blob: f76fa3bb6c6114422668d5e929360f513a6d603d [file] [log] [blame]
//===- LoopPeel.cpp -------------------------------------------------------===//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
// Loop Peeling Utilities.
#include "llvm/Transforms/Utils/LoopPeel.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Analysis/Loads.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/LoopIterator.h"
#include "llvm/Analysis/ScalarEvolution.h"
#include "llvm/Analysis/ScalarEvolutionExpressions.h"
#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/IR/BasicBlock.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/LLVMContext.h"
#include "llvm/IR/MDBuilder.h"
#include "llvm/IR/PatternMatch.h"
#include "llvm/IR/ProfDataUtils.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Cloning.h"
#include "llvm/Transforms/Utils/LoopSimplify.h"
#include "llvm/Transforms/Utils/LoopUtils.h"
#include "llvm/Transforms/Utils/ValueMapper.h"
#include <algorithm>
#include <cassert>
#include <cstdint>
#include <optional>
using namespace llvm;
using namespace llvm::PatternMatch;
#define DEBUG_TYPE "loop-peel"
STATISTIC(NumPeeled, "Number of loops peeled");
static cl::opt<unsigned> UnrollPeelCount(
"unroll-peel-count", cl::Hidden,
cl::desc("Set the unroll peeling count, for testing purposes"));
static cl::opt<bool>
UnrollAllowPeeling("unroll-allow-peeling", cl::init(true), cl::Hidden,
cl::desc("Allows loops to be peeled when the dynamic "
"trip count is known to be low."));
static cl::opt<bool>
cl::init(false), cl::Hidden,
cl::desc("Allows loop nests to be peeled."));
static cl::opt<unsigned> UnrollPeelMaxCount(
"unroll-peel-max-count", cl::init(7), cl::Hidden,
cl::desc("Max average trip count which will cause loop peeling."));
static cl::opt<unsigned> UnrollForcePeelCount(
"unroll-force-peel-count", cl::init(0), cl::Hidden,
cl::desc("Force a peel count regardless of profiling information."));
static cl::opt<bool> DisableAdvancedPeeling(
"disable-advanced-peeling", cl::init(false), cl::Hidden,
"Disable advance peeling. Issues for convergent targets (D134803)."));
static const char *PeeledCountMetaData = "llvm.loop.peeled.count";
// Check whether we are capable of peeling this loop.
bool llvm::canPeel(const Loop *L) {
// Make sure the loop is in simplified form
if (!L->isLoopSimplifyForm())
return false;
if (!DisableAdvancedPeeling)
return true;
SmallVector<BasicBlock *, 4> Exits;
// The latch must either be the only exiting block or all non-latch exit
// blocks have either a deopt or unreachable terminator or compose a chain of
// blocks where the last one is either deopt or unreachable terminated. Both
// deopt and unreachable terminators are a strong indication they are not
// taken. Note that this is a profitability check, not a legality check. Also
// note that LoopPeeling currently can only update the branch weights of latch
// blocks and branch weights to blocks with deopt or unreachable do not need
// updating.
return llvm::all_of(Exits, IsBlockFollowedByDeoptOrUnreachable);
namespace {
// As a loop is peeled, it may be the case that Phi nodes become
// loop-invariant (ie, known because there is only one choice).
// For example, consider the following function:
// void g(int);
// void binary() {
// int x = 0;
// int y = 0;
// int a = 0;
// for(int i = 0; i <100000; ++i) {
// g(x);
// x = y;
// g(a);
// y = a + 1;
// a = 5;
// }
// }
// Peeling 3 iterations is beneficial because the values for x, y and a
// become known. The IR for this loop looks something like the following:
// %i = phi i32 [ 0, %entry ], [ %inc, %if.end ]
// %a = phi i32 [ 0, %entry ], [ 5, %if.end ]
// %y = phi i32 [ 0, %entry ], [ %add, %if.end ]
// %x = phi i32 [ 0, %entry ], [ %y, %if.end ]
// ...
// tail call void @_Z1gi(i32 signext %x)
// tail call void @_Z1gi(i32 signext %a)
// %add = add nuw nsw i32 %a, 1
// %inc = add nuw nsw i32 %i, 1
// %exitcond = icmp eq i32 %inc, 100000
// br i1 %exitcond, label %for.cond.cleanup, label %for.body
// The arguments for the calls to g will become known after 3 iterations
// of the loop, because the phi nodes values become known after 3 iterations
// of the loop (ie, they are known on the 4th iteration, so peel 3 iterations).
// The first iteration has g(0), g(0); the second has g(0), g(5); the
// third has g(1), g(5) and the fourth (and all subsequent) have g(6), g(5).
// Now consider the phi nodes:
// %a is a phi with constants so it is determined after iteration 1.
// %y is a phi based on a constant and %a so it is determined on
// the iteration after %a is determined, so iteration 2.
// %x is a phi based on a constant and %y so it is determined on
// the iteration after %y, so iteration 3.
// %i is based on itself (and is an induction variable) so it is
// never determined.
// This means that peeling off 3 iterations will result in being able to
// remove the phi nodes for %a, %y, and %x. The arguments for the
// corresponding calls to g are determined and the code for computing
// x, y, and a can be removed.
