//===- lib/Linker/IRMover.cpp ---------------------------------------------===// // // The LLVM Compiler Infrastructure // // This file is distributed under the University of Illinois Open Source // License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// #include "llvm/Linker/IRMover.h" #include "LinkDiagnosticInfo.h" #include "llvm/ADT/SetVector.h" #include "llvm/ADT/SmallString.h" #include "llvm/ADT/Triple.h" #include "llvm/IR/Constants.h" #include "llvm/IR/DebugInfo.h" #include "llvm/IR/DiagnosticPrinter.h" #include "llvm/IR/GVMaterializer.h" #include "llvm/IR/TypeFinder.h" #include "llvm/Transforms/Utils/Cloning.h" using namespace llvm; //===----------------------------------------------------------------------===// // TypeMap implementation. //===----------------------------------------------------------------------===// namespace { class TypeMapTy : public ValueMapTypeRemapper { /// This is a mapping from a source type to a destination type to use. DenseMap MappedTypes; /// When checking to see if two subgraphs are isomorphic, we speculatively /// add types to MappedTypes, but keep track of them here in case we need to /// roll back. SmallVector SpeculativeTypes; SmallVector SpeculativeDstOpaqueTypes; /// This is a list of non-opaque structs in the source module that are mapped /// to an opaque struct in the destination module. SmallVector SrcDefinitionsToResolve; /// This is the set of opaque types in the destination modules who are /// getting a body from the source module. SmallPtrSet DstResolvedOpaqueTypes; public: TypeMapTy(IRMover::IdentifiedStructTypeSet &DstStructTypesSet) : DstStructTypesSet(DstStructTypesSet) {} IRMover::IdentifiedStructTypeSet &DstStructTypesSet; /// Indicate that the specified type in the destination module is conceptually /// equivalent to the specified type in the source module. void addTypeMapping(Type *DstTy, Type *SrcTy); /// Produce a body for an opaque type in the dest module from a type /// definition in the source module. void linkDefinedTypeBodies(); /// Return the mapped type to use for the specified input type from the /// source module. Type *get(Type *SrcTy); Type *get(Type *SrcTy, SmallPtrSet &Visited); void finishType(StructType *DTy, StructType *STy, ArrayRef ETypes); FunctionType *get(FunctionType *T) { return cast(get((Type *)T)); } private: Type *remapType(Type *SrcTy) override { return get(SrcTy); } bool areTypesIsomorphic(Type *DstTy, Type *SrcTy); }; } void TypeMapTy::addTypeMapping(Type *DstTy, Type *SrcTy) { assert(SpeculativeTypes.empty()); assert(SpeculativeDstOpaqueTypes.empty()); // Check to see if these types are recursively isomorphic and establish a // mapping between them if so. if (!areTypesIsomorphic(DstTy, SrcTy)) { // Oops, they aren't isomorphic. Just discard this request by rolling out // any speculative mappings we've established. for (Type *Ty : SpeculativeTypes) MappedTypes.erase(Ty); SrcDefinitionsToResolve.resize(SrcDefinitionsToResolve.size() - SpeculativeDstOpaqueTypes.size()); for (StructType *Ty : SpeculativeDstOpaqueTypes) DstResolvedOpaqueTypes.erase(Ty); } else { for (Type *Ty : SpeculativeTypes) if (auto *STy = dyn_cast(Ty)) if (STy->hasName()) STy->setName(""); } SpeculativeTypes.clear(); SpeculativeDstOpaqueTypes.clear(); } /// Recursively walk this pair of types, returning true if they are isomorphic, /// false if they are not. bool TypeMapTy::areTypesIsomorphic(Type *DstTy, Type *SrcTy) { // Two types with differing kinds are clearly not isomorphic. if (DstTy->getTypeID() != SrcTy->getTypeID()) return false; // If we have an entry in the MappedTypes table, then we have our answer. Type *&Entry = MappedTypes[SrcTy]; if (Entry) return Entry == DstTy; // Two identical types are clearly isomorphic. Remember this // non-speculatively. if (DstTy == SrcTy) { Entry = DstTy; return true; } // Okay, we have two types with identical kinds that we haven't seen before. // If this is an opaque struct type, special case it. if (StructType *SSTy = dyn_cast(SrcTy)) { // Mapping an opaque type to any struct, just keep the dest struct. if (SSTy->isOpaque()) { Entry = DstTy; SpeculativeTypes.push_back(SrcTy); return true; } // Mapping a non-opaque source type to an opaque dest. If this is the first // type that we're mapping onto this destination type then we succeed. Keep // the dest, but fill it in later. If this is the second (different) type // that we're trying to map onto the same opaque type then we fail. if (cast(DstTy)->isOpaque()) { // We can only map one source type onto the opaque destination type. if (!DstResolvedOpaqueTypes.insert(cast(DstTy)).second) return false; SrcDefinitionsToResolve.push_back(SSTy); SpeculativeTypes.push_back(SrcTy); SpeculativeDstOpaqueTypes.push_back(cast(DstTy)); Entry = DstTy; return true; } } // If the number of subtypes disagree between the two types, then we fail. if (SrcTy->getNumContainedTypes() != DstTy->getNumContainedTypes()) return false; // Fail if any of the extra properties (e.g. array size) of the type disagree. if (isa(DstTy)) return false; // bitwidth disagrees. if (PointerType *PT = dyn_cast(DstTy)) { if (PT->getAddressSpace() != cast(SrcTy)->getAddressSpace()) return false; } else if (FunctionType *FT = dyn_cast(DstTy)) { if (FT->isVarArg() != cast(SrcTy)->isVarArg()) return false; } else if (StructType *DSTy = dyn_cast(DstTy)) { StructType *SSTy = cast(SrcTy); if (DSTy->isLiteral() != SSTy->isLiteral() || DSTy->isPacked() != SSTy->isPacked()) return false; } else if (ArrayType *DATy = dyn_cast(DstTy)) { if (DATy->getNumElements() != cast(SrcTy)->getNumElements()) return false; } else if (VectorType *DVTy = dyn_cast(DstTy)) { if (DVTy->getNumElements() != cast(SrcTy)->getNumElements()) return false; } // Otherwise, we speculate that these two types will line up and recursively // check the subelements. Entry = DstTy; SpeculativeTypes.push_back(SrcTy); for (unsigned I = 0, E = SrcTy->getNumContainedTypes(); I != E; ++I) if (!areTypesIsomorphic(DstTy->getContainedType(I), SrcTy->getContainedType(I))) return false; // If everything seems to have lined up, then everything is great. return true; } void TypeMapTy::linkDefinedTypeBodies() { SmallVector Elements; for (StructType *SrcSTy : SrcDefinitionsToResolve) { StructType *DstSTy = cast(MappedTypes[SrcSTy]); assert(DstSTy->isOpaque()); // Map the body of the source type over to a new body for the dest type. Elements.resize(SrcSTy->getNumElements()); for (unsigned I = 0, E = Elements.size(); I != E; ++I) Elements[I] = get(SrcSTy->getElementType(I)); DstSTy->setBody(Elements, SrcSTy->isPacked()); DstStructTypesSet.switchToNonOpaque(DstSTy); } SrcDefinitionsToResolve.clear(); DstResolvedOpaqueTypes.clear(); } void TypeMapTy::finishType(StructType *DTy, StructType *STy, ArrayRef ETypes) { DTy->setBody(ETypes, STy->isPacked()); // Steal STy's name. if (STy->hasName()) { SmallString<16> TmpName = STy->getName(); STy->setName(""); DTy->setName(TmpName); } DstStructTypesSet.addNonOpaque(DTy); } Type *TypeMapTy::get(Type *Ty) { SmallPtrSet Visited; return get(Ty, Visited); } Type *TypeMapTy::get(Type *Ty, SmallPtrSet &Visited) { // If we already have an entry for this type, return it. Type **Entry = &MappedTypes[Ty]; if (*Entry) return *Entry; // These are types that LLVM itself will unique. bool IsUniqued = !isa(Ty) || cast(Ty)->isLiteral(); #ifndef NDEBUG if (!IsUniqued) { for (auto &Pair : MappedTypes) { assert(!(Pair.first != Ty && Pair.second == Ty) && "mapping to a source type"); } } #endif if (!IsUniqued && !Visited.insert(cast(Ty)).second) { StructType *DTy = StructType::create(Ty->getContext()); return *Entry = DTy; } // If this is not a recursive type, then just map all of the elements and // then rebuild the type from inside out. SmallVector ElementTypes; // If there are no element types to map, then the type is itself. This is // true for the anonymous {} struct, things like 'float', integers, etc. if (Ty->getNumContainedTypes() == 0 && IsUniqued) return *Entry = Ty; // Remap all of the elements, keeping track of whether any of them change. bool AnyChange = false; ElementTypes.resize(Ty->getNumContainedTypes()); for (unsigned I = 0, E = Ty->getNumContainedTypes(); I != E; ++I) { ElementTypes[I] = get(Ty->getContainedType(I), Visited); AnyChange |= ElementTypes[I] != Ty->getContainedType(I); } // If we found our type while recursively processing stuff, just use it. Entry = &MappedTypes[Ty]; if (*Entry) { if (auto *DTy = dyn_cast(*Entry)) { if (DTy->isOpaque()) { auto *STy = cast(Ty); finishType(DTy, STy, ElementTypes); } } return *Entry; } // If all of the element types mapped directly over and the type is not // a nomed struct, then the type is usable as-is. if (!AnyChange && IsUniqued) return *Entry = Ty; // Otherwise, rebuild a modified type. switch (Ty->getTypeID()) { default: llvm_unreachable("unknown derived type to remap"); case Type::ArrayTyID: return *Entry = ArrayType::get(ElementTypes[0], cast(Ty)->getNumElements()); case Type::VectorTyID: return *Entry = VectorType::get(ElementTypes[0], cast(Ty)->getNumElements()); case Type::PointerTyID: return *Entry = PointerType::get(ElementTypes[0], cast(Ty)->getAddressSpace()); case Type::FunctionTyID: return *Entry = FunctionType::get(ElementTypes[0], makeArrayRef(ElementTypes).slice(1), cast(Ty)->isVarArg()); case Type::StructTyID: { auto *STy = cast(Ty); bool IsPacked = STy->isPacked(); if (IsUniqued) return *Entry = StructType::get(Ty->getContext(), ElementTypes, IsPacked); // If the type is opaque, we can just use it directly. if (STy->isOpaque()) { DstStructTypesSet.addOpaque(STy); return *Entry = Ty; } if (StructType *OldT = DstStructTypesSet.findNonOpaque(ElementTypes, IsPacked)) { STy->setName(""); return *Entry = OldT; } if (!AnyChange) { DstStructTypesSet.addNonOpaque(STy); return *Entry = Ty; } StructType *DTy = StructType::create(Ty->getContext()); finishType(DTy, STy, ElementTypes); return *Entry = DTy; } } } LinkDiagnosticInfo::LinkDiagnosticInfo(DiagnosticSeverity Severity, const Twine &Msg) : DiagnosticInfo(DK_Linker, Severity), Msg(Msg) {} void LinkDiagnosticInfo::print(DiagnosticPrinter &DP) const { DP << Msg; } //===----------------------------------------------------------------------===// // IRLinker implementation. //===----------------------------------------------------------------------===// namespace { class IRLinker; /// Creates prototypes for functions that are lazily linked on the fly. This /// speeds up linking for modules with many/ lazily