Support for ThinLTO function importing and symbol linking.

[tejohnson]

Summary

Support for necessary linkage changes and symbol renaming during
ThinLTO function importing.

Also includes llvm-link support for manually importing functions
and associated llvm-link based tests.

Note that this does not include support for intelligently importing
metadata, which is currently imported duplicate times. That support will
be in the follow-on patch, and currently is ignored by the tests.

[tejohnson]

Ping.
Thanks,
Teresa

[davidxl]

include/llvm/IR/FunctionInfo.h

235

comdat functions should have been merged at this point, right?

lib/Linker/LinkModules.cpp

441

Do we want to support importing a list of functions in a batch mode?

450

Is it better to add another flag to indicate whether it is for primary module or imported modules.

466

assert dstM == srcM?

573

How about just getName or getLinkName? The name may or may not be new.

676

Unless the address taken bit is set (to handle the case address is used in comparison).

688

For non promoted constant local globals, the name also needs to be uniqued -- otherwise there may be conflicts in the assembly file (some constants are materialized using memory read).

719

isPerformingImport() check is redundant here.

745

I think it should stay as WeakODRLinkage -- there is no guarantee that it will be emitted in other modules (if there is no reference).

771

The comment seems to mean the opposite.

[tejohnson]

Thanks for the comments. Responses below.

include/llvm/IR/FunctionInfo.h

235

Merged where? At the point we invoke this we are just getting a module ready to import from - we would not have merged comdats across modules. And the FunctionInfoList may contain multiple copies of a comdat, unless we decided in the stage-2 global indexing step to merge their entries in the combined index. Presumably due to the way this is being used to guard renaming and promotion of locals, we couldn't remove an entry for any comdat copy that accessed a local - but I doubt that could happen?

lib/Linker/LinkModules.cpp

441

Good question, and something I was pondering. For now I think we should support one at a time, and I can extend this to support importing a set in the future if it seems useful. Part of the issue is that until you import a function, you won't know what other function in the same module are called by it that you may now want to import. But presumably you may have multiple functions from the same source module called from the current primary module that you know you want to import, and batch mode could help there. But I would prefer to add that handling as a follow-on if it is necessary.

450

That is implied by whether FuncToImport is null or not. A flag would be redundant.

466

That should never be the case - the dstM is the "combined" module being created by the link. Even if we don't import, the module being linked is not the same as the srcM.

573

Ok, I can change to getName. I will accordingly change getNewLinkage below to getLinkage.

676

Good catch. And unfortunately for GVars there is no address taken bit or easy check for this. For functions, there is a hasAddressTaken function that walks the uses, but not for GVars. But that isn't useful here in the importing context in any case, since we haven't even parsed all of the functions where it may be address taken. It will have to be recorded in the function index in the front end. For now I will simply promote all locals (and leave a comment about what would need to be done to reduce this scope).

688

Good point. Since I will now promote all locals, this is somewhat moot, but will resurface if we do something more narrow to identify the address taken locals that need promotion. I will change this to rename all locals when we are importing or exporting.

719

Will remove.

745

WeakODR is not discardable, unlike LinkOnceODR. So it will be emitted in the original module.

771

You're right, will fix the comment.

[davidxl]

include/llvm/IR/FunctionInfo.h

235

yes, In theory, the combined index generator can do comdat de-duplication before writing out the combined summary -- but since that step is mandatory, handling duplication is still required here.

lib/Linker/LinkModules.cpp

441

ok.

450

ok.

466

sounds like this can be done in place (making dstM the same as srcM), but as it is existing code, ok.

676

When promoting Constant Global, make sure the initializer is made available so that const prop can still happen.

688

ok.

745

ok. I got it confused with LinkOnceODR.

[tobiasvk]

lib/Linker/LinkModules.cpp

676

Could you use the unnamed_addr bit?

[tejohnson]

lib/Linker/LinkModules.cpp

676

Thanks for the tip!

I am not familiar with this attribute, but see it documented at http://llvm.org/docs/LangRef.html#global-variables:

"Global variables can be marked with unnamed_addr which indicates that the address is not significant, only the content. Constants marked like this can be merged with other constants if they have the same initializer. Note that a constant with significant address can be merged with a unnamed_addr constant, the result being a constant whose address is significant."

