LoopUnrollPass.cpp

Go to the documentation of this file.

//===- LoopUnroll.cpp - Loop unroller pass --------------------------------===//
//
// 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 implements a simple loop unroller.  It works best when loops have
// been canonicalized by the -indvars pass, allowing it to determine the trip
// counts of loops easily.
//===----------------------------------------------------------------------===//
 
#include "llvm/Transforms/Scalar/LoopUnrollPass.h"
#include "ScalarOptions.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/DenseMapInfo.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SetVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/Analysis/AssumptionCache.h"
#include "llvm/Analysis/BlockFrequencyInfo.h"
#include "llvm/Analysis/CodeMetrics.h"
#include "llvm/Analysis/LoopAnalysisManager.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/LoopUnrollAnalyzer.h"
#include "llvm/Analysis/MemorySSA.h"
#include "llvm/Analysis/OptimizationRemarkEmitter.h"
#include "llvm/Analysis/ProfileSummaryInfo.h"
#include "llvm/Analysis/ScalarEvolution.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/Analysis/UniformityAnalysis.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/CFG.h"
#include "llvm/IR/Constant.h"
#include "llvm/IR/Constants.h"
#include "llvm/IR/DiagnosticInfo.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/Function.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/Instructions.h"
#include "llvm/IR/Metadata.h"
#include "llvm/IR/PassManager.h"
#include "llvm/InitializePasses.h"
#include "llvm/Pass.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Transforms/Scalar/LoopPassManager.h"
#include "llvm/Transforms/Utils.h"
#include "llvm/Transforms/Utils/LoopPeel.h"
#include "llvm/Transforms/Utils/LoopSimplify.h"
#include "llvm/Transforms/Utils/LoopUtils.h"
#include "llvm/Transforms/Utils/ScalarEvolutionExpander.h"
#include "llvm/Transforms/Utils/SizeOpts.h"
#include "llvm/Transforms/Utils/UnrollLoop.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <limits>
#include <optional>
#include <string>
#include <tuple>
#include <utility>
 
using namespace llvm;
 
#define DEBUG_TYPE "loop-unroll"
 
bool llvm::getForgetSCEVInLoopUnroll(
    const cl::LibraryOptionsContext &LibraryOptions) {
  return LibraryOptions.get<ScalarOptions>().forget_scev_loop_unroll;
}
 
/// A magic value for use with the Threshold parameter to indicate
/// that the loop unroll should be performed regardless of how much
/// code expansion would result.
static const unsigned NoThreshold = std::numeric_limits<unsigned>::max();
 
/// Gather the various unrolling parameters based on the defaults, compiler
/// flags, TTI overrides and user specified parameters.
TargetTransformInfo::UnrollingPreferences llvm::gatherUnrollingPreferences(
    Loop *L, ScalarEvolution &SE, const TargetTransformInfo &TTI,
    BlockFrequencyInfo *BFI, ProfileSummaryInfo *PSI,
    OptimizationRemarkEmitter &ORE, int OptLevel,
    std::optional<unsigned> UserThreshold, std::optional<bool> UserAllowPartial,
    std::optional<bool> UserRuntime, std::optional<bool> UserUpperBound,
    std::optional<unsigned> UserFullUnrollMaxCount) {
  const ScalarOptions &Opts = getScalarOptions(*L);
  TargetTransformInfo::UnrollingPreferences UP;
 
  // Set up the defaults
  UP.Threshold = OptLevel > 2 ? Opts.unroll_threshold_aggressive
                              : Opts.unroll_threshold_default;
  UP.MaxPercentThresholdBoost = 400;
  UP.OptSizeThreshold = Opts.unroll_optsize_threshold;
  UP.PartialThreshold = 150;
  UP.PartialOptSizeThreshold = Opts.unroll_optsize_threshold;
  UP.DefaultUnrollRuntimeCount = 8;
  UP.MaxCount = std::numeric_limits<unsigned>::max();
  UP.MaxUpperBound = 8;
  UP.FullUnrollMaxCount = std::numeric_limits<unsigned>::max();
  UP.BEInsns = 2;
  UP.Partial = false;
  UP.Runtime = false;
  UP.AllowRemainder = true;
  UP.UnrollRemainder = false;
  UP.AllowExpensiveTripCount = false;
  UP.Force = false;
  UP.UpperBound = false;
  UP.UnrollAndJam = false;
  UP.UnrollAndJamInnerLoopThreshold = 60;
  UP.MaxIterationsCountToAnalyze = 10;
  UP.SCEVExpansionBudget = SCEVCheapExpansionBudget;
  UP.RuntimeUnrollMultiExit = false;
  UP.AddAdditionalAccumulators = false;
 
  // Override with any target specific settings
  TTI.getUnrollingPreferences(L, SE, UP, &ORE);
 
  // Apply size attributes
  bool OptForSize = L->getHeader()->getParent()->hasOptSize() ||
                    // Let unroll hints / pragmas take precedence over PGSO.
                    (hasUnrollTransformation(L) != TM_ForcedByUser &&
                     llvm::shouldOptimizeForSize(L->getHeader(), PSI, BFI,
                                                 PGSOQueryType::IRPass));
  if (OptForSize) {
    UP.Threshold = UP.OptSizeThreshold;
    UP.PartialThreshold = UP.PartialOptSizeThreshold;
    UP.MaxPercentThresholdBoost = 100;
  }
 
  // Apply any user values specified by cl::opt
  if (Opts.unroll_threshold)
    UP.Threshold = *Opts.unroll_threshold;
  if (Opts.unroll_partial_threshold)
    UP.PartialThreshold = *Opts.unroll_partial_threshold;
  if (Opts.unroll_max_percent_threshold_boost)
    UP.MaxPercentThresholdBoost = *Opts.unroll_max_percent_threshold_boost;
  if (Opts.unroll_max_count)
    UP.MaxCount = *Opts.unroll_max_count;
  if (Opts.unroll_max_upperbound)
    UP.MaxUpperBound = *Opts.unroll_max_upperbound;
  if (Opts.unroll_full_max_count)
    UP.FullUnrollMaxCount = *Opts.unroll_full_max_count;
  UP.Partial = valueOr(Opts.unroll_allow_partial, UP.Partial);
  UP.AllowRemainder = valueOr(Opts.unroll_allow_remainder, UP.AllowRemainder);
  UP.Runtime = valueOr(Opts.unroll_runtime, UP.Runtime);
  if (Opts.unroll_max_upperbound == 0)
    UP.UpperBound = false;
  UP.UnrollRemainder = valueOr(Opts.unroll_remainder, UP.UnrollRemainder);
  if (Opts.unroll_max_iteration_count_to_analyze)
    UP.MaxIterationsCountToAnalyze =
        *Opts.unroll_max_iteration_count_to_analyze;
 
  // Apply user values provided by argument
  if (UserThreshold) {
    UP.Threshold = *UserThreshold;
    UP.PartialThreshold = *UserThreshold;
  }
  if (UserAllowPartial)
    UP.Partial = *UserAllowPartial;
  if (UserRuntime)
    UP.Runtime = *UserRuntime;
  if (UserUpperBound)
    UP.UpperBound = *UserUpperBound;
  if (UserFullUnrollMaxCount)
    UP.FullUnrollMaxCount = *UserFullUnrollMaxCount;
 
  return UP;
}
 
namespace {
 
/// A struct to densely store the state of an instruction after unrolling at
/// each iteration.
///
/// This is designed to work like a tuple of <Instruction *, int> for the
/// purposes of hashing and lookup, but to be able to associate two boolean
/// states with each key.
struct UnrolledInstState {
  Instruction *I;
  int Iteration : 30;
  unsigned IsFree : 1;
  unsigned IsCounted : 1;
};
 
/// Hashing and equality testing for a set of the instruction states.
struct UnrolledInstStateKeyInfo {
  using PairInfo = DenseMapInfo<std::pair<Instruction *, int>>;
 
  static inline unsigned getHashValue(const UnrolledInstState &S) {
    return PairInfo::getHashValue({S.I, S.Iteration});
  }
 
  static inline bool isEqual(const UnrolledInstState &LHS,
                             const UnrolledInstState &RHS) {
    return PairInfo::isEqual({LHS.I, LHS.Iteration}, {RHS.I, RHS.Iteration});
  }
};
 
struct EstimatedUnrollCost {
  /// The estimated cost after unrolling.
  unsigned UnrolledCost;
 
  /// The estimated dynamic cost of executing the instructions in the
  /// rolled form.
  unsigned RolledDynamicCost;
};
 
} // end anonymous namespace
 
/// Figure out if the loop is worth full unrolling.
///
/// Complete loop unrolling can make some loads constant, and we need to know
/// if that would expose any further optimization opportunities.  This routine
/// estimates this optimization.  It computes cost of unrolled loop
/// (UnrolledCost) and dynamic cost of the original loop (RolledDynamicCost). By
/// dynamic cost we mean that we won't count costs of blocks that are known not
/// to be executed (i.e. if we have a branch in the loop and we know that at the
/// given iteration its condition would be resolved to true, we won't add up the
/// cost of the 'false'-block).
/// \returns Optional value, holding the RolledDynamicCost and UnrolledCost. If
/// the analysis failed (no benefits expected from the unrolling, or the loop is
/// too big to analyze), the returned value is std::nullopt.
static std::optional<EstimatedUnrollCost> analyzeLoopUnrollCost(
    const Loop *L, unsigned TripCount, DominatorTree &DT, ScalarEvolution &SE,
    const SmallPtrSetImpl<const Value *> &EphValues,
    const TargetTransformInfo &TTI, unsigned MaxUnrolledLoopSize,
    unsigned MaxIterationsCountToAnalyze) {
  // We want to be able to scale offsets by the trip count and add more offsets
  // to them without checking for overflows, and we already don't want to
  // analyze *massive* trip counts, so we force the max to be reasonably small.
  assert(MaxIterationsCountToAnalyze <
             (unsigned)(std::numeric_limits<int>::max() / 2) &&
         "The unroll iterations max is too large!");
 
