//===- SimplifyCFG.cpp - Code to perform CFG simplification ---------------===// // // 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 // //===----------------------------------------------------------------------===// // // Peephole optimize the CFG. // //===----------------------------------------------------------------------===// #include "llvm/ADT/APInt.h" #include "llvm/ADT/ArrayRef.h" #include "llvm/ADT/DenseMap.h" #include "llvm/ADT/MapVector.h" #include "llvm/ADT/STLExtras.h" #include "llvm/ADT/Sequence.h" #include "llvm/ADT/SetOperations.h" #include "llvm/ADT/SetVector.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/ADT/Statistic.h" #include "llvm/ADT/StringRef.h" #include "llvm/Analysis/AssumptionCache.h" #include "llvm/Analysis/CaptureTracking.h" #include "llvm/Analysis/ConstantFolding.h" #include "llvm/Analysis/DomTreeUpdater.h" #include "llvm/Analysis/GuardUtils.h" #include "llvm/Analysis/InstructionSimplify.h" #include "llvm/Analysis/MemorySSA.h" #include "llvm/Analysis/MemorySSAUpdater.h" #include "llvm/Analysis/TargetTransformInfo.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/IR/Attributes.h" #include "llvm/IR/BasicBlock.h" #include "llvm/IR/CFG.h" #include "llvm/IR/Constant.h" #include "llvm/IR/ConstantRange.h" #include "llvm/IR/Constants.h" #include "llvm/IR/DataLayout.h" #include "llvm/IR/DebugInfo.h" #include "llvm/IR/DerivedTypes.h" #include "llvm/IR/Function.h" #include "llvm/IR/GlobalValue.h" #include "llvm/IR/GlobalVariable.h" #include "llvm/IR/IRBuilder.h" #include "llvm/IR/InstrTypes.h" #include "llvm/IR/Instruction.h" #include "llvm/IR/Instructions.h" #include "llvm/IR/IntrinsicInst.h" #include "llvm/IR/LLVMContext.h" #include "llvm/IR/MDBuilder.h" #include "llvm/IR/MemoryModelRelaxationAnnotations.h" #include "llvm/IR/Metadata.h" #include "llvm/IR/Module.h" #include "llvm/IR/NoFolder.h" #include "llvm/IR/Operator.h" #include "llvm/IR/PatternMatch.h" #include "llvm/IR/ProfDataUtils.h" #include "llvm/IR/Type.h" #include "llvm/IR/Use.h" #include "llvm/IR/User.h" #include "llvm/IR/Value.h" #include "llvm/IR/ValueHandle.h" #include "llvm/Support/BranchProbability.h" #include "llvm/Support/Casting.h" #include "llvm/Support/CommandLine.h" #include "llvm/Support/Debug.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/KnownBits.h" #include "llvm/Support/MathExtras.h" #include "llvm/Support/raw_ostream.h" #include "llvm/Transforms/Utils/BasicBlockUtils.h" #include "llvm/Transforms/Utils/Local.h" #include "llvm/Transforms/Utils/ValueMapper.h" #include #include #include #include #include #include #include #include #include #include #include #include using namespace llvm; using namespace PatternMatch; #define DEBUG_TYPE "simplifycfg" cl::opt llvm::RequireAndPreserveDomTree( "simplifycfg-require-and-preserve-domtree", cl::Hidden, cl::desc("Temorary development switch used to gradually uplift SimplifyCFG " "into preserving DomTree,")); // Chosen as 2 so as to be cheap, but still to have enough power to fold // a select, so the "clamp" idiom (of a min followed by a max) will be caught. // To catch this, we need to fold a compare and a select, hence '2' being the // minimum reasonable default. static cl::opt PHINodeFoldingThreshold( "phi-node-folding-threshold", cl::Hidden, cl::init(2), cl::desc( "Control the amount of phi node folding to perform (default = 2)")); static cl::opt TwoEntryPHINodeFoldingThreshold( "two-entry-phi-node-folding-threshold", cl::Hidden, cl::init(4), cl::desc("Control the maximal total instruction cost that we are willing " "to speculatively execute to fold a 2-entry PHI node into a " "select (default = 4)")); static cl::opt HoistCommon("simplifycfg-hoist-common", cl::Hidden, cl::init(true), cl::desc("Hoist common instructions up to the parent block")); static cl::opt HoistCommonSkipLimit("simplifycfg-hoist-common-skip-limit", cl::Hidden, cl::init(20), cl::desc("Allow reordering across at most this many " "instructions when hoisting")); static cl::opt SinkCommon("simplifycfg-sink-common", cl::Hidden, cl::init(true), cl::desc("Sink common instructions down to the end block")); static cl::opt HoistCondStores( "simplifycfg-hoist-cond-stores", cl::Hidden, cl::init(true), cl::desc("Hoist conditional stores if an unconditional store precedes")); static cl::opt MergeCondStores( "simplifycfg-merge-cond-stores", cl::Hidden, cl::init(true), cl::desc("Hoist conditional stores even if an unconditional store does not " "precede - hoist multiple conditional stores into a single " "predicated store")); static cl::opt MergeCondStoresAggressively( "simplifycfg-merge-cond-stores-aggressively", cl::Hidden, cl::init(false), cl::desc("When merging conditional stores, do so even if the resultant " "basic blocks are unlikely to be if-converted as a result")); static cl::opt SpeculateOneExpensiveInst( "speculate-one-expensive-inst", cl::Hidden, cl::init(true), cl::desc("Allow exactly one expensive instruction to be speculatively " "executed")); static cl::opt MaxSpeculationDepth( "max-speculation-depth", cl::Hidden, cl::init(10), cl::desc("Limit maximum recursion depth when calculating costs of " "speculatively executed instructions")); static cl::opt MaxSmallBlockSize("simplifycfg-max-small-block-size", cl::Hidden, cl::init(10), cl::desc("Max size of a block which is still considered " "small enough to thread through")); // Two is chosen to allow one negation and a logical combine. static cl::opt BranchFoldThreshold("simplifycfg-branch-fold-threshold", cl::Hidden, cl::init(2), cl::desc("Maximum cost of combining conditions when " "folding branches")); static cl::opt BranchFoldToCommonDestVectorMultiplier( "simplifycfg-branch-fold-common-dest-vector-multiplier", cl::Hidden, cl::init(2), cl::desc("Multiplier to apply to threshold when determining whether or not " "to fold branch to common destination when vector operations are " "present")); static cl::opt EnableMergeCompatibleInvokes( "simplifycfg-merge-compatible-invokes", cl::Hidden, cl::init(true), cl::desc("Allow SimplifyCFG to merge invokes together when appropriate")); static cl::opt MaxSwitchCasesPerResult( "max-switch-cases-per-result", cl::Hidden, cl::init(16), cl::desc("Limit cases to analyze when converting a switch to select")); STATISTIC(NumBitMaps, "Number of switch instructions turned into bitmaps"); STATISTIC(NumLinearMaps, "Number of switch instructions turned into linear mapping"); STATISTIC(NumLookupTables, "Number of switch instructions turned into lookup tables"); STATISTIC( NumLookupTablesHoles, "Number of switch instructions turned into lookup tables (holes checked)"); STATISTIC(NumTableCmpReuses, "Number of reused switch table lookup compares"); STATISTIC(NumFoldValueComparisonIntoPredecessors, "Number of value comparisons folded into predecessor basic blocks"); STATISTIC(NumFoldBranchToCommonDest, "Number of branches folded into predecessor basic block"); STATISTIC( NumHoistCommonCode, "Number of common instruction 'blocks' hoisted up to the begin block"); STATISTIC(NumHoistCommonInstrs, "Number of common instructions hoisted up to the begin block"); STATISTIC(NumSinkCommonCode, "Number of common instruction 'blocks' sunk down to the end block"); STATISTIC(NumSinkCommonInstrs, "Number of common instructions sunk down to the end block"); STATISTIC(NumSpeculations, "Number of speculative executed instructions"); STATISTIC(NumInvokes, "Number of invokes with empty resume blocks simplified into calls"); STATISTIC(NumInvokesMerged, "Number of invokes that were merged together"); STATISTIC(NumInvokeSetsFormed, "Number of invoke sets that were formed"); namespace { // The first field contains the value that the switch produces when a certain // case group is selected, and the second field is a vector containing the // cases composing the case group. using SwitchCaseResultVectorTy = SmallVector>, 2>; // The first field contains the phi node that generates a result of the switch // and the second field contains the value generated for a certain case in the // switch for that PHI. using SwitchCaseResultsTy = SmallVector, 4>; /// ValueEqualityComparisonCase - Represents a case of a switch. struct ValueEqualityComparisonCase { ConstantInt *Value; BasicBlock *Dest; ValueEqualityComparisonCase(ConstantInt *Value, BasicBlock *Dest) : Value(Value), Dest(Dest) {} bool operator<(ValueEqualityComparisonCase RHS) const { // Comparing pointers is ok as we only rely on the order for uniquing. return Value < RHS.Value; } bool operator==(BasicBlock *RHSDest) const { return Dest == RHSDest; } }; class SimplifyCFGOpt { const TargetTransformInfo &TTI; DomTreeUpdater *DTU; const DataLayout &DL; ArrayRef LoopHeaders; const SimplifyCFGOptions &Options; bool Resimplify; Value *isValueEqualityComparison(Instruction *TI); BasicBlock *GetValueEqualityComparisonCases( Instruction *TI, std::vector &Cases); bool SimplifyEqualityComparisonWithOnlyPredecessor(Instruction *TI, BasicBlock *Pred, IRBuilder<> &Builder); bool PerformValueComparisonIntoPredecessorFolding(Instruction *TI, Value *&CV, Instruction *PTI, IRBuilder<> &Builder); bool FoldValueComparisonIntoPredecessors(Instruction *TI, IRBuilder<> &Builder); bool simplifyResume(ResumeInst *RI, IRBuilder<> &Builder); bool simplifySingleResume(ResumeInst *RI); bool simplifyCommonResume(ResumeInst *RI); bool simplifyCleanupReturn(CleanupReturnInst *RI); bool simplifyUnreachable(UnreachableInst *UI); bool simplifySwitch(SwitchInst *SI, IRBuilder<> &Builder); bool simplifyIndirectBr(IndirectBrInst *IBI); bool simplifyBranch(BranchInst *Branch, IRBuilder<> &Builder); bool simplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder); bool simplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder); bool tryToSimplifyUncondBranchWithICmpInIt(ICmpInst *ICI, IRBuilder<> &Builder); bool hoistCommonCodeFromSuccessors(BasicBlock *BB, bool EqTermsOnly); bool hoistSuccIdenticalTerminatorToSwitchOrIf( Instruction *TI, Instruction *I1, SmallVectorImpl &OtherSuccTIs); bool SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB); bool SimplifyTerminatorOnSelect(Instruction *OldTerm, Value *Cond, BasicBlock *TrueBB, BasicBlock *FalseBB, uint32_t TrueWeight, uint32_t FalseWeight); bool SimplifyBranchOnICmpChain(BranchInst *BI, IRBuilder<> &Builder, const DataLayout &DL); bool SimplifySwitchOnSelect(SwitchInst *SI, SelectInst *Select); bool SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, SelectInst *SI); bool TurnSwitchRangeIntoICmp(SwitchInst *SI, IRBuilder<> &Builder); public: SimplifyCFGOpt(const TargetTransformInfo &TTI, DomTreeUpdater *DTU, const DataLayout &DL, ArrayRef LoopHeaders, const SimplifyCFGOptions &Opts) : TTI(TTI), DTU(DTU), DL(DL), LoopHeaders(LoopHeaders), Options(Opts) { assert((!DTU || !DTU->hasPostDomTree()) && "SimplifyCFG is not yet capable of maintaining validity of a " "PostDomTree, so don't ask for it."); } bool simplifyOnce(BasicBlock *BB); bool run(BasicBlock *BB); // Helper to set Resimplify and return change indication. bool requestResimplify() { Resimplify = true; return true; } }; } // end anonymous namespace /// Return true if all the PHI nodes in the basic block \p BB /// receive compatible (identical) incoming values when coming from /// all of the predecessor blocks that are specified in \p IncomingBlocks. /// /// Note that if the values aren't exactly identical, but \p EquivalenceSet /// is provided, and *both* of the values are present in the set, /// then they are considered equal. static bool IncomingValuesAreCompatible( BasicBlock *BB, ArrayRef IncomingBlocks, SmallPtrSetImpl *EquivalenceSet = nullptr) { assert(IncomingBlocks.size() == 2 && "Only for a pair of incoming blocks at the time!"); // FIXME: it is okay if one of the incoming values is an `undef` value, // iff the other incoming value is guaranteed to be a non-poison value. // FIXME: it is okay if one of the incoming values is a `poison` value. return all_of(BB->phis(), [IncomingBlocks, EquivalenceSet](PHINode &PN) { Value *IV0 = PN.getIncomingValueForBlock(IncomingBlocks[0]); Value *IV1 = PN.getIncomingValueForBlock(IncomingBlocks[1]); if (IV0 == IV1) return true; if (EquivalenceSet && EquivalenceSet->contains(IV0) && EquivalenceSet->contains(IV1)) return true; return false; }); } /// Return true if it is safe to merge these two /// terminator instructions together. static bool SafeToMergeTerminators(Instruction *SI1, Instruction *SI2, SmallSetVector *FailBlocks = nullptr) { if (SI1 == SI2) return false; // Can't merge with self! // It is not safe to merge these two switch instructions if they have a common // successor, and if that successor has a PHI node, and if *that* PHI node has // conflicting incoming values from the two switch blocks. BasicBlock *SI1BB = SI1->getParent(); BasicBlock *SI2BB = SI2->getParent(); SmallPtrSet SI1Succs(succ_begin(SI1BB), succ_end(SI1BB)); bool Fail = false; for (BasicBlock *Succ : successors(SI2BB)) { if (!SI1Succs.count(Succ)) continue; if (IncomingValuesAreCompatible(Succ, {SI1BB, SI2BB})) continue; Fail = true; if (FailBlocks) FailBlocks->insert(Succ); else break; } return !Fail; } /// Update PHI nodes in Succ to indicate that there will now be entries in it /// from the 'NewPred' block. The values that will be flowing into the PHI nodes /// will be the same as those coming in from ExistPred, an existing predecessor /// of Succ. static void AddPredecessorToBlock(BasicBlock *Succ, BasicBlock *NewPred, BasicBlock *ExistPred, MemorySSAUpdater *MSSAU = nullptr) { for (PHINode &PN : Succ->phis()) PN.addIncoming(PN.getIncomingValueForBlock(ExistPred), NewPred); if (MSSAU) if (auto *MPhi = MSSAU->getMemorySSA()->getMemoryAccess(Succ)) MPhi->addIncoming(MPhi->getIncomingValueForBlock(ExistPred), NewPred); } /// Compute an abstract "cost" of speculating the given instruction, /// which is assumed to be safe to speculate. TCC_Free means cheap, /// TCC_Basic means less cheap, and TCC_Expensive means prohibitively /// expensive. static InstructionCost computeSpeculationCost(const User *I, const TargetTransformInfo &TTI) { assert((!isa(I) || isSafeToSpeculativelyExecute(cast(I))) && "Instruction is not safe to speculatively execute!"); return TTI.getInstructionCost(I, TargetTransformInfo::TCK_SizeAndLatency); } /// If we have a merge point of an "if condition" as accepted above, /// return true if the specified value dominates the block. We /// don't handle the true generality of domination here, just a special case /// which works well enough for us. /// /// If AggressiveInsts is non-null, and if V does not dominate BB, we check to /// see if V (which must be an instruction) and its recursive operands /// that do not dominate BB have a combined cost lower than Budget and /// are non-trapping. If both are true, the instruction is inserted into the /// set and true is returned. /// /// The cost for most non-trapping instructions is defined as 1 except for /// Select whose cost is 2. /// /// After this function returns, Cost is increased by the cost of /// V plus its non-dominating operands. If that cost is greater than /// Budget, false is returned and Cost is undefined. static bool dominatesMergePoint(Value *V, BasicBlock *BB, SmallPtrSetImpl &AggressiveInsts, InstructionCost &Cost, InstructionCost Budget, const TargetTransformInfo &TTI, unsigned Depth = 0) { // It is possible to hit a zero-cost cycle (phi/gep instructions for example), // so limit the recursion depth. // TODO: While this recursion limit does prevent pathological behavior, it // would be better to track visited instructions to avoid cycles. if (Depth == MaxSpeculationDepth) return false; Instruction *I = dyn_cast(V); if (!I) { // Non-instructions dominate all instructions and can be executed // unconditionally. return true; } BasicBlock *PBB = I->getParent(); // We don't want to allow weird loops that might have the "if condition" in // the bottom of this block. if (PBB == BB) return false; // If this instruction is defined in a block that contains an unconditional // branch to BB, then it must be in the 'conditional' part of the "if // statement". If not, it definitely dominates the region. BranchInst *BI = dyn_cast(PBB->getTerminator()); if (!BI || BI->isConditional() || BI->getSuccessor(0) != BB) return true; // If we have seen this instruction before, don't count it again. if (AggressiveInsts.count(I)) return true; // Okay, it looks like the instruction IS in the "condition". Check to // see if it's a cheap instruction to unconditionally compute, and if it // only uses stuff defined outside of the condition. If so, hoist it out. if (!isSafeToSpeculativelyExecute(I)) return false; Cost += computeSpeculationCost(I, TTI); // Allow exactly one instruction to be speculated regardless of its cost // (as long as it is safe to do so). // This is intended to flatten the CFG even if the instruction is a division // or other expensive operation. The speculation of an expensive instruction // is expected to be undone in CodeGenPrepare if the speculation has not // enabled further IR optimizations. if (Cost > Budget && (!SpeculateOneExpensiveInst || !AggressiveInsts.empty() || Depth > 0 || !Cost.isValid())) return false; // Okay, we can only really hoist these out if their operands do // not take us over the cost threshold. for (Use &Op : I->operands()) if (!dominatesMergePoint(Op, BB, AggressiveInsts, Cost, Budget, TTI, Depth + 1)) return false; // Okay, it's safe to do this! Remember this instruction. AggressiveInsts.insert(I); return true; } /// Extract ConstantInt from value, looking through IntToPtr /// and PointerNullValue. Return NULL if value is not a constant int. static ConstantInt *GetConstantInt(Value *V, const DataLayout &DL) { // Normal constant int. ConstantInt *CI = dyn_cast(V); if (CI || !isa(V) || !V->getType()->isPointerTy() || DL.isNonIntegralPointerType(V->getType())) return CI; // This is some kind of pointer constant. Turn it into a pointer-sized // ConstantInt if possible. IntegerType *PtrTy = cast(DL.getIntPtrType(V->getType())); // Null pointer means 0, see SelectionDAGBuilder::getValue(const Value*). if (isa(V)) return ConstantInt::get(PtrTy, 0); // IntToPtr const int. if (ConstantExpr *CE = dyn_cast(V)) if (CE->getOpcode() == Instruction::IntToPtr) if (ConstantInt *CI = dyn_cast(CE->getOperand(0))) { // The constant is very likely to have the right type already. if (CI->getType() == PtrTy) return CI; else return cast( ConstantFoldIntegerCast(CI, PtrTy, /*isSigned=*/false, DL)); } return nullptr; } namespace { /// Given a chain of or (||) or and (&&) comparison of a value against a /// constant, this will try to recover the information required for a switch /// structure. /// It will depth-first traverse the chain of comparison, seeking for patterns /// like %a == 12 or %a < 4 and combine them to produce a set of integer /// representing the different cases for the switch. /// Note that if the chain is composed of '||' it will build the set of elements /// that matches the comparisons (i.e. any of this value validate the chain) /// while for a chain of '&&' it will build the set elements that make the test /// fail. struct ConstantComparesGatherer { const DataLayout &DL; /// Value found for the switch comparison Value *CompValue = nullptr; /// Extra clause to be checked before the switch Value *Extra = nullptr; /// Set of integers to match in switch SmallVector Vals; /// Number of comparisons matched in the and/or chain unsigned UsedICmps = 0; /// Construct and compute the result for the comparison instruction Cond ConstantComparesGatherer(Instruction *Cond, const DataLayout &DL) : DL(DL) { gather(Cond); } ConstantComparesGatherer(const ConstantComparesGatherer &) = delete; ConstantComparesGatherer & operator=(const ConstantComparesGatherer &) = delete; private: /// Try to set the current value used for the comparison, it succeeds only if /// it wasn't set before or if the new value is the same as the old one bool setValueOnce(Value *NewVal) { if (CompValue && CompValue != NewVal) return false; CompValue = NewVal; return (CompValue != nullptr); } /// Try to match Instruction "I" as a comparison against a constant and /// populates the array Vals with the set of values that match (or do not /// match depending on isEQ). /// Return false on failure. On success, the Value the comparison matched /// against is placed in CompValue. /// If CompValue is already set, the function is expected to fail if a match /// is found but the value compared to is different. bool matchInstruction(Instruction *I, bool isEQ) { // If this is an icmp against a constant, handle this as one of the cases. ICmpInst *ICI; ConstantInt *C; if (!((ICI = dyn_cast(I)) && (C = GetConstantInt(I->getOperand(1), DL)))) { return false; } Value *RHSVal; const APInt *RHSC; // Pattern match a special case // (x & ~2^z) == y --> x == y || x == y|2^z // This undoes a transformation done by instcombine to fuse 2 compares. if (ICI->getPredicate() == (isEQ ? ICmpInst::ICMP_EQ : ICmpInst::ICMP_NE)) { // It's a little bit hard to see why the following transformations are // correct. Here is a CVC3 program to verify them for 64-bit values: /* ONE : BITVECTOR(64) = BVZEROEXTEND(0bin1, 63); x : BITVECTOR(64); y : BITVECTOR(64); z : BITVECTOR(64); mask : BITVECTOR(64) = BVSHL(ONE, z); QUERY( (y & ~mask = y) => ((x & ~mask = y) <=> (x = y OR x = (y | mask))) ); QUERY( (y | mask = y) => ((x | mask = y) <=> (x = y OR x = (y & ~mask))) ); */ // Please note that each pattern must be a dual implication (<--> or // iff). One directional implication can create spurious matches. If the // implication is only one-way, an unsatisfiable condition on the left // side can imply a satisfiable condition on the right side. Dual // implication ensures that satisfiable conditions are transformed to // other satisfiable conditions and unsatisfiable conditions are // transformed to other unsatisfiable conditions. // Here is a concrete example of a unsatisfiable condition on the left // implying a satisfiable condition on the right: // // mask = (1 << z) // (x & ~mask) == y --> (x == y || x == (y | mask)) // // Substituting y = 3, z = 0 yields: // (x & -2) == 3 --> (x == 3 || x == 2) // Pattern match a special case: /* QUERY( (y & ~mask = y) => ((x & ~mask = y) <=> (x = y OR x = (y | mask))) ); */ if (match(ICI->getOperand(0), m_And(m_Value(RHSVal), m_APInt(RHSC)))) { APInt Mask = ~*RHSC; if (Mask.isPowerOf2() && (C->getValue() & ~Mask) == C->getValue()) { // If we already have a value for the switch, it has to match! if (!setValueOnce(RHSVal)) return false; Vals.push_back(C); Vals.push_back( ConstantInt::get(C->getContext(), C->getValue() | Mask)); UsedICmps++; return true; } } // Pattern match a special case: /* QUERY( (y | mask = y) => ((x | mask = y) <=> (x = y OR x = (y & ~mask))) ); */ if (match(ICI->getOperand(0), m_Or(m_Value(RHSVal), m_APInt(RHSC)))) { APInt Mask = *RHSC; if (Mask.isPowerOf2() && (C->getValue() | Mask) == C->getValue()) { // If we already have a value for the switch, it has to match! if (!setValueOnce(RHSVal)) return false; Vals.push_back(C); Vals.push_back(ConstantInt::get(C->getContext(), C->getValue() & ~Mask)); UsedICmps++; return true; } } // If we already have a value for the switch, it has to match! if (!setValueOnce(ICI->getOperand(0))) return false; UsedICmps++; Vals.push_back(C); return ICI->getOperand(0); } // If we have "x ult 3", for example, then we can add 0,1,2 to the set. ConstantRange Span = ConstantRange::makeExactICmpRegion(ICI->getPredicate(), C->getValue()); // Shift the range if the compare is fed by an add. This is the range // compare idiom as emitted by instcombine. Value *CandidateVal = I->getOperand(0); if (match(I->getOperand(0), m_Add(m_Value(RHSVal), m_APInt(RHSC)))) { Span = Span.subtract(*RHSC); CandidateVal = RHSVal; } // If this is an and/!= check, then we are looking to build the set of // value that *don't* pass the and chain. I.e. to turn "x ugt 2" into // x != 0 && x != 1. if (!isEQ) Span = Span.inverse(); // If there are a ton of values, we don't want to make a ginormous switch. if (Span.isSizeLargerThan(8) || Span.isEmptySet()) { return false; } // If we already have a value for the switch, it has to match! if (!setValueOnce(CandidateVal)) return false; // Add all values from the range to the set for (APInt Tmp = Span.getLower(); Tmp != Span.getUpper(); ++Tmp) Vals.push_back(ConstantInt::get(I->getContext(), Tmp)); UsedICmps++; return true; } /// Given a potentially 'or'd or 'and'd together collection of icmp /// eq/ne/lt/gt instructions that compare a value against a constant, extract /// the value being compared, and stick the list constants into the Vals /// vector. /// One "Extra" case is allowed to differ from the other. void gather(Value *V) { bool isEQ = match(V, m_LogicalOr(m_Value(), m_Value())); // Keep a stack (SmallVector for efficiency) for depth-first traversal SmallVector DFT; SmallPtrSet Visited; // Initialize Visited.insert(V); DFT.push_back(V); while (!DFT.empty()) { V = DFT.pop_back_val(); if (Instruction *I = dyn_cast(V)) { // If it is a || (or && depending on isEQ), process the operands. Value *Op0, *Op1; if (isEQ ? match(I, m_LogicalOr(m_Value(Op0), m_Value(Op1))) : match(I, m_LogicalAnd(m_Value(Op0), m_Value(Op1)))) { if (Visited.insert(Op1).second) DFT.push_back(Op1); if (Visited.insert(Op0).second) DFT.push_back(Op0); continue; } // Try to match the current instruction if (matchInstruction(I, isEQ)) // Match succeed, continue the loop continue; } // One element of the sequence of || (or &&) could not be match as a // comparison against the same value as the others. // We allow only one "Extra" case to be checked before the switch if (!Extra) { Extra = V; continue; } // Failed to parse a proper sequence, abort now CompValue = nullptr; break; } } }; } // end anonymous namespace static void EraseTerminatorAndDCECond(Instruction *TI, MemorySSAUpdater *MSSAU = nullptr) { Instruction *Cond = nullptr; if (SwitchInst *SI = dyn_cast(TI)) { Cond = dyn_cast(SI->getCondition()); } else if (BranchInst *BI = dyn_cast(TI)) { if (BI->isConditional()) Cond = dyn_cast(BI->getCondition()); } else if (IndirectBrInst *IBI = dyn_cast(TI)) { Cond = dyn_cast(IBI->getAddress()); } TI->eraseFromParent(); if (Cond) RecursivelyDeleteTriviallyDeadInstructions(Cond, nullptr, MSSAU); } /// Return true if the specified terminator checks /// to see if a value is equal to constant integer value. Value *SimplifyCFGOpt::isValueEqualityComparison(Instruction *TI) { Value *CV = nullptr; if (SwitchInst *SI = dyn_cast(TI)) { // Do not permit merging of large switch instructions into their // predecessors unless there is only one predecessor. if (!SI->getParent()->hasNPredecessorsOrMore(128 / SI->getNumSuccessors())) CV = SI->getCondition(); } else if (BranchInst *BI = dyn_cast(TI)) if (BI->isConditional() && BI->getCondition()->hasOneUse()) if (ICmpInst *ICI = dyn_cast(BI->getCondition())) { if (ICI->isEquality() && GetConstantInt(ICI->getOperand(1), DL)) CV = ICI->getOperand(0); } // Unwrap any lossless ptrtoint cast. if (CV) { if (PtrToIntInst *PTII = dyn_cast(CV)) { Value *Ptr = PTII->getPointerOperand(); if (PTII->getType() == DL.getIntPtrType(Ptr->getType())) CV = Ptr; } } return CV; } /// Given a value comparison instruction, /// decode all of the 'cases' that it represents and return the 'default' block. BasicBlock *SimplifyCFGOpt::GetValueEqualityComparisonCases( Instruction *TI, std::vector &Cases) { if (SwitchInst *SI = dyn_cast(TI)) { Cases.reserve(SI->getNumCases()); for (auto Case : SI->cases()) Cases.push_back(ValueEqualityComparisonCase(Case.getCaseValue(), Case.getCaseSuccessor())); return SI->getDefaultDest(); } BranchInst *BI = cast(TI); ICmpInst *ICI = cast(BI->getCondition()); BasicBlock *Succ = BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_NE); Cases.push_back(ValueEqualityComparisonCase( GetConstantInt(ICI->getOperand(1), DL), Succ)); return BI->getSuccessor(ICI->getPredicate() == ICmpInst::ICMP_EQ); } /// Given a vector of bb/value pairs, remove any entries /// in the list that match the specified block. static void EliminateBlockCases(BasicBlock *BB, std::vector &Cases) { llvm::erase(Cases, BB); } /// Return true if there are any keys in C1 that exist in C2 as well. static bool ValuesOverlap(std::vector &C1, std::vector &C2) { std::vector *V1 = &C1, *V2 = &C2; // Make V1 be smaller than V2. if (V1->size() > V2->size()) std::swap(V1, V2); if (V1->empty()) return false; if (V1->size() == 1) { // Just scan V2. ConstantInt *TheVal = (*V1)[0].Value; for (const ValueEqualityComparisonCase &VECC : *V2) if (TheVal == VECC.Value) return true; } // Otherwise, just sort both lists and compare element by element. array_pod_sort(V1->begin(), V1->end()); array_pod_sort(V2->begin(), V2->end()); unsigned i1 = 0, i2 = 0, e1 = V1->size(), e2 = V2->size(); while (i1 != e1 && i2 != e2) { if ((*V1)[i1].Value == (*V2)[i2].Value) return true; if ((*V1)[i1].Value < (*V2)[i2].Value) ++i1; else ++i2; } return false; } // Set branch weights on SwitchInst. This sets the metadata if there is at // least one non-zero weight. static void setBranchWeights(SwitchInst *SI, ArrayRef Weights, bool IsExpected) { // Check that there is at least one non-zero weight. Otherwise, pass // nullptr to setMetadata which will erase the existing metadata. MDNode *N = nullptr; if (llvm::any_of(Weights, [](uint32_t W) { return W != 0; })) N = MDBuilder(SI->getParent()->getContext()) .createBranchWeights(Weights, IsExpected); SI->setMetadata(LLVMContext::MD_prof, N); } // Similar to the above, but for branch and select instructions that take // exactly 2 weights. static void setBranchWeights(Instruction *I, uint32_t TrueWeight, uint32_t FalseWeight, bool IsExpected) { assert(isa(I) || isa(I)); // Check that there is at least one non-zero weight. Otherwise, pass // nullptr to setMetadata which will erase the existing metadata. MDNode *N = nullptr; if (TrueWeight || FalseWeight) N = MDBuilder(I->getParent()->getContext()) .createBranchWeights(TrueWeight, FalseWeight, IsExpected); I->setMetadata(LLVMContext::MD_prof, N); } /// If TI is known to be a terminator instruction and its block is known to /// only have a single predecessor block, check to see if that predecessor is /// also a value comparison with the same value, and if that comparison /// determines the outcome of this comparison. If so, simplify TI. This does a /// very limited form of jump threading. bool SimplifyCFGOpt::SimplifyEqualityComparisonWithOnlyPredecessor( Instruction *TI, BasicBlock *Pred, IRBuilder<> &Builder) { Value *PredVal = isValueEqualityComparison(Pred->getTerminator()); if (!PredVal) return false; // Not a value comparison in predecessor. Value *ThisVal = isValueEqualityComparison(TI); assert(ThisVal && "This isn't a value comparison!!"); if (ThisVal != PredVal) return false; // Different predicates. // TODO: Preserve branch weight metadata, similarly to how // FoldValueComparisonIntoPredecessors preserves it. // Find out information about when control will move from Pred to TI's block. std::vector PredCases; BasicBlock *PredDef = GetValueEqualityComparisonCases(Pred->getTerminator(), PredCases); EliminateBlockCases(PredDef, PredCases); // Remove default from cases. // Find information about how control leaves this block. std::vector ThisCases; BasicBlock *ThisDef = GetValueEqualityComparisonCases(TI, ThisCases); EliminateBlockCases(ThisDef, ThisCases); // Remove default from cases. // If TI's block is the default block from Pred's comparison, potentially // simplify TI based on this knowledge. if (PredDef == TI->getParent()) { // If we are here, we know that the value is none of those cases listed in // PredCases. If there are any cases in ThisCases that are in PredCases, we // can simplify TI. if (!ValuesOverlap(PredCases, ThisCases)) return false; if (isa(TI)) { // Okay, one of the successors of this condbr is dead. Convert it to a // uncond br. assert(ThisCases.size() == 1 && "Branch can only have one case!"); // Insert the new branch. Instruction *NI = Builder.CreateBr(ThisDef); (void)NI; // Remove PHI node entries for the dead edge. ThisCases[0].Dest->removePredecessor(PredDef); LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator() << "Through successor TI: " << *TI << "Leaving: " << *NI << "\n"); EraseTerminatorAndDCECond(TI); if (DTU) DTU->applyUpdates( {{DominatorTree::Delete, PredDef, ThisCases[0].Dest}}); return true; } SwitchInstProfUpdateWrapper SI = *cast(TI); // Okay, TI has cases that are statically dead, prune them away. SmallPtrSet DeadCases; for (unsigned i = 0, e = PredCases.size(); i != e; ++i) DeadCases.insert(PredCases[i].Value); LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator() << "Through successor TI: " << *TI); SmallDenseMap NumPerSuccessorCases; for (SwitchInst::CaseIt i = SI->case_end(), e = SI->case_begin(); i != e;) { --i; auto *Successor = i->getCaseSuccessor(); if (DTU) ++NumPerSuccessorCases[Successor]; if (DeadCases.count(i->getCaseValue())) { Successor->removePredecessor(PredDef); SI.removeCase(i); if (DTU) --NumPerSuccessorCases[Successor]; } } if (DTU) { std::vector Updates; for (const std::pair &I : NumPerSuccessorCases) if (I.second == 0) Updates.push_back({DominatorTree::Delete, PredDef, I.first}); DTU->applyUpdates(Updates); } LLVM_DEBUG(dbgs() << "Leaving: " << *TI << "\n"); return true; } // Otherwise, TI's block must correspond to some matched value. Find out // which value (or set of values) this is. ConstantInt *TIV = nullptr; BasicBlock *TIBB = TI->getParent(); for (unsigned i = 0, e = PredCases.size(); i != e; ++i) if (PredCases[i].Dest == TIBB) { if (TIV) return false; // Cannot handle multiple values coming to this block. TIV = PredCases[i].Value; } assert(TIV && "No edge from pred to succ?"); // Okay, we found the one constant that our value can be if we get into TI's // BB. Find out which successor will unconditionally be branched to. BasicBlock *TheRealDest = nullptr; for (unsigned i = 0, e = ThisCases.size(); i != e; ++i) if (ThisCases[i].Value == TIV) { TheRealDest = ThisCases[i].Dest; break; } // If not handled by any explicit cases, it is handled by the default case. if (!TheRealDest) TheRealDest = ThisDef; SmallPtrSet RemovedSuccs; // Remove PHI node entries for dead edges. BasicBlock *CheckEdge = TheRealDest; for (BasicBlock *Succ : successors(TIBB)) if (Succ != CheckEdge) { if (Succ != TheRealDest) RemovedSuccs.insert(Succ); Succ->removePredecessor(TIBB); } else CheckEdge = nullptr; // Insert the new branch. Instruction *NI = Builder.CreateBr(TheRealDest); (void)NI; LLVM_DEBUG(dbgs() << "Threading pred instr: " << *Pred->getTerminator() << "Through successor TI: " << *TI << "Leaving: " << *NI << "\n"); EraseTerminatorAndDCECond(TI); if (DTU) { SmallVector Updates; Updates.reserve(RemovedSuccs.size()); for (auto *RemovedSucc : RemovedSuccs) Updates.push_back({DominatorTree::Delete, TIBB, RemovedSucc}); DTU->applyUpdates(Updates); } return true; } namespace { /// This class implements a stable ordering of constant /// integers that does not depend on their address. This is important for /// applications that sort ConstantInt's to ensure uniqueness. struct ConstantIntOrdering { bool operator()(const ConstantInt *LHS, const ConstantInt *RHS) const { return LHS->getValue().ult(RHS->getValue()); } }; } // end anonymous namespace static int ConstantIntSortPredicate(ConstantInt *const *P1, ConstantInt *const *P2) { const ConstantInt *LHS = *P1; const ConstantInt *RHS = *P2; if (LHS == RHS) return 0; return LHS->getValue().ult(RHS->getValue()) ? 