X86ISelLowering.cpp

//===-- X86ISelLowering.cpp - X86 DAG Lowering Implementation -------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file was developed by Chris Lattner and is distributed under
// the University of Illinois Open Source License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file defines the interfaces that X86 uses to lower LLVM code into a
// selection DAG.
//
//===----------------------------------------------------------------------===//

#include "X86.h"
#include "X86CodeEmitter.h"
#include "X86InstrBuilder.h"
#include "X86ISelLowering.h"
#include "X86MachineFunctionInfo.h"
#include "X86TargetMachine.h"
#include "llvm/CallingConv.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/GlobalVariable.h"
#include "llvm/Function.h"
#include "llvm/Intrinsics.h"
#include "llvm/ADT/VectorExtras.h"
#include "llvm/Analysis/ScalarEvolutionExpressions.h"
#include "llvm/CodeGen/CallingConvLower.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SSARegMap.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ParameterAttributes.h"
using namespace llvm;

X86TargetLowering::X86TargetLowering(TargetMachine &TM)
  : TargetLowering(TM) {
  Subtarget = &TM.getSubtarget<X86Subtarget>();
  X86ScalarSSE = Subtarget->hasSSE2();
  X86StackPtr = Subtarget->is64Bit() ? X86::RSP : X86::ESP;

  RegInfo = TM.getRegisterInfo();

  // Set up the TargetLowering object.

  // X86 is weird, it always uses i8 for shift amounts and setcc results.
  setShiftAmountType(MVT::i8);
  setSetCCResultType(MVT::i8);
  setSetCCResultContents(ZeroOrOneSetCCResult);
  setSchedulingPreference(SchedulingForRegPressure);
  setShiftAmountFlavor(Mask);   // shl X, 32 == shl X, 0
  setStackPointerRegisterToSaveRestore(X86StackPtr);

  if (Subtarget->isTargetDarwin()) {
    // Darwin should use _setjmp/_longjmp instead of setjmp/longjmp.
    setUseUnderscoreSetJmp(false);
    setUseUnderscoreLongJmp(false);
  } else if (Subtarget->isTargetMingw()) {
    // MS runtime is weird: it exports _setjmp, but longjmp!
    setUseUnderscoreSetJmp(true);
    setUseUnderscoreLongJmp(false);
  } else {
    setUseUnderscoreSetJmp(true);
    setUseUnderscoreLongJmp(true);
  }
  
  // Set up the register classes.
  addRegisterClass(MVT::i8, X86::GR8RegisterClass);
  addRegisterClass(MVT::i16, X86::GR16RegisterClass);
  addRegisterClass(MVT::i32, X86::GR32RegisterClass);
  if (Subtarget->is64Bit())
    addRegisterClass(MVT::i64, X86::GR64RegisterClass);

  setLoadXAction(ISD::SEXTLOAD, MVT::i1, Expand);

  // Promote all UINT_TO_FP to larger SINT_TO_FP's, as X86 doesn't have this
  // operation.
  setOperationAction(ISD::UINT_TO_FP       , MVT::i1   , Promote);
  setOperationAction(ISD::UINT_TO_FP       , MVT::i8   , Promote);
  setOperationAction(ISD::UINT_TO_FP       , MVT::i16  , Promote);

  if (Subtarget->is64Bit()) {
    setOperationAction(ISD::UINT_TO_FP     , MVT::i64  , Expand);
    setOperationAction(ISD::UINT_TO_FP     , MVT::i32  , Promote);
  } else {
    if (X86ScalarSSE)
      // If SSE i64 SINT_TO_FP is not available, expand i32 UINT_TO_FP.
      setOperationAction(ISD::UINT_TO_FP   , MVT::i32  , Expand);
    else
      setOperationAction(ISD::UINT_TO_FP   , MVT::i32  , Promote);
  }

  // Promote i1/i8 SINT_TO_FP to larger SINT_TO_FP's, as X86 doesn't have
  // this operation.
  setOperationAction(ISD::SINT_TO_FP       , MVT::i1   , Promote);
  setOperationAction(ISD::SINT_TO_FP       , MVT::i8   , Promote);
  // SSE has no i16 to fp conversion, only i32
  if (X86ScalarSSE)
    setOperationAction(ISD::SINT_TO_FP     , MVT::i16  , Promote);
  else {
    setOperationAction(ISD::SINT_TO_FP     , MVT::i16  , Custom);
    setOperationAction(ISD::SINT_TO_FP     , MVT::i32  , Custom);
  }

  if (!Subtarget->is64Bit()) {
    // Custom lower SINT_TO_FP and FP_TO_SINT from/to i64 in 32-bit mode.
    setOperationAction(ISD::SINT_TO_FP     , MVT::i64  , Custom);
    setOperationAction(ISD::FP_TO_SINT     , MVT::i64  , Custom);
  }

  // Promote i1/i8 FP_TO_SINT to larger FP_TO_SINTS's, as X86 doesn't have
  // this operation.
  setOperationAction(ISD::FP_TO_SINT       , MVT::i1   , Promote);
  setOperationAction(ISD::FP_TO_SINT       , MVT::i8   , Promote);

  if (X86ScalarSSE) {
    setOperationAction(ISD::FP_TO_SINT     , MVT::i16  , Promote);
  } else {
    setOperationAction(ISD::FP_TO_SINT     , MVT::i16  , Custom);
    setOperationAction(ISD::FP_TO_SINT     , MVT::i32  , Custom);
  }