// The PhiAnalyzer class calculates how many times a loop should be
// peeled based on the above analysis of the phi nodes in the loop while
// respecting the maximum specified.
class PhiAnalyzer {
PhiAnalyzer(const Loop &L, unsigned MaxIterations);
// Calculate the sufficient minimum number of iterations of the loop to peel
// such that phi instructions become determined (subject to allowable limits)
std::optional<unsigned> calculateIterationsToPeel();
using PeelCounter = std::optional<unsigned>;
const PeelCounter Unknown = std::nullopt;
// Add 1 respecting Unknown and return Unknown if result over MaxIterations
PeelCounter addOne(PeelCounter PC) const {
if (PC == Unknown)
return Unknown;
return (*PC + 1 <= MaxIterations) ? PeelCounter{*PC + 1} : Unknown;
// Calculate the number of iterations after which the given value
// becomes an invariant.
PeelCounter calculate(const Value &);
const Loop &L;
const unsigned MaxIterations;
// Map of Values to number of iterations to invariance
SmallDenseMap<const Value *, PeelCounter> IterationsToInvariance;
PhiAnalyzer::PhiAnalyzer(const Loop &L, unsigned MaxIterations)
: L(L), MaxIterations(MaxIterations) {
assert(canPeel(&L) && "loop is not suitable for peeling");
assert(MaxIterations > 0 && "no peeling is allowed?");
// This function calculates the number of iterations after which the value
// becomes an invariant. The pre-calculated values are memorized in a map.
// N.B. This number will be Unknown or <= MaxIterations.
// The function is calculated according to the following definition:
// Given %x = phi <Inputs from above the loop>, ..., [%y, %back.edge].
// F(%x) = G(%y) + 1 (N.B. [MaxIterations | Unknown] + 1 => Unknown)
// G(%y) = 0 if %y is a loop invariant
// G(%y) = G(%BackEdgeValue) if %y is a phi in the header block
// G(%y) = TODO: if %y is an expression based on phis and loop invariants
// The example looks like:
// %x = phi(0, %a) <-- becomes invariant starting from 3rd iteration.
// %y = phi(0, 5)
// %a = %y + 1
// G(%y) = Unknown otherwise (including phi not in header block)
PhiAnalyzer::PeelCounter PhiAnalyzer::calculate(const Value &V) {
// If we already know the answer, take it from the map.
auto I = IterationsToInvariance.find(&V);
if (I != IterationsToInvariance.end())
return I->second;
// Place Unknown to map to avoid infinite recursion. Such
// cycles can never stop on an invariant.
IterationsToInvariance[&V] = Unknown;
if (L.isLoopInvariant(&V))
// Loop invariant so known at start.
return (IterationsToInvariance[&V] = 0);
if (const PHINode *Phi = dyn_cast<PHINode>(&V)) {
if (Phi->getParent() != L.getHeader()) {
// Phi is not in header block so Unknown.
assert(IterationsToInvariance[&V] == Unknown && "unexpected value saved");
return Unknown;
// We need to analyze the input from the back edge and add 1.
Value *Input = Phi->getIncomingValueForBlock(L.getLoopLatch());
PeelCounter Iterations = calculate(*Input);
assert(IterationsToInvariance[Input] == Iterations &&
"unexpected value saved");
return (IterationsToInvariance[Phi] = addOne(Iterations));
if (const Instruction *I = dyn_cast<Instruction>(&V)) {
if (isa<CmpInst>(I) || I->isBinaryOp()) {
// Binary instructions get the max of the operands.
PeelCounter LHS = calculate(*I->getOperand(0));
if (LHS == Unknown)
return Unknown;
PeelCounter RHS = calculate(*I->getOperand(1));
if (RHS == Unknown)
return Unknown;
return (IterationsToInvariance[I] = {std::max(*LHS, *RHS)});
if (I->isCast())
// Cast instructions get the value of the operand.
return (IterationsToInvariance[I] = calculate(*I->getOperand(0)));
// TODO: handle more expressions
// Everything else is Unknown.
assert(IterationsToInvariance[&V] == Unknown && "unexpected value saved");
return Unknown;
std::optional<unsigned> PhiAnalyzer::calculateIterationsToPeel() {
unsigned Iterations = 0;
for (auto &PHI : L.getHeader()->phis()) {
PeelCounter ToInvariance = calculate(PHI);
if (ToInvariance != Unknown) {
assert(*ToInvariance <= MaxIterations && "bad result in phi analysis");
Iterations = std::max(Iterations, *ToInvariance);
if (Iterations == MaxIterations)
assert((Iterations <= MaxIterations) && "bad result in phi analysis");
return Iterations ? std::optional<unsigned>(Iterations) : std::nullopt;
} // unnamed namespace
// Try to find any invariant memory reads that will become dereferenceable in
// the remainder loop after peeling. The load must also be used (transitively)
// by an exit condition. Returns the number of iterations to peel off (at the
// moment either 0 or 1).
static unsigned peelToTurnInvariantLoadsDerefencebale(Loop &L,
DominatorTree &DT,
AssumptionCache *AC) {
// Skip loops with a single exiting block, because there should be no benefit
// for the heuristic below.
if (L.getExitingBlock())
return 0;
// All non-latch exit blocks must have an UnreachableInst terminator.