  // Only analyze inner loops. We can't properly estimate cost of nested loops
  // and we won't visit inner loops again anyway.
  if (!L->isInnermost()) {
    LLVM_DEBUG(dbgs().indent(3)
               << "Not analyzing loop cost: not an innermost loop.\n");
    return std::nullopt;
  }
 
  // Don't simulate loops with a big or unknown tripcount
  if (!TripCount || TripCount > MaxIterationsCountToAnalyze) {
    LLVM_DEBUG(dbgs().indent(3)
               << "Not analyzing loop cost: trip count "
               << (TripCount ? "too large" : "unknown") << ".\n");
    return std::nullopt;
  }
 
  SmallSetVector<BasicBlock *, 16> BBWorklist;
  SmallSetVector<std::pair<BasicBlock *, BasicBlock *>, 4> ExitWorklist;
  DenseMap<Value *, Value *> SimplifiedValues;
  SmallVector<std::pair<Value *, Value *>, 4> SimplifiedInputValues;
 
  // The estimated cost of the unrolled form of the loop. We try to estimate
  // this by simplifying as much as we can while computing the estimate.
  InstructionCost UnrolledCost = 0;
 
  // We also track the estimated dynamic (that is, actually executed) cost in
  // the rolled form. This helps identify cases when the savings from unrolling
  // aren't just exposing dead control flows, but actual reduced dynamic
  // instructions due to the simplifications which we expect to occur after
  // unrolling.
  InstructionCost RolledDynamicCost = 0;
 
  // We track the simplification of each instruction in each iteration. We use
  // this to recursively merge costs into the unrolled cost on-demand so that
  // we don't count the cost of any dead code. This is essentially a map from
  // <instruction, int> to <bool, bool>, but stored as a densely packed struct.
  DenseSet<UnrolledInstState, UnrolledInstStateKeyInfo> InstCostMap;
 
  // A small worklist used to accumulate cost of instructions from each
  // observable and reached root in the loop.
  SmallVector<Instruction *, 16> CostWorklist;
 
  // PHI-used worklist used between iterations while accumulating cost.
  SmallVector<Instruction *, 4> PHIUsedList;
 
  // Helper function to accumulate cost for instructions in the loop.
  auto AddCostRecursively = [&](Instruction &RootI, int Iteration) {
    assert(Iteration >= 0 && "Cannot have a negative iteration!");
    assert(CostWorklist.empty() && "Must start with an empty cost list");
    assert(PHIUsedList.empty() && "Must start with an empty phi used list");
    CostWorklist.push_back(&RootI);
    TargetTransformInfo::TargetCostKind CostKind =
      RootI.getFunction()->hasMinSize() ?
      TargetTransformInfo::TCK_CodeSize :
      TargetTransformInfo::TCK_SizeAndLatency;
    for (;; --Iteration) {
      do {
        Instruction *I = CostWorklist.pop_back_val();
 
        // InstCostMap only uses I and Iteration as a key, the other two values
        // don't matter here.
        auto CostIter = InstCostMap.find({I, Iteration, 0, 0});
        if (CostIter == InstCostMap.end())
          // If an input to a PHI node comes from a dead path through the loop
          // we may have no cost data for it here. What that actually means is
          // that it is free.
          continue;
        auto &Cost = *CostIter;
        if (Cost.IsCounted)
          // Already counted this instruction.
          continue;
 
        // Mark that we are counting the cost of this instruction now.
        Cost.IsCounted = true;
 
        // If this is a PHI node in the loop header, just add it to the PHI set.
        if (auto *PhiI = dyn_cast<PHINode>(I))
          if (PhiI->getParent() == L->getHeader()) {
            assert(Cost.IsFree && "Loop PHIs shouldn't be evaluated as they "
                                  "inherently simplify during unrolling.");
            if (Iteration == 0)
              continue;
 
            // Push the incoming value from the backedge into the PHI used list
            // if it is an in-loop instruction. We'll use this to populate the
            // cost worklist for the next iteration (as we count backwards).
            if (auto *OpI = dyn_cast<Instruction>(
                    PhiI->getIncomingValueForBlock(L->getLoopLatch())))
              if (L->contains(OpI))
                PHIUsedList.push_back(OpI);
            continue;
          }
 
        // First accumulate the cost of this instruction.
        if (!Cost.IsFree) {
          // Consider simplified operands in instruction cost.
          SmallVector<Value *, 4> Operands;
          transform(I->operands(), std::back_inserter(Operands),
                    [&](Value *Op) {
                      if (auto Res = SimplifiedValues.lookup(Op))
                        return Res;
                      return Op;
                    });
          UnrolledCost += TTI.getInstructionCost(I, Operands, CostKind);
          LLVM_DEBUG(dbgs().indent(3)
                     << "Adding cost of instruction (iteration " << Iteration
                     << "): ");
          LLVM_DEBUG(I->dump());
        }
 
        // We must count the cost of every operand which is not free,
        // recursively. If we reach a loop PHI node, simply add it to the set
        // to be considered on the next iteration (backwards!).
        for (Value *Op : I->operands()) {
          // Check whether this operand is free due to being a constant or
          // outside the loop.
          auto *OpI = dyn_cast<Instruction>(Op);
          if (!OpI || !L->contains(OpI))
            continue;
 
          // Otherwise accumulate its cost.
          CostWorklist.push_back(OpI);
        }
      } while (!CostWorklist.empty());
 
      if (PHIUsedList.empty())
        // We've exhausted the search.
        break;
 
      assert(Iteration > 0 &&
             "Cannot track PHI-used values past the first iteration!");
      CostWorklist.append(PHIUsedList.begin(), PHIUsedList.end());
      PHIUsedList.clear();
    }
  };
 
  // Ensure that we don't violate the loop structure invariants relied on by
  // this analysis.
  assert(L->isLoopSimplifyForm() && "Must put loop into normal form first.");
  assert(L->isLCSSAForm(DT) &&
         "Must have loops in LCSSA form to track live-out values.");
 
  LLVM_DEBUG(dbgs().indent(3)
             << "Starting LoopUnroll profitability analysis...\n");
 
  TargetTransformInfo::TargetCostKind CostKind =
    L->getHeader()->getParent()->hasMinSize() ?
    TargetTransformInfo::TCK_CodeSize : TargetTransformInfo::TCK_SizeAndLatency;
  // Simulate execution of each iteration of the loop counting instructions,
  // which would be simplified.
  // Since the same load will take different values on different iterations,
  // we literally have to go through all loop's iterations.
  for (unsigned Iteration = 0; Iteration < TripCount; ++Iteration) {
    LLVM_DEBUG(dbgs().indent(3) << "Analyzing iteration " << Iteration << "\n");
 
    // Prepare for the iteration by collecting any simplified entry or backedge
    // inputs.
    for (Instruction &I : *L->getHeader()) {
      auto *PHI = dyn_cast<PHINode>(&I);
      if (!PHI)
        break;
 
      // The loop header PHI nodes must have exactly two input: one from the
      // loop preheader and one from the loop latch.
      assert(
          PHI->getNumIncomingValues() == 2 &&
          "Must have an incoming value only for the preheader and the latch.");
 
      Value *V = PHI->getIncomingValueForBlock(
          Iteration == 0 ? L->getLoopPreheader() : L->getLoopLatch());
      if (Iteration != 0 && SimplifiedValues.contains(V))
        V = SimplifiedValues.lookup(V);
      SimplifiedInputValues.push_back({PHI, V});
    }
 
    // Now clear and re-populate the map for the next iteration.
    SimplifiedValues.clear();
    while (!SimplifiedInputValues.empty())
      SimplifiedValues.insert(SimplifiedInputValues.pop_back_val());
 
    UnrolledInstAnalyzer Analyzer(Iteration, SimplifiedValues, SE, L);
 
    BBWorklist.clear();
    BBWorklist.insert(L->getHeader());
    // Note that we *must not* cache the size, this loop grows the worklist.
    for (unsigned Idx = 0; Idx != BBWorklist.size(); ++Idx) {
      BasicBlock *BB = BBWorklist[Idx];
 
      // Visit all instructions in the given basic block and try to simplify
      // it.  We don't change the actual IR, just count optimization
      // opportunities.
      for (Instruction &I : *BB) {
        // These won't get into the final code - don't even try calculating the
        // cost for them.
        if (EphValues.contains(&I))
          continue;
 
        // Track this instruction's expected baseline cost when executing the
        // rolled loop form.
        RolledDynamicCost += TTI.getInstructionCost(&I, CostKind);
 
        // Visit the instruction to analyze its loop cost after unrolling,
        // and if the visitor returns true, mark the instruction as free after
        // unrolling and continue.
        bool IsFree = Analyzer.visit(I);
        bool Inserted = InstCostMap.insert({&I, (int)Iteration,
                                           (unsigned)IsFree,
                                           /*IsCounted*/ false}).second;
        (void)Inserted;
        assert(Inserted && "Cannot have a state for an unvisited instruction!");
 
        if (IsFree)
          continue;
 
        // Can't properly model a cost of a call.
        // FIXME: With a proper cost model we should be able to do it.
        if (auto *CI = dyn_cast<CallInst>(&I)) {
          const Function *Callee = CI->getCalledFunction();
          if (!Callee || TTI.isLoweredToCall(Callee)) {
            LLVM_DEBUG(dbgs().indent(3)
                       << "Can't analyze cost of loop with call\n");
            return std::nullopt;
          }
        }
 
        // If the instruction might have a side-effect recursively account for
        // the cost of it and all the instructions leading up to it.
        if (I.mayHaveSideEffects())
          AddCostRecursively(I, Iteration);
 
        // If unrolled body turns out to be too big, bail out.
        if (UnrolledCost > MaxUnrolledLoopSize) {
          LLVM_DEBUG({
            dbgs().indent(3) << "Exceeded threshold.. exiting.\n";
            dbgs().indent(3)
                << "UnrolledCost: " << UnrolledCost
                << ", MaxUnrolledLoopSize: " << MaxUnrolledLoopSize << "\n";
          });
          return std::nullopt;
        }
      }
 
      Instruction *TI = BB->getTerminator();
 
      auto getSimplifiedConstant = [&](Value *V) -> Constant * {
        if (SimplifiedValues.contains(V))
          V = SimplifiedValues.lookup(V);
        return dyn_cast<Constant>(V);
      };
 