So it sounds like unnamed_addr variables are safe to be treated as non-address taken. I'm not sure under what context it is set though or if it covers all non-address taken const variables?

[tejohnson]

include/llvm/IR/FunctionInfo.h

235

Right, that is what this loop is handling.

lib/Linker/LinkModules.cpp

676

When promoting Constant Global, make sure the initializer is made available so that const prop can still happen.

Right, the initializers are imported along with the variables.

[tobiasvk]

lib/Linker/LinkModules.cpp

676

So it sounds like unnamed_addr variables are safe to be treated as non-address taken. I'm not sure under what context it is set though or if it covers all non-address taken const variables?

It's set by the GlobalOpt pass (mainly). GlobalOpt uses a helper, GlobalStatus, that does the analysis. I think you're safe to assume that it's non-address taken *if* the flag is set, but obviously GlobalOpt may not have been run.

[tejohnson]

lib/Linker/LinkModules.cpp

676

Great, I will guard the handling of constant variables by hasUnnamedAddr as well so we can catch those cases where GlobalOpt has been run and set it.
Thanks.

[tejohnson]

Updated as per review comments from davidxl, dexonsmith and tobiasvk,
including updating test to use new llvm-lto functionality instead of gold.

[tejohnson]

• Handle aliases that were already imported from the same module, and add a test for that scenario.

[tejohnson]

Also, pinging the patch review. Duncan, did you have any more comments?

Thanks,
Teresa

include/llvm/IR/FunctionInfo.h

public:
   }
 
   /// Add a new module path, mapped to the given module Id, and return StringRef
   /// owned by string table map.
   StringRef addModulePath(StringRef ModPath, uint64_t ModId) {
     return ModulePathStringTable.insert(std::make_pair(ModPath, ModId))
         .first->first();
   }
+
+  /// Check if the given Module has any functions available for exporting
+  /// in the index.
+  bool hasExportedFunctions(const Module *M) {
+    for (const Function &SF : *M) {
+      std::string FuncName = SF.getName().str();
+      if (SF.hasLocalLinkage())
+        FuncName = getGlobalNameForLocal(SF.getName(),
+                                         getModuleId(M->getModuleIdentifier()));
+      for (const auto &FI : getFunctionInfoList(FuncName)) {
+        if (M->getModuleIdentifier() == FI->functionSummary()->modulePath())
+          return true;
+      }
+    }
+    return false;
+  }
 };
 
 }  // End llvm namespace
 
 #endif

include/llvm/Linker/Linker.h

 //===- Linker.h - Module Linker Interface -----------------------*- C++ -*-===//
 //
 //                     The LLVM Compiler Infrastructure
 //
 // This file is distributed under the University of Illinois Open Source
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_LINKER_LINKER_H
 #define LLVM_LINKER_LINKER_H
 
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/IR/DiagnosticInfo.h"
+#include "llvm/IR/FunctionInfo.h"
 
 namespace llvm {
 class Module;
 class StructType;
 class Type;
 
 /// This class provides the core functionality of linking in LLVM. It keeps a
 /// pointer to the merged module so far. It doesn't take ownership of the
public:
   Linker(Module *M);
 
   Module *getModule() const { return Composite; }
   void deleteModule();
 
   /// \brief Link \p Src into the composite. The source is destroyed.
   /// Passing OverrideSymbols as true will have symbols from Src
   /// shadow those in the Dest.
+  /// For ThinLTO function importing/exporting the \p FunctionInfoIndex
+  /// is passed. If a \p FuncToImport is provided, only that single
+  /// function is imported from the source module.
   /// Returns true on error.
-  bool linkInModule(Module *Src, unsigned Flags = Flags::None);
+  bool linkInModule(Module *Src, unsigned Flags = Flags::None,
+                    FunctionInfoIndex *Index = nullptr,
+                    Function *FuncToImport = nullptr);
 
