[oota-llvm.git] / lib / CodeGen / SelectionDAG / SelectionDAGBuilder.cpp

//===-- SelectionDAGBuilder.cpp - Selection-DAG building ------------------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This implements routines for translating from LLVM IR into SelectionDAG IR.
//
//===----------------------------------------------------------------------===//

#include "SelectionDAGBuilder.h"
#include "SDNodeDbgValue.h"
#include "llvm/ADT/BitVector.h"
#include "llvm/ADT/Optional.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/BranchProbabilityInfo.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/FastISel.h"
#include "llvm/CodeGen/FunctionLoweringInfo.h"
#include "llvm/CodeGen/GCMetadata.h"
#include "llvm/CodeGen/GCStrategy.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/StackMaps.h"
#include "llvm/CodeGen/WinEHFuncInfo.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DebugInfo.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GlobalVariable.h"
#include "llvm/IR/InlineAsm.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Statepoint.h"
#include "llvm/MC/MCSymbol.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetFrameLowering.h"
#include "llvm/Target/TargetInstrInfo.h"
#include "llvm/Target/TargetIntrinsicInfo.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Target/TargetOptions.h"
#include "llvm/Target/TargetSelectionDAGInfo.h"
#include "llvm/Target/TargetSubtargetInfo.h"
#include <algorithm>
using namespace llvm;

#define DEBUG_TYPE "isel"

/// LimitFloatPrecision - Generate low-precision inline sequences for
/// some float libcalls (6, 8 or 12 bits).
static unsigned LimitFloatPrecision;

static cl::opt<unsigned, true>
LimitFPPrecision("limit-float-precision",
                 cl::desc("Generate low-precision inline sequences "
                          "for some float libcalls"),
                 cl::location(LimitFloatPrecision),
                 cl::init(0));

// Limit the width of DAG chains. This is important in general to prevent
// prevent DAG-based analysis from blowing up. For example, alias analysis and
// load clustering may not complete in reasonable time. It is difficult to
// recognize and avoid this situation within each individual analysis, and
// future analyses are likely to have the same behavior. Limiting DAG width is
// the safe approach, and will be especially important with global DAGs.
//
// MaxParallelChains default is arbitrarily high to avoid affecting
// optimization, but could be lowered to improve compile time. Any ld-ld-st-st
// sequence over this should have been converted to llvm.memcpy by the
// frontend. It easy to induce this behavior with .ll code such as:
// %buffer = alloca [4096 x i8]
// %data = load [4096 x i8]* %argPtr
// store [4096 x i8] %data, [4096 x i8]* %buffer
static const unsigned MaxParallelChains = 64;

static SDValue getCopyFromPartsVector(SelectionDAG &DAG, SDLoc DL,
                                      const SDValue *Parts, unsigned NumParts,
                                      MVT PartVT, EVT ValueVT, const Value *V);

/// getCopyFromParts - Create a value that contains the specified legal parts
/// combined into the value they represent.  If the parts combine to a type
/// larger then ValueVT then AssertOp can be used to specify whether the extra
/// bits are known to be zero (ISD::AssertZext) or sign extended from ValueVT
/// (ISD::AssertSext).
static SDValue getCopyFromParts(SelectionDAG &DAG, SDLoc DL,
                                const SDValue *Parts,
                                unsigned NumParts, MVT PartVT, EVT ValueVT,
                                const Value *V,
                                ISD::NodeType AssertOp = ISD::DELETED_NODE) {
  if (ValueVT.isVector())
    return getCopyFromPartsVector(DAG, DL, Parts, NumParts,
                                  PartVT, ValueVT, V);

  assert(NumParts > 0 && "No parts to assemble!");
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue Val = Parts[0];

  if (NumParts > 1) {
    // Assemble the value from multiple parts.
    if (ValueVT.isInteger()) {
      unsigned PartBits = PartVT.getSizeInBits();
      unsigned ValueBits = ValueVT.getSizeInBits();

      // Assemble the power of 2 part.
      unsigned RoundParts = NumParts & (NumParts - 1) ?
        1 << Log2_32(NumParts) : NumParts;
      unsigned RoundBits = PartBits * RoundParts;
      EVT RoundVT = RoundBits == ValueBits ?
        ValueVT : EVT::getIntegerVT(*DAG.getContext(), RoundBits);
      SDValue Lo, Hi;

      EVT HalfVT = EVT::getIntegerVT(*DAG.getContext(), RoundBits/2);

      if (RoundParts > 2) {
        Lo = getCopyFromParts(DAG, DL, Parts, RoundParts / 2,
                              PartVT, HalfVT, V);
        Hi = getCopyFromParts(DAG, DL, Parts + RoundParts / 2,
                              RoundParts / 2, PartVT, HalfVT, V);
      } else {
        Lo = DAG.getNode(ISD::BITCAST, DL, HalfVT, Parts[0]);
        Hi = DAG.getNode(ISD::BITCAST, DL, HalfVT, Parts[1]);
      }

      if (TLI.isBigEndian())
        std::swap(Lo, Hi);

      Val = DAG.getNode(ISD::BUILD_PAIR, DL, RoundVT, Lo, Hi);

      if (RoundParts < NumParts) {
        // Assemble the trailing non-power-of-2 part.
        unsigned OddParts = NumParts - RoundParts;
        EVT OddVT = EVT::getIntegerVT(*DAG.getContext(), OddParts * PartBits);
        Hi = getCopyFromParts(DAG, DL,
                              Parts + RoundParts, OddParts, PartVT, OddVT, V);

        // Combine the round and odd parts.
        Lo = Val;
        if (TLI.isBigEndian())
          std::swap(Lo, Hi);
        EVT TotalVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
        Hi = DAG.getNode(ISD::ANY_EXTEND, DL, TotalVT, Hi);
        Hi = DAG.getNode(ISD::SHL, DL, TotalVT, Hi,
                         DAG.getConstant(Lo.getValueType().getSizeInBits(),
                                         TLI.getPointerTy()));
        Lo = DAG.getNode(ISD::ZERO_EXTEND, DL, TotalVT, Lo);
        Val = DAG.getNode(ISD::OR, DL, TotalVT, Lo, Hi);
      }
    } else if (PartVT.isFloatingPoint()) {
      // FP split into multiple FP parts (for ppcf128)
      assert(ValueVT == EVT(MVT::ppcf128) && PartVT == MVT::f64 &&
             "Unexpected split");
      SDValue Lo, Hi;
      Lo = DAG.getNode(ISD::BITCAST, DL, EVT(MVT::f64), Parts[0]);
      Hi = DAG.getNode(ISD::BITCAST, DL, EVT(MVT::f64), Parts[1]);
      if (TLI.hasBigEndianPartOrdering(ValueVT))
        std::swap(Lo, Hi);
      Val = DAG.getNode(ISD::BUILD_PAIR, DL, ValueVT, Lo, Hi);
    } else {
      // FP split into integer parts (soft fp)
      assert(ValueVT.isFloatingPoint() && PartVT.isInteger() &&
             !PartVT.isVector() && "Unexpected split");
      EVT IntVT = EVT::getIntegerVT(*DAG.getContext(), ValueVT.getSizeInBits());
      Val = getCopyFromParts(DAG, DL, Parts, NumParts, PartVT, IntVT, V);
    }
  }

  // There is now one part, held in Val.  Correct it to match ValueVT.
  EVT PartEVT = Val.getValueType();

  if (PartEVT == ValueVT)
    return Val;

  if (PartEVT.isInteger() && ValueVT.isInteger()) {
    if (ValueVT.bitsLT(PartEVT)) {
      // For a truncate, see if we have any information to
      // indicate whether the truncated bits will always be
      // zero or sign-extension.
      if (AssertOp != ISD::DELETED_NODE)
        Val = DAG.getNode(AssertOp, DL, PartEVT, Val,
                          DAG.getValueType(ValueVT));
      return DAG.getNode(ISD::TRUNCATE, DL, ValueVT, Val);
    }
    return DAG.getNode(ISD::ANY_EXTEND, DL, ValueVT, Val);
  }

  if (PartEVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
    // FP_ROUND's are always exact here.
    if (ValueVT.bitsLT(Val.getValueType()))
      return DAG.getNode(ISD::FP_ROUND, DL, ValueVT, Val,
                         DAG.getTargetConstant(1, TLI.getPointerTy()));

    return DAG.getNode(ISD::FP_EXTEND, DL, ValueVT, Val);
  }

  if (PartEVT.getSizeInBits() == ValueVT.getSizeInBits())
    return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);

  llvm_unreachable("Unknown mismatch!");
}

static void diagnosePossiblyInvalidConstraint(LLVMContext &Ctx, const Value *V,
                                              const Twine &ErrMsg) {
  const Instruction *I = dyn_cast_or_null<Instruction>(V);
  if (!V)
    return Ctx.emitError(ErrMsg);

  const char *AsmError = ", possible invalid constraint for vector type";
  if (const CallInst *CI = dyn_cast<CallInst>(I))
    if (isa<InlineAsm>(CI->getCalledValue()))
      return Ctx.emitError(I, ErrMsg + AsmError);

  return Ctx.emitError(I, ErrMsg);
}

/// getCopyFromPartsVector - Create a value that contains the specified legal
/// parts combined into the value they represent.  If the parts combine to a
/// type larger then ValueVT then AssertOp can be used to specify whether the
/// extra bits are known to be zero (ISD::AssertZext) or sign extended from
/// ValueVT (ISD::AssertSext).
static SDValue getCopyFromPartsVector(SelectionDAG &DAG, SDLoc DL,
                                      const SDValue *Parts, unsigned NumParts,
                                      MVT PartVT, EVT ValueVT, const Value *V) {
  assert(ValueVT.isVector() && "Not a vector value");
  assert(NumParts > 0 && "No parts to assemble!");
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue Val = Parts[0];

  // Handle a multi-element vector.
  if (NumParts > 1) {
    EVT IntermediateVT;
    MVT RegisterVT;
    unsigned NumIntermediates;
    unsigned NumRegs =
    TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT, IntermediateVT,
                               NumIntermediates, RegisterVT);
    assert(NumRegs == NumParts && "Part count doesn't match vector breakdown!");
    NumParts = NumRegs; // Silence a compiler warning.
    assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!");
    assert(RegisterVT == Parts[0].getSimpleValueType() &&
           "Part type doesn't match part!");

    // Assemble the parts into intermediate operands.
    SmallVector<SDValue, 8> Ops(NumIntermediates);
    if (NumIntermediates == NumParts) {
      // If the register was not expanded, truncate or copy the value,
      // as appropriate.
      for (unsigned i = 0; i != NumParts; ++i)
        Ops[i] = getCopyFromParts(DAG, DL, &Parts[i], 1,
                                  PartVT, IntermediateVT, V);
    } else if (NumParts > 0) {
      // If the intermediate type was expanded, build the intermediate
      // operands from the parts.
      assert(NumParts % NumIntermediates == 0 &&
             "Must expand into a divisible number of parts!");
      unsigned Factor = NumParts / NumIntermediates;
      for (unsigned i = 0; i != NumIntermediates; ++i)
        Ops[i] = getCopyFromParts(DAG, DL, &Parts[i * Factor], Factor,
                                  PartVT, IntermediateVT, V);
    }

    // Build a vector with BUILD_VECTOR or CONCAT_VECTORS from the
    // intermediate operands.
    Val = DAG.getNode(IntermediateVT.isVector() ? ISD::CONCAT_VECTORS
                                                : ISD::BUILD_VECTOR,
                      DL, ValueVT, Ops);
  }

  // There is now one part, held in Val.  Correct it to match ValueVT.
  EVT PartEVT = Val.getValueType();

  if (PartEVT == ValueVT)
    return Val;

  if (PartEVT.isVector()) {
    // If the element type of the source/dest vectors are the same, but the
    // parts vector has more elements than the value vector, then we have a
    // vector widening case (e.g. <2 x float> -> <4 x float>).  Extract the
    // elements we want.
    if (PartEVT.getVectorElementType() == ValueVT.getVectorElementType()) {
      assert(PartEVT.getVectorNumElements() > ValueVT.getVectorNumElements() &&
             "Cannot narrow, it would be a lossy transformation");
      return DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL, ValueVT, Val,
                         DAG.getConstant(0, TLI.getVectorIdxTy()));
    }

    // Vector/Vector bitcast.
    if (ValueVT.getSizeInBits() == PartEVT.getSizeInBits())
      return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);

    assert(PartEVT.getVectorNumElements() == ValueVT.getVectorNumElements() &&
      "Cannot handle this kind of promotion");
    // Promoted vector extract
    bool Smaller = ValueVT.bitsLE(PartEVT);
    return DAG.getNode((Smaller ? ISD::TRUNCATE : ISD::ANY_EXTEND),
                       DL, ValueVT, Val);

  }

  // Trivial bitcast if the types are the same size and the destination
  // vector type is legal.
  if (PartEVT.getSizeInBits() == ValueVT.getSizeInBits() &&
      TLI.isTypeLegal(ValueVT))
    return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);

  // Handle cases such as i8 -> <1 x i1>
  if (ValueVT.getVectorNumElements() != 1) {
    diagnosePossiblyInvalidConstraint(*DAG.getContext(), V,
                                      "non-trivial scalar-to-vector conversion");
    return DAG.getUNDEF(ValueVT);
  }

  if (ValueVT.getVectorNumElements() == 1 &&
      ValueVT.getVectorElementType() != PartEVT) {
    bool Smaller = ValueVT.bitsLE(PartEVT);
    Val = DAG.getNode((Smaller ? ISD::TRUNCATE : ISD::ANY_EXTEND),
                       DL, ValueVT.getScalarType(), Val);
  }

  return DAG.getNode(ISD::BUILD_VECTOR, DL, ValueVT, Val);
}

static void getCopyToPartsVector(SelectionDAG &DAG, SDLoc dl,
                                 SDValue Val, SDValue *Parts, unsigned NumParts,
                                 MVT PartVT, const Value *V);

/// getCopyToParts - Create a series of nodes that contain the specified value
/// split into legal parts.  If the parts contain more bits than Val, then, for
/// integers, ExtendKind can be used to specify how to generate the extra bits.
static void getCopyToParts(SelectionDAG &DAG, SDLoc DL,
                           SDValue Val, SDValue *Parts, unsigned NumParts,
                           MVT PartVT, const Value *V,
                           ISD::NodeType ExtendKind = ISD::ANY_EXTEND) {
  EVT ValueVT = Val.getValueType();

  // Handle the vector case separately.
  if (ValueVT.isVector())
    return getCopyToPartsVector(DAG, DL, Val, Parts, NumParts, PartVT, V);

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  unsigned PartBits = PartVT.getSizeInBits();
  unsigned OrigNumParts = NumParts;
  assert(TLI.isTypeLegal(PartVT) && "Copying to an illegal type!");

  if (NumParts == 0)
    return;

  assert(!ValueVT.isVector() && "Vector case handled elsewhere");
  EVT PartEVT = PartVT;
  if (PartEVT == ValueVT) {
    assert(NumParts == 1 && "No-op copy with multiple parts!");
    Parts[0] = Val;
    return;
  }

  if (NumParts * PartBits > ValueVT.getSizeInBits()) {
    // If the parts cover more bits than the value has, promote the value.
    if (PartVT.isFloatingPoint() && ValueVT.isFloatingPoint()) {
      assert(NumParts == 1 && "Do not know what to promote to!");
      Val = DAG.getNode(ISD::FP_EXTEND, DL, PartVT, Val);
    } else {
      assert((PartVT.isInteger() || PartVT == MVT::x86mmx) &&
             ValueVT.isInteger() &&
             "Unknown mismatch!");
      ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
      Val = DAG.getNode(ExtendKind, DL, ValueVT, Val);
      if (PartVT == MVT::x86mmx)
        Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
    }
  } else if (PartBits == ValueVT.getSizeInBits()) {
    // Different types of the same size.
    assert(NumParts == 1 && PartEVT != ValueVT);
    Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
  } else if (NumParts * PartBits < ValueVT.getSizeInBits()) {
    // If the parts cover less bits than value has, truncate the value.
    assert((PartVT.isInteger() || PartVT == MVT::x86mmx) &&
           ValueVT.isInteger() &&
           "Unknown mismatch!");
    ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
    Val = DAG.getNode(ISD::TRUNCATE, DL, ValueVT, Val);
    if (PartVT == MVT::x86mmx)
      Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
  }

  // The value may have changed - recompute ValueVT.
  ValueVT = Val.getValueType();
  assert(NumParts * PartBits == ValueVT.getSizeInBits() &&
         "Failed to tile the value with PartVT!");

  if (NumParts == 1) {
    if (PartEVT != ValueVT)
      diagnosePossiblyInvalidConstraint(*DAG.getContext(), V,
                                        "scalar-to-vector conversion failed");

    Parts[0] = Val;
    return;
  }

  // Expand the value into multiple parts.
  if (NumParts & (NumParts - 1)) {
    // The number of parts is not a power of 2.  Split off and copy the tail.
    assert(PartVT.isInteger() && ValueVT.isInteger() &&
           "Do not know what to expand to!");
    unsigned RoundParts = 1 << Log2_32(NumParts);
    unsigned RoundBits = RoundParts * PartBits;
    unsigned OddParts = NumParts - RoundParts;
    SDValue OddVal = DAG.getNode(ISD::SRL, DL, ValueVT, Val,
                                 DAG.getIntPtrConstant(RoundBits));
    getCopyToParts(DAG, DL, OddVal, Parts + RoundParts, OddParts, PartVT, V);

    if (TLI.isBigEndian())
      // The odd parts were reversed by getCopyToParts - unreverse them.
      std::reverse(Parts + RoundParts, Parts + NumParts);

    NumParts = RoundParts;
    ValueVT = EVT::getIntegerVT(*DAG.getContext(), NumParts * PartBits);
    Val = DAG.getNode(ISD::TRUNCATE, DL, ValueVT, Val);
  }

  // The number of parts is a power of 2.  Repeatedly bisect the value using
  // EXTRACT_ELEMENT.
  Parts[0] = DAG.getNode(ISD::BITCAST, DL,
                         EVT::getIntegerVT(*DAG.getContext(),
                                           ValueVT.getSizeInBits()),
                         Val);

  for (unsigned StepSize = NumParts; StepSize > 1; StepSize /= 2) {
    for (unsigned i = 0; i < NumParts; i += StepSize) {
      unsigned ThisBits = StepSize * PartBits / 2;
      EVT ThisVT = EVT::getIntegerVT(*DAG.getContext(), ThisBits);
      SDValue &Part0 = Parts[i];
      SDValue &Part1 = Parts[i+StepSize/2];

      Part1 = DAG.getNode(ISD::EXTRACT_ELEMENT, DL,
                          ThisVT, Part0, DAG.getIntPtrConstant(1));
      Part0 = DAG.getNode(ISD::EXTRACT_ELEMENT, DL,
                          ThisVT, Part0, DAG.getIntPtrConstant(0));

      if (ThisBits == PartBits && ThisVT != PartVT) {
        Part0 = DAG.getNode(ISD::BITCAST, DL, PartVT, Part0);
        Part1 = DAG.getNode(ISD::BITCAST, DL, PartVT, Part1);
      }
    }
  }

  if (TLI.isBigEndian())
    std::reverse(Parts, Parts + OrigNumParts);
}


/// getCopyToPartsVector - Create a series of nodes that contain the specified
/// value split into legal parts.
static void getCopyToPartsVector(SelectionDAG &DAG, SDLoc DL,
                                 SDValue Val, SDValue *Parts, unsigned NumParts,
                                 MVT PartVT, const Value *V) {
  EVT ValueVT = Val.getValueType();
  assert(ValueVT.isVector() && "Not a vector");
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();

  if (NumParts == 1) {
    EVT PartEVT = PartVT;
    if (PartEVT == ValueVT) {
      // Nothing to do.
    } else if (PartVT.getSizeInBits() == ValueVT.getSizeInBits()) {
      // Bitconvert vector->vector case.
      Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
    } else if (PartVT.isVector() &&
               PartEVT.getVectorElementType() == ValueVT.getVectorElementType() &&
               PartEVT.getVectorNumElements() > ValueVT.getVectorNumElements()) {
      EVT ElementVT = PartVT.getVectorElementType();
      // Vector widening case, e.g. <2 x float> -> <4 x float>.  Shuffle in
      // undef elements.
      SmallVector<SDValue, 16> Ops;
      for (unsigned i = 0, e = ValueVT.getVectorNumElements(); i != e; ++i)
        Ops.push_back(DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
                                  ElementVT, Val, DAG.getConstant(i,
                                                  TLI.getVectorIdxTy())));

      for (unsigned i = ValueVT.getVectorNumElements(),
           e = PartVT.getVectorNumElements(); i != e; ++i)
        Ops.push_back(DAG.getUNDEF(ElementVT));

      Val = DAG.getNode(ISD::BUILD_VECTOR, DL, PartVT, Ops);

      // FIXME: Use CONCAT for 2x -> 4x.

