SelectionDAGBuilder.cpp

Go to the documentation of this file.

//===- 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/APFloat.h"
#include "llvm/ADT/APInt.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/BitVector.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/None.h"
#include "llvm/ADT/Optional.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/Triple.h"
#include "llvm/ADT/Twine.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/BranchProbabilityInfo.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/EHPersonalities.h"
#include "llvm/Analysis/Loads.h"
#include "llvm/Analysis/MemoryLocation.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/Analysis/VectorUtils.h"
#include "llvm/CodeGen/Analysis.h"
#include "llvm/CodeGen/FunctionLoweringInfo.h"
#include "llvm/CodeGen/GCMetadata.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/MachineBasicBlock.h"
#include "llvm/CodeGen/MachineFrameInfo.h"
#include "llvm/CodeGen/MachineFunction.h"
#include "llvm/CodeGen/MachineInstr.h"
#include "llvm/CodeGen/MachineInstrBuilder.h"
#include "llvm/CodeGen/MachineJumpTableInfo.h"
#include "llvm/CodeGen/MachineMemOperand.h"
#include "llvm/CodeGen/MachineModuleInfo.h"
#include "llvm/CodeGen/MachineOperand.h"
#include "llvm/CodeGen/MachineRegisterInfo.h"
#include "llvm/CodeGen/RuntimeLibcalls.h"
#include "llvm/CodeGen/SelectionDAG.h"
#include "llvm/CodeGen/SelectionDAGNodes.h"
#include "llvm/CodeGen/SelectionDAGTargetInfo.h"
#include "llvm/CodeGen/StackMaps.h"
#include "llvm/CodeGen/TargetFrameLowering.h"
#include "llvm/CodeGen/TargetInstrInfo.h"
#include "llvm/CodeGen/TargetLowering.h"
#include "llvm/CodeGen/TargetOpcodes.h"
#include "llvm/CodeGen/TargetRegisterInfo.h"
#include "llvm/CodeGen/TargetSubtargetInfo.h"
#include "llvm/CodeGen/ValueTypes.h"
#include "llvm/CodeGen/WinEHFuncInfo.h"
#include "llvm/IR/Argument.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/CFG.h"
#include "llvm/IR/CallSite.h"
#include "llvm/IR/CallingConv.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/ConstantRange.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/DebugInfoMetadata.h"
#include "llvm/IR/DebugLoc.h"
#include "llvm/IR/DerivedTypes.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/GetElementPtrTypeIterator.h"
#include "llvm/IR/InlineAsm.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/LLVMContext.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/IR/Statepoint.h"
#include "llvm/IR/Type.h"
#include "llvm/IR/User.h"
#include "llvm/IR/Value.h"
#include "llvm/MC/MCContext.h"
#include "llvm/MC/MCSymbol.h"
#include "llvm/Support/AtomicOrdering.h"
#include "llvm/Support/BranchProbability.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CodeGen.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/MachineValueType.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Target/TargetIntrinsicInfo.h"
#include "llvm/Target/TargetMachine.h"
#include "llvm/Target/TargetOptions.h"
#include <algorithm>
#include <cassert>
#include <cstddef>
#include <cstdint>
#include <cstring>
#include <iterator>
#include <limits>
#include <numeric>
#include <tuple>
#include <utility>
#include <vector>

using namespace llvm;
using namespace PatternMatch;

#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::Hidden,
                     cl::init(0));

static cl::opt<unsigned> SwitchPeelThreshold(
    "switch-peel-threshold", cl::Hidden, cl::init(66),
    cl::desc("Set the case probability threshold for peeling the case from a "
             "switch statement. A value greater than 100 will void this "
             "optimization"));

// Limit the width of DAG chains. This is important in general to 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 is 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;

// Return the calling convention if the Value passed requires ABI mangling as it
// is a parameter to a function or a return value from a function which is not
// an intrinsic.
static Optional<CallingConv::ID> getABIRegCopyCC(const Value *V) {
  if (auto *R = dyn_cast<ReturnInst>(V))
    return R->getParent()->getParent()->getCallingConv();

  if (auto *CI = dyn_cast<CallInst>(V)) {
    const bool IsInlineAsm = CI->isInlineAsm();
    const bool IsIndirectFunctionCall =
        !IsInlineAsm && !CI->getCalledFunction();

    // It is possible that the call instruction is an inline asm statement or an
    // indirect function call in which case the return value of
    // getCalledFunction() would be nullptr.
    const bool IsInstrinsicCall =
        !IsInlineAsm && !IsIndirectFunctionCall &&
        CI->getCalledFunction()->getIntrinsicID() != Intrinsic::not_intrinsic;

    if (!IsInlineAsm && !IsInstrinsicCall)
      return CI->getCallingConv();
  }

  return None;
}

static SDValue getCopyFromPartsVector(SelectionDAG &DAG, const SDLoc &DL,
                                      const SDValue *Parts, unsigned NumParts,
                                      MVT PartVT, EVT ValueVT, const Value *V,
                                      Optional<CallingConv::ID> CC);

/// getCopyFromParts - Create a value that contains the specified legal parts
/// combined into the value they represent.  If the parts combine to a type
/// larger than 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, const SDLoc &DL,
                                const SDValue *Parts, unsigned NumParts,
                                MVT PartVT, EVT ValueVT, const Value *V,
                                Optional<CallingConv::ID> CC = None,
                                Optional<ISD::NodeType> AssertOp = None) {
  if (ValueVT.isVector())
    return getCopyFromPartsVector(DAG, DL, Parts, NumParts, PartVT, ValueVT, V,
                                  CC);

  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 (DAG.getDataLayout().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, CC);

        // Combine the round and odd parts.
        Lo = Val;
        if (DAG.getDataLayout().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.getValueSizeInBits(), DL,
                                        TLI.getPointerTy(DAG.getDataLayout())));
        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, DAG.getDataLayout()))
        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, CC);
    }
  }

  // There is now one part, held in Val.  Correct it to match ValueVT.
  // PartEVT is the type of the register class that holds the value.
  // ValueVT is the type of the inline asm operation.
  EVT PartEVT = Val.getValueType();

  if (PartEVT == ValueVT)
    return Val;

  if (PartEVT.isInteger() && ValueVT.isFloatingPoint() &&
      ValueVT.bitsLT(PartEVT)) {
    // For an FP value in an integer part, we need to truncate to the right
    // width first.
    PartEVT = EVT::getIntegerVT(*DAG.getContext(),  ValueVT.getSizeInBits());
    Val = DAG.getNode(ISD::TRUNCATE, DL, PartEVT, Val);
  }

  // Handle types that have the same size.
  if (PartEVT.getSizeInBits() == ValueVT.getSizeInBits())
    return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);

  // Handle types with different sizes.
  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.hasValue())
        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, DL, TLI.getPointerTy(DAG.getDataLayout())));

    return DAG.getNode(ISD::FP_EXTEND, 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 than 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, const SDLoc &DL,
                                      const SDValue *Parts, unsigned NumParts,
                                      MVT PartVT, EVT ValueVT, const Value *V,
                                      Optional<CallingConv::ID> CallConv) {
  assert(ValueVT.isVector() && "Not a vector value");
  assert(NumParts > 0 && "No parts to assemble!");
  const bool IsABIRegCopy = CallConv.hasValue();

  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;

    if (IsABIRegCopy) {
      NumRegs = TLI.getVectorTypeBreakdownForCallingConv(
          *DAG.getContext(), CallConv.getValue(), ValueVT, IntermediateVT,
          NumIntermediates, RegisterVT);
    } else {
      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.getSizeInBits() ==
           Parts[0].getSimpleValueType().getSizeInBits() &&
           "Part type sizes don't match!");

    // 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.
    EVT BuiltVectorTy =
        EVT::getVectorVT(*DAG.getContext(), IntermediateVT.getScalarType(),
                         (IntermediateVT.isVector()
                              ? IntermediateVT.getVectorNumElements() * NumParts
                              : NumIntermediates));
    Val = DAG.getNode(IntermediateVT.isVector() ? ISD::CONCAT_VECTORS
                                                : ISD::BUILD_VECTOR,
                      DL, BuiltVectorTy, 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, DL, TLI.getVectorIdxTy(DAG.getDataLayout())));
    }

    // 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
    return DAG.getAnyExtOrTrunc(Val, DL, ValueVT);

  }

  // 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);

  if (ValueVT.getVectorNumElements() != 1) {
     // Certain ABIs require that vectors are passed as integers. For vectors
     // are the same size, this is an obvious bitcast.
     if (ValueVT.getSizeInBits() == PartEVT.getSizeInBits()) {
       return DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);
     } else if (ValueVT.getSizeInBits() < PartEVT.getSizeInBits()) {
       // Bitcast Val back the original type and extract the corresponding
       // vector we want.
       unsigned Elts = PartEVT.getSizeInBits() / ValueVT.getScalarSizeInBits();
       EVT WiderVecType = EVT::getVectorVT(*DAG.getContext(),
                                           ValueVT.getVectorElementType(), Elts);
       Val = DAG.getBitcast(WiderVecType, Val);
       return DAG.getNode(
           ISD::EXTRACT_SUBVECTOR, DL, ValueVT, Val,
           DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout())));
     }

     diagnosePossiblyInvalidConstraint(
         *DAG.getContext(), V, "non-trivial scalar-to-vector conversion");
     return DAG.getUNDEF(ValueVT);
  }

  // Handle cases such as i8 -> <1 x i1>
  EVT ValueSVT = ValueVT.getVectorElementType();
  if (ValueVT.getVectorNumElements() == 1 && ValueSVT != PartEVT)
    Val = ValueVT.isFloatingPoint() ? DAG.getFPExtendOrRound(Val, DL, ValueSVT)
                                    : DAG.getAnyExtOrTrunc(Val, DL, ValueSVT);

  return DAG.getBuildVector(ValueVT, DL, Val);
}

static void getCopyToPartsVector(SelectionDAG &DAG, const SDLoc &dl,
                                 SDValue Val, SDValue *Parts, unsigned NumParts,
                                 MVT PartVT, const Value *V,
                                 Optional<CallingConv::ID> CallConv);

/// 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, const SDLoc &DL, SDValue Val,
                           SDValue *Parts, unsigned NumParts, MVT PartVT,
                           const Value *V,
                           Optional<CallingConv::ID> CallConv = None,
                           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,
                                CallConv);

  unsigned PartBits = PartVT.getSizeInBits();
  unsigned OrigNumParts = NumParts;
  assert(DAG.getTargetLoweringInfo().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 {
      if (ValueVT.isFloatingPoint()) {
        // FP values need to be bitcast, then extended if they are being put
        // into a larger container.
        ValueVT = EVT::getIntegerVT(*DAG.getContext(),  ValueVT.getSizeInBits());
        Val = DAG.getNode(ISD::BITCAST, DL, ValueVT, Val);
      }
      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");
      Val = DAG.getNode(ISD::BITCAST, DL, PartVT, Val);
    }

    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, DL));
    getCopyToParts(DAG, DL, OddVal, Parts + RoundParts, OddParts, PartVT, V,
                   CallConv);

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

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

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

static SDValue widenVectorToPartType(SelectionDAG &DAG,
                                     SDValue Val, const SDLoc &DL, EVT PartVT) {
  if (!PartVT.isVector())
    return SDValue();

  EVT ValueVT = Val.getValueType();
  unsigned PartNumElts = PartVT.getVectorNumElements();
  unsigned ValueNumElts = ValueVT.getVectorNumElements();
  if (PartNumElts > ValueNumElts &&
      PartVT.getVectorElementType() == ValueVT.getVectorElementType()) {
    EVT ElementVT = PartVT.getVectorElementType();
    // Vector widening case, e.g. <2 x float> -> <4 x float>.  Shuffle in
    // undef elements.
    SmallVector<SDValue, 16> Ops;
    DAG.ExtractVectorElements(Val, Ops);
    SDValue EltUndef = DAG.getUNDEF(ElementVT);
    for (unsigned i = ValueNumElts, e = PartNumElts; i != e; ++i)
      Ops.push_back(EltUndef);

    // FIXME: Use CONCAT for 2x -> 4x.
    return DAG.getBuildVector(PartVT, DL, Ops);
  }

  return SDValue();
}

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

  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 (SDValue Widened = widenVectorToPartType(DAG, Val, DL, PartVT)) {
      Val = Widened;
    } else if (PartVT.isVector() &&
               PartEVT.getVectorElementType().bitsGE(
                 ValueVT.getVectorElementType()) &&
               PartEVT.getVectorNumElements() == ValueVT.getVectorNumElements()) {

      // Promoted vector extract
      Val = DAG.getAnyExtOrTrunc(Val, DL, PartVT);
    } else {
      if (ValueVT.getVectorNumElements() == 1) {
        Val = DAG.getNode(
            ISD::EXTRACT_VECTOR_ELT, DL, PartVT, Val,
            DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout())));
      } else {
        assert(PartVT.getSizeInBits() > ValueVT.getSizeInBits() &&
               "lossy conversion of vector to scalar type");
        EVT IntermediateType =
            EVT::getIntegerVT(*DAG.getContext(), ValueVT.getSizeInBits());
        Val = DAG.getBitcast(IntermediateType, Val);
        Val = DAG.getAnyExtOrTrunc(Val, DL, PartVT);
      }
    }

    assert(Val.getValueType() == PartVT && "Unexpected vector part value type");
    Parts[0] = Val;
    return;
  }

  // Handle a multi-element vector.
  EVT IntermediateVT;
  MVT RegisterVT;
  unsigned NumIntermediates;
  unsigned NumRegs;
  if (IsABIRegCopy) {
    NumRegs = TLI.getVectorTypeBreakdownForCallingConv(
        *DAG.getContext(), CallConv.getValue(), ValueVT, IntermediateVT,
        NumIntermediates, RegisterVT);
  } else {
    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!");

  unsigned IntermediateNumElts = IntermediateVT.isVector() ?
    IntermediateVT.getVectorNumElements() : 1;

  // Convert the vector to the appropiate type if necessary.
  unsigned DestVectorNoElts = NumIntermediates * IntermediateNumElts;

  EVT BuiltVectorTy = EVT::getVectorVT(
      *DAG.getContext(), IntermediateVT.getScalarType(), DestVectorNoElts);
  MVT IdxVT = TLI.getVectorIdxTy(DAG.getDataLayout());
  if (ValueVT != BuiltVectorTy) {
    if (SDValue Widened = widenVectorToPartType(DAG, Val, DL, BuiltVectorTy))
      Val = Widened;

    Val = DAG.getNode(ISD::BITCAST, DL, BuiltVectorTy, Val);
  }

  // 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 * IntermediateNumElts, DL, IdxVT));
    } else {
      Ops[i] = DAG.getNode(
          ISD::EXTRACT_VECTOR_ELT, DL, IntermediateVT, Val,
          DAG.getConstant(i, DL, IdxVT));
    }
  }

  // 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, CallConv);
  } 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,
                     CallConv);
  }
}

RegsForValue::RegsForValue(const SmallVector<unsigned, 4> &regs, MVT regvt,
                           EVT valuevt, Optional<CallingConv::ID> CC)
    : ValueVTs(1, valuevt), RegVTs(1, regvt), Regs(regs),
      RegCount(1, regs.size()), CallConv(CC) {}

RegsForValue::RegsForValue(LLVMContext &Context, const TargetLowering &TLI,
                           const DataLayout &DL, unsigned Reg, Type *Ty,
                           Optional<CallingConv::ID> CC) {
  ComputeValueVTs(TLI, DL, Ty, ValueVTs);

  CallConv = CC;

  for (EVT ValueVT : ValueVTs) {
    unsigned NumRegs =
        isABIMangled()
            ? TLI.getNumRegistersForCallingConv(Context, CC.getValue(), ValueVT)
            : TLI.getNumRegisters(Context, ValueVT);
    MVT RegisterVT =
        isABIMangled()
            ? TLI.getRegisterTypeForCallingConv(Context, CC.getValue(), ValueVT)
            : TLI.getRegisterType(Context, ValueVT);
    for (unsigned i = 0; i != NumRegs; ++i)
      Regs.push_back(Reg + i);
    RegVTs.push_back(RegisterVT);
    RegCount.push_back(NumRegs);
    Reg += NumRegs;
  }
}

SDValue RegsForValue::getCopyFromRegs(SelectionDAG &DAG,
                                      FunctionLoweringInfo &FuncInfo,
                                      const 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 = RegCount[Value];
    MVT RegisterVT = isABIMangled() ? TLI.getRegisterTypeForCallingConv(
                                          *DAG.getContext(),
                                          CallConv.getValue(), RegVTs[Value])
                                    : 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())
        continue;

