X-Git-Url: http://plrg.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FLinker%2FLinkModules.cpp;h=9585fba6e4f1d6d40d195d09cec1669158432fee;hb=ac844bf562ab6adee5160e0ad1fbbfa67ad5c393;hp=6c025d685f3738d8c633eb022f14decd7cfe8ed4;hpb=f97b79e09e1cbefe1934f8f9fa43ed77cd9b363a;p=oota-llvm.git diff --git a/lib/Linker/LinkModules.cpp b/lib/Linker/LinkModules.cpp index 6c025d685f3..9585fba6e4f 100644 --- a/lib/Linker/LinkModules.cpp +++ b/lib/Linker/LinkModules.cpp @@ -13,10 +13,14 @@ #include "llvm/Linker/Linker.h" #include "llvm-c/Linker.h" +#include "llvm/ADT/Hashing.h" #include "llvm/ADT/Optional.h" #include "llvm/ADT/SetVector.h" #include "llvm/ADT/SmallString.h" +#include "llvm/ADT/Statistic.h" +#include "llvm/ADT/Triple.h" #include "llvm/IR/Constants.h" +#include "llvm/IR/DebugInfo.h" #include "llvm/IR/DiagnosticInfo.h" #include "llvm/IR/DiagnosticPrinter.h" #include "llvm/IR/LLVMContext.h" @@ -36,8 +40,6 @@ using namespace llvm; //===----------------------------------------------------------------------===// namespace { -typedef SmallPtrSet TypeSet; - class TypeMapTy : public ValueMapTypeRemapper { /// This is a mapping from a source type to a destination type to use. DenseMap MappedTypes; @@ -58,9 +60,10 @@ class TypeMapTy : public ValueMapTypeRemapper { SmallPtrSet DstResolvedOpaqueTypes; public: - TypeMapTy(TypeSet &Set) : DstStructTypesSet(Set) {} + TypeMapTy(Linker::IdentifiedStructTypeSet &DstStructTypesSet) + : DstStructTypesSet(DstStructTypesSet) {} - TypeSet &DstStructTypesSet; + Linker::IdentifiedStructTypeSet &DstStructTypesSet; /// Indicate that the specified type in the destination module is conceptually /// equivalent to the specified type in the source module. void addTypeMapping(Type *DstTy, Type *SrcTy); @@ -72,6 +75,9 @@ public: /// Return the mapped type to use for the specified input type from the /// source module. Type *get(Type *SrcTy); + Type *get(Type *SrcTy, SmallPtrSet &Visited); + + void finishType(StructType *DTy, StructType *STy, ArrayRef ETypes); FunctionType *get(FunctionType *T) { return cast(get((Type *)T)); @@ -225,124 +231,127 @@ void TypeMapTy::linkDefinedTypeBodies() { DstResolvedOpaqueTypes.clear(); } -Type *TypeMapTy::get(Type *Ty) { -#ifndef NDEBUG - for (auto &Pair : MappedTypes) { - assert(!(Pair.first != Ty && Pair.second == Ty) && - "mapping to a source type"); +void TypeMapTy::finishType(StructType *DTy, StructType *STy, + ArrayRef ETypes) { + DTy->setBody(ETypes, STy->isPacked()); + + // Steal STy's name. + if (STy->hasName()) { + SmallString<16> TmpName = STy->getName(); + STy->setName(""); + DTy->setName(TmpName); } -#endif + DstStructTypesSet.addNonOpaque(DTy); +} + +Type *TypeMapTy::get(Type *Ty) { + SmallPtrSet Visited; + return get(Ty, Visited); +} + +Type *TypeMapTy::get(Type *Ty, SmallPtrSet &Visited) { // If we already have an entry for this type, return it. Type **Entry = &MappedTypes[Ty]; if (*Entry) return *Entry; - // 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; + // These are types that LLVM itself will unique. + bool IsUniqued = !isa(Ty) || cast(Ty)->isLiteral(); - // 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] = get(Ty->getContainedType(I)); - AnyChange |= ElementTypes[I] != Ty->getContainedType(I); +#ifndef NDEBUG + if (!IsUniqued) { + for (auto &Pair : MappedTypes) { + assert(!(Pair.first != Ty && Pair.second == Ty) && + "mapping to a source type"); } + } +#endif - // If we found our type while recursively processing stuff, just use it. - Entry = &MappedTypes[Ty]; - if (*Entry) - return *Entry; + if (!IsUniqued && !Visited.insert(cast(Ty)).second) { + StructType *DTy = StructType::create(Ty->getContext()); + return *Entry = DTy; + } - // If all of the element types mapped directly over, then the type is usable - // as-is. - if (!AnyChange) - return *Entry = Ty; + // If this is not a recursive type, then just map all of the elements and + // then rebuild the type from inside out. + SmallVector ElementTypes; + + // If there are no element types to map, then the type is itself. This is + // true for the anonymous {} struct, things like 'float', integers, etc. + if (Ty->getNumContainedTypes() == 0 && IsUniqued) + return *Entry = Ty; + + // Remap all of the elements, keeping track of whether any of them change. + bool AnyChange = false; + ElementTypes.resize(Ty->getNumContainedTypes()); + for (unsigned I = 0, E = Ty->getNumContainedTypes(); I != E; ++I) { + ElementTypes[I] = get(Ty->getContainedType(I), Visited); + AnyChange |= ElementTypes[I] != Ty->getContainedType(I); + } - // 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()); + // If we found our type while