linked functions of which /// few get used. class GlobalValueMaterializer final : public ValueMaterializer { IRLinker &TheIRLinker; public: GlobalValueMaterializer(IRLinker &TheIRLinker) : TheIRLinker(TheIRLinker) {} Value *materializeDeclFor(Value *V) override; void materializeInitFor(GlobalValue *New, GlobalValue *Old) override; }; class LocalValueMaterializer final : public ValueMaterializer { IRLinker &TheIRLinker; public: LocalValueMaterializer(IRLinker &TheIRLinker) : TheIRLinker(TheIRLinker) {} Value *materializeDeclFor(Value *V) override; void materializeInitFor(GlobalValue *New, GlobalValue *Old) override; }; /// Type of the Metadata map in \a ValueToValueMapTy. typedef DenseMap MDMapT; /// This is responsible for keeping track of the state used for moving data /// from SrcM to DstM. class IRLinker { Module &DstM; std::unique_ptr SrcM; /// See IRMover::move(). std::function AddLazyFor; TypeMapTy TypeMap; GlobalValueMaterializer GValMaterializer; LocalValueMaterializer LValMaterializer; /// A metadata map that's shared between IRLinker instances. MDMapT &SharedMDs; /// Mapping of values from what they used to be in Src, to what they are now /// in DstM. ValueToValueMapTy is a ValueMap, which involves some overhead /// due to the use of Value handles which the Linker doesn't actually need, /// but this allows us to reuse the ValueMapper code. ValueToValueMapTy ValueMap; ValueToValueMapTy AliasValueMap; DenseSet ValuesToLink; std::vector Worklist; void maybeAdd(GlobalValue *GV) { if (ValuesToLink.insert(GV).second) Worklist.push_back(GV); } /// Set to true when all global value body linking is complete (including /// lazy linking). Used to prevent metadata linking from creating new /// references. bool DoneLinkingBodies = false; bool HasError = false; /// Entry point for mapping values and alternate context for mapping aliases. ValueMapper Mapper; unsigned AliasMCID; /// Handles cloning of a global values from the source module into /// the destination module, including setting the attributes and visibility. GlobalValue *copyGlobalValueProto(const GlobalValue *SGV, bool ForDefinition); /// Helper method for setting a message and returning an error code. bool emitError(const Twine &Message) { SrcM->getContext().diagnose(LinkDiagnosticInfo(DS_Error, Message)); HasError = true; return true; } void emitWarning(const Twine &Message) { SrcM->getContext().diagnose(LinkDiagnosticInfo(DS_Warning, Message)); } /// Given a global in the source module, return the global in the /// destination module that is being linked to, if any. GlobalValue *getLinkedToGlobal(const GlobalValue *SrcGV) { // If the source has no name it can't link. If it has local linkage, // there is no name match-up going on. if (!SrcGV->hasName() || SrcGV->hasLocalLinkage()) return nullptr; // Otherwise see if we have a match in the destination module's symtab. GlobalValue *DGV = DstM.getNamedValue(SrcGV->getName()); if (!DGV) return nullptr; // If we found a global with the same name in the dest module, but it has // internal linkage, we are really not doing any linkage here. if (DGV->hasLocalLinkage()) return nullptr; // Otherwise, we do in fact link to the destination global. return DGV; } void computeTypeMapping(); Constant *linkAppendingVarProto(GlobalVariable *DstGV, const GlobalVariable *SrcGV); /// Given the GlobaValue \p SGV in the source module, and the matching /// GlobalValue \p DGV (if any), return true if the linker will pull \p SGV /// into the destination module. /// /// Note this code may call the client-provided \p AddLazyFor. bool shouldLink(GlobalValue *DGV, GlobalValue &SGV); Constant *linkGlobalValueProto(GlobalValue *GV, bool ForAlias); bool linkModuleFlagsMetadata(); void linkGlobalInit(GlobalVariable &Dst, GlobalVariable &Src); bool linkFunctionBody(Function &Dst, Function &Src); void linkAliasBody(GlobalAlias &Dst, GlobalAlias &Src); bool linkGlobalValueBody(GlobalValue &Dst, GlobalValue &Src); /// Functions that take care of cloning a specific global value type /// into the destination module. GlobalVariable *copyGlobalVariableProto(const GlobalVariable *SGVar); Function *copyFunctionProto(const Function *SF); GlobalValue *copyGlobalAliasProto(const GlobalAlias *SGA); void linkNamedMDNodes(); public: IRLinker(Module &DstM, MDMapT &SharedMDs, IRMover::IdentifiedStructTypeSet &Set, std::unique_ptr SrcM, ArrayRef ValuesToLink, std::function AddLazyFor) : DstM(DstM), SrcM(std::move(SrcM)), AddLazyFor(AddLazyFor), TypeMap(Set), GValMaterializer(*this), LValMaterializer(*this), SharedMDs(SharedMDs), Mapper(ValueMap, RF_MoveDistinctMDs | RF_IgnoreMissingLocals, &TypeMap, &GValMaterializer), AliasMCID(Mapper.registerAlternateMappingContext(AliasValueMap, &LValMaterializer)) { ValueMap.getMDMap() = std::move(SharedMDs); for (GlobalValue *GV : ValuesToLink) maybeAdd(GV); } ~IRLinker() { SharedMDs = std::move(*ValueMap.getMDMap()); } bool run(); Value *materializeDeclFor(Value *V, bool ForAlias); void materializeInitFor(GlobalValue *New, GlobalValue *Old, bool ForAlias); }; } /// The LLVM SymbolTable class autorenames globals that conflict in the symbol /// table. This is good for all clients except for us. Go through the trouble /// to force this back. static void