Transforms/IPO/PartialInlining.cpp

//===- PartialInlining.cpp - Inline parts of functions --------------------===//
//
// 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 performs partial inlining, typically by inlining an if statement
// that surrounds the body of the function.
//
//===----------------------------------------------------------------------===//

#include "llvm/Transforms/IPO/PartialInlining.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/DepthFirstIterator.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/BlockFrequencyInfo.h"
#include "llvm/Analysis/BranchProbabilityInfo.h"
#include "llvm/Analysis/InlineCost.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/OptimizationRemarkEmitter.h"
#include "llvm/Analysis/ProfileSummaryInfo.h"
#include "llvm/Analysis/TargetLibraryInfo.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/IR/Attributes.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/CFG.h"
#include "llvm/IR/DebugLoc.h"
#include "llvm/IR/DiagnosticInfo.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/InstrTypes.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/Module.h"
#include "llvm/IR/Operator.h"
#include "llvm/IR/ProfDataUtils.h"
#include "llvm/IR/User.h"
#include "llvm/Support/BlockFrequency.h"
#include "llvm/Support/BranchProbability.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Transforms/IPO.h"
#include "llvm/Transforms/Utils/Cloning.h"
#include "llvm/Transforms/Utils/CodeExtractor.h"
#include "llvm/Transforms/Utils/ValueMapper.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <memory>
#include <tuple>
#include <vector>

using namespace llvm;

#define DEBUG_TYPE "partial-inlining"

STATISTIC(NumPartialInlined,
          "Number of callsites functions partially inlined into.");
STATISTIC(NumColdOutlinePartialInlined, "Number of times functions with "
                                        "cold outlined regions were partially "
                                        "inlined into its caller(s).");
STATISTIC(NumColdRegionsFound,
           "Number of cold single entry/exit regions found.");
STATISTIC(NumColdRegionsOutlined,
           "Number of cold single entry/exit regions outlined.");

// Command line option to disable partial-inlining. The default is false:
static cl::opt<bool>
    DisablePartialInlining("disable-partial-inlining", cl::init(false),
                           cl::Hidden, cl::desc("Disable partial inlining"));
// Command line option to disable multi-region partial-inlining. The default is
// false:
static cl::opt<bool> DisableMultiRegionPartialInline(
    "disable-mr-partial-inlining", cl::init(false), cl::Hidden,
    cl::desc("Disable multi-region partial inlining"));

// Command line option to force outlining in regions with live exit variables.
// The default is false:
static cl::opt<bool>
    ForceLiveExit("pi-force-live-exit-outline", cl::init(false), cl::Hidden,
               cl::desc("Force outline regions with live exits"));

// Command line option to enable marking outline functions with Cold Calling
// Convention. The default is false:
static cl::opt<bool>
    MarkOutlinedColdCC("pi-mark-coldcc", cl::init(false), cl::Hidden,
                       cl::desc("Mark outline function calls with ColdCC"));

// This is an option used by testing:
static cl::opt<bool> SkipCostAnalysis("skip-partial-inlining-cost-analysis",

                                      cl::ReallyHidden,
                                      cl::desc("Skip Cost Analysis"));
// Used to determine if a cold region is worth outlining based on
// its inlining cost compared to the original function.  Default is set at 10%.
// ie. if the cold region reduces the inlining cost of the original function by
// at least 10%.
static cl::opt<float> MinRegionSizeRatio(
    "min-region-size-ratio", cl::init(0.1), cl::Hidden,
    cl::desc("Minimum ratio comparing relative sizes of each "
             "outline candidate and original function"));
// Used to tune the minimum number of execution counts needed in the predecessor
// block to the cold edge. ie. confidence interval.
static cl::opt<unsigned>
    MinBlockCounterExecution("min-block-execution", cl::init(100), cl::Hidden,
                             cl::desc("Minimum block executions to consider "
                                      "its BranchProbabilityInfo valid"));
// Used to determine when an edge is considered cold. Default is set to 10%. ie.
// if the branch probability is 10% or less, then it is deemed as 'cold'.
static cl::opt<float> ColdBranchRatio(
    "cold-branch-ratio", cl::init(0.1), cl::Hidden,
    cl::desc("Minimum BranchProbability to consider a region cold."));

static cl::opt<unsigned> MaxNumInlineBlocks(
    "max-num-inline-blocks", cl::init(5), cl::Hidden,
    cl::desc("Max number of blocks to be partially inlined"));

// Command line option to set the maximum number of partial inlining allowed
// for the module. The default value of -1 means no limit.
static cl::opt<int> MaxNumPartialInlining(
    "max-partial-inlining", cl::init(-1), cl::Hidden,
    cl::desc("Max number of partial inlining. The default is unlimited"));

// Used only when PGO or user annotated branch data is absent. It is
// the least value that is used to weigh the outline region. If BFI
// produces larger value, the BFI value will be used.
static cl::opt<int>
    OutlineRegionFreqPercent("outline-region-freq-percent", cl::init(75),
                             cl::Hidden,
                             cl::desc("Relative frequency of outline region to "
                                      "the entry block"));

static cl::opt<unsigned> ExtraOutliningPenalty(
    "partial-inlining-extra-penalty", cl::init(0), cl::Hidden,
    cl::desc("A debug option to add additional penalty to the computed one."));

namespace {

struct FunctionOutliningInfo {
  FunctionOutliningInfo() = default;

  // Returns the number of blocks to be inlined including all blocks
  // in Entries and one return block.
  unsigned getNumInlinedBlocks() const { return Entries.size() + 1; }

  // A set of blocks including the function entry that guard
  // the region to be outlined.
  SmallVector<BasicBlock *, 4> Entries;

  // The return block that is not included in the outlined region.
  BasicBlock *ReturnBlock = nullptr;

  // The dominating block of the region to be outlined.
  BasicBlock *NonReturnBlock = nullptr;

  // The set of blocks in Entries that are predecessors to ReturnBlock
  SmallVector<BasicBlock *, 4> ReturnBlockPreds;
};

struct FunctionOutliningMultiRegionInfo {
  FunctionOutliningMultiRegionInfo() = default;

  // Container for outline regions
  struct OutlineRegionInfo {
    OutlineRegionInfo(ArrayRef<BasicBlock *> Region, BasicBlock *EntryBlock,
                      BasicBlock *ExitBlock, BasicBlock *ReturnBlock)
        : Region(Region), EntryBlock(EntryBlock), ExitBlock(ExitBlock),
          ReturnBlock(ReturnBlock) {}
    SmallVector<BasicBlock *, 8> Region;
    BasicBlock *EntryBlock;
    BasicBlock *ExitBlock;
    BasicBlock *ReturnBlock;
  };

  SmallVector<OutlineRegionInfo, 4> ORI;
};

struct PartialInlinerImpl {

  PartialInlinerImpl(
      function_ref<AssumptionCache &(Function &)> GetAC,
      function_ref<AssumptionCache *(Function &)> LookupAC,
      function_ref<TargetTransformInfo &(Function &)> GTTI,
      function_ref<const TargetLibraryInfo &(Function &)> GTLI,
      ProfileSummaryInfo &ProfSI,
      function_ref<BlockFrequencyInfo &(Function &)> GBFI = nullptr)
      : GetAssumptionCache(GetAC), LookupAssumptionCache(LookupAC),
        GetTTI(GTTI), GetBFI(GBFI), GetTLI(GTLI), PSI(ProfSI) {}

  bool run(Module &M);
  // Main part of the transformation that calls helper functions to find
  // outlining candidates, clone & outline the function, and attempt to
  // partially inline the resulting function. Returns true if
  // inlining was successful, false otherwise.  Also returns the outline
  // function (only if we partially inlined early returns) as there is a
  // possibility to further "peel" early return statements that were left in the
  // outline function due to code size.
  std::pair<bool, Function *> unswitchFunction(Function &F);