recursively processing stuff, just use it. + Entry = &MappedTypes[Ty]; + if (*Entry) { + if (auto *DTy = dyn_cast(*Entry)) { + if (DTy->isOpaque()) { + auto *STy = cast(Ty); + finishType(DTy, STy, ElementTypes); + } } + return *Entry; } - // 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()) { - // A named structure type from src module is used. Add it to the Set of - // identified structs in the destination module. - DstStructTypesSet.insert(STy); - return *Entry = STy; - } + // If all of the element types mapped directly over and the type is not + // a nomed struct, then the type is usable as-is. + if (!AnyChange && IsUniqued) + return *Entry = Ty; + + // Otherwise, rebuild a modified type. + switch (Ty->getTypeID()) { + default: + llvm_unreachable("unknown derived type to remap"); + case Type::ArrayTyID: + return *Entry = ArrayType::get(ElementTypes[0], + cast(Ty)->getNumElements()); + case Type::VectorTyID: + return *Entry = VectorType::get(ElementTypes[0], + cast(Ty)->getNumElements()); + case Type::PointerTyID: + return *Entry = PointerType::get(ElementTypes[0], + cast(Ty)->getAddressSpace()); + case Type::FunctionTyID: + return *Entry = FunctionType::get(ElementTypes[0], + makeArrayRef(ElementTypes).slice(1), + cast(Ty)->isVarArg()); + case Type::StructTyID: { + auto *STy = cast(Ty); + bool IsPacked = STy->isPacked(); + if (IsUniqued) + return *Entry = StructType::get(Ty->getContext(), ElementTypes, IsPacked); + + // If the type is opaque, we can just use it directly. + if (STy->isOpaque()) { + DstStructTypesSet.addOpaque(STy); + return *Entry = Ty; + } - // Otherwise we create a new type. - StructType *DTy = StructType::create(STy->getContext()); - // A new identified structure type was created. Add it to the set of - // identified structs in the destination module. - DstStructTypesSet.insert(DTy); - *Entry = DTy; + if (StructType *OldT = + DstStructTypesSet.findNonOpaque(ElementTypes, IsPacked)) { + STy->setName(""); + return *Entry = OldT; + } - SmallVector ElementTypes; - ElementTypes.resize(STy->getNumElements()); - for (unsigned I = 0, E = ElementTypes.size(); I != E; ++I) - ElementTypes[I] = get(STy->getElementType(I)); - DTy->setBody(ElementTypes, STy->isPacked()); + if (!AnyChange) { + DstStructTypesSet.addNonOpaque(STy); + return *Entry = Ty; + } - // Steal STy's name. - if (STy->hasName()) { - SmallString<16> TmpName = STy->getName(); - STy->setName(""); - DTy->setName(TmpName); + StructType *DTy = StructType::create(Ty->getContext()); + finishType(DTy, STy, ElementTypes); + return *Entry = DTy; + } } - - return DTy; } //===----------------------------------------------------------------------===// @@ -358,13 +367,13 @@ class ModuleLinker; class ValueMaterializerTy : public ValueMaterializer { TypeMapTy &TypeMap; Module *DstM; - std::vector &LazilyLinkFunctions; + std::vector &LazilyLinkGlobalValues; public: ValueMaterializerTy(TypeMapTy &TypeMap, Module *DstM, - std::vector &LazilyLinkFunctions) + std::vector &LazilyLinkGlobalValues) : ValueMaterializer(), TypeMap(TypeMap), DstM(DstM), - LazilyLinkFunctions(LazilyLinkFunctions) {} + LazilyLinkGlobalValues(LazilyLinkGlobalValues) {} Value *materializeValueFor(Value *V) override; }; @@ -406,16 +415,19 @@ class ModuleLinker { // Set of items not to link in from source. SmallPtrSet DoNotLinkFromSource; - // Vector of functions to lazily link in. - std::vector LazilyLinkFunctions; + // Vector of GlobalValues to lazily link in. + std::vector LazilyLinkGlobalValues; + + /// Functions that have replaced other functions. + SmallPtrSet OverridingFunctions; - Linker::DiagnosticHandlerFunction DiagnosticHandler; + DiagnosticHandlerFunction DiagnosticHandler; public: - ModuleLinker(Module *dstM, TypeSet &Set, Module *srcM, - Linker::DiagnosticHandlerFunction DiagnosticHandler) + ModuleLinker(Module *dstM, Linker::IdentifiedStructTypeSet &Set, Module *srcM, + DiagnosticHandlerFunction DiagnosticHandler) : DstM(dstM), SrcM(srcM), TypeMap(Set), - ValMaterializer(TypeMap, DstM, LazilyLinkFunctions), + ValMaterializer(TypeMap, DstM, LazilyLinkGlobalValues), DiagnosticHandler(DiagnosticHandler) {} bool run(); @@ -477,20 +489,17 @@ private: const GlobalVariable *SrcGV); bool linkGlobalValueProto(GlobalValue *GV); - GlobalValue *linkGlobalVariableProto(const GlobalVariable *SGVar, - GlobalValue *DGV, bool LinkFromSrc); - GlobalValue *linkFunctionProto(const Function *SF, GlobalValue *DGV, - bool LinkFromSrc); - GlobalValue *linkGlobalAliasProto(const GlobalAlias *SGA, GlobalValue *DGV, - bool LinkFromSrc); - bool linkModuleFlagsMetadata(); void linkAppendingVarInit(const AppendingVarInfo &AVI); - void