// Otherwise the heuristic below may not be profitable.
SmallVector<BasicBlock *, 4> Exits;
if (any_of(Exits, [](const BasicBlock *BB) {
return !isa<UnreachableInst>(BB->getTerminator());
return 0;
// Now look for invariant loads that dominate the latch and are not known to
// be dereferenceable. If there are such loads and no writes, they will become
// dereferenceable in the loop if the first iteration is peeled off. Also
// collect the set of instructions controlled by such loads. Only peel if an
// exit condition uses (transitively) such a load.
BasicBlock *Header = L.getHeader();
BasicBlock *Latch = L.getLoopLatch();
SmallPtrSet<Value *, 8> LoadUsers;
const DataLayout &DL = L.getHeader()->getModule()->getDataLayout();
for (BasicBlock *BB : L.blocks()) {
for (Instruction &I : *BB) {
if (I.mayWriteToMemory())
return 0;
auto Iter = LoadUsers.find(&I);
if (Iter != LoadUsers.end()) {
for (Value *U : I.users())
// Do not look for reads in the header; they can already be hoisted
// without peeling.
if (BB == Header)
if (auto *LI = dyn_cast<LoadInst>(&I)) {
Value *Ptr = LI->getPointerOperand();
if (DT.dominates(BB, Latch) && L.isLoopInvariant(Ptr) &&
!isDereferenceablePointer(Ptr, LI->getType(), DL, LI, AC, &DT))
for (Value *U : I.users())
SmallVector<BasicBlock *> ExitingBlocks;
if (any_of(ExitingBlocks, [&LoadUsers](BasicBlock *Exiting) {
return LoadUsers.contains(Exiting->getTerminator());
return 1;
return 0;
// Return the number of iterations to peel off that make conditions in the
// body true/false. For example, if we peel 2 iterations off the loop below,
// the condition i < 2 can be evaluated at compile time.
// for (i = 0; i < n; i++)
// if (i < 2)
// ..
// else
// ..
// }
static unsigned countToEliminateCompares(Loop &L, unsigned MaxPeelCount,
ScalarEvolution &SE) {
assert(L.isLoopSimplifyForm() && "Loop needs to be in loop simplify form");
unsigned DesiredPeelCount = 0;
// Do not peel the entire loop.
const SCEV *BE = SE.getConstantMaxBackedgeTakenCount(&L);
if (const SCEVConstant *SC = dyn_cast<SCEVConstant>(BE))
MaxPeelCount =
std::min((unsigned)SC->getAPInt().getLimitedValue() - 1, MaxPeelCount);
const unsigned MaxDepth = 4;
std::function<void(Value *, unsigned)> ComputePeelCount =
[&](Value *Condition, unsigned Depth) -> void {
if (!Condition->getType()->isIntegerTy() || Depth >= MaxDepth)
Value *LeftVal, *RightVal;
if (match(Condition, m_And(m_Value(LeftVal), m_Value(RightVal))) ||
match(Condition, m_Or(m_Value(LeftVal), m_Value(RightVal)))) {
ComputePeelCount(LeftVal, Depth + 1);
ComputePeelCount(RightVal, Depth + 1);
CmpInst::Predicate Pred;
if (!match(Condition, m_ICmp(Pred, m_Value(LeftVal), m_Value(RightVal))))
const SCEV *LeftSCEV = SE.getSCEV(LeftVal);
const SCEV *RightSCEV = SE.getSCEV(RightVal);
// Do not consider predicates that are known to be true or false
// independently of the loop iteration.
if (SE.evaluatePredicate(Pred, LeftSCEV, RightSCEV))
// Check if we have a condition with one AddRec and one non AddRec
// expression. Normalize LeftSCEV to be the AddRec.
if (!isa<SCEVAddRecExpr>(LeftSCEV)) {
if (isa<SCEVAddRecExpr>(RightSCEV)) {
std::swap(LeftSCEV, RightSCEV);
Pred = ICmpInst::getSwappedPredicate(Pred);
} else
const SCEVAddRecExpr *LeftAR = cast<SCEVAddRecExpr>(LeftSCEV);
// Avoid huge SCEV computations in the loop below, make sure we only
// consider AddRecs of the loop we are trying to peel.
if (!LeftAR->isAffine() || LeftAR->getLoop() != &L)
if (!(ICmpInst::isEquality(Pred) && LeftAR->hasNoSelfWrap()) &&
!SE.getMonotonicPredicateType(LeftAR, Pred))
// Check if extending the current DesiredPeelCount lets us evaluate Pred
// or !Pred in the loop body statically.
unsigned NewPeelCount = DesiredPeelCount;
const SCEV *IterVal = LeftAR->evaluateAtIteration(
SE.getConstant(LeftSCEV->getType(), NewPeelCount), SE);
// If the original condition is not known, get the negated predicate
// (which holds on the else branch) and check if it is known. This allows
// us to peel of iterations that make the original condition false.