   /// \brief Set the composite to the passed-in module.
   void setModule(Module *Dst);
 
   static bool LinkModules(Module *Dest, Module *Src,
                           DiagnosticHandlerFunction DiagnosticHandler,
                           unsigned Flags = Flags::None);
 

lib/Linker/LinkModules.cpp

 
 /// Creates prototypes for functions that are lazily linked on the fly. This
 /// speeds up linking for modules with many/ lazily linked functions of which
 /// few get used.
 class ValueMaterializerTy : public ValueMaterializer {
   TypeMapTy &TypeMap;
   Module *DstM;
   std::vector<GlobalValue *> &LazilyLinkGlobalValues;
+  ModuleLinker *ModLinker;
 
 public:
-  ValueMaterializerTy(TypeMapTy &TypeMap, Module *DstM,
-                      std::vector<GlobalValue *> &LazilyLinkGlobalValues)
-      : ValueMaterializer(), TypeMap(TypeMap), DstM(DstM),
-        LazilyLinkGlobalValues(LazilyLinkGlobalValues) {}
+                     std::vector<GlobalValue *> &LazilyLinkGlobalValues,
+                     ModuleLinker *ModLinker)
+     : ValueMaterializer(),
+       TypeMap(TypeMap),
+       DstM(DstM),
+       LazilyLinkGlobalValues(LazilyLinkGlobalValues),
+       ModLinker(ModLinker) {}
 
   Value *materializeValueFor(Value *V) override;
 };
 
 class LinkDiagnosticInfo : public DiagnosticInfo {
   const Twine &Msg;
 
 public:
class ModuleLinker {
   /// Functions that have replaced other functions.
   SmallPtrSet<const Function *, 16> OverridingFunctions;
 
   DiagnosticHandlerFunction DiagnosticHandler;
 
   /// For symbol clashes, prefer those from Src.
   unsigned Flags;
 
+  /// Function index passed into ModuleLinker for using in function
+  /// importing/exporting handling.
+  FunctionInfoIndex *ImportIndex;
+
+  /// Function to import from source module, all other functions are
+  /// imported as declarations instead of definitions.
+  Function *ImportFunction;
+
+  /// Set to true if the given FunctionInfoIndex contains any functions
+  /// from this source module, in which case we must conservatively assume
+  /// that any of its functions may be imported into another module
+  /// as part of a different backend compilation process.
+  bool HasExportedFunctions;
+
 public:
-  ModuleLinker(Module *dstM, Linker::IdentifiedStructTypeSet &Set, Module *srcM,
-               DiagnosticHandlerFunction DiagnosticHandler, unsigned Flags)
-      : DstM(dstM), SrcM(srcM), TypeMap(Set),
-        ValMaterializer(TypeMap, DstM, LazilyLinkGlobalValues),
-        DiagnosticHandler(DiagnosticHandler), Flags(Flags) {}
+              DiagnosticHandlerFunction DiagnosticHandler, unsigned Flags,
+              FunctionInfoIndex *Index = nullptr,
+              Function *FuncToImport = nullptr)
+     : DstM(dstM),
+       SrcM(srcM),
+       TypeMap(Set),
+       ValMaterializer(TypeMap, DstM, LazilyLinkGlobalValues, this),
+       DiagnosticHandler(DiagnosticHandler),
+       Flags(Flags),
+       ImportIndex(Index),
+       ImportFunction(FuncToImport) {
+   assert(ImportIndex ||
+          !ImportFunction && "Expect a FunctionInfoIndex when importing");
+   // If we have a FunctionInfoIndex but no function to import,
+   // then this is the primary module being compiled in a ThinLTO
+   // backend compilation, and we need to see if it has functions that
+   // may be exported to another backend compilation.
+   if (ImportIndex && !ImportFunction)
+     HasExportedFunctions = ImportIndex->hasExportedFunctions(SrcM);
+ }
 
   bool run();
 
   bool shouldOverrideFromSrc() { return Flags & Linker::OverrideFromSrc; }
   bool shouldLinkOnlyNeeded() { return Flags & Linker::LinkOnlyNeeded; }
   bool shouldInternalizeLinkedSymbols() {
     return Flags & Linker::InternalizeLinkedSymbols;
   }
 