linkGlobalInits(); - void linkFunctionBody(Function *Dst, Function *Src); - void linkAliasBodies(); + + void linkGlobalInit(GlobalVariable &Dst, GlobalVariable &Src); + bool linkFunctionBody(Function &Dst, Function &Src); + void linkAliasBody(GlobalAlias &Dst, GlobalAlias &Src); + bool linkGlobalValueBody(GlobalValue &Src); + void linkNamedMDNodes(); + void stripReplacedSubprograms(); }; } @@ -518,17 +527,7 @@ static void forceRenaming(GlobalValue *GV, StringRef Name) { /// copy additional attributes (those not needed to construct a GlobalValue) /// from the SrcGV to the DestGV. static void copyGVAttributes(GlobalValue *DestGV, const GlobalValue *SrcGV) { - // Use the maximum alignment, rather than just copying the alignment of SrcGV. - auto *DestGO = dyn_cast(DestGV); - unsigned Alignment; - if (DestGO) - Alignment = std::max(DestGO->getAlignment(), SrcGV->getAlignment()); - DestGV->copyAttributesFrom(SrcGV); - - if (DestGO) - DestGO->setAlignment(Alignment); - forceRenaming(DestGV, SrcGV->getName()); } @@ -545,22 +544,71 @@ static bool isLessConstraining(GlobalValue::VisibilityTypes a, return false; } +/// Loop through the global variables in the src module and merge them into the +/// dest module. +static GlobalVariable *copyGlobalVariableProto(TypeMapTy &TypeMap, Module &DstM, + const GlobalVariable *SGVar) { + // No linking to be performed or linking from the source: simply create an + // identical version of the symbol over in the dest module... the + // initializer will be filled in later by LinkGlobalInits. + GlobalVariable *NewDGV = new GlobalVariable( + DstM, TypeMap.get(SGVar->getType()->getElementType()), + SGVar->isConstant(), SGVar->getLinkage(), /*init*/ nullptr, + SGVar->getName(), /*insertbefore*/ nullptr, SGVar->getThreadLocalMode(), + SGVar->getType()->getAddressSpace()); + + return NewDGV; +} + +/// Link the function in the source module into the destination module if +/// needed, setting up mapping information. +static Function *copyFunctionProto(TypeMapTy &TypeMap, Module &DstM, + const Function *SF) { + // If there is no linkage to be performed or we are linking from the source, + // bring SF over. + return Function::Create(TypeMap.get(SF->getFunctionType()), SF->getLinkage(), + SF->getName(), &DstM); +} + +/// Set up prototypes for any aliases that come over from the source module. +static GlobalAlias *copyGlobalAliasProto(TypeMapTy &TypeMap, Module &DstM, + const GlobalAlias *SGA) { + // If there is no linkage to be performed or we're linking from the source, + // bring over SGA. + auto *PTy = cast(TypeMap.get(SGA->getType())); + return GlobalAlias::create(PTy->getElementType(), PTy->getAddressSpace(), + SGA->getLinkage(), SGA->getName(), &DstM); +} + +static GlobalValue *copyGlobalValueProto(TypeMapTy &TypeMap, Module &DstM, + const GlobalValue *SGV) { + GlobalValue *NewGV; + if (auto *SGVar = dyn_cast(SGV)) + NewGV = copyGlobalVariableProto(TypeMap, DstM, SGVar); + else if (auto *SF = dyn_cast(SGV)) + NewGV = copyFunctionProto(TypeMap, DstM, SF); + else + NewGV = copyGlobalAliasProto(TypeMap, DstM, cast(SGV)); + copyGVAttributes(NewGV, SGV); + return NewGV; +} + Value *ValueMaterializerTy::materializeValueFor(Value *V) { - Function *SF = dyn_cast(V); - if (!SF) + auto *SGV = dyn_cast(V); + if (!SGV) return nullptr; - Function *DF = Function::Create(TypeMap.get(SF->getFunctionType()), - SF->getLinkage(), SF->getName(), DstM); - copyGVAttributes(DF, SF); + GlobalValue *DGV = copyGlobalValueProto(TypeMap, *DstM, SGV); - if (Comdat *SC = SF->getComdat()) { - Comdat *DC = DstM->getOrInsertComdat(SC->getName()); - DF->setComdat(DC); + if (Comdat *SC = SGV->getComdat()) { + if (auto *DGO = dyn_cast(DGV)) { + Comdat *DC = DstM->getOrInsertComdat(SC->getName()); + DGO->setComdat(DC); + } } - LazilyLinkFunctions.push_back(SF); - return DF; + LazilyLinkGlobalValues.push_back(SGV); + return DGV; } bool ModuleLinker::getComdatLeader(Module *M, StringRef ComdatName, @@ -791,10 +839,8 @@ void ModuleLinker::computeTypeMapping() { // 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 }". - TypeFinder SrcStructTypes; - SrcStructTypes.run(*SrcM, true); - - for (StructType *ST : SrcStructTypes) { + std::vector Types = SrcM->getIdentifiedStructTypes(); + for (StructType *ST : Types) { if (!ST->hasName()) continue; @@ -827,7 +873,7 @@ void ModuleLinker::computeTypeMapping() { // we prefer to take the '%C' version. So we are then left with both // '%C.1' and '%C' being used for the same types. This leads to some // variables using one type and some using the other. - if (TypeMap.DstStructTypesSet.count(DST)) + if (TypeMap.DstStructTypesSet.hasType(DST)) TypeMap.addTypeMapping(DST, ST); } @@ -1026,118 +1072,53 @@ bool ModuleLinker::linkGlobalValueProto(GlobalValue *SGV) { return false; GlobalValue *NewGV; - if (auto *SGVar = dyn_cast(SGV)) { - NewGV = linkGlobalVariableProto(SGVar, DGV, LinkFromSrc); - if (!NewGV) - return true; - } else if (auto *SF = dyn_cast(SGV)) { - NewGV = linkFunctionProto(SF, DGV, LinkFromSrc); + if (!LinkFromSrc) { + NewGV = DGV; } else { - NewGV = linkGlobalAliasProto(cast(SGV), DGV, LinkFromSrc); - } - - if (NewGV) { - if (NewGV != DGV) - copyGVAttributes(NewGV, SGV); - - NewGV->setUnnamedAddr(HasUnnamedAddr); - NewGV->setVisibility(Visibility); - - if (auto *NewGO = dyn_cast(NewGV)) { - if (C) - NewGO->setComdat(C); + // If the GV is to be lazily linked, don't create it just yet. + // The ValueMaterializerTy will deal with creating it if it's used. + if (!DGV && (SGV->hasLocalLinkage() || SGV->hasLinkOnceLinkage() || + SGV->hasAvailableExternallyLinkage())) { + DoNotLinkFromSource.insert(SGV); + return false; } - // Make sure to remember this mapping. - if (NewGV != DGV) { - if (DGV) { - DGV->replaceAllUsesWith( - ConstantExpr::getBitCast(NewGV, DGV->getType())); - DGV->eraseFromParent(); - } - ValueMap[SGV] = NewGV; - } - } + NewGV = copyGlobalValueProto(TypeMap, *DstM, SGV); - return false; -} + if (DGV && isa(DGV)) + if (auto *NewF = dyn_cast(NewGV)) + OverridingFunctions.insert(NewF); + } -/// Loop through the global variables in the src module and merge them into the -/// dest module. -GlobalValue *ModuleLinker::linkGlobalVariableProto(const GlobalVariable *SGVar, - GlobalValue *DGV, - bool LinkFromSrc) { - unsigned Alignment = 0; - bool ClearConstant = false; + NewGV->setUnnamedAddr(HasUnnamedAddr); + NewGV->setVisibility(Visibility); - if (DGV) { - if (DGV->hasCommonLinkage() && SGVar->hasCommonLinkage()) - Alignment = std::max(SGVar->getAlignment(), DGV->getAlignment()); + if (auto *NewGO = dyn_cast(NewGV)) { + if (C) + NewGO->setComdat(C); - auto *DGVar = dyn_cast(DGV); - if (!SGVar->isConstant() || (DGVar && !DGVar->isConstant())) - ClearConstant = true; + if (DGV && DGV->hasCommonLinkage() && SGV->hasCommonLinkage()) + NewGO->setAlignment(std::max(DGV->getAlignment(), SGV->getAlignment())); } - if (!LinkFromSrc) { - if (auto *NewGVar = dyn_cast(DGV)) { - if (Alignment) - NewGVar->setAlignment(Alignment); - if (NewGVar->isDeclaration() && ClearConstant) - NewGVar->setConstant(false); - } - return DGV; + if (auto *NewGVar = dyn_cast(NewGV)) { + auto *DGVar = dyn_cast_or_null(DGV); + auto *SGVar = dyn_cast(SGV); + if (DGVar && SGVar && DGVar->isDeclaration() && SGVar->isDeclaration() && + (!DGVar->isConstant() || !SGVar->isConstant())) + NewGVar->setConstant(false); } - // No linking to be performed or linking from the source: simply create an - // identical version of the symbol over in the dest module... the - // initializer will be filled in later by LinkGlobalInits. - GlobalVariable *NewDGV = new GlobalVariable( - *DstM, TypeMap.get(SGVar->getType()->getElementType()), - SGVar->isConstant(), SGVar->getLinkage(), /*init*/ nullptr, - SGVar->getName(), /*insertbefore*/ nullptr, SGVar->getThreadLocalMode(), - SGVar->getType()->getAddressSpace()); - - if (Alignment) - NewDGV->setAlignment(Alignment); - - return NewDGV; -} - -/// Link the function in the source module into the destination module if -/// needed, setting up mapping information. -GlobalValue *ModuleLinker::linkFunctionProto(const Function *SF, - GlobalValue *DGV, - bool LinkFromSrc) { - if (!LinkFromSrc) - return DGV; - - // If the function is to be lazily linked, don't create it just yet. - // The ValueMaterializerTy will deal with creating it if it's used. - if (!DGV && (SF->hasLocalLinkage() || SF->hasLinkOnceLinkage() || - SF->hasAvailableExternallyLinkage())) { - DoNotLinkFromSource.insert(SF); - return nullptr; + // Make sure to remember this mapping. + if (NewGV != DGV) { + if (DGV) { + DGV->replaceAllUsesWith(ConstantExpr::getBitCast(NewGV, DGV->getType())); + DGV->eraseFromParent(); + } + ValueMap[SGV] = NewGV; } - // If there is no linkage to be performed or we are linking from the source, - // bring SF over. - return Function::Create(TypeMap.get(SF->getFunctionType()), SF->getLinkage(), - SF->getName(), DstM); -} - -/// Set up prototypes for any aliases that come over from the source module. -GlobalValue *ModuleLinker::linkGlobalAliasProto(const GlobalAlias *SGA, - GlobalValue *DGV, - bool LinkFromSrc) { - if (!LinkFromSrc) - return DGV; - - // If there is no linkage to be performed or we're linking from the source, - // bring over SGA. - auto *PTy = cast(TypeMap.get(SGA->getType())); - return