if (!SE.isKnownPredicate(Pred, IterVal, RightSCEV))
Pred = ICmpInst::getInversePredicate(Pred);
const SCEV *Step = LeftAR->getStepRecurrence(SE);
const SCEV *NextIterVal = SE.getAddExpr(IterVal, Step);
auto PeelOneMoreIteration = [&IterVal, &NextIterVal, &SE, Step,
&NewPeelCount]() {
IterVal = NextIterVal;
NextIterVal = SE.getAddExpr(IterVal, Step);
auto CanPeelOneMoreIteration = [&NewPeelCount, &MaxPeelCount]() {
return NewPeelCount < MaxPeelCount;
while (CanPeelOneMoreIteration() &&
SE.isKnownPredicate(Pred, IterVal, RightSCEV))
// With *that* peel count, does the predicate !Pred become known in the
// first iteration of the loop body after peeling?
if (!SE.isKnownPredicate(ICmpInst::getInversePredicate(Pred), IterVal,
return; // If not, give up.
// However, for equality comparisons, that isn't always sufficient to
// eliminate the comparsion in loop body, we may need to peel one more
// iteration. See if that makes !Pred become unknown again.
if (ICmpInst::isEquality(Pred) &&
!SE.isKnownPredicate(ICmpInst::getInversePredicate(Pred), NextIterVal,
RightSCEV) &&
!SE.isKnownPredicate(Pred, IterVal, RightSCEV) &&
SE.isKnownPredicate(Pred, NextIterVal, RightSCEV)) {
if (!CanPeelOneMoreIteration())
return; // Need to peel one more iteration, but can't. Give up.
PeelOneMoreIteration(); // Great!
DesiredPeelCount = std::max(DesiredPeelCount, NewPeelCount);
for (BasicBlock *BB : L.blocks()) {
for (Instruction &I : *BB) {
if (SelectInst *SI = dyn_cast<SelectInst>(&I))
ComputePeelCount(SI->getCondition(), 0);
auto *BI = dyn_cast<BranchInst>(BB->getTerminator());
if (!BI || BI->isUnconditional())
// Ignore loop exit condition.
if (L.getLoopLatch() == BB)
ComputePeelCount(BI->getCondition(), 0);
return DesiredPeelCount;
/// This "heuristic" exactly matches implicit behavior which used to exist
/// inside getLoopEstimatedTripCount. It was added here to keep an
/// improvement inside that API from causing peeling to become more aggressive.
/// This should probably be removed.
static bool violatesLegacyMultiExitLoopCheck(Loop *L) {
BasicBlock *Latch = L->getLoopLatch();
if (!Latch)
return true;
BranchInst *LatchBR = dyn_cast<BranchInst>(Latch->getTerminator());
if (!LatchBR || LatchBR->getNumSuccessors() != 2 || !L->isLoopExiting(Latch))
return true;
assert((LatchBR->getSuccessor(0) == L->getHeader() ||
LatchBR->getSuccessor(1) == L->getHeader()) &&
"At least one edge out of the latch must go to the header");
SmallVector<BasicBlock *, 4> ExitBlocks;
return any_of(ExitBlocks, [](const BasicBlock *EB) {
return !EB->getTerminatingDeoptimizeCall();
// Return the number of iterations we want to peel off.
void llvm::computePeelCount(Loop *L, unsigned LoopSize,
TargetTransformInfo::PeelingPreferences &PP,
unsigned TripCount, DominatorTree &DT,
ScalarEvolution &SE, AssumptionCache *AC,
unsigned Threshold) {
assert(LoopSize > 0 && "Zero loop size is not allowed!");
// Save the PP.PeelCount value set by the target in
// TTI.getPeelingPreferences or by the flag -unroll-peel-count.
unsigned TargetPeelCount = PP.PeelCount;
PP.PeelCount = 0;
if (!canPeel(L))
// Only try to peel innermost loops by default.
// The constraint can be relaxed by the target in TTI.getPeelingPreferences
// or by the flag -unroll-allow-loop-nests-peeling.
if (!PP.AllowLoopNestsPeeling && !L->isInnermost())
// If the user provided a peel count, use that.
bool UserPeelCount = UnrollForcePeelCount.getNumOccurrences() > 0;
if (UserPeelCount) {
LLVM_DEBUG(dbgs() << "Force-peeling first " << UnrollForcePeelCount
<< " iterations.\n");
PP.PeelCount = UnrollForcePeelCount;
PP.PeelProfiledIterations = true;
// Skip peeling if it's disabled.
if (!PP.AllowPeeling)
// Check that we can peel at least one iteration.
if (2 * LoopSize > Threshold)
unsigned AlreadyPeeled = 0;
if (auto Peeled = getOptionalIntLoopAttribute(L, PeeledCountMetaData))
AlreadyPeeled = *Peeled;
// Stop if we already peeled off the maximum number of iterations.
if (AlreadyPeeled >= UnrollPeelMaxCount)
// Pay respect to limitations implied by loop size and the max peel count.