  // This class speculatively clones the function to be partial inlined.
  // At the end of partial inlining, the remaining callsites to the cloned
  // function that are not partially inlined will be fixed up to reference
  // the original function, and the cloned function will be erased.
  struct FunctionCloner {
    // Two constructors, one for single region outlining, the other for
    // multi-region outlining.
    FunctionCloner(Function *F, FunctionOutliningInfo *OI,
                   OptimizationRemarkEmitter &ORE,
                   function_ref<AssumptionCache *(Function &)> LookupAC,
                   function_ref<TargetTransformInfo &(Function &)> GetTTI);
    FunctionCloner(Function *F, FunctionOutliningMultiRegionInfo *OMRI,
                   OptimizationRemarkEmitter &ORE,
                   function_ref<AssumptionCache *(Function &)> LookupAC,
                   function_ref<TargetTransformInfo &(Function &)> GetTTI);

    ~FunctionCloner();

    // Prepare for function outlining: making sure there is only
    // one incoming edge from the extracted/outlined region to
    // the return block.
    void normalizeReturnBlock() const;

    // Do function outlining for cold regions.
    bool doMultiRegionFunctionOutlining();
    // Do function outlining for region after early return block(s).
    // NOTE: For vararg functions that do the vararg handling in the outlined
    //       function, we temporarily generate IR that does not properly
    //       forward varargs to the outlined function. Calling InlineFunction
    //       will update calls to the outlined functions to properly forward
    //       the varargs.
    Function *doSingleRegionFunctionOutlining();

    Function *OrigFunc = nullptr;
    Function *ClonedFunc = nullptr;

    typedef std::pair<Function *, BasicBlock *> FuncBodyCallerPair;
    // Keep track of Outlined Functions and the basic block they're called from.
    SmallVector<FuncBodyCallerPair, 4> OutlinedFunctions;

    // ClonedFunc is inlined in one of its callers after function
    // outlining.
    bool IsFunctionInlined = false;
    // The cost of the region to be outlined.
    InstructionCost OutlinedRegionCost = 0;
    // ClonedOI is specific to outlining non-early return blocks.
    std::unique_ptr<FunctionOutliningInfo> ClonedOI = nullptr;
    // ClonedOMRI is specific to outlining cold regions.
    std::unique_ptr<FunctionOutliningMultiRegionInfo> ClonedOMRI = nullptr;
    std::unique_ptr<BlockFrequencyInfo> ClonedFuncBFI = nullptr;
    OptimizationRemarkEmitter &ORE;
    function_ref<AssumptionCache *(Function &)> LookupAC;
    function_ref<TargetTransformInfo &(Function &)> GetTTI;
  };

private:
  int NumPartialInlining = 0;
  function_ref<AssumptionCache &(Function &)> GetAssumptionCache;
  function_ref<AssumptionCache *(Function &)> LookupAssumptionCache;
  function_ref<TargetTransformInfo &(Function &)> GetTTI;
  function_ref<BlockFrequencyInfo &(Function &)> GetBFI;
  function_ref<const TargetLibraryInfo &(Function &)> GetTLI;
  ProfileSummaryInfo &PSI;

  // Return the frequency of the OutlininingBB relative to F's entry point.
  // The result is no larger than 1 and is represented using BP.
  // (Note that the outlined region's 'head' block can only have incoming
  // edges from the guarding entry blocks).
  BranchProbability
  getOutliningCallBBRelativeFreq(FunctionCloner &Cloner) const;

  // Return true if the callee of CB should be partially inlined with
  // profit.
  bool shouldPartialInline(CallBase &CB, FunctionCloner &Cloner,
                           BlockFrequency WeightedOutliningRcost,
                           OptimizationRemarkEmitter &ORE) const;

  // Try to inline DuplicateFunction (cloned from F with call to
  // the OutlinedFunction into its callers. Return true
  // if there is any successful inlining.
  bool tryPartialInline(FunctionCloner &Cloner);

  // Compute the mapping from use site of DuplicationFunction to the enclosing
  // BB's profile count.
  void
  computeCallsiteToProfCountMap(Function *DuplicateFunction,
                                DenseMap<User *, uint64_t> &SiteCountMap) const;

  bool isLimitReached() const {
    return (MaxNumPartialInlining != -1 &&
            NumPartialInlining >= MaxNumPartialInlining);
  }

  static CallBase *getSupportedCallBase(User *U) {
    if (isa<CallInst>(U) || isa<InvokeInst>(U))
      return cast<CallBase>(U);
    llvm_unreachable("All uses must be calls");
    return nullptr;
  }

  static CallBase *getOneCallSiteTo(Function &F) {
    User *User = *F.user_begin();
    return getSupportedCallBase(User);
  }

  std::tuple<DebugLoc, BasicBlock *> getOneDebugLoc(Function &F) const {
    CallBase *CB = getOneCallSiteTo(F);
    DebugLoc DLoc = CB->getDebugLoc();
    BasicBlock *Block = CB->getParent();
    return std::make_tuple(DLoc, Block);
  }

  // Returns the costs associated with function outlining:
  // - The first value is the non-weighted runtime cost for making the call
  //   to the outlined function, including the addtional  setup cost in the
  //    outlined function itself;
  // - The second value is the estimated size of the new call sequence in
  //   basic block Cloner.OutliningCallBB;
  std::tuple<InstructionCost, InstructionCost>
  computeOutliningCosts(FunctionCloner &Cloner) const;

  // Compute the 'InlineCost' of block BB. InlineCost is a proxy used to
  // approximate both the size and runtime cost (Note that in the current
  // inline cost analysis, there is no clear distinction there either).
  static InstructionCost computeBBInlineCost(BasicBlock *BB,
                                             TargetTransformInfo *TTI);

  std::unique_ptr<FunctionOutliningInfo>
  computeOutliningInfo(Function &F) const;

  std::unique_ptr<FunctionOutliningMultiRegionInfo>
  computeOutliningColdRegionsInfo(Function &F,
                                  OptimizationRemarkEmitter &ORE) const;
};

} // end anonymous namespace

std::unique_ptr<FunctionOutliningMultiRegionInfo>
PartialInlinerImpl::computeOutliningColdRegionsInfo(
    Function &F, OptimizationRemarkEmitter &ORE) const {
  BasicBlock *EntryBlock = &F.front();

  DominatorTree DT(F);
  LoopInfo LI(DT);
  BranchProbabilityInfo BPI(F, LI);
  std::unique_ptr<BlockFrequencyInfo> ScopedBFI;
  BlockFrequencyInfo *BFI;
  if (!GetBFI) {
    ScopedBFI.reset(new BlockFrequencyInfo(F, BPI, LI));
    BFI = ScopedBFI.get();
  } else
    BFI = &(GetBFI(F));

  // Return if we don't have profiling information.
  if (!PSI.hasInstrumentationProfile())
    return std::unique_ptr<FunctionOutliningMultiRegionInfo>();

  std::unique_ptr<FunctionOutliningMultiRegionInfo> OutliningInfo =
      std::make_unique<FunctionOutliningMultiRegionInfo>();

  auto IsSingleExit =
      [&ORE](SmallVectorImpl<BasicBlock *> &BlockList) -> BasicBlock * {
    BasicBlock *ExitBlock = nullptr;
    for (auto *Block : BlockList) {
      for (BasicBlock *Succ : successors(Block)) {
        if (!is_contained(BlockList, Succ)) {
          if (ExitBlock) {
            ORE.emit([&]() {
              return OptimizationRemarkMissed(DEBUG_TYPE, "MultiExitRegion",
                                              &Succ->front())
                     << "Region dominated by "
                     << ore::NV("Block", BlockList.front()->getName())
                     << " has more than one region exit edge.";
            });
            return nullptr;
          }

