llvm/lib/Transforms/Scalar/ExpandMemCmp.cpp

//===--- ExpandMemCmp.cpp - Expand memcmp() to load/stores ----------------===//
//
// 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
//
//===----------------------------------------------------------------------===//
//
// This pass tries to expand memcmp() calls into optimally-sized loads and
// compares for the target.
//
//===----------------------------------------------------------------------===//

#include "llvm/Transforms/Scalar/ExpandMemCmp.h"
#include "ScalarOptions.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/DomTreeUpdater.h"
#include "llvm/Analysis/LazyBlockFrequencyInfo.h"
#include "llvm/Analysis/ProfileSummaryInfo.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Analysis/ValueTracking.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/InstIterator.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/IR/ProfDataUtils.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/SizeOpts.h"
#include <optional>

using namespace llvm;
using namespace llvm::PatternMatch;

#define DEBUG_TYPE "expand-memcmp"

STATISTIC(NumMemCmpCalls, "Number of memcmp calls");
STATISTIC(NumMemCmpNotConstant, "Number of memcmp calls without constant size");
STATISTIC(NumMemCmpGreaterThanMax,
          "Number of memcmp calls with size greater than max size");
STATISTIC(NumMemCmpInlined, "Number of inlined memcmp calls");

namespace {

// Return the known alignment of the pointer argument \p ArgNo of \p CI,
// combining the alignment of the underlying pointer value with any align
// attribute on the call site itself.
static Align getMemCmpArgAlignment(const CallInst *CI, unsigned ArgNo,
                                   const DataLayout &DL) {
  Align A = CI->getArgOperand(ArgNo)->getPointerAlignment(DL);
  if (MaybeAlign ParamAlign = CI->getParamAlign(ArgNo))
    A = std::max(A, *ParamAlign);
  return A;
}

// This class provides helper functions to expand a memcmp library call into an
// inline expansion.
class MemCmpExpansion {
  struct LoadPair {
    Value *Lhs = nullptr;
    Value *Rhs = nullptr;
  };

  struct ResultBlock {
    BasicBlock *BB = nullptr;
    PHINode *PhiSrc1 = nullptr;
    PHINode *PhiSrc2 = nullptr;

    ResultBlock() = default;
  };

  CallInst *const CI = nullptr;
  ResultBlock ResBlock;
  const uint64_t Size;
  unsigned MaxLoadSize = 0;
  const uint64_t NumLoadsPerBlock;
  const unsigned MaxBytesPerBlock;
  unsigned MaxBlockSize = 0;
  std::vector<BasicBlock *> LoadCmpBlocks;
  BasicBlock *EndBlock = nullptr;
  PHINode *PhiRes = nullptr;
  const bool IsUsedForZeroCmp;
  const DataLayout &DL;
  const TargetTransformInfo &TTI;
  // The known common alignment of the two source pointers.
  const Align CommonAlign;
  DomTreeUpdater *DTU = nullptr;
  IRBuilder<> Builder;
  // Represents the decomposition in blocks of the expansion. For example,
  // comparing 33 bytes on X86+sse can be done with 2x16-byte loads and
  // 1x1-byte load, which would be represented as [{16, 0}, {16, 16}, {1, 32}.
  struct LoadEntry {
    LoadEntry(unsigned LoadSize, uint64_t Offset)
        : LoadSize(LoadSize), Offset(Offset) {
    }

    // The size of the load for this block, in bytes.
    unsigned LoadSize;
    // The offset of this load from the base pointer, in bytes.
    uint64_t Offset;
  };
  using LoadEntryVector = SmallVector<LoadEntry, 8>;
  LoadEntryVector LoadSequence;

  void createLoadCmpBlocks();
  void createResultBlock();
  void setupResultBlockPHINodes();
  void setupEndBlockPHINodes();
  Value *getCompareLoadPairs(unsigned BlockIndex, unsigned &LoadIndex);
  LoadPair getPackedLoadPair(unsigned BlockIndex, unsigned &LoadIndex);
  void emitLoadCompareBlock(unsigned BlockIndex, unsigned &LoadIndex);
  void emitLoadCompareBlockMultipleLoads(unsigned BlockIndex,
                                         unsigned &LoadIndex);
  void emitLoadCompareByteBlock(unsigned BlockIndex, unsigned OffsetBytes);
  void emitMemCmpResultBlock();
  Value *getMemCmpExpansionZeroCase();
  Value *getMemCmpEqZeroOneBlock();
  Value *getMemCmpOneBlock();
  Value *getMemCmpOneBlockMultipleLoads();
  Value *getMemCmpResult(const LoadPair &Loads);
  LoadPair getLoadPair(Type *LoadSizeType, Type *BSwapSizeType,
                       Type *CmpSizeType, unsigned OffsetBytes);

  // Return true if a load of `LoadSize` bytes at `Offset` from the base
  // pointers is accessible on the target: either it is naturally aligned given
  // the known common base alignment, or the target allows a misaligned access
  // of that width.
  bool isAccessAllowed(unsigned LoadSize, uint64_t Offset) const;

  static LoadEntryVector
  computeGreedyLoadSequence(uint64_t Size, llvm::ArrayRef<unsigned> LoadSizes,
                            unsigned MaxNumLoads);
  LoadEntryVector computeOverlappingLoadSequence(uint64_t Size,
                                                 unsigned MaxLoadSize,
                                                 unsigned MaxNumLoads) const;

  void optimiseLoadSequence(
      LoadEntryVector &LoadSequence,
      const TargetTransformInfo::MemCmpExpansionOptions &Options,
      bool IsUsedForZeroCmp) const;

public:
  MemCmpExpansion(CallInst *CI, uint64_t Size,
                  const TargetTransformInfo::MemCmpExpansionOptions &Options,
                  const bool IsUsedForZeroCmp, const DataLayout &TheDataLayout,
                  DomTreeUpdater *DTU, const TargetTransformInfo &TTI,
                  Align CommonAlign, unsigned MaxBytesPerBlock);

  unsigned getNumBlocks();
  unsigned getNumLoadsInBlock(unsigned LoadIndex) const;
  unsigned getNumBytesInBlock(unsigned LoadIndex, unsigned NumLoads) const;
  uint64_t getNumLoads() const { return LoadSequence.size(); }

  Value *getMemCmpExpansion();
};