unsigned MaxPeelCount = UnrollPeelMaxCount;
MaxPeelCount = std::min(MaxPeelCount, Threshold / LoopSize - 1);
// Start the max computation with the PP.PeelCount value set by the target
// in TTI.getPeelingPreferences or by the flag -unroll-peel-count.
unsigned DesiredPeelCount = TargetPeelCount;
// Here we try to get rid of Phis which become invariants after 1, 2, ..., N
// iterations of the loop. For this we compute the number for iterations after
// which every Phi is guaranteed to become an invariant, and try to peel the
// maximum number of iterations among these values, thus turning all those
// Phis into invariants.
if (MaxPeelCount > DesiredPeelCount) {
// Check how many iterations are useful for resolving Phis
auto NumPeels = PhiAnalyzer(*L, MaxPeelCount).calculateIterationsToPeel();
if (NumPeels)
DesiredPeelCount = std::max(DesiredPeelCount, *NumPeels);
DesiredPeelCount = std::max(DesiredPeelCount,
countToEliminateCompares(*L, MaxPeelCount, SE));
if (DesiredPeelCount == 0)
DesiredPeelCount = peelToTurnInvariantLoadsDerefencebale(*L, DT, AC);
if (DesiredPeelCount > 0) {
DesiredPeelCount = std::min(DesiredPeelCount, MaxPeelCount);
// Consider max peel count limitation.
assert(DesiredPeelCount > 0 && "Wrong loop size estimation?");
if (DesiredPeelCount + AlreadyPeeled <= UnrollPeelMaxCount) {
LLVM_DEBUG(dbgs() << "Peel " << DesiredPeelCount
<< " iteration(s) to turn"
<< " some Phis into invariants.\n");
PP.PeelCount = DesiredPeelCount;
PP.PeelProfiledIterations = false;
// Bail if we know the statically calculated trip count.
// In this case we rather prefer partial unrolling.
if (TripCount)
// Do not apply profile base peeling if it is disabled.
if (!PP.PeelProfiledIterations)
// If we don't know the trip count, but have reason to believe the average
// trip count is low, peeling should be beneficial, since we will usually
// hit the peeled section.
// We only do this in the presence of profile information, since otherwise
// our estimates of the trip count are not reliable enough.
if (L->getHeader()->getParent()->hasProfileData()) {
if (violatesLegacyMultiExitLoopCheck(L))
std::optional<unsigned> EstimatedTripCount = getLoopEstimatedTripCount(L);
if (!EstimatedTripCount)
LLVM_DEBUG(dbgs() << "Profile-based estimated trip count is "
<< *EstimatedTripCount << "\n");
if (*EstimatedTripCount) {
if (*EstimatedTripCount + AlreadyPeeled <= MaxPeelCount) {
unsigned PeelCount = *EstimatedTripCount;
LLVM_DEBUG(dbgs() << "Peeling first " << PeelCount << " iterations.\n");
PP.PeelCount = PeelCount;
LLVM_DEBUG(dbgs() << "Already peel count: " << AlreadyPeeled << "\n");
LLVM_DEBUG(dbgs() << "Max peel count: " << UnrollPeelMaxCount << "\n");
LLVM_DEBUG(dbgs() << "Loop cost: " << LoopSize << "\n");
LLVM_DEBUG(dbgs() << "Max peel cost: " << Threshold << "\n");
LLVM_DEBUG(dbgs() << "Max peel count by cost: "
<< (Threshold / LoopSize - 1) << "\n");
struct WeightInfo {
// Weights for current iteration.
SmallVector<uint32_t> Weights;
// Weights to subtract after each iteration.
const SmallVector<uint32_t> SubWeights;
/// Update the branch weights of an exiting block of a peeled-off loop
/// iteration.
/// Let F is a weight of the edge to continue (fallthrough) into the loop.
/// Let E is a weight of the edge to an exit.
/// F/(F+E) is a probability to go to loop and E/(F+E) is a probability to
/// go to exit.
/// Then, Estimated ExitCount = F / E.
/// For I-th (counting from 0) peeled off iteration we set the weights for
/// the peeled exit as (EC - I, 1). It gives us reasonable distribution,
/// The probability to go to exit 1/(EC-I) increases. At the same time
/// the estimated exit count in the remainder loop reduces by I.
/// To avoid dealing with division rounding we can just multiple both part
/// of weights to E and use weight as (F - I * E, E).
static void updateBranchWeights(Instruction *Term, WeightInfo &Info) {
setBranchWeights(*Term, Info.Weights);
for (auto [Idx, SubWeight] : enumerate(Info.SubWeights))
if (SubWeight != 0)
// Don't set the probability of taking the edge from latch to loop header
// to less than 1:1 ratio (meaning Weight should not be lower than
// SubWeight), as this could significantly reduce the loop's hotness,
// which would be incorrect in the case of underestimating the trip count.
