X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FAnalysis%2FBasicAliasAnalysis.cpp;h=4b7d2f5d9dc68296dfaae21f31f5423443b2e509;hb=6b3ae4638bc5a3fb3bad286f96a1234b8a53053a;hp=7ce9d1651f73c25a422f8ec31e8e7cb5792428e4;hpb=4ba8cfc5a4e98dbe55529bb2a0de17565e90c128;p=oota-llvm.git diff --git a/lib/Analysis/BasicAliasAnalysis.cpp b/lib/Analysis/BasicAliasAnalysis.cpp index 7ce9d1651f7..4b7d2f5d9dc 100644 --- a/lib/Analysis/BasicAliasAnalysis.cpp +++ b/lib/Analysis/BasicAliasAnalysis.cpp @@ -1,15 +1,15 @@ -//===- BasicAliasAnalysis.cpp - Local Alias Analysis Impl -----------------===// +//===- BasicAliasAnalysis.cpp - Stateless Alias Analysis Impl -------------===// // // The LLVM Compiler Infrastructure // -// This file was developed by the LLVM research group and is distributed under -// the University of Illinois Open Source License. See LICENSE.TXT for details. +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // -// This file defines the default implementation of the Alias Analysis interface -// that simply implements a few identities (two different globals cannot alias, -// etc), but otherwise does no analysis. +// 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. // //===----------------------------------------------------------------------===// @@ -18,794 +18,1238 @@ #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" #include "llvm/Function.h" -#include "llvm/ParameterAttributes.h" +#include "llvm/GlobalAlias.h" #include "llvm/GlobalVariable.h" #include "llvm/Instructions.h" -#include "llvm/Intrinsics.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/Target/TargetLibraryInfo.h" +#include "llvm/ADT/SmallPtrSet.h" #include "llvm/ADT/SmallVector.h" -#include "llvm/ADT/STLExtras.h" -#include "llvm/Support/Compiler.h" +#include "llvm/Support/ErrorHandling.h" #include "llvm/Support/GetElementPtrTypeIterator.h" -#include "llvm/Support/ManagedStatic.h" #include using namespace llvm; -namespace { - /// NoAA - This class implements the -no-aa pass, which always returns "I - /// don't know" for alias queries. NoAA is unlike other alias analysis - /// implementations, in that it does not chain to a previous analysis. As - /// such it doesn't follow many of the rules that other alias analyses must. - /// - struct VISIBILITY_HIDDEN NoAA : public ImmutablePass, public AliasAnalysis { - static char ID; // Class identification, replacement for typeinfo - NoAA() : ImmutablePass((intptr_t)&ID) {} - explicit NoAA(intptr_t PID) : ImmutablePass(PID) { } +//===----------------------------------------------------------------------===// +// Useful predicates +//===----------------------------------------------------------------------===// - virtual void getAnalysisUsage(AnalysisUsage &AU) const { - AU.addRequired(); - } +/// 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; +} - virtual void initializePass() { - TD = &getAnalysis(); +/// 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; +} - virtual AliasResult alias(const Value *V1, unsigned V1Size, - const Value *V2, unsigned V2Size) { - return MayAlias; +/// 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) { + 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; +} - virtual ModRefBehavior getModRefBehavior(Function *F, CallSite CS, - std::vector *Info) { - return UnknownModRefBehavior; +/// 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; } - virtual void getArgumentAccesses(Function *F, CallSite CS, - std::vector &Info) { - assert(0 && "This method may not be called on this function!"); + const GEPOperator *GEPOp = dyn_cast(Op); + if (GEPOp == 0) { + // If it's not a GEP, hand it off to SimplifyInstruction to see if it + // can come up with something. This matches what GetUnderlyingObject does. + 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; + } + + 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 (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, + static_cast(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; +} - virtual void getMustAliases(Value *P, std::vector &RetVals) { } - virtual bool pointsToConstantMemory(const Value *P) { return false; } - virtual ModRefResult getModRefInfo(CallSite CS, Value *P, unsigned Size) { - return ModRef; +/// 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; } - virtual ModRefResult getModRefInfo(CallSite CS1, CallSite CS2) { - return ModRef; + + // 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); } - virtual bool hasNoModRefInfoForCalls() const { return true; } + } +} - virtual void deleteValue(Value *V) {} - virtual void copyValue(Value *From, Value *To) {} - }; +//===----------------------------------------------------------------------===// +// BasicAliasAnalysis Pass +//===----------------------------------------------------------------------===// - // Register this pass... - char NoAA::ID = 0; - RegisterPass - U("no-aa", "No Alias Analysis (always returns 'may' alias)"); +#ifndef NDEBUG +static const Function *getParent(const