//===- InductiveRangeCheckElimination.cpp - -------------------------------===// // // 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 // //===----------------------------------------------------------------------===// // // The InductiveRangeCheckElimination pass splits a loop's iteration space into // three disjoint ranges. It does that in a way such that the loop running in // the middle loop provably does not need range checks. As an example, it will // convert // // len = < known positive > // for (i = 0; i < n; i++) { // if (0 <= i && i < len) { // do_something(); // } else { // throw_out_of_bounds(); // } // } // // to // // len = < known positive > // limit = smin(n, len) // // no first segment // for (i = 0; i < limit; i++) { // if (0 <= i && i < len) { // this check is fully redundant // do_something(); // } else { // throw_out_of_bounds(); // } // } // for (i = limit; i < n; i++) { // if (0 <= i && i < len) { // do_something(); // } else { // throw_out_of_bounds(); // } // } // //===----------------------------------------------------------------------===// #include "llvm/Transforms/Scalar/InductiveRangeCheckElimination.h" #include "llvm/ADT/APInt.h" #include "llvm/ADT/ArrayRef.h" #include "llvm/ADT/PriorityWorklist.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/StringRef.h" #include "llvm/ADT/Twine.h" #include "llvm/Analysis/BlockFrequencyInfo.h" #include "llvm/Analysis/BranchProbabilityInfo.h" #include "llvm/Analysis/LoopAnalysisManager.h" #include "llvm/Analysis/LoopInfo.h" #include "llvm/Analysis/ScalarEvolution.h" #include "llvm/Analysis/ScalarEvolutionExpressions.h" #include "llvm/IR/BasicBlock.h" #include "llvm/IR/CFG.h" #include "llvm/IR/Constants.h" #include "llvm/IR/DerivedTypes.h" #include "llvm/IR/Dominators.h" #include "llvm/IR/Function.h" #include "llvm/IR/IRBuilder.h" #include "llvm/IR/InstrTypes.h" #include "llvm/IR/Instructions.h" #include "llvm/IR/Metadata.h" #include "llvm/IR/Module.h" #include "llvm/IR/PatternMatch.h" #include "llvm/IR/Type.h" #include "llvm/IR/Use.h" #include "llvm/IR/User.h" #include "llvm/IR/Value.h" #include "llvm/Support/BranchProbability.h" #include "llvm/Support/Casting.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Compiler.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/Cloning.h" #include "llvm/Transforms/Utils/LoopConstrainer.h" #include "llvm/Transforms/Utils/LoopSimplify.h" #include "llvm/Transforms/Utils/LoopUtils.h" #include "llvm/Transforms/Utils/ScalarEvolutionExpander.h" #include "llvm/Transforms/Utils/ValueMapper.h" #include #include #include #include #include using namespace llvm; using namespace llvm::PatternMatch; static cl::opt LoopSizeCutoff("irce-loop-size-cutoff", cl::Hidden, cl::init(64)); static cl::opt PrintChangedLoops("irce-print-changed-loops", cl::Hidden, cl::init(false)); static cl::opt PrintRangeChecks("irce-print-range-checks", cl::Hidden, cl::init(false)); static cl::opt SkipProfitabilityChecks("irce-skip-profitability-checks", cl::Hidden, cl::init(false)); static cl::opt MinRuntimeIterations("irce-min-runtime-iterations", cl::Hidden, cl::init(10)); static cl::opt AllowUnsignedLatchCondition("irce-allow-unsigned-latch", cl::Hidden, cl::init(true)); static cl::opt AllowNarrowLatchCondition( "irce-allow-narrow-latch", cl::Hidden, cl::init(true), cl::desc("If set to true, IRCE may eliminate wide range checks in loops " "with narrow latch condition.")); static cl::opt MaxTypeSizeForOverflowCheck( "irce-max-type-size-for-overflow-check", cl::Hidden, cl::init(32), cl::desc( "Maximum size of range check type for which can be produced runtime " "overflow check of its limit's computation")); static cl::opt PrintScaledBoundaryRangeChecks("irce-print-scaled-boundary-range-checks", cl::Hidden, cl::init(false)); #define DEBUG_TYPE "irce" namespace { /// An inductive range check is conditional branch in a loop with /// /// 1. a very cold successor (i.e. the branch jumps to that successor very /// rarely) /// /// and /// /// 2. a condition that is provably true for some contiguous range of values /// taken by the containing loop's induction variable. /// class InductiveRangeCheck { const SCEV *Begin = nullptr; const SCEV *Step = nullptr; const SCEV *End = nullptr; Use *CheckUse = nullptr; static bool parseRangeCheckICmp(Loop *L, ICmpInst *ICI, ScalarEvolution &SE, const SCEVAddRecExpr *&Index, const SCEV *&End); static void