+  /// Handles cloning of a global values from the source module into
+  /// the destination module, including setting the attributes and visibility.
+  GlobalValue *copyGlobalValueProto(TypeMapTy &TypeMap, const GlobalValue *SGV,
+                                    const GlobalValue *DGV = nullptr);
+
+  /// Check if we should promote the given local value to global scope.
+  bool doPromoteLocalToGlobal(const GlobalValue *SGV);
+
 private:
   bool shouldLinkFromSource(bool &LinkFromSrc, const GlobalValue &Dest,
                             const GlobalValue &Src);
 
   /// Helper method for setting a message and returning an error code.
   bool emitError(const Twine &Message) {
     DiagnosticHandler(LinkDiagnosticInfo(DS_Error, Message));
     return true;
private:
 
   void linkAppendingVarInit(const AppendingVarInfo &AVI);
 
   void linkGlobalInit(GlobalVariable &Dst, GlobalVariable &Src);
   bool linkFunctionBody(Function &Dst, Function &Src);
   void linkAliasBody(GlobalAlias &Dst, GlobalAlias &Src);
   bool linkGlobalValueBody(GlobalValue &Src);
 
+  /// Functions that take care of cloning a specific global value type
+  /// into the destination module.
+  GlobalVariable *copyGlobalVariableProto(TypeMapTy &TypeMap,
+                                          const GlobalVariable *SGVar);
+  Function *copyFunctionProto(TypeMapTy &TypeMap, const Function *SF);
+  GlobalValue *copyGlobalAliasProto(TypeMapTy &TypeMap, const GlobalAlias *SGA);
+
+  /// Helper methods to check if we are importing from or potentially
+  /// exporting from the current source module.
+  bool isPerformingImport() { return ImportFunction != nullptr; }
+  bool isModuleExporting() { return HasExportedFunctions; }
+
+  /// If we are importing from the source module, checks if we should
+  /// import SGV as a definition, otherwise import as a declaration.
+  bool doImportAsDefinition(const GlobalValue *SGV);
+
+  /// Get the name for SGV that should be used in the linked destination
+  /// module. Specifically, this handles the case where we need to rename
+  /// a local that is being promoted to global scope.
+  StringRef getNewName(const GlobalValue *SGV);
+
+  /// Get the new linkage for SGV that should be used in the linked destination
+  /// module. Specifically, for ThinLTO importing or exporting it may need
+  /// to be adjusted.
+  GlobalValue::LinkageTypes getNewLinkage(const GlobalValue *SGV);
+
+  /// Copies the necessary global value attributes and name from the source
+  /// to the newly cloned global value.
+  void copyGVAttributes(GlobalValue *NewGV, const GlobalValue *SrcGV);
+
+  /// Updates the visibility for the new global cloned from the source
+  /// and, if applicable, linked with an existing destination global.
+  /// Handles visibility change required for promoted locals.
+  void setVisibility(GlobalValue *NewGV, const GlobalValue *SGV,
+                     const GlobalValue *DGV = nullptr);
+
   void linkNamedMDNodes();
   void stripReplacedSubprograms();
 };
 }
 
 /// The LLVM SymbolTable class autorenames globals that conflict in the symbol
 /// table. This is good for all clients except for us. Go through the trouble
 /// to force this back.
 static void forceRenaming(GlobalValue *GV, StringRef Name) {
   // If the global doesn't force its name or if it already has the right name,
   // there is nothing for us to do.
+  // Note that any required local to global promotion should already be done,
+  // so promoted locals will not skip this handling as their linkage is no
+  // longer local.
   if (GV->hasLocalLinkage() || GV->getName() == Name)
     return;
 
   Module *M = GV->getParent();
 
   // If there is a conflict, rename the conflict.
   if (GlobalValue *ConflictGV = M->getNamedValue(Name)) {
     GV->takeName(ConflictGV);
     ConflictGV->setName(Name);    // This will cause ConflictGV to get renamed
     assert(ConflictGV->getName() != Name && "forceRenaming didn't work");
   } else {
     GV->setName(Name);              // Force the name back
   }
 }
 