Value *V) { + if (const Instruction *inst = dyn_cast(V)) + return inst->getParent()->getParent(); - // Declare that we implement the AliasAnalysis interface - RegisterAnalysisGroup V(U); -} // End of anonymous namespace + if (const Argument *arg = dyn_cast(V)) + return arg->getParent(); + + return NULL; +} -ImmutablePass *llvm::createNoAAPass() { return new NoAA(); } +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 default alias analysis implementation. - /// Because it doesn't chain to a previous alias analysis (like -no-aa), it - /// derives from the NoAA class. - struct VISIBILITY_HIDDEN BasicAliasAnalysis : public NoAA { + /// BasicAliasAnalysis - This is the primary alias analysis implementation. + struct BasicAliasAnalysis : public ImmutablePass, public AliasAnalysis { static char ID; // Class identification, replacement for typeinfo - BasicAliasAnalysis() : NoAA((intptr_t)&ID) { } - AliasResult alias(const Value *V1, unsigned V1Size, - const Value *V2, unsigned V2Size); + BasicAliasAnalysis() : ImmutablePass(ID), + // AliasCache rarely has more than 1 or 2 elements, + // so start it off fairly small so that clear() + // doesn't have to tromp through 64 (the default) + // elements on each alias query. This really wants + // something like a SmallDenseMap. + AliasCache(8) { + initializeBasicAliasAnalysisPass(*PassRegistry::getPassRegistry()); + } - ModRefResult getModRefInfo(CallSite CS, Value *P, unsigned Size); - ModRefResult getModRefInfo(CallSite CS1, CallSite CS2) { - return NoAA::getModRefInfo(CS1,CS2); + virtual void initializePass() { + InitializeAliasAnalysis(this); } - /// hasNoModRefInfoForCalls - We can provide mod/ref information against - /// non-escaping allocations. - virtual bool hasNoModRefInfoForCalls() const { return false; } + virtual void getAnalysisUsage(AnalysisUsage &AU) const { + AU.addRequired(); + AU.addRequired(); + } + + virtual AliasResult alias(const Location &LocA, + const Location &LocB) { + assert(AliasCache.empty() && "AliasCache 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); + AliasCache.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. - bool pointsToConstantMemory(const Value *P); - + 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: - // CheckGEPInstructions - Check two GEP instructions with known - // must-aliasing base pointers. This checks to see if the index expressions - // preclude the pointers from aliasing... - AliasResult - CheckGEPInstructions(const Type* BasePtr1Ty, - Value **GEP1Ops, unsigned NumGEP1Ops, unsigned G1Size, - const Type *BasePtr2Ty, - Value **GEP2Ops, unsigned NumGEP2Ops, unsigned G2Size); + // AliasCache - Track alias queries to guard against recursion. + typedef std::pair LocPair; + typedef DenseMap AliasCacheTy; + AliasCacheTy AliasCache; + + // Visited - Track instructions visited by pointsToConstantMemory. + 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; - RegisterPass - X("basicaa", "Basic Alias Analysis (default AA impl)"); +// Register this pass... +char BasicAliasAnalysis::ID = 0; +INITIALIZE_AG_PASS_BEGIN(BasicAliasAnalysis, AliasAnalysis, "basicaa", + "Basic Alias Analysis (stateless AA impl)", + false, true, false) +INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo) +INITIALIZE_AG_PASS_END(BasicAliasAnalysis, AliasAnalysis, "basicaa", + "Basic Alias Analysis (stateless AA impl)", + false, true, false) - // Declare that we implement the AliasAnalysis interface - RegisterAnalysisGroup Y(X); -} // End of anonymous namespace ImmutablePass *llvm::createBasicAliasAnalysisPass() { return new BasicAliasAnalysis(); } -// getUnderlyingObject - This traverses the use chain to figure out what object -// the specified value points to. If the value points to, or is derived from, a -// unique object or an argument, return it. -static const Value *getUnderlyingObject(const Value *V) { - if (!isa(V->getType())) return 0; +/// 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); + } - // If we are at some type of object, return it. GlobalValues and Allocations - // have unique addresses. - if (isa(V) || isa(V) || isa(V)) - return V; + // 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; + } - // Traverse through different addressing mechanisms... - if (const Instruction *I = dyn_cast(V)) { - if (isa(I) || isa(I)) - return getUnderlyingObject(I->getOperand(0)); - } else if (const ConstantExpr *CE = dyn_cast(V)) { - if (CE->getOpcode() == Instruction::BitCast || - CE->getOpcode() == Instruction::GetElementPtr) - return getUnderlyingObject(CE->getOperand(0)); - } - return 0; -} + // 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; + } -static const User *isGEP(const Value *V) { - if (isa(V) || - (isa(V) && - cast(V)->getOpcode() == Instruction::GetElementPtr)) - return cast(V); - return 0; -} + // 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; + } -static const Value *GetGEPOperands(const Value *V, - SmallVector &GEPOps){ - assert(GEPOps.empty() && "Expect empty