//===- LowerMemIntrinsics.cpp ----------------------------------*- C++ -*--===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//

#include "llvm/Transforms/Utils/LowerMemIntrinsics.h"
#include "llvm/Analysis/ScalarEvolution.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/MDBuilder.h"
#include "llvm/IR/ProfDataUtils.h"
#include "llvm/ProfileData/InstrProf.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/MathExtras.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/LoopUtils.h"
#include <limits>
#include <optional>

#define DEBUG_TYPE "lower-mem-intrinsics"

using namespace llvm;

namespace llvm {
extern cl::opt<bool> ProfcheckDisableMetadataFixes;
}

/// \returns \p Len urem \p OpSize, checking for optimization opportunities.
/// \p OpSizeVal must be the integer value of the \c ConstantInt \p OpSize.
static Value *getRuntimeLoopRemainder(IRBuilderBase &B, Value *Len,
                                      Value *OpSize, unsigned OpSizeVal) {
  // For powers of 2, we can and by (OpSizeVal - 1) instead of using urem.
  if (isPowerOf2_32(OpSizeVal))
    return B.CreateAnd(Len, OpSizeVal - 1);
  return B.CreateURem(Len, OpSize);
}

/// \returns (\p Len udiv \p OpSize) mul \p OpSize, checking for optimization
/// opportunities.
/// If \p RTLoopRemainder is provided, it must be the result of
/// \c getRuntimeLoopRemainder() with the same arguments.
static Value *getRuntimeLoopUnits(IRBuilderBase &B, Value *Len, Value *OpSize,
                                  unsigned OpSizeVal,
                                  Value *RTLoopRemainder = nullptr) {
  if (!RTLoopRemainder)
    RTLoopRemainder = getRuntimeLoopRemainder(B, Len, OpSize, OpSizeVal);
  return B.CreateSub(Len, RTLoopRemainder);
}

namespace {
/// Container for the return values of insertLoopExpansion.
struct LoopExpansionInfo {
  /// The instruction at the end of the main loop body.
  Instruction *MainLoopIP = nullptr;

  /// The unit index in the main loop body.
  Value *MainLoopIndex = nullptr;

  /// The instruction at the end of the residual loop body. Can be nullptr if no
  /// residual is required.
  Instruction *ResidualLoopIP = nullptr;

  /// The unit index in the residual loop body. Can be nullptr if no residual is
  /// required.
  Value *ResidualLoopIndex = nullptr;
};

std::optional<uint64_t> getAverageMemOpLoopTripCount(const MemIntrinsic &I) {
  if (ProfcheckDisableMetadataFixes)
    return std::nullopt;
  if (std::optional<Function::ProfileCount> EC =
          I.getFunction()->getEntryCount();
      !EC || !EC->getCount())
    return std::nullopt;
  if (const auto Len = I.getLengthInBytes())
    return Len->getZExtValue();
  uint64_t Total = 0;
  SmallVector<InstrProfValueData> ProfData =
      getValueProfDataFromInst(I, InstrProfValueKind::IPVK_MemOPSize,
                               std::numeric_limits<uint32_t>::max(), Total);
  if (!Total)
    return std::nullopt;
  uint64_t TripCount = 0;
  for (const auto &P : ProfData)
    TripCount += P.Count * P.Value;
  return std::round(1.0 * TripCount / Total);
}

} // namespace

/// Insert the control flow and loop counters for a memcpy/memset loop
/// expansion.
///
/// This function inserts IR corresponding to the following C code before
/// \p InsertBefore:
/// \code
/// LoopUnits = (Len / MainLoopStep) * MainLoopStep;
/// ResidualUnits = Len - LoopUnits;
/// MainLoopIndex = 0;
/// if (LoopUnits > 0) {
///   do {
///     // MainLoopIP
///     MainLoopIndex += MainLoopStep;
///   } while (MainLoopIndex < LoopUnits);
/// }
/// for (size_t i = 0; i < ResidualUnits; i += ResidualLoopStep) {
///   ResidualLoopIndex = LoopUnits + i;
///   // ResidualLoopIP
/// }
/// \endcode
///
/// \p MainLoopStep and \p ResidualLoopStep determine by how many "units" the
/// loop index is increased in each iteration of the main and residual loops,
/// respectively. In most cases, the "unit" will be bytes, but larger units are
/// useful for lowering memset.pattern.
///
/// The computation of \c LoopUnits and \c ResidualUnits is performed at compile
/// time if \p Len is a \c ConstantInt.
/// The second (residual) loop is omitted if \p ResidualLoopStep is 0 or equal
/// to \p MainLoopStep.
/// The generated \c MainLoopIP, \c MainLoopIndex, \c ResidualLoopIP, and
/// \c ResidualLoopIndex are returned in a \c LoopExpansionInfo object.
static LoopExpansionInfo
insertLoopExpansion(Instruction *InsertBefore, Value *Len,
                    unsigned MainLoopStep, unsigned ResidualLoopStep,
                    StringRef BBNamePrefix,
                    std::optional<uint64_t> AverageTripCount) {
  assert((ResidualLoopStep == 0 || MainLoopStep % ResidualLoopStep == 0) &&
         "ResidualLoopStep must divide MainLoopStep if specified");
  assert(ResidualLoopStep <= MainLoopStep &&
         "ResidualLoopStep cannot be larger than MainLoopStep");
  assert(MainLoopStep > 0 && "MainLoopStep must be non-zero");
  LoopExpansionInfo LEI;
  BasicBlock *PreLoopBB = InsertBefore->getParent();
  BasicBlock *PostLoopBB = PreLoopBB->splitBasicBlock(
      InsertBefore, BBNamePrefix + "-post-expansion");
  Function *ParentFunc = PreLoopBB->getParent();
  LLVMContext &Ctx = PreLoopBB->getContext();
  const DebugLoc &DbgLoc = InsertBefore->getStableDebugLoc();
  IRBuilder<> PreLoopBuilder(PreLoopBB->getTerminator());
  PreLoopBuilder.SetCurrentDebugLocation(DbgLoc);

  // Calculate the main loop trip count and remaining units to cover after the
  // loop.
  Type *LenType = Len->getType();
  IntegerType *ILenType = cast<IntegerType>(LenType);
  ConstantInt *CIMainLoopStep = ConstantInt::get(ILenType, MainLoopStep);

  Value *LoopUnits = Len;
  Value *ResidualUnits = nullptr;
  // We can make a conditional branch unconditional if we know that the
  // MainLoop must be executed at least once.
  bool MustTakeMainLoop = false;
  if (MainLoopStep != 1) {
    if (auto *CLen = dyn_cast<ConstantInt>(Len)) {
      uint64_t TotalUnits = CLen->getZExtValue();
      uint64_t LoopEndCount = alignDown(TotalUnits, MainLoopStep);
      uint64_t ResidualCount = TotalUnits - LoopEndCount;
      LoopUnits = ConstantInt::get(LenType, LoopEndCount);
      ResidualUnits = ConstantInt::get(LenType, ResidualCount);
      MustTakeMainLoop = LoopEndCount > 0;
      // As an optimization, we could skip generating the residual loop if
      // ResidualCount is known to be 0. However, current uses of this function
      // don't request a residual loop if the length is constant (they generate
      // a (potentially empty) sequence of loads and stores instead), so this
      // optimization would have no effect here.
    } else {
      ResidualUnits = getRuntimeLoopRemainder(PreLoopBuilder, Len,
                                              CIMainLoopStep, MainLoopStep);
      LoopUnits = getRuntimeLoopUnits(PreLoopBuilder, Len, CIMainLoopStep,
                                      MainLoopStep, ResidualUnits);
    }
  } else if (auto *CLen = dyn_cast<ConstantInt>(Len)) {
    MustTakeMainLoop = CLen->getZExtValue() > 0;
  }

  BasicBlock *MainLoopBB = BasicBlock::Create(
      Ctx, BBNamePrefix + "-expansion-main-body", ParentFunc, PostLoopBB);
  IRBuilder<> LoopBuilder(MainLoopBB);
  LoopBuilder.SetCurrentDebugLocation(DbgLoc);

  PHINode *LoopIndex = LoopBuilder.CreatePHI(LenType, 2, "loop-index");
  LEI.MainLoopIndex = LoopIndex;
  LoopIndex->addIncoming(ConstantInt::get(LenType, 0U), PreLoopBB);

  Value *NewIndex =
      LoopBuilder.CreateAdd(LoopIndex, ConstantInt::get(LenType, MainLoopStep));
  LoopIndex->addIncoming(NewIndex, MainLoopBB);

  // One argument of the addition is a loop-variant PHI, so it must be an
  // Instruction (i.e., it cannot be a Constant).
  LEI.MainLoopIP = cast<Instruction>(NewIndex);

  if (ResidualLoopStep > 0 && ResidualLoopStep < MainLoopStep) {
    // Loop body for the residual accesses.
    BasicBlock *ResLoopBB =
        BasicBlock::Create(Ctx, BBNamePrefix + "-expansion-residual-body",
                           PreLoopBB->getParent(), PostLoopBB);
    // BB to check if the residual loop is needed.
    BasicBlock *ResidualCondBB =
        BasicBlock::Create(Ctx, BBNamePrefix + "-expansion-residual-cond",
                           PreLoopBB->getParent(), ResLoopBB);