extractRangeChecksFromCond(Loop *L, ScalarEvolution &SE, Use &ConditionUse, SmallVectorImpl &Checks, SmallPtrSetImpl &Visited); static bool parseIvAgaisntLimit(Loop *L, Value *LHS, Value *RHS, ICmpInst::Predicate Pred, ScalarEvolution &SE, const SCEVAddRecExpr *&Index, const SCEV *&End); static bool reassociateSubLHS(Loop *L, Value *VariantLHS, Value *InvariantRHS, ICmpInst::Predicate Pred, ScalarEvolution &SE, const SCEVAddRecExpr *&Index, const SCEV *&End); public: const SCEV *getBegin() const { return Begin; } const SCEV *getStep() const { return Step; } const SCEV *getEnd() const { return End; } void print(raw_ostream &OS) const { OS << "InductiveRangeCheck:\n"; OS << " Begin: "; Begin->print(OS); OS << " Step: "; Step->print(OS); OS << " End: "; End->print(OS); OS << "\n CheckUse: "; getCheckUse()->getUser()->print(OS); OS << " Operand: " << getCheckUse()->getOperandNo() << "\n"; } LLVM_DUMP_METHOD void dump() { print(dbgs()); } Use *getCheckUse() const { return CheckUse; } /// Represents an signed integer range [Range.getBegin(), Range.getEnd()). If /// R.getEnd() le R.getBegin(), then R denotes the empty range. class Range { const SCEV *Begin; const SCEV *End; public: Range(const SCEV *Begin, const SCEV *End) : Begin(Begin), End(End) { assert(Begin->getType() == End->getType() && "ill-typed range!"); } Type *getType() const { return Begin->getType(); } const SCEV *getBegin() const { return Begin; } const SCEV *getEnd() const { return End; } bool isEmpty(ScalarEvolution &SE, bool IsSigned) const { if (Begin == End) return true; if (IsSigned) return SE.isKnownPredicate(ICmpInst::ICMP_SGE, Begin, End); else return SE.isKnownPredicate(ICmpInst::ICMP_UGE, Begin, End); } }; /// This is the value the condition of the branch needs to evaluate to for the /// branch to take the hot successor (see (1) above). bool getPassingDirection() { return true; } /// Computes a range for the induction variable (IndVar) in which the range /// check is redundant and can be constant-folded away. The induction /// variable is not required to be the canonical {0,+,1} induction variable. std::optional computeSafeIterationSpace(ScalarEvolution &SE, const SCEVAddRecExpr *IndVar, bool IsLatchSigned) const; /// Parse out a set of inductive range checks from \p BI and append them to \p /// Checks. /// /// NB! There may be conditions feeding into \p BI that aren't inductive range /// checks, and hence don't end up in \p Checks. static void extractRangeChecksFromBranch( BranchInst *BI, Loop *L, ScalarEvolution &SE, BranchProbabilityInfo *BPI, SmallVectorImpl &Checks, bool &Changed); }; class InductiveRangeCheckElimination { ScalarEvolution &SE; BranchProbabilityInfo *BPI; DominatorTree &DT; LoopInfo &LI; using GetBFIFunc = std::optional>; GetBFIFunc GetBFI; // Returns true if it is profitable to do a transform basing on estimation of // number of iterations. bool isProfitableToTransform(const Loop &L, LoopStructure &LS); public: InductiveRangeCheckElimination(ScalarEvolution &SE, BranchProbabilityInfo *BPI, DominatorTree &DT, LoopInfo &LI, GetBFIFunc GetBFI = std::nullopt) : SE(SE), BPI(BPI), DT(DT), LI(LI), GetBFI(GetBFI) {} bool run(Loop *L, function_ref LPMAddNewLoop); }; } // end anonymous namespace /// Parse a single ICmp instruction, `ICI`, into a range check. If `ICI` cannot /// be interpreted as a range check, return false. Otherwise set `Index` to the /// SCEV being range checked, and set `End` to the upper or lower limit `Index` /// is being range checked. bool InductiveRangeCheck::parseRangeCheckICmp(Loop *L, ICmpInst *ICI, ScalarEvolution &SE, const SCEVAddRecExpr *&Index, const SCEV *&End) { auto IsLoopInvariant = [&SE, L](Value *V) { return SE.isLoopInvariant(SE.getSCEV(V), L); }; ICmpInst::Predicate Pred = ICI->getPredicate(); Value *LHS = ICI->getOperand(0); Value *RHS = ICI->getOperand(1); if (!LHS->getType()->isIntegerTy()) return false; // Canonicalize to the `Index Pred Invariant` comparison if (IsLoopInvariant(LHS)) { std::swap(LHS, RHS); Pred = CmpInst::getSwappedPredicate(Pred); } else if (!IsLoopInvariant(RHS)) // Both LHS and RHS are loop variant return false; if (parseIvAgaisntLimit(L, LHS, RHS, Pred, SE, Index, End)) return true; if (reassociateSubLHS(L, LHS, RHS, Pred, SE, Index, End)) return true; // TODO: support ReassociateAddLHS return false; } // Try to parse range check in the form of "IV vs Limit" bool