// Return true if a load of `LoadSize` bytes at `Offset` from the base pointers
// is accessible on the target: either it is naturally aligned given the known
// common base alignment, or the target allows a misaligned access of that
// width. We query whether the access is *allowed*, not whether it is *fast*,
// matching the historical behavior of forming unaligned loads whenever the
// target permits them.
static bool isAccessAllowed(const CallInst *CI, const TargetTransformInfo &TTI,
                            Align CommonAlign, unsigned LoadSize,
                            uint64_t Offset) {
  // The access is naturally aligned when the known alignment is at least the
  // load width. LoadSize is not necessarily a power of two here: some targets
  // like RISC-V add non-power-of-two load sizes for vector memcmp, so compare
  // against the raw width rather than constructing an Align, which would
  // require a power of two.
  Align AccessAlign = commonAlignment(CommonAlign, Offset);
  if (AccessAlign.value() >= LoadSize)
    return true;
  unsigned AS = CI->getArgOperand(0)->getType()->getPointerAddressSpace();
  return TTI.allowsMisalignedMemoryAccesses(CI->getContext(), LoadSize * 8, AS,
                                            AccessAlign);
}

// Return true if a load of `LoadSize` bytes at `Offset` from the base pointers
// is accessible on the target given the known common base alignment. This gates
// the (power-of-two) overlapping loads; tail expansions are always legalized by
// the backend and skip this check.
bool MemCmpExpansion::isAccessAllowed(unsigned LoadSize,
                                      uint64_t Offset) const {
  return ::isAccessAllowed(CI, TTI, CommonAlign, LoadSize, Offset);
}

MemCmpExpansion::LoadEntryVector
MemCmpExpansion::computeGreedyLoadSequence(uint64_t Size,
                                           llvm::ArrayRef<unsigned> LoadSizes,
                                           const unsigned MaxNumLoads) {
  LoadEntryVector LoadSequence;
  uint64_t Offset = 0;
  while (Size && !LoadSizes.empty()) {
    const unsigned LoadSize = LoadSizes.front();
    const uint64_t NumLoadsForThisSize = Size / LoadSize;
    if (LoadSequence.size() + NumLoadsForThisSize > MaxNumLoads) {
      // Do not expand if the total number of loads is larger than what the
      // target allows. Note that it's important that we exit before completing
      // the expansion to avoid using a ton of memory to store the expansion for
      // large sizes.
      return {};
    }
    if (NumLoadsForThisSize > 0) {
      for (uint64_t I = 0; I < NumLoadsForThisSize; ++I) {
        LoadSequence.push_back({LoadSize, Offset});
        Offset += LoadSize;
      }
      Size = Size % LoadSize;
    }
    LoadSizes = LoadSizes.drop_front();
  }
  return LoadSequence;
}

MemCmpExpansion::LoadEntryVector
MemCmpExpansion::computeOverlappingLoadSequence(
    uint64_t Size, const unsigned MaxLoadSize,
    const unsigned MaxNumLoads) const {
  // These are already handled by the greedy approach.
  if (Size < 2 || MaxLoadSize < 2)
    return {};

  // We try to do as many non-overlapping loads as possible starting from the
  // beginning.
  const uint64_t NumNonOverlappingLoads = Size / MaxLoadSize;
  assert(NumNonOverlappingLoads && "there must be at least one load");
  // There remain 0 to (MaxLoadSize - 1) bytes to load, this will be done with
  // an overlapping load.
  Size = Size - NumNonOverlappingLoads * MaxLoadSize;
  // Bail if we do not need an overloapping store, this is already handled by
  // the greedy approach.
  if (Size == 0)
    return {};
  // Bail if the number of loads (non-overlapping + potential overlapping one)
  // is larger than the max allowed.
  if ((NumNonOverlappingLoads + 1) > MaxNumLoads)
    return {};

  // Add non-overlapping loads.
  LoadEntryVector LoadSequence;
  uint64_t Offset = 0;
  for (uint64_t I = 0; I < NumNonOverlappingLoads; ++I) {
    LoadSequence.push_back({MaxLoadSize, Offset});
    Offset += MaxLoadSize;
  }

  // Add the last overlapping load. Its offset is not a multiple of the load
  // size, so it may be misaligned; bail if the target cannot access it.
  assert(Size > 0 && Size < MaxLoadSize && "broken invariant");
  uint64_t OverlapOffset = Offset - (MaxLoadSize - Size);
  if (!isAccessAllowed(MaxLoadSize, OverlapOffset))
    return {};

  LoadSequence.push_back({MaxLoadSize, OverlapOffset});
  return LoadSequence;
}

void MemCmpExpansion::optimiseLoadSequence(
    LoadEntryVector &LoadSequence,
    const TargetTransformInfo::MemCmpExpansionOptions &Options,
    bool IsUsedForZeroCmp) const {
  // This part of code attempts to optimize the LoadSequence by merging allowed
  // subsequences into single loads of allowed sizes from
  // `MemCmpExpansionOptions::AllowedTailExpansions`. If it is for zero
  // comparison or if no allowed tail expansions are specified, we exit early.
  if (IsUsedForZeroCmp || Options.AllowedTailExpansions.empty())
    return;

  while (LoadSequence.size() >= 2) {
    auto Last = LoadSequence[LoadSequence.size() - 1];
    auto PreLast = LoadSequence[LoadSequence.size() - 2];

    // Exit the loop if the two sequences are not contiguous
    if (PreLast.Offset + PreLast.LoadSize != Last.Offset)
      break;

    auto LoadSize = Last.LoadSize + PreLast.LoadSize;
    if (find(Options.AllowedTailExpansions, LoadSize) ==
        Options.AllowedTailExpansions.end())
      break;

    // A merged load wider than MaxLoadSize can only be emitted when it is the
    // sole load (getMemCmpOneBlock); in a multi-block expansion
    // emitLoadCompareBlock requires every load to fit in MaxLoadSize (the
    // result-block phis are sized to it). The per-call-site alignment filter
    // can shrink MaxLoadSize, so stop merging when the result would still be
    // multi-block and the merged load exceeds it.
    if (LoadSize > MaxLoadSize && LoadSequence.size() > 2)
      break;

    // Remove the last two sequences and replace with the combined sequence
    LoadSequence.pop_back();
    LoadSequence.pop_back();
    LoadSequence.emplace_back(LoadSize, PreLast.Offset);
  }
}