    // Enter the MainLoop unless no main loop iteration is required.
    ConstantInt *Zero = ConstantInt::get(ILenType, 0U);
    if (MustTakeMainLoop)
      PreLoopBuilder.CreateBr(MainLoopBB);
    else {
      auto *BR = PreLoopBuilder.CreateCondBr(
          PreLoopBuilder.CreateICmpNE(LoopUnits, Zero), MainLoopBB,
          ResidualCondBB);
      if (AverageTripCount.has_value()) {
        MDBuilder MDB(ParentFunc->getContext());
        setFittedBranchWeights(*BR,
                               {AverageTripCount.value() % MainLoopStep, 1},
                               /*IsExpected=*/false);
      } else {
        setExplicitlyUnknownBranchWeightsIfProfiled(*BR, DEBUG_TYPE);
      }
    }
    PreLoopBB->getTerminator()->eraseFromParent();

    // Stay in the MainLoop until we have handled all the LoopUnits. Then go to
    // the residual condition BB.
    LoopBuilder.CreateCondBr(LoopBuilder.CreateICmpULT(NewIndex, LoopUnits),
                             MainLoopBB, ResidualCondBB);

    // Determine if we need to branch to the residual loop or bypass it.
    IRBuilder<> RCBuilder(ResidualCondBB);
    RCBuilder.SetCurrentDebugLocation(DbgLoc);
    RCBuilder.CreateCondBr(RCBuilder.CreateICmpNE(ResidualUnits, Zero),
                           ResLoopBB, PostLoopBB);

    IRBuilder<> ResBuilder(ResLoopBB);
    ResBuilder.SetCurrentDebugLocation(DbgLoc);
    PHINode *ResidualIndex =
        ResBuilder.CreatePHI(LenType, 2, "residual-loop-index");
    ResidualIndex->addIncoming(Zero, ResidualCondBB);

    // Add the offset at the end of the main loop to the loop counter of the
    // residual loop to get the proper index.
    Value *FullOffset = ResBuilder.CreateAdd(LoopUnits, ResidualIndex);
    LEI.ResidualLoopIndex = FullOffset;

    Value *ResNewIndex = ResBuilder.CreateAdd(
        ResidualIndex, ConstantInt::get(LenType, ResidualLoopStep));
    ResidualIndex->addIncoming(ResNewIndex, ResLoopBB);

    // One argument of the addition is a loop-variant PHI, so it must be an
    // Instruction (i.e., it cannot be a Constant).
    LEI.ResidualLoopIP = cast<Instruction>(ResNewIndex);

    // Stay in the residual loop until all ResidualUnits are handled.
    ResBuilder.CreateCondBr(
        ResBuilder.CreateICmpULT(ResNewIndex, ResidualUnits), ResLoopBB,
        PostLoopBB);
  } else {
    // There is no need for a residual loop after the main loop. We do however
    // need to patch up the control flow by creating the terminators for the
    // preloop block and the main loop.

    // Enter the MainLoop unless no main loop iteration is required.
    if (MustTakeMainLoop) {
      PreLoopBuilder.CreateBr(MainLoopBB);
    } else {
      ConstantInt *Zero = ConstantInt::get(ILenType, 0U);
      MDBuilder B(ParentFunc->getContext());
      PreLoopBuilder.CreateCondBr(PreLoopBuilder.CreateICmpNE(LoopUnits, Zero),
                                  MainLoopBB, PostLoopBB,
                                  B.createLikelyBranchWeights());
    }
    PreLoopBB->getTerminator()->eraseFromParent();
    // Stay in the MainLoop until we have handled all the LoopUnits.
    auto *Br = LoopBuilder.CreateCondBr(
        LoopBuilder.CreateICmpULT(NewIndex, LoopUnits), MainLoopBB, PostLoopBB);
    if (AverageTripCount.has_value())
      setFittedBranchWeights(*Br, {AverageTripCount.value() / MainLoopStep, 1},
                             /*IsExpected=*/false);
    else
      setExplicitlyUnknownBranchWeightsIfProfiled(*Br, DEBUG_TYPE);
  }
  return LEI;
}

void llvm::createMemCpyLoopKnownSize(Instruction *InsertBefore, Value *SrcAddr,
                                     Value *DstAddr, ConstantInt *CopyLen,
                                     Align SrcAlign, Align DstAlign,
                                     bool SrcIsVolatile, bool DstIsVolatile,
                                     bool CanOverlap,
                                     const TargetTransformInfo &TTI,
                                     std::optional<uint32_t> AtomicElementSize,
                                     std::optional<uint64_t> AverageTripCount) {
  // No need to expand zero length copies.
  if (CopyLen->isZero())
    return;

  BasicBlock *PreLoopBB = InsertBefore->getParent();
  Function *ParentFunc = PreLoopBB->getParent();
  LLVMContext &Ctx = PreLoopBB->getContext();
  const DataLayout &DL = ParentFunc->getDataLayout();
  MDBuilder MDB(Ctx);
  MDNode *NewDomain = MDB.createAnonymousAliasScopeDomain("MemCopyDomain");
  StringRef Name = "MemCopyAliasScope";
  MDNode *NewScope = MDB.createAnonymousAliasScope(NewDomain, Name);

  unsigned SrcAS = cast<PointerType>(SrcAddr->getType())->getAddressSpace();
  unsigned DstAS = cast<PointerType>(DstAddr->getType())->getAddressSpace();

  Type *TypeOfCopyLen = CopyLen->getType();
  Type *LoopOpType = TTI.getMemcpyLoopLoweringType(
      Ctx, CopyLen, SrcAS, DstAS, SrcAlign, DstAlign, AtomicElementSize);
  assert((!AtomicElementSize || !LoopOpType->isVectorTy()) &&
         "Atomic memcpy lowering is not supported for vector operand type");

  Type *Int8Type = Type::getInt8Ty(Ctx);
  TypeSize LoopOpSize = DL.getTypeStoreSize(LoopOpType);
  assert(LoopOpSize.isFixed() && "LoopOpType cannot be a scalable vector type");
  assert((!AtomicElementSize || LoopOpSize % *AtomicElementSize == 0) &&
         "Atomic memcpy lowering is not supported for selected operand size");

  uint64_t LoopEndCount =
      alignDown(CopyLen->getZExtValue(), LoopOpSize.getFixedValue());

  // Skip the loop expansion entirely if the loop would never be taken.
  if (LoopEndCount != 0) {
    LoopExpansionInfo LEI =
        insertLoopExpansion(InsertBefore, CopyLen, LoopOpSize, 0,
                            "static-memcpy", AverageTripCount);

    // Fill MainLoopBB
    IRBuilder<> MainLoopBuilder(LEI.MainLoopIP);
    Align PartDstAlign(commonAlignment(DstAlign, LoopOpSize));
    Align PartSrcAlign(commonAlignment(SrcAlign, LoopOpSize));

    // If we used LoopOpType as GEP element type, we would iterate over the
    // buffers in TypeStoreSize strides while copying TypeAllocSize bytes, i.e.,
    // we would miss bytes if TypeStoreSize != TypeAllocSize. Therefore, use
    // byte offsets computed from the TypeStoreSize.
    Value *SrcGEP =
        MainLoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr, LEI.MainLoopIndex);
    LoadInst *Load = MainLoopBuilder.CreateAlignedLoad(
        LoopOpType, SrcGEP, PartSrcAlign, SrcIsVolatile);
    if (!CanOverlap) {
      // Set alias scope for loads.
      Load->setMetadata(LLVMContext::MD_alias_scope,
                        MDNode::get(Ctx, NewScope));
    }
    Value *DstGEP =
        MainLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr, LEI.MainLoopIndex);
    StoreInst *Store = MainLoopBuilder.CreateAlignedStore(
        Load, DstGEP, PartDstAlign, DstIsVolatile);
    if (!CanOverlap) {
      // Indicate that stores don't overlap loads.
      Store->setMetadata(LLVMContext::MD_noalias, MDNode::get(Ctx, NewScope));
    }
    if (AtomicElementSize) {
      Load->setAtomic(AtomicOrdering::Unordered);
      Store->setAtomic(AtomicOrdering::Unordered);
    }
    assert(!LEI.ResidualLoopIP && !LEI.ResidualLoopIndex &&
           "No residual loop was requested");
  }

  // Copy the remaining bytes with straight-line code.
  uint64_t BytesCopied = LoopEndCount;
  uint64_t RemainingBytes = CopyLen->getZExtValue() - BytesCopied;
  if (RemainingBytes == 0)
    return;

  IRBuilder<> RBuilder(InsertBefore);
  SmallVector<Type *, 5> RemainingOps;
  TTI.getMemcpyLoopResidualLoweringType(RemainingOps, Ctx, RemainingBytes,
                                        SrcAS, DstAS, SrcAlign, DstAlign,
                                        AtomicElementSize);

  for (auto *OpTy : RemainingOps) {
    Align PartSrcAlign(commonAlignment(SrcAlign, BytesCopied));
    Align PartDstAlign(commonAlignment(DstAlign, BytesCopied));