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

      unsigned RegSize = RegisterVT.getScalarSizeInBits();
      unsigned NumSignBits = LOI->NumSignBits;
      unsigned NumZeroBits = LOI->Known.countMinLeadingZeros();

      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, dl, RegisterVT);
        continue;
      }

      // FIXME: We capture more information than the dag can represent.  For
      // now, just use the tightest assertzext/assertsext possible.
      bool isSExt;
      EVT FromVT(MVT::Other);
      if (NumZeroBits) {
        FromVT = EVT::getIntegerVT(*DAG.getContext(), RegSize - NumZeroBits);
        isSExt = false;
      } else if (NumSignBits > 1) {
        FromVT =
            EVT::getIntegerVT(*DAG.getContext(), RegSize - NumSignBits + 1);
        isSExt = true;
      } 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, CallConv);
    Part += NumRegs;
    Parts.clear();
  }

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

void RegsForValue::getCopyToRegs(SDValue Val, SelectionDAG &DAG,
                                 const 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) {
    unsigned NumParts = RegCount[Value];

    MVT RegisterVT = isABIMangled() ? TLI.getRegisterTypeForCallingConv(
                                          *DAG.getContext(),
                                          CallConv.getValue(), RegVTs[Value])
                                    : 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, CallConv, 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);
}

void RegsForValue::AddInlineAsmOperands(unsigned Code, bool HasMatching,
                                        unsigned MatchingIdx, const SDLoc &dl,
                                        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, dl, MVT::i32);
  Ops.push_back(Res);

  if (Code == InlineAsm::Kind_Clobber) {
    // Clobbers should always have a 1:1 mapping with registers, and may
    // reference registers that have illegal (e.g. vector) types. Hence, we
    // shouldn't try to apply any sort of splitting logic to them.
    assert(Regs.size() == RegVTs.size() && Regs.size() == ValueVTs.size() &&
           "No 1:1 mapping from clobbers to regs?");
    unsigned SP = TLI.getStackPointerRegisterToSaveRestore();
    (void)SP;
    for (unsigned I = 0, E = ValueVTs.size(); I != E; ++I) {
      Ops.push_back(DAG.getRegister(Regs[I], RegVTs[I]));
      assert(
          (Regs[I] != SP ||
           DAG.getMachineFunction().getFrameInfo().hasOpaqueSPAdjustment()) &&
          "If we clobbered the stack pointer, MFI should know about it.");
    }
    return;
  }

  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));
    }
  }
}

SmallVector<std::pair<unsigned, unsigned>, 4>
RegsForValue::getRegsAndSizes() const {
  SmallVector<std::pair<unsigned, unsigned>, 4> OutVec;
  unsigned I = 0;
  for (auto CountAndVT : zip_first(RegCount, RegVTs)) {
    unsigned RegCount = std::get<0>(CountAndVT);
    MVT RegisterVT = std::get<1>(CountAndVT);
    unsigned RegisterSize = RegisterVT.getSizeInBits();
    for (unsigned E = I + RegCount; I != E; ++I)
      OutVec.push_back(std::make_pair(Regs[I], RegisterSize));
  }
  return OutVec;
}

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

void SelectionDAGBuilder::clear() {
  NodeMap.clear();
  UnusedArgNodeMap.clear();
  PendingLoads.clear();
  PendingExports.clear();
  CurInst = nullptr;
  HasTailCall = false;
  SDNodeOrder = LowestSDNodeOrder;
  StatepointLowering.clear();
}

void SelectionDAGBuilder::clearDanglingDebugInfo() {
  DanglingDebugInfoMap.clear();
}

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.
  // If we have >= 2^16 loads then split across multiple token factors as
  // there's a 64k limit on the number of SDNode operands.
  SDValue Root;
  size_t Limit = (1 << 16) - 1;
  while (PendingLoads.size() > Limit) {
    unsigned SliceIdx = PendingLoads.size() - Limit;
    auto ExtractedTFs = ArrayRef<SDValue>(PendingLoads).slice(SliceIdx, Limit);
    SDValue NewTF =
        DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other, ExtractedTFs);
    PendingLoads.erase(PendingLoads.begin() + SliceIdx, PendingLoads.end());
    PendingLoads.emplace_back(NewTF);
  }
  Root = DAG.getNode(ISD::TokenFactor, getCurSDLoc(), MVT::Other, PendingLoads);
  PendingLoads.clear();
  DAG.setRoot(Root);
  return Root;
}

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 (I.isTerminator()) {
    HandlePHINodesInSuccessorBlocks(I.getParent());
  }

  // Increase the SDNodeOrder if dealing with a non-debug instruction.
  if (!isa<DbgInfoIntrinsic>(I))
    ++SDNodeOrder;

  CurInst = &I;

  visit(I.getOpcode(), I);

  if (auto *FPMO = dyn_cast<FPMathOperator>(&I)) {
    // Propagate the fast-math-flags of this IR instruction to the DAG node that
    // maps to this instruction.
    // TODO: We could handle all flags (nsw, etc) here.
    // TODO: If an IR instruction maps to >1 node, only the final node will have
    //       flags set.
    if (SDNode *Node = getNodeForIRValue(&I)) {
      SDNodeFlags IncomingFlags;
      IncomingFlags.copyFMF(*FPMO);
      if (!Node->getFlags().isDefined())
        Node->setFlags(IncomingFlags);
      else
        Node->intersectFlagsWith(IncomingFlags);
    }
  }

  if (!I.isTerminator() && !HasTailCall &&
      !isStatepoint(&I)) // statepoints handle their exports internally
    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"
  }
}

void SelectionDAGBuilder::dropDanglingDebugInfo(const DILocalVariable *Variable,
                                                const DIExpression *Expr) {
  auto isMatchingDbgValue = [&](DanglingDebugInfo &DDI) {
    const DbgValueInst *DI = DDI.getDI();
    DIVariable *DanglingVariable = DI->getVariable();
    DIExpression *DanglingExpr = DI->getExpression();
    if (DanglingVariable == Variable && Expr->fragmentsOverlap(DanglingExpr)) {
      LLVM_DEBUG(dbgs() << "Dropping dangling debug info for " << *DI << "\n");
      return true;
    }
    return false;
  };

  for (auto &DDIMI : DanglingDebugInfoMap) {
    DanglingDebugInfoVector &DDIV = DDIMI.second;
    DDIV.erase(remove_if(DDIV, isMatchingDbgValue), DDIV.end());
  }
}

// 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) {
  auto DanglingDbgInfoIt = DanglingDebugInfoMap.find(V);
  if (DanglingDbgInfoIt == DanglingDebugInfoMap.end())
    return;

  DanglingDebugInfoVector &DDIV = DanglingDbgInfoIt->second;
  for (auto &DDI : DDIV) {
    const DbgValueInst *DI = DDI.getDI();
    assert(DI && "Ill-formed DanglingDebugInfo");
    DebugLoc dl = DDI.getdl();
    unsigned ValSDNodeOrder = Val.getNode()->getIROrder();
    unsigned DbgSDNodeOrder = DDI.getSDNodeOrder();
    DILocalVariable *Variable = DI->getVariable();
    DIExpression *Expr = DI->getExpression();
    assert(Variable->isValidLocationForIntrinsic(dl) &&
           "Expected inlined-at fields to agree");
    SDDbgValue *SDV;
    if (Val.getNode()) {
      if (!EmitFuncArgumentDbgValue(V, Variable, Expr, dl, false, Val)) {
        LLVM_DEBUG(dbgs() << "Resolve dangling debug info [order="
                          << DbgSDNodeOrder << "] for:\n  " << *DI << "\n");
        LLVM_DEBUG(dbgs() << "  By mapping to:\n    "; Val.dump());
        // Increase the SDNodeOrder for the DbgValue here to make sure it is
        // inserted after the definition of Val when emitting the instructions
        // after ISel. An alternative could be to teach
        // ScheduleDAGSDNodes::EmitSchedule to delay the insertion properly.
        LLVM_DEBUG(if (ValSDNodeOrder > DbgSDNodeOrder) dbgs()
                   << "changing SDNodeOrder from " << DbgSDNodeOrder << " to "
                   << ValSDNodeOrder << "\n");
        SDV = getDbgValue(Val, Variable, Expr, dl,
                          std::max(DbgSDNodeOrder, ValSDNodeOrder));
        DAG.AddDbgValue(SDV, Val.getNode(), false);
      } else
        LLVM_DEBUG(dbgs() << "Resolved dangling debug info for " << *DI
                          << "in EmitFuncArgumentDbgValue\n");
    } else
      LLVM_DEBUG(dbgs() << "Dropping debug info for " << *DI << "\n");
  }
  DDIV.clear();
}

/// 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 Result;

  if (It != FuncInfo.ValueMap.end()) {
    unsigned InReg = It->second;

    RegsForValue RFV(*DAG.getContext(), DAG.getTargetLoweringInfo(),
                     DAG.getDataLayout(), InReg, Ty,
                     None); // This is not an ABI copy.
    SDValue Chain = DAG.getEntryNode();
    Result = RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr,
                                 V);
    resolveDanglingDebugInfo(V, Result);
  }

  return Result;
}

/// 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.
  if (SDValue copyFromReg = getCopyFromRegs(V, V->getType()))
    return copyFromReg;

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

// Return true if SDValue exists for the given Value
bool SelectionDAGBuilder::findValue(const Value *V) const {
  return (NodeMap.find(V) != NodeMap.end()) ||
    (FuncInfo.ValueMap.find(V) != FuncInfo.ValueMap.end());
}

/// 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()) {
    if (isa<ConstantSDNode>(N) || isa<ConstantFPSDNode>(N)) {
      // Remove the debug location from the node as the node is about to be used
      // in a location which may differ from the original debug location.  This
      // is relevant to Constant and ConstantFP nodes because they can appear
      // as constant expressions inside PHI nodes.
      N->setDebugLoc(DebugLoc());
    }
    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(DAG.getDataLayout(), V->getType(), true);

    if (const ConstantInt *CI = dyn_cast<ConstantInt>(C))
      return DAG.getConstant(*CI, getCurSDLoc(), 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, getCurSDLoc(),
                             TLI.getPointerTy(DAG.getDataLayout(), AS));
    }

    if (const ConstantFP *CFP = dyn_cast<ConstantFP>(C))
      return DAG.getConstantFP(*CFP, getCurSDLoc(), 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.getBuildVector(VT, getCurSDLoc(), 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, DAG.getDataLayout(), 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, getCurSDLoc(), EltVT);
        else
          Constants[i] = DAG.getConstant(0, getCurSDLoc(), 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(DAG.getDataLayout(), VecTy->getElementType());

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

    // Create a BUILD_VECTOR node.
    return NodeMap[V] = DAG.getBuildVector(VT, getCurSDLoc(), 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.getFrameIndexTy(DAG.getDataLayout()));
  }

  // 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, DAG.getDataLayout(), InReg,
                     Inst->getType(), getABIRegCopyCC(V));
    SDValue Chain = DAG.getEntryNode();
    return RFV.getCopyFromRegs(DAG, FuncInfo, getCurSDLoc(), Chain, nullptr, V);
  }

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

void SelectionDAGBuilder::visitCatchPad(const CatchPadInst &I) {
  auto Pers = classifyEHPersonality(FuncInfo.Fn->getPersonalityFn());
  bool IsMSVCCXX = Pers == EHPersonality::MSVC_CXX;
  bool IsCoreCLR = Pers == EHPersonality::CoreCLR;
  bool IsSEH = isAsynchronousEHPersonality(Pers);
  bool IsWasmCXX = Pers == EHPersonality::Wasm_CXX;
  MachineBasicBlock *CatchPadMBB = FuncInfo.MBB;
  if (!IsSEH)
    CatchPadMBB->setIsEHScopeEntry();
  // In MSVC C++ and CoreCLR, catchblocks are funclets and need prologues.
  if (IsMSVCCXX || IsCoreCLR)
    CatchPadMBB->setIsEHFuncletEntry();
  // Wasm does not need catchpads anymore
  if (!IsWasmCXX)
    DAG.setRoot(DAG.getNode(ISD::CATCHPAD, getCurSDLoc(), MVT::Other,
                            getControlRoot()));
}

void SelectionDAGBuilder::visitCatchRet(const CatchReturnInst &I) {
  // Update machine-CFG edge.
  MachineBasicBlock *TargetMBB = FuncInfo.MBBMap[I.getSuccessor()];
  FuncInfo.MBB->addSuccessor(TargetMBB);

  auto Pers = classifyEHPersonality(FuncInfo.Fn->getPersonalityFn());
  bool IsSEH = isAsynchronousEHPersonality(Pers);
  if (IsSEH) {
    // If this is not a fall-through branch or optimizations are switched off,
    // emit the branch.
    if (TargetMBB != NextBlock(FuncInfo.MBB) ||
        TM.getOptLevel() == CodeGenOpt::None)
      DAG.setRoot(DAG.getNode(ISD::BR, getCurSDLoc(), MVT::Other,
                              getControlRoot(), DAG.getBasicBlock(TargetMBB)));
    return;
  }

  // Figure out the funclet membership for the catchret's successor.
  // This will be used by the FuncletLayout pass to determine how to order the
  // BB's.
  // A 'catchret' returns to the outer scope's color.
  Value *ParentPad = I.getCatchSwitchParentPad();
  const BasicBlock *SuccessorColor;
  if (isa<ConstantTokenNone>(ParentPad))
    SuccessorColor = &FuncInfo.Fn->getEntryBlock();
  else
    SuccessorColor = cast<Instruction>(ParentPad)->getParent();
  assert(SuccessorColor && "No parent funclet for catchret!");
  MachineBasicBlock *SuccessorColorMBB = FuncInfo.MBBMap[SuccessorColor];
  assert(SuccessorColorMBB && "No MBB for SuccessorColor!");

  // Create the terminator node.
  SDValue Ret = DAG.getNode(ISD::CATCHRET, getCurSDLoc(), MVT::Other,
                            getControlRoot(), DAG.getBasicBlock(TargetMBB),
                            DAG.getBasicBlock(SuccessorColorMBB));
  DAG.setRoot(Ret);
}

void SelectionDAGBuilder::visitCleanupPad(const CleanupPadInst &CPI) {
  // Don't emit any special code for the cleanuppad instruction. It just marks
  // the start of an EH scope/funclet.
  FuncInfo.MBB->setIsEHScopeEntry();
  auto Pers = classifyEHPersonality(FuncInfo.Fn->getPersonalityFn());
  if (Pers != EHPersonality::Wasm_CXX) {
    FuncInfo.MBB->setIsEHFuncletEntry();
    FuncInfo.MBB->setIsCleanupFuncletEntry();
  }
}

/// When an invoke or a cleanupret unwinds to the next EH pad, there are
/// many places it could ultimately go. In the IR, we have a single unwind
/// destination, but in the machine CFG, we enumerate all the possible blocks.
/// This function skips over imaginary basic blocks that hold catchswitch
/// instructions, and finds all the "real" machine
/// basic block destinations. As those destinations may not be successors of
/// EHPadBB, here we also calculate the edge probability to those destinations.
/// The passed-in Prob is the edge probability to EHPadBB.
static void findUnwindDestinations(
    FunctionLoweringInfo &FuncInfo, const BasicBlock *EHPadBB,
    BranchProbability Prob,
    SmallVectorImpl<std::pair<MachineBasicBlock *, BranchProbability>>
        &UnwindDests) {
  EHPersonality Personality =
    classifyEHPersonality(FuncInfo.Fn->getPersonalityFn());
  bool IsMSVCCXX = Personality == EHPersonality::MSVC_CXX;
  bool IsCoreCLR = Personality == EHPersonality::CoreCLR;
  bool IsWasmCXX = Personality == EHPersonality::Wasm_CXX;
  bool IsSEH = isAsynchronousEHPersonality(Personality);

  while (EHPadBB) {
    const Instruction *Pad = EHPadBB->getFirstNonPHI();
    BasicBlock *NewEHPadBB = nullptr;
    if (isa<LandingPadInst>(Pad)) {
      // Stop on landingpads. They are not funclets.
      UnwindDests.emplace_back(FuncInfo.MBBMap[EHPadBB], Prob);
      break;
    } else if (isa<CleanupPadInst>(Pad)) {
      // Stop on cleanup pads. Cleanups are always funclet entries for all known
      // personalities.
      UnwindDests.emplace_back(FuncInfo.MBBMap[EHPadBB], Prob);
      UnwindDests.back().first->setIsEHScopeEntry();
      if (!IsWasmCXX)
        UnwindDests.back().first->setIsEHFuncletEntry();
      break;
    } else if (auto *CatchSwitch = dyn_cast<CatchSwitchInst>(Pad)) {
      // Add the catchpad handlers to the possible destinations.
      for (const BasicBlock *CatchPadBB : CatchSwitch->handlers()) {
        UnwindDests.emplace_back(FuncInfo.MBBMap[CatchPadBB], Prob);
        // For MSVC++ and the CLR, catchblocks are funclets and need prologues.
        if (IsMSVCCXX || IsCoreCLR)
          UnwindDests.back().first->setIsEHFuncletEntry();
        if (!IsSEH)
          UnwindDests.back().first->setIsEHScopeEntry();
      }
      NewEHPadBB = CatchSwitch->getUnwindDest();
    } else {
      continue;
    }

