X86ISelLowering.cpp

        if (ChainVal->getOperand(i).Val == LdVal) {
          TokenFactorIndex = i;
          Ld = cast<LoadSDNode>(St->getValue());
        } else
          Ops.push_back(ChainVal->getOperand(i));
      }
    }
    if (Ld) {
      // If we are a 64-bit capable x86, lower to a single movq load/store pair.
      if (Subtarget->is64Bit()) {
        SDOperand NewLd = DAG.getLoad(MVT::i64, Ld->getChain(), 
                                      Ld->getBasePtr(), Ld->getSrcValue(), 
                                      Ld->getSrcValueOffset(), Ld->isVolatile(),
                                      Ld->getAlignment());
        SDOperand NewChain = NewLd.getValue(1);
        if (TokenFactorIndex != -1) {
          Ops.push_back(NewLd);
          NewChain = DAG.getNode(ISD::TokenFactor, MVT::Other, &Ops[0], 
                                 Ops.size());
        }
        return DAG.getStore(NewChain, NewLd, St->getBasePtr(),
                            St->getSrcValue(), St->getSrcValueOffset(),
                            St->isVolatile(), St->getAlignment());
      }

      // Otherwise, lower to two 32-bit copies.
      SDOperand LoAddr = Ld->getBasePtr();
      SDOperand HiAddr = DAG.getNode(ISD::ADD, MVT::i32, LoAddr,
                                     DAG.getConstant(MVT::i32, 4));

      SDOperand LoLd = DAG.getLoad(MVT::i32, Ld->getChain(), LoAddr,
                                   Ld->getSrcValue(), Ld->getSrcValueOffset(),
                                   Ld->isVolatile(), Ld->getAlignment());
      SDOperand HiLd = DAG.getLoad(MVT::i32, Ld->getChain(), HiAddr,
                                   Ld->getSrcValue(), Ld->getSrcValueOffset()+4,
                                   Ld->isVolatile(), 
                                   MinAlign(Ld->getAlignment(), 4));

      SDOperand NewChain = LoLd.getValue(1);
      if (TokenFactorIndex != -1) {
        Ops.push_back(LoLd);
        Ops.push_back(HiLd);
        NewChain = DAG.getNode(ISD::TokenFactor, MVT::Other, &Ops[0], 
                               Ops.size());
      }

      LoAddr = St->getBasePtr();
      HiAddr = DAG.getNode(ISD::ADD, MVT::i32, LoAddr,
                           DAG.getConstant(MVT::i32, 4));

      SDOperand LoSt = DAG.getStore(NewChain, LoLd, LoAddr,
                          St->getSrcValue(), St->getSrcValueOffset(),
                          St->isVolatile(), St->getAlignment());
      SDOperand HiSt = DAG.getStore(NewChain, HiLd, HiAddr,
                                    St->getSrcValue(), St->getSrcValueOffset()+4,
                                    St->isVolatile(), 
                                    MinAlign(St->getAlignment(), 4));
      return DAG.getNode(ISD::TokenFactor, MVT::Other, LoSt, HiSt);
    }
  }
  return SDOperand();
}

/// PerformFORCombine - Do target-specific dag combines on X86ISD::FOR and
/// X86ISD::FXOR nodes.
static SDOperand PerformFORCombine(SDNode *N, SelectionDAG &DAG) {
  assert(N->getOpcode() == X86ISD::FOR || N->getOpcode() == X86ISD::FXOR);
  // F[X]OR(0.0, x) -> x
  // F[X]OR(x, 0.0) -> x
  if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(N->getOperand(0)))
    if (C->getValueAPF().isPosZero())
      return N->getOperand(1);
  if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(N->getOperand(1)))
    if (C->getValueAPF().isPosZero())
      return N->getOperand(0);
  return SDOperand();
}

/// PerformFANDCombine - Do target-specific dag combines on X86ISD::FAND nodes.
static SDOperand PerformFANDCombine(SDNode *N, SelectionDAG &DAG) {
  // FAND(0.0, x) -> 0.0
  // FAND(x, 0.0) -> 0.0
  if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(N->getOperand(0)))
    if (C->getValueAPF().isPosZero())
      return N->getOperand(0);
  if (ConstantFPSDNode *C = dyn_cast<ConstantFPSDNode>(N->getOperand(1)))
    if (C->getValueAPF().isPosZero())
      return N->getOperand(1);
  return SDOperand();
}


SDOperand X86TargetLowering::PerformDAGCombine(SDNode *N,
                                               DAGCombinerInfo &DCI) const {
  SelectionDAG &DAG = DCI.DAG;
  switch (N->getOpcode()) {
  default: break;
  case ISD::VECTOR_SHUFFLE: return PerformShuffleCombine(N, DAG, Subtarget);
  case ISD::SELECT:         return PerformSELECTCombine(N, DAG, Subtarget);
  case ISD::STORE:          
      return PerformSTORECombine(cast<StoreSDNode>(N), DAG, Subtarget);
  case X86ISD::FXOR:
  case X86ISD::FOR:         return PerformFORCombine(N, DAG);
  case X86ISD::FAND:        return PerformFANDCombine(N, DAG);
  }

  return SDOperand();
}

//===----------------------------------------------------------------------===//
//                           X86 Inline Assembly Support
//===----------------------------------------------------------------------===//