    TypeSize OperandSize = DL.getTypeStoreSize(OpTy);
    assert((!AtomicElementSize || OperandSize % *AtomicElementSize == 0) &&
           "Atomic memcpy lowering is not supported for selected operand size");

    Value *SrcGEP = RBuilder.CreateInBoundsGEP(
        Int8Type, SrcAddr, ConstantInt::get(TypeOfCopyLen, BytesCopied));
    LoadInst *Load =
        RBuilder.CreateAlignedLoad(OpTy, SrcGEP, PartSrcAlign, SrcIsVolatile);
    if (!CanOverlap) {
      // Set alias scope for loads.
      Load->setMetadata(LLVMContext::MD_alias_scope,
                        MDNode::get(Ctx, NewScope));
    }
    Value *DstGEP = RBuilder.CreateInBoundsGEP(
        Int8Type, DstAddr, ConstantInt::get(TypeOfCopyLen, BytesCopied));
    StoreInst *Store =
        RBuilder.CreateAlignedStore(Load, DstGEP, PartDstAlign, DstIsVolatile);
    if (!CanOverlap) {
      // Indicate that stores don't overlap loads.
      Store->setMetadata(LLVMContext::MD_noalias, MDNode::get(Ctx, NewScope));
    }
    if (AtomicElementSize) {
      Load->setAtomic(AtomicOrdering::Unordered);
      Store->setAtomic(AtomicOrdering::Unordered);
    }
    BytesCopied += OperandSize;
  }
  assert(BytesCopied == CopyLen->getZExtValue() &&
         "Bytes copied should match size in the call!");
}

void llvm::createMemCpyLoopUnknownSize(
    Instruction *InsertBefore, Value *SrcAddr, Value *DstAddr, Value *CopyLen,
    Align SrcAlign, Align DstAlign, bool SrcIsVolatile, bool DstIsVolatile,
    bool CanOverlap, const TargetTransformInfo &TTI,
    std::optional<uint32_t> AtomicElementSize,
    std::optional<uint64_t> AverageTripCount) {
  BasicBlock *PreLoopBB = InsertBefore->getParent();
  Function *ParentFunc = PreLoopBB->getParent();
  const DataLayout &DL = ParentFunc->getDataLayout();
  LLVMContext &Ctx = PreLoopBB->getContext();
  MDBuilder MDB(Ctx);
  MDNode *NewDomain = MDB.createAnonymousAliasScopeDomain("MemCopyDomain");
  StringRef Name = "MemCopyAliasScope";
  MDNode *NewScope = MDB.createAnonymousAliasScope(NewDomain, Name);

  unsigned SrcAS = cast<PointerType>(SrcAddr->getType())->getAddressSpace();
  unsigned DstAS = cast<PointerType>(DstAddr->getType())->getAddressSpace();

  Type *LoopOpType = TTI.getMemcpyLoopLoweringType(
      Ctx, CopyLen, SrcAS, DstAS, SrcAlign, DstAlign, AtomicElementSize);
  assert((!AtomicElementSize || !LoopOpType->isVectorTy()) &&
         "Atomic memcpy lowering is not supported for vector operand type");
  TypeSize LoopOpSize = DL.getTypeStoreSize(LoopOpType);
  assert((!AtomicElementSize || LoopOpSize % *AtomicElementSize == 0) &&
         "Atomic memcpy lowering is not supported for selected operand size");

  Type *Int8Type = Type::getInt8Ty(Ctx);

  Type *ResidualLoopOpType = AtomicElementSize
                                 ? Type::getIntNTy(Ctx, *AtomicElementSize * 8)
                                 : Int8Type;
  TypeSize ResidualLoopOpSize = DL.getTypeStoreSize(ResidualLoopOpType);
  assert(ResidualLoopOpSize == (AtomicElementSize ? *AtomicElementSize : 1) &&
         "Store size is expected to match type size");

  LoopExpansionInfo LEI =
      insertLoopExpansion(InsertBefore, CopyLen, LoopOpSize, ResidualLoopOpSize,
                          "dynamic-memcpy", AverageTripCount);

  // Fill MainLoopBB
  IRBuilder<> MainLoopBuilder(LEI.MainLoopIP);
  Align PartSrcAlign(commonAlignment(SrcAlign, LoopOpSize));
  Align PartDstAlign(commonAlignment(DstAlign, LoopOpSize));

  // If we used LoopOpType as GEP element type, we would iterate over the
  // buffers in TypeStoreSize strides while copying TypeAllocSize bytes, i.e.,
  // we would miss bytes if TypeStoreSize != TypeAllocSize. Therefore, use byte
  // offsets computed from the TypeStoreSize.
  Value *SrcGEP =
      MainLoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr, LEI.MainLoopIndex);
  LoadInst *Load = MainLoopBuilder.CreateAlignedLoad(
      LoopOpType, SrcGEP, PartSrcAlign, SrcIsVolatile);
  if (!CanOverlap) {
    // Set alias scope for loads.
    Load->setMetadata(LLVMContext::MD_alias_scope, MDNode::get(Ctx, NewScope));
  }
  Value *DstGEP =
      MainLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr, LEI.MainLoopIndex);
  StoreInst *Store = MainLoopBuilder.CreateAlignedStore(
      Load, DstGEP, PartDstAlign, DstIsVolatile);
  if (!CanOverlap) {
    // Indicate that stores don't overlap loads.
    Store->setMetadata(LLVMContext::MD_noalias, MDNode::get(Ctx, NewScope));
  }
  if (AtomicElementSize) {
    Load->setAtomic(AtomicOrdering::Unordered);
    Store->setAtomic(AtomicOrdering::Unordered);
  }

  // Fill ResidualLoopBB.
  if (!LEI.ResidualLoopIP)
    return;

  Align ResSrcAlign(commonAlignment(PartSrcAlign, ResidualLoopOpSize));
  Align ResDstAlign(commonAlignment(PartDstAlign, ResidualLoopOpSize));

  IRBuilder<> ResLoopBuilder(LEI.ResidualLoopIP);
  Value *ResSrcGEP = ResLoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr,
                                                      LEI.ResidualLoopIndex);
  LoadInst *ResLoad = ResLoopBuilder.CreateAlignedLoad(
      ResidualLoopOpType, ResSrcGEP, ResSrcAlign, SrcIsVolatile);
  if (!CanOverlap) {
    // Set alias scope for loads.
    ResLoad->setMetadata(LLVMContext::MD_alias_scope,
                         MDNode::get(Ctx, NewScope));
  }
  Value *ResDstGEP = ResLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr,
                                                      LEI.ResidualLoopIndex);
  StoreInst *ResStore = ResLoopBuilder.CreateAlignedStore(
      ResLoad, ResDstGEP, ResDstAlign, DstIsVolatile);
  if (!CanOverlap) {
    // Indicate that stores don't overlap loads.
    ResStore->setMetadata(LLVMContext::MD_noalias, MDNode::get(Ctx, NewScope));
  }
  if (AtomicElementSize) {
    ResLoad->setAtomic(AtomicOrdering::Unordered);
    ResStore->setAtomic(AtomicOrdering::Unordered);
  }
}

// If \p Addr1 and \p Addr2 are pointers to different address spaces, create an
// addresspacecast to obtain a pair of pointers in the same addressspace. The
// caller needs to ensure that addrspacecasting is possible.
// No-op if the pointers are in the same address space.
static std::pair<Value *, Value *>
tryInsertCastToCommonAddrSpace(IRBuilderBase &B, Value *Addr1, Value *Addr2,
                               const TargetTransformInfo &TTI) {
  Value *ResAddr1 = Addr1;
  Value *ResAddr2 = Addr2;

  unsigned AS1 = cast<PointerType>(Addr1->getType())->getAddressSpace();
  unsigned AS2 = cast<PointerType>(Addr2->getType())->getAddressSpace();
  if (AS1 != AS2) {
    if (TTI.isValidAddrSpaceCast(AS2, AS1))
      ResAddr2 = B.CreateAddrSpaceCast(Addr2, Addr1->getType());
    else if (TTI.isValidAddrSpaceCast(AS1, AS2))
      ResAddr1 = B.CreateAddrSpaceCast(Addr1, Addr2->getType());
    else
      llvm_unreachable("Can only lower memmove between address spaces if they "
                       "support addrspacecast");
  }
  return {ResAddr1, ResAddr2};
}

// Lower memmove to IR. memmove is required to correctly copy overlapping memory
// regions; therefore, it has to check the relative positions of the source and
// destination pointers and choose the copy direction accordingly.
//
// The code below is an IR rendition of this C function:
//
// void* memmove(void* dst, const void* src, size_t n) {
//   unsigned char* d = dst;
//   const unsigned char* s = src;
//   if (s < d) {
//     // copy backwards
//     while (n--) {
//       d[n] = s[n];
//     }
//   } else {
//     // copy forward
//     for (size_t i = 0; i < n; ++i) {
//       d[i] = s[i];
//     }
//   }
//   return dst;
// }
//
// If the TargetTransformInfo specifies a wider MemcpyLoopLoweringType, it is
// used for the memory accesses in the loops. Then, additional loops with
// byte-wise accesses are added for the remaining bytes.
static void createMemMoveLoopUnknownSize(Instruction *InsertBefore,
                                         Value *SrcAddr, Value *DstAddr,
                                         Value *CopyLen, Align SrcAlign,
                                         Align DstAlign, bool SrcIsVolatile,
                                         bool DstIsVolatile,
                                         const TargetTransformInfo &TTI) {
  Type *TypeOfCopyLen = CopyLen->getType();
  BasicBlock *OrigBB = InsertBefore->getParent();
  Function *F = OrigBB->getParent();
  const DataLayout &DL = F->getDataLayout();
  LLVMContext &Ctx = OrigBB->getContext();
  unsigned SrcAS = cast<PointerType>(SrcAddr->getType())->getAddressSpace();
  unsigned DstAS = cast<PointerType>(DstAddr->getType())->getAddressSpace();