          ExitBlock = Block;
        }
      }
    }
    return ExitBlock;
  };

  auto BBProfileCount = [BFI](BasicBlock *BB) {
    return BFI->getBlockProfileCount(BB).value_or(0);
  };

  // Use the same computeBBInlineCost function to compute the cost savings of
  // the outlining the candidate region.
  TargetTransformInfo *FTTI = &GetTTI(F);
  InstructionCost OverallFunctionCost = 0;
  for (auto &BB : F)
    OverallFunctionCost += computeBBInlineCost(&BB, FTTI);

  LLVM_DEBUG(dbgs() << "OverallFunctionCost = " << OverallFunctionCost
                    << "\n";);

  InstructionCost MinOutlineRegionCost = OverallFunctionCost.map(
      [&](auto Cost) { return Cost * MinRegionSizeRatio; });

  BranchProbability MinBranchProbability(
      static_cast<int>(ColdBranchRatio * MinBlockCounterExecution),
      MinBlockCounterExecution);
  bool ColdCandidateFound = false;
  BasicBlock *CurrEntry = EntryBlock;
  std::vector<BasicBlock *> DFS;
  SmallPtrSet<BasicBlock *, 8> VisitedSet;
  DFS.push_back(CurrEntry);
  VisitedSet.insert(CurrEntry);

  // Use Depth First Search on the basic blocks to find CFG edges that are
  // considered cold.
  // Cold regions considered must also have its inline cost compared to the
  // overall inline cost of the original function.  The region is outlined only
  // if it reduced the inline cost of the function by 'MinOutlineRegionCost' or
  // more.
  while (!DFS.empty()) {
    auto *ThisBB = DFS.back();
    DFS.pop_back();
    // Only consider regions with predecessor blocks that are considered
    // not-cold (default: part of the top 99.99% of all block counters)
    // AND greater than our minimum block execution count (default: 100).
    if (PSI.isColdBlock(ThisBB, BFI) ||
        BBProfileCount(ThisBB) < MinBlockCounterExecution)
      continue;
    for (auto SI = succ_begin(ThisBB); SI != succ_end(ThisBB); ++SI) {
      if (!VisitedSet.insert(*SI).second)
        continue;
      DFS.push_back(*SI);
      // If branch isn't cold, we skip to the next one.
      BranchProbability SuccProb = BPI.getEdgeProbability(ThisBB, *SI);
      if (SuccProb > MinBranchProbability)
        continue;

      LLVM_DEBUG(dbgs() << "Found cold edge: " << ThisBB->getName() << "->"
                        << SI->getName()
                        << "\nBranch Probability = " << SuccProb << "\n";);

      SmallVector<BasicBlock *, 8> DominateVector;
      DT.getDescendants(*SI, DominateVector);
      assert(!DominateVector.empty() &&
             "SI should be reachable and have at least itself as descendant");

      // We can only outline single entry regions (for now).
      if (!DominateVector.front()->hasNPredecessors(1)) {
        LLVM_DEBUG(dbgs() << "ABORT: Block " << SI->getName()
                          << " doesn't have a single predecessor in the "
                             "dominator tree\n";);
        continue;
      }

      BasicBlock *ExitBlock = nullptr;
      // We can only outline single exit regions (for now).
      if (!(ExitBlock = IsSingleExit(DominateVector))) {
        LLVM_DEBUG(dbgs() << "ABORT: Block " << SI->getName()
                          << " doesn't have a unique successor\n";);
        continue;
      }

      InstructionCost OutlineRegionCost = 0;
      for (auto *BB : DominateVector)
        OutlineRegionCost += computeBBInlineCost(BB, &GetTTI(*BB->getParent()));

      LLVM_DEBUG(dbgs() << "OutlineRegionCost = " << OutlineRegionCost
                        << "\n";);

      if (!SkipCostAnalysis && OutlineRegionCost < MinOutlineRegionCost) {
        ORE.emit([&]() {
          return OptimizationRemarkAnalysis(DEBUG_TYPE, "TooCostly",
                                            &SI->front())
                 << ore::NV("Callee", &F)
                 << " inline cost-savings smaller than "
                 << ore::NV("Cost", MinOutlineRegionCost);
        });

        LLVM_DEBUG(dbgs() << "ABORT: Outline region cost is smaller than "
                          << MinOutlineRegionCost << "\n";);
        continue;
      }

      // For now, ignore blocks that belong to a SISE region that is a
      // candidate for outlining.  In the future, we may want to look
      // at inner regions because the outer region may have live-exit
      // variables.
      VisitedSet.insert_range(DominateVector);

      // ReturnBlock here means the block after the outline call
      BasicBlock *ReturnBlock = ExitBlock->getSingleSuccessor();
      FunctionOutliningMultiRegionInfo::OutlineRegionInfo RegInfo(
          DominateVector, DominateVector.front(), ExitBlock, ReturnBlock);
      OutliningInfo->ORI.push_back(RegInfo);
      LLVM_DEBUG(dbgs() << "Found Cold Candidate starting at block: "
                        << DominateVector.front()->getName() << "\n";);
      ColdCandidateFound = true;
      NumColdRegionsFound++;
    }
  }

  if (ColdCandidateFound)
    return OutliningInfo;

  return std::unique_ptr<FunctionOutliningMultiRegionInfo>();
}

std::unique_ptr<FunctionOutliningInfo>
PartialInlinerImpl::computeOutliningInfo(Function &F) const {
  BasicBlock *EntryBlock = &F.front();
  BranchInst *BR = dyn_cast<BranchInst>(EntryBlock->getTerminator());
  if (!BR || BR->isUnconditional())
    return std::unique_ptr<FunctionOutliningInfo>();

  // Returns true if Succ is BB's successor
  auto IsSuccessor = [](BasicBlock *Succ, BasicBlock *BB) {
    return is_contained(successors(BB), Succ);
  };

  auto IsReturnBlock = [](BasicBlock *BB) {
    Instruction *TI = BB->getTerminator();
    return isa<ReturnInst>(TI);
  };

  auto GetReturnBlock = [&](BasicBlock *Succ1, BasicBlock *Succ2) {
    if (IsReturnBlock(Succ1))
      return std::make_tuple(Succ1, Succ2);
    if (IsReturnBlock(Succ2))
      return std::make_tuple(Succ2, Succ1);

    return std::make_tuple<BasicBlock *, BasicBlock *>(nullptr, nullptr);
  };

  // Detect a triangular shape:
  auto GetCommonSucc = [&](BasicBlock *Succ1, BasicBlock *Succ2) {
    if (IsSuccessor(Succ1, Succ2))
      return std::make_tuple(Succ1, Succ2);
    if (IsSuccessor(Succ2, Succ1))
      return std::make_tuple(Succ2, Succ1);

    return std::make_tuple<BasicBlock *, BasicBlock *>(nullptr, nullptr);
  };

  std::unique_ptr<FunctionOutliningInfo> OutliningInfo =
      std::make_unique<FunctionOutliningInfo>();

  BasicBlock *CurrEntry = EntryBlock;
  bool CandidateFound = false;
  do {
    // The number of blocks to be inlined has already reached
    // the limit. When MaxNumInlineBlocks is set to 0 or 1, this
    // disables partial inlining for the function.
    if (OutliningInfo->getNumInlinedBlocks() >= MaxNumInlineBlocks)
      break;

    if (succ_size(CurrEntry) != 2)
      break;