      //SDValue UndefElts = DAG.getUNDEF(VectorTy);
      //Val = DAG.getNode(ISD::CONCAT_VECTORS, DL, PartVT, Val, UndefElts);
    } else if (PartVT.isVector() &&
               PartEVT.getVectorElementType().bitsGE(
                 ValueVT.getVectorElementType()) &&
               PartEVT.getVectorNumElements() == ValueVT.getVectorNumElements()) {

      // Promoted vector extract
      bool Smaller = PartEVT.bitsLE(ValueVT);
      Val = DAG.getNode((Smaller ? ISD::TRUNCATE : ISD::ANY_EXTEND),
                        DL, PartVT, Val);
    } else{
      // Vector -> scalar conversion.
      assert(ValueVT.getVectorNumElements() == 1 &&
             "Only trivial vector-to-scalar conversions should get here!");
      Val = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
                        PartVT, Val, DAG.getConstant(0, TLI.getVectorIdxTy()));

      bool Smaller = ValueVT.bitsLE(PartVT);
      Val = DAG.getNode((Smaller ? ISD::TRUNCATE : ISD::ANY_EXTEND),
                         DL, PartVT, Val);
    }

    Parts[0] = Val;
    return;
  }

  // Handle a multi-element vector.
  EVT IntermediateVT;
  MVT RegisterVT;
  unsigned NumIntermediates;
  unsigned NumRegs = TLI.getVectorTypeBreakdown(*DAG.getContext(), ValueVT,
                                                IntermediateVT,
                                                NumIntermediates, RegisterVT);
  unsigned NumElements = ValueVT.getVectorNumElements();

  assert(NumRegs == NumParts && "Part count doesn't match vector breakdown!");
  NumParts = NumRegs; // Silence a compiler warning.
  assert(RegisterVT == PartVT && "Part type doesn't match vector breakdown!");

  // Split the vector into intermediate operands.
  SmallVector<SDValue, 8> Ops(NumIntermediates);
  for (unsigned i = 0; i != NumIntermediates; ++i) {
    if (IntermediateVT.isVector())
      Ops[i] = DAG.getNode(ISD::EXTRACT_SUBVECTOR, DL,
                           IntermediateVT, Val,
                   DAG.getConstant(i * (NumElements / NumIntermediates),
                                   TLI.getVectorIdxTy()));
    else
      Ops[i] = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL,
                           IntermediateVT, Val,
                           DAG.getConstant(i, TLI.getVectorIdxTy()));
  }

  // Split the intermediate operands into legal parts.
  if (NumParts == NumIntermediates) {
    // If the register was not expanded, promote or copy the value,
    // as appropriate.
    for (unsigned i = 0; i != NumParts; ++i)
      getCopyToParts(DAG, DL, Ops[i], &Parts[i], 1, PartVT, V);
  } else if (NumParts > 0) {
    // If the intermediate type was expanded, split each the value into
    // legal parts.
    assert(NumIntermediates != 0 && "division by zero");
    assert(NumParts % NumIntermediates == 0 &&
           "Must expand into a divisible number of parts!");
    unsigned Factor = NumParts / NumIntermediates;
    for (unsigned i = 0; i != NumIntermediates; ++i)
      getCopyToParts(DAG, DL, Ops[i], &Parts[i*Factor], Factor, PartVT, V);
  }
}

namespace {
  /// RegsForValue - This struct represents the registers (physical or virtual)
  /// that a particular set of values is assigned, and the type information
  /// about the value. The most common situation is to represent one value at a
  /// time, but struct or array values are handled element-wise as multiple
  /// values.  The splitting of aggregates is performed recursively, so that we
  /// never have aggregate-typed registers. The values at this point do not
  /// necessarily have legal types, so each value may require one or more
  /// registers of some legal type.
  ///
  struct RegsForValue {
    /// ValueVTs - The value types of the values, which may not be legal, and
    /// may need be promoted or synthesized from one or more registers.
    ///
    SmallVector<EVT, 4> ValueVTs;

    /// RegVTs - The value types of the registers. This is the same size as
    /// ValueVTs and it records, for each value, what the type of the assigned
    /// register or registers are. (Individual values are never synthesized
    /// from more than one type of register.)
    ///
    /// With virtual registers, the contents of RegVTs is redundant with TLI's
    /// getRegisterType member function, however when with physical registers
    /// it is necessary to have a separate record of the types.
    ///
    SmallVector<MVT, 4> RegVTs;

    /// Regs - This list holds the registers assigned to the values.
    /// Each legal or promoted value requires one register, and each
    /// expanded value requires multiple registers.
    ///
    SmallVector<unsigned, 4> Regs;

    RegsForValue() {}

    RegsForValue(const SmallVector<unsigned, 4> &regs,
                 MVT regvt, EVT valuevt)
      : ValueVTs(1, valuevt), RegVTs(1, regvt), Regs(regs) {}

    RegsForValue(LLVMContext &Context, const TargetLowering &tli,
                 unsigned Reg, Type *Ty) {
      ComputeValueVTs(tli, Ty, ValueVTs);

      for (unsigned Value = 0, e = ValueVTs.size(); Value != e; ++Value) {
        EVT ValueVT = ValueVTs[Value];
        unsigned NumRegs = tli.getNumRegisters(Context, ValueVT);
        MVT RegisterVT = tli.getRegisterType(Context, ValueVT);
        for (unsigned i = 0; i != NumRegs; ++i)
          Regs.push_back(Reg + i);
        RegVTs.push_back(RegisterVT);
        Reg += NumRegs;
      }
    }

    /// append - Add the specified values to this one.
    void append(const RegsForValue &RHS) {
      ValueVTs.append(RHS.ValueVTs.begin(), RHS.ValueVTs.end());
      RegVTs.append(RHS.RegVTs.begin(), RHS.RegVTs.end());
      Regs.append(RHS.Regs.begin(), RHS.Regs.end());
    }

    /// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
    /// this value and returns the result as a ValueVTs value.  This uses
    /// Chain/Flag as the input and updates them for the output Chain/Flag.
    /// If the Flag pointer is NULL, no flag is used.
    SDValue getCopyFromRegs(SelectionDAG &DAG, FunctionLoweringInfo &FuncInfo,
                            SDLoc dl,
                            SDValue &Chain, SDValue *Flag,
                            const Value *V = nullptr) const;

    /// getCopyToRegs - Emit a series of CopyToReg nodes that copies the
    /// specified value into the registers specified by this object.  This uses
    /// Chain/Flag as the input and updates them for the output Chain/Flag.
    /// If the Flag pointer is NULL, no flag is used.
    void
    getCopyToRegs(SDValue Val, SelectionDAG &DAG, SDLoc dl, SDValue &Chain,
                  SDValue *Flag, const Value *V,
                  ISD::NodeType PreferredExtendType = ISD::ANY_EXTEND) const;

    /// AddInlineAsmOperands - Add this value to the specified inlineasm node
    /// operand list.  This adds the code marker, matching input operand index
    /// (if applicable), and includes the number of values added into it.
    void AddInlineAsmOperands(unsigned Kind,
                              bool HasMatching, unsigned MatchingIdx,
                              SelectionDAG &DAG,
                              std::vector<SDValue> &Ops) const;
  };
}

/// getCopyFromRegs - Emit a series of CopyFromReg nodes that copies from
/// this value and returns the result as a ValueVT value.  This uses
/// Chain/Flag as the input and updates them for the output Chain/Flag.
/// If the Flag pointer is NULL, no flag is used.
SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG,
                                      FunctionLoweringInfo &FuncInfo,
                                      SDLoc dl,
                                      SDValue &Chain, SDValue *Flag,
                                      const Value *V) const {
  // A Value with type {} or [0 x %t] needs no registers.
  if (ValueVTs.empty())
    return SDValue();

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();

  // Assemble the legal parts into the final values.
  SmallVector<SDValue, 4> Values(ValueVTs.size());
  SmallVector<SDValue, 8> Parts;
  for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) {
    // Copy the legal parts from the registers.
    EVT ValueVT = ValueVTs[Value];
    unsigned NumRegs = TLI.getNumRegisters(*DAG.getContext(), ValueVT);
    MVT RegisterVT = RegVTs[Value];

    Parts.resize(NumRegs);
    for (unsigned i = 0; i != NumRegs; ++i) {
      SDValue P;
      if (!Flag) {
        P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT);
      } else {
        P = DAG.getCopyFromReg(Chain, dl, Regs[Part+i], RegisterVT, *Flag);
        *Flag = P.getValue(2);
      }

      Chain = P.getValue(1);
      Parts[i] = P;

      // If the source register was virtual and if we know something about it,
      // add an assert node.
      if (!TargetRegisterInfo::isVirtualRegister(Regs[Part+i]) ||
          !RegisterVT.isInteger() || RegisterVT.isVector())
        continue;

      const FunctionLoweringInfo::LiveOutInfo *LOI =
        FuncInfo.GetLiveOutRegInfo(Regs[Part+i]);
      if (!LOI)
        continue;

      unsigned RegSize = RegisterVT.getSizeInBits();
      unsigned NumSignBits = LOI->NumSignBits;
      unsigned NumZeroBits = LOI->KnownZero.countLeadingOnes();

      if (NumZeroBits == RegSize) {
        // The current value is a zero.
        // Explicitly express that as it would be easier for
        // optimizations to kick in.
        Parts[i] = DAG.getConstant(0, RegisterVT);
        continue;
      }

      // FIXME: We capture more information than the dag can represent.  For
      // now, just use the tightest assertzext/assertsext possible.
      bool isSExt = true;
      EVT FromVT(MVT::Other);
      if (NumSignBits == RegSize)
        isSExt = true, FromVT = MVT::i1;   // ASSERT SEXT 1
      else if (NumZeroBits >= RegSize-1)
        isSExt = false, FromVT = MVT::i1;  // ASSERT ZEXT 1
      else if (NumSignBits > RegSize-8)
        isSExt = true, FromVT = MVT::i8;   // ASSERT SEXT 8
      else if (NumZeroBits >= RegSize-8)
        isSExt = false, FromVT = MVT::i8;  // ASSERT ZEXT 8
      else if (NumSignBits > RegSize-16)
        isSExt = true, FromVT = MVT::i16;  // ASSERT SEXT 16
      else if (NumZeroBits >= RegSize-16)
        isSExt = false, FromVT = MVT::i16; // ASSERT ZEXT 16
      else if (NumSignBits > RegSize-32)
        isSExt = true, FromVT = MVT::i32;  // ASSERT SEXT 32
      else if (NumZeroBits >= RegSize-32)
        isSExt = false, FromVT = MVT::i32; // ASSERT ZEXT 32
      else
        continue;

      // Add an assertion node.
      assert(FromVT != MVT::Other);
      Parts[i] = DAG.getNode(isSExt ? ISD::AssertSext : ISD::AssertZext, dl,
                             RegisterVT, P, DAG.getValueType(FromVT));
    }

    Values[Value] = getCopyFromParts(DAG, dl, Parts.begin(),
                                     NumRegs, RegisterVT, ValueVT, V);
    Part += NumRegs;
    Parts.clear();
  }

  return DAG.getNode(ISD::MERGE_VALUES, dl, DAG.getVTList(ValueVTs), Values);
}

/// getCopyToRegs - Emit a series of CopyToReg nodes that copies the
/// specified value into the registers specified by this object.  This uses
/// Chain/Flag as the input and updates them for the output Chain/Flag.
/// If the Flag pointer is NULL, no flag is used.
void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG, SDLoc dl,
                                 SDValue &Chain, SDValue *Flag, const Value *V,
                                 ISD::NodeType PreferredExtendType) const {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  ISD::NodeType ExtendKind = PreferredExtendType;

  // Get the list of the values's legal parts.
  unsigned NumRegs = Regs.size();
  SmallVector<SDValue, 8> Parts(NumRegs);
  for (unsigned Value = 0, Part = 0, e = ValueVTs.size(); Value != e; ++Value) {
    EVT ValueVT = ValueVTs[Value];
    unsigned NumParts = TLI.getNumRegisters(*DAG.getContext(), ValueVT);
    MVT RegisterVT = RegVTs[Value];

    if (ExtendKind == ISD::ANY_EXTEND && TLI.isZExtFree(Val, RegisterVT))
      ExtendKind = ISD::ZERO_EXTEND;

    getCopyToParts(DAG, dl, Val.getValue(Val.getResNo() + Value),
                   &Parts[Part], NumParts, RegisterVT, V, ExtendKind);
    Part += NumParts;
  }

  // Copy the parts into the registers.
  SmallVector<SDValue, 8> Chains(NumRegs);
  for (unsigned i = 0; i != NumRegs; ++i) {
    SDValue Part;
    if (!Flag) {
      Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i]);
    } else {
      Part = DAG.getCopyToReg(Chain, dl, Regs[i], Parts[i], *Flag);
      *Flag = Part.getValue(1);
    }

    Chains[i] = Part.getValue(0);
  }

  if (NumRegs == 1 || Flag)
    // If NumRegs > 1 && Flag is used then the use of the last CopyToReg is
    // flagged to it. That is the CopyToReg nodes and the user are considered
    // a single scheduling unit. If we create a TokenFactor and return it as
    // chain, then the TokenFactor is both a predecessor (operand) of the
    // user as well as a successor (the TF operands are flagged to the user).
    // c1, f1 = CopyToReg
    // c2, f2 = CopyToReg
    // c3     = TokenFactor c1, c2
    // ...
    //        = op c3, ..., f2
    Chain = Chains[NumRegs-1];
  else
    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Chains);
}

/// AddInlineAsmOperands - Add this value to the specified inlineasm node
/// operand list.  This adds the code marker and includes the number of
/// values added into it.
void RegsForValue::AddInlineAsmOperands(unsigned Code, bool HasMatching,
                                        unsigned MatchingIdx,
                                        SelectionDAG &DAG,
                                        std::vector<SDValue> &Ops) const {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();

  unsigned Flag = InlineAsm::getFlagWord(Code, Regs.size());
  if (HasMatching)
    Flag = InlineAsm::getFlagWordForMatchingOp(Flag, MatchingIdx);
  else if (!Regs.empty() &&
           TargetRegisterInfo::isVirtualRegister(Regs.front())) {
    // Put the register class of the virtual registers in the flag word.  That
    // way, later passes can recompute register class constraints for inline
    // assembly as well as normal instructions.
    // Don't do this for tied operands that can use the regclass information
    // from the def.
    const MachineRegisterInfo &MRI = DAG.getMachineFunction().getRegInfo();
    const TargetRegisterClass *RC = MRI.getRegClass(Regs.front());
    Flag = InlineAsm::getFlagWordForRegClass(Flag, RC->getID());
  }

  SDValue Res = DAG.getTargetConstant(Flag, MVT::i32);
  Ops.push_back(Res);

  unsigned SP = TLI.getStackPointerRegisterToSaveRestore();
  for (unsigned Value = 0, Reg = 0, e = ValueVTs.size(); Value != e; ++Value) {
    unsigned NumRegs = TLI.getNumRegisters(*DAG.getContext(), ValueVTs[Value]);
    MVT RegisterVT = RegVTs[Value];
    for (unsigned i = 0; i != NumRegs; ++i) {
      assert(Reg < Regs.size() && "Mismatch in # registers expected");
      unsigned TheReg = Regs[Reg++];
      Ops.push_back(DAG.getRegister(TheReg, RegisterVT));

      if (TheReg == SP && Code == InlineAsm::Kind_Clobber) {
        // If we clobbered the stack pointer, MFI should know about it.
        assert(DAG.getMachineFunction().getFrameInfo()->
            hasInlineAsmWithSPAdjust());
      }
    }
  }
}

void SelectionDAGBuilder::init(GCFunctionInfo *gfi, AliasAnalysis &aa,
                               const TargetLibraryInfo *li) {
  AA = &aa;
  GFI = gfi;
  LibInfo = li;
  DL = DAG.getTarget().getDataLayout();
  Context = DAG.getContext();
  LPadToCallSiteMap.clear();
}

/// clear - Clear out the current SelectionDAG and the associated
/// state and prepare this SelectionDAGBuilder object to be used
/// for a new block. This doesn't clear out information about
/// additional blocks that are needed to complete switch lowering
/// or PHI node updating; that information is cleared out as it is
/// consumed.
void SelectionDAGBuilder::clear() {
  NodeMap.clear();
  UnusedArgNodeMap.clear();
  PendingLoads.clear();
  PendingExports.clear();
  CurInst = nullptr;
  HasTailCall = false;
  SDNodeOrder = LowestSDNodeOrder;
  StatepointLowering.clear();
}

/// clearDanglingDebugInfo - Clear the dangling debug information
/// map. This function is separated from the clear so that debug
/// information that is dangling in a basic block can be properly
/// resolved in a different basic block. This allows the
/// SelectionDAG to resolve dangling debug information attached
/// to PHI nodes.
void SelectionDAGBuilder::clearDanglingDebugInfo() {
  DanglingDebugInfoMap.clear();
}

/// getRoot - Return the current virtual root of the Selection DAG,
/// flushing any PendingLoad items. This must be done before emitting
/// a store or any other node that may need to be ordered after any
/// prior load instructions.
///
SDValue SelectionDAGBuilder::getRoot() {
  if (PendingLoads.empty())
    return DAG.getRoot();

  if (PendingLoads.size() == 1) {
    SDValue Root = PendingLoads[0];
    DAG.setRoot(Root);
    PendingLoads.clear();
    return Root;
  }

  // Otherwise, we have to make a token factor node.
  SDValue Root = DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other,
                             PendingLoads);
  PendingLoads.clear();
  DAG.setRoot(Root);
  return Root;
}

/// getControlRoot - Similar to getRoot, but instead of flushing all the
/// PendingLoad items, flush all the PendingExports items. It is necessary
/// to do this before emitting a terminator instruction.
///
SDValue SelectionDAGBuilder::getControlRoot() {
  SDValue Root = DAG.getRoot();

  if (PendingExports.empty())
    return Root;