  Type *LoopOpType = TTI.getMemcpyLoopLoweringType(Ctx, CopyLen, SrcAS, DstAS,
                                                   SrcAlign, DstAlign);
  TypeSize LoopOpSize = DL.getTypeStoreSize(LoopOpType);
  Type *Int8Type = Type::getInt8Ty(Ctx);
  bool LoopOpIsInt8 = LoopOpType == Int8Type;

  // If the memory accesses are wider than one byte, residual loops with
  // i8-accesses are required to move remaining bytes.
  bool RequiresResidual = !LoopOpIsInt8;

  Type *ResidualLoopOpType = Int8Type;
  TypeSize ResidualLoopOpSize = DL.getTypeStoreSize(ResidualLoopOpType);

  // Calculate the loop trip count and remaining bytes to copy after the loop.
  IntegerType *ILengthType = cast<IntegerType>(TypeOfCopyLen);
  ConstantInt *CILoopOpSize = ConstantInt::get(ILengthType, LoopOpSize);
  ConstantInt *CIResidualLoopOpSize =
      ConstantInt::get(ILengthType, ResidualLoopOpSize);
  ConstantInt *Zero = ConstantInt::get(ILengthType, 0);

  const DebugLoc &DbgLoc = InsertBefore->getStableDebugLoc();
  IRBuilder<> PLBuilder(InsertBefore);
  PLBuilder.SetCurrentDebugLocation(DbgLoc);

  Value *RuntimeLoopBytes = CopyLen;
  Value *RuntimeLoopRemainder = nullptr;
  Value *SkipResidualCondition = nullptr;
  if (RequiresResidual) {
    RuntimeLoopRemainder =
        getRuntimeLoopRemainder(PLBuilder, CopyLen, CILoopOpSize, LoopOpSize);
    RuntimeLoopBytes = getRuntimeLoopUnits(PLBuilder, CopyLen, CILoopOpSize,
                                           LoopOpSize, RuntimeLoopRemainder);
    SkipResidualCondition =
        PLBuilder.CreateICmpEQ(RuntimeLoopRemainder, Zero, "skip_residual");
  }
  Value *SkipMainCondition =
      PLBuilder.CreateICmpEQ(RuntimeLoopBytes, Zero, "skip_main");

  // Create the a comparison of src and dst, based on which we jump to either
  // the forward-copy part of the function (if src >= dst) or the backwards-copy
  // part (if src < dst).
  // SplitBlockAndInsertIfThenElse conveniently creates the basic if-then-else
  // structure. Its block terminators (unconditional branches) are replaced by
  // the appropriate conditional branches when the loop is built.
  // If the pointers are in different address spaces, they need to be converted
  // to a compatible one. Cases where memory ranges in the different address
  // spaces cannot overlap are lowered as memcpy and not handled here.
  auto [CmpSrcAddr, CmpDstAddr] =
      tryInsertCastToCommonAddrSpace(PLBuilder, SrcAddr, DstAddr, TTI);
  Value *PtrCompare =
      PLBuilder.CreateICmpULT(CmpSrcAddr, CmpDstAddr, "compare_src_dst");
  Instruction *ThenTerm, *ElseTerm;
  SplitBlockAndInsertIfThenElse(PtrCompare, InsertBefore->getIterator(),
                                &ThenTerm, &ElseTerm);

  // If the LoopOpSize is greater than 1, each part of the function consists of
  // four blocks:
  //   memmove_copy_backwards:
  //       skip the residual loop when 0 iterations are required
  //   memmove_bwd_residual_loop:
  //       copy the last few bytes individually so that the remaining length is
  //       a multiple of the LoopOpSize
  //   memmove_bwd_middle: skip the main loop when 0 iterations are required
  //   memmove_bwd_main_loop: the actual backwards loop BB with wide accesses
  //   memmove_copy_forward: skip the main loop when 0 iterations are required
  //   memmove_fwd_main_loop: the actual forward loop BB with wide accesses
  //   memmove_fwd_middle: skip the residual loop when 0 iterations are required
  //   memmove_fwd_residual_loop: copy the last few bytes individually
  //
  // The main and residual loop are switched between copying forward and
  // backward so that the residual loop always operates on the end of the moved
  // range. This is based on the assumption that buffers whose start is aligned
  // with the LoopOpSize are more common than buffers whose end is.
  //
  // If the LoopOpSize is 1, each part of the function consists of two blocks:
  //   memmove_copy_backwards: skip the loop when 0 iterations are required
  //   memmove_bwd_main_loop: the actual backwards loop BB
  //   memmove_copy_forward: skip the loop when 0 iterations are required
  //   memmove_fwd_main_loop: the actual forward loop BB
  BasicBlock *CopyBackwardsBB = ThenTerm->getParent();
  CopyBackwardsBB->setName("memmove_copy_backwards");
  BasicBlock *CopyForwardBB = ElseTerm->getParent();
  CopyForwardBB->setName("memmove_copy_forward");
  BasicBlock *ExitBB = InsertBefore->getParent();
  ExitBB->setName("memmove_done");

  Align PartSrcAlign(commonAlignment(SrcAlign, LoopOpSize));
  Align PartDstAlign(commonAlignment(DstAlign, LoopOpSize));

  // Accesses in the residual loops do not share the same alignment as those in
  // the main loops.
  Align ResidualSrcAlign(commonAlignment(PartSrcAlign, ResidualLoopOpSize));
  Align ResidualDstAlign(commonAlignment(PartDstAlign, ResidualLoopOpSize));

  // Copying backwards.
  {
    BasicBlock *MainLoopBB = BasicBlock::Create(
        F->getContext(), "memmove_bwd_main_loop", F, CopyForwardBB);

    // The predecessor of the memmove_bwd_main_loop. Updated in the
    // following if a residual loop is emitted first.
    BasicBlock *PredBB = CopyBackwardsBB;

    if (RequiresResidual) {
      // backwards residual loop
      BasicBlock *ResidualLoopBB = BasicBlock::Create(
          F->getContext(), "memmove_bwd_residual_loop", F, MainLoopBB);
      IRBuilder<> ResidualLoopBuilder(ResidualLoopBB);
      ResidualLoopBuilder.SetCurrentDebugLocation(DbgLoc);
      PHINode *ResidualLoopPhi = ResidualLoopBuilder.CreatePHI(ILengthType, 0);
      Value *ResidualIndex = ResidualLoopBuilder.CreateSub(
          ResidualLoopPhi, CIResidualLoopOpSize, "bwd_residual_index");
      // If we used LoopOpType as GEP element type, we would iterate over the
      // buffers in TypeStoreSize strides while copying TypeAllocSize bytes,
      // i.e., we would miss bytes if TypeStoreSize != TypeAllocSize. Therefore,
      // use byte offsets computed from the TypeStoreSize.
      Value *LoadGEP = ResidualLoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr,
                                                             ResidualIndex);
      Value *Element = ResidualLoopBuilder.CreateAlignedLoad(
          ResidualLoopOpType, LoadGEP, ResidualSrcAlign, SrcIsVolatile,
          "element");
      Value *StoreGEP = ResidualLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr,
                                                              ResidualIndex);
      ResidualLoopBuilder.CreateAlignedStore(Element, StoreGEP,
                                             ResidualDstAlign, DstIsVolatile);

      // After the residual loop, go to an intermediate block.
      BasicBlock *IntermediateBB = BasicBlock::Create(
          F->getContext(), "memmove_bwd_middle", F, MainLoopBB);
      // Later code expects a terminator in the PredBB.
      IRBuilder<> IntermediateBuilder(IntermediateBB);
      IntermediateBuilder.SetCurrentDebugLocation(DbgLoc);
      IntermediateBuilder.CreateUnreachable();
      ResidualLoopBuilder.CreateCondBr(
          ResidualLoopBuilder.CreateICmpEQ(ResidualIndex, RuntimeLoopBytes),
          IntermediateBB, ResidualLoopBB);

      ResidualLoopPhi->addIncoming(ResidualIndex, ResidualLoopBB);
      ResidualLoopPhi->addIncoming(CopyLen, CopyBackwardsBB);

      // How to get to the residual:
      BranchInst *BrInst =
          BranchInst::Create(IntermediateBB, ResidualLoopBB,
                             SkipResidualCondition, ThenTerm->getIterator());
      BrInst->setDebugLoc(DbgLoc);
      ThenTerm->eraseFromParent();

