass vector
    if (VA.isRegLoc()) {
      if (VA.getLocVT() != MVT::f32) {
        RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
        continue;
      }
      Arg = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Arg);
      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
      continue;
    }

    assert(VA.isMemLoc());

    // Create a store off the stack pointer for this argument.
    SDValue StackPtr = DAG.getRegister(SP::O6, MVT::i32);
    SDValue PtrOff = DAG.getIntPtrConstant(VA.getLocMemOffset()+StackOffset);
    PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
    MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
                                       MachinePointerInfo(),
                                       false, false, 0));
  }


  // Emit all stores, make sure the occur before any copies into physregs.
  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.
  // The InFlag in necessary since all emitted instructions must be
  // stuck together.
  SDValue InFlag;
  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
    unsigned Reg = RegsToPass[i].first;
    // Remap I0->I7 -> O0->O7.
    if (Reg >= SP::I0 && Reg <= SP::I7)
      Reg = Reg-SP::I0+SP::O0;

    Chain = DAG.getCopyToReg(Chain, dl, Reg, RegsToPass[i].second, InFlag);
    InFlag = Chain.getValue(1);
  }

  unsigned SRetArgSize = (hasStructRetAttr)? getSRetArgSize(DAG, Callee):0;

  // 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.
  // Likewise ExternalSymbol -> TargetExternalSymbol.
  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
    Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, MVT::i32);
  else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee))
    Callee = DAG.getTargetExternalSymbol(E->getSymbol(), MVT::i32);

  // Returns a chain & a flag for retval copy to use
  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
  SmallVector<SDValue, 8> Ops;
  Ops.push_back(Chain);
  Ops.push_back(Callee);
  if (hasStructRetAttr)
    Ops.push_back(DAG.getTargetConstant(SRetArgSize, MVT::i32));
  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
    unsigned Reg = RegsToPass[i].first;
    if (Reg >= SP::I0 && Reg <= SP::I7)
      Reg = Reg-SP::I0+SP::O0;

    Ops.push_back(DAG.getRegister(Reg, RegsToPass[i].second.getValueType()));
  }
  if (InFlag.getNode())
    Ops.push_back(InFlag);

  Chain = DAG.getNode(SPISD::CALL, dl, NodeTys, &Ops[0], Ops.size());
  InFlag = Chain.getValue(1);

  Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(ArgsSize, true),
                             DAG.getIntPtrConstant(0, true), InFlag);
  InFlag = Chain.getValue(1);

  // Assign locations to each value returned by this call.
  SmallVector<CCValAssign, 16> RVLocs;
  CCState RVInfo(CallConv, isVarArg, DAG.getTarget(),
                 RVLocs, *DAG.getContext());

  RVInfo.AnalyzeCallResult(Ins, RetCC_Sparc32);

  // Copy all of the result registers out of their specified physreg.
  for (unsigned i = 0; i != RVLocs.size(); ++i) {
    unsigned Reg = RVLocs[i].getLocReg();

    // Remap I0->I7 -> O0->O7.
    if (Reg >= SP::I0 && Reg <= SP::I7)
      Reg = Reg-SP::I0+SP::O0;

    Chain = DAG.getCopyFromReg(Chain, dl, Reg,
                               RVLocs[i].getValVT(), InFlag).getValue(1);
    InFlag = Chain.getValue(2);
    InVals.push_back(Chain.getValue(0));
  }

  return Chain;
}
// Generate code to call a function
SDValue
VectorProcTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
	SmallVectorImpl<SDValue> &InVals) const 
{
	SelectionDAG &DAG = CLI.DAG;
	DebugLoc &dl = CLI.DL;
	SmallVector<ISD::OutputArg, 32> &Outs = CLI.Outs;
	SmallVector<SDValue, 32> &OutVals = CLI.OutVals;
	SmallVector<ISD::InputArg, 32> &Ins = CLI.Ins;
	SDValue Chain = CLI.Chain;
	SDValue Callee = CLI.Callee;
	CallingConv::ID CallConv = CLI.CallConv;
	bool isVarArg = CLI.IsVarArg;
	
	// We do not support tail calls. This flag must be cleared in order
	// to indicate that to subsequent passes.
	CLI.IsTailCall = false;

	MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();

	// Analyze operands of the call, assigning locations to each operand.
	// VectorProcCallingConv.td will auto-generate CC_VectorProc32, which 
	// knows how to handle operands (what go in registers vs. stack, etc).
	SmallVector<CCValAssign, 16> ArgLocs;
	CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(),
		DAG.getTarget(), ArgLocs, *DAG.getContext());
	CCInfo.AnalyzeCallOperands(Outs, CC_VectorProc32);

	// Get the size of the outgoing arguments stack space requirement.
	unsigned ArgsSize = CCInfo.getNextStackOffset();

	// We always keep the stack pointer 64 byte aligned so we can use block
	// loads/stores for vector arguments
	ArgsSize = (ArgsSize + 63) & ~63;

	// Create local copies for all arguments that are passed by value
	SmallVector<SDValue, 8> ByValArgs;
	for (unsigned i = 0, e = Outs.size(); i != e; ++i) {
		ISD::ArgFlagsTy Flags = Outs[i].Flags;
		if (!Flags.isByVal())
			continue;