    BasicBlock *Succ1 = *succ_begin(CurrEntry);
    BasicBlock *Succ2 = *(succ_begin(CurrEntry) + 1);

    BasicBlock *ReturnBlock, *NonReturnBlock;
    std::tie(ReturnBlock, NonReturnBlock) = GetReturnBlock(Succ1, Succ2);

    if (ReturnBlock) {
      OutliningInfo->Entries.push_back(CurrEntry);
      OutliningInfo->ReturnBlock = ReturnBlock;
      OutliningInfo->NonReturnBlock = NonReturnBlock;
      CandidateFound = true;
      break;
    }

    BasicBlock *CommSucc, *OtherSucc;
    std::tie(CommSucc, OtherSucc) = GetCommonSucc(Succ1, Succ2);

    if (!CommSucc)
      break;

    OutliningInfo->Entries.push_back(CurrEntry);
    CurrEntry = OtherSucc;
  } while (true);

  if (!CandidateFound)
    return std::unique_ptr<FunctionOutliningInfo>();

  // There should not be any successors (not in the entry set) other than
  // {ReturnBlock, NonReturnBlock}
  assert(OutliningInfo->Entries[0] == &F.front() &&
         "Function Entry must be the first in Entries vector");
  DenseSet<BasicBlock *> Entries(llvm::from_range, OutliningInfo->Entries);

  // Returns true of BB has Predecessor which is not
  // in Entries set.
  auto HasNonEntryPred = [Entries](BasicBlock *BB) {
    for (auto *Pred : predecessors(BB)) {
      if (!Entries.count(Pred))
        return true;
    }
    return false;
  };
  auto CheckAndNormalizeCandidate =
      [Entries, HasNonEntryPred](FunctionOutliningInfo *OutliningInfo) {
        for (BasicBlock *E : OutliningInfo->Entries) {
          for (auto *Succ : successors(E)) {
            if (Entries.count(Succ))
              continue;
            if (Succ == OutliningInfo->ReturnBlock)
              OutliningInfo->ReturnBlockPreds.push_back(E);
            else if (Succ != OutliningInfo->NonReturnBlock)
              return false;
          }
          // There should not be any outside incoming edges either:
          if (HasNonEntryPred(E))
            return false;
        }
        return true;
      };

  if (!CheckAndNormalizeCandidate(OutliningInfo.get()))
    return std::unique_ptr<FunctionOutliningInfo>();

  // Now further growing the candidate's inlining region by
  // peeling off dominating blocks from the outlining region:
  while (OutliningInfo->getNumInlinedBlocks() < MaxNumInlineBlocks) {
    BasicBlock *Cand = OutliningInfo->NonReturnBlock;
    if (succ_size(Cand) != 2)
      break;

    if (HasNonEntryPred(Cand))
      break;

    BasicBlock *Succ1 = *succ_begin(Cand);
    BasicBlock *Succ2 = *(succ_begin(Cand) + 1);

    BasicBlock *ReturnBlock, *NonReturnBlock;
    std::tie(ReturnBlock, NonReturnBlock) = GetReturnBlock(Succ1, Succ2);
    if (!ReturnBlock || ReturnBlock != OutliningInfo->ReturnBlock)
      break;

    if (NonReturnBlock->getSinglePredecessor() != Cand)
      break;

    // Now grow and update OutlininigInfo:
    OutliningInfo->Entries.push_back(Cand);
    OutliningInfo->NonReturnBlock = NonReturnBlock;
    OutliningInfo->ReturnBlockPreds.push_back(Cand);
    Entries.insert(Cand);
  }

  return OutliningInfo;
}

// Check if there is PGO data or user annotated branch data:
static bool hasProfileData(const Function &F, const FunctionOutliningInfo &OI) {
  if (F.hasProfileData())
    return true;
  // Now check if any of the entry block has MD_prof data:
  for (auto *E : OI.Entries) {
    BranchInst *BR = dyn_cast<BranchInst>(E->getTerminator());
    if (!BR || BR->isUnconditional())
      continue;
    if (hasBranchWeightMD(*BR))
      return true;
  }
  return false;
}

BranchProbability PartialInlinerImpl::getOutliningCallBBRelativeFreq(
    FunctionCloner &Cloner) const {
  BasicBlock *OutliningCallBB = Cloner.OutlinedFunctions.back().second;
  auto EntryFreq =
      Cloner.ClonedFuncBFI->getBlockFreq(&Cloner.ClonedFunc->getEntryBlock());
  auto OutliningCallFreq =
      Cloner.ClonedFuncBFI->getBlockFreq(OutliningCallBB);
  // FIXME Hackery needed because ClonedFuncBFI is based on the function BEFORE
  // we outlined any regions, so we may encounter situations where the
  // OutliningCallFreq is *slightly* bigger than the EntryFreq.
  if (OutliningCallFreq.getFrequency() > EntryFreq.getFrequency())
    OutliningCallFreq = EntryFreq;

  auto OutlineRegionRelFreq = BranchProbability::getBranchProbability(
      OutliningCallFreq.getFrequency(), EntryFreq.getFrequency());

  if (hasProfileData(*Cloner.OrigFunc, *Cloner.ClonedOI))
    return OutlineRegionRelFreq;

  // When profile data is not available, we need to be conservative in
  // estimating the overall savings. Static branch prediction can usually
  // guess the branch direction right (taken/non-taken), but the guessed
  // branch probability is usually not biased enough. In case when the
  // outlined region is predicted to be likely, its probability needs
  // to be made higher (more biased) to not under-estimate the cost of
  // function outlining. On the other hand, if the outlined region
  // is predicted to be less likely, the predicted probablity is usually
  // higher than the actual. For instance, the actual probability of the
  // less likely target is only 5%, but the guessed probablity can be
  // 40%. In the latter case, there is no need for further adjustment.
  // FIXME: add an option for this.
  if (OutlineRegionRelFreq < BranchProbability(45, 100))
    return OutlineRegionRelFreq;

  OutlineRegionRelFreq = std::max(
      OutlineRegionRelFreq, BranchProbability(OutlineRegionFreqPercent, 100));

  return OutlineRegionRelFreq;
}

bool PartialInlinerImpl::shouldPartialInline(
    CallBase &CB, FunctionCloner &Cloner, BlockFrequency WeightedOutliningRcost,
    OptimizationRemarkEmitter &ORE) const {
  using namespace ore;

  Function *Callee = CB.getCalledFunction();
  assert(Callee == Cloner.ClonedFunc);

  if (SkipCostAnalysis)
    return isInlineViable(*Callee).isSuccess();

  Function *Caller = CB.getCaller();
  auto &CalleeTTI = GetTTI(*Callee);
  bool RemarksEnabled =
      Callee->getContext().getDiagHandlerPtr()->isMissedOptRemarkEnabled(
          DEBUG_TYPE);
  InlineCost IC =
      getInlineCost(CB, getInlineParams(), CalleeTTI, GetAssumptionCache,
                    GetTLI, GetBFI, &PSI, RemarksEnabled ? &ORE : nullptr);

  if (IC.isAlways()) {
    ORE.emit([&]() {
      return OptimizationRemarkAnalysis(DEBUG_TYPE, "AlwaysInline", &CB)
             << NV("Callee", Cloner.OrigFunc)
             << " should always be fully inlined, not partially";
    });
    return false;
  }

  if (IC.isNever()) {
    ORE.emit([&]() {
      return OptimizationRemarkMissed(DEBUG_TYPE, "NeverInline", &CB)
             << NV("Callee", Cloner.OrigFunc) << " not partially inlined into "
             << NV("Caller", Caller)
             << " because it should never be inlined (cost=never)";
    });
    return false;
  }

  if (!IC) {
    ORE.emit([&]() {
      return OptimizationRemarkAnalysis(DEBUG_TYPE, "TooCostly", &CB)
             << NV("Callee", Cloner.OrigFunc) << " not partially inlined into "
             << NV("Caller", Caller) << " because too costly to inline (cost="
             << NV("Cost", IC.getCost()) << ", threshold="
             << NV("Threshold", IC.getCostDelta() + IC.getCost()) << ")";
    });
    return false;
  }
  const DataLayout &DL = Caller->getDataLayout();