  // Handle FP_TO_UINT by promoting the destination to a larger signed
  // conversion.
  setOperationAction(ISD::FP_TO_UINT       , MVT::i1   , Promote);
  setOperationAction(ISD::FP_TO_UINT       , MVT::i8   , Promote);
  setOperationAction(ISD::FP_TO_UINT       , MVT::i16  , Promote);

  if (Subtarget->is64Bit()) {
    setOperationAction(ISD::FP_TO_UINT     , MVT::i64  , Expand);
    setOperationAction(ISD::FP_TO_UINT     , MVT::i32  , Promote);
  } else {
    if (X86ScalarSSE && !Subtarget->hasSSE3())
      // Expand FP_TO_UINT into a select.
      // FIXME: We would like to use a Custom expander here eventually to do
      // the optimal thing for SSE vs. the default expansion in the legalizer.
      setOperationAction(ISD::FP_TO_UINT   , MVT::i32  , Expand);
    else
      // With SSE3 we can use fisttpll to convert to a signed i64.
      setOperationAction(ISD::FP_TO_UINT   , MVT::i32  , Promote);
  }

  // TODO: when we have SSE, these could be more efficient, by using movd/movq.
  if (!X86ScalarSSE) {
    setOperationAction(ISD::BIT_CONVERT      , MVT::f32  , Expand);
    setOperationAction(ISD::BIT_CONVERT      , MVT::i32  , Expand);
  }

  setOperationAction(ISD::BR_JT            , MVT::Other, Expand);
  setOperationAction(ISD::BRCOND           , MVT::Other, Custom);
  setOperationAction(ISD::BR_CC            , MVT::Other, Expand);
  setOperationAction(ISD::SELECT_CC        , MVT::Other, Expand);
  setOperationAction(ISD::MEMMOVE          , MVT::Other, Expand);
  if (Subtarget->is64Bit())
    setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i32, Expand);
  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16  , Expand);
  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8   , Expand);
  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1   , Expand);
  setOperationAction(ISD::FP_ROUND_INREG   , MVT::f32  , Expand);
  setOperationAction(ISD::FREM             , MVT::f64  , Expand);

  setOperationAction(ISD::CTPOP            , MVT::i8   , Expand);
  setOperationAction(ISD::CTTZ             , MVT::i8   , Expand);
  setOperationAction(ISD::CTLZ             , MVT::i8   , Expand);
  setOperationAction(ISD::CTPOP            , MVT::i16  , Expand);
  setOperationAction(ISD::CTTZ             , MVT::i16  , Expand);
  setOperationAction(ISD::CTLZ             , MVT::i16  , Expand);
  setOperationAction(ISD::CTPOP            , MVT::i32  , Expand);
  setOperationAction(ISD::CTTZ             , MVT::i32  , Expand);
  setOperationAction(ISD::CTLZ             , MVT::i32  , Expand);
  if (Subtarget->is64Bit()) {
    setOperationAction(ISD::CTPOP          , MVT::i64  , Expand);
    setOperationAction(ISD::CTTZ           , MVT::i64  , Expand);
    setOperationAction(ISD::CTLZ           , MVT::i64  , Expand);
  }

  setOperationAction(ISD::READCYCLECOUNTER , MVT::i64  , Custom);
  setOperationAction(ISD::BSWAP            , MVT::i16  , Expand);

  // These should be promoted to a larger select which is supported.
  setOperationAction(ISD::SELECT           , MVT::i1   , Promote);
  setOperationAction(ISD::SELECT           , MVT::i8   , Promote);
  // X86 wants to expand cmov itself.
  setOperationAction(ISD::SELECT          , MVT::i16  , Custom);
  setOperationAction(ISD::SELECT          , MVT::i32  , Custom);
  setOperationAction(ISD::SELECT          , MVT::f32  , Custom);
  setOperationAction(ISD::SELECT          , MVT::f64  , Custom);
  setOperationAction(ISD::SETCC           , MVT::i8   , Custom);
  setOperationAction(ISD::SETCC           , MVT::i16  , Custom);
  setOperationAction(ISD::SETCC           , MVT::i32  , Custom);
  setOperationAction(ISD::SETCC           , MVT::f32  , Custom);
  setOperationAction(ISD::SETCC           , MVT::f64  , Custom);
  if (Subtarget->is64Bit()) {
    setOperationAction(ISD::SELECT        , MVT::i64  , Custom);
    setOperationAction(ISD::SETCC         , MVT::i64  , Custom);
  }
  // X86 ret instruction may pop stack.
  setOperationAction(ISD::RET             , MVT::Other, Custom);
  if (!Subtarget->is64Bit())
    setOperationAction(ISD::EH_RETURN       , MVT::Other, Custom);

  // Darwin ABI issue.
  setOperationAction(ISD::ConstantPool    , MVT::i32  , Custom);
  setOperationAction(ISD::JumpTable       , MVT::i32  , Custom);
  setOperationAction(ISD::GlobalAddress   , MVT::i32  , Custom);
  setOperationAction(ISD::GlobalTLSAddress, MVT::i32  , Custom);
  setOperationAction(ISD::ExternalSymbol  , MVT::i32  , Custom);
  if (Subtarget->is64Bit()) {
    setOperationAction(ISD::ConstantPool  , MVT::i64  , Custom);
    setOperationAction(ISD::JumpTable     , MVT::i64  , Custom);
    setOperationAction(ISD::GlobalAddress , MVT::i64  , Custom);
    setOperationAction(ISD::ExternalSymbol, MVT::i64  , Custom);
  }
  // 64-bit addm sub, shl, sra, srl (iff 32-bit x86)
  setOperationAction(ISD::SHL_PARTS       , MVT::i32  , Custom);
  setOperationAction(ISD::SRA_PARTS       , MVT::i32  , Custom);
  setOperationAction(ISD::SRL_PARTS       , MVT::i32  , Custom);
  // X86 wants to expand memset / memcpy itself.
  setOperationAction(ISD::MEMSET          , MVT::Other, Custom);
  setOperationAction(ISD::MEMCPY          , MVT::Other, Custom);