// Initialize the basic block structure required for expansion of memcmp call
// with given maximum load size and memcmp size parameter.
// This structure includes:
// 1. A list of load compare blocks - LoadCmpBlocks.
// 2. An EndBlock, split from original instruction point, which is the block to
// return from.
// 3. ResultBlock, block to branch to for early exit when a
// LoadCmpBlock finds a difference.
MemCmpExpansion::MemCmpExpansion(
    CallInst *const CI, uint64_t Size,
    const TargetTransformInfo::MemCmpExpansionOptions &Options,
    const bool IsUsedForZeroCmp, const DataLayout &TheDataLayout,
    DomTreeUpdater *DTU, const TargetTransformInfo &TTI, Align CommonAlign,
    unsigned MaxBytesPerBlock)
    : CI(CI), Size(Size), NumLoadsPerBlock(Options.NumLoadsPerBlock),
      MaxBytesPerBlock(MaxBytesPerBlock), IsUsedForZeroCmp(IsUsedForZeroCmp),
      DL(TheDataLayout), TTI(TTI), CommonAlign(CommonAlign), DTU(DTU),
      Builder(CI) {
  assert(Size > 0 && "zero blocks");
  assert(NumLoadsPerBlock > 0 && "zero loads per block");
  // Scale the max size down if the target can load more bytes than we need.
  llvm::ArrayRef<unsigned> LoadSizes(Options.LoadSizes);
  while (!LoadSizes.empty() && LoadSizes.front() > Size) {
    LoadSizes = LoadSizes.drop_front();
  }
  assert(!LoadSizes.empty() && "cannot load Size bytes");
  MaxLoadSize = LoadSizes.front();
  // Compute the decomposition.
  LoadSequence =
      computeGreedyLoadSequence(Size, LoadSizes, Options.MaxNumLoads);
  assert(LoadSequence.size() <= Options.MaxNumLoads && "broken invariant");
  // If we allow overlapping loads and the load sequence is not already optimal,
  // use overlapping loads.
  if (Options.AllowOverlappingLoads &&
      (LoadSequence.empty() || LoadSequence.size() > 2)) {
    auto OverlappingLoads =
        computeOverlappingLoadSequence(Size, MaxLoadSize, Options.MaxNumLoads);
    if (!OverlappingLoads.empty() &&
        (LoadSequence.empty() ||
         OverlappingLoads.size() < LoadSequence.size())) {
      LoadSequence = OverlappingLoads;
    }
  }
  assert(LoadSequence.size() <= Options.MaxNumLoads && "broken invariant");
  optimiseLoadSequence(LoadSequence, Options, IsUsedForZeroCmp);

  unsigned LoadIndex = 0;
  while (LoadIndex < getNumLoads()) {
    unsigned NumLoads = getNumLoadsInBlock(LoadIndex);
    MaxBlockSize =
        std::max(MaxBlockSize, getNumBytesInBlock(LoadIndex, NumLoads));
    LoadIndex += NumLoads;
  }
  if (MaxBlockSize)
    MaxBlockSize = PowerOf2Ceil(MaxBlockSize);
}

unsigned MemCmpExpansion::getNumLoadsInBlock(unsigned LoadIndex) const {
  if (IsUsedForZeroCmp)
    return std::min<uint64_t>(getNumLoads() - LoadIndex, NumLoadsPerBlock);

  unsigned NumLoads = 0;
  unsigned NumBytes = 0;
  while (LoadIndex + NumLoads < getNumLoads() && NumLoads < NumLoadsPerBlock &&
         NumBytes + LoadSequence[LoadIndex + NumLoads].LoadSize <=
             MaxBytesPerBlock) {
    NumBytes += LoadSequence[LoadIndex + NumLoads].LoadSize;
    ++NumLoads;
  }
  assert(NumLoads && "at least one load must fit in a block");
  return NumLoads;
}

unsigned MemCmpExpansion::getNumBytesInBlock(unsigned LoadIndex,
                                             unsigned NumLoads) const {
  unsigned NumBytes = 0;
  for (unsigned I = 0; I != NumLoads; ++I)
    NumBytes += LoadSequence[LoadIndex + I].LoadSize;
  return NumBytes;
}

unsigned MemCmpExpansion::getNumBlocks() {
  unsigned NumBlocks = 0;
  for (unsigned LoadIndex = 0; LoadIndex < getNumLoads(); ++NumBlocks)
    LoadIndex += getNumLoadsInBlock(LoadIndex);
  return NumBlocks;
}

void MemCmpExpansion::createLoadCmpBlocks() {
  for (unsigned i = 0; i < getNumBlocks(); i++) {
    BasicBlock *BB = BasicBlock::Create(CI->getContext(), "loadbb",
                                        EndBlock->getParent(), EndBlock);
    LoadCmpBlocks.push_back(BB);
  }
}

void MemCmpExpansion::createResultBlock() {
  ResBlock.BB = BasicBlock::Create(CI->getContext(), "res_block",
                                   EndBlock->getParent(), EndBlock);
}

MemCmpExpansion::LoadPair MemCmpExpansion::getLoadPair(Type *LoadSizeType,
                                                       Type *BSwapSizeType,
                                                       Type *CmpSizeType,
                                                       unsigned OffsetBytes) {
  // Get the memory source at offset `OffsetBytes`.
  Value *LhsSource = CI->getArgOperand(0);
  Value *RhsSource = CI->getArgOperand(1);
  Align LhsAlign = getMemCmpArgAlignment(CI, 0, DL);
  Align RhsAlign = getMemCmpArgAlignment(CI, 1, DL);
  if (OffsetBytes > 0) {
    auto *ByteType = Type::getInt8Ty(CI->getContext());
    LhsSource = Builder.CreateConstGEP1_64(ByteType, LhsSource, OffsetBytes);
    RhsSource = Builder.CreateConstGEP1_64(ByteType, RhsSource, OffsetBytes);
    LhsAlign = commonAlignment(LhsAlign, OffsetBytes);
    RhsAlign = commonAlignment(RhsAlign, OffsetBytes);
  }

  // Create a constant or a load from the source.
  Value *Lhs = nullptr;
  if (auto *C = dyn_cast<Constant>(LhsSource))
    Lhs = ConstantFoldLoadFromConstPtr(C, LoadSizeType, DL);
  if (!Lhs)
    Lhs = Builder.CreateAlignedLoad(LoadSizeType, LhsSource, LhsAlign);

  Value *Rhs = nullptr;
  if (auto *C = dyn_cast<Constant>(RhsSource))
    Rhs = ConstantFoldLoadFromConstPtr(C, LoadSizeType, DL);
  if (!Rhs)
    Rhs = Builder.CreateAlignedLoad(LoadSizeType, RhsSource, RhsAlign);

  // Zero extend if Byte Swap intrinsic has different type
  if (BSwapSizeType && LoadSizeType != BSwapSizeType) {
    Lhs = Builder.CreateZExt(Lhs, BSwapSizeType);
    Rhs = Builder.CreateZExt(Rhs, BSwapSizeType);
  }