  // The savings of eliminating the call:
  int NonWeightedSavings = getCallsiteCost(CalleeTTI, CB, DL);
  BlockFrequency NormWeightedSavings(NonWeightedSavings);

  // Weighted saving is smaller than weighted cost, return false
  if (NormWeightedSavings < WeightedOutliningRcost) {
    ORE.emit([&]() {
      return OptimizationRemarkAnalysis(DEBUG_TYPE, "OutliningCallcostTooHigh",
                                        &CB)
             << NV("Callee", Cloner.OrigFunc) << " not partially inlined into "
             << NV("Caller", Caller) << " runtime overhead (overhead="
             << NV("Overhead", (unsigned)WeightedOutliningRcost.getFrequency())
             << ", savings="
             << NV("Savings", (unsigned)NormWeightedSavings.getFrequency())
             << ")"
             << " of making the outlined call is too high";
    });

    return false;
  }

  ORE.emit([&]() {
    return OptimizationRemarkAnalysis(DEBUG_TYPE, "CanBePartiallyInlined", &CB)
           << NV("Callee", Cloner.OrigFunc) << " can be partially inlined into "
           << NV("Caller", Caller) << " with cost=" << NV("Cost", IC.getCost())
           << " (threshold="
           << NV("Threshold", IC.getCostDelta() + IC.getCost()) << ")";
  });
  return true;
}

// TODO: Ideally  we should share Inliner's InlineCost Analysis code.
// For now use a simplified version. The returned 'InlineCost' will be used
// to esimate the size cost as well as runtime cost of the BB.
InstructionCost
PartialInlinerImpl::computeBBInlineCost(BasicBlock *BB,
                                        TargetTransformInfo *TTI) {
  InstructionCost InlineCost = 0;
  const DataLayout &DL = BB->getDataLayout();
  int InstrCost = InlineConstants::getInstrCost();
  for (Instruction &I : BB->instructionsWithoutDebug()) {
    // Skip free instructions.
    switch (I.getOpcode()) {
    case Instruction::BitCast:
    case Instruction::PtrToInt:
    case Instruction::IntToPtr:
    case Instruction::Alloca:
    case Instruction::PHI:
      continue;
    case Instruction::GetElementPtr:
      if (cast<GetElementPtrInst>(&I)->hasAllZeroIndices())
        continue;
      break;
    default:
      break;
    }

    if (I.isLifetimeStartOrEnd())
      continue;

    if (auto *II = dyn_cast<IntrinsicInst>(&I)) {
      Intrinsic::ID IID = II->getIntrinsicID();
      SmallVector<Type *, 4> Tys;
      FastMathFlags FMF;
      for (Value *Val : II->args())
        Tys.push_back(Val->getType());

      if (auto *FPMO = dyn_cast<FPMathOperator>(II))
        FMF = FPMO->getFastMathFlags();

      IntrinsicCostAttributes ICA(IID, II->getType(), Tys, FMF);
      InlineCost += TTI->getIntrinsicInstrCost(ICA, TTI::TCK_SizeAndLatency);
      continue;
    }

    if (CallInst *CI = dyn_cast<CallInst>(&I)) {
      InlineCost += getCallsiteCost(*TTI, *CI, DL);
      continue;
    }

    if (InvokeInst *II = dyn_cast<InvokeInst>(&I)) {
      InlineCost += getCallsiteCost(*TTI, *II, DL);
      continue;
    }

    if (SwitchInst *SI = dyn_cast<SwitchInst>(&I)) {
      InlineCost += (SI->getNumCases() + 1) * InstrCost;
      continue;
    }
    InlineCost += InstrCost;
  }

  return InlineCost;
}

std::tuple<InstructionCost, InstructionCost>
PartialInlinerImpl::computeOutliningCosts(FunctionCloner &Cloner) const {
  InstructionCost OutliningFuncCallCost = 0, OutlinedFunctionCost = 0;
  for (auto FuncBBPair : Cloner.OutlinedFunctions) {
    Function *OutlinedFunc = FuncBBPair.first;
    BasicBlock* OutliningCallBB = FuncBBPair.second;
    // Now compute the cost of the call sequence to the outlined function
    // 'OutlinedFunction' in BB 'OutliningCallBB':
    auto *OutlinedFuncTTI = &GetTTI(*OutlinedFunc);
    OutliningFuncCallCost +=
        computeBBInlineCost(OutliningCallBB, OutlinedFuncTTI);

    // Now compute the cost of the extracted/outlined function itself:
    for (BasicBlock &BB : *OutlinedFunc)
      OutlinedFunctionCost += computeBBInlineCost(&BB, OutlinedFuncTTI);
  }
  assert(OutlinedFunctionCost >= Cloner.OutlinedRegionCost &&
         "Outlined function cost should be no less than the outlined region");

  // The code extractor introduces a new root and exit stub blocks with
  // additional unconditional branches. Those branches will be eliminated
  // later with bb layout. The cost should be adjusted accordingly:
  OutlinedFunctionCost -=
      2 * InlineConstants::getInstrCost() * Cloner.OutlinedFunctions.size();

  InstructionCost OutliningRuntimeOverhead =
      OutliningFuncCallCost +
      (OutlinedFunctionCost - Cloner.OutlinedRegionCost) +
      ExtraOutliningPenalty.getValue();

  return std::make_tuple(OutliningFuncCallCost, OutliningRuntimeOverhead);
}

// Create the callsite to profile count map which is
// used to update the original function's entry count,
// after the function is partially inlined into the callsite.
void PartialInlinerImpl::computeCallsiteToProfCountMap(
    Function *DuplicateFunction,
    DenseMap<User *, uint64_t> &CallSiteToProfCountMap) const {
  std::vector<User *> Users(DuplicateFunction->user_begin(),
                            DuplicateFunction->user_end());
  Function *CurrentCaller = nullptr;
  std::unique_ptr<BlockFrequencyInfo> TempBFI;
  BlockFrequencyInfo *CurrentCallerBFI = nullptr;

  auto ComputeCurrBFI = [&,this](Function *Caller) {
      // For the old pass manager:
      if (!GetBFI) {
        DominatorTree DT(*Caller);
        LoopInfo LI(DT);
        BranchProbabilityInfo BPI(*Caller, LI);
        TempBFI.reset(new BlockFrequencyInfo(*Caller, BPI, LI));
        CurrentCallerBFI = TempBFI.get();
      } else {
        // New pass manager:
        CurrentCallerBFI = &(GetBFI(*Caller));
      }
  };

  for (User *User : Users) {
    CallBase *CB = getSupportedCallBase(User);
    Function *Caller = CB->getCaller();
    if (CurrentCaller != Caller) {
      CurrentCaller = Caller;
      ComputeCurrBFI(Caller);
    } else {
      assert(CurrentCallerBFI && "CallerBFI is not set");
    }
    BasicBlock *CallBB = CB->getParent();
    auto Count = CurrentCallerBFI->getBlockProfileCount(CallBB);
    if (Count)
      CallSiteToProfCountMap[User] = *Count;
    else
      CallSiteToProfCountMap[User] = 0;
  }
}

