X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FLinker%2FLinkModules.cpp;h=7293f3d0e8a07828cc521124d90236c039386ee0;hb=8c3f33d337acf0cf2015fa04e0c7aafe3931fbd3;hp=f2187c0988234b07c8c372338823be1469e03deb;hpb=c47a474802172608f0ddb68948e766155b5e40d4;p=oota-llvm.git diff --git a/lib/Linker/LinkModules.cpp b/lib/Linker/LinkModules.cpp index f2187c09882..7293f3d0e8a 100644 --- a/lib/Linker/LinkModules.cpp +++ b/lib/Linker/LinkModules.cpp @@ -9,472 +9,506 @@ // // This file implements the LLVM module linker. // -// Specifically, this: -// * Merges global variables between the two modules -// * Uninit + Uninit = Init, Init + Uninit = Init, Init + Init = Error if != -// * Merges functions between two modules -// //===----------------------------------------------------------------------===// #include "llvm/Linker.h" #include "llvm/Constants.h" #include "llvm/DerivedTypes.h" -#include "llvm/LLVMContext.h" -#include "llvm/Module.h" -#include "llvm/TypeSymbolTable.h" -#include "llvm/ValueSymbolTable.h" #include "llvm/Instructions.h" -#include "llvm/Assembly/Writer.h" +#include "llvm/Module.h" +#include "llvm/ADT/DenseSet.h" +#include "llvm/ADT/Optional.h" +#include "llvm/ADT/SetVector.h" +#include "llvm/ADT/SmallPtrSet.h" #include "llvm/Support/Debug.h" -#include "llvm/Support/ErrorHandling.h" +#include "llvm/Support/Path.h" #include "llvm/Support/raw_ostream.h" -#include "llvm/System/Path.h" -#include "llvm/ADT/DenseMap.h" +#include "llvm/Transforms/Utils/Cloning.h" +#include "llvm/Transforms/Utils/ValueMapper.h" +#include "llvm-c/Linker.h" +#include using namespace llvm; -// Error - Simple wrapper function to conditionally assign to E and return true. -// This just makes error return conditions a little bit simpler... -static inline bool Error(std::string *E, const Twine &Message) { - if (E) *E = Message.str(); - return true; -} - -// Function: ResolveTypes() -// -// Description: -// Attempt to link the two specified types together. -// -// Inputs: -// DestTy - The type to which we wish to resolve. -// SrcTy - The original type which we want to resolve. -// -// Outputs: -// DestST - The symbol table in which the new type should be placed. -// -// Return value: -// true - There is an error and the types cannot yet be linked. -// false - No errors. -// -static bool ResolveTypes(const Type *DestTy, const Type *SrcTy) { - if (DestTy == SrcTy) return false; // If already equal, noop - assert(DestTy && SrcTy && "Can't handle null types"); - - if (const OpaqueType *OT = dyn_cast(DestTy)) { - // Type _is_ in module, just opaque... - const_cast(OT)->refineAbstractTypeTo(SrcTy); - } else if (const OpaqueType *OT = dyn_cast(SrcTy)) { - const_cast(OT)->refineAbstractTypeTo(DestTy); - } else { - return true; // Cannot link types... not-equal and neither is opaque. - } - return false; -} +//===----------------------------------------------------------------------===// +// TypeMap implementation. +//===----------------------------------------------------------------------===// -/// LinkerTypeMap - This implements a map of types that is stable -/// even if types are resolved/refined to other types. This is not a general -/// purpose map, it is specific to the linker's use. namespace { -class LinkerTypeMap : public AbstractTypeUser { - typedef DenseMap TheMapTy; - TheMapTy TheMap; +class TypeMapTy : public ValueMapTypeRemapper { + /// MappedTypes - This is a mapping from a source type to a destination type + /// to use. + DenseMap MappedTypes; + + /// SpeculativeTypes - 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; + + /// SrcDefinitionsToResolve - 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; + + /// DstResolvedOpaqueTypes - This is the set of opaque types in the + /// destination modules who are getting a body from the source module. + SmallPtrSet DstResolvedOpaqueTypes; - LinkerTypeMap(const LinkerTypeMap&); // DO NOT IMPLEMENT - void operator=(const LinkerTypeMap&); // DO NOT IMPLEMENT public: - LinkerTypeMap() {} - ~LinkerTypeMap() { - for (DenseMap::iterator I = TheMap.begin(), - E = TheMap.end(); I != E; ++I) - I->first->removeAbstractTypeUser(this); - } - - /// lookup - Return the value for the specified type or null if it doesn't - /// exist. - const Type *lookup(const Type *Ty) const { - TheMapTy::const_iterator I = TheMap.find(Ty); - if (I != TheMap.end()) return I->second; - return 0; - } - - /// erase - Remove the specified type, returning true if it was in the set. - bool erase(const Type *Ty) { - if (!TheMap.erase(Ty)) - return false; - if (Ty->isAbstract()) - Ty->removeAbstractTypeUser(this); - return true; - } - - /// insert - This returns true if the pointer was new to the set, false if it - /// was already in the set. - bool insert(const Type *Src, const Type *Dst) { - if (!TheMap.insert(std::make_pair(Src, PATypeHolder(Dst))).second) - return false; // Already in map. - if (Src->isAbstract()) - Src->addAbstractTypeUser(this); - return true; + /// addTypeMapping - 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); + + /// linkDefinedTypeBodies - Produce a body for an opaque type in the dest + /// module from a type definition in the source module. + void linkDefinedTypeBodies(); + + /// get - Return the mapped type to use for the specified input type from the + /// source module. + Type *get(Type *SrcTy); + + FunctionType *get(FunctionType *T) {return cast(get((Type*)T));} + + /// dump - Dump out the type map for debugging purposes. + void dump() const { + for (DenseMap::const_iterator + I = MappedTypes.begin(), E = MappedTypes.end(); I != E; ++I) { + dbgs() << "TypeMap: "; + I->first->dump(); + dbgs() << " => "; + I->second->dump(); + dbgs() << '\n'; + } } -protected: - /// refineAbstractType - The callback method invoked when an abstract type is - /// resolved to another type. An object must override this method to update - /// its internal state to reference NewType instead of OldType. - /// - virtual void refineAbstractType(const DerivedType *OldTy, - const Type *NewTy) { - TheMapTy::iterator I = TheMap.find(OldTy); - const Type *DstTy = I->second; - - TheMap.erase(I); - if (OldTy->isAbstract()) - OldTy->removeAbstractTypeUser(this); - - // Don't reinsert into the map if the key is concrete now. - if (NewTy->isAbstract()) - insert(NewTy, DstTy); +private: + Type *getImpl(Type *T); + /// remapType - Implement the ValueMapTypeRemapper interface. + Type *remapType(Type *SrcTy) { + return get(SrcTy); } + + bool areTypesIsomorphic(Type *DstTy, Type *SrcTy); +}; +} - /// The other case which AbstractTypeUsers must be aware of is when a type - /// makes the transition from being abstract (where it has clients on it's - /// AbstractTypeUsers list) to concrete (where it does not). This method - /// notifies ATU's when this occurs for a type. - virtual void typeBecameConcrete(const DerivedType *AbsTy) { - TheMap.erase(AbsTy); - AbsTy->removeAbstractTypeUser(this); +void TypeMapTy::addTypeMapping(Type *DstTy, Type *SrcTy) { + Type *&Entry = MappedTypes[SrcTy]; + if (Entry) return; + + if (DstTy == SrcTy) { + Entry = DstTy; + return; } - - // for debugging... - virtual void dump() const { - dbgs() << "AbstractTypeSet!\n"; + + // 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 (unsigned i = 0, e = SpeculativeTypes.size(); i != e; ++i) + MappedTypes.erase(SpeculativeTypes[i]); } -}; + SpeculativeTypes.clear(); } - -// RecursiveResolveTypes - This is just like ResolveTypes, except that it -// recurses down into derived types, merging the used types if the parent types -// are compatible. -static bool RecursiveResolveTypesI(const Type *DstTy, const Type *SrcTy, - LinkerTypeMap &Pointers) { - if (DstTy == SrcTy) return false; // If already equal, noop - - // If we found our opaque type, resolve it now! - if (DstTy->isOpaqueTy() || SrcTy->isOpaqueTy()) - return ResolveTypes(DstTy, SrcTy); - - // Two types cannot be resolved together if they are of different primitive - // type. For example, we cannot resolve an int to a float. - if (DstTy->getTypeID() != SrcTy->getTypeID()) return true; - - // If neither type is abstract, then they really are just different types. - if (!DstTy->isAbstract() && !SrcTy->isAbstract()) - return true; - - // Otherwise, resolve the used type used by this derived type... - switch (DstTy->getTypeID()) { - default: +/// areTypesIsomorphic - 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; - case Type::FunctionTyID: { - const FunctionType *DstFT = cast(DstTy); - const FunctionType *SrcFT = cast(SrcTy); - if (DstFT->isVarArg() != SrcFT->isVarArg() || - DstFT->getNumContainedTypes() != SrcFT->getNumContainedTypes()) - return true; - - // Use TypeHolder's so recursive resolution won't break us. - PATypeHolder ST(SrcFT), DT(DstFT); - for (unsigned i = 0, e = DstFT->getNumContainedTypes(); i != e; ++i) { - const Type *SE = ST->getContainedType(i), *DE = DT->getContainedType(i); - if (SE != DE && RecursiveResolveTypesI(DE, SE, Pointers)) - return true; - } - return false; } - case Type::StructTyID: { - const StructType *DstST = cast(DstTy); - const StructType *SrcST = cast(SrcTy); - if (DstST->getNumContainedTypes() != SrcST->getNumContainedTypes()) + + // 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; - - PATypeHolder ST(SrcST), DT(DstST); - for (unsigned i = 0, e = DstST->getNumContainedTypes(); i != e; ++i) { - const Type *SE = ST->getContainedType(i), *DE = DT->getContainedType(i); - if (SE != DE && RecursiveResolveTypesI(DE, SE, Pointers)) - return true; } - return false; - } - case Type::ArrayTyID: { - const ArrayType *DAT = cast(DstTy); - const ArrayType *SAT = cast(SrcTy); - if (DAT->getNumElements() != SAT->getNumElements()) return true; - return RecursiveResolveTypesI(DAT->getElementType(), SAT->getElementType(), - Pointers); - } - case Type::VectorTyID: { - const VectorType *DVT = cast(DstTy); - const VectorType *SVT = cast(SrcTy); - if (DVT->getNumElements() != SVT->getNumElements()) return true; - return RecursiveResolveTypesI(DVT->getElementType(), SVT->getElementType(), - Pointers); - } - case Type::PointerTyID: { - const PointerType *DstPT = cast(DstTy); - const PointerType *SrcPT = cast(SrcTy); - if (DstPT->getAddressSpace() != SrcPT->getAddressSpace()) + // 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. This doesn't need to be speculative. 