    BranchProbabilityInfo *BPI = FuncInfo.BPI;
    if (BPI && NewEHPadBB)
      Prob *= BPI->getEdgeProbability(EHPadBB, NewEHPadBB);
    EHPadBB = NewEHPadBB;
  }
}

void SelectionDAGBuilder::visitCleanupRet(const CleanupReturnInst &I) {
  // Update successor info.
  SmallVector<std::pair<MachineBasicBlock *, BranchProbability>, 1> UnwindDests;
  auto UnwindDest = I.getUnwindDest();
  BranchProbabilityInfo *BPI = FuncInfo.BPI;
  BranchProbability UnwindDestProb =
      (BPI && UnwindDest)
          ? BPI->getEdgeProbability(FuncInfo.MBB->getBasicBlock(), UnwindDest)
          : BranchProbability::getZero();
  findUnwindDestinations(FuncInfo, UnwindDest, UnwindDestProb, UnwindDests);
  for (auto &UnwindDest : UnwindDests) {
    UnwindDest.first->setIsEHPad();
    addSuccessorWithProb(FuncInfo.MBB, UnwindDest.first, UnwindDest.second);
  }
  FuncInfo.MBB->normalizeSuccProbs();

  // Create the terminator node.
  SDValue Ret =
      DAG.getNode(ISD::CLEANUPRET, getCurSDLoc(), MVT::Other, getControlRoot());
  DAG.setRoot(Ret);
}

void SelectionDAGBuilder::visitCatchSwitch(const CatchSwitchInst &CSI) {
  report_fatal_error("visitCatchSwitch not yet implemented!");
}

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

  // Calls to @llvm.experimental.deoptimize don't generate a return value, so
  // lower
  //
  //   %val = call <ty> @llvm.experimental.deoptimize()
  //   ret <ty> %val
  //
  // differently.
  if (I.getParent()->getTerminatingDeoptimizeCall()) {
    LowerDeoptimizingReturn();
    return;
  }

  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, DL,
                    F->getReturnType()->getPointerTo(
                        DAG.getDataLayout().getAllocaAddrSpace()),
                    PtrValueVTs);

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

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

    SmallVector<SDValue, 4> Chains(NumValues);
    for (unsigned i = 0; i != NumValues; ++i) {
      // An aggregate return value cannot wrap around the address space, so
      // offsets to its parts don't wrap either.
      SDValue Ptr = DAG.getObjectPtrOffset(getCurSDLoc(), RetPtr, Offsets[i]);
      Chains[i] = DAG.getStore(
          Chain, getCurSDLoc(), SDValue(RetOp.getNode(), RetOp.getResNo() + i),
          // FIXME: better loc info would be nice.
          Ptr, MachinePointerInfo::getUnknownStack(DAG.getMachineFunction()));
    }

    Chain = DAG.getNode(ISD::TokenFactor, getCurSDLoc(),
                        MVT::Other, Chains);
  } else if (I.getNumOperands() != 0) {
    SmallVector<EVT, 4> ValueVTs;
    ComputeValueVTs(TLI, DL, 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(AttributeList::ReturnIndex,
                                          Attribute::SExt))
        ExtendKind = ISD::SIGN_EXTEND;
      else if (F->getAttributes().hasAttribute(AttributeList::ReturnIndex,
                                               Attribute::ZExt))
        ExtendKind = ISD::ZERO_EXTEND;

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

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

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

        CallingConv::ID CC = F->getCallingConv();

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

  // Push in swifterror virtual register as the last element of Outs. This makes
  // sure swifterror virtual register will be returned in the swifterror
  // physical register.
  const Function *F = I.getParent()->getParent();
  if (TLI.supportSwiftError() &&
      F->getAttributes().hasAttrSomewhere(Attribute::SwiftError)) {
    assert(FuncInfo.SwiftErrorArg && "Need a swift error argument");
    ISD::ArgFlagsTy Flags = ISD::ArgFlagsTy();
    Flags.setSwiftError();
    Outs.push_back(ISD::OutputArg(Flags, EVT(TLI.getPointerTy(DL)) /*vt*/,
                                  EVT(TLI.getPointerTy(DL)) /*argvt*/,
                                  true /*isfixed*/, 1 /*origidx*/,
                                  0 /*partOffs*/));
    // Create SDNode for the swifterror virtual register.
    OutVals.push_back(
        DAG.getRegister(FuncInfo.getOrCreateSwiftErrorVRegUseAt(
                            &I, FuncInfo.MBB, FuncInfo.SwiftErrorArg).first,
                        EVT(TLI.getPointerTy(DL))));
  }

  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.
BranchProbability
SelectionDAGBuilder::getEdgeProbability(const MachineBasicBlock *Src,
                                        const MachineBasicBlock *Dst) const {
  BranchProbabilityInfo *BPI = FuncInfo.BPI;
  const BasicBlock *SrcBB = Src->getBasicBlock();
  const BasicBlock *DstBB = Dst->getBasicBlock();
  if (!BPI) {
    // If BPI is not available, set the default probability as 1 / N, where N is
    // the number of successors.
    auto SuccSize = std::max<uint32_t>(succ_size(SrcBB), 1);
    return BranchProbability(1, SuccSize);
  }
  return BPI->getEdgeProbability(SrcBB, DstBB);
}

void SelectionDAGBuilder::addSuccessorWithProb(MachineBasicBlock *Src,
                                               MachineBasicBlock *Dst,
                                               BranchProbability Prob) {
  if (!FuncInfo.BPI)
    Src->addSuccessorWithoutProb(Dst);
  else {
    if (Prob.isUnknown())
      Prob = getEdgeProbability(Src, Dst);
    Src->addSuccessor(Dst, Prob);
  }
}

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,
                                                  BranchProbability TProb,
                                                  BranchProbability FProb,
                                                  bool InvertCond) {
  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)) {
        ICmpInst::Predicate Pred =
            InvertCond ? IC->getInversePredicate() : IC->getPredicate();
        Condition = getICmpCondCode(Pred);
      } else {
        const FCmpInst *FC = cast<FCmpInst>(Cond);
        FCmpInst::Predicate Pred =
            InvertCond ? FC->getInversePredicate() : FC->getPredicate();
        Condition = getFCmpCondCode(Pred);
        if (TM.Options.NoNaNsFPMath)
          Condition = getFCmpCodeWithoutNaN(Condition);
      }

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

  // Create a CaseBlock record representing this branch.
  ISD::CondCode Opc = InvertCond ? ISD::SETNE : ISD::SETEQ;
  CaseBlock CB(Opc, Cond, ConstantInt::getTrue(*DAG.getContext()),
               nullptr, TBB, FBB, CurBB, getCurSDLoc(), TProb, FProb);
  SwitchCases.push_back(CB);
}

void SelectionDAGBuilder::FindMergedConditions(const Value *Cond,
                                               MachineBasicBlock *TBB,
                                               MachineBasicBlock *FBB,
                                               MachineBasicBlock *CurBB,
                                               MachineBasicBlock *SwitchBB,
                                               Instruction::BinaryOps Opc,
                                               BranchProbability TProb,
                                               BranchProbability FProb,
                                               bool InvertCond) {
  // Skip over not part of the tree and remember to invert op and operands at
  // next level.
  Value *NotCond;
  if (match(Cond, m_OneUse(m_Not(m_Value(NotCond)))) &&
      InBlock(NotCond, CurBB->getBasicBlock())) {
    FindMergedConditions(NotCond, TBB, FBB, CurBB, SwitchBB, Opc, TProb, FProb,
                         !InvertCond);
    return;
  }

  const Instruction *BOp = dyn_cast<Instruction>(Cond);
  // Compute the effective opcode for Cond, taking into account whether it needs
  // to be inverted, e.g.
  //   and (not (or A, B)), C
  // gets lowered as
  //   and (and (not A, not B), C)
  unsigned BOpc = 0;
  if (BOp) {
    BOpc = BOp->getOpcode();
    if (InvertCond) {
      if (BOpc == Instruction::And)
        BOpc = Instruction::Or;
      else if (BOpc == Instruction::Or)
        BOpc = Instruction::And;
    }
  }

  // If this node is not part of the or/and tree, emit it as a branch.
  if (!BOp || !(isa<BinaryOperator>(BOp) || isa<CmpInst>(BOp)) ||
      BOpc != unsigned(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,
                                 TProb, FProb, InvertCond);
    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 original BB.
    // Assuming the original probabilities are A and B, one choice is to set
    // BB1's probabilities to A/2 and A/2+B, and set TmpBB's probabilities to
    // A/(1+B) and 2B/(1+B). 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.

    auto NewTrueProb = TProb / 2;
    auto NewFalseProb = TProb / 2 + FProb;
    // Emit the LHS condition.
    FindMergedConditions(BOp->getOperand(0), TBB, TmpBB, CurBB, SwitchBB, Opc,
                         NewTrueProb, NewFalseProb, InvertCond);

    // Normalize A/2 and B to get A/(1+B) and 2B/(1+B).
    SmallVector<BranchProbability, 2> Probs{TProb / 2, FProb};
    BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
    // Emit the RHS condition into TmpBB.
    FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc,
                         Probs[0], Probs[1], InvertCond);
  } 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 original BB.
    // Assuming the original probabilities are A and B, one choice is to set
    // BB1's probabilities to A+B/2 and B/2, and set TmpBB's probabilities to
    // 2A/(1+A) and B/(1+A). This choice assumes that FalseProb for BB1 ==
    // TrueProb for BB1 * FalseProb for TmpBB.

    auto NewTrueProb = TProb + FProb / 2;
    auto NewFalseProb = FProb / 2;
    // Emit the LHS condition.
    FindMergedConditions(BOp->getOperand(0), TmpBB, FBB, CurBB, SwitchBB, Opc,
                         NewTrueProb, NewFalseProb, InvertCond);

    // Normalize A and B/2 to get 2A/(1+A) and B/(1+A).
    SmallVector<BranchProbability, 2> Probs{TProb, FProb / 2};
    BranchProbability::normalizeProbabilities(Probs.begin(), Probs.end());
    // Emit the RHS condition into TmpBB.
    FindMergedConditions(BOp->getOperand(1), TBB, FBB, TmpBB, SwitchBB, Opc,
                         Probs[0], Probs[1], InvertCond);
  }
}

/// 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)) {
    Instruction::BinaryOps Opcode = BOp->getOpcode();
    if (!DAG.getTargetLoweringInfo().isJumpExpensive() && BOp->hasOneUse() &&
        !I.getMetadata(LLVMContext::MD_unpredictable) &&
        (Opcode == Instruction::And || Opcode == Instruction::Or)) {
      FindMergedConditions(BOp, Succ0MBB, Succ1MBB, BrMBB, BrMBB,
                           Opcode,
                           getEdgeProbability(BrMBB, Succ0MBB),
                           getEdgeProbability(BrMBB, Succ1MBB),
                           /*InvertCond=*/false);
      // 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, getCurSDLoc());

  // 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 = CB.DL;

  // 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, dl, 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, dl, VT),
                          ISD::SETLE);
    } else {
      SDValue SUB = DAG.getNode(ISD::SUB, dl,
                                VT, CmpOp, DAG.getConstant(Low, dl, VT));
      Cond = DAG.getSetCC(dl, MVT::i1, SUB,
                          DAG.getConstant(High-Low, dl, VT), ISD::SETULE);
    }
  }

  // Update successor info
  addSuccessorWithProb(SwitchBB, CB.TrueBB, CB.TrueProb);
  // 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)
    addSuccessorWithProb(SwitchBB, CB.FalseBB, CB.FalseProb);
  SwitchBB->normalizeSuccProbs();

  // 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, dl, 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(DAG.getDataLayout());
  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) {
  SDLoc dl = getCurSDLoc();

  // 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, dl, VT, SwitchOp,
                            DAG.getConstant(JTH.First, dl, 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, dl, TLI.getPointerTy(DAG.getDataLayout()));

  unsigned JumpTableReg =
      FuncInfo.CreateReg(TLI.getPointerTy(DAG.getDataLayout()));
  SDValue CopyTo = DAG.getCopyToReg(getControlRoot(), dl,
                                    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(
      dl, TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(),
                                 Sub.getValueType()),
      Sub, DAG.getConstant(JTH.Last - JTH.First, dl, VT), ISD::SETUGT);

  SDValue BrCond = DAG.getNode(ISD::BRCOND, dl,
                               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, dl, MVT::Other, BrCond,
                         DAG.getBasicBlock(JT.MBB));

  DAG.setRoot(BrCond);
}

/// Create a LOAD_STACK_GUARD node, and let it carry the target specific global
/// variable if there exists one.
static SDValue getLoadStackGuard(SelectionDAG &DAG, const SDLoc &DL,
                                 SDValue &Chain) {
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT PtrTy = TLI.getPointerTy(DAG.getDataLayout());
  MachineFunction &MF = DAG.getMachineFunction();
  Value *Global = TLI.getSDagStackGuard(*MF.getFunction().getParent());
  MachineSDNode *Node =
      DAG.getMachineNode(TargetOpcode::LOAD_STACK_GUARD, DL, PtrTy, Chain);
  if (Global) {
    MachinePointerInfo MPInfo(Global);
    auto Flags = MachineMemOperand::MOLoad | MachineMemOperand::MOInvariant |
                 MachineMemOperand::MODereferenceable;
    MachineMemOperand *MemRef = MF.getMachineMemOperand(
        MPInfo, Flags, PtrTy.getSizeInBits() / 8, DAG.getEVTAlignment(PtrTy));
    DAG.setNodeMemRefs(Node, {MemRef});
  }
  return SDValue(Node, 0);
}

/// 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(DAG.getDataLayout());

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

  SDValue Guard;
  SDLoc dl = getCurSDLoc();
  SDValue StackSlotPtr = DAG.getFrameIndex(FI, PtrTy);
  const Module &M = *ParentBB->getParent()->getFunction().getParent();
  unsigned Align = DL->getPrefTypeAlignment(Type::getInt8PtrTy(M.getContext()));

  // Generate code to load the content of the guard slot.
  SDValue GuardVal = DAG.getLoad(
      PtrTy, dl, DAG.getEntryNode(), StackSlotPtr,
      MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI), Align,
      MachineMemOperand::MOVolatile);

  if (TLI.useStackGuardXorFP())
    GuardVal = TLI.emitStackGuardXorFP(DAG, GuardVal, dl);

  // Retrieve guard check function, nullptr if instrumentation is inlined.
  if (const Value *GuardCheck = TLI.getSSPStackGuardCheck(M)) {
    // The target provides a guard check function to validate the guard value.
    // Generate a call to that function with the content of the guard slot as
    // argument.
    auto *Fn = cast<Function>(GuardCheck);
    FunctionType *FnTy = Fn->getFunctionType();
    assert(FnTy->getNumParams() == 1 && "Invalid function signature");

    TargetLowering::ArgListTy Args;
    TargetLowering::ArgListEntry Entry;
    Entry.Node = GuardVal;
    Entry.Ty = FnTy->getParamType(0);
    if (Fn->hasAttribute(1, Attribute::AttrKind::InReg))
      Entry.IsInReg = true;
    Args.push_back(Entry);

    TargetLowering::CallLoweringInfo CLI(DAG);
    CLI.setDebugLoc(getCurSDLoc())
      .setChain(DAG.getEntryNode())
      .setCallee(Fn->getCallingConv(), FnTy->getReturnType(),
                 getValue(GuardCheck), std::move(Args));

    std::pair<SDValue, SDValue> Result = TLI.LowerCallTo(CLI);
    DAG.setRoot(Result.second);
    return;
  }

  // If useLoadStackGuardNode returns true, generate LOAD_STACK_GUARD.
  // Otherwise, emit a volatile load to retrieve the stack guard value.
  SDValue Chain = DAG.getEntryNode();
  if (TLI.useLoadStackGuardNode()) {
    Guard = getLoadStackGuard(DAG, dl, Chain);
  } else {
    const Value *IRGuard = TLI.getSDagStackGuard(M);
    SDValue GuardPtr = getValue(IRGuard);

    Guard =
        DAG.getLoad(PtrTy, dl, Chain, GuardPtr, MachinePointerInfo(IRGuard, 0),
                    Align, MachineMemOperand::MOVolatile);
  }