list to populate!"); - GEPOps.insert(GEPOps.end(), cast(V)->op_begin()+1, - cast(V)->op_end()); - - // Accumulate all of the chained indexes into the operand array - V = cast(V)->getOperand(0); - - while (const User *G = isGEP(V)) { - if (!isa(GEPOps[0]) || isa(GEPOps[0]) || - !cast(GEPOps[0])->isNullValue()) - break; // Don't handle folding arbitrary pointer offsets yet... - GEPOps.erase(GEPOps.begin()); // Drop the zero index - GEPOps.insert(GEPOps.begin(), G->op_begin()+1, G->op_end()); - V = G->getOperand(0); - } - return V; -} + // Otherwise be conservative. + Visited.clear(); + return AliasAnalysis::pointsToConstantMemory(Loc, OrLocal); -/// pointsToConstantMemory - Chase pointers until we find a (constant -/// global) or not. -bool BasicAliasAnalysis::pointsToConstantMemory(const Value *P) { - if (const Value *V = getUnderlyingObject(P)) - if (const GlobalVariable *GV = dyn_cast(V)) - return GV->isConstant(); - return false; -} + } while (!Worklist.empty() && --MaxLookup); -// Determine if an AllocationInst instruction escapes from the function it is -// contained in. If it does not escape, there is no way for another function to -// mod/ref it. We do this by looking at its uses and determining if the uses -// can escape (recursively). -static bool AddressMightEscape(const Value *V) { - for (Value::use_const_iterator UI = V->use_begin(), E = V->use_end(); - UI != E; ++UI) { - const Instruction *I = cast(*UI); - switch (I->getOpcode()) { - case Instruction::Load: - break; //next use. - case Instruction::Store: - if (I->getOperand(0) == V) - return true; // Escapes if the pointer is stored. - break; // next use. - case Instruction::GetElementPtr: - if (AddressMightEscape(I)) - return true; - break; // next use. - case Instruction::BitCast: - if (!isa(I->getType())) - return true; - if (AddressMightEscape(I)) - return true; - break; // next use - case Instruction::Ret: - // If returned, the address will escape to calling functions, but no - // callees could modify it. - break; // next use - default: - return true; - } - } - return false; + Visited.clear(); + return Worklist.empty(); } -// 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(CallSite CS, Value *P, unsigned Size) { - if (!isa(P)) - if (const AllocationInst *AI = - dyn_cast_or_null(getUnderlyingObject(P))) { - // Okay, the pointer is to a stack allocated object. If we can prove that - // the pointer never "escapes", then we know the call cannot clobber it, - // because it simply can't get its address. - if (!AddressMightEscape(AI)) - return NoModRef; +/// 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; - // If this is a tail call and P points to a stack location, we know that - // the tail call cannot access or modify the local stack. - if (CallInst *CI = dyn_cast(CS.getInstruction())) - if (CI->isTailCall() && isa(AI)) - return NoModRef; - } + 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 AliasAnalysis::getModRefInfo(CS, P, Size); + return ModRefBehavior(AliasAnalysis::getModRefBehavior(CS) & Min); } -static bool isNoAliasArgument(const Argument *Arg) { - const Function *Func = Arg->getParent(); - const ParamAttrsList *Attr = Func->getParamAttrs(); - if (Attr) { - unsigned Idx = 1; - for (Function::const_arg_iterator I = Func->arg_begin(), - E = Func->arg_end(); I != E; ++I, ++Idx) { - if (&(*I) == Arg && - Attr->paramHasAttr(Idx, ParamAttr::NoAlias)) - return true; - } +/// 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 } - return false; -} -// alias - Provide a bunch of ad-hoc rules to disambiguate in common cases, such -// as array references. Note that this function is heavily tail recursive. -// Hopefully we have a smart C++ compiler. :) -// -AliasAnalysis::AliasResult -BasicAliasAnalysis::alias(const Value *V1, unsigned V1Size, - const Value *V2, unsigned V2Size) { - // Strip off any constant expression casts if they exist - if (const ConstantExpr *CE = dyn_cast(V1)) - if (CE->isCast() && isa(CE->getOperand(0)->getType())) - V1 = CE->getOperand(0); - if (const ConstantExpr *CE = dyn_cast(V2)) - if (CE->isCast() && isa(CE->getOperand(0)->getType())) - V2 = CE->getOperand(0); + ModRefBehavior Min = UnknownModRefBehavior; - // Are we checking for alias of the same value? - if (V1 == V2) return MustAlias; + // If the function declares it only reads memory, go with that. + if (F->onlyReadsMemory()) + Min = OnlyReadsMemory; - if ((!isa(V1->getType()) || !isa(V2->getType())) && - V1->getType() != Type::Int64Ty && V2->getType() != Type::Int64Ty) - return NoAlias; // Scalars cannot alias each other + // Otherwise be conservative. + return ModRefBehavior(AliasAnalysis::getModRefBehavior(F) & Min); +} - // Strip off cast instructions... - if (const BitCastInst *I = dyn_cast(V1)) - return alias(I->getOperand(0), V1Size, V2, V2Size); - if (const BitCastInst *I = dyn_cast(V2)) - return alias(V1, V1Size, I->getOperand(0), V2Size); - - // Figure out what objects these things are pointing to if we