  // We don't have line number support yet.
  setOperationAction(ISD::LOCATION, MVT::Other, Expand);
  setOperationAction(ISD::DEBUG_LOC, MVT::Other, Expand);
  // FIXME - use subtarget debug flags
  if (!Subtarget->isTargetDarwin() &&
      !Subtarget->isTargetELF() &&
      !Subtarget->isTargetCygMing())
    setOperationAction(ISD::LABEL, MVT::Other, Expand);

  setOperationAction(ISD::EXCEPTIONADDR, MVT::i64, Expand);
  setOperationAction(ISD::EHSELECTION,   MVT::i64, Expand);
  setOperationAction(ISD::EXCEPTIONADDR, MVT::i32, Expand);
  setOperationAction(ISD::EHSELECTION,   MVT::i32, Expand);
  if (Subtarget->is64Bit()) {
    // FIXME: Verify
    setExceptionPointerRegister(X86::RAX);
    setExceptionSelectorRegister(X86::RDX);
  } else {
    setExceptionPointerRegister(X86::EAX);
    setExceptionSelectorRegister(X86::EDX);
  }
  
  setOperationAction(ISD::ADJUST_TRAMP, MVT::i32,   Expand);
  setOperationAction(ISD::ADJUST_TRAMP, MVT::i64,   Expand);
  setOperationAction(ISD::TRAMPOLINE,   MVT::Other, Custom);

  // VASTART needs to be custom lowered to use the VarArgsFrameIndex
  setOperationAction(ISD::VASTART           , MVT::Other, Custom);
  setOperationAction(ISD::VAARG             , MVT::Other, Expand);
  setOperationAction(ISD::VAEND             , MVT::Other, Expand);
  if (Subtarget->is64Bit())
    setOperationAction(ISD::VACOPY          , MVT::Other, Custom);
  else
    setOperationAction(ISD::VACOPY          , MVT::Other, Expand);

  setOperationAction(ISD::STACKSAVE,          MVT::Other, Expand);
  setOperationAction(ISD::STACKRESTORE,       MVT::Other, Expand);
  if (Subtarget->is64Bit())
    setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);
  if (Subtarget->isTargetCygMing())
    setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
  else
    setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);

  if (X86ScalarSSE) {
    // Set up the FP register classes.
    addRegisterClass(MVT::f32, X86::FR32RegisterClass);
    addRegisterClass(MVT::f64, X86::FR64RegisterClass);

    // Use ANDPD to simulate FABS.
    setOperationAction(ISD::FABS , MVT::f64, Custom);
    setOperationAction(ISD::FABS , MVT::f32, Custom);

    // Use XORP to simulate FNEG.
    setOperationAction(ISD::FNEG , MVT::f64, Custom);
    setOperationAction(ISD::FNEG , MVT::f32, Custom);

    // Use ANDPD and ORPD to simulate FCOPYSIGN.
    setOperationAction(ISD::FCOPYSIGN, MVT::f64, Custom);
    setOperationAction(ISD::FCOPYSIGN, MVT::f32, Custom);

    // We don't support sin/cos/fmod
    setOperationAction(ISD::FSIN , MVT::f64, Expand);
    setOperationAction(ISD::FCOS , MVT::f64, Expand);
    setOperationAction(ISD::FREM , MVT::f64, Expand);
    setOperationAction(ISD::FSIN , MVT::f32, Expand);
    setOperationAction(ISD::FCOS , MVT::f32, Expand);
    setOperationAction(ISD::FREM , MVT::f32, Expand);

    // Expand FP immediates into loads from the stack, except for the special
    // cases we handle.
    setOperationAction(ISD::ConstantFP, MVT::f64, Expand);
    setOperationAction(ISD::ConstantFP, MVT::f32, Expand);
    addLegalFPImmediate(+0.0); // xorps / xorpd
  } else {
    // Set up the FP register classes.
    addRegisterClass(MVT::f64, X86::RFP64RegisterClass);
    addRegisterClass(MVT::f32, X86::RFP32RegisterClass);

    setOperationAction(ISD::UNDEF,     MVT::f64, Expand);
    setOperationAction(ISD::UNDEF,     MVT::f32, Expand);
    setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
    setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
    setOperationAction(ISD::FP_ROUND,  MVT::f32, Expand);

    if (!UnsafeFPMath) {
      setOperationAction(ISD::FSIN           , MVT::f64  , Expand);
      setOperationAction(ISD::FCOS           , MVT::f64  , Expand);
    }

    setOperationAction(ISD::ConstantFP, MVT::f64, Expand);
    setOperationAction(ISD::ConstantFP, MVT::f32, Expand);
    addLegalFPImmediate(+0.0); // FLD0
    addLegalFPImmediate(+1.0); // FLD1
    addLegalFPImmediate(-0.0); // FLD0/FCHS
    addLegalFPImmediate(-1.0); // FLD1/FCHS
  }

  // Long double always uses X87.
  addRegisterClass(MVT::f80, X86::RFP80RegisterClass);

  // First set operation action for all vector types to expand. Then we
  // will selectively turn on ones that can be effectively codegen'd.
  for (unsigned VT = (unsigned)MVT::FIRST_VECTOR_VALUETYPE;
       VT <= (unsigned)MVT::LAST_VECTOR_VALUETYPE; ++VT) {
    setOperationAction(ISD::ADD , (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::SUB , (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FADD, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FNEG, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FSUB, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::MUL , (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FMUL, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::SDIV, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::UDIV, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FDIV, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::SREM, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::UREM, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::LOAD, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::VECTOR_SHUFFLE,     (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::EXTRACT_VECTOR_ELT, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::INSERT_VECTOR_ELT,  (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FABS, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FSIN, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FCOS, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FREM, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FPOWI, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FSQRT, (MVT::ValueType)VT, Expand);
    setOperationAction(ISD::FCOPYSIGN, (MVT::ValueType)VT, Expand);
  }

  if (Subtarget->hasMMX()) {
    addRegisterClass(MVT::v8i8,  X86::VR64RegisterClass);
    addRegisterClass(MVT::v4i16, X86::VR64RegisterClass);
    addRegisterClass(MVT::v2i32, X86::VR64RegisterClass);
    addRegisterClass(MVT::v1i64, X86::VR64RegisterClass);