Info.Weights[Idx] =
Info.Weights[Idx] > SubWeight
? std::max(Info.Weights[Idx] - SubWeight, SubWeight)
: SubWeight;
/// Initialize the weights for all exiting blocks.
static void initBranchWeights(DenseMap<Instruction *, WeightInfo> &WeightInfos,
Loop *L) {
SmallVector<BasicBlock *> ExitingBlocks;
for (BasicBlock *ExitingBlock : ExitingBlocks) {
Instruction *Term = ExitingBlock->getTerminator();
SmallVector<uint32_t> Weights;
if (!extractBranchWeights(*Term, Weights))
// See the comment on updateBranchWeights() for an explanation of what we
// do here.
uint32_t FallThroughWeights = 0;
uint32_t ExitWeights = 0;
for (auto [Succ, Weight] : zip(successors(Term), Weights)) {
if (L->contains(Succ))
FallThroughWeights += Weight;
ExitWeights += Weight;
// Don't try to update weights for degenerate case.
if (FallThroughWeights == 0)
SmallVector<uint32_t> SubWeights;
for (auto [Succ, Weight] : zip(successors(Term), Weights)) {
if (!L->contains(Succ)) {
// Exit weights stay the same.
// Subtract exit weights on each iteration, distributed across all
// fallthrough edges.
double W = (double)Weight / (double)FallThroughWeights;
SubWeights.push_back((uint32_t)(ExitWeights * W));
WeightInfos.insert({Term, {std::move(Weights), std::move(SubWeights)}});
/// Clones the body of the loop L, putting it between \p InsertTop and \p
/// InsertBot.
/// \param IterNumber The serial number of the iteration currently being
/// peeled off.
/// \param ExitEdges The exit edges of the original loop.
/// \param[out] NewBlocks A list of the blocks in the newly created clone
/// \param[out] VMap The value map between the loop and the new clone.
/// \param LoopBlocks A helper for DFS-traversal of the loop.
/// \param LVMap A value-map that maps instructions from the original loop to
/// instructions in the last peeled-off iteration.
static void cloneLoopBlocks(
Loop *L, unsigned IterNumber, BasicBlock *InsertTop, BasicBlock *InsertBot,
SmallVectorImpl<std::pair<BasicBlock *, BasicBlock *>> &ExitEdges,
SmallVectorImpl<BasicBlock *> &NewBlocks, LoopBlocksDFS &LoopBlocks,
ValueToValueMapTy &VMap, ValueToValueMapTy &LVMap, DominatorTree *DT,
LoopInfo *LI, ArrayRef<MDNode *> LoopLocalNoAliasDeclScopes,
ScalarEvolution &SE) {
BasicBlock *Header = L->getHeader();
BasicBlock *Latch = L->getLoopLatch();
BasicBlock *PreHeader = L->getLoopPreheader();
Function *F = Header->getParent();
LoopBlocksDFS::RPOIterator BlockBegin = LoopBlocks.beginRPO();
LoopBlocksDFS::RPOIterator BlockEnd = LoopBlocks.endRPO();
Loop *ParentLoop = L->getParentLoop();
// For each block in the original loop, create a new copy,
// and update the value map with the newly created values.
for (LoopBlocksDFS::RPOIterator BB = BlockBegin; BB != BlockEnd; ++BB) {
BasicBlock *NewBB = CloneBasicBlock(*BB, VMap, ".peel", F);
// If an original block is an immediate child of the loop L, its copy
// is a child of a ParentLoop after peeling. If a block is a child of
// a nested loop, it is handled in the cloneLoop() call below.
if (ParentLoop && LI->getLoopFor(*BB) == L)
ParentLoop->addBasicBlockToLoop(NewBB, *LI);
VMap[*BB] = NewBB;
// If dominator tree is available, insert nodes to represent cloned blocks.
if (DT) {
if (Header == *BB)
DT->addNewBlock(NewBB, InsertTop);
else {
DomTreeNode *IDom = DT->getNode(*BB)->getIDom();
// VMap must contain entry for IDom, as the iteration order is RPO.
DT->addNewBlock(NewBB, cast<BasicBlock>(VMap[IDom->getBlock()]));
// Identify what other metadata depends on the cloned version. After
// cloning, replace the metadata with the corrected version for both
// memory instructions and noalias intrinsics.
std::string Ext = (Twine("Peel") + Twine(IterNumber)).str();
cloneAndAdaptNoAliasScopes(LoopLocalNoAliasDeclScopes, NewBlocks,
Header->getContext(), Ext);
// Recursively create the new Loop objects for nested loops, if any,
// to preserve LoopInfo.
for (Loop *ChildLoop : *L) {
cloneLoop(ChildLoop, ParentLoop, VMap, LI, nullptr);
// Hook-up the control flow for the newly inserted blocks.
// The new header is hooked up directly to the "top", which is either
// the original loop preheader (for the first iteration) or the previous
// iteration's exiting block (for every other iteration)
InsertTop->getTerminator()->setSuccessor(0, cast<BasicBlock>(VMap[Header]));
// Similarly, for the latch:
// The original exiting edge is still hooked up to the loop exit.