1 : -1; } /// Get Weights of a given terminator, the default weight is at the front /// of the vector. If TI is a conditional eq, we need to swap the branch-weight /// metadata. static void GetBranchWeights(Instruction *TI, SmallVectorImpl &Weights) { MDNode *MD = TI->getMetadata(LLVMContext::MD_prof); assert(MD && "Invalid branch-weight metadata"); extractFromBranchWeightMD64(MD, Weights); // If TI is a conditional eq, the default case is the false case, // and the corresponding branch-weight data is at index 2. We swap the // default weight to be the first entry. if (BranchInst *BI = dyn_cast(TI)) { assert(Weights.size() == 2); ICmpInst *ICI = cast(BI->getCondition()); if (ICI->getPredicate() == ICmpInst::ICMP_EQ) std::swap(Weights.front(), Weights.back()); } } /// Keep halving the weights until all can fit in uint32_t. static void FitWeights(MutableArrayRef Weights) { uint64_t Max = *llvm::max_element(Weights); if (Max > UINT_MAX) { unsigned Offset = 32 - llvm::countl_zero(Max); for (uint64_t &I : Weights) I >>= Offset; } } static void CloneInstructionsIntoPredecessorBlockAndUpdateSSAUses( BasicBlock *BB, BasicBlock *PredBlock, ValueToValueMapTy &VMap) { Instruction *PTI = PredBlock->getTerminator(); // If we have bonus instructions, clone them into the predecessor block. // Note that there may be multiple predecessor blocks, so we cannot move // bonus instructions to a predecessor block. for (Instruction &BonusInst : *BB) { if (BonusInst.isTerminator()) continue; Instruction *NewBonusInst = BonusInst.clone(); if (!isa(BonusInst) && PTI->getDebugLoc() != NewBonusInst->getDebugLoc()) { // Unless the instruction has the same !dbg location as the original // branch, drop it. When we fold the bonus instructions we want to make // sure we reset their debug locations in order to avoid stepping on // dead code caused by folding dead branches. NewBonusInst->setDebugLoc(DebugLoc()); } RemapInstruction(NewBonusInst, VMap, RF_NoModuleLevelChanges | RF_IgnoreMissingLocals); // If we speculated an instruction, we need to drop any metadata that may // result in undefined behavior, as the metadata might have been valid // only given the branch precondition. // Similarly strip attributes on call parameters that may cause UB in // location the call is moved to. NewBonusInst->dropUBImplyingAttrsAndMetadata(); NewBonusInst->insertInto(PredBlock, PTI->getIterator()); auto Range = NewBonusInst->cloneDebugInfoFrom(&BonusInst); RemapDbgRecordRange(NewBonusInst->getModule(), Range, VMap, RF_NoModuleLevelChanges | RF_IgnoreMissingLocals); if (isa(BonusInst)) continue; NewBonusInst->takeName(&BonusInst); BonusInst.setName(NewBonusInst->getName() + ".old"); VMap[&BonusInst] = NewBonusInst; // Update (liveout) uses of bonus instructions, // now that the bonus instruction has been cloned into predecessor. // Note that we expect to be in a block-closed SSA form for this to work! for (Use &U : make_early_inc_range(BonusInst.uses())) { auto *UI = cast(U.getUser()); auto *PN = dyn_cast(UI); if (!PN) { assert(UI->getParent() == BB && BonusInst.comesBefore(UI) && "If the user is not a PHI node, then it should be in the same " "block as, and come after, the original bonus instruction."); continue; // Keep using the original bonus instruction. } // Is this the block-closed SSA form PHI node? if (PN->getIncomingBlock(U) == BB) continue; // Great, keep using the original bonus instruction. // The only other alternative is an "use" when coming from // the predecessor block - here we should refer to the cloned bonus instr. assert(PN->getIncomingBlock(U) == PredBlock && "Not in block-closed SSA form?"); U.set(NewBonusInst); } } } bool SimplifyCFGOpt::PerformValueComparisonIntoPredecessorFolding( Instruction *TI, Value *&CV, Instruction *PTI, IRBuilder<> &Builder) { BasicBlock *BB = TI->getParent(); BasicBlock *Pred = PTI->getParent(); SmallVector Updates; // Figure out which 'cases' to copy from SI to PSI. std::vector BBCases; BasicBlock *BBDefault = GetValueEqualityComparisonCases(TI, BBCases); std::vector PredCases; BasicBlock *PredDefault = GetValueEqualityComparisonCases(PTI, PredCases); // Based on whether the default edge from PTI goes to BB or not, fill in // PredCases and PredDefault with the new switch cases we would like to // build. SmallMapVector NewSuccessors; // Update the branch weight metadata along the way SmallVector Weights; bool PredHasWeights = hasBranchWeightMD(*PTI); bool SuccHasWeights = hasBranchWeightMD(*TI); if (PredHasWeights) { GetBranchWeights(PTI, Weights); // branch-weight metadata is inconsistent here. if (Weights.size() != 1 + PredCases.size()) PredHasWeights = SuccHasWeights = false; } else if (SuccHasWeights) // If there are no predecessor weights but there are successor weights, // populate Weights with 1, which will later be scaled to the sum of // successor's weights Weights.assign(1 + PredCases.size(), 1); SmallVector SuccWeights; if (SuccHasWeights) { GetBranchWeights(TI, SuccWeights); // branch-weight metadata is inconsistent here. if (SuccWeights.size() != 1 + BBCases.size()) PredHasWeights = SuccHasWeights = false; } else if (PredHasWeights) SuccWeights.assign(1 + BBCases.size(), 1); if (PredDefault == BB) { // If this is the default destination from PTI, only the edges in TI // that don't occur in PTI, or that branch to BB will be activated. std::set PTIHandled; for (unsigned i = 0, e = PredCases.size(); i != e; ++i) if (PredCases[i].Dest != BB) PTIHandled.insert(PredCases[i].Value); else { // The default destination is BB, we don't need explicit targets. std::swap(PredCases[i], PredCases.back()); if (PredHasWeights || SuccHasWeights) { // Increase weight for the default case. Weights[0] += Weights[i + 1]; std::swap(Weights[i + 1], Weights.back()); Weights.pop_back(); } PredCases.pop_back(); --i; --e; } // Reconstruct the new switch statement we will be building. if (PredDefault != BBDefault) { PredDefault->removePredecessor(Pred); if (DTU && PredDefault != BB) Updates.push_back({DominatorTree::Delete, Pred, PredDefault}); PredDefault = BBDefault; ++NewSuccessors[BBDefault]; } unsigned CasesFromPred = Weights.size(); uint64_t ValidTotalSuccWeight = 0; for (unsigned i = 0, e = BBCases.size(); i != e; ++i) if (!PTIHandled.count(BBCases[i].Value) && BBCases[i].Dest != BBDefault) { PredCases.push_back(BBCases[i]); ++NewSuccessors[BBCases[i].Dest]; if (SuccHasWeights || PredHasWeights) { // The default weight is at index 0, so weight for the ith case // should be at index i+1. Scale the cases from successor by // PredDefaultWeight (Weights[0]). Weights.push_back(Weights[0] * SuccWeights[i + 1]); ValidTotalSuccWeight += SuccWeights[i + 1]; } } if (SuccHasWeights || PredHasWeights) { ValidTotalSuccWeight += SuccWeights[0]; // Scale the cases from predecessor by ValidTotalSuccWeight. for (unsigned i = 1; i < CasesFromPred; ++i) Weights[i] *= ValidTotalSuccWeight; // Scale the default weight by SuccDefaultWeight (SuccWeights[0]). Weights[0] *= SuccWeights[0]; } } else { // If this is not the default destination from PSI, only the edges // in SI that occur in PSI with a destination of BB will be // activated. std::set PTIHandled; std::map WeightsForHandled; for (unsigned i = 0, e = PredCases.size(); i != e; ++i) if (PredCases[i].Dest == BB) { PTIHandled.insert(PredCases[i].Value); if (PredHasWeights || SuccHasWeights) { WeightsForHandled[PredCases[i].Value] = Weights[i + 1]; std::swap(Weights[i + 1], Weights.back()); Weights.pop_back(); } std::swap(PredCases[i], PredCases.back()); PredCases.pop_back(); --i; --e; } // Okay, now we know which constants were sent to BB from the // predecessor. Figure out where they will all go now. for (unsigned i = 0, e = BBCases.size(); i != e; ++i) if (PTIHandled.count(BBCases[i].Value)) { // If this is one we are capable of getting... if (PredHasWeights || SuccHasWeights) Weights.push_back(WeightsForHandled[BBCases[i].Value]); PredCases.push_back(BBCases[i]); ++NewSuccessors[BBCases[i].Dest]; PTIHandled.erase(BBCases[i].Value); // This constant is taken care of } // If there are any constants vectored to BB that TI doesn't handle, // they must go to the default destination of TI. for (ConstantInt *I : PTIHandled) { if (PredHasWeights || SuccHasWeights) Weights.push_back(WeightsForHandled[I]); PredCases.push_back(ValueEqualityComparisonCase(I, BBDefault)); ++NewSuccessors[BBDefault]; } } // Okay, at this point, we know which new successor Pred will get. Make // sure we update the number of entries in the PHI nodes for these // successors. SmallPtrSet SuccsOfPred; if (DTU) { SuccsOfPred = {succ_begin(Pred), succ_end(Pred)}; Updates.reserve(Updates.size() + NewSuccessors.size()); } for (const std::pair &NewSuccessor : NewSuccessors) { for (auto I : seq(NewSuccessor.second)) { (void)I; AddPredecessorToBlock(NewSuccessor.first, Pred, BB); } if (DTU && !SuccsOfPred.contains(NewSuccessor.first)) Updates.push_back({DominatorTree::Insert, Pred, NewSuccessor.first}); } Builder.SetInsertPoint(PTI); // Convert pointer to int before we switch. if (CV->getType()->isPointerTy()) { CV = Builder.CreatePtrToInt(CV, DL.getIntPtrType(CV->getType()), "magicptr"); } // Now that the successors are updated, create the new Switch instruction. SwitchInst *NewSI = Builder.CreateSwitch(CV, PredDefault, PredCases.size()); NewSI->setDebugLoc(PTI->getDebugLoc()); for (ValueEqualityComparisonCase &V : PredCases) NewSI->addCase(V.Value, V.Dest); if (PredHasWeights || SuccHasWeights) { // Halve the weights if any of them cannot fit in an uint32_t FitWeights(Weights); SmallVector MDWeights(Weights.begin(), Weights.end()); setBranchWeights(NewSI, MDWeights, /*IsExpected=*/false); } EraseTerminatorAndDCECond(PTI); // Okay, last check. If BB is still a successor of PSI, then we must // have an infinite loop case. If so, add an infinitely looping block // to handle the case to preserve the behavior of the code. BasicBlock *InfLoopBlock = nullptr; for (unsigned i = 0, e = NewSI->getNumSuccessors(); i != e; ++i) if (NewSI->getSuccessor(i) == BB) { if (!InfLoopBlock) { // Insert it at the end of the function, because it's either code, // or it won't matter if it's hot. :) InfLoopBlock = BasicBlock::Create(BB->getContext(), "infloop", BB->getParent()); BranchInst::Create(InfLoopBlock, InfLoopBlock); if (DTU) Updates.push_back( {DominatorTree::Insert, InfLoopBlock, InfLoopBlock}); } NewSI->setSuccessor(i, InfLoopBlock); } if (DTU) { if (InfLoopBlock) Updates.push_back({DominatorTree::Insert, Pred, InfLoopBlock}); Updates.push_back({DominatorTree::Delete, Pred, BB}); DTU->applyUpdates(Updates); } ++NumFoldValueComparisonIntoPredecessors; return true; } /// The specified terminator is a value equality comparison instruction /// (either a switch or a branch on "X == c"). /// See if any of the predecessors of the terminator block are value comparisons /// on the same value. If so, and if safe to do so, fold them together. bool SimplifyCFGOpt::FoldValueComparisonIntoPredecessors(Instruction *TI, IRBuilder<> &Builder) { BasicBlock *BB = TI->getParent(); Value *CV = isValueEqualityComparison(TI); // CondVal assert(CV && "Not a comparison?"); bool Changed = false; SmallSetVector Preds(pred_begin(BB), pred_end(BB)); while (!Preds.empty()) { BasicBlock *Pred = Preds.pop_back_val(); Instruction *PTI = Pred->getTerminator(); // Don't try to fold into itself. if (Pred == BB) continue; // See if the predecessor is a comparison with the same value. Value *PCV = isValueEqualityComparison(PTI); // PredCondVal if (PCV != CV) continue; SmallSetVector FailBlocks; if (!SafeToMergeTerminators(TI, PTI, &FailBlocks)) { for (auto *Succ : FailBlocks) { if (!SplitBlockPredecessors(Succ, TI->getParent(), ".fold.split", DTU)) return false; } } PerformValueComparisonIntoPredecessorFolding(TI, CV, PTI, Builder); Changed = true; } return Changed; } // If we would need to insert a select that uses the value of this invoke // (comments in hoistSuccIdenticalTerminatorToSwitchOrIf explain why we would // need to do this), we can't hoist the invoke, as there is nowhere to put the // select in this case. static bool isSafeToHoistInvoke(BasicBlock *BB1, BasicBlock *BB2, Instruction *I1, Instruction *I2) { for (BasicBlock *Succ : successors(BB1)) { for (const PHINode &PN : Succ->phis()) { Value *BB1V = PN.getIncomingValueForBlock(BB1); Value *BB2V = PN.getIncomingValueForBlock(BB2); if (BB1V != BB2V && (BB1V == I1 || BB2V == I2)) { return false; } } } return true; } // Get interesting characteristics of instructions that // `hoistCommonCodeFromSuccessors` didn't hoist. They restrict what kind of // instructions can be reordered across. enum SkipFlags { SkipReadMem = 1, SkipSideEffect = 2, SkipImplicitControlFlow = 4 }; static unsigned skippedInstrFlags(Instruction *I) { unsigned Flags = 0; if (I->mayReadFromMemory()) Flags |= SkipReadMem; // We can't arbitrarily move around allocas, e.g. moving allocas (especially // inalloca) across stacksave/stackrestore boundaries. if (I->mayHaveSideEffects() || isa(I)) Flags |= SkipSideEffect; if (!isGuaranteedToTransferExecutionToSuccessor(I)) Flags |= SkipImplicitControlFlow; return Flags; } // Returns true if it is safe to reorder an instruction across preceding // instructions in a basic block. static bool isSafeToHoistInstr(Instruction *I, unsigned Flags) { // Don't reorder a store over a load. if ((Flags & SkipReadMem) && I->mayWriteToMemory()) return false; // If we have seen an instruction with side effects, it's unsafe to reorder an // instruction which reads memory or itself has side effects. if ((Flags & SkipSideEffect) && (I->mayReadFromMemory() || I->mayHaveSideEffects() || isa(I))) return false; // Reordering across an instruction which does not necessarily transfer // control to the next instruction is speculation. if ((Flags & SkipImplicitControlFlow) && !isSafeToSpeculativelyExecute(I)) return false; // Hoisting of llvm.deoptimize is only legal together with the next return // instruction, which this pass is not always able to do. if (auto *CB = dyn_cast(I)) if (CB->getIntrinsicID() == Intrinsic::experimental_deoptimize) return false; // It's also unsafe/illegal to hoist an instruction above its instruction // operands BasicBlock *BB = I->getParent(); for (Value *Op : I->operands()) { if (auto *J = dyn_cast(Op)) if (J->getParent() == BB) return false; } return true; } static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I, bool PtrValueMayBeModified = false); /// Helper function for hoistCommonCodeFromSuccessors. Return true if identical /// instructions \p I1 and \p I2 can and should be hoisted. static bool shouldHoistCommonInstructions(Instruction *I1, Instruction *I2, const TargetTransformInfo &TTI) { // If we're going to hoist a call, make sure that the two instructions // we're commoning/hoisting are both marked with musttail, or neither of // them is marked as such. Otherwise, we might end up in a situation where // we hoist from a block where the terminator is a `ret` to a block where // the terminator is a `br`, and `musttail` calls expect to be followed by // a return. auto *C1 = dyn_cast(I1); auto *C2 = dyn_cast(I2); if (C1 && C2) if (C1->isMustTailCall() != C2->isMustTailCall()) return false; if (!TTI.isProfitableToHoist(I1) || !TTI.isProfitableToHoist(I2)) return false; // If any of the two call sites has nomerge or convergent attribute, stop // hoisting. if (const auto *CB1 = dyn_cast(I1)) if (CB1->cannotMerge() || CB1->isConvergent()) return false; if (const auto *CB2 = dyn_cast(I2)) if (CB2->cannotMerge() || CB2->isConvergent()) return false; return true; } /// Hoists DbgVariableRecords from \p I1 and \p OtherInstrs that are identical /// in lock-step to \p TI. This matches how dbg.* intrinsics are hoisting in /// hoistCommonCodeFromSuccessors. e.g. The input: /// I1 DVRs: { x, z }, /// OtherInsts: { I2 DVRs: { x, y, z } } /// would result in hoisting only DbgVariableRecord x. static void hoistLockstepIdenticalDbgVariableRecords( Instruction *TI, Instruction *I1, SmallVectorImpl &OtherInsts) { if (!I1->hasDbgRecords()) return; using CurrentAndEndIt = std::pair; // Vector of {Current, End} iterators. SmallVector Itrs; Itrs.reserve(OtherInsts.size() + 1); // Helper lambdas for lock-step checks: // Return true if this Current == End. auto atEnd = [](const CurrentAndEndIt &Pair) { return Pair.first == Pair.second; }; // Return true if all Current are identical. auto allIdentical = [](const SmallVector &Itrs) { return all_of(make_first_range(ArrayRef(Itrs).drop_front()), [&](DbgRecord::self_iterator I) { return Itrs[0].first->isIdenticalToWhenDefined(*I); }); }; // Collect the iterators. Itrs.push_back( {I1->getDbgRecordRange().begin(), I1->getDbgRecordRange().end()}); for (Instruction *Other : OtherInsts) { if (!Other->hasDbgRecords()) return; Itrs.push_back( {Other->getDbgRecordRange().begin(), Other->getDbgRecordRange().end()}); } // Iterate in lock-step until any of the DbgRecord lists are exausted. If // the lock-step DbgRecord are identical, hoist all of them to TI. // This replicates the dbg.* intrinsic behaviour in // hoistCommonCodeFromSuccessors. while (none_of(Itrs, atEnd)) { bool HoistDVRs = allIdentical(Itrs); for (CurrentAndEndIt &Pair : Itrs) { // Increment Current iterator now as we may be about to move the // DbgRecord. DbgRecord &DR = *Pair.first++; if (HoistDVRs) { DR.removeFromParent(); TI->getParent()->insertDbgRecordBefore(&DR, TI->getIterator()); } } } } /// Hoist any common code in the successor blocks up into the block. This /// function guarantees that BB dominates all successors. If EqTermsOnly is /// given, only perform hoisting in case both blocks only contain a terminator. /// In that case, only the original BI will be replaced and selects for PHIs are /// added. bool SimplifyCFGOpt::hoistCommonCodeFromSuccessors(BasicBlock *BB, bool EqTermsOnly) { // This does very trivial matching, with limited scanning, to find identical // instructions in the two blocks. In particular, we don't want to get into // O(N1*N2*...) situations here where Ni are the sizes of these successors. As // such, we currently just scan for obviously identical instructions in an // identical order, possibly separated by the same number of non-identical // instructions. unsigned int SuccSize = succ_size(BB); if (SuccSize < 2) return false; // If either of the blocks has it's address taken, then we can't do this fold, // because the code we'd hoist would no longer run when we jump into the block // by it's address. for (auto *Succ : successors(BB)) if (Succ->hasAddressTaken() || !Succ->getSinglePredecessor()) return false; auto *TI = BB->getTerminator(); // The second of pair is a SkipFlags bitmask. using SuccIterPair = std::pair; SmallVector SuccIterPairs; for (auto *Succ : successors(BB)) { BasicBlock::iterator SuccItr = Succ->begin(); if (isa(*SuccItr)) return false; SuccIterPairs.push_back(SuccIterPair(SuccItr, 0)); } // Check if only hoisting terminators is allowed. This does not add new // instructions to the hoist location. if (EqTermsOnly) { // Skip any debug intrinsics, as they are free to hoist. for (auto &SuccIter : make_first_range(SuccIterPairs)) { auto *INonDbg = &*skipDebugIntrinsics(SuccIter); if (!INonDbg->isTerminator()) return false; } // Now we know that we only need to hoist debug intrinsics and the // terminator. Let the loop below handle those 2 cases. } // Count how many instructions were not hoisted so far. There's a limit on how // many instructions we skip, serving as a compilation time control as well as // preventing excessive increase of life ranges. unsigned NumSkipped = 0; // If we find an unreachable instruction at the beginning of a basic block, we // can still hoist instructions from the rest of the basic blocks. if (SuccIterPairs.size() > 2) { erase_if(SuccIterPairs, [](const auto &Pair) { return isa(Pair.first); }); if (SuccIterPairs.size() < 2) return false; } bool Changed = false; for (;;) { auto *SuccIterPairBegin = SuccIterPairs.begin(); auto &BB1ItrPair = *SuccIterPairBegin++; auto OtherSuccIterPairRange = iterator_range(SuccIterPairBegin, SuccIterPairs.end()); auto OtherSuccIterRange = make_first_range(OtherSuccIterPairRange); Instruction *I1 = &*BB1ItrPair.first; // Skip debug info if it is not identical. bool AllDbgInstsAreIdentical = all_of(OtherSuccIterRange, [I1](auto &Iter) { Instruction *I2 = &*Iter; return I1->isIdenticalToWhenDefined(I2); }); if (!AllDbgInstsAreIdentical) { while (isa(I1)) I1 = &*++BB1ItrPair.first; for (auto &SuccIter : OtherSuccIterRange) { Instruction *I2 = &*SuccIter; while (isa(I2)) I2 = &*++SuccIter; } } bool AllInstsAreIdentical = true; bool HasTerminator = I1->isTerminator(); for (auto &SuccIter : OtherSuccIterRange) { Instruction *I2 = &*SuccIter; HasTerminator |= I2->isTerminator(); if (AllInstsAreIdentical && (!I1->isIdenticalToWhenDefined(I2) || MMRAMetadata(*I1) != MMRAMetadata(*I2))) AllInstsAreIdentical = false; } SmallVector OtherInsts; for (auto &SuccIter : OtherSuccIterRange) OtherInsts.push_back(&*SuccIter); // If we are hoisting the terminator instruction, don't move one (making a // broken BB), instead clone it, and remove BI. if (HasTerminator) { // Even if BB, which contains only one unreachable instruction, is ignored // at the beginning of the loop, we can hoist the terminator instruction. // If any instructions remain in the block, we cannot hoist terminators. if (NumSkipped || !AllInstsAreIdentical) { hoistLockstepIdenticalDbgVariableRecords(TI, I1, OtherInsts); return Changed; } return hoistSuccIdenticalTerminatorToSwitchOrIf(TI, I1, OtherInsts) || Changed; } if (AllInstsAreIdentical) { unsigned SkipFlagsBB1 = BB1ItrPair.second; AllInstsAreIdentical = isSafeToHoistInstr(I1, SkipFlagsBB1) && all_of(OtherSuccIterPairRange, [=](const auto &Pair) { Instruction *I2 = &*Pair.first; unsigned SkipFlagsBB2 = Pair.second; // Even if the instructions are identical, it may not // be safe to hoist them if we have skipped over // instructions with side effects or their operands // weren't hoisted. return isSafeToHoistInstr(I2, SkipFlagsBB2) && shouldHoistCommonInstructions(I1, I2, TTI); }); } if (AllInstsAreIdentical) { BB1ItrPair.first++; if (isa(I1)) { // The debug location is an integral part of a debug info intrinsic // and can't be separated from it or replaced. Instead of attempting // to merge locations, simply hoist both copies of the intrinsic. hoistLockstepIdenticalDbgVariableRecords(TI, I1, OtherInsts); // We've just hoisted DbgVariableRecords; move I1 after them (before TI) // and leave any that were not hoisted behind (by calling moveBefore // rather than moveBeforePreserving). I1->moveBefore(TI); for (auto &SuccIter : OtherSuccIterRange) { auto *I2 = &*SuccIter++; assert(isa(I2)); I2->moveBefore(TI); } } else { // For a normal instruction, we just move one to right before the // branch, then replace all uses of the other with the first. Finally, // we remove the now redundant second instruction. hoistLockstepIdenticalDbgVariableRecords(TI, I1, OtherInsts); // We've just hoisted DbgVariableRecords; move I1 after them (before TI) // and leave any that were not hoisted behind (by calling moveBefore // rather than moveBeforePreserving). I1->moveBefore(TI); for (auto &SuccIter : OtherSuccIterRange) { Instruction *I2 = &*SuccIter++; assert(I2 != I1); if (!I2->use_empty()) I2->replaceAllUsesWith(I1); I1->andIRFlags(I2); combineMetadataForCSE(I1, I2, true); // I1 and I2 are being combined into a single instruction. Its debug // location is the merged locations of the original instructions. I1->applyMergedLocation(I1->getDebugLoc(), I2->getDebugLoc()); I2->eraseFromParent(); } } if (!Changed) NumHoistCommonCode += SuccIterPairs.size(); Changed = true; NumHoistCommonInstrs += SuccIterPairs.size(); } else { if (NumSkipped >= HoistCommonSkipLimit) { hoistLockstepIdenticalDbgVariableRecords(TI, I1, OtherInsts); return Changed; } // We are about to skip over a pair of non-identical instructions. Record // if any have characteristics that would prevent reordering instructions // across them. for (auto &SuccIterPair : SuccIterPairs) { Instruction *I = &*SuccIterPair.first++; SuccIterPair.second |= skippedInstrFlags(I); } ++NumSkipped; } } } bool SimplifyCFGOpt::hoistSuccIdenticalTerminatorToSwitchOrIf( Instruction *TI, Instruction *I1, SmallVectorImpl &OtherSuccTIs) { auto *BI = dyn_cast(TI); bool Changed = false; BasicBlock *TIParent = TI->getParent(); BasicBlock *BB1 = I1->getParent(); // Use only for an if statement. auto *I2 = *OtherSuccTIs.begin(); auto *BB2 = I2->getParent(); if (BI) { assert(OtherSuccTIs.size() == 1); assert(BI->getSuccessor(0) == I1->getParent()); assert(BI->getSuccessor(1) == I2->getParent()); } // In the case of an if statement, we try to hoist an invoke. // FIXME: Can we define a safety predicate for CallBr? // FIXME: Test case llvm/test/Transforms/SimplifyCFG/2009-06-15-InvokeCrash.ll // removed in 4c923b3b3fd0ac1edebf0603265ca3ba51724937 commit? if (isa(I1) && (!BI || !isSafeToHoistInvoke(BB1, BB2, I1, I2))) return false; // TODO: callbr hoisting currently disabled pending further study. if (isa(I1)) return false; for (BasicBlock *Succ : successors(BB1)) { for (PHINode &PN : Succ->phis()) { Value *BB1V = PN.getIncomingValueForBlock(BB1); for (Instruction *OtherSuccTI : OtherSuccTIs) { Value *BB2V = PN.getIncomingValueForBlock(OtherSuccTI->getParent()); if (BB1V == BB2V) continue; // In the case of an if statement, check for // passingValueIsAlwaysUndefined here because we would rather eliminate // undefined control flow then converting it to a select. if (!BI || passingValueIsAlwaysUndefined(BB1V, &PN) || passingValueIsAlwaysUndefined(BB2V, &PN)) return false; } } } // Hoist DbgVariableRecords attached to the terminator to match dbg.