  // Turn all of the CopyToReg chains into one factored node.
  if (Root.getOpcode() != ISD::EntryToken) {
    unsigned i = 0, e = PendingExports.size();
    for (; i != e; ++i) {
      assert(PendingExports[i].getNode()->getNumOperands() > 1);
      if (PendingExports[i].getNode()->getOperand(0) == Root)
        break;  // Don't add the root if we already indirectly depend on it.
    }

    if (i == e)
      PendingExports.push_back(Root);
  }

  Root = DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other,
                     PendingExports);
  PendingExports.clear();
  DAG.setRoot(Root);
  return Root;
}

void SelectionDAGBuilder::visit(const Instruction &I) {
  // Set up outgoing PHI node register values before emitting the terminator.
  if (isa<TerminatorInst>(&I))
    HandlePHINodesInSuccessorBlocks(I.getParent());

  ++SDNodeOrder;

  CurInst = &I;

  visit(I.getOpcode(), I);

  if (!isa<TerminatorInst>(&I) && !HasTailCall)
    CopyToExportRegsIfNeeded(&I);

  CurInst = nullptr;
}

void SelectionDAGBuilder::visitPHI(const PHINode &) {
  llvm_unreachable("SelectionDAGBuilder shouldn't visit PHI nodes!");
}

void SelectionDAGBuilder::visit(unsigned Opcode, const User &I) {
  // Note: this doesn't use InstVisitor, because it has to work with
  // ConstantExpr's in addition to instructions.
  switch (Opcode) {
  default: llvm_unreachable("Unknown instruction type encountered!");
    // Build the switch statement using the Instruction.def file.
#define HANDLE_INST(NUM, OPCODE, CLASS) \
    case Instruction::OPCODE: visit##OPCODE((const CLASS&)I); break;
#include "llvm/IR/Instruction.def"
  }
}

// resolveDanglingDebugInfo - if we saw an earlier dbg_value referring to V,
// generate the debug data structures now that we've seen its definition.
void SelectionDAGBuilder::resolveDanglingDebugInfo(const Value *V,
                                                   SDValue Val) {
  DanglingDebugInfo &DDI = DanglingDebugInfoMap[V];
  if (DDI.getDI()) {
    const DbgValueInst *DI = DDI.getDI();
    DebugLoc dl = DDI.getdl();
    unsigned DbgSDNodeOrder = DDI.getSDNodeOrder();
    MDLocalVariable *Variable = DI->getVariable();
    MDExpression *Expr = DI->getExpression();
    assert(Variable->isValidLocationForIntrinsic(dl) &&
           "Expected inlined-at fields to agree");
    uint64_t Offset = DI->getOffset();
    // A dbg.value for an alloca is always indirect.
    bool IsIndirect = isa<AllocaInst>(V) || Offset != 0;
    SDDbgValue *SDV;
    if (Val.getNode()) {
      if (!EmitFuncArgumentDbgValue(V, Variable, Expr, dl, Offset, IsIndirect,
                                    Val)) {
        SDV = DAG.getDbgValue(Variable, Expr, Val.getNode(), Val.getResNo(),
                              IsIndirect, Offset, dl, DbgSDNodeOrder);
        DAG.AddDbgValue(SDV, Val.getNode(), false);
      }
    } else
      DEBUG(dbgs() << "Dropping debug info for " << *DI << "\n");
    DanglingDebugInfoMap[V] = DanglingDebugInfo();
  }
}

/// getCopyFromRegs - If there was virtual register allocated for the value V
/// emit CopyFromReg of the specified type Ty. Return empty SDValue() otherwise.
SDValue SelectionDAGBuilder::getCopyFromRegs(const Value *V, Type *Ty) {
  DenseMap<const Value *, unsigned>::iterator It = FuncInfo.ValueMap.find(V);
  SDValue res;

  if (It != FuncInfo.ValueMap.end()) {
    unsigned InReg = It->second;
    RegsForValue RFV(*DAG.getContext(), DAG.getTargetLoweringInfo(), InReg,
                     Ty);
    SDValue Chain = DAG.getEntryNode();
    res = RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V);
    resolveDanglingDebugInfo(V, res);
  }

  return res;
}

/// getValue - Return an SDValue for the given Value.
SDValue SelectionDAGBuilder::getValue(const Value *V) {
  // If we already have an SDValue for this value, use it. It's important
  // to do this first, so that we don't create a CopyFromReg if we already
  // have a regular SDValue.
  SDValue &N = NodeMap[V];
  if (N.getNode()) return N;

  // If there's a virtual register allocated and initialized for this
  // value, use it.
  SDValue copyFromReg = getCopyFromRegs(V, V->getType());
  if (copyFromReg.getNode()) {
    return copyFromReg;
  }

  // Otherwise create a new SDValue and remember it.
  SDValue Val = getValueImpl(V);
  NodeMap[V] = Val;
  resolveDanglingDebugInfo(V, Val);
  return Val;
}

/// getNonRegisterValue - Return an SDValue for the given Value, but
/// don't look in FuncInfo.ValueMap for a virtual register.
SDValue SelectionDAGBuilder::getNonRegisterValue(const Value *V) {
  // If we already have an SDValue for this value, use it.
  SDValue &N = NodeMap[V];
  if (N.getNode()) return N;

  // Otherwise create a new SDValue and remember it.
  SDValue Val = getValueImpl(V);
  NodeMap[V] = Val;
  resolveDanglingDebugInfo(V, Val);
  return Val;
}

/// getValueImpl - Helper function for getValue and getNonRegisterValue.
/// Create an SDValue for the given value.
SDValue SelectionDAGBuilder::getValueImpl(const Value *V) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();

  if (const Constant *C = dyn_cast<Constant>(V)) {
    EVT VT = TLI.getValueType(V->getType(), true);

    if (const ConstantInt *CI = dyn_cast<ConstantInt>(C))
      return DAG.getConstant(*CI, VT);

    if (const GlobalValue *GV = dyn_cast<GlobalValue>(C))
      return DAG.getGlobalAddress(GV, getCurSDLoc(), VT);

    if (isa<ConstantPointerNull>(C)) {
      unsigned AS = V->getType()->getPointerAddressSpace();
      return DAG.getConstant(0, TLI.getPointerTy(AS));
    }

    if (const ConstantFP *CFP = dyn_cast<ConstantFP>(C))
      return DAG.getConstantFP(*CFP, VT);

    if (isa<UndefValue>(C) && !V->getType()->isAggregateType())
      return DAG.getUNDEF(VT);

    if (const ConstantExpr *CE = dyn_cast<ConstantExpr>(C)) {
      visit(CE->getOpcode(), *CE);
      SDValue N1 = NodeMap[V];
      assert(N1.getNode() && "visit didn't populate the NodeMap!");
      return N1;
    }

    if (isa<ConstantStruct>(C) || isa<ConstantArray>(C)) {
      SmallVector<SDValue, 4> Constants;
      for (User::const_op_iterator OI = C->op_begin(), OE = C->op_end();
           OI != OE; ++OI) {
        SDNode *Val = getValue(*OI).getNode();
        // If the operand is an empty aggregate, there are no values.
        if (!Val) continue;
        // Add each leaf value from the operand to the Constants list
        // to form a flattened list of all the values.
        for (unsigned i = 0, e = Val->getNumValues(); i != e; ++i)
          Constants.push_back(SDValue(Val, i));
      }

      return DAG.getMergeValues(Constants, getCurSDLoc());
    }

    if (const ConstantDataSequential *CDS =
          dyn_cast<ConstantDataSequential>(C)) {
      SmallVector<SDValue, 4> Ops;
      for (unsigned i = 0, e = CDS->getNumElements(); i != e; ++i) {
        SDNode *Val = getValue(CDS->getElementAsConstant(i)).getNode();
        // Add each leaf value from the operand to the Constants list
        // to form a flattened list of all the values.
        for (unsigned i = 0, e = Val->getNumValues(); i != e; ++i)
          Ops.push_back(SDValue(Val, i));
      }

      if (isa<ArrayType>(CDS->getType()))
        return DAG.getMergeValues(Ops, getCurSDLoc());
      return NodeMap[V] = DAG.getNode(ISD::BUILD_VECTOR, getCurSDLoc(),
                                      VT, Ops);
    }

    if (C->getType()->isStructTy() || C->getType()->isArrayTy()) {
      assert((isa<ConstantAggregateZero>(C) || isa<UndefValue>(C)) &&
             "Unknown struct or array constant!");

      SmallVector<EVT, 4> ValueVTs;
      ComputeValueVTs(TLI, C->getType(), ValueVTs);
      unsigned NumElts = ValueVTs.size();
      if (NumElts == 0)
        return SDValue(); // empty struct
      SmallVector<SDValue, 4> Constants(NumElts);
      for (unsigned i = 0; i != NumElts; ++i) {
        EVT EltVT = ValueVTs[i];
        if (isa<UndefValue>(C))
          Constants[i] = DAG.getUNDEF(EltVT);
        else if (EltVT.isFloatingPoint())
          Constants[i] = DAG.getConstantFP(0, EltVT);
        else
          Constants[i] = DAG.getConstant(0, EltVT);
      }

      return DAG.getMergeValues(Constants, getCurSDLoc());
    }

    if (const BlockAddress *BA = dyn_cast<BlockAddress>(C))
      return DAG.getBlockAddress(BA, VT);

    VectorType *VecTy = cast<VectorType>(V->getType());
    unsigned NumElements = VecTy->getNumElements();

    // Now that we know the number and type of the elements, get that number of
    // elements into the Ops array based on what kind of constant it is.
    SmallVector<SDValue, 16> Ops;
    if (const ConstantVector *CV = dyn_cast<ConstantVector>(C)) {
      for (unsigned i = 0; i != NumElements; ++i)
        Ops.push_back(getValue(CV->getOperand(i)));
    } else {
      assert(isa<ConstantAggregateZero>(C) && "Unknown vector constant!");
      EVT EltVT = TLI.getValueType(VecTy->getElementType());

      SDValue Op;
      if (EltVT.isFloatingPoint())
        Op = DAG.getConstantFP(0, EltVT);
      else
        Op = DAG.getConstant(0, EltVT);
      Ops.assign(NumElements, Op);
    }

    // Create a BUILD_VECTOR node.
    return NodeMap[V] = DAG.getNode(ISD::BUILD_VECTOR, getCurSDLoc(), VT, Ops);
  }

  // If this is a static alloca, generate it as the frameindex instead of
  // computation.
  if (const AllocaInst *AI = dyn_cast<AllocaInst>(V)) {
    DenseMap<const AllocaInst*, int>::iterator SI =
      FuncInfo.StaticAllocaMap.find(AI);
    if (SI != FuncInfo.StaticAllocaMap.end())
      return DAG.getFrameIndex(SI->second, TLI.getPointerTy());
  }

  // If this is an instruction which fast-isel has deferred, select it now.
  if (const Instruction *Inst = dyn_cast<Instruction>(V)) {
    unsigned InReg = FuncInfo.InitializeRegForValue(Inst);
    RegsForValue RFV(*DAG.getContext(), TLI, InReg, Inst->getType());
    SDValue Chain = DAG.getEntryNode();
    return RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V);
  }

  llvm_unreachable("Can't get register for value!");
}

void SelectionDAGBuilder::visitRet(const ReturnInst &I) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue Chain = getControlRoot();
  SmallVector<ISD::OutputArg, 8> Outs;
  SmallVector<SDValue, 8> OutVals;

  if (!FuncInfo.CanLowerReturn) {
    unsigned DemoteReg = FuncInfo.DemoteRegister;
    const Function *F = I.getParent()->getParent();

    // Emit a store of the return value through the virtual register.
    // Leave Outs empty so that LowerReturn won't try to load return
    // registers the usual way.
    SmallVector<EVT, 1> PtrValueVTs;
    ComputeValueVTs(TLI, PointerType::getUnqual(F->getReturnType()),
                    PtrValueVTs);

    SDValue RetPtr = DAG.getRegister(DemoteReg, PtrValueVTs[0]);
    SDValue RetOp = getValue(I.getOperand(0));

    SmallVector<EVT, 4> ValueVTs;
    SmallVector<uint64_t, 4> Offsets;
    ComputeValueVTs(TLI, I.getOperand(0)->getType(), ValueVTs, &Offsets);
    unsigned NumValues = ValueVTs.size();

    SmallVector<SDValue, 4> Chains(NumValues);
    for (unsigned i = 0; i != NumValues; ++i) {
      SDValue Add = DAG.getNode(ISD::ADD, getCurSDLoc(),
                                RetPtr.getValueType(), RetPtr,
                                DAG.getIntPtrConstant(Offsets[i]));
      Chains[i] =
        DAG.getStore(Chain, getCurSDLoc(),
                     SDValue(RetOp.getNode(), RetOp.getResNo() + i),
                     // FIXME: better loc info would be nice.
                     Add, MachinePointerInfo(), false, false, 0);
    }

    Chain = DAG.getNode(ISD::TokenFactor, getCurSDLoc(),
                        MVT::Other, Chains);
  } else if (I.getNumOperands() != 0) {
    SmallVector<EVT, 4> ValueVTs;
    ComputeValueVTs(TLI, I.getOperand(0)->getType(), ValueVTs);
    unsigned NumValues = ValueVTs.size();
    if (NumValues) {
      SDValue RetOp = getValue(I.getOperand(0));

      const Function *F = I.getParent()->getParent();

      ISD::NodeType ExtendKind = ISD::ANY_EXTEND;
      if (F->getAttributes().hasAttribute(AttributeSet::ReturnIndex,
                                          Attribute::SExt))
        ExtendKind = ISD::SIGN_EXTEND;
      else if (F->getAttributes().hasAttribute(AttributeSet::ReturnIndex,
                                               Attribute::ZExt))
        ExtendKind = ISD::ZERO_EXTEND;

      LLVMContext &Context = F->getContext();
      bool RetInReg = F->getAttributes().hasAttribute(AttributeSet::ReturnIndex,
                                                      Attribute::InReg);

      for (unsigned j = 0; j != NumValues; ++j) {
        EVT VT = ValueVTs[j];

        if (ExtendKind != ISD::ANY_EXTEND && VT.isInteger())
          VT = TLI.getTypeForExtArgOrReturn(Context, VT, ExtendKind);

        unsigned NumParts = TLI.getNumRegisters(Context, VT);
        MVT PartVT = TLI.getRegisterType(Context, VT);
        SmallVector<SDValue, 4> Parts(NumParts);
        getCopyToParts(DAG, getCurSDLoc(),
                       SDValue(RetOp.getNode(), RetOp.getResNo() + j),
                       &Parts[0], NumParts, PartVT, &I, ExtendKind);

        // 'inreg' on function refers to return value
        ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
        if (RetInReg)
          Flags.setInReg();

        // Propagate extension type if any
        if (ExtendKind == ISD::SIGN_EXTEND)
          Flags.setSExt();
        else if (ExtendKind == ISD::ZERO_EXTEND)
          Flags.setZExt();

        for (unsigned i = 0; i < NumParts; ++i) {
          Outs.push_back(ISD::OutputArg(Flags, Parts[i].getValueType(),
                                        VT, /*isfixed=*/true, 0, 0));
          OutVals.push_back(Parts[i]);
        }
      }
    }
  }

  bool isVarArg = DAG.getMachineFunction().getFunction()->isVarArg();
  CallingConv::ID CallConv =
    DAG.getMachineFunction().getFunction()->getCallingConv();
  Chain = DAG.getTargetLoweringInfo().LowerReturn(
      Chain, CallConv, isVarArg, Outs, OutVals, getCurSDLoc(), DAG);

  // Verify that the target's LowerReturn behaved as expected.
  assert(Chain.getNode() && Chain.getValueType() == MVT::Other &&
         "LowerReturn didn't return a valid chain!");

  // Update the DAG with the new chain value resulting from return lowering.
  DAG.setRoot(Chain);
}

/// CopyToExportRegsIfNeeded - If the given value has virtual registers
/// created for it, emit nodes to copy the value into the virtual
/// registers.
void SelectionDAGBuilder::CopyToExportRegsIfNeeded(const Value *V) {
  // Skip empty types
  if (V->getType()->isEmptyTy())
    return;

  DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
  if (VMI != FuncInfo.ValueMap.end()) {
    assert(!V->use_empty() && "Unused value assigned virtual registers!");
    CopyValueToVirtualRegister(V, VMI->second);
  }
}

/// ExportFromCurrentBlock - If this condition isn't known to be exported from
/// the current basic block, add it to ValueMap now so that we'll get a
/// CopyTo/FromReg.
void SelectionDAGBuilder::ExportFromCurrentBlock(const Value *V) {
  // No need to export constants.
  if (!isa<Instruction>(V) && !isa<Argument>(V)) return;

  // Already exported?
  if (FuncInfo.isExportedInst(V)) return;

  unsigned Reg = FuncInfo.InitializeRegForValue(V);
  CopyValueToVirtualRegister(V, Reg);
}

bool SelectionDAGBuilder::isExportableFromCurrentBlock(const Value *V,
                                                     const BasicBlock *FromBB) {
  // The operands of the setcc have to be in this block.  We don't know
  // how to export them from some other block.
  if (const Instruction *VI = dyn_cast<Instruction>(V)) {
    // Can export from current BB.
    if (VI->getParent() == FromBB)
      return true;

    // Is already exported, noop.
    return FuncInfo.isExportedInst(V);
  }

  // If this is an argument, we can export it if the BB is the entry block or
  // if it is already exported.
  if (isa<Argument>(V)) {
    if (FromBB == &FromBB->getParent()->getEntryBlock())
      return true;

    // Otherwise, can only export this if it is already exported.
    return FuncInfo.isExportedInst(V);
  }

  // Otherwise, constants can always be exported.
  return true;
}

/// Return branch probability calculated by BranchProbabilityInfo for IR blocks.
uint32_t SelectionDAGBuilder::getEdgeWeight(const MachineBasicBlock *Src,
                                            const MachineBasicBlock *Dst) const {
  BranchProbabilityInfo *BPI = FuncInfo.BPI;
  if (!BPI)
    return 0;
  const BasicBlock *SrcBB = Src->getBasicBlock();
  const BasicBlock *DstBB = Dst->getBasicBlock();
  return BPI->getEdgeWeight(SrcBB, DstBB);
}

void SelectionDAGBuilder::
addSuccessorWithWeight(MachineBasicBlock *Src, MachineBasicBlock *Dst,
                       uint32_t Weight /* = 0 */) {
  if (!Weight)
    Weight = getEdgeWeight(Src, Dst);
  Src->addSuccessor(Dst, Weight);
}


static bool InBlock(const Value *V, const BasicBlock *BB) {
  if (const Instruction *I = dyn_cast<Instruction>(V))
    return I->getParent() == BB;
  return true;
}

/// EmitBranchForMergedCondition - Helper method for FindMergedConditions.
/// This function emits a branch and is used at the leaves of an OR or an
/// AND operator tree.
///
void
SelectionDAGBuilder::EmitBranchForMergedCondition(const Value *Cond,
                                                  MachineBasicBlock *TBB,
                                                  MachineBasicBlock *FBB,
                                                  MachineBasicBlock *CurBB,
                                                  MachineBasicBlock *SwitchBB,
                                                  uint32_t TWeight,
                                                  uint32_t FWeight) {
  const BasicBlock *BB = CurBB->getBasicBlock();