  // Swap bytes if required.
  if (BSwapSizeType) {
    Function *Bswap = Intrinsic::getOrInsertDeclaration(
        CI->getModule(), Intrinsic::bswap, BSwapSizeType);
    Lhs = Builder.CreateCall(Bswap, Lhs);
    Rhs = Builder.CreateCall(Bswap, Rhs);
  }

  // Zero extend if required.
  if (CmpSizeType != nullptr && CmpSizeType != Lhs->getType()) {
    Lhs = Builder.CreateZExt(Lhs, CmpSizeType);
    Rhs = Builder.CreateZExt(Rhs, CmpSizeType);
  }
  return {Lhs, Rhs};
}

// This function creates the IR instructions for loading and comparing 1 byte.
// It loads 1 byte from each source of the memcmp parameters with the given
// GEPIndex. It then subtracts the two loaded values and adds this result to the
// final phi node for selecting the memcmp result.
void MemCmpExpansion::emitLoadCompareByteBlock(unsigned BlockIndex,
                                               unsigned OffsetBytes) {
  BasicBlock *BB = LoadCmpBlocks[BlockIndex];
  Builder.SetInsertPoint(BB);
  const LoadPair Loads =
      getLoadPair(Type::getInt8Ty(CI->getContext()), nullptr,
                  Type::getInt32Ty(CI->getContext()), OffsetBytes);
  Value *Diff = Builder.CreateSub(Loads.Lhs, Loads.Rhs);

  PhiRes->addIncoming(Diff, BB);

  if (BlockIndex < (LoadCmpBlocks.size() - 1)) {
    // Early exit branch if difference found to EndBlock. Otherwise, continue to
    // next LoadCmpBlock,
    Value *Cmp = Builder.CreateICmp(ICmpInst::ICMP_NE, Diff,
                                    ConstantInt::get(Diff->getType(), 0));
    Builder.CreateCondBr(Cmp, EndBlock, LoadCmpBlocks[BlockIndex + 1]);
    if (DTU)
      DTU->applyUpdates(
          {{DominatorTree::Insert, BB, EndBlock},
           {DominatorTree::Insert, BB, LoadCmpBlocks[BlockIndex + 1]}});
  } else {
    // The last block has an unconditional branch to EndBlock.
    Builder.CreateBr(EndBlock);
    if (DTU)
      DTU->applyUpdates({{DominatorTree::Insert, BB, EndBlock}});
  }
}

/// Generate an equality comparison for one or more pairs of loaded values.
/// This is used in the case where the memcmp() call is compared equal or not
/// equal to zero.
Value *MemCmpExpansion::getCompareLoadPairs(unsigned BlockIndex,
                                            unsigned &LoadIndex) {
  assert(LoadIndex < getNumLoads() &&
         "getCompareLoadPairs() called with no remaining loads");
  std::vector<Value *> XorList, OrList;
  Value *Diff = nullptr;

  const unsigned NumLoads = getNumLoadsInBlock(LoadIndex);

  // For a single-block expansion, start inserting before the memcmp call.
  if (LoadCmpBlocks.empty())
    Builder.SetInsertPoint(CI);
  else
    Builder.SetInsertPoint(LoadCmpBlocks[BlockIndex]);

  Value *Cmp = nullptr;
  // If we have multiple loads per block, we need to generate a composite
  // comparison using xor+or. The type for the combinations is the largest load
  // type.
  IntegerType *const MaxLoadType =
      NumLoads == 1 ? nullptr
                    : IntegerType::get(CI->getContext(), MaxLoadSize * 8);

  for (unsigned i = 0; i < NumLoads; ++i, ++LoadIndex) {
    const LoadEntry &CurLoadEntry = LoadSequence[LoadIndex];
    const LoadPair Loads = getLoadPair(
        IntegerType::get(CI->getContext(), CurLoadEntry.LoadSize * 8), nullptr,
        MaxLoadType, CurLoadEntry.Offset);

    if (NumLoads != 1) {
      // If we have multiple loads per block, we need to generate a composite
      // comparison using xor+or.
      Diff = Builder.CreateXor(Loads.Lhs, Loads.Rhs);
      Diff = Builder.CreateZExt(Diff, MaxLoadType);
      XorList.push_back(Diff);
    } else {
      // If there's only one load per block, we just compare the loaded values.
      Cmp = Builder.CreateICmpNE(Loads.Lhs, Loads.Rhs);
    }
  }

  auto pairWiseOr = [&](std::vector<Value *> &InList) -> std::vector<Value *> {
    std::vector<Value *> OutList;
    for (unsigned i = 0; i < InList.size() - 1; i = i + 2) {
      Value *Or = Builder.CreateOr(InList[i], InList[i + 1]);
      OutList.push_back(Or);
    }
    if (InList.size() % 2 != 0)
      OutList.push_back(InList.back());
    return OutList;
  };

  if (!Cmp) {
    // Pairwise OR the XOR results.
    OrList = pairWiseOr(XorList);

    // Pairwise OR the OR results until one result left.
    while (OrList.size() != 1) {
      OrList = pairWiseOr(OrList);
    }

    assert(Diff && "Failed to find comparison diff");
    Cmp = Builder.CreateICmpNE(OrList[0], ConstantInt::get(Diff->getType(), 0));
  }

  return Cmp;
}

MemCmpExpansion::LoadPair
MemCmpExpansion::getPackedLoadPair(unsigned BlockIndex, unsigned &LoadIndex) {
  assert(LoadIndex < getNumLoads() &&
         "getPackedLoadPair() called with no remaining loads");
  if (LoadCmpBlocks.empty())
    Builder.SetInsertPoint(CI);
  else
    Builder.SetInsertPoint(LoadCmpBlocks[BlockIndex]);

  const unsigned NumLoads = getNumLoadsInBlock(LoadIndex);
  const unsigned NumBytes = getNumBytesInBlock(LoadIndex, NumLoads);
  // Pack the loads so that the byte at the lowest address occupies the most
  // significant bits. An unsigned comparison of the packed values therefore
  // has the same lexicographic ordering as memcmp.
  auto *BlockType = IntegerType::get(CI->getContext(), MaxBlockSize * 8);
  Value *PackedLhs = ConstantInt::get(BlockType, 0);
  Value *PackedRhs = ConstantInt::get(BlockType, 0);
  unsigned RemainingBytes = NumBytes;

  for (unsigned I = 0; I != NumLoads; ++I, ++LoadIndex) {
    const LoadEntry &Entry = LoadSequence[LoadIndex];
    auto *LoadType = IntegerType::get(CI->getContext(), Entry.LoadSize * 8);
    auto *BSwapType = DL.isLittleEndian() && Entry.LoadSize != 1
                          ? IntegerType::get(CI->getContext(),
                                             PowerOf2Ceil(Entry.LoadSize * 8))
                          : nullptr;
    LoadPair Loads = getLoadPair(LoadType, BSwapType, BlockType, Entry.Offset);

    if (BSwapType && BSwapType->getIntegerBitWidth() != Entry.LoadSize * 8) {
      unsigned Padding = BSwapType->getIntegerBitWidth() - Entry.LoadSize * 8;
      Loads.Lhs = Builder.CreateLShr(Loads.Lhs, Padding);
      Loads.Rhs = Builder.CreateLShr(Loads.Rhs, Padding);
    }