      // Add in the live successors by first checking whether we have terminator
      // that may be simplified based on the values simplified by this call.
      BasicBlock *KnownSucc = nullptr;
      if (CondBrInst *BI = dyn_cast<CondBrInst>(TI)) {
        if (auto *SimpleCond = getSimplifiedConstant(BI->getCondition())) {
          // Just take the first successor if condition is undef
          if (isa<UndefValue>(SimpleCond))
            KnownSucc = BI->getSuccessor(0);
          else if (ConstantInt *SimpleCondVal =
                       dyn_cast<ConstantInt>(SimpleCond))
            KnownSucc = BI->getSuccessor(SimpleCondVal->isZero() ? 1 : 0);
        }
      } else if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
        if (auto *SimpleCond = getSimplifiedConstant(SI->getCondition())) {
          // Just take the first successor if condition is undef
          if (isa<UndefValue>(SimpleCond))
            KnownSucc = SI->getSuccessor(0);
          else if (ConstantInt *SimpleCondVal =
                       dyn_cast<ConstantInt>(SimpleCond))
            KnownSucc = SI->findCaseValue(SimpleCondVal)->getCaseSuccessor();
        }
      }
      if (KnownSucc) {
        if (L->contains(KnownSucc))
          BBWorklist.insert(KnownSucc);
        else
          ExitWorklist.insert({BB, KnownSucc});
        continue;
      }
 
      // Add BB's successors to the worklist.
      for (BasicBlock *Succ : successors(BB))
        if (L->contains(Succ))
          BBWorklist.insert(Succ);
        else
          ExitWorklist.insert({BB, Succ});
      AddCostRecursively(*TI, Iteration);
    }
 
    // If we found no optimization opportunities on the first iteration, we
    // won't find them on later ones too.
    if (UnrolledCost == RolledDynamicCost) {
      LLVM_DEBUG({
        dbgs().indent(3) << "No opportunities found.. exiting.\n";
        dbgs().indent(3) << "UnrolledCost: " << UnrolledCost << "\n";
      });
      return std::nullopt;
    }
  }
 
  while (!ExitWorklist.empty()) {
    BasicBlock *ExitingBB, *ExitBB;
    std::tie(ExitingBB, ExitBB) = ExitWorklist.pop_back_val();
 
    for (Instruction &I : *ExitBB) {
      auto *PN = dyn_cast<PHINode>(&I);
      if (!PN)
        break;
 
      Value *Op = PN->getIncomingValueForBlock(ExitingBB);
      if (auto *OpI = dyn_cast<Instruction>(Op))
        if (L->contains(OpI))
          AddCostRecursively(*OpI, TripCount - 1);
    }
  }
 
  assert(UnrolledCost.isValid() && RolledDynamicCost.isValid() &&
         "All instructions must have a valid cost, whether the "
         "loop is rolled or unrolled.");
 
  LLVM_DEBUG({
    dbgs().indent(3) << "Analysis finished:\n";
    dbgs().indent(3) << "UnrolledCost: " << UnrolledCost
                     << ", RolledDynamicCost: " << RolledDynamicCost << "\n";
  });
  return {{unsigned(UnrolledCost.getValue()),
           unsigned(RolledDynamicCost.getValue())}};
}
 
UnrollCostEstimator::UnrollCostEstimator(
    const Loop *L, const TargetTransformInfo &TTI,
    const SmallPtrSetImpl<const Value *> &EphValues, unsigned BEInsns,
    bool PrepareForLTO, bool TripCountIsUniform) {
  CodeMetrics Metrics;
  for (BasicBlock *BB : L->blocks())
    Metrics.analyzeBasicBlock(BB, TTI, EphValues, PrepareForLTO, L);
  NumInlineCandidates = Metrics.NumInlineCandidates;
  NotDuplicatable = Metrics.notDuplicatable;
  Convergence = Metrics.Convergence;
  LoopSize = Metrics.NumInsts;
  // Convergent operations make the remainder prelude unsafe by adding a
  // control-flow dependency, unless the trip count is uniform per
  // UniformityInfo, in which case all paths agree and the remainder is safe.
  ConvergenceAllowsRuntime =
      (Metrics.Convergence != ConvergenceKind::Uncontrolled &&
       !getLoopConvergenceHeart(L)) ||
      TripCountIsUniform;
 
  // Don't allow an estimate of size zero.  This would allows unrolling of loops
  // with huge iteration counts, which is a compile time problem even if it's
  // not a problem for code quality. Also, the code using this size may assume
  // that each loop has at least three instructions (likely a conditional
  // branch, a comparison feeding that branch, and some kind of loop increment
  // feeding that comparison instruction).
  if (LoopSize.isValid() && LoopSize < BEInsns + 1)
    // This is an open coded max() on InstructionCost
    LoopSize = BEInsns + 1;
}
 
bool UnrollCostEstimator::canUnroll(OptimizationRemarkEmitter *ORE,
                                    const Loop *L) const {
  auto ReportCannotUnroll = [&](StringRef Reason) {
    LLVM_DEBUG(dbgs().indent(1) << "Not unrolling: " << Reason << ".\n");
    if (ORE && L)
      ORE->emit([&]() {
        return OptimizationRemarkMissed(DEBUG_TYPE, "CannotUnrollLoop",
                                        L->getStartLoc(), L->getHeader())
               << "unable to unroll loop: " << Reason;
      });
  };
 
  if (Convergence == ConvergenceKind::ExtendedLoop) {
    ReportCannotUnroll("contains convergent operations");
    return false;
  }
  if (!LoopSize.isValid()) {
    ReportCannotUnroll("loop size could not be computed");
    return false;
  }
  if (NotDuplicatable) {
    ReportCannotUnroll("contains non-duplicatable instructions");
    return false;
  }
  return true;
}
 
uint64_t UnrollCostEstimator::getUnrolledLoopSize(
    const TargetTransformInfo::UnrollingPreferences &UP, unsigned Count) const {
  unsigned LS = LoopSize.getValue();
  assert(LS >= UP.BEInsns && "LoopSize should not be less than BEInsns!");
  return static_cast<uint64_t>(LS - UP.BEInsns) * Count + UP.BEInsns;
}
 
// Returns true if the loop has an unroll(full) pragma.
static bool hasUnrollFullPragma(const Loop *L) {
  return getUnrollMetadataForLoop(L, "llvm.loop.unroll.full");
}
 
// Returns true if the loop has an unroll(enable) pragma. This metadata is used
// for both "#pragma unroll" and "#pragma clang loop unroll(enable)" directives.
static bool hasUnrollEnablePragma(const Loop *L) {
  return getUnrollMetadataForLoop(L, "llvm.loop.unroll.enable");
}
 
// Returns true if the loop has a runtime unroll(disable) pragma.
static bool hasRuntimeUnrollDisablePragma(const Loop *L) {
  return getUnrollMetadataForLoop(L, "llvm.loop.unroll.runtime.disable");
}
 
/// Returns true if the SCEV expression is uniform, i.e., all threads in a
/// convergent execution agree on its value. Recursively checks operands.
/// Returns false if the SCEV could not be computed.
static bool isSCEVUniform(const SCEV *S, UniformityInfo &UI) {
  if (isa<SCEVCouldNotCompute>(S))
    return false;
  if (isa<SCEVConstant>(S))
    return true;
  if (auto *U = dyn_cast<SCEVUnknown>(S))
    return UI.isUniformAtDef(U->getValue());
  for (const SCEV *Op : S->operands()) {
    if (!isSCEVUniform(Op, UI))
      return false;
  }
  return true;
}
 
// If loop has an unroll_count pragma return the (necessarily
// positive) value from the pragma.  Otherwise return 0.
static unsigned unrollCountPragmaValue(const Loop *L) {
  MDNode *MD = getUnrollMetadataForLoop(L, "llvm.loop.unroll.count");
  if (MD) {
    assert(MD->getNumOperands() == 2 &&
           "Unroll count hint metadata should have two operands.");
    unsigned Count =
        mdconst::extract<ConstantInt>(MD->getOperand(1))->getZExtValue();
    assert(Count >= 1 && "Unroll count must be positive.");
    return Count;
  }
  return 0;
}
 
UnrollPragmaInfo::UnrollPragmaInfo(const Loop *L)
    : UserUnrollCount(getScalarOptions(*L).unroll_count.has_value()),
      PragmaFullUnroll(hasUnrollFullPragma(L)),
      PragmaCount(unrollCountPragmaValue(L)),
      PragmaEnableUnroll(hasUnrollEnablePragma(L)),
      PragmaRuntimeUnrollDisable(hasRuntimeUnrollDisablePragma(L)),
      ExplicitUnroll(PragmaCount > 0 || PragmaFullUnroll ||
                     PragmaEnableUnroll || UserUnrollCount) {}
 
// Computes the boosting factor for complete unrolling.
// If fully unrolling the loop would save a lot of RolledDynamicCost, it would
// be beneficial to fully unroll the loop even if unrolledcost is large. We
// use (RolledDynamicCost / UnrolledCost) to model the unroll benefits to adjust
// the unroll threshold.
static unsigned getFullUnrollBoostingFactor(const EstimatedUnrollCost &Cost,
                                            unsigned MaxPercentThresholdBoost) {
  if (Cost.RolledDynamicCost >= std::numeric_limits<unsigned>::max() / 100)
    return 100;
  else if (Cost.UnrolledCost != 0)
    // The boosting factor is RolledDynamicCost / UnrolledCost
    return std::min(100 * Cost.RolledDynamicCost / Cost.UnrolledCost,
                    MaxPercentThresholdBoost);
  else
    return MaxPercentThresholdBoost;
}
 
static std::optional<unsigned>
shouldPragmaUnroll(const ScalarOptions &Opts, Loop *L,
                   const UnrollPragmaInfo &PInfo, const unsigned TripMultiple,
                   const unsigned TripCount, unsigned MaxTripCount,
                   const UnrollCostEstimator UCE,
                   const TargetTransformInfo::UnrollingPreferences &UP,
                   OptimizationRemarkEmitter *ORE) {
 