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))) + return false; + SrcDefinitionsToResolve.push_back(SSTy); + Entry = DstTy; return true; - - // If this is a pointer type, check to see if we have already seen it. If - // so, we are in a recursive branch. Cut off the search now. We cannot use - // an associative container for this search, because the type pointers (keys - // in the container) change whenever types get resolved. - if (SrcPT->isAbstract()) - if (const Type *ExistingDestTy = Pointers.lookup(SrcPT)) - return ExistingDestTy != DstPT; - - if (DstPT->isAbstract()) - if (const Type *ExistingSrcTy = Pointers.lookup(DstPT)) - return ExistingSrcTy != SrcPT; - // Otherwise, add the current pointers to the vector to stop recursion on - // this pair. - if (DstPT->isAbstract()) - Pointers.insert(DstPT, SrcPT); - if (SrcPT->isAbstract()) - Pointers.insert(SrcPT, DstPT); - - return RecursiveResolveTypesI(DstPT->getElementType(), - SrcPT->getElementType(), Pointers); - } - } -} - -static bool RecursiveResolveTypes(const Type *DestTy, const Type *SrcTy) { - LinkerTypeMap PointerTypes; - return RecursiveResolveTypesI(DestTy, SrcTy, PointerTypes); -} - - -// LinkTypes - Go through the symbol table of the Src module and see if any -// types are named in the src module that are not named in the Dst module. -// Make sure there are no type name conflicts. -static bool LinkTypes(Module *Dest, const Module *Src, std::string *Err) { - TypeSymbolTable *DestST = &Dest->getTypeSymbolTable(); - const TypeSymbolTable *SrcST = &Src->getTypeSymbolTable(); - - // Look for a type plane for Type's... - TypeSymbolTable::const_iterator TI = SrcST->begin(); - TypeSymbolTable::const_iterator TE = SrcST->end(); - if (TI == TE) return false; // No named types, do nothing. - - // Some types cannot be resolved immediately because they depend on other - // types being resolved to each other first. This contains a list of types we - // are waiting to recheck. - std::vector DelayedTypesToResolve; - - for ( ; TI != TE; ++TI ) { - const std::string &Name = TI->first; - const Type *RHS = TI->second; - - // Check to see if this type name is already in the dest module. - Type *Entry = DestST->lookup(Name); - - // If the name is just in the source module, bring it over to the dest. - if (Entry == 0) { - if (!Name.empty()) - DestST->insert(Name, const_cast(RHS)); - } else if (ResolveTypes(Entry, RHS)) { - // They look different, save the types 'till later to resolve. - DelayedTypesToResolve.push_back(Name); } } + + // 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; + } - // Iteratively resolve types while we can... - while (!DelayedTypesToResolve.empty()) { - // Loop over all of the types, attempting to resolve them if possible... - unsigned OldSize = DelayedTypesToResolve.size(); - - // Try direct resolution by name... - for (unsigned i = 0; i != DelayedTypesToResolve.size(); ++i) { - const std::string &Name = DelayedTypesToResolve[i]; - Type *T1 = SrcST->lookup(Name); - Type *T2 = DestST->lookup(Name); - if (!ResolveTypes(T2, T1)) { - // We are making progress! - DelayedTypesToResolve.erase(DelayedTypesToResolve.begin()+i); - --i; - } - } + // Otherwise, we speculate that these two types will line up and recursively + // check the subelements. + Entry = DstTy; + SpeculativeTypes.push_back(SrcTy); - // Did we not eliminate any types? - if (DelayedTypesToResolve.size() == OldSize) { - // Attempt to resolve subelements of types. This allows us to merge these - // two types: { int* } and { opaque* } - for (unsigned i = 0, e = DelayedTypesToResolve.size(); i != e; ++i) { - const std::string &Name = DelayedTypesToResolve[i]; - if (!RecursiveResolveTypes(SrcST->lookup(Name), DestST->lookup(Name))) { - // We are making progress! - DelayedTypesToResolve.erase(DelayedTypesToResolve.begin()+i); - - // Go back to the main loop, perhaps we can resolve directly by name - // now... - break; - } - } + 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; +} - // If we STILL cannot resolve the types, then there is something wrong. - if (DelayedTypesToResolve.size() == OldSize) { - // Remove the symbol name from the destination. - DelayedTypesToResolve.pop_back(); - } +/// linkDefinedTypeBodies - Produce a body for an opaque type in the dest +/// module from a type definition in the source module. +void TypeMapTy::linkDefinedTypeBodies() { + SmallVector Elements; + SmallString<16> TmpName; + + // Note that processing entries in this loop (calling 'get') can add new + // entries to the SrcDefinitionsToResolve vector. + while (!SrcDefinitionsToResolve.empty()) { + StructType *SrcSTy = SrcDefinitionsToResolve.pop_back_val(); + StructType *DstSTy = cast(MappedTypes[SrcSTy]); + + // TypeMap is a many-to-one mapping, if there were multiple types that + // provide a body for DstSTy then previous iterations of this loop may have + // already handled it. Just ignore this case. + if (!DstSTy->isOpaque()) continue; + assert(!SrcSTy->isOpaque() && "Not resolving a definition?"); + + // 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] = getImpl(SrcSTy->getElementType(i)); + + DstSTy->setBody(Elements, SrcSTy->isPacked()); + + // If DstSTy has no name or has a longer name than STy, then viciously steal + // STy's name. + if (!SrcSTy->hasName()) continue; + StringRef SrcName = SrcSTy->getName(); + + if (!DstSTy->hasName() || DstSTy->getName().size() > SrcName.size()) { + TmpName.insert(TmpName.end(), SrcName.begin(), SrcName.end()); + SrcSTy->setName(""); + DstSTy->setName(TmpName.str()); + TmpName.clear(); } } - - - return false; + + DstResolvedOpaqueTypes.clear(); } -#ifndef NDEBUG -static void PrintMap(const std::map &M) { - for (std::map::const_iterator I = M.begin(), E =M.end(); - I != E; ++I) { - dbgs() << " Fr: " << (const void*)I->first << " "; - I->first->dump(); - dbgs() << " To: " << (const void*)I->second << " "; - I->second->dump(); - dbgs() << "\n"; - } +/// get - Return the mapped type to use for the specified input type from the +/// source module. +Type *TypeMapTy::get(Type *Ty) { + Type *Result = getImpl(Ty); + + // If this caused a reference to any struct type, resolve it before returning. + if (!SrcDefinitionsToResolve.empty()) + linkDefinedTypeBodies(); + return Result; } -#endif - - -// RemapOperand - Use ValueMap to convert constants from one module to another. -static Value *RemapOperand(const Value *In, - std::map &ValueMap) { - std::map::const_iterator I = ValueMap.find(In); - if (I != ValueMap.end()) - return I->second; - - // Check to see if it's a constant that we are interested in transforming. - Value *Result = 0; - if (const Constant *CPV = dyn_cast(In)) { - if ((!isa(CPV->getType()) && !isa(CPV)) || - isa(CPV) || isa(CPV)) - return const_cast(CPV); // Simple constants stay identical. - - if (const ConstantArray *CPA = dyn_cast(CPV)) { - std::vector Operands(CPA->getNumOperands()); - for (unsigned i = 0, e = CPA->getNumOperands(); i != e; ++i) - Operands[i] =cast(RemapOperand(CPA->getOperand(i), ValueMap)); - Result = ConstantArray::get(cast(CPA->getType()), Operands); - } else if (const ConstantStruct *CPS = dyn_cast(CPV)) { - std::vector Operands(CPS->getNumOperands()); - for (unsigned i = 0, e = CPS->getNumOperands(); i != e; ++i) - Operands[i] =cast(RemapOperand(CPS->getOperand(i), ValueMap)); - Result = ConstantStruct::get(cast(CPS->getType()), Operands); - } else if (isa(CPV) || isa(CPV)) { - Result = const_cast(CPV); - } else if (const ConstantVector *CP = dyn_cast(CPV)) { - std::vector Operands(CP->getNumOperands()); - for (unsigned i = 0, e = CP->getNumOperands(); i != e; ++i) - Operands[i] = cast(RemapOperand(CP->getOperand(i), ValueMap)); - Result = ConstantVector::get(Operands); - } else if (const ConstantExpr *CE = dyn_cast(CPV)) { - std::vector Ops; - for (unsigned i = 0, e = CE->getNumOperands(); i != e; ++i) - Ops.push_back(cast(RemapOperand(CE->getOperand(i),ValueMap))); - Result = CE->getWithOperands(Ops); - } else if (const BlockAddress *CE = dyn_cast(CPV)) { - Result = BlockAddress::get( - cast(RemapOperand(CE->getFunction(), ValueMap)), - CE->getBasicBlock()); - } else { - assert(!isa(CPV) && "Unmapped global?"); - llvm_unreachable("Unknown type of derived type constant value!"); + +/// getImpl - This is the recursive version of get(). +Type *TypeMapTy::getImpl(Type *Ty) { + // If we already have an entry for this type, return it. + Type **Entry = &MappedTypes[Ty]; + if (*Entry) return *Entry; + + // If this is not a named struct type, then just map all of the elements and + // then rebuild the type from inside out. + if (!isa(Ty) || cast(Ty)->isLiteral()) { + // 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) + return *Entry = Ty; + + // Remap all of the elements, keeping track of whether any of them change. + bool AnyChange = false; + SmallVector ElementTypes; + ElementTypes.resize(Ty->getNumContainedTypes()); + for (unsigned i = 0, e = Ty->getNumContainedTypes(); i != e; ++i) { + ElementTypes[i] = getImpl(Ty->getContainedType(i)); + AnyChange |= ElementTypes[i] != Ty->getContainedType(i); } - } else if (const MDNode *MD = dyn_cast(In)) { - if (MD->isFunctionLocal()) { - SmallVector Elts; - for (unsigned i = 0, e = MD->getNumOperands(); i != e; ++i) { - if (MD->getOperand(i)) - Elts.push_back(RemapOperand(MD->getOperand(i), ValueMap)); - else - Elts.push_back(NULL); - } - Result = MDNode::get(In->getContext(), Elts.data(), MD->getNumOperands()); - } else { - Result = const_cast(In); + + // If we found our type while recursively processing stuff, just use it. + Entry = &MappedTypes[Ty]; + if (*Entry) return *Entry; + + // If all of the element types mapped directly over, then the type is usable + // as-is. + if (!AnyChange) + 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: + // Note that this is only reached for anonymous structs. + return *Entry = StructType::get(Ty->getContext(), ElementTypes, + cast(Ty)->isPacked()); } - } else if (isa(In) || isa(In) || isa(In)) { - Result = const_cast(In); } - // Cache the mapping in our local map structure - if (Result) { - ValueMap[In] = Result; - return Result; - } + // Otherwise, this is an unmapped named struct. If the struct can be directly + // mapped over, just use it as-is. This happens in a case when the linked-in + // module has something like: + // %T = type {%T*, i32} + // @GV = global %T* null + // where T does not exist at all in the destination module. + // + // The other case we watch for is when the type is not in the destination + // module, but that it has to be rebuilt because it refers to something that + // is already mapped. For example, if the destination module has: + // %A = type { i32 } + // and the source module has something like + // %A' = type { i32 } + // %B = type { %A'* } + // @GV = global %B* null + // then we want to create a new type: "%B = type { %A*}" and have it take the + // pristine "%B" name from the source module. + // + // To determine which case this is, we have to recursively walk the type graph + // speculating that we'll be able to reuse it unmodified. Only