forceRenaming(GlobalValue *GV, StringRef Name) { // If the global doesn't force its name or if it already has the right name, // there is nothing for us to do. if (GV->hasLocalLinkage() || GV->getName() == Name) return; Module *M = GV->getParent(); // If there is a conflict, rename the conflict. if (GlobalValue *ConflictGV = M->getNamedValue(Name)) { GV->takeName(ConflictGV); ConflictGV->setName(Name); // This will cause ConflictGV to get renamed assert(ConflictGV->getName() != Name && "forceRenaming didn't work"); } else { GV->setName(Name); // Force the name back } } Value *GlobalValueMaterializer::materializeDeclFor(Value *V) { return TheIRLinker.materializeDeclFor(V, false); } void GlobalValueMaterializer::materializeInitFor(GlobalValue *New, GlobalValue *Old) { TheIRLinker.materializeInitFor(New, Old, false); } Value *LocalValueMaterializer::materializeDeclFor(Value *V) { return TheIRLinker.materializeDeclFor(V, true); } void LocalValueMaterializer::materializeInitFor(GlobalValue *New, GlobalValue *Old) { TheIRLinker.materializeInitFor(New, Old, true); } Value *IRLinker::materializeDeclFor(Value *V, bool ForAlias) { auto *SGV = dyn_cast(V); if (!SGV) return nullptr; return linkGlobalValueProto(SGV, ForAlias); } void IRLinker::materializeInitFor(GlobalValue *New, GlobalValue *Old, bool ForAlias) { // If we already created the body, just return. if (auto *F = dyn_cast(New)) { if (!F->isDeclaration()) return; } else if (auto *V = dyn_cast(New)) { if (V->hasInitializer() || V->hasAppendingLinkage()) return; } else { auto *A = cast(New); if (A->getAliasee()) return; } if (ForAlias || shouldLink(New, *Old)) linkGlobalValueBody(*New, *Old); } /// Loop through the global variables in the src module and merge them into the /// dest module. GlobalVariable *IRLinker::copyGlobalVariableProto(const GlobalVariable *SGVar) { // No linking to be performed or linking from the source: simply create an // identical version of the symbol over in the dest module... the // initializer will be filled in later by LinkGlobalInits. GlobalVariable *NewDGV = new GlobalVariable(DstM, TypeMap.get(SGVar->getValueType()), SGVar->isConstant(), GlobalValue::ExternalLinkage, /*init*/ nullptr, SGVar->getName(), /*insertbefore*/ nullptr, SGVar->getThreadLocalMode(), SGVar->getType()->getAddressSpace()); NewDGV->setAlignment(SGVar->getAlignment()); return NewDGV; } /// Link the function in the source module into the destination module if /// needed, setting up mapping information. Function *IRLinker::copyFunctionProto(const Function *SF) { // If there is no linkage to be performed or we are linking from the source, // bring SF over. return Function::Create(TypeMap.get(SF->getFunctionType()), GlobalValue::ExternalLinkage, SF->getName(), &DstM); } /// Set up prototypes for any aliases that come over from the source module. GlobalValue *IRLinker::copyGlobalAliasProto(const GlobalAlias *SGA) { // If there is no linkage to be performed or we're linking from the source, // bring over SGA. auto *Ty = TypeMap.get(SGA->getValueType()); return GlobalAlias::create(Ty, SGA->getType()->getPointerAddressSpace(), GlobalValue::ExternalLinkage, SGA->getName(), &DstM); } GlobalValue *IRLinker::copyGlobalValueProto(const GlobalValue *SGV, bool ForDefinition) { GlobalValue *NewGV; if (auto *SGVar = dyn_cast(SGV)) { NewGV = copyGlobalVariableProto(SGVar); } else if (auto *SF = dyn_cast(SGV)) { NewGV = copyFunctionProto(SF); } else { if (ForDefinition) NewGV = copyGlobalAliasProto(cast(SGV)); else NewGV = new GlobalVariable( DstM, TypeMap.get(SGV->getValueType()), /*isConstant*/ false, GlobalValue::ExternalLinkage, /*init*/ nullptr, SGV->getName(), /*insertbefore*/ nullptr, SGV->getThreadLocalMode(), SGV->getType()->getAddressSpace()); } if (ForDefinition) NewGV->setLinkage(SGV->getLinkage()); else if (SGV->hasExternalWeakLinkage()) NewGV->setLinkage(GlobalValue::ExternalWeakLinkage); NewGV->copyAttributesFrom(SGV); // Remove these copied constants in case this stays a declaration, since // they point to the source module. If the def is linked the values will // be mapped in during linkFunctionBody. if (auto *NewF = dyn_cast(NewGV)) { NewF->setPersonalityFn(nullptr); NewF->setPrefixData(nullptr); NewF->setPrologueData(nullptr); } return NewGV; } /// Loop over all of the linked values to compute type mappings. For example, /// if we link "extern Foo *x" and "Foo *x = NULL", then we have two struct /// types 'Foo' but one got renamed when the module was loaded into the same /// LLVMContext. void IRLinker::computeTypeMapping() { for (GlobalValue &SGV : SrcM->globals()) { GlobalValue *DGV = getLinkedToGlobal(&SGV); if (!DGV) continue; if (!DGV->hasAppendingLinkage() || !SGV.hasAppendingLinkage()) { TypeMap.addTypeMapping(DGV->getType(), SGV.getType()); continue; } // Unify the element type of appending arrays. ArrayType *DAT = cast(DGV->getValueType()); ArrayType *SAT = cast(SGV.getValueType()); TypeMap.addTypeMapping(DAT->getElementType(), SAT->getElementType()); } for (GlobalValue &SGV : *SrcM) if (GlobalValue *DGV = getLinkedToGlobal(&SGV)) TypeMap.addTypeMapping(DGV->getType(), SGV.getType()); for (GlobalValue &SGV : SrcM->aliases()) if (GlobalValue *DGV = getLinkedToGlobal(&SGV)) TypeMap.addTypeMapping(DGV->getType(), SGV.getType()); // Incorporate