PartialInlinerImpl::FunctionCloner::FunctionCloner(
    Function *F, FunctionOutliningInfo *OI, OptimizationRemarkEmitter &ORE,
    function_ref<AssumptionCache *(Function &)> LookupAC,
    function_ref<TargetTransformInfo &(Function &)> GetTTI)
    : OrigFunc(F), ORE(ORE), LookupAC(LookupAC), GetTTI(GetTTI) {
  ClonedOI = std::make_unique<FunctionOutliningInfo>();

  // Clone the function, so that we can hack away on it.
  ValueToValueMapTy VMap;
  ClonedFunc = CloneFunction(F, VMap);

  ClonedOI->ReturnBlock = cast<BasicBlock>(VMap[OI->ReturnBlock]);
  ClonedOI->NonReturnBlock = cast<BasicBlock>(VMap[OI->NonReturnBlock]);
  for (BasicBlock *BB : OI->Entries)
    ClonedOI->Entries.push_back(cast<BasicBlock>(VMap[BB]));

  for (BasicBlock *E : OI->ReturnBlockPreds) {
    BasicBlock *NewE = cast<BasicBlock>(VMap[E]);
    ClonedOI->ReturnBlockPreds.push_back(NewE);
  }
  // Go ahead and update all uses to the duplicate, so that we can just
  // use the inliner functionality when we're done hacking.
  F->replaceAllUsesWith(ClonedFunc);
}

PartialInlinerImpl::FunctionCloner::FunctionCloner(
    Function *F, FunctionOutliningMultiRegionInfo *OI,
    OptimizationRemarkEmitter &ORE,
    function_ref<AssumptionCache *(Function &)> LookupAC,
    function_ref<TargetTransformInfo &(Function &)> GetTTI)
    : OrigFunc(F), ORE(ORE), LookupAC(LookupAC), GetTTI(GetTTI) {
  ClonedOMRI = std::make_unique<FunctionOutliningMultiRegionInfo>();

  // Clone the function, so that we can hack away on it.
  ValueToValueMapTy VMap;
  ClonedFunc = CloneFunction(F, VMap);

  // Go through all Outline Candidate Regions and update all BasicBlock
  // information.
  for (const FunctionOutliningMultiRegionInfo::OutlineRegionInfo &RegionInfo :
       OI->ORI) {
    SmallVector<BasicBlock *, 8> Region;
    for (BasicBlock *BB : RegionInfo.Region)
      Region.push_back(cast<BasicBlock>(VMap[BB]));

    BasicBlock *NewEntryBlock = cast<BasicBlock>(VMap[RegionInfo.EntryBlock]);
    BasicBlock *NewExitBlock = cast<BasicBlock>(VMap[RegionInfo.ExitBlock]);
    BasicBlock *NewReturnBlock = nullptr;
    if (RegionInfo.ReturnBlock)
      NewReturnBlock = cast<BasicBlock>(VMap[RegionInfo.ReturnBlock]);
    FunctionOutliningMultiRegionInfo::OutlineRegionInfo MappedRegionInfo(
        Region, NewEntryBlock, NewExitBlock, NewReturnBlock);
    ClonedOMRI->ORI.push_back(MappedRegionInfo);
  }
  // Go ahead and update all uses to the duplicate, so that we can just
  // use the inliner functionality when we're done hacking.
  F->replaceAllUsesWith(ClonedFunc);
}

void PartialInlinerImpl::FunctionCloner::normalizeReturnBlock() const {
  auto GetFirstPHI = [](BasicBlock *BB) {
    BasicBlock::iterator I = BB->begin();
    PHINode *FirstPhi = nullptr;
    while (I != BB->end()) {
      PHINode *Phi = dyn_cast<PHINode>(I);
      if (!Phi)
        break;
      if (!FirstPhi) {
        FirstPhi = Phi;
        break;
      }
    }
    return FirstPhi;
  };

  // Shouldn't need to normalize PHIs if we're not outlining non-early return
  // blocks.
  if (!ClonedOI)
    return;

  // Special hackery is needed with PHI nodes that have inputs from more than
  // one extracted block.  For simplicity, just split the PHIs into a two-level
  // sequence of PHIs, some of which will go in the extracted region, and some
  // of which will go outside.
  BasicBlock *PreReturn = ClonedOI->ReturnBlock;
  // only split block when necessary:
  PHINode *FirstPhi = GetFirstPHI(PreReturn);
  unsigned NumPredsFromEntries = ClonedOI->ReturnBlockPreds.size();

  if (!FirstPhi || FirstPhi->getNumIncomingValues() <= NumPredsFromEntries + 1)
    return;

  auto IsTrivialPhi = [](PHINode *PN) -> Value * {
    if (llvm::all_equal(PN->incoming_values()))
      return PN->getIncomingValue(0);
    return nullptr;
  };

  ClonedOI->ReturnBlock = ClonedOI->ReturnBlock->splitBasicBlock(
      ClonedOI->ReturnBlock->getFirstNonPHIIt());
  BasicBlock::iterator I = PreReturn->begin();
  BasicBlock::iterator Ins = ClonedOI->ReturnBlock->begin();
  SmallVector<Instruction *, 4> DeadPhis;
  while (I != PreReturn->end()) {
    PHINode *OldPhi = dyn_cast<PHINode>(I);
    if (!OldPhi)
      break;

    PHINode *RetPhi =
        PHINode::Create(OldPhi->getType(), NumPredsFromEntries + 1, "");
    RetPhi->insertBefore(Ins);
    OldPhi->replaceAllUsesWith(RetPhi);
    Ins = ClonedOI->ReturnBlock->getFirstNonPHIIt();

    RetPhi->addIncoming(&*I, PreReturn);
    for (BasicBlock *E : ClonedOI->ReturnBlockPreds) {
      RetPhi->addIncoming(OldPhi->getIncomingValueForBlock(E), E);
      OldPhi->removeIncomingValue(E);
    }

    // After incoming values splitting, the old phi may become trivial.
    // Keeping the trivial phi can introduce definition inside the outline
    // region which is live-out, causing necessary overhead (load, store
    // arg passing etc).
    if (auto *OldPhiVal = IsTrivialPhi(OldPhi)) {
      OldPhi->replaceAllUsesWith(OldPhiVal);
      DeadPhis.push_back(OldPhi);
    }
    ++I;
  }
  for (auto *DP : DeadPhis)
    DP->eraseFromParent();

  for (auto *E : ClonedOI->ReturnBlockPreds)
    E->getTerminator()->replaceUsesOfWith(PreReturn, ClonedOI->ReturnBlock);
}

bool PartialInlinerImpl::FunctionCloner::doMultiRegionFunctionOutlining() {

  auto ComputeRegionCost =
      [&](SmallVectorImpl<BasicBlock *> &Region) -> InstructionCost {
    InstructionCost Cost = 0;
    for (BasicBlock* BB : Region)
      Cost += computeBBInlineCost(BB, &GetTTI(*BB->getParent()));
    return Cost;
  };

  assert(ClonedOMRI && "Expecting OutlineInfo for multi region outline");

  if (ClonedOMRI->ORI.empty())
    return false;

  // The CodeExtractor needs a dominator tree.
  DominatorTree DT;
  DT.recalculate(*ClonedFunc);

  // Manually calculate a BlockFrequencyInfo and BranchProbabilityInfo.
  LoopInfo LI(DT);
  BranchProbabilityInfo BPI(*ClonedFunc, LI);
  ClonedFuncBFI.reset(new BlockFrequencyInfo(*ClonedFunc, BPI, LI));

  // Cache and recycle the CodeExtractor analysis to avoid O(n^2) compile-time.
  CodeExtractorAnalysisCache CEAC(*ClonedFunc);