  // Using unroll pragma
  // 1st priority is unroll count set by "unroll-count" option.
 
  if (PInfo.UserUnrollCount) {
    if (UP.AllowRemainder &&
        UCE.getUnrolledLoopSize(UP, *Opts.unroll_count) < UP.Threshold) {
      LLVM_DEBUG(dbgs().indent(2) << "Unrolling with user-specified count: "
                                  << *Opts.unroll_count << ".\n");
      return *Opts.unroll_count;
    }
    LLVM_DEBUG(dbgs().indent(2)
               << "Not unrolling with user count " << *Opts.unroll_count << ": "
               << (UP.AllowRemainder ? "exceeds threshold"
                                     : "remainder not allowed")
               << ".\n");
  }
 
  // 2nd priority is unroll count set by pragma.
  if (PInfo.PragmaCount > 0) {
    if ((UP.AllowRemainder || (TripMultiple % PInfo.PragmaCount == 0))) {
      LLVM_DEBUG(dbgs().indent(2) << "Unrolling with pragma count: "
                                  << PInfo.PragmaCount << ".\n");
      return PInfo.PragmaCount;
    }
    LLVM_DEBUG(dbgs().indent(2)
               << "Not unrolling with pragma count " << PInfo.PragmaCount
               << ": remainder not allowed, count does not divide trip "
               << "multiple " << TripMultiple << ".\n");
    ORE->emit([&]() {
      return OptimizationRemarkAnalysis(DEBUG_TYPE, "PragmaUnrollCountRejected",
                                        L->getStartLoc(), L->getHeader())
             << "may be unable to unroll loop with count "
             << ore::NV("PragmaCount", PInfo.PragmaCount)
             << ": remainder loop is not allowed and count does not divide "
                "trip multiple "
             << ore::NV("TripMultiple", TripMultiple);
    });
  }
 
  if (PInfo.PragmaFullUnroll) {
    if (TripCount != 0) {
      // Certain cases with UBSAN can cause trip count to be calculated as
      // INT_MAX, Block full unrolling at a reasonable limit so that the
      // compiler doesn't hang trying to unroll the loop. See PR77842
      if (TripCount > Opts.pragma_unroll_full_max_iterations) {
        LLVM_DEBUG(dbgs().indent(2)
                   << "Won't unroll; trip count is too large.\n");
        ORE->emit([&]() {
          return OptimizationRemarkAnalysis(DEBUG_TYPE,
                                            "PragmaFullUnrollTripCountTooLarge",
                                            L->getStartLoc(), L->getHeader())
                 << "may be unable to fully unroll loop: trip count "
                 << ore::NV("TripCount", TripCount) << " exceeds limit "
                 << ore::NV("Limit", Opts.pragma_unroll_full_max_iterations);
        });
        return std::nullopt;
      }
 
      LLVM_DEBUG(dbgs().indent(2)
                 << "Fully unrolling with trip count: " << TripCount << ".\n");
      return TripCount;
    }
    LLVM_DEBUG(dbgs().indent(2)
               << "Not fully unrolling: unknown trip count.\n");
    ORE->emit([&]() {
      return OptimizationRemarkAnalysis(DEBUG_TYPE,
                                        "PragmaFullUnrollUnknownTripCount",
                                        L->getStartLoc(), L->getHeader())
             << "may be unable to fully unroll loop: trip count is unknown";
    });
  }
 
  if (PInfo.PragmaEnableUnroll && !TripCount && MaxTripCount &&
      MaxTripCount <= UP.MaxUpperBound) {
    LLVM_DEBUG(dbgs().indent(2)
               << "Unrolling with max trip count: " << MaxTripCount << ".\n");
    return MaxTripCount;
  }
 
  return std::nullopt;
}
 
static std::optional<unsigned> shouldFullUnroll(
    Loop *L, const TargetTransformInfo &TTI, DominatorTree &DT,
    ScalarEvolution &SE, const SmallPtrSetImpl<const Value *> &EphValues,
    const unsigned FullUnrollTripCount, const UnrollCostEstimator UCE,
    const TargetTransformInfo::UnrollingPreferences &UP) {
  assert(FullUnrollTripCount && "should be non-zero!");
 
  if (FullUnrollTripCount > UP.FullUnrollMaxCount) {
    LLVM_DEBUG(dbgs().indent(2)
               << "Not unrolling: trip count " << FullUnrollTripCount
               << " exceeds max count " << UP.FullUnrollMaxCount << ".\n");
    return std::nullopt;
  }
 
  // When computing the unrolled size, note that BEInsns are not replicated
  // like the rest of the loop body.
  uint64_t UnrolledSize = UCE.getUnrolledLoopSize(UP, FullUnrollTripCount);
  if (UnrolledSize < UP.Threshold) {
    LLVM_DEBUG(dbgs().indent(2) << "Unrolling: size " << UnrolledSize
                                << " < threshold " << UP.Threshold << ".\n");
    return FullUnrollTripCount;
  }
 
  LLVM_DEBUG(dbgs().indent(2)
             << "Unrolled size " << UnrolledSize << " exceeds threshold "
             << UP.Threshold << "; checking for cost benefit.\n");
 
  // The loop isn't that small, but we still can fully unroll it if that
  // helps to remove a significant number of instructions.
  // To check that, run additional analysis on the loop.
  if (std::optional<EstimatedUnrollCost> Cost = analyzeLoopUnrollCost(
          L, FullUnrollTripCount, DT, SE, EphValues, TTI,
          UP.Threshold * UP.MaxPercentThresholdBoost / 100,
          UP.MaxIterationsCountToAnalyze)) {
    unsigned Boost =
        getFullUnrollBoostingFactor(*Cost, UP.MaxPercentThresholdBoost);
    unsigned BoostedThreshold = UP.Threshold * Boost / 100;
    if (Cost->UnrolledCost < BoostedThreshold) {
      LLVM_DEBUG(dbgs().indent(2) << "Profitable after cost analysis.\n");
      return FullUnrollTripCount;
    }
    LLVM_DEBUG(dbgs().indent(2)
               << "Not unrolling: cost " << Cost->UnrolledCost
               << " >= boosted threshold " << BoostedThreshold << ".\n");
  }
 
  return std::nullopt;
}
 
static std::optional<unsigned>
shouldPartialUnroll(const unsigned LoopSize, const unsigned TripCount,
                    const UnrollCostEstimator UCE,
                    const TargetTransformInfo::UnrollingPreferences &UP) {
 
  if (!TripCount)
    return std::nullopt;
 
  if (!UP.Partial) {
    LLVM_DEBUG(dbgs().indent(2) << "Will not try to unroll partially because "
                                << "-unroll-allow-partial not given\n");
    return 0;
  }
  unsigned Count = TripCount;
  if (UP.PartialThreshold != NoThreshold) {
    // Reduce unroll count to be modulo of TripCount for partial unrolling.
    if (UCE.getUnrolledLoopSize(UP, Count) > UP.PartialThreshold) {
      unsigned NewCount =
          (std::max(UP.PartialThreshold, UP.BEInsns + 1) - UP.BEInsns) /
          (LoopSize - UP.BEInsns);
      LLVM_DEBUG(dbgs().indent(2)
                 << "Unrolled size exceeds threshold; reducing count "
                 << "from " << Count << " to " << NewCount << ".\n");
      Count = NewCount;
    }
    if (Count > UP.MaxCount)
      Count = UP.MaxCount;
    while (Count != 0 && TripCount % Count != 0)
      Count--;
    if (UP.AllowRemainder && Count <= 1) {
      // If there is no Count that is modulo of TripCount, set Count to
      // largest power-of-two factor that satisfies the threshold limit.
      // As we'll create fixup loop, do the type of unrolling only if
      // remainder loop is allowed.
      // Note: DefaultUnrollRuntimeCount is used as a reasonable starting point
      // even though this is partial unrolling (not runtime unrolling).
      Count = UP.DefaultUnrollRuntimeCount;
      while (Count != 0 &&
             UCE.getUnrolledLoopSize(UP, Count) > UP.PartialThreshold)
        Count >>= 1;
    }
    if (Count < 2) {
      LLVM_DEBUG(dbgs().indent(2)
                 << "Will not partially unroll: no profitable count.\n");
      Count = 0;
    }
  } else {
    Count = TripCount;
  }
  if (Count > UP.MaxCount)
    Count = UP.MaxCount;
 
  LLVM_DEBUG(dbgs().indent(2)
             << "Partially unrolling with count: " << Count << "\n");
 
  return Count;
}
 
static unsigned
shouldRuntimeUnroll(Loop *L, unsigned MaxTripCount, unsigned TripMultiple,
                    const UnrollCostEstimator &UCE,
                    const UnrollPragmaInfo &PInfo, const ScalarOptions &Opts,
                    TargetTransformInfo::UnrollingPreferences &UP) {
  // Don't unroll a runtime trip count loop when it is disabled.
  if (PInfo.PragmaRuntimeUnrollDisable) {
    LLVM_DEBUG(dbgs().indent(2)
               << "Not runtime unrolling: disabled by pragma.\n");
    return 0;
  }
 
  // Don't unroll a small upper bound loop unless user or TTI asked to do so.
  if (MaxTripCount && !UP.Force && MaxTripCount <= UP.MaxUpperBound) {
    LLVM_DEBUG(dbgs().indent(2) << "Not runtime unrolling: max trip count "
                                << MaxTripCount << " is small (<= "
                                << UP.MaxUpperBound << ") and not forced.\n");
    return 0;
  }
 
  // Check if the runtime trip count is too small when profile is available.
  if (L->getHeader()->getParent()->hasProfileData()) {
    if (auto ProfileTripCount = getLoopEstimatedTripCount(L)) {
      if (*ProfileTripCount < Opts.flat_loop_tripcount_threshold)
        return 0;
      else
        UP.AllowExpensiveTripCount = true;
    }
  }
  if (!UP.Runtime) {
    LLVM_DEBUG(dbgs().indent(2)
               << "Will not try to unroll loop with runtime trip count "
               << "because -unroll-runtime not given\n");
    return 0;
  }
 
  unsigned Count = UP.DefaultUnrollRuntimeCount;
 