  // If the leaf of the tree is a comparison, merge the condition into
  // the caseblock.
  if (const CmpInst *BOp = dyn_cast<CmpInst>(Cond)) {
    // The operands of the cmp have to be in this block.  We don't know
    // how to export them from some other block.  If this is the first block
    // of the sequence, no exporting is needed.
    if (CurBB == SwitchBB ||
        (isExportableFromCurrentBlock(BOp->getOperand(0), BB) &&
         isExportableFromCurrentBlock(BOp->getOperand(1), BB))) {
      ISD::CondCode Condition;
      if (const ICmpInst *IC = dyn_cast<ICmpInst>(Cond)) {
        Condition = getICmpCondCode(IC->getPredicate());
      } else if (const FCmpInst *FC = dyn_cast<FCmpInst>(Cond)) {
        Condition = getFCmpCondCode(FC->getPredicate());
        if (TM.Options.NoNaNsFPMath)
          Condition = getFCmpCodeWithoutNaN(Condition);
      } else {
        (void)Condition; // silence warning.
        llvm_unreachable("Unknown compare instruction");
      }

      CaseBlock CB(Condition, BOp->getOperand(0), BOp->getOperand(1), nullptr,
                   TBB, FBB, CurBB, TWeight, FWeight);
      SwitchCases.push_back(CB);
      return;
    }
  }

  // Create a CaseBlock record representing this branch.
  CaseBlock CB(ISD::SETEQ, Cond, ConstantInt::getTrue(*DAG.getContext()),
               nullptr, TBB, FBB, CurBB, TWeight, FWeight);
  SwitchCases.push_back(CB);
}

/// Scale down both weights to fit into uint32_t.
static void ScaleWeights(uint64_t &NewTrue, uint64_t &NewFalse) {
  uint64_t NewMax = (NewTrue > NewFalse) ? NewTrue : NewFalse;
  uint32_t Scale = (NewMax / UINT32_MAX) + 1;
  NewTrue = NewTrue / Scale;
  NewFalse = NewFalse / Scale;
}

/// FindMergedConditions - If Cond is an expression like
void SelectionDAGBuilder::FindMergedConditions(const Value *Cond,
                                               MachineBasicBlock *TBB,
                                               MachineBasicBlock *FBB,
                                               MachineBasicBlock *CurBB,
                                               MachineBasicBlock *SwitchBB,
                                               unsigned Opc, uint32_t TWeight,
                                               uint32_t FWeight) {
  // If this node is not part of the or/and tree, emit it as a branch.
  const Instruction *BOp = dyn_cast<Instruction>(Cond);
  if (!BOp || !(isa<BinaryOperator>(BOp) || isa<CmpInst>(BOp)) ||
      (unsigned)BOp->getOpcode() != Opc || !BOp->hasOneUse() ||
      BOp->getParent() != CurBB->getBasicBlock() ||
      !InBlock(BOp->getOperand(0), CurBB->getBasicBlock()) ||
      !InBlock(BOp->getOperand(1), CurBB->getBasicBlock())) {
    EmitBranchForMergedCondition(Cond, TBB, FBB, CurBB, SwitchBB,
                                 TWeight, FWeight);
    return;
  }

  //  Create TmpBB after CurBB.
  MachineFunction::iterator BBI = CurBB;
  MachineFunction &MF = DAG.getMachineFunction();
  MachineBasicBlock *TmpBB = MF.CreateMachineBasicBlock(CurBB->getBasicBlock());
  CurBB->getParent()->insert(++BBI, TmpBB);

  if (Opc == Instruction::Or) {
    // Codegen X | Y as:
    // BB1:
    //   jmp_if_X TBB
    //   jmp TmpBB
    // TmpBB:
    //   jmp_if_Y TBB
    //   jmp FBB
    //

    // We have flexibility in setting Prob for BB1 and Prob for TmpBB.
    // The requirement is that
    //   TrueProb for BB1 + (FalseProb for BB1 * TrueProb for TmpBB)
    //     = TrueProb for orignal BB.
    // Assuming the orignal weights are A and B, one choice is to set BB1's
    // weights to A and A+2B, and set TmpBB's weights to A and 2B. This choice
    // assumes that
    //   TrueProb for BB1 == FalseProb for BB1 * TrueProb for TmpBB.
    // Another choice is to assume TrueProb for BB1 equals to TrueProb for
    // TmpBB, but the math is more complicated.

    uint64_t NewTrueWeight = TWeight;
    uint64_t NewFalseWeight = (uint64_t)TWeight + 2 * (uint64_t)FWeight;
    ScaleWeights(NewTrueWeight, NewFalseWeight);
    // Emit the LHS condition.
    FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, SwitchBB, Opc,
                         NewTrueWeight, NewFalseWeight);

    NewTrueWeight = TWeight;
    NewFalseWeight = 2 * (uint64_t)FWeight;
    ScaleWeights(NewTrueWeight, NewFalseWeight);
    // Emit the RHS condition into TmpBB.
    FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc,
                         NewTrueWeight, NewFalseWeight);
  } else {
    assert(Opc == Instruction::And && "Unknown merge op!");
    // Codegen X & Y as:
    // BB1:
    //   jmp_if_X TmpBB
    //   jmp FBB
    // TmpBB:
    //   jmp_if_Y TBB
    //   jmp FBB
    //
    //  This requires creation of TmpBB after CurBB.

    // We have flexibility in setting Prob for BB1 and Prob for TmpBB.
    // The requirement is that
    //   FalseProb for BB1 + (TrueProb for BB1 * FalseProb for TmpBB)
    //     = FalseProb for orignal BB.
    // Assuming the orignal weights are A and B, one choice is to set BB1's
    // weights to 2A+B and B, and set TmpBB's weights to 2A and B. This choice
    // assumes that
    //   FalseProb for BB1 == TrueProb for BB1 * FalseProb for TmpBB.

    uint64_t NewTrueWeight = 2 * (uint64_t)TWeight + (uint64_t)FWeight;
    uint64_t NewFalseWeight = FWeight;
    ScaleWeights(NewTrueWeight, NewFalseWeight);
    // Emit the LHS condition.
    FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, SwitchBB, Opc,
                         NewTrueWeight, NewFalseWeight);

    NewTrueWeight = 2 * (uint64_t)TWeight;
    NewFalseWeight = FWeight;
    ScaleWeights(NewTrueWeight, NewFalseWeight);
    // Emit the RHS condition into TmpBB.
    FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc,
                         NewTrueWeight, NewFalseWeight);
  }
}

/// If the set of cases should be emitted as a series of branches, return true.
/// If we should emit this as a bunch of and/or'd together conditions, return
/// false.
bool
SelectionDAGBuilder::ShouldEmitAsBranches(const std::vector<CaseBlock> &Cases) {
  if (Cases.size() != 2) return true;

  // If this is two comparisons of the same values or'd or and'd together, they
  // will get folded into a single comparison, so don't emit two blocks.
  if ((Cases[0].CmpLHS == Cases[1].CmpLHS &&
       Cases[0].CmpRHS == Cases[1].CmpRHS) ||
      (Cases[0].CmpRHS == Cases[1].CmpLHS &&
       Cases[0].CmpLHS == Cases[1].CmpRHS)) {
    return false;
  }

  // Handle: (X != null) | (Y != null) --> (X|Y) != 0
  // Handle: (X == null) & (Y == null) --> (X|Y) == 0
  if (Cases[0].CmpRHS == Cases[1].CmpRHS &&
      Cases[0].CC == Cases[1].CC &&
      isa<Constant>(Cases[0].CmpRHS) &&
      cast<Constant>(Cases[0].CmpRHS)->isNullValue()) {
    if (Cases[0].CC == ISD::SETEQ && Cases[0].TrueBB == Cases[1].ThisBB)
      return false;
    if (Cases[0].CC == ISD::SETNE && Cases[0].FalseBB == Cases[1].ThisBB)
      return false;
  }

  return true;
}

void SelectionDAGBuilder::visitBr(const BranchInst &I) {
  MachineBasicBlock *BrMBB = FuncInfo.MBB;

  // Update machine-CFG edges.
  MachineBasicBlock *Succ0MBB = FuncInfo.MBBMap[I.getSuccessor(0)];

  if (I.isUnconditional()) {
    // Update machine-CFG edges.
    BrMBB->addSuccessor(Succ0MBB);

    // If this is not a fall-through branch or optimizations are switched off,
    // emit the branch.
    if (Succ0MBB != NextBlock(BrMBB) || TM.getOptLevel() == CodeGenOpt::None)
      DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(),
                              MVT::Other, getControlRoot(),
                              DAG.getBasicBlock(Succ0MBB)));

    return;
  }

  // If this condition is one of the special cases we handle, do special stuff
  // now.
  const Value *CondVal = I.getCondition();
  MachineBasicBlock *Succ1MBB = FuncInfo.MBBMap[I.getSuccessor(1)];

  // If this is a series of conditions that are or'd or and'd together, emit
  // this as a sequence of branches instead of setcc's with and/or operations.
  // As long as jumps are not expensive, this should improve performance.
  // For example, instead of something like:
  //     cmp A, B
  //     C = seteq
  //     cmp D, E
  //     F = setle
  //     or C, F
  //     jnz foo
  // Emit:
  //     cmp A, B
  //     je foo
  //     cmp D, E
  //     jle foo
  //
  if (const BinaryOperator *BOp = dyn_cast<BinaryOperator>(CondVal)) {
    if (!DAG.getTargetLoweringInfo().isJumpExpensive() &&
        BOp->hasOneUse() && (BOp->getOpcode() == Instruction::And ||
                             BOp->getOpcode() == Instruction::Or)) {
      FindMergedConditions(BOp, Succ0MBB, Succ1MBB, BrMBB, BrMBB,
                           BOp->getOpcode(), getEdgeWeight(BrMBB, Succ0MBB),
                           getEdgeWeight(BrMBB, Succ1MBB));
      // If the compares in later blocks need to use values not currently
      // exported from this block, export them now.  This block should always
      // be the first entry.
      assert(SwitchCases[0].ThisBB == BrMBB && "Unexpected lowering!");

      // Allow some cases to be rejected.
      if (ShouldEmitAsBranches(SwitchCases)) {
        for (unsigned i = 1, e = SwitchCases.size(); i != e; ++i) {
          ExportFromCurrentBlock(SwitchCases[i].CmpLHS);
          ExportFromCurrentBlock(SwitchCases[i].CmpRHS);
        }

        // Emit the branch for this block.
        visitSwitchCase(SwitchCases[0], BrMBB);
        SwitchCases.erase(SwitchCases.begin());
        return;
      }

      // Okay, we decided not to do this, remove any inserted MBB's and clear
      // SwitchCases.
      for (unsigned i = 1, e = SwitchCases.size(); i != e; ++i)
        FuncInfo.MF->erase(SwitchCases[i].ThisBB);

      SwitchCases.clear();
    }
  }

  // Create a CaseBlock record representing this branch.
  CaseBlock CB(ISD::SETEQ, CondVal, ConstantInt::getTrue(*DAG.getContext()),
               nullptr, Succ0MBB, Succ1MBB, BrMBB);

  // Use visitSwitchCase to actually insert the fast branch sequence for this
  // cond branch.
  visitSwitchCase(CB, BrMBB);
}

/// visitSwitchCase - Emits the necessary code to represent a single node in
/// the binary search tree resulting from lowering a switch instruction.
void SelectionDAGBuilder::visitSwitchCase(CaseBlock &CB,
                                          MachineBasicBlock *SwitchBB) {
  SDValue Cond;
  SDValue CondLHS = getValue(CB.CmpLHS);
  SDLoc dl = getCurSDLoc();

  // Build the setcc now.
  if (!CB.CmpMHS) {
    // Fold "(X == true)" to X and "(X == false)" to !X to
    // handle common cases produced by branch lowering.
    if (CB.CmpRHS == ConstantInt::getTrue(*DAG.getContext()) &&
        CB.CC == ISD::SETEQ)
      Cond = CondLHS;
    else if (CB.CmpRHS == ConstantInt::getFalse(*DAG.getContext()) &&
             CB.CC == ISD::SETEQ) {
      SDValue True = DAG.getConstant(1, CondLHS.getValueType());
      Cond = DAG.getNode(ISD::XOR, dl, CondLHS.getValueType(), CondLHS, True);
    } else
      Cond = DAG.getSetCC(dl, MVT::i1, CondLHS, getValue(CB.CmpRHS), CB.CC);
  } else {
    assert(CB.CC == ISD::SETLE && "Can handle only LE ranges now");

    const APInt& Low = cast<ConstantInt>(CB.CmpLHS)->getValue();
    const APInt& High = cast<ConstantInt>(CB.CmpRHS)->getValue();

    SDValue CmpOp = getValue(CB.CmpMHS);
    EVT VT = CmpOp.getValueType();

    if (cast<ConstantInt>(CB.CmpLHS)->isMinValue(true)) {
      Cond = DAG.getSetCC(dl, MVT::i1, CmpOp, DAG.getConstant(High, VT),
                          ISD::SETLE);
    } else {
      SDValue SUB = DAG.getNode(ISD::SUB, dl,
                                VT, CmpOp, DAG.getConstant(Low, VT));
      Cond = DAG.getSetCC(dl, MVT::i1, SUB,
                          DAG.getConstant(High-Low, VT), ISD::SETULE);
    }
  }

  // Update successor info
  addSuccessorWithWeight(SwitchBB, CB.TrueBB, CB.TrueWeight);
  // TrueBB and FalseBB are always different unless the incoming IR is
  // degenerate. This only happens when running llc on weird IR.
  if (CB.TrueBB != CB.FalseBB)
    addSuccessorWithWeight(SwitchBB, CB.FalseBB, CB.FalseWeight);

  // If the lhs block is the next block, invert the condition so that we can
  // fall through to the lhs instead of the rhs block.
  if (CB.TrueBB == NextBlock(SwitchBB)) {
    std::swap(CB.TrueBB, CB.FalseBB);
    SDValue True = DAG.getConstant(1, Cond.getValueType());
    Cond = DAG.getNode(ISD::XOR, dl, Cond.getValueType(), Cond, True);
  }

  SDValue BrCond = DAG.getNode(ISD::BRCOND, dl,
                               MVT::Other, getControlRoot(), Cond,
                               DAG.getBasicBlock(CB.TrueBB));

  // Insert the false branch. Do this even if it's a fall through branch,
  // this makes it easier to do DAG optimizations which require inverting
  // the branch condition.
  BrCond = DAG.getNode(ISD::BR, dl, MVT::Other, BrCond,
                       DAG.getBasicBlock(CB.FalseBB));

  DAG.setRoot(BrCond);
}

/// visitJumpTable - Emit JumpTable node in the current MBB
void SelectionDAGBuilder::visitJumpTable(JumpTable &JT) {
  // Emit the code for the jump table
  assert(JT.Reg != -1U && "Should lower JT Header first!");
  EVT PTy = DAG.getTargetLoweringInfo().getPointerTy();
  SDValue Index = DAG.getCopyFromReg(getControlRoot(), getCurSDLoc(),
                                     JT.Reg, PTy);
  SDValue Table = DAG.getJumpTable(JT.JTI, PTy);
  SDValue BrJumpTable = DAG.getNode(ISD::BR_JT, getCurSDLoc(),
                                    MVT::Other, Index.getValue(1),
                                    Table, Index);
  DAG.setRoot(BrJumpTable);
}

/// visitJumpTableHeader - This function emits necessary code to produce index
/// in the JumpTable from switch case.
void SelectionDAGBuilder::visitJumpTableHeader(JumpTable &JT,
                                               JumpTableHeader &JTH,
                                               MachineBasicBlock *SwitchBB) {
  // Subtract the lowest switch case value from the value being switched on and
  // conditional branch to default mbb if the result is greater than the
  // difference between smallest and largest cases.
  SDValue SwitchOp = getValue(JTH.SValue);
  EVT VT = SwitchOp.getValueType();
  SDValue Sub = DAG.getNode(ISD::SUB, getCurSDLoc(), VT, SwitchOp,
                            DAG.getConstant(JTH.First, VT));

  // The SDNode we just created, which holds the value being switched on minus
  // the smallest case value, needs to be copied to a virtual register so it
  // can be used as an index into the jump table in a subsequent basic block.
  // This value may be smaller or larger than the target's pointer type, and
  // therefore require extension or truncating.
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SwitchOp = DAG.getZExtOrTrunc(Sub, getCurSDLoc(), TLI.getPointerTy());

  unsigned JumpTableReg = FuncInfo.CreateReg(TLI.getPointerTy());
  SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurSDLoc(),
                                    JumpTableReg, SwitchOp);
  JT.Reg = JumpTableReg;

  // Emit the range check for the jump table, and branch to the default block
  // for the switch statement if the value being switched on exceeds the largest
  // case in the switch.
  SDValue CMP =
      DAG.getSetCC(getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(),
                                                         Sub.getValueType()),
                   Sub, DAG.getConstant(JTH.Last - JTH.First, VT), ISD::SETUGT);

  SDValue BrCond = DAG.getNode(ISD::BRCOND, getCurSDLoc(),
                               MVT::Other, CopyTo, CMP,
                               DAG.getBasicBlock(JT.Default));

  // Avoid emitting unnecessary branches to the next block.
  if (JT.MBB != NextBlock(SwitchBB))
    BrCond = DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, BrCond,
                         DAG.getBasicBlock(JT.MBB));

  DAG.setRoot(BrCond);
}

/// Codegen a new tail for a stack protector check ParentMBB which has had its
/// tail spliced into a stack protector check success bb.
///
/// For a high level explanation of how this fits into the stack protector
/// generation see the comment on the declaration of class
/// StackProtectorDescriptor.
void SelectionDAGBuilder::visitSPDescriptorParent(StackProtectorDescriptor &SPD,
                                                  MachineBasicBlock *ParentBB) {

  // First create the loads to the guard/stack slot for the comparison.
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT PtrTy = TLI.getPointerTy();

  MachineFrameInfo *MFI = ParentBB->getParent()->getFrameInfo();
  int FI = MFI->getStackProtectorIndex();

  const Value *IRGuard = SPD.getGuard();
  SDValue GuardPtr = getValue(IRGuard);
  SDValue StackSlotPtr = DAG.getFrameIndex(FI, PtrTy);

  unsigned Align =
    TLI.getDataLayout()->getPrefTypeAlignment(IRGuard->getType());

  SDValue Guard;

  // If GuardReg is set and useLoadStackGuardNode returns true, retrieve the
  // guard value from the virtual register holding the value. Otherwise, emit a
  // volatile load to retrieve the stack guard value.
  unsigned GuardReg = SPD.getGuardReg();

  if (GuardReg && TLI.useLoadStackGuardNode())
    Guard = DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(), GuardReg,
                               PtrTy);
  else
    Guard = DAG.getLoad(PtrTy, getCurSDLoc(), DAG.getEntryNode(),
                        GuardPtr, MachinePointerInfo(IRGuard, 0),
                        true, false, false, Align);

  SDValue StackSlot = DAG.getLoad(PtrTy, getCurSDLoc(), DAG.getEntryNode(),
                                  StackSlotPtr,
                                  MachinePointerInfo::getFixedStack(FI),
                                  true, false, false, Align);

  // Perform the comparison via a subtract/getsetcc.
  EVT VT = Guard.getValueType();
  SDValue Sub = DAG.getNode(ISD::SUB, getCurSDLoc(), VT, Guard, StackSlot);

  SDValue Cmp =
      DAG.getSetCC(getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(),
                                                         Sub.getValueType()),
                   Sub, DAG.getConstant(0, VT), ISD::SETNE);