// The backedge now goes to the "bottom", which is either the loop's real
// header (for the last peeled iteration) or the copied header of the next
// iteration (for every other iteration)
BasicBlock *NewLatch = cast<BasicBlock>(VMap[Latch]);
auto *LatchTerm = cast<Instruction>(NewLatch->getTerminator());
for (unsigned idx = 0, e = LatchTerm->getNumSuccessors(); idx < e; ++idx)
if (LatchTerm->getSuccessor(idx) == Header) {
LatchTerm->setSuccessor(idx, InsertBot);
if (DT)
DT->changeImmediateDominator(InsertBot, NewLatch);
// The new copy of the loop body starts with a bunch of PHI nodes
// that pick an incoming value from either the preheader, or the previous
// loop iteration. Since this copy is no longer part of the loop, we
// resolve this statically:
// For the first iteration, we use the value from the preheader directly.
// For any other iteration, we replace the phi with the value generated by
// the immediately preceding clone of the loop body (which represents
// the previous iteration).
for (BasicBlock::iterator I = Header->begin(); isa<PHINode>(I); ++I) {
PHINode *NewPHI = cast<PHINode>(VMap[&*I]);
if (IterNumber == 0) {
VMap[&*I] = NewPHI->getIncomingValueForBlock(PreHeader);
} else {
Value *LatchVal = NewPHI->getIncomingValueForBlock(Latch);
Instruction *LatchInst = dyn_cast<Instruction>(LatchVal);
if (LatchInst && L->contains(LatchInst))
VMap[&*I] = LVMap[LatchInst];
VMap[&*I] = LatchVal;
// Fix up the outgoing values - we need to add a value for the iteration
// we've just created. Note that this must happen *after* the incoming
// values are adjusted, since the value going out of the latch may also be
// a value coming into the header.
for (auto Edge : ExitEdges)
for (PHINode &PHI : Edge.second->phis()) {
Value *LatchVal = PHI.getIncomingValueForBlock(Edge.first);
Instruction *LatchInst = dyn_cast<Instruction>(LatchVal);
if (LatchInst && L->contains(LatchInst))
LatchVal = VMap[LatchVal];
PHI.addIncoming(LatchVal, cast<BasicBlock>(VMap[Edge.first]));
// LastValueMap is updated with the values for the current loop
// which are used the next time this function is called.
for (auto KV : VMap)
LVMap[KV.first] = KV.second;
llvm::gatherPeelingPreferences(Loop *L, ScalarEvolution &SE,
const TargetTransformInfo &TTI,
std::optional<bool> UserAllowPeeling,
std::optional<bool> UserAllowProfileBasedPeeling,
bool UnrollingSpecficValues) {
TargetTransformInfo::PeelingPreferences PP;
// Set the default values.
PP.PeelCount = 0;
PP.AllowPeeling = true;
PP.AllowLoopNestsPeeling = false;
PP.PeelProfiledIterations = true;
// Get the target specifc values.
TTI.getPeelingPreferences(L, SE, PP);
// User specified values using cl::opt.
if (UnrollingSpecficValues) {
if (UnrollPeelCount.getNumOccurrences() > 0)
PP.PeelCount = UnrollPeelCount;
if (UnrollAllowPeeling.getNumOccurrences() > 0)
PP.AllowPeeling = UnrollAllowPeeling;
if (UnrollAllowLoopNestsPeeling.getNumOccurrences() > 0)
PP.AllowLoopNestsPeeling = UnrollAllowLoopNestsPeeling;
// User specifed values provided by argument.
if (UserAllowPeeling)
PP.AllowPeeling = *UserAllowPeeling;
if (UserAllowProfileBasedPeeling)
PP.PeelProfiledIterations = *UserAllowProfileBasedPeeling;
return PP;
/// Peel off the first \p PeelCount iterations of loop \p L.
/// Note that this does not peel them off as a single straight-line block.
/// Rather, each iteration is peeled off separately, and needs to check the
/// exit condition.
/// For loops that dynamically execute \p PeelCount iterations or less
/// this provides a benefit, since the peeled off iterations, which account
/// for the bulk of dynamic execution, can be further simplified by scalar
/// optimizations.
bool llvm::peelLoop(Loop *L, unsigned PeelCount, LoopInfo *LI,
ScalarEvolution *SE, DominatorTree &DT, AssumptionCache *AC,
bool PreserveLCSSA, ValueToValueMapTy &LVMap) {
assert(PeelCount > 0 && "Attempt to peel out zero iterations?");
assert(canPeel(L) && "Attempt to peel a loop which is not peelable?");
LoopBlocksDFS LoopBlocks(L);
BasicBlock *Header = L->getHeader();
BasicBlock *PreHeader = L->getLoopPreheader();
BasicBlock *Latch = L->getLoopLatch();
SmallVector<std::pair<BasicBlock *, BasicBlock *>, 4> ExitEdges;
// Remember dominators of blocks we might reach through exits to change them
// later. Immediate dominator of such block might change, because we add more
// routes which can lead to the exit: we can reach it from the peeled
// iterations too.
DenseMap<BasicBlock *, BasicBlock *> NonLoopBlocksIDom;
for (auto *BB : L->blocks()) {
auto *BBDomNode = DT.getNode(BB);
SmallVector<BasicBlock *, 16> ChildrenToUpdate;
for (auto *ChildDomNode : BBDomNode->children()) {
auto *ChildBB = ChildDomNode->getBlock();
if (!L->contains(ChildBB))
// The new idom of the block will be the nearest common dominator
// of all copies of the previous idom. This is equivalent to the
// nearest common dominator of the previous idom and the first latch,
// which dominates all copies of the previous idom.