  // Reduce unroll count to be the largest power-of-two factor of
  // the original count which satisfies the threshold limit.
  while (Count != 0 && UCE.getUnrolledLoopSize(UP, Count) > UP.PartialThreshold)
    Count >>= 1;
 
#ifndef NDEBUG
  unsigned OrigCount = Count;
#endif
 
  if (!UP.AllowRemainder && Count != 0 && (TripMultiple % Count) != 0) {
    while (Count != 0 && TripMultiple % Count != 0)
      Count >>= 1;
    LLVM_DEBUG(dbgs().indent(2)
               << "Remainder loop is restricted (that could be architecture "
                  "specific or because the loop contains a convergent "
                  "instruction), so unroll count must divide the trip "
                  "multiple, "
               << TripMultiple << ".  Reducing unroll count from " << OrigCount
               << " to " << Count << ".\n");
  }
 
  if (Count > UP.MaxCount)
    Count = UP.MaxCount;
 
  if (MaxTripCount && Count > MaxTripCount)
    Count = MaxTripCount;
 
  if (Count < 2)
    Count = 0;
  else
    LLVM_DEBUG(dbgs().indent(2)
               << "Runtime unrolling with count: " << Count << "\n");
  return Count;
}
 
// Calculates and returns the unroll count, using metadata and command-line
// options that are specific to the LoopUnroll pass (which, for instance, are
// irrelevant for the LoopUnrollAndJam pass).
// FIXME: This function is used by LoopUnroll and LoopUnrollAndJam, but consumes
// many LoopUnroll-specific options. The shared functionality should be
// refactored into it own function.
unsigned llvm::computeUnrollCount(
    Loop *L, const TargetTransformInfo &TTI, DominatorTree &DT, LoopInfo *LI,
    AssumptionCache *AC, ScalarEvolution &SE,
    const SmallPtrSetImpl<const Value *> &EphValues,
    OptimizationRemarkEmitter *ORE, const unsigned TripCount,
    const unsigned MaxTripCount, const bool MaxOrZero,
    const unsigned TripMultiple, const UnrollCostEstimator &UCE,
    TargetTransformInfo::UnrollingPreferences &UP,
    TargetTransformInfo::PeelingPreferences &PP) {
  const ScalarOptions &Opts = getScalarOptions(*L);
 
  unsigned LoopSize = UCE.getRolledLoopSize();
 
  LLVM_DEBUG(dbgs().indent(1) << "Computing unroll count: TripCount="
                              << TripCount << ", MaxTripCount=" << MaxTripCount
                              << (MaxOrZero ? " (MaxOrZero)" : "")
                              << ", TripMultiple=" << TripMultiple << "\n");
 
  UnrollPragmaInfo PInfo(L);
  LLVM_DEBUG({
    if (PInfo.ExplicitUnroll) {
      dbgs().indent(1) << "Explicit unroll requested:";
      if (PInfo.UserUnrollCount)
        dbgs() << " user-count";
      if (PInfo.PragmaFullUnroll)
        dbgs() << " pragma-full";
      if (PInfo.PragmaCount > 0)
        dbgs() << " pragma-count(" << PInfo.PragmaCount << ")";
      if (PInfo.PragmaEnableUnroll)
        dbgs() << " pragma-enable";
      dbgs() << "\n";
    }
  });
 
  // Use an explicit peel count that has been specified for testing. In this
  // case it's not permitted to also specify an explicit unroll count.
  if (PP.PeelCount) {
    if (Opts.unroll_count) {
      reportFatalUsageError("Cannot specify both explicit peel count and "
                            "explicit unroll count");
    }
    LLVM_DEBUG(dbgs().indent(2)
               << "Using explicit peel count: " << PP.PeelCount << ".\n");
    UP.Runtime = false;
    return 1;
  }
 
  // If a user provided an explicit unroll pragma (with or without count),
  // enable runtime unrolling and override expensive trip count checks.
  if (PInfo.PragmaEnableUnroll || PInfo.PragmaCount > 0) {
    UP.AllowExpensiveTripCount = true;
    UP.Runtime = true;
  }
 
  // Check for an explicit unroll count.
  // 1st priority is unroll count set by "unroll-count" option.
  // 2nd priority is unroll count set by pragma.
  LLVM_DEBUG(dbgs().indent(1) << "Trying pragma unroll...\n");
  if (auto UnrollFactor =
          shouldPragmaUnroll(Opts, L, PInfo, TripMultiple, TripCount,
                             MaxTripCount, UCE, UP, ORE)) {
    if (PInfo.UserUnrollCount || (PInfo.PragmaCount > 0)) {
      UP.AllowExpensiveTripCount = true;
      UP.Force = true;
    }
    return *UnrollFactor;
  } else {
    if (PInfo.ExplicitUnroll && TripCount != 0) {
      // If the loop has an unrolling pragma, we want to be more aggressive with
      // unrolling limits. Set thresholds to at least the PragmaUnrollThreshold
      // value which is larger than the default limits.
      UP.Threshold = std::max(UP.Threshold, Opts.pragma_unroll_threshold);
      UP.PartialThreshold =
          std::max(UP.PartialThreshold, Opts.pragma_unroll_threshold);
    }
  }
 
  // 3rd priority is exact full unrolling.  This will eliminate all copies
  // of some exit test.
  LLVM_DEBUG(dbgs().indent(1) << "Trying full unroll...\n");
  if (TripCount) {
    if (auto UnrollFactor =
            shouldFullUnroll(L, TTI, DT, SE, EphValues, TripCount, UCE, UP))
      return *UnrollFactor;
  }
 
  // 4th priority is bounded unrolling.
  // We can unroll by the upper bound amount if it's generally allowed or if
  // we know that the loop is executed either the upper bound or zero times.
  // (MaxOrZero unrolling keeps only the first loop test, so the number of
  // loop tests remains the same compared to the non-unrolled version, whereas
  // the generic upper bound unrolling keeps all but the last loop test so the
  // number of loop tests goes up which may end up being worse on targets with
  // constrained branch predictor resources so is controlled by an option.)
  // In addition we only unroll small upper bounds.
  // Note that the cost of bounded unrolling is always strictly greater than
  // cost of exact full unrolling.  As such, if we have an exact count and
  // found it unprofitable, we'll never chose to bounded unroll.
  LLVM_DEBUG(dbgs().indent(1) << "Trying upper-bound unroll...\n");
  if (!TripCount && MaxTripCount && (UP.UpperBound || MaxOrZero) &&
      MaxTripCount <= UP.MaxUpperBound) {
    if (auto UnrollFactor =
            shouldFullUnroll(L, TTI, DT, SE, EphValues, MaxTripCount, UCE, UP))
      return *UnrollFactor;
  }
 
  // 5th priority is loop peeling.
  LLVM_DEBUG(dbgs().indent(1) << "Trying loop peeling...\n");
  computePeelCount(L, LoopSize, PP, TripCount, DT, SE, TTI, AC, UP.Threshold);
  if (PP.PeelCount) {
    LLVM_DEBUG(dbgs().indent(2)
               << "Peeling with count: " << PP.PeelCount << ".\n");
    UP.Runtime = false;
    return 1;
  }
 
  // Before starting partial unrolling, set UP.Partial to true,
  // if user explicitly asked for unrolling.
  if (TripCount)
    UP.Partial |= PInfo.ExplicitUnroll;
 
  // 6th priority is partial unrolling.
  // Try partial unroll only when TripCount could be statically calculated.
  LLVM_DEBUG(dbgs().indent(1) << "Trying partial unroll...\n");
  if (auto UnrollFactor = shouldPartialUnroll(LoopSize, TripCount, UCE, UP))
    return *UnrollFactor;
  assert(TripCount == 0 &&
         "All cases when TripCount is constant should be covered here.");
 
  // 7th priority is runtime unrolling.
  LLVM_DEBUG(dbgs().indent(1) << "Trying runtime unroll...\n");
  return shouldRuntimeUnroll(L, MaxTripCount, TripMultiple, UCE, PInfo, Opts,
                             UP);
}
 
static LoopUnrollResult
tryToUnrollLoop(Loop *L, DominatorTree &DT, LoopInfo *LI, ScalarEvolution &SE,
                const TargetTransformInfo &TTI, AssumptionCache &AC,
                OptimizationRemarkEmitter &ORE, BlockFrequencyInfo *BFI,
                ProfileSummaryInfo *PSI, bool PreserveLCSSA, int OptLevel,
                bool OnlyFullUnroll, bool OnlyWhenForced, bool ForgetAllSCEV,
                bool PrepareForLTO, std::optional<unsigned> ProvidedThreshold,
                std::optional<bool> ProvidedAllowPartial,
                std::optional<bool> ProvidedRuntime,
                std::optional<bool> ProvidedUpperBound,
                std::optional<bool> ProvidedAllowPeeling,
                std::optional<bool> ProvidedAllowProfileBasedPeeling,
                std::optional<unsigned> ProvidedFullUnrollMaxCount,
                UniformityInfo *UI = nullptr, AAResults *AA = nullptr) {
 
  LLVM_DEBUG(dbgs() << "Loop Unroll: F["
                    << L->getHeader()->getParent()->getName() << "] Loop %"
                    << L->getHeader()->getName()
                    << " (depth=" << L->getLoopDepth() << ")\n");
  TransformationMode TM = hasUnrollTransformation(L);
  if (TM & TM_Disable) {
    LLVM_DEBUG(dbgs().indent(1) << "Not unrolling: transformation disabled by "
                                << "metadata.\n");
    return LoopUnrollResult::Unmodified;
  }
 
  // If this loop isn't forced to be unrolled, avoid unrolling it when the
  // parent loop has an explicit unroll-and-jam pragma. This is to prevent
  // automatic unrolling from interfering with the user requested
  // transformation.
  Loop *ParentL = L->getParentLoop();
  if (ParentL != nullptr &&
      hasUnrollAndJamTransformation(ParentL) == TM_ForcedByUser &&
      hasUnrollTransformation(L) != TM_ForcedByUser) {
    LLVM_DEBUG(dbgs().indent(1) << "Not unrolling loop since parent loop has"
                                << " llvm.loop.unroll_and_jam.\n");
    return LoopUnrollResult::Unmodified;
  }
 