  // If the sub is not 0, then we know the guard/stackslot do not equal, so
  // branch to failure MBB.
  SDValue BrCond = DAG.getNode(ISD::BRCOND, getCurSDLoc(),
                               MVT::Other, StackSlot.getOperand(0),
                               Cmp, DAG.getBasicBlock(SPD.getFailureMBB()));
  // Otherwise branch to success MBB.
  SDValue Br = DAG.getNode(ISD::BR, getCurSDLoc(),
                           MVT::Other, BrCond,
                           DAG.getBasicBlock(SPD.getSuccessMBB()));

  DAG.setRoot(Br);
}

/// Codegen the failure basic block for a stack protector check.
///
/// A failure stack protector machine basic block consists simply of a call to
/// __stack_chk_fail().
///
/// For a high level explanation of how this fits into the stack protector
/// generation see the comment on the declaration of class
/// StackProtectorDescriptor.
void
SelectionDAGBuilder::visitSPDescriptorFailure(StackProtectorDescriptor &SPD) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue Chain =
      TLI.makeLibCall(DAG, RTLIB::STACKPROTECTOR_CHECK_FAIL, MVT::isVoid,
                      nullptr, 0, false, getCurSDLoc(), false, false).second;
  DAG.setRoot(Chain);
}

/// visitBitTestHeader - This function emits necessary code to produce value
/// suitable for "bit tests"
void SelectionDAGBuilder::visitBitTestHeader(BitTestBlock &B,
                                             MachineBasicBlock *SwitchBB) {
  // Subtract the minimum value
  SDValue SwitchOp = getValue(B.SValue);
  EVT VT = SwitchOp.getValueType();
  SDValue Sub = DAG.getNode(ISD::SUB, getCurSDLoc(), VT, SwitchOp,
                            DAG.getConstant(B.First, VT));

  // Check range
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue RangeCmp =
      DAG.getSetCC(getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(),
                                                         Sub.getValueType()),
                   Sub, DAG.getConstant(B.Range, VT), ISD::SETUGT);

  // Determine the type of the test operands.
  bool UsePtrType = false;
  if (!TLI.isTypeLegal(VT))
    UsePtrType = true;
  else {
    for (unsigned i = 0, e = B.Cases.size(); i != e; ++i)
      if (!isUIntN(VT.getSizeInBits(), B.Cases[i].Mask)) {
        // Switch table case range are encoded into series of masks.
        // Just use pointer type, it's guaranteed to fit.
        UsePtrType = true;
        break;
      }
  }
  if (UsePtrType) {
    VT = TLI.getPointerTy();
    Sub = DAG.getZExtOrTrunc(Sub, getCurSDLoc(), VT);
  }

  B.RegVT = VT.getSimpleVT();
  B.Reg = FuncInfo.CreateReg(B.RegVT);
  SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), getCurSDLoc(),
                                    B.Reg, Sub);

  MachineBasicBlock* MBB = B.Cases[0].ThisBB;

  addSuccessorWithWeight(SwitchBB, B.Default);
  addSuccessorWithWeight(SwitchBB, MBB);

  SDValue BrRange = DAG.getNode(ISD::BRCOND, getCurSDLoc(),
                                MVT::Other, CopyTo, RangeCmp,
                                DAG.getBasicBlock(B.Default));

  // Avoid emitting unnecessary branches to the next block.
  if (MBB != NextBlock(SwitchBB))
    BrRange = DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, CopyTo,
                          DAG.getBasicBlock(MBB));

  DAG.setRoot(BrRange);
}

/// visitBitTestCase - this function produces one "bit test"
void SelectionDAGBuilder::visitBitTestCase(BitTestBlock &BB,
                                           MachineBasicBlock* NextMBB,
                                           uint32_t BranchWeightToNext,
                                           unsigned Reg,
                                           BitTestCase &B,
                                           MachineBasicBlock *SwitchBB) {
  MVT VT = BB.RegVT;
  SDValue ShiftOp = DAG.getCopyFromReg(getControlRoot(), getCurSDLoc(),
                                       Reg, VT);
  SDValue Cmp;
  unsigned PopCount = countPopulation(B.Mask);
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  if (PopCount == 1) {
    // Testing for a single bit; just compare the shift count with what it
    // would need to be to shift a 1 bit in that position.
    Cmp = DAG.getSetCC(
        getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(), VT), ShiftOp,
        DAG.getConstant(countTrailingZeros(B.Mask), VT), ISD::SETEQ);
  } else if (PopCount == BB.Range) {
    // There is only one zero bit in the range, test for it directly.
    Cmp = DAG.getSetCC(
        getCurSDLoc(), TLI.getSetCCResultType(*DAG.getContext(), VT), ShiftOp,
        DAG.getConstant(countTrailingOnes(B.Mask), VT), ISD::SETNE);
  } else {
    // Make desired shift
    SDValue SwitchVal = DAG.getNode(ISD::SHL, getCurSDLoc(), VT,
                                    DAG.getConstant(1, VT), ShiftOp);

    // Emit bit tests and jumps
    SDValue AndOp = DAG.getNode(ISD::AND, getCurSDLoc(),
                                VT, SwitchVal, DAG.getConstant(B.Mask, VT));
    Cmp = DAG.getSetCC(getCurSDLoc(),
                       TLI.getSetCCResultType(*DAG.getContext(), VT), AndOp,
                       DAG.getConstant(0, VT), ISD::SETNE);
  }

  // The branch weight from SwitchBB to B.TargetBB is B.ExtraWeight.
  addSuccessorWithWeight(SwitchBB, B.TargetBB, B.ExtraWeight);
  // The branch weight from SwitchBB to NextMBB is BranchWeightToNext.
  addSuccessorWithWeight(SwitchBB, NextMBB, BranchWeightToNext);

  SDValue BrAnd = DAG.getNode(ISD::BRCOND, getCurSDLoc(),
                              MVT::Other, getControlRoot(),
                              Cmp, DAG.getBasicBlock(B.TargetBB));

  // Avoid emitting unnecessary branches to the next block.
  if (NextMBB != NextBlock(SwitchBB))
    BrAnd = DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, BrAnd,
                        DAG.getBasicBlock(NextMBB));

  DAG.setRoot(BrAnd);
}

void SelectionDAGBuilder::visitInvoke(const InvokeInst &I) {
  MachineBasicBlock *InvokeMBB = FuncInfo.MBB;

  // Retrieve successors.
  MachineBasicBlock *Return = FuncInfo.MBBMap[I.getSuccessor(0)];
  MachineBasicBlock *LandingPad = FuncInfo.MBBMap[I.getSuccessor(1)];

  const Value *Callee(I.getCalledValue());
  const Function *Fn = dyn_cast<Function>(Callee);
  if (isa<InlineAsm>(Callee))
    visitInlineAsm(&I);
  else if (Fn && Fn->isIntrinsic()) {
    switch (Fn->getIntrinsicID()) {
    default:
      llvm_unreachable("Cannot invoke this intrinsic");
    case Intrinsic::donothing:
      // Ignore invokes to @llvm.donothing: jump directly to the next BB.
      break;
    case Intrinsic::experimental_patchpoint_void:
    case Intrinsic::experimental_patchpoint_i64:
      visitPatchpoint(&I, LandingPad);
      break;
    case Intrinsic::experimental_gc_statepoint:
      LowerStatepoint(ImmutableStatepoint(&I), LandingPad);
      break;
    }
  } else
    LowerCallTo(&I, getValue(Callee), false, LandingPad);

  // If the value of the invoke is used outside of its defining block, make it
  // available as a virtual register.
  // We already took care of the exported value for the statepoint instruction
  // during call to the LowerStatepoint.
  if (!isStatepoint(I)) {
    CopyToExportRegsIfNeeded(&I);
  }

  // Update successor info
  addSuccessorWithWeight(InvokeMBB, Return);
  addSuccessorWithWeight(InvokeMBB, LandingPad);

  // Drop into normal successor.
  DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(),
                          MVT::Other, getControlRoot(),
                          DAG.getBasicBlock(Return)));
}

void SelectionDAGBuilder::visitResume(const ResumeInst &RI) {
  llvm_unreachable("SelectionDAGBuilder shouldn't visit resume instructions!");
}

void SelectionDAGBuilder::visitLandingPad(const LandingPadInst &LP) {
  assert(FuncInfo.MBB->isLandingPad() &&
         "Call to landingpad not in landing pad!");

  MachineBasicBlock *MBB = FuncInfo.MBB;
  MachineModuleInfo &MMI = DAG.getMachineFunction().getMMI();
  AddLandingPadInfo(LP, MMI, MBB);

  // If there aren't registers to copy the values into (e.g., during SjLj
  // exceptions), then don't bother to create these DAG nodes.
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  if (TLI.getExceptionPointerRegister() == 0 &&
      TLI.getExceptionSelectorRegister() == 0)
    return;

  SmallVector<EVT, 2> ValueVTs;
  ComputeValueVTs(TLI, LP.getType(), ValueVTs);
  assert(ValueVTs.size() == 2 && "Only two-valued landingpads are supported");

  // Get the two live-in registers as SDValues. The physregs have already been
  // copied into virtual registers.
  SDValue Ops[2];
  if (FuncInfo.ExceptionPointerVirtReg) {
    Ops[0] = DAG.getZExtOrTrunc(
        DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(),
                           FuncInfo.ExceptionPointerVirtReg, TLI.getPointerTy()),
        getCurSDLoc(), ValueVTs[0]);
  } else {
    Ops[0] = DAG.getConstant(0, TLI.getPointerTy());
  }
  Ops[1] = DAG.getZExtOrTrunc(
      DAG.getCopyFromReg(DAG.getEntryNode(), getCurSDLoc(),
                         FuncInfo.ExceptionSelectorVirtReg, TLI.getPointerTy()),
      getCurSDLoc(), ValueVTs[1]);

  // Merge into one.
  SDValue Res = DAG.getNode(ISD::MERGE_VALUES, getCurSDLoc(),
                            DAG.getVTList(ValueVTs), Ops);
  setValue(&LP, Res);
}

unsigned
SelectionDAGBuilder::visitLandingPadClauseBB(GlobalValue *ClauseGV,
                                             MachineBasicBlock *LPadBB) {
  SDValue Chain = getControlRoot();

  // Get the typeid that we will dispatch on later.
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  const TargetRegisterClass *RC = TLI.getRegClassFor(TLI.getPointerTy());
  unsigned VReg = FuncInfo.MF->getRegInfo().createVirtualRegister(RC);
  unsigned TypeID = DAG.getMachineFunction().getMMI().getTypeIDFor(ClauseGV);
  SDValue Sel = DAG.getConstant(TypeID, TLI.getPointerTy());
  Chain = DAG.getCopyToReg(Chain, getCurSDLoc(), VReg, Sel);

  // Branch to the main landing pad block.
  MachineBasicBlock *ClauseMBB = FuncInfo.MBB;
  ClauseMBB->addSuccessor(LPadBB);
  DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other, Chain,
                          DAG.getBasicBlock(LPadBB)));
  return VReg;
}

/// handleSmallSwitchCaseRange - Emit a series of specific tests (suitable for
/// small case ranges).
bool SelectionDAGBuilder::handleSmallSwitchRange(CaseRec& CR,
                                                 CaseRecVector& WorkList,
                                                 const Value* SV,
                                                 MachineBasicBlock *Default,
                                                 MachineBasicBlock *SwitchBB) {
  // Size is the number of Cases represented by this range.
  size_t Size = CR.Range.second - CR.Range.first;
  if (Size > 3)
    return false;

  // Get the MachineFunction which holds the current MBB.  This is used when
  // inserting any additional MBBs necessary to represent the switch.
  MachineFunction *CurMF = FuncInfo.MF;

  // Figure out which block is immediately after the current one.
  MachineBasicBlock *NextMBB = nullptr;
  MachineFunction::iterator BBI = CR.CaseBB;
  if (++BBI != FuncInfo.MF->end())
    NextMBB = BBI;

  BranchProbabilityInfo *BPI = FuncInfo.BPI;
  // If any two of the cases has the same destination, and if one value
  // is the same as the other, but has one bit unset that the other has set,
  // use bit manipulation to do two compares at once.  For example:
  // "if (X == 6 || X == 4)" -> "if ((X|2) == 6)"
  // TODO: This could be extended to merge any 2 cases in switches with 3 cases.
  // TODO: Handle cases where CR.CaseBB != SwitchBB.
  if (Size == 2 && CR.CaseBB == SwitchBB) {
    Case &Small = *CR.Range.first;
    Case &Big = *(CR.Range.second-1);

    if (Small.Low == Small.High && Big.Low == Big.High && Small.BB == Big.BB) {
      const APInt& SmallValue = Small.Low->getValue();
      const APInt& BigValue = Big.Low->getValue();

      // Check that there is only one bit different.
      if (BigValue.countPopulation() == SmallValue.countPopulation() + 1 &&
          (SmallValue | BigValue) == BigValue) {
        // Isolate the common bit.
        APInt CommonBit = BigValue & ~SmallValue;
        assert((SmallValue | CommonBit) == BigValue &&
               CommonBit.countPopulation() == 1 && "Not a common bit?");

        SDValue CondLHS = getValue(SV);
        EVT VT = CondLHS.getValueType();
        SDLoc DL = getCurSDLoc();

        SDValue Or = DAG.getNode(ISD::OR, DL, VT, CondLHS,
                                 DAG.getConstant(CommonBit, VT));
        SDValue Cond = DAG.getSetCC(DL, MVT::i1,
                                    Or, DAG.getConstant(BigValue, VT),
                                    ISD::SETEQ);

        // Update successor info.
        // Both Small and Big will jump to Small.BB, so we sum up the weights.
        addSuccessorWithWeight(SwitchBB, Small.BB,
                               Small.ExtraWeight + Big.ExtraWeight);
        addSuccessorWithWeight(SwitchBB, Default,
          // The default destination is the first successor in IR.
          BPI ? BPI->getEdgeWeight(SwitchBB->getBasicBlock(), (unsigned)0) : 0);

        // Insert the true branch.
        SDValue BrCond = DAG.getNode(ISD::BRCOND, DL, MVT::Other,
                                     getControlRoot(), Cond,
                                     DAG.getBasicBlock(Small.BB));

        // Insert the false branch.
        BrCond = DAG.getNode(ISD::BR, DL, MVT::Other, BrCond,
                             DAG.getBasicBlock(Default));

        DAG.setRoot(BrCond);
        return true;
      }
    }
  }

  // Order cases by weight so the most likely case will be checked first.
  uint32_t UnhandledWeights = 0;
  if (BPI) {
    for (CaseItr I = CR.Range.first, IE = CR.Range.second; I != IE; ++I) {
      uint32_t IWeight = I->ExtraWeight;
      UnhandledWeights += IWeight;
      for (CaseItr J = CR.Range.first; J < I; ++J) {
        uint32_t JWeight = J->ExtraWeight;
        if (IWeight > JWeight)
          std::swap(*I, *J);
      }
    }
  }
  // Rearrange the case blocks so that the last one falls through if possible.
  Case &BackCase = *(CR.Range.second-1);
  if (Size > 1 && NextMBB && Default != NextMBB && BackCase.BB != NextMBB) {
    // The last case block won't fall through into 'NextMBB' if we emit the
    // branches in this order.  See if rearranging a case value would help.
    // We start at the bottom as it's the case with the least weight.
    for (Case *I = &*(CR.Range.second-2), *E = &*CR.Range.first-1; I != E; --I)
      if (I->BB == NextMBB) {
        std::swap(*I, BackCase);
        break;
      }
  }

  // Create a CaseBlock record representing a conditional branch to
  // the Case's target mbb if the value being switched on SV is equal
  // to C.
  MachineBasicBlock *CurBlock = CR.CaseBB;
  for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) {
    MachineBasicBlock *FallThrough;
    if (I != E-1) {
      FallThrough = CurMF->CreateMachineBasicBlock(CurBlock->getBasicBlock());
      CurMF->insert(BBI, FallThrough);

      // Put SV in a virtual register to make it available from the new blocks.
      ExportFromCurrentBlock(SV);
    } else {
      // If the last case doesn't match, go to the default block.
      FallThrough = Default;
    }

    const Value *RHS, *LHS, *MHS;
    ISD::CondCode CC;
    if (I->High == I->Low) {
      // This is just small small case range :) containing exactly 1 case
      CC = ISD::SETEQ;
      LHS = SV; RHS = I->High; MHS = nullptr;
    } else {
      CC = ISD::SETLE;
      LHS = I->Low; MHS = SV; RHS = I->High;
    }

    // The false weight should be sum of all un-handled cases.
    UnhandledWeights -= I->ExtraWeight;
    CaseBlock CB(CC, LHS, RHS, MHS, /* truebb */ I->BB, /* falsebb */ FallThrough,
                 /* me */ CurBlock,
                 /* trueweight */ I->ExtraWeight,
                 /* falseweight */ UnhandledWeights);

    // If emitting the first comparison, just call visitSwitchCase to emit the
    // code into the current block.  Otherwise, push the CaseBlock onto the
    // vector to be later processed by SDISel, and insert the node's MBB
    // before the next MBB.
    if (CurBlock == SwitchBB)
      visitSwitchCase(CB, SwitchBB);
    else
      SwitchCases.push_back(CB);

    CurBlock = FallThrough;
  }

  return true;
}

static inline bool areJTsAllowed(const TargetLowering &TLI) {
  return TLI.isOperationLegalOrCustom(ISD::BR_JT, MVT::Other) ||
         TLI.isOperationLegalOrCustom(ISD::BRIND, MVT::Other);
}

static APInt ComputeRange(const APInt &First, const APInt &Last) {
  uint32_t BitWidth = std::max(Last.getBitWidth(), First.getBitWidth()) + 1;
  APInt LastExt = Last.sext(BitWidth), FirstExt = First.sext(BitWidth);
  return (LastExt - FirstExt + 1ULL);
}

/// handleJTSwitchCase - Emit jumptable for current switch case range
bool SelectionDAGBuilder::handleJTSwitchCase(CaseRec &CR,
                                             CaseRecVector &WorkList,
                                             const Value *SV,
                                             MachineBasicBlock *Default,
                                             MachineBasicBlock *SwitchBB) {
  Case& FrontCase = *CR.Range.first;
  Case& BackCase  = *(CR.Range.second-1);

  const APInt &First = FrontCase.Low->getValue();
  const APInt &Last  = BackCase.High->getValue();

  APInt TSize(First.getBitWidth(), 0);
  for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I)
    TSize += I->size();

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  if (!areJTsAllowed(TLI) || TSize.ult(TLI.getMinimumJumpTableEntries()))
    return false;

  APInt Range = ComputeRange(First, Last);
  // The density is TSize / Range. Require at least 40%.
  // It should not be possible for IntTSize to saturate for sane code, but make
  // sure we handle Range saturation correctly.
  uint64_t IntRange = Range.getLimitedValue(UINT64_MAX/10);
  uint64_t IntTSize = TSize.getLimitedValue(UINT64_MAX/10);
  if (IntTSize * 10 < IntRange * 4)
    return false;

  DEBUG(dbgs() << "Lowering jump table\n"
               << "First entry: " << First << ". Last entry: " << Last << '\n'
               << "Range: " << Range << ". Size: " << TSize << ".\n\n");

  // Get the MachineFunction which holds the current MBB.  This is used when
  // inserting any additional MBBs necessary to represent the switch.
  MachineFunction *CurMF = FuncInfo.MF;

  // Figure out which block is immediately after the current one.
  MachineFunction::iterator BBI = CR.CaseBB;
  ++BBI;

  const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock();