  SetVector<Value *> Inputs, Outputs, Sinks;
  for (FunctionOutliningMultiRegionInfo::OutlineRegionInfo RegionInfo :
       ClonedOMRI->ORI) {
    InstructionCost CurrentOutlinedRegionCost =
        ComputeRegionCost(RegionInfo.Region);

    CodeExtractor CE(RegionInfo.Region, &DT, /*AggregateArgs*/ false,
                     ClonedFuncBFI.get(), &BPI,
                     LookupAC(*RegionInfo.EntryBlock->getParent()),
                     /* AllowVarargs */ false);

    CE.findInputsOutputs(Inputs, Outputs, Sinks);

    LLVM_DEBUG({
      dbgs() << "inputs: " << Inputs.size() << "\n";
      dbgs() << "outputs: " << Outputs.size() << "\n";
      for (Value *value : Inputs)
        dbgs() << "value used in func: " << *value << "\n";
      for (Value *output : Outputs)
        dbgs() << "instr used in func: " << *output << "\n";
    });

    // Do not extract regions that have live exit variables.
    if (Outputs.size() > 0 && !ForceLiveExit)
      continue;

    if (Function *OutlinedFunc = CE.extractCodeRegion(CEAC)) {
      CallBase *OCS = PartialInlinerImpl::getOneCallSiteTo(*OutlinedFunc);
      BasicBlock *OutliningCallBB = OCS->getParent();
      assert(OutliningCallBB->getParent() == ClonedFunc);
      OutlinedFunctions.push_back(std::make_pair(OutlinedFunc,OutliningCallBB));
      NumColdRegionsOutlined++;
      OutlinedRegionCost += CurrentOutlinedRegionCost;

      if (MarkOutlinedColdCC) {
        OutlinedFunc->setCallingConv(CallingConv::Cold);
        OCS->setCallingConv(CallingConv::Cold);
      }
    } else
      ORE.emit([&]() {
        return OptimizationRemarkMissed(DEBUG_TYPE, "ExtractFailed",
                                        &RegionInfo.Region.front()->front())
               << "Failed to extract region at block "
               << ore::NV("Block", RegionInfo.Region.front());
      });
  }

  return !OutlinedFunctions.empty();
}

Function *
PartialInlinerImpl::FunctionCloner::doSingleRegionFunctionOutlining() {
  // Returns true if the block is to be partial inlined into the caller
  // (i.e. not to be extracted to the out of line function)
  auto ToBeInlined = [&, this](BasicBlock *BB) {
    return BB == ClonedOI->ReturnBlock ||
           llvm::is_contained(ClonedOI->Entries, BB);
  };

  assert(ClonedOI && "Expecting OutlineInfo for single region outline");
  // The CodeExtractor needs a dominator tree.
  DominatorTree DT;
  DT.recalculate(*ClonedFunc);

  // Manually calculate a BlockFrequencyInfo and BranchProbabilityInfo.
  LoopInfo LI(DT);
  BranchProbabilityInfo BPI(*ClonedFunc, LI);
  ClonedFuncBFI.reset(new BlockFrequencyInfo(*ClonedFunc, BPI, LI));

  // Gather up the blocks that we're going to extract.
  std::vector<BasicBlock *> ToExtract;
  auto *ClonedFuncTTI = &GetTTI(*ClonedFunc);
  ToExtract.push_back(ClonedOI->NonReturnBlock);
  OutlinedRegionCost += PartialInlinerImpl::computeBBInlineCost(
      ClonedOI->NonReturnBlock, ClonedFuncTTI);
  for (BasicBlock *BB : depth_first(&ClonedFunc->getEntryBlock()))
    if (!ToBeInlined(BB) && BB != ClonedOI->NonReturnBlock) {
      ToExtract.push_back(BB);
      // FIXME: the code extractor may hoist/sink more code
      // into the outlined function which may make the outlining
      // overhead (the difference of the outlined function cost
      // and OutliningRegionCost) look larger.
      OutlinedRegionCost += computeBBInlineCost(BB, ClonedFuncTTI);
    }

  // Extract the body of the if.
  CodeExtractorAnalysisCache CEAC(*ClonedFunc);
  Function *OutlinedFunc =
      CodeExtractor(ToExtract, &DT, /*AggregateArgs*/ false,
                    ClonedFuncBFI.get(), &BPI, LookupAC(*ClonedFunc),
                    /* AllowVarargs */ true)
          .extractCodeRegion(CEAC);

  if (OutlinedFunc) {
    BasicBlock *OutliningCallBB =
        PartialInlinerImpl::getOneCallSiteTo(*OutlinedFunc)->getParent();
    assert(OutliningCallBB->getParent() == ClonedFunc);
    OutlinedFunctions.push_back(std::make_pair(OutlinedFunc, OutliningCallBB));
  } else
    ORE.emit([&]() {
      return OptimizationRemarkMissed(DEBUG_TYPE, "ExtractFailed",
                                      &ToExtract.front()->front())
             << "Failed to extract region at block "
             << ore::NV("Block", ToExtract.front());
    });

  return OutlinedFunc;
}

PartialInlinerImpl::FunctionCloner::~FunctionCloner() {
  // Ditch the duplicate, since we're done with it, and rewrite all remaining
  // users (function pointers, etc.) back to the original function.
  ClonedFunc->replaceAllUsesWith(OrigFunc);
  ClonedFunc->eraseFromParent();
  if (!IsFunctionInlined) {
    // Remove each function that was speculatively created if there is no
    // reference.
    for (auto FuncBBPair : OutlinedFunctions) {
      Function *Func = FuncBBPair.first;
      Func->eraseFromParent();
    }
  }
}

std::pair<bool, Function *> PartialInlinerImpl::unswitchFunction(Function &F) {
  if (F.hasAddressTaken())
    return {false, nullptr};

  // Let inliner handle it
  if (F.hasFnAttribute(Attribute::AlwaysInline))
    return {false, nullptr};

  if (F.hasFnAttribute(Attribute::NoInline))
    return {false, nullptr};

  if (PSI.isFunctionEntryCold(&F))
    return {false, nullptr};

  if (F.users().empty())
    return {false, nullptr};

  OptimizationRemarkEmitter ORE(&F);

  // Only try to outline cold regions if we have a profile summary, which
  // implies we have profiling information.
  if (PSI.hasProfileSummary() && F.hasProfileData() &&
      !DisableMultiRegionPartialInline) {
    std::unique_ptr<FunctionOutliningMultiRegionInfo> OMRI =
        computeOutliningColdRegionsInfo(F, ORE);
    if (OMRI) {
      FunctionCloner Cloner(&F, OMRI.get(), ORE, LookupAssumptionCache, GetTTI);

      LLVM_DEBUG({
        dbgs() << "HotCountThreshold = " << PSI.getHotCountThreshold() << "\n";
        dbgs() << "ColdCountThreshold = " << PSI.getColdCountThreshold()
               << "\n";
      });

      bool DidOutline = Cloner.doMultiRegionFunctionOutlining();

      if (DidOutline) {
        LLVM_DEBUG({
          dbgs() << ">>>>>> Outlined (Cloned) Function >>>>>>\n";
          Cloner.ClonedFunc->print(dbgs());
          dbgs() << "<<<<<< Outlined (Cloned) Function <<<<<<\n";
        });

        if (tryPartialInline(Cloner))
          return {true, nullptr};
      }
    }
  }

  // Fall-thru to regular partial inlining if we:
  //    i) can't find any cold regions to outline, or
  //   ii) can't inline the outlined function anywhere.
  std::unique_ptr<FunctionOutliningInfo> OI = computeOutliningInfo(F);
  if (!OI)
    return {false, nullptr};