InductiveRangeCheck::parseIvAgaisntLimit(Loop *L, Value *LHS, Value *RHS, ICmpInst::Predicate Pred, ScalarEvolution &SE, const SCEVAddRecExpr *&Index, const SCEV *&End) { auto SIntMaxSCEV = [&](Type *T) { unsigned BitWidth = cast(T)->getBitWidth(); return SE.getConstant(APInt::getSignedMaxValue(BitWidth)); }; const auto *AddRec = dyn_cast(SE.getSCEV(LHS)); if (!AddRec) return false; // We strengthen "0 <= I" to "0 <= I < INT_SMAX" and "I < L" to "0 <= I < L". // We can potentially do much better here. // If we want to adjust upper bound for the unsigned range check as we do it // for signed one, we will need to pick Unsigned max switch (Pred) { default: return false; case ICmpInst::ICMP_SGE: if (match(RHS, m_ConstantInt<0>())) { Index = AddRec; End = SIntMaxSCEV(Index->getType()); return true; } return false; case ICmpInst::ICMP_SGT: if (match(RHS, m_ConstantInt<-1>())) { Index = AddRec; End = SIntMaxSCEV(Index->getType()); return true; } return false; case ICmpInst::ICMP_SLT: case ICmpInst::ICMP_ULT: Index = AddRec; End = SE.getSCEV(RHS); return true; case ICmpInst::ICMP_SLE: case ICmpInst::ICMP_ULE: const SCEV *One = SE.getOne(RHS->getType()); const SCEV *RHSS = SE.getSCEV(RHS); bool Signed = Pred == ICmpInst::ICMP_SLE; if (SE.willNotOverflow(Instruction::BinaryOps::Add, Signed, RHSS, One)) { Index = AddRec; End = SE.getAddExpr(RHSS, One); return true; } return false; } llvm_unreachable("default clause returns!"); } // Try to parse range check in the form of "IV - Offset vs Limit" or "Offset - // IV vs Limit" bool InductiveRangeCheck::reassociateSubLHS( Loop *L, Value *VariantLHS, Value *InvariantRHS, ICmpInst::Predicate Pred, ScalarEvolution &SE, const SCEVAddRecExpr *&Index, const SCEV *&End) { Value *LHS, *RHS; if (!match(VariantLHS, m_Sub(m_Value(LHS), m_Value(RHS)))) return false; const SCEV *IV = SE.getSCEV(LHS); const SCEV *Offset = SE.getSCEV(RHS); const SCEV *Limit = SE.getSCEV(InvariantRHS); bool OffsetSubtracted = false; if (SE.isLoopInvariant(IV, L)) // "Offset - IV vs Limit" std::swap(IV, Offset); else if (SE.isLoopInvariant(Offset, L)) // "IV - Offset vs Limit" OffsetSubtracted = true; else return false; const auto *AddRec = dyn_cast(IV); if (!AddRec) return false; // In order to turn "IV - Offset < Limit" into "IV < Limit + Offset", we need // to be able to freely move values from left side of inequality to right side // (just as in normal linear arithmetics). Overflows make things much more // complicated, so we want to avoid this. // // Let's prove that the initial subtraction doesn't overflow with all IV's // values from the safe range constructed for that check. // // [Case 1] IV - Offset < Limit // It doesn't overflow if: // SINT_MIN <= IV - Offset <= SINT_MAX // In terms of scaled SINT we need to prove: // SINT_MIN + Offset <= IV <= SINT_MAX + Offset // Safe range will be constructed: // 0 <= IV < Limit + Offset // It means that 'IV - Offset' doesn't underflow, because: // SINT_MIN + Offset < 0 <= IV // and doesn't overflow: // IV < Limit + Offset <= SINT_MAX + Offset // // [Case 2] Offset - IV > Limit // It doesn't overflow if: // SINT_MIN <= Offset - IV <= SINT_MAX // In terms of scaled SINT we need to prove: // -SINT_MIN >= IV - Offset >= -SINT_MAX // Offset - SINT_MIN >= IV >= Offset - SINT_MAX // Safe range will be constructed: // 0 <= IV < Offset - Limit // It means that 'Offset - IV' doesn't underflow, because // Offset - SINT_MAX < 0 <= IV // and doesn't overflow: // IV < Offset - Limit <= Offset - SINT_MIN // // For the computed upper boundary of the IV's range (Offset +/- Limit) we // don't know exactly whether it overflows or not. So if we can't prove this // fact at compile time, we scale boundary computations to a wider type with // the intention to add runtime overflow check. auto getExprScaledIfOverflow = [&](Instruction::BinaryOps BinOp, const SCEV *LHS, const SCEV *RHS) -> const SCEV * { const SCEV *(ScalarEvolution::*Operation)(const SCEV *, const SCEV *, SCEV::NoWrapFlags, unsigned); switch (BinOp) { default: llvm_unreachable("Unsupported binary op"); case Instruction::Add: Operation = &ScalarEvolution::getAddExpr; break; case Instruction::Sub: Operation = &ScalarEvolution::getMinusSCEV; break; } if (SE.willNotOverflow(BinOp, ICmpInst::isSigned(Pred), LHS, RHS, cast(VariantLHS))) return (SE.