  // Create a new basic block to hold the code for loading the address
  // of the jump table, and jumping to it.  Update successor information;
  // we will either branch to the default case for the switch, or the jump
  // table.
  MachineBasicBlock *JumpTableBB = CurMF->CreateMachineBasicBlock(LLVMBB);
  CurMF->insert(BBI, JumpTableBB);

  addSuccessorWithWeight(CR.CaseBB, Default);
  addSuccessorWithWeight(CR.CaseBB, JumpTableBB);

  // Build a vector of destination BBs, corresponding to each target
  // of the jump table. If the value of the jump table slot corresponds to
  // a case statement, push the case's BB onto the vector, otherwise, push
  // the default BB.
  std::vector<MachineBasicBlock*> DestBBs;
  APInt TEI = First;
  for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++TEI) {
    const APInt &Low = I->Low->getValue();
    const APInt &High = I->High->getValue();

    if (Low.sle(TEI) && TEI.sle(High)) {
      DestBBs.push_back(I->BB);
      if (TEI==High)
        ++I;
    } else {
      DestBBs.push_back(Default);
    }
  }

  // Calculate weight for each unique destination in CR.
  DenseMap<MachineBasicBlock*, uint32_t> DestWeights;
  if (FuncInfo.BPI) {
    for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I)
      DestWeights[I->BB] += I->ExtraWeight;
  }

  // Update successor info. Add one edge to each unique successor.
  BitVector SuccsHandled(CR.CaseBB->getParent()->getNumBlockIDs());
  for (MachineBasicBlock *DestBB : DestBBs) {
    if (!SuccsHandled[DestBB->getNumber()]) {
      SuccsHandled[DestBB->getNumber()] = true;
      auto I = DestWeights.find(DestBB);
      addSuccessorWithWeight(JumpTableBB, DestBB,
                             I != DestWeights.end() ? I->second : 0);
    }
  }

  // Create a jump table index for this jump table.
  unsigned JTEncoding = TLI.getJumpTableEncoding();
  unsigned JTI = CurMF->getOrCreateJumpTableInfo(JTEncoding)
                       ->createJumpTableIndex(DestBBs);

  // Set the jump table information so that we can codegen it as a second
  // MachineBasicBlock
  JumpTable JT(-1U, JTI, JumpTableBB, Default);
  JumpTableHeader JTH(First, Last, SV, CR.CaseBB, (CR.CaseBB == SwitchBB));
  if (CR.CaseBB == SwitchBB)
    visitJumpTableHeader(JT, JTH, SwitchBB);

  JTCases.push_back(JumpTableBlock(JTH, JT));
  return true;
}

/// handleBTSplitSwitchCase - emit comparison and split binary search tree into
/// 2 subtrees.
bool SelectionDAGBuilder::handleBTSplitSwitchCase(CaseRec& CR,
                                                  CaseRecVector& WorkList,
                                                  const Value* SV,
                                                  MachineBasicBlock* SwitchBB) {
  Case& FrontCase = *CR.Range.first;
  Case& BackCase  = *(CR.Range.second-1);

  // Size is the number of Cases represented by this range.
  unsigned Size = CR.Range.second - CR.Range.first;

  const APInt &First = FrontCase.Low->getValue();
  const APInt &Last  = BackCase.High->getValue();
  double FMetric = 0;
  CaseItr Pivot = CR.Range.first + Size/2;

  // Select optimal pivot, maximizing sum density of LHS and RHS. This will
  // (heuristically) allow us to emit JumpTable's later.
  APInt TSize(First.getBitWidth(), 0);
  for (CaseItr I = CR.Range.first, E = CR.Range.second;
       I!=E; ++I)
    TSize += I->size();

  APInt LSize = FrontCase.size();
  APInt RSize = TSize-LSize;
  DEBUG(dbgs() << "Selecting best pivot: \n"
               << "First: " << First << ", Last: " << Last <<'\n'
               << "LSize: " << LSize << ", RSize: " << RSize << '\n');
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  for (CaseItr I = CR.Range.first, J=I+1, E = CR.Range.second;
       J!=E; ++I, ++J) {
    const APInt &LEnd = I->High->getValue();
    const APInt &RBegin = J->Low->getValue();
    APInt Range = ComputeRange(LEnd, RBegin);
    assert((Range - 2ULL).isNonNegative() &&
           "Invalid case distance");
    // Use volatile double here to avoid excess precision issues on some hosts,
    // e.g. that use 80-bit X87 registers.
    // Only consider the density of sub-ranges that actually have sufficient
    // entries to be lowered as a jump table.
    volatile double LDensity =
        LSize.ult(TLI.getMinimumJumpTableEntries())
            ? 0.0
            : LSize.roundToDouble() / (LEnd - First + 1ULL).roundToDouble();
    volatile double RDensity =
        RSize.ult(TLI.getMinimumJumpTableEntries())
            ? 0.0
            : RSize.roundToDouble() / (Last - RBegin + 1ULL).roundToDouble();
    volatile double Metric = Range.logBase2() * (LDensity + RDensity);
    // Should always split in some non-trivial place
    DEBUG(dbgs() <<"=>Step\n"
                 << "LEnd: " << LEnd << ", RBegin: " << RBegin << '\n'
                 << "LDensity: " << LDensity
                 << ", RDensity: " << RDensity << '\n'
                 << "Metric: " << Metric << '\n');
    if (FMetric < Metric) {
      Pivot = J;
      FMetric = Metric;
      DEBUG(dbgs() << "Current metric set to: " << FMetric << '\n');
    }

    LSize += J->size();
    RSize -= J->size();
  }

  if (FMetric == 0 || !areJTsAllowed(TLI))
    Pivot = CR.Range.first + Size/2;
  splitSwitchCase(CR, Pivot, WorkList, SV, SwitchBB);
  return true;
}

void SelectionDAGBuilder::splitSwitchCase(CaseRec &CR, CaseItr Pivot,
                                          CaseRecVector &WorkList,
                                          const Value *SV,
                                          MachineBasicBlock *SwitchBB) {
  // Get the MachineFunction which holds the current MBB.  This is used when
  // inserting any additional MBBs necessary to represent the switch.
  MachineFunction *CurMF = FuncInfo.MF;

  // Figure out which block is immediately after the current one.
  MachineFunction::iterator BBI = CR.CaseBB;
  ++BBI;

  const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock();

  CaseRange LHSR(CR.Range.first, Pivot);
  CaseRange RHSR(Pivot, CR.Range.second);
  const ConstantInt *C = Pivot->Low;
  MachineBasicBlock *FalseBB = nullptr, *TrueBB = nullptr;

  // We know that we branch to the LHS if the Value being switched on is
  // less than the Pivot value, C.  We use this to optimize our binary
  // tree a bit, by recognizing that if SV is greater than or equal to the
  // LHS's Case Value, and that Case Value is exactly one less than the
  // Pivot's Value, then we can branch directly to the LHS's Target,
  // rather than creating a leaf node for it.
  if ((LHSR.second - LHSR.first) == 1 && LHSR.first->High == CR.GE &&
      C->getValue() == (CR.GE->getValue() + 1LL)) {
    TrueBB = LHSR.first->BB;
  } else {
    TrueBB = CurMF->CreateMachineBasicBlock(LLVMBB);
    CurMF->insert(BBI, TrueBB);
    WorkList.push_back(CaseRec(TrueBB, C, CR.GE, LHSR));

    // Put SV in a virtual register to make it available from the new blocks.
    ExportFromCurrentBlock(SV);
  }

  // Similar to the optimization above, if the Value being switched on is
  // known to be less than the Constant CR.LT, and the current Case Value
  // is CR.LT - 1, then we can branch directly to the target block for
  // the current Case Value, rather than emitting a RHS leaf node for it.
  if ((RHSR.second - RHSR.first) == 1 && CR.LT &&
      RHSR.first->Low->getValue() == (CR.LT->getValue() - 1LL)) {
    FalseBB = RHSR.first->BB;
  } else {
    FalseBB = CurMF->CreateMachineBasicBlock(LLVMBB);
    CurMF->insert(BBI, FalseBB);
    WorkList.push_back(CaseRec(FalseBB, CR.LT, C, RHSR));

    // Put SV in a virtual register to make it available from the new blocks.
    ExportFromCurrentBlock(SV);
  }

  // Create a CaseBlock record representing a conditional branch to
  // the LHS node if the value being switched on SV is less than C.
  // Otherwise, branch to LHS.
  CaseBlock CB(ISD::SETLT, SV, C, nullptr, TrueBB, FalseBB, CR.CaseBB);

  if (CR.CaseBB == SwitchBB)
    visitSwitchCase(CB, SwitchBB);
  else
    SwitchCases.push_back(CB);
}

/// handleBitTestsSwitchCase - if current case range has few destination and
/// range span less, than machine word bitwidth, encode case range into series
/// of masks and emit bit tests with these masks.
bool SelectionDAGBuilder::handleBitTestsSwitchCase(CaseRec& CR,
                                                   CaseRecVector& WorkList,
                                                   const Value* SV,
                                                   MachineBasicBlock* Default,
                                                   MachineBasicBlock* SwitchBB) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT PTy = TLI.getPointerTy();
  unsigned IntPtrBits = PTy.getSizeInBits();

  Case& FrontCase = *CR.Range.first;
  Case& BackCase  = *(CR.Range.second-1);

  // Get the MachineFunction which holds the current MBB.  This is used when
  // inserting any additional MBBs necessary to represent the switch.
  MachineFunction *CurMF = FuncInfo.MF;

  // If target does not have legal shift left, do not emit bit tests at all.
  if (!TLI.isOperationLegal(ISD::SHL, PTy))
    return false;

  size_t numCmps = 0;
  for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) {
    // Single case counts one, case range - two.
    numCmps += (I->Low == I->High ? 1 : 2);
  }

  // Count unique destinations
  SmallSet<MachineBasicBlock*, 4> Dests;
  for (CaseItr I = CR.Range.first, E = CR.Range.second; I != E; ++I) {
    Dests.insert(I->BB);
    if (Dests.size() > 3)
      // Don't bother the code below, if there are too much unique destinations
      return false;
  }
  DEBUG(dbgs() << "Total number of unique destinations: "
        << Dests.size() << '\n'
        << "Total number of comparisons: " << numCmps << '\n');

  // Compute span of values.
  const APInt& minValue = FrontCase.Low->getValue();
  const APInt& maxValue = BackCase.High->getValue();
  APInt cmpRange = maxValue - minValue;

  DEBUG(dbgs() << "Compare range: " << cmpRange << '\n'
               << "Low bound: " << minValue << '\n'
               << "High bound: " << maxValue << '\n');

  if (cmpRange.uge(IntPtrBits) ||
      (!(Dests.size() == 1 && numCmps >= 3) &&
       !(Dests.size() == 2 && numCmps >= 5) &&
       !(Dests.size() >= 3 && numCmps >= 6)))
    return false;

  DEBUG(dbgs() << "Emitting bit tests\n");
  APInt lowBound = APInt::getNullValue(cmpRange.getBitWidth());

  // Optimize the case where all the case values fit in a
  // word without having to subtract minValue. In this case,
  // we can optimize away the subtraction.
  if (minValue.isNonNegative() && maxValue.slt(IntPtrBits)) {
    cmpRange = maxValue;
  } else {
    lowBound = minValue;
  }

  CaseBitsVector CasesBits;
  unsigned i, count = 0;

  for (CaseItr I = CR.Range.first, E = CR.Range.second; I!=E; ++I) {
    MachineBasicBlock* Dest = I->BB;
    for (i = 0; i < count; ++i)
      if (Dest == CasesBits[i].BB)
        break;

    if (i == count) {
      assert((count < 3) && "Too much destinations to test!");
      CasesBits.push_back(CaseBits(0, Dest, 0, 0/*Weight*/));
      count++;
    }

    const APInt& lowValue = I->Low->getValue();
    const APInt& highValue = I->High->getValue();

    uint64_t lo = (lowValue - lowBound).getZExtValue();
    uint64_t hi = (highValue - lowBound).getZExtValue();
    CasesBits[i].ExtraWeight += I->ExtraWeight;

    for (uint64_t j = lo; j <= hi; j++) {
      CasesBits[i].Mask |=  1ULL << j;
      CasesBits[i].Bits++;
    }

  }
  std::sort(CasesBits.begin(), CasesBits.end(), CaseBitsCmp());

  BitTestInfo BTC;

  // Figure out which block is immediately after the current one.
  MachineFunction::iterator BBI = CR.CaseBB;
  ++BBI;

  const BasicBlock *LLVMBB = CR.CaseBB->getBasicBlock();

  DEBUG(dbgs() << "Cases:\n");
  for (unsigned i = 0, e = CasesBits.size(); i!=e; ++i) {
    DEBUG(dbgs() << "Mask: " << CasesBits[i].Mask
                 << ", Bits: " << CasesBits[i].Bits
                 << ", BB: " << CasesBits[i].BB << '\n');

    MachineBasicBlock *CaseBB = CurMF->CreateMachineBasicBlock(LLVMBB);
    CurMF->insert(BBI, CaseBB);
    BTC.push_back(BitTestCase(CasesBits[i].Mask,
                              CaseBB,
                              CasesBits[i].BB, CasesBits[i].ExtraWeight));

    // Put SV in a virtual register to make it available from the new blocks.
    ExportFromCurrentBlock(SV);
  }

  BitTestBlock BTB(lowBound, cmpRange, SV,
                   -1U, MVT::Other, (CR.CaseBB == SwitchBB),
                   CR.CaseBB, Default, std::move(BTC));

  if (CR.CaseBB == SwitchBB)
    visitBitTestHeader(BTB, SwitchBB);

  BitTestCases.push_back(std::move(BTB));

  return true;
}

void SelectionDAGBuilder::Clusterify(CaseVector &Cases, const SwitchInst *SI) {
  BranchProbabilityInfo *BPI = FuncInfo.BPI;

  // Extract cases from the switch and sort them.
  typedef std::pair<const ConstantInt*, unsigned> CasePair;
  std::vector<CasePair> Sorted;
  Sorted.reserve(SI->getNumCases());
  for (auto I : SI->cases())
    Sorted.push_back(std::make_pair(I.getCaseValue(), I.getSuccessorIndex()));
  std::sort(Sorted.begin(), Sorted.end(), [](CasePair a, CasePair b) {
    return a.first->getValue().slt(b.first->getValue());
  });

  // Merge adjacent cases with the same destination, build Cases vector.
  assert(Cases.empty() && "Cases should be empty before Clusterify;");
  Cases.reserve(SI->getNumCases());
  MachineBasicBlock *PreviousSucc = nullptr;
  for (CasePair &CP : Sorted) {
    const ConstantInt *CaseVal = CP.first;
    unsigned SuccIndex = CP.second;
    MachineBasicBlock *Succ = FuncInfo.MBBMap[SI->getSuccessor(SuccIndex)];
    uint32_t Weight = BPI ? BPI->getEdgeWeight(SI->getParent(), SuccIndex) : 0;

    if (PreviousSucc == Succ &&
        (CaseVal->getValue() - Cases.back().High->getValue()) == 1) {
      // If this case has the same successor and is a neighbour, merge it into
      // the previous cluster.
      Cases.back().High = CaseVal;
      Cases.back().ExtraWeight += Weight;
    } else {
      Cases.push_back(Case(CaseVal, CaseVal, Succ, Weight));
    }

    PreviousSucc = Succ;
  }

  DEBUG({
      size_t numCmps = 0;
      for (auto &I : Cases)
        // A range counts double, since it requires two compares.
        numCmps += I.Low != I.High ? 2 : 1;

      dbgs() << "Clusterify finished. Total clusters: " << Cases.size()
             << ". Total compares: " << numCmps << '\n';
    });
}

void SelectionDAGBuilder::UpdateSplitBlock(MachineBasicBlock *First,
                                           MachineBasicBlock *Last) {
  // Update JTCases.
  for (unsigned i = 0, e = JTCases.size(); i != e; ++i)
    if (JTCases[i].first.HeaderBB == First)
      JTCases[i].first.HeaderBB = Last;

  // Update BitTestCases.
  for (unsigned i = 0, e = BitTestCases.size(); i != e; ++i)
    if (BitTestCases[i].Parent == First)
      BitTestCases[i].Parent = Last;
}

void SelectionDAGBuilder::visitSwitch(const SwitchInst &SI) {
  MachineBasicBlock *SwitchMBB = FuncInfo.MBB;

  // Create a vector of Cases, sorted so that we can efficiently create a binary
  // search tree from them.
  CaseVector Cases;
  Clusterify(Cases, &SI);

  // Get the default destination MBB.
  MachineBasicBlock *Default = FuncInfo.MBBMap[SI.getDefaultDest()];

  if (isa<UnreachableInst>(SI.getDefaultDest()->getFirstNonPHIOrDbg()) &&
      !Cases.empty()) {
    // Replace an unreachable default destination with the most popular case
    // destination.
    DenseMap<const BasicBlock *, unsigned> Popularity;
    unsigned MaxPop = 0;
    const BasicBlock *MaxBB = nullptr;
    for (auto I : SI.cases()) {
      const BasicBlock *BB = I.getCaseSuccessor();
      if (++Popularity[BB] > MaxPop) {
        MaxPop = Popularity[BB];
        MaxBB = BB;
      }
    }

    // Set new default.
    assert(MaxPop > 0);
    assert(MaxBB);
    Default = FuncInfo.MBBMap[MaxBB];

    // Remove cases that were pointing to the destination that is now the default.
    Cases.erase(std::remove_if(Cases.begin(), Cases.end(),
                               [&](const Case &C) { return C.BB == Default; }),
                Cases.end());
  }

  // If there is only the default destination, go there directly.
  if (Cases.empty()) {
    // Update machine-CFG edges.
    SwitchMBB->addSuccessor(Default);

    // If this is not a fall-through branch, emit the branch.
    if (Default != NextBlock(SwitchMBB)) {
      DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other,
                              getControlRoot(), DAG.getBasicBlock(Default)));
    }
    return;
  }

  // Get the Value to be switched on.
  const Value *SV = SI.getCondition();

  // Push the initial CaseRec onto the worklist
  CaseRecVector WorkList;
  WorkList.push_back(CaseRec(SwitchMBB,nullptr,nullptr,
                             CaseRange(Cases.begin(),Cases.end())));

  while (!WorkList.empty()) {
    // Grab a record representing a case range to process off the worklist
    CaseRec CR = WorkList.back();
    WorkList.pop_back();

    if (handleBitTestsSwitchCase(CR, WorkList, SV, Default, SwitchMBB))
      continue;

    // If the range has few cases (two or less) emit a series of specific
    // tests.
    if (handleSmallSwitchRange(CR, WorkList, SV, Default, SwitchMBB))
      continue;

    // If the switch has more than N blocks, and is at least 40% dense, and the
    // target supports indirect branches, then emit a jump table rather than
    // lowering the switch to a binary tree of conditional branches.
    // N defaults to 4 and is controlled via TLS.getMinimumJumpTableEntries().
    if (handleJTSwitchCase(CR, WorkList, SV, Default, SwitchMBB))
      continue;

    // Emit binary tree. We need to pick a pivot, and push left and right ranges
    // onto the worklist. Leafs are handled via handleSmallSwitchRange() call.
    handleBTSplitSwitchCase(CR, WorkList, SV, SwitchMBB);
  }
}

void SelectionDAGBuilder::visitIndirectBr(const IndirectBrInst &I) {
  MachineBasicBlock *IndirectBrMBB = FuncInfo.MBB;

  // Update machine-CFG edges with unique successors.
  SmallSet<BasicBlock*, 32> Done;
  for (unsigned i = 0, e = I.getNumSuccessors(); i != e; ++i) {
    BasicBlock *BB = I.getSuccessor(i);
    bool Inserted = Done.insert(BB).second;
    if (!Inserted)
        continue;