		SDValue Arg = OutVals[i];
		unsigned Size = Flags.getByValSize();
		unsigned Align = Flags.getByValAlign();

		int FI = MFI->CreateStackObject(Size, Align, false);
		SDValue FIPtr = DAG.getFrameIndex(FI, getPointerTy());
		SDValue SizeNode = DAG.getConstant(Size, MVT::i32);
		Chain = DAG.getMemcpy(Chain, dl, FIPtr, Arg, SizeNode, Align,
			false,        //isVolatile,
			(Size <= 32), //AlwaysInline if size <= 32
			MachinePointerInfo(), MachinePointerInfo());

		ByValArgs.push_back(FIPtr);
	}

	// CALLSEQ_START will decrement the stack to reserve space
	Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(ArgsSize, true));

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

	// Walk through arguments, storing each one to the proper palce
	bool hasStructRetAttr = false;
	for (unsigned i = 0, realArgIdx = 0, byvalArgIdx = 0, e = ArgLocs.size();
		i != e; ++i, ++realArgIdx) {

		CCValAssign &VA = ArgLocs[i];
		SDValue Arg = OutVals[realArgIdx];

		ISD::ArgFlagsTy Flags = Outs[realArgIdx].Flags;

		// Use the local copy we created above if this is passed by value
		if (Flags.isByVal())
			Arg = ByValArgs[byvalArgIdx++];

		// Promote the value if needed.
		switch (VA.getLocInfo()) {
			case CCValAssign::Full: 
				break;
				
			case CCValAssign::SExt:
				Arg = DAG.getNode(ISD::SIGN_EXTEND, dl, VA.getLocVT(), Arg);
				break;

			case CCValAssign::ZExt:
				Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
				break;

			case CCValAssign::AExt:
				Arg = DAG.getNode(ISD::ANY_EXTEND, dl, VA.getLocVT(), Arg);
				break;

			case CCValAssign::BCvt:
				Arg = DAG.getNode(ISD::BITCAST, dl, VA.getLocVT(), Arg);
				break;

			default: 
				llvm_unreachable("Unknown loc info!");
		}

		if (Flags.isSRet()) {
			// Structure return
			assert(VA.needsCustom());
			SDValue StackPtr = DAG.getRegister(VectorProc::SP_REG, MVT::i32);
			SDValue PtrOff = DAG.getIntPtrConstant(64);
			PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
			MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
				MachinePointerInfo(), false, false, 0));
			hasStructRetAttr = true;
			continue;
		}

		// Arguments that can be passed on register must be kept at
		// RegsToPass vector
		if (VA.isRegLoc()) {
			RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
			continue;
		}

		// This needs to be pushed on the stack
		assert(VA.isMemLoc());

		// Create a store off the stack pointer for this argument.
		SDValue StackPtr = DAG.getRegister(VectorProc::SP_REG, MVT::i32);
		SDValue PtrOff = DAG.getIntPtrConstant(VA.getLocMemOffset());
		PtrOff = DAG.getNode(ISD::ADD, dl, MVT::i32, StackPtr, PtrOff);
		MemOpChains.push_back(DAG.getStore(Chain, dl, Arg, PtrOff,
			MachinePointerInfo(), false, false, 0));
	}

	// Emit all stores, make sure the occur before any copies into physregs.
	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.
	// The InFlag in necessary since all emitted instructions must be
	// stuck together.
	SDValue InFlag;
	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);
	}

	unsigned SRetArgSize = hasStructRetAttr ? getSRetArgSize(DAG, Callee) : 0;

	// Get the function address.
	// 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.
	// Likewise ExternalSymbol -> TargetExternalSymbol.
	if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee))
		Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl, MVT::i32);
	else if (ExternalSymbolSDNode *E = dyn_cast<ExternalSymbolSDNode>(Callee))
		Callee = DAG.getTargetExternalSymbol(E->getSymbol(), MVT::i32);

	// Returns a chain & a flag for retval copy to use
	SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
	SmallVector<SDValue, 8> Ops;
	Ops.push_back(Chain);
	Ops.push_back(Callee);
	if (hasStructRetAttr)
		Ops.push_back(DAG.getTargetConstant(SRetArgSize, MVT::i32));

	for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i)
	{
		Ops.push_back(DAG.getRegister(RegsToPass[i].first,
			RegsToPass[i].second.getValueType()));
	}
	
	// Add a register mask operand representing the call-preserved registers.
	const TargetRegisterInfo *TRI = getTargetMachine().getRegisterInfo();
	const uint32_t *Mask = TRI->getCallPreservedMask(CLI.CallConv);
	assert(Mask && "Missing call preserved mask for calling convention");
	Ops.push_back(CLI.DAG.getRegisterMask(Mask));

	if (InFlag.getNode())
		Ops.push_back(InFlag);

	Chain = DAG.getNode(VectorProcISD::CALL, dl, NodeTys, &Ops[0], Ops.size());
	InFlag = Chain.getValue(1);

	Chain = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(ArgsSize, true),
		DAG.getIntPtrConstant(0, true), InFlag);
	InFlag = Chain.getValue(1);

	// The call has returned, handle return values
	SmallVector<CCValAssign, 16> RVLocs;
	CCState RVInfo(CallConv, isVarArg, DAG.getMachineFunction(),
		DAG.getTarget(), RVLocs, *DAG.getContext());

	RVInfo.AnalyzeCallResult(Ins, RetCC_VectorProc32);

	// Copy all of the result registers out of their specified physreg.
	for (unsigned i = 0; i != RVLocs.size(); ++i) {
		Chain = DAG.getCopyFromReg(Chain, dl, RVLocs[i].getLocReg(),
			RVLocs[i].getValVT(), InFlag).getValue(1);
		InFlag = Chain.getValue(2);
		InVals.push_back(Chain.getValue(0));
	}

	return Chain;
}