//===- BasicAliasAnalysis.cpp - Stateless Alias Analysis Impl -------------===// // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // // This file defines the primary stateless implementation of the // Alias Analysis interface that implements identities (two different // globals cannot alias, etc), but does no stateful analysis. // //===----------------------------------------------------------------------===// #include "llvm/Analysis/AliasAnalysis.h" #include "llvm/Analysis/Passes.h" #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Function.h" #include "llvm/GlobalAlias.h" #include "llvm/GlobalVariable.h" #include "llvm/Instructions.h" #include "llvm/IntrinsicInst.h" #include "llvm/LLVMContext.h" #include "llvm/Operator.h" #include "llvm/Pass.h" #include "llvm/Analysis/CaptureTracking.h" #include "llvm/Analysis/MemoryBuiltins.h" #include "llvm/Analysis/InstructionSimplify.h" #include "llvm/Analysis/ValueTracking.h" #include "llvm/Target/TargetData.h" #include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" #include "llvm/Support/ErrorHandling.h" #include "llvm/Support/GetElementPtrTypeIterator.h" #include using namespace llvm; //===----------------------------------------------------------------------===// // Useful predicates //===----------------------------------------------------------------------===// /// isKnownNonNull - Return true if we know that the specified value is never /// null. static bool isKnownNonNull(const Value *V) { // Alloca never returns null, malloc might. if (isa(V)) return true; // A byval argument is never null. if (const Argument *A = dyn_cast(V)) return A->hasByValAttr(); // Global values are not null unless extern weak. if (const GlobalValue *GV = dyn_cast(V)) return !GV->hasExternalWeakLinkage(); return false; } /// isNonEscapingLocalObject - Return true if the pointer is to a function-local /// object that never escapes from the function. static bool isNonEscapingLocalObject(const Value *V) { // If this is a local allocation, check to see if it escapes. if (isa(V) || isNoAliasCall(V)) // Set StoreCaptures to True so that we can assume in our callers that the // pointer is not the result of a load instruction. Currently // PointerMayBeCaptured doesn't have any special analysis for the // StoreCaptures=false case; if it did, our callers could be refined to be // more precise. return !PointerMayBeCaptured(V, false, /*StoreCaptures=*/true); // If this is an argument that corresponds to a byval or noalias argument, // then it has not escaped before entering the function. Check if it escapes // inside the function. if (const Argument *A = dyn_cast(V)) if (A->hasByValAttr() || A->hasNoAliasAttr()) { // Don't bother analyzing arguments already known not to escape. if (A->hasNoCaptureAttr()) return true; return !PointerMayBeCaptured(V, false, /*StoreCaptures=*/true); } return false; } /// isEscapeSource - Return true if the pointer is one which would have /// been considered an escape by isNonEscapingLocalObject. static bool isEscapeSource(const Value *V) { if (isa(V) || isa(V) || isa(V)) return true; // The load case works because isNonEscapingLocalObject considers all // stores to be escapes (it passes true for the StoreCaptures argument // to PointerMayBeCaptured). if (isa(V)) return true; return false; } /// getObjectSize - Return the size of the object specified by V, or /// UnknownSize if unknown. static uint64_t getObjectSize(const Value *V, const TargetData &TD) { const Type *AccessTy; if (const GlobalVariable *GV = dyn_cast(V)) { if (!GV->hasDefinitiveInitializer()) return AliasAnalysis::UnknownSize; AccessTy = GV->getType()->getElementType(); } else if (const AllocaInst *AI = dyn_cast(V)) { if (!AI->isArrayAllocation()) AccessTy = AI->getType()->getElementType(); else return AliasAnalysis::UnknownSize; } else if (const CallInst* CI = extractMallocCall(V)) { if (!isArrayMalloc(V, &TD)) // The size is the argument to the malloc call. if (const ConstantInt* C = dyn_cast(CI->getArgOperand(0))) return C->getZExtValue(); return AliasAnalysis::UnknownSize; } else if (const Argument *A = dyn_cast(V)) { if (A->hasByValAttr()) AccessTy = cast(A->getType())->getElementType(); else return AliasAnalysis::UnknownSize; } else { return AliasAnalysis::UnknownSize; } if (AccessTy->isSized()) return TD.getTypeAllocSize(AccessTy); return AliasAnalysis::UnknownSize; } /// isObjectSmallerThan - Return true if we can prove that the object specified /// by V is smaller than Size. static bool isObjectSmallerThan(const Value *V, uint64_t Size, const TargetData &TD) { uint64_t ObjectSize = getObjectSize(V, TD); return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize < Size; } /// isObjectSize - Return true if we can prove that the object specified /// by V has size Size. static bool isObjectSize(const Value *V, uint64_t Size, const