      PredBB = IntermediateBB;
    }

    // main loop
    IRBuilder<> MainLoopBuilder(MainLoopBB);
    MainLoopBuilder.SetCurrentDebugLocation(DbgLoc);
    PHINode *MainLoopPhi = MainLoopBuilder.CreatePHI(ILengthType, 0);
    Value *MainIndex =
        MainLoopBuilder.CreateSub(MainLoopPhi, CILoopOpSize, "bwd_main_index");
    Value *LoadGEP =
        MainLoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr, MainIndex);
    Value *Element = MainLoopBuilder.CreateAlignedLoad(
        LoopOpType, LoadGEP, PartSrcAlign, SrcIsVolatile, "element");
    Value *StoreGEP =
        MainLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr, MainIndex);
    MainLoopBuilder.CreateAlignedStore(Element, StoreGEP, PartDstAlign,
                                       DstIsVolatile);
    MainLoopBuilder.CreateCondBr(MainLoopBuilder.CreateICmpEQ(MainIndex, Zero),
                                 ExitBB, MainLoopBB);
    MainLoopPhi->addIncoming(MainIndex, MainLoopBB);
    MainLoopPhi->addIncoming(RuntimeLoopBytes, PredBB);

    // How to get to the main loop:
    Instruction *PredBBTerm = PredBB->getTerminator();
    BranchInst *BrInst = BranchInst::Create(
        ExitBB, MainLoopBB, SkipMainCondition, PredBBTerm->getIterator());
    BrInst->setDebugLoc(DbgLoc);
    PredBBTerm->eraseFromParent();
  }

  // Copying forward.
  // main loop
  {
    BasicBlock *MainLoopBB =
        BasicBlock::Create(F->getContext(), "memmove_fwd_main_loop", F, ExitBB);
    IRBuilder<> MainLoopBuilder(MainLoopBB);
    MainLoopBuilder.SetCurrentDebugLocation(DbgLoc);
    PHINode *MainLoopPhi =
        MainLoopBuilder.CreatePHI(ILengthType, 0, "fwd_main_index");
    Value *LoadGEP =
        MainLoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr, MainLoopPhi);
    Value *Element = MainLoopBuilder.CreateAlignedLoad(
        LoopOpType, LoadGEP, PartSrcAlign, SrcIsVolatile, "element");
    Value *StoreGEP =
        MainLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr, MainLoopPhi);
    MainLoopBuilder.CreateAlignedStore(Element, StoreGEP, PartDstAlign,
                                       DstIsVolatile);
    Value *MainIndex = MainLoopBuilder.CreateAdd(MainLoopPhi, CILoopOpSize);
    MainLoopPhi->addIncoming(MainIndex, MainLoopBB);
    MainLoopPhi->addIncoming(Zero, CopyForwardBB);

    Instruction *CopyFwdBBTerm = CopyForwardBB->getTerminator();
    BasicBlock *SuccessorBB = ExitBB;
    if (RequiresResidual)
      SuccessorBB =
          BasicBlock::Create(F->getContext(), "memmove_fwd_middle", F, ExitBB);

    // leaving or staying in the main loop
    MainLoopBuilder.CreateCondBr(
        MainLoopBuilder.CreateICmpEQ(MainIndex, RuntimeLoopBytes), SuccessorBB,
        MainLoopBB);

    // getting in or skipping the main loop
    BranchInst *BrInst =
        BranchInst::Create(SuccessorBB, MainLoopBB, SkipMainCondition,
                           CopyFwdBBTerm->getIterator());
    BrInst->setDebugLoc(DbgLoc);
    CopyFwdBBTerm->eraseFromParent();

    if (RequiresResidual) {
      BasicBlock *IntermediateBB = SuccessorBB;
      IRBuilder<> IntermediateBuilder(IntermediateBB);
      IntermediateBuilder.SetCurrentDebugLocation(DbgLoc);
      BasicBlock *ResidualLoopBB = BasicBlock::Create(
          F->getContext(), "memmove_fwd_residual_loop", F, ExitBB);
      IntermediateBuilder.CreateCondBr(SkipResidualCondition, ExitBB,
                                       ResidualLoopBB);

      // Residual loop
      IRBuilder<> ResidualLoopBuilder(ResidualLoopBB);
      ResidualLoopBuilder.SetCurrentDebugLocation(DbgLoc);
      PHINode *ResidualLoopPhi =
          ResidualLoopBuilder.CreatePHI(ILengthType, 0, "fwd_residual_index");
      Value *LoadGEP = ResidualLoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr,
                                                             ResidualLoopPhi);
      Value *Element = ResidualLoopBuilder.CreateAlignedLoad(
          ResidualLoopOpType, LoadGEP, ResidualSrcAlign, SrcIsVolatile,
          "element");
      Value *StoreGEP = ResidualLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr,
                                                              ResidualLoopPhi);
      ResidualLoopBuilder.CreateAlignedStore(Element, StoreGEP,
                                             ResidualDstAlign, DstIsVolatile);
      Value *ResidualIndex =
          ResidualLoopBuilder.CreateAdd(ResidualLoopPhi, CIResidualLoopOpSize);
      ResidualLoopBuilder.CreateCondBr(
          ResidualLoopBuilder.CreateICmpEQ(ResidualIndex, CopyLen), ExitBB,
          ResidualLoopBB);
      ResidualLoopPhi->addIncoming(ResidualIndex, ResidualLoopBB);
      ResidualLoopPhi->addIncoming(RuntimeLoopBytes, IntermediateBB);
    }
  }
}

// Similar to createMemMoveLoopUnknownSize, only the trip counts are computed at
// compile time, obsolete loops and branches are omitted, and the residual code
// is straight-line code instead of a loop.
static void createMemMoveLoopKnownSize(Instruction *InsertBefore,
                                       Value *SrcAddr, Value *DstAddr,
                                       ConstantInt *CopyLen, Align SrcAlign,
                                       Align DstAlign, bool SrcIsVolatile,
                                       bool DstIsVolatile,
                                       const TargetTransformInfo &TTI) {
  // No need to expand zero length moves.
  if (CopyLen->isZero())
    return;

  Type *TypeOfCopyLen = CopyLen->getType();
  BasicBlock *OrigBB = InsertBefore->getParent();
  Function *F = OrigBB->getParent();
  const DataLayout &DL = F->getDataLayout();
  LLVMContext &Ctx = OrigBB->getContext();
  unsigned SrcAS = cast<PointerType>(SrcAddr->getType())->getAddressSpace();
  unsigned DstAS = cast<PointerType>(DstAddr->getType())->getAddressSpace();

  Type *LoopOpType = TTI.getMemcpyLoopLoweringType(Ctx, CopyLen, SrcAS, DstAS,
                                                   SrcAlign, DstAlign);
  TypeSize LoopOpSize = DL.getTypeStoreSize(LoopOpType);
  assert(LoopOpSize.isFixed() && "LoopOpType cannot be a scalable vector type");
  Type *Int8Type = Type::getInt8Ty(Ctx);

  // Calculate the loop trip count and remaining bytes to copy after the loop.
  uint64_t BytesCopiedInLoop =
      alignDown(CopyLen->getZExtValue(), LoopOpSize.getFixedValue());
  uint64_t RemainingBytes = CopyLen->getZExtValue() - BytesCopiedInLoop;

  IntegerType *ILengthType = cast<IntegerType>(TypeOfCopyLen);
  ConstantInt *Zero = ConstantInt::get(ILengthType, 0);
  ConstantInt *LoopBound = ConstantInt::get(ILengthType, BytesCopiedInLoop);
  ConstantInt *CILoopOpSize = ConstantInt::get(ILengthType, LoopOpSize);

  const DebugLoc &DbgLoc = InsertBefore->getStableDebugLoc();
  IRBuilder<> PLBuilder(InsertBefore);
  PLBuilder.SetCurrentDebugLocation(DbgLoc);

  auto [CmpSrcAddr, CmpDstAddr] =
      tryInsertCastToCommonAddrSpace(PLBuilder, SrcAddr, DstAddr, TTI);
  Value *PtrCompare =
      PLBuilder.CreateICmpULT(CmpSrcAddr, CmpDstAddr, "compare_src_dst");
  Instruction *ThenTerm, *ElseTerm;
  SplitBlockAndInsertIfThenElse(PtrCompare, InsertBefore->getIterator(),
                                &ThenTerm, &ElseTerm);

  BasicBlock *CopyBackwardsBB = ThenTerm->getParent();
  BasicBlock *CopyForwardBB = ElseTerm->getParent();
  BasicBlock *ExitBB = InsertBefore->getParent();
  ExitBB->setName("memmove_done");

  Align PartSrcAlign(commonAlignment(SrcAlign, LoopOpSize));
  Align PartDstAlign(commonAlignment(DstAlign, LoopOpSize));

  // Helper function to generate a load/store pair of a given type in the
  // residual. Used in the forward and backward branches.
  auto GenerateResidualLdStPair = [&](Type *OpTy, IRBuilderBase &Builder,
                                      uint64_t &BytesCopied) {
    Align ResSrcAlign(commonAlignment(SrcAlign, BytesCopied));
    Align ResDstAlign(commonAlignment(DstAlign, BytesCopied));

    TypeSize OperandSize = DL.getTypeStoreSize(OpTy);