    // FIXME: add MMX packed arithmetics

    setOperationAction(ISD::ADD,                MVT::v8i8,  Legal);
    setOperationAction(ISD::ADD,                MVT::v4i16, Legal);
    setOperationAction(ISD::ADD,                MVT::v2i32, Legal);
    setOperationAction(ISD::ADD,                MVT::v1i64, Legal);

    setOperationAction(ISD::SUB,                MVT::v8i8,  Legal);
    setOperationAction(ISD::SUB,                MVT::v4i16, Legal);
    setOperationAction(ISD::SUB,                MVT::v2i32, Legal);

    setOperationAction(ISD::MULHS,              MVT::v4i16, Legal);
    setOperationAction(ISD::MUL,                MVT::v4i16, Legal);

    setOperationAction(ISD::AND,                MVT::v8i8,  Promote);
    AddPromotedToType (ISD::AND,                MVT::v8i8,  MVT::v1i64);
    setOperationAction(ISD::AND,                MVT::v4i16, Promote);
    AddPromotedToType (ISD::AND,                MVT::v4i16, MVT::v1i64);
    setOperationAction(ISD::AND,                MVT::v2i32, Promote);
    AddPromotedToType (ISD::AND,                MVT::v2i32, MVT::v1i64);
    setOperationAction(ISD::AND,                MVT::v1i64, Legal);

    setOperationAction(ISD::OR,                 MVT::v8i8,  Promote);
    AddPromotedToType (ISD::OR,                 MVT::v8i8,  MVT::v1i64);
    setOperationAction(ISD::OR,                 MVT::v4i16, Promote);
    AddPromotedToType (ISD::OR,                 MVT::v4i16, MVT::v1i64);
    setOperationAction(ISD::OR,                 MVT::v2i32, Promote);
    AddPromotedToType (ISD::OR,                 MVT::v2i32, MVT::v1i64);
    setOperationAction(ISD::OR,                 MVT::v1i64, Legal);

    setOperationAction(ISD::XOR,                MVT::v8i8,  Promote);
    AddPromotedToType (ISD::XOR,                MVT::v8i8,  MVT::v1i64);
    setOperationAction(ISD::XOR,                MVT::v4i16, Promote);
    AddPromotedToType (ISD::XOR,                MVT::v4i16, MVT::v1i64);
    setOperationAction(ISD::XOR,                MVT::v2i32, Promote);
    AddPromotedToType (ISD::XOR,                MVT::v2i32, MVT::v1i64);
    setOperationAction(ISD::XOR,                MVT::v1i64, Legal);

    setOperationAction(ISD::LOAD,               MVT::v8i8,  Promote);
    AddPromotedToType (ISD::LOAD,               MVT::v8i8,  MVT::v1i64);
    setOperationAction(ISD::LOAD,               MVT::v4i16, Promote);
    AddPromotedToType (ISD::LOAD,               MVT::v4i16, MVT::v1i64);
    setOperationAction(ISD::LOAD,               MVT::v2i32, Promote);
    AddPromotedToType (ISD::LOAD,               MVT::v2i32, MVT::v1i64);
    setOperationAction(ISD::LOAD,               MVT::v1i64, Legal);

    setOperationAction(ISD::BUILD_VECTOR,       MVT::v8i8,  Custom);
    setOperationAction(ISD::BUILD_VECTOR,       MVT::v4i16, Custom);
    setOperationAction(ISD::BUILD_VECTOR,       MVT::v2i32, Custom);
    setOperationAction(ISD::BUILD_VECTOR,       MVT::v1i64, Custom);

    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v8i8,  Custom);
    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v4i16, Custom);
    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v2i32, Custom);
    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v1i64, Custom);

    setOperationAction(ISD::SCALAR_TO_VECTOR,   MVT::v8i8,  Custom);
    setOperationAction(ISD::SCALAR_TO_VECTOR,   MVT::v4i16, Custom);
    setOperationAction(ISD::SCALAR_TO_VECTOR,   MVT::v2i32, Custom);
    setOperationAction(ISD::SCALAR_TO_VECTOR,   MVT::v1i64, Custom);
  }

  if (Subtarget->hasSSE1()) {
    addRegisterClass(MVT::v4f32, X86::VR128RegisterClass);

    setOperationAction(ISD::FADD,               MVT::v4f32, Legal);
    setOperationAction(ISD::FSUB,               MVT::v4f32, Legal);
    setOperationAction(ISD::FMUL,               MVT::v4f32, Legal);
    setOperationAction(ISD::FDIV,               MVT::v4f32, Legal);
    setOperationAction(ISD::FSQRT,              MVT::v4f32, Legal);
    setOperationAction(ISD::FNEG,               MVT::v4f32, Custom);
    setOperationAction(ISD::LOAD,               MVT::v4f32, Legal);
    setOperationAction(ISD::BUILD_VECTOR,       MVT::v4f32, Custom);
    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v4f32, Custom);
    setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v4f32, Custom);
    setOperationAction(ISD::SELECT,             MVT::v4f32, Custom);
  }

  if (Subtarget->hasSSE2()) {
    addRegisterClass(MVT::v2f64, X86::VR128RegisterClass);
    addRegisterClass(MVT::v16i8, X86::VR128RegisterClass);
    addRegisterClass(MVT::v8i16, X86::VR128RegisterClass);
    addRegisterClass(MVT::v4i32, X86::VR128RegisterClass);
    addRegisterClass(MVT::v2i64, X86::VR128RegisterClass);