can... - const Value *O1 = getUnderlyingObject(V1); - const Value *O2 = getUnderlyingObject(V2); - - // Pointing at a discernible object? - if (O1) { - if (O2) { - if (const Argument *O1Arg = dyn_cast(O1)) { - // Incoming argument cannot alias locally allocated object! - if (isa(O2)) return NoAlias; - - // If they are two different objects, and one is a noalias argument - // then they do not alias. - if (O1 != O2 && isNoAliasArgument(O1Arg)) - return NoAlias; - - // Otherwise, nothing is known... - } - - if (const Argument *O2Arg = dyn_cast(O2)) { - // Incoming argument cannot alias locally allocated object! - if (isa(O1)) return NoAlias; - - // If they are two different objects, and one is a noalias argument - // then they do not alias. - if (O1 != O2 && isNoAliasArgument(O2Arg)) - return NoAlias; - - // Otherwise, nothing is known... +/// 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 or byval pointer arguments. If this + // pointer were passed to arguments that were neither of these, then it + // couldn't be no-capture. + if (!(*CI)->getType()->isPointerTy() || + (!CS.paramHasAttr(ArgNo+1, Attribute::NoCapture) && + !CS.paramHasAttr(ArgNo+1, Attribute::ByVal))) + continue; - } else if (O1 != O2) { - if (!isa(O1)) - // If they are two different objects, and neither is an argument, - // we know that we have no alias... - return NoAlias; + // 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 they are the same object, they we can look at the indexes. If they - // index off of the object is the same for both pointers, they must alias. - // If they are provably different, they must not alias. Otherwise, we - // can't tell anything. } + + if (!PassedAsArg) + return NoModRef; + } - - if (!isa(O1) && isa(V2)) - return NoAlias; // Unique values don't alias null - - if (isa(O1) || - (isa(O1) && - !cast(O1)->isArrayAllocation())) - if (cast(O1->getType())->getElementType()->isSized()) { - // If the size of the other access is larger than the total size of the - // global/alloca/malloc, it cannot be accessing the global (it's - // undefined to load or store bytes before or after an object). - const Type *ElTy = cast(O1->getType())->getElementType(); - unsigned GlobalSize = getTargetData().getABITypeSize(ElTy); - if (GlobalSize < V2Size && V2Size != ~0U) - return NoAlias; + const TargetLibraryInfo &TLI = getAnalysis(); + 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; + } + case Intrinsic::arm_neon_vld1: { + // LLVM's vld1 and vst1 intrinsics currently only support a single + // vector register. + uint64_t Size = + TD ? TD->getTypeStoreSize(II->getType()) : UnknownSize; + if (isNoAlias(Location(II->getArgOperand(0), Size, + II->getMetadata(LLVMContext::MD_tbaa)), + Loc)) + return NoModRef; + break; + } + case Intrinsic::arm_neon_vst1: { + uint64_t Size = + TD ? TD->getTypeStoreSize(II->getArgOperand(1)->getType()) : UnknownSize; + if (isNoAlias(Location(II->getArgOperand(0), Size, + II->getMetadata(LLVMContext::MD_tbaa)), + Loc)) + return NoModRef; + break; + } + } - if (O2) { - if (!isa(O2) && isa(V1)) - return NoAlias; // Unique values don't alias null - - if (isa(O2) || - (isa(O2) && - !cast(O2)->isArrayAllocation())) - if (cast(O2->getType())->getElementType()->isSized()) { - // If the size of the other access is larger than the total size of the - // global/alloca/malloc, it cannot be accessing the object (it's - // undefined to load or store bytes before or after an object). - const Type *ElTy = cast(O2->getType())->getElementType(); - unsigned GlobalSize = getTargetData().getABITypeSize(ElTy); - if (GlobalSize < V1Size && V1Size != ~0U) - return NoAlias; + // We can bound the aliasing properties of memset_pattern16 just as we can + // for memcpy/memset. This is particularly important because the + // LoopIdiomRecognizer likes to turn loops into calls to memset_pattern16 + // whenever possible. + else if (TLI.has(LibFunc::memset_pattern16) && + CS.getCalledFunction() && + CS.getCalledFunction()->getName() == "memset_pattern16") { + const Function *MS = CS.getCalledFunction(); + FunctionType *MemsetType = MS->getFunctionType(); + if (!MemsetType->isVarArg() && MemsetType->getNumParams() == 3 && + isa(MemsetType->getParamType(0)) && + isa(MemsetType->getParamType(1)) && + isa(MemsetType->getParamType(2))) { + uint64_t Len = UnknownSize; + if (const ConstantInt *LenCI = dyn_cast(CS.getArgument(2))) + Len = LenCI->getZExtValue(); + const Value *Dest = CS.getArgument(0); + const Value *Src = CS.getArgument(1); + // If it can't overlap the source dest, then it doesn't modref the loc. + if (isNoAlias(Location(Dest, Len), Loc)) { + // Always reads 16 bytes of the source. + if (isNoAlias(Location(Src, 16), Loc)) + return NoModRef; + // If it can't overlap the dest, then worst case it reads the loc. + Min = Ref; + // Always reads 16 bytes of the source. + } else if (isNoAlias(Location(Src, 16), Loc)) { + // If it can't overlap the source, then worst case it mutates the loc. + Min = Mod; } + } } + // 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) { + 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. - // Note that we also handle chains of getelementptr instructions as well as - // constant expression getelementptrs here. - // - if (isGEP(V1) && isGEP(V2)) { - // Drill down into the first non-gep value, to test for must-aliasing of - // the base pointers. - const User *G = cast(V1); - while (isGEP(G->getOperand(0)) && - G->getOperand(1) == - Constant::getNullValue(G->getOperand(1)->getType())) - G = cast(G->getOperand(0)); - const Value *BasePtr1 = G->getOperand(0); - - G = cast(V2); - while (isGEP(G->getOperand(0)) && - G->getOperand(1) == - Constant::getNullValue(G->getOperand(1)->getType())) - G = cast(G->getOperand(0)); - const Value *BasePtr2 = G->getOperand(0); - + if (const GEPOperator *GEP2 = dyn_cast(V2)) { // Do the base pointers alias? - AliasResult BaseAlias = alias(BasePtr1, ~0U, BasePtr2, ~0U); - if (BaseAlias == NoAlias) return NoAlias; - if (BaseAlias == MustAlias) { - // If the base pointers alias each other exactly, check to see if we can - // figure out anything about the resultant pointers, to try to prove - // non-aliasing. - - // Collect all of the chained GEP operands together into one simple place - SmallVector GEP1Ops, GEP2Ops; - BasePtr1 = GetGEPOperands(V1, GEP1Ops); - BasePtr2 = GetGEPOperands(V2, GEP2Ops); - - // If GetGEPOperands were able to fold to the same must-aliased pointer, - // do the comparison. - if (BasePtr1 == BasePtr2) { - AliasResult GAlias = - CheckGEPInstructions(BasePtr1->getType(), - &GEP1Ops[0], GEP1Ops.size(), V1Size, - BasePtr2->getType(), - &GEP2Ops[0], GEP2Ops.size(), V2Size); - if (GAlias != MayAlias) - return GAlias; + 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 constant 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 the difference is + // greater, we know they do not overlap. + if (GEP1BaseOffset != 0 && GEP1VariableIndices.empty()) { + if (GEP1BaseOffset >= 0) { + if (V2Size != UnknownSize) { + if ((uint64_t)GEP1BaseOffset < V2Size) + return PartialAlias; + return NoAlias; + } + } else { + if (V1Size != UnknownSize) { + if (-(uint64_t)GEP1BaseOffset < V1Size) + return PartialAlias; + return NoAlias; } } } - // Check to see if these two pointers are related by a getelementptr - // instruction. If one pointer is a GEP with a non-zero index of the other - // pointer, we know they cannot alias. - // - if (isGEP(V2)) { - std::swap(V1, V2); - std::swap(V1Size, V2Size); + // Try to distinguish something like &A[i][1] against &A[42][0]. + // Grab the least significant bit set in any of the scales. + if (!GEP1VariableIndices.empty()) { + uint64_t Modulo = 0; + for (unsigned i = 0, e = GEP1VariableIndices.size(); i != e; ++i) + Modulo |= (uint64_t)GEP1VariableIndices[i].Scale; + Modulo = Modulo ^ (Modulo & (Modulo - 1)); + + // We can compute the difference between the two addresses + // mod Modulo. Check whether that difference guarantees that the + // two locations do not alias. + uint64_t ModOffset = (uint64_t)GEP1BaseOffset & (Modulo - 1); + if (V1Size != UnknownSize && V2Size != UnknownSize && + ModOffset >= V2Size && V1Size <= Modulo - ModOffset) + return NoAlias; } - if (V1Size != ~0U && V2Size != ~0U) - if (isGEP(V1)) { - SmallVector GEPOperands; - const Value *BasePtr = GetGEPOperands(V1, GEPOperands); - - AliasResult R = alias(BasePtr, V1Size, V2, V2Size); - if (R == MustAlias) { - // If there is at least one non-zero constant index, we know they cannot - // alias. - bool ConstantFound = false; - bool AllZerosFound = true; - for (unsigned i = 0, e = GEPOperands.size(); i != e; ++i) - if (const Constant *C = dyn_cast(GEPOperands[i])) { - if (!C->isNullValue()) { - ConstantFound = true; - AllZerosFound = false; - break; - } - } else { - AllZerosFound = false; - } - - // If we have getelementptr , 0, 0, 0, 0, ... and V2 must aliases - // the ptr, the end result is a must alias also. - if (AllZerosFound) - return MustAlias; - - if (ConstantFound) { - if (V2Size <= 1 && V1Size <= 1) // Just pointer check? - return NoAlias; - - // Otherwise we have to check to see that the distance is more than - // the size of the argument... build an index vector that is equal to - // the arguments provided, except substitute 0's for any variable - // indexes we find... - if (cast( - BasePtr->getType())->getElementType()->isSized()) { - for (unsigned i = 0; i != GEPOperands.size(); ++i) - if (!isa(GEPOperands[i])) - GEPOperands[i] = - Constant::getNullValue(GEPOperands[i]->getType()); - int64_t Offset = - getTargetData().getIndexedOffset(BasePtr->getType(), - &GEPOperands[0], - GEPOperands.size()); - - if (Offset >= (int64_t)V2Size || Offset <= -(int64_t)V1Size) - return NoAlias; - } - } - } - } + // Statically, we can see that the base objects are the same, but the + // pointers have dynamic offsets which we can't resolve. And none of our + // little tricks above worked. + // + // TODO: Returning