  // If this loop isn't forced to be unrolled, avoid unrolling it when the
  // loop has an explicit unroll-and-jam pragma. This is to prevent automatic
  // unrolling from interfering with the user requested transformation.
  if (hasUnrollAndJamTransformation(L) == TM_ForcedByUser &&
      hasUnrollTransformation(L) != TM_ForcedByUser) {
    LLVM_DEBUG(
        dbgs().indent(1)
        << "Not unrolling loop since it has llvm.loop.unroll_and_jam.\n");
    return LoopUnrollResult::Unmodified;
  }
 
  if (!L->isLoopSimplifyForm()) {
    LLVM_DEBUG(dbgs().indent(1)
               << "Not unrolling loop which is not in loop-simplify form.\n");
    if (TM & TM_ForcedByUser) {
      ORE.emit([&]() {
        return OptimizationRemarkMissed(DEBUG_TYPE, "NotInLoopSimplifyForm",
                                        L->getStartLoc(), L->getHeader())
               << "unable to unroll loop: not in loop-simplify form";
      });
    }
    return LoopUnrollResult::Unmodified;
  }
 
  // When automatic unrolling is disabled, do not unroll unless overridden for
  // this loop.
  if (OnlyWhenForced && !(TM & TM_Enable)) {
    LLVM_DEBUG(dbgs().indent(1) << "Not unrolling: automatic unrolling "
                                << "disabled and loop not explicitly "
                                << "enabled.\n");
    return LoopUnrollResult::Unmodified;
  }
 
  bool OptForSize = L->getHeader()->getParent()->hasOptSize();
  TargetTransformInfo::UnrollingPreferences UP = gatherUnrollingPreferences(
      L, SE, TTI, BFI, PSI, ORE, OptLevel, ProvidedThreshold,
      ProvidedAllowPartial, ProvidedRuntime, ProvidedUpperBound,
      ProvidedFullUnrollMaxCount);
  TargetTransformInfo::PeelingPreferences PP = gatherPeelingPreferences(
      L, SE, TTI, ProvidedAllowPeeling, ProvidedAllowProfileBasedPeeling, true);
 
  // Exit early if unrolling is disabled. For OptForSize, we pick the loop size
  // as threshold later on.
  if (UP.Threshold == 0 && (!UP.Partial || UP.PartialThreshold == 0) &&
      !OptForSize) {
    LLVM_DEBUG(dbgs().indent(1) << "Not unrolling: all thresholds are zero.\n");
    if (TM & TM_ForcedByUser) {
      ORE.emit([&]() {
        return OptimizationRemarkMissed(DEBUG_TYPE, "UnrollThresholdsZero",
                                        L->getStartLoc(), L->getHeader())
               << "unable to unroll loop: unroll threshold is zero";
      });
    }
    return LoopUnrollResult::Unmodified;
  }
 
  SmallPtrSet<const Value *, 32> EphValues;
  CodeMetrics::collectEphemeralValues(L, &AC, EphValues);
 
  // Check if the backedge-taken count is uniform before constructing UCE.
  // This is used to allow runtime unrolling with a remainder for convergent
  // loops when all threads agree on the trip count.
  const SCEV *BTC = SE.getBackedgeTakenCount(L);
  bool TripCountIsUniform = UI && isSCEVUniform(BTC, *UI);
  UnrollCostEstimator UCE(L, TTI, EphValues, UP.BEInsns, PrepareForLTO,
                          TripCountIsUniform);
  if (!UCE.canUnroll((TM & TM_ForcedByUser) ? &ORE : nullptr, L))
    return LoopUnrollResult::Unmodified;
 
  unsigned LoopSize = UCE.getRolledLoopSize();
  LLVM_DEBUG(dbgs() << "Loop Size = " << LoopSize << "\n");
 
  // When optimizing for size, use LoopSize + 1 as threshold (we use < Threshold
  // later), to (fully) unroll loops, if it does not increase code size.
  if (OptForSize)
    UP.Threshold = std::max(UP.Threshold, LoopSize + 1);
 
  if (UCE.NumInlineCandidates != 0) {
    LLVM_DEBUG(dbgs().indent(1)
               << "Not unrolling loop with inlinable calls.\n");
    if (TM & TM_ForcedByUser) {
      ORE.emit([&]() {
        return OptimizationRemarkMissed(DEBUG_TYPE,
                                        "InlineCandidatesPreventUnroll",
                                        L->getStartLoc(), L->getHeader())
               << "unable to unroll loop: contains inlinable calls";
      });
    }
    return LoopUnrollResult::Unmodified;
  }
 
  // Find the smallest exact trip count for any exit. This is an upper bound
  // on the loop trip count, but an exit at an earlier iteration is still
  // possible. An unroll by the smallest exact trip count guarantees that all
  // branches relating to at least one exit can be eliminated. This is unlike
  // the max trip count, which only guarantees that the backedge can be broken.
  unsigned TripCount = 0;
  unsigned TripMultiple = 1;
  SmallVector<BasicBlock *, 8> ExitingBlocks;
  L->getExitingBlocks(ExitingBlocks);
  for (BasicBlock *ExitingBlock : ExitingBlocks)
    if (unsigned TC = SE.getSmallConstantTripCount(L, ExitingBlock))
      if (!TripCount || TC < TripCount)
        TripCount = TripMultiple = TC;
 
  if (!TripCount) {
    // If no exact trip count is known, determine the trip multiple of either
    // the loop latch or the single exiting block.
    // TODO: Relax for multiple exits.
    BasicBlock *ExitingBlock = L->getLoopLatch();
    if (!ExitingBlock || !L->isLoopExiting(ExitingBlock))
      ExitingBlock = L->getExitingBlock();
    if (ExitingBlock)
      TripMultiple = SE.getSmallConstantTripMultiple(L, ExitingBlock);
  }
 
  // If the loop contains a convergent operation, the prelude we'd add
  // to do the first few instructions before we hit the unrolled loop
  // is unsafe -- it adds a control-flow dependency to the convergent
  // operation. Therefore restrict remainder loop (try unrolling without).
  UP.AllowRemainder &= UCE.ConvergenceAllowsRuntime;
 
  // Try to find the trip count upper bound if we cannot find the exact trip
  // count.
  unsigned MaxTripCount = 0;
  bool MaxOrZero = false;
  if (!TripCount) {
    MaxTripCount = SE.getSmallConstantMaxTripCount(L);
    MaxOrZero = SE.isBackedgeTakenCountMaxOrZero(L);
  }
 
  // computeUnrollCount() decides whether it is beneficial to use upper bound to
  // fully unroll the loop.
  unsigned Count =
      computeUnrollCount(L, TTI, DT, LI, &AC, SE, EphValues, &ORE, TripCount,
                         MaxTripCount, MaxOrZero, TripMultiple, UCE, UP, PP);
  if (!Count) {
    LLVM_DEBUG(dbgs().indent(1)
               << "Not unrolling: no viable strategy found.\n");
    if (TM & TM_ForcedByUser) {
      ORE.emit([&]() {
        return OptimizationRemarkMissed(DEBUG_TYPE, "NoUnrollStrategy",
                                        L->getStartLoc(), L->getHeader())
               << "unable to unroll loop: no viable unroll count found";
      });
    }
    return LoopUnrollResult::Unmodified;
  }
 
  UP.Runtime &= UCE.ConvergenceAllowsRuntime;
 
  if (PP.PeelCount) {
    assert(Count == 1 && "Cannot perform peel and unroll in the same step");
    LLVM_DEBUG(dbgs() << "PEELING loop %" << L->getHeader()->getName()
                      << " with iteration count " << PP.PeelCount << "!\n");
    ORE.emit([&]() {
      return OptimizationRemark(DEBUG_TYPE, "Peeled", L->getStartLoc(),
                                L->getHeader())
             << "peeled loop by " << ore::NV("PeelCount", PP.PeelCount)
             << " iterations";
    });
 
    ValueToValueMapTy VMap;
    peelLoop(L, PP.PeelCount, PP.PeelLast, LI, &SE, DT, &AC, PreserveLCSSA,
             VMap);
    simplifyLoopAfterUnroll(L, true, LI, &SE, &DT, &AC, &TTI, L->getBlocks(),
                            nullptr);
    // If the loop was peeled, we already "used up" the profile information
    // we had, so we don't want to unroll or peel again.
    if (PP.PeelProfiledIterations)
      L->setLoopAlreadyUnrolled();
    return LoopUnrollResult::PartiallyUnrolled;
  }
 
  // Do not attempt partial/runtime unrolling in FullLoopUnrolling
  if (OnlyFullUnroll && ((!TripCount && !MaxTripCount) || Count < TripCount ||
                         Count < MaxTripCount)) {
    LLVM_DEBUG(dbgs().indent(1)
               << "Not attempting partial/runtime unroll in FullLoopUnroll.\n");
    return LoopUnrollResult::Unmodified;
  }
 
  // At this point, UP.Runtime indicates that run-time unrolling is allowed.
  // However, we only want to actually perform it if we don't know the trip
  // count and the unroll count doesn't divide the known trip multiple.
  // TODO: This decision should probably be pushed up into
  // computeUnrollCount().
  UP.Runtime &= TripCount == 0 && TripMultiple % Count != 0;
 
  // Save loop properties before it is transformed.
  MDNode *OrigLoopID = L->getLoopID();
  UnrollPragmaInfo PInfo(L);
  DebugLoc LoopStartLoc = L->getStartLoc();
  BasicBlock *LoopHeader = L->getHeader();
 