    setOperationAction(ISD::ADD,                MVT::v16i8, Legal);
    setOperationAction(ISD::ADD,                MVT::v8i16, Legal);
    setOperationAction(ISD::ADD,                MVT::v4i32, Legal);
    setOperationAction(ISD::ADD,                MVT::v2i64, Legal);
    setOperationAction(ISD::SUB,                MVT::v16i8, Legal);
    setOperationAction(ISD::SUB,                MVT::v8i16, Legal);
    setOperationAction(ISD::SUB,                MVT::v4i32, Legal);
    setOperationAction(ISD::SUB,                MVT::v2i64, Legal);
    setOperationAction(ISD::MUL,                MVT::v8i16, Legal);
    setOperationAction(ISD::FADD,               MVT::v2f64, Legal);
    setOperationAction(ISD::FSUB,               MVT::v2f64, Legal);
    setOperationAction(ISD::FMUL,               MVT::v2f64, Legal);
    setOperationAction(ISD::FDIV,               MVT::v2f64, Legal);
    setOperationAction(ISD::FSQRT,              MVT::v2f64, Legal);
    setOperationAction(ISD::FNEG,               MVT::v2f64, Custom);

    setOperationAction(ISD::SCALAR_TO_VECTOR,   MVT::v16i8, Custom);
    setOperationAction(ISD::SCALAR_TO_VECTOR,   MVT::v8i16, Custom);
    setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v8i16, Custom);
    setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v4i32, Custom);
    // Implement v4f32 insert_vector_elt in terms of SSE2 v8i16 ones.
    setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v4f32, Custom);

    // Custom lower build_vector, vector_shuffle, and extract_vector_elt.
    for (unsigned VT = (unsigned)MVT::v16i8; VT != (unsigned)MVT::v2i64; VT++) {
      setOperationAction(ISD::BUILD_VECTOR,        (MVT::ValueType)VT, Custom);
      setOperationAction(ISD::VECTOR_SHUFFLE,      (MVT::ValueType)VT, Custom);
      setOperationAction(ISD::EXTRACT_VECTOR_ELT,  (MVT::ValueType)VT, Custom);
    }
    setOperationAction(ISD::BUILD_VECTOR,       MVT::v2f64, Custom);
    setOperationAction(ISD::BUILD_VECTOR,       MVT::v2i64, Custom);
    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v2f64, Custom);
    setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v2i64, Custom);
    setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2f64, Custom);
    setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2i64, Custom);

    // Promote v16i8, v8i16, v4i32 load, select, and, or, xor to v2i64.
    for (unsigned VT = (unsigned)MVT::v16i8; VT != (unsigned)MVT::v2i64; VT++) {
      setOperationAction(ISD::AND,    (MVT::ValueType)VT, Promote);
      AddPromotedToType (ISD::AND,    (MVT::ValueType)VT, MVT::v2i64);
      setOperationAction(ISD::OR,     (MVT::ValueType)VT, Promote);
      AddPromotedToType (ISD::OR,     (MVT::ValueType)VT, MVT::v2i64);
      setOperationAction(ISD::XOR,    (MVT::ValueType)VT, Promote);
      AddPromotedToType (ISD::XOR,    (MVT::ValueType)VT, MVT::v2i64);
      setOperationAction(ISD::LOAD,   (MVT::ValueType)VT, Promote);
      AddPromotedToType (ISD::LOAD,   (MVT::ValueType)VT, MVT::v2i64);
      setOperationAction(ISD::SELECT, (MVT::ValueType)VT, Promote);
      AddPromotedToType (ISD::SELECT, (MVT::ValueType)VT, MVT::v2i64);
    }

    // Custom lower v2i64 and v2f64 selects.
    setOperationAction(ISD::LOAD,               MVT::v2f64, Legal);
    setOperationAction(ISD::LOAD,               MVT::v2i64, Legal);
    setOperationAction(ISD::SELECT,             MVT::v2f64, Custom);
    setOperationAction(ISD::SELECT,             MVT::v2i64, Custom);
  }

  // We want to custom lower some of our intrinsics.
  setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);

  // We have target-specific dag combine patterns for the following nodes:
  setTargetDAGCombine(ISD::VECTOR_SHUFFLE);
  setTargetDAGCombine(ISD::SELECT);

  computeRegisterProperties();

  // FIXME: These should be based on subtarget info. Plus, the values should
  // be smaller when we are in optimizing for size mode.
  maxStoresPerMemset = 16; // For %llvm.memset -> sequence of stores
  maxStoresPerMemcpy = 16; // For %llvm.memcpy -> sequence of stores
  maxStoresPerMemmove = 16; // For %llvm.memmove -> sequence of stores
  allowUnalignedMemoryAccesses = true; // x86 supports it!
}


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

#include "X86GenCallingConv.inc"
    
/// LowerRET - Lower an ISD::RET node.
SDOperand X86TargetLowering::LowerRET(SDOperand Op, SelectionDAG &DAG) {
  assert((Op.getNumOperands() & 1) == 1 && "ISD::RET should have odd # args");
  
  SmallVector<CCValAssign, 16> RVLocs;
  unsigned CC = DAG.getMachineFunction().getFunction()->getCallingConv();
  bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg();
  CCState CCInfo(CC, isVarArg, getTargetMachine(), RVLocs);
  CCInfo.AnalyzeReturn(Op.Val, 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().liveout_empty()) {
    for (unsigned i = 0; i != RVLocs.size(); ++i)
      if (RVLocs[i].isRegLoc())
        DAG.getMachineFunction().addLiveOut(RVLocs[i].getLocReg());
  }
  
  SDOperand Chain = Op.getOperand(0);
  SDOperand Flag;
  
  // Copy the result values into the output registers.
  if (RVLocs.size() != 1 || !RVLocs[0].isRegLoc() ||
      RVLocs[0].getLocReg() != X86::ST0) {
    for (unsigned i = 0; i != RVLocs.size(); ++i) {
      CCValAssign &VA = RVLocs[i];
      assert(VA.isRegLoc() && "Can only return in registers!");
      Chain = DAG.getCopyToReg(Chain, VA.getLocReg(), Op.getOperand(i*2+1),
                               Flag);
      Flag = Chain.getValue(1);
    }
  } else {
    // We need to handle a destination of ST0 specially, because it isn't really
    // a register.
    SDOperand Value = Op.getOperand(1);
    