    // If we used LoopOpType as GEP element type, we would iterate over the
    // buffers in TypeStoreSize strides while copying TypeAllocSize bytes, i.e.,
    // we would miss bytes if TypeStoreSize != TypeAllocSize. Therefore, use
    // byte offsets computed from the TypeStoreSize.
    Value *SrcGEP = Builder.CreateInBoundsGEP(
        Int8Type, SrcAddr, ConstantInt::get(TypeOfCopyLen, BytesCopied));
    LoadInst *Load =
        Builder.CreateAlignedLoad(OpTy, SrcGEP, ResSrcAlign, SrcIsVolatile);
    Value *DstGEP = Builder.CreateInBoundsGEP(
        Int8Type, DstAddr, ConstantInt::get(TypeOfCopyLen, BytesCopied));
    Builder.CreateAlignedStore(Load, DstGEP, ResDstAlign, DstIsVolatile);
    BytesCopied += OperandSize;
  };

  // Copying backwards.
  if (RemainingBytes != 0) {
    CopyBackwardsBB->setName("memmove_bwd_residual");
    uint64_t BytesCopied = BytesCopiedInLoop;

    // Residual code is required to move the remaining bytes. We need the same
    // instructions as in the forward case, only in reverse. So we generate code
    // the same way, except that we change the IRBuilder insert point for each
    // load/store pair so that each one is inserted before the previous one
    // instead of after it.
    IRBuilder<> BwdResBuilder(CopyBackwardsBB,
                              CopyBackwardsBB->getFirstNonPHIIt());
    BwdResBuilder.SetCurrentDebugLocation(DbgLoc);
    SmallVector<Type *, 5> RemainingOps;
    TTI.getMemcpyLoopResidualLoweringType(RemainingOps, Ctx, RemainingBytes,
                                          SrcAS, DstAS, PartSrcAlign,
                                          PartDstAlign);
    for (auto *OpTy : RemainingOps) {
      // reverse the order of the emitted operations
      BwdResBuilder.SetInsertPoint(CopyBackwardsBB,
                                   CopyBackwardsBB->getFirstNonPHIIt());
      GenerateResidualLdStPair(OpTy, BwdResBuilder, BytesCopied);
    }
  }
  if (BytesCopiedInLoop != 0) {
    BasicBlock *LoopBB = CopyBackwardsBB;
    BasicBlock *PredBB = OrigBB;
    if (RemainingBytes != 0) {
      // if we introduce residual code, it needs its separate BB
      LoopBB = CopyBackwardsBB->splitBasicBlock(
          CopyBackwardsBB->getTerminator(), "memmove_bwd_loop");
      PredBB = CopyBackwardsBB;
    } else {
      CopyBackwardsBB->setName("memmove_bwd_loop");
    }
    IRBuilder<> LoopBuilder(LoopBB->getTerminator());
    LoopBuilder.SetCurrentDebugLocation(DbgLoc);
    PHINode *LoopPhi = LoopBuilder.CreatePHI(ILengthType, 0);
    Value *Index = LoopBuilder.CreateSub(LoopPhi, CILoopOpSize, "bwd_index");
    Value *LoadGEP = LoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr, Index);
    Value *Element = LoopBuilder.CreateAlignedLoad(
        LoopOpType, LoadGEP, PartSrcAlign, SrcIsVolatile, "element");
    Value *StoreGEP = LoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr, Index);
    LoopBuilder.CreateAlignedStore(Element, StoreGEP, PartDstAlign,
                                   DstIsVolatile);

    // Replace the unconditional branch introduced by
    // SplitBlockAndInsertIfThenElse to turn LoopBB into a loop.
    Instruction *UncondTerm = LoopBB->getTerminator();
    LoopBuilder.CreateCondBr(LoopBuilder.CreateICmpEQ(Index, Zero), ExitBB,
                             LoopBB);
    UncondTerm->eraseFromParent();

    LoopPhi->addIncoming(Index, LoopBB);
    LoopPhi->addIncoming(LoopBound, PredBB);
  }

  // Copying forward.
  BasicBlock *FwdResidualBB = CopyForwardBB;
  if (BytesCopiedInLoop != 0) {
    CopyForwardBB->setName("memmove_fwd_loop");
    BasicBlock *LoopBB = CopyForwardBB;
    BasicBlock *SuccBB = ExitBB;
    if (RemainingBytes != 0) {
      // if we introduce residual code, it needs its separate BB
      SuccBB = CopyForwardBB->splitBasicBlock(CopyForwardBB->getTerminator(),
                                              "memmove_fwd_residual");
      FwdResidualBB = SuccBB;
    }
    IRBuilder<> LoopBuilder(LoopBB->getTerminator());
    LoopBuilder.SetCurrentDebugLocation(DbgLoc);
    PHINode *LoopPhi = LoopBuilder.CreatePHI(ILengthType, 0, "fwd_index");
    Value *LoadGEP = LoopBuilder.CreateInBoundsGEP(Int8Type, SrcAddr, LoopPhi);
    Value *Element = LoopBuilder.CreateAlignedLoad(
        LoopOpType, LoadGEP, PartSrcAlign, SrcIsVolatile, "element");
    Value *StoreGEP = LoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr, LoopPhi);
    LoopBuilder.CreateAlignedStore(Element, StoreGEP, PartDstAlign,
                                   DstIsVolatile);
    Value *Index = LoopBuilder.CreateAdd(LoopPhi, CILoopOpSize);
    LoopPhi->addIncoming(Index, LoopBB);
    LoopPhi->addIncoming(Zero, OrigBB);

    // Replace the unconditional branch to turn LoopBB into a loop.
    Instruction *UncondTerm = LoopBB->getTerminator();
    LoopBuilder.CreateCondBr(LoopBuilder.CreateICmpEQ(Index, LoopBound), SuccBB,
                             LoopBB);
    UncondTerm->eraseFromParent();
  }

  if (RemainingBytes != 0) {
    uint64_t BytesCopied = BytesCopiedInLoop;

    // Residual code is required to move the remaining bytes. In the forward
    // case, we emit it in the normal order.
    IRBuilder<> FwdResBuilder(FwdResidualBB->getTerminator());
    FwdResBuilder.SetCurrentDebugLocation(DbgLoc);
    SmallVector<Type *, 5> RemainingOps;
    TTI.getMemcpyLoopResidualLoweringType(RemainingOps, Ctx, RemainingBytes,
                                          SrcAS, DstAS, PartSrcAlign,
                                          PartDstAlign);
    for (auto *OpTy : RemainingOps)
      GenerateResidualLdStPair(OpTy, FwdResBuilder, BytesCopied);
  }
}

/// Create a Value of \p DstType that consists of a sequence of copies of
/// \p SetValue, using bitcasts and a vector splat.
static Value *createMemSetSplat(const DataLayout &DL, IRBuilderBase &B,
                                Value *SetValue, Type *DstType) {
  TypeSize DstSize = DL.getTypeStoreSize(DstType);
  Type *SetValueType = SetValue->getType();
  TypeSize SetValueSize = DL.getTypeStoreSize(SetValueType);
  assert(SetValueSize == DL.getTypeAllocSize(SetValueType) &&
         "Store size and alloc size of SetValue's type must match");
  assert(SetValueSize != 0 && DstSize % SetValueSize == 0 &&
         "DstType size must be a multiple of SetValue size");

  Value *Result = SetValue;
  if (DstSize != SetValueSize) {
    if (!SetValueType->isIntegerTy() && !SetValueType->isFloatingPointTy()) {
      // If the type cannot be put into a vector, bitcast to iN first.
      LLVMContext &Ctx = SetValue->getContext();
      Result = B.CreateBitCast(Result, Type::getIntNTy(Ctx, SetValueSize * 8),
                               "setvalue.toint");
    }
    // Form a sufficiently large vector consisting of SetValue, repeated.
    Result =
        B.CreateVectorSplat(DstSize / SetValueSize, Result, "setvalue.splat");
  }

  // The value has the right size, but we might have to bitcast it to the right
  // type.
  Result = B.CreateBitCast(Result, DstType, "setvalue.splat.cast");
  return Result;
}

static void
createMemSetLoopKnownSize(Instruction *InsertBefore, Value *DstAddr,
                          ConstantInt *Len, Value *SetValue, Align DstAlign,
                          bool IsVolatile, const TargetTransformInfo *TTI,
                          std::optional<uint64_t> AverageTripCount) {
  // No need to expand zero length memsets.
  if (Len->isZero())
    return;

  BasicBlock *PreLoopBB = InsertBefore->getParent();
  Function *ParentFunc = PreLoopBB->getParent();
  const DataLayout &DL = ParentFunc->getDataLayout();
  LLVMContext &Ctx = PreLoopBB->getContext();

  unsigned DstAS = cast<PointerType>(DstAddr->getType())->getAddressSpace();

  Type *TypeOfLen = Len->getType();
  Type *Int8Type = Type::getInt8Ty(Ctx);
  assert(SetValue->getType() == Int8Type && "Can only set bytes");