*Operation)(LHS, RHS, SCEV::FlagAnyWrap, 0); // We couldn't prove that the expression does not overflow. // Than scale it to a wider type to check overflow at runtime. auto *Ty = cast(LHS->getType()); if (Ty->getBitWidth() > MaxTypeSizeForOverflowCheck) return nullptr; auto WideTy = IntegerType::get(Ty->getContext(), Ty->getBitWidth() * 2); return (SE.*Operation)(SE.getSignExtendExpr(LHS, WideTy), SE.getSignExtendExpr(RHS, WideTy), SCEV::FlagAnyWrap, 0); }; if (OffsetSubtracted) // "IV - Offset < Limit" -> "IV" < Offset + Limit Limit = getExprScaledIfOverflow(Instruction::BinaryOps::Add, Offset, Limit); else { // "Offset - IV > Limit" -> "IV" < Offset - Limit Limit = getExprScaledIfOverflow(Instruction::BinaryOps::Sub, Offset, Limit); Pred = ICmpInst::getSwappedPredicate(Pred); } if (Pred == ICmpInst::ICMP_SLT || Pred == ICmpInst::ICMP_SLE) { // "Expr <= Limit" -> "Expr < Limit + 1" if (Pred == ICmpInst::ICMP_SLE && Limit) Limit = getExprScaledIfOverflow(Instruction::BinaryOps::Add, Limit, SE.getOne(Limit->getType())); if (Limit) { Index = AddRec; End = Limit; return true; } } return false; } void InductiveRangeCheck::extractRangeChecksFromCond( Loop *L, ScalarEvolution &SE, Use &ConditionUse, SmallVectorImpl &Checks, SmallPtrSetImpl &Visited) { Value *Condition = ConditionUse.get(); if (!Visited.insert(Condition).second) return; // TODO: Do the same for OR, XOR, NOT etc? if (match(Condition, m_LogicalAnd(m_Value(), m_Value()))) { extractRangeChecksFromCond(L, SE, cast(Condition)->getOperandUse(0), Checks, Visited); extractRangeChecksFromCond(L, SE, cast(Condition)->getOperandUse(1), Checks, Visited); return; } ICmpInst *ICI = dyn_cast(Condition); if (!ICI) return; const SCEV *End = nullptr; const SCEVAddRecExpr *IndexAddRec = nullptr; if (!parseRangeCheckICmp(L, ICI, SE, IndexAddRec, End)) return; assert(IndexAddRec && "IndexAddRec was not computed"); assert(End && "End was not computed"); if ((IndexAddRec->getLoop() != L) || !IndexAddRec->isAffine()) return; InductiveRangeCheck IRC; IRC.End = End; IRC.Begin = IndexAddRec->getStart(); IRC.Step = IndexAddRec->getStepRecurrence(SE); IRC.CheckUse = &ConditionUse; Checks.push_back(IRC); } void InductiveRangeCheck::extractRangeChecksFromBranch( BranchInst *BI, Loop *L, ScalarEvolution &SE, BranchProbabilityInfo *BPI, SmallVectorImpl &Checks, bool &Changed) { if (BI->isUnconditional() || BI->getParent() == L->getLoopLatch()) return; unsigned IndexLoopSucc = L->contains(BI->getSuccessor(0)) ? 0 : 1; assert(L->contains(BI->getSuccessor(IndexLoopSucc)) && "No edges coming to loop?"); BranchProbability LikelyTaken(15, 16); if (!SkipProfitabilityChecks && BPI && BPI->getEdgeProbability(BI->getParent(), IndexLoopSucc) < LikelyTaken) return; // IRCE expects branch's true edge comes to loop. Invert branch for opposite // case. if (IndexLoopSucc != 0) { IRBuilder<> Builder(BI); InvertBranch(BI, Builder); if (BPI) BPI->swapSuccEdgesProbabilities(BI->getParent()); Changed = true; } SmallPtrSet Visited; InductiveRangeCheck::extractRangeChecksFromCond(L, SE, BI->getOperandUse(0), Checks, Visited); } /// If the type of \p S matches with \p Ty, return \p S. Otherwise, return /// signed or unsigned extension of \p S to type \p Ty. static const SCEV *NoopOrExtend(const SCEV *S, Type *Ty, ScalarEvolution &SE, bool Signed) { return Signed ? SE.getNoopOrSignExtend(S, Ty) : SE.getNoopOrZeroExtend(S, Ty); } // Compute a safe set of limits for the main loop to run in -- effectively the // intersection of `Range' and the iteration space of the original loop. // Return std::nullopt if unable to compute the set of subranges. static std::optional calculateSubRanges(ScalarEvolution &SE, const Loop &L, InductiveRangeCheck::Range &Range, const LoopStructure &MainLoopStructure) { auto *RTy = cast(Range.getType()); // We only support wide range checks and narrow latches. if (!AllowNarrowLatchCondition && RTy != MainLoopStructure.ExitCountTy) return std::nullopt; if (RTy->getBitWidth() < MainLoopStructure.ExitCountTy->getBitWidth()) return std::nullopt; LoopConstrainer::SubRanges Result; bool IsSignedPredicate = MainLoopStructure.IsSignedPredicate; // I think we can be more aggressive here and make this nuw / nsw if the // addition that feeds into the icmp for the latch's terminating branch is nuw // / nsw. In any case, a wrapping 2's complement addition is safe. const SCEV *Start = NoopOrExtend(SE.getSCEV(MainLoopStructure.IndVarStart), RTy, SE, IsSignedPredicate); const SCEV *End = NoopOrExtend(SE.getSCEV(MainLoopStructure.LoopExitAt), RTy, SE, IsSignedPredicate); bool Increasing = MainLoopStructure.IndVarIncreasing; // We compute `Smallest` and `Greatest` such that [Smallest, Greatest), or // [Smallest, GreatestSeen] is the range of values the induction variable // takes. const SCEV *Smallest = nullptr, *Greatest = nullptr, *GreatestSeen = nullptr; const SCEV *One = SE.getOne(RTy); if (Increasing) { Smallest = Start; Greatest = End; // No overflow, because the range [Smallest, GreatestSeen] is not empty. GreatestSeen = SE.getMinusSCEV(End, One); } else { // These two computations may sign-overflow. Here is why that is okay: // // We know that the induction variable does not sign-overflow on any // iteration except the last one, and it starts at `Start` and ends at // `End`, decrementing by one every time. // // * if `Smallest` sign-overflows we know `End` is `INT_SMAX`. Since the // induction variable is decreasing we know that the smallest value // the loop body is actually executed with is `INT_SMIN` == `Smallest`. // // * if `Greatest` sign-overflows, we know it can only be `INT_SMIN`. In // that case, `Clamp` will always return `Smallest` and // [`Result.LowLimit`, `Result.HighLimit`) = [`Smallest`, `Smallest`) // will be an empty range. Returning an empty range is always safe. Smallest = SE.getAddExpr(End, One); Greatest = SE.getAddExpr(Start, One); GreatestSeen = Start; } auto Clamp = [&SE, Smallest, Greatest, IsSignedPredicate](const SCEV *S) { return IsSignedPredicate ? SE.getSMaxExpr(Smallest, SE.getSMinExpr(Greatest, S)) : SE.getUMaxExpr(Smallest, SE.getUMinExpr(Greatest, S)); }; // In some cases we can prove that we don't need a pre or post loop. ICmpInst::Predicate PredLE = IsSignedPredicate ? ICmpInst::ICMP_SLE : ICmpInst::ICMP_ULE; ICmpInst::Predicate PredLT = IsSignedPredicate ? ICmpInst::ICMP_SLT : ICmpInst::ICMP_ULT; bool ProvablyNoPreloop = SE.isKnownPredicate(PredLE, Range.getBegin(), Smallest); if (!ProvablyNoPreloop) Result.LowLimit = Clamp(Range.getBegin()); bool ProvablyNoPostLoop = SE.isKnownPredicate(PredLT, GreatestSeen, Range.getEnd()); if (!ProvablyNoPostLoop) Result.HighLimit = Clamp(Range.getEnd()); return Result; } /// Computes and returns a range of values for the induction variable (IndVar) /// in which the range check can be safely elided. If it cannot compute such a /// range, returns std::nullopt. std::optional InductiveRangeCheck::computeSafeIterationSpace(ScalarEvolution &SE, const SCEVAddRecExpr *IndVar, bool IsLatchSigned) const { // We can deal when types of latch check and range checks don't match in case // if latch check is more narrow. auto *IVType = dyn_cast(IndVar->getType()); auto *RCType = dyn_cast(getBegin()->getType()); auto *EndType = dyn_cast(getEnd()->getType()); // Do not work with pointer types. if (!IVType || !RCType) return std::nullopt; if (IVType->getBitWidth() > RCType->getBitWidth()) return std::nullopt; // IndVar is of the form "A + B * I" (where "I" is the canonical induction // variable, that may or may not exist as a real llvm::Value in the loop) and // this inductive range check is a range check on the "C + D * I" ("C" is // getBegin() and "D" is getStep()). We rewrite the value being range // checked to "M + N * IndVar" where "N" = "D * B^(-1)" and "M" = "C - NA". // // The actual inequalities we solve are of the form // // 0 <= M + 1 * IndVar < L given L >= 0 (i.e. N == 1) // // Here L stands for upper limit of the safe iteration space. // The inequality is satisfied by (0 - M) <= IndVar < (L - M). To avoid // overflows when calculating (0 - M) and (L - M) we, depending on type of // IV's iteration space, limit the calculations by borders of the iteration // space. For example, if IndVar is unsigned, (0 - M) overflows for any M > 0. // If we figured out that "anything greater than (-M) is safe", we strengthen // this to "everything greater than 0 is safe", assuming that values between // -M and 0 just do not exist in unsigned iteration space, and we don't want // to deal with overflown values. if (!IndVar->isAffine()) return std::nullopt; const SCEV *A = NoopOrExtend(IndVar->getStart(), RCType, SE, IsLatchSigned); const SCEVConstant *B = dyn_cast( NoopOrExtend(IndVar->getStepRecurrence(SE), RCType, SE, IsLatchSigned)); if (!B) return std::nullopt; assert(!B->isZero() && "Recurrence with zero step?"); const SCEV *C = getBegin(); const