    MachineBasicBlock *Succ = FuncInfo.MBBMap[BB];
    addSuccessorWithWeight(IndirectBrMBB, Succ);
  }

  DAG.setRoot(DAG.getNode(ISD::BRIND, getCurSDLoc(),
                          MVT::Other, getControlRoot(),
                          getValue(I.getAddress())));
}

void SelectionDAGBuilder::visitUnreachable(const UnreachableInst &I) {
  if (DAG.getTarget().Options.TrapUnreachable)
    DAG.setRoot(DAG.getNode(ISD::TRAP, getCurSDLoc(), MVT::Other, DAG.getRoot()));
}

void SelectionDAGBuilder::visitFSub(const User &I) {
  // -0.0 - X --> fneg
  Type *Ty = I.getType();
  if (isa<Constant>(I.getOperand(0)) &&
      I.getOperand(0) == ConstantFP::getZeroValueForNegation(Ty)) {
    SDValue Op2 = getValue(I.getOperand(1));
    setValue(&I, DAG.getNode(ISD::FNEG, getCurSDLoc(),
                             Op2.getValueType(), Op2));
    return;
  }

  visitBinary(I, ISD::FSUB);
}

void SelectionDAGBuilder::visitBinary(const User &I, unsigned OpCode) {
  SDValue Op1 = getValue(I.getOperand(0));
  SDValue Op2 = getValue(I.getOperand(1));

  bool nuw = false;
  bool nsw = false;
  bool exact = false;
  if (const OverflowingBinaryOperator *OFBinOp =
          dyn_cast<const OverflowingBinaryOperator>(&I)) {
    nuw = OFBinOp->hasNoUnsignedWrap();
    nsw = OFBinOp->hasNoSignedWrap();
  }
  if (const PossiblyExactOperator *ExactOp =
          dyn_cast<const PossiblyExactOperator>(&I))
    exact = ExactOp->isExact();

  SDValue BinNodeValue = DAG.getNode(OpCode, getCurSDLoc(), Op1.getValueType(),
                                     Op1, Op2, nuw, nsw, exact);
  setValue(&I, BinNodeValue);
}

void SelectionDAGBuilder::visitShift(const User &I, unsigned Opcode) {
  SDValue Op1 = getValue(I.getOperand(0));
  SDValue Op2 = getValue(I.getOperand(1));

  EVT ShiftTy =
      DAG.getTargetLoweringInfo().getShiftAmountTy(Op2.getValueType());

  // Coerce the shift amount to the right type if we can.
  if (!I.getType()->isVectorTy() && Op2.getValueType() != ShiftTy) {
    unsigned ShiftSize = ShiftTy.getSizeInBits();
    unsigned Op2Size = Op2.getValueType().getSizeInBits();
    SDLoc DL = getCurSDLoc();

    // If the operand is smaller than the shift count type, promote it.
    if (ShiftSize > Op2Size)
      Op2 = DAG.getNode(ISD::ZERO_EXTEND, DL, ShiftTy, Op2);

    // If the operand is larger than the shift count type but the shift
    // count type has enough bits to represent any shift value, truncate
    // it now. This is a common case and it exposes the truncate to
    // optimization early.
    else if (ShiftSize >= Log2_32_Ceil(Op2.getValueType().getSizeInBits()))
      Op2 = DAG.getNode(ISD::TRUNCATE, DL, ShiftTy, Op2);
    // Otherwise we'll need to temporarily settle for some other convenient
    // type.  Type legalization will make adjustments once the shiftee is split.
    else
      Op2 = DAG.getZExtOrTrunc(Op2, DL, MVT::i32);
  }

  bool nuw = false;
  bool nsw = false;
  bool exact = false;

  if (Opcode == ISD::SRL || Opcode == ISD::SRA || Opcode == ISD::SHL) {

    if (const OverflowingBinaryOperator *OFBinOp =
            dyn_cast<const OverflowingBinaryOperator>(&I)) {
      nuw = OFBinOp->hasNoUnsignedWrap();
      nsw = OFBinOp->hasNoSignedWrap();
    }
    if (const PossiblyExactOperator *ExactOp =
            dyn_cast<const PossiblyExactOperator>(&I))
      exact = ExactOp->isExact();
  }

  SDValue Res = DAG.getNode(Opcode, getCurSDLoc(), Op1.getValueType(), Op1, Op2,
                            nuw, nsw, exact);
  setValue(&I, Res);
}

void SelectionDAGBuilder::visitSDiv(const User &I) {
  SDValue Op1 = getValue(I.getOperand(0));
  SDValue Op2 = getValue(I.getOperand(1));

  // Turn exact SDivs into multiplications.
  // FIXME: This should be in DAGCombiner, but it doesn't have access to the
  // exact bit.
  if (isa<BinaryOperator>(&I) && cast<BinaryOperator>(&I)->isExact() &&
      !isa<ConstantSDNode>(Op1) &&
      isa<ConstantSDNode>(Op2) && !cast<ConstantSDNode>(Op2)->isNullValue())
    setValue(&I, DAG.getTargetLoweringInfo()
                     .BuildExactSDIV(Op1, Op2, getCurSDLoc(), DAG));
  else
    setValue(&I, DAG.getNode(ISD::SDIV, getCurSDLoc(), Op1.getValueType(),
                             Op1, Op2));
}

void SelectionDAGBuilder::visitICmp(const User &I) {
  ICmpInst::Predicate predicate = ICmpInst::BAD_ICMP_PREDICATE;
  if (const ICmpInst *IC = dyn_cast<ICmpInst>(&I))
    predicate = IC->getPredicate();
  else if (const ConstantExpr *IC = dyn_cast<ConstantExpr>(&I))
    predicate = ICmpInst::Predicate(IC->getPredicate());
  SDValue Op1 = getValue(I.getOperand(0));
  SDValue Op2 = getValue(I.getOperand(1));
  ISD::CondCode Opcode = getICmpCondCode(predicate);

  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getSetCC(getCurSDLoc(), DestVT, Op1, Op2, Opcode));
}

void SelectionDAGBuilder::visitFCmp(const User &I) {
  FCmpInst::Predicate predicate = FCmpInst::BAD_FCMP_PREDICATE;
  if (const FCmpInst *FC = dyn_cast<FCmpInst>(&I))
    predicate = FC->getPredicate();
  else if (const ConstantExpr *FC = dyn_cast<ConstantExpr>(&I))
    predicate = FCmpInst::Predicate(FC->getPredicate());
  SDValue Op1 = getValue(I.getOperand(0));
  SDValue Op2 = getValue(I.getOperand(1));
  ISD::CondCode Condition = getFCmpCondCode(predicate);
  if (TM.Options.NoNaNsFPMath)
    Condition = getFCmpCodeWithoutNaN(Condition);
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getSetCC(getCurSDLoc(), DestVT, Op1, Op2, Condition));
}

void SelectionDAGBuilder::visitSelect(const User &I) {
  SmallVector<EVT, 4> ValueVTs;
  ComputeValueVTs(DAG.getTargetLoweringInfo(), I.getType(), ValueVTs);
  unsigned NumValues = ValueVTs.size();
  if (NumValues == 0) return;

  SmallVector<SDValue, 4> Values(NumValues);
  SDValue Cond     = getValue(I.getOperand(0));
  SDValue TrueVal  = getValue(I.getOperand(1));
  SDValue FalseVal = getValue(I.getOperand(2));
  ISD::NodeType OpCode = Cond.getValueType().isVector() ?
    ISD::VSELECT : ISD::SELECT;

  for (unsigned i = 0; i != NumValues; ++i)
    Values[i] = DAG.getNode(OpCode, getCurSDLoc(),
                            TrueVal.getNode()->getValueType(TrueVal.getResNo()+i),
                            Cond,
                            SDValue(TrueVal.getNode(),
                                    TrueVal.getResNo() + i),
                            SDValue(FalseVal.getNode(),
                                    FalseVal.getResNo() + i));

  setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurSDLoc(),
                           DAG.getVTList(ValueVTs), Values));
}

void SelectionDAGBuilder::visitTrunc(const User &I) {
  // TruncInst cannot be a no-op cast because sizeof(src) > sizeof(dest).
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::TRUNCATE, getCurSDLoc(), DestVT, N));
}

void SelectionDAGBuilder::visitZExt(const User &I) {
  // ZExt cannot be a no-op cast because sizeof(src) < sizeof(dest).
  // ZExt also can't be a cast to bool for same reason. So, nothing much to do
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::ZERO_EXTEND, getCurSDLoc(), DestVT, N));
}

void SelectionDAGBuilder::visitSExt(const User &I) {
  // SExt cannot be a no-op cast because sizeof(src) < sizeof(dest).
  // SExt also can't be a cast to bool for same reason. So, nothing much to do
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::SIGN_EXTEND, getCurSDLoc(), DestVT, N));
}

void SelectionDAGBuilder::visitFPTrunc(const User &I) {
  // FPTrunc is never a no-op cast, no need to check
  SDValue N = getValue(I.getOperand(0));
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT DestVT = TLI.getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::FP_ROUND, getCurSDLoc(), DestVT, N,
                           DAG.getTargetConstant(0, TLI.getPointerTy())));
}

void SelectionDAGBuilder::visitFPExt(const User &I) {
  // FPExt is never a no-op cast, no need to check
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::FP_EXTEND, getCurSDLoc(), DestVT, N));
}

void SelectionDAGBuilder::visitFPToUI(const User &I) {
  // FPToUI is never a no-op cast, no need to check
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::FP_TO_UINT, getCurSDLoc(), DestVT, N));
}

void SelectionDAGBuilder::visitFPToSI(const User &I) {
  // FPToSI is never a no-op cast, no need to check
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::FP_TO_SINT, getCurSDLoc(), DestVT, N));
}

void SelectionDAGBuilder::visitUIToFP(const User &I) {
  // UIToFP is never a no-op cast, no need to check
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::UINT_TO_FP, getCurSDLoc(), DestVT, N));
}

void SelectionDAGBuilder::visitSIToFP(const User &I) {
  // SIToFP is never a no-op cast, no need to check
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getNode(ISD::SINT_TO_FP, getCurSDLoc(), DestVT, N));
}

void SelectionDAGBuilder::visitPtrToInt(const User &I) {
  // What to do depends on the size of the integer and the size of the pointer.
  // We can either truncate, zero extend, or no-op, accordingly.
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getZExtOrTrunc(N, getCurSDLoc(), DestVT));
}

void SelectionDAGBuilder::visitIntToPtr(const User &I) {
  // What to do depends on the size of the integer and the size of the pointer.
  // We can either truncate, zero extend, or no-op, accordingly.
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());
  setValue(&I, DAG.getZExtOrTrunc(N, getCurSDLoc(), DestVT));
}

void SelectionDAGBuilder::visitBitCast(const User &I) {
  SDValue N = getValue(I.getOperand(0));
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(I.getType());

  // BitCast assures us that source and destination are the same size so this is
  // either a BITCAST or a no-op.
  if (DestVT != N.getValueType())
    setValue(&I, DAG.getNode(ISD::BITCAST, getCurSDLoc(),
                             DestVT, N)); // convert types.
  // Check if the original LLVM IR Operand was a ConstantInt, because getValue()
  // might fold any kind of constant expression to an integer constant and that
  // is not what we are looking for. Only regcognize a bitcast of a genuine
  // constant integer as an opaque constant.
  else if(ConstantInt *C = dyn_cast<ConstantInt>(I.getOperand(0)))
    setValue(&I, DAG.getConstant(C->getValue(), DestVT, /*isTarget=*/false,
                                 /*isOpaque*/true));
  else
    setValue(&I, N);            // noop cast.
}

void SelectionDAGBuilder::visitAddrSpaceCast(const User &I) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  const Value *SV = I.getOperand(0);
  SDValue N = getValue(SV);
  EVT DestVT = TLI.getValueType(I.getType());

  unsigned SrcAS = SV->getType()->getPointerAddressSpace();
  unsigned DestAS = I.getType()->getPointerAddressSpace();

  if (!TLI.isNoopAddrSpaceCast(SrcAS, DestAS))
    N = DAG.getAddrSpaceCast(getCurSDLoc(), DestVT, N, SrcAS, DestAS);

  setValue(&I, N);
}

void SelectionDAGBuilder::visitInsertElement(const User &I) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue InVec = getValue(I.getOperand(0));
  SDValue InVal = getValue(I.getOperand(1));
  SDValue InIdx = DAG.getSExtOrTrunc(getValue(I.getOperand(2)),
                                     getCurSDLoc(), TLI.getVectorIdxTy());
  setValue(&I, DAG.getNode(ISD::INSERT_VECTOR_ELT, getCurSDLoc(),
                           TLI.getValueType(I.getType()), InVec, InVal, InIdx));
}

void SelectionDAGBuilder::visitExtractElement(const User &I) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue InVec = getValue(I.getOperand(0));
  SDValue InIdx = DAG.getSExtOrTrunc(getValue(I.getOperand(1)),
                                     getCurSDLoc(), TLI.getVectorIdxTy());
  setValue(&I, DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurSDLoc(),
                           TLI.getValueType(I.getType()), InVec, InIdx));
}

// Utility for visitShuffleVector - Return true if every element in Mask,
// beginning from position Pos and ending in Pos+Size, falls within the
// specified sequential range [L, L+Pos). or is undef.
static bool isSequentialInRange(const SmallVectorImpl<int> &Mask,
                                unsigned Pos, unsigned Size, int Low) {
  for (unsigned i = Pos, e = Pos+Size; i != e; ++i, ++Low)
    if (Mask[i] >= 0 && Mask[i] != Low)
      return false;
  return true;
}

void SelectionDAGBuilder::visitShuffleVector(const User &I) {
  SDValue Src1 = getValue(I.getOperand(0));
  SDValue Src2 = getValue(I.getOperand(1));

  SmallVector<int, 8> Mask;
  ShuffleVectorInst::getShuffleMask(cast<Constant>(I.getOperand(2)), Mask);
  unsigned MaskNumElts = Mask.size();

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT VT = TLI.getValueType(I.getType());
  EVT SrcVT = Src1.getValueType();
  unsigned SrcNumElts = SrcVT.getVectorNumElements();

  if (SrcNumElts == MaskNumElts) {
    setValue(&I, DAG.getVectorShuffle(VT, getCurSDLoc(), Src1, Src2,
                                      &Mask[0]));
    return;
  }

  // Normalize the shuffle vector since mask and vector length don't match.
  if (SrcNumElts < MaskNumElts && MaskNumElts % SrcNumElts == 0) {
    // Mask is longer than the source vectors and is a multiple of the source
    // vectors.  We can use concatenate vector to make the mask and vectors
    // lengths match.
    if (SrcNumElts*2 == MaskNumElts) {
      // First check for Src1 in low and Src2 in high
      if (isSequentialInRange(Mask, 0, SrcNumElts, 0) &&
          isSequentialInRange(Mask, SrcNumElts, SrcNumElts, SrcNumElts)) {
        // The shuffle is concatenating two vectors together.
        setValue(&I, DAG.getNode(ISD::CONCAT_VECTORS, getCurSDLoc(),
                                 VT, Src1, Src2));
        return;
      }
      // Then check for Src2 in low and Src1 in high
      if (isSequentialInRange(Mask, 0, SrcNumElts, SrcNumElts) &&
          isSequentialInRange(Mask, SrcNumElts, SrcNumElts, 0)) {
        // The shuffle is concatenating two vectors together.
        setValue(&I, DAG.getNode(ISD::CONCAT_VECTORS, getCurSDLoc(),
                                 VT, Src2, Src1));
        return;
      }
    }

    // Pad both vectors with undefs to make them the same length as the mask.
    unsigned NumConcat = MaskNumElts / SrcNumElts;
    bool Src1U = Src1.getOpcode() == ISD::UNDEF;
    bool Src2U = Src2.getOpcode() == ISD::UNDEF;
    SDValue UndefVal = DAG.getUNDEF(SrcVT);

    SmallVector<SDValue, 8> MOps1(NumConcat, UndefVal);
    SmallVector<SDValue, 8> MOps2(NumConcat, UndefVal);
    MOps1[0] = Src1;
    MOps2[0] = Src2;

    Src1 = Src1U ? DAG.getUNDEF(VT) : DAG.getNode(ISD::CONCAT_VECTORS,
                                                  getCurSDLoc(), VT, MOps1);
    Src2 = Src2U ? DAG.getUNDEF(VT) : DAG.getNode(ISD::CONCAT_VECTORS,
                                                  getCurSDLoc(), VT, MOps2);

    // Readjust mask for new input vector length.
    SmallVector<int, 8> MappedOps;
    for (unsigned i = 0; i != MaskNumElts; ++i) {
      int Idx = Mask[i];
      if (Idx >= (int)SrcNumElts)
        Idx -= SrcNumElts - MaskNumElts;
      MappedOps.push_back(Idx);
    }

    setValue(&I, DAG.getVectorShuffle(VT, getCurSDLoc(), Src1, Src2,
                                      &MappedOps[0]));
    return;
  }

  if (SrcNumElts > MaskNumElts) {
    // Analyze the access pattern of the vector to see if we can extract
    // two subvectors and do the shuffle. The analysis is done by calculating
    // the range of elements the mask access on both vectors.
    int MinRange[2] = { static_cast<int>(SrcNumElts),
                        static_cast<int>(SrcNumElts)};
    int MaxRange[2] = {-1, -1};

    for (unsigned i = 0; i != MaskNumElts; ++i) {
      int Idx = Mask[i];
      unsigned Input = 0;
      if (Idx < 0)
        continue;

      if (Idx >= (int)SrcNumElts) {
        Input = 1;
        Idx -= SrcNumElts;
      }
      if (Idx > MaxRange[Input])
        MaxRange[Input] = Idx;
      if (Idx < MinRange[Input])
        MinRange[Input] = Idx;
    }

    // Check if the access is smaller than the vector size and can we find
    // a reasonable extract index.
    int RangeUse[2] = { -1, -1 };  // 0 = Unused, 1 = Extract, -1 = Can not
                                   // Extract.
    int StartIdx[2];  // StartIdx to extract from
    for (unsigned Input = 0; Input < 2; ++Input) {
      if (MinRange[Input] >= (int)SrcNumElts && MaxRange[Input] < 0) {
        RangeUse[Input] = 0; // Unused
        StartIdx[Input] = 0;
        continue;
      }

      // Find a good start index that is a multiple of the mask length. Then
      // see if the rest of the elements are in range.
      StartIdx[Input] = (MinRange[Input]/MaskNumElts)*MaskNumElts;
      if (MaxRange[Input] - StartIdx[Input] < (int)MaskNumElts &&
          StartIdx[Input] + MaskNumElts <= SrcNumElts)
        RangeUse[Input] = 1; // Extract from a multiple of the mask length.
    }

    if (RangeUse[0] == 0 && RangeUse[1] == 0) {
      setValue(&I, DAG.getUNDEF(VT)); // Vectors are not used.
      return;
    }
    if (RangeUse[0] >= 0 && RangeUse[1] >= 0) {
      // Extract appropriate subvector and generate a vector shuffle
      for (unsigned Input = 0; Input < 2; ++Input) {
        SDValue &Src = Input == 0 ? Src1 : Src2;
        if (RangeUse[Input] == 0)
          Src = DAG.getUNDEF(VT);
        else
          Src = DAG.getNode(
              ISD::EXTRACT_SUBVECTOR, getCurSDLoc(), VT, Src,
              DAG.getConstant(StartIdx[Input], TLI.getVectorIdxTy()));
      }