  Type *LoopOpType = Int8Type;
  if (TTI) {
    // Use the same memory access type as for a memcpy with the same Dst and Src
    // alignment and address space.
    LoopOpType = TTI->getMemcpyLoopLoweringType(
        Ctx, Len, DstAS, DstAS, DstAlign, DstAlign, std::nullopt);
  }
  TypeSize LoopOpSize = DL.getTypeStoreSize(LoopOpType);
  assert(LoopOpSize.isFixed() && "LoopOpType cannot be a scalable vector type");

  uint64_t LoopEndCount =
      alignDown(Len->getZExtValue(), LoopOpSize.getFixedValue());

  if (LoopEndCount != 0) {
    Value *SplatSetValue = nullptr;
    {
      IRBuilder<> PreLoopBuilder(InsertBefore);
      SplatSetValue =
          createMemSetSplat(DL, PreLoopBuilder, SetValue, LoopOpType);
    }

    // Don't generate a residual loop, the remaining bytes are set with
    // straight-line code.
    LoopExpansionInfo LEI = insertLoopExpansion(
        InsertBefore, Len, LoopOpSize, 0, "static-memset", AverageTripCount);

    // Fill MainLoopBB
    IRBuilder<> MainLoopBuilder(LEI.MainLoopIP);
    Align PartDstAlign(commonAlignment(DstAlign, LoopOpSize));

    Value *DstGEP =
        MainLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr, LEI.MainLoopIndex);

    MainLoopBuilder.CreateAlignedStore(SplatSetValue, DstGEP, PartDstAlign,
                                       IsVolatile);

    assert(!LEI.ResidualLoopIP && !LEI.ResidualLoopIndex &&
           "No residual loop was requested");
  }

  uint64_t BytesSet = LoopEndCount;
  uint64_t RemainingBytes = Len->getZExtValue() - BytesSet;
  if (RemainingBytes == 0)
    return;

  IRBuilder<> RBuilder(InsertBefore);

  assert(TTI && "there cannot be a residual loop without TTI");
  SmallVector<Type *, 5> RemainingOps;
  TTI->getMemcpyLoopResidualLoweringType(RemainingOps, Ctx, RemainingBytes,
                                         DstAS, DstAS, DstAlign, DstAlign,
                                         std::nullopt);

  Type *PreviousOpTy = nullptr;
  Value *SplatSetValue = nullptr;
  for (auto *OpTy : RemainingOps) {
    TypeSize OperandSize = DL.getTypeStoreSize(OpTy);
    assert(OperandSize.isFixed() &&
           "Operand types cannot be scalable vector types");
    Align PartDstAlign(commonAlignment(DstAlign, BytesSet));

    // Avoid recomputing the splat SetValue if it's the same as for the last
    // iteration.
    if (OpTy != PreviousOpTy)
      SplatSetValue = createMemSetSplat(DL, RBuilder, SetValue, OpTy);

    Value *DstGEP = RBuilder.CreateInBoundsGEP(
        Int8Type, DstAddr, ConstantInt::get(TypeOfLen, BytesSet));
    RBuilder.CreateAlignedStore(SplatSetValue, DstGEP, PartDstAlign,
                                IsVolatile);
    BytesSet += OperandSize;
    PreviousOpTy = OpTy;
  }
  assert(BytesSet == Len->getZExtValue() &&
         "Bytes set should match size in the call!");
}

static void
createMemSetLoopUnknownSize(Instruction *InsertBefore, Value *DstAddr,
                            Value *Len, Value *SetValue, Align DstAlign,
                            bool IsVolatile, const TargetTransformInfo *TTI,
                            std::optional<uint64_t> AverageTripCount) {
  BasicBlock *PreLoopBB = InsertBefore->getParent();
  Function *ParentFunc = PreLoopBB->getParent();
  const DataLayout &DL = ParentFunc->getDataLayout();
  LLVMContext &Ctx = PreLoopBB->getContext();

  unsigned DstAS = cast<PointerType>(DstAddr->getType())->getAddressSpace();

  Type *Int8Type = Type::getInt8Ty(Ctx);
  assert(SetValue->getType() == Int8Type && "Can only set bytes");

  Type *LoopOpType = Int8Type;
  if (TTI) {
    LoopOpType = TTI->getMemcpyLoopLoweringType(
        Ctx, Len, DstAS, DstAS, DstAlign, DstAlign, std::nullopt);
  }
  TypeSize LoopOpSize = DL.getTypeStoreSize(LoopOpType);
  assert(LoopOpSize.isFixed() && "LoopOpType cannot be a scalable vector type");

  Type *ResidualLoopOpType = Int8Type;
  TypeSize ResidualLoopOpSize = DL.getTypeStoreSize(ResidualLoopOpType);

  Value *SplatSetValue = SetValue;
  {
    IRBuilder<> PreLoopBuilder(InsertBefore);
    SplatSetValue = createMemSetSplat(DL, PreLoopBuilder, SetValue, LoopOpType);
  }

  LoopExpansionInfo LEI =
      insertLoopExpansion(InsertBefore, Len, LoopOpSize, ResidualLoopOpSize,
                          "dynamic-memset", AverageTripCount);

  // Fill MainLoopBB
  IRBuilder<> MainLoopBuilder(LEI.MainLoopIP);
  Align PartDstAlign(commonAlignment(DstAlign, LoopOpSize));

  Value *DstGEP =
      MainLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr, LEI.MainLoopIndex);
  MainLoopBuilder.CreateAlignedStore(SplatSetValue, DstGEP, PartDstAlign,
                                     IsVolatile);

  // Fill ResidualLoopBB
  if (!LEI.ResidualLoopIP)
    return;

  Align ResDstAlign(commonAlignment(PartDstAlign, ResidualLoopOpSize));

  IRBuilder<> ResLoopBuilder(LEI.ResidualLoopIP);

  Value *ResDstGEP = ResLoopBuilder.CreateInBoundsGEP(Int8Type, DstAddr,
                                                      LEI.ResidualLoopIndex);
  ResLoopBuilder.CreateAlignedStore(SetValue, ResDstGEP, ResDstAlign,
                                    IsVolatile);
}

static void createMemSetLoop(Instruction *InsertBefore, Value *DstAddr,
                             Value *CopyLen, Value *SetValue, Align DstAlign,
                             std::optional<uint64_t> AverageTripCount,
                             bool IsVolatile) {
  // Currently no longer used for memset, only for memset.pattern.
  // TODO: Update the memset.pattern lowering to also use the loop expansion
  //       framework and remove this function.
  Type *TypeOfCopyLen = CopyLen->getType();
  BasicBlock *OrigBB = InsertBefore->getParent();
  Function *F = OrigBB->getParent();
  const DataLayout &DL = F->getDataLayout();
  BasicBlock *NewBB =
      OrigBB->splitBasicBlock(InsertBefore, "split");
  BasicBlock *LoopBB
    = BasicBlock::Create(F->getContext(), "loadstoreloop", F, NewBB);

  const DebugLoc &DbgLoc = InsertBefore->getStableDebugLoc();
  IRBuilder<> Builder(OrigBB->getTerminator());
  Builder.SetCurrentDebugLocation(DbgLoc);

  auto *ToLoopBR = Builder.CreateCondBr(
      Builder.CreateICmpEQ(ConstantInt::get(TypeOfCopyLen, 0), CopyLen), NewBB,
      LoopBB);
  MDBuilder MDB(F->getContext());
  if (AverageTripCount.has_value())
    ToLoopBR->setMetadata(LLVMContext::MD_prof,
                          MDB.createLikelyBranchWeights());
  else
    setExplicitlyUnknownBranchWeightsIfProfiled(*ToLoopBR, DEBUG_TYPE);

  OrigBB->getTerminator()->eraseFromParent();

  TypeSize PartSize = DL.getTypeStoreSize(SetValue->getType());
  Align PartAlign(commonAlignment(DstAlign, PartSize));

  IRBuilder<> LoopBuilder(LoopBB);
  LoopBuilder.SetCurrentDebugLocation(DbgLoc);
  PHINode *LoopIndex = LoopBuilder.CreatePHI(TypeOfCopyLen, 0);
  LoopIndex->addIncoming(ConstantInt::get(TypeOfCopyLen, 0), OrigBB);

  LoopBuilder.CreateAlignedStore(
      SetValue,
      LoopBuilder.CreateInBoundsGEP(SetValue->getType(), DstAddr, LoopIndex),
      PartAlign, IsVolatile);