TargetData &TD) { uint64_t ObjectSize = getObjectSize(V, TD); return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize == Size; } //===----------------------------------------------------------------------===// // GetElementPtr Instruction Decomposition and Analysis //===----------------------------------------------------------------------===// namespace { enum ExtensionKind { EK_NotExtended, EK_SignExt, EK_ZeroExt }; struct VariableGEPIndex { const Value *V; ExtensionKind Extension; int64_t Scale; }; } /// GetLinearExpression - Analyze the specified value as a linear expression: /// "A*V + B", where A and B are constant integers. Return the scale and offset /// values as APInts and return V as a Value*, and return whether we looked /// through any sign or zero extends. The incoming Value is known to have /// IntegerType and it may already be sign or zero extended. /// /// Note that this looks through extends, so the high bits may not be /// represented in the result. static Value *GetLinearExpression(Value *V, APInt &Scale, APInt &Offset, ExtensionKind &Extension, const TargetData &TD, unsigned Depth) { assert(V->getType()->isIntegerTy() && "Not an integer value"); // Limit our recursion depth. if (Depth == 6) { Scale = 1; Offset = 0; return V; } if (BinaryOperator *BOp = dyn_cast(V)) { if (ConstantInt *RHSC = dyn_cast(BOp->getOperand(1))) { switch (BOp->getOpcode()) { default: break; case Instruction::Or: // X|C == X+C if all the bits in C are unset in X. Otherwise we can't // analyze it. if (!MaskedValueIsZero(BOp->getOperand(0), RHSC->getValue(), &TD)) break; // FALL THROUGH. case Instruction::Add: V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension, TD, Depth+1); Offset += RHSC->getValue(); return V; case Instruction::Mul: V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension, TD, Depth+1); Offset *= RHSC->getValue(); Scale *= RHSC->getValue(); return V; case Instruction::Shl: V = GetLinearExpression(BOp->getOperand(0), Scale, Offset, Extension, TD, Depth+1); Offset <<= RHSC->getValue().getLimitedValue(); Scale <<= RHSC->getValue().getLimitedValue(); return V; } } } // Since GEP indices are sign extended anyway, we don't care about the high // bits of a sign or zero extended value - just scales and offsets. The // extensions have to be consistent though. if ((isa(V) && Extension != EK_ZeroExt) || (isa(V) && Extension != EK_SignExt)) { Value *CastOp = cast(V)->getOperand(0); unsigned OldWidth = Scale.getBitWidth(); unsigned SmallWidth = CastOp->getType()->getPrimitiveSizeInBits(); Scale = Scale.trunc(SmallWidth); Offset = Offset.trunc(SmallWidth); Extension = isa(V) ? EK_SignExt : EK_ZeroExt; Value *Result = GetLinearExpression(CastOp, Scale, Offset, Extension, TD, Depth+1); Scale = Scale.zext(OldWidth); Offset = Offset.zext(OldWidth); return Result; } Scale = 1; Offset = 0; return V; } /// DecomposeGEPExpression - If V is a symbolic pointer expression, decompose it /// into a base pointer with a constant offset and a number of scaled symbolic /// offsets. /// /// The scaled symbolic offsets (represented by pairs of a Value* and a scale in /// the VarIndices vector) are Value*'s that are known to be scaled by the /// specified amount, but which may have other unrepresented high bits. As such, /// the gep cannot necessarily be reconstructed from its decomposed form. /// /// When TargetData is around, this function is capable of analyzing everything /// that GetUnderlyingObject can look through. When not, it just looks /// through pointer casts. /// static const Value * DecomposeGEPExpression(const Value *V, int64_t &BaseOffs, SmallVectorImpl &VarIndices, const TargetData *TD) { // Limit recursion depth to limit compile time in crazy cases. unsigned MaxLookup = 6; BaseOffs = 0; do { // See if this is a bitcast or GEP. const Operator *Op = dyn_cast(V); if (Op == 0) { // The only non-operator case we can handle are GlobalAliases. if (const GlobalAlias *GA = dyn_cast(V)) { if (!GA->mayBeOverridden()) { V = GA->getAliasee(); continue; } } return V; } if (Op->getOpcode() == Instruction::BitCast) { V = Op->getOperand(0); continue; } if (const Instruction *I = dyn_cast(V)) // TODO: Get a DominatorTree and use it here. if (const Value *Simplified = SimplifyInstruction(const_cast(I), TD)) { V = Simplified; continue; } const GEPOperator *GEPOp = dyn_cast(Op); if (GEPOp == 0) return V; // Don't attempt to analyze GEPs over unsized objects. if (!cast(GEPOp->getOperand(0)->getType()) ->getElementType()->isSized()) return V; // If we are lacking TargetData information, we can't compute the offets of // elements computed by GEPs. However, we can handle bitcast equivalent // GEPs. if (TD == 0) { if (!GEPOp->hasAllZeroIndices()) return V; V = GEPOp->getOperand(0); continue; } // Walk the indices of the GEP, accumulating them into