GlobalAlias::create(PTy->getElementType(), PTy->getAddressSpace(), - SGA->getLinkage(), SGA->getName(), DstM); + return false; } static void getArrayElements(const Constant *C, @@ -1182,70 +1163,80 @@ void ModuleLinker::linkAppendingVarInit(const AppendingVarInfo &AVI) { /// 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, &ValMaterializer)); - } +void ModuleLinker::linkGlobalInit(GlobalVariable &Dst, GlobalVariable &Src) { + // Figure out what the initializer looks like in the dest module. + Dst.setInitializer(MapValue(Src.getInitializer(), ValueMap, RF_None, &TypeMap, + &ValMaterializer)); } /// 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()); +bool ModuleLinker::linkFunctionBody(Function &Dst, Function &Src) { + assert(Dst.isDeclaration() && !Src.isDeclaration()); + + // Materialize if needed. + if (std::error_code EC = Src.materialize()) + return emitError(EC.message()); + + // Link in the prefix data. + if (Src.hasPrefixData()) + Dst.setPrefixData(MapValue(Src.getPrefixData(), ValueMap, RF_None, &TypeMap, + &ValMaterializer)); + + // Link in the prologue data. + if (Src.hasPrologueData()) + Dst.setPrologueData(MapValue(Src.getPrologueData(), ValueMap, RF_None, + &TypeMap, &ValMaterializer)); // Go through and convert function arguments over, remembering the mapping. - 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 over. + Function::arg_iterator DI = Dst.arg_begin(); + for (Argument &Arg : Src.args()) { + DI->setName(Arg.getName()); // Copy the name over. // Add a mapping to our mapping. - ValueMap[I] = DI; + ValueMap[&Arg] = DI; + ++DI; } // Splice the body of the source function into the dest function. - Dst->getBasicBlockList().splice(Dst->end(), Src->getBasicBlockList()); + 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, + for (BasicBlock &BB : Dst) + for (Instruction &I : BB) + RemapInstruction(&I, ValueMap, RF_IgnoreMissingEntries, &TypeMap, &ValMaterializer); // 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); + for (Argument &Arg : Src.args()) + ValueMap.erase(&Arg); + Src.Dematerialize(); + return false; } -/// 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]); - Constant *Val = - MapValue(Aliasee, ValueMap, RF_None, &TypeMap, &ValMaterializer); - DA->setAliasee(Val); - } +void ModuleLinker::linkAliasBody(GlobalAlias &Dst, GlobalAlias &Src) { + Constant *Aliasee = Src.getAliasee(); + Constant *Val = + MapValue(Aliasee, ValueMap, RF_None, &TypeMap, &ValMaterializer); + Dst.setAliasee(Val); +} + +bool ModuleLinker::linkGlobalValueBody(GlobalValue &Src) { + Value *Dst = ValueMap[&Src]; + assert(Dst); + if (auto *F = dyn_cast(&Src)) + return linkFunctionBody(cast(*Dst), *F); + if (auto *GVar = dyn_cast(&Src)) { + linkGlobalInit(cast(*Dst), *GVar); + return false; } + linkAliasBody(cast(*Dst), cast(Src)); + return false; } /// Insert all of the named MDNodes in Src into the Dest module. @@ -1258,8 +1249,50 @@ void ModuleLinker::linkNamedMDNodes() { 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, &ValMaterializer)); + DestNMD->addOperand(MapMetadata(I->getOperand(i), ValueMap, RF_None, + &TypeMap, &ValMaterializer)); + } +} + +/// Drop DISubprograms that have been superseded. +/// +/// FIXME: this creates an asymmetric result: we strip losing subprograms from +/// DstM, but leave losing subprograms in SrcM. Instead we should also strip +/// losers from SrcM, but this requires extra plumbing in MapMetadata. +void ModuleLinker::stripReplacedSubprograms() { + // Avoid quadratic runtime by returning early when there's nothing to do. + if (OverridingFunctions.empty()) + return; + + // Move the functions now, so the set gets cleared even on early returns. + auto Functions = std::move(OverridingFunctions); + OverridingFunctions.clear(); + + // Drop subprograms whose functions have been overridden by the new compile + // unit. + NamedMDNode *CompileUnits = DstM->getNamedMetadata("llvm.dbg.cu"); + if (!CompileUnits) + return; + for (unsigned I = 0, E = CompileUnits->getNumOperands(); I != E; ++I) { + DICompileUnit CU(CompileUnits->getOperand(I)); + assert(CU && "Expected valid compile unit"); + + DITypedArray SPs(CU.getSubprograms()); + assert(SPs && "Expected valid subprogram array"); + + SmallVector NewSPs; + NewSPs.reserve(SPs.getNumElements()); + for (unsigned S = 0, SE = SPs.getNumElements(); S != SE; ++S) { + DISubprogram SP = SPs.getElement(S); + if (SP && SP.getFunction() && Functions.count(SP.getFunction())) + continue; + + NewSPs.push_back(SP); + } + + // Redirect operand to the overriding subprogram. + if (NewSPs.size() != SPs.getNumElements()) + CU.replaceSubprograms(DIArray(MDNode::get(DstM->getContext(), NewSPs))); } } @@ -1280,17 +1313,17 @@ bool ModuleLinker::linkModuleFlagsMetadata() { } // First build a map of the existing module flags and requirements. - DenseMap Flags; + DenseMap> Flags; SmallSetVector Requirements; for (unsigned I = 0, E = DstModFlags->getNumOperands(); I != E; ++I) { MDNode *Op = DstModFlags->getOperand(I); - ConstantInt *Behavior = cast(Op->getOperand(0)); + ConstantInt *Behavior = mdconst::extract(Op->getOperand(0)); MDString *ID = cast(Op->getOperand(1)); if (Behavior->getZExtValue() == Module::Require) { Requirements.insert(cast(Op->getOperand(2))); } else { - Flags[ID] = Op; + Flags[ID] = std::make_pair(Op, I); } } @@ -1299,9 +1332,12 @@ bool ModuleLinker::linkModuleFlagsMetadata() { bool HasErr = false; for (unsigned I = 0, E = SrcModFlags->getNumOperands(); I != E; ++I) { MDNode *SrcOp = SrcModFlags->getOperand(I); - ConstantInt *SrcBehavior = cast(SrcOp->getOperand(0)); + ConstantInt *SrcBehavior = + mdconst::extract(SrcOp->getOperand(0)); MDString *ID = cast(SrcOp->getOperand(1)); - MDNode *DstOp = Flags.lookup(ID); + MDNode *DstOp; + unsigned DstIndex; + std::tie(DstOp, DstIndex) = Flags.lookup(ID); unsigned SrcBehaviorValue = SrcBehavior->getZExtValue(); // If this is a requirement, add it and continue. @@ -1316,13 +1352,14 @@ bool ModuleLinker::linkModuleFlagsMetadata() { // If there is no existing flag with this ID, just add it. if (!DstOp) { - Flags[ID] = SrcOp; + Flags[ID] = std::make_pair(SrcOp, DstModFlags->getNumOperands()); DstModFlags->addOperand(SrcOp); continue; } // Otherwise, perform a merge. - ConstantInt *DstBehavior = cast(DstOp->getOperand(0)); + ConstantInt *DstBehavior = + mdconst::extract(DstOp->getOperand(0)); unsigned DstBehaviorValue = DstBehavior->getZExtValue(); // If either flag has override behavior, handle it first. @@ -1336,8 +1373,8 @@ bool ModuleLinker::linkModuleFlagsMetadata() { continue; } else if (SrcBehaviorValue == Module::Override) { // Update the destination flag to that of the source. - DstOp->replaceOperandWith(0, SrcBehavior); - DstOp->replaceOperandWith(2, SrcOp->getOperand(2)); + DstModFlags->setOperand(DstIndex, SrcOp); + Flags[ID].first = SrcOp; continue; } @@ -1348,6 +1385,13 @@ bool ModuleLinker::linkModuleFlagsMetadata() { continue; } + auto replaceDstValue = [&](MDNode *New) { + Metadata *FlagOps[] = {DstOp->getOperand(0), ID, New}; + MDNode *Flag = MDNode::get(DstM->getContext(), FlagOps); + DstModFlags->setOperand(DstIndex, Flag); + Flags[ID].first = Flag; + }; + // Perform the merge for standard behavior types. switch (SrcBehaviorValue) { case Module::Require: @@ -1371,29 +1415,27 @@ bool ModuleLinker::linkModuleFlagsMetadata() { case Module::Append: { MDNode *DstValue = cast(DstOp->getOperand(2)); MDNode *SrcValue = cast(SrcOp->getOperand(2)); - unsigned NumOps = DstValue->getNumOperands() + SrcValue->getNumOperands(); - Value **VP, **Values = VP = new Value*[NumOps]; - for (unsigned i = 0, e = DstValue->getNumOperands(); i != e; ++i, ++VP) - *VP = DstValue->getOperand(i); - for (unsigned i = 0, e = SrcValue->getNumOperands(); i != e; ++i, ++VP) - *VP = SrcValue->getOperand(i); - DstOp->replaceOperandWith(2, MDNode::get(DstM->getContext(), - ArrayRef(Values, - NumOps))); - delete[] Values; + SmallVector MDs; + MDs.reserve(DstValue->getNumOperands() + SrcValue->getNumOperands()); + for (unsigned i = 0, e = DstValue->getNumOperands(); i != e; ++i) + MDs.push_back(DstValue->getOperand(i)); + for (unsigned i = 0, e = SrcValue->getNumOperands(); i != e; ++i) + MDs.push_back(SrcValue->getOperand(i)); + + replaceDstValue(MDNode::get(DstM->getContext(), MDs)); break; } case Module::AppendUnique: { - SmallSetVector Elts; + SmallSetVector Elts; MDNode *DstValue = cast(DstOp->getOperand(2)); MDNode *SrcValue = cast(SrcOp->getOperand(2)); for (unsigned i = 0, e = DstValue->getNumOperands(); i != e; ++i) Elts.insert(DstValue->getOperand(i)); for (unsigned i = 0, e = SrcValue->getNumOperands(); i != e; ++i) Elts.insert(SrcValue->getOperand(i)); - DstOp->replaceOperandWith(2, MDNode::get(DstM->getContext(), - ArrayRef(Elts.begin(), - Elts.end()))); + + replaceDstValue(MDNode::get(DstM->getContext(), + makeArrayRef(Elts.begin(), Elts.end()))); break; } } @@ -1403,9 +1445,9 @@ bool ModuleLinker::linkModuleFlagsMetadata() { for (unsigned I = 0, E = Requirements.size(); I != E; ++I) { MDNode *Requirement = Requirements[I]; MDString *Flag = cast(Requirement->getOperand(0)); - Value *ReqValue = Requirement->getOperand(1); + Metadata *ReqValue = Requirement->getOperand(1); - MDNode *Op = Flags[Flag]; + MDNode *Op = Flags[Flag].first; if (!Op || Op->getOperand(2) != ReqValue) { HasErr |= emitError("linking module flags '" + Flag->getString() + "': does not have the required value"); @@ -1416,6 +1458,28 @@ bool ModuleLinker::linkModuleFlagsMetadata() { return HasErr; } +// This function returns true if the triples match. +static bool triplesMatch(const Triple &T0, const Triple &T1) { + // If vendor is apple, ignore the version number. + if (T0.getVendor() == Triple::Apple) + return T0.getArch() == T1.getArch() && + T0.getSubArch() == T1.getSubArch() && + T0.getVendor() == T1.getVendor() && + T0.getOS() == T1.getOS(); + + return T0 == T1; +} + +// This function returns the merged triple. +static std::string mergeTriples(const Triple &SrcTriple, const Triple &DstTriple) { + // If vendor is apple, pick the triple with the larger version number. + if (SrcTriple.getVendor() == Triple::Apple) + if (DstTriple.isOSVersionLT(SrcTriple)) + return SrcTriple.str(); + + return DstTriple.str(); +} + bool ModuleLinker::run() { assert(DstM && "Null destination module"); assert(SrcM && "Null source module"); @@ -1425,10 +1489,6 @@ bool ModuleLinker::run() { if (!DstM->getDataLayout() && SrcM->getDataLayout()) DstM->setDataLayout(SrcM->getDataLayout()); - // Copy the target triple from the source to dest if the dest's is empty. - if (DstM->getTargetTriple().empty() && !SrcM->getTargetTriple().empty()) - DstM->setTargetTriple(SrcM->getTargetTriple()); - if (SrcM->getDataLayout() && DstM->getDataLayout() && *SrcM->getDataLayout() != *DstM->getDataLayout()) { emitWarning("Linking two modules of different data layouts: '" + @@ -1437,14 +1497,21 @@ bool ModuleLinker::run() { DstM->getModuleIdentifier() + "' is '" + DstM->getDataLayoutStr() + "'\n"); } - if (!SrcM->getTargetTriple().empty() && - DstM->getTargetTriple() != SrcM->getTargetTriple()) { + + // Copy the target triple from the source to dest if the dest's is empty. + if (DstM->getTargetTriple().empty() && !SrcM->getTargetTriple().empty()) + DstM->setTargetTriple(SrcM->getTargetTriple()); + + Triple SrcTriple(SrcM->getTargetTriple()), DstTriple(DstM->getTargetTriple()); + + if (!SrcM->getTargetTriple().empty() && !triplesMatch(SrcTriple, DstTriple)) emitWarning("Linking two modules of different target triples: " + SrcM->getModuleIdentifier() + "' is '" + SrcM->getTargetTriple() + "' whereas '" + DstM->getModuleIdentifier() + "' is '" + DstM->getTargetTriple() + "'\n"); - } + + DstM->setTargetTriple(mergeTriples(SrcTriple, DstTriple)); // Append the module inline asm string. if (!SrcM->getModuleInlineAsm().empty()) { @@ -1498,33 +1565,39 @@ bool ModuleLinker::run() { for (unsigned i = 0, e = AppendingVars.size(); i != e; ++i) linkAppendingVarInit(AppendingVars[i]); + for (const auto &Entry : DstM->getComdatSymbolTable()) { + const Comdat &C = Entry.getValue(); + if (C.getSelectionKind() == Comdat::Any) + continue; + const GlobalValue *GV = SrcM->getNamedValue(C.getName()); + assert(GV); + MapValue(GV, ValueMap, RF_None, &TypeMap, &ValMaterializer); + } + // 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; - - Function *DF = cast(ValueMap[SF]); - if (SF->hasPrefixData()) { - // Link in the prefix data. - DF->setPrefixData(MapValue( - SF->getPrefixData(), ValueMap, RF_None, &TypeMap, &ValMaterializer)); - } - - // Materialize if needed. - if (std::error_code EC = SF->materialize()) - return emitError(EC.message()); - + for (Function &SF : *SrcM) { // Skip if no body (function is external). - if (SF->isDeclaration()) + if (SF.isDeclaration()) + continue; + + // Skip if not linking from source. + if (DoNotLinkFromSource.count(&SF)) continue; - linkFunctionBody(DF, SF); - SF->Dematerialize(); + if (linkGlobalValueBody(SF)) + return true; } // Resolve all uses of aliases with aliasees. - linkAliasBodies(); + for (GlobalAlias &Src : SrcM->aliases()) { + if (DoNotLinkFromSource.count(&Src)) + continue; + linkGlobalValueBody(Src); + } + + // Strip replaced subprograms before linking together compile units. + stripReplacedSubprograms(); // Remap all of the named MDNodes in Src into the DstM module. We do this // after linking GlobalValues so that MDNodes that reference GlobalValues @@ -1537,53 +1610,106 @@ bool ModuleLinker::run() { // Update the initializers in the DstM module now that all globals that may // be referenced are in DstM. - linkGlobalInits(); + for (GlobalVariable &Src : SrcM->globals()) { + // Only process initialized GV's or ones not already in dest. + if (!Src.hasInitializer() || DoNotLinkFromSource.count(&Src)) + continue; + linkGlobalValueBody(Src); + } // 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) { - Function *SF = *I; - if (!SF) - continue; + while (!LazilyLinkGlobalValues.empty()) { + GlobalValue *SGV = LazilyLinkGlobalValues.back(); + LazilyLinkGlobalValues.pop_back(); - Function *DF = cast(ValueMap[SF]); - if (SF->hasPrefixData()) { - // Link in the prefix data. - DF->setPrefixData(MapValue(SF->getPrefixData(), - ValueMap, - RF_None, - &TypeMap, - &ValMaterializer)); - } + assert(!SGV->isDeclaration() && "users should not pass down decls"); + if (linkGlobalValueBody(*SGV)) + return true; + } - // Materialize if needed. - if (std::error_code EC = SF->materialize()) - return emitError(EC.message()); + return false; +} - // Skip if no body (function is external). - if (SF->isDeclaration()) - continue; +Linker::StructTypeKeyInfo::KeyTy::KeyTy(ArrayRef E, bool P) + : ETypes(E), IsPacked(P) {} - // Erase from vector *before* the function body is linked - linkFunctionBody could - // invalidate I. - LazilyLinkFunctions.erase(I); +Linker::StructTypeKeyInfo::KeyTy::KeyTy(const StructType *ST) + : ETypes(ST->elements()), IsPacked(ST->isPacked()) {} - // Link in function body. - linkFunctionBody(DF, SF); - SF->Dematerialize(); +bool Linker::StructTypeKeyInfo::KeyTy::operator==(const KeyTy &That) const { + if (IsPacked != That.IsPacked) + return false; + if (ETypes != That.ETypes) + return false; + return true; +} - // Set flag to indicate we may have more functions to lazily link in - // since we linked in a function. - LinkedInAnyFunctions = true; - break; - } - } while (LinkedInAnyFunctions); +bool Linker::StructTypeKeyInfo::KeyTy::operator!=(const KeyTy &That) const { + return !this->operator==(That); +} - return false; +StructType *Linker::StructTypeKeyInfo::getEmptyKey() { + return DenseMapInfo::getEmptyKey(); +} + +StructType *Linker::StructTypeKeyInfo::getTombstoneKey() { + return DenseMapInfo::getTombstoneKey(); +} + +unsigned Linker::StructTypeKeyInfo::getHashValue(const KeyTy &Key) { + return hash_combine(hash_combine_range(Key.ETypes.begin(), Key.ETypes.end()), + Key.IsPacked); +} + +unsigned Linker::StructTypeKeyInfo::getHashValue(const StructType *ST) { + return getHashValue(KeyTy(ST)); +} + +bool Linker::StructTypeKeyInfo::isEqual(const KeyTy &LHS, + const StructType *RHS) { + if (RHS == getEmptyKey() || RHS == getTombstoneKey()) + return false; + return LHS == KeyTy(RHS); +} + +bool Linker::StructTypeKeyInfo::isEqual(const StructType *LHS, + const StructType *RHS) { + if (RHS == getEmptyKey()) + return LHS == getEmptyKey(); + + if (RHS == getTombstoneKey()) + return LHS == getTombstoneKey(); + + return KeyTy(LHS) == KeyTy(RHS); +} + +void Linker::IdentifiedStructTypeSet::addNonOpaque(StructType *Ty) { + assert(!Ty->isOpaque()); + NonOpaqueStructTypes.insert(Ty); +} + +void Linker::IdentifiedStructTypeSet::addOpaque(StructType *Ty) { + assert(Ty->isOpaque()); + OpaqueStructTypes.insert(Ty); +} + +StructType * +Linker::IdentifiedStructTypeSet::findNonOpaque(ArrayRef ETypes, + bool IsPacked) { + Linker::StructTypeKeyInfo::KeyTy Key(ETypes, IsPacked); + auto I = NonOpaqueStructTypes.find_as(Key); + if (I == NonOpaqueStructTypes.end()) + return nullptr; + return *I; +} + +bool Linker::IdentifiedStructTypeSet::hasType(StructType *Ty) { + if (Ty->isOpaque()) + return OpaqueStructTypes.count(Ty); + auto I = NonOpaqueStructTypes.find(Ty); + if (I == NonOpaqueStructTypes.end()) + return false; + return *I == Ty; } void Linker::init(Module *M, DiagnosticHandlerFunction DiagnosticHandler) { @@ -1592,7 +1718,12 @@ void Linker::init(Module *M, DiagnosticHandlerFunction DiagnosticHandler) { TypeFinder StructTypes; StructTypes.run(*M, true); - IdentifiedStructTypes.insert(StructTypes.begin(), StructTypes.end()); + for (StructType *Ty : StructTypes) { + if (Ty->isOpaque()) + IdentifiedStructTypes.addOpaque(Ty); + else + IdentifiedStructTypes.addNonOpaque(Ty); + } } Linker::Linker(Module *M, DiagnosticHandlerFunction DiagnosticHandler) { @@ -1616,7 +1747,9 @@ void Linker::deleteModule() { bool Linker::linkInModule(Module *Src) { ModuleLinker TheLinker(Composite, IdentifiedStructTypes, Src, DiagnosticHandler); - return TheLinker.run(); + bool RetCode = TheLinker.run(); + Composite->dropTriviallyDeadConstantArrays(); + return RetCode; } //===----------------------------------------------------------------------===// @@ -1644,7 +1777,7 @@ bool Linker::LinkModules(Module *Dest, Module *Src) { //===----------------------------------------------------------------------===// LLVMBool LLVMLinkModules(LLVMModuleRef Dest, LLVMModuleRef Src, - LLVMLinkerMode Mode, char **OutMessages) { + unsigned Unused, char **OutMessages) { Module *D = unwrap(Dest); std::string Message; raw_string_ostream Stream(Message);