 /// copy additional attributes (those not needed to construct a GlobalValue)
 /// from the SrcGV to the DestGV.
-static void copyGVAttributes(GlobalValue *DestGV, const GlobalValue *SrcGV) {
-  DestGV->copyAttributesFrom(SrcGV);
-  forceRenaming(DestGV, SrcGV->getName());
+void ModuleLinker::copyGVAttributes(GlobalValue *NewGV,
+                                    const GlobalValue *SrcGV) {
+  auto *GA = dyn_cast<GlobalAlias>(SrcGV);
+  // Check for the special case of converting an alias (definition) to a
+  // non-alias (declaration). This can happen when we are importing and
+  // encounter a weak_any alias (weak_any defs may not be imported, see
+  // comments in getNewLinkage) or an alias whose base object is being
+  // imported as a declaration. In that case copy the attributes from the base
+  // object.
+  if (GA && !dyn_cast<GlobalAlias>(NewGV)) {
+    assert(isPerformingImport() &&
+           (GA->hasWeakAnyLinkage() ||
+            !doImportAsDefinition(GA->getBaseObject())));
+    NewGV->copyAttributesFrom(GA->getBaseObject());
+  } else
+    NewGV->copyAttributesFrom(SrcGV);
+  forceRenaming(NewGV, getNewName(SrcGV));
 }
 