    // If this is an FP return with ScalarSSE, we need to move the value from
    // an XMM register onto the fp-stack.
    if (X86ScalarSSE) {
      SDOperand MemLoc;
      
      // If this is a load into a scalarsse value, don't store the loaded value
      // back to the stack, only to reload it: just replace the scalar-sse load.
      if (ISD::isNON_EXTLoad(Value.Val) &&
          (Chain == Value.getValue(1) || Chain == Value.getOperand(0))) {
        Chain  = Value.getOperand(0);
        MemLoc = Value.getOperand(1);
      } else {
        // Spill the value to memory and reload it into top of stack.
        unsigned Size = MVT::getSizeInBits(RVLocs[0].getValVT())/8;
        MachineFunction &MF = DAG.getMachineFunction();
        int SSFI = MF.getFrameInfo()->CreateStackObject(Size, Size);
        MemLoc = DAG.getFrameIndex(SSFI, getPointerTy());
        Chain = DAG.getStore(Op.getOperand(0), Value, MemLoc, NULL, 0);
      }
      SDVTList Tys = DAG.getVTList(RVLocs[0].getValVT(), MVT::Other);
      SDOperand Ops[] = {Chain, MemLoc, DAG.getValueType(RVLocs[0].getValVT())};
      Value = DAG.getNode(X86ISD::FLD, Tys, Ops, 3);
      Chain = Value.getValue(1);
    }
    
    SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Flag);
    SDOperand Ops[] = { Chain, Value };
    Chain = DAG.getNode(X86ISD::FP_SET_RESULT, Tys, Ops, 2);
    Flag = Chain.getValue(1);
  }
  
  SDOperand BytesToPop = DAG.getConstant(getBytesToPopOnReturn(), MVT::i16);
  if (Flag.Val)
    return DAG.getNode(X86ISD::RET_FLAG, MVT::Other, Chain, BytesToPop, Flag);
  else
    return DAG.getNode(X86ISD::RET_FLAG, MVT::Other, Chain, BytesToPop);
}


/// LowerCallResult - Lower the result values of an ISD::CALL into the
/// appropriate copies out of appropriate physical registers.  This assumes that
/// Chain/InFlag are the input chain/flag to use, and that TheCall is the call
/// being lowered.  The returns a SDNode with the same number of values as the
/// ISD::CALL.
SDNode *X86TargetLowering::
LowerCallResult(SDOperand Chain, SDOperand InFlag, SDNode *TheCall, 
                unsigned CallingConv, SelectionDAG &DAG) {
  
  // Assign locations to each value returned by this call.
  SmallVector<CCValAssign, 16> RVLocs;
  bool isVarArg = cast<ConstantSDNode>(TheCall->getOperand(2))->getValue() != 0;
  CCState CCInfo(CallingConv, isVarArg, getTargetMachine(), RVLocs);
  CCInfo.AnalyzeCallResult(TheCall, RetCC_X86);

  
  SmallVector<SDOperand, 8> ResultVals;
  
  // Copy all of the result registers out of their specified physreg.
  if (RVLocs.size() != 1 || RVLocs[0].getLocReg() != X86::ST0) {
    for (unsigned i = 0; i != RVLocs.size(); ++i) {
      Chain = DAG.getCopyFromReg(Chain, RVLocs[i].getLocReg(),
                                 RVLocs[i].getValVT(), InFlag).getValue(1);
      InFlag = Chain.getValue(2);
      ResultVals.push_back(Chain.getValue(0));
    }
  } else {
    // Copies from the FP stack are special, as ST0 isn't a valid register
    // before the fp stackifier runs.
    
    // Copy ST0 into an RFP register with FP_GET_RESULT.
    SDVTList Tys = DAG.getVTList(RVLocs[0].getValVT(), MVT::Other, MVT::Flag);
    SDOperand GROps[] = { Chain, InFlag };
    SDOperand RetVal = DAG.getNode(X86ISD::FP_GET_RESULT, Tys, GROps, 2);
    Chain  = RetVal.getValue(1);
    InFlag = RetVal.getValue(2);
    
    // If we are using ScalarSSE, store ST(0) to the stack and reload it into
    // an XMM register.
    if (X86ScalarSSE) {
      // FIXME: Currently the FST is flagged to the FP_GET_RESULT. This
      // shouldn't be necessary except that RFP cannot be live across
      // multiple blocks. When stackifier is fixed, they can be uncoupled.
      MachineFunction &MF = DAG.getMachineFunction();
      int SSFI = MF.getFrameInfo()->CreateStackObject(8, 8);
      SDOperand StackSlot = DAG.getFrameIndex(SSFI, getPointerTy());
      SDOperand Ops[] = {
        Chain, RetVal, StackSlot, DAG.getValueType(RVLocs[0].getValVT()), InFlag
      };
      Chain = DAG.getNode(X86ISD::FST, MVT::Other, Ops, 5);
      RetVal = DAG.getLoad(RVLocs[0].getValVT(), Chain, StackSlot, NULL, 0);
      Chain = RetVal.getValue(1);
    }
    ResultVals.push_back(RetVal);
  }
  
  // Merge everything together with a MERGE_VALUES node.
  ResultVals.push_back(Chain);
  return DAG.getNode(ISD::MERGE_VALUES, TheCall->getVTList(),
                     &ResultVals[0], ResultVals.size()).Val;
}


//===----------------------------------------------------------------------===//
//                C & StdCall 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.