    RemainingBytes -= Entry.LoadSize;
    unsigned Shift = RemainingBytes * 8;
    PackedLhs =
        Builder.CreateOr(PackedLhs, Builder.CreateShl(Loads.Lhs, Shift));
    PackedRhs =
        Builder.CreateOr(PackedRhs, Builder.CreateShl(Loads.Rhs, Shift));
  }

  return {PackedLhs, PackedRhs};
}

void MemCmpExpansion::emitLoadCompareBlockMultipleLoads(unsigned BlockIndex,
                                                        unsigned &LoadIndex) {
  Value *Cmp = getCompareLoadPairs(BlockIndex, LoadIndex);

  BasicBlock *NextBB = (BlockIndex == (LoadCmpBlocks.size() - 1))
                           ? EndBlock
                           : LoadCmpBlocks[BlockIndex + 1];
  // Early exit branch if difference found to ResultBlock. Otherwise,
  // continue to next LoadCmpBlock or EndBlock.
  BasicBlock *BB = Builder.GetInsertBlock();
  CondBrInst *CmpBr = Builder.CreateCondBr(Cmp, ResBlock.BB, NextBB);
  setExplicitlyUnknownBranchWeightsIfProfiled(*CmpBr, DEBUG_TYPE,
                                              CI->getFunction());
  if (DTU)
    DTU->applyUpdates({{DominatorTree::Insert, BB, ResBlock.BB},
                       {DominatorTree::Insert, BB, NextBB}});

  // Add a phi edge for the last LoadCmpBlock to Endblock with a value of 0
  // since early exit to ResultBlock was not taken (no difference was found in
  // any of the bytes).
  if (BlockIndex == LoadCmpBlocks.size() - 1) {
    Value *Zero = ConstantInt::get(Type::getInt32Ty(CI->getContext()), 0);
    PhiRes->addIncoming(Zero, LoadCmpBlocks[BlockIndex]);
  }
}

// This function creates the IR intructions for loading and comparing using the
// given LoadSize. It loads the number of bytes specified by LoadSize from each
// source of the memcmp parameters. It then does a subtract to see if there was
// a difference in the loaded values. If a difference is found, it branches
// with an early exit to the ResultBlock for calculating which source was
// larger. Otherwise, it falls through to the either the next LoadCmpBlock or
// the EndBlock if this is the last LoadCmpBlock. Loading 1 byte is handled with
// a special case through emitLoadCompareByteBlock. The special handling can
// simply subtract the loaded values and add it to the result phi node.
void MemCmpExpansion::emitLoadCompareBlock(unsigned BlockIndex,
                                           unsigned &LoadIndex) {
  const unsigned NumLoads = getNumLoadsInBlock(LoadIndex);
  if (NumLoads == 1 && LoadSequence[LoadIndex].LoadSize == 1) {
    MemCmpExpansion::emitLoadCompareByteBlock(BlockIndex,
                                              LoadSequence[LoadIndex].Offset);
    ++LoadIndex;
    return;
  }

  LoadPair Loads;
  if (NumLoads == 1) {
    const LoadEntry &Entry = LoadSequence[LoadIndex++];
    auto *LoadType = IntegerType::get(CI->getContext(), Entry.LoadSize * 8);
    auto *BSwapType = DL.isLittleEndian()
                          ? IntegerType::get(CI->getContext(),
                                             PowerOf2Ceil(Entry.LoadSize * 8))
                          : nullptr;
    auto *CmpType = IntegerType::get(CI->getContext(), MaxBlockSize * 8);
    Builder.SetInsertPoint(LoadCmpBlocks[BlockIndex]);
    Loads = getLoadPair(LoadType, BSwapType, CmpType, Entry.Offset);
  } else {
    Loads = getPackedLoadPair(BlockIndex, LoadIndex);
  }

  // Add the loaded values to the phi nodes for calculating memcmp result only
  // if result is not used in a zero equality.
  if (!IsUsedForZeroCmp) {
    ResBlock.PhiSrc1->addIncoming(Loads.Lhs, LoadCmpBlocks[BlockIndex]);
    ResBlock.PhiSrc2->addIncoming(Loads.Rhs, LoadCmpBlocks[BlockIndex]);
  }

  Value *Cmp = Builder.CreateICmp(ICmpInst::ICMP_EQ, Loads.Lhs, Loads.Rhs);
  BasicBlock *NextBB = (BlockIndex == (LoadCmpBlocks.size() - 1))
                           ? EndBlock
                           : LoadCmpBlocks[BlockIndex + 1];
  // Early exit branch if difference found to ResultBlock. Otherwise, continue
  // to next LoadCmpBlock or EndBlock.
  BasicBlock *BB = Builder.GetInsertBlock();
  CondBrInst *CmpBr = Builder.CreateCondBr(Cmp, NextBB, ResBlock.BB);
  setExplicitlyUnknownBranchWeightsIfProfiled(*CmpBr, DEBUG_TYPE,
                                              CI->getFunction());
  if (DTU)
    DTU->applyUpdates({{DominatorTree::Insert, BB, NextBB},
                       {DominatorTree::Insert, BB, ResBlock.BB}});

  // Add a phi edge for the last LoadCmpBlock to Endblock with a value of 0
  // since early exit to ResultBlock was not taken (no difference was found in
  // any of the bytes).
  if (BlockIndex == LoadCmpBlocks.size() - 1) {
    Value *Zero = ConstantInt::get(Type::getInt32Ty(CI->getContext()), 0);
    PhiRes->addIncoming(Zero, LoadCmpBlocks[BlockIndex]);
  }
}

// This function populates the ResultBlock with a sequence to calculate the
// memcmp result. It compares the two loaded source values and returns -1 if
// src1 < src2 and 1 if src1 > src2.
void MemCmpExpansion::emitMemCmpResultBlock() {
  // Special case: if memcmp result is used in a zero equality, result does not
  // need to be calculated and can simply return 1.
  if (IsUsedForZeroCmp) {
    BasicBlock::iterator InsertPt = ResBlock.BB->getFirstInsertionPt();
    Builder.SetInsertPoint(InsertPt);
    Value *Res = ConstantInt::get(Type::getInt32Ty(CI->getContext()), 1);
    PhiRes->addIncoming(Res, ResBlock.BB);
    Builder.CreateBr(EndBlock);
    if (DTU)
      DTU->applyUpdates({{DominatorTree::Insert, ResBlock.BB, EndBlock}});
    return;
  }
  BasicBlock::iterator InsertPt = ResBlock.BB->getFirstInsertionPt();
  Builder.SetInsertPoint(InsertPt);