 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;
 }
 
+bool ModuleLinker::doImportAsDefinition(const GlobalValue *SGV) {
+  if (!isPerformingImport()) return false;
+  // Always import GlobalVariable definitions. The linkage changes
+  // described in getNewLinkage ensure the correct behavior (e.g.
+  // global variables with external linkage are transformed to
+  // available_externally defintions, which are ultimately turned into
+  // declaratios after the EliminateAvailableExternally pass).
+  if (dyn_cast<GlobalVariable>(SGV) && !SGV->isDeclaration()) return true;
+  // Only import the function requested for importing.
+  auto *SF = dyn_cast<Function>(SGV);
+  if (SF && SF == ImportFunction) return true;
+  // Otherwise no.
+  return false;
+}
+
+bool ModuleLinker::doPromoteLocalToGlobal(const GlobalValue *SGV) {
+  assert(SGV->hasLocalLinkage());
+  // Both the imported references and the original local variable must
+  // be promoted.
+  if (!isPerformingImport() && !isModuleExporting()) return false;
+  auto *GVar = dyn_cast<GlobalVariable>(SGV);
+  // Local const variables never need to be promoted, the imported uses
+  // can simply use the clone created in this module.
+  if (GVar && GVar->isConstant()) return false;
+  return true;
+}
+
+StringRef ModuleLinker::getNewName(const GlobalValue *SGV) {
+  // For locals that must be promoted to global scope, ensure that
+  // the promoted name uniquely identifies the copy in the original module,
+  // using the ID assigned during combined index creation.
+  if (SGV->hasLocalLinkage() && doPromoteLocalToGlobal(SGV))
+    return FunctionInfoIndex::getGlobalNameForLocal(
+        SGV->getName(),
+        ImportIndex->getModuleId(SGV->getParent()->getModuleIdentifier()));
+  return SGV->getName();
+}
+
+GlobalValue::LinkageTypes ModuleLinker::getNewLinkage(const GlobalValue *SGV) {
+  // Any local variable that is referenced by an exported function needs
+  // to be promoted to global scope. Since we don't currently know which
+  // functions reference which local variables/functions, we must treat
+  // all as potentially exported if this module is exporting anything.
+  if (isModuleExporting()) {
+    if (SGV->hasLocalLinkage() && doPromoteLocalToGlobal(SGV))
+      return GlobalValue::ExternalLinkage;
+    return SGV->getLinkage();
+  }
+
+  // Otherwise, if we aren't importing, no linkage change is needed.
+  if (!isPerformingImport()) return SGV->getLinkage();
+
+  switch (SGV->getLinkage()) {
+    case GlobalValue::ExternalLinkage:
+      // External defnitions are converted to available_externally
+      // definitions upon import, so that they are available for inlining
+      // and/or optimization, but are turned into declarations later
+      // during the EliminateAvailableExternally pass.
+      if (doImportAsDefinition(SGV))
+        return GlobalValue::AvailableExternallyLinkage;
+      // An imported external declaration stays external.
+      return SGV->getLinkage();
+
+    case GlobalValue::AvailableExternallyLinkage:
+      // An imported available_externally definition converts
+      // to external if imported as a declaration.
+      if (isPerformingImport() && !doImportAsDefinition(SGV))
+        return GlobalValue::ExternalLinkage;
+      // An imported available_externally declaration stays that way.
+      return SGV->getLinkage();
+
+    case GlobalValue::LinkOnceAnyLinkage:
+    case GlobalValue::LinkOnceODRLinkage:
+      // These both stay the same when importing the definition.
+      // The ThinLTO pass will eventually force-import their definitions.
+      return SGV->getLinkage();
+
+    case GlobalValue::WeakAnyLinkage:
+      // Can't import weak_any definitions correctly, or we might change the
+      // program semantics, since the linker will pick the first weak_any
+      // definition and importing would change the order they are seen by the
+      // linker. The module linking caller needs to enforce this.
+      assert(!doImportAsDefinition(SGV));
+      // If imported as a declaration, it becomes external_weak.
+      return GlobalValue::ExternalWeakLinkage;
+
+    case GlobalValue::WeakODRLinkage:
+      // For weak_odr linkage, there is a guarantee that all copies will be
+      // equivalent, so the issue described above for weak_any does not exist,
+      // and the definition can be imported. It can be treated similarly
+      // to an imported externally visible global value.
+      if (doImportAsDefinition(SGV))
+        return GlobalValue::AvailableExternallyLinkage;
+      else
+        return GlobalValue::ExternalLinkage;
+
+    case GlobalValue::AppendingLinkage:
+      // It would be incorrect to import an appending linkage variable,
+      // since it would cause global constructors/destructors to be
+      // executed multiple times. This should have already been handled
+      // by linkGlobalValueProto.
+      assert(false && "Cannot import appending linkage variable");
+
+    case GlobalValue::InternalLinkage:
+    case GlobalValue::PrivateLinkage:
+      // If we are promoting the local to global scope, it is handled
+      // similarly to a normal externally visible global.
+      if (doPromoteLocalToGlobal(SGV)) {
+        if (doImportAsDefinition(SGV))
+          return GlobalValue::AvailableExternallyLinkage;
+        else
+          return GlobalValue::ExternalLinkage;
+      }
+      // A non-promoted imported local definition stays local.
+      // The ThinLTO pass will eventually force-import their definitions.
+      return SGV->getLinkage();
+
+    case GlobalValue::ExternalWeakLinkage:
+      // External weak doesn't apply to declarations, must be a definition.
+      assert(!doImportAsDefinition(SGV));
+      // Linkage stays external_weak.
+      return SGV->getLinkage();
+
+    case GlobalValue::CommonLinkage:
+      // Linkage stays common on definitions.
+      // The ThinLTO pass will eventually force-import their definitions.
+      return SGV->getLinkage();
+  }
+
+  llvm_unreachable("unknown linkage type");
+}
+
 /// 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) {
+GlobalVariable *ModuleLinker::copyGlobalVariableProto(
+    TypeMapTy &TypeMap, 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(),
+      *DstM, TypeMap.get(SGVar->getType()->getElementType()),
+      SGVar->isConstant(), getNewLinkage(SGVar), /*init*/ nullptr,
+      getNewName(SGVar), /*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,
+Function *ModuleLinker::copyFunctionProto(TypeMapTy &TypeMap,
                                           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);
+  return Function::Create(TypeMap.get(SF->getFunctionType()), getNewLinkage(SF),
+                          getNewName(SF), DstM);
 }
 