BaseOff/VarIndices. gep_type_iterator GTI = gep_type_begin(GEPOp); for (User::const_op_iterator I = GEPOp->op_begin()+1, E = GEPOp->op_end(); I != E; ++I) { Value *Index = *I; // Compute the (potentially symbolic) offset in bytes for this index. if (const StructType *STy = dyn_cast(*GTI++)) { // For a struct, add the member offset. unsigned FieldNo = cast(Index)->getZExtValue(); if (FieldNo == 0) continue; BaseOffs += TD->getStructLayout(STy)->getElementOffset(FieldNo); continue; } // For an array/pointer, add the element offset, explicitly scaled. if (ConstantInt *CIdx = dyn_cast(Index)) { if (CIdx->isZero()) continue; BaseOffs += TD->getTypeAllocSize(*GTI)*CIdx->getSExtValue(); continue; } uint64_t Scale = TD->getTypeAllocSize(*GTI); ExtensionKind Extension = EK_NotExtended; // If the integer type is smaller than the pointer size, it is implicitly // sign extended to pointer size. unsigned Width = cast(Index->getType())->getBitWidth(); if (TD->getPointerSizeInBits() > Width) Extension = EK_SignExt; // Use GetLinearExpression to decompose the index into a C1*V+C2 form. APInt IndexScale(Width, 0), IndexOffset(Width, 0); Index = GetLinearExpression(Index, IndexScale, IndexOffset, Extension, *TD, 0); // The GEP index scale ("Scale") scales C1*V+C2, yielding (C1*V+C2)*Scale. // This gives us an aggregate computation of (C1*Scale)*V + C2*Scale. BaseOffs += IndexOffset.getSExtValue()*Scale; Scale *= IndexScale.getSExtValue(); // If we already had an occurrence of this index variable, merge this // scale into it. For example, we want to handle: // A[x][x] -> x*16 + x*4 -> x*20 // This also ensures that 'x' only appears in the index list once. for (unsigned i = 0, e = VarIndices.size(); i != e; ++i) { if (VarIndices[i].V == Index && VarIndices[i].Extension == Extension) { Scale += VarIndices[i].Scale; VarIndices.erase(VarIndices.begin()+i); break; } } // Make sure that we have a scale that makes sense for this target's // pointer size. if (unsigned ShiftBits = 64-TD->getPointerSizeInBits()) { Scale <<= ShiftBits; Scale = (int64_t)Scale >> ShiftBits; } if (Scale) { VariableGEPIndex Entry = {Index, Extension, Scale}; VarIndices.push_back(Entry); } } // Analyze the base pointer next. V = GEPOp->getOperand(0); } while (--MaxLookup); // If the chain of expressions is too deep, just return early. return V; } /// GetIndexDifference - Dest and Src are the variable indices from two /// decomposed GetElementPtr instructions GEP1 and GEP2 which have common base /// pointers. Subtract the GEP2 indices from GEP1 to find the symbolic /// difference between the two pointers. static void GetIndexDifference(SmallVectorImpl &Dest, const SmallVectorImpl &Src) { if (Src.empty()) return; for (unsigned i = 0, e = Src.size(); i != e; ++i) { const Value *V = Src[i].V; ExtensionKind Extension = Src[i].Extension; int64_t Scale = Src[i].Scale; // Find V in Dest. This is N^2, but pointer indices almost never have more // than a few variable indexes. for (unsigned j = 0, e = Dest.size(); j != e; ++j) { if (Dest[j].V != V || Dest[j].Extension != Extension) continue; // If we found it, subtract off Scale V's from the entry in Dest. If it // goes to zero, remove the entry. if (Dest[j].Scale != Scale) Dest[j].Scale -= Scale; else Dest.erase(Dest.begin()+j); Scale = 0; break; } // If we didn't consume this entry, add it to the end of the Dest list. if (Scale) { VariableGEPIndex Entry = { V, Extension, -Scale }; Dest.push_back(Entry); } } } //===----------------------------------------------------------------------===// // BasicAliasAnalysis Pass //===----------------------------------------------------------------------===// #ifndef NDEBUG static const Function *getParent(const Value *V) { if (const Instruction *inst = dyn_cast(V)) return inst->getParent()->getParent(); if (const Argument *arg = dyn_cast(V)) return arg->getParent(); return NULL; } static bool notDifferentParent(const Value *O1, const Value *O2) { const Function *F1 = getParent(O1); const Function *F2 = getParent(O2); return !F1 || !F2 || F1 == F2; } #endif namespace { /// BasicAliasAnalysis - This is the primary alias analysis implementation. struct BasicAliasAnalysis : public ImmutablePass, public AliasAnalysis { static char ID; // Class identification, replacement for typeinfo BasicAliasAnalysis() : ImmutablePass(ID) { initializeBasicAliasAnalysisPass(*PassRegistry::getPassRegistry()); } virtual void initializePass() { InitializeAliasAnalysis(this); } virtual void getAnalysisUsage(AnalysisUsage &AU) const { AU.addRequired(); } virtual AliasResult alias(const Location &LocA, const Location &LocB) { assert(Visited.empty() && "Visited must be cleared after use!"); assert(notDifferentParent(LocA.Ptr, LocB.Ptr) && "BasicAliasAnalysis doesn't support interprocedural queries."); AliasResult Alias = aliasCheck(LocA.Ptr, LocA.Size, LocA.TBAATag, LocB.Ptr, LocB.Size, LocB.TBAATag); Visited.clear(); return