/// AddLiveIn - This helper function adds the specified physical register to the
/// MachineFunction as a live in value.  It also creates a corresponding virtual
/// register for it.
static unsigned AddLiveIn(MachineFunction &MF, unsigned PReg,
                          const TargetRegisterClass *RC) {
  assert(RC->contains(PReg) && "Not the correct regclass!");
  unsigned VReg = MF.getSSARegMap()->createVirtualRegister(RC);
  MF.addLiveIn(PReg, VReg);
  return VReg;
}

SDOperand X86TargetLowering::LowerCCCArguments(SDOperand Op, SelectionDAG &DAG,
                                               bool isStdCall) {
  unsigned NumArgs = Op.Val->getNumValues() - 1;
  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  SDOperand Root = Op.getOperand(0);
  bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;

  // Assign locations to all of the incoming arguments.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(MF.getFunction()->getCallingConv(), isVarArg,
                 getTargetMachine(), ArgLocs);
  CCInfo.AnalyzeFormalArguments(Op.Val, CC_X86_32_C);
   
  SmallVector<SDOperand, 8> ArgValues;
  unsigned LastVal = ~0U;
  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()) {
      MVT::ValueType RegVT = VA.getLocVT();
      TargetRegisterClass *RC;
      if (RegVT == MVT::i32)
        RC = X86::GR32RegisterClass;
      else {
        assert(MVT::isVector(RegVT));
        RC = X86::VR128RegisterClass;
      }
      
      unsigned Reg = AddLiveIn(DAG.getMachineFunction(), VA.getLocReg(), RC);
      SDOperand ArgValue = DAG.getCopyFromReg(Root, 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, RegVT, ArgValue,
                               DAG.getValueType(VA.getValVT()));
      else if (VA.getLocInfo() == CCValAssign::ZExt)
        ArgValue = DAG.getNode(ISD::AssertZext, RegVT, ArgValue,
                               DAG.getValueType(VA.getValVT()));
      
      if (VA.getLocInfo() != CCValAssign::Full)
        ArgValue = DAG.getNode(ISD::TRUNCATE, VA.getValVT(), ArgValue);
      
      ArgValues.push_back(ArgValue);
    } else {
      assert(VA.isMemLoc());
      
      // Create the nodes corresponding to a load from this parameter slot.
      int FI = MFI->CreateFixedObject(MVT::getSizeInBits(VA.getValVT())/8,
                                      VA.getLocMemOffset());
      SDOperand FIN = DAG.getFrameIndex(FI, getPointerTy());
      ArgValues.push_back(DAG.getLoad(VA.getValVT(), Root, FIN, NULL, 0));
    }
  }
  
  unsigned StackSize = CCInfo.getNextStackOffset();

  ArgValues.push_back(Root);

  // 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)
    VarArgsFrameIndex = MFI->CreateFixedObject(1, StackSize);

  if (isStdCall && !isVarArg) {
    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 (NumArgs &&
        (cast<ConstantSDNode>(Op.getOperand(3))->getValue() &
         ISD::ParamFlags::StructReturn))
      BytesToPopOnReturn = 4;  
    
    BytesCallerReserves = StackSize;
  }
  
  RegSaveFrameIndex = 0xAAAAAAA;  // X86-64 only.
  ReturnAddrIndex = 0;            // No return address slot generated yet.

  MF.getInfo<X86MachineFunctionInfo>()
    ->setBytesToPopOnReturn(BytesToPopOnReturn);

  // Return the new list of results.
  return DAG.getNode(ISD::MERGE_VALUES, Op.Val->getVTList(),
                     &ArgValues[0], ArgValues.size()).getValue(Op.ResNo);
}

SDOperand X86TargetLowering::LowerCCCCallTo(SDOperand Op, SelectionDAG &DAG,
                                            unsigned CC) {
  SDOperand Chain     = Op.getOperand(0);
  bool isVarArg       = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;
  bool isTailCall     = cast<ConstantSDNode>(Op.getOperand(3))->getValue() != 0;
  SDOperand Callee    = Op.getOperand(4);
  unsigned NumOps     = (Op.getNumOperands() - 5) / 2;

  // Analyze operands of the call, assigning locations to each operand.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(CC, isVarArg, getTargetMachine(), ArgLocs);
  CCInfo.AnalyzeCallOperands(Op.Val, CC_X86_32_C);
  
  // Get a count of how many bytes are to be pushed on the stack.
  unsigned NumBytes = CCInfo.getNextStackOffset();

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

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

  SDOperand StackPtr;

  // Walk the register/memloc assignments, inserting copies/loads.
  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
    CCValAssign &VA = ArgLocs[i];
    SDOperand Arg = Op.getOperand(5+2*VA.getValNo());
    
    // Promote the value if needed.
    switch (VA.getLocInfo()) {
    default: assert(0 && "Unknown loc info!");
    case CCValAssign::Full: break;
    case CCValAssign::SExt:
      Arg = DAG.getNode(ISD::SIGN_EXTEND, VA.getLocVT(), Arg);
      break;
    case CCValAssign::ZExt:
      Arg = DAG.getNode(ISD::ZERO_EXTEND, VA.getLocVT(), Arg);
      break;
    case CCValAssign::AExt:
      Arg = DAG.getNode(ISD::ANY_EXTEND, VA.getLocVT(), Arg);
      break;
    }
    
    if (VA.isRegLoc()) {
      RegsToPass.push_back(std::make_pair(VA.getLocReg(), Arg));
    } else {
      assert(VA.isMemLoc());
      if (StackPtr.Val == 0)
        StackPtr = DAG.getRegister(getStackPtrReg(), getPointerTy());
      SDOperand PtrOff = DAG.getConstant(VA.getLocMemOffset(), getPointerTy());
      PtrOff = DAG.getNode(ISD::ADD, getPointerTy(), StackPtr, PtrOff);
      MemOpChains.push_back(DAG.getStore(Chain, Arg, PtrOff, NULL, 0));
    }
  }