types by name, scanning all the types in the source module. // At this point, the destination module may have a type "%foo = { i32 }" for // example. When the source module got loaded into the same LLVMContext, if // it had the same type, it would have been renamed to "%foo.42 = { i32 }". std::vector Types = SrcM->getIdentifiedStructTypes(); for (StructType *ST : Types) { if (!ST->hasName()) continue; // Check to see if there is a dot in the name followed by a digit. size_t DotPos = ST->getName().rfind('.'); if (DotPos == 0 || DotPos == StringRef::npos || ST->getName().back() == '.' || !isdigit(static_cast(ST->getName()[DotPos + 1]))) continue; // Check to see if the destination module has a struct with the prefix name. StructType *DST = DstM.getTypeByName(ST->getName().substr(0, DotPos)); if (!DST) continue; // Don't use it if this actually came from the source module. They're in // the same LLVMContext after all. Also don't use it unless the type is // actually used in the destination module. This can happen in situations // like this: // // Module A Module B // -------- -------- // %Z = type { %A } %B = type { %C.1 } // %A = type { %B.1, [7 x i8] } %C.1 = type { i8* } // %B.1 = type { %C } %A.2 = type { %B.3, [5 x i8] } // %C = type { i8* } %B.3 = type { %C.1 } // // When we link Module B with Module A, the '%B' in Module B is // used. However, that would then use '%C.1'. But when we process '%C.1', // we prefer to take the '%C' version. So we are then left with both // '%C.1' and '%C' being used for the same types. This leads to some // variables using one type and some using the other. if (TypeMap.DstStructTypesSet.hasType(DST)) TypeMap.addTypeMapping(DST, ST); } // Now that we have discovered all of the type equivalences, get a body for // any 'opaque' types in the dest module that are now resolved. TypeMap.linkDefinedTypeBodies(); } static void getArrayElements(const Constant *C, SmallVectorImpl &Dest) { unsigned NumElements = cast(C->getType())->getNumElements(); for (unsigned i = 0; i != NumElements; ++i) Dest.push_back(C->getAggregateElement(i)); } /// If there were any appending global variables, link them together now. /// Return true on error. Constant *IRLinker::linkAppendingVarProto(GlobalVariable *DstGV, const GlobalVariable *SrcGV) { Type *EltTy = cast(TypeMap.get(SrcGV->getValueType())) ->getElementType(); // FIXME: This upgrade is done during linking to support the C API. Once the // old form is deprecated, we should move this upgrade to // llvm::UpgradeGlobalVariable() and simplify the logic here and in // Mapper::mapAppendingVariable() in ValueMapper.cpp. StringRef Name = SrcGV->getName(); bool IsNewStructor = false; bool IsOldStructor = false; if (Name == "llvm.global_ctors" || Name == "llvm.global_dtors") { if (cast(EltTy)->getNumElements() == 3) IsNewStructor = true; else IsOldStructor = true; } PointerType *VoidPtrTy = Type::getInt8Ty(SrcGV->getContext())->getPointerTo(); if (IsOldStructor) { auto &ST = *cast(EltTy); Type *Tys[3] = {ST.getElementType(0), ST.getElementType(1), VoidPtrTy}; EltTy = StructType::get(SrcGV->getContext(), Tys, false); } uint64_t DstNumElements = 0; if (DstGV) { ArrayType *DstTy = cast(DstGV->getValueType()); DstNumElements = DstTy->getNumElements(); if (!SrcGV->hasAppendingLinkage() || !DstGV->hasAppendingLinkage()) { emitError( "Linking globals named '" + SrcGV->getName() + "': can only link appending global with another appending global!"); return nullptr; } // Check to see that they two arrays agree on type. if (EltTy != DstTy->getElementType()) { emitError("Appending variables with different element types!"); return nullptr; } if (DstGV->isConstant() != SrcGV->isConstant()) { emitError("Appending variables linked with different const'ness!"); return nullptr; } if (DstGV->getAlignment() != SrcGV->getAlignment()) { emitError( "Appending variables with different alignment need to be linked!"); return nullptr; } if (DstGV->getVisibility() != SrcGV->getVisibility()) { emitError( "Appending variables with different visibility need to be linked!"); return nullptr; } if (DstGV->hasUnnamedAddr() != SrcGV->hasUnnamedAddr()) { emitError( "Appending variables with different unnamed_addr need to be linked!"); return nullptr; } if (StringRef(DstGV->getSection()) != SrcGV->getSection()) { emitError( "Appending variables with different section name need to be linked!"); return nullptr; } } SmallVector SrcElements; getArrayElements(SrcGV->getInitializer(), SrcElements); if (IsNewStructor) SrcElements.erase( std::remove_if(SrcElements.begin(), SrcElements.end(), [this](Constant *E) { auto *Key = dyn_cast( E->getAggregateElement(2)->stripPointerCasts()); if (!Key) return false; GlobalValue *DGV = getLinkedToGlobal(Key); return !shouldLink(DGV, *Key); }), SrcElements.end()); uint64_t NewSize = DstNumElements + SrcElements.size(); ArrayType *NewType = ArrayType::get(EltTy, NewSize); // Create the new global variable. GlobalVariable *NG = new GlobalVariable( DstM, NewType, SrcGV->isConstant(), SrcGV->getLinkage(), /*init*/ nullptr, /*name*/ "", DstGV, SrcGV->getThreadLocalMode(), SrcGV->getType()->getAddressSpace()); NG->copyAttributesFrom(SrcGV); forceRenaming(NG, SrcGV->getName()); Constant *Ret = ConstantExpr::getBitCast(NG, TypeMap.get(SrcGV->getType())); Mapper.scheduleMapAppendingVariable(*NG, DstGV ? DstGV->getInitializer() : nullptr, IsOldStructor, SrcElements); // Replace any uses of the two global variables with uses of the new // global. if (DstGV) { DstGV->replaceAllUsesWith(ConstantExpr::getBitCast(NG, DstGV->getType())); DstGV->eraseFromParent(); } return Ret; } bool IRLinker::shouldLink(GlobalValue *DGV, GlobalValue &SGV) { if (ValuesToLink.count(&SGV)) return true; if (SGV.hasLocalLinkage()) return true; if (DGV && !DGV->isDeclarationForLinker()) return false; if (SGV.hasAvailableExternallyLinkage()) return true; if (SGV.isDeclaration()) return false; if (DoneLinkingBodies) return false; // Callback to the client to give a chance to lazily add the Global to the // list of value to link. bool LazilyAdded = false; AddLazyFor(SGV, [this, &LazilyAdded](GlobalValue &GV) { maybeAdd(&GV); LazilyAdded = true; }); return LazilyAdded; } Constant *IRLinker::linkGlobalValueProto(GlobalValue *SGV, bool ForAlias) { GlobalValue *DGV = getLinkedToGlobal(SGV); bool ShouldLink = shouldLink(DGV, *SGV); // just missing from map if (ShouldLink) { auto I = ValueMap.find(SGV); if (I != ValueMap.end()) return cast(I->second); I = AliasValueMap.find(SGV); if (I != AliasValueMap.end()) return cast(I->second); } if (!ShouldLink && ForAlias) DGV = nullptr; // Handle the ultra special appending linkage case first. assert(!DGV || SGV->hasAppendingLinkage() == DGV->hasAppendingLinkage()); if (SGV->hasAppendingLinkage()) return linkAppendingVarProto(cast_or_null(DGV), cast(SGV)); GlobalValue *NewGV; if (DGV && !ShouldLink) { NewGV = DGV; } else { // If we are done linking global value bodies (i.e. we are performing // metadata linking), don't link in the global value due to this // reference, simply map it to null. if (DoneLinkingBodies) return nullptr; NewGV = copyGlobalValueProto(SGV, ShouldLink); if (ShouldLink || !ForAlias) forceRenaming(NewGV, SGV->getName()); } if (ShouldLink || ForAlias) { if (const Comdat *SC = SGV->getComdat()) { if (auto *GO = dyn_cast(NewGV)) { Comdat *DC = DstM.getOrInsertComdat(SC->getName()); DC->setSelectionKind(SC->getSelectionKind()); GO->setComdat(DC); } } } if (!ShouldLink && ForAlias) NewGV->setLinkage(GlobalValue::InternalLinkage); Constant *C = NewGV; if (DGV) C = ConstantExpr::getBitCast(NewGV, TypeMap.get(SGV->getType())); if (DGV && NewGV != DGV) { DGV->replaceAllUsesWith(ConstantExpr::getBitCast(NewGV, DGV->getType())); DGV->eraseFromParent(); } return C; } /// Update the initializers in the Dest module now that all globals that may be /// referenced are in Dest. void IRLinker::linkGlobalInit(GlobalVariable &Dst, GlobalVariable &Src) { // Figure out what the initializer looks like in the dest module. Mapper.scheduleMapGlobalInitializer(Dst, *Src.getInitializer()); } /// Copy the source function over into the dest function and fix up references /// to values. At this point we know that Dest is an external function, and /// that Src is not. bool IRLinker::linkFunctionBody(Function &Dst, Function &Src) { assert(Dst.isDeclaration() && !Src.isDeclaration()); // Materialize if needed. if (std::error_code EC = Src.materialize()) return emitError(EC.message()); // Link in the operands without remapping. if (Src.hasPrefixData()) Dst.setPrefixData(Src.getPrefixData()); if (Src.hasPrologueData()) Dst.setPrologueData(Src.getPrologueData()); if (Src.hasPersonalityFn()) Dst.setPersonalityFn(Src.getPersonalityFn()); // Copy over the metadata attachments without remapping. SmallVector, 8> MDs; Src.getAllMetadata(MDs); for (const auto &I : MDs) Dst.setMetadata(I.first, I.second); // Steal arguments and splice the body of Src into Dst. Dst.stealArgumentListFrom(Src); Dst.getBasicBlockList().splice(Dst.end(), Src.getBasicBlockList()); // Everything has been moved over. Remap it. Mapper.scheduleRemapFunction(Dst); return false; } void IRLinker::linkAliasBody(GlobalAlias &Dst, GlobalAlias &Src) { Mapper.scheduleMapGlobalAliasee(Dst, *Src.getAliasee(), AliasMCID); } bool IRLinker::linkGlobalValueBody(GlobalValue &Dst, GlobalValue &Src) { if (auto *F = dyn_cast(&Src)) return linkFunctionBody(cast(Dst), *F); if (auto *GVar = dyn_cast(&Src)) { linkGlobalInit(cast(Dst), *GVar); return false; } linkAliasBody(cast(Dst), cast(Src)); return false; } /// Insert all of the named MDNodes in Src into the Dest module. void IRLinker::linkNamedMDNodes() { const NamedMDNode *SrcModFlags = SrcM->getModuleFlagsMetadata(); for (const NamedMDNode &NMD : SrcM->named_metadata()) { // Don't link module flags here. Do them separately. if (&NMD == SrcModFlags) continue; NamedMDNode *DestNMD = DstM.getOrInsertNamedMetadata(NMD.getName()); // Add Src elements into Dest node. for (const MDNode *Op : NMD.operands()) DestNMD->addOperand(Mapper.mapMDNode(*Op)); } } /// Merge the linker flags in Src into the Dest module. bool IRLinker::linkModuleFlagsMetadata() { // If the source module has no module flags, we are done. const NamedMDNode *SrcModFlags = SrcM->getModuleFlagsMetadata(); if (!SrcModFlags) return false; // If the destination module doesn't have module flags yet, then just copy // over the source module's flags. NamedMDNode *DstModFlags = DstM.getOrInsertModuleFlagsMetadata(); if (DstModFlags->getNumOperands() == 0) { for (unsigned I = 0, E = SrcModFlags->getNumOperands(); I != E; ++I) DstModFlags->addOperand(SrcModFlags->getOperand(I)); return false; } // First build a map of the existing module flags and requirements. DenseMap> Flags; SmallSetVector Requirements; for (unsigned I = 0, E = DstModFlags->getNumOperands(); I != E; ++I) { MDNode *Op = DstModFlags->getOperand(I); ConstantInt *Behavior = mdconst::extract(Op->getOperand(0)); MDString *ID = cast(Op->getOperand(1)); if (Behavior->getZExtValue() == Module::Require) { Requirements.insert(cast(Op->getOperand(2))); } else { Flags[ID] = std::make_pair(Op, I); } } // Merge in the flags from the source module, and also collect its set of // requirements. for (unsigned I = 0, E = SrcModFlags->getNumOperands(); I != E; ++I) { MDNode *SrcOp = SrcModFlags->getOperand(I); ConstantInt *SrcBehavior = mdconst::extract(SrcOp->getOperand(0)); MDString *ID = cast(SrcOp->getOperand(1)); MDNode *DstOp; unsigned DstIndex; std::tie(DstOp, DstIndex) = Flags.lookup(ID); unsigned SrcBehaviorValue = SrcBehavior->getZExtValue(); // If this is a requirement, add it and continue. if (SrcBehaviorValue == Module::Require) { // If the destination module does not already have this requirement, add // it. if (Requirements.insert(cast(SrcOp->getOperand(2)))) { DstModFlags->addOperand(SrcOp); } continue; } // If there is no existing flag with this ID, just add it. if (!DstOp) { Flags[ID] = std::make_pair(SrcOp, DstModFlags->getNumOperands()); DstModFlags->addOperand(SrcOp); continue; } // Otherwise, perform a merge. ConstantInt *DstBehavior = mdconst::extract(DstOp->getOperand(0)); unsigned DstBehaviorValue = DstBehavior->getZExtValue(); // If either flag has override behavior, handle it first. if (DstBehaviorValue == Module::Override) { // Diagnose inconsistent flags which both have override behavior. if (SrcBehaviorValue == Module::Override && SrcOp->getOperand(2) != DstOp->getOperand(2)) { emitError("linking module flags '" + ID->getString() + "': IDs have conflicting override values"); } continue; } else if (SrcBehaviorValue == Module::Override) { // Update the destination flag to that of the source. DstModFlags->setOperand(DstIndex, SrcOp); Flags[ID].first = SrcOp; continue; } // Diagnose inconsistent merge behavior types. if (SrcBehaviorValue != DstBehaviorValue) { emitError("linking module flags '" + ID->getString() + "': IDs have conflicting behaviors"); continue; } auto replaceDstValue = [&](MDNode *New) { Metadata *FlagOps[] = {DstOp->getOperand(0), ID, New}; MDNode *Flag = MDNode::get(DstM.getContext(), FlagOps); DstModFlags->setOperand(DstIndex, Flag); Flags[ID].first = Flag; }; // Perform the merge for standard behavior types. switch (SrcBehaviorValue) { case Module::Require: case Module::Override: llvm_unreachable("not possible"); case Module::Error: { // Emit an error if the values differ. if (SrcOp->getOperand(2) != DstOp->getOperand(2)) { emitError("linking module flags '" + ID->getString() + "': IDs have conflicting values"); } continue; } case Module::Warning: { // Emit a warning if the values differ. if (SrcOp->getOperand(2) != DstOp->getOperand(2)) { emitWarning("linking module flags '" + ID->getString() + "': IDs have conflicting values"); } continue; } case Module::Append: { MDNode *DstValue = cast(DstOp->getOperand(2)); MDNode *SrcValue = cast(SrcOp->getOperand(2)); SmallVector MDs; MDs.reserve(DstValue->getNumOperands() + SrcValue->getNumOperands()); MDs.append(DstValue->op_begin(), DstValue->op_end()); MDs.append(SrcValue->op_begin(), SrcValue->op_end()); replaceDstValue(MDNode::get(DstM.getContext(), MDs)); break; } case Module::AppendUnique: { SmallSetVector Elts; MDNode *DstValue = cast(DstOp->getOperand(2)); MDNode *SrcValue = cast(SrcOp->getOperand(2)); Elts.insert(DstValue->op_begin(), DstValue->op_end()); Elts.insert(SrcValue->op_begin(), SrcValue->op_end()); replaceDstValue(MDNode::get(DstM.getContext(), makeArrayRef(Elts.begin(), Elts.end()))); break; } } } // Check all of the requirements. for (unsigned I = 0, E = Requirements.size(); I != E; ++I) { MDNode *Requirement = Requirements[I]; MDString *Flag = cast(Requirement->getOperand(0)); Metadata *ReqValue = Requirement->getOperand(1); MDNode *Op = Flags[Flag].first; if (!Op || Op->getOperand(2) != ReqValue) { emitError("linking module flags '" + Flag->getString() + "': does not have the required value"); continue; } } return HasError; } // This function returns true if the triples match. static bool triplesMatch(const Triple &T0, const Triple &T1) { // If vendor is apple, ignore the version number. if (T0.getVendor() == Triple::Apple) return T0.getArch() == T1.getArch() && T0.getSubArch() == T1.getSubArch() && T0.getVendor() == T1.getVendor() && T0.getOS() == T1.getOS(); return T0 == T1; } // This function returns the merged triple. static std::string mergeTriples(const Triple &SrcTriple, const Triple &DstTriple) { // If vendor is apple, pick the triple with the larger version number. if (SrcTriple.getVendor() == Triple::Apple) if (DstTriple.isOSVersionLT(SrcTriple)) return SrcTriple.str(); return DstTriple.str(); } bool IRLinker::run() { // Ensure metadata materialized before value