/// getConstraintType - Given a constraint letter, return the type of
/// constraint it is for this target.
X86TargetLowering::ConstraintType
X86TargetLowering::getConstraintType(const std::string &Constraint) const {
  if (Constraint.size() == 1) {
    switch (Constraint[0]) {
    case 'A':
    case 'r':
    case 'R':
    case 'l':
    case 'q':
    case 'Q':
    case 'x':
    case 'Y':
      return C_RegisterClass;
    default:
      break;
    }
  }
  return TargetLowering::getConstraintType(Constraint);
}

/// LowerXConstraint - try to replace an X constraint, which matches anything,
/// with another that has more specific requirements based on the type of the
/// corresponding operand.
void X86TargetLowering::lowerXConstraint(MVT::ValueType ConstraintVT, 
                                         std::string& s) const {
  if (MVT::isFloatingPoint(ConstraintVT)) {
    if (Subtarget->hasSSE2())
      s = "Y";
    else if (Subtarget->hasSSE1())
      s = "x";
    else
      s = "f";
  } else
    return TargetLowering::lowerXConstraint(ConstraintVT, s);
}

/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
/// vector.  If it is invalid, don't add anything to Ops.
void X86TargetLowering::LowerAsmOperandForConstraint(SDOperand Op,
                                                     char Constraint,
                                                     std::vector<SDOperand>&Ops,
                                                     SelectionDAG &DAG) {
  SDOperand Result(0, 0);
  
  switch (Constraint) {
  default: break;
  case 'I':
    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op)) {
      if (C->getValue() <= 31) {
        Result = DAG.getTargetConstant(C->getValue(), Op.getValueType());
        break;
      }
    }
    return;
  case 'N':
    if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op)) {
      if (C->getValue() <= 255) {
        Result = DAG.getTargetConstant(C->getValue(), Op.getValueType());
        break;
      }
    }
    return;
  case 'i': {
    // Literal immediates are always ok.
    if (ConstantSDNode *CST = dyn_cast<ConstantSDNode>(Op)) {
      Result = DAG.getTargetConstant(CST->getValue(), Op.getValueType());
      break;
    }

    // If we are in non-pic codegen mode, we allow the address of a global (with
    // an optional displacement) to be used with 'i'.
    GlobalAddressSDNode *GA = dyn_cast<GlobalAddressSDNode>(Op);
    int64_t Offset = 0;
    
    // Match either (GA) or (GA+C)
    if (GA) {
      Offset = GA->getOffset();
    } else if (Op.getOpcode() == ISD::ADD) {
      ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op.getOperand(1));
      GA = dyn_cast<GlobalAddressSDNode>(Op.getOperand(0));
      if (C && GA) {
        Offset = GA->getOffset()+C->getValue();
      } else {
        C = dyn_cast<ConstantSDNode>(Op.getOperand(1));
        GA = dyn_cast<GlobalAddressSDNode>(Op.getOperand(0));
        if (C && GA)
          Offset = GA->getOffset()+C->getValue();
        else
          C = 0, GA = 0;
      }
    }
    
    if (GA) {
      // If addressing this global requires a load (e.g. in PIC mode), we can't
      // match.
      if (Subtarget->GVRequiresExtraLoad(GA->getGlobal(), getTargetMachine(),
                                         false))
        return;

      Op = DAG.getTargetGlobalAddress(GA->getGlobal(), GA->getValueType(0),
                                      Offset);
      Result = Op;
      break;
    }

    // Otherwise, not valid for this mode.
    return;
  }
  }
  
  if (Result.Val) {
    Ops.push_back(Result);
    return;
  }
  return TargetLowering::LowerAsmOperandForConstraint(Op, Constraint, Ops, DAG);
}

std::vector<unsigned> X86TargetLowering::
getRegClassForInlineAsmConstraint(const std::string &Constraint,
                                  MVT::ValueType VT) const {
  if (Constraint.size() == 1) {
    // FIXME: not handling fp-stack yet!
    switch (Constraint[0]) {      // GCC X86 Constraint Letters
    default: break;  // Unknown constraint letter
    case 'A':   // EAX/EDX
      if (VT == MVT::i32 || VT == MVT::i64)
        return make_vector<unsigned>(X86::EAX, X86::EDX, 0);
      break;
    case 'q':   // Q_REGS (GENERAL_REGS in 64-bit mode)
    case 'Q':   // Q_REGS
      if (VT == MVT::i32)
        return make_vector<unsigned>(X86::EAX, X86::EDX, X86::ECX, X86::EBX, 0);
      else if (VT == MVT::i16)
        return make_vector<unsigned>(X86::AX, X86::DX, X86::CX, X86::BX, 0);
      else if (VT == MVT::i8)
        return make_vector<unsigned>(X86::AL, X86::DL, X86::CL, X86::BL, 0);
      else if (VT == MVT::i64)
        return make_vector<unsigned>(X86::RAX, X86::RDX, X86::RCX, X86::RBX, 0);
      break;
    }
  }