SCEVConstant *D = dyn_cast(getStep()); if (D != B) return std::nullopt; assert(!D->getValue()->isZero() && "Recurrence with zero step?"); unsigned BitWidth = RCType->getBitWidth(); const SCEV *SIntMax = SE.getConstant(APInt::getSignedMaxValue(BitWidth)); const SCEV *SIntMin = SE.getConstant(APInt::getSignedMinValue(BitWidth)); // Subtract Y from X so that it does not go through border of the IV // iteration space. Mathematically, it is equivalent to: // // ClampedSubtract(X, Y) = min(max(X - Y, INT_MIN), INT_MAX). [1] // // In [1], 'X - Y' is a mathematical subtraction (result is not bounded to // any width of bit grid). But after we take min/max, the result is // guaranteed to be within [INT_MIN, INT_MAX]. // // In [1], INT_MAX and INT_MIN are respectively signed and unsigned max/min // values, depending on type of latch condition that defines IV iteration // space. auto ClampedSubtract = [&](const SCEV *X, const SCEV *Y) { // FIXME: The current implementation assumes that X is in [0, SINT_MAX]. // This is required to ensure that SINT_MAX - X does not overflow signed and // that X - Y does not overflow unsigned if Y is negative. Can we lift this // restriction and make it work for negative X either? if (IsLatchSigned) { // X is a number from signed range, Y is interpreted as signed. // Even if Y is SINT_MAX, (X - Y) does not reach SINT_MIN. So the only // thing we should care about is that we didn't cross SINT_MAX. // So, if Y is positive, we subtract Y safely. // Rule 1: Y > 0 ---> Y. // If 0 <= -Y <= (SINT_MAX - X), we subtract Y safely. // Rule 2: Y >=s (X - SINT_MAX) ---> Y. // If 0 <= (SINT_MAX - X) < -Y, we can only subtract (X - SINT_MAX). // Rule 3: Y (X - SINT_MAX). // It gives us smax(Y, X - SINT_MAX) to subtract in all cases. const SCEV *XMinusSIntMax = SE.getMinusSCEV(X, SIntMax); return SE.getMinusSCEV(X, SE.getSMaxExpr(Y, XMinusSIntMax), SCEV::FlagNSW); } else // X is a number from unsigned range, Y is interpreted as signed. // Even if Y is SINT_MIN, (X - Y) does not reach UINT_MAX. So the only // thing we should care about is that we didn't cross zero. // So, if Y is negative, we subtract Y safely. // Rule 1: Y Y. // If 0 <= Y <= X, we subtract Y safely. // Rule 2: Y <=s X ---> Y. // If 0 <= X < Y, we should stop at 0 and can only subtract X. // Rule 3: Y >s X ---> X. // It gives us smin(X, Y) to subtract in all cases. return SE.getMinusSCEV(X, SE.getSMinExpr(X, Y), SCEV::FlagNUW); }; const SCEV *M = SE.getMinusSCEV(C, A); const SCEV *Zero = SE.getZero(M->getType()); // This function returns SCEV equal to 1 if X is non-negative 0 otherwise. auto SCEVCheckNonNegative = [&](const SCEV *X) { const Loop *L = IndVar->getLoop(); const SCEV *Zero = SE.getZero(X->getType()); const SCEV *One = SE.getOne(X->getType()); // Can we trivially prove that X is a non-negative or negative value? if (isKnownNonNegativeInLoop(X, L, SE)) return One; else if (isKnownNegativeInLoop(X, L, SE)) return Zero; // If not, we will have to figure it out during the execution. // Function smax(smin(X, 0), -1) + 1 equals to 1 if X >= 0 and 0 if X < 0. const SCEV *NegOne = SE.getNegativeSCEV(One); return SE.getAddExpr(SE.getSMaxExpr(SE.getSMinExpr(X, Zero), NegOne), One); }; // This function returns SCEV equal to 1 if X will not overflow in terms of // range check type, 0 otherwise. auto SCEVCheckWillNotOverflow = [&](const SCEV *X) { // X doesn't overflow if SINT_MAX >= X. // Then if (SINT_MAX - X) >= 0, X doesn't overflow const SCEV *SIntMaxExt = SE.getSignExtendExpr(SIntMax, X->getType()); const SCEV *OverflowCheck = SCEVCheckNonNegative(SE.getMinusSCEV(SIntMaxExt, X)); // X doesn't underflow if X >= SINT_MIN. // Then if (X - SINT_MIN) >= 0, X doesn't underflow const SCEV *SIntMinExt = SE.getSignExtendExpr(SIntMin, X->getType()); const SCEV *UnderflowCheck = SCEVCheckNonNegative(SE.getMinusSCEV(X, SIntMinExt)); return SE.getMulExpr(OverflowCheck, UnderflowCheck); }; // FIXME: Current implementation of ClampedSubtract implicitly assumes that // X is non-negative (in sense of a signed value). We need to re-implement // this function in a way that it will correctly handle negative X as well. // We use it twice: for X = 0 everything is fine, but for X = getEnd() we can // end up with a negative X and produce wrong results. So currently we ensure // that if getEnd() is negative then both ends of the safe range are zero. // Note that this