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

  SDValue Cmp = DAG.getSetCC(dl, TLI.getSetCCResultType(DAG.getDataLayout(),
                                                        *DAG.getContext(),
                                                        Sub.getValueType()),
                             Sub, DAG.getConstant(0, dl, 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, dl,
                               MVT::Other, GuardVal.getOperand(0),
                               Cmp, DAG.getBasicBlock(SPD.getFailureMBB()));
  // Otherwise branch to success MBB.
  SDValue Br = DAG.getNode(ISD::BR, dl,
                           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,
                      None, 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) {
  SDLoc dl = getCurSDLoc();

  // Subtract the minimum value
  SDValue SwitchOp = getValue(B.SValue);
  EVT VT = SwitchOp.getValueType();
  SDValue Sub = DAG.getNode(ISD::SUB, dl, VT, SwitchOp,
                            DAG.getConstant(B.First, dl, VT));

  // Check range
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue RangeCmp = DAG.getSetCC(
      dl, TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(),
                                 Sub.getValueType()),
      Sub, DAG.getConstant(B.Range, dl, 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(DAG.getDataLayout());
    Sub = DAG.getZExtOrTrunc(Sub, dl, VT);
  }

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

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

  addSuccessorWithProb(SwitchBB, B.Default, B.DefaultProb);
  addSuccessorWithProb(SwitchBB, MBB, B.Prob);
  SwitchBB->normalizeSuccProbs();

  SDValue BrRange = DAG.getNode(ISD::BRCOND, dl,
                                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, dl, MVT::Other, BrRange,
                          DAG.getBasicBlock(MBB));

  DAG.setRoot(BrRange);
}

/// visitBitTestCase - this function produces one "bit test"
void SelectionDAGBuilder::visitBitTestCase(BitTestBlock &BB,
                                           MachineBasicBlock* NextMBB,
                                           BranchProbability BranchProbToNext,
                                           unsigned Reg,
                                           BitTestCase &B,
                                           MachineBasicBlock *SwitchBB) {
  SDLoc dl = getCurSDLoc();
  MVT VT = BB.RegVT;
  SDValue ShiftOp = DAG.getCopyFromReg(getControlRoot(), dl, 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(
        dl, TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT),
        ShiftOp, DAG.getConstant(countTrailingZeros(B.Mask), dl, VT),
        ISD::SETEQ);
  } else if (PopCount == BB.Range) {
    // There is only one zero bit in the range, test for it directly.
    Cmp = DAG.getSetCC(
        dl, TLI.getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT),
        ShiftOp, DAG.getConstant(countTrailingOnes(B.Mask), dl, VT),
        ISD::SETNE);
  } else {
    // Make desired shift
    SDValue SwitchVal = DAG.getNode(ISD::SHL, dl, VT,
                                    DAG.getConstant(1, dl, VT), ShiftOp);

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

  // The branch probability from SwitchBB to B.TargetBB is B.ExtraProb.
  addSuccessorWithProb(SwitchBB, B.TargetBB, B.ExtraProb);
  // The branch probability from SwitchBB to NextMBB is BranchProbToNext.
  addSuccessorWithProb(SwitchBB, NextMBB, BranchProbToNext);
  // It is not guaranteed that the sum of B.ExtraProb and BranchProbToNext is
  // one as they are relative probabilities (and thus work more like weights),
  // and hence we need to normalize them to let the sum of them become one.
  SwitchBB->normalizeSuccProbs();

  SDValue BrAnd = DAG.getNode(ISD::BRCOND, dl,
                              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, dl, MVT::Other, BrAnd,
                        DAG.getBasicBlock(NextMBB));

  DAG.setRoot(BrAnd);
}

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

  // Retrieve successors. Look through artificial IR level blocks like
  // catchswitch for successors.
  MachineBasicBlock *Return = FuncInfo.MBBMap[I.getSuccessor(0)];
  const BasicBlock *EHPadBB = I.getSuccessor(1);

  // Deopt bundles are lowered in LowerCallSiteWithDeoptBundle, and we don't
  // have to do anything here to lower funclet bundles.
  assert(!I.hasOperandBundlesOtherThan(
             {LLVMContext::OB_deopt, LLVMContext::OB_funclet}) &&
         "Cannot lower invokes with arbitrary operand bundles yet!");

  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, EHPadBB);
      break;
    case Intrinsic::experimental_gc_statepoint:
      LowerStatepoint(ImmutableStatepoint(&I), EHPadBB);
      break;
    }
  } else if (I.countOperandBundlesOfType(LLVMContext::OB_deopt)) {
    // Currently we do not lower any intrinsic calls with deopt operand bundles.
    // Eventually we will support lowering the @llvm.experimental.deoptimize
    // intrinsic, and right now there are no plans to support other intrinsics
    // with deopt state.
    LowerCallSiteWithDeoptBundle(&I, getValue(Callee), EHPadBB);
  } else {
    LowerCallTo(&I, getValue(Callee), false, EHPadBB);
  }

  // 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);
  }

  SmallVector<std::pair<MachineBasicBlock *, BranchProbability>, 1> UnwindDests;
  BranchProbabilityInfo *BPI = FuncInfo.BPI;
  BranchProbability EHPadBBProb =
      BPI ? BPI->getEdgeProbability(InvokeMBB->getBasicBlock(), EHPadBB)
          : BranchProbability::getZero();
  findUnwindDestinations(FuncInfo, EHPadBB, EHPadBBProb, UnwindDests);

  // Update successor info.
  addSuccessorWithProb(InvokeMBB, Return);
  for (auto &UnwindDest : UnwindDests) {
    UnwindDest.first->setIsEHPad();
    addSuccessorWithProb(InvokeMBB, UnwindDest.first, UnwindDest.second);
  }
  InvokeMBB->normalizeSuccProbs();

  // 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->isEHPad() &&
         "Call to landingpad not in landing pad!");

  // 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();
  const Constant *PersonalityFn = FuncInfo.Fn->getPersonalityFn();
  if (TLI.getExceptionPointerRegister(PersonalityFn) == 0 &&
      TLI.getExceptionSelectorRegister(PersonalityFn) == 0)
    return;

  // If landingpad's return type is token type, we don't create DAG nodes
  // for its exception pointer and selector value. The extraction of exception
  // pointer or selector value from token type landingpads is not currently
  // supported.
  if (LP.getType()->isTokenTy())
    return;

  SmallVector<EVT, 2> ValueVTs;
  SDLoc dl = getCurSDLoc();
  ComputeValueVTs(TLI, DAG.getDataLayout(), 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(), dl,
                           FuncInfo.ExceptionPointerVirtReg,
                           TLI.getPointerTy(DAG.getDataLayout())),
        dl, ValueVTs[0]);
  } else {
    Ops[0] = DAG.getConstant(0, dl, TLI.getPointerTy(DAG.getDataLayout()));
  }
  Ops[1] = DAG.getZExtOrTrunc(
      DAG.getCopyFromReg(DAG.getEntryNode(), dl,
                         FuncInfo.ExceptionSelectorVirtReg,
                         TLI.getPointerTy(DAG.getDataLayout())),
      dl, ValueVTs[1]);

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

void SelectionDAGBuilder::sortAndRangeify(CaseClusterVector &Clusters) {
#ifndef NDEBUG
  for (const CaseCluster &CC : Clusters)
    assert(CC.Low == CC.High && "Input clusters must be single-case");
#endif

  llvm::sort(Clusters, [](const CaseCluster &a, const CaseCluster &b) {
    return a.Low->getValue().slt(b.Low->getValue());
  });

  // Merge adjacent clusters with the same destination.
  const unsigned N = Clusters.size();
  unsigned DstIndex = 0;
  for (unsigned SrcIndex = 0; SrcIndex < N; ++SrcIndex) {
    CaseCluster &CC = Clusters[SrcIndex];
    const ConstantInt *CaseVal = CC.Low;
    MachineBasicBlock *Succ = CC.MBB;

    if (DstIndex != 0 && Clusters[DstIndex - 1].MBB == Succ &&
        (CaseVal->getValue() - Clusters[DstIndex - 1].High->getValue()) == 1) {
      // If this case has the same successor and is a neighbour, merge it into
      // the previous cluster.
      Clusters[DstIndex - 1].High = CaseVal;
      Clusters[DstIndex - 1].Prob += CC.Prob;
    } else {
      std::memmove(&Clusters[DstIndex++], &Clusters[SrcIndex],
                   sizeof(Clusters[SrcIndex]));
    }
  }
  Clusters.resize(DstIndex);
}

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::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];
    addSuccessorWithProb(IndirectBrMBB, Succ);
  }
  IndirectBrMBB->normalizeSuccProbs();

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

void SelectionDAGBuilder::visitUnreachable(const UnreachableInst &I) {
  if (!DAG.getTarget().Options.TrapUnreachable)
    return;

  // We may be able to ignore unreachable behind a noreturn call.
  if (DAG.getTarget().Options.NoTrapAfterNoreturn) {
    const BasicBlock &BB = *I.getParent();
    if (&I != &BB.front()) {
      BasicBlock::const_iterator PredI =
        std::prev(BasicBlock::const_iterator(&I));
      if (const CallInst *Call = dyn_cast<CallInst>(&*PredI)) {
        if (Call->doesNotReturn())
          return;
      }
    }
  }

  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);
}

/// Checks if the given instruction performs a vector reduction, in which case
/// we have the freedom to alter the elements in the result as long as the
/// reduction of them stays unchanged.
static bool isVectorReductionOp(const User *I) {
  const Instruction *Inst = dyn_cast<Instruction>(I);
  if (!Inst || !Inst->getType()->isVectorTy())
    return false;

  auto OpCode = Inst->getOpcode();
  switch (OpCode) {
  case Instruction::Add:
  case Instruction::Mul:
  case Instruction::And:
  case Instruction::Or:
  case Instruction::Xor:
    break;
  case Instruction::FAdd:
  case Instruction::FMul:
    if (const FPMathOperator *FPOp = dyn_cast<const FPMathOperator>(Inst))
      if (FPOp->getFastMathFlags().isFast())
        break;
    LLVM_FALLTHROUGH;
  default:
    return false;
  }

  unsigned ElemNum = Inst->getType()->getVectorNumElements();
  // Ensure the reduction size is a power of 2.
  if (!isPowerOf2_32(ElemNum))
    return false;

  unsigned ElemNumToReduce = ElemNum;

  // Do DFS search on the def-use chain from the given instruction. We only
  // allow four kinds of operations during the search until we reach the
  // instruction that extracts the first element from the vector:
  //
  //   1. The reduction operation of the same opcode as the given instruction.
  //
  //   2. PHI node.
  //
  //   3. ShuffleVector instruction together with a reduction operation that
  //      does a partial reduction.
  //
  //   4. ExtractElement that extracts the first element from the vector, and we
  //      stop searching the def-use chain here.
  //
  // 3 & 4 above perform a reduction on all elements of the vector. We push defs
  // from 1-3 to the stack to continue the DFS. The given instruction is not
  // a reduction operation if we meet any other instructions other than those
  // listed above.

  SmallVector<const User *, 16> UsersToVisit{Inst};
  SmallPtrSet<const User *, 16> Visited;
  bool ReduxExtracted = false;

  while (!UsersToVisit.empty()) {
    auto User = UsersToVisit.back();
    UsersToVisit.pop_back();
    if (!Visited.insert(User).second)
      continue;

    for (const auto &U : User->users()) {
      auto Inst = dyn_cast<Instruction>(U);
      if (!Inst)
        return false;

      if (Inst->getOpcode() == OpCode || isa<PHINode>(U)) {
        if (const FPMathOperator *FPOp = dyn_cast<const FPMathOperator>(Inst))
          if (!isa<PHINode>(FPOp) && !FPOp->getFastMathFlags().isFast())
            return false;
        UsersToVisit.push_back(U);
      } else if (const ShuffleVectorInst *ShufInst =
                     dyn_cast<ShuffleVectorInst>(U)) {
        // Detect the following pattern: A ShuffleVector instruction together
        // with a reduction that do partial reduction on the first and second
        // ElemNumToReduce / 2 elements, and store the result in
        // ElemNumToReduce / 2 elements in another vector.

        unsigned ResultElements = ShufInst->getType()->getVectorNumElements();
        if (ResultElements < ElemNum)
          return false;

        if (ElemNumToReduce == 1)
          return false;
        if (!isa<UndefValue>(U->getOperand(1)))
          return false;
        for (unsigned i = 0; i < ElemNumToReduce / 2; ++i)
          if (ShufInst->getMaskValue(i) != int(i + ElemNumToReduce / 2))
            return false;
        for (unsigned i = ElemNumToReduce / 2; i < ElemNum; ++i)
          if (ShufInst->getMaskValue(i) != -1)
            return false;

        // There is only one user of this ShuffleVector instruction, which
        // must be a reduction operation.
        if (!U->hasOneUse())
          return false;

        auto U2 = dyn_cast<Instruction>(*U->user_begin());
        if (!U2 || U2->getOpcode() != OpCode)
          return false;

        // Check operands of the reduction operation.
        if ((U2->getOperand(0) == U->getOperand(0) && U2->getOperand(1) == U) ||
            (U2->getOperand(1) == U->getOperand(0) && U2->getOperand(0) == U)) {
          UsersToVisit.push_back(U2);
          ElemNumToReduce /= 2;
        } else
          return false;
      } else if (isa<ExtractElementInst>(U)) {
        // At this moment we should have reduced all elements in the vector.
        if (ElemNumToReduce != 1)
          return false;

        const ConstantInt *Val = dyn_cast<ConstantInt>(U->getOperand(1));
        if (!Val || !Val->isZero())
          return false;

        ReduxExtracted = true;
      } else
        return false;
    }
  }
  return ReduxExtracted;
}

void SelectionDAGBuilder::visitUnary(const User &I, unsigned Opcode) {
  SDNodeFlags Flags;

  SDValue Op = getValue(I.getOperand(0));
  SDValue UnNodeValue = DAG.getNode(Opcode, getCurSDLoc(), Op.getValueType(),
                                    Op, Flags);
  setValue(&I, UnNodeValue);
}

void SelectionDAGBuilder::visitBinary(const User &I, unsigned Opcode) {
  SDNodeFlags Flags;
  if (auto *OFBinOp = dyn_cast<OverflowingBinaryOperator>(&I)) {
    Flags.setNoSignedWrap(OFBinOp->hasNoSignedWrap());
    Flags.setNoUnsignedWrap(OFBinOp->hasNoUnsignedWrap());
  }
  if (auto *ExactOp = dyn_cast<PossiblyExactOperator>(&I)) {
    Flags.setExact(ExactOp->isExact());
  }
  if (isVectorReductionOp(&I)) {
    Flags.setVectorReduction(true);
    LLVM_DEBUG(dbgs() << "Detected a reduction operation:" << I << "\n");
  }

  SDValue Op1 = getValue(I.getOperand(0));
  SDValue Op2 = getValue(I.getOperand(1));
  SDValue BinNodeValue = DAG.getNode(Opcode, getCurSDLoc(), Op1.getValueType(),
                                     Op1, Op2, Flags);
  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(
      Op1.getValueType(), DAG.getDataLayout());

  // 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.getValueSizeInBits();
    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.getValueSizeInBits()))
      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();
  }
  SDNodeFlags Flags;
  Flags.setExact(exact);
  Flags.setNoSignedWrap(nsw);
  Flags.setNoUnsignedWrap(nuw);
  SDValue Res = DAG.getNode(Opcode, getCurSDLoc(), Op1.getValueType(), Op1, Op2,
                            Flags);
  setValue(&I, Res);
}

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

  SDNodeFlags Flags;
  Flags.setExact(isa<PossiblyExactOperator>(&I) &&
                 cast<PossiblyExactOperator>(&I)->isExact());
  setValue(&I, DAG.getNode(ISD::SDIV, getCurSDLoc(), Op1.getValueType(), Op1,
                           Op2, Flags));
}

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(DAG.getDataLayout(),
                                                        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);
  auto *FPMO = dyn_cast<FPMathOperator>(&I);
  if ((FPMO && FPMO->hasNoNaNs()) || TM.Options.NoNaNsFPMath)
    Condition = getFCmpCodeWithoutNaN(Condition);

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

// Check if the condition of the select has one use or two users that are both
// selects with the same condition.
static bool hasOnlySelectUsers(const Value *Cond) {
  return llvm::all_of(Cond->users(), [](const Value *V) {
    return isa<SelectInst>(V);
  });
}

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

  SmallVector<SDValue, 4> Values(NumValues);
  SDValue Cond     = getValue(I.getOperand(0));
  SDValue LHSVal   = getValue(I.getOperand(1));
  SDValue RHSVal   = getValue(I.getOperand(2));
  auto BaseOps = {Cond};
  ISD::NodeType OpCode = Cond.getValueType().isVector() ?
    ISD::VSELECT : ISD::SELECT;