  return std::vector<unsigned>();
}

std::pair<unsigned, const TargetRegisterClass*>
X86TargetLowering::getRegForInlineAsmConstraint(const std::string &Constraint,
                                                MVT::ValueType VT) const {
  // First, see if this is a constraint that directly corresponds to an LLVM
  // register class.
  if (Constraint.size() == 1) {
    // GCC Constraint Letters
    switch (Constraint[0]) {
    default: break;
    case 'r':   // GENERAL_REGS
    case 'R':   // LEGACY_REGS
    case 'l':   // INDEX_REGS
      if (VT == MVT::i64 && Subtarget->is64Bit())
        return std::make_pair(0U, X86::GR64RegisterClass);
      if (VT == MVT::i32)
        return std::make_pair(0U, X86::GR32RegisterClass);
      else if (VT == MVT::i16)
        return std::make_pair(0U, X86::GR16RegisterClass);
      else if (VT == MVT::i8)
        return std::make_pair(0U, X86::GR8RegisterClass);
      break;
    case 'y':   // MMX_REGS if MMX allowed.
      if (!Subtarget->hasMMX()) break;
      return std::make_pair(0U, X86::VR64RegisterClass);
      break;
    case 'Y':   // SSE_REGS if SSE2 allowed
      if (!Subtarget->hasSSE2()) break;
      // FALL THROUGH.
    case 'x':   // SSE_REGS if SSE1 allowed
      if (!Subtarget->hasSSE1()) break;
      
      switch (VT) {
      default: break;
      // Scalar SSE types.
      case MVT::f32:
      case MVT::i32:
        return std::make_pair(0U, X86::FR32RegisterClass);
      case MVT::f64:
      case MVT::i64:
        return std::make_pair(0U, X86::FR64RegisterClass);
      // Vector types.
      case MVT::v16i8:
      case MVT::v8i16:
      case MVT::v4i32:
      case MVT::v2i64:
      case MVT::v4f32:
      case MVT::v2f64:
        return std::make_pair(0U, X86::VR128RegisterClass);
      }
      break;
    }
  }
  
  // Use the default implementation in TargetLowering to convert the register
  // constraint into a member of a register class.
  std::pair<unsigned, const TargetRegisterClass*> Res;
  Res = TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);

  // Not found as a standard register?
  if (Res.second == 0) {
    // GCC calls "st(0)" just plain "st".
    if (StringsEqualNoCase("{st}", Constraint)) {
      Res.first = X86::ST0;
      Res.second = X86::RFP80RegisterClass;
    }

    return Res;
  }

  // Otherwise, check to see if this is a register class of the wrong value
  // type.  For example, we want to map "{ax},i32" -> {eax}, we don't want it to
  // turn into {ax},{dx}.
  if (Res.second->hasType(VT))
    return Res;   // Correct type already, nothing to do.

  // All of the single-register GCC register classes map their values onto
  // 16-bit register pieces "ax","dx","cx","bx","si","di","bp","sp".  If we
  // really want an 8-bit or 32-bit register, map to the appropriate register
  // class and return the appropriate register.
  if (Res.second != X86::GR16RegisterClass)
    return Res;

  if (VT == MVT::i8) {
    unsigned DestReg = 0;
    switch (Res.first) {
    default: break;
    case X86::AX: DestReg = X86::AL; break;
    case X86::DX: DestReg = X86::DL; break;
    case X86::CX: DestReg = X86::CL; break;
    case X86::BX: DestReg = X86::BL; break;
    }
    if (DestReg) {
      Res.first = DestReg;
      Res.second = Res.second = X86::GR8RegisterClass;
    }
  } else if (VT == MVT::i32) {
    unsigned DestReg = 0;
    switch (Res.first) {
    default: break;
    case X86::AX: DestReg = X86::EAX; break;
    case X86::DX: DestReg = X86::EDX; break;
    case X86::CX: DestReg = X86::ECX; break;
    case X86::BX: DestReg = X86::EBX; break;
    case X86::SI: DestReg = X86::ESI; break;
    case X86::DI: DestReg = X86::EDI; break;
    case X86::BP: DestReg = X86::EBP; break;
    case X86::SP: DestReg = X86::ESP; break;
    }
    if (DestReg) {
      Res.first = DestReg;
      Res.second = Res.second = X86::GR32RegisterClass;
    }
  } else if (VT == MVT::i64) {
    unsigned DestReg = 0;
    switch (Res.first) {
    default: break;
    case X86::AX: DestReg = X86::RAX; break;
    case X86::DX: DestReg = X86::RDX; break;
    case X86::CX: DestReg = X86::RCX; break;
    case X86::BX: DestReg = X86::RBX; break;
    case X86::SI: DestReg = X86::RSI; break;
    case X86::DI: DestReg = X86::RDI; break;
    case X86::BP: DestReg = X86::RBP; break;
    case X86::SP: DestReg = X86::RSP; break;
    }
    if (DestReg) {
      Res.first = DestReg;
      Res.second = Res.second = X86::GR64RegisterClass;
    }
  }

  return Res;
}