  // If the first argument is an sret pointer, remember it.
  bool isSRet = NumOps &&
    (cast<ConstantSDNode>(Op.getOperand(6))->getValue() &
     ISD::ParamFlags::StructReturn);
  
  if (!MemOpChains.empty())
    Chain = DAG.getNode(ISD::TokenFactor, 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.
  SDOperand InFlag;
  for (unsigned i = 0, e = RegsToPass.size(); i != e; ++i) {
    Chain = DAG.getCopyToReg(Chain, RegsToPass[i].first, RegsToPass[i].second,
                             InFlag);
    InFlag = Chain.getValue(1);
  }

  // ELF / PIC requires GOT in the EBX register before function calls via PLT
  // GOT pointer.
  if (getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
      Subtarget->isPICStyleGOT()) {
    Chain = DAG.getCopyToReg(Chain, X86::EBX,
                             DAG.getNode(X86ISD::GlobalBaseReg, getPointerTy()),
                             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.
  if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
    // We should use extra load for direct calls to dllimported functions in
    // non-JIT mode.
    if (!Subtarget->GVRequiresExtraLoad(G->getGlobal(),
                                        getTargetMachine(), true))
      Callee = DAG.getTargetGlobalAddress(G->getGlobal(), getPointerTy());
  } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee))
    Callee = DAG.getTargetExternalSymbol(S->getSymbol(), getPointerTy());

  // Returns a chain & a flag for retval copy to use.
  SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
  SmallVector<SDOperand, 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 an implicit use GOT pointer in EBX.
  if (getTargetMachine().getRelocationModel() == Reloc::PIC_ &&
      Subtarget->isPICStyleGOT())
    Ops.push_back(DAG.getRegister(X86::EBX, getPointerTy()));
  
  if (InFlag.Val)
    Ops.push_back(InFlag);

  Chain = DAG.getNode(isTailCall ? X86ISD::TAILCALL : X86ISD::CALL,
                      NodeTys, &Ops[0], Ops.size());
  InFlag = Chain.getValue(1);

  // Create the CALLSEQ_END node.
  unsigned NumBytesForCalleeToPush = 0;

  if (CC == CallingConv::X86_StdCall) {
    if (isVarArg)
      NumBytesForCalleeToPush = isSRet ? 4 : 0;
    else
      NumBytesForCalleeToPush = NumBytes;
  } else {
    // If this is is a call to a struct-return function, the callee
    // pops the hidden struct pointer, so we have to push it back.
    // This is common for Darwin/X86, Linux & Mingw32 targets.
    NumBytesForCalleeToPush = isSRet ? 4 : 0;
  }
  
  NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
  Ops.clear();
  Ops.push_back(Chain);
  Ops.push_back(DAG.getConstant(NumBytes, getPointerTy()));
  Ops.push_back(DAG.getConstant(NumBytesForCalleeToPush, getPointerTy()));
  Ops.push_back(InFlag);
  Chain = DAG.getNode(ISD::CALLSEQ_END, NodeTys, &Ops[0], Ops.size());
  InFlag = Chain.getValue(1);

  // Handle result values, copying them out of physregs into vregs that we
  // return.
  return SDOperand(LowerCallResult(Chain, InFlag, Op.Val, CC, DAG), Op.ResNo);
}


//===----------------------------------------------------------------------===//
//                   FastCall Calling Convention implementation
//===----------------------------------------------------------------------===//
//
// The X86 'fastcall' calling convention passes up to two integer arguments in
// registers (an appropriate portion of ECX/EDX), passes arguments in C order,
// and requires that the callee pop its arguments off the stack (allowing proper
// tail calls), and has the same return value conventions as C calling convs.
//
// This calling convention always arranges for the callee pop value to be 8n+4
// bytes, which is needed for tail recursion elimination and stack alignment
// reasons.
SDOperand
X86TargetLowering::LowerFastCCArguments(SDOperand Op, SelectionDAG &DAG) {
  MachineFunction &MF = DAG.getMachineFunction();
  MachineFrameInfo *MFI = MF.getFrameInfo();
  SDOperand Root = Op.getOperand(0);
  bool isVarArg = cast<ConstantSDNode>(Op.getOperand(2))->getValue() != 0;

  // Assign locations to all of the incoming arguments.
  SmallVector<CCValAssign, 16> ArgLocs;
  CCState CCInfo(MF.getFunction()->getCallingConv(), isVarArg,
                 getTargetMachine(), ArgLocs);
  CCInfo.AnalyzeFormalArguments(Op.Val, CC_X86_32_FastCall);
  
  SmallVector<SDOperand, 8> ArgValues;
  unsigned LastVal = ~0U;
  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()) {
      MVT::ValueType RegVT = VA.getLocVT();
      TargetRegisterClass *RC;
      if (RegVT == MVT::i32)
        RC = X86::GR32RegisterClass;
      else {
        assert(MVT::isVector(RegVT));
        RC = X86::VR128RegisterClass;
      }
      
      unsigned Reg = AddLiveIn(DAG.getMachineFunction(), VA.getLocReg(), RC);
      SDOperand ArgValue = DAG.getCopyFromReg(Root, 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, RegVT, ArgValue,
                               DAG.getValueType(VA.getValVT()));
      else if (VA.getLocInfo() == CCValAssign::ZExt)
        ArgValue = DAG.getNode(ISD::AssertZext, RegVT, ArgValue,
                               DAG.getValueType(VA.getValVT()));
      
      if (VA.getLocInfo() != CCValAssign::Full)
        ArgValue = DAG.getNode(ISD::TRUNCATE, VA.getValVT(), ArgValue);
      
      ArgValues.push_back(ArgValue);
    } else {
      assert(VA.isMemLoc());
      
      // Create the nodes corresponding to a load from this parameter slot.
      int FI = MFI->CreateFixedObject(MVT::getSizeInBits(VA.getValVT())/8,
                                      VA.getLocMemOffset());
      SDOperand FIN = DAG.getFrameIndex(FI, getPointerTy());