  Value *Cmp = Builder.CreateICmp(ICmpInst::ICMP_ULT, ResBlock.PhiSrc1,
                                  ResBlock.PhiSrc2);

  Value *Res =
      Builder.CreateSelect(Cmp, Constant::getAllOnesValue(Builder.getInt32Ty()),
                           ConstantInt::get(Builder.getInt32Ty(), 1));
  setExplicitlyUnknownBranchWeightsIfProfiled(*cast<Instruction>(Res),
                                              DEBUG_TYPE, CI->getFunction());

  PhiRes->addIncoming(Res, ResBlock.BB);
  Builder.CreateBr(EndBlock);
  if (DTU)
    DTU->applyUpdates({{DominatorTree::Insert, ResBlock.BB, EndBlock}});
}

void MemCmpExpansion::setupResultBlockPHINodes() {
  Type *MaxLoadType = IntegerType::get(CI->getContext(), MaxBlockSize * 8);
  Builder.SetInsertPoint(ResBlock.BB);
  ResBlock.PhiSrc1 = Builder.CreatePHI(MaxLoadType, getNumBlocks(), "phi.src1");
  ResBlock.PhiSrc2 = Builder.CreatePHI(MaxLoadType, getNumBlocks(), "phi.src2");
}

void MemCmpExpansion::setupEndBlockPHINodes() {
  Builder.SetInsertPoint(EndBlock->begin());
  PhiRes = Builder.CreatePHI(Type::getInt32Ty(CI->getContext()), 2, "phi.res");
}

Value *MemCmpExpansion::getMemCmpExpansionZeroCase() {
  unsigned LoadIndex = 0;
  // This loop populates each of the LoadCmpBlocks with the IR sequence to
  // handle multiple loads per block.
  for (unsigned I = 0; I < getNumBlocks(); ++I) {
    emitLoadCompareBlockMultipleLoads(I, LoadIndex);
  }

  emitMemCmpResultBlock();
  return PhiRes;
}

/// A memcmp expansion that compares equality with 0 and only has one block of
/// load and compare can bypass the compare, branch, and phi IR that is required
/// in the general case.
Value *MemCmpExpansion::getMemCmpEqZeroOneBlock() {
  unsigned LoadIndex = 0;
  Value *Cmp = getCompareLoadPairs(0, LoadIndex);
  assert(LoadIndex == getNumLoads() && "some entries were not consumed");
  return Builder.CreateZExt(Cmp, Type::getInt32Ty(CI->getContext()));
}

/// A memcmp expansion that only has one block of load and compare can bypass
/// the compare, branch, and phi IR that is required in the general case.
/// This function also analyses users of memcmp, and if there is only one user
/// from which we can conclude that only 2 out of 3 memcmp outcomes really
/// matter, then it generates more efficient code with only one comparison.
Value *MemCmpExpansion::getMemCmpOneBlock() {
  bool NeedsBSwap = DL.isLittleEndian() && Size != 1;
  Type *LoadSizeType = IntegerType::get(CI->getContext(), Size * 8);
  Type *BSwapSizeType =
      NeedsBSwap ? IntegerType::get(CI->getContext(), PowerOf2Ceil(Size * 8))
                 : nullptr;
  Type *MaxLoadType =
      IntegerType::get(CI->getContext(),
                       std::max(MaxLoadSize, (unsigned)PowerOf2Ceil(Size)) * 8);

  // The i8 and i16 cases don't need compares. We zext the loaded values and
  // subtract them to get the suitable negative, zero, or positive i32 result.
  if (Size == 1 || Size == 2) {
    const LoadPair Loads = getLoadPair(LoadSizeType, BSwapSizeType,
                                       Builder.getInt32Ty(), /*Offset*/ 0);
    return Builder.CreateSub(Loads.Lhs, Loads.Rhs);
  }

  const LoadPair Loads = getLoadPair(LoadSizeType, BSwapSizeType, MaxLoadType,
                                     /*Offset*/ 0);

  return getMemCmpResult(Loads);
}

Value *MemCmpExpansion::getMemCmpOneBlockMultipleLoads() {
  unsigned LoadIndex = 0;
  LoadPair Loads = getPackedLoadPair(/*BlockIndex=*/0, LoadIndex);
  assert(LoadIndex == getNumLoads() && "some entries were not consumed");
  return getMemCmpResult(Loads);
}

Value *MemCmpExpansion::getMemCmpResult(const LoadPair &Loads) {
  // If a user of memcmp cares only about two outcomes, for example:
  //    bool result = memcmp(a, b, NBYTES) > 0;
  // We can generate more optimal code with a smaller number of operations
  if (CI->hasOneUser()) {
    auto *UI = cast<Instruction>(*CI->user_begin());
    CmpPredicate Pred = ICmpInst::Predicate::BAD_ICMP_PREDICATE;
    bool NeedsZExt = false;
    // This is a special case because instead of checking if the result is less
    // than zero:
    //    bool result = memcmp(a, b, NBYTES) < 0;
    // Compiler is clever enough to generate the following code:
    //    bool result = memcmp(a, b, NBYTES) >> 31;
    if (match(UI,
              m_LShr(m_Value(),
                     m_SpecificInt(CI->getType()->getIntegerBitWidth() - 1)))) {
      Pred = ICmpInst::ICMP_SLT;
      NeedsZExt = true;
    } else if (match(UI, m_SpecificICmp(ICmpInst::ICMP_SGT, m_Specific(CI),
                                        m_AllOnes()))) {
      // Adjust predicate as if it compared with 0.
      Pred = ICmpInst::ICMP_SGE;
    } else if (match(UI, m_SpecificICmp(ICmpInst::ICMP_SLT, m_Specific(CI),
                                        m_One()))) {
      // Adjust predicate as if it compared with 0.
      Pred = ICmpInst::ICMP_SLE;
    } else {
      // In case of a successful match this call will set `Pred` variable
      match(UI, m_ICmp(Pred, m_Specific(CI), m_Zero()));
    }
    // Generate new code and remove the original memcmp call and the user
    if (ICmpInst::isSigned(Pred)) {
      Value *Cmp = Builder.CreateICmp(ICmpInst::getUnsignedPredicate(Pred),
                                      Loads.Lhs, Loads.Rhs);
      auto *Result = NeedsZExt ? Builder.CreateZExt(Cmp, UI->getType()) : Cmp;
      UI->replaceAllUsesWith(Result);
      UI->eraseFromParent();
      CI->eraseFromParent();
      return nullptr;
    }
  }