if this is + // safe would we map the entire thing over. Because this is an optimization, + // and is not required for the prettiness of the linked module, we just skip + // it and always rebuild a type here. + StructType *STy = cast(Ty); + + // If the type is opaque, we can just use it directly. + if (STy->isOpaque()) + return *Entry = STy; + + // Otherwise we create a new type and resolve its body later. This will be + // resolved by the top level of get(). + SrcDefinitionsToResolve.push_back(STy); + StructType *DTy = StructType::create(STy->getContext()); + DstResolvedOpaqueTypes.insert(DTy); + return *Entry = DTy; +} -#ifndef NDEBUG - dbgs() << "LinkModules ValueMap: \n"; - PrintMap(ValueMap); +//===----------------------------------------------------------------------===// +// ModuleLinker implementation. +//===----------------------------------------------------------------------===// - dbgs() << "Couldn't remap value: " << (const void*)In << " " << *In << "\n"; - llvm_unreachable("Couldn't remap value!"); -#endif - return 0; +namespace { + /// ModuleLinker - This is an implementation class for the LinkModules + /// function, which is the entrypoint for this file. + class ModuleLinker { + Module *DstM, *SrcM; + + TypeMapTy TypeMap; + + /// ValueMap - 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; + + struct AppendingVarInfo { + GlobalVariable *NewGV; // New aggregate global in dest module. + Constant *DstInit; // Old initializer from dest module. + Constant *SrcInit; // Old initializer from src module. + }; + + std::vector AppendingVars; + + unsigned Mode; // Mode to treat source module. + + // Set of items not to link in from source. + SmallPtrSet DoNotLinkFromSource; + + // Vector of functions to lazily link in. + std::vector LazilyLinkFunctions; + + public: + std::string ErrorMsg; + + ModuleLinker(Module *dstM, Module *srcM, unsigned mode) + : DstM(dstM), SrcM(srcM), Mode(mode) { } + + bool run(); + + private: + /// emitError - Helper method for setting a message and returning an error + /// code. + bool emitError(const Twine &Message) { + ErrorMsg = Message.str(); + return true; + } + + /// getLinkageResult - This analyzes the two global values and determines + /// what the result will look like in the destination module. + bool getLinkageResult(GlobalValue *Dest, const GlobalValue *Src, + GlobalValue::LinkageTypes <, + GlobalValue::VisibilityTypes &Vis, + bool &LinkFromSrc); + + /// getLinkedToGlobal - Given a global in the source module, return the + /// global in the destination module that is being linked to, if any. + GlobalValue *getLinkedToGlobal(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 0; + + // Otherwise see if we have a match in the destination module's symtab. + GlobalValue *DGV = DstM->getNamedValue(SrcGV->getName()); + if (DGV == 0) return 0; + + // 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 0; + + // Otherwise, we do in fact link to the destination global. + return DGV; + } + + void computeTypeMapping(); + bool categorizeModuleFlagNodes(const NamedMDNode *ModFlags, + DenseMap &ErrorNode, + DenseMap &WarningNode, + DenseMap &OverrideNode, + DenseMap > &RequireNodes, + SmallSetVector &SeenIDs); + + bool linkAppendingVarProto(GlobalVariable *DstGV, GlobalVariable *SrcGV); + bool linkGlobalProto(GlobalVariable *SrcGV); + bool linkFunctionProto(Function *SrcF); + bool linkAliasProto(GlobalAlias *SrcA); + bool linkModuleFlagsMetadata(); + + void linkAppendingVarInit(const AppendingVarInfo &AVI); + void linkGlobalInits(); + void linkFunctionBody(Function *Dst, Function *Src); + void linkAliasBodies(); + void linkNamedMDNodes(); + }; } -/// ForceRenaming - The LLVM SymbolTable class autorenames globals that conflict +/// forceRenaming - 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, const std::string &Name) { - assert(GV->getName() != Name && "Can't force rename to self"); - ValueSymbolTable &ST = GV->getParent()->getValueSymbolTable(); +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 = cast_or_null(ST.lookup(Name))) { - assert(ConflictGV->hasLocalLinkage() && - "Not conflicting with a static global, should link instead!"); + 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"); + assert(ConflictGV->getName() != Name && "forceRenaming didn't work"); } else { GV->setName(Name); // Force the name back } } -/// CopyGVAttributes - copy additional attributes (those not needed to construct +/// copyGVAttributes - copy additional attributes (those not needed to construct /// a GlobalValue) from the SrcGV to the DestGV. -static void CopyGVAttributes(GlobalValue *DestGV, const GlobalValue *SrcGV) { +static void copyGVAttributes(GlobalValue *DestGV, const GlobalValue *SrcGV) { // Use the maximum alignment, rather than just copying the alignment of SrcGV. unsigned Alignment = std::max(DestGV->getAlignment(), SrcGV->getAlignment()); DestGV->copyAttributesFrom(SrcGV); DestGV->setAlignment(Alignment); + + forceRenaming(DestGV, SrcGV->getName()); +} + +static bool isLessConstraining(GlobalValue::VisibilityTypes a, + GlobalValue::VisibilityTypes b) { + if (a == GlobalValue::HiddenVisibility) + return false; + if (b == GlobalValue::HiddenVisibility) + return true; + if (a == GlobalValue::ProtectedVisibility) + return false; + if (b == GlobalValue::ProtectedVisibility) + return true; + return false; } -/// GetLinkageResult - This analyzes the two global values and determines what +/// getLinkageResult - This analyzes the two global values and determines what /// the result will look like in the destination module. In particular, it -/// computes the resultant linkage type, computes whether the global in the -/// source should be copied over to the destination (replacing the existing -/// one), and computes whether this linkage is an error or not. It also performs -/// visibility checks: we cannot link together two symbols with different -/// visibilities. -static bool GetLinkageResult(GlobalValue *Dest, const GlobalValue *Src, - GlobalValue::LinkageTypes <, bool &LinkFromSrc, - std::string *Err) { - assert((!Dest || !Src->hasLocalLinkage()) && +/// computes the resultant linkage type and visibility, computes whether the +/// global in the source should be copied over to the destination (replacing +/// the existing one), and computes whether this linkage is an error or not. +bool ModuleLinker::getLinkageResult(GlobalValue *Dest, const GlobalValue *Src, + GlobalValue::LinkageTypes <, + GlobalValue::VisibilityTypes &Vis, + bool &LinkFromSrc) { + assert(Dest && "Must have two globals being queried"); + assert(!Src->hasLocalLinkage() && "If Src has internal linkage, Dest shouldn't be set!"); - if (!Dest) { - // Linking something to nothing. - LinkFromSrc = true; - LT = Src->getLinkage(); - } else if (Src->isDeclaration()) { + + bool SrcIsDeclaration = Src->isDeclaration() && !Src->isMaterializable(); + bool DestIsDeclaration = Dest->isDeclaration(); + + if (SrcIsDeclaration) { // If Src is external or if both Src & Dest are external.. Just link the // external globals, we aren't adding anything. if (Src->hasDLLImportLinkage()) { // If one of GVs has DLLImport linkage, result should be dllimport'ed. - if (Dest->isDeclaration()) { + if (DestIsDeclaration) { LinkFromSrc = true; LT = Src->getLinkage(); } @@ -486,16 +520,10 @@ static bool GetLinkageResult(GlobalValue *Dest, const GlobalValue *Src, LinkFromSrc = false; LT = Dest->getLinkage(); } - } else if (Dest->isDeclaration() && !Dest->hasDLLImportLinkage()) { + } else if (DestIsDeclaration && !Dest->hasDLLImportLinkage()) { // If Dest is external but Src is not: LinkFromSrc = true; LT = Src->getLinkage(); - } else if (Src->hasAppendingLinkage() || Dest->hasAppendingLinkage()) { - if (Src->getLinkage() != Dest->getLinkage()) - return Error(Err, "Linking globals named '" + Src->getName() + - "': can only link appending global with another appending global!"); - LinkFromSrc = true; // Special cased. - LT = Src->getLinkage(); } else if (Src->isWeakForLinker()) { // At this point we know that Dest has LinkOnce, External*, Weak, Common, // or DLL* linkage. @@ -519,861 +547,806 @@ static bool GetLinkageResult(GlobalValue *Dest, const GlobalValue *Src, LT = GlobalValue::ExternalLinkage; } } else { - assert((Dest->hasExternalLinkage() || - Dest->hasDLLImportLinkage() || - Dest->hasDLLExportLinkage() || - Dest->hasExternalWeakLinkage()) && - (Src->hasExternalLinkage() || - Src->hasDLLImportLinkage() || - Src->hasDLLExportLinkage() || - Src->hasExternalWeakLinkage()) && + assert((Dest->hasExternalLinkage() || Dest->hasDLLImportLinkage() || + Dest->hasDLLExportLinkage() || Dest->hasExternalWeakLinkage()) && + (Src->hasExternalLinkage() || Src->hasDLLImportLinkage() || + Src->hasDLLExportLinkage() || Src->hasExternalWeakLinkage()) && "Unexpected linkage type!"); - return Error(Err, "Linking globals named '" + Src->getName() + + return emitError("Linking globals named '" + Src->getName() + "': symbol multiply defined!"); } - // Check visibility - if (Dest && Src->getVisibility() != Dest->getVisibility()) - if (!Src->isDeclaration() && !Dest->isDeclaration()) - return Error(Err, "Linking globals named '" + Src->getName() + - "': symbols have different visibilities!"); + // Compute the visibility. We follow the rules in the System V Application + // Binary Interface. + Vis = isLessConstraining(Src->getVisibility(), Dest->getVisibility()) ? + Dest->getVisibility() : Src->getVisibility(); return false; } -// Insert all of the named mdnoes in Src into the Dest module. -static void LinkNamedMDNodes(Module *Dest, Module *Src) { - for (Module::const_named_metadata_iterator I = Src->named_metadata_begin(), - E = Src->named_metadata_end(); I != E; ++I) { - const NamedMDNode *SrcNMD = I; - NamedMDNode *DestNMD = Dest->getOrInsertNamedMetadata(SrcNMD->getName()); - // Add Src elements into Dest node. - for (unsigned i = 0, e = SrcNMD->getNumOperands(); i != e; ++i) - DestNMD->addOperand(SrcNMD->getOperand(i)); +/// computeTypeMapping - 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 ModuleLinker::computeTypeMapping() { + // Incorporate globals. + for (Module::global_iterator I = SrcM->global_begin(), + E = SrcM->global_end(); I != E; ++I) { + GlobalValue *DGV = getLinkedToGlobal(I); + if (DGV == 0) continue; + + if (!DGV->hasAppendingLinkage() || !I->hasAppendingLinkage()) { + TypeMap.addTypeMapping(DGV->getType(), I->getType()); + continue; + } + + // Unify the element type of appending arrays. + ArrayType *DAT = cast(DGV->getType()->getElementType()); + ArrayType *SAT = cast(I->getType()->getElementType()); + TypeMap.addTypeMapping(DAT->getElementType(), SAT->getElementType()); + } + + // Incorporate functions. + for (Module::iterator I = SrcM->begin(), E = SrcM->end(); I != E; ++I) { + if (GlobalValue *DGV = getLinkedToGlobal(I)) + TypeMap.addTypeMapping(DGV->getType(), I->getType()); } -} - -// LinkGlobals - Loop through the global variables in the src module and merge -// them into the dest module. -static bool LinkGlobals(Module *Dest, const Module *Src, - std::map &ValueMap, - std::multimap &AppendingVars, - std::string *Err) { - ValueSymbolTable &DestSymTab = Dest->getValueSymbolTable(); - - // Loop over all of the globals in the src module, mapping them over as we go - for (Module::const_global_iterator I = Src->global_begin(), - E = Src->global_end(); I != E; ++I) { - const GlobalVariable *SGV = I; - GlobalValue *DGV = 0; - - // Check to see if may have to link the global with the global, alias or - // function. - if (SGV->hasName() && !SGV->hasLocalLinkage()) - DGV = cast_or_null(DestSymTab.lookup(SGV->getName())); - // 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 && DGV->hasLocalLinkage()) - DGV = 0; + // 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 SrcStructTypes; + SrcM->findUsedStructTypes(SrcStructTypes, true); + SmallPtrSet SrcStructTypesSet(SrcStructTypes.begin(), + SrcStructTypes.end()); + + std::vector DstStructTypes; + DstM->findUsedStructTypes(DstStructTypes, true); + SmallPtrSet DstStructTypesSet(DstStructTypes.begin(), + DstStructTypes.end()); + + for (unsigned i = 0, e = SrcStructTypes.size(); i != e; ++i) { + StructType *ST = SrcStructTypes[i]; + 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(ST->getName()[DotPos+1])) + continue; + + // Check to see if the destination module has a struct with the prefix name. + if (StructType *DST = DstM->getTypeByName(ST->getName().substr(0, DotPos))) + // 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 (!SrcStructTypesSet.count(DST) && DstStructTypesSet.count(DST)) + TypeMap.addTypeMapping(DST, ST); + } - // If types don't agree due to opaque types, try to resolve them. - if (DGV && DGV->getType() != SGV->getType()) - RecursiveResolveTypes(SGV->getType(), DGV->getType()); + // Don't bother incorporating aliases, they aren't generally typed well. + + // 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(); +} - assert((SGV->hasInitializer() || SGV->hasExternalWeakLinkage() || - SGV->hasExternalLinkage() || SGV->hasDLLImportLinkage()) && - "Global must either be external or have an initializer!"); +/// linkAppendingVarProto - If there were any appending global variables, link +/// them together now. Return true on error. +bool ModuleLinker::linkAppendingVarProto(GlobalVariable *DstGV, + GlobalVariable *SrcGV) { + + if (!SrcGV->hasAppendingLinkage() || !DstGV->hasAppendingLinkage()) + return emitError("Linking globals named '" + SrcGV->getName() + + "': can only link appending global with another appending global!"); + + ArrayType *DstTy = cast(DstGV->getType()->getElementType()); + ArrayType *SrcTy = + cast(TypeMap.get(SrcGV->getType()->getElementType())); + Type *EltTy = DstTy->getElementType(); + + // Check to see that they two arrays agree on type. + if (EltTy != SrcTy->getElementType()) + return emitError("Appending variables with different element types!"); + if (DstGV->isConstant() != SrcGV->isConstant()) + return emitError("Appending variables linked with different const'ness!"); + + if (DstGV->getAlignment() != SrcGV->getAlignment()) + return emitError( + "Appending variables with different alignment need to be linked!"); + + if (DstGV->getVisibility() != SrcGV->getVisibility()) + return emitError( + "Appending variables with different visibility need to be linked!"); + + if (DstGV->getSection() != SrcGV->getSection()) + return emitError( + "Appending variables with different section name need to be linked!"); + + uint64_t NewSize = DstTy->getNumElements() + SrcTy->getNumElements(); + ArrayType *NewType = ArrayType::get(EltTy, NewSize); + + // Create the new global variable. + GlobalVariable *NG = + new GlobalVariable(*DstGV->getParent(), NewType, SrcGV->isConstant(), + DstGV->getLinkage(), /*init*/0, /*name*/"", DstGV, + DstGV->isThreadLocal(), + DstGV->getType()->getAddressSpace()); + + // Propagate alignment, visibility and section info. + copyGVAttributes(NG, DstGV); + + AppendingVarInfo AVI; + AVI.NewGV = NG; + AVI.DstInit = DstGV->getInitializer(); + AVI.SrcInit = SrcGV->getInitializer(); + AppendingVars.push_back(AVI); + + // Replace any uses of the two global variables with uses of the new + // global. + ValueMap[SrcGV] = ConstantExpr::getBitCast(NG, TypeMap.get(SrcGV->getType())); + + DstGV->replaceAllUsesWith(ConstantExpr::getBitCast(NG, DstGV->getType())); + DstGV->eraseFromParent(); + + // Track the source variable so we don't try to link it. + DoNotLinkFromSource.insert(SrcGV); + + return false; +} +/// linkGlobalProto - Loop through the global variables in the src module and +/// merge them into the dest module. +bool ModuleLinker::linkGlobalProto(GlobalVariable *SGV) { + GlobalValue *DGV = getLinkedToGlobal(SGV); + llvm::Optional NewVisibility; + + if (DGV) { + // Concatenation of appending linkage variables is magic and handled later. + if (DGV->hasAppendingLinkage() || SGV->hasAppendingLinkage()) + return linkAppendingVarProto(cast(DGV), SGV); + + // Determine whether linkage of these two globals follows the source + // module's definition or the destination module's definition. GlobalValue::LinkageTypes NewLinkage = GlobalValue::InternalLinkage; + GlobalValue::VisibilityTypes NV; bool LinkFromSrc = false; - if (GetLinkageResult(DGV, SGV, NewLinkage, LinkFromSrc, Err)) + if (getLinkageResult(DGV, SGV, NewLinkage, NV, LinkFromSrc)) return true; - - if (DGV == 0) { - // No linking to be performed, 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(*Dest, SGV->getType()->getElementType(), - SGV->isConstant(), SGV->getLinkage(), /*init*/0, - SGV->getName(), 0, false, - SGV->getType()->getAddressSpace()); - // Propagate alignment, visibility and section info. - CopyGVAttributes(NewDGV, SGV); - - // If the LLVM runtime renamed the global, but it is an externally visible - // symbol, DGV must be an existing global with internal linkage. Rename - // it. - if (!NewDGV->hasLocalLinkage() && NewDGV->getName() != SGV->getName()) - ForceRenaming(NewDGV, SGV->getName()); - - // Make sure to remember this mapping. - ValueMap[SGV] = NewDGV; - - // Keep track that this is an appending variable. - if (SGV->hasAppendingLinkage()) - AppendingVars.insert(std::make_pair(SGV->getName(), NewDGV)); - continue; - } - - // If the visibilities of the symbols disagree and the destination is a - // prototype, take the visibility of its input. - if (DGV->isDeclaration()) - DGV->setVisibility(SGV->getVisibility()); - - if (DGV->hasAppendingLinkage()) { - // No linking is performed yet. Just insert a new copy of the global, and - // keep track of the fact that it is an appending variable in the - // AppendingVars map. The name is cleared out so that no linkage is - // performed. - GlobalVariable *NewDGV = - new GlobalVariable(*Dest, SGV->getType()->getElementType(), - SGV->isConstant(), SGV->getLinkage(), /*init*/0, - "", 0, false, - SGV->getType()->getAddressSpace()); - - // Set alignment allowing CopyGVAttributes merge it with alignment of SGV. - NewDGV->setAlignment(DGV->getAlignment()); - // Propagate alignment, section and visibility info. - CopyGVAttributes(NewDGV, SGV); - - // Make sure to remember this mapping... - ValueMap[SGV] = NewDGV; - - // Keep track that this is an appending variable... - AppendingVars.insert(std::make_pair(SGV->getName(), NewDGV)); - continue; - } - - if (LinkFromSrc) { - if (isa(DGV)) - return Error(Err, "Global-Alias Collision on '" + SGV->getName() + - "': symbol multiple defined"); - - // If the types don't match, and if we are to link from the source, nuke - // DGV and create a new one of the appropriate type. Note that the thing - // we are replacing may be a function (if a prototype, weak, etc) or a - // global variable. - GlobalVariable *NewDGV = - new GlobalVariable(*Dest, SGV->getType()->getElementType(), - SGV->isConstant(), NewLinkage, /*init*/0, - DGV->getName(), 0, false, - SGV->getType()->getAddressSpace()); - - // Propagate alignment, section, and visibility info. - CopyGVAttributes(NewDGV, SGV); - DGV->replaceAllUsesWith(ConstantExpr::getBitCast(NewDGV, - DGV->getType())); - - // DGV will conflict with NewDGV because they both had the same - // name. We must erase this now so ForceRenaming doesn't assert - // because DGV might not have internal linkage. - if (GlobalVariable *Var = dyn_cast(DGV)) - Var->eraseFromParent(); - else - cast(DGV)->eraseFromParent(); - - // If the symbol table renamed the global, but it is an externally visible - // symbol, DGV must be an existing global with internal linkage. Rename. - if (NewDGV->getName() != SGV->getName() && !NewDGV->hasLocalLinkage()) - ForceRenaming(NewDGV, SGV->getName()); - - // Inherit const as appropriate. - NewDGV->setConstant(SGV->isConstant()); + NewVisibility = NV; + + // If we're not linking from the source, then keep the definition that we + // have. + if (!LinkFromSrc) { + // Special case for const propagation. + if (GlobalVariable *DGVar = dyn_cast(DGV)) + if (DGVar->isDeclaration() && SGV->isConstant() && !DGVar->isConstant()) + DGVar->setConstant(true); + + // Set calculated linkage and visibility. + DGV->setLinkage(NewLinkage); + DGV->setVisibility(*NewVisibility); // Make sure to remember this mapping. - ValueMap[SGV] = NewDGV; - continue; - } - - // Not "link from source", keep the one in the DestModule and remap the - // input onto it. - - // Special case for const propagation. - if (GlobalVariable *DGVar = dyn_cast(DGV)) - if (DGVar->isDeclaration() && SGV->isConstant() && !DGVar->isConstant()) - DGVar->setConstant(true); - - // SGV is global, but DGV is alias. - if (isa(DGV)) { - // The only valid mappings are: - // - SGV is external declaration, which is effectively a no-op. - // - SGV is weak, when we just need to throw SGV out. - if (!SGV->isDeclaration() && !SGV->isWeakForLinker()) - return Error(Err, "Global-Alias Collision on '" + SGV->getName() + - "': symbol multiple defined"); + ValueMap[SGV] = ConstantExpr::getBitCast(DGV,TypeMap.get(SGV->getType())); + + // Track the source global so that we don't attempt to copy it over when + // processing global initializers. + DoNotLinkFromSource.insert(SGV); + + return false; } - - // Set calculated linkage - DGV->setLinkage(NewLinkage); - - // Make sure to remember this mapping... - ValueMap[SGV] = ConstantExpr::getBitCast(DGV, SGV->getType()); } - return false; -} - -static GlobalValue::LinkageTypes -CalculateAliasLinkage(const GlobalValue *SGV, const GlobalValue *DGV) { - GlobalValue::LinkageTypes SL = SGV->getLinkage(); - GlobalValue::LinkageTypes DL = DGV->getLinkage(); - if (SL == GlobalValue::ExternalLinkage || DL == GlobalValue::ExternalLinkage) - return GlobalValue::ExternalLinkage; - else if (SL == GlobalValue::WeakAnyLinkage || - DL == GlobalValue::WeakAnyLinkage) - return GlobalValue::WeakAnyLinkage; - else if (SL == GlobalValue::WeakODRLinkage || - DL == GlobalValue::WeakODRLinkage) - return GlobalValue::WeakODRLinkage; - else if (SL == GlobalValue::InternalLinkage && - DL == GlobalValue::InternalLinkage) - return GlobalValue::InternalLinkage; - else if (SL == GlobalValue::LinkerPrivateLinkage && - DL == GlobalValue::LinkerPrivateLinkage) - return GlobalValue::LinkerPrivateLinkage; - else { - assert (SL == GlobalValue::PrivateLinkage && - DL == GlobalValue::PrivateLinkage && "Unexpected linkage type"); - return GlobalValue::PrivateLinkage; + + // 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(SGV->getType()->getElementType()), + SGV->isConstant(), SGV->getLinkage(), /*init*/0, + SGV->getName(), /*insertbefore*/0, + SGV->isThreadLocal(), + SGV->getType()->getAddressSpace()); + // Propagate alignment, visibility and section info. + copyGVAttributes(NewDGV, SGV); + if (NewVisibility) + NewDGV->setVisibility(*NewVisibility); + + if (DGV) { + DGV->replaceAllUsesWith(ConstantExpr::getBitCast(NewDGV, DGV->getType())); + DGV->eraseFromParent(); } + + // Make sure to remember this mapping. + ValueMap[SGV] = NewDGV; + return false; } -// LinkAlias - Loop through the alias in the src module and link them into the -// dest module. We're assuming, that all functions/global variables were already -// linked in. -static bool LinkAlias(Module *Dest, const Module *Src, - std::map &ValueMap, - std::string *Err) { - // Loop over all alias in the src module - for (Module::const_alias_iterator I = Src->alias_begin(), - E = Src->alias_end(); I != E; ++I) { - const GlobalAlias *SGA = I; - const GlobalValue *SAliasee = SGA->getAliasedGlobal(); - GlobalAlias *NewGA = NULL; - - // Globals were already linked, thus we can just query ValueMap for variant - // of SAliasee in Dest. - std::map::const_iterator VMI = ValueMap.find(SAliasee); - assert(VMI != ValueMap.end() && "Aliasee not linked"); - GlobalValue* DAliasee = cast(VMI->second); - GlobalValue* DGV = NULL; - - // Try to find something 'similar' to SGA in destination module. - if (!DGV && !SGA->hasLocalLinkage()) { - DGV = Dest->getNamedAlias(SGA->getName()); - - // If types don't agree due to opaque types, try to resolve them. - if (DGV && DGV->getType() != SGA->getType()) - RecursiveResolveTypes(SGA->getType(), DGV->getType()); - } - - if (!DGV && !SGA->hasLocalLinkage()) { - DGV = Dest->getGlobalVariable(SGA->getName()); - - // If types don't agree due to opaque types, try to resolve them. - if (DGV && DGV->getType() != SGA->getType()) - RecursiveResolveTypes(SGA->getType(), DGV->getType()); - } +/// linkFunctionProto - Link the function in the source module into the +/// destination module if needed, setting up mapping information. +bool ModuleLinker::linkFunctionProto(Function *SF) { + GlobalValue *DGV = getLinkedToGlobal(SF); + llvm::Optional NewVisibility; - if (!DGV && !SGA->hasLocalLinkage()) { - DGV = Dest->getFunction(SGA->getName()); + if (DGV) { + GlobalValue::LinkageTypes NewLinkage = GlobalValue::InternalLinkage; + bool LinkFromSrc = false; + GlobalValue::VisibilityTypes NV; + if (getLinkageResult(DGV, SF, NewLinkage, NV, LinkFromSrc)) + return true; + NewVisibility = NV; - // If types don't agree due to opaque types, try to resolve them. - if (DGV && DGV->getType() != SGA->getType()) - RecursiveResolveTypes(SGA->getType(), DGV->getType()); - } + if (!LinkFromSrc) { + // Set calculated linkage + DGV->setLinkage(NewLinkage); + DGV->setVisibility(*NewVisibility); - // No linking to be performed on internal stuff. - if (DGV && DGV->hasLocalLinkage()) - DGV = NULL; - - if (GlobalAlias *DGA = dyn_cast_or_null(DGV)) { - // Types are known to be the same, check whether aliasees equal. As - // globals are already linked we just need query ValueMap to find the - // mapping. - if (DAliasee == DGA->getAliasedGlobal()) { - // This is just two copies of the same alias. Propagate linkage, if - // necessary. - DGA->setLinkage(CalculateAliasLinkage(SGA, DGA)); - - NewGA = DGA; - // Proceed to 'common' steps - } else - return Error(Err, "Alias Collision on '" + SGA->getName()+ - "': aliases have different aliasees"); - } else if (GlobalVariable *DGVar = dyn_cast_or_null(DGV)) { - // The only allowed way is to link alias with external declaration or weak - // symbol.. - if (DGVar->isDeclaration() || DGVar->isWeakForLinker()) { - // But only if aliasee is global too... - if (!isa(DAliasee)) - return Error(Err, "Global-Alias Collision on '" + SGA->getName() + - "': aliasee is not global variable"); - - NewGA = new GlobalAlias(SGA->getType(), SGA->getLinkage(), - SGA->getName(), DAliasee, Dest); - CopyGVAttributes(NewGA, SGA); - - // Any uses of DGV need to change to NewGA, with cast, if needed. - if (SGA->getType() != DGVar->getType()) - DGVar->replaceAllUsesWith(ConstantExpr::getBitCast(NewGA, - DGVar->getType())); - else - DGVar->replaceAllUsesWith(NewGA); - - // DGVar will conflict with NewGA because they both had the same - // name. We must erase this now so ForceRenaming doesn't assert - // because DGV might not have internal linkage. - DGVar->eraseFromParent(); - - // Proceed to 'common' steps - } else - return Error(Err, "Global-Alias Collision on '" + SGA->getName() + - "': symbol multiple defined"); - } else if (Function *DF = dyn_cast_or_null(DGV)) { - // The only allowed way is to link alias with external declaration or weak - // symbol... - if (DF->isDeclaration() || DF->isWeakForLinker()) { - // But only if aliasee is function too... - if (!isa(DAliasee)) - return Error(Err, "Function-Alias Collision on '" + SGA->getName() + - "': aliasee is not function"); - - NewGA = new GlobalAlias(SGA->getType(), SGA->getLinkage(), - SGA->getName(), DAliasee, Dest); - CopyGVAttributes(NewGA, SGA); - - // Any uses of DF need to change to NewGA, with cast, if needed. - if (SGA->getType() != DF->getType()) - DF->replaceAllUsesWith(ConstantExpr::getBitCast(NewGA, - DF->getType())); - else - DF->replaceAllUsesWith(NewGA); - - // DF will conflict with NewGA because they both had the same - // name. We must erase this now so ForceRenaming doesn't assert - // because DF might not have internal linkage. - DF->eraseFromParent(); - - // Proceed to 'common' steps - } else - return Error(Err, "Function-Alias Collision on '" + SGA->getName() + - "': symbol multiple defined"); - } else { - // No linking to be performed, simply create an identical version of the - // alias over in the dest module... - Constant *Aliasee = DAliasee; - // Fixup aliases to bitcasts. Note that aliases to GEPs are still broken - // by this, but aliases to GEPs are broken to a lot of other things, so - // it's less important. - if (SGA->getType() != DAliasee->getType()) - Aliasee = ConstantExpr::getBitCast(DAliasee, SGA->getType()); - NewGA = new GlobalAlias(SGA->getType(), SGA->getLinkage(), - SGA->getName(), Aliasee, Dest); - CopyGVAttributes(NewGA, SGA); - - // Proceed to 'common' steps + // Make sure to remember this mapping. + ValueMap[SF] = ConstantExpr::getBitCast(DGV, TypeMap.get(SF->getType())); + + // Track the function from the source module so we don't attempt to remap + // it. + DoNotLinkFromSource.insert(SF); + + return false; } - - assert(NewGA && "No alias was created in destination module!"); - - // If the symbol table renamed the alias, but it is an externally visible - // symbol, DGA must be an global value with internal linkage. Rename it. - if (NewGA->getName() != SGA->getName() && - !NewGA->hasLocalLinkage()) - ForceRenaming(NewGA, SGA->getName()); - - // Remember this mapping so uses in the source module get remapped - // later by RemapOperand. - ValueMap[SGA] = NewGA; } - - return false; -} - - -// LinkGlobalInits - Update the initializers in the Dest module now that all -// globals that may be referenced are in Dest. -static bool LinkGlobalInits(Module *Dest, const Module *Src, - std::map &ValueMap, - std::string *Err) { - // Loop over all of the globals in the src module, mapping them over as we go - for (Module::const_global_iterator I = Src->global_begin(), - E = Src->global_end(); I != E; ++I) { - const GlobalVariable *SGV = I; - - if (SGV->hasInitializer()) { // Only process initialized GV's - // Figure out what the initializer looks like in the dest module... - Constant *SInit = - cast(RemapOperand(SGV->getInitializer(), ValueMap)); - // Grab destination global variable or alias. - GlobalValue *DGV = cast(ValueMap[SGV]->stripPointerCasts()); - - // If dest if global variable, check that initializers match. - if (GlobalVariable *DGVar = dyn_cast(DGV)) { - if (DGVar->hasInitializer()) { - if (SGV->hasExternalLinkage()) { - if (DGVar->getInitializer() != SInit) - return Error(Err, "Global Variable Collision on '" + - SGV->getName() + - "': global variables have different initializers"); - } else if (DGVar->isWeakForLinker()) { - // Nothing is required, mapped values will take the new global - // automatically. - } else if (SGV->isWeakForLinker()) { - // Nothing is required, mapped values will take the new global - // automatically. - } else if (DGVar->hasAppendingLinkage()) { - llvm_unreachable("Appending linkage unimplemented!"); - } else { - llvm_unreachable("Unknown linkage!"); - } - } else { - // Copy the initializer over now... - DGVar->setInitializer(SInit); - } - } else { - // Destination is alias, the only valid situation is when source is - // weak. Also, note, that we already checked linkage in LinkGlobals(), - // thus we assert here. - // FIXME: Should we weaken this assumption, 'dereference' alias and - // check for initializer of aliasee? - assert(SGV->isWeakForLinker()); - } + + // If there is no linkage to be performed or we are linking from the source, + // bring SF over. + Function *NewDF = Function::Create(TypeMap.get(SF->getFunctionType()), + SF->getLinkage(), SF->getName(), DstM); + copyGVAttributes(NewDF, SF); + if (NewVisibility) + NewDF->setVisibility(*NewVisibility); + + if (DGV) { + // Any uses of DF need to change to NewDF, with cast. + DGV->replaceAllUsesWith(ConstantExpr::getBitCast(NewDF, DGV->getType())); + DGV->eraseFromParent(); + } else { + // Internal, LO_ODR, or LO linkage - stick in set to ignore and lazily link. + if (SF->hasLocalLinkage() || SF->hasLinkOnceLinkage() || + SF->hasAvailableExternallyLinkage()) { + DoNotLinkFromSource.insert(SF); + LazilyLinkFunctions.push_back(SF); } } + + ValueMap[SF] = NewDF; return