  // Unroll the loop.
  Loop *RemainderLoop = nullptr;
  UnrollLoopOptions ULO;
  ULO.Count = Count;
  ULO.Force = UP.Force;
  ULO.AllowExpensiveTripCount = UP.AllowExpensiveTripCount;
  ULO.UnrollRemainder = UP.UnrollRemainder;
  ULO.Runtime = UP.Runtime;
  ULO.ForgetAllSCEV = ForgetAllSCEV;
  ULO.Heart = getLoopConvergenceHeart(L);
  ULO.SCEVExpansionBudget = UP.SCEVExpansionBudget;
  ULO.RuntimeUnrollMultiExit = UP.RuntimeUnrollMultiExit;
  ULO.AddAdditionalAccumulators = UP.AddAdditionalAccumulators;
  LoopUnrollResult UnrollResult = UnrollLoop(
      L, ULO, LI, &SE, &DT, &AC, &TTI, &ORE, PreserveLCSSA, &RemainderLoop, AA);
  if (UnrollResult == LoopUnrollResult::Unmodified) {
    if (PInfo.ExplicitUnroll) {
      LLVM_DEBUG(dbgs().indent(1)
                 << "Failed to unroll loop as explicitly requested.\n");
      ORE.emit([&]() {
        return OptimizationRemarkMissed(DEBUG_TYPE, "FailedToUnrollAsRequested",
                                        LoopStartLoc, LoopHeader)
               << "failed to unroll loop as explicitly requested";
      });
    }
    return LoopUnrollResult::Unmodified;
  }
 
  if (PInfo.PragmaFullUnroll && ULO.Count != TripCount) {
    ORE.emit([&]() {
      return OptimizationRemarkMissed(DEBUG_TYPE, "FullUnrollAsDirectedFailed",
                                      LoopStartLoc, LoopHeader)
             << "unable to fully unroll loop as directed; "
             << "unrolled by factor " << ore::NV("UnrollCount", ULO.Count);
    });
  }
  if (PInfo.PragmaCount > 0 && ULO.Count != PInfo.PragmaCount) {
    ORE.emit([&]() {
      return OptimizationRemarkMissed(DEBUG_TYPE, "UnrollCountDiffers",
                                      LoopStartLoc, LoopHeader)
             << "unable to unroll loop with requested count "
             << ore::NV("RequestedCount", PInfo.PragmaCount)
             << "; unrolled by factor " << ore::NV("UnrollCount", ULO.Count);
    });
  }
 
  if (RemainderLoop) {
    std::optional<MDNode *> RemainderLoopID =
        makeFollowupLoopID(OrigLoopID, {LLVMLoopUnrollFollowupAll,
                                        LLVMLoopUnrollFollowupRemainder});
    if (RemainderLoopID)
      RemainderLoop->setLoopID(*RemainderLoopID);
  }
 
  if (UnrollResult != LoopUnrollResult::FullyUnrolled) {
    std::optional<MDNode *> NewLoopID =
        makeFollowupLoopID(OrigLoopID, {LLVMLoopUnrollFollowupAll,
                                        LLVMLoopUnrollFollowupUnrolled});
    if (NewLoopID) {
      L->setLoopID(*NewLoopID);
 
      // Do not setLoopAlreadyUnrolled if loop attributes have been specified
      // explicitly.
      return UnrollResult;
    }
  }
 
  // If loop has an unroll count pragma or unrolled by explicitly set count
  // mark loop as unrolled to prevent unrolling beyond that requested.
  if (UnrollResult != LoopUnrollResult::FullyUnrolled && PInfo.ExplicitUnroll)
    L->setLoopAlreadyUnrolled();
 
  return UnrollResult;
}
 
namespace {
 
class LoopUnroll : public LoopPass {
public:
  static char ID; // Pass ID, replacement for typeid
 
  int OptLevel;
 
  /// If false, use a cost model to determine whether unrolling of a loop is
  /// profitable. If true, only loops that explicitly request unrolling via
  /// metadata are considered. All other loops are skipped.
  bool OnlyWhenForced;
 
  /// If false, when SCEV is invalidated, only forget everything in the
  /// top-most loop (call forgetTopMostLoop), of the loop being processed.
  /// Otherwise, forgetAllLoops and rebuild when needed next.
  bool ForgetAllSCEV;
 
  std::optional<unsigned> ProvidedThreshold;
  std::optional<bool> ProvidedAllowPartial;
  std::optional<bool> ProvidedRuntime;
  std::optional<bool> ProvidedUpperBound;
  std::optional<bool> ProvidedAllowPeeling;
  std::optional<bool> ProvidedAllowProfileBasedPeeling;
  std::optional<unsigned> ProvidedFullUnrollMaxCount;
 
  LoopUnroll(int OptLevel = 2, bool OnlyWhenForced = false,
             bool ForgetAllSCEV = false,
             std::optional<unsigned> Threshold = std::nullopt,
             std::optional<bool> AllowPartial = std::nullopt,
             std::optional<bool> Runtime = std::nullopt,
             std::optional<bool> UpperBound = std::nullopt,
             std::optional<bool> AllowPeeling = std::nullopt,
             std::optional<bool> AllowProfileBasedPeeling = std::nullopt,
             std::optional<unsigned> ProvidedFullUnrollMaxCount = std::nullopt)
      : LoopPass(ID), OptLevel(OptLevel), OnlyWhenForced(OnlyWhenForced),
        ForgetAllSCEV(ForgetAllSCEV), ProvidedThreshold(Threshold),
        ProvidedAllowPartial(AllowPartial), ProvidedRuntime(Runtime),
        ProvidedUpperBound(UpperBound), ProvidedAllowPeeling(AllowPeeling),
        ProvidedAllowProfileBasedPeeling(AllowProfileBasedPeeling),
        ProvidedFullUnrollMaxCount(ProvidedFullUnrollMaxCount) {
    initializeLoopUnrollPass(*PassRegistry::getPassRegistry());
  }
 
  bool runOnLoop(Loop *L, LPPassManager &LPM) override {
    if (skipLoop(L))
      return false;
 
    Function &F = *L->getHeader()->getParent();
 
    auto &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
    LoopInfo *LI = &getAnalysis<LoopInfoWrapperPass>().getLoopInfo();
    ScalarEvolution &SE = getAnalysis<ScalarEvolutionWrapperPass>().getSE();
    const TargetTransformInfo &TTI =
        getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
    auto &AC = getAnalysis<AssumptionCacheTracker>().getAssumptionCache(F);
    UniformityInfo *UI =
        TTI.hasBranchDivergence(&F)
            ? &getAnalysis<UniformityInfoWrapperPass>().getUniformityInfo()
            : nullptr;
    // For the old PM, we can't use OptimizationRemarkEmitter as an analysis
    // pass.  Function analyses need to be preserved across loop transformations
    // but ORE cannot be preserved (see comment before the pass definition).
    OptimizationRemarkEmitter ORE(&F);
    bool PreserveLCSSA = mustPreserveAnalysisID(LCSSAID);
 
    LoopUnrollResult Result = tryToUnrollLoop(
        L, DT, LI, SE, TTI, AC, ORE, nullptr, nullptr, PreserveLCSSA, OptLevel,
        /*OnlyFullUnroll*/ false, OnlyWhenForced, ForgetAllSCEV,
        /*PrepareForLTO*/ false, ProvidedThreshold, ProvidedAllowPartial,
        ProvidedRuntime, ProvidedUpperBound, ProvidedAllowPeeling,
        ProvidedAllowProfileBasedPeeling, ProvidedFullUnrollMaxCount, UI);
 
    if (Result == LoopUnrollResult::FullyUnrolled)
      LPM.markLoopAsDeleted(*L);
 
    return Result != LoopUnrollResult::Unmodified;
  }
 
  /// This transformation requires natural loop information & requires that
  /// loop preheaders be inserted into the CFG...
  void getAnalysisUsage(AnalysisUsage &AU) const override {
    AU.addRequired<AssumptionCacheTracker>();
    AU.addRequired<TargetTransformInfoWrapperPass>();
    AU.addRequired<UniformityInfoWrapperPass>();
    // FIXME: Loop passes are required to preserve domtree, and for now we just
    // recreate dom info if anything gets unrolled.
    getLoopAnalysisUsage(AU);
  }
};
 
} // end anonymous namespace
 
char LoopUnroll::ID = 0;
 
INITIALIZE_PASS_BEGIN(LoopUnroll, "loop-unroll", "Unroll loops", false, false)
INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker)
INITIALIZE_PASS_DEPENDENCY(LoopPass)
INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
INITIALIZE_PASS_DEPENDENCY(UniformityInfoWrapperPass)
INITIALIZE_PASS_END(LoopUnroll, "loop-unroll", "Unroll loops", false, false)
 
Pass *llvm::createLoopUnrollPass(int OptLevel, bool OnlyWhenForced,
                                 bool ForgetAllSCEV, int Threshold,
                                 int AllowPartial, int Runtime, int UpperBound,
                                 int AllowPeeling) {
  // TODO: It would make more sense for this function to take the optionals
  // directly, but that's dangerous since it would silently break out of tree
  // callers.
  return new LoopUnroll(
      OptLevel, OnlyWhenForced, ForgetAllSCEV,
      Threshold == -1 ? std::nullopt : std::optional<unsigned>(Threshold),
      AllowPartial == -1 ? std::nullopt : std::optional<bool>(AllowPartial),
      Runtime == -1 ? std::nullopt : std::optional<bool>(Runtime),
      UpperBound == -1 ? std::nullopt : std::optional<bool>(UpperBound),
      AllowPeeling == -1 ? std::nullopt : std::optional<bool>(AllowPeeling));
}
 
PreservedAnalyses LoopFullUnrollPass::run(Loop &L, LoopAnalysisManager &AM,
                                          LoopStandardAnalysisResults &AR,
                                          LPMUpdater &Updater) {
  // For the new PM, we can't use OptimizationRemarkEmitter as an analysis
  // pass. Function analyses need to be preserved across loop transformations
  // but ORE cannot be preserved (see comment before the pass definition).
  OptimizationRemarkEmitter ORE(L.getHeader()->getParent());
  const ScalarOptions &Opts = getScalarOptions(L);
 