  // Min/max matching is only viable if all output VTs are the same.
  if (is_splat(ValueVTs)) {
    EVT VT = ValueVTs[0];
    LLVMContext &Ctx = *DAG.getContext();
    auto &TLI = DAG.getTargetLoweringInfo();

    // We care about the legality of the operation after it has been type
    // legalized.
    while (TLI.getTypeAction(Ctx, VT) != TargetLoweringBase::TypeLegal &&
           VT != TLI.getTypeToTransformTo(Ctx, VT))
      VT = TLI.getTypeToTransformTo(Ctx, VT);

    // If the vselect is legal, assume we want to leave this as a vector setcc +
    // vselect. Otherwise, if this is going to be scalarized, we want to see if
    // min/max is legal on the scalar type.
    bool UseScalarMinMax = VT.isVector() &&
      !TLI.isOperationLegalOrCustom(ISD::VSELECT, VT);

    Value *LHS, *RHS;
    auto SPR = matchSelectPattern(const_cast<User*>(&I), LHS, RHS);
    ISD::NodeType Opc = ISD::DELETED_NODE;
    switch (SPR.Flavor) {
    case SPF_UMAX:    Opc = ISD::UMAX; break;
    case SPF_UMIN:    Opc = ISD::UMIN; break;
    case SPF_SMAX:    Opc = ISD::SMAX; break;
    case SPF_SMIN:    Opc = ISD::SMIN; break;
    case SPF_FMINNUM:
      switch (SPR.NaNBehavior) {
      case SPNB_NA: llvm_unreachable("No NaN behavior for FP op?");
      case SPNB_RETURNS_NAN:   Opc = ISD::FMINIMUM; break;
      case SPNB_RETURNS_OTHER: Opc = ISD::FMINNUM; break;
      case SPNB_RETURNS_ANY: {
        if (TLI.isOperationLegalOrCustom(ISD::FMINNUM, VT))
          Opc = ISD::FMINNUM;
        else if (TLI.isOperationLegalOrCustom(ISD::FMINIMUM, VT))
          Opc = ISD::FMINIMUM;
        else if (UseScalarMinMax)
          Opc = TLI.isOperationLegalOrCustom(ISD::FMINNUM, VT.getScalarType()) ?
            ISD::FMINNUM : ISD::FMINIMUM;
        break;
      }
      }
      break;
    case SPF_FMAXNUM:
      switch (SPR.NaNBehavior) {
      case SPNB_NA: llvm_unreachable("No NaN behavior for FP op?");
      case SPNB_RETURNS_NAN:   Opc = ISD::FMAXIMUM; break;
      case SPNB_RETURNS_OTHER: Opc = ISD::FMAXNUM; break;
      case SPNB_RETURNS_ANY:

        if (TLI.isOperationLegalOrCustom(ISD::FMAXNUM, VT))
          Opc = ISD::FMAXNUM;
        else if (TLI.isOperationLegalOrCustom(ISD::FMAXIMUM, VT))
          Opc = ISD::FMAXIMUM;
        else if (UseScalarMinMax)
          Opc = TLI.isOperationLegalOrCustom(ISD::FMAXNUM, VT.getScalarType()) ?
            ISD::FMAXNUM : ISD::FMAXIMUM;
        break;
      }
      break;
    default: break;
    }

    if (Opc != ISD::DELETED_NODE &&
        (TLI.isOperationLegalOrCustom(Opc, VT) ||
         (UseScalarMinMax &&
          TLI.isOperationLegalOrCustom(Opc, VT.getScalarType()))) &&
        // If the underlying comparison instruction is used by any other
        // instruction, the consumed instructions won't be destroyed, so it is
        // not profitable to convert to a min/max.
        hasOnlySelectUsers(cast<SelectInst>(I).getCondition())) {
      OpCode = Opc;
      LHSVal = getValue(LHS);
      RHSVal = getValue(RHS);
      BaseOps = {};
    }
  }

  for (unsigned i = 0; i != NumValues; ++i) {
    SmallVector<SDValue, 3> Ops(BaseOps.begin(), BaseOps.end());
    Ops.push_back(SDValue(LHSVal.getNode(), LHSVal.getResNo() + i));
    Ops.push_back(SDValue(RHSVal.getNode(), RHSVal.getResNo() + i));
    Values[i] = DAG.getNode(OpCode, getCurSDLoc(),
                            LHSVal.getNode()->getValueType(LHSVal.getResNo()+i),
                            Ops);
  }

  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(DAG.getDataLayout(),
                                                        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(DAG.getDataLayout(),
                                                        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(DAG.getDataLayout(),
                                                        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));
  SDLoc dl = getCurSDLoc();
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT DestVT = TLI.getValueType(DAG.getDataLayout(), I.getType());
  setValue(&I, DAG.getNode(ISD::FP_ROUND, dl, DestVT, N,
                           DAG.getTargetConstant(
                               0, dl, TLI.getPointerTy(DAG.getDataLayout()))));
}

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(DAG.getDataLayout(),
                                                        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(DAG.getDataLayout(),
                                                        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(DAG.getDataLayout(),
                                                        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(DAG.getDataLayout(),
                                                        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(DAG.getDataLayout(),
                                                        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(DAG.getDataLayout(),
                                                        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(DAG.getDataLayout(),
                                                        I.getType());
  setValue(&I, DAG.getZExtOrTrunc(N, getCurSDLoc(), DestVT));
}

void SelectionDAGBuilder::visitBitCast(const User &I) {
  SDValue N = getValue(I.getOperand(0));
  SDLoc dl = getCurSDLoc();
  EVT DestVT = DAG.getTargetLoweringInfo().getValueType(DAG.getDataLayout(),
                                                        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, dl,
                             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 recognize 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(), dl, 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(DAG.getDataLayout(), 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(DAG.getDataLayout()));
  setValue(&I, DAG.getNode(ISD::INSERT_VECTOR_ELT, getCurSDLoc(),
                           TLI.getValueType(DAG.getDataLayout(), 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(DAG.getDataLayout()));
  setValue(&I, DAG.getNode(ISD::EXTRACT_VECTOR_ELT, getCurSDLoc(),
                           TLI.getValueType(DAG.getDataLayout(), I.getType()),
                           InVec, InIdx));
}

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

  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(DAG.getDataLayout(), I.getType());
  EVT SrcVT = Src1.getValueType();
  unsigned SrcNumElts = SrcVT.getVectorNumElements();

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

  // Normalize the shuffle vector since mask and vector length don't match.
  if (SrcNumElts < MaskNumElts) {
    // Mask is longer than the source vectors. We can use concatenate vector to
    // make the mask and vectors lengths match.

    if (MaskNumElts % SrcNumElts == 0) {
      // Mask length is a multiple of the source vector length.
      // Check if the shuffle is some kind of concatenation of the input
      // vectors.
      unsigned NumConcat = MaskNumElts / SrcNumElts;
      bool IsConcat = true;
      SmallVector<int, 8> ConcatSrcs(NumConcat, -1);
      for (unsigned i = 0; i != MaskNumElts; ++i) {
        int Idx = Mask[i];
        if (Idx < 0)
          continue;
        // Ensure the indices in each SrcVT sized piece are sequential and that
        // the same source is used for the whole piece.
        if ((Idx % SrcNumElts != (i % SrcNumElts)) ||
            (ConcatSrcs[i / SrcNumElts] >= 0 &&
             ConcatSrcs[i / SrcNumElts] != (int)(Idx / SrcNumElts))) {
          IsConcat = false;
          break;
        }
        // Remember which source this index came from.
        ConcatSrcs[i / SrcNumElts] = Idx / SrcNumElts;
      }

      // The shuffle is concatenating multiple vectors together. Just emit
      // a CONCAT_VECTORS operation.
      if (IsConcat) {
        SmallVector<SDValue, 8> ConcatOps;
        for (auto Src : ConcatSrcs) {
          if (Src < 0)
            ConcatOps.push_back(DAG.getUNDEF(SrcVT));
          else if (Src == 0)
            ConcatOps.push_back(Src1);
          else
            ConcatOps.push_back(Src2);
        }
        setValue(&I, DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, ConcatOps));
        return;
      }
    }

    unsigned PaddedMaskNumElts = alignTo(MaskNumElts, SrcNumElts);
    unsigned NumConcat = PaddedMaskNumElts / SrcNumElts;
    EVT PaddedVT = EVT::getVectorVT(*DAG.getContext(), VT.getScalarType(),
                                    PaddedMaskNumElts);

    // Pad both vectors with undefs to make them the same length as the mask.
    SDValue UndefVal = DAG.getUNDEF(SrcVT);

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

    Src1 = Src1.isUndef()
               ? DAG.getUNDEF(PaddedVT)
               : DAG.getNode(ISD::CONCAT_VECTORS, DL, PaddedVT, MOps1);
    Src2 = Src2.isUndef()
               ? DAG.getUNDEF(PaddedVT)
               : DAG.getNode(ISD::CONCAT_VECTORS, DL, PaddedVT, MOps2);

    // Readjust mask for new input vector length.
    SmallVector<int, 8> MappedOps(PaddedMaskNumElts, -1);
    for (unsigned i = 0; i != MaskNumElts; ++i) {
      int Idx = Mask[i];
      if (Idx >= (int)SrcNumElts)
        Idx -= SrcNumElts - PaddedMaskNumElts;
      MappedOps[i] = Idx;
    }

    SDValue Result = DAG.getVectorShuffle(PaddedVT, DL, Src1, Src2, MappedOps);

    // If the concatenated vector was padded, extract a subvector with the
    // correct number of elements.
    if (MaskNumElts != PaddedMaskNumElts)
      Result = DAG.getNode(
          ISD::EXTRACT_SUBVECTOR, DL, VT, Result,
          DAG.getConstant(0, DL, TLI.getVectorIdxTy(DAG.getDataLayout())));

    setValue(&I, Result);
    return;
  }

  if (SrcNumElts > MaskNumElts) {
    // Analyze the access pattern of the vector to see if we can extract
    // two subvectors and do the shuffle.
    int StartIdx[2] = { -1, -1 };  // StartIdx to extract from
    bool CanExtract = true;
    for (int Idx : Mask) {
      unsigned Input = 0;
      if (Idx < 0)
        continue;

      if (Idx >= (int)SrcNumElts) {
        Input = 1;
        Idx -= SrcNumElts;
      }

      // If all the indices come from the same MaskNumElts sized portion of
      // the sources we can use extract. Also make sure the extract wouldn't
      // extract past the end of the source.
      int NewStartIdx = alignDown(Idx, MaskNumElts);
      if (NewStartIdx + MaskNumElts > SrcNumElts ||
          (StartIdx[Input] >= 0 && StartIdx[Input] != NewStartIdx))
        CanExtract = false;
      // Make sure we always update StartIdx as we use it to track if all
      // elements are undef.
      StartIdx[Input] = NewStartIdx;
    }

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

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

      setValue(&I, DAG.getVectorShuffle(VT, DL, Src1, Src2, MappedOps));
      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(DAG.getDataLayout());
  SmallVector<SDValue,8> Ops;
  for (int Idx : Mask) {
    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, DL,
                        EltVT, Src, DAG.getConstant(Idx, DL, IdxVT));
    }

    Ops.push_back(Res);
  }

  setValue(&I, DAG.getBuildVector(VT, DL, Ops));
}

void SelectionDAGBuilder::visitInsertValue(const User &I) {
  ArrayRef<unsigned> Indices;
  if (const InsertValueInst *IV = dyn_cast<InsertValueInst>(&I))
    Indices = IV->getIndices();
  else
    Indices = cast<ConstantExpr>(&I)->getIndices();

  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, Indices);

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SmallVector<EVT, 4> AggValueVTs;
  ComputeValueVTs(TLI, DAG.getDataLayout(), AggTy, AggValueVTs);
  SmallVector<EVT, 4> ValValueVTs;
  ComputeValueVTs(TLI, DAG.getDataLayout(), 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 User &I) {
  ArrayRef<unsigned> Indices;
  if (const ExtractValueInst *EV = dyn_cast<ExtractValueInst>(&I))
    Indices = EV->getIndices();
  else
    Indices = cast<ConstantExpr>(&I)->getIndices();

  const Value *Op0 = I.getOperand(0);
  Type *AggTy = Op0->getType();
  Type *ValTy = I.getType();
  bool OutOfUndef = isa<UndefValue>(Op0);

  unsigned LinearIndex = ComputeLinearIndex(AggTy, Indices);

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SmallVector<EVT, 4> ValValueVTs;
  ComputeValueVTs(TLI, DAG.getDataLayout(), 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.
  unsigned AS = Op0->getType()->getScalarType()->getPointerAddressSpace();
  SDValue N = getValue(Op0);
  SDLoc dl = getCurSDLoc();

  // Normalize Vector GEP - all scalar operands should be converted to the
  // splat vector.
  unsigned VectorWidth = I.getType()->isVectorTy() ?
    cast<VectorType>(I.getType())->getVectorNumElements() : 0;

  if (VectorWidth && !N.getValueType().isVector()) {
    LLVMContext &Context = *DAG.getContext();
    EVT VT = EVT::getVectorVT(Context, N.getValueType(), VectorWidth);
    N = DAG.getSplatBuildVector(VT, dl, N);
  }

  for (gep_type_iterator GTI = gep_type_begin(&I), E = gep_type_end(&I);
       GTI != E; ++GTI) {
    const Value *Idx = GTI.getOperand();
    if (StructType *StTy = GTI.getStructTypeOrNull()) {
      unsigned Field = cast<Constant>(Idx)->getUniqueInteger().getZExtValue();
      if (Field) {
        // N = N + Offset
        uint64_t Offset = DL->getStructLayout(StTy)->getElementOffset(Field);

        // In an inbounds GEP with an offset that is nonnegative even when
        // interpreted as signed, assume there is no unsigned overflow.
        SDNodeFlags Flags;
        if (int64_t(Offset) >= 0 && cast<GEPOperator>(I).isInBounds())
          Flags.setNoUnsignedWrap(true);

        N = DAG.getNode(ISD::ADD, dl, N.getValueType(), N,
                        DAG.getConstant(Offset, dl, N.getValueType()), Flags);
      }
    } else {
      unsigned IdxSize = DAG.getDataLayout().getIndexSizeInBits(AS);
      MVT IdxTy = MVT::getIntegerVT(IdxSize);
      APInt ElementSize(IdxSize, DL->getTypeAllocSize(GTI.getIndexedType()));

      // If this is a scalar constant or a splat vector of constants,
      // handle it quickly.
      const auto *CI = dyn_cast<ConstantInt>(Idx);
      if (!CI && isa<ConstantDataVector>(Idx) &&
          cast<ConstantDataVector>(Idx)->getSplatValue())
        CI = cast<ConstantInt>(cast<ConstantDataVector>(Idx)->getSplatValue());

      if (CI) {
        if (CI->isZero())
          continue;
        APInt Offs = ElementSize * CI->getValue().sextOrTrunc(IdxSize);
        LLVMContext &Context = *DAG.getContext();
        SDValue OffsVal = VectorWidth ?
          DAG.getConstant(Offs, dl, EVT::getVectorVT(Context, IdxTy, VectorWidth)) :
          DAG.getConstant(Offs, dl, IdxTy);

        // In an inbouds GEP with an offset that is nonnegative even when
        // interpreted as signed, assume there is no unsigned overflow.
        SDNodeFlags Flags;
        if (Offs.isNonNegative() && cast<GEPOperator>(I).isInBounds())
          Flags.setNoUnsignedWrap(true);

        N = DAG.getNode(ISD::ADD, dl, N.getValueType(), N, OffsVal, Flags);
        continue;
      }

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

      if (!IdxN.getValueType().isVector() && VectorWidth) {
        EVT VT = EVT::getVectorVT(*Context, IdxN.getValueType(), VectorWidth);
        IdxN = DAG.getSplatBuildVector(VT, dl, IdxN);
      }

      // If the index is smaller or larger than intptr_t, truncate or extend
      // it.
      IdxN = DAG.getSExtOrTrunc(IdxN, dl, 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, dl,
                             N.getValueType(), IdxN,
                             DAG.getConstant(Amt, dl, IdxN.getValueType()));
        } else {
          SDValue Scale = DAG.getConstant(ElementSize, dl, IdxN.getValueType());
          IdxN = DAG.getNode(ISD::MUL, dl,
                             N.getValueType(), IdxN, Scale);
        }
      }

      N = DAG.getNode(ISD::ADD, dl,
                      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.