  // The result of memcmp is negative, zero, or positive.
  return Builder.CreateIntrinsic(Builder.getInt32Ty(), Intrinsic::ucmp,
                                 {Loads.Lhs, Loads.Rhs});
}

// This function expands the memcmp call into an inline expansion and returns
// the memcmp result. Returns nullptr if the memcmp is already replaced.
Value *MemCmpExpansion::getMemCmpExpansion() {
  // Create the basic block framework for a multi-block expansion.
  if (getNumBlocks() != 1) {
    BasicBlock *StartBlock = CI->getParent();
    EndBlock = SplitBlock(StartBlock, CI, DTU, /*LI=*/nullptr,
                          /*MSSAU=*/nullptr, "endblock");
    setupEndBlockPHINodes();
    createResultBlock();

    // If return value of memcmp is not used in a zero equality, we need to
    // calculate which source was larger. The calculation requires the
    // two loaded source values of each load compare block.
    // These will be saved in the phi nodes created by setupResultBlockPHINodes.
    if (!IsUsedForZeroCmp) setupResultBlockPHINodes();

    // Create the number of required load compare basic blocks.
    createLoadCmpBlocks();

    // Update the terminator added by SplitBlock to branch to the first
    // LoadCmpBlock.
    StartBlock->getTerminator()->setSuccessor(0, LoadCmpBlocks[0]);
    if (DTU)
      DTU->applyUpdates({{DominatorTree::Insert, StartBlock, LoadCmpBlocks[0]},
                         {DominatorTree::Delete, StartBlock, EndBlock}});
  }

  Builder.SetCurrentDebugLocation(CI->getDebugLoc());

  if (IsUsedForZeroCmp)
    return getNumBlocks() == 1 ? getMemCmpEqZeroOneBlock()
                               : getMemCmpExpansionZeroCase();

  if (getNumBlocks() == 1)
    return getNumLoads() == 1 ? getMemCmpOneBlock()
                              : getMemCmpOneBlockMultipleLoads();

  unsigned LoadIndex = 0;
  for (unsigned I = 0; I < getNumBlocks(); ++I) {
    emitLoadCompareBlock(I, LoadIndex);
  }

  emitMemCmpResultBlock();
  return PhiRes;
}

// This function checks to see if an expansion of memcmp can be generated.
// It checks for constant compare size that is less than the max inline size.
// If an expansion cannot occur, returns false to leave as a library call.
// Otherwise, the library call is replaced with a new IR instruction sequence.
/// We want to transform:
/// %call = call signext i32 @memcmp(i8* %0, i8* %1, i64 15)
/// To:
/// loadbb:
///  %0 = bitcast i32* %buffer2 to i8*
///  %1 = bitcast i32* %buffer1 to i8*
///  %2 = bitcast i8* %1 to i64*
///  %3 = bitcast i8* %0 to i64*
///  %4 = load i64, i64* %2
///  %5 = load i64, i64* %3
///  %6 = call i64 @llvm.bswap.i64(i64 %4)
///  %7 = call i64 @llvm.bswap.i64(i64 %5)
///  %8 = sub i64 %6, %7
///  %9 = icmp ne i64 %8, 0
///  br i1 %9, label %res_block, label %loadbb1
/// res_block:                                        ; preds = %loadbb2,
/// %loadbb1, %loadbb
///  %phi.src1 = phi i64 [ %6, %loadbb ], [ %22, %loadbb1 ], [ %36, %loadbb2 ]
///  %phi.src2 = phi i64 [ %7, %loadbb ], [ %23, %loadbb1 ], [ %37, %loadbb2 ]
///  %10 = icmp ult i64 %phi.src1, %phi.src2
///  %11 = select i1 %10, i32 -1, i32 1
///  br label %endblock
/// loadbb1:                                          ; preds = %loadbb
///  %12 = bitcast i32* %buffer2 to i8*
///  %13 = bitcast i32* %buffer1 to i8*
///  %14 = bitcast i8* %13 to i32*
///  %15 = bitcast i8* %12 to i32*
///  %16 = getelementptr i32, i32* %14, i32 2
///  %17 = getelementptr i32, i32* %15, i32 2
///  %18 = load i32, i32* %16
///  %19 = load i32, i32* %17
///  %20 = call i32 @llvm.bswap.i32(i32 %18)
///  %21 = call i32 @llvm.bswap.i32(i32 %19)
///  %22 = zext i32 %20 to i64
///  %23 = zext i32 %21 to i64
///  %24 = sub i64 %22, %23
///  %25 = icmp ne i64 %24, 0
///  br i1 %25, label %res_block, label %loadbb2
/// loadbb2:                                          ; preds = %loadbb1
///  %26 = bitcast i32* %buffer2 to i8*
///  %27 = bitcast i32* %buffer1 to i8*
///  %28 = bitcast i8* %27 to i16*
///  %29 = bitcast i8* %26 to i16*
///  %30 = getelementptr i16, i16* %28, i16 6
///  %31 = getelementptr i16, i16* %29, i16 6
///  %32 = load i16, i16* %30
///  %33 = load i16, i16* %31
///  %34 = call i16 @llvm.bswap.i16(i16 %32)
///  %35 = call i16 @llvm.bswap.i16(i16 %33)
///  %36 = zext i16 %34 to i64
///  %37 = zext i16 %35 to i64
///  %38 = sub i64 %36, %37
///  %39 = icmp ne i64 %38, 0
///  br i1 %39, label %res_block, label %loadbb3
/// loadbb3:                                          ; preds = %loadbb2
///  %40 = bitcast i32* %buffer2 to i8*
///  %41 = bitcast i32* %buffer1 to i8*
///  %42 = getelementptr i8, i8* %41, i8 14
///  %43 = getelementptr i8, i8* %40, i8 14
///  %44 = load i8, i8* %42
///  %45 = load i8, i8* %43
///  %46 = zext i8 %44 to i32
///  %47 = zext i8 %45 to i32
///  %48 = sub i32 %46, %47
///  br label %endblock
/// endblock:                                         ; preds = %res_block,
/// %loadbb3
///  %phi.res = phi i32 [ %48, %loadbb3 ], [ %11, %res_block ]
///  ret i32 %phi.res
static bool expandMemCmp(const ScalarOptions &Opts, CallInst *CI,
                         const TargetTransformInfo *TTI, const DataLayout *DL,
                         ProfileSummaryInfo *PSI, BlockFrequencyInfo *BFI,
                         DomTreeUpdater *DTU, const bool IsBCmp) {
  NumMemCmpCalls++;