may pessimize elimination of unsigned range checks against // negative values. const SCEV *REnd = getEnd(); const SCEV *EndWillNotOverflow = SE.getOne(RCType); auto PrintRangeCheck = [&](raw_ostream &OS) { auto L = IndVar->getLoop(); OS << "irce: in function "; OS << L->getHeader()->getParent()->getName(); OS << ", in "; L->print(OS); OS << "there is range check with scaled boundary:\n"; print(OS); }; if (EndType->getBitWidth() > RCType->getBitWidth()) { assert(EndType->getBitWidth() == RCType->getBitWidth() * 2); if (PrintScaledBoundaryRangeChecks) PrintRangeCheck(errs()); // End is computed with extended type but will be truncated to a narrow one // type of range check. Therefore we need a check that the result will not // overflow in terms of narrow type. EndWillNotOverflow = SE.getTruncateExpr(SCEVCheckWillNotOverflow(REnd), RCType); REnd = SE.getTruncateExpr(REnd, RCType); } const SCEV *RuntimeChecks = SE.getMulExpr(SCEVCheckNonNegative(REnd), EndWillNotOverflow); const SCEV *Begin = SE.getMulExpr(ClampedSubtract(Zero, M), RuntimeChecks); const SCEV *End = SE.getMulExpr(ClampedSubtract(REnd, M), RuntimeChecks); return InductiveRangeCheck::Range(Begin, End); } static std::optional IntersectSignedRange(ScalarEvolution &SE, const std::optional &R1, const InductiveRangeCheck::Range &R2) { if (R2.isEmpty(SE, /* IsSigned */ true)) return std::nullopt; if (!R1) return R2; auto &R1Value = *R1; // We never return empty ranges from this function, and R1 is supposed to be // a result of intersection. Thus, R1 is never empty. assert(!R1Value.isEmpty(SE, /* IsSigned */ true) && "We should never have empty R1!"); // TODO: we could widen the smaller range and have this work; but for now we // bail out to keep things simple. if (R1Value.getType() != R2.getType()) return std::nullopt; const SCEV *NewBegin = SE.getSMaxExpr(R1Value.getBegin(), R2.getBegin()); const SCEV *NewEnd = SE.getSMinExpr(R1Value.getEnd(), R2.getEnd()); // If the resulting range is empty, just return std::nullopt. auto Ret = InductiveRangeCheck::Range(NewBegin, NewEnd); if (Ret.isEmpty(SE, /* IsSigned */ true)) return std::nullopt; return Ret; } static std::optional IntersectUnsignedRange(ScalarEvolution &SE, const std::optional &R1, const InductiveRangeCheck::Range &R2) { if (R2.isEmpty(SE, /* IsSigned */ false)) return std::nullopt; if (!R1) return R2; auto &R1Value = *R1; // We never return empty ranges from this function, and R1 is supposed to be // a result of intersection. Thus, R1 is never empty. assert(!R1Value.isEmpty(SE, /* IsSigned */ false) && "We should never have empty R1!"); // TODO: we could widen the smaller range and have this work; but for now we // bail out to keep things simple. if (R1Value.getType() != R2.getType()) return std::nullopt; const SCEV *NewBegin = SE.getUMaxExpr(R1Value.getBegin(), R2.getBegin()); const SCEV *NewEnd = SE.getUMinExpr(R1Value.getEnd(), R2.getEnd()); // If the resulting range is empty, just return std::nullopt. auto Ret = InductiveRangeCheck::Range(NewBegin, NewEnd); if (Ret.isEmpty(SE, /* IsSigned */ false)) return std::nullopt; return Ret; } PreservedAnalyses IRCEPass::run(Function &F, FunctionAnalysisManager &AM) { auto &DT = AM.getResult(F); LoopInfo &LI = AM.getResult(F); // There are no loops in the function. Return before computing other expensive // analyses. if (LI.empty()) return PreservedAnalyses::all(); auto &SE = AM.getResult(F); auto &BPI = AM.getResult(F); // Get BFI analysis result on demand. Please note that modification of // CFG invalidates this analysis and we should handle it. auto getBFI = [&F, &AM ]()->BlockFrequencyInfo & { return AM.getResult(F); }; InductiveRangeCheckElimination IRCE(SE, &BPI, DT, LI, { getBFI }); bool Changed = false; { bool CFGChanged = false; for (const auto &L : LI) { CFGChanged |= simplifyLoop(L, &DT, &LI, &SE, nullptr, nullptr, /*PreserveLCSSA=*/false); Changed |= formLCSSARecursively(*L, DT, &LI, &SE); } Changed |= CFGChanged; if (CFGChanged && !SkipProfitabilityChecks) { PreservedAnalyses PA = PreservedAnalyses::all(); PA.abandon(); AM.invalidate(F, PA); } } SmallPriorityWorklist Worklist; appendLoopsToWorklist(LI, Worklist); auto LPMAddNewLoop = [&Worklist](Loop *NL, bool IsSubloop) { if (!IsSubloop) appendLoopsToWorklist(*NL, Worklist); }; while (!Worklist.empty()) { Loop *L = Worklist.pop_back_val(); if (IRCE.run(L, LPMAddNewLoop)) { Changed = true; if (!SkipProfitabilityChecks) { PreservedAnalyses PA = PreservedAnalyses::all(); PA.abandon(); AM.invalidate(F, PA); } } } if (!Changed) return PreservedAnalyses::all(); return getLoopPassPreservedAnalyses(); } bool InductiveRangeCheckElimination::isProfitableToTransform(const Loop &L, LoopStructure &LS) { if (SkipProfitabilityChecks) return true; if (GetBFI) { BlockFrequencyInfo &BFI = (*GetBFI)(); uint64_t hFreq = BFI.getBlockFreq(LS.Header).getFrequency(); uint64_t phFreq = BFI.getBlockFreq(L.getLoopPreheader()).getFrequency(); if (phFreq != 0 && hFreq != 0 && (hFreq / phFreq < MinRuntimeIterations)) { LLVM_DEBUG(dbgs() << "irce: could not prove profitability: " << "the estimated number of iterations basing on " "frequency info is " << (hFreq / phFreq) << "\n";); return false; } return true; } if (!BPI) return true; BranchProbability ExitProbability = BPI->getEdgeProbability(LS.Latch, LS.LatchBrExitIdx); if (ExitProbability > BranchProbability(1, MinRuntimeIterations)) { LLVM_DEBUG(dbgs() << "irce: could not prove profitability: " << "the exit probability is too big " << ExitProbability << "\n";); return false; } return true; } bool InductiveRangeCheckElimination::run( Loop *L, function_ref LPMAddNewLoop) { if (L->getBlocks().size() >= LoopSizeCutoff) { LLVM_DEBUG(dbgs() << "irce: giving up constraining loop, too large\n"); return false; } BasicBlock *Preheader = L->getLoopPreheader(); if (!Preheader) { LLVM_DEBUG(dbgs() << "irce: loop has no preheader, leaving\n"); return false; } LLVMContext &Context = Preheader->getContext(); SmallVector RangeChecks; bool Changed = false; for (auto *BBI : L->getBlocks()) if (BranchInst *TBI = dyn_cast(BBI->getTerminator())) InductiveRangeCheck::extractRangeChecksFromBranch(TBI, L, SE, BPI, RangeChecks, Changed); if (RangeChecks.empty()) return Changed; auto PrintRecognizedRangeChecks = [&](raw_ostream &OS) { OS << "irce: looking at loop "; L->print(OS); OS << "irce: loop has " << RangeChecks.size() << " inductive range checks: \n"; for (InductiveRangeCheck &IRC : RangeChecks) IRC.print(OS); }; LLVM_DEBUG(PrintRecognizedRangeChecks(dbgs())); if (PrintRangeChecks) PrintRecognizedRangeChecks(errs()); const char *FailureReason = nullptr; std::optional MaybeLoopStructure = LoopStructure::parseLoopStructure(SE, *L, AllowUnsignedLatchCondition, FailureReason); if (!MaybeLoopStructure) { LLVM_DEBUG(dbgs() << "irce: could not parse loop structure: " << FailureReason << "\n";); return Changed; } LoopStructure LS = *MaybeLoopStructure; if (!isProfitableToTransform(*L, LS)) return Changed; const SCEVAddRecExpr *IndVar = cast(SE.getMinusSCEV(SE.getSCEV(LS.IndVarBase), SE.getSCEV(LS.IndVarStep))); std::optional SafeIterRange; SmallVector RangeChecksToEliminate; // Basing on the type of latch predicate, we interpret the IV iteration range // as signed or unsigned range. We use different min/max functions (signed or // unsigned) when intersecting this range with safe iteration ranges implied // by range checks. auto IntersectRange = LS.IsSignedPredicate ? IntersectSignedRange : IntersectUnsignedRange; for (InductiveRangeCheck &IRC : RangeChecks) { auto Result = IRC.computeSafeIterationSpace(SE, IndVar, LS.IsSignedPredicate); if (Result) { auto MaybeSafeIterRange = IntersectRange(SE, SafeIterRange, *Result); if (MaybeSafeIterRange) { assert(!MaybeSafeIterRange->isEmpty(SE, LS.IsSignedPredicate) && "We should never return empty ranges!"); RangeChecksToEliminate.push_back(IRC); SafeIterRange = *MaybeSafeIterRange; } } } if (!SafeIterRange) return Changed; std::optional MaybeSR = calculateSubRanges(SE, *L, *SafeIterRange, LS); if (!MaybeSR) { LLVM_DEBUG(dbgs() << "irce: could not compute subranges\n"); return false; } LoopConstrainer LC(*L, LI, LPMAddNewLoop, LS, SE, DT, SafeIterRange->getBegin()->getType(), *MaybeSR); if (LC.run()) { Changed = true; auto PrintConstrainedLoopInfo = [L]() { dbgs() << "irce: in function "; dbgs() << L->getHeader()->getParent()->getName() << ": "; dbgs() << "constrained "; L->print(dbgs()); }; LLVM_DEBUG(PrintConstrainedLoopInfo()); if (PrintChangedLoops) PrintConstrainedLoopInfo(); // Optimize away the now-redundant range checks. for (InductiveRangeCheck &IRC : RangeChecksToEliminate) { ConstantInt *FoldedRangeCheck = IRC.getPassingDirection() ? ConstantInt::getTrue(Context) : ConstantInt::getFalse(Context); IRC.getCheckUse()->set(FoldedRangeCheck); } } return Changed; }