* // intrinsic hoisting behaviour in hoistCommonCodeFromSuccessors. hoistLockstepIdenticalDbgVariableRecords(TI, I1, OtherSuccTIs); // Clone the terminator and hoist it into the pred, without any debug info. Instruction *NT = I1->clone(); NT->insertInto(TIParent, TI->getIterator()); if (!NT->getType()->isVoidTy()) { I1->replaceAllUsesWith(NT); for (Instruction *OtherSuccTI : OtherSuccTIs) OtherSuccTI->replaceAllUsesWith(NT); NT->takeName(I1); } Changed = true; NumHoistCommonInstrs += OtherSuccTIs.size() + 1; // Ensure terminator gets a debug location, even an unknown one, in case // it involves inlinable calls. SmallVector Locs; Locs.push_back(I1->getDebugLoc()); for (auto *OtherSuccTI : OtherSuccTIs) Locs.push_back(OtherSuccTI->getDebugLoc()); NT->setDebugLoc(DILocation::getMergedLocations(Locs)); // PHIs created below will adopt NT's merged DebugLoc. IRBuilder Builder(NT); // In the case of an if statement, hoisting one of the terminators from our // successor is a great thing. Unfortunately, the successors of the if/else // blocks may have PHI nodes in them. If they do, all PHI entries for BB1/BB2 // must agree for all PHI nodes, so we insert select instruction to compute // the final result. if (BI) { std::map, SelectInst *> InsertedSelects; for (BasicBlock *Succ : successors(BB1)) { for (PHINode &PN : Succ->phis()) { Value *BB1V = PN.getIncomingValueForBlock(BB1); Value *BB2V = PN.getIncomingValueForBlock(BB2); if (BB1V == BB2V) continue; // These values do not agree. Insert a select instruction before NT // that determines the right value. SelectInst *&SI = InsertedSelects[std::make_pair(BB1V, BB2V)]; if (!SI) { // Propagate fast-math-flags from phi node to its replacement select. IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); if (isa(PN)) Builder.setFastMathFlags(PN.getFastMathFlags()); SI = cast(Builder.CreateSelect( BI->getCondition(), BB1V, BB2V, BB1V->getName() + "." + BB2V->getName(), BI)); } // Make the PHI node use the select for all incoming values for BB1/BB2 for (unsigned i = 0, e = PN.getNumIncomingValues(); i != e; ++i) if (PN.getIncomingBlock(i) == BB1 || PN.getIncomingBlock(i) == BB2) PN.setIncomingValue(i, SI); } } } SmallVector Updates; // Update any PHI nodes in our new successors. for (BasicBlock *Succ : successors(BB1)) { AddPredecessorToBlock(Succ, TIParent, BB1); if (DTU) Updates.push_back({DominatorTree::Insert, TIParent, Succ}); } if (DTU) for (BasicBlock *Succ : successors(TI)) Updates.push_back({DominatorTree::Delete, TIParent, Succ}); EraseTerminatorAndDCECond(TI); if (DTU) DTU->applyUpdates(Updates); return Changed; } // Check lifetime markers. static bool isLifeTimeMarker(const Instruction *I) { if (auto II = dyn_cast(I)) { switch (II->getIntrinsicID()) { default: break; case Intrinsic::lifetime_start: case Intrinsic::lifetime_end: return true; } } return false; } // TODO: Refine this. This should avoid cases like turning constant memcpy sizes // into variables. static bool replacingOperandWithVariableIsCheap(const Instruction *I, int OpIdx) { return !isa(I); } // All instructions in Insts belong to different blocks that all unconditionally // branch to a common successor. Analyze each instruction and return true if it // would be possible to sink them into their successor, creating one common // instruction instead. For every value that would be required to be provided by // PHI node (because an operand varies in each input block), add to PHIOperands. static bool canSinkInstructions( ArrayRef Insts, DenseMap> &PHIOperands) { // Prune out obviously bad instructions to move. Each instruction must have // the same number of uses, and we check later that the uses are consistent. std::optional NumUses; for (auto *I : Insts) { // These instructions may change or break semantics if moved. if (isa(I) || I->isEHPad() || isa(I) || I->getType()->isTokenTy()) return false; // Do not try to sink an instruction in an infinite loop - it can cause // this algorithm to infinite loop. if (I->getParent()->getSingleSuccessor() == I->getParent()) return false; // Conservatively return false if I is an inline-asm instruction. Sinking // and merging inline-asm instructions can potentially create arguments // that cannot satisfy the inline-asm constraints. // If the instruction has nomerge or convergent attribute, return false. if (const auto *C = dyn_cast(I)) if (C->isInlineAsm() || C->cannotMerge() || C->isConvergent()) return false; if (!NumUses) NumUses = I->getNumUses(); else if (NumUses != I->getNumUses()) return false; } const Instruction *I0 = Insts.front(); const auto I0MMRA = MMRAMetadata(*I0); for (auto *I : Insts) { if (!I->isSameOperationAs(I0)) return false; // swifterror pointers can only be used by a load or store; sinking a load // or store would require introducing a select for the pointer operand, // which isn't allowed for swifterror pointers. if (isa(I) && I->getOperand(1)->isSwiftError()) return false; if (isa(I) && I->getOperand(0)->isSwiftError()) return false; // Treat MMRAs conservatively. This pass can be quite aggressive and // could drop a lot of MMRAs otherwise. if (MMRAMetadata(*I) != I0MMRA) return false; } // Uses must be consistent: If I0 is used in a phi node in the sink target, // then the other phi operands must match the instructions from Insts. This // also has to hold true for any phi nodes that would be created as a result // of sinking. Both of these cases are represented by PhiOperands. for (const Use &U : I0->uses()) { auto It = PHIOperands.find(&U); if (It == PHIOperands.end()) // There may be uses in other blocks when sinking into a loop header. return false; if (!equal(Insts, It->second)) return false; } // Because SROA can't handle speculating stores of selects, try not to sink // loads, stores or lifetime markers of allocas when we'd have to create a // PHI for the address operand. Also, because it is likely that loads or // stores of allocas will disappear when Mem2Reg/SROA is run, don't sink // them. // This can cause code churn which can have unintended consequences down // the line - see https://llvm.org/bugs/show_bug.cgi?id=30244. // FIXME: This is a workaround for a deficiency in SROA - see // https://llvm.org/bugs/show_bug.cgi?id=30188 if (isa(I0) && any_of(Insts, [](const Instruction *I) { return isa(I->getOperand(1)->stripPointerCasts()); })) return false; if (isa(I0) && any_of(Insts, [](const Instruction *I) { return isa(I->getOperand(0)->stripPointerCasts()); })) return false; if (isLifeTimeMarker(I0) && any_of(Insts, [](const Instruction *I) { return isa(I->getOperand(1)->stripPointerCasts()); })) return false; // For calls to be sinkable, they must all be indirect, or have same callee. // I.e. if we have two direct calls to different callees, we don't want to // turn that into an indirect call. Likewise, if we have an indirect call, // and a direct call, we don't actually want to have a single indirect call. if (isa(I0)) { auto IsIndirectCall = [](const Instruction *I) { return cast(I)->isIndirectCall(); }; bool HaveIndirectCalls = any_of(Insts, IsIndirectCall); bool AllCallsAreIndirect = all_of(Insts, IsIndirectCall); if (HaveIndirectCalls) { if (!AllCallsAreIndirect) return false; } else { // All callees must be identical. Value *Callee = nullptr; for (const Instruction *I : Insts) { Value *CurrCallee = cast(I)->getCalledOperand(); if (!Callee) Callee = CurrCallee; else if (Callee != CurrCallee) return false; } } } for (unsigned OI = 0, OE = I0->getNumOperands(); OI != OE; ++OI) { Value *Op = I0->getOperand(OI); if (Op->getType()->isTokenTy()) // Don't touch any operand of token type. return false; auto SameAsI0 = [&I0, OI](const Instruction *I) { assert(I->getNumOperands() == I0->getNumOperands()); return I->getOperand(OI) == I0->getOperand(OI); }; if (!all_of(Insts, SameAsI0)) { if ((isa(Op) && !replacingOperandWithVariableIsCheap(I0, OI)) || !canReplaceOperandWithVariable(I0, OI)) // We can't create a PHI from this GEP. return false; auto &Ops = PHIOperands[&I0->getOperandUse(OI)]; for (auto *I : Insts) Ops.push_back(I->getOperand(OI)); } } return true; } // Assuming canSinkInstructions(Blocks) has returned true, sink the last // instruction of every block in Blocks to their common successor, commoning // into one instruction. static void sinkLastInstruction(ArrayRef Blocks) { auto *BBEnd = Blocks[0]->getTerminator()->getSuccessor(0); // canSinkInstructions returning true guarantees that every block has at // least one non-terminator instruction. SmallVector Insts; for (auto *BB : Blocks) { Instruction *I = BB->getTerminator(); do { I = I->getPrevNode(); } while (isa(I) && I != &BB->front()); if (!isa(I)) Insts.push_back(I); } // We don't need to do any more checking here; canSinkInstructions should // have done it all for us. SmallVector NewOperands; Instruction *I0 = Insts.front(); for (unsigned O = 0, E = I0->getNumOperands(); O != E; ++O) { // This check is different to that in canSinkInstructions. There, we // cared about the global view once simplifycfg (and instcombine) have // completed - it takes into account PHIs that become trivially // simplifiable. However here we need a more local view; if an operand // differs we create a PHI and rely on instcombine to clean up the very // small mess we may make. bool NeedPHI = any_of(Insts, [&I0, O](const Instruction *I) { return I->getOperand(O) != I0->getOperand(O); }); if (!NeedPHI) { NewOperands.push_back(I0->getOperand(O)); continue; } // Create a new PHI in the successor block and populate it. auto *Op = I0->getOperand(O); assert(!Op->getType()->isTokenTy() && "Can't PHI tokens!"); auto *PN = PHINode::Create(Op->getType(), Insts.size(), Op->getName() + ".sink"); PN->insertBefore(BBEnd->begin()); for (auto *I : Insts) PN->addIncoming(I->getOperand(O), I->getParent()); NewOperands.push_back(PN); } // Arbitrarily use I0 as the new "common" instruction; remap its operands // and move it to the start of the successor block. for (unsigned O = 0, E = I0->getNumOperands(); O != E; ++O) I0->getOperandUse(O).set(NewOperands[O]); I0->moveBefore(*BBEnd, BBEnd->getFirstInsertionPt()); // Update metadata and IR flags, and merge debug locations. for (auto *I : Insts) if (I != I0) { // The debug location for the "common" instruction is the merged locations // of all the commoned instructions. We start with the original location // of the "common" instruction and iteratively merge each location in the // loop below. // This is an N-way merge, which will be inefficient if I0 is a CallInst. // However, as N-way merge for CallInst is rare, so we use simplified API // instead of using complex API for N-way merge. I0->applyMergedLocation(I0->getDebugLoc(), I->getDebugLoc()); combineMetadataForCSE(I0, I, true); I0->andIRFlags(I); } for (User *U : make_early_inc_range(I0->users())) { // canSinkLastInstruction checked that all instructions are only used by // phi nodes in a way that allows replacing the phi node with the common // instruction. auto *PN = cast(U); PN->replaceAllUsesWith(I0); PN->eraseFromParent(); } // Finally nuke all instructions apart from the common instruction. for (auto *I : Insts) { if (I == I0) continue; // The remaining uses are debug users, replace those with the common inst. // In most (all?) cases this just introduces a use-before-def. assert(I->user_empty() && "Inst unexpectedly still has non-dbg users"); I->replaceAllUsesWith(I0); I->eraseFromParent(); } } namespace { // LockstepReverseIterator - Iterates through instructions // in a set of blocks in reverse order from the first non-terminator. // For example (assume all blocks have size n): // LockstepReverseIterator I([B1, B2, B3]); // *I-- = [B1[n], B2[n], B3[n]]; // *I-- = [B1[n-1], B2[n-1], B3[n-1]]; // *I-- = [B1[n-2], B2[n-2], B3[n-2]]; // ... class LockstepReverseIterator { ArrayRef Blocks; SmallVector Insts; bool Fail; public: LockstepReverseIterator(ArrayRef Blocks) : Blocks(Blocks) { reset(); } void reset() { Fail = false; Insts.clear(); for (auto *BB : Blocks) { Instruction *Inst = BB->getTerminator(); for (Inst = Inst->getPrevNode(); Inst && isa(Inst);) Inst = Inst->getPrevNode(); if (!Inst) { // Block wasn't big enough. Fail = true; return; } Insts.push_back(Inst); } } bool isValid() const { return !Fail; } void operator--() { if (Fail) return; for (auto *&Inst : Insts) { for (Inst = Inst->getPrevNode(); Inst && isa(Inst);) Inst = Inst->getPrevNode(); // Already at beginning of block. if (!Inst) { Fail = true; return; } } } void operator++() { if (Fail) return; for (auto *&Inst : Insts) { for (Inst = Inst->getNextNode(); Inst && isa(Inst);) Inst = Inst->getNextNode(); // Already at end of block. if (!Inst) { Fail = true; return; } } } ArrayRef operator * () const { return Insts; } }; } // end anonymous namespace /// Check whether BB's predecessors end with unconditional branches. If it is /// true, sink any common code from the predecessors to BB. static bool SinkCommonCodeFromPredecessors(BasicBlock *BB, DomTreeUpdater *DTU) { // We support two situations: // (1) all incoming arcs are unconditional // (2) there are non-unconditional incoming arcs // // (2) is very common in switch defaults and // else-if patterns; // // if (a) f(1); // else if (b) f(2); // // produces: // // [if] // / \ // [f(1)] [if] // | | \ // | | | // | [f(2)]| // \ | / // [ end ] // // [end] has two unconditional predecessor arcs and one conditional. The // conditional refers to the implicit empty 'else' arc. This conditional // arc can also be caused by an empty default block in a switch. // // In this case, we attempt to sink code from all *unconditional* arcs. // If we can sink instructions from these arcs (determined during the scan // phase below) we insert a common successor for all unconditional arcs and // connect that to [end], to enable sinking: // // [if] // / \ // [x(1)] [if] // | | \ // | | \ // | [x(2)] | // \ / | // [sink.split] | // \ / // [ end ] // SmallVector UnconditionalPreds; bool HaveNonUnconditionalPredecessors = false; for (auto *PredBB : predecessors(BB)) { auto *PredBr = dyn_cast(PredBB->getTerminator()); if (PredBr && PredBr->isUnconditional()) UnconditionalPreds.push_back(PredBB); else HaveNonUnconditionalPredecessors = true; } if (UnconditionalPreds.size() < 2) return false; // We take a two-step approach to tail sinking. First we scan from the end of // each block upwards in lockstep. If the n'th instruction from the end of each // block can be sunk, those instructions are added to ValuesToSink and we // carry on. If we can sink an instruction but need to PHI-merge some operands // (because they're not identical in each instruction) we add these to // PHIOperands. // We prepopulate PHIOperands with the phis that already exist in BB. DenseMap> PHIOperands; for (PHINode &PN : BB->phis()) { SmallDenseMap IncomingVals; for (const Use &U : PN.incoming_values()) IncomingVals.insert({PN.getIncomingBlock(U), &U}); auto &Ops = PHIOperands[IncomingVals[UnconditionalPreds[0]]]; for (BasicBlock *Pred : UnconditionalPreds) Ops.push_back(*IncomingVals[Pred]); } int ScanIdx = 0; SmallPtrSet InstructionsToSink; LockstepReverseIterator LRI(UnconditionalPreds); while (LRI.isValid() && canSinkInstructions(*LRI, PHIOperands)) { LLVM_DEBUG(dbgs() << "SINK: instruction can be sunk: " << *(*LRI)[0] << "\n"); InstructionsToSink.insert((*LRI).begin(), (*LRI).end()); ++ScanIdx; --LRI; } // If no instructions can be sunk, early-return. if (ScanIdx == 0) return false; bool followedByDeoptOrUnreachable = IsBlockFollowedByDeoptOrUnreachable(BB); if (!followedByDeoptOrUnreachable) { // Okay, we *could* sink last ScanIdx instructions. But how many can we // actually sink before encountering instruction that is unprofitable to // sink? auto ProfitableToSinkInstruction = [&](LockstepReverseIterator &LRI) { unsigned NumPHIInsts = 0; for (Use &U : (*LRI)[0]->operands()) { auto It = PHIOperands.find(&U); if (It != PHIOperands.end() && !all_of(It->second, [&](Value *V) { return InstructionsToSink.contains(V); })) { ++NumPHIInsts; // FIXME: this check is overly optimistic. We may end up not sinking // said instruction, due to the very same profitability check. // See @creating_too_many_phis in sink-common-code.ll. } } LLVM_DEBUG(dbgs() << "SINK: #phi insts: " << NumPHIInsts << "\n"); return NumPHIInsts <= 1; }; // We've determined that we are going to sink last ScanIdx instructions, // and recorded them in InstructionsToSink. Now, some instructions may be // unprofitable to sink. But that determination depends on the instructions // that we are going to sink. // First, forward scan: find the first instruction unprofitable to sink, // recording all the ones that are profitable to sink. // FIXME: would it be better, after we detect that not all are profitable. // to either record the profitable ones, or erase the unprofitable ones? // Maybe we need to choose (at runtime) the one that will touch least // instrs? LRI.reset(); int Idx = 0; SmallPtrSet InstructionsProfitableToSink; while (Idx < ScanIdx) { if (!ProfitableToSinkInstruction(LRI)) { // Too many PHIs would be created. LLVM_DEBUG( dbgs() << "SINK: stopping here, too many PHIs would be created!\n"); break; } InstructionsProfitableToSink.insert((*LRI).begin(), (*LRI).end()); --LRI; ++Idx; } // If no instructions can be sunk, early-return. if (Idx == 0) return false; // Did we determine that (only) some instructions are unprofitable to sink? if (Idx < ScanIdx) { // Okay, some instructions are unprofitable. ScanIdx = Idx; InstructionsToSink = InstructionsProfitableToSink; // But, that may make other instructions unprofitable, too. // So, do a backward scan, do any earlier instructions become // unprofitable? assert( !ProfitableToSinkInstruction(LRI) && "We already know that the last instruction is unprofitable to sink"); ++LRI; --Idx; while (Idx >= 0) { // If we detect that an instruction becomes unprofitable to sink, // all earlier instructions won't be sunk either, // so preemptively keep InstructionsProfitableToSink in sync. // FIXME: is this the most performant approach? for (auto *I : *LRI) InstructionsProfitableToSink.erase(I); if (!ProfitableToSinkInstruction(LRI)) { // Everything starting with this instruction won't be sunk. ScanIdx = Idx; InstructionsToSink = InstructionsProfitableToSink; } ++LRI; --Idx; } } // If no instructions can be sunk, early-return. if (ScanIdx == 0) return false; } bool Changed = false; if (HaveNonUnconditionalPredecessors) { if (!followedByDeoptOrUnreachable) { // It is always legal to sink common instructions from unconditional // predecessors. However, if not all predecessors are unconditional, // this transformation might be pessimizing. So as a rule of thumb, // don't do it unless we'd sink at least one non-speculatable instruction. // See https://bugs.llvm.org/show_bug.cgi?id=30244 LRI.reset(); int Idx = 0; bool Profitable = false; while (Idx < ScanIdx) { if (!isSafeToSpeculativelyExecute((*LRI)[0])) { Profitable = true; break; } --LRI; ++Idx; } if (!Profitable) return false; } LLVM_DEBUG(dbgs() << "SINK: Splitting edge\n"); // We have a conditional edge and we're going to sink some instructions. // Insert a new block postdominating all blocks we're going to sink from. if (!SplitBlockPredecessors(BB, UnconditionalPreds, ".sink.split", DTU)) // Edges couldn't be split. return false; Changed = true; } // Now that we've analyzed all potential sinking candidates, perform the // actual sink. We iteratively sink the last non-terminator of the source // blocks into their common successor unless doing so would require too // many PHI instructions to be generated (currently only one PHI is allowed // per sunk instruction). // // We can use InstructionsToSink to discount values needing PHI-merging that will // actually be sunk in a later iteration. This allows us to be more // aggressive in what we sink. This does allow a false positive where we // sink presuming a later value will also be sunk, but stop half way through // and never actually sink it which means we produce more PHIs than intended. // This is unlikely in practice though. int SinkIdx = 0; for (; SinkIdx != ScanIdx; ++SinkIdx) { LLVM_DEBUG(dbgs() << "SINK: Sink: " << *UnconditionalPreds[0]->getTerminator()->getPrevNode() << "\n"); // Because we've sunk every instruction in turn, the current instruction to // sink is always at index 0. LRI.reset(); sinkLastInstruction(UnconditionalPreds); NumSinkCommonInstrs++; Changed = true; } if (SinkIdx != 0) ++NumSinkCommonCode; return Changed; } namespace { struct CompatibleSets { using SetTy = SmallVector; SmallVector Sets; static bool shouldBelongToSameSet(ArrayRef Invokes); SetTy &getCompatibleSet(InvokeInst *II); void insert(InvokeInst *II); }; CompatibleSets::SetTy &CompatibleSets::getCompatibleSet(InvokeInst *II) { // Perform a linear scan over all the existing sets, see if the new `invoke` // is compatible with any particular set. Since we know that all the `invokes` // within a set are compatible, only check the first `invoke` in each set. // WARNING: at worst, this has quadratic complexity. for (CompatibleSets::SetTy &Set : Sets) { if (CompatibleSets::shouldBelongToSameSet({Set.front(), II})) return Set; } // Otherwise, we either had no sets yet, or this invoke forms a new set. return Sets.emplace_back(); } void CompatibleSets::insert(InvokeInst *II) { getCompatibleSet(II).emplace_back(II); } bool CompatibleSets::shouldBelongToSameSet(ArrayRef Invokes) { assert(Invokes.size() == 2 && "Always called with exactly two candidates."); // Can we theoretically merge these `invoke`s? auto IsIllegalToMerge = [](InvokeInst *II) { return II->cannotMerge() || II->isInlineAsm(); }; if (any_of(Invokes, IsIllegalToMerge)) return false; // Either both `invoke`s must be direct, // or both `invoke`s must be indirect. auto IsIndirectCall = [](InvokeInst *II) { return II->isIndirectCall(); }; bool HaveIndirectCalls = any_of(Invokes, IsIndirectCall); bool AllCallsAreIndirect = all_of(Invokes, IsIndirectCall); if (HaveIndirectCalls) { if (!AllCallsAreIndirect) return false; } else { // All callees must be identical. Value *Callee = nullptr; for (InvokeInst *II : Invokes) { Value *CurrCallee = II->getCalledOperand(); assert(CurrCallee && "There is always a called operand."); if (!Callee) Callee = CurrCallee; else if (Callee != CurrCallee) return false; } } // Either both `invoke`s must not have a normal destination, // or both `invoke`s must have a normal destination, auto HasNormalDest = [](InvokeInst *II) { return !isa(II->getNormalDest()->getFirstNonPHIOrDbg()); }; if (any_of(Invokes, HasNormalDest)) { // Do not merge `invoke` that does not have a normal destination with one // that does have a normal destination, even though doing so would be legal. if (!all_of(Invokes, HasNormalDest)) return false; // All normal destinations must be identical. BasicBlock *NormalBB = nullptr; for (InvokeInst *II : Invokes) { BasicBlock *CurrNormalBB = II->getNormalDest(); assert(CurrNormalBB && "There is always a 'continue to' basic block."); if (!NormalBB) NormalBB = CurrNormalBB; else if (NormalBB != CurrNormalBB) return false; } // In the normal destination, the incoming values for these two `invoke`s // must be compatible. SmallPtrSet EquivalenceSet(Invokes.begin(), Invokes.end()); if (!IncomingValuesAreCompatible( NormalBB, {Invokes[0]->getParent(), Invokes[1]->getParent()}, &EquivalenceSet)) return false; } #ifndef NDEBUG // All unwind destinations must be identical. // We know that because we have started from said unwind destination. BasicBlock *UnwindBB = nullptr; for (InvokeInst *II : Invokes) { BasicBlock *CurrUnwindBB = II->getUnwindDest(); assert(CurrUnwindBB && "There is always an 'unwind to' basic block."); if (!UnwindBB) UnwindBB = CurrUnwindBB; else assert(UnwindBB == CurrUnwindBB && "Unexpected unwind destination."); } #endif // In the unwind destination, the incoming values for these two `invoke`s // must be compatible. if (!IncomingValuesAreCompatible( Invokes.front()->getUnwindDest(), {Invokes[0]->getParent(), Invokes[1]->getParent()})) return false; // Ignoring arguments, these `invoke`s must be identical, // including operand bundles. const InvokeInst *II0 = Invokes.front(); for (auto *II : Invokes.drop_front()) if (!II->isSameOperationAs(II0)) return false; // Can we theoretically form the data operands for the merged `invoke`? auto IsIllegalToMergeArguments = [](auto Ops) { Use &U0 = std::get<0>(Ops); Use &U1 = std::get<1>(Ops); if (U0 == U1) return false; return U0->getType()->isTokenTy() || !canReplaceOperandWithVariable(cast(U0.getUser()), U0.getOperandNo()); }; assert(Invokes.size() == 2 && "Always called with exactly two candidates."); if (any_of(zip(Invokes[0]->data_ops(), Invokes[1]->data_ops()), IsIllegalToMergeArguments)) return false; return true; } } // namespace // Merge all invokes in the provided set, all of which are compatible // as per the `CompatibleSets::shouldBelongToSameSet()`. static void MergeCompatibleInvokesImpl(ArrayRef Invokes, DomTreeUpdater *DTU) { assert(Invokes.size() >= 2 && "Must have at least two invokes to merge."); SmallVector Updates; if (DTU) Updates.reserve(2 + 3 * Invokes.size()); bool HasNormalDest = !isa(Invokes[0]->getNormalDest()->getFirstNonPHIOrDbg()); // Clone one of the invokes into a new basic block. // Since they are all compatible, it doesn't matter which invoke is cloned. InvokeInst *MergedInvoke = [&Invokes, HasNormalDest]() { InvokeInst *II0 = Invokes.front(); BasicBlock *II0BB = II0->getParent(); BasicBlock *InsertBeforeBlock = II0->getParent()->getIterator()->getNextNode(); Function *Func = II0BB->getParent(); LLVMContext &Ctx = II0->getContext(); BasicBlock *MergedInvokeBB = BasicBlock::Create( Ctx, II0BB->getName() + ".invoke", Func, InsertBeforeBlock); auto *MergedInvoke = cast(II0->clone()); // NOTE: all invokes have the same attributes, so no handling needed. MergedInvoke->insertInto(MergedInvokeBB, MergedInvokeBB->end()); if (!HasNormalDest) { // This set does not have a normal destination, // so just form a new block with unreachable terminator. BasicBlock *MergedNormalDest = BasicBlock::Create( Ctx, II0BB->getName() + ".cont", Func, InsertBeforeBlock); new UnreachableInst(Ctx, MergedNormalDest); MergedInvoke->setNormalDest(MergedNormalDest); } // The unwind destination, however, remainds identical for all invokes here. return MergedInvoke; }(); if (DTU) { // Predecessor blocks that contained these invokes will now branch to // the new block that contains the merged invoke, ... for (InvokeInst *II : Invokes) Updates.push_back( {DominatorTree::Insert, II->getParent(), MergedInvoke->getParent()}); // ... which has the new `unreachable` block as normal destination, // or unwinds to the (same for all `invoke`s in this set) `landingpad`, for (BasicBlock *SuccBBOfMergedInvoke : successors(MergedInvoke)) Updates.push_back({DominatorTree::Insert, MergedInvoke->getParent(), SuccBBOfMergedInvoke}); // Since predecessor blocks now unconditionally branch to a new block, // they no longer branch to their original successors. for (InvokeInst *II : Invokes) for (BasicBlock *SuccOfPredBB : successors(II->getParent())) Updates.push_back( {DominatorTree::Delete, II->getParent(), SuccOfPredBB}); } bool IsIndirectCall = Invokes[0]->isIndirectCall(); // Form the merged operands for the merged invoke. for (Use &U : MergedInvoke->operands()) { // Only PHI together the indirect callees and data operands. if (MergedInvoke->isCallee(&U)) { if (!IsIndirectCall) continue; } else if (!MergedInvoke->isDataOperand(&U)) continue; // Don't create trivial PHI's with all-identical incoming values. bool NeedPHI = any_of(Invokes, [&U](InvokeInst *II) { return II->getOperand(U.getOperandNo()) != U.get(); }); if (!NeedPHI) continue; // Form a PHI out of all the data ops under this index. PHINode *PN = PHINode::Create( U->getType(), /*NumReservedValues=*/Invokes.size(), "", MergedInvoke->getIterator()); for (InvokeInst *II : Invokes) PN->addIncoming(II->getOperand(U.getOperandNo()), II->getParent()); U.set(PN); } // We've ensured that each PHI node has compatible (identical) incoming values // when coming from each of the `invoke`s in the current merge set, // so update the PHI nodes accordingly. for (BasicBlock *Succ : successors(MergedInvoke)) AddPredecessorToBlock(Succ, /*NewPred=*/MergedInvoke->getParent(), /*ExistPred=*/Invokes.front()->getParent()); // And finally, replace the original `invoke`s with an unconditional branch // to the block with the merged `invoke`. Also, give that merged `invoke` // the merged debugloc of all the original `invoke`s. DILocation *MergedDebugLoc = nullptr; for (InvokeInst *II : Invokes) { // Compute the debug location common to all the original `invoke`s. if (!MergedDebugLoc) MergedDebugLoc = II->getDebugLoc(); else MergedDebugLoc = DILocation::getMergedLocation(MergedDebugLoc, II->getDebugLoc()); // And replace the old `invoke` with an unconditionally branch // to the block with the merged `invoke`. for (BasicBlock *OrigSuccBB : successors(II->getParent())) OrigSuccBB->removePredecessor(II->getParent()); BranchInst::Create(MergedInvoke->getParent(), II->getParent()); II->replaceAllUsesWith(MergedInvoke); II->eraseFromParent(); ++NumInvokesMerged; } MergedInvoke->setDebugLoc(MergedDebugLoc); ++NumInvokeSetsFormed; if (DTU) DTU->applyUpdates(Updates); } /// If this block is a `landingpad` exception handling block, categorize all /// the predecessor `invoke`s into sets, with all `invoke`s in each set /// being "mergeable" together, and then merge invokes in each set together. /// /// This is a weird mix of hoisting and sinking. Visually, it goes from: /// [...] [...] /// | | /// [invoke0] [invoke1] /// / \ / \ /// [cont0] [landingpad] [cont1] /// to: /// [...] [...] /// \ / /// [invoke] /// / \ /// [cont] [landingpad] /// /// But of course we can only do that if the invokes share the `landingpad`, /// edges invoke0->cont0 and invoke1->cont1 are "compatible", /// and the invoked functions are "compatible". static bool MergeCompatibleInvokes(BasicBlock *BB, DomTreeUpdater *DTU) { if (!EnableMergeCompatibleInvokes) return false; bool Changed = false; // FIXME: generalize to all exception handling blocks? if (!BB->isLandingPad()) return Changed; CompatibleSets Grouper; // Record all the predecessors of this `landingpad`. As per verifier, // the only allowed predecessor is the unwind edge of an `invoke`. // We want to group "compatible" `invokes` into the same set to be merged. for (BasicBlock *PredBB : predecessors(BB)) Grouper.insert(cast(PredBB->getTerminator())); // And now, merge `invoke`s that were grouped togeter. for (ArrayRef Invokes : Grouper.Sets) { if (Invokes.size() < 2) continue; Changed = true; MergeCompatibleInvokesImpl(Invokes, DTU); } return Changed; } namespace { /// Track ephemeral values, which should be ignored for cost-modelling /// purposes. Requires walking instructions in reverse order. class EphemeralValueTracker { SmallPtrSet EphValues; bool isEphemeral(const Instruction *I) { if (isa(I)) return true; return !I->mayHaveSideEffects() && !I->isTerminator() && all_of(I->users(), [&](const User *U) { return EphValues.count(cast(U)); }); } public: bool track(const Instruction *I) { if (isEphemeral(I)) { EphValues.insert(I); return true; } return false; } bool contains(const Instruction *I) const { return EphValues.contains(I); } }; } // namespace /// Determine if we can hoist sink a sole store instruction out of a /// conditional block. /// /// We are looking for code like the following: /// BrBB: /// store i32 %add, i32* %arrayidx2 /// ... // No other stores or function calls (we could be calling a memory /// ... // function). /// %cmp = icmp ult %x, %y /// br i1 %cmp, label %EndBB, label %ThenBB /// ThenBB: /// store i32 %add5, i32* %arrayidx2 /// br label EndBB /// EndBB: /// ... /// We are going to transform this into: /// BrBB: /// store i32 %add, i32* %arrayidx2 /// ... // /// %cmp = icmp ult %x, %y /// %add.add5 = select i1 %cmp, i32 %add, %add5 /// store i32 %add.add5, i32* %arrayidx2 /// ... /// /// \return The pointer to the value of the previous store if the store can be /// hoisted into the predecessor block. 0 otherwise. static Value *isSafeToSpeculateStore(Instruction *I, BasicBlock *BrBB, BasicBlock *StoreBB, BasicBlock *EndBB) { StoreInst *StoreToHoist = dyn_cast(I); if (!StoreToHoist) return nullptr; // Volatile or atomic. if (!StoreToHoist->isSimple()) return nullptr; Value *StorePtr = StoreToHoist->getPointerOperand(); Type *StoreTy = StoreToHoist->getValueOperand()->getType(); // Look for a store to the same pointer in BrBB. unsigned MaxNumInstToLookAt = 9; // Skip pseudo probe intrinsic calls which are not really killing any memory // accesses. for (Instruction &CurI : reverse(BrBB->instructionsWithoutDebug(true))) { if (!MaxNumInstToLookAt) break; --MaxNumInstToLookAt; // Could be calling an instruction that affects memory like free(). if (CurI.mayWriteToMemory() && !isa(CurI)) return nullptr; if (auto *SI = dyn_cast(&CurI)) { // Found the previous store to same location and type. Make sure it is // simple, to avoid introducing a spurious non-atomic write after an // atomic write. if (SI->getPointerOperand() == StorePtr && SI->getValueOperand()->getType() == StoreTy && SI->isSimple() && SI->getAlign() >= StoreToHoist->getAlign()) // Found the previous store, return its value operand. return SI->getValueOperand(); return nullptr; // Unknown store. } if (auto *LI = dyn_cast(&CurI)) { if (LI->getPointerOperand() == StorePtr && LI->getType() == StoreTy && LI->isSimple() && LI->getAlign() >= StoreToHoist->getAlign()) { // Local objects (created by an `alloca` instruction) are always // writable, so once we are past a read from a location it is valid to // also write to that same location. // If the address of the local object never escapes the function, that // means it's never concurrently read or written, hence moving the store // from under the condition will not introduce a data race. auto *AI = dyn_cast(getUnderlyingObject(StorePtr)); if (AI && !PointerMayBeCaptured(AI, false, true)) // Found a previous load, return it. return LI; } // The load didn't work out, but we may still find a store. } } return nullptr; } /// Estimate the cost of the insertion(s) and check that the PHI nodes can be /// converted to selects. static bool validateAndCostRequiredSelects(BasicBlock *BB, BasicBlock *ThenBB, BasicBlock *EndBB, unsigned &SpeculatedInstructions, InstructionCost &Cost, const TargetTransformInfo &TTI) { TargetTransformInfo::TargetCostKind CostKind = BB->getParent()->hasMinSize() ? TargetTransformInfo::TCK_CodeSize : TargetTransformInfo::TCK_SizeAndLatency; bool HaveRewritablePHIs = false; for (PHINode &PN : EndBB->phis()) { Value *OrigV = PN.getIncomingValueForBlock(BB); Value *ThenV = PN.getIncomingValueForBlock(ThenBB); // FIXME: Try to remove some of the duplication with // hoistCommonCodeFromSuccessors. Skip PHIs which are trivial. if (ThenV == OrigV) continue; Cost += TTI.getCmpSelInstrCost(Instruction::Select, PN.getType(), nullptr, CmpInst::BAD_ICMP_PREDICATE, CostKind); // Don't convert to selects if we could remove undefined behavior instead. if (passingValueIsAlwaysUndefined(OrigV, &PN) || passingValueIsAlwaysUndefined(ThenV, &PN)) return false; HaveRewritablePHIs = true; ConstantExpr *OrigCE = dyn_cast(OrigV); ConstantExpr *ThenCE = dyn_cast(ThenV); if (!OrigCE && !ThenCE) continue; // Known cheap (FIXME: Maybe not true for aggregates). InstructionCost OrigCost = OrigCE ? computeSpeculationCost(OrigCE, TTI) : 0; InstructionCost ThenCost = ThenCE ? computeSpeculationCost(ThenCE, TTI) : 0; InstructionCost MaxCost = 2 * PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; if (OrigCost + ThenCost > MaxCost) return false; // Account for the cost of an unfolded ConstantExpr which could end up // getting expanded into Instructions. // FIXME: This doesn't account for how many operations are combined in the // constant expression. ++SpeculatedInstructions; if (SpeculatedInstructions > 1) return false; } return HaveRewritablePHIs; } /// Speculate a conditional basic block flattening the CFG. /// /// Note that this is a very risky transform currently. Speculating /// instructions like this is most often not desirable. Instead, there is an MI /// pass which can do it with full awareness of the resource constraints. /// However, some cases are "obvious" and we should do directly. An example of /// this is speculating a single, reasonably cheap instruction. /// /// There is only one distinct advantage to flattening the CFG at the IR level: /// it makes very common but simplistic optimizations such as are common in /// instcombine and the DAG combiner more powerful by removing CFG edges and /// modeling their effects with easier to reason about SSA value graphs. /// /// /// An illustration of this transform is turning this IR: /// \code /// BB: /// %cmp = icmp ult %x, %y /// br i1 %cmp, label %EndBB, label %ThenBB /// ThenBB: /// %sub = sub %x, %y /// br label BB2 /// EndBB: /// %phi = phi [ %sub, %ThenBB ], [ 0, %EndBB ] /// ... /// \endcode /// /// Into this IR: /// \code /// BB: /// %cmp = icmp ult %x, %y /// %sub = sub %x, %y /// %cond = select i1 %cmp, 0, %sub /// ... /// \endcode /// /// \returns true if the conditional block is removed. bool SimplifyCFGOpt::SpeculativelyExecuteBB(BranchInst *BI, BasicBlock *ThenBB) { if (!Options.SpeculateBlocks) return false; // Be conservative for now. FP select instruction can often be expensive. Value *BrCond = BI->getCondition(); if (isa(BrCond)) return false; BasicBlock *BB = BI->getParent(); BasicBlock *EndBB = ThenBB->getTerminator()->getSuccessor(0); InstructionCost Budget = PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; // If ThenBB is actually on the false edge of the conditional branch, remember // to swap the select operands later. bool Invert = false; if (ThenBB != BI->getSuccessor(0)) { assert(ThenBB == BI->getSuccessor(1) && "No edge from 'if' block?"); Invert = true; } assert(EndBB == BI->getSuccessor(!Invert) && "No edge from to end block"); // If the branch is non-unpredictable, and is predicted to *not* branch to // the `then` block, then avoid speculating it. if (!BI->getMetadata(LLVMContext::MD_unpredictable)) { uint64_t TWeight, FWeight; if (extractBranchWeights(*BI, TWeight, FWeight) && (TWeight + FWeight) != 0) { uint64_t EndWeight = Invert ? TWeight : FWeight; BranchProbability BIEndProb = BranchProbability::getBranchProbability(EndWeight, TWeight + FWeight); BranchProbability Likely = TTI.getPredictableBranchThreshold(); if (BIEndProb >= Likely) return false; } } // Keep a count of how many times instructions are used within ThenBB when // they are candidates for sinking into ThenBB. Specifically: // - They are defined in BB, and // - They have no side effects, and // - All of their uses are in ThenBB. SmallDenseMap SinkCandidateUseCounts; SmallVector SpeculatedDbgIntrinsics; unsigned SpeculatedInstructions = 0; Value *SpeculatedStoreValue = nullptr; StoreInst *SpeculatedStore = nullptr; EphemeralValueTracker EphTracker; for (Instruction &I : reverse(drop_end(*ThenBB))) { // Skip debug info. if (isa(I)) { SpeculatedDbgIntrinsics.push_back(&I); continue; } // Skip pseudo probes. The consequence is we lose track of the branch // probability for ThenBB, which is fine since the optimization here takes // place regardless of the branch probability. if (isa(I)) { // The probe should be deleted so that it will not be over-counted when // the samples collected on the non-conditional path are counted towards // the conditional path. We leave it for the counts inference algorithm to // figure out a proper count for an unknown probe. SpeculatedDbgIntrinsics.push_back(&I); continue; } // Ignore ephemeral values, they will be dropped by the transform. if (EphTracker.track(&I)) continue; // Only speculatively execute a single instruction (not counting the // terminator) for now. ++SpeculatedInstructions; if (SpeculatedInstructions > 1) return false; // Don't hoist the instruction if it's unsafe or expensive. if (!isSafeToSpeculativelyExecute(&I) && !