  // Keep track of the previous loop structure so we can identify new loops
  // created by unrolling.
  Loop *ParentL = L.getParentLoop();
  SmallPtrSet<Loop *, 4> OldLoops;
  if (ParentL)
    OldLoops.insert_range(*ParentL);
  else
    OldLoops.insert_range(AR.LI);
 
  std::string LoopName = std::string(L.getName());
 
  bool Changed =
      tryToUnrollLoop(&L, AR.DT, &AR.LI, AR.SE, AR.TTI, AR.AC, ORE,
                      /*BFI*/ nullptr, /*PSI*/ nullptr,
                      /*PreserveLCSSA*/ true, OptLevel, /*OnlyFullUnroll*/ true,
                      OnlyWhenForced, ForgetSCEV, PrepareForLTO,
                      /*Threshold*/ std::nullopt, /*AllowPartial*/ false,
                      /*Runtime*/ false, /*UpperBound*/ false,
                      /*AllowPeeling*/ true,
                      /*AllowProfileBasedPeeling*/ false,
                      /*FullUnrollMaxCount*/ std::nullopt) !=
      LoopUnrollResult::Unmodified;
  if (!Changed)
    return PreservedAnalyses::all();
 
  // The parent must not be damaged by unrolling!
#ifndef NDEBUG
  if (ParentL)
    ParentL->verifyLoop();
#endif
 
  // Unrolling can do several things to introduce new loops into a loop nest:
  // - Full unrolling clones child loops within the current loop but then
  //   removes the current loop making all of the children appear to be new
  //   sibling loops.
  //
  // When a new loop appears as a sibling loop after fully unrolling,
  // its nesting structure has fundamentally changed and we want to revisit
  // it to reflect that.
  //
  // When unrolling has removed the current loop, we need to tell the
  // infrastructure that it is gone.
  //
  // Finally, we support a debugging/testing mode where we revisit child loops
  // as well. These are not expected to require further optimizations as either
  // they or the loop they were cloned from have been directly visited already.
  // But the debugging mode allows us to check this assumption.
  bool IsCurrentLoopValid = false;
  SmallVector<Loop *, 4> SibLoops;
  if (ParentL)
    SibLoops.append(ParentL->begin(), ParentL->end());
  else
    SibLoops.append(AR.LI.begin(), AR.LI.end());
  erase_if(SibLoops, [&](Loop *SibLoop) {
    if (SibLoop == &L) {
      IsCurrentLoopValid = true;
      return true;
    }
 
    // Otherwise erase the loop from the list if it was in the old loops.
    return OldLoops.contains(SibLoop);
  });
  Updater.addSiblingLoops(SibLoops);
 
  if (!IsCurrentLoopValid) {
    Updater.markLoopAsDeleted(L, LoopName);
  } else {
    // We can only walk child loops if the current loop remained valid.
    if (Opts.unroll_revisit_child_loops) {
      // Walk *all* of the child loops.
      SmallVector<Loop *, 4> ChildLoops(L.begin(), L.end());
      Updater.addChildLoops(ChildLoops);
    }
  }
 
  return getLoopPassPreservedAnalyses();
}
 
PreservedAnalyses LoopUnrollPass::run(Function &F,
                                      FunctionAnalysisManager &AM) {
  auto &LI = AM.getResult<LoopAnalysis>(F);
  // There are no loops in the function. Return before computing other expensive
  // analyses.
  if (LI.empty())
    return PreservedAnalyses::all();
  auto &SE = AM.getResult<ScalarEvolutionAnalysis>(F);
  auto &TTI = AM.getResult<TargetIRAnalysis>(F);
  auto &DT = AM.getResult<DominatorTreeAnalysis>(F);
  auto &AC = AM.getResult<AssumptionAnalysis>(F);
  auto &ORE = AM.getResult<OptimizationRemarkEmitterAnalysis>(F);
  AAResults &AA = AM.getResult<AAManager>(F);
 
  UniformityInfo *UI = TTI.hasBranchDivergence(&F)
                           ? &AM.getResult<UniformityInfoAnalysis>(F)
                           : nullptr;
 
  LoopAnalysisManager *LAM = nullptr;
  if (auto *LAMProxy = AM.getCachedResult<LoopAnalysisManagerFunctionProxy>(F))
    LAM = &LAMProxy->getManager();
 
  auto &MAMProxy = AM.getResult<ModuleAnalysisManagerFunctionProxy>(F);
  ProfileSummaryInfo *PSI =
      MAMProxy.getCachedResult<ProfileSummaryAnalysis>(*F.getParent());
  auto *BFI = (PSI && PSI->hasProfileSummary()) ?
      &AM.getResult<BlockFrequencyAnalysis>(F) : nullptr;
 
  bool Changed = false;
 
  // The unroller requires loops to be in simplified form, and also needs LCSSA.
  // Since simplification may add new inner loops, it has to run before the
  // legality and profitability checks. This means running the loop unroller
  // will simplify all loops, regardless of whether anything end up being
  // unrolled.
  for (const auto &L : LI) {
    Changed |=
        simplifyLoop(L, &DT, &LI, &SE, &AC, nullptr, false /* PreserveLCSSA */);
    Changed |= formLCSSARecursively(*L, DT, &LI, &SE);
  }
 
  // Add the loop nests in the reverse order of LoopInfo. See method
  // declaration.
  SmallPriorityWorklist<Loop *, 4> Worklist;
  appendLoopsToWorklist(LI, Worklist);
 
  // Cache the number of loops in each top-level nest. The gated transforms
  // never add loops to the worklist, so the counts stay a safe upper bound.
  const unsigned MaxNestLoops = getScalarOptions(F).unroll_max_nest_loops;
  DenseMap<const Loop *, unsigned> TopLevelNestSize;
  if (MaxNestLoops)
    for (const Loop *TopL : LI)
      TopLevelNestSize[TopL] = TopL->getLoopsInPreorder().size();
 
  while (!Worklist.empty()) {
    // Because the LoopInfo stores the loops in RPO, we walk the worklist
    // from back to front so that we work forward across the CFG, which
    // for unrolling is only needed to get optimization remarks emitted in
    // a forward order.
    Loop &L = *Worklist.pop_back_val();
#ifndef NDEBUG
    Loop *ParentL = L.getParentLoop();
#endif
 
    // Check if the profile summary indicates that the profiled application
    // has a huge working set size, in which case we disable peeling to avoid
    // bloating it further.
    std::optional<bool> LocalAllowPeeling = UnrollOpts.AllowPeeling;
    if (PSI && PSI->hasHugeWorkingSetSize())
      LocalAllowPeeling = false;
 
    // Suppress heuristic runtime/partial unrolling and peeling for loops in a
    // pathologically large nest, where each such transform forces a whole-nest
    // SCEV invalidation (quadratic compile time). Explicit unroll pragmas still
    // win: gatherUnrollingPreferences applies these overrides before
    // computeUnrollCount re-enables unrolling for pragma-marked loops.
    std::optional<bool> LocalAllowRuntime = UnrollOpts.AllowRuntime;
    std::optional<bool> LocalAllowPartial = UnrollOpts.AllowPartial;
    if (MaxNestLoops) {
      unsigned NestSize = TopLevelNestSize.lookup(L.getOutermostLoop());
      if (NestSize > MaxNestLoops) {
        LocalAllowRuntime = false;
        LocalAllowPartial = false;
        LocalAllowPeeling = false;
        LLVM_DEBUG(dbgs().indent(1)
                   << "Disabling runtime/partial unroll and peeling: "
                      "outermost nest has "
                   << NestSize << " loops (> " << MaxNestLoops << ").\n");
      }
    }
 
    std::string LoopName = std::string(L.getName());
    // The API here is quite complex to call and we allow to select some
    // flavors of unrolling during construction time (by setting UnrollOpts).
    LoopUnrollResult Result =
        tryToUnrollLoop(&L, DT, &LI, SE, TTI, AC, ORE, BFI, PSI,
                        /*PreserveLCSSA*/ true, UnrollOpts.OptLevel,
                        /*OnlyFullUnroll*/ false, UnrollOpts.OnlyWhenForced,
                        UnrollOpts.ForgetSCEV, UnrollOpts.PrepareForLTO,
                        /*Threshold*/ std::nullopt, LocalAllowPartial,
                        LocalAllowRuntime, UnrollOpts.AllowUpperBound,
                        LocalAllowPeeling, UnrollOpts.AllowProfileBasedPeeling,
                        UnrollOpts.FullUnrollMaxCount, UI, &AA);
    Changed |= Result != LoopUnrollResult::Unmodified;
 
    // The parent must not be damaged by unrolling!
#ifndef NDEBUG
    if (Result != LoopUnrollResult::Unmodified && ParentL)
      ParentL->verifyLoop();
#endif
 
    // Clear any cached analysis results for L if we removed it completely.
    if (LAM && Result == LoopUnrollResult::FullyUnrolled)
      LAM->clear(L, LoopName);
  }
 
  if (!Changed)
    return PreservedAnalyses::all();
 
  return getLoopPassPreservedAnalyses();
}
 
void LoopUnrollPass::printPipeline(
    raw_ostream &OS, function_ref<StringRef(StringRef)> MapClassName2PassName) {
  static_cast<PassInfoMixin<LoopUnrollPass> *>(this)->printPipeline(
      OS, MapClassName2PassName);
  OS << '<';
  if (UnrollOpts.AllowPartial != std::nullopt)
    OS << (*UnrollOpts.AllowPartial ? "" : "no-") << "partial;";
  if (UnrollOpts.AllowPeeling != std::nullopt)
    OS << (*UnrollOpts.AllowPeeling ? "" : "no-") << "peeling;";
  if (UnrollOpts.AllowRuntime != std::nullopt)
    OS << (*UnrollOpts.AllowRuntime ? "" : "no-") << "runtime;";
  if (UnrollOpts.AllowUpperBound != std::nullopt)
    OS << (*UnrollOpts.AllowUpperBound ? "" : "no-") << "upperbound;";
  if (UnrollOpts.AllowProfileBasedPeeling != std::nullopt)
    OS << (*UnrollOpts.AllowProfileBasedPeeling ? "" : "no-")
       << "profile-peeling;";
  if (UnrollOpts.FullUnrollMaxCount != std::nullopt)
    OS << "full-unroll-max=" << UnrollOpts.FullUnrollMaxCount << ';';
  if (UnrollOpts.PrepareForLTO)
    OS << "prepare-for-lto;";
  OS << 'O' << UnrollOpts.OptLevel;
  OS << '>';
}