Alias; } virtual ModRefResult getModRefInfo(ImmutableCallSite CS, const Location &Loc); virtual ModRefResult getModRefInfo(ImmutableCallSite CS1, ImmutableCallSite CS2) { // The AliasAnalysis base class has some smarts, lets use them. return AliasAnalysis::getModRefInfo(CS1, CS2); } /// pointsToConstantMemory - Chase pointers until we find a (constant /// global) or not. virtual bool pointsToConstantMemory(const Location &Loc, bool OrLocal); /// getModRefBehavior - Return the behavior when calling the given /// call site. virtual ModRefBehavior getModRefBehavior(ImmutableCallSite CS); /// getModRefBehavior - Return the behavior when calling the given function. /// For use when the call site is not known. virtual ModRefBehavior getModRefBehavior(const Function *F); /// getAdjustedAnalysisPointer - This method is used when a pass implements /// an analysis interface through multiple inheritance. If needed, it /// should override this to adjust the this pointer as needed for the /// specified pass info. virtual void *getAdjustedAnalysisPointer(const void *ID) { if (ID == &AliasAnalysis::ID) return (AliasAnalysis*)this; return this; } private: // Visited - Track instructions visited by a aliasPHI, aliasSelect(), and aliasGEP(). SmallPtrSet Visited; // aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP // instruction against another. AliasResult aliasGEP(const GEPOperator *V1, uint64_t V1Size, const Value *V2, uint64_t V2Size, const MDNode *V2TBAAInfo, const Value *UnderlyingV1, const Value *UnderlyingV2); // aliasPHI - Provide a bunch of ad-hoc rules to disambiguate a PHI // instruction against another. AliasResult aliasPHI(const PHINode *PN, uint64_t PNSize, const MDNode *PNTBAAInfo, const Value *V2, uint64_t V2Size, const MDNode *V2TBAAInfo); /// aliasSelect - Disambiguate a Select instruction against another value. AliasResult aliasSelect(const SelectInst *SI, uint64_t SISize, const MDNode *SITBAAInfo, const Value *V2, uint64_t V2Size, const MDNode *V2TBAAInfo); AliasResult aliasCheck(const Value *V1, uint64_t V1Size, const MDNode *V1TBAATag, const Value *V2, uint64_t V2Size, const MDNode *V2TBAATag); }; } // End of anonymous namespace // Register this pass... char BasicAliasAnalysis::ID = 0; INITIALIZE_AG_PASS(BasicAliasAnalysis, AliasAnalysis, "basicaa", "Basic Alias Analysis (stateless AA impl)", false, true, false) ImmutablePass *llvm::createBasicAliasAnalysisPass() { return new BasicAliasAnalysis(); } /// pointsToConstantMemory - Returns whether the given pointer value /// points to memory that is local to the function, with global constants being /// considered local to all functions. bool BasicAliasAnalysis::pointsToConstantMemory(const Location &Loc, bool OrLocal) { assert(Visited.empty() && "Visited must be cleared after use!"); unsigned MaxLookup = 8; SmallVector Worklist; Worklist.push_back(Loc.Ptr); do { const Value *V = GetUnderlyingObject(Worklist.pop_back_val(), TD); if (!Visited.insert(V)) { Visited.clear(); return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal); } // An alloca instruction defines local memory. if (OrLocal && isa(V)) continue; // A global constant counts as local memory for our purposes. if (const GlobalVariable *GV = dyn_cast(V)) { // Note: this doesn't require GV to be "ODR" because it isn't legal for a // global to be marked constant in some modules and non-constant in // others. GV may even be a declaration, not a definition. if (!GV->isConstant()) { Visited.clear(); return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal); } continue; } // If both select values point to local memory, then so does the select. if (const SelectInst *SI = dyn_cast(V)) { Worklist.push_back(SI->getTrueValue()); Worklist.push_back(SI->getFalseValue()); continue; } // If all values incoming to a phi node point to local memory, then so does // the phi. if (const PHINode *PN = dyn_cast(V)) { // Don't bother inspecting phi nodes with many operands. if (PN->getNumIncomingValues() > MaxLookup) { Visited.clear(); return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal); } for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) Worklist.push_back(PN->getIncomingValue(i)); continue; } // Otherwise be conservative. Visited.clear(); return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal); } while (!Worklist.empty() && --MaxLookup); Visited.clear(); return Worklist.empty(); } /// getModRefBehavior - Return the behavior when calling the given call site. AliasAnalysis::ModRefBehavior BasicAliasAnalysis::getModRefBehavior(ImmutableCallSite CS) { if (CS.doesNotAccessMemory()) // Can't do better than this. return DoesNotAccessMemory; ModRefBehavior Min = UnknownModRefBehavior; // If the callsite knows it only reads memory, don't return worse // than that. if (CS.onlyReadsMemory()) Min = OnlyReadsMemory; // The AliasAnalysis base class has some smarts, lets use them. return ModRefBehavior(AliasAnalysis::getModRefBehavior(CS) & Min); } /// getModRefBehavior - Return the behavior when calling the given function. /// For use when the call site is not known. AliasAnalysis::ModRefBehavior BasicAliasAnalysis::getModRefBehavior(const Function *F) { // If the function declares it doesn't access memory, we can't do better. if (F->doesNotAccessMemory()) return DoesNotAccessMemory; // For intrinsics, we can check the table. if (unsigned iid = F->getIntrinsicID()) { #define GET_INTRINSIC_MODREF_BEHAVIOR #include "llvm/Intrinsics.gen" #undef GET_INTRINSIC_MODREF_BEHAVIOR } ModRefBehavior Min = UnknownModRefBehavior; // If the function declares it only reads memory, go with that. if (F->onlyReadsMemory()) Min = OnlyReadsMemory; // Otherwise be conservative. return ModRefBehavior(AliasAnalysis::getModRefBehavior(F) & Min); } /// getModRefInfo - Check to see if the specified callsite can clobber the /// specified memory object. Since we only look at local properties of this /// function, we really can't say much about this query. We do, however, use /// simple "address taken" analysis on local objects. AliasAnalysis::ModRefResult BasicAliasAnalysis::getModRefInfo(ImmutableCallSite CS, const Location &Loc) { assert(notDifferentParent(CS.getInstruction(), Loc.Ptr) && "AliasAnalysis query involving multiple functions!"); const Value *Object = GetUnderlyingObject(Loc.Ptr, TD); // If this is a tail call and Loc.Ptr points to a stack location, we know that // the tail call cannot access or modify the local stack. // We cannot exclude byval arguments here; these belong to the caller of // the current function not to the current function, and a tail callee // may reference them. if (isa(Object)) if (const CallInst *CI = dyn_cast(CS.getInstruction())) if (CI->isTailCall()) return NoModRef; // If the pointer is to a locally allocated object that does not escape, // then the call can not mod/ref the pointer unless the call takes the pointer // as an argument, and itself doesn't capture it. if (!isa(Object) && CS.getInstruction() != Object && isNonEscapingLocalObject(Object)) { bool PassedAsArg = false; unsigned ArgNo = 0; for (ImmutableCallSite::arg_iterator CI = CS.arg_begin(), CE = CS.arg_end(); CI != CE; ++CI, ++ArgNo) { // Only look at the no-capture pointer arguments. if (!(*CI)->getType()->isPointerTy() || !CS.paramHasAttr(ArgNo+1, Attribute::NoCapture)) continue; // If this is a no-capture pointer argument, see if we can tell that it // is impossible to alias the pointer we're checking. If not, we have to // assume that the call could touch the pointer, even though it doesn't // escape. if (!isNoAlias(Location(cast(CI)), Loc)) { PassedAsArg = true; break; } } if (!PassedAsArg) return NoModRef; } ModRefResult Min = ModRef; // Finally, handle specific knowledge of intrinsics. const IntrinsicInst *II = dyn_cast(CS.getInstruction()); if (II != 0) switch (II->getIntrinsicID()) { default: break; case Intrinsic::memcpy: case Intrinsic::memmove: { uint64_t Len = UnknownSize; if (ConstantInt *LenCI = dyn_cast(II->getArgOperand(2))) Len = LenCI->getZExtValue(); Value *Dest = II->getArgOperand(0); Value *Src = II->getArgOperand(1); // If it can't overlap the source dest, then it doesn't modref the loc. if (isNoAlias(Location(Dest, Len), Loc)) { if (isNoAlias(Location(Src, Len), Loc)) return NoModRef; // If it can't overlap the dest, then worst case it reads the loc. Min = Ref; } else if (isNoAlias(Location(Src, Len), Loc)) { // If it can't overlap the source, then worst case it mutates the loc. Min = Mod; } break; } case Intrinsic::memset: // Since memset is 'accesses arguments' only, the AliasAnalysis base class // will handle it for the variable length case. if (ConstantInt *LenCI = dyn_cast(II->getArgOperand(2))) { uint64_t Len = LenCI->getZExtValue(); Value *Dest = II->getArgOperand(0); if (isNoAlias(Location(Dest, Len), Loc)) return NoModRef; } // We know that memset doesn't load anything. Min = Mod; break; case Intrinsic::atomic_cmp_swap: case Intrinsic::atomic_swap: case Intrinsic::atomic_load_add: case Intrinsic::atomic_load_sub: case Intrinsic::atomic_load_and: case Intrinsic::atomic_load_nand: case Intrinsic::atomic_load_or: case Intrinsic::atomic_load_xor: case Intrinsic::atomic_load_max: case Intrinsic::atomic_load_min: case Intrinsic::atomic_load_umax: case Intrinsic::atomic_load_umin: if (TD) { Value *Op1 = II->getArgOperand(0); uint64_t Op1Size = TD->getTypeStoreSize(Op1->getType()); MDNode *Tag = II->getMetadata(LLVMContext::MD_tbaa); if (isNoAlias(Location(Op1, Op1Size, Tag), Loc)) return NoModRef; } break; case Intrinsic::lifetime_start: case Intrinsic::lifetime_end: case Intrinsic::invariant_start: { uint64_t PtrSize = cast(II->getArgOperand(0))->getZExtValue(); if (isNoAlias(Location(II->getArgOperand(1), PtrSize, II->getMetadata(LLVMContext::MD_tbaa)), Loc)) return NoModRef; break; } case Intrinsic::invariant_end: { uint64_t PtrSize = cast(II->getArgOperand(1))->getZExtValue(); if (isNoAlias(Location(II->getArgOperand(2), PtrSize, II->getMetadata(LLVMContext::MD_tbaa)), Loc)) return NoModRef; break; } } // The AliasAnalysis base class has some smarts, lets use them. return ModRefResult(AliasAnalysis::getModRefInfo(CS, Loc) & Min); } /// aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP instruction /// against another pointer. We know that V1 is a GEP, but we don't know /// anything about V2. UnderlyingV1 is GetUnderlyingObject(GEP1, TD), /// UnderlyingV2 is the same for V2. /// AliasAnalysis::AliasResult BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size, const Value *V2, uint64_t V2Size, const MDNode *V2TBAAInfo, const Value *UnderlyingV1, const Value *UnderlyingV2) { // If this GEP has been visited before, we're on a use-def cycle. // Such cycles are only valid when PHI nodes are involved or in unreachable // code. The visitPHI function catches cycles containing PHIs, but there // could still be a cycle without PHIs in unreachable code. if (!Visited.insert(GEP1)) return MayAlias; int64_t GEP1BaseOffset; SmallVector GEP1VariableIndices; // If we have two gep instructions with must-alias'ing base pointers, figure // out if the indexes to the GEP tell us anything about the derived pointer. if (const GEPOperator *GEP2 = dyn_cast(V2)) { // Do the base pointers alias? AliasResult BaseAlias = aliasCheck(UnderlyingV1, UnknownSize, 0, UnderlyingV2, UnknownSize, 0); // If we get a No or May, then return it immediately, no amount of analysis // will improve this situation. if (BaseAlias != MustAlias) return BaseAlias; // Otherwise, we have a MustAlias. Since the base pointers alias each other // exactly, see if the computed offset from the common pointer tells us // about the relation of the resulting pointer. const Value *GEP1BasePtr = DecomposeGEPExpression(GEP1, GEP1BaseOffset, GEP1VariableIndices, TD); int64_t GEP2BaseOffset; SmallVector GEP2VariableIndices; const Value *GEP2BasePtr = DecomposeGEPExpression(GEP2, GEP2BaseOffset, GEP2VariableIndices, TD); // If DecomposeGEPExpression isn't able to look all the way through the // addressing operation, we must not have TD and this is too complex for us // to handle without it. if (GEP1BasePtr != UnderlyingV1 || GEP2BasePtr != UnderlyingV2) { assert(TD == 0 && "DecomposeGEPExpression and GetUnderlyingObject disagree!"); return MayAlias; } // Subtract the GEP2 pointer from the GEP1 pointer to find out their // symbolic difference. GEP1BaseOffset -= GEP2BaseOffset; GetIndexDifference(GEP1VariableIndices, GEP2VariableIndices); } else { // Check to see if these two pointers are related by the getelementptr // instruction. If one pointer is a GEP with a non-zero index of the other // pointer, we know they cannot alias. // If both accesses are unknown size, we can't do anything useful here. if (V1Size == UnknownSize && V2Size == UnknownSize) return MayAlias; AliasResult R = aliasCheck(UnderlyingV1, UnknownSize, 0, V2, V2Size, V2TBAAInfo); if (R != MustAlias) // If V2 may alias GEP base pointer, conservatively returns MayAlias. // If V2 is known not to alias GEP base pointer, then the two values // cannot alias per GEP semantics: "A pointer value formed from a // getelementptr instruction is associated with the addresses associated // with the first operand of the getelementptr". return R; const Value *GEP1BasePtr = DecomposeGEPExpression(GEP1, GEP1BaseOffset, GEP1VariableIndices, TD); // If DecomposeGEPExpression isn't able to look all the way through the // addressing operation, we must not have TD and this is too complex for us // to handle without it. if (GEP1BasePtr != UnderlyingV1) { assert(TD == 0 && "DecomposeGEPExpression and GetUnderlyingObject disagree!"); return MayAlias; } } // In the two GEP Case, if there is no difference in the offsets of the // computed pointers, the resultant pointers are a must alias. This // hapens when we have two lexically identical GEP's (for example). // // In the other case, if we have getelementptr , 0, 0, 0, 0, ... and V2 // must aliases the GEP, the end result is a must alias also. if (GEP1BaseOffset == 0 && GEP1VariableIndices.empty()) return MustAlias; // If there is a difference between the pointers, but the difference is // less than the size of the associated memory object, then we know // that the objects are partially overlapping. if (GEP1BaseOffset != 0 && GEP1VariableIndices.empty()) { if (GEP1BaseOffset >= 0 ? (V2Size != UnknownSize && (uint64_t)GEP1BaseOffset < V2Size) : (V1Size != UnknownSize && -(uint64_t)GEP1BaseOffset < V1Size && GEP1BaseOffset != INT64_MIN)) return PartialAlias; } // If