  FunctionCloner Cloner(&F, OI.get(), ORE, LookupAssumptionCache, GetTTI);
  Cloner.normalizeReturnBlock();

  Function *OutlinedFunction = Cloner.doSingleRegionFunctionOutlining();

  if (!OutlinedFunction)
    return {false, nullptr};

  if (tryPartialInline(Cloner))
    return {true, OutlinedFunction};

  return {false, nullptr};
}

bool PartialInlinerImpl::tryPartialInline(FunctionCloner &Cloner) {
  if (Cloner.OutlinedFunctions.empty())
    return false;

  auto OutliningCosts = computeOutliningCosts(Cloner);

  InstructionCost SizeCost = std::get<0>(OutliningCosts);
  InstructionCost NonWeightedRcost = std::get<1>(OutliningCosts);

  assert(SizeCost.isValid() && NonWeightedRcost.isValid() &&
         "Expected valid costs");

  // Only calculate RelativeToEntryFreq when we are doing single region
  // outlining.
  BranchProbability RelativeToEntryFreq;
  if (Cloner.ClonedOI)
    RelativeToEntryFreq = getOutliningCallBBRelativeFreq(Cloner);
  else
    // RelativeToEntryFreq doesn't make sense when we have more than one
    // outlined call because each call will have a different relative frequency
    // to the entry block.  We can consider using the average, but the
    // usefulness of that information is questionable. For now, assume we never
    // execute the calls to outlined functions.
    RelativeToEntryFreq = BranchProbability(0, 1);

  BlockFrequency WeightedRcost =
      BlockFrequency(NonWeightedRcost.getValue()) * RelativeToEntryFreq;

  // The call sequence(s) to the outlined function(s) are larger than the sum of
  // the original outlined region size(s), it does not increase the chances of
  // inlining the function with outlining (The inliner uses the size increase to
  // model the cost of inlining a callee).
  if (!SkipCostAnalysis && Cloner.OutlinedRegionCost < SizeCost) {
    OptimizationRemarkEmitter OrigFuncORE(Cloner.OrigFunc);
    DebugLoc DLoc;
    BasicBlock *Block;
    std::tie(DLoc, Block) = getOneDebugLoc(*Cloner.ClonedFunc);
    OrigFuncORE.emit([&]() {
      return OptimizationRemarkAnalysis(DEBUG_TYPE, "OutlineRegionTooSmall",
                                        DLoc, Block)
             << ore::NV("Function", Cloner.OrigFunc)
             << " not partially inlined into callers (Original Size = "
             << ore::NV("OutlinedRegionOriginalSize", Cloner.OutlinedRegionCost)
             << ", Size of call sequence to outlined function = "
             << ore::NV("NewSize", SizeCost) << ")";
    });
    return false;
  }

  assert(Cloner.OrigFunc->users().empty() &&
         "F's users should all be replaced!");

  std::vector<User *> Users(Cloner.ClonedFunc->user_begin(),
                            Cloner.ClonedFunc->user_end());

  DenseMap<User *, uint64_t> CallSiteToProfCountMap;
  auto CalleeEntryCount = Cloner.OrigFunc->getEntryCount();
  if (CalleeEntryCount)
    computeCallsiteToProfCountMap(Cloner.ClonedFunc, CallSiteToProfCountMap);

  uint64_t CalleeEntryCountV =
      (CalleeEntryCount ? CalleeEntryCount->getCount() : 0);

  bool AnyInline = false;
  for (User *User : Users) {
    CallBase *CB = getSupportedCallBase(User);

    if (isLimitReached())
      continue;

    OptimizationRemarkEmitter CallerORE(CB->getCaller());
    if (!shouldPartialInline(*CB, Cloner, WeightedRcost, CallerORE))
      continue;

    // Construct remark before doing the inlining, as after successful inlining
    // the callsite is removed.
    OptimizationRemark OR(DEBUG_TYPE, "PartiallyInlined", CB);
    OR << ore::NV("Callee", Cloner.OrigFunc) << " partially inlined into "
       << ore::NV("Caller", CB->getCaller());

    InlineFunctionInfo IFI(GetAssumptionCache, &PSI);
    // We can only forward varargs when we outlined a single region, else we
    // bail on vararg functions.
    if (!InlineFunction(*CB, IFI, /*MergeAttributes=*/false, nullptr, true,
                        (Cloner.ClonedOI ? Cloner.OutlinedFunctions.back().first
                                         : nullptr))
             .isSuccess())
      continue;

    CallerORE.emit(OR);

    // Now update the entry count:
    if (CalleeEntryCountV) {
      if (auto It = CallSiteToProfCountMap.find(User);
          It != CallSiteToProfCountMap.end()) {
        uint64_t CallSiteCount = It->second;
        CalleeEntryCountV -= std::min(CalleeEntryCountV, CallSiteCount);
      }
    }

    AnyInline = true;
    NumPartialInlining++;
    // Update the stats
    if (Cloner.ClonedOI)
      NumPartialInlined++;
    else
      NumColdOutlinePartialInlined++;
  }

  if (AnyInline) {
    Cloner.IsFunctionInlined = true;
    if (CalleeEntryCount)
      Cloner.OrigFunc->setEntryCount(Function::ProfileCount(
          CalleeEntryCountV, CalleeEntryCount->getType()));
    OptimizationRemarkEmitter OrigFuncORE(Cloner.OrigFunc);
    OrigFuncORE.emit([&]() {
      return OptimizationRemark(DEBUG_TYPE, "PartiallyInlined", Cloner.OrigFunc)
             << "Partially inlined into at least one caller";
    });
  }

  return AnyInline;
}

bool PartialInlinerImpl::run(Module &M) {
  if (DisablePartialInlining)
    return false;

  std::vector<Function *> Worklist;
  Worklist.reserve(M.size());
  for (Function &F : M)
    if (!F.use_empty() && !F.isDeclaration())
      Worklist.push_back(&F);

  bool Changed = false;
  while (!Worklist.empty()) {
    Function *CurrFunc = Worklist.back();
    Worklist.pop_back();

    if (CurrFunc->use_empty())
      continue;

    std::pair<bool, Function *> Result = unswitchFunction(*CurrFunc);
    if (Result.second)
      Worklist.push_back(Result.second);
    Changed |= Result.first;
  }

  return Changed;
}

PreservedAnalyses PartialInlinerPass::run(Module &M,
                                          ModuleAnalysisManager &AM) {
  auto &FAM = AM.getResult<FunctionAnalysisManagerModuleProxy>(M).getManager();

  auto GetAssumptionCache = [&FAM](Function &F) -> AssumptionCache & {
    return FAM.getResult<AssumptionAnalysis>(F);
  };

  auto LookupAssumptionCache = [&FAM](Function &F) -> AssumptionCache * {
    return FAM.getCachedResult<AssumptionAnalysis>(F);
  };

  auto GetBFI = [&FAM](Function &F) -> BlockFrequencyInfo & {
    return FAM.getResult<BlockFrequencyAnalysis>(F);
  };

  auto GetTTI = [&FAM](Function &F) -> TargetTransformInfo & {
    return FAM.getResult<TargetIRAnalysis>(F);
  };

  auto GetTLI = [&FAM](Function &F) -> TargetLibraryInfo & {
    return FAM.getResult<TargetLibraryAnalysis>(F);
  };

  ProfileSummaryInfo &PSI = AM.getResult<ProfileSummaryAnalysis>(M);

  if (PartialInlinerImpl(GetAssumptionCache, LookupAssumptionCache, GetTTI,
                         GetTLI, PSI, GetBFI)
          .run(M))
    return PreservedAnalyses::none();
  return PreservedAnalyses::all();
}