mapping. if (SrcM->getMaterializer() && SrcM->getMaterializer()->materializeMetadata()) return true; // Inherit the target data from the source module if the destination module // doesn't have one already. if (DstM.getDataLayout().isDefault()) DstM.setDataLayout(SrcM->getDataLayout()); if (SrcM->getDataLayout() != DstM.getDataLayout()) { emitWarning("Linking two modules of different data layouts: '" + SrcM->getModuleIdentifier() + "' is '" + SrcM->getDataLayoutStr() + "' whereas '" + DstM.getModuleIdentifier() + "' is '" + DstM.getDataLayoutStr() + "'\n"); } // Copy the target triple from the source to dest if the dest's is empty. if (DstM.getTargetTriple().empty() && !SrcM->getTargetTriple().empty()) DstM.setTargetTriple(SrcM->getTargetTriple()); Triple SrcTriple(SrcM->getTargetTriple()), DstTriple(DstM.getTargetTriple()); if (!SrcM->getTargetTriple().empty() && !triplesMatch(SrcTriple, DstTriple)) emitWarning("Linking two modules of different target triples: " + SrcM->getModuleIdentifier() + "' is '" + SrcM->getTargetTriple() + "' whereas '" + DstM.getModuleIdentifier() + "' is '" + DstM.getTargetTriple() + "'\n"); DstM.setTargetTriple(mergeTriples(SrcTriple, DstTriple)); // Append the module inline asm string. if (!SrcM->getModuleInlineAsm().empty()) { if (DstM.getModuleInlineAsm().empty()) DstM.setModuleInlineAsm(SrcM->getModuleInlineAsm()); else DstM.setModuleInlineAsm(DstM.getModuleInlineAsm() + "\n" + SrcM->getModuleInlineAsm()); } // Loop over all of the linked values to compute type mappings. computeTypeMapping(); std::reverse(Worklist.begin(), Worklist.end()); while (!Worklist.empty()) { GlobalValue *GV = Worklist.back(); Worklist.pop_back(); // Already mapped. if (ValueMap.find(GV) != ValueMap.end() || AliasValueMap.find(GV) != AliasValueMap.end()) continue; assert(!GV->isDeclaration()); Mapper.mapValue(*GV); if (HasError) return true; } // Note that we are done linking global value bodies. This prevents // metadata linking from creating new references. DoneLinkingBodies = true; Mapper.addFlags(RF_NullMapMissingGlobalValues); // Remap all of the named MDNodes in Src into the DstM module. We do this // after linking GlobalValues so that MDNodes that reference GlobalValues // are properly remapped. linkNamedMDNodes(); // Merge the module flags into the DstM module. if (linkModuleFlagsMetadata()) return true; return false; } IRMover::StructTypeKeyInfo::KeyTy::KeyTy(ArrayRef E, bool P) : ETypes(E), IsPacked(P) {} IRMover::StructTypeKeyInfo::KeyTy::KeyTy(const StructType *ST) : ETypes(ST->elements()), IsPacked(ST->isPacked()) {} bool IRMover::StructTypeKeyInfo::KeyTy::operator==(const KeyTy &That) const { if (IsPacked != That.IsPacked) return false; if (ETypes != That.ETypes) return false; return true; } bool IRMover::StructTypeKeyInfo::KeyTy::operator!=(const KeyTy &That) const { return !this->operator==(That); } StructType *IRMover::StructTypeKeyInfo::getEmptyKey() { return DenseMapInfo::getEmptyKey(); } StructType *IRMover::StructTypeKeyInfo::getTombstoneKey() { return DenseMapInfo::getTombstoneKey(); } unsigned IRMover::StructTypeKeyInfo::getHashValue(const KeyTy &Key) { return hash_combine(hash_combine_range(Key.ETypes.begin(), Key.ETypes.end()), Key.IsPacked); } unsigned IRMover::StructTypeKeyInfo::getHashValue(const StructType *ST) { return getHashValue(KeyTy(ST)); } bool IRMover::StructTypeKeyInfo::isEqual(const KeyTy &LHS, const StructType *RHS) { if (RHS == getEmptyKey() || RHS == getTombstoneKey()) return false; return LHS == KeyTy(RHS); } bool IRMover::StructTypeKeyInfo::isEqual(const StructType *LHS, const StructType *RHS) { if (RHS == getEmptyKey()) return LHS == getEmptyKey(); if (RHS == getTombstoneKey()) return LHS == getTombstoneKey(); return KeyTy(LHS) == KeyTy(RHS); } void IRMover::IdentifiedStructTypeSet::addNonOpaque(StructType *Ty) { assert(!Ty->isOpaque()); NonOpaqueStructTypes.insert(Ty); } void IRMover::IdentifiedStructTypeSet::switchToNonOpaque(StructType *Ty) { assert(!Ty->isOpaque()); NonOpaqueStructTypes.insert(Ty); bool Removed = OpaqueStructTypes.erase(Ty); (void)Removed; assert(Removed); } void IRMover::IdentifiedStructTypeSet::addOpaque(StructType *Ty) { assert(Ty->isOpaque()); OpaqueStructTypes.insert(Ty); } StructType * IRMover::IdentifiedStructTypeSet::findNonOpaque(ArrayRef ETypes, bool IsPacked) { IRMover::StructTypeKeyInfo::KeyTy Key(ETypes, IsPacked); auto I = NonOpaqueStructTypes.find_as(Key); if (I == NonOpaqueStructTypes.end()) return nullptr; return *I; } bool IRMover::IdentifiedStructTypeSet::hasType(StructType *Ty) { if (Ty->isOpaque()) return OpaqueStructTypes.count(Ty); auto I = NonOpaqueStructTypes.find(Ty); if (I == NonOpaqueStructTypes.end()) return false; return *I == Ty; } IRMover::IRMover(Module &M) : Composite(M) { TypeFinder StructTypes; StructTypes.run(M, true); for (StructType *Ty : StructTypes) { if (Ty->isOpaque()) IdentifiedStructTypes.addOpaque(Ty); else IdentifiedStructTypes.addNonOpaque(Ty); } } bool IRMover::move( std::unique_ptr Src, ArrayRef ValuesToLink, std::function AddLazyFor) { IRLinker TheIRLinker(Composite, SharedMDs, IdentifiedStructTypes, std::move(Src), ValuesToLink, AddLazyFor); bool RetCode = TheIRLinker.run(); Composite.dropTriviallyDeadConstantArrays(); return RetCode; }