  Value *NewIndex =
      LoopBuilder.CreateAdd(LoopIndex, ConstantInt::get(TypeOfCopyLen, 1));
  LoopIndex->addIncoming(NewIndex, LoopBB);

  auto *LoopBR = LoopBuilder.CreateCondBr(
      LoopBuilder.CreateICmpULT(NewIndex, CopyLen), LoopBB, NewBB);
  if (AverageTripCount.has_value())
    setFittedBranchWeights(*LoopBR, {AverageTripCount.value(), 1},
                           /*IsExpected=*/false);
  else
    setExplicitlyUnknownBranchWeightsIfProfiled(*LoopBR, DEBUG_TYPE);
}

template <typename T>
static bool canOverlap(MemTransferBase<T> *Memcpy, ScalarEvolution *SE) {
  if (SE) {
    const SCEV *SrcSCEV = SE->getSCEV(Memcpy->getRawSource());
    const SCEV *DestSCEV = SE->getSCEV(Memcpy->getRawDest());
    if (SE->isKnownPredicateAt(CmpInst::ICMP_NE, SrcSCEV, DestSCEV, Memcpy))
      return false;
  }
  return true;
}

void llvm::expandMemCpyAsLoop(MemCpyInst *Memcpy,
                              const TargetTransformInfo &TTI,
                              ScalarEvolution *SE) {
  bool CanOverlap = canOverlap(Memcpy, SE);
  auto TripCount = getAverageMemOpLoopTripCount(*Memcpy);
  if (ConstantInt *CI = dyn_cast<ConstantInt>(Memcpy->getLength())) {
    createMemCpyLoopKnownSize(
        /*InsertBefore=*/Memcpy,
        /*SrcAddr=*/Memcpy->getRawSource(),
        /*DstAddr=*/Memcpy->getRawDest(),
        /*CopyLen=*/CI,
        /*SrcAlign=*/Memcpy->getSourceAlign().valueOrOne(),
        /*DstAlign=*/Memcpy->getDestAlign().valueOrOne(),
        /*SrcIsVolatile=*/Memcpy->isVolatile(),
        /*DstIsVolatile=*/Memcpy->isVolatile(),
        /*CanOverlap=*/CanOverlap,
        /*TTI=*/TTI,
        /*AtomicElementSize=*/std::nullopt,
        /*AverageTripCount=*/TripCount);
  } else {
    createMemCpyLoopUnknownSize(
        /*InsertBefore=*/Memcpy,
        /*SrcAddr=*/Memcpy->getRawSource(),
        /*DstAddr=*/Memcpy->getRawDest(),
        /*CopyLen=*/Memcpy->getLength(),
        /*SrcAlign=*/Memcpy->getSourceAlign().valueOrOne(),
        /*DstAlign=*/Memcpy->getDestAlign().valueOrOne(),
        /*SrcIsVolatile=*/Memcpy->isVolatile(),
        /*DstIsVolatile=*/Memcpy->isVolatile(),
        /*CanOverlap=*/CanOverlap,
        /*TTI=*/TTI,
        /*AtomicElementSize=*/std::nullopt,
        /*AverageTripCount=*/TripCount);
  }
}

bool llvm::expandMemMoveAsLoop(MemMoveInst *Memmove,
                               const TargetTransformInfo &TTI) {
  Value *CopyLen = Memmove->getLength();
  Value *SrcAddr = Memmove->getRawSource();
  Value *DstAddr = Memmove->getRawDest();
  Align SrcAlign = Memmove->getSourceAlign().valueOrOne();
  Align DstAlign = Memmove->getDestAlign().valueOrOne();
  bool SrcIsVolatile = Memmove->isVolatile();
  bool DstIsVolatile = SrcIsVolatile;
  IRBuilder<> CastBuilder(Memmove);
  CastBuilder.SetCurrentDebugLocation(Memmove->getStableDebugLoc());

  unsigned SrcAS = SrcAddr->getType()->getPointerAddressSpace();
  unsigned DstAS = DstAddr->getType()->getPointerAddressSpace();
  if (SrcAS != DstAS) {
    if (!TTI.addrspacesMayAlias(SrcAS, DstAS)) {
      // We may not be able to emit a pointer comparison, but we don't have
      // to. Expand as memcpy.
      auto AverageTripCount = getAverageMemOpLoopTripCount(*Memmove);
      if (ConstantInt *CI = dyn_cast<ConstantInt>(CopyLen)) {
        createMemCpyLoopKnownSize(
            /*InsertBefore=*/Memmove, SrcAddr, DstAddr, CI, SrcAlign, DstAlign,
            SrcIsVolatile, DstIsVolatile,
            /*CanOverlap=*/false, TTI, std::nullopt, AverageTripCount);
      } else {
        createMemCpyLoopUnknownSize(
            /*InsertBefore=*/Memmove, SrcAddr, DstAddr, CopyLen, SrcAlign,
            DstAlign, SrcIsVolatile, DstIsVolatile,
            /*CanOverlap=*/false, TTI, std::nullopt, AverageTripCount);
      }

      return true;
    }

    if (!(TTI.isValidAddrSpaceCast(DstAS, SrcAS) ||
          TTI.isValidAddrSpaceCast(SrcAS, DstAS))) {
      // We don't know generically if it's legal to introduce an
      // addrspacecast. We need to know either if it's legal to insert an
      // addrspacecast, or if the address spaces cannot alias.
      LLVM_DEBUG(
          dbgs() << "Do not know how to expand memmove between different "
                    "address spaces\n");
      return false;
    }
  }

  if (ConstantInt *CI = dyn_cast<ConstantInt>(CopyLen)) {
    createMemMoveLoopKnownSize(
        /*InsertBefore=*/Memmove, SrcAddr, DstAddr, CI, SrcAlign, DstAlign,
        SrcIsVolatile, DstIsVolatile, TTI);
  } else {
    createMemMoveLoopUnknownSize(
        /*InsertBefore=*/Memmove, SrcAddr, DstAddr, CopyLen, SrcAlign, DstAlign,
        SrcIsVolatile, DstIsVolatile, TTI);
  }
  return true;
}

void llvm::expandMemSetAsLoop(MemSetInst *Memset,
                              const TargetTransformInfo *TTI) {
  auto AverageTripCount = getAverageMemOpLoopTripCount(*Memset);
  if (ConstantInt *CI = dyn_cast<ConstantInt>(Memset->getLength())) {
    createMemSetLoopKnownSize(
        /*InsertBefore=*/Memset,
        /*DstAddr=*/Memset->getRawDest(),
        /*Len=*/CI,
        /*SetValue=*/Memset->getValue(),
        /*DstAlign=*/Memset->getDestAlign().valueOrOne(),
        /*IsVolatile=*/Memset->isVolatile(),
        /*TTI=*/TTI,
        /*AverageTripCount=*/AverageTripCount);
  } else {
    createMemSetLoopUnknownSize(
        /*InsertBefore=*/Memset,
        /*DstAddr=*/Memset->getRawDest(),
        /*Len=*/Memset->getLength(),
        /*SetValue=*/Memset->getValue(),
        /*DstAlign=*/Memset->getDestAlign().valueOrOne(),
        /*IsVolatile=*/Memset->isVolatile(),
        /*TTI=*/TTI,
        /*AverageTripCount=*/AverageTripCount);
  }
}

void llvm::expandMemSetAsLoop(MemSetInst *MemSet,
                              const TargetTransformInfo &TTI) {
  expandMemSetAsLoop(MemSet, &TTI);
}

void llvm::expandMemSetPatternAsLoop(MemSetPatternInst *Memset) {
  createMemSetLoop(/*InsertBefore=*/Memset,
                   /*DstAddr=*/Memset->getRawDest(),
                   /*CopyLen=*/Memset->getLength(),
                   /*SetValue=*/Memset->getValue(),
                   /*DstAlign=*/Memset->getDestAlign().valueOrOne(),
                   /*AverageTripCount=*/getAverageMemOpLoopTripCount(*Memset),
                   /*IsVolatile=*/Memset->isVolatile());
}

void llvm::expandAtomicMemCpyAsLoop(AnyMemCpyInst *AtomicMemcpy,
                                    const TargetTransformInfo &TTI,
                                    ScalarEvolution *SE) {
  assert(AtomicMemcpy->isAtomic());
  if (ConstantInt *CI = dyn_cast<ConstantInt>(AtomicMemcpy->getLength())) {
    createMemCpyLoopKnownSize(
        /*InsertBefore=*/AtomicMemcpy,
        /*SrcAddr=*/AtomicMemcpy->getRawSource(),
        /*DstAddr=*/AtomicMemcpy->getRawDest(),
        /*CopyLen=*/CI,
        /*SrcAlign=*/AtomicMemcpy->getSourceAlign().valueOrOne(),
        /*DstAlign=*/AtomicMemcpy->getDestAlign().valueOrOne(),
        /*SrcIsVolatile=*/AtomicMemcpy->isVolatile(),
        /*DstIsVolatile=*/AtomicMemcpy->isVolatile(),
        /*CanOverlap=*/false, // SrcAddr & DstAddr may not overlap by spec.
        /*TTI=*/TTI,
        /*AtomicElementSize=*/AtomicMemcpy->getElementSizeInBytes());
  } else {
    createMemCpyLoopUnknownSize(
        /*InsertBefore=*/AtomicMemcpy,
        /*SrcAddr=*/AtomicMemcpy->getRawSource(),
        /*DstAddr=*/AtomicMemcpy->getRawDest(),
        /*CopyLen=*/AtomicMemcpy->getLength(),
        /*SrcAlign=*/AtomicMemcpy->getSourceAlign().valueOrOne(),
        /*DstAlign=*/AtomicMemcpy->getDestAlign().valueOrOne(),
        /*SrcIsVolatile=*/AtomicMemcpy->isVolatile(),
        /*DstIsVolatile=*/AtomicMemcpy->isVolatile(),
        /*CanOverlap=*/false, // SrcAddr & DstAddr may not overlap by spec.
        /*TargetTransformInfo=*/TTI,
        /*AtomicElementSize=*/AtomicMemcpy->getElementSizeInBytes());
  }
}