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

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

  EVT IntPtr = TLI.getPointerTy(DAG.getDataLayout(), DL.getAllocaAddrSpace());
  if (AllocSize.getValueType() != IntPtr)
    AllocSize = DAG.getZExtOrTrunc(AllocSize, dl, IntPtr);

  AllocSize = DAG.getNode(ISD::MUL, dl, IntPtr,
                          AllocSize,
                          DAG.getConstant(TySize, dl, 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. This doesn't overflow because we're computing
  // an address inside an alloca.
  SDNodeFlags Flags;
  Flags.setNoUnsignedWrap(true);
  AllocSize = DAG.getNode(ISD::ADD, dl, AllocSize.getValueType(), AllocSize,
                          DAG.getConstant(StackAlign - 1, dl, IntPtr), Flags);

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

  SDValue Ops[] = {getRoot(), AllocSize, DAG.getConstant(Align, dl, IntPtr)};
  SDVTList VTs = DAG.getVTList(AllocSize.getValueType(), MVT::Other);
  SDValue DSA = DAG.getNode(ISD::DYNAMIC_STACKALLOC, dl, 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 TargetLowering &TLI = DAG.getTargetLoweringInfo();
  const Value *SV = I.getOperand(0);
  if (TLI.supportSwiftError()) {
    // Swifterror values can come from either a function parameter with
    // swifterror attribute or an alloca with swifterror attribute.
    if (const Argument *Arg = dyn_cast<Argument>(SV)) {
      if (Arg->hasSwiftErrorAttr())
        return visitLoadFromSwiftError(I);
    }

    if (const AllocaInst *Alloca = dyn_cast<AllocaInst>(SV)) {
      if (Alloca->isSwiftError())
        return visitLoadFromSwiftError(I);
    }
  }

  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;
  bool isDereferenceable = isDereferenceablePointer(SV, DAG.getDataLayout());
  unsigned Alignment = I.getAlignment();

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

  SmallVector<EVT, 4> ValueVTs;
  SmallVector<uint64_t, 4> Offsets;
  ComputeValueVTs(TLI, DAG.getDataLayout(), 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 && AA->pointsToConstantMemory(MemoryLocation(
               SV, DAG.getDataLayout().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();
  }

  SDLoc dl = getCurSDLoc();

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

  // An aggregate load cannot wrap around the address space, so offsets to its
  // parts don't wrap either.
  SDNodeFlags Flags;
  Flags.setNoUnsignedWrap(true);

  SmallVector<SDValue, 4> Values(NumValues);
  SmallVector<SDValue, 4> Chains(std::min(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, dl, MVT::Other,
                                  makeArrayRef(Chains.data(), ChainI));
      Root = Chain;
      ChainI = 0;
    }
    SDValue A = DAG.getNode(ISD::ADD, dl,
                            PtrVT, Ptr,
                            DAG.getConstant(Offsets[i], dl, PtrVT),
                            Flags);
    auto MMOFlags = MachineMemOperand::MONone;
    if (isVolatile)
      MMOFlags |= MachineMemOperand::MOVolatile;
    if (isNonTemporal)
      MMOFlags |= MachineMemOperand::MONonTemporal;
    if (isInvariant)
      MMOFlags |= MachineMemOperand::MOInvariant;
    if (isDereferenceable)
      MMOFlags |= MachineMemOperand::MODereferenceable;
    MMOFlags |= TLI.getMMOFlags(I);

    SDValue L = DAG.getLoad(ValueVTs[i], dl, Root, A,
                            MachinePointerInfo(SV, Offsets[i]), Alignment,
                            MMOFlags, AAInfo, Ranges);

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

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

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

void SelectionDAGBuilder::visitStoreToSwiftError(const StoreInst &I) {
  assert(DAG.getTargetLoweringInfo().supportSwiftError() &&
         "call visitStoreToSwiftError when backend supports swifterror");

  SmallVector<EVT, 4> ValueVTs;
  SmallVector<uint64_t, 4> Offsets;
  const Value *SrcV = I.getOperand(0);
  ComputeValueVTs(DAG.getTargetLoweringInfo(), DAG.getDataLayout(),
                  SrcV->getType(), ValueVTs, &Offsets);
  assert(ValueVTs.size() == 1 && Offsets[0] == 0 &&
         "expect a single EVT for swifterror");

  SDValue Src = getValue(SrcV);
  // Create a virtual register, then update the virtual register.
  unsigned VReg; bool CreatedVReg;
  std::tie(VReg, CreatedVReg) = FuncInfo.getOrCreateSwiftErrorVRegDefAt(&I);
  // Chain, DL, Reg, N or Chain, DL, Reg, N, Glue
  // Chain can be getRoot or getControlRoot.
  SDValue CopyNode = DAG.getCopyToReg(getRoot(), getCurSDLoc(), VReg,
                                      SDValue(Src.getNode(), Src.getResNo()));
  DAG.setRoot(CopyNode);
  if (CreatedVReg)
    FuncInfo.setCurrentSwiftErrorVReg(FuncInfo.MBB, I.getOperand(1), VReg);
}

void SelectionDAGBuilder::visitLoadFromSwiftError(const LoadInst &I) {
  assert(DAG.getTargetLoweringInfo().supportSwiftError() &&
         "call visitLoadFromSwiftError when backend supports swifterror");

  assert(!I.isVolatile() &&
         I.getMetadata(LLVMContext::MD_nontemporal) == nullptr &&
         I.getMetadata(LLVMContext::MD_invariant_load) == nullptr &&
         "Support volatile, non temporal, invariant for load_from_swift_error");

  const Value *SV = I.getOperand(0);
  Type *Ty = I.getType();
  AAMDNodes AAInfo;
  I.getAAMetadata(AAInfo);
  assert((!AA || !AA->pointsToConstantMemory(MemoryLocation(
             SV, DAG.getDataLayout().getTypeStoreSize(Ty), AAInfo))) &&
         "load_from_swift_error should not be constant memory");

  SmallVector<EVT, 4> ValueVTs;
  SmallVector<uint64_t, 4> Offsets;
  ComputeValueVTs(DAG.getTargetLoweringInfo(), DAG.getDataLayout(), Ty,
                  ValueVTs, &Offsets);
  assert(ValueVTs.size() == 1 && Offsets[0] == 0 &&
         "expect a single EVT for swifterror");

  // Chain, DL, Reg, VT, Glue or Chain, DL, Reg, VT
  SDValue L = DAG.getCopyFromReg(
      getRoot(), getCurSDLoc(),
      FuncInfo.getOrCreateSwiftErrorVRegUseAt(&I, FuncInfo.MBB, SV).first,
      ValueVTs[0]);

  setValue(&I, L);
}

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

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

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  if (TLI.supportSwiftError()) {
    // Swifterror values can come from either a function parameter with
    // swifterror attribute or an alloca with swifterror attribute.
    if (const Argument *Arg = dyn_cast<Argument>(PtrV)) {
      if (Arg->hasSwiftErrorAttr())
        return visitStoreToSwiftError(I);
    }

    if (const AllocaInst *Alloca = dyn_cast<AllocaInst>(PtrV)) {
      if (Alloca->isSwiftError())
        return visitStoreToSwiftError(I);
    }
  }

  SmallVector<EVT, 4> ValueVTs;
  SmallVector<uint64_t, 4> Offsets;
  ComputeValueVTs(DAG.getTargetLoweringInfo(), DAG.getDataLayout(),
                  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(MaxParallelChains, NumValues));
  SDLoc dl = getCurSDLoc();
  EVT PtrVT = Ptr.getValueType();
  unsigned Alignment = I.getAlignment();
  AAMDNodes AAInfo;
  I.getAAMetadata(AAInfo);

  auto MMOFlags = MachineMemOperand::MONone;
  if (I.isVolatile())
    MMOFlags |= MachineMemOperand::MOVolatile;
  if (I.getMetadata(LLVMContext::MD_nontemporal) != nullptr)
    MMOFlags |= MachineMemOperand::MONonTemporal;
  MMOFlags |= TLI.getMMOFlags(I);

  // An aggregate load cannot wrap around the address space, so offsets to its
  // parts don't wrap either.
  SDNodeFlags Flags;
  Flags.setNoUnsignedWrap(true);

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

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

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

  auto getMaskedStoreOps = [&](Value* &Ptr, Value* &Mask, Value* &Src0,
                           unsigned& Alignment) {
    // llvm.masked.store.*(Src0, Ptr, alignment, Mask)
    Src0 = I.getArgOperand(0);
    Ptr = I.getArgOperand(1);
    Alignment = cast<ConstantInt>(I.getArgOperand(2))->getZExtValue();
    Mask = I.getArgOperand(3);
  };
  auto getCompressingStoreOps = [&](Value* &Ptr, Value* &Mask, Value* &Src0,
                           unsigned& Alignment) {
    // llvm.masked.compressstore.*(Src0, Ptr, Mask)
    Src0 = I.getArgOperand(0);
    Ptr = I.getArgOperand(1);
    Mask = I.getArgOperand(2);
    Alignment = 0;
  };

  Value  *PtrOperand, *MaskOperand, *Src0Operand;
  unsigned Alignment;
  if (IsCompressing)
    getCompressingStoreOps(PtrOperand, MaskOperand, Src0Operand, Alignment);
  else
    getMaskedStoreOps(PtrOperand, MaskOperand, Src0Operand, Alignment);

  SDValue Ptr = getValue(PtrOperand);
  SDValue Src0 = getValue(Src0Operand);
  SDValue Mask = getValue(MaskOperand);

  EVT VT = Src0.getValueType();
  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 /* Truncating */,
                                         IsCompressing);
  DAG.setRoot(StoreNode);
  setValue(&I, StoreNode);
}

// Get a uniform base for the Gather/Scatter intrinsic.
// The first argument of the Gather/Scatter intrinsic is a vector of pointers.
// We try to represent it as a base pointer + vector of indices.
// Usually, the vector of pointers comes from a 'getelementptr' instruction.
// The first operand of the GEP may be a single pointer or a vector of pointers
// Example:
//   %gep.ptr = getelementptr i32, <8 x i32*> %vptr, <8 x i32> %ind
//  or
//   %gep.ptr = getelementptr i32, i32* %ptr,        <8 x i32> %ind
// %res = call <8 x i32> @llvm.masked.gather.v8i32(<8 x i32*> %gep.ptr, ..
//
// When the first GEP operand is a single pointer - it is the uniform base we
// are looking for. If first operand of the GEP is a splat vector - we
// extract the splat value and use it as a uniform base.
// In all other cases the function returns 'false'.
static bool getUniformBase(const Value* &Ptr, SDValue& Base, SDValue& Index,
                           SDValue &Scale, SelectionDAGBuilder* SDB) {
  SelectionDAG& DAG = SDB->DAG;
  LLVMContext &Context = *DAG.getContext();

  assert(Ptr->getType()->isVectorTy() && "Uexpected pointer type");
  const GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Ptr);
  if (!GEP)
    return false;

  const Value *GEPPtr = GEP->getPointerOperand();
  if (!GEPPtr->getType()->isVectorTy())
    Ptr = GEPPtr;
  else if (!(Ptr = getSplatValue(GEPPtr)))
    return false;

  unsigned FinalIndex = GEP->getNumOperands() - 1;
  Value *IndexVal = GEP->getOperand(FinalIndex);

  // Ensure all the other indices are 0.
  for (unsigned i = 1; i < FinalIndex; ++i) {
    auto *C = dyn_cast<ConstantInt>(GEP->getOperand(i));
    if (!C || !C->isZero())
      return false;
  }

  // The operands of the GEP may be defined in another basic block.
  // In this case we'll not find nodes for the operands.
  if (!SDB->findValue(Ptr) || !SDB->findValue(IndexVal))
    return false;

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  const DataLayout &DL = DAG.getDataLayout();
  Scale = DAG.getTargetConstant(DL.getTypeAllocSize(GEP->getResultElementType()),
                                SDB->getCurSDLoc(), TLI.getPointerTy(DL));
  Base = SDB->getValue(Ptr);
  Index = SDB->getValue(IndexVal);

  if (!Index.getValueType().isVector()) {
    unsigned GEPWidth = GEP->getType()->getVectorNumElements();
    EVT VT = EVT::getVectorVT(Context, Index.getValueType(), GEPWidth);
    Index = DAG.getSplatBuildVector(VT, SDLoc(Index), Index);
  }
  return true;
}

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

  // llvm.masked.scatter.*(Src0, Ptrs, alignemt, Mask)
  const Value *Ptr = I.getArgOperand(1);
  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);
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();

  AAMDNodes AAInfo;
  I.getAAMetadata(AAInfo);

  SDValue Base;
  SDValue Index;
  SDValue Scale;
  const Value *BasePtr = Ptr;
  bool UniformBase = getUniformBase(BasePtr, Base, Index, Scale, this);

  const Value *MemOpBasePtr = UniformBase ? BasePtr : nullptr;
  MachineMemOperand *MMO = DAG.getMachineFunction().
    getMachineMemOperand(MachinePointerInfo(MemOpBasePtr),
                         MachineMemOperand::MOStore,  VT.getStoreSize(),
                         Alignment, AAInfo);
  if (!UniformBase) {
    Base = DAG.getConstant(0, sdl, TLI.getPointerTy(DAG.getDataLayout()));
    Index = getValue(Ptr);
    Scale = DAG.getTargetConstant(1, sdl, TLI.getPointerTy(DAG.getDataLayout()));
  }
  SDValue Ops[] = { getRoot(), Src0, Mask, Base, Index, Scale };
  SDValue Scatter = DAG.getMaskedScatter(DAG.getVTList(MVT::Other), VT, sdl,
                                         Ops, MMO);
  DAG.setRoot(Scatter);
  setValue(&I, Scatter);
}

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

  auto getMaskedLoadOps = [&](Value* &Ptr, Value* &Mask, Value* &Src0,
                           unsigned& Alignment) {
    // @llvm.masked.load.*(Ptr, alignment, Mask, Src0)
    Ptr = I.getArgOperand(0);
    Alignment = cast<ConstantInt>(I.getArgOperand(1))->getZExtValue();
    Mask = I.getArgOperand(2);
    Src0 = I.getArgOperand(3);
  };
  auto getExpandingLoadOps = [&](Value* &Ptr, Value* &Mask, Value* &Src0,
                           unsigned& Alignment) {
    // @llvm.masked.expandload.*(Ptr, Mask, Src0)
    Ptr = I.getArgOperand(0);
    Alignment = 0;
    Mask = I.getArgOperand(1);
    Src0 = I.getArgOperand(2);
  };

  Value  *PtrOperand, *MaskOperand, *Src0Operand;
  unsigned Alignment;
  if (IsExpanding)
    getExpandingLoadOps(PtrOperand, MaskOperand, Src0Operand, Alignment);
  else
    getMaskedLoadOps(PtrOperand, MaskOperand, Src0Operand, Alignment);

  SDValue Ptr = getValue(PtrOperand);
  SDValue Src0 = getValue(Src0Operand);
  SDValue Mask = getValue(MaskOperand);

  EVT VT = Src0.getValueType();
  if (!Alignment)
    Alignment = DAG.getEVTAlignment(VT);

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

  // Do not serialize masked loads of constant memory with anything.
  bool AddToChain = !AA || !AA->pointsToConstantMemory(MemoryLocation(
      PtrOperand, DAG.getDataLayout().getTypeStoreSize(I.getType()), AAInfo));
  SDValue InChain = AddToChain ? DAG.getRoot() : 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, IsExpanding);
  if (AddToChain)
    PendingLoads.push_back(Load.getValue(1));
  setValue(&I, Load);
}

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

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

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT VT = TLI.getValueType(DAG.getDataLayout(), 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 Root = DAG.getRoot();
  SDValue Base;
  SDValue Index;
  SDValue Scale;
  const Value *BasePtr = Ptr;
  bool UniformBase = getUniformBase(BasePtr, Base, Index, Scale, this);
  bool ConstantMemory = false;
  if (UniformBase &&
      AA && AA->pointsToConstantMemory(MemoryLocation(
          BasePtr, DAG.getDataLayout().getTypeStoreSize(I.getType()),
          AAInfo))) {
    // Do not serialize (non-volatile) loads of constant memory with anything.
    Root = DAG.getEntryNode();
    ConstantMemory = true;
  }

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

  if (!UniformBase) {
    Base = DAG.getConstant(0, sdl, TLI.getPointerTy(DAG.getDataLayout()));
    Index = getValue(Ptr);
    Scale = DAG.getTargetConstant(1, sdl, TLI.getPointerTy(DAG.getDataLayout()));
  }
  SDValue Ops[] = { Root, Src0, Mask, Base, Index, Scale };
  SDValue Gather = DAG.getMaskedGather(DAG.getVTList(VT, MVT::Other), VT, sdl,
                                       Ops, MMO);