      // Calculate new mask.
      SmallVector<int, 8> MappedOps;
      for (unsigned i = 0; i != MaskNumElts; ++i) {
        int Idx = Mask[i];
        if (Idx >= 0) {
          if (Idx < (int)SrcNumElts)
            Idx -= StartIdx[0];
          else
            Idx -= SrcNumElts + StartIdx[1] - MaskNumElts;
        }
        MappedOps.push_back(Idx);
      }

      setValue(&I, DAG.getVectorShuffle(VT, getCurSDLoc(), Src1, Src2,
                                        &MappedOps[0]));
      return;
    }
  }

  // We can't use either concat vectors or extract subvectors so fall back to
  // replacing the shuffle with extract and build vector.
  // to insert and build vector.
  EVT EltVT = VT.getVectorElementType();
  EVT IdxVT = TLI.getVectorIdxTy();
  SmallVector<SDValue,8> Ops;
  for (unsigned i = 0; i != MaskNumElts; ++i) {
    int Idx = Mask[i];
    SDValue Res;

    if (Idx < 0) {
      Res = DAG.getUNDEF(EltVT);
    } else {
      SDValue &Src = Idx < (int)SrcNumElts ? Src1 : Src2;
      if (Idx >= (int)SrcNumElts) Idx -= SrcNumElts;

      Res = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurSDLoc(),
                        EltVT, Src, DAG.getConstant(Idx, IdxVT));
    }

    Ops.push_back(Res);
  }

  setValue(&I, DAG.getNode(ISD::BUILD_VECTOR, getCurSDLoc(), VT, Ops));
}

void SelectionDAGBuilder::visitInsertValue(const InsertValueInst &I) {
  const Value *Op0 = I.getOperand(0);
  const Value *Op1 = I.getOperand(1);
  Type *AggTy = I.getType();
  Type *ValTy = Op1->getType();
  bool IntoUndef = isa<UndefValue>(Op0);
  bool FromUndef = isa<UndefValue>(Op1);

  unsigned LinearIndex = ComputeLinearIndex(AggTy, I.getIndices());

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SmallVector<EVT, 4> AggValueVTs;
  ComputeValueVTs(TLI, AggTy, AggValueVTs);
  SmallVector<EVT, 4> ValValueVTs;
  ComputeValueVTs(TLI, ValTy, ValValueVTs);

  unsigned NumAggValues = AggValueVTs.size();
  unsigned NumValValues = ValValueVTs.size();
  SmallVector<SDValue, 4> Values(NumAggValues);

  // Ignore an insertvalue that produces an empty object
  if (!NumAggValues) {
    setValue(&I, DAG.getUNDEF(MVT(MVT::Other)));
    return;
  }

  SDValue Agg = getValue(Op0);
  unsigned i = 0;
  // Copy the beginning value(s) from the original aggregate.
  for (; i != LinearIndex; ++i)
    Values[i] = IntoUndef ? DAG.getUNDEF(AggValueVTs[i]) :
                SDValue(Agg.getNode(), Agg.getResNo() + i);
  // Copy values from the inserted value(s).
  if (NumValValues) {
    SDValue Val = getValue(Op1);
    for (; i != LinearIndex + NumValValues; ++i)
      Values[i] = FromUndef ? DAG.getUNDEF(AggValueVTs[i]) :
                  SDValue(Val.getNode(), Val.getResNo() + i - LinearIndex);
  }
  // Copy remaining value(s) from the original aggregate.
  for (; i != NumAggValues; ++i)
    Values[i] = IntoUndef ? DAG.getUNDEF(AggValueVTs[i]) :
                SDValue(Agg.getNode(), Agg.getResNo() + i);

  setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurSDLoc(),
                           DAG.getVTList(AggValueVTs), Values));
}

void SelectionDAGBuilder::visitExtractValue(const ExtractValueInst &I) {
  const Value *Op0 = I.getOperand(0);
  Type *AggTy = Op0->getType();
  Type *ValTy = I.getType();
  bool OutOfUndef = isa<UndefValue>(Op0);

  unsigned LinearIndex = ComputeLinearIndex(AggTy, I.getIndices());

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SmallVector<EVT, 4> ValValueVTs;
  ComputeValueVTs(TLI, ValTy, ValValueVTs);

  unsigned NumValValues = ValValueVTs.size();

  // Ignore a extractvalue that produces an empty object
  if (!NumValValues) {
    setValue(&I, DAG.getUNDEF(MVT(MVT::Other)));
    return;
  }

  SmallVector<SDValue, 4> Values(NumValValues);

  SDValue Agg = getValue(Op0);
  // Copy out the selected value(s).
  for (unsigned i = LinearIndex; i != LinearIndex + NumValValues; ++i)
    Values[i - LinearIndex] =
      OutOfUndef ?
        DAG.getUNDEF(Agg.getNode()->getValueType(Agg.getResNo() + i)) :
        SDValue(Agg.getNode(), Agg.getResNo() + i);

  setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurSDLoc(),
                           DAG.getVTList(ValValueVTs), Values));
}

void SelectionDAGBuilder::visitGetElementPtr(const User &I) {
  Value *Op0 = I.getOperand(0);
  // Note that the pointer operand may be a vector of pointers. Take the scalar
  // element which holds a pointer.
  Type *Ty = Op0->getType()->getScalarType();
  unsigned AS = Ty->getPointerAddressSpace();
  SDValue N = getValue(Op0);

  for (GetElementPtrInst::const_op_iterator OI = I.op_begin()+1, E = I.op_end();
       OI != E; ++OI) {
    const Value *Idx = *OI;
    if (StructType *StTy = dyn_cast<StructType>(Ty)) {
      unsigned Field = cast<Constant>(Idx)->getUniqueInteger().getZExtValue();
      if (Field) {
        // N = N + Offset
        uint64_t Offset = DL->getStructLayout(StTy)->getElementOffset(Field);
        N = DAG.getNode(ISD::ADD, getCurSDLoc(), N.getValueType(), N,
                        DAG.getConstant(Offset, N.getValueType()));
      }

      Ty = StTy->getElementType(Field);
    } else {
      Ty = cast<SequentialType>(Ty)->getElementType();
      MVT PtrTy = DAG.getTargetLoweringInfo().getPointerTy(AS);
      unsigned PtrSize = PtrTy.getSizeInBits();
      APInt ElementSize(PtrSize, DL->getTypeAllocSize(Ty));

      // If this is a constant subscript, handle it quickly.
      if (const auto *CI = dyn_cast<ConstantInt>(Idx)) {
        if (CI->isZero())
          continue;
        APInt Offs = ElementSize * CI->getValue().sextOrTrunc(PtrSize);
        SDValue OffsVal = DAG.getConstant(Offs, PtrTy);
        N = DAG.getNode(ISD::ADD, getCurSDLoc(), N.getValueType(), N, OffsVal);
        continue;
      }

      // N = N + Idx * ElementSize;
      SDValue IdxN = getValue(Idx);

      // If the index is smaller or larger than intptr_t, truncate or extend
      // it.
      IdxN = DAG.getSExtOrTrunc(IdxN, getCurSDLoc(), N.getValueType());

      // If this is a multiply by a power of two, turn it into a shl
      // immediately.  This is a very common case.
      if (ElementSize != 1) {
        if (ElementSize.isPowerOf2()) {
          unsigned Amt = ElementSize.logBase2();
          IdxN = DAG.getNode(ISD::SHL, getCurSDLoc(),
                             N.getValueType(), IdxN,
                             DAG.getConstant(Amt, IdxN.getValueType()));
        } else {
          SDValue Scale = DAG.getConstant(ElementSize, IdxN.getValueType());
          IdxN = DAG.getNode(ISD::MUL, getCurSDLoc(),
                             N.getValueType(), IdxN, Scale);
        }
      }

      N = DAG.getNode(ISD::ADD, getCurSDLoc(),
                      N.getValueType(), N, IdxN);
    }
  }

  setValue(&I, N);
}

void SelectionDAGBuilder::visitAlloca(const AllocaInst &I) {
  // If this is a fixed sized alloca in the entry block of the function,
  // allocate it statically on the stack.
  if (FuncInfo.StaticAllocaMap.count(&I))
    return;   // getValue will auto-populate this.

  Type *Ty = I.getAllocatedType();
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  uint64_t TySize = TLI.getDataLayout()->getTypeAllocSize(Ty);
  unsigned Align =
      std::max((unsigned)TLI.getDataLayout()->getPrefTypeAlignment(Ty),
               I.getAlignment());

  SDValue AllocSize = getValue(I.getArraySize());

  EVT IntPtr = TLI.getPointerTy();
  if (AllocSize.getValueType() != IntPtr)
    AllocSize = DAG.getZExtOrTrunc(AllocSize, getCurSDLoc(), IntPtr);

  AllocSize = DAG.getNode(ISD::MUL, getCurSDLoc(), IntPtr,
                          AllocSize,
                          DAG.getConstant(TySize, IntPtr));

  // Handle alignment.  If the requested alignment is less than or equal to
  // the stack alignment, ignore it.  If the size is greater than or equal to
  // the stack alignment, we note this in the DYNAMIC_STACKALLOC node.
  unsigned StackAlign =
      DAG.getSubtarget().getFrameLowering()->getStackAlignment();
  if (Align <= StackAlign)
    Align = 0;

  // Round the size of the allocation up to the stack alignment size
  // by add SA-1 to the size.
  AllocSize = DAG.getNode(ISD::ADD, getCurSDLoc(),
                          AllocSize.getValueType(), AllocSize,
                          DAG.getIntPtrConstant(StackAlign-1));

  // Mask out the low bits for alignment purposes.
  AllocSize = DAG.getNode(ISD::AND, getCurSDLoc(),
                          AllocSize.getValueType(), AllocSize,
                          DAG.getIntPtrConstant(~(uint64_t)(StackAlign-1)));

  SDValue Ops[] = { getRoot(), AllocSize, DAG.getIntPtrConstant(Align) };
  SDVTList VTs = DAG.getVTList(AllocSize.getValueType(), MVT::Other);
  SDValue DSA = DAG.getNode(ISD::DYNAMIC_STACKALLOC, getCurSDLoc(), VTs, Ops);
  setValue(&I, DSA);
  DAG.setRoot(DSA.getValue(1));

  assert(FuncInfo.MF->getFrameInfo()->hasVarSizedObjects());
}

void SelectionDAGBuilder::visitLoad(const LoadInst &I) {
  if (I.isAtomic())
    return visitAtomicLoad(I);

  const Value *SV = I.getOperand(0);
  SDValue Ptr = getValue(SV);

  Type *Ty = I.getType();

  bool isVolatile = I.isVolatile();
  bool isNonTemporal = I.getMetadata(LLVMContext::MD_nontemporal) != nullptr;
  bool isInvariant = I.getMetadata(LLVMContext::MD_invariant_load) != nullptr;
  unsigned Alignment = I.getAlignment();

  AAMDNodes AAInfo;
  I.getAAMetadata(AAInfo);
  const MDNode *Ranges = I.getMetadata(LLVMContext::MD_range);

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SmallVector<EVT, 4> ValueVTs;
  SmallVector<uint64_t, 4> Offsets;
  ComputeValueVTs(TLI, Ty, ValueVTs, &Offsets);
  unsigned NumValues = ValueVTs.size();
  if (NumValues == 0)
    return;

  SDValue Root;
  bool ConstantMemory = false;
  if (isVolatile || NumValues > MaxParallelChains)
    // Serialize volatile loads with other side effects.
    Root = getRoot();
  else if (AA->pointsToConstantMemory(
             AliasAnalysis::Location(SV, AA->getTypeStoreSize(Ty), AAInfo))) {
    // Do not serialize (non-volatile) loads of constant memory with anything.
    Root = DAG.getEntryNode();
    ConstantMemory = true;
  } else {
    // Do not serialize non-volatile loads against each other.
    Root = DAG.getRoot();
  }

  if (isVolatile)
    Root = TLI.prepareVolatileOrAtomicLoad(Root, getCurSDLoc(), DAG);

  SmallVector<SDValue, 4> Values(NumValues);
  SmallVector<SDValue, 4> Chains(std::min(unsigned(MaxParallelChains),
                                          NumValues));
  EVT PtrVT = Ptr.getValueType();
  unsigned ChainI = 0;
  for (unsigned i = 0; i != NumValues; ++i, ++ChainI) {
    // Serializing loads here may result in excessive register pressure, and
    // TokenFactor places arbitrary choke points on the scheduler. SD scheduling
    // could recover a bit by hoisting nodes upward in the chain by recognizing
    // they are side-effect free or do not alias. The optimizer should really
    // avoid this case by converting large object/array copies to llvm.memcpy
    // (MaxParallelChains should always remain as failsafe).
    if (ChainI == MaxParallelChains) {
      assert(PendingLoads.empty() && "PendingLoads must be serialized first");
      SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other,
                                  makeArrayRef(Chains.data(), ChainI));
      Root = Chain;
      ChainI = 0;
    }
    SDValue A = DAG.getNode(ISD::ADD, getCurSDLoc(),
                            PtrVT, Ptr,
                            DAG.getConstant(Offsets[i], PtrVT));
    SDValue L = DAG.getLoad(ValueVTs[i], getCurSDLoc(), Root,
                            A, MachinePointerInfo(SV, Offsets[i]), isVolatile,
                            isNonTemporal, isInvariant, Alignment, AAInfo,
                            Ranges);

    Values[i] = L;
    Chains[ChainI] = L.getValue(1);
  }

  if (!ConstantMemory) {
    SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other,
                                makeArrayRef(Chains.data(), ChainI));
    if (isVolatile)
      DAG.setRoot(Chain);
    else
      PendingLoads.push_back(Chain);
  }

  setValue(&I, DAG.getNode(ISD::MERGE_VALUES, getCurSDLoc(),
                           DAG.getVTList(ValueVTs), Values));
}

void SelectionDAGBuilder::visitStore(const StoreInst &I) {
  if (I.isAtomic())
    return visitAtomicStore(I);

  const Value *SrcV = I.getOperand(0);
  const Value *PtrV = I.getOperand(1);

  SmallVector<EVT, 4> ValueVTs;
  SmallVector<uint64_t, 4> Offsets;
  ComputeValueVTs(DAG.getTargetLoweringInfo(), SrcV->getType(),
                  ValueVTs, &Offsets);
  unsigned NumValues = ValueVTs.size();
  if (NumValues == 0)
    return;

  // Get the lowered operands. Note that we do this after
  // checking if NumResults is zero, because with zero results
  // the operands won't have values in the map.
  SDValue Src = getValue(SrcV);
  SDValue Ptr = getValue(PtrV);

  SDValue Root = getRoot();
  SmallVector<SDValue, 4> Chains(std::min(unsigned(MaxParallelChains),
                                          NumValues));
  EVT PtrVT = Ptr.getValueType();
  bool isVolatile = I.isVolatile();
  bool isNonTemporal = I.getMetadata(LLVMContext::MD_nontemporal) != nullptr;
  unsigned Alignment = I.getAlignment();

  AAMDNodes AAInfo;
  I.getAAMetadata(AAInfo);

  unsigned ChainI = 0;
  for (unsigned i = 0; i != NumValues; ++i, ++ChainI) {
    // See visitLoad comments.
    if (ChainI == MaxParallelChains) {
      SDValue Chain = DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other,
                                  makeArrayRef(Chains.data(), ChainI));
      Root = Chain;
      ChainI = 0;
    }
    SDValue Add = DAG.getNode(ISD::ADD, getCurSDLoc(), PtrVT, Ptr,
                              DAG.getConstant(Offsets[i], PtrVT));
    SDValue St = DAG.getStore(Root, getCurSDLoc(),
                              SDValue(Src.getNode(), Src.getResNo() + i),
                              Add, MachinePointerInfo(PtrV, Offsets[i]),
                              isVolatile, isNonTemporal, Alignment, AAInfo);
    Chains[ChainI] = St;
  }

  SDValue StoreNode = DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other,
                                  makeArrayRef(Chains.data(), ChainI));
  DAG.setRoot(StoreNode);
}

void SelectionDAGBuilder::visitMaskedStore(const CallInst &I) {
  SDLoc sdl = getCurSDLoc();

  // llvm.masked.store.*(Src0, Ptr, alignemt, Mask)
  Value  *PtrOperand = I.getArgOperand(1);
  SDValue Ptr = getValue(PtrOperand);
  SDValue Src0 = getValue(I.getArgOperand(0));
  SDValue Mask = getValue(I.getArgOperand(3));
  EVT VT = Src0.getValueType();
  unsigned Alignment = (cast<ConstantInt>(I.getArgOperand(2)))->getZExtValue();
  if (!Alignment)
    Alignment = DAG.getEVTAlignment(VT);

  AAMDNodes AAInfo;
  I.getAAMetadata(AAInfo);

  MachineMemOperand *MMO =
    DAG.getMachineFunction().
    getMachineMemOperand(MachinePointerInfo(PtrOperand),
                          MachineMemOperand::MOStore,  VT.getStoreSize(),
                          Alignment, AAInfo);
  SDValue StoreNode = DAG.getMaskedStore(getRoot(), sdl, Src0, Ptr, Mask, VT,
                                         MMO, false);
  DAG.setRoot(StoreNode);
  setValue(&I, StoreNode);
}

void SelectionDAGBuilder::visitMaskedLoad(const CallInst &I) {
  SDLoc sdl = getCurSDLoc();

  // @llvm.masked.load.*(Ptr, alignment, Mask, Src0)
  Value  *PtrOperand = I.getArgOperand(0);
  SDValue Ptr = getValue(PtrOperand);
  SDValue Src0 = getValue(I.getArgOperand(3));
  SDValue Mask = getValue(I.getArgOperand(2));

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT VT = TLI.getValueType(I.getType());
  unsigned Alignment = (cast<ConstantInt>(I.getArgOperand(1)))->getZExtValue();
  if (!Alignment)
    Alignment = DAG.getEVTAlignment(VT);

  AAMDNodes AAInfo;
  I.getAAMetadata(AAInfo);
  const MDNode *Ranges = I.getMetadata(LLVMContext::MD_range);

  SDValue InChain = DAG.getRoot();
  if (AA->pointsToConstantMemory(
      AliasAnalysis::Location(PtrOperand,
                              AA->getTypeStoreSize(I.getType()),
                              AAInfo))) {
    // Do not serialize (non-volatile) loads of constant memory with anything.
    InChain = DAG.getEntryNode();
  }

  MachineMemOperand *MMO =
    DAG.getMachineFunction().
    getMachineMemOperand(MachinePointerInfo(PtrOperand),
                          MachineMemOperand::MOLoad,  VT.getStoreSize(),
                          Alignment, AAInfo, Ranges);

  SDValue Load = DAG.getMaskedLoad(VT, sdl, InChain, Ptr, Mask, Src0, VT, MMO,
                                   ISD::NON_EXTLOAD);
  SDValue OutChain = Load.getValue(1);
  DAG.setRoot(OutChain);
  setValue(&I, Load);
}

void SelectionDAGBuilder::visitAtomicCmpXchg(const AtomicCmpXchgInst &I) {
  SDLoc dl = getCurSDLoc();
  AtomicOrdering SuccessOrder = I.getSuccessOrdering();
  AtomicOrdering FailureOrder = I.getFailureOrdering();
  SynchronizationScope Scope = I.getSynchScope();

  SDValue InChain = getRoot();

&nb