BasicBlock *NewIDom = DT.findNearestCommonDominator(BB, Latch);
for (auto *ChildBB : ChildrenToUpdate)
NonLoopBlocksIDom[ChildBB] = NewIDom;
Function *F = Header->getParent();
// Set up all the necessary basic blocks. It is convenient to split the
// preheader into 3 parts - two blocks to anchor the peeled copy of the loop
// body, and a new preheader for the "real" loop.
// Peeling the first iteration transforms.
// PreHeader:
// ...
// Header:
// LoopBody
// If (cond) goto Header
// Exit:
// into
// InsertTop:
// LoopBody
// If (!cond) goto Exit
// InsertBot:
// NewPreHeader:
// ...
// Header:
// LoopBody
// If (cond) goto Header
// Exit:
// Each following iteration will split the current bottom anchor in two,
// and put the new copy of the loop body between these two blocks. That is,
// after peeling another iteration from the example above, we'll split
// InsertBot, and get:
// InsertTop:
// LoopBody
// If (!cond) goto Exit
// InsertBot:
// LoopBody
// If (!cond) goto Exit
// NewPreHeader:
// ...
// Header:
// LoopBody
// If (cond) goto Header
// Exit:
BasicBlock *InsertTop = SplitEdge(PreHeader, Header, &DT, LI);
BasicBlock *InsertBot =
SplitBlock(InsertTop, InsertTop->getTerminator(), &DT, LI);
BasicBlock *NewPreHeader =
SplitBlock(InsertBot, InsertBot->getTerminator(), &DT, LI);
InsertTop->setName(Header->getName() + ".peel.begin");
InsertBot->setName(Header->getName() + "");
NewPreHeader->setName(PreHeader->getName() + ".peel.newph");
Instruction *LatchTerm =
// If we have branch weight information, we'll want to update it for the
// newly created branches.
DenseMap<Instruction *, WeightInfo> Weights;
initBranchWeights(Weights, L);
// Identify what noalias metadata is inside the loop: if it is inside the
// loop, the associated metadata must be cloned for each iteration.
SmallVector<MDNode *, 6> LoopLocalNoAliasDeclScopes;
identifyNoAliasScopesToClone(L->getBlocks(), LoopLocalNoAliasDeclScopes);
// For each peeled-off iteration, make a copy of the loop.
for (unsigned Iter = 0; Iter < PeelCount; ++Iter) {
SmallVector<BasicBlock *, 8> NewBlocks;
ValueToValueMapTy VMap;
cloneLoopBlocks(L, Iter, InsertTop, InsertBot, ExitEdges, NewBlocks,
LoopBlocks, VMap, LVMap, &DT, LI,
LoopLocalNoAliasDeclScopes, *SE);
// Remap to use values from the current iteration instead of the
// previous one.
remapInstructionsInBlocks(NewBlocks, VMap);
// Update IDoms of the blocks reachable through exits.
if (Iter == 0)
for (auto BBIDom : NonLoopBlocksIDom)
for (auto &[Term, Info] : Weights) {
auto *TermCopy = cast<Instruction>(VMap[Term]);
updateBranchWeights(TermCopy, Info);
// Remove Loop metadata from the latch branch instruction
// because it is not the Loop's latch branch anymore.
auto *LatchTermCopy = cast<Instruction>(VMap[LatchTerm]);
LatchTermCopy->setMetadata(LLVMContext::MD_loop, nullptr);
InsertTop = InsertBot;
InsertBot = SplitBlock(InsertBot, InsertBot->getTerminator(), &DT, LI);
InsertBot->setName(Header->getName() + "");
F->splice(InsertTop->getIterator(), F, NewBlocks[0]->getIterator(),
// Now adjust the phi nodes in the loop header to get their initial values
// from the last peeled-off iteration instead of the preheader.
for (BasicBlock::iterator I = Header->begin(); isa<PHINode>(I); ++I) {
PHINode *PHI = cast<PHINode>(I);
Value *NewVal = PHI->getIncomingValueForBlock(Latch);
Instruction *LatchInst = dyn_cast<Instruction>(NewVal);
if (LatchInst && L->contains(LatchInst))
NewVal = LVMap[LatchInst];
PHI->setIncomingValueForBlock(NewPreHeader, NewVal);
for (const auto &[Term, Info] : Weights) {
setBranchWeights(*Term, Info.Weights);
// Update Metadata for count of peeled off iterations.
unsigned AlreadyPeeled = 0;
if (auto Peeled = getOptionalIntLoopAttribute(L, PeeledCountMetaData))
AlreadyPeeled = *Peeled;
addStringMetadataToLoop(L, PeeledCountMetaData, AlreadyPeeled + PeelCount);
if (Loop *ParentLoop = L->getParentLoop())
L = ParentLoop;
// We modified the loop, update SE.
// Finally DomtTree must be correct.
// FIXME: Incrementally update loop-simplify
simplifyLoop(L, &DT, LI, SE, AC, nullptr, PreserveLCSSA);
return true;