false; } -// LinkFunctionProtos - Link the functions together between the two modules, -// without doing function bodies... this just adds external function prototypes -// to the Dest function... -// -static bool LinkFunctionProtos(Module *Dest, const Module *Src, - std::map &ValueMap, - std::string *Err) { - ValueSymbolTable &DestSymTab = Dest->getValueSymbolTable(); - - // Loop over all of the functions in the src module, mapping them over - for (Module::const_iterator I = Src->begin(), E = Src->end(); I != E; ++I) { - const Function *SF = I; // SrcFunction - GlobalValue *DGV = 0; - - // Check to see if may have to link the function with the global, alias or - // function. - if (SF->hasName() && !SF->hasLocalLinkage()) - DGV = cast_or_null(DestSymTab.lookup(SF->getName())); - - // 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 && DGV->hasLocalLinkage()) - DGV = 0; - - // If types don't agree due to opaque types, try to resolve them. - if (DGV && DGV->getType() != SF->getType()) - RecursiveResolveTypes(SF->getType(), DGV->getType()); +/// LinkAliasProto - Set up prototypes for any aliases that come over from the +/// source module. +bool ModuleLinker::linkAliasProto(GlobalAlias *SGA) { + GlobalValue *DGV = getLinkedToGlobal(SGA); + llvm::Optional NewVisibility; + if (DGV) { GlobalValue::LinkageTypes NewLinkage = GlobalValue::InternalLinkage; + GlobalValue::VisibilityTypes NV; bool LinkFromSrc = false; - if (GetLinkageResult(DGV, SF, NewLinkage, LinkFromSrc, Err)) + if (getLinkageResult(DGV, SGA, NewLinkage, NV, LinkFromSrc)) return true; + NewVisibility = NV; - // If there is no linkage to be performed, just bring over SF without - // modifying it. - if (DGV == 0) { - // Function does not already exist, simply insert an function signature - // identical to SF into the dest module. - Function *NewDF = Function::Create(SF->getFunctionType(), - SF->getLinkage(), - SF->getName(), Dest); - CopyGVAttributes(NewDF, SF); - - // If the LLVM runtime renamed the function, but it is an externally - // visible symbol, DF must be an existing function with internal linkage. - // Rename it. - if (!NewDF->hasLocalLinkage() && NewDF->getName() != SF->getName()) - ForceRenaming(NewDF, SF->getName()); - - // ... and remember this mapping... - ValueMap[SF] = NewDF; - continue; - } + if (!LinkFromSrc) { + // Set calculated linkage. + DGV->setLinkage(NewLinkage); + DGV->setVisibility(*NewVisibility); - // If the visibilities of the symbols disagree and the destination is a - // prototype, take the visibility of its input. - if (DGV->isDeclaration()) - DGV->setVisibility(SF->getVisibility()); - - if (LinkFromSrc) { - if (isa(DGV)) - return Error(Err, "Function-Alias Collision on '" + SF->getName() + - "': symbol multiple defined"); - - // We have a definition of the same name but different type in the - // source module. Copy the prototype to the destination and replace - // uses of the destination's prototype with the new prototype. - Function *NewDF = Function::Create(SF->getFunctionType(), NewLinkage, - SF->getName(), Dest); - CopyGVAttributes(NewDF, SF); - - // Any uses of DF need to change to NewDF, with cast - DGV->replaceAllUsesWith(ConstantExpr::getBitCast(NewDF, - DGV->getType())); - - // DF will conflict with NewDF because they both had the same. We must - // erase this now so ForceRenaming doesn't assert because DF might - // not have internal linkage. - if (GlobalVariable *Var = dyn_cast(DGV)) - Var->eraseFromParent(); - else - cast(DGV)->eraseFromParent(); - - // If the symbol table renamed the function, but it is an externally - // visible symbol, DF must be an existing function with internal - // linkage. Rename it. - if (NewDF->getName() != SF->getName() && !NewDF->hasLocalLinkage()) - ForceRenaming(NewDF, SF->getName()); - - // Remember this mapping so uses in the source module get remapped - // later by RemapOperand. - ValueMap[SF] = NewDF; - continue; + // Make sure to remember this mapping. + ValueMap[SGA] = ConstantExpr::getBitCast(DGV,TypeMap.get(SGA->getType())); + + // Track the alias from the source module so we don't attempt to remap it. + DoNotLinkFromSource.insert(SGA); + + return false; } + } + + // If there is no linkage to be performed or we're linking from the source, + // bring over SGA. + GlobalAlias *NewDA = new GlobalAlias(TypeMap.get(SGA->getType()), + SGA->getLinkage(), SGA->getName(), + /*aliasee*/0, DstM); + copyGVAttributes(NewDA, SGA); + if (NewVisibility) + NewDA->setVisibility(*NewVisibility); + + if (DGV) { + // Any uses of DGV need to change to NewDA, with cast. + DGV->replaceAllUsesWith(ConstantExpr::getBitCast(NewDA, DGV->getType())); + DGV->eraseFromParent(); + } + + ValueMap[SGA] = NewDA; + return false; +} - // Not "link from source", keep the one in the DestModule and remap the - // input onto it. - - if (isa(DGV)) { - // The only valid mappings are: - // - SF is external declaration, which is effectively a no-op. - // - SF is weak, when we just need to throw SF out. - if (!SF->isDeclaration() && !SF->isWeakForLinker()) - return Error(Err, "Function-Alias Collision on '" + SF->getName() + - "': symbol multiple defined"); - } +static void getArrayElements(Constant *C, SmallVectorImpl &Dest) { + unsigned NumElements = cast(C->getType())->getNumElements(); - // Set calculated linkage - DGV->setLinkage(NewLinkage); + for (unsigned i = 0; i != NumElements; ++i) + Dest.push_back(C->getAggregateElement(i)); +} + +void ModuleLinker::linkAppendingVarInit(const AppendingVarInfo &AVI) { + // Merge the initializer. + SmallVector Elements; + getArrayElements(AVI.DstInit, Elements); + + Constant *SrcInit = MapValue(AVI.SrcInit, ValueMap, RF_None, &TypeMap); + getArrayElements(SrcInit, Elements); + + ArrayType *NewType = cast(AVI.NewGV->getType()->getElementType()); + AVI.NewGV->setInitializer(ConstantArray::get(NewType, Elements)); +} - // Make sure to remember this mapping. - ValueMap[SF] = ConstantExpr::getBitCast(DGV, SF->getType()); +/// linkGlobalInits - Update the initializers in the Dest module now that all +/// globals that may be referenced are in Dest. +void ModuleLinker::linkGlobalInits() { + // Loop over all of the globals in the src module, mapping them over as we go + for (Module::const_global_iterator I = SrcM->global_begin(), + E = SrcM->global_end(); I != E; ++I) { + + // Only process initialized GV's or ones not already in dest. + if (!I->hasInitializer() || DoNotLinkFromSource.count(I)) continue; + + // Grab destination global variable. + GlobalVariable *DGV = cast(ValueMap[I]); + // Figure out what the initializer looks like in the dest module. + DGV->setInitializer(MapValue(I->getInitializer(), ValueMap, + RF_None, &TypeMap)); } - return false; } -// LinkFunctionBody - 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. -static bool LinkFunctionBody(Function *Dest, Function *Src, - std::map &ValueMap, - std::string *Err) { - assert(Src && Dest && Dest->isDeclaration() && !Src->isDeclaration()); +/// linkFunctionBody - 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. +void ModuleLinker::linkFunctionBody(Function *Dst, Function *Src) { + assert(Src && Dst && Dst->isDeclaration() && !Src->isDeclaration()); // Go through and convert function arguments over, remembering the mapping. - Function::arg_iterator DI = Dest->arg_begin(); + Function::arg_iterator DI = Dst->arg_begin(); for (Function::arg_iterator I = Src->arg_begin(), E = Src->arg_end(); I != E; ++I, ++DI) { - DI->setName(I->getName()); // Copy the name information over... + DI->setName(I->getName()); // Copy the name over. - // Add a mapping to our local map + // Add a mapping to our mapping. ValueMap[I] = DI; } - // Splice the body of the source function into the dest function. - Dest->getBasicBlockList().splice(Dest->end(), Src->getBasicBlockList()); - - // At this point, all of the instructions and values of the function are now - // copied over. The only problem is that they are still referencing values in - // the Source function as operands. Loop through all of the operands of the - // functions and patch them up to point to the local versions... - // - for (Function::iterator BB = Dest->begin(), BE = Dest->end(); BB != BE; ++BB) - for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) - for (Instruction::op_iterator OI = I->op_begin(), OE = I->op_end(); - OI != OE; ++OI) - if (!isa(*OI) && !isa(*OI)) - *OI = RemapOperand(*OI, ValueMap); - + if (Mode == Linker::DestroySource) { + // Splice the body of the source function into the dest function. + Dst->getBasicBlockList().splice(Dst->end(), Src->getBasicBlockList()); + + // At this point, all of the instructions and values of the function are now + // copied over. The only problem is that they are still referencing values in + // the Source function as operands. Loop through all of the operands of the + // functions and patch them up to point to the local versions. + for (Function::iterator BB = Dst->begin(), BE = Dst->end(); BB != BE; ++BB) + for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I) + RemapInstruction(I, ValueMap, RF_IgnoreMissingEntries, &TypeMap); + + } else { + // Clone the body of the function into the dest function. + SmallVector Returns; // Ignore returns. + CloneFunctionInto(Dst, Src, ValueMap, false, Returns, "", NULL, &TypeMap); + } + // There is no need to map the arguments anymore. for (Function::arg_iterator I = Src->arg_begin(), E = Src->arg_end(); I != E; ++I) ValueMap.erase(I); - - return false; + } - -// LinkFunctionBodies - Link in the function bodies that are defined in the -// source module into the DestModule. This consists basically of copying the -// function over and fixing up references to values. -static bool LinkFunctionBodies(Module *Dest, Module *Src, - std::map &ValueMap, - std::string *Err) { - - // Loop over all of the functions in the src module, mapping them over as we - // go - for (Module::iterator SF = Src->begin(), E = Src->end(); SF != E; ++SF) { - if (!SF->isDeclaration()) { // No body if function is external - Function *DF = dyn_cast(ValueMap[SF]); // Destination function - - // DF not external SF external? - if (DF && DF->isDeclaration()) - // Only provide the function body if there isn't one already. - if (LinkFunctionBody(DF, SF, ValueMap, Err)) - return true; +/// linkAliasBodies - Insert all of the aliases in Src into the Dest module. +void ModuleLinker::linkAliasBodies() { + for (Module::alias_iterator I = SrcM->alias_begin(), E = SrcM->alias_end(); + I != E; ++I) { + if (DoNotLinkFromSource.count(I)) + continue; + if (Constant *Aliasee = I->getAliasee()) { + GlobalAlias *DA = cast(ValueMap[I]); + DA->setAliasee(MapValue(Aliasee, ValueMap, RF_None, &TypeMap)); } } - return false; } -// LinkAppendingVars - If there were any appending global variables, link them -// together now. Return true on error. -static bool LinkAppendingVars(Module *M, - std::multimap &AppendingVars, - std::string *ErrorMsg) { - if (AppendingVars.empty()) return false; // Nothing to do. - - // Loop over the multimap of appending vars, processing any variables with the - // same name, forming a new appending global variable with both of the - // initializers merged together, then rewrite references to the old variables - // and delete them. - std::vector Inits; - while (AppendingVars.size() > 1) { - // Get the first two elements in the map... - std::multimap::iterator Second = AppendingVars.begin(), First=Second++; - - // If the first two elements are for different names, there is no pair... - // Otherwise there is a pair, so link them together... - if (First->first == Second->first) { - GlobalVariable *G1 = First->second, *G2 = Second->second; - const ArrayType *T1 = cast(G1->getType()->getElementType()); - const ArrayType *T2 = cast(G2->getType()->getElementType()); - - // Check to see that they two arrays agree on type... - if (T1->getElementType() != T2->getElementType()) - return Error(ErrorMsg, - "Appending variables with different element types need to be linked!"); - if (G1->isConstant() != G2->isConstant()) - return Error(ErrorMsg, - "Appending variables linked with different const'ness!"); - - if (G1->getAlignment() != G2->getAlignment()) - return Error(ErrorMsg, - "Appending variables with different alignment need to be linked!"); - - if (G1->getVisibility() != G2->getVisibility()) - return Error(ErrorMsg, - "Appending variables with different visibility need to be linked!"); - - if (G1->getSection() != G2->getSection()) - return Error(ErrorMsg, - "Appending variables with different section name need to be linked!"); - - unsigned NewSize = T1->getNumElements() + T2->getNumElements(); - ArrayType *NewType = ArrayType::get(T1->getElementType(), - NewSize); - - G1->setName(""); // Clear G1's name in case of a conflict! - - // Create the new global variable... - GlobalVariable *NG = - new GlobalVariable(*M, NewType, G1->isConstant(), G1->getLinkage(), - /*init*/0, First->first, 0, G1->isThreadLocal(), - G1->getType()->getAddressSpace()); - - // Propagate alignment, visibility and section info. - CopyGVAttributes(NG, G1); - - // Merge the initializer... - Inits.reserve(NewSize); - if (ConstantArray *I = dyn_cast(G1->getInitializer())) { - for (unsigned i = 0, e = T1->getNumElements(); i != e; ++i) - Inits.push_back(I->getOperand(i)); - } else { - assert(isa(G1->getInitializer())); - Constant *CV = Constant::getNullValue(T1->getElementType()); - for (unsigned i = 0, e = T1->getNumElements(); i != e; ++i) - Inits.push_back(CV); - } - if (ConstantArray *I = dyn_cast(G2->getInitializer())) { - for (unsigned i = 0, e = T2->getNumElements(); i != e; ++i) - Inits.push_back(I->getOperand(i)); - } else { - assert(isa(G2->getInitializer())); - Constant *CV = Constant::getNullValue(T2->getElementType()); - for (unsigned i = 0, e = T2->getNumElements(); i != e; ++i) - Inits.push_back(CV); - } - NG->setInitializer(ConstantArray::get(NewType, Inits)); - Inits.clear(); - - // Replace any uses of the two global variables with uses of the new - // global... - - // FIXME: This should rewrite simple/straight-forward uses such as - // getelementptr instructions to not use the Cast! - G1->replaceAllUsesWith(ConstantExpr::getBitCast(NG, - G1->getType())); - G2->replaceAllUsesWith(ConstantExpr::getBitCast(NG, - G2->getType())); - - // Remove the two globals from the module now... - M->getGlobalList().erase(G1); - M->getGlobalList().erase(G2); - - // Put the new global into the AppendingVars map so that we can handle - // linking of more than two vars... - Second->second = NG; - } - AppendingVars.erase(First); +/// linkNamedMDNodes - Insert all of the named MDNodes in Src into the Dest +/// module. +void ModuleLinker::linkNamedMDNodes() { + const NamedMDNode *SrcModFlags = SrcM->getModuleFlagsMetadata(); + for (Module::const_named_metadata_iterator I = SrcM->named_metadata_begin(), + E = SrcM->named_metadata_end(); I != E; ++I) { + // Don't link module flags here. Do them separately. + if (&*I == SrcModFlags) continue; + NamedMDNode *DestNMD = DstM->getOrInsertNamedMetadata(I->getName()); + // Add Src elements into Dest node. + for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) + DestNMD->addOperand(MapValue(I->getOperand(i), ValueMap, + RF_None, &TypeMap)); } - - return false; } -static bool ResolveAliases(Module *Dest) { - for (Module::alias_iterator I = Dest->alias_begin(), E = Dest->alias_end(); - I != E; ++I) - // We can't sue resolveGlobalAlias here because we need to preserve - // bitcasts and GEPs. - if (const Constant *C = I->getAliasee()) { - while (dyn_cast(C)) - C = cast(C)->getAliasee(); - const GlobalValue *GV = dyn_cast(C); - if (C != I && !(GV && GV->isDeclaration())) - I->replaceAllUsesWith(const_cast(C)); +/// categorizeModuleFlagNodes - Categorize the module flags according to their +/// type: Error, Warning, Override, and Require. +bool ModuleLinker:: +categorizeModuleFlagNodes(const NamedMDNode *ModFlags, + DenseMap &ErrorNode, + DenseMap &WarningNode, + DenseMap &OverrideNode, + DenseMap > &RequireNodes, + SmallSetVector &SeenIDs) { + bool HasErr = false; + + for (unsigned I = 0, E = ModFlags->getNumOperands(); I != E; ++I) { + MDNode *Op = ModFlags->getOperand(I); + assert(Op->getNumOperands() == 3 && "Invalid module flag metadata!"); + assert(isa(Op->getOperand(0)) && + "Module flag's first operand must be an integer!"); + assert(isa(Op->getOperand(1)) && + "Module flag's second operand must be an MDString!"); + + ConstantInt *Behavior = cast(Op->getOperand(0)); + MDString *ID = cast(Op->getOperand(1)); + Value *Val = Op->getOperand(2); + switch (Behavior->getZExtValue()) { + default: + assert(false && "Invalid behavior in module flag metadata!"); + break; + case Module::Error: { + MDNode *&ErrNode = ErrorNode[ID]; + if (!ErrNode) ErrNode = Op; + if (ErrNode->getOperand(2) != Val) + HasErr = emitError("linking module flags '" + ID->getString() + + "': IDs have conflicting values"); + break; + } + case Module::Warning: { + MDNode *&WarnNode = WarningNode[ID]; + if (!WarnNode) WarnNode = Op; + if (WarnNode->getOperand(2) != Val) + errs() << "WARNING: linking module flags '" << ID->getString() + << "': IDs have conflicting values"; + break; + } + case Module::Require: RequireNodes[ID].insert(Op); break; + case Module::Override: { + MDNode *&OvrNode = OverrideNode[ID]; + if (!OvrNode) OvrNode = Op; + if (OvrNode->getOperand(2) != Val) + HasErr = emitError("linking module flags '" + ID->getString() + + "': IDs have conflicting override values"); + break; + } } - return false; + SeenIDs.insert(ID); + } + + return HasErr; } -// LinkModules - This function links two modules together, with the resulting -// left module modified to be the composite of the two input modules. If an -// error occurs, true is returned and ErrorMsg (if not null) is set to indicate -// the problem. Upon failure, the Dest module could be in a modified state, and -// shouldn't be relied on to be consistent. -bool -Linker::LinkModules(Module *Dest, Module *Src, std::string *ErrorMsg) { - assert(Dest != 0 && "Invalid Destination module"); - assert(Src != 0 && "Invalid Source Module"); - - if (Dest->getDataLayout().empty()) { - if (!Src->getDataLayout().empty()) { - Dest->setDataLayout(Src->getDataLayout()); - } else { - std::string DataLayout; +/// linkModuleFlagsMetadata - Merge the linker flags in Src into the Dest +/// module. +bool ModuleLinker::linkModuleFlagsMetadata() { + const NamedMDNode *SrcModFlags = SrcM->getModuleFlagsMetadata(); + if (!SrcModFlags) return false; - if (Dest->getEndianness() == Module::AnyEndianness) { - if (Src->getEndianness() == Module::BigEndian) - DataLayout.append("E"); - else if (Src->getEndianness() == Module::LittleEndian) - DataLayout.append("e"); - } + NamedMDNode *DstModFlags = DstM->getOrInsertModuleFlagsMetadata(); + + // If the destination module doesn't have module flags yet, then just copy + // over the source module's flags. + if (DstModFlags->getNumOperands() == 0) { + for (unsigned I = 0, E = SrcModFlags->getNumOperands(); I != E; ++I) + DstModFlags->addOperand(SrcModFlags->getOperand(I)); + + return false; + } + + bool HasErr = false; + + // Otherwise, we have to merge them based on their behaviors. First, + // categorize all of the nodes in the modules' module flags. If an error or + // warning occurs, then emit the appropriate message(s). + DenseMap ErrorNode; + DenseMap WarningNode; + DenseMap OverrideNode; + DenseMap > RequireNodes; + SmallSetVector SeenIDs; + + HasErr |= categorizeModuleFlagNodes(SrcModFlags, ErrorNode, WarningNode, + OverrideNode, RequireNodes, SeenIDs); + HasErr |= categorizeModuleFlagNodes(DstModFlags, ErrorNode, WarningNode, + OverrideNode, RequireNodes, SeenIDs); + + // Check that there isn't both an error and warning node for a flag. + for (SmallSetVector::iterator + I = SeenIDs.begin(), E = SeenIDs.end(); I != E; ++I) { + MDString *ID = *I; + if (ErrorNode[ID] && WarningNode[ID]) + HasErr = emitError("linking module flags '" + ID->getString() + + "': IDs have conflicting behaviors"); + } + + // Early exit if we had an error. + if (HasErr) return true; + + // Get the destination's module flags ready for new operands. + DstModFlags->dropAllReferences(); + + // Add all of the module flags to the destination module. + DenseMap > AddedNodes; + for (SmallSetVector::iterator + I = SeenIDs.begin(), E = SeenIDs.end(); I != E; ++I) { + MDString *ID = *I; + if (OverrideNode[ID]) { + DstModFlags->addOperand(OverrideNode[ID]); + AddedNodes[ID].push_back(OverrideNode[ID]); + } else if (ErrorNode[ID]) { + DstModFlags->addOperand(ErrorNode[ID]); + AddedNodes[ID].push_back(ErrorNode[ID]); + } else if (WarningNode[ID]) { + DstModFlags->addOperand(WarningNode[ID]); + AddedNodes[ID].push_back(WarningNode[ID]); + } + + for (SmallSetVector::iterator + II = RequireNodes[ID].begin(), IE = RequireNodes[ID].end(); + II != IE; ++II) + DstModFlags->addOperand(*II); + } - if (Dest->getPointerSize() == Module::AnyPointerSize) { - if (Src->getPointerSize() == Module::Pointer64) - DataLayout.append(DataLayout.length() == 0 ? "p:64:64" : "-p:64:64"); - else if (Src->getPointerSize() == Module::Pointer32) - DataLayout.append(DataLayout.length() == 0 ? "p:32:32" : "-p:32:32"); + // Now check that all of the requirements have been satisfied. + for (SmallSetVector::iterator + I = SeenIDs.begin(), E = SeenIDs.end(); I != E; ++I) { + MDString *ID = *I; + SmallSetVector &Set = RequireNodes[ID]; + + for (SmallSetVector::iterator + II = Set.begin(), IE = Set.end(); II != IE; ++II) { + MDNode *Node = *II; + assert(isa(Node->getOperand(2)) && + "Module flag's third operand must be an MDNode!"); + MDNode *Val = cast(Node->getOperand(2)); + + MDString *ReqID = cast(Val->getOperand(0)); + Value *ReqVal = Val->getOperand(1); + + bool HasValue = false; + for (SmallVectorImpl::iterator + RI = AddedNodes[ReqID].begin(), RE = AddedNodes[ReqID].end(); + RI != RE; ++RI) { + MDNode *ReqNode = *RI; + if (ReqNode->getOperand(2) == ReqVal) { + HasValue = true; + break; + } } - Dest->setDataLayout(DataLayout); + + if (!HasValue) + HasErr = emitError("linking module flags '" + ReqID->getString() + + "': does not have the required value"); } } + return HasErr; +} + +bool ModuleLinker::run() { + assert(DstM && "Null destination module"); + assert(SrcM && "Null source module"); + + // Inherit the target data from the source module if the destination module + // doesn't have one already. + if (DstM->getDataLayout().empty() && !SrcM->getDataLayout().empty()) + DstM->setDataLayout(SrcM->getDataLayout()); + // Copy the target triple from the source to dest if the dest's is empty. - if (Dest->getTargetTriple().empty() && !Src->getTargetTriple().empty()) - Dest->setTargetTriple(Src->getTargetTriple()); + if (DstM->getTargetTriple().empty() && !SrcM->getTargetTriple().empty()) + DstM->setTargetTriple(SrcM->getTargetTriple()); - if (!Src->getDataLayout().empty() && !Dest->getDataLayout().empty() && - Src->getDataLayout() != Dest->getDataLayout()) + if (!SrcM->getDataLayout().empty() && !DstM->getDataLayout().empty() && + SrcM->getDataLayout() != DstM->getDataLayout()) errs() << "WARNING: Linking two modules of different data layouts!\n"; - if (!Src->getTargetTriple().empty() && - Dest->getTargetTriple() != Src->getTargetTriple()) - errs() << "WARNING: Linking two modules of different target triples!\n"; + if (!SrcM->getTargetTriple().empty() && + DstM->getTargetTriple() != SrcM->getTargetTriple()) { + errs() << "WARNING: Linking two modules of different target triples: "; + if (!SrcM->getModuleIdentifier().empty()) + errs() << SrcM->getModuleIdentifier() << ": "; + errs() << "'" << SrcM->getTargetTriple() << "' and '" + << DstM->getTargetTriple() << "'\n"; + } // Append the module inline asm string. - if (!Src->getModuleInlineAsm().empty()) { - if (Dest->getModuleInlineAsm().empty()) - Dest->setModuleInlineAsm(Src->getModuleInlineAsm()); + if (!SrcM->getModuleInlineAsm().empty()) { + if (DstM->getModuleInlineAsm().empty()) + DstM->setModuleInlineAsm(SrcM->getModuleInlineAsm()); else - Dest->setModuleInlineAsm(Dest->getModuleInlineAsm()+"\n"+ - Src->getModuleInlineAsm()); + DstM->setModuleInlineAsm(DstM->getModuleInlineAsm()+"\n"+ + SrcM->getModuleInlineAsm()); } // Update the destination module's dependent libraries list with the libraries // from the source module. There's no opportunity for duplicates here as the // Module ensures that duplicate insertions are discarded. - for (Module::lib_iterator SI = Src->lib_begin(), SE = Src->lib_end(); + for (Module::lib_iterator SI = SrcM->lib_begin(), SE = SrcM->lib_end(); SI != SE; ++SI) - Dest->addLibrary(*SI); - - // LinkTypes - Go through the symbol table of the Src module and see if any - // types are named in the src module that are not named in the Dst module. - // Make sure there are no type name conflicts. - if (LinkTypes(Dest, Src, ErrorMsg)) - return true; - - // ValueMap - Mapping of values from what they used to be in Src, to what they - // are now in Dest. - std::map ValueMap; - - // AppendingVars - Keep track of global variables in the destination module - // with appending linkage. After the module is linked together, they are - // appended and the module is rewritten. - std::multimap AppendingVars; - for (Module::global_iterator I = Dest->global_begin(), E = Dest->global_end(); - I != E; ++I) { - // Add all of the appending globals already in the Dest module to - // AppendingVars. - if (I->hasAppendingLinkage()) - AppendingVars.insert(std::make_pair(I->getName(), I)); - } - - // Insert all of the named mdnoes in Src into the Dest module. - LinkNamedMDNodes(Dest, Src); - - // Insert all of the globals in src into the Dest module... without linking + DstM->addLibrary(*SI); + + // If the source library's module id is in the dependent library list of the + // destination library, remove it since that module is now linked in. + StringRef ModuleId = SrcM->getModuleIdentifier(); + if (!ModuleId.empty()) + DstM->removeLibrary(sys::path::stem(ModuleId)); + + // Loop over all of the linked values to compute type mappings. + computeTypeMapping(); + + // Insert all of the globals in src into the DstM module... without linking // initializers (which could refer to functions not yet mapped over). - if (LinkGlobals(Dest, Src, ValueMap, AppendingVars, ErrorMsg)) - return true; + for (Module::global_iterator I = SrcM->global_begin(), + E = SrcM->global_end(); I != E; ++I) + if (linkGlobalProto(I)) + return true; // Link the functions together between the two modules, without doing function - // bodies... this just adds external function prototypes to the Dest + // bodies... this just adds external function prototypes to the DstM // function... We do this so that when we begin processing function bodies, // all of the global values that may be referenced are available in our // ValueMap. - if (LinkFunctionProtos(Dest, Src, ValueMap, ErrorMsg)) - return true; + for (Module::iterator I = SrcM->begin(), E = SrcM->end(); I != E; ++I) + if (linkFunctionProto(I)) + return true; - // If there were any alias, link them now. We really need to do this now, - // because all of the aliases that may be referenced need to be available in - // ValueMap - if (LinkAlias(Dest, Src, ValueMap, ErrorMsg)) return true; + // If there were any aliases, link them now. + for (Module::alias_iterator I = SrcM->alias_begin(), + E = SrcM->alias_end(); I != E; ++I) + if (linkAliasProto(I)) + return true; - // Update the initializers in the Dest module now that all globals that may - // be referenced are in Dest. - if (LinkGlobalInits(Dest, Src, ValueMap, ErrorMsg)) return true; + for (unsigned i = 0, e = AppendingVars.size(); i != e; ++i) + linkAppendingVarInit(AppendingVars[i]); + + // Update the initializers in the DstM module now that all globals that may + // be referenced are in DstM. + linkGlobalInits(); + + // Link in the function bodies that are defined in the source module into + // DstM. + for (Module::iterator SF = SrcM->begin(), E = SrcM->end(); SF != E; ++SF) { + // Skip if not linking from source. + if (DoNotLinkFromSource.count(SF)) continue; + + // Skip if no body (function is external) or materialize. + if (SF->isDeclaration()) { + if (!SF->isMaterializable()) + continue; + if (SF->Materialize(&ErrorMsg)) + return true; + } + + linkFunctionBody(cast(ValueMap[SF]), SF); + SF->Dematerialize(); + } - // Link in the function bodies that are defined in the source module into the - // DestModule. This consists basically of copying the function over and - // fixing up references to values. - if (LinkFunctionBodies(Dest, Src, ValueMap, ErrorMsg)) return true; + // Resolve all uses of aliases with aliasees. + linkAliasBodies(); - // If there were any appending global variables, link them together now. - if (LinkAppendingVars(Dest, AppendingVars, ErrorMsg)) return true; + // 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(); - // Resolve all uses of aliases with aliasees - if (ResolveAliases(Dest)) return true; + // Merge the module flags into the DstM module. + if (linkModuleFlagsMetadata()) + return true; - // If the source library's module id is in the dependent library list of the - // destination library, remove it since that module is now linked in. - sys::Path modId; - modId.set(Src->getModuleIdentifier()); - if (!modId.isEmpty()) - Dest->removeLibrary(modId.getBasename()); + // Process vector of lazily linked in functions. + bool LinkedInAnyFunctions; + do { + LinkedInAnyFunctions = false; + + for(std::vector::iterator I = LazilyLinkFunctions.begin(), + E = LazilyLinkFunctions.end(); I != E; ++I) { + if (!*I) + continue; + + Function *SF = *I; + Function *DF = cast(ValueMap[SF]); + + if (!DF->use_empty()) { + + // Materialize if necessary. + if (SF->isDeclaration()) { + if (!SF->isMaterializable()) + continue; + if (SF->Materialize(&ErrorMsg)) + return true; + } + + // Link in function body. + linkFunctionBody(DF, SF); + SF->Dematerialize(); + + // "Remove" from vector by setting the element to 0. + *I = 0; + + // Set flag to indicate we may have more functions to lazily link in + // since we linked in a function. + LinkedInAnyFunctions = true; + } + } + } while (LinkedInAnyFunctions); + + // Remove any prototypes of functions that were not actually linked in. + for(std::vector::iterator I = LazilyLinkFunctions.begin(), + E = LazilyLinkFunctions.end(); I != E; ++I) { + if (!*I) + continue; + + Function *SF = *I; + Function *DF = cast(ValueMap[SF]); + if (DF->use_empty()) + DF->eraseFromParent(); + } + + // Now that all of the types from the source are used, resolve any structs + // copied over to the dest that didn't exist there. + TypeMap.linkDefinedTypeBodies(); + + return false; +} + +//===----------------------------------------------------------------------===// +// LinkModules entrypoint. +//===----------------------------------------------------------------------===// + +/// LinkModules - This function links two modules together, with the resulting +/// left module modified to be the composite of the two input modules. If an +/// error occurs, true is returned and ErrorMsg (if not null) is set to indicate +/// the problem. Upon failure, the Dest module could be in a modified state, +/// and shouldn't be relied on to be consistent. +bool Linker::LinkModules(Module *Dest, Module *Src, unsigned Mode, + std::string *ErrorMsg) { + ModuleLinker TheLinker(Dest, Src, Mode); + if (TheLinker.run()) { + if (ErrorMsg) *ErrorMsg = TheLinker.ErrorMsg; + return true; + } return false; } -// vim: sw=2 +//===----------------------------------------------------------------------===// +// C API. +//===----------------------------------------------------------------------===// + +LLVMBool LLVMLinkModules(LLVMModuleRef Dest, LLVMModuleRef Src, + LLVMLinkerMode Mode, char **OutMessages) { + std::string Messages; + LLVMBool Result = Linker::LinkModules(unwrap(Dest), unwrap(Src), + Mode, OutMessages? &Messages : 0); + if (OutMessages) + *OutMessages = strdup(Messages.c_str()); + return Result; +}