(HoistCondStores && (SpeculatedStoreValue = isSafeToSpeculateStore( &I, BB, ThenBB, EndBB)))) return false; if (!SpeculatedStoreValue && computeSpeculationCost(&I, TTI) > PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic) return false; // Store the store speculation candidate. if (SpeculatedStoreValue) SpeculatedStore = cast(&I); // Do not hoist the instruction if any of its operands are defined but not // used in BB. The transformation will prevent the operand from // being sunk into the use block. for (Use &Op : I.operands()) { Instruction *OpI = dyn_cast(Op); if (!OpI || OpI->getParent() != BB || OpI->mayHaveSideEffects()) continue; // Not a candidate for sinking. ++SinkCandidateUseCounts[OpI]; } } // Consider any sink candidates which are only used in ThenBB as costs for // speculation. Note, while we iterate over a DenseMap here, we are summing // and so iteration order isn't significant. for (const auto &[Inst, Count] : SinkCandidateUseCounts) if (Inst->hasNUses(Count)) { ++SpeculatedInstructions; if (SpeculatedInstructions > 1) return false; } // Check that we can insert the selects and that it's not too expensive to do // so. bool Convert = SpeculatedStore != nullptr; InstructionCost Cost = 0; Convert |= validateAndCostRequiredSelects(BB, ThenBB, EndBB, SpeculatedInstructions, Cost, TTI); if (!Convert || Cost > Budget) return false; // If we get here, we can hoist the instruction and if-convert. LLVM_DEBUG(dbgs() << "SPECULATIVELY EXECUTING BB" << *ThenBB << "\n";); // Insert a select of the value of the speculated store. if (SpeculatedStoreValue) { IRBuilder Builder(BI); Value *OrigV = SpeculatedStore->getValueOperand(); Value *TrueV = SpeculatedStore->getValueOperand(); Value *FalseV = SpeculatedStoreValue; if (Invert) std::swap(TrueV, FalseV); Value *S = Builder.CreateSelect( BrCond, TrueV, FalseV, "spec.store.select", BI); SpeculatedStore->setOperand(0, S); SpeculatedStore->applyMergedLocation(BI->getDebugLoc(), SpeculatedStore->getDebugLoc()); // The value stored is still conditional, but the store itself is now // unconditonally executed, so we must be sure that any linked dbg.assign // intrinsics are tracking the new stored value (the result of the // select). If we don't, and the store were to be removed by another pass // (e.g. DSE), then we'd eventually end up emitting a location describing // the conditional value, unconditionally. // // === Before this transformation === // pred: // store %one, %x.dest, !DIAssignID !1 // dbg.assign %one, "x", ..., !1, ... // br %cond if.then // // if.then: // store %two, %x.dest, !DIAssignID !2 // dbg.assign %two, "x", ..., !2, ... // // === After this transformation === // pred: // store %one, %x.dest, !DIAssignID !1 // dbg.assign %one, "x", ..., !1 /// ... // %merge = select %cond, %two, %one // store %merge, %x.dest, !DIAssignID !2 // dbg.assign %merge, "x", ..., !2 auto replaceVariable = [OrigV, S](auto *DbgAssign) { if (llvm::is_contained(DbgAssign->location_ops(), OrigV)) DbgAssign->replaceVariableLocationOp(OrigV, S); }; for_each(at::getAssignmentMarkers(SpeculatedStore), replaceVariable); for_each(at::getDVRAssignmentMarkers(SpeculatedStore), replaceVariable); } // Metadata can be dependent on the condition we are hoisting above. // Strip all UB-implying metadata on the instruction. Drop the debug loc // to avoid making it appear as if the condition is a constant, which would // be misleading while debugging. // Similarly strip attributes that maybe dependent on condition we are // hoisting above. for (auto &I : make_early_inc_range(*ThenBB)) { if (!SpeculatedStoreValue || &I != SpeculatedStore) { // Don't update the DILocation of dbg.assign intrinsics. if (!isa(&I)) I.setDebugLoc(DebugLoc()); } I.dropUBImplyingAttrsAndMetadata(); // Drop ephemeral values. if (EphTracker.contains(&I)) { I.replaceAllUsesWith(PoisonValue::get(I.getType())); I.eraseFromParent(); } } // Hoist the instructions. // In "RemoveDIs" non-instr debug-info mode, drop DbgVariableRecords attached // to these instructions, in the same way that dbg.value intrinsics are // dropped at the end of this block. for (auto &It : make_range(ThenBB->begin(), ThenBB->end())) for (DbgRecord &DR : make_early_inc_range(It.getDbgRecordRange())) // Drop all records except assign-kind DbgVariableRecords (dbg.assign // equivalent). if (DbgVariableRecord *DVR = dyn_cast(&DR); !DVR || !DVR->isDbgAssign()) It.dropOneDbgRecord(&DR); BB->splice(BI->getIterator(), ThenBB, ThenBB->begin(), std::prev(ThenBB->end())); // Insert selects and rewrite the PHI operands. IRBuilder Builder(BI); for (PHINode &PN : EndBB->phis()) { unsigned OrigI = PN.getBasicBlockIndex(BB); unsigned ThenI = PN.getBasicBlockIndex(ThenBB); Value *OrigV = PN.getIncomingValue(OrigI); Value *ThenV = PN.getIncomingValue(ThenI); // Skip PHIs which are trivial. if (OrigV == ThenV) continue; // Create a select whose true value is the speculatively executed value and // false value is the pre-existing value. Swap them if the branch // destinations were inverted. Value *TrueV = ThenV, *FalseV = OrigV; if (Invert) std::swap(TrueV, FalseV); Value *V = Builder.CreateSelect(BrCond, TrueV, FalseV, "spec.select", BI); PN.setIncomingValue(OrigI, V); PN.setIncomingValue(ThenI, V); } // Remove speculated dbg intrinsics. // FIXME: Is it possible to do this in a more elegant way? Moving/merging the // dbg value for the different flows and inserting it after the select. for (Instruction *I : SpeculatedDbgIntrinsics) { // We still want to know that an assignment took place so don't remove // dbg.assign intrinsics. if (!isa(I)) I->eraseFromParent(); } ++NumSpeculations; return true; } /// Return true if we can thread a branch across this block. static bool BlockIsSimpleEnoughToThreadThrough(BasicBlock *BB) { int Size = 0; EphemeralValueTracker EphTracker; // Walk the loop in reverse so that we can identify ephemeral values properly // (values only feeding assumes). for (Instruction &I : reverse(BB->instructionsWithoutDebug(false))) { // Can't fold blocks that contain noduplicate or convergent calls. if (CallInst *CI = dyn_cast(&I)) if (CI->cannotDuplicate() || CI->isConvergent()) return false; // Ignore ephemeral values which are deleted during codegen. // We will delete Phis while threading, so Phis should not be accounted in // block's size. if (!EphTracker.track(&I) && !isa(I)) { if (Size++ > MaxSmallBlockSize) return false; // Don't clone large BB's. } // We can only support instructions that do not define values that are // live outside of the current basic block. for (User *U : I.users()) { Instruction *UI = cast(U); if (UI->getParent() != BB || isa(UI)) return false; } // Looks ok, continue checking. } return true; } static ConstantInt *getKnownValueOnEdge(Value *V, BasicBlock *From, BasicBlock *To) { // Don't look past the block defining the value, we might get the value from // a previous loop iteration. auto *I = dyn_cast(V); if (I && I->getParent() == To) return nullptr; // We know the value if the From block branches on it. auto *BI = dyn_cast(From->getTerminator()); if (BI && BI->isConditional() && BI->getCondition() == V && BI->getSuccessor(0) != BI->getSuccessor(1)) return BI->getSuccessor(0) == To ? ConstantInt::getTrue(BI->getContext()) : ConstantInt::getFalse(BI->getContext()); return nullptr; } /// If we have a conditional branch on something for which we know the constant /// value in predecessors (e.g. a phi node in the current block), thread edges /// from the predecessor to their ultimate destination. static std::optional FoldCondBranchOnValueKnownInPredecessorImpl(BranchInst *BI, DomTreeUpdater *DTU, const DataLayout &DL, AssumptionCache *AC) { SmallMapVector, 2> KnownValues; BasicBlock *BB = BI->getParent(); Value *Cond = BI->getCondition(); PHINode *PN = dyn_cast(Cond); if (PN && PN->getParent() == BB) { // Degenerate case of a single entry PHI. if (PN->getNumIncomingValues() == 1) { FoldSingleEntryPHINodes(PN->getParent()); return true; } for (Use &U : PN->incoming_values()) if (auto *CB = dyn_cast(U)) KnownValues[CB].insert(PN->getIncomingBlock(U)); } else { for (BasicBlock *Pred : predecessors(BB)) { if (ConstantInt *CB = getKnownValueOnEdge(Cond, Pred, BB)) KnownValues[CB].insert(Pred); } } if (KnownValues.empty()) return false; // Now we know that this block has multiple preds and two succs. // Check that the block is small enough and values defined in the block are // not used outside of it. if (!BlockIsSimpleEnoughToThreadThrough(BB)) return false; for (const auto &Pair : KnownValues) { // Okay, we now know that all edges from PredBB should be revectored to // branch to RealDest. ConstantInt *CB = Pair.first; ArrayRef PredBBs = Pair.second.getArrayRef(); BasicBlock *RealDest = BI->getSuccessor(!CB->getZExtValue()); if (RealDest == BB) continue; // Skip self loops. // Skip if the predecessor's terminator is an indirect branch. if (any_of(PredBBs, [](BasicBlock *PredBB) { return isa(PredBB->getTerminator()); })) continue; LLVM_DEBUG({ dbgs() << "Condition " << *Cond << " in " << BB->getName() << " has value " << *Pair.first << " in predecessors:\n"; for (const BasicBlock *PredBB : Pair.second) dbgs() << " " << PredBB->getName() << "\n"; dbgs() << "Threading to destination " << RealDest->getName() << ".\n"; }); // Split the predecessors we are threading into a new edge block. We'll // clone the instructions into this block, and then redirect it to RealDest. BasicBlock *EdgeBB = SplitBlockPredecessors(BB, PredBBs, ".critedge", DTU); // TODO: These just exist to reduce test diff, we can drop them if we like. EdgeBB->setName(RealDest->getName() + ".critedge"); EdgeBB->moveBefore(RealDest); // Update PHI nodes. AddPredecessorToBlock(RealDest, EdgeBB, BB); // BB may have instructions that are being threaded over. Clone these // instructions into EdgeBB. We know that there will be no uses of the // cloned instructions outside of EdgeBB. BasicBlock::iterator InsertPt = EdgeBB->getFirstInsertionPt(); DenseMap TranslateMap; // Track translated values. TranslateMap[Cond] = CB; // RemoveDIs: track instructions that we optimise away while folding, so // that we can copy DbgVariableRecords from them later. BasicBlock::iterator SrcDbgCursor = BB->begin(); for (BasicBlock::iterator BBI = BB->begin(); &*BBI != BI; ++BBI) { if (PHINode *PN = dyn_cast(BBI)) { TranslateMap[PN] = PN->getIncomingValueForBlock(EdgeBB); continue; } // Clone the instruction. Instruction *N = BBI->clone(); // Insert the new instruction into its new home. N->insertInto(EdgeBB, InsertPt); if (BBI->hasName()) N->setName(BBI->getName() + ".c"); // Update operands due to translation. for (Use &Op : N->operands()) { DenseMap::iterator PI = TranslateMap.find(Op); if (PI != TranslateMap.end()) Op = PI->second; } // Check for trivial simplification. if (Value *V = simplifyInstruction(N, {DL, nullptr, nullptr, AC})) { if (!BBI->use_empty()) TranslateMap[&*BBI] = V; if (!N->mayHaveSideEffects()) { N->eraseFromParent(); // Instruction folded away, don't need actual // inst N = nullptr; } } else { if (!BBI->use_empty()) TranslateMap[&*BBI] = N; } if (N) { // Copy all debug-info attached to instructions from the last we // successfully clone, up to this instruction (they might have been // folded away). for (; SrcDbgCursor != BBI; ++SrcDbgCursor) N->cloneDebugInfoFrom(&*SrcDbgCursor); SrcDbgCursor = std::next(BBI); // Clone debug-info on this instruction too. N->cloneDebugInfoFrom(&*BBI); // Register the new instruction with the assumption cache if necessary. if (auto *Assume = dyn_cast(N)) if (AC) AC->registerAssumption(Assume); } } for (; &*SrcDbgCursor != BI; ++SrcDbgCursor) InsertPt->cloneDebugInfoFrom(&*SrcDbgCursor); InsertPt->cloneDebugInfoFrom(BI); BB->removePredecessor(EdgeBB); BranchInst *EdgeBI = cast(EdgeBB->getTerminator()); EdgeBI->setSuccessor(0, RealDest); EdgeBI->setDebugLoc(BI->getDebugLoc()); if (DTU) { SmallVector Updates; Updates.push_back({DominatorTree::Delete, EdgeBB, BB}); Updates.push_back({DominatorTree::Insert, EdgeBB, RealDest}); DTU->applyUpdates(Updates); } // For simplicity, we created a separate basic block for the edge. Merge // it back into the predecessor if possible. This not only avoids // unnecessary SimplifyCFG iterations, but also makes sure that we don't // bypass the check for trivial cycles above. MergeBlockIntoPredecessor(EdgeBB, DTU); // Signal repeat, simplifying any other constants. return std::nullopt; } return false; } static bool FoldCondBranchOnValueKnownInPredecessor(BranchInst *BI, DomTreeUpdater *DTU, const DataLayout &DL, AssumptionCache *AC) { std::optional Result; bool EverChanged = false; do { // Note that None means "we changed things, but recurse further." Result = FoldCondBranchOnValueKnownInPredecessorImpl(BI, DTU, DL, AC); EverChanged |= Result == std::nullopt || *Result; } while (Result == std::nullopt); return EverChanged; } /// Given a BB that starts with the specified two-entry PHI node, /// see if we can eliminate it. static bool FoldTwoEntryPHINode(PHINode *PN, const TargetTransformInfo &TTI, DomTreeUpdater *DTU, const DataLayout &DL, bool SpeculateUnpredictables) { // Ok, this is a two entry PHI node. Check to see if this is a simple "if // statement", which has a very simple dominance structure. Basically, we // are trying to find the condition that is being branched on, which // subsequently causes this merge to happen. We really want control // dependence information for this check, but simplifycfg can't keep it up // to date, and this catches most of the cases we care about anyway. BasicBlock *BB = PN->getParent(); BasicBlock *IfTrue, *IfFalse; BranchInst *DomBI = GetIfCondition(BB, IfTrue, IfFalse); if (!DomBI) return false; Value *IfCond = DomBI->getCondition(); // Don't bother if the branch will be constant folded trivially. if (isa(IfCond)) return false; BasicBlock *DomBlock = DomBI->getParent(); SmallVector IfBlocks; llvm::copy_if( PN->blocks(), std::back_inserter(IfBlocks), [](BasicBlock *IfBlock) { return cast(IfBlock->getTerminator())->isUnconditional(); }); assert((IfBlocks.size() == 1 || IfBlocks.size() == 2) && "Will have either one or two blocks to speculate."); // If the branch is non-unpredictable, see if we either predictably jump to // the merge bb (if we have only a single 'then' block), or if we predictably // jump to one specific 'then' block (if we have two of them). // It isn't beneficial to speculatively execute the code // from the block that we know is predictably not entered. bool IsUnpredictable = DomBI->getMetadata(LLVMContext::MD_unpredictable); if (!IsUnpredictable) { uint64_t TWeight, FWeight; if (extractBranchWeights(*DomBI, TWeight, FWeight) && (TWeight + FWeight) != 0) { BranchProbability BITrueProb = BranchProbability::getBranchProbability(TWeight, TWeight + FWeight); BranchProbability Likely = TTI.getPredictableBranchThreshold(); BranchProbability BIFalseProb = BITrueProb.getCompl(); if (IfBlocks.size() == 1) { BranchProbability BIBBProb = DomBI->getSuccessor(0) == BB ? BITrueProb : BIFalseProb; if (BIBBProb >= Likely) return false; } else { if (BITrueProb >= Likely || BIFalseProb >= Likely) return false; } } } // Don't try to fold an unreachable block. For example, the phi node itself // can't be the candidate if-condition for a select that we want to form. if (auto *IfCondPhiInst = dyn_cast(IfCond)) if (IfCondPhiInst->getParent() == BB) return false; // Okay, we found that we can merge this two-entry phi node into a select. // Doing so would require us to fold *all* two entry phi nodes in this block. // At some point this becomes non-profitable (particularly if the target // doesn't support cmov's). Only do this transformation if there are two or // fewer PHI nodes in this block. unsigned NumPhis = 0; for (BasicBlock::iterator I = BB->begin(); isa(I); ++NumPhis, ++I) if (NumPhis > 2) return false; // Loop over the PHI's seeing if we can promote them all to select // instructions. While we are at it, keep track of the instructions // that need to be moved to the dominating block. SmallPtrSet AggressiveInsts; InstructionCost Cost = 0; InstructionCost Budget = TwoEntryPHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; if (SpeculateUnpredictables && IsUnpredictable) Budget += TTI.getBranchMispredictPenalty(); bool Changed = false; for (BasicBlock::iterator II = BB->begin(); isa(II);) { PHINode *PN = cast(II++); if (Value *V = simplifyInstruction(PN, {DL, PN})) { PN->replaceAllUsesWith(V); PN->eraseFromParent(); Changed = true; continue; } if (!dominatesMergePoint(PN->getIncomingValue(0), BB, AggressiveInsts, Cost, Budget, TTI) || !dominatesMergePoint(PN->getIncomingValue(1), BB, AggressiveInsts, Cost, Budget, TTI)) return Changed; } // If we folded the first phi, PN dangles at this point. Refresh it. If // we ran out of PHIs then we simplified them all. PN = dyn_cast(BB->begin()); if (!PN) return true; // Return true if at least one of these is a 'not', and another is either // a 'not' too, or a constant. auto CanHoistNotFromBothValues = [](Value *V0, Value *V1) { if (!match(V0, m_Not(m_Value()))) std::swap(V0, V1); auto Invertible = m_CombineOr(m_Not(m_Value()), m_AnyIntegralConstant()); return match(V0, m_Not(m_Value())) && match(V1, Invertible); }; // Don't fold i1 branches on PHIs which contain binary operators or // (possibly inverted) select form of or/ands, unless one of // the incoming values is an 'not' and another one is freely invertible. // These can often be turned into switches and other things. auto IsBinOpOrAnd = [](Value *V) { return match( V, m_CombineOr( m_BinOp(), m_CombineOr(m_Select(m_Value(), m_ImmConstant(), m_Value()), m_Select(m_Value(), m_Value(), m_ImmConstant())))); }; if (PN->getType()->isIntegerTy(1) && (IsBinOpOrAnd(PN->getIncomingValue(0)) || IsBinOpOrAnd(PN->getIncomingValue(1)) || IsBinOpOrAnd(IfCond)) && !CanHoistNotFromBothValues(PN->getIncomingValue(0), PN->getIncomingValue(1))) return Changed; // If all PHI nodes are promotable, check to make sure that all instructions // in the predecessor blocks can be promoted as well. If not, we won't be able // to get rid of the control flow, so it's not worth promoting to select // instructions. for (BasicBlock *IfBlock : IfBlocks) for (BasicBlock::iterator I = IfBlock->begin(); !I->isTerminator(); ++I) if (!AggressiveInsts.count(&*I) && !I->isDebugOrPseudoInst()) { // This is not an aggressive instruction that we can promote. // Because of this, we won't be able to get rid of the control flow, so // the xform is not worth it. return Changed; } // If either of the blocks has it's address taken, we can't do this fold. if (any_of(IfBlocks, [](BasicBlock *IfBlock) { return IfBlock->hasAddressTaken(); })) return Changed; LLVM_DEBUG(dbgs() << "FOUND IF CONDITION! " << *IfCond; if (IsUnpredictable) dbgs() << " (unpredictable)"; dbgs() << " T: " << IfTrue->getName() << " F: " << IfFalse->getName() << "\n"); // If we can still promote the PHI nodes after this gauntlet of tests, // do all of the PHI's now. // Move all 'aggressive' instructions, which are defined in the // conditional parts of the if's up to the dominating block. for (BasicBlock *IfBlock : IfBlocks) hoistAllInstructionsInto(DomBlock, DomBI, IfBlock); IRBuilder Builder(DomBI); // Propagate fast-math-flags from phi nodes to replacement selects. IRBuilder<>::FastMathFlagGuard FMFGuard(Builder); while (PHINode *PN = dyn_cast(BB->begin())) { if (isa(PN)) Builder.setFastMathFlags(PN->getFastMathFlags()); // Change the PHI node into a select instruction. Value *TrueVal = PN->getIncomingValueForBlock(IfTrue); Value *FalseVal = PN->getIncomingValueForBlock(IfFalse); Value *Sel = Builder.CreateSelect(IfCond, TrueVal, FalseVal, "", DomBI); PN->replaceAllUsesWith(Sel); Sel->takeName(PN); PN->eraseFromParent(); } // At this point, all IfBlocks are empty, so our if statement // has been flattened. Change DomBlock to jump directly to our new block to // avoid other simplifycfg's kicking in on the diamond. Builder.CreateBr(BB); SmallVector Updates; if (DTU) { Updates.push_back({DominatorTree::Insert, DomBlock, BB}); for (auto *Successor : successors(DomBlock)) Updates.push_back({DominatorTree::Delete, DomBlock, Successor}); } DomBI->eraseFromParent(); if (DTU) DTU->applyUpdates(Updates); return true; } static Value *createLogicalOp(IRBuilderBase &Builder, Instruction::BinaryOps Opc, Value *LHS, Value *RHS, const Twine &Name = "") { // Try to relax logical op to binary op. if (impliesPoison(RHS, LHS)) return Builder.CreateBinOp(Opc, LHS, RHS, Name); if (Opc == Instruction::And) return Builder.CreateLogicalAnd(LHS, RHS, Name); if (Opc == Instruction::Or) return Builder.CreateLogicalOr(LHS, RHS, Name); llvm_unreachable("Invalid logical opcode"); } /// Return true if either PBI or BI has branch weight available, and store /// the weights in {Pred|Succ}{True|False}Weight. If one of PBI and BI does /// not have branch weight, use 1:1 as its weight. static bool extractPredSuccWeights(BranchInst *PBI, BranchInst *BI, uint64_t &PredTrueWeight, uint64_t &PredFalseWeight, uint64_t &SuccTrueWeight, uint64_t &SuccFalseWeight) { bool PredHasWeights = extractBranchWeights(*PBI, PredTrueWeight, PredFalseWeight); bool SuccHasWeights = extractBranchWeights(*BI, SuccTrueWeight, SuccFalseWeight); if (PredHasWeights || SuccHasWeights) { if (!PredHasWeights) PredTrueWeight = PredFalseWeight = 1; if (!SuccHasWeights) SuccTrueWeight = SuccFalseWeight = 1; return true; } else { return false; } } /// Determine if the two branches share a common destination and deduce a glue /// that joins the branches' conditions to arrive at the common destination if /// that would be profitable. static std::optional> shouldFoldCondBranchesToCommonDestination(BranchInst *BI, BranchInst *PBI, const TargetTransformInfo *TTI) { assert(BI && PBI && BI->isConditional() && PBI->isConditional() && "Both blocks must end with a conditional branches."); assert(is_contained(predecessors(BI->getParent()), PBI->getParent()) && "PredBB must be a predecessor of BB."); // We have the potential to fold the conditions together, but if the // predecessor branch is predictable, we may not want to merge them. uint64_t PTWeight, PFWeight; BranchProbability PBITrueProb, Likely; if (TTI && !PBI->getMetadata(LLVMContext::MD_unpredictable) && extractBranchWeights(*PBI, PTWeight, PFWeight) && (PTWeight + PFWeight) != 0) { PBITrueProb = BranchProbability::getBranchProbability(PTWeight, PTWeight + PFWeight); Likely = TTI->getPredictableBranchThreshold(); } if (PBI->getSuccessor(0) == BI->getSuccessor(0)) { // Speculate the 2nd condition unless the 1st is probably true. if (PBITrueProb.isUnknown() || PBITrueProb < Likely) return {{BI->getSuccessor(0), Instruction::Or, false}}; } else if (PBI->getSuccessor(1) == BI->getSuccessor(1)) { // Speculate the 2nd condition unless the 1st is probably false. if (PBITrueProb.isUnknown() || PBITrueProb.getCompl() < Likely) return {{BI->getSuccessor(1), Instruction::And, false}}; } else if (PBI->getSuccessor(0) == BI->getSuccessor(1)) { // Speculate the 2nd condition unless the 1st is probably true. if (PBITrueProb.isUnknown() || PBITrueProb < Likely) return {{BI->getSuccessor(1), Instruction::And, true}}; } else if (PBI->getSuccessor(1) == BI->getSuccessor(0)) { // Speculate the 2nd condition unless the 1st is probably false. if (PBITrueProb.isUnknown() || PBITrueProb.getCompl() < Likely) return {{BI->getSuccessor(0), Instruction::Or, true}}; } return std::nullopt; } static bool performBranchToCommonDestFolding(BranchInst *BI, BranchInst *PBI, DomTreeUpdater *DTU, MemorySSAUpdater *MSSAU, const TargetTransformInfo *TTI) { BasicBlock *BB = BI->getParent(); BasicBlock *PredBlock = PBI->getParent(); // Determine if the two branches share a common destination. BasicBlock *CommonSucc; Instruction::BinaryOps Opc; bool InvertPredCond; std::tie(CommonSucc, Opc, InvertPredCond) = *shouldFoldCondBranchesToCommonDestination(BI, PBI, TTI); LLVM_DEBUG(dbgs() << "FOLDING BRANCH TO COMMON DEST:\n" << *PBI << *BB); IRBuilder<> Builder(PBI); // The builder is used to create instructions to eliminate the branch in BB. // If BB's terminator has !annotation metadata, add it to the new // instructions. Builder.CollectMetadataToCopy(BB->getTerminator(), {LLVMContext::MD_annotation}); // If we need to invert the condition in the pred block to match, do so now. if (InvertPredCond) { InvertBranch(PBI, Builder); } BasicBlock *UniqueSucc = PBI->getSuccessor(0) == BB ? BI->getSuccessor(0) : BI->getSuccessor(1); // Before cloning instructions, notify the successor basic block that it // is about to have a new predecessor. This will update PHI nodes, // which will allow us to update live-out uses of bonus instructions. AddPredecessorToBlock(UniqueSucc, PredBlock, BB, MSSAU); // Try to update branch weights. uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight; if (extractPredSuccWeights(PBI, BI, PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight)) { SmallVector NewWeights; if (PBI->getSuccessor(0) == BB) { // PBI: br i1 %x, BB, FalseDest // BI: br i1 %y, UniqueSucc, FalseDest // TrueWeight is TrueWeight for PBI * TrueWeight for BI. NewWeights.push_back(PredTrueWeight * SuccTrueWeight); // FalseWeight is FalseWeight for PBI * TotalWeight for BI + // TrueWeight for PBI * FalseWeight for BI. // We assume that total weights of a BranchInst can fit into 32 bits. // Therefore, we will not have overflow using 64-bit arithmetic. NewWeights.push_back(PredFalseWeight * (SuccFalseWeight + SuccTrueWeight) + PredTrueWeight * SuccFalseWeight); } else { // PBI: br i1 %x, TrueDest, BB // BI: br i1 %y, TrueDest, UniqueSucc // TrueWeight is TrueWeight for PBI * TotalWeight for BI + // FalseWeight for PBI * TrueWeight for BI. NewWeights.push_back(PredTrueWeight * (SuccFalseWeight + SuccTrueWeight) + PredFalseWeight * SuccTrueWeight); // FalseWeight is FalseWeight for PBI * FalseWeight for BI. NewWeights.push_back(PredFalseWeight * SuccFalseWeight); } // Halve the weights if any of them cannot fit in an uint32_t FitWeights(NewWeights); SmallVector MDWeights(NewWeights.begin(), NewWeights.end()); setBranchWeights(PBI, MDWeights[0], MDWeights[1], /*IsExpected=*/false); // TODO: If BB is reachable from all paths through PredBlock, then we // could replace PBI's branch probabilities with BI's. } else PBI->setMetadata(LLVMContext::MD_prof, nullptr); // Now, update the CFG. PBI->setSuccessor(PBI->getSuccessor(0) != BB, UniqueSucc); if (DTU) DTU->applyUpdates({{DominatorTree::Insert, PredBlock, UniqueSucc}, {DominatorTree::Delete, PredBlock, BB}}); // If BI was a loop latch, it may have had associated loop metadata. // We need to copy it to the new latch, that is, PBI. if (MDNode *LoopMD = BI->getMetadata(LLVMContext::MD_loop)) PBI->setMetadata(LLVMContext::MD_loop, LoopMD); ValueToValueMapTy VMap; // maps original values to cloned values CloneInstructionsIntoPredecessorBlockAndUpdateSSAUses(BB, PredBlock, VMap); Module *M = BB->getModule(); if (PredBlock->IsNewDbgInfoFormat) { PredBlock->getTerminator()->cloneDebugInfoFrom(BB->getTerminator()); for (DbgVariableRecord &DVR : filterDbgVars(PredBlock->getTerminator()->getDbgRecordRange())) { RemapDbgRecord(M, &DVR, VMap, RF_NoModuleLevelChanges | RF_IgnoreMissingLocals); } } // Now that the Cond was cloned into the predecessor basic block, // or/and the two conditions together. Value *BICond = VMap[BI->getCondition()]; PBI->setCondition( createLogicalOp(Builder, Opc, PBI->getCondition(), BICond, "or.cond")); ++NumFoldBranchToCommonDest; return true; } /// Return if an instruction's type or any of its operands' types are a vector /// type. static bool isVectorOp(Instruction &I) { return I.getType()->isVectorTy() || any_of(I.operands(), [](Use &U) { return U->getType()->isVectorTy(); }); } /// If this basic block is simple enough, and if a predecessor branches to us /// and one of our successors, fold the block into the predecessor and use /// logical operations to pick the right destination. bool llvm::FoldBranchToCommonDest(BranchInst *BI, DomTreeUpdater *DTU, MemorySSAUpdater *MSSAU, const TargetTransformInfo *TTI, unsigned BonusInstThreshold) { // If this block ends with an unconditional branch, // let SpeculativelyExecuteBB() deal with it. if (!BI->isConditional()) return false; BasicBlock *BB = BI->getParent(); TargetTransformInfo::TargetCostKind CostKind = BB->getParent()->hasMinSize() ? TargetTransformInfo::TCK_CodeSize : TargetTransformInfo::TCK_SizeAndLatency; Instruction *Cond = dyn_cast(BI->getCondition()); if (!Cond || (!isa(Cond) && !isa(Cond) && !isa(Cond)) || Cond->getParent() != BB || !Cond->hasOneUse()) return false; // Finally, don't infinitely unroll conditional loops. if (is_contained(successors(BB), BB)) return false; // With which predecessors will we want to deal with? SmallVector Preds; for (BasicBlock *PredBlock : predecessors(BB)) { BranchInst *PBI = dyn_cast(PredBlock->getTerminator()); // Check that we have two conditional branches. If there is a PHI node in // the common successor, verify that the same value flows in from both // blocks. if (!PBI || PBI->isUnconditional() || !SafeToMergeTerminators(BI, PBI)) continue; // Determine if the two branches share a common destination. BasicBlock *CommonSucc; Instruction::BinaryOps Opc; bool InvertPredCond; if (auto Recipe = shouldFoldCondBranchesToCommonDestination(BI, PBI, TTI)) std::tie(CommonSucc, Opc, InvertPredCond) = *Recipe; else continue; // Check the cost of inserting the necessary logic before performing the // transformation. if (TTI) { Type *Ty = BI->getCondition()->getType(); InstructionCost Cost = TTI->getArithmeticInstrCost(Opc, Ty, CostKind); if (InvertPredCond && (!PBI->getCondition()->hasOneUse() || !isa(PBI->getCondition()))) Cost += TTI->getArithmeticInstrCost(Instruction::Xor, Ty, CostKind); if (Cost > BranchFoldThreshold) continue; } // Ok, we do want to deal with this predecessor. Record it. Preds.emplace_back(PredBlock); } // If there aren't any predecessors into which we can fold, // don't bother checking the cost. if (Preds.empty()) return false; // Only allow this transformation if computing the condition doesn't involve // too many instructions and these involved instructions can be executed // unconditionally. We denote all involved instructions except the condition // as "bonus instructions", and only allow this transformation when the // number of the bonus instructions we'll need to create when cloning into // each predecessor does not exceed a certain threshold. unsigned NumBonusInsts = 0; bool SawVectorOp = false; const unsigned PredCount = Preds.size(); for (Instruction &I : *BB) { // Don't check the branch condition comparison itself. if (&I == Cond) continue; // Ignore dbg intrinsics, and the terminator. if (isa(I) || isa(I)) continue; // I must be safe to execute unconditionally. if (!isSafeToSpeculativelyExecute(&I)) return false; SawVectorOp |= isVectorOp(I); // Account for the cost of duplicating this instruction into each // predecessor. Ignore free instructions. if (!TTI || TTI->getInstructionCost(&I, CostKind) != TargetTransformInfo::TCC_Free) { NumBonusInsts += PredCount; // Early exits once we reach the limit. if (NumBonusInsts > BonusInstThreshold * BranchFoldToCommonDestVectorMultiplier) return false; } auto IsBCSSAUse = [BB, &I](Use &U) { auto *UI = cast(U.getUser()); if (auto *PN = dyn_cast(UI)) return PN->getIncomingBlock(U) == BB; return UI->getParent() == BB && I.comesBefore(UI); }; // Does this instruction require rewriting of uses? if (!all_of(I.uses(), IsBCSSAUse)) return false; } if (NumBonusInsts > BonusInstThreshold * (SawVectorOp ? BranchFoldToCommonDestVectorMultiplier : 1)) return false; // Ok, we have the budget. Perform the transformation. for (BasicBlock *PredBlock : Preds) { auto *PBI = cast(PredBlock->getTerminator()); return performBranchToCommonDestFolding(BI, PBI, DTU, MSSAU, TTI); } return false; } // If there is only one store in BB1 and BB2, return it, otherwise return // nullptr. static StoreInst *findUniqueStoreInBlocks(BasicBlock *BB1, BasicBlock *BB2) { StoreInst *S = nullptr; for (auto *BB : {BB1, BB2}) { if (!BB) continue; for (auto &I : *BB) if (auto *SI = dyn_cast(&I)) { if (S) // Multiple stores seen. return nullptr; else S = SI; } } return S; } static Value *ensureValueAvailableInSuccessor(Value *V, BasicBlock *BB, Value *AlternativeV = nullptr) { // PHI is going to be a PHI node that allows the value V that is defined in // BB to be referenced in BB's only successor. // // If AlternativeV is nullptr, the only value we care about in PHI is V. It // doesn't matter to us what the other operand is (it'll never get used). We // could just create a new PHI with an undef incoming value, but that could // increase register pressure if EarlyCSE/InstCombine can't fold it with some // other PHI. So here we directly look for some PHI in BB's successor with V // as an incoming operand. If we find one, we use it, else we create a new // one. // // If AlternativeV is not nullptr, we care about both incoming values in PHI. // PHI must be exactly: phi [ %BB, %V ], [ %OtherBB, %AlternativeV] // where OtherBB is the single other predecessor of BB's only successor. PHINode *PHI = nullptr; BasicBlock *Succ = BB->getSingleSuccessor(); for (auto I = Succ->begin(); isa(I); ++I) if (cast(I)->getIncomingValueForBlock(BB) == V) { PHI = cast(I); if (!AlternativeV) break; assert(Succ->hasNPredecessors(2)); auto PredI = pred_begin(Succ); BasicBlock *OtherPredBB = *PredI == BB ? *++PredI : *PredI; if (PHI->getIncomingValueForBlock(OtherPredBB) == AlternativeV) break; PHI = nullptr; } if (PHI) return PHI; // If V is not an instruction defined in BB, just return it. if (!AlternativeV && (!isa(V) || cast(V)->getParent() != BB)) return V; PHI = PHINode::Create(V->getType(), 2, "simplifycfg.merge"); PHI->insertBefore(Succ->begin()); PHI->addIncoming(V, BB); for (BasicBlock *PredBB : predecessors(Succ)) if (PredBB != BB) PHI->addIncoming( AlternativeV ? AlternativeV : PoisonValue::get(V->getType()), PredBB); return PHI; } static bool mergeConditionalStoreToAddress( BasicBlock *PTB, BasicBlock *PFB, BasicBlock *QTB, BasicBlock *QFB, BasicBlock *PostBB, Value *Address, bool InvertPCond, bool InvertQCond, DomTreeUpdater *DTU, const DataLayout &DL, const TargetTransformInfo &TTI) { // For every pointer, there must be exactly two stores, one coming from // PTB or PFB, and the other from QTB or QFB. We don't support more than one // store (to any address) in PTB,PFB or QTB,QFB. // FIXME: We could relax this restriction with a bit more work and performance // testing. StoreInst *PStore = findUniqueStoreInBlocks(PTB, PFB); StoreInst *QStore = findUniqueStoreInBlocks(QTB, QFB); if (!PStore || !QStore) return false; // Now check the stores are compatible. if (!QStore->isUnordered() || !PStore->isUnordered() || PStore->getValueOperand()->getType() != QStore->getValueOperand()->getType()) return false; // Check that sinking the store won't cause program behavior changes. Sinking // the store out of the Q blocks won't change any behavior as we're sinking // from a block to its unconditional successor. But we're moving a store from // the P blocks down through the middle block (QBI) and past both QFB and QTB. // So we need to check that there are no aliasing loads or stores in // QBI, QTB and QFB. We also need to check there are no conflicting memory // operations between PStore and the end of its parent block. // // The ideal way to do this is to query AliasAnalysis, but we don't // preserve AA currently so that is dangerous. Be super safe and just // check there are no other memory operations at all. for (auto &I : *QFB->getSinglePredecessor()) if (I.mayReadOrWriteMemory()) return false; for (auto &I : *QFB) if (&I != QStore && I.mayReadOrWriteMemory()) return false; if (QTB) for (auto &I : *QTB) if (&I != QStore && I.mayReadOrWriteMemory()) return false; for (auto I = BasicBlock::iterator(PStore), E = PStore->getParent()->end(); I != E; ++I) if (&*I != PStore && I->mayReadOrWriteMemory()) return false; // If we're not in aggressive mode, we only optimize if we have some // confidence that by optimizing we'll allow P and/or Q to be if-converted. auto IsWorthwhile = [&](BasicBlock *BB, ArrayRef FreeStores) { if (!BB) return true; // Heuristic: if the block can be if-converted/phi-folded and the // instructions inside are all cheap (arithmetic/GEPs), it's worthwhile to // thread this store. InstructionCost Cost = 0; InstructionCost Budget = PHINodeFoldingThreshold * TargetTransformInfo::TCC_Basic; for (auto &I : BB->instructionsWithoutDebug(false)) { // Consider terminator instruction to be free. if (I.isTerminator()) continue; // If this is one the stores that we want to speculate out of this BB, // then don't count it's cost, consider it to be free. if (auto *S = dyn_cast(&I)) if (llvm::find(FreeStores, S)) continue; // Else, we have a white-list of instructions that we are ak speculating. if (!isa(I) && !isa(I)) return false; // Not in white-list - not worthwhile folding. // And finally, if this is a non-free instruction that we are okay // speculating, ensure that we consider the speculation budget. Cost += TTI.getInstructionCost(&I, TargetTransformInfo::TCK_SizeAndLatency); if (Cost > Budget) return false; // Eagerly refuse to fold as soon as we're out of budget. } assert(Cost <= Budget && "When we run out of budget we will eagerly return from within the " "per-instruction loop."); return true; }; const std::array FreeStores = {PStore, QStore}; if (!MergeCondStoresAggressively && (!IsWorthwhile(PTB, FreeStores) || !IsWorthwhile(PFB, FreeStores) || !IsWorthwhile(QTB, FreeStores) || !IsWorthwhile(QFB, FreeStores))) return false; // If PostBB has more than two predecessors, we need to split it so we can // sink the store. if (std::next(pred_begin(PostBB), 2) != pred_end(PostBB)) { // We know that QFB's only successor is PostBB. And QFB has a single // predecessor. If QTB exists, then its only successor is also PostBB. // If QTB does not exist, then QFB's only predecessor has a conditional // branch to QFB and PostBB. BasicBlock *TruePred = QTB ? QTB : QFB->getSinglePredecessor(); BasicBlock *NewBB = SplitBlockPredecessors(PostBB, {QFB, TruePred}, "condstore.split", DTU); if (!NewBB) return false; PostBB = NewBB; } // OK, we're going to sink the stores to PostBB. The store has to be // conditional though, so first create the predicate. Value *PCond = cast(PFB->getSinglePredecessor()->getTerminator()) ->getCondition(); Value *QCond = cast(QFB->getSinglePredecessor()->getTerminator()) ->getCondition(); Value *PPHI = ensureValueAvailableInSuccessor(PStore->getValueOperand(), PStore->getParent()); Value *QPHI = ensureValueAvailableInSuccessor(QStore->getValueOperand(), QStore->getParent(), PPHI); BasicBlock::iterator PostBBFirst = PostBB->getFirstInsertionPt(); IRBuilder<> QB(PostBB, PostBBFirst); QB.SetCurrentDebugLocation(PostBBFirst->getStableDebugLoc()); Value *PPred = PStore->getParent() == PTB ? PCond : QB.CreateNot(PCond); Value *QPred = QStore->getParent() == QTB ? QCond : QB.CreateNot(QCond); if (InvertPCond) PPred = QB.CreateNot(PPred); if (InvertQCond) QPred = QB.CreateNot(QPred); Value *CombinedPred = QB.CreateOr(PPred, QPred); BasicBlock::iterator InsertPt = QB.GetInsertPoint(); auto *T = SplitBlockAndInsertIfThen(CombinedPred, InsertPt, /*Unreachable=*/false, /*BranchWeights=*/nullptr, DTU); QB.SetInsertPoint(T); StoreInst *SI = cast(QB.CreateStore(QPHI, Address)); SI->setAAMetadata(PStore->getAAMetadata().merge(QStore->getAAMetadata())); // Choose the minimum alignment. If we could prove both stores execute, we // could use biggest one. In this case, though, we only know that one of the // stores executes. And we don't know it's safe to take the alignment from a // store that doesn't execute. SI->setAlignment(std::min(PStore->getAlign(), QStore->getAlign())); QStore->eraseFromParent(); PStore->eraseFromParent(); return true; } static bool mergeConditionalStores(BranchInst *PBI, BranchInst *QBI, DomTreeUpdater *DTU, const DataLayout &DL, const TargetTransformInfo &TTI) { // The intention here is to find diamonds or triangles (see below) where each // conditional block contains a store to the same address. Both of these // stores are conditional, so they can't be unconditionally sunk. But it may // be profitable to speculatively sink the stores into one merged store at the // end, and predicate the merged store on the union of the two conditions of // PBI and QBI. // // This can reduce the number of stores executed if both of the conditions are // true, and can allow the blocks to become small enough to be if-converted. // This optimization will also chain, so that ladders of test-and-set // sequences can be if-converted away. // // We only deal with simple diamonds or triangles: // // PBI or PBI or a combination of the two // / \ | \ // PTB PFB | PFB // \ / | / // QBI QBI // / \ | \ // QTB QFB | QFB // \ / | / // PostBB PostBB // // We model triangles as a type of diamond with a nullptr "true" block. // Triangles are canonicalized so that the fallthrough edge is represented by // a true condition, as in the diagram above. BasicBlock *PTB = PBI->getSuccessor(0); BasicBlock *PFB = PBI->getSuccessor(1); BasicBlock *QTB = QBI->getSuccessor(0); BasicBlock *QFB = QBI->getSuccessor(1); BasicBlock *PostBB = QFB->getSingleSuccessor(); // Make sure we have a good guess for PostBB. If QTB's only successor is // QFB, then QFB is a better PostBB. if (QTB->getSingleSuccessor() == QFB) PostBB = QFB; // If we couldn't find a good PostBB, stop. if (!PostBB) return false; bool InvertPCond = false, InvertQCond = false; // Canonicalize fallthroughs to the true branches. if (PFB == QBI->getParent()) { std::swap(PFB, PTB); InvertPCond = true; } if (QFB == PostBB) { std::swap(QFB, QTB); InvertQCond = true; } // From this point on we can assume PTB or QTB may be fallthroughs but PFB // and QFB may not. Model fallthroughs as a nullptr block. if (PTB == QBI->getParent()) PTB = nullptr; if (QTB == PostBB) QTB = nullptr; // Legality bailouts. We must have at least the non-fallthrough blocks and // the post-dominating block, and the non-fallthroughs must only have one // predecessor. auto HasOnePredAndOneSucc = [](BasicBlock *BB, BasicBlock *P, BasicBlock *S) { return BB->getSinglePredecessor() == P && BB->getSingleSuccessor() == S; }; if (!HasOnePredAndOneSucc(PFB, PBI->getParent(), QBI->getParent()) || !HasOnePredAndOneSucc(QFB, QBI->getParent(), PostBB)) return false; if ((PTB && !HasOnePredAndOneSucc(PTB, PBI->getParent(), QBI->getParent())) || (QTB && !HasOnePredAndOneSucc(QTB, QBI->getParent(), PostBB))) return false; if (!QBI->getParent()->hasNUses(2)) return false; // OK, this is a sequence of two diamonds or triangles. // Check if there are stores in PTB or PFB that are repeated in QTB or QFB. SmallPtrSet PStoreAddresses, QStoreAddresses; for (auto *BB : {PTB, PFB}) { if (!BB) continue; for (auto &I : *BB) if (StoreInst *SI = dyn_cast(&I)) PStoreAddresses.insert(SI->getPointerOperand()); } for (auto *BB : {QTB, QFB}) { if (!BB) continue; for (auto &I : *BB) if (StoreInst *SI = dyn_cast(&I)) QStoreAddresses.insert(SI->getPointerOperand()); } set_intersect(PStoreAddresses, QStoreAddresses); // set_intersect mutates PStoreAddresses in place. Rename it here to make it // clear what it contains. auto &CommonAddresses = PStoreAddresses; bool Changed = false; for (auto *Address : CommonAddresses) Changed |= mergeConditionalStoreToAddress(PTB, PFB, QTB, QFB, PostBB, Address, InvertPCond, InvertQCond, DTU, DL, TTI); return Changed; } /// If the previous block ended with a widenable branch, determine if reusing /// the target block is profitable and legal. This will have the effect of /// "widening" PBI, but doesn't require us to reason about hosting safety. static bool tryWidenCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI, DomTreeUpdater *DTU) { // TODO: This can be generalized in two important ways: // 1) We can allow phi nodes in IfFalseBB and simply reuse all the input // values from the PBI edge. // 2) We can sink side effecting instructions into BI's fallthrough // successor provided they doesn't contribute to computation of // BI's condition. BasicBlock *IfTrueBB = PBI->getSuccessor(0); BasicBlock *IfFalseBB = PBI->getSuccessor(1); if (!isWidenableBranch(PBI) || IfTrueBB != BI->getParent() || !BI->getParent()->getSinglePredecessor()) return false; if (!IfFalseBB->phis().empty()) return false; // TODO // This helps avoid infinite loop with SimplifyCondBranchToCondBranch which // may undo the transform done here. // TODO: There might be a more fine-grained solution to this. if (!llvm::succ_empty(IfFalseBB)) return false; // Use lambda to lazily compute expensive condition after cheap ones. auto NoSideEffects = [](BasicBlock &BB) { return llvm::none_of(BB, [](const Instruction &I) { return I.mayWriteToMemory() || I.mayHaveSideEffects(); }); }; if (BI->getSuccessor(1) != IfFalseBB && // no inf looping BI->getSuccessor(1)->getTerminatingDeoptimizeCall() && // profitability NoSideEffects(*BI->getParent())) { auto *OldSuccessor = BI->getSuccessor(1); OldSuccessor->removePredecessor(BI->getParent()); BI->setSuccessor(1, IfFalseBB); if (DTU) DTU->applyUpdates( {{DominatorTree::Insert, BI->getParent(), IfFalseBB}, {DominatorTree::Delete, BI->getParent(), OldSuccessor}}); return true; } if (BI->getSuccessor(0) != IfFalseBB && // no inf looping BI->getSuccessor(0)->getTerminatingDeoptimizeCall() && // profitability NoSideEffects(*BI->getParent())) { auto *OldSuccessor = BI->getSuccessor(0); OldSuccessor->removePredecessor(BI->getParent()); BI->setSuccessor(0, IfFalseBB); if (DTU) DTU->applyUpdates( {{DominatorTree::Insert, BI->getParent(), IfFalseBB}, {DominatorTree::Delete, BI->getParent(), OldSuccessor}}); return true; } return false; } /// If we have a conditional branch as a predecessor of another block, /// this function tries to simplify it. We know /// that PBI and BI are both conditional branches, and BI is in one of the /// successor blocks of PBI - PBI branches to BI. static bool SimplifyCondBranchToCondBranch(BranchInst *PBI, BranchInst *BI, DomTreeUpdater *DTU, const DataLayout &DL, const TargetTransformInfo &TTI) { assert(PBI->isConditional() && BI->isConditional()); BasicBlock *BB = BI->getParent(); // If this block ends with a branch instruction, and if there is a // predecessor that ends on a branch of the same condition, make // this conditional branch redundant. if (PBI->getCondition() == BI->getCondition() && PBI->getSuccessor(0) != PBI->getSuccessor(1)) { // Okay, the outcome of this conditional branch is statically // knowable. If this block had a single pred, handle specially, otherwise // FoldCondBranchOnValueKnownInPredecessor() will handle it. if (BB->getSinglePredecessor()) { // Turn this into a branch on constant. bool CondIsTrue = PBI->getSuccessor(0) == BB; BI->setCondition( ConstantInt::get(Type::getInt1Ty(BB->getContext()), CondIsTrue)); return true; // Nuke the branch on constant. } } // If the previous block ended with a widenable branch, determine if reusing // the target block is profitable and legal. This will have the effect of // "widening" PBI, but doesn't require us to reason about hosting safety. if (tryWidenCondBranchToCondBranch(PBI, BI, DTU)) return true; // If both branches are conditional and both contain stores to the same // address, remove the stores from the conditionals and create a conditional // merged store at the end. if (MergeCondStores && mergeConditionalStores(PBI, BI, DTU, DL, TTI)) return true; // If this is a conditional branch in an empty block, and if any // predecessors are a conditional branch to one of our destinations, // fold the conditions into logical ops and one cond br. // Ignore dbg intrinsics. if (&*BB->instructionsWithoutDebug(false).begin() != BI) return false; int PBIOp, BIOp; if (PBI->getSuccessor(0) == BI->getSuccessor(0)) { PBIOp = 0; BIOp = 0; } else if (PBI->getSuccessor(0) == BI->getSuccessor(1)) { PBIOp = 0; BIOp = 1; } else if (PBI->getSuccessor(1) == BI->getSuccessor(0)) { PBIOp = 1; BIOp = 0; } else if (PBI->getSuccessor(1) == BI->getSuccessor(1)) { PBIOp = 1; BIOp = 1; } else { return false; } // Check to make sure that the other destination of this branch // isn't BB itself. If so, this is an infinite loop that will // keep getting unwound. if (PBI->getSuccessor(PBIOp) == BB) return false; // If predecessor's branch probability to BB is too low don't merge branches. SmallVector PredWeights; if (!PBI->getMetadata(LLVMContext::MD_unpredictable) && extractBranchWeights(*PBI, PredWeights) && (static_cast(PredWeights[0]) + PredWeights[1]) != 0) { BranchProbability CommonDestProb = BranchProbability::getBranchProbability( PredWeights[PBIOp], static_cast(PredWeights[0]) + PredWeights[1]); BranchProbability Likely = TTI.getPredictableBranchThreshold(); if (CommonDestProb >= Likely) return false; } // Do not perform this transformation if it would require // insertion of a large number of select instructions. For targets // without predication/cmovs, this is a big pessimization. BasicBlock *CommonDest = PBI->getSuccessor(PBIOp); BasicBlock *RemovedDest = PBI->getSuccessor(PBIOp ^ 1); unsigned NumPhis = 0; for (BasicBlock::iterator II = CommonDest->begin(); isa(II); ++II, ++NumPhis) { if (NumPhis > 2) // Disable this xform. return false; } // Finally, if everything is ok, fold the branches to logical ops. BasicBlock *OtherDest = BI->getSuccessor(BIOp ^ 1); LLVM_DEBUG(dbgs() << "FOLDING BRs:" << *PBI->getParent() << "AND: " << *BI->getParent()); SmallVector Updates; // If OtherDest *is* BB, then BB is a basic block with a single conditional // branch in it, where one edge (OtherDest) goes back to itself but the other // exits. We don't *know* that the program avoids the infinite loop // (even though that seems likely). If we do this xform naively, we'll end up // recursively unpeeling the loop. Since we know that (after the xform is // done) that the block *is* infinite if reached, we just make it an obviously // infinite loop with no cond branch. if (OtherDest == BB) { // Insert it at the end of the function, because it's either code, // or it won't matter if it's hot. :) BasicBlock *InfLoopBlock = BasicBlock::Create(BB->getContext(), "infloop", BB->getParent()); BranchInst::Create(InfLoopBlock, InfLoopBlock); if (DTU) Updates.push_back({DominatorTree::Insert, InfLoopBlock, InfLoopBlock}); OtherDest = InfLoopBlock; } LLVM_DEBUG(dbgs() << *PBI->getParent()->getParent()); // BI may have other predecessors. Because of this, we leave // it alone, but modify PBI. // Make sure we get to CommonDest on True&True directions. Value *PBICond = PBI->getCondition(); IRBuilder Builder(PBI); if (PBIOp) PBICond = Builder.CreateNot(PBICond, PBICond->getName() + ".not"); Value *BICond = BI->getCondition(); if (BIOp) BICond = Builder.CreateNot(BICond, BICond->getName() + ".not"); // Merge the conditions. Value *Cond = createLogicalOp(Builder, Instruction::Or, PBICond, BICond, "brmerge"); // Modify PBI to branch on the new condition to the new dests. PBI->setCondition(Cond); PBI->setSuccessor(0, CommonDest); PBI->setSuccessor(1, OtherDest); if (DTU) { Updates.push_back({DominatorTree::Insert, PBI->getParent(), OtherDest}); Updates.push_back({DominatorTree::Delete, PBI->getParent(), RemovedDest}); DTU->applyUpdates(Updates); } // Update branch weight for PBI. uint64_t PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight; uint64_t PredCommon, PredOther, SuccCommon, SuccOther; bool HasWeights = extractPredSuccWeights(PBI, BI, PredTrueWeight, PredFalseWeight, SuccTrueWeight, SuccFalseWeight); if (HasWeights) { PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight; PredOther = PBIOp ? PredTrueWeight : PredFalseWeight; SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight; SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight; // The weight to CommonDest should be PredCommon * SuccTotal + // PredOther * SuccCommon. // The weight to OtherDest should be PredOther * SuccOther. uint64_t NewWeights[2] = {PredCommon * (SuccCommon + SuccOther) + PredOther * SuccCommon, PredOther * SuccOther}; // Halve the weights if any of them cannot fit in an uint32_t FitWeights(NewWeights); setBranchWeights(PBI, NewWeights[0], NewWeights[1], /*IsExpected=*/false); } // OtherDest may have phi nodes. If so, add an entry from PBI's // block that are identical to the entries for BI's block. AddPredecessorToBlock(OtherDest, PBI->getParent(), BB); // We know that the CommonDest already had an edge from PBI to // it. If it has PHIs though, the PHIs may have different // entries for BB and PBI's BB. If so, insert a select to make // them agree. for (PHINode &PN : CommonDest->phis()) { Value *BIV = PN.getIncomingValueForBlock(BB); unsigned PBBIdx = PN.getBasicBlockIndex(PBI->getParent()); Value *PBIV = PN.getIncomingValue(PBBIdx); if (BIV != PBIV) { // Insert a select in PBI to pick the right value. SelectInst *NV = cast( Builder.CreateSelect(PBICond, PBIV, BIV, PBIV->getName() + ".mux")); PN.setIncomingValue(PBBIdx, NV); // Although the select has the same condition as PBI, the original branch // weights for PBI do not apply to the new select because the select's // 'logical' edges are incoming edges of the phi that is eliminated, not // the outgoing edges of PBI. if (HasWeights) { uint64_t PredCommon = PBIOp ? PredFalseWeight : PredTrueWeight; uint64_t PredOther = PBIOp ? PredTrueWeight : PredFalseWeight; uint64_t SuccCommon = BIOp ? SuccFalseWeight : SuccTrueWeight; uint64_t SuccOther = BIOp ? SuccTrueWeight : SuccFalseWeight; // The weight to PredCommonDest should be PredCommon * SuccTotal. // The weight to PredOtherDest should be PredOther * SuccCommon. uint64_t NewWeights[2] = {PredCommon * (SuccCommon + SuccOther), PredOther * SuccCommon}; FitWeights(NewWeights); setBranchWeights(NV, NewWeights[0], NewWeights[1], /*IsExpected=*/false); } } } LLVM_DEBUG(dbgs() << "INTO: " << *PBI->getParent()); LLVM_DEBUG(dbgs() << *PBI->getParent()->getParent()); // This basic block is probably dead. We know it has at least // one fewer predecessor. return true; } // Simplifies a terminator by replacing it with a branch to TrueBB if Cond is // true or to FalseBB if Cond is false. // Takes care of updating the successors and removing the old terminator. // Also makes sure not to introduce new successors by assuming that edges to // non-successor TrueBBs and FalseBBs aren't reachable. bool SimplifyCFGOpt::SimplifyTerminatorOnSelect(Instruction *OldTerm, Value *Cond, BasicBlock *TrueBB, BasicBlock *FalseBB, uint32_t TrueWeight, uint32_t FalseWeight) { auto *BB = OldTerm->getParent(); // Remove any superfluous successor edges from the CFG. // First, figure out which successors to preserve. // If TrueBB and FalseBB are equal, only try to preserve one copy of that // successor. BasicBlock *KeepEdge1 = TrueBB; BasicBlock *KeepEdge2 = TrueBB != FalseBB ? FalseBB : nullptr; SmallSetVector RemovedSuccessors; // Then remove the rest. for (BasicBlock *Succ : successors(OldTerm)) { // Make sure only to keep exactly one copy of each edge. if (Succ == KeepEdge1) KeepEdge1 = nullptr; else if (Succ == KeepEdge2) KeepEdge2 = nullptr; else { Succ->removePredecessor(BB, /*KeepOneInputPHIs=*/true); if (Succ != TrueBB && Succ != FalseBB) RemovedSuccessors.insert(Succ); } } IRBuilder<> Builder(OldTerm); Builder.SetCurrentDebugLocation(OldTerm->getDebugLoc()); // Insert an appropriate new terminator. if (!KeepEdge1 && !KeepEdge2) { if (TrueBB == FalseBB) { // We were only looking for one successor, and it was present. // Create an unconditional branch to it. Builder.CreateBr(TrueBB); } else { // We found both of the successors we were looking for. // Create a conditional branch sharing the condition of the select. BranchInst *NewBI = Builder.CreateCondBr(Cond, TrueBB, FalseBB); if (TrueWeight != FalseWeight) setBranchWeights(NewBI, TrueWeight, FalseWeight, /*IsExpected=*/false); } } else if (KeepEdge1 && (KeepEdge2 || TrueBB == FalseBB)) { // Neither of the selected blocks were successors, so this // terminator must be unreachable. new UnreachableInst(OldTerm->getContext(), OldTerm->getIterator()); } else { // One of the selected values was a successor, but the other wasn't. // Insert an unconditional branch to the one that was found; // the edge to the one that wasn't must be unreachable. if (!KeepEdge1) { // Only TrueBB was found. Builder.CreateBr(TrueBB); } else { // Only FalseBB was found. Builder.CreateBr(FalseBB); } } EraseTerminatorAndDCECond(OldTerm); if (DTU) { SmallVector Updates; Updates.reserve(RemovedSuccessors.size()); for (auto *RemovedSuccessor : RemovedSuccessors) Updates.push_back({DominatorTree::Delete, BB, RemovedSuccessor}); DTU->applyUpdates(Updates); } return true; } // Replaces // (switch (select cond, X, Y)) on constant X, Y // with a branch - conditional if X and Y lead to distinct BBs, // unconditional otherwise. bool SimplifyCFGOpt::SimplifySwitchOnSelect(SwitchInst *SI, SelectInst *Select) { // Check for constant integer values in the select. ConstantInt *TrueVal = dyn_cast(Select->getTrueValue()); ConstantInt *FalseVal = dyn_cast(Select->getFalseValue()); if (!TrueVal || !FalseVal) return false; // Find the relevant condition and destinations. Value *Condition = Select->getCondition(); BasicBlock *TrueBB = SI->findCaseValue(TrueVal)->getCaseSuccessor(); BasicBlock *FalseBB = SI->findCaseValue(FalseVal)->getCaseSuccessor(); // Get weight for TrueBB and FalseBB. uint32_t TrueWeight = 0, FalseWeight = 0; SmallVector Weights; bool HasWeights = hasBranchWeightMD(*SI); if (HasWeights) { GetBranchWeights(SI, Weights); if (Weights.size() == 1 + SI->getNumCases()) { TrueWeight = (uint32_t)Weights[SI->findCaseValue(TrueVal)->getSuccessorIndex()]; FalseWeight = (uint32_t)Weights[SI->findCaseValue(FalseVal)->getSuccessorIndex()]; } } // Perform the actual simplification. return SimplifyTerminatorOnSelect(SI, Condition, TrueBB, FalseBB, TrueWeight, FalseWeight); } // Replaces // (indirectbr (select cond, blockaddress(@fn, BlockA), // blockaddress(@fn, BlockB))) // with // (br cond, BlockA, BlockB). bool SimplifyCFGOpt::SimplifyIndirectBrOnSelect(IndirectBrInst *IBI, SelectInst *SI) { // Check that both operands of the select are block addresses. BlockAddress *TBA = dyn_cast(SI->getTrueValue()); BlockAddress *FBA = dyn_cast(SI->getFalseValue()); if (!TBA || !FBA) return false; // Extract the actual blocks. BasicBlock *TrueBB = TBA->getBasicBlock(); BasicBlock *FalseBB = FBA->getBasicBlock(); // Perform the actual simplification. return SimplifyTerminatorOnSelect(IBI, SI->getCondition(), TrueBB, FalseBB, 0, 0); } /// This is called when we find an icmp instruction /// (a seteq/setne with a constant) as the only instruction in a /// block that ends with an uncond branch. We are looking for a very specific /// pattern that occurs when "A == 1 || A == 2 || A == 3" gets simplified. In /// this case, we merge the first two "or's of icmp" into a switch, but then the /// default value goes to an uncond block with a seteq in it, we get something /// like: /// /// switch i8 %A, label %DEFAULT [ i8 1, label %end i8 2, label %end ] /// DEFAULT: /// %tmp = icmp eq i8 %A, 92 /// br label %end /// end: /// ... = phi i1 [ true, %entry ], [ %tmp, %DEFAULT ], [ true, %entry ] /// /// We prefer to split the edge to 'end' so that there is a true/false entry to /// the PHI, merging the third icmp into the switch. bool SimplifyCFGOpt::tryToSimplifyUncondBranchWithICmpInIt( ICmpInst *ICI, IRBuilder<> &Builder) { BasicBlock *BB = ICI->getParent(); // If the block has any PHIs in it or the icmp has multiple uses, it is too // complex. if (isa(BB->begin()) || !ICI->hasOneUse()) return false; Value *V = ICI->getOperand(0); ConstantInt *Cst = cast(ICI->getOperand(1)); // The pattern we're looking for is where our only predecessor is a switch on // 'V' and this block is the default case for the switch. In this case we can // fold the compared value into the switch to simplify things. BasicBlock *Pred = BB->getSinglePredecessor(); if (!Pred || !isa(Pred->getTerminator())) return false; SwitchInst *SI = cast(Pred->getTerminator()); if (SI->getCondition() != V) return false; // If BB is reachable on a non-default case, then we simply know the value of // V in this block. Substitute it and constant fold the icmp instruction // away. if (SI->getDefaultDest() != BB) { ConstantInt *VVal = SI->findCaseDest(BB); assert(VVal && "Should have a unique destination value"); ICI->setOperand(0, VVal); if (Value *V = simplifyInstruction(ICI, {DL, ICI})) { ICI->replaceAllUsesWith(V); ICI->eraseFromParent(); } // BB is now empty, so it is likely to simplify away. return requestResimplify(); } // Ok, the block is reachable from the default dest. If the constant we're // comparing exists in one of the other edges, then we can constant fold ICI // and zap it. if (SI->findCaseValue(Cst) != SI->case_default()) { Value *V; if (ICI->getPredicate() == ICmpInst::ICMP_EQ) V = ConstantInt::getFalse(BB->getContext()); else V = ConstantInt::getTrue(BB->getContext()); ICI->replaceAllUsesWith(V); ICI->eraseFromParent(); // BB is now empty, so it is likely to simplify away. return requestResimplify(); } // The use of the icmp has to be in the 'end' block, by the only PHI node in // the block. BasicBlock *SuccBlock = BB->getTerminator()->getSuccessor(0); PHINode *PHIUse = dyn_cast(ICI->user_back()); if (PHIUse == nullptr || PHIUse != &SuccBlock->front() || isa(++BasicBlock::iterator(PHIUse))) return false; // If the icmp is a SETEQ, then the default dest gets false, the new edge gets // true in the PHI. Constant *DefaultCst = ConstantInt::getTrue(BB->getContext()); Constant *NewCst = ConstantInt::getFalse(BB->getContext()); if (ICI->getPredicate() == ICmpInst::ICMP_EQ) std::swap(DefaultCst, NewCst); // Replace ICI (which is used by the PHI for the default value) with true or // false depending on if it is EQ or NE. ICI->replaceAllUsesWith(DefaultCst); ICI->eraseFromParent(); SmallVector Updates; // Okay, the switch goes to this block on a default value. Add an edge from // the switch to the merge point on the compared value. BasicBlock *NewBB = BasicBlock::Create(BB->getContext(), "switch.edge", BB->getParent(), BB); { SwitchInstProfUpdateWrapper SIW(*SI); auto W0 = SIW.getSuccessorWeight(0); SwitchInstProfUpdateWrapper::CaseWeightOpt NewW; if (W0) { NewW = ((uint64_t(*W0) + 1) >> 1); SIW.setSuccessorWeight(0, *NewW); } SIW.addCase(Cst, NewBB, NewW); if (DTU) Updates.push_back({DominatorTree::Insert, Pred, NewBB}); } // NewBB branches to the phi block, add the uncond branch and the phi entry. Builder.SetInsertPoint(NewBB); Builder.SetCurrentDebugLocation(SI->getDebugLoc()); Builder.CreateBr(SuccBlock); PHIUse->addIncoming(NewCst, NewBB); if (DTU) { Updates.push_back({DominatorTree::Insert, NewBB, SuccBlock}); DTU->applyUpdates(Updates); } return true; } /// The specified branch is a conditional branch. /// Check to see if it is branching on an or/and chain of icmp instructions, and /// fold it into a switch instruction if so. bool SimplifyCFGOpt::SimplifyBranchOnICmpChain(BranchInst *BI, IRBuilder<> &Builder, const DataLayout &DL) { Instruction *Cond = dyn_cast(BI->getCondition()); if (!Cond) return false; // Change br (X == 0 | X == 1), T, F into a switch instruction. // If this is a bunch of seteq's or'd together, or if it's a bunch of // 'setne's and'ed together, collect them. // Try to gather values from a chain of and/or to be turned into a switch ConstantComparesGatherer ConstantCompare(Cond, DL); // Unpack the result SmallVectorImpl &Values = ConstantCompare.Vals; Value *CompVal = ConstantCompare.CompValue; unsigned UsedICmps = ConstantCompare.UsedICmps; Value *ExtraCase = ConstantCompare.Extra; // If we didn't have a multiply compared value, fail. if (!CompVal) return false; // Avoid turning single icmps into a switch. if (UsedICmps <= 1) return