  SDValue OutChain = Gather.getValue(1);
  if (!ConstantMemory)
    PendingLoads.push_back(OutChain);
  setValue(&I, Gather);
}

void SelectionDAGBuilder::visitAtomicCmpXchg(const AtomicCmpXchgInst &I) {
  SDLoc dl = getCurSDLoc();
  AtomicOrdering SuccessOrder = I.getSuccessOrdering();
  AtomicOrdering FailureOrder = I.getFailureOrdering();
  SyncScope::ID SSID = I.getSyncScopeID();

  SDValue InChain = getRoot();

  MVT MemVT = getValue(I.getCompareOperand()).getSimpleValueType();
  SDVTList VTs = DAG.getVTList(MemVT, MVT::i1, MVT::Other);
  SDValue L = DAG.getAtomicCmpSwap(
      ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS, dl, MemVT, VTs, InChain,
      getValue(I.getPointerOperand()), getValue(I.getCompareOperand()),
      getValue(I.getNewValOperand()), MachinePointerInfo(I.getPointerOperand()),
      /*Alignment=*/ 0, SuccessOrder, FailureOrder, SSID);

  SDValue OutChain = L.getValue(2);

  setValue(&I, L);
  DAG.setRoot(OutChain);
}

void SelectionDAGBuilder::visitAtomicRMW(const AtomicRMWInst &I) {
  SDLoc dl = getCurSDLoc();
  ISD::NodeType NT;
  switch (I.getOperation()) {
  default: llvm_unreachable("Unknown atomicrmw operation");
  case AtomicRMWInst::Xchg: NT = ISD::ATOMIC_SWAP; break;
  case AtomicRMWInst::Add:  NT = ISD::ATOMIC_LOAD_ADD; break;
  case AtomicRMWInst::Sub:  NT = ISD::ATOMIC_LOAD_SUB; break;
  case AtomicRMWInst::And:  NT = ISD::ATOMIC_LOAD_AND; break;
  case AtomicRMWInst::Nand: NT = ISD::ATOMIC_LOAD_NAND; break;
  case AtomicRMWInst::Or:   NT = ISD::ATOMIC_LOAD_OR; break;
  case AtomicRMWInst::Xor:  NT = ISD::ATOMIC_LOAD_XOR; break;
  case AtomicRMWInst::Max:  NT = ISD::ATOMIC_LOAD_MAX; break;
  case AtomicRMWInst::Min:  NT = ISD::ATOMIC_LOAD_MIN; break;
  case AtomicRMWInst::UMax: NT = ISD::ATOMIC_LOAD_UMAX; break;
  case AtomicRMWInst::UMin: NT = ISD::ATOMIC_LOAD_UMIN; break;
  }
  AtomicOrdering Order = I.getOrdering();
  SyncScope::ID SSID = I.getSyncScopeID();

  SDValue InChain = getRoot();

  SDValue L =
    DAG.getAtomic(NT, dl,
                  getValue(I.getValOperand()).getSimpleValueType(),
                  InChain,
                  getValue(I.getPointerOperand()),
                  getValue(I.getValOperand()),
                  I.getPointerOperand(),
                  /* Alignment=*/ 0, Order, SSID);

  SDValue OutChain = L.getValue(1);

  setValue(&I, L);
  DAG.setRoot(OutChain);
}

void SelectionDAGBuilder::visitFence(const FenceInst &I) {
  SDLoc dl = getCurSDLoc();
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  SDValue Ops[3];
  Ops[0] = getRoot();
  Ops[1] = DAG.getConstant((unsigned)I.getOrdering(), dl,
                           TLI.getFenceOperandTy(DAG.getDataLayout()));
  Ops[2] = DAG.getConstant(I.getSyncScopeID(), dl,
                           TLI.getFenceOperandTy(DAG.getDataLayout()));
  DAG.setRoot(DAG.getNode(ISD::ATOMIC_FENCE, dl, MVT::Other, Ops));
}

void SelectionDAGBuilder::visitAtomicLoad(const LoadInst &I) {
  SDLoc dl = getCurSDLoc();
  AtomicOrdering Order = I.getOrdering();
  SyncScope::ID SSID = I.getSyncScopeID();

  SDValue InChain = getRoot();

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT VT = TLI.getValueType(DAG.getDataLayout(), I.getType());

  if (!TLI.supportsUnalignedAtomics() &&
      I.getAlignment() < VT.getStoreSize())
    report_fatal_error("Cannot generate unaligned atomic load");

  MachineMemOperand *MMO =
      DAG.getMachineFunction().
      getMachineMemOperand(MachinePointerInfo(I.getPointerOperand()),
                           MachineMemOperand::MOVolatile |
                           MachineMemOperand::MOLoad,
                           VT.getStoreSize(),
                           I.getAlignment() ? I.getAlignment() :
                                              DAG.getEVTAlignment(VT),
                           AAMDNodes(), nullptr, SSID, Order);

  InChain = TLI.prepareVolatileOrAtomicLoad(InChain, dl, DAG);
  SDValue L =
      DAG.getAtomic(ISD::ATOMIC_LOAD, dl, VT, VT, InChain,
                    getValue(I.getPointerOperand()), MMO);

  SDValue OutChain = L.getValue(1);

  setValue(&I, L);
  DAG.setRoot(OutChain);
}

void SelectionDAGBuilder::visitAtomicStore(const StoreInst &I) {
  SDLoc dl = getCurSDLoc();

  AtomicOrdering Order = I.getOrdering();
  SyncScope::ID SSID = I.getSyncScopeID();

  SDValue InChain = getRoot();

  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  EVT VT =
      TLI.getValueType(DAG.getDataLayout(), I.getValueOperand()->getType());

  if (I.getAlignment() < VT.getStoreSize())
    report_fatal_error("Cannot generate unaligned atomic store");

  SDValue OutChain =
    DAG.getAtomic(ISD::ATOMIC_STORE, dl, VT,
                  InChain,
                  getValue(I.getPointerOperand()),
                  getValue(I.getValueOperand()),
                  I.getPointerOperand(), I.getAlignment(),
                  Order, SSID);

  DAG.setRoot(OutChain);
}

/// visitTargetIntrinsic - Lower a call of a target intrinsic to an INTRINSIC
/// node.
void SelectionDAGBuilder::visitTargetIntrinsic(const CallInst &I,
                                               unsigned Intrinsic) {
  // Ignore the callsite's attributes. A specific call site may be marked with
  // readnone, but the lowering code will expect the chain based on the
  // definition.
  const Function *F = I.getCalledFunction();
  bool HasChain = !F->doesNotAccessMemory();
  bool OnlyLoad = HasChain && F->onlyReadsMemory();

  // Build the operand list.
  SmallVector<SDValue, 8> Ops;
  if (HasChain) {  // If this intrinsic has side-effects, chainify it.
    if (OnlyLoad) {
      // We don't need to serialize loads against other loads.
      Ops.push_back(DAG.getRoot());
    } else {
      Ops.push_back(getRoot());
    }
  }

  // Info is set by getTgtMemInstrinsic
  TargetLowering::IntrinsicInfo Info;
  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
  bool IsTgtIntrinsic = TLI.getTgtMemIntrinsic(Info, I,
                                               DAG.getMachineFunction(),
                                               Intrinsic);

  // Add the intrinsic ID as an integer operand if it's not a target intrinsic.
  if (!IsTgtIntrinsic || Info.opc == ISD::INTRINSIC_VOID ||
      Info.opc == ISD::INTRINSIC_W_CHAIN)
    Ops.push_back(DAG.getTargetConstant(Intrinsic, getCurSDLoc(),
                                        TLI.getPointerTy(DAG.getDataLayout())));

  // Add all operands of the call to the operand list.
  for (unsigned i = 0, e = I.getNumArgOperands(); i != e; ++i) {
    SDValue Op = getValue(I.getArgOperand(i));
    Ops.push_back(Op);
  }

  SmallVector<EVT, 4> ValueVTs;
  ComputeValueVTs(TLI, DAG.getDataLayout(), I.getType(), ValueVTs);

  if (HasChain)
    ValueVTs.push_back(MVT::Other);

  SDVTList VTs = DAG.getVTList(ValueVTs);

  // Create the node.
  SDValue Result;
  if (IsTgtIntrinsic) {
    // This is target intrinsic that touches memory
    Result = DAG.getMemIntrinsicNode(Info.opc, getCurSDLoc(), VTs,
      Ops, Info.memVT,
      MachinePointerInfo(Info.ptrVal, Info.offset), Info.align,
      Info.flags, Info.size);
  } else if (!HasChain) {
    Result = DAG.getNode(ISD::INTRINSIC_WO_CHAIN, getCurSDLoc(), VTs, Ops);
  } else if (!I.getType()->isVoidTy()) {
    Result = DAG.getNode(ISD::INTRINSIC_W_CHAIN, getCurSDLoc(), VTs, Ops);
  } else {
    Result = DAG.getNode(ISD::INTRINSIC_VOID, getCurSDLoc(), VTs, Ops);
  }

  if (HasChain) {
    SDValue Chain = Result.getValue(Result.getNode()->getNumValues()-1);
    if (OnlyLoad)
      PendingLoads.push_back(Chain);
    else
      DAG.setRoot(Chain);
  }

  if (!I.getType()->isVoidTy()) {
    if (VectorType *PTy = dyn_cast<VectorType>(I.getType())) {
      EVT VT = TLI.getValueType(DAG.getDataLayout(), PTy);
      Result = DAG.getNode(ISD::BITCAST, getCurSDLoc(), VT, Result);
    } else
      Result = lowerRangeToAssertZExt(DAG, I, Result);

    setValue(&I, Result);
  }
}

/// GetSignificand - Get the significand and build it into a floating-point
/// number with exponent of 1:
///
///   Op = (Op & 0x007fffff) | 0x3f800000;
///
/// where Op is the hexadecimal representation of floating point value.
static SDValue GetSignificand(SelectionDAG &DAG, SDValue Op, const SDLoc &dl) {
  SDValue t1 = DAG.getNode(ISD::AND, dl, MVT::i32, Op,
                           DAG.getConstant(0x007fffff, dl, MVT::i32));
  SDValue t2 = DAG.getNode(ISD::OR, dl, MVT::i32, t1,
                           DAG.getConstant(0x3f800000, dl, MVT::i32));
  return DAG.getNode(ISD::BITCAST, dl, MVT::f32, t2);
}

/// GetExponent - Get the exponent:
///
///   (float)(int)(((Op & 0x7f800000) >> 23) - 127);
///
/// where Op is the hexadecimal representation of floating point value.
static SDValue GetExponent(SelectionDAG &DAG, SDValue Op,
                           const TargetLowering &TLI, const SDLoc &dl) {
  SDValue t0 = DAG.getNode(ISD::AND, dl, MVT::i32, Op,
                           DAG.getConstant(0x7f800000, dl, MVT::i32));
  SDValue t1 = DAG.getNode(
      ISD::SRL, dl, MVT::i32, t0,
      DAG.getConstant(23, dl, TLI.getPointerTy(DAG.getDataLayout())));
  SDValue t2 = DAG.getNode(ISD::SUB, dl, MVT::i32, t1,
                           DAG.getConstant(127, dl, MVT::i32));
  return DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, t2);
}

/// getF32Constant - Get 32-bit floating point constant.
static SDValue getF32Constant(SelectionDAG &DAG, unsigned Flt,
                              const SDLoc &dl) {
  return DAG.getConstantFP(APFloat(APFloat::IEEEsingle(), APInt(32, Flt)), dl,
                           MVT::f32);
}

static SDValue getLimitedPrecisionExp2(SDValue t0, const SDLoc &dl,
                                       SelectionDAG &DAG) {
  // TODO: What fast-math-flags should be set on the floating-point nodes?

  //   IntegerPartOfX = ((int32_t)(t0);
  SDValue IntegerPartOfX = DAG.getNode(ISD::FP_TO_SINT, dl, MVT::i32, t0);

  //   FractionalPartOfX = t0 - (float)IntegerPartOfX;
  SDValue t1 = DAG.getNode(ISD::SINT_TO_FP, dl, MVT::f32, IntegerPartOfX);
  SDValue X = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0, t1);

  //   IntegerPartOfX <<= 23;
  IntegerPartOfX = DAG.getNode(
      ISD::SHL, dl, MVT::i32, IntegerPartOfX,
      DAG.getConstant(23, dl, DAG.getTargetLoweringInfo().getPointerTy(
                                  DAG.getDataLayout())));

  SDValue TwoToFractionalPartOfX;
  if (LimitFloatPrecision <= 6) {
    // For floating-point precision of 6:
    //
    //   TwoToFractionalPartOfX =
    //     0.997535578f +
    //       (0.735607626f + 0.252464424f * x) * x;
    //
    // error 0.0144103317, which is 6 bits
    SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                             getF32Constant(DAG, 0x3e814304, dl));
    SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
                             getF32Constant(DAG, 0x3f3c50c8, dl));
    SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
    TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
                                         getF32Constant(DAG, 0x3f7f5e7e, dl));
  } else if (LimitFloatPrecision <= 12) {
    // For floating-point precision of 12:
    //
    //   TwoToFractionalPartOfX =
    //     0.999892986f +
    //       (0.696457318f +
    //         (0.224338339f + 0.792043434e-1f * x) * x) * x;
    //
    // error 0.000107046256, which is 13 to 14 bits
    SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                             getF32Constant(DAG, 0x3da235e3, dl));
    SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
                             getF32Constant(DAG, 0x3e65b8f3, dl));
    SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
    SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
                             getF32Constant(DAG, 0x3f324b07, dl));
    SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
    TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
                                         getF32Constant(DAG, 0x3f7ff8fd, dl));
  } else { // LimitFloatPrecision <= 18
    // For floating-point precision of 18:
    //
    //   TwoToFractionalPartOfX =
    //     0.999999982f +
    //       (0.693148872f +
    //         (0.240227044f +
    //           (0.554906021e-1f +
    //             (0.961591928e-2f +
    //               (0.136028312e-2f + 0.157059148e-3f *x)*x)*x)*x)*x)*x;
    // error 2.47208000*10^(-7), which is better than 18 bits
    SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                             getF32Constant(DAG, 0x3924b03e, dl));
    SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
                             getF32Constant(DAG, 0x3ab24b87, dl));
    SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
    SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
                             getF32Constant(DAG, 0x3c1d8c17, dl));
    SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
    SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
                             getF32Constant(DAG, 0x3d634a1d, dl));
    SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
    SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
                             getF32Constant(DAG, 0x3e75fe14, dl));
    SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
    SDValue t11 = DAG.getNode(ISD::FADD, dl, MVT::f32, t10,
                              getF32Constant(DAG, 0x3f317234, dl));
    SDValue t12 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t11, X);
    TwoToFractionalPartOfX = DAG.getNode(ISD::FADD, dl, MVT::f32, t12,
                                         getF32Constant(DAG, 0x3f800000, dl));
  }

  // Add the exponent into the result in integer domain.
  SDValue t13 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, TwoToFractionalPartOfX);
  return DAG.getNode(ISD::BITCAST, dl, MVT::f32,
                     DAG.getNode(ISD::ADD, dl, MVT::i32, t13, IntegerPartOfX));
}

/// expandExp - Lower an exp intrinsic. Handles the special sequences for
/// limited-precision mode.
static SDValue expandExp(const SDLoc &dl, SDValue Op, SelectionDAG &DAG,
                         const TargetLowering &TLI) {
  if (Op.getValueType() == MVT::f32 &&
      LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {

    // Put the exponent in the right bit position for later addition to the
    // final result:
    //
    //   #define LOG2OFe 1.4426950f
    //   t0 = Op * LOG2OFe

    // TODO: What fast-math-flags should be set here?
    SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, Op,
                             getF32Constant(DAG, 0x3fb8aa3b, dl));
    return getLimitedPrecisionExp2(t0, dl, DAG);
  }

  // No special expansion.
  return DAG.getNode(ISD::FEXP, dl, Op.getValueType(), Op);
}

/// expandLog - Lower a log intrinsic. Handles the special sequences for
/// limited-precision mode.
static SDValue expandLog(const SDLoc &dl, SDValue Op, SelectionDAG &DAG,
                         const TargetLowering &TLI) {
  // TODO: What fast-math-flags should be set on the floating-point nodes?

  if (Op.getValueType() == MVT::f32 &&
      LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
    SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);

    // Scale the exponent by log(2) [0.69314718f].
    SDValue Exp = GetExponent(DAG, Op1, TLI, dl);
    SDValue LogOfExponent = DAG.getNode(ISD::FMUL, dl, MVT::f32, Exp,
                                        getF32Constant(DAG, 0x3f317218, dl));

    // Get the significand and build it into a floating-point number with
    // exponent of 1.
    SDValue X = GetSignificand(DAG, Op1, dl);