PartialAlias instead of MayAlias is a mild hack; the + // practical effect of this is protecting TBAA in the case of dynamic + // indices into arrays of unions. An alternative way to solve this would + // be to have clang emit extra metadata for unions and/or union accesses. + // A union-specific solution wouldn't handle the problem for malloc'd + // memory however. + return PartialAlias; +} - return MayAlias; +static AliasAnalysis::AliasResult +MergeAliasResults(AliasAnalysis::AliasResult A, AliasAnalysis::AliasResult B) { + // If the results agree, take it. + if (A == B) + return A; + // A mix of PartialAlias and MustAlias is PartialAlias. + if ((A == AliasAnalysis::PartialAlias && B == AliasAnalysis::MustAlias) || + (B == AliasAnalysis::PartialAlias && A == AliasAnalysis::MustAlias)) + return AliasAnalysis::PartialAlias; + // Otherwise, we don't know anything. + return AliasAnalysis::MayAlias; } -// This function is used to determin if the indices of two GEP instructions are -// equal. V1 and V2 are the indices. -static bool IndexOperandsEqual(Value *V1, Value *V2) { - if (V1->getType() == V2->getType()) - return V1 == V2; - if (Constant *C1 = dyn_cast(V1)) - if (Constant *C2 = dyn_cast(V2)) { - // Sign extend the constants to long types, if necessary - if (C1->getType() != Type::Int64Ty) - C1 = ConstantExpr::getSExt(C1, Type::Int64Ty); - if (C2->getType() != Type::Int64Ty) - C2 = ConstantExpr::getSExt(C2, Type::Int64Ty); - return C1 == C2; +/// 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 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); + return MergeAliasResults(ThisAlias, Alias); } - return false; -} -/// CheckGEPInstructions - Check two GEP instructions with known must-aliasing -/// base pointers. This checks to see if the index expressions preclude the -/// pointers from aliasing... -AliasAnalysis::AliasResult -BasicAliasAnalysis::CheckGEPInstructions( - const Type* BasePtr1Ty, Value **GEP1Ops, unsigned NumGEP1Ops, unsigned G1S, - const Type *BasePtr2Ty, Value **GEP2Ops, unsigned NumGEP2Ops, unsigned G2S) { - // We currently can't handle the case when the base pointers have different - // primitive types. Since this is uncommon anyway, we are happy being - // extremely conservative. - if (BasePtr1Ty != BasePtr2Ty) + // 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; - const PointerType *GEPPointerTy = cast(BasePtr1Ty); - - // Find the (possibly empty) initial sequence of equal values... which are not - // necessarily constants. - unsigned NumGEP1Operands = NumGEP1Ops, NumGEP2Operands = NumGEP2Ops; - unsigned MinOperands = std::min(NumGEP1Operands, NumGEP2Operands); - unsigned MaxOperands = std::max(NumGEP1Operands, NumGEP2Operands); - unsigned UnequalOper = 0; - while (UnequalOper != MinOperands && - IndexOperandsEqual(GEP1Ops[UnequalOper], GEP2Ops[UnequalOper])) { - // Advance through the type as we go... - ++UnequalOper; - if (const CompositeType *CT = dyn_cast(BasePtr1Ty)) - BasePtr1Ty = CT->getTypeAtIndex(GEP1Ops[UnequalOper-1]); - else { - // If all operands equal each other, then the derived pointers must - // alias each other... - BasePtr1Ty = 0; - assert(UnequalOper == NumGEP1Operands && UnequalOper == NumGEP2Operands && - "Ran out of type nesting, but not out of operands?"); - return MustAlias; - } - } + AliasResult ThisAlias = + aliasCheck(V2, V2Size, V2TBAAInfo, SI->getFalseValue(), SISize, SITBAAInfo); + return MergeAliasResults(ThisAlias, Alias); +} - // If we have seen all constant operands, and run out of indexes on one of the - // getelementptrs, check to see if the tail of the leftover one is all zeros. - // If so, return mustalias. - if (UnequalOper == MinOperands) { - if (NumGEP1Ops < NumGEP2Ops) { - std::swap(GEP1Ops, GEP2Ops); - std::swap(NumGEP1Ops, NumGEP2Ops); +// 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) { + // 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); + Alias = MergeAliasResults(ThisAlias, Alias); + if (Alias == MayAlias) + break; + } + return Alias; } - bool AllAreZeros = true; - for (unsigned i = UnequalOper; i != MaxOperands; ++i) - if (!isa(GEP1Ops[i]) || - !cast(GEP1Ops[i])->isNullValue()) { - AllAreZeros = false; - break; - } - if (AllAreZeros) return MustAlias; + 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; - // So now we know that the indexes derived from the base pointers, - // which are known to alias, are different. We can still determine a - // no-alias result if there are differing constant pairs in the index - // chain. For example: - // A[i][0] != A[j][1] iff (&A[0][1]-&A[0][0] >= std::max(G1S, G2S)) - // - // We have to be careful here about array accesses. In particular, consider: - // A[1][0] vs A[0][i] - // In this case, we don't *know* that the array will be accessed in bounds: - // the index could even be negative. Because of this, we have to - // conservatively *give up* and return may alias. We disregard differing - // array subscripts that are followed by a variable index