we have a known constant offset, see if this offset is larger than the // access size being queried. If so, and if no variable indices can remove // pieces of this constant, then we know we have a no-alias. For example, // &A[100] != &A. // In order to handle cases like &A[100][i] where i is an out of range // subscript, we have to ignore all constant offset pieces that are a multiple // of a scaled index. Do this by removing constant offsets that are a // multiple of any of our variable indices. This allows us to transform // things like &A[i][1] because i has a stride of (e.g.) 8 bytes but the 1 // provides an offset of 4 bytes (assuming a <= 4 byte access). for (unsigned i = 0, e = GEP1VariableIndices.size(); i != e && GEP1BaseOffset;++i) if (int64_t RemovedOffset = GEP1BaseOffset/GEP1VariableIndices[i].Scale) GEP1BaseOffset -= RemovedOffset*GEP1VariableIndices[i].Scale; // If our known offset is bigger than the access size, we know we don't have // an alias. if (GEP1BaseOffset) { if (GEP1BaseOffset >= 0 ? (V2Size != UnknownSize && (uint64_t)GEP1BaseOffset >= V2Size) : (V1Size != UnknownSize && -(uint64_t)GEP1BaseOffset >= V1Size && GEP1BaseOffset != INT64_MIN)) return NoAlias; } return MayAlias; } /// aliasSelect - Provide a bunch of ad-hoc rules to disambiguate a Select /// instruction against another. AliasAnalysis::AliasResult BasicAliasAnalysis::aliasSelect(const SelectInst *SI, uint64_t SISize, const MDNode *SITBAAInfo, const Value *V2, uint64_t V2Size, const MDNode *V2TBAAInfo) { // If this select has been visited before, we're on a use-def cycle. // Such cycles are only valid when PHI nodes are involved or in unreachable // code. The visitPHI function catches cycles containing PHIs, but there // could still be a cycle without PHIs in unreachable code. if (!Visited.insert(SI)) return MayAlias; // If the values are Selects with the same condition, we can do a more precise // check: just check for aliases between the values on corresponding arms. if (const SelectInst *SI2 = dyn_cast(V2)) if (SI->getCondition() == SI2->getCondition()) { AliasResult Alias = aliasCheck(SI->getTrueValue(), SISize, SITBAAInfo, SI2->getTrueValue(), V2Size, V2TBAAInfo); if (Alias == MayAlias) return MayAlias; AliasResult ThisAlias = aliasCheck(SI->getFalseValue(), SISize, SITBAAInfo, SI2->getFalseValue(), V2Size, V2TBAAInfo); if (ThisAlias != Alias) return MayAlias; return Alias; } // If both arms of the Select node NoAlias or MustAlias V2, then returns // NoAlias / MustAlias. Otherwise, returns MayAlias. AliasResult Alias = aliasCheck(V2, V2Size, V2TBAAInfo, SI->getTrueValue(), SISize, SITBAAInfo); if (Alias == MayAlias) return MayAlias; // If V2 is visited, the recursive case will have been caught in the // above aliasCheck call, so these subsequent calls to aliasCheck // don't need to assume that V2 is being visited recursively. Visited.erase(V2); AliasResult ThisAlias = aliasCheck(V2, V2Size, V2TBAAInfo, SI->getFalseValue(), SISize, SITBAAInfo); if (ThisAlias != Alias) return MayAlias; return Alias; } // aliasPHI - Provide a bunch of ad-hoc rules to disambiguate a PHI instruction // against another. AliasAnalysis::AliasResult BasicAliasAnalysis::aliasPHI(const PHINode *PN, uint64_t PNSize, const MDNode *PNTBAAInfo, const Value *V2, uint64_t V2Size, const MDNode *V2TBAAInfo) { // The PHI node has already been visited, avoid recursion any further. if (!Visited.insert(PN)) return MayAlias; // If the values are PHIs in the same block, we can do a more precise // as well as efficient check: just check for aliases between the values // on corresponding edges. if (const PHINode *PN2 = dyn_cast(V2)) if (PN2->getParent() == PN->getParent()) { AliasResult Alias = aliasCheck(PN->getIncomingValue(0), PNSize, PNTBAAInfo, PN2->getIncomingValueForBlock(PN->getIncomingBlock(0)), V2Size, V2TBAAInfo); if (Alias == MayAlias) return MayAlias; for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) { AliasResult ThisAlias = aliasCheck(PN->getIncomingValue(i), PNSize, PNTBAAInfo, PN2->getIncomingValueForBlock(PN->getIncomingBlock(i)), V2Size, V2TBAAInfo); if (ThisAlias != Alias) return MayAlias; } return Alias; } SmallPtrSet UniqueSrc; SmallVector V1Srcs; for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i) { Value *PV1 = PN->getIncomingValue(i); if (isa(PV1)) // If any of the source itself is a PHI, return MayAlias conservatively // to avoid compile time explosion. The worst possible case is if both // sides are PHI nodes. In which case, this is O(m x n) time where 'm' // and 'n' are the number of PHI sources. return MayAlias; if (UniqueSrc.insert(PV1)) V1Srcs.push_back(PV1); } AliasResult Alias = aliasCheck(V2, V2Size, V2TBAAInfo, V1Srcs[0], PNSize, PNTBAAInfo); // Early exit if the check of the first PHI source against V2 is MayAlias. // Other results are not possible. if (Alias == MayAlias) return MayAlias; // If all sources of the PHI node NoAlias or MustAlias V2, then returns // NoAlias / MustAlias. Otherwise, returns MayAlias. for (unsigned i = 1, e = V1Srcs.size(); i != e; ++i) { Value *V = V1Srcs[i]; // If V2 is visited, the recursive case will have been caught in the // above aliasCheck call, so these subsequent calls to aliasCheck // don't need to assume that V2 is being visited recursively. Visited.erase(V2); AliasResult ThisAlias = aliasCheck(V2, V2Size, V2TBAAInfo, V, PNSize, PNTBAAInfo); if (ThisAlias != Alias || ThisAlias == MayAlias) return MayAlias; } return Alias; } // aliasCheck - Provide a bunch of ad-hoc rules to disambiguate in common cases, // such as array references. // AliasAnalysis::AliasResult BasicAliasAnalysis::aliasCheck(const Value *V1, uint64_t V1Size, const MDNode *V1TBAAInfo, const Value *V2, uint64_t V2Size, const MDNode *V2TBAAInfo) { // If either of the memory references is empty, it doesn't matter what the // pointer values are. if (V1Size == 0 || V2Size == 0) return NoAlias; // Strip off any casts if they exist. V1 = V1->stripPointerCasts(); V2 = V2->stripPointerCasts(); // Are we checking for alias of the same value? if (V1 == V2) return MustAlias; if (!V1->getType()->isPointerTy() || !V2->getType()->isPointerTy()) return NoAlias; // Scalars cannot alias each other // Figure out what objects these things are pointing to if we can. const Value *O1 = GetUnderlyingObject(V1, TD); const Value *O2 = GetUnderlyingObject(V2, TD); // Null values in the default address space don't point to any object, so they // don't alias any other pointer. if (const ConstantPointerNull *CPN = dyn_cast(O1)) if (CPN->getType()->getAddressSpace() == 0) return NoAlias; if (const ConstantPointerNull *CPN = dyn_cast(O2)) if (CPN->getType()->getAddressSpace() == 0) return NoAlias; if (O1 != O2) { // If V1/V2 point to two different objects we know that we have no alias. if (isIdentifiedObject(O1) && isIdentifiedObject(O2)) return NoAlias; // Constant pointers can't alias with non-const isIdentifiedObject objects. if ((isa(O1) && isIdentifiedObject(O2) && !isa(O2)) || (isa(O2) && isIdentifiedObject(O1) && !isa(O1))) return NoAlias; // Arguments can't alias with local allocations or noalias calls // in the same function. if (((isa(O1) && (isa(O2) || isNoAliasCall(O2))) || (isa(O2) && (isa(O1) || isNoAliasCall(O1))))) return NoAlias; // Most objects can't alias null. if ((isa(O2) && isKnownNonNull(O1)) || (isa(O1) && isKnownNonNull(O2))) return NoAlias; // If one pointer is the result of a call/invoke or load and the other is a // non-escaping local object within the same function, then we know the // object couldn't escape to a point where the call could return it. // // Note that if the pointers are in different functions, there are a // variety of complications. A call with a nocapture argument may still // temporary store the nocapture argument's value in a temporary memory // location if that memory location doesn't escape. Or it may pass a // nocapture value to other functions as long as they don't capture it. if (isEscapeSource(O1) && isNonEscapingLocalObject(O2)) return NoAlias; if (isEscapeSource(O2) && isNonEscapingLocalObject(O1)) return NoAlias; } // If the size of one access is larger than the entire object on the other // side, then we know such behavior is undefined and can assume no alias. if (TD) if ((V1Size != UnknownSize && isObjectSmallerThan(O2, V1Size, *TD)) || (V2Size != UnknownSize && isObjectSmallerThan(O1, V2Size, *TD))) return NoAlias; // FIXME: This isn't aggressively handling alias(GEP, PHI) for example: if the // GEP can't simplify, we don't even look at the PHI cases. if (!isa(V1) && isa(V2)) { std::swap(V1, V2); std::swap(V1Size, V2Size); std::swap(O1, O2); } if (const GEPOperator *GV1 = dyn_cast(V1)) { AliasResult Result = aliasGEP(GV1, V1Size, V2, V2Size, V2TBAAInfo, O1, O2); if (Result != MayAlias) return Result; } if (isa(V2) && !isa(V1)) { std::swap(V1, V2); std::swap(V1Size, V2Size); } if (const PHINode *PN = dyn_cast(V1)) { AliasResult Result = aliasPHI(PN, V1Size, V1TBAAInfo, V2, V2Size, V2TBAAInfo); if (Result != MayAlias) return Result; } if (isa(V2) && !isa(V1)) { std::swap(V1, V2); std::swap(V1Size, V2Size); } if (const SelectInst *S1 = dyn_cast(V1)) { AliasResult Result = aliasSelect(S1, V1Size, V1TBAAInfo, V2, V2Size, V2TBAAInfo); if (Result != MayAlias) return Result; } // If both pointers are pointing into the same object and one of them // accesses is accessing the entire object, then the accesses must // overlap in some way. if (TD && O1 == O2) if ((V1Size != UnknownSize && isObjectSize(O1, V1Size, *TD)) || (V2Size != UnknownSize && isObjectSize(O2, V2Size, *TD))) return PartialAlias; return AliasAnalysis::alias(Location(V1, V1Size, V1TBAAInfo), Location(V2, V2Size, V2TBAAInfo)); }