  // Early exit from expansion if -Oz.
  if (CI->getFunction()->hasMinSize())
    return false;

  // Early exit from expansion if size is not a constant.
  ConstantInt *SizeCast = dyn_cast<ConstantInt>(CI->getArgOperand(2));
  if (!SizeCast) {
    NumMemCmpNotConstant++;
    return false;
  }
  const uint64_t SizeVal = SizeCast->getZExtValue();

  if (SizeVal == 0) {
    return false;
  }
  // TTI call to check if target would like to expand memcmp. Also, get the
  // available load sizes.
  const bool IsUsedForZeroCmp =
      IsBCmp || isOnlyUsedInZeroEqualityComparison(CI);
  bool OptForSize = llvm::shouldOptimizeForSize(CI->getParent(), PSI, BFI);
  auto Options = TTI->enableMemCmpExpansion(OptForSize,
                                            IsUsedForZeroCmp);
  if (!Options) return false;

  if (Opts.memcmp_num_loads_per_block)
    Options.NumLoadsPerBlock = *Opts.memcmp_num_loads_per_block;

  if (OptForSize && Opts.max_loads_per_memcmp_opt_size)
    Options.MaxNumLoads = *Opts.max_loads_per_memcmp_opt_size;

  if (!OptForSize && Opts.max_loads_per_memcmp)
    Options.MaxNumLoads = *Opts.max_loads_per_memcmp;

  // Keep only the load sizes the target can access at the base alignment:
  // either the access is naturally aligned, or the target allows a misaligned
  // access of that width. This lets strict-alignment targets expand compares
  // whose pointers happen to be sufficiently aligned, while still falling back
  // to the libcall when no load size fits. Because the greedy load sequence
  // only places a load of size S at an offset that is a multiple of S, a size
  // kept here is always accessible in that sequence; overlapping loads and
  // merged tail expansions are checked separately against their actual offsets
  // in MemCmpExpansion.
  const Align CommonAlign = std::min(getMemCmpArgAlignment(CI, 0, *DL),
                                     getMemCmpArgAlignment(CI, 1, *DL));
  // Remember the target's preferred load width before filtering inaccessible
  // sizes. It is also the maximum width of the value formed by packing the
  // surviving loads in one ordering-compare block.
  const unsigned MaxBytesPerBlock = Options.LoadSizes.front();
  llvm::erase_if(Options.LoadSizes, [&](unsigned LoadSize) {
    return !isAccessAllowed(CI, *TTI, CommonAlign, LoadSize, /*Offset=*/0);
  });
  // If the filter removed every load size, bail out to the libcall: the
  // MemCmpExpansion constructor asserts that at least one load size remains.
  // In practice all in-tree targets include a byte load size, which is
  // accessible at any alignment and therefore always survives the filter.
  if (Options.LoadSizes.empty())
    return false;

  MemCmpExpansion Expansion(CI, SizeVal, Options, IsUsedForZeroCmp, *DL, DTU,
                            *TTI, CommonAlign, MaxBytesPerBlock);

  // Don't expand if this will require more loads than desired by the target.
  if (Expansion.getNumLoads() == 0) {
    NumMemCmpGreaterThanMax++;
    return false;
  }

  NumMemCmpInlined++;

  if (Value *Res = Expansion.getMemCmpExpansion()) {
    // Replace call with result of expansion and erase call.
    CI->replaceAllUsesWith(Res);
    CI->eraseFromParent();
  }

  return true;
}

static PreservedAnalyses runImpl(Function &F, const TargetLibraryInfo *TLI,
                                 const TargetTransformInfo *TTI,
                                 ProfileSummaryInfo *PSI,
                                 BlockFrequencyInfo *BFI, DominatorTree *DT) {
  std::optional<DomTreeUpdater> DTU;
  if (DT)
    DTU.emplace(DT, DomTreeUpdater::UpdateStrategy::Lazy);

  const DataLayout& DL = F.getDataLayout();
  SmallVector<std::pair<CallInst *, LibFunc>, 8> MemCmpCalls;
  for (Instruction &I : instructions(F)) {
    if (auto *CI = dyn_cast<CallInst>(&I)) {
      LibFunc Func = TLI->getLibFunc(*CI);
      if (Func == LibFunc_memcmp || Func == LibFunc_bcmp)
        MemCmpCalls.push_back({CI, Func});
    }
  }

  const ScalarOptions &Opts =
      F.getContext().getLibraryOptions().get<ScalarOptions>();
  bool MadeChanges = false;
  for (const auto &[CI, Func] : MemCmpCalls) {
    if (expandMemCmp(Opts, CI, TTI, &DL, PSI, BFI, DTU ? &*DTU : nullptr,
                     Func == LibFunc_bcmp))
      MadeChanges = true;
  }

  if (MadeChanges)
    for (BasicBlock &BB : F)
      SimplifyInstructionsInBlock(&BB);
  if (!MadeChanges)
    return PreservedAnalyses::all();
  PreservedAnalyses PA;
  PA.preserve<DominatorTreeAnalysis>();
  return PA;
}

} // namespace

PreservedAnalyses ExpandMemCmpPass::run(Function &F,
                                        FunctionAnalysisManager &FAM) {
  // Don't expand memcmp in sanitized functions — sanitizers intercept memcmp
  // calls to check for memory errors, and expanding would bypass that.
  if (F.hasFnAttribute(Attribute::SanitizeAddress) ||
      F.hasFnAttribute(Attribute::SanitizeMemory) ||
      F.hasFnAttribute(Attribute::SanitizeThread) ||
      F.hasFnAttribute(Attribute::SanitizeHWAddress))
    return PreservedAnalyses::all();

  const auto &TLI = FAM.getResult<TargetLibraryAnalysis>(F);
  const auto &TTI = FAM.getResult<TargetIRAnalysis>(F);
  auto *PSI = FAM.getResult<ModuleAnalysisManagerFunctionProxy>(F)
                  .getCachedResult<ProfileSummaryAnalysis>(*F.getParent());
  BlockFrequencyInfo *BFI = (PSI && PSI->hasProfileSummary())
                                ? &FAM.getResult<BlockFrequencyAnalysis>(F)
                                : nullptr;
  auto *DT = FAM.getCachedResult<DominatorTreeAnalysis>(F);

  return runImpl(F, &TLI, &TTI, PSI, BFI, DT);
}