without going - // through a struct. - // - unsigned SizeMax = std::max(G1S, G2S); - if (SizeMax == ~0U) return MayAlias; // Avoid frivolous work. - - // Scan for the first operand that is constant and unequal in the - // two getelementptrs... - unsigned FirstConstantOper = UnequalOper; - for (; FirstConstantOper != MinOperands; ++FirstConstantOper) { - const Value *G1Oper = GEP1Ops[FirstConstantOper]; - const Value *G2Oper = GEP2Ops[FirstConstantOper]; - - if (G1Oper != G2Oper) // Found non-equal constant indexes... - if (Constant *G1OC = dyn_cast(const_cast(G1Oper))) - if (Constant *G2OC = dyn_cast(const_cast(G2Oper))){ - if (G1OC->getType() != G2OC->getType()) { - // Sign extend both operands to long. - if (G1OC->getType() != Type::Int64Ty) - G1OC = ConstantExpr::getSExt(G1OC, Type::Int64Ty); - if (G2OC->getType() != Type::Int64Ty) - G2OC = ConstantExpr::getSExt(G2OC, Type::Int64Ty); - GEP1Ops[FirstConstantOper] = G1OC; - GEP2Ops[FirstConstantOper] = G2OC; - } - - if (G1OC != G2OC) { - // Handle the "be careful" case above: if this is an array/vector - // subscript, scan for a subsequent variable array index. - if (isa(BasePtr1Ty)) { - const Type *NextTy = - cast(BasePtr1Ty)->getElementType(); - bool isBadCase = false; - - for (unsigned Idx = FirstConstantOper+1; - Idx != MinOperands && isa(NextTy); ++Idx) { - const Value *V1 = GEP1Ops[Idx], *V2 = GEP2Ops[Idx]; - if (!isa(V1) || !isa(V2)) { - isBadCase = true; - break; - } - NextTy = cast(NextTy)->getElementType(); - } - - if (isBadCase) G1OC = 0; - } - - // Make sure they are comparable (ie, not constant expressions), and - // make sure the GEP with the smaller leading constant is GEP1. - if (G1OC) { - Constant *Compare = ConstantExpr::getICmp(ICmpInst::ICMP_SGT, - G1OC, G2OC); - if (ConstantInt *CV = dyn_cast(Compare)) { - if (CV->getZExtValue()) { // If they are comparable and G2 > G1 - std::swap(GEP1Ops, GEP2Ops); // Make GEP1 < GEP2 - std::swap(NumGEP1Ops, NumGEP2Ops); - } - break; - } - } - } - } - BasePtr1Ty = cast(BasePtr1Ty)->getTypeAtIndex(G1Oper); + // 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]; + + AliasResult ThisAlias = aliasCheck(V2, V2Size, V2TBAAInfo, + V, PNSize, PNTBAAInfo); + Alias = MergeAliasResults(ThisAlias, Alias); + if (Alias == MayAlias) + break; } - // No shared constant operands, and we ran out of common operands. At this - // point, the GEP instructions have run through all of their operands, and we - // haven't found evidence that there are any deltas between the GEP's. - // However, one GEP may have more operands than the other. If this is the - // case, there may still be hope. Check this now. - if (FirstConstantOper == MinOperands) { - // Make GEP1Ops be the longer one if there is a longer one. - if (NumGEP1Ops < NumGEP2Ops) { - std::swap(GEP1Ops, GEP2Ops); - std::swap(NumGEP1Ops, NumGEP2Ops); - } - - // Is there anything to check? - if (NumGEP1Ops > MinOperands) { - for (unsigned i = FirstConstantOper; i != MaxOperands; ++i) - if (isa(GEP1Ops[i]) && - !cast(GEP1Ops[i])->isZero()) { - // Yup, there's a constant in the tail. Set all variables to - // constants in the GEP instruction to make it suiteable for - // TargetData::getIndexedOffset. - for (i = 0; i != MaxOperands; ++i) - if (!isa(GEP1Ops[i])) - GEP1Ops[i] = Constant::getNullValue(GEP1Ops[i]->getType()); - // Okay, now get the offset. This is the relative offset for the full - // instruction. - const TargetData &TD = getTargetData(); - int64_t Offset1 = TD.getIndexedOffset(GEPPointerTy, GEP1Ops, - NumGEP1Ops); - - // Now check without any constants at the end. - int64_t Offset2 = TD.getIndexedOffset(GEPPointerTy, GEP1Ops, - MinOperands); - - // If the tail provided a bit enough offset, return noalias! - if ((uint64_t)(Offset2-Offset1) >= SizeMax) - return NoAlias; - } - } + return Alias; +} - // Couldn't find anything useful. - return MayAlias; - } +// 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(); - // If there are non-equal constants arguments, then we can figure - // out a minimum known delta between the two index expressions... at - // this point we know that the first constant index of GEP1 is less - // than the first constant index of GEP2. - - // Advance BasePtr[12]Ty over this first differing constant operand. - BasePtr2Ty = cast(BasePtr1Ty)-> - getTypeAtIndex(GEP2Ops[FirstConstantOper]); - BasePtr1Ty = cast(BasePtr1Ty)-> - getTypeAtIndex(GEP1Ops[FirstConstantOper]); - - // We are going to be using TargetData::getIndexedOffset to determine the - // offset that each of the GEP's is reaching. To do this, we have to convert - // all variable references to constant references. To do this, we convert the - // initial sequence of array subscripts into constant zeros to start with. - const Type *ZeroIdxTy = GEPPointerTy; - for (unsigned i = 0; i != FirstConstantOper; ++i) { - if (!isa(ZeroIdxTy)) - GEP1Ops[i] = GEP2Ops[i] = Constant::getNullValue(Type::Int32Ty); - - if (const CompositeType *CT = dyn_cast(ZeroIdxTy)) - ZeroIdxTy = CT->getTypeAtIndex(GEP1Ops[i]); - } + // Are we checking for alias of the same value? + if (V1 == V2) return MustAlias; - // We know that GEP1Ops[FirstConstantOper] & GEP2Ops[FirstConstantOper] are ok - - // Loop over the rest of the operands... - for (unsigned i = FirstConstantOper+1; i != MaxOperands; ++i) { - const Value *Op1 = i < NumGEP1Ops ? GEP1Ops[i] : 0; - const Value *Op2 = i < NumGEP2Ops ? GEP2Ops[i] : 0; - // If they are equal, use a zero index... - if (Op1 == Op2 && BasePtr1Ty == BasePtr2Ty) { - if (!isa(Op1)) - GEP1Ops[i] = GEP2Ops[i] = Constant::getNullValue(Op1->getType()); - // Otherwise, just keep the constants we have. - } else { - if (Op1) { - if (const ConstantInt *Op1C = dyn_cast(Op1)) { - // If this is an array index, make sure the array element is in range. - if (const ArrayType *AT = dyn_cast(BasePtr1Ty)) { - if (Op1C->getZExtValue() >= AT->getNumElements()) - return MayAlias; // Be conservative with out-of-range accesses - } else if (const VectorType *VT = dyn_cast(BasePtr1Ty)) { - if (Op1C->getZExtValue() >= VT->getNumElements()) - return MayAlias; // Be conservative with out-of-range accesses - } - - } else { - // GEP1 is known to produce a value less than GEP2. To be - // conservatively correct, we must assume the largest possible - // constant is used in this position. This cannot be the initial - // index to the GEP instructions (because we know we have at least one - // element before this one with the different constant arguments), so - // we know that the current index must be into either a struct or - // array. Because we know it's not constant, this cannot be a - // structure index. Because of this, we can calculate the maximum - // value possible. - // - if (const ArrayType *AT = dyn_cast(BasePtr1Ty)) - GEP1Ops[i] = ConstantInt::get(Type::Int64Ty,AT->getNumElements()-1); - else if (const VectorType *VT = dyn_cast(BasePtr1Ty)) - GEP1Ops[i] = ConstantInt::get(Type::Int64Ty,VT->getNumElements()-1); - } - } + if (!V1->getType()->isPointerTy() || !V2->getType()->isPointerTy()) + return NoAlias; // Scalars cannot alias each other - if (Op2) { - if (const ConstantInt *Op2C = dyn_cast(Op2)) { - // If this is an array index, make sure the array element is in range. - if (const ArrayType *AT = dyn_cast(BasePtr2Ty)) { - if (Op2C->getZExtValue() >= AT->getNumElements()) - return MayAlias; // Be conservative with out-of-range accesses - } else if (const VectorType *VT = dyn_cast(BasePtr2Ty)) { - if (Op2C->getZExtValue() >= VT->getNumElements()) - return MayAlias; // Be conservative with out-of-range accesses - } - } else { // Conservatively assume the minimum value for this index - GEP2Ops[i] = Constant::getNullValue(Op2->getType()); - } - } - } + // Figure out what objects these things are pointing to if we can. + const Value *O1 = GetUnderlyingObject(V1, TD); + const Value *O2 = GetUnderlyingObject(V2, TD); - if (BasePtr1Ty && Op1) { - if (const CompositeType *CT = dyn_cast(BasePtr1Ty)) - BasePtr1Ty = CT->getTypeAtIndex(GEP1Ops[i]); - else - BasePtr1Ty = 0; - } + // 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 (BasePtr2Ty && Op2) { - if (const CompositeType *CT = dyn_cast(BasePtr2Ty)) - BasePtr2Ty = CT->getTypeAtIndex(GEP2Ops[i]); - else - BasePtr2Ty = 0; - } + 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 (GEPPointerTy->getElementType()->isSized()) { - int64_t Offset1 = - getTargetData().getIndexedOffset(GEPPointerTy, GEP1Ops, NumGEP1Ops); - int64_t Offset2 = - getTargetData().getIndexedOffset(GEPPointerTy, GEP2Ops, NumGEP2Ops); - assert(Offset1 != Offset2 && - "There is at least one different constant here!"); - - // Make sure we compare the absolute difference. - if (Offset1 > Offset2) - std::swap(Offset1, Offset2); - - if ((uint64_t)(Offset2-Offset1) >= SizeMax) { - //cerr << "Determined that these two GEP's don't alias [" - // << SizeMax << " bytes]: \n" << *GEP1 << *GEP2; + // 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; - } + + // Check the cache before climbing up use-def chains. This also terminates + // otherwise infinitely recursive queries. + LocPair Locs(Location(V1, V1Size, V1TBAAInfo), + Location(V2, V2Size, V2TBAAInfo)); + if (V1 > V2) + std::swap(Locs.first, Locs.second); + std::pair Pair = + AliasCache.insert(std::make_pair(Locs, MayAlias)); + if (!Pair.second) + return Pair.first->second; + + // 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 AliasCache[Locs] = 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 AliasCache[Locs] = 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 AliasCache[Locs] = Result; } - return MayAlias; -} -// Make sure that anything that uses AliasAnalysis pulls in this file... -DEFINING_FILE_FOR(BasicAliasAnalysis) + // 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 AliasCache[Locs] = PartialAlias; + + AliasResult Result = + AliasAnalysis::alias(Location(V1, V1Size, V1TBAAInfo), + Location(V2, V2Size, V2TBAAInfo)); + return AliasCache[Locs] = Result; +}