    SDValue LogOfMantissa;
    if (LimitFloatPrecision <= 6) {
      // For floating-point precision of 6:
      //
      //   LogofMantissa =
      //     -1.1609546f +
      //       (1.4034025f - 0.23903021f * x) * x;
      //
      // error 0.0034276066, which is better than 8 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0xbe74c456, dl));
      SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x3fb3a2b1, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
                                  getF32Constant(DAG, 0x3f949a29, dl));
    } else if (LimitFloatPrecision <= 12) {
      // For floating-point precision of 12:
      //
      //   LogOfMantissa =
      //     -1.7417939f +
      //       (2.8212026f +
      //         (-1.4699568f +
      //           (0.44717955f - 0.56570851e-1f * x) * x) * x) * x;
      //
      // error 0.000061011436, which is 14 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0xbd67b6d6, dl));
      SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x3ee4f4b8, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
                               getF32Constant(DAG, 0x3fbc278b, dl));
      SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
      SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
                               getF32Constant(DAG, 0x40348e95, dl));
      SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
      LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
                                  getF32Constant(DAG, 0x3fdef31a, dl));
    } else { // LimitFloatPrecision <= 18
      // For floating-point precision of 18:
      //
      //   LogOfMantissa =
      //     -2.1072184f +
      //       (4.2372794f +
      //         (-3.7029485f +
      //           (2.2781945f +
      //             (-0.87823314f +
      //               (0.19073739f - 0.17809712e-1f * x) * x) * x) * x) * x)*x;
      //
      // error 0.0000023660568, which is better than 18 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0xbc91e5ac, dl));
      SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x3e4350aa, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
                               getF32Constant(DAG, 0x3f60d3e3, dl));
      SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
      SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
                               getF32Constant(DAG, 0x4011cdf0, dl));
      SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
      SDValue t7 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
                               getF32Constant(DAG, 0x406cfd1c, dl));
      SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
      SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
                               getF32Constant(DAG, 0x408797cb, dl));
      SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
      LogOfMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t10,
                                  getF32Constant(DAG, 0x4006dcab, dl));
    }

    return DAG.getNode(ISD::FADD, dl, MVT::f32, LogOfExponent, LogOfMantissa);
  }

  // No special expansion.
  return DAG.getNode(ISD::FLOG, dl, Op.getValueType(), Op);
}

/// expandLog2 - Lower a log2 intrinsic. Handles the special sequences for
/// limited-precision mode.
static SDValue expandLog2(const SDLoc &dl, SDValue Op, SelectionDAG &DAG,
                          const TargetLowering &TLI) {
  // TODO: What fast-math-flags should be set on the floating-point nodes?

  if (Op.getValueType() == MVT::f32 &&
      LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
    SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);

    // Get the exponent.
    SDValue LogOfExponent = GetExponent(DAG, Op1, TLI, dl);

    // Get the significand and build it into a floating-point number with
    // exponent of 1.
    SDValue X = GetSignificand(DAG, Op1, dl);

    // Different possible minimax approximations of significand in
    // floating-point for various degrees of accuracy over [1,2].
    SDValue Log2ofMantissa;
    if (LimitFloatPrecision <= 6) {
      // For floating-point precision of 6:
      //
      //   Log2ofMantissa = -1.6749035f + (2.0246817f - .34484768f * x) * x;
      //
      // error 0.0049451742, which is more than 7 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0xbeb08fe0, dl));
      SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x40019463, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
                                   getF32Constant(DAG, 0x3fd6633d, dl));
    } else if (LimitFloatPrecision <= 12) {
      // For floating-point precision of 12:
      //
      //   Log2ofMantissa =
      //     -2.51285454f +
      //       (4.07009056f +
      //         (-2.12067489f +
      //           (.645142248f - 0.816157886e-1f * x) * x) * x) * x;
      //
      // error 0.0000876136000, which is better than 13 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0xbda7262e, dl));
      SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x3f25280b, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
                               getF32Constant(DAG, 0x4007b923, dl));
      SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
      SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
                               getF32Constant(DAG, 0x40823e2f, dl));
      SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
      Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
                                   getF32Constant(DAG, 0x4020d29c, dl));
    } else { // LimitFloatPrecision <= 18
      // For floating-point precision of 18:
      //
      //   Log2ofMantissa =
      //     -3.0400495f +
      //       (6.1129976f +
      //         (-5.3420409f +
      //           (3.2865683f +
      //             (-1.2669343f +
      //               (0.27515199f -
      //                 0.25691327e-1f * x) * x) * x) * x) * x) * x;
      //
      // error 0.0000018516, which is better than 18 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0xbcd2769e, dl));
      SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x3e8ce0b9, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      SDValue t3 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
                               getF32Constant(DAG, 0x3fa22ae7, dl));
      SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
      SDValue t5 = DAG.getNode(ISD::FADD, dl, MVT::f32, t4,
                               getF32Constant(DAG, 0x40525723, dl));
      SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
      SDValue t7 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t6,
                               getF32Constant(DAG, 0x40aaf200, dl));
      SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
      SDValue t9 = DAG.getNode(ISD::FADD, dl, MVT::f32, t8,
                               getF32Constant(DAG, 0x40c39dad, dl));
      SDValue t10 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t9, X);
      Log2ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t10,
                                   getF32Constant(DAG, 0x4042902c, dl));
    }

    return DAG.getNode(ISD::FADD, dl, MVT::f32, LogOfExponent, Log2ofMantissa);
  }

  // No special expansion.
  return DAG.getNode(ISD::FLOG2, dl, Op.getValueType(), Op);
}

/// expandLog10 - Lower a log10 intrinsic. Handles the special sequences for
/// limited-precision mode.
static SDValue expandLog10(const SDLoc &dl, SDValue Op, SelectionDAG &DAG,
                           const TargetLowering &TLI) {
  // TODO: What fast-math-flags should be set on the floating-point nodes?

  if (Op.getValueType() == MVT::f32 &&
      LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
    SDValue Op1 = DAG.getNode(ISD::BITCAST, dl, MVT::i32, Op);

    // Scale the exponent by log10(2) [0.30102999f].
    SDValue Exp = GetExponent(DAG, Op1, TLI, dl);
    SDValue LogOfExponent = DAG.getNode(ISD::FMUL, dl, MVT::f32, Exp,
                                        getF32Constant(DAG, 0x3e9a209a, dl));

    // Get the significand and build it into a floating-point number with
    // exponent of 1.
    SDValue X = GetSignificand(DAG, Op1, dl);

    SDValue Log10ofMantissa;
    if (LimitFloatPrecision <= 6) {
      // For floating-point precision of 6:
      //
      //   Log10ofMantissa =
      //     -0.50419619f +
      //       (0.60948995f - 0.10380950f * x) * x;
      //
      // error 0.0014886165, which is 6 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0xbdd49a13, dl));
      SDValue t1 = DAG.getNode(ISD::FADD, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x3f1c0789, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t2,
                                    getF32Constant(DAG, 0x3f011300, dl));
    } else if (LimitFloatPrecision <= 12) {
      // For floating-point precision of 12:
      //
      //   Log10ofMantissa =
      //     -0.64831180f +
      //       (0.91751397f +
      //         (-0.31664806f + 0.47637168e-1f * x) * x) * x;
      //
      // error 0.00019228036, which is better than 12 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0x3d431f31, dl));
      SDValue t1 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x3ea21fb2, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
                               getF32Constant(DAG, 0x3f6ae232, dl));
      SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
      Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t4,
                                    getF32Constant(DAG, 0x3f25f7c3, dl));
    } else { // LimitFloatPrecision <= 18
      // For floating-point precision of 18:
      //
      //   Log10ofMantissa =
      //     -0.84299375f +
      //       (1.5327582f +
      //         (-1.0688956f +
      //           (0.49102474f +
      //             (-0.12539807f + 0.13508273e-1f * x) * x) * x) * x) * x;
      //
      // error 0.0000037995730, which is better than 18 bits
      SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, X,
                               getF32Constant(DAG, 0x3c5d51ce, dl));
      SDValue t1 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t0,
                               getF32Constant(DAG, 0x3e00685a, dl));
      SDValue t2 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t1, X);
      SDValue t3 = DAG.getNode(ISD::FADD, dl, MVT::f32, t2,
                               getF32Constant(DAG, 0x3efb6798, dl));
      SDValue t4 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t3, X);
      SDValue t5 = DAG.getNode(ISD::FSUB, dl, MVT::f32, t4,
                               getF32Constant(DAG, 0x3f88d192, dl));
      SDValue t6 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t5, X);
      SDValue t7 = DAG.getNode(ISD::FADD, dl, MVT::f32, t6,
                               getF32Constant(DAG, 0x3fc4316c, dl));
      SDValue t8 = DAG.getNode(ISD::FMUL, dl, MVT::f32, t7, X);
      Log10ofMantissa = DAG.getNode(ISD::FSUB, dl, MVT::f32, t8,
                                    getF32Constant(DAG, 0x3f57ce70, dl));
    }

    return DAG.getNode(ISD::FADD, dl, MVT::f32, LogOfExponent, Log10ofMantissa);
  }

  // No special expansion.
  return DAG.getNode(ISD::FLOG10, dl, Op.getValueType(), Op);
}

/// expandExp2 - Lower an exp2 intrinsic. Handles the special sequences for
/// limited-precision mode.
static SDValue expandExp2(const SDLoc &dl, SDValue Op, SelectionDAG &DAG,
                          const TargetLowering &TLI) {
  if (Op.getValueType() == MVT::f32 &&
      LimitFloatPrecision > 0 && LimitFloatPrecision <= 18)
    return getLimitedPrecisionExp2(Op, dl, DAG);

  // No special expansion.
  return DAG.getNode(ISD::FEXP2, dl, Op.getValueType(), Op);
}

/// visitPow - Lower a pow intrinsic. Handles the special sequences for
/// limited-precision mode with x == 10.0f.
static SDValue expandPow(const SDLoc &dl, SDValue LHS, SDValue RHS,
                         SelectionDAG &DAG, const TargetLowering &TLI) {
  bool IsExp10 = false;
  if (LHS.getValueType() == MVT::f32 && RHS.getValueType() == MVT::f32 &&
      LimitFloatPrecision > 0 && LimitFloatPrecision <= 18) {
    if (ConstantFPSDNode *LHSC = dyn_cast<ConstantFPSDNode>(LHS)) {
      APFloat Ten(10.0f);
      IsExp10 = LHSC->isExactlyValue(Ten);
    }
  }

  // TODO: What fast-math-flags should be set on the FMUL node?
  if (IsExp10) {
    // Put the exponent in the right bit position for later addition to the
    // final result:
    //
    //   #define LOG2OF10 3.3219281f
    //   t0 = Op * LOG2OF10;
    SDValue t0 = DAG.getNode(ISD::FMUL, dl, MVT::f32, RHS,
                             getF32Constant(DAG, 0x40549a78, dl));
    return getLimitedPrecisionExp2(t0, dl, DAG);
  }

  // No special expansion.
  return DAG.getNode(ISD::FPOW, dl, LHS.getValueType(), LHS, RHS);
}

/// ExpandPowI - Expand a llvm.powi intrinsic.
static SDValue ExpandPowI(const SDLoc &DL, SDValue LHS, SDValue RHS,
                          SelectionDAG &DAG) {
  // If RHS is a constant, we can expand this out to a multiplication tree,
  // otherwise we end up lowering to a call to __powidf2 (for example).  When
  // optimizing for size, we only want to do this if the expansion would produce
  // a small number of multiplies, otherwise we do the full expansion.
  if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS)) {
    // Get the exponent as a positive value.
    unsigned Val = RHSC->getSExtValue();
    if ((int)Val < 0) Val = -Val;

    // powi(x, 0) -> 1.0
    if (Val == 0)
      return DAG.getConstantFP(1.0, DL, LHS.getValueType());

    const Function &F = DAG.getMachineFunction().getFunction();
    if (!F.optForSize() ||
        // If optimizing for size, don't insert too many multiplies.
        // This inserts up to 5 multiplies.
        countPopulation(Val) + Log2_32(Val) < 7) {
      // We use the simple binary decomposition method to generate the multiply
      // sequence.  There are more optimal ways to do this (for example,
      // powi(x,15) generates one more multiply than it should), but this has
      // the benefit of being both really simple and much better than a libcall.
      SDValue Res;  // Logically starts equal to 1.0
      SDValue CurSquare = LHS;
      // TODO: Intrinsics should have fast-math-flags that propagate to these
      // nodes.
      while (Val) {
        if (Val & 1) {
          if (Res.getNode())
            Res = DAG.getNode(ISD::FMUL, DL,Res.getValueType(), Res, CurSquare);
          else
            Res = CurSquare;  // 1.0*CurSquare.
        }

        CurSquare = DAG.getNode(ISD::FMUL, DL, CurSquare.getValueType(),
                                CurSquare, CurSquare);
        Val >>= 1;
      }

      // If the original was negative, invert the result, producing 1/(x*x*x).
      if (RHSC->getSExtValue() < 0)
        Res = DAG.getNode(ISD::FDIV, DL, LHS.getValueType(),
                          DAG.getConstantFP(1.0, DL, LHS.getValueType()), Res);
      return Res;
    }
  }

  // Otherwise, expand to a libcall.
  return DAG.getNode(ISD::FPOWI, DL, LHS.getValueType(), LHS, RHS);
}

// getUnderlyingArgReg - Find underlying register used for a truncated or
// bitcasted argument.
static unsigned getUnderlyingArgReg(const SDValue &N) {
  switch (N.getOpcode()) {
  case ISD::CopyFromReg:
    return cast<RegisterSDNode>(N.getOperand(1))->getReg();
  case ISD::BITCAST:
  case ISD::AssertZext:
  case ISD::AssertSext:
  case ISD::TRUNCATE:
    return getUnderlyingArgReg(N.getOperand(0));
  default:
    return 0;
  }
}

/// If the DbgValueInst is a dbg_value of a function argument, create the
/// corresponding DBG_VALUE machine instruction for it now.  At the end of
/// instruction selection, they will be inserted to the entry BB.
bool SelectionDAGBuilder::EmitFuncArgumentDbgValue(
    const Value *V, DILocalVariable *Variable, DIExpression *Expr,
    DILocation *DL, bool IsDbgDeclare, const SDValue &N) {
  const Argument *Arg = dyn_cast<Argument>(V);
  if (!Arg)
    return false;

  MachineFunction &MF = DAG.getMachineFunction();
  const TargetInstrInfo *TII = DAG.getSubtarget().getInstrInfo();

  bool IsIndirect = false;
  Optional<MachineOperand> Op;
  // Some arguments' frame index is recorded during argument lowering.
  int FI = FuncInfo.getArgumentFrameIndex(Arg);
  if (FI != std::numeric_limits<int>::max())
    Op = MachineOperand::CreateFI(FI);

  if (!Op && N.getNode()) {
    unsigned Reg = getUnderlyingArgReg(N);
    if (Reg && TargetRegisterInfo::isVirtualRegister(Reg)) {
      MachineRegisterInfo &RegInfo = MF.getRegInfo();
      unsigned PR = RegInfo.getLiveInPhysReg(Reg);
      if (PR)
        Reg = PR;
    }
    if (Reg) {
      Op = MachineOperand::CreateReg(Reg, false);
      IsIndirect = IsDbgDeclare;
    }
  }

  if (!Op && N.getNode())
    // Check if frame index is available.
    if (LoadSDNode *LNode = dyn_cast<LoadSDNode>(N.getNode()))
      if (FrameIndexSDNode *FINode =
          dyn_cast<FrameIndexSDNode>(LNode->getBasePtr().getNode()))
        Op = MachineOperand::CreateFI(FINode->getIndex());

  if (!Op) {
    // Check if ValueMap has reg number.
    DenseMap<const Value *, unsigned>::iterator VMI = FuncInfo.ValueMap.find(V);
    if (VMI != FuncInfo.ValueMap.end()) {
      const auto &TLI = DAG.getTargetLoweringInfo();
      RegsForValue RFV(V->getContext(), TLI, DAG.getDataLayout(), VMI->second,
                       V->getType(), getABIRegCopyCC(V));
      if (RFV.occupiesMultipleRegs()) {
        unsigned Offset = 0;
        for (auto RegAndSize : RFV.getRegsAndSizes()) {
          Op = MachineOperand::CreateReg(RegAndSize.first, false);
          auto FragmentExpr = DIExpression::createFragmentExpression(
              Expr, Offset, RegAndSize.second);
          if (!FragmentExpr)
            continue;
          FuncInfo.ArgDbgValues.push_back(
              BuildMI(MF, DL, TII->get(TargetOpcode::DBG_VALUE), IsDbgDeclare,
                      Op->