 /// Set up prototypes for any aliases that come over from the source module.
-static GlobalAlias *copyGlobalAliasProto(TypeMapTy &TypeMap, Module &DstM,
+GlobalValue *ModuleLinker::copyGlobalAliasProto(TypeMapTy &TypeMap,
                                                 const GlobalAlias *SGA) {
+  // If we are importing and encounter a weak_any alias, or an alias to
+  // an object being imported as a declaration, we must import the alias
+  // as a declaration as well, which involves converting it to a non-alias.
+  // See comments in getNewLinkage for why we cannot import weak_any defintions.
+  if (isPerformingImport() && (SGA->hasWeakAnyLinkage() ||
+                               !doImportAsDefinition(SGA->getBaseObject()))) {
+    // Need to convert to declaration. All aliases must be definitions.
+    const GlobalValue *GVal = SGA->getBaseObject();
+    GlobalValue *NewGV;
+    if (auto *GVar = dyn_cast<GlobalVariable>(GVal))
+      NewGV = copyGlobalVariableProto(TypeMap, GVar);
+    else {
+      auto *F = dyn_cast<Function>(GVal);
+      assert(F);
+      NewGV = copyFunctionProto(TypeMap, F);
+    }
+    // Set the linkage to ExternalWeak, see also comments in getNewLinkage.
+    if (SGA->hasWeakAnyLinkage())
+      NewGV->setLinkage(GlobalValue::ExternalWeakLinkage);
+    // Don't attempt to link body, needs to be a declaration.
+    DoNotLinkFromSource.insert(SGA);
+    return NewGV;
+  }
   // If there is no linkage to be performed or we're linking from the source,
   // bring over SGA.
   auto *Ty = TypeMap.get(SGA->getValueType());
   return GlobalAlias::create(Ty, SGA->getType()->getPointerAddressSpace(),
-                             SGA->getLinkage(), SGA->getName(), &DstM);
+                             getNewLinkage(SGA), getNewName(SGA), DstM);
+}
+
+void ModuleLinker::setVisibility(GlobalValue *NewGV, const GlobalValue *SGV,
+                                 const GlobalValue *DGV) {
+  GlobalValue::VisibilityTypes Visibility = SGV->getVisibility();
+  if (DGV)
+    Visibility = isLessConstraining(Visibility, DGV->getVisibility())
+                     ? DGV->getVisibility()
+                     : Visibility;
+  // For promoted locals, mark them hidden so that they can later be
+  // stripped from the symbol table to reduce bloat.
+  if (SGV->hasLocalLinkage() && doPromoteLocalToGlobal(SGV))
+    Visibility = GlobalValue::HiddenVisibility;
+  NewGV->setVisibility(Visibility);
 }
 
-static GlobalValue *copyGlobalValueProto(TypeMapTy &TypeMap, Module &DstM,
-                                         const GlobalValue *SGV) {
+GlobalValue *ModuleLinker::copyGlobalValueProto(TypeMapTy &TypeMap,
+                                                const GlobalValue *SGV,
+                                                const GlobalValue *DGV) {
   GlobalValue *NewGV;
   if (auto *SGVar = dyn_cast<GlobalVariable>(SGV))
-    NewGV = copyGlobalVariableProto(TypeMap, DstM, SGVar);
+    NewGV = copyGlobalVariableProto(TypeMap, SGVar);
   else if (auto *SF = dyn_cast<Function>(SGV))
-    NewGV = copyFunctionProto(TypeMap, DstM, SF);
+    NewGV = copyFunctionProto(TypeMap, SF);
   else
-    NewGV = copyGlobalAliasProto(TypeMap, DstM, cast<GlobalAlias>(SGV));
+    NewGV = copyGlobalAliasProto(TypeMap, cast<GlobalAlias>(SGV));
   copyGVAttributes(NewGV, SGV);
+  setVisibility(NewGV, SGV, DGV);
   return NewGV;
 }
 
 Value *ValueMaterializerTy::materializeValueFor(Value *V) {
   auto *SGV = dyn_cast<GlobalValue>(V);
   if (!SGV)
     return nullptr;
 
-  GlobalValue *DGV = copyGlobalValueProto(TypeMap, *DstM, SGV);
+  GlobalValue *DGV = ModLinker->copyGlobalValueProto(TypeMap, SGV);
 
   if (Comdat *SC = SGV->getComdat()) {
     if (auto *DGO = dyn_cast<GlobalObject>(DGV)) {
       Comdat *DC = DstM->getOrInsertComdat(SC->getName());
       DGO->setComdat(DC);
     }
   }
 
bool ModuleLinker::shouldLinkFromSource(bool &LinkFromSrc,
   // Should we unconditionally use the Src?
   if (shouldOverrideFromSrc()) {
     LinkFromSrc = true;
     return false;
   }
 
   // We always have to add Src if it has appending linkage.
   if (Src.hasAppendingLinkage()) {
+    // Caller should have already determined that we can't link from source
+    // when importing (see comments in linkGlobalValueProto).
+    assert(!isPerformingImport());
     LinkFromSrc = true;
     return false;
   }
 
+  if (isPerformingImport() && isa<Function>(&Src)) {
+    // For functions, LinkFromSrc iff this is the function requested
+    // for importing. For variables, decide below normally.
+    LinkFromSrc = (&Src == ImportFunction);
+    return false;
+  }
+
   bool SrcIsDeclaration = Src.isDeclarationForLinker();
   bool DestIsDeclaration = Dest.isDeclarationForLinker();
 
   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.hasDLLImportStorageClass()) {
       // If one of GVs is marked as DLLImport, result should be dllimport'ed.
bool ModuleLinker::linkAppendingVarProto(GlobalVariable *DstGV,
 
   return false;
 }
 
 bool ModuleLinker::linkGlobalValueProto(GlobalValue *SGV) {
   GlobalValue *DGV = getLinkedToGlobal(SGV);
 
   // Handle the ultra special appending linkage case first.
+  assert(!DGV || SGV->hasAppendingLinkage() == DGV->hasAppendingLinkage());
+  if (SGV->hasAppendingLinkage() && isPerformingImport()) {
+    // Don't want to append to global_ctors list, for example, when we
+    // are importing for ThinLTO, otherwise the global ctors and dtors
+    // get executed multiple times for local variables (the latter causing
+    // double frees).
+    DoNotLinkFromSource.insert(SGV);
+    return false;
+  }
   if (DGV && DGV->hasAppendingLinkage())
     return linkAppendingVarProto(cast<GlobalVariable>(DGV),
                                  cast<GlobalVariable>(SGV));
 
   bool LinkFromSrc = true;
   Comdat *C = nullptr;
-  GlobalValue::VisibilityTypes Visibility = SGV->getVisibility();
   bool HasUnnamedAddr = SGV->hasUnnamedAddr();
 
   if (const Comdat *SC = SGV->getComdat()) {
     Comdat::SelectionKind SK;
     std::tie(SK, LinkFromSrc) = ComdatsChosen[SC];
     C = DstM->getOrInsertComdat(SC->getName());
     C->setSelectionKind(SK);
   } else if (DGV) {
     if (shouldLinkFromSource(LinkFromSrc, *DGV, *SGV))
       return true;
   }
 
   if (!LinkFromSrc) {
     // Track the source global so that we don't attempt to copy it over when
     // processing global initializers.
     DoNotLinkFromSource.insert(SGV);
 
     if (DGV)
       // Make sure to remember this mapping.
       ValueMap[SGV] =
           ConstantExpr::getBitCast(DGV, TypeMap.get(SGV->getType()));
   }
 
-  if (DGV) {
-    Visibility = isLessConstraining(Visibility, DGV->getVisibility())
-                     ? DGV->getVisibility()
-                     : Visibility;
-    HasUnnamedAddr = HasUnnamedAddr && DGV->hasUnnamedAddr();
-  }
+  if (DGV) HasUnnamedAddr = HasUnnamedAddr && DGV->hasUnnamedAddr();
 
   if (!LinkFromSrc && !DGV)
     return false;
 
   GlobalValue *NewGV;
   if (!LinkFromSrc) {
     NewGV = DGV;
+    // When linking from source we setVisibility from copyGlobalValueProto.
+    setVisibility(NewGV, SGV, DGV);
   } else {
     // 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 && !shouldOverrideFromSrc() &&
+    if (!DGV && !shouldOverrideFromSrc() && SGV != ImportFunction &&