false; bool TrueWhenEqual = match(Cond, m_LogicalOr(m_Value(), m_Value())); // There might be duplicate constants in the list, which the switch // instruction can't handle, remove them now. array_pod_sort(Values.begin(), Values.end(), ConstantIntSortPredicate); Values.erase(llvm::unique(Values), Values.end()); // If Extra was used, we require at least two switch values to do the // transformation. A switch with one value is just a conditional branch. if (ExtraCase && Values.size() < 2) return false; // TODO: Preserve branch weight metadata, similarly to how // FoldValueComparisonIntoPredecessors preserves it. // Figure out which block is which destination. BasicBlock *DefaultBB = BI->getSuccessor(1); BasicBlock *EdgeBB = BI->getSuccessor(0); if (!TrueWhenEqual) std::swap(DefaultBB, EdgeBB); BasicBlock *BB = BI->getParent(); LLVM_DEBUG(dbgs() << "Converting 'icmp' chain with " << Values.size() << " cases into SWITCH. BB is:\n" << *BB); SmallVector Updates; // If there are any extra values that couldn't be folded into the switch // then we evaluate them with an explicit branch first. Split the block // right before the condbr to handle it. if (ExtraCase) { BasicBlock *NewBB = SplitBlock(BB, BI, DTU, /*LI=*/nullptr, /*MSSAU=*/nullptr, "switch.early.test"); // Remove the uncond branch added to the old block. Instruction *OldTI = BB->getTerminator(); Builder.SetInsertPoint(OldTI); // There can be an unintended UB if extra values are Poison. Before the // transformation, extra values may not be evaluated according to the // condition, and it will not raise UB. But after transformation, we are // evaluating extra values before checking the condition, and it will raise // UB. It can be solved by adding freeze instruction to extra values. AssumptionCache *AC = Options.AC; if (!isGuaranteedNotToBeUndefOrPoison(ExtraCase, AC, BI, nullptr)) ExtraCase = Builder.CreateFreeze(ExtraCase); if (TrueWhenEqual) Builder.CreateCondBr(ExtraCase, EdgeBB, NewBB); else Builder.CreateCondBr(ExtraCase, NewBB, EdgeBB); OldTI->eraseFromParent(); if (DTU) Updates.push_back({DominatorTree::Insert, BB, EdgeBB}); // If there are PHI nodes in EdgeBB, then we need to add a new entry to them // for the edge we just added. AddPredecessorToBlock(EdgeBB, BB, NewBB); LLVM_DEBUG(dbgs() << " ** 'icmp' chain unhandled condition: " << *ExtraCase << "\nEXTRABB = " << *BB); BB = NewBB; } Builder.SetInsertPoint(BI); // Convert pointer to int before we switch. if (CompVal->getType()->isPointerTy()) { CompVal = Builder.CreatePtrToInt( CompVal, DL.getIntPtrType(CompVal->getType()), "magicptr"); } // Create the new switch instruction now. SwitchInst *New = Builder.CreateSwitch(CompVal, DefaultBB, Values.size()); // Add all of the 'cases' to the switch instruction. for (unsigned i = 0, e = Values.size(); i != e; ++i) New->addCase(Values[i], EdgeBB); // We added edges from PI to the EdgeBB. As such, if there were any // PHI nodes in EdgeBB, they need entries to be added corresponding to // the number of edges added. for (BasicBlock::iterator BBI = EdgeBB->begin(); isa(BBI); ++BBI) { PHINode *PN = cast(BBI); Value *InVal = PN->getIncomingValueForBlock(BB); for (unsigned i = 0, e = Values.size() - 1; i != e; ++i) PN->addIncoming(InVal, BB); } // Erase the old branch instruction. EraseTerminatorAndDCECond(BI); if (DTU) DTU->applyUpdates(Updates); LLVM_DEBUG(dbgs() << " ** 'icmp' chain result is:\n" << *BB << '\n'); return true; } bool SimplifyCFGOpt::simplifyResume(ResumeInst *RI, IRBuilder<> &Builder) { if (isa(RI->getValue())) return simplifyCommonResume(RI); else if (isa(RI->getParent()->getFirstNonPHI()) && RI->getValue() == RI->getParent()->getFirstNonPHI()) // The resume must unwind the exception that caused control to branch here. return simplifySingleResume(RI); return false; } // Check if cleanup block is empty static bool isCleanupBlockEmpty(iterator_range R) { for (Instruction &I : R) { auto *II = dyn_cast(&I); if (!II) return false; Intrinsic::ID IntrinsicID = II->getIntrinsicID(); switch (IntrinsicID) { case Intrinsic::dbg_declare: case Intrinsic::dbg_value: case Intrinsic::dbg_label: case Intrinsic::lifetime_end: break; default: return false; } } return true; } // Simplify resume that is shared by several landing pads (phi of landing pad). bool SimplifyCFGOpt::simplifyCommonResume(ResumeInst *RI) { BasicBlock *BB = RI->getParent(); // Check that there are no other instructions except for debug and lifetime // intrinsics between the phi's and resume instruction. if (!isCleanupBlockEmpty( make_range(RI->getParent()->getFirstNonPHI(), BB->getTerminator()))) return false; SmallSetVector TrivialUnwindBlocks; auto *PhiLPInst = cast(RI->getValue()); // Check incoming blocks to see if any of them are trivial. for (unsigned Idx = 0, End = PhiLPInst->getNumIncomingValues(); Idx != End; Idx++) { auto *IncomingBB = PhiLPInst->getIncomingBlock(Idx); auto *IncomingValue = PhiLPInst->getIncomingValue(Idx); // If the block has other successors, we can not delete it because // it has other dependents. if (IncomingBB->getUniqueSuccessor() != BB) continue; auto *LandingPad = dyn_cast(IncomingBB->getFirstNonPHI()); // Not the landing pad that caused the control to branch here. if (IncomingValue != LandingPad) continue; if (isCleanupBlockEmpty( make_range(LandingPad->getNextNode(), IncomingBB->getTerminator()))) TrivialUnwindBlocks.insert(IncomingBB); } // If no trivial unwind blocks, don't do any simplifications. if (TrivialUnwindBlocks.empty()) return false; // Turn all invokes that unwind here into calls. for (auto *TrivialBB : TrivialUnwindBlocks) { // Blocks that will be simplified should be removed from the phi node. // Note there could be multiple edges to the resume block, and we need // to remove them all. while (PhiLPInst->getBasicBlockIndex(TrivialBB) != -1) BB->removePredecessor(TrivialBB, true); for (BasicBlock *Pred : llvm::make_early_inc_range(predecessors(TrivialBB))) { removeUnwindEdge(Pred, DTU); ++NumInvokes; } // In each SimplifyCFG run, only the current processed block can be erased. // Otherwise, it will break the iteration of SimplifyCFG pass. So instead // of erasing TrivialBB, we only remove the branch to the common resume // block so that we can later erase the resume block since it has no // predecessors. TrivialBB->getTerminator()->eraseFromParent(); new UnreachableInst(RI->getContext(), TrivialBB); if (DTU) DTU->applyUpdates({{DominatorTree::Delete, TrivialBB, BB}}); } // Delete the resume block if all its predecessors have been removed. if (pred_empty(BB)) DeleteDeadBlock(BB, DTU); return !TrivialUnwindBlocks.empty(); } // Simplify resume that is only used by a single (non-phi) landing pad. bool SimplifyCFGOpt::simplifySingleResume(ResumeInst *RI) { BasicBlock *BB = RI->getParent(); auto *LPInst = cast(BB->getFirstNonPHI()); assert(RI->getValue() == LPInst && "Resume must unwind the exception that caused control to here"); // Check that there are no other instructions except for debug intrinsics. if (!isCleanupBlockEmpty( make_range(LPInst->getNextNode(), RI))) return false; // Turn all invokes that unwind here into calls and delete the basic block. for (BasicBlock *Pred : llvm::make_early_inc_range(predecessors(BB))) { removeUnwindEdge(Pred, DTU); ++NumInvokes; } // The landingpad is now unreachable. Zap it. DeleteDeadBlock(BB, DTU); return true; } static bool removeEmptyCleanup(CleanupReturnInst *RI, DomTreeUpdater *DTU) { // If this is a trivial cleanup pad that executes no instructions, it can be // eliminated. If the cleanup pad continues to the caller, any predecessor // that is an EH pad will be updated to continue to the caller and any // predecessor that terminates with an invoke instruction will have its invoke // instruction converted to a call instruction. If the cleanup pad being // simplified does not continue to the caller, each predecessor will be // updated to continue to the unwind destination of the cleanup pad being // simplified. BasicBlock *BB = RI->getParent(); CleanupPadInst *CPInst = RI->getCleanupPad(); if (CPInst->getParent() != BB) // This isn't an empty cleanup. return false; // We cannot kill the pad if it has multiple uses. This typically arises // from unreachable basic blocks. if (!CPInst->hasOneUse()) return false; // Check that there are no other instructions except for benign intrinsics. if (!isCleanupBlockEmpty( make_range(CPInst->getNextNode(), RI))) return false; // If the cleanup return we are simplifying unwinds to the caller, this will // set UnwindDest to nullptr. BasicBlock *UnwindDest = RI->getUnwindDest(); Instruction *DestEHPad = UnwindDest ? UnwindDest->getFirstNonPHI() : nullptr; // We're about to remove BB from the control flow. Before we do, sink any // PHINodes into the unwind destination. Doing this before changing the // control flow avoids some potentially slow checks, since we can currently // be certain that UnwindDest and BB have no common predecessors (since they // are both EH pads). if (UnwindDest) { // First, go through the PHI nodes in UnwindDest and update any nodes that // reference the block we are removing for (PHINode &DestPN : UnwindDest->phis()) { int Idx = DestPN.getBasicBlockIndex(BB); // Since BB unwinds to UnwindDest, it has to be in the PHI node. assert(Idx != -1); // This PHI node has an incoming value that corresponds to a control // path through the cleanup pad we are removing. If the incoming // value is in the cleanup pad, it must be a PHINode (because we // verified above that the block is otherwise empty). Otherwise, the // value is either a constant or a value that dominates the cleanup // pad being removed. // // Because BB and UnwindDest are both EH pads, all of their // predecessors must unwind to these blocks, and since no instruction // can have multiple unwind destinations, there will be no overlap in // incoming blocks between SrcPN and DestPN. Value *SrcVal = DestPN.getIncomingValue(Idx); PHINode *SrcPN = dyn_cast(SrcVal); bool NeedPHITranslation = SrcPN && SrcPN->getParent() == BB; for (auto *Pred : predecessors(BB)) { Value *Incoming = NeedPHITranslation ? SrcPN->getIncomingValueForBlock(Pred) : SrcVal; DestPN.addIncoming(Incoming, Pred); } } // Sink any remaining PHI nodes directly into UnwindDest. Instruction *InsertPt = DestEHPad; for (PHINode &PN : make_early_inc_range(BB->phis())) { if (PN.use_empty() || !PN.isUsedOutsideOfBlock(BB)) // If the PHI node has no uses or all of its uses are in this basic // block (meaning they are debug or lifetime intrinsics), just leave // it. It will be erased when we erase BB below. continue; // Otherwise, sink this PHI node into UnwindDest. // Any predecessors to UnwindDest which are not already represented // must be back edges which inherit the value from the path through // BB. In this case, the PHI value must reference itself. for (auto *pred : predecessors(UnwindDest)) if (pred != BB) PN.addIncoming(&PN, pred); PN.moveBefore(InsertPt); // Also, add a dummy incoming value for the original BB itself, // so that the PHI is well-formed until we drop said predecessor. PN.addIncoming(PoisonValue::get(PN.getType()), BB); } } std::vector Updates; // We use make_early_inc_range here because we will remove all predecessors. for (BasicBlock *PredBB : llvm::make_early_inc_range(predecessors(BB))) { if (UnwindDest == nullptr) { if (DTU) { DTU->applyUpdates(Updates); Updates.clear(); } removeUnwindEdge(PredBB, DTU); ++NumInvokes; } else { BB->removePredecessor(PredBB); Instruction *TI = PredBB->getTerminator(); TI->replaceUsesOfWith(BB, UnwindDest); if (DTU) { Updates.push_back({DominatorTree::Insert, PredBB, UnwindDest}); Updates.push_back({DominatorTree::Delete, PredBB, BB}); } } } if (DTU) DTU->applyUpdates(Updates); DeleteDeadBlock(BB, DTU); return true; } // Try to merge two cleanuppads together. static bool mergeCleanupPad(CleanupReturnInst *RI) { // Skip any cleanuprets which unwind to caller, there is nothing to merge // with. BasicBlock *UnwindDest = RI->getUnwindDest(); if (!UnwindDest) return false; // This cleanupret isn't the only predecessor of this cleanuppad, it wouldn't // be safe to merge without code duplication. if (UnwindDest->getSinglePredecessor() != RI->getParent()) return false; // Verify that our cleanuppad's unwind destination is another cleanuppad. auto *SuccessorCleanupPad = dyn_cast(&UnwindDest->front()); if (!SuccessorCleanupPad) return false; CleanupPadInst *PredecessorCleanupPad = RI->getCleanupPad(); // Replace any uses of the successor cleanupad with the predecessor pad // The only cleanuppad uses should be this cleanupret, it's cleanupret and // funclet bundle operands. SuccessorCleanupPad->replaceAllUsesWith(PredecessorCleanupPad); // Remove the old cleanuppad. SuccessorCleanupPad->eraseFromParent(); // Now, we simply replace the cleanupret with a branch to the unwind // destination. BranchInst::Create(UnwindDest, RI->getParent()); RI->eraseFromParent(); return true; } bool SimplifyCFGOpt::simplifyCleanupReturn(CleanupReturnInst *RI) { // It is possible to transiantly have an undef cleanuppad operand because we // have deleted some, but not all, dead blocks. // Eventually, this block will be deleted. if (isa(RI->getOperand(0))) return false; if (mergeCleanupPad(RI)) return true; if (removeEmptyCleanup(RI, DTU)) return true; return false; } // WARNING: keep in sync with InstCombinerImpl::visitUnreachableInst()! bool SimplifyCFGOpt::simplifyUnreachable(UnreachableInst *UI) { BasicBlock *BB = UI->getParent(); bool Changed = false; // Ensure that any debug-info records that used to occur after the Unreachable // are moved to in front of it -- otherwise they'll "dangle" at the end of // the block. BB->flushTerminatorDbgRecords(); // Debug-info records on the unreachable inst itself should be deleted, as // below we delete everything past the final executable instruction. UI->dropDbgRecords(); // If there are any instructions immediately before the unreachable that can // be removed, do so. while (UI->getIterator() != BB->begin()) { BasicBlock::iterator BBI = UI->getIterator(); --BBI; if (!isGuaranteedToTransferExecutionToSuccessor(&*BBI)) break; // Can not drop any more instructions. We're done here. // Otherwise, this instruction can be freely erased, // even if it is not side-effect free. // Note that deleting EH's here is in fact okay, although it involves a bit // of subtle reasoning. If this inst is an EH, all the predecessors of this // block will be the unwind edges of Invoke/CatchSwitch/CleanupReturn, // and we can therefore guarantee this block will be erased. // If we're deleting this, we're deleting any subsequent debug info, so // delete DbgRecords. BBI->dropDbgRecords(); // Delete this instruction (any uses are guaranteed to be dead) BBI->replaceAllUsesWith(PoisonValue::get(BBI->getType())); BBI->eraseFromParent(); Changed = true; } // If the unreachable instruction is the first in the block, take a gander // at all of the predecessors of this instruction, and simplify them. if (&BB->front() != UI) return Changed; std::vector Updates; SmallSetVector Preds(pred_begin(BB), pred_end(BB)); for (BasicBlock *Predecessor : Preds) { Instruction *TI = Predecessor->getTerminator(); IRBuilder<> Builder(TI); if (auto *BI = dyn_cast(TI)) { // We could either have a proper unconditional branch, // or a degenerate conditional branch with matching destinations. if (all_of(BI->successors(), [BB](auto *Successor) { return Successor == BB; })) { new UnreachableInst(TI->getContext(), TI->getIterator()); TI->eraseFromParent(); Changed = true; } else { assert(BI->isConditional() && "Can't get here with an uncond branch."); Value* Cond = BI->getCondition(); assert(BI->getSuccessor(0) != BI->getSuccessor(1) && "The destinations are guaranteed to be different here."); CallInst *Assumption; if (BI->getSuccessor(0) == BB) { Assumption = Builder.CreateAssumption(Builder.CreateNot(Cond)); Builder.CreateBr(BI->getSuccessor(1)); } else { assert(BI->getSuccessor(1) == BB && "Incorrect CFG"); Assumption = Builder.CreateAssumption(Cond); Builder.CreateBr(BI->getSuccessor(0)); } if (Options.AC) Options.AC->registerAssumption(cast(Assumption)); EraseTerminatorAndDCECond(BI); Changed = true; } if (DTU) Updates.push_back({DominatorTree::Delete, Predecessor, BB}); } else if (auto *SI = dyn_cast(TI)) { SwitchInstProfUpdateWrapper SU(*SI); for (auto i = SU->case_begin(), e = SU->case_end(); i != e;) { if (i->getCaseSuccessor() != BB) { ++i; continue; } BB->removePredecessor(SU->getParent()); i = SU.removeCase(i); e = SU->case_end(); Changed = true; } // Note that the default destination can't be removed! if (DTU && SI->getDefaultDest() != BB) Updates.push_back({DominatorTree::Delete, Predecessor, BB}); } else if (auto *II = dyn_cast(TI)) { if (II->getUnwindDest() == BB) { if (DTU) { DTU->applyUpdates(Updates); Updates.clear(); } auto *CI = cast(removeUnwindEdge(TI->getParent(), DTU)); if (!CI->doesNotThrow()) CI->setDoesNotThrow(); Changed = true; } } else if (auto *CSI = dyn_cast(TI)) { if (CSI->getUnwindDest() == BB) { if (DTU) { DTU->applyUpdates(Updates); Updates.clear(); } removeUnwindEdge(TI->getParent(), DTU); Changed = true; continue; } for (CatchSwitchInst::handler_iterator I = CSI->handler_begin(), E = CSI->handler_end(); I != E; ++I) { if (*I == BB) { CSI->removeHandler(I); --I; --E; Changed = true; } } if (DTU) Updates.push_back({DominatorTree::Delete, Predecessor, BB}); if (CSI->getNumHandlers() == 0) { if (CSI->hasUnwindDest()) { // Redirect all predecessors of the block containing CatchSwitchInst // to instead branch to the CatchSwitchInst's unwind destination. if (DTU) { for (auto *PredecessorOfPredecessor : predecessors(Predecessor)) { Updates.push_back({DominatorTree::Insert, PredecessorOfPredecessor, CSI->getUnwindDest()}); Updates.push_back({DominatorTree::Delete, PredecessorOfPredecessor, Predecessor}); } } Predecessor->replaceAllUsesWith(CSI->getUnwindDest()); } else { // Rewrite all preds to unwind to caller (or from invoke to call). if (DTU) { DTU->applyUpdates(Updates); Updates.clear(); } SmallVector EHPreds(predecessors(Predecessor)); for (BasicBlock *EHPred : EHPreds) removeUnwindEdge(EHPred, DTU); } // The catchswitch is no longer reachable. new UnreachableInst(CSI->getContext(), CSI->getIterator()); CSI->eraseFromParent(); Changed = true; } } else if (auto *CRI = dyn_cast(TI)) { (void)CRI; assert(CRI->hasUnwindDest() && CRI->getUnwindDest() == BB && "Expected to always have an unwind to BB."); if (DTU) Updates.push_back({DominatorTree::Delete, Predecessor, BB}); new UnreachableInst(TI->getContext(), TI->getIterator()); TI->eraseFromParent(); Changed = true; } } if (DTU) DTU->applyUpdates(Updates); // If this block is now dead, remove it. if (pred_empty(BB) && BB != &BB->getParent()->getEntryBlock()) { DeleteDeadBlock(BB, DTU); return true; } return Changed; } static bool CasesAreContiguous(SmallVectorImpl &Cases) { assert(Cases.size() >= 1); array_pod_sort(Cases.begin(), Cases.end(), ConstantIntSortPredicate); for (size_t I = 1, E = Cases.size(); I != E; ++I) { if (Cases[I - 1]->getValue() != Cases[I]->getValue() + 1) return false; } return true; } static void createUnreachableSwitchDefault(SwitchInst *Switch, DomTreeUpdater *DTU, bool RemoveOrigDefaultBlock = true) { LLVM_DEBUG(dbgs() << "SimplifyCFG: switch default is dead.\n"); auto *BB = Switch->getParent(); auto *OrigDefaultBlock = Switch->getDefaultDest(); if (RemoveOrigDefaultBlock) OrigDefaultBlock->removePredecessor(BB); BasicBlock *NewDefaultBlock = BasicBlock::Create( BB->getContext(), BB->getName() + ".unreachabledefault", BB->getParent(), OrigDefaultBlock); new UnreachableInst(Switch->getContext(), NewDefaultBlock); Switch->setDefaultDest(&*NewDefaultBlock); if (DTU) { SmallVector Updates; Updates.push_back({DominatorTree::Insert, BB, &*NewDefaultBlock}); if (RemoveOrigDefaultBlock && !is_contained(successors(BB), OrigDefaultBlock)) Updates.push_back({DominatorTree::Delete, BB, &*OrigDefaultBlock}); DTU->applyUpdates(Updates); } } /// Turn a switch into an integer range comparison and branch. /// Switches with more than 2 destinations are ignored. /// Switches with 1 destination are also ignored. bool SimplifyCFGOpt::TurnSwitchRangeIntoICmp(SwitchInst *SI, IRBuilder<> &Builder) { assert(SI->getNumCases() > 1 && "Degenerate switch?"); bool HasDefault = !isa(SI->getDefaultDest()->getFirstNonPHIOrDbg()); auto *BB = SI->getParent(); // Partition the cases into two sets with different destinations. BasicBlock *DestA = HasDefault ? SI->getDefaultDest() : nullptr; BasicBlock *DestB = nullptr; SmallVector CasesA; SmallVector CasesB; for (auto Case : SI->cases()) { BasicBlock *Dest = Case.getCaseSuccessor(); if (!DestA) DestA = Dest; if (Dest == DestA) { CasesA.push_back(Case.getCaseValue()); continue; } if (!DestB) DestB = Dest; if (Dest == DestB) { CasesB.push_back(Case.getCaseValue()); continue; } return false; // More than two destinations. } if (!DestB) return false; // All destinations are the same and the default is unreachable assert(DestA && DestB && "Single-destination switch should have been folded."); assert(DestA != DestB); assert(DestB != SI->getDefaultDest()); assert(!CasesB.empty() && "There must be non-default cases."); assert(!CasesA.empty() || HasDefault); // Figure out if one of the sets of cases form a contiguous range. SmallVectorImpl *ContiguousCases = nullptr; BasicBlock *ContiguousDest = nullptr; BasicBlock *OtherDest = nullptr; if (!CasesA.empty() && CasesAreContiguous(CasesA)) { ContiguousCases = &CasesA; ContiguousDest = DestA; OtherDest = DestB; } else if (CasesAreContiguous(CasesB)) { ContiguousCases = &CasesB; ContiguousDest = DestB; OtherDest = DestA; } else return false; // Start building the compare and branch. Constant *Offset = ConstantExpr::getNeg(ContiguousCases->back()); Constant *NumCases = ConstantInt::get(Offset->getType(), ContiguousCases->size()); Value *Sub = SI->getCondition(); if (!Offset->isNullValue()) Sub = Builder.CreateAdd(Sub, Offset, Sub->getName() + ".off"); Value *Cmp; // If NumCases overflowed, then all possible values jump to the successor. if (NumCases->isNullValue() && !ContiguousCases->empty()) Cmp = ConstantInt::getTrue(SI->getContext()); else Cmp = Builder.CreateICmpULT(Sub, NumCases, "switch"); BranchInst *NewBI = Builder.CreateCondBr(Cmp, ContiguousDest, OtherDest); // Update weight for the newly-created conditional branch. if (hasBranchWeightMD(*SI)) { SmallVector Weights; GetBranchWeights(SI, Weights); if (Weights.size() == 1 + SI->getNumCases()) { uint64_t TrueWeight = 0; uint64_t FalseWeight = 0; for (size_t I = 0, E = Weights.size(); I != E; ++I) { if (SI->getSuccessor(I) == ContiguousDest) TrueWeight += Weights[I]; else FalseWeight += Weights[I]; } while (TrueWeight > UINT32_MAX || FalseWeight > UINT32_MAX) { TrueWeight /= 2; FalseWeight /= 2; } setBranchWeights(NewBI, TrueWeight, FalseWeight, /*IsExpected=*/false); } } // Prune obsolete incoming values off the successors' PHI nodes. for (auto BBI = ContiguousDest->begin(); isa(BBI); ++BBI) { unsigned PreviousEdges = ContiguousCases->size(); if (ContiguousDest == SI->getDefaultDest()) ++PreviousEdges; for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I) cast(BBI)->removeIncomingValue(SI->getParent()); } for (auto BBI = OtherDest->begin(); isa(BBI); ++BBI) { unsigned PreviousEdges = SI->getNumCases() - ContiguousCases->size(); if (OtherDest == SI->getDefaultDest()) ++PreviousEdges; for (unsigned I = 0, E = PreviousEdges - 1; I != E; ++I) cast(BBI)->removeIncomingValue(SI->getParent()); } // Clean up the default block - it may have phis or other instructions before // the unreachable terminator. if (!HasDefault) createUnreachableSwitchDefault(SI, DTU); auto *UnreachableDefault = SI->getDefaultDest(); // Drop the switch. SI->eraseFromParent(); if (!HasDefault && DTU) DTU->applyUpdates({{DominatorTree::Delete, BB, UnreachableDefault}}); return true; } /// Compute masked bits for the condition of a switch /// and use it to remove dead cases. static bool eliminateDeadSwitchCases(SwitchInst *SI, DomTreeUpdater *DTU, AssumptionCache *AC, const DataLayout &DL) { Value *Cond = SI->getCondition(); KnownBits Known = computeKnownBits(Cond, DL, 0, AC, SI); // We can also eliminate cases by determining that their values are outside of // the limited range of the condition based on how many significant (non-sign) // bits are in the condition value. unsigned MaxSignificantBitsInCond = ComputeMaxSignificantBits(Cond, DL, 0, AC, SI); // Gather dead cases. SmallVector DeadCases; SmallDenseMap NumPerSuccessorCases; SmallVector UniqueSuccessors; for (const auto &Case : SI->cases()) { auto *Successor = Case.getCaseSuccessor(); if (DTU) { if (!NumPerSuccessorCases.count(Successor)) UniqueSuccessors.push_back(Successor); ++NumPerSuccessorCases[Successor]; } const APInt &CaseVal = Case.getCaseValue()->getValue(); if (Known.Zero.intersects(CaseVal) || !Known.One.isSubsetOf(CaseVal) || (CaseVal.getSignificantBits() > MaxSignificantBitsInCond)) { DeadCases.push_back(Case.getCaseValue()); if (DTU) --NumPerSuccessorCases[Successor]; LLVM_DEBUG(dbgs() << "SimplifyCFG: switch case " << CaseVal << " is dead.\n"); } } // If we can prove that the cases must cover all possible values, the // default destination becomes dead and we can remove it. If we know some // of the bits in the value, we can use that to more precisely compute the // number of possible unique case values. bool HasDefault = !isa(SI->getDefaultDest()->getFirstNonPHIOrDbg()); const unsigned NumUnknownBits = Known.getBitWidth() - (Known.Zero | Known.One).popcount(); assert(NumUnknownBits <= Known.getBitWidth()); if (HasDefault && DeadCases.empty() && NumUnknownBits < 64 /* avoid overflow */) { uint64_t AllNumCases = 1ULL << NumUnknownBits; if (SI->getNumCases() == AllNumCases) { createUnreachableSwitchDefault(SI, DTU); return true; } // When only one case value is missing, replace default with that case. // Eliminating the default branch will provide more opportunities for // optimization, such as lookup tables. if (SI->getNumCases() == AllNumCases - 1) { assert(NumUnknownBits > 1 && "Should be canonicalized to a branch"); IntegerType *CondTy = cast(Cond->getType()); if (CondTy->getIntegerBitWidth() > 64 || !DL.fitsInLegalInteger(CondTy->getIntegerBitWidth())) return false; uint64_t MissingCaseVal = 0; for (const auto &Case : SI->cases()) MissingCaseVal ^= Case.getCaseValue()->getValue().getLimitedValue(); auto *MissingCase = cast(ConstantInt::get(Cond->getType(), MissingCaseVal)); SwitchInstProfUpdateWrapper SIW(*SI); SIW.addCase(MissingCase, SI->getDefaultDest(), SIW.getSuccessorWeight(0)); createUnreachableSwitchDefault(SI, DTU, /*RemoveOrigDefaultBlock*/ false); SIW.setSuccessorWeight(0, 0); return true; } } if (DeadCases.empty()) return false; SwitchInstProfUpdateWrapper SIW(*SI); for (ConstantInt *DeadCase : DeadCases) { SwitchInst::CaseIt CaseI = SI->findCaseValue(DeadCase); assert(CaseI != SI->case_default() && "Case was not found. Probably mistake in DeadCases forming."); // Prune unused values from PHI nodes. CaseI->getCaseSuccessor()->removePredecessor(SI->getParent()); SIW.removeCase(CaseI); } if (DTU) { std::vector Updates; for (auto *Successor : UniqueSuccessors) if (NumPerSuccessorCases[Successor] == 0) Updates.push_back({DominatorTree::Delete, SI->getParent(), Successor}); DTU->applyUpdates(Updates); } return true; } /// If BB would be eligible for simplification by /// TryToSimplifyUncondBranchFromEmptyBlock (i.e. it is empty and terminated /// by an unconditional branch), look at the phi node for BB in the successor /// block and see if the incoming value is equal to CaseValue. If so, return /// the phi node, and set PhiIndex to BB's index in the phi node. static PHINode *FindPHIForConditionForwarding(ConstantInt *CaseValue, BasicBlock *BB, int *PhiIndex) { if (BB->getFirstNonPHIOrDbg() != BB->getTerminator()) return nullptr; // BB must be empty to be a candidate for simplification. if (!BB->getSinglePredecessor()) return nullptr; // BB must be dominated by the switch. BranchInst *Branch = dyn_cast(BB->getTerminator()); if (!Branch || !Branch->isUnconditional()) return nullptr; // Terminator must be unconditional branch. BasicBlock *Succ = Branch->getSuccessor(0); for (PHINode &PHI : Succ->phis()) { int Idx = PHI.getBasicBlockIndex(BB); assert(Idx >= 0 && "PHI has no entry for predecessor?"); Value *InValue = PHI.getIncomingValue(Idx); if (InValue != CaseValue) continue; *PhiIndex = Idx; return &PHI; } return nullptr; } /// Try to forward the condition of a switch instruction to a phi node /// dominated by the switch, if that would mean that some of the destination /// blocks of the switch can be folded away. Return true if a change is made. static bool ForwardSwitchConditionToPHI(SwitchInst *SI) { using ForwardingNodesMap = DenseMap>; ForwardingNodesMap ForwardingNodes; BasicBlock *SwitchBlock = SI->getParent(); bool Changed = false; for (const auto &Case : SI->cases()) { ConstantInt *CaseValue = Case.getCaseValue(); BasicBlock *CaseDest = Case.getCaseSuccessor(); // Replace phi operands in successor blocks that are using the constant case // value rather than the switch condition variable: // switchbb: // switch i32 %x, label %default [ // i32 17, label %succ // ... // succ: // %r = phi i32 ... [ 17, %switchbb ] ... // --> // %r = phi i32 ... [ %x, %switchbb ] ... for (PHINode &Phi : CaseDest->phis()) { // This only works if there is exactly 1 incoming edge from the switch to // a phi. If there is >1, that means multiple cases of the switch map to 1 // value in the phi, and that phi value is not the switch condition. Thus, // this transform would not make sense (the phi would be invalid because // a phi can't have different incoming values from the same block). int SwitchBBIdx = Phi.getBasicBlockIndex(SwitchBlock); if (Phi.getIncomingValue(SwitchBBIdx) == CaseValue && count(Phi.blocks(), SwitchBlock) == 1) { Phi.setIncomingValue(SwitchBBIdx, SI->getCondition()); Changed = true; } } // Collect phi nodes that are indirectly using this switch's case constants. int PhiIdx; if (auto *Phi = FindPHIForConditionForwarding(CaseValue, CaseDest, &PhiIdx)) ForwardingNodes[Phi].push_back(PhiIdx); } for (auto &ForwardingNode : ForwardingNodes) { PHINode *Phi = ForwardingNode.first; SmallVectorImpl &Indexes = ForwardingNode.second; // Check if it helps to fold PHI. if (Indexes.size() < 2 && !llvm::is_contained(Phi->incoming_values(), SI->getCondition())) continue; for (int Index : Indexes) Phi->setIncomingValue(Index, SI->getCondition()); Changed = true; } return Changed; } /// Return true if the backend will be able to handle /// initializing an array of constants like C. static bool ValidLookupTableConstant(Constant *C, const TargetTransformInfo &TTI) { if (C->isThreadDependent()) return false; if (C->isDLLImportDependent()) return false; if (!isa(C) && !isa(C) && !isa(C) && !isa(C) && !isa(C) && !isa(C)) return false; if (ConstantExpr *CE = dyn_cast(C)) { // Pointer casts and in-bounds GEPs will not prohibit the backend from // materializing the array of constants. Constant *StrippedC = cast(CE->stripInBoundsConstantOffsets()); if (StrippedC == C || !ValidLookupTableConstant(StrippedC, TTI)) return false; } if (!TTI.shouldBuildLookupTablesForConstant(C)) return false; return true; } /// If V is a Constant, return it. Otherwise, try to look up /// its constant value in ConstantPool, returning 0 if it's not there. static Constant * LookupConstant(Value *V, const SmallDenseMap &ConstantPool) { if (Constant *C = dyn_cast(V)) return C; return ConstantPool.lookup(V); } /// Try to fold instruction I into a constant. This works for /// simple instructions such as binary operations where both operands are /// constant or can be replaced by constants from the ConstantPool. Returns the /// resulting constant on success, 0 otherwise. static Constant * ConstantFold(Instruction *I, const DataLayout &DL, const SmallDenseMap &ConstantPool) { if (SelectInst *Select = dyn_cast(I)) { Constant *A = LookupConstant(Select->getCondition(), ConstantPool); if (!A) return nullptr; if (A->isAllOnesValue()) return LookupConstant(Select->getTrueValue(), ConstantPool); if (A->isNullValue()) return LookupConstant(Select->getFalseValue(), ConstantPool); return nullptr; } SmallVector COps; for (unsigned N = 0, E = I->getNumOperands(); N != E; ++N) { if (Constant *A = LookupConstant(I->getOperand(N), ConstantPool)) COps.push_back(A); else return nullptr; } return ConstantFoldInstOperands(I, COps, DL); } /// Try to determine the resulting constant values in phi nodes /// at the common destination basic block, *CommonDest, for one of the case /// destionations CaseDest corresponding to value CaseVal (0 for the default /// case), of a switch instruction SI. static bool getCaseResults(SwitchInst *SI, ConstantInt *CaseVal, BasicBlock *CaseDest, BasicBlock **CommonDest, SmallVectorImpl> &Res, const DataLayout &DL, const TargetTransformInfo &TTI) { // The block from which we enter the common destination. BasicBlock *Pred = SI->getParent(); // If CaseDest is empty except for some side-effect free instructions through // which we can constant-propagate the CaseVal, continue to its successor. SmallDenseMap ConstantPool; ConstantPool.insert(std::make_pair(SI->getCondition(), CaseVal)); for (Instruction &I : CaseDest->instructionsWithoutDebug(false)) { if (I.isTerminator()) { // If the terminator is a simple branch, continue to the next block. if (I.getNumSuccessors() != 1 || I.isSpecialTerminator()) return false; Pred = CaseDest; CaseDest = I.getSuccessor(0); } else if (Constant *C = ConstantFold(&I, DL, ConstantPool)) { // Instruction is side-effect free and constant. // If the instruction has uses outside this block or a phi node slot for // the block, it is not safe to bypass the instruction since it would then // no longer dominate all its uses. for (auto &Use : I.uses()) { User *User = Use.getUser(); if (Instruction *I = dyn_cast(User)) if (I->getParent() == CaseDest) continue; if (PHINode *Phi = dyn_cast(User)) if (Phi->getIncomingBlock(Use) == CaseDest) continue; return false; } ConstantPool.insert(std::make_pair(&I, C)); } else { break; } } // If we did not have a CommonDest before, use the current one. if (!*CommonDest) *CommonDest = CaseDest; // If the destination isn't the common one, abort. if (CaseDest != *CommonDest) return false; // Get the values for this case from phi nodes in the destination block. for (PHINode &PHI : (*CommonDest)->phis()) { int Idx = PHI.getBasicBlockIndex(Pred); if (Idx == -1) continue; Constant *ConstVal = LookupConstant(PHI.getIncomingValue(Idx), ConstantPool); if (!ConstVal) return false; // Be conservative about which kinds of constants we support. if (!ValidLookupTableConstant(ConstVal, TTI)) return false; Res.push_back(std::make_pair(&PHI, ConstVal)); } return Res.size() > 0; } // Helper function used to add CaseVal to the list of cases that generate // Result. Returns the updated number of cases that generate this result. static size_t mapCaseToResult(ConstantInt *CaseVal, SwitchCaseResultVectorTy &UniqueResults, Constant *Result) { for (auto &I : UniqueResults) { if (I.first == Result) { I.second.push_back(CaseVal); return I.second.size(); } } UniqueResults.push_back( std::make_pair(Result, SmallVector(1, CaseVal))); return 1; } // Helper function that initializes a map containing // results for the PHI node of the common destination block for a switch // instruction. Returns false if multiple PHI nodes have been found or if // there is not a common destination block for the switch. static bool initializeUniqueCases(SwitchInst *SI, PHINode *&PHI, BasicBlock *&CommonDest, SwitchCaseResultVectorTy &UniqueResults, Constant *&DefaultResult, const DataLayout &DL, const TargetTransformInfo &TTI, uintptr_t MaxUniqueResults) { for (const auto &I : SI->cases()) { ConstantInt *CaseVal = I.getCaseValue(); // Resulting value at phi nodes for this case value. SwitchCaseResultsTy Results; if (!getCaseResults(SI, CaseVal, I.getCaseSuccessor(), &CommonDest, Results, DL, TTI)) return false; // Only one value per case is permitted. if (Results.size() > 1) return false; // Add the case->result mapping to UniqueResults. const size_t NumCasesForResult = mapCaseToResult(CaseVal, UniqueResults, Results.begin()->second); // Early out if there are too many cases for this result. if (NumCasesForResult > MaxSwitchCasesPerResult) return false; // Early out if there are too many unique results. if (UniqueResults.size() > MaxUniqueResults) return false; // Check the PHI consistency. if (!PHI) PHI = Results[0].first; else if (PHI != Results[0].first) return false; } // Find the default result value. SmallVector, 1> DefaultResults; BasicBlock *DefaultDest = SI->getDefaultDest(); getCaseResults(SI, nullptr, SI->getDefaultDest(), &CommonDest, DefaultResults, DL, TTI); // If the default value is not found abort unless the default destination // is unreachable. DefaultResult = DefaultResults.size() == 1 ? DefaultResults.begin()->second : nullptr; if ((!DefaultResult && !isa(DefaultDest->getFirstNonPHIOrDbg()))) return false; return true; } // Helper function that checks if it is possible to transform a switch with only // two cases (or two cases + default) that produces a result into a select. // TODO: Handle switches with more than 2 cases that map to the same result. static Value *foldSwitchToSelect(const SwitchCaseResultVectorTy &ResultVector, Constant *DefaultResult, Value *Condition, IRBuilder<> &Builder) { // If we are selecting between only two cases transform into a simple // select or a two-way select if default is possible. // Example: // switch (a) { %0 = icmp eq i32 %a, 10 // case 10: return 42; %1 = select i1 %0, i32 42, i32 4 // case 20: return 2; ----> %2 = icmp eq i32 %a, 20 // default: return 4; %3 = select i1 %2, i32 2, i32 %1 // } if (ResultVector.size() == 2 && ResultVector[0].second.size() == 1 && ResultVector[1].second.size() == 1) { ConstantInt *FirstCase = ResultVector[0].second[0]; ConstantInt *SecondCase = ResultVector[1].second[0]; Value *SelectValue = ResultVector[1].first; if (DefaultResult) { Value *ValueCompare = Builder.CreateICmpEQ(Condition, SecondCase, "switch.selectcmp"); SelectValue = Builder.CreateSelect(ValueCompare, ResultVector[1].first, DefaultResult, "switch.select"); } Value *ValueCompare = Builder.CreateICmpEQ(Condition, FirstCase, "switch.selectcmp"); return Builder.CreateSelect(ValueCompare, ResultVector[0].first, SelectValue, "switch.select"); } // Handle the degenerate case where two cases have the same result value. if (ResultVector.size() == 1 && DefaultResult) { ArrayRef CaseValues = ResultVector[0].second; unsigned CaseCount = CaseValues.size(); // n bits group cases map to the same result: // case 0,4 -> Cond & 0b1..1011 == 0 ? result : default // case 0,2,4,6 -> Cond & 0b1..1001 == 0 ? result : default // case 0,2,8,10 -> Cond & 0b1..0101 == 0 ? result : default if (isPowerOf2_32(CaseCount)) { ConstantInt *MinCaseVal = CaseValues[0]; // Find mininal value. for (auto *Case : CaseValues) if (Case->getValue().slt(MinCaseVal->getValue())) MinCaseVal = Case; // Mark the bits case number touched. APInt BitMask = APInt::getZero(MinCaseVal->getBitWidth()); for (auto *Case : CaseValues) BitMask |= (Case->getValue() - MinCaseVal->getValue()); // Check if cases with the same result can cover all number // in touched bits. if (BitMask.popcount() == Log2_32(CaseCount)) { if (!MinCaseVal->isNullValue()) Condition = Builder.CreateSub(Condition, MinCaseVal); Value *And = Builder.CreateAnd(Condition, ~BitMask, "switch.and"); Value *Cmp = Builder.CreateICmpEQ( And, Constant::getNullValue(And->getType()), "switch.selectcmp"); return Builder.CreateSelect(Cmp, ResultVector[0].first, DefaultResult); } } // Handle the degenerate case where two cases have the same value. if (CaseValues.size() == 2) { Value *Cmp1 = Builder.CreateICmpEQ(Condition, CaseValues[0], "switch.selectcmp.case1"); Value *Cmp2 = Builder.CreateICmpEQ(Condition, CaseValues[1], "switch.selectcmp.case2"); Value *Cmp = Builder.CreateOr(Cmp1, Cmp2, "switch.selectcmp"); return Builder.CreateSelect(Cmp, ResultVector[0].first, DefaultResult); } } return nullptr; } // Helper function to cleanup a switch instruction that has been converted into // a select, fixing up PHI nodes and basic blocks. static void removeSwitchAfterSelectFold(SwitchInst *SI, PHINode *PHI, Value *SelectValue, IRBuilder<> &Builder, DomTreeUpdater *DTU) { std::vector Updates; BasicBlock *SelectBB = SI->getParent(); BasicBlock *DestBB = PHI->getParent(); if (DTU && !is_contained(predecessors(DestBB), SelectBB)) Updates.push_back({DominatorTree::Insert, SelectBB, DestBB}); Builder.CreateBr(DestBB); // Remove the switch. PHI->removeIncomingValueIf( [&](unsigned Idx) { return PHI->getIncomingBlock(Idx) == SelectBB; }); PHI->addIncoming(SelectValue, SelectBB); SmallPtrSet RemovedSuccessors; for (unsigned i = 0, e = SI->getNumSuccessors(); i < e; ++i) { BasicBlock *Succ = SI->getSuccessor(i); if (Succ == DestBB) continue; Succ->removePredecessor(SelectBB); if (DTU && RemovedSuccessors.insert(Succ).second) Updates.push_back({DominatorTree::Delete, SelectBB, Succ}); } SI->eraseFromParent(); if (DTU) DTU->applyUpdates(Updates); } /// If a switch is only used to initialize one or more phi nodes in a common /// successor block with only two different constant values, try to replace the /// switch with a select. Returns true if the fold was made. static bool trySwitchToSelect(SwitchInst *SI, IRBuilder<> &Builder, DomTreeUpdater *DTU, const DataLayout &DL, const TargetTransformInfo &TTI) { Value *const Cond = SI->getCondition(); PHINode *PHI = nullptr; BasicBlock *CommonDest = nullptr; Constant *DefaultResult; SwitchCaseResultVectorTy UniqueResults; // Collect all the cases that will deliver the same value from the switch. if (!initializeUniqueCases(SI, PHI, CommonDest, UniqueResults, DefaultResult, DL, TTI, /*MaxUniqueResults*/ 2)) return false; assert(PHI != nullptr && "PHI for value select not found"); Builder.SetInsertPoint(SI); Value *SelectValue = foldSwitchToSelect(UniqueResults, DefaultResult, Cond, Builder); if (!SelectValue) return false; removeSwitchAfterSelectFold(SI, PHI, SelectValue, Builder, DTU); return true; } namespace { /// This class represents a lookup table that can be used to replace a switch. class SwitchLookupTable { public: /// Create a lookup table to use as a switch replacement with the contents /// of Values, using DefaultValue to fill any holes in the table. SwitchLookupTable( Module &M, uint64_t TableSize, ConstantInt *Offset, const SmallVectorImpl> &Values, Constant *DefaultValue, const DataLayout &DL, const StringRef &FuncName); /// Build instructions with Builder to retrieve the value at /// the position given by Index in the lookup table. Value *BuildLookup(Value *Index, IRBuilder<> &Builder); /// Return true if a table with TableSize elements of /// type ElementType would fit in a target-legal register. static bool WouldFitInRegister(const DataLayout &DL, uint64_t TableSize, Type *ElementType); private: // Depending on the contents of the table, it can be represented in // different ways. enum { // For tables where each element contains the same value, we just have to // store that single value and return it for each lookup. SingleValueKind, // For tables where there is a linear relationship between table index // and values. We calculate the result with a simple multiplication // and addition instead of a table lookup. LinearMapKind, // For small tables with integer elements, we can pack them into a bitmap // that fits into a target-legal register. Values are retrieved by // shift and mask operations. BitMapKind, // The table is stored as an array of values. Values are retrieved by load // instructions from the table. ArrayKind } Kind; // For SingleValueKind, this is the single value. Constant *SingleValue = nullptr; // For BitMapKind, this is the bitmap. ConstantInt *BitMap = nullptr; IntegerType *BitMapElementTy = nullptr; // For LinearMapKind, these are the constants used to derive the value. ConstantInt *LinearOffset = nullptr; ConstantInt *LinearMultiplier = nullptr; bool LinearMapValWrapped = false; // For ArrayKind, this is the array. GlobalVariable *Array = nullptr; }; } // end anonymous namespace SwitchLookupTable::SwitchLookupTable( Module &M, uint64_t TableSize, ConstantInt *Offset, const SmallVectorImpl> &Values, Constant *DefaultValue, const DataLayout &DL, const StringRef &FuncName) { assert(Values.size() && "Can't build lookup table without values!"); assert(TableSize >= Values.size() && "Can't fit values in table!"); // If all values in the table are equal, this is that value. SingleValue = Values.begin()->second; Type *ValueType = Values.begin()->second->getType(); // Build up the table contents. SmallVector TableContents(TableSize); for (size_t I = 0, E = Values.size(); I != E; ++I) { ConstantInt *CaseVal = Values[I].first; Constant *CaseRes = Values[I].second; assert(CaseRes->getType() == ValueType); uint64_t Idx = (CaseVal->getValue() - Offset->getValue()).getLimitedValue(); TableContents[Idx] = CaseRes; if (CaseRes != SingleValue) SingleValue = nullptr; } // Fill in any holes in the table with the default result. if (Values.size() < TableSize) { assert(DefaultValue && "Need a default value to fill the lookup table holes."); assert(DefaultValue->getType() == ValueType); for (uint64_t I = 0; I < TableSize; ++I) { if (!TableContents[I]) TableContents[I] = DefaultValue; } if (DefaultValue != SingleValue) SingleValue = nullptr; } // If each element in the table contains the same value, we only need to store // that single value. if (SingleValue) { Kind = SingleValueKind; return; } // Check if we can derive the value with a linear transformation from the // table index. if (isa(ValueType)) { bool LinearMappingPossible = true; APInt PrevVal; APInt DistToPrev; // When linear map is monotonic and signed overflow doesn't happen on // maximum index, we can attach nsw on Add and Mul. bool NonMonotonic = false; assert(TableSize >= 2 && "Should be a SingleValue table."); // Check if there is the same distance between two consecutive values. for (uint64_t I = 0; I < TableSize; ++I) { ConstantInt *ConstVal = dyn_cast(TableContents[I]); if (!ConstVal) { // This is an undef. We could deal with it, but undefs in lookup tables // are very seldom. It's probably not worth the additional complexity. LinearMappingPossible = false; break; } const APInt &Val = ConstVal->getValue(); if (I != 0) { APInt Dist = Val - PrevVal; if (I == 1) { DistToPrev = Dist; } else if (Dist != DistToPrev) { LinearMappingPossible = false; break; } NonMonotonic |= Dist.isStrictlyPositive() ? Val.sle(PrevVal) : Val.sgt(PrevVal); } PrevVal = Val; } if (LinearMappingPossible) { LinearOffset = cast(TableContents[0]); LinearMultiplier = ConstantInt::get(M.getContext(), DistToPrev); bool MayWrap = false; APInt M = LinearMultiplier->getValue(); (void)M.smul_ov(APInt(M.getBitWidth(), TableSize - 1), MayWrap); LinearMapValWrapped = NonMonotonic || MayWrap; Kind = LinearMapKind; ++NumLinearMaps; return; } } // If the type is integer and the table fits in a register, build a bitmap. if (WouldFitInRegister(DL, TableSize, ValueType)) { IntegerType *IT = cast(ValueType); APInt TableInt(TableSize * IT->getBitWidth(), 0); for (uint64_t I = TableSize; I > 0; --I) { TableInt <<= IT->getBitWidth(); // Insert values into the bitmap. Undef values are set to zero. if (!isa(TableContents[I - 1])) { ConstantInt *Val = cast(TableContents[I - 1]); TableInt |= Val->getValue().zext(TableInt.getBitWidth()); } } BitMap = ConstantInt::get(M.getContext(), TableInt); BitMapElementTy = IT; Kind = BitMapKind; ++NumBitMaps; return; } // Store the table in an array. ArrayType *ArrayTy = ArrayType::get(ValueType, TableSize); Constant *Initializer = ConstantArray::get(ArrayTy, TableContents); Array = new GlobalVariable(M, ArrayTy, /*isConstant=*/true, GlobalVariable::PrivateLinkage, Initializer, "switch.table." + FuncName); Array->setUnnamedAddr(GlobalValue::UnnamedAddr::Global); // Set the alignment to that of an array items. We will be only loading one // value out of it. Array->setAlignment(DL.getPrefTypeAlign(ValueType)); Kind = ArrayKind; } Value *SwitchLookupTable::BuildLookup(Value *Index, IRBuilder<> &Builder) { switch (Kind) { case SingleValueKind: return SingleValue; case LinearMapKind: { // Derive the result value from the input value. Value *Result = Builder.CreateIntCast(Index, LinearMultiplier->getType(), false, "switch.idx.cast"); if (!LinearMultiplier->isOne()) Result = Builder.CreateMul(Result, LinearMultiplier, "switch.idx.mult", /*HasNUW = */ false, /*HasNSW = */ !LinearMapValWrapped); if (!LinearOffset->isZero()) Result = Builder.CreateAdd(Result, LinearOffset, "switch.offset", /*HasNUW = */ false, /*HasNSW = */ !LinearMapValWrapped); return Result; } case BitMapKind: { // Type of the bitmap (e.g. i59). IntegerType *MapTy = BitMap->getIntegerType(); // Cast Index to the same type as the bitmap. // Note: The Index is <= the number of elements in the table, so // truncating it to the width of the bitmask is safe. Value *ShiftAmt = Builder.CreateZExtOrTrunc(Index, MapTy, "switch.cast"); // Multiply the shift amount by the element width. NUW/NSW can always be // set, because WouldFitInRegister guarantees Index * ShiftAmt is in // BitMap's bit width. ShiftAmt = Builder.CreateMul( ShiftAmt, ConstantInt::get(MapTy, BitMapElementTy->getBitWidth()), "switch.shiftamt",/*HasNUW =*/true,/*HasNSW =*/true); // Shift down. Value *DownShifted = Builder.CreateLShr(BitMap, ShiftAmt, "switch.downshift"); // Mask off. return Builder.CreateTrunc(DownShifted, BitMapElementTy, "switch.masked"); } case ArrayKind: { // Make sure the table index will not overflow when treated as signed. IntegerType *IT = cast(Index->getType()); uint64_t TableSize = Array->getInitializer()->getType()->getArrayNumElements(); if (TableSize > (1ULL << std::min(IT->getBitWidth() - 1, 63u))) Index = Builder.CreateZExt( Index, IntegerType::get(IT->getContext(), IT->getBitWidth() + 1), "switch.tableidx.zext"); Value *GEPIndices[] = {Builder.getInt32(0), Index}; Value *GEP = Builder.CreateInBoundsGEP(Array->getValueType(), Array, GEPIndices, "switch.gep"); return Builder.CreateLoad( cast(Array->getValueType())->getElementType(), GEP, "switch.load"); } } llvm_unreachable("Unknown lookup table kind!"); } bool SwitchLookupTable::WouldFitInRegister(const DataLayout &DL, uint64_t TableSize, Type *ElementType) { auto *IT = dyn_cast(ElementType); if (!IT) return false; // FIXME: If the type is wider than it needs to be, e.g. i8 but all values // are <= 15, we could try to narrow the type. // Avoid overflow, fitsInLegalInteger uses unsigned int for the width. if (TableSize >= UINT_MAX / IT->getBitWidth()) return false; return DL.fitsInLegalInteger(TableSize * IT->getBitWidth()); } static bool isTypeLegalForLookupTable(Type *Ty, const TargetTransformInfo &TTI, const DataLayout &DL) { // Allow any legal type. if (TTI.isTypeLegal(Ty)) return true; auto *IT = dyn_cast(Ty); if (!IT) return false; // Also allow power of 2 integer types that have at least 8 bits and fit in // a register. These types are common in frontend languages and targets // usually support loads of these types. // TODO: We could relax this to any integer that fits in a register and rely // on ABI alignment and padding in the table to allow the load to be widened. // Or we could widen the constants and truncate the load. unsigned BitWidth = IT->getBitWidth(); return BitWidth >= 8 && isPowerOf2_32(BitWidth) && DL.fitsInLegalInteger(IT->getBitWidth()); } static bool isSwitchDense(uint64_t NumCases, uint64_t CaseRange) { // 40% is the default density for building a jump table in optsize/minsize // mode. See also TargetLoweringBase::isSuitableForJumpTable(), which this // function was based on. const uint64_t MinDensity = 40; if (CaseRange >= UINT64_MAX / 100) return false; // Avoid multiplication overflows below. return NumCases * 100 >= CaseRange * MinDensity; } static bool isSwitchDense(ArrayRef Values) { uint64_t Diff = (uint64_t)Values.back() - (uint64_t)Values.front(); uint64_t Range = Diff + 1; if (Range < Diff) return false; // Overflow. return isSwitchDense(Values.size(), Range); } /// Determine whether a lookup table should be built for this switch, based on /// the number of cases, size of the table, and the types of the results. // TODO: We could support larger than legal types by limiting based on the // number of loads required and/or table size. If the constants are small we // could use smaller table entries and extend after the load. static bool ShouldBuildLookupTable(SwitchInst *SI, uint64_t TableSize, const TargetTransformInfo &TTI, const DataLayout &DL, const SmallDenseMap &ResultTypes) { if (SI->getNumCases() > TableSize) return false; // TableSize overflowed. bool AllTablesFitInRegister = true; bool HasIllegalType = false; for (const auto &I : ResultTypes) { Type *Ty = I.second; // Saturate this flag to true. HasIllegalType = HasIllegalType || !isTypeLegalForLookupTable(Ty, TTI, DL); // Saturate this flag to false. AllTablesFitInRegister = AllTablesFitInRegister && SwitchLookupTable::WouldFitInRegister(DL, TableSize, Ty); // If both flags saturate, we're done. NOTE: This *only* works with // saturating flags, and all flags have to saturate first due to the // non-deterministic behavior of iterating over a dense map. if (HasIllegalType && !AllTablesFitInRegister) break; } // If each table would fit in a register, we should build it anyway. if (AllTablesFitInRegister) return true; // Don't build a table that doesn't fit in-register if it has illegal types. if (HasIllegalType) return false; return isSwitchDense(SI->getNumCases(), TableSize); } static bool ShouldUseSwitchConditionAsTableIndex( ConstantInt &MinCaseVal, const ConstantInt &MaxCaseVal, bool HasDefaultResults, const SmallDenseMap &ResultTypes, const DataLayout &DL, const TargetTransformInfo &TTI) { if (MinCaseVal.isNullValue()) return true; if (MinCaseVal.isNegative() || MaxCaseVal.getLimitedValue() == std::numeric_limits::max() || !HasDefaultResults) return false; return all_of(ResultTypes, [&](const auto &KV) { return SwitchLookupTable::WouldFitInRegister( DL, MaxCaseVal.getLimitedValue() + 1 /* TableSize */, KV.second /* ResultType */); }); } /// Try to reuse the switch table index compare. Following pattern: /// \code /// if (idx < tablesize) /// r = table[idx]; // table does not contain default_value /// else /// r = default_value; /// if (r != default_value) /// ... /// \endcode /// Is optimized to: /// \code /// cond = idx < tablesize; /// if (cond) /// r = table[idx]; /// else /// r = default_value; /// if (cond) /// ... /// \endcode /// Jump threading will then eliminate the second if(cond). static void reuseTableCompare( User *PhiUser, BasicBlock *PhiBlock, BranchInst *RangeCheckBranch, Constant *DefaultValue, const SmallVectorImpl> &Values) { ICmpInst *CmpInst = dyn_cast(PhiUser); if (!CmpInst) return; // We require that the compare is in the same block as the phi so that jump // threading can do its work afterwards. if (CmpInst->getParent() != PhiBlock) return; Constant *CmpOp1 = dyn_cast(CmpInst->getOperand(1)); if (!CmpOp1) return; Value *RangeCmp = RangeCheckBranch->getCondition(); Constant *TrueConst = ConstantInt::getTrue(RangeCmp->getType()); Constant *FalseConst = ConstantInt::getFalse(RangeCmp->getType()); // Check if the compare with the default value is constant true or false. const DataLayout &DL = PhiBlock->getDataLayout(); Constant *DefaultConst = ConstantFoldCompareInstOperands( CmpInst->getPredicate(), DefaultValue, CmpOp1, DL); if (DefaultConst != TrueConst && DefaultConst != FalseConst) return; // Check if the compare with the case values is distinct from the default // compare result. for (auto ValuePair : Values) { Constant *CaseConst = ConstantFoldCompareInstOperands( CmpInst->getPredicate(), ValuePair.second, CmpOp1, DL); if (!CaseConst || CaseConst == DefaultConst || (CaseConst != TrueConst && CaseConst != FalseConst)) return; } // Check if the branch instruction dominates the phi node. It's a simple // dominance check, but sufficient for our needs. // Although this check is invariant in the calling loops, it's better to do it // at this late stage. Practically we do it at most once for a switch. BasicBlock *BranchBlock = RangeCheckBranch->getParent(); for (BasicBlock *Pred : predecessors(PhiBlock)) { if (Pred != BranchBlock && Pred->getUniquePredecessor() != BranchBlock) return; } if (DefaultConst == FalseConst) { // The compare yields the same result. We can replace it. CmpInst->replaceAllUsesWith(RangeCmp); ++NumTableCmpReuses; } else { // The compare yields the same result, just inverted. We can replace it. Value *InvertedTableCmp = BinaryOperator::CreateXor( RangeCmp, ConstantInt::get(RangeCmp->getType(), 1), "inverted.cmp", RangeCheckBranch->getIterator()); CmpInst->replaceAllUsesWith(InvertedTableCmp); ++NumTableCmpReuses; } } /// If the switch is only used to initialize one or more phi nodes in a common /// successor block with different constant values, replace the switch with /// lookup tables. static bool SwitchToLookupTable(SwitchInst *SI, IRBuilder<> &Builder, DomTreeUpdater *DTU, const DataLayout &DL, const TargetTransformInfo &TTI) { assert(SI->getNumCases() > 1 && "Degenerate switch?"); BasicBlock *BB = SI->getParent(); Function *Fn = BB->getParent(); // Only build lookup table when we have a target that supports it or the // attribute is not set. if (!TTI.shouldBuildLookupTables() || (Fn->getFnAttribute("no-jump-tables").getValueAsBool())) return false; // FIXME: If the switch is too sparse for a lookup table, perhaps we could // split off a dense part and build a lookup table for that. // FIXME: This creates arrays of GEPs to constant strings, which means each // GEP needs a runtime relocation in PIC code. We should just build one big // string and lookup indices into that. // Ignore switches with less than three cases. Lookup tables will not make // them faster, so we don't analyze them. if (SI->getNumCases() < 3) return false; // Figure out the corresponding result for each case value and phi node in the // common destination, as well as the min and max case values. assert(!SI->cases().empty()); SwitchInst::CaseIt CI = SI->case_begin(); ConstantInt *MinCaseVal = CI->getCaseValue(); ConstantInt *MaxCaseVal = CI->getCaseValue(); BasicBlock *CommonDest = nullptr; using ResultListTy = SmallVector, 4>; SmallDenseMap ResultLists; SmallDenseMap DefaultResults; SmallDenseMap ResultTypes; SmallVector PHIs; for (SwitchInst::CaseIt E = SI->case_end(); CI != E; ++CI) { ConstantInt *CaseVal = CI->getCaseValue(); if (CaseVal->getValue().slt(MinCaseVal->getValue())) MinCaseVal = CaseVal; if (CaseVal->getValue().sgt(MaxCaseVal->getValue())) MaxCaseVal = CaseVal; // Resulting value at phi nodes for this case value. using ResultsTy = SmallVector, 4>; ResultsTy Results; if (!getCaseResults(SI, CaseVal, CI->getCaseSuccessor(), &CommonDest, Results, DL, TTI)) return false; // Append the result from this case to the list for each phi. for (const auto &I : Results) { PHINode *PHI = I.first; Constant *Value = I.second; if (!ResultLists.count(PHI)) PHIs.push_back(PHI); ResultLists[PHI].push_back(std::make_pair(CaseVal, Value)); } } // Keep track of the result types. for (PHINode *PHI : PHIs) { ResultTypes[PHI] = ResultLists[PHI][0].second->getType(); } uint64_t NumResults = ResultLists[PHIs[0]].size(); // If the table has holes, we need a constant result for the default case // or a bitmask that fits in a register. SmallVector, 4> DefaultResultsList; bool HasDefaultResults = getCaseResults(SI, nullptr, SI->getDefaultDest(), &CommonDest, DefaultResultsList, DL, TTI); for (const auto &I : DefaultResultsList) { PHINode *PHI = I.first; Constant *Result = I.second; DefaultResults[PHI] = Result; } bool UseSwitchConditionAsTableIndex = ShouldUseSwitchConditionAsTableIndex( *MinCaseVal, *MaxCaseVal, HasDefaultResults, ResultTypes, DL, TTI); uint64_t TableSize; if (UseSwitchConditionAsTableIndex) TableSize = MaxCaseVal->getLimitedValue() + 1; else TableSize = (MaxCaseVal->getValue() - MinCaseVal->getValue()).getLimitedValue() + 1; // If the default destination is unreachable, or if the lookup table covers // all values of the conditional variable, branch directly to the lookup table // BB. Otherwise, check that the condition is within the case range. bool DefaultIsReachable = !SI->defaultDestUndefined(); bool TableHasHoles = (NumResults < TableSize); // If the table has holes but the default destination doesn't produce any // constant results, the lookup table entries corresponding to the holes will // contain undefined values. bool AllHolesAreUndefined = TableHasHoles && !HasDefaultResults; // If the default destination doesn't produce a constant result but is still // reachable, and the lookup table has holes, we need to use a mask to // determine if the current index should load from the lookup table or jump // to the default case. // The mask is unnecessary if the table has holes but the default destination // is unreachable, as in that case the holes must also be unreachable. bool NeedMask = AllHolesAreUndefined && DefaultIsReachable; if (NeedMask) { // As an extra penalty for the validity test we require more cases. if (SI->getNumCases() < 4) // FIXME: Find best threshold value (benchmark). return false; if (!DL.fitsInLegalInteger(TableSize)) return false; } if (!ShouldBuildLookupTable(SI, TableSize, TTI, DL, ResultTypes)) return false; std::vector Updates; // Compute the maximum table size representable by the integer type we are // switching upon. unsigned CaseSize = MinCaseVal->getType()->getPrimitiveSizeInBits(); uint64_t MaxTableSize = CaseSize > 63 ? UINT64_MAX : 1ULL << CaseSize; assert(MaxTableSize >= TableSize && "It is impossible for a switch to have more entries than the max " "representable value of its input integer type's size."); // Create the BB that does the lookups. Module &Mod = *CommonDest->getParent()->getParent(); BasicBlock *LookupBB = BasicBlock::Create( Mod.getContext(), "switch.lookup", CommonDest->getParent(), CommonDest); // Compute the table index value. Builder.SetInsertPoint(SI); Value *TableIndex; ConstantInt *TableIndexOffset; if (UseSwitchConditionAsTableIndex) { TableIndexOffset = ConstantInt::get(MaxCaseVal->getIntegerType(), 0); TableIndex = SI->getCondition(); } else { TableIndexOffset = MinCaseVal; // If the default is unreachable, all case values are s>= MinCaseVal. Then // we can try to attach nsw. bool MayWrap = true; if (!DefaultIsReachable) { APInt Res = MaxCaseVal->getValue().ssub_ov(MinCaseVal->getValue(), MayWrap); (void)Res; } TableIndex = Builder.CreateSub(SI->getCondition(), TableIndexOffset, "switch.tableidx", /*HasNUW =*/false, /*HasNSW =*/!MayWrap); } BranchInst *RangeCheckBranch = nullptr; // Grow the table to cover all possible index values to avoid the range check. // It will use the default result to fill in the table hole later, so make // sure it exist. if (UseSwitchConditionAsTableIndex && HasDefaultResults) { ConstantRange CR = computeConstantRange(TableIndex, /* ForSigned */ false); // Grow the table shouldn't have any size impact by checking // WouldFitInRegister. // TODO: Consider growing the table also when it doesn't fit in a register // if no optsize is specified. const uint64_t UpperBound = CR.getUpper().getLimitedValue(); if (!CR.isUpperWrapped() && all_of(ResultTypes, [&](const auto &KV) { return SwitchLookupTable::WouldFitInRegister( DL, UpperBound, KV.second /* ResultType */); })) { // There may be some case index larger than the UpperBound (unreachable // case), so make sure the table size does not get smaller. TableSize = std::max(UpperBound, TableSize); // The default branch is unreachable after we enlarge the lookup table. // Adjust DefaultIsReachable to reuse code path. DefaultIsReachable = false; } } const bool GeneratingCoveredLookupTable = (MaxTableSize == TableSize); if (!DefaultIsReachable || GeneratingCoveredLookupTable) { Builder.CreateBr(LookupBB); if (DTU) Updates.push_back({DominatorTree::Insert, BB, LookupBB}); // Note: We call removeProdecessor later since we need to be able to get the // PHI value for the default case in case we're using a bit mask. } else { Value *Cmp = Builder.CreateICmpULT( TableIndex, ConstantInt::get(MinCaseVal->getType(), TableSize)); RangeCheckBranch = Builder.CreateCondBr(Cmp, LookupBB, SI->getDefaultDest()); if (DTU) Updates.push_back({DominatorTree::Insert, BB, LookupBB}); } // Populate the BB that does the lookups. Builder.SetInsertPoint(LookupBB); if (NeedMask) { // Before doing the lookup, we do the hole check. The LookupBB is therefore // re-purposed to do the hole check, and we create a new LookupBB. BasicBlock *MaskBB = LookupBB; MaskBB->setName("switch.hole_check"); LookupBB = BasicBlock::Create(Mod.getContext(), "switch.lookup", CommonDest->getParent(), CommonDest); // Make the mask's bitwidth at least 8-bit and a power-of-2 to avoid // unnecessary illegal types. uint64_t TableSizePowOf2 = NextPowerOf2(std::max(7ULL, TableSize - 1ULL)); APInt MaskInt(TableSizePowOf2, 0); APInt One(TableSizePowOf2, 1); // Build bitmask; fill in a 1 bit for every case. const ResultListTy &ResultList = ResultLists[PHIs[0]]; for (size_t I = 0, E = ResultList.size(); I != E; ++I) { uint64_t Idx = (ResultList[I].first->getValue() - TableIndexOffset->getValue()) .getLimitedValue(); MaskInt |= One << Idx; } ConstantInt *TableMask = ConstantInt::get(Mod.getContext(), MaskInt); // Get the TableIndex'th bit of the bitmask. // If this bit is 0 (meaning hole) jump to the default destination, // else continue with table lookup. IntegerType *MapTy = TableMask->getIntegerType(); Value *MaskIndex = Builder.CreateZExtOrTrunc(TableIndex, MapTy, "switch.maskindex"); Value *Shifted = Builder.CreateLShr(TableMask, MaskIndex, "switch.shifted"); Value *LoBit = Builder.CreateTrunc( Shifted, Type::getInt1Ty(Mod.getContext()), "switch.lobit"); Builder.CreateCondBr(LoBit, LookupBB, SI->getDefaultDest()); if (DTU) { Updates.push_back({DominatorTree::Insert, MaskBB, LookupBB}); Updates.push_back({DominatorTree::Insert, MaskBB, SI->getDefaultDest()}); } Builder.SetInsertPoint(LookupBB); AddPredecessorToBlock(SI->getDefaultDest(), MaskBB, BB); } if (!DefaultIsReachable || GeneratingCoveredLookupTable) { // We cached PHINodes in PHIs. To avoid accessing deleted PHINodes later, // do not delete PHINodes here. SI->getDefaultDest()->removePredecessor(BB, /*KeepOneInputPHIs=*/true); if (DTU) Updates.push_back({DominatorTree::Delete, BB, SI->getDefaultDest()}); } for (PHINode *PHI : PHIs) { const ResultListTy &ResultList = ResultLists[PHI]; // Use any value to fill the lookup table holes. Constant *DV = AllHolesAreUndefined ? ResultLists[PHI][0].second : DefaultResults[PHI]; StringRef FuncName = Fn->getName(); SwitchLookupTable Table(Mod, TableSize, TableIndexOffset, ResultList, DV, DL, FuncName); Value *Result = Table.BuildLookup(TableIndex, Builder); // Do a small peephole optimization: re-use the switch table compare if // possible. if (!TableHasHoles && HasDefaultResults && RangeCheckBranch) { BasicBlock *PhiBlock = PHI->getParent(); // Search for compare instructions which use the phi. for (auto *User : PHI->users()) { reuseTableCompare(User, PhiBlock, RangeCheckBranch, DV, ResultList); } } PHI->addIncoming(Result, LookupBB); } Builder.CreateBr(CommonDest); if (DTU) Updates.push_back({DominatorTree::Insert, LookupBB, CommonDest}); // Remove the switch. SmallPtrSet RemovedSuccessors; for (unsigned i = 0, e = SI->getNumSuccessors(); i < e; ++i) { BasicBlock *Succ = SI->getSuccessor(i); if (Succ == SI->getDefaultDest()) continue; Succ->removePredecessor(BB); if (DTU && RemovedSuccessors.insert(Succ).second) Updates.push_back({DominatorTree::Delete, BB, Succ}); } SI->eraseFromParent(); if (DTU) DTU->applyUpdates(Updates); ++NumLookupTables; if (NeedMask) ++NumLookupTablesHoles; return true; } /// Try to transform a switch that has "holes" in it to a contiguous sequence /// of cases. /// /// A switch such as: switch(i) {case 5: case 9: case 13: case 17:} can be /// range-reduced to: switch ((i-5) / 4) {case 0: case 1: case 2: case 3:}. /// /// This converts a sparse switch into a dense switch which allows better /// lowering and could also allow transforming into a lookup table. static bool ReduceSwitchRange(SwitchInst *SI, IRBuilder<> &Builder, const DataLayout &DL, const TargetTransformInfo &TTI) { auto *CondTy = cast(SI->getCondition()->getType()); if (CondTy->getIntegerBitWidth() > 64 || !DL.fitsInLegalInteger(CondTy->getIntegerBitWidth())) return false; // Only bother with this optimization if there are more than 3 switch cases; // SDAG will only bother creating jump tables for 4 or more cases. if (SI->getNumCases() < 4) return false; // This transform is agnostic to the signedness of the input or case values. We // can treat the case values as signed or unsigned. We can optimize more common // cases such as a sequence crossing zero {-4,0,4,8} if we interpret case values // as signed. SmallVector Values; for (const auto &C : SI->cases()) Values.push_back(C.getCaseValue()->getValue().getSExtValue()); llvm::sort(Values); // If the switch is already dense, there's nothing useful to do here. if (isSwitchDense(Values)) return false; // First, transform the values such that they start at zero and ascend. int64_t Base = Values[0]; for (auto &V : Values) V -= (uint64_t)(Base); // Now we have signed numbers that have been shifted so that, given enough // precision, there are no negative values. Since the rest of the transform // is bitwise only, we switch now to an unsigned representation. // This transform can be done speculatively because it is so cheap - it // results in a single rotate operation being inserted. // countTrailingZeros(0) returns 64. As Values is guaranteed to have more than // one element and LLVM disallows duplicate cases, Shift is guaranteed to be // less than 64. unsigned Shift = 64; for (auto &V : Values) Shift = std::min(Shift, (unsigned)llvm::countr_zero((uint64_t)V)); assert(Shift < 64); if (Shift > 0) for (auto &V : Values) V = (int64_t)((uint64_t)V >> Shift); if (!isSwitchDense(Values)) // Transform didn't create a dense switch. return false; // The obvious transform is to shift the switch condition right and emit a // check that the condition actually cleanly divided by GCD, i.e. // C & (1 << Shift - 1) == 0 // inserting a new CFG edge to handle the case where it didn't divide cleanly. // // A cheaper way of doing this is a simple ROTR(C, Shift). This performs the // shift and puts the shifted-off bits in the uppermost bits. If any of these // are nonzero then the switch condition will be very large and will hit the // default case. auto *Ty = cast(SI->getCondition()->getType()); Builder.SetInsertPoint(SI); Value *Sub = Builder.CreateSub(SI->getCondition(), ConstantInt::get(Ty, Base)); Value *Rot = Builder.CreateIntrinsic( Ty, Intrinsic::fshl, {Sub, Sub, ConstantInt::get(Ty, Ty->getBitWidth() - Shift)}); SI->replaceUsesOfWith(SI->getCondition(), Rot); for (auto Case : SI->cases()) { auto *Orig = Case.getCaseValue(); auto Sub = Orig->getValue() - APInt(Ty->getBitWidth(), Base); Case.setValue(cast(ConstantInt::get(Ty, Sub.lshr(Shift)))); } return true; } /// Tries to transform switch of powers of two to reduce switch range. /// For example, switch like: /// switch (C) { case 1: case 2: case 64: case 128: } /// will be transformed to: /// switch (count_trailing_zeros(C)) { case 0: case 1: case 6: case 7: } /// /// This transformation allows better lowering and could allow transforming into /// a lookup table. static bool simplifySwitchOfPowersOfTwo(SwitchInst *SI, IRBuilder<> &Builder, const DataLayout &DL, const TargetTransformInfo &TTI) { Value *Condition = SI->getCondition(); LLVMContext &Context = SI->getContext(); auto *CondTy = cast(Condition->getType()); if (CondTy->getIntegerBitWidth() > 64 || !DL.fitsInLegalInteger(CondTy->getIntegerBitWidth())) return false; const auto CttzIntrinsicCost = TTI.getIntrinsicInstrCost( IntrinsicCostAttributes(Intrinsic::cttz, CondTy, {Condition, ConstantInt::getTrue(Context)}), TTI::TCK_SizeAndLatency); if (CttzIntrinsicCost > TTI::TCC_Basic) // Inserting intrinsic is too expensive. return false; // Only bother with this optimization if there are more than 3 switch cases. // SDAG will only bother creating jump tables for 4 or more cases. if (SI->getNumCases() < 4) return false; // We perform this optimization only for switches with // unreachable default case. // This assumtion will save us from checking if `Condition` is a power of two. if (!isa(SI->getDefaultDest()->getFirstNonPHIOrDbg())) return false; // Check that switch cases are powers of two. SmallVector Values; for (const auto &Case : SI->cases()) { uint64_t CaseValue = Case.getCaseValue()->getValue().getZExtValue(); if (llvm::has_single_bit(CaseValue)) Values.push_back(CaseValue); else return false; } // isSwichDense requires case values to be sorted. llvm::sort(Values); if (!isSwitchDense(Values.size(), llvm::countr_zero(Values.back()) - llvm::countr_zero(Values.front()) + 1)) // Transform is unable to generate dense switch. return false; Builder.SetInsertPoint(SI); // Replace each case with its trailing zeros number. for (auto &Case : SI->cases()) { auto *OrigValue = Case.getCaseValue(); Case.setValue(ConstantInt::get(OrigValue->getIntegerType(), OrigValue->getValue().countr_zero())); } // Replace condition with its trailing zeros number. auto *ConditionTrailingZeros = Builder.CreateIntrinsic( Intrinsic::cttz, {CondTy}, {Condition, ConstantInt::getTrue(Context)}); SI->setCondition(ConditionTrailingZeros); return true; } bool SimplifyCFGOpt::simplifySwitch(SwitchInst *SI, IRBuilder<> &Builder) { BasicBlock *BB = SI->getParent(); if (isValueEqualityComparison(SI)) { // If we only have one predecessor, and if it is a branch on this value, // see if that predecessor totally determines the outcome of this switch. if (BasicBlock *OnlyPred = BB->getSinglePredecessor()) if (SimplifyEqualityComparisonWithOnlyPredecessor(SI, OnlyPred, Builder)) return requestResimplify(); Value *Cond = SI->getCondition(); if (SelectInst *Select = dyn_cast(Cond)) if (SimplifySwitchOnSelect(SI, Select)) return requestResimplify(); // If the block only contains the switch, see if we can fold the block // away into any preds. if (SI == &*BB->instructionsWithoutDebug(false).begin()) if (FoldValueComparisonIntoPredecessors(SI, Builder)) return requestResimplify(); } // Try to transform the switch into an icmp and a branch. // The conversion from switch to comparison may lose information on // impossible switch values, so disable it early in the pipeline. if (Options.ConvertSwitchRangeToICmp && TurnSwitchRangeIntoICmp(SI, Builder)) return requestResimplify(); // Remove unreachable cases. if (eliminateDeadSwitchCases(SI, DTU, Options.AC, DL)) return requestResimplify(); if (trySwitchToSelect(SI, Builder, DTU, DL, TTI)) return requestResimplify(); if (Options.ForwardSwitchCondToPhi && ForwardSwitchConditionToPHI(SI)) return requestResimplify(); // The conversion from switch to lookup tables results in difficult-to-analyze // code and makes pruning branches much harder. This is a problem if the // switch expression itself can still be restricted as a result of inlining or // CVP. Therefore, only apply this transformation during late stages of the // optimisation pipeline. if (Options.ConvertSwitchToLookupTable && SwitchToLookupTable(SI, Builder, DTU, DL, TTI)) return requestResimplify(); if (simplifySwitchOfPowersOfTwo(SI, Builder, DL, TTI)) return requestResimplify(); if (ReduceSwitchRange(SI, Builder, DL, TTI)) return requestResimplify(); if (HoistCommon && hoistCommonCodeFromSuccessors(SI->getParent(), !Options.HoistCommonInsts)) return requestResimplify(); return false; } bool SimplifyCFGOpt::simplifyIndirectBr(IndirectBrInst *IBI) { BasicBlock *BB = IBI->getParent(); bool Changed = false; // Eliminate redundant destinations. SmallPtrSet Succs; SmallSetVector RemovedSuccs; for (unsigned i = 0, e = IBI->getNumDestinations(); i != e; ++i) { BasicBlock *Dest = IBI->getDestination(i); if (!Dest->hasAddressTaken() || !Succs.insert(Dest).second) { if (!Dest->hasAddressTaken()) RemovedSuccs.insert(Dest); Dest->removePredecessor(BB); IBI->removeDestination(i); --i; --e; Changed = true; } } if (DTU) { std::vector Updates; Updates.reserve(RemovedSuccs.size()); for (auto *RemovedSucc : RemovedSuccs) Updates.push_back({DominatorTree::Delete, BB, RemovedSucc}); DTU->applyUpdates(Updates); } if (IBI->getNumDestinations() == 0) { // If the indirectbr has no successors, change it to unreachable. new UnreachableInst(IBI->getContext(), IBI->getIterator()); EraseTerminatorAndDCECond(IBI); return true; } if (IBI->getNumDestinations() == 1) { // If the indirectbr has one successor, change it to a direct branch. BranchInst::Create(IBI->getDestination(0), IBI->getIterator()); EraseTerminatorAndDCECond(IBI); return true; } if (SelectInst *SI = dyn_cast(IBI->getAddress())) { if (SimplifyIndirectBrOnSelect(IBI, SI)) return requestResimplify(); } return Changed; } /// Given an block with only a single landing pad and a unconditional branch /// try to find another basic block which this one can be merged with. This /// handles cases where we have multiple invokes with unique landing pads, but /// a shared handler. /// /// We specifically choose to not worry about merging non-empty blocks /// here. That is a PRE/scheduling problem and is best solved elsewhere. In /// practice, the optimizer produces empty landing pad blocks quite frequently /// when dealing with exception dense code. (see: instcombine, gvn, if-else /// sinking in this file) /// /// This is primarily a code size optimization. We need to avoid performing /// any transform which might inhibit optimization (such as our ability to /// specialize a particular handler via tail commoning). We do this by not /// merging any blocks which require us to introduce a phi. Since the same /// values are flowing through both blocks, we don't lose any ability to /// specialize. If anything, we make such specialization more likely. /// /// TODO - This transformation could remove entries from a phi in the target /// block when the inputs in the phi are the same for the two blocks being /// merged. In some cases, this could result in removal of the PHI entirely. static bool TryToMergeLandingPad(LandingPadInst *LPad, BranchInst *BI, BasicBlock *BB, DomTreeUpdater *DTU) { auto Succ = BB->getUniqueSuccessor(); assert(Succ); // If there's a phi in the successor block, we'd likely have to introduce // a phi into the merged landing pad block. if (isa(*Succ->begin())) return false; for (BasicBlock *OtherPred : predecessors(Succ)) { if (BB == OtherPred) continue; BasicBlock::iterator I = OtherPred->begin(); LandingPadInst *LPad2 = dyn_cast(I); if (!LPad2 || !LPad2->isIdenticalTo(LPad)) continue; for (++I; isa(I); ++I) ; BranchInst *BI2 = dyn_cast(I); if (!BI2 || !BI2->isIdenticalTo(BI)) continue; std::vector Updates; // We've found an identical block. Update our predecessors to take that // path instead and make ourselves dead. SmallSetVector UniquePreds(pred_begin(BB), pred_end(BB)); for (BasicBlock *Pred : UniquePreds) { InvokeInst *II = cast(Pred->getTerminator()); assert(II->getNormalDest() != BB && II->getUnwindDest() == BB && "unexpected successor"); II->setUnwindDest(OtherPred); if (DTU) { Updates.push_back({DominatorTree::Insert, Pred, OtherPred}); Updates.push_back({DominatorTree::Delete, Pred, BB}); } } // The debug info in OtherPred doesn't cover the merged control flow that // used to go through BB. We need to delete it or update it. for (Instruction &Inst : llvm::make_early_inc_range(*OtherPred)) if (isa(Inst)) Inst.eraseFromParent(); SmallSetVector UniqueSuccs(succ_begin(BB), succ_end(BB)); for (BasicBlock *Succ : UniqueSuccs) { Succ->removePredecessor(BB); if (DTU) Updates.push_back({DominatorTree::Delete, BB, Succ}); } IRBuilder<> Builder(BI); Builder.CreateUnreachable(); BI->eraseFromParent(); if (DTU) DTU->applyUpdates(Updates); return true; } return false; } bool SimplifyCFGOpt::simplifyBranch(BranchInst *Branch, IRBuilder<> &Builder) { return Branch->isUnconditional() ? simplifyUncondBranch(Branch, Builder) : simplifyCondBranch(Branch, Builder); } bool SimplifyCFGOpt::simplifyUncondBranch(BranchInst *BI, IRBuilder<> &Builder) { BasicBlock *BB = BI->getParent(); BasicBlock *Succ = BI->getSuccessor(0); // If the Terminator is the only non-phi instruction, simplify the block. // If LoopHeader is provided, check if the block or its successor is a loop // header. (This is for early invocations before loop simplify and // vectorization to keep canonical loop forms for nested loops. These blocks // can be eliminated when the pass is invoked later in the back-end.) // Note that if BB has only one predecessor then we do not introduce new // backedge, so we can eliminate BB. bool NeedCanonicalLoop = Options.NeedCanonicalLoop && (!LoopHeaders.empty() && BB->hasNPredecessorsOrMore(2) && (is_contained(LoopHeaders, BB) || is_contained(LoopHeaders, Succ))); BasicBlock::iterator I = BB->getFirstNonPHIOrDbg(true)->getIterator(); if (I->isTerminator() && BB != &BB->getParent()->getEntryBlock() && !NeedCanonicalLoop && TryToSimplifyUncondBranchFromEmptyBlock(BB, DTU)) return true; // If the only instruction in the block is a seteq/setne comparison against a // constant, try to simplify the block. if (ICmpInst *ICI = dyn_cast(I)) if (ICI->isEquality() && isa(ICI->getOperand(1))) { for (++I; isa(I); ++I) ; if (I->isTerminator() && tryToSimplifyUncondBranchWithICmpInIt(ICI, Builder)) return true; } // See if we can merge an empty landing pad block with another which is // equivalent. if (LandingPadInst *LPad = dyn_cast(I)) { for (++I; isa(I); ++I) ; if (I->isTerminator() && TryToMergeLandingPad(LPad, BI, BB, DTU)) return true; } // If this basic block is ONLY a compare and a branch, and if a predecessor // branches to us and our successor, fold the comparison into the // predecessor and use logical operations to update the incoming value // for PHI nodes in common successor. if (Options.SpeculateBlocks && FoldBranchToCommonDest(BI, DTU, /*MSSAU=*/nullptr, &TTI, Options.BonusInstThreshold)) return requestResimplify(); return false; } static BasicBlock *allPredecessorsComeFromSameSource(BasicBlock *BB) { BasicBlock *PredPred = nullptr; for (auto *P : predecessors(BB)) { BasicBlock *PPred = P->getSinglePredecessor(); if (!PPred || (PredPred && PredPred != PPred)) return nullptr; PredPred = PPred; } return PredPred; } /// Fold the following pattern: /// bb0: /// br i1 %cond1, label %bb1, label %bb2 /// bb1: /// br i1 %cond2, label %bb3, label %bb4 /// bb2: /// br i1 %cond2, label %bb4, label %bb3 /// bb3: /// ... /// bb4: /// ... /// into /// bb0: /// %cond = xor i1 %cond1, %cond2 /// br i1 %cond, label %bb4, label %bb3 /// bb3: /// ... /// bb4: /// ... /// NOTE: %cond2 always dominates the terminator of bb0. static bool mergeNestedCondBranch(BranchInst *BI, DomTreeUpdater *DTU) { BasicBlock *BB = BI->getParent(); BasicBlock *BB1 = BI->getSuccessor(0); BasicBlock *BB2 = BI->getSuccessor(1); auto IsSimpleSuccessor = [BB](BasicBlock *Succ, BranchInst *&SuccBI) { if (Succ == BB) return false; if (&Succ->front() != Succ->getTerminator()) return false; SuccBI = dyn_cast(Succ->getTerminator()); if (!SuccBI || !SuccBI->isConditional()) return false; BasicBlock *Succ1 = SuccBI->getSuccessor(0); BasicBlock *Succ2 = SuccBI->getSuccessor(1); return Succ1 != Succ && Succ2 != Succ && Succ1 != BB && Succ2 != BB && !isa(Succ1->front()) && !isa(Succ2->front()); }; BranchInst *BB1BI, *BB2BI; if (!IsSimpleSuccessor(BB1, BB1BI) || !IsSimpleSuccessor(BB2, BB2BI)) return false; if (BB1BI->getCondition() != BB2BI->getCondition() || BB1BI->getSuccessor(0) != BB2BI->getSuccessor(1) || BB1BI->getSuccessor(1) != BB2BI->getSuccessor(0)) return false; BasicBlock *BB3 = BB1BI->getSuccessor(0); BasicBlock *BB4 = BB1BI->getSuccessor(1); IRBuilder<> Builder(BI); BI->setCondition( Builder.CreateXor(BI->getCondition(), BB1BI->getCondition())); BB1->removePredecessor(BB); BI->setSuccessor(0, BB4); BB2->removePredecessor(BB); BI->setSuccessor(1, BB3); if (DTU) { SmallVector Updates; Updates.push_back({DominatorTree::Delete, BB, BB1}); Updates.push_back({DominatorTree::Insert, BB, BB4}); Updates.push_back({DominatorTree::Delete, BB, BB2}); Updates.push_back({DominatorTree::Insert, BB, BB3}); DTU->applyUpdates(Updates); } bool HasWeight = false; uint64_t BBTWeight, BBFWeight; if (extractBranchWeights(*BI, BBTWeight, BBFWeight)) HasWeight = true; else BBTWeight = BBFWeight = 1; uint64_t BB1TWeight, BB1FWeight; if (extractBranchWeights(*BB1BI, BB1TWeight, BB1FWeight)) HasWeight = true; else BB1TWeight = BB1FWeight = 1; uint64_t BB2TWeight, BB2FWeight; if (extractBranchWeights(*BB2BI, BB2TWeight, BB2FWeight)) HasWeight = true; else BB2TWeight = BB2FWeight = 1; if (HasWeight) { uint64_t Weights[2] = {BBTWeight * BB1FWeight + BBFWeight * BB2TWeight, BBTWeight * BB1TWeight + BBFWeight * BB2FWeight}; FitWeights(Weights); setBranchWeights(BI, Weights[0], Weights[1], /*IsExpected=*/false); } return true; } bool SimplifyCFGOpt::simplifyCondBranch(BranchInst *BI, IRBuilder<> &Builder) { assert( !isa(BI->getCondition()) && BI->getSuccessor(0) != BI->getSuccessor(1) && "Tautological conditional branch should have been eliminated already."); BasicBlock *BB = BI->getParent(); if (!Options.SimplifyCondBranch || BI->getFunction()->hasFnAttribute(Attribute::OptForFuzzing)) return false; // Conditional branch if (isValueEqualityComparison(BI)) { // If we only have one predecessor, and if it is a branch on this value, // see if that predecessor totally determines the outcome of this // switch. if (BasicBlock *OnlyPred = BB->getSinglePredecessor()) if (SimplifyEqualityComparisonWithOnlyPredecessor(BI, OnlyPred, Builder)) return requestResimplify(); // This block must be empty, except for the setcond inst, if it exists. // Ignore dbg and pseudo intrinsics. auto I = BB->instructionsWithoutDebug(true).begin(); if (&*I == BI) { if (FoldValueComparisonIntoPredecessors(BI, Builder)) return requestResimplify(); } else if (&*I == cast(BI->getCondition())) { ++I; if (&*I == BI && FoldValueComparisonIntoPredecessors(BI, Builder)) return requestResimplify(); } } // Try to turn "br (X == 0 | X == 1), T, F" into a switch instruction. if (SimplifyBranchOnICmpChain(BI, Builder, DL)) return true; // If this basic block has dominating predecessor blocks and the dominating // blocks' conditions imply BI's condition, we know the direction of BI. std::optional Imp = isImpliedByDomCondition(BI->getCondition(), BI, DL); if (Imp) { // Turn this into a branch on constant. auto *OldCond = BI->getCondition(); ConstantInt *TorF = *Imp ? ConstantInt::getTrue(BB->getContext()) : ConstantInt::getFalse(BB->getContext()); BI->setCondition(TorF); RecursivelyDeleteTriviallyDeadInstructions(OldCond); return requestResimplify(); } // If this basic block is ONLY a compare and a branch, and if a predecessor // branches to us and one of our successors, fold the comparison into the // predecessor and use logical operations to pick the right destination. if (Options.SpeculateBlocks && FoldBranchToCommonDest(BI, DTU, /*MSSAU=*/nullptr, &TTI, Options.BonusInstThreshold)) return requestResimplify(); // We have a conditional branch to two blocks that are only reachable // from BI. We know that the condbr dominates the two blocks, so see if // there is any identical code in the "then" and "else" blocks. If so, we // can hoist it up to the branching block. if (BI->getSuccessor(0)->getSinglePredecessor()) { if (BI->getSuccessor(1)->getSinglePredecessor()) { if (HoistCommon && hoistCommonCodeFromSuccessors( BI->getParent(), !Options.HoistCommonInsts)) return requestResimplify(); } else { // If Successor #1 has multiple preds, we may be able to conditionally // execute Successor #0 if it branches to Successor #1. Instruction *Succ0TI = BI->getSuccessor(0)->getTerminator(); if (Succ0TI->getNumSuccessors() == 1 && Succ0TI->getSuccessor(0) == BI->getSuccessor(1)) if (SpeculativelyExecuteBB(BI, BI->getSuccessor(0))) return requestResimplify(); } } else if (BI->getSuccessor(1)->getSinglePredecessor()) { // If Successor #0 has multiple preds, we may be able to conditionally // execute Successor #1 if it branches to Successor #0. Instruction *Succ1TI = BI->getSuccessor(1)->getTerminator(); if (Succ1TI->getNumSuccessors() == 1 && Succ1TI->getSuccessor(0) == BI->getSuccessor(0)) if (SpeculativelyExecuteBB(BI, BI->getSuccessor(1))) return requestResimplify(); } // If this is a branch on something for which we know the constant value in // predecessors (e.g. a phi node in the current block), thread control // through this block. if (FoldCondBranchOnValueKnownInPredecessor(BI, DTU, DL, Options.AC)) return requestResimplify(); // Scan predecessor blocks for conditional branches. for (BasicBlock *Pred : predecessors(BB)) if (BranchInst *PBI = dyn_cast(Pred->getTerminator())) if (PBI != BI && PBI->isConditional()) if (SimplifyCondBranchToCondBranch(PBI, BI, DTU, DL, TTI)) return requestResimplify(); // Look for diamond patterns. if (MergeCondStores) if (BasicBlock *PrevBB = allPredecessorsComeFromSameSource(BB)) if (BranchInst *PBI = dyn_cast(PrevBB->getTerminator())) if (PBI != BI && PBI->isConditional()) if (mergeConditionalStores(PBI, BI, DTU, DL, TTI)) return requestResimplify(); // Look for nested conditional branches. if (mergeNestedCondBranch(BI, DTU)) return requestResimplify(); return false; } /// Check if passing a value to an instruction will cause undefined behavior. static bool passingValueIsAlwaysUndefined(Value *V, Instruction *I, bool PtrValueMayBeModified) { Constant *C = dyn_cast(V); if (!C) return false; if (I->use_empty()) return false; if (C->isNullValue() || isa(C)) { // Only look at the first use we can handle, avoid hurting compile time with // long uselists auto FindUse = llvm::find_if(I->users(), [](auto *U) { auto *Use = cast(U); // Change this list when we want to add new instructions. switch (Use->getOpcode()) { default: return false; case Instruction::GetElementPtr: case Instruction::Ret: case Instruction::BitCast: case Instruction::Load: case Instruction::Store: case Instruction::Call: case Instruction::CallBr: case Instruction::Invoke: return true; } }); if (FindUse == I->user_end()) return false; auto *Use = cast(*FindUse); // Bail out if Use is not in the same BB as I or Use == I or Use comes // before I in the block. The latter two can be the case if Use is a // PHI node. if (Use->getParent() != I->getParent() || Use == I || Use->comesBefore(I)) return false; // Now make sure that there are no instructions in between that can alter // control flow (eg. calls) auto InstrRange = make_range(std::next(I->getIterator()), Use->getIterator()); if (any_of(InstrRange, [](Instruction &I) { return !isGuaranteedToTransferExecutionToSuccessor(&I); })) return false; // Look through GEPs. A load from a GEP derived from NULL is still undefined if (GetElementPtrInst *GEP = dyn_cast(Use)) if (GEP->getPointerOperand() == I) { // The current base address is null, there are four cases to consider: // getelementptr (TY, null, 0) -> null // getelementptr (TY, null, not zero) -> may be modified // getelementptr inbounds (TY, null, 0) -> null // getelementptr inbounds (TY, null, not zero) -> poison iff null is // undefined? if (!GEP->hasAllZeroIndices() && (!GEP->isInBounds() || NullPointerIsDefined(GEP->getFunction(), GEP->getPointerAddressSpace()))) PtrValueMayBeModified = true; return passingValueIsAlwaysUndefined(V, GEP, PtrValueMayBeModified); } // Look through return. if (ReturnInst *Ret = dyn_cast(Use)) { bool HasNoUndefAttr = Ret->getFunction()->hasRetAttribute(Attribute::NoUndef); // Return undefined to a noundef return value is undefined. if (isa(C) && HasNoUndefAttr) return true; // Return null to a nonnull+noundef return value is undefined. if (C->isNullValue() && HasNoUndefAttr && Ret->getFunction()->hasRetAttribute(Attribute::NonNull)) { return !PtrValueMayBeModified; } } // Look through bitcasts. if (BitCastInst *BC = dyn_cast(Use)) return passingValueIsAlwaysUndefined(V, BC, PtrValueMayBeModified); // Load from null is undefined. if (LoadInst *LI = dyn_cast(Use)) if (!LI->isVolatile()) return !NullPointerIsDefined(LI->getFunction(), LI->getPointerAddressSpace()); // Store to null is undefined. if (StoreInst *SI = dyn_cast(Use)) if (!SI->isVolatile()) return (!NullPointerIsDefined(SI->getFunction(), SI->getPointerAddressSpace())) && SI->getPointerOperand() == I; // llvm.assume(false/undef) always triggers immediate UB. if (auto *Assume = dyn_cast(Use)) { // Ignore assume operand bundles. if (I == Assume->getArgOperand(0)) return true; } if (auto *CB = dyn_cast(Use)) { if (C->isNullValue() && NullPointerIsDefined(CB->getFunction())) return false; // A call to null is undefined. if (CB->getCalledOperand() == I) return true; if (C->isNullValue()) { for (const llvm::Use &Arg : CB->args()) if (Arg == I) { unsigned ArgIdx = CB->getArgOperandNo(&Arg); if (CB->isPassingUndefUB(ArgIdx) && CB->paramHasAttr(ArgIdx, Attribute::NonNull)) { // Passing null to a nonnnull+noundef argument is undefined. return !PtrValueMayBeModified; } } } else if (isa(C)) { // Passing undef to a noundef argument is undefined. for (const llvm::Use &Arg : CB->args()) if (Arg == I) { unsigned ArgIdx = CB->getArgOperandNo(&Arg); if (CB->isPassingUndefUB(ArgIdx)) { // Passing undef to a noundef argument is undefined. return true; } } } } } return false; } /// If BB has an incoming value that will always trigger undefined behavior /// (eg. null pointer dereference), remove the branch leading here. static bool removeUndefIntroducingPredecessor(BasicBlock *BB, DomTreeUpdater *DTU, AssumptionCache *AC) { for (PHINode &PHI : BB->phis()) for (unsigned i = 0, e = PHI.getNumIncomingValues(); i != e; ++i) if (passingValueIsAlwaysUndefined(PHI.getIncomingValue(i), &PHI)) { BasicBlock *Predecessor = PHI.getIncomingBlock(i); Instruction *T = Predecessor->getTerminator(); IRBuilder<> Builder(T); if (BranchInst *BI = dyn_cast(T)) { BB->removePredecessor(Predecessor); // Turn unconditional branches into unreachables and remove the dead // destination from conditional branches. if (BI->isUnconditional()) Builder.CreateUnreachable(); else { // Preserve guarding condition in assume, because it might not be // inferrable from any dominating condition. Value *Cond = BI->getCondition(); CallInst *Assumption; if (BI->getSuccessor(0) == BB) Assumption = Builder.CreateAssumption(Builder.CreateNot(Cond)); else Assumption = Builder.CreateAssumption(Cond); if (AC) AC->registerAssumption(cast(Assumption)); Builder.CreateBr(BI->getSuccessor(0) == BB ? BI->getSuccessor(1) : BI->getSuccessor(0)); } BI->eraseFromParent(); if (DTU) DTU->applyUpdates({{DominatorTree::Delete, Predecessor, BB}}); return true; } else if (SwitchInst *SI = dyn_cast(T)) { // Redirect all branches leading to UB into // a newly created unreachable block. BasicBlock *Unreachable = BasicBlock::Create( Predecessor->getContext(), "unreachable", BB->getParent(), BB); Builder.SetInsertPoint(Unreachable); // The new block contains only one instruction: Unreachable Builder.CreateUnreachable(); for (const auto &Case : SI->cases()) if (Case.getCaseSuccessor() == BB) { BB->removePredecessor(Predecessor); Case.setSuccessor(Unreachable); } if (SI->getDefaultDest() == BB) { BB->removePredecessor(Predecessor); SI->setDefaultDest(Unreachable); } if (DTU) DTU->applyUpdates( { { DominatorTree::Insert, Predecessor, Unreachable }, { DominatorTree::Delete, Predecessor, BB } }); return true; } } return false; } bool SimplifyCFGOpt::simplifyOnce(BasicBlock *BB) { bool Changed = false; assert(BB && BB->getParent() && "Block not embedded in function!"); assert(BB->getTerminator() && "Degenerate basic block encountered!"); // Remove basic blocks that have no predecessors (except the entry block)... // or that just have themself as a predecessor. These are unreachable. if ((pred_empty(BB) && BB != &BB->getParent()->getEntryBlock()) || BB->getSinglePredecessor() == BB) { LLVM_DEBUG(dbgs() << "Removing BB: \n" << *BB); DeleteDeadBlock(BB, DTU); return true; } // Check to see if we can constant propagate this terminator instruction // away... Changed |= ConstantFoldTerminator(BB, /*DeleteDeadConditions=*/true, /*TLI=*/nullptr, DTU); // Check for and eliminate duplicate PHI nodes in this block. Changed |= EliminateDuplicatePHINodes(BB); // Check for and remove branches that will always cause undefined behavior. if (removeUndefIntroducingPredecessor(BB, DTU, Options.AC)) return requestResimplify(); // Merge basic blocks into their predecessor if there is only one distinct // pred, and if there is only one distinct successor of the predecessor, and // if there are no PHI nodes. if (MergeBlockIntoPredecessor(BB, DTU)) return true; if (SinkCommon && Options.SinkCommonInsts) if (SinkCommonCodeFromPredecessors(BB, DTU) || MergeCompatibleInvokes(BB, DTU)) { // SinkCommonCodeFromPredecessors() does not automatically CSE PHI's, // so we may now how duplicate PHI's. // Let's rerun EliminateDuplicatePHINodes() first, // before FoldTwoEntryPHINode() potentially converts them into select's, // after which we'd need a whole EarlyCSE pass run to cleanup them. return true; } IRBuilder<> Builder(BB); if (Options.SpeculateBlocks && !BB->getParent()->hasFnAttribute(Attribute::OptForFuzzing)) { // If there is a trivial two-entry PHI node in this basic block, and we can // eliminate it, do so now. if (auto *PN = dyn_cast(BB->begin())) if (PN->getNumIncomingValues() == 2) if (FoldTwoEntryPHINode(PN, TTI, DTU, DL, Options.SpeculateUnpredictables)) return true; } Instruction *Terminator = BB->getTerminator(); Builder.SetInsertPoint(Terminator); switch (Terminator->getOpcode()) { case Instruction::Br: Changed |= simplifyBranch(cast(Terminator), Builder); break; case Instruction::Resume: Changed |= simplifyResume(cast(Terminator), Builder); break; case Instruction::CleanupRet: Changed |= simplifyCleanupReturn(cast(Terminator)); break; case Instruction::Switch: Changed |= simplifySwitch(cast(Terminator), Builder); break; case Instruction::Unreachable: Changed |= simplifyUnreachable(cast(Terminator)); break; case Instruction::IndirectBr: Changed |= simplifyIndirectBr(cast(Terminator)); break; } return Changed; } bool SimplifyCFGOpt::run(BasicBlock *BB) { bool Changed = false; // Repeated simplify BB as long as resimplification is requested. do { Resimplify = false; // Perform one round of simplifcation. Resimplify flag will be set if // another iteration is requested. Changed |= simplifyOnce(BB); } while (Resimplify); return Changed; } bool llvm::simplifyCFG(BasicBlock *BB, const TargetTransformInfo &TTI, DomTreeUpdater *DTU, const SimplifyCFGOptions &Options, ArrayRef LoopHeaders) { return SimplifyCFGOpt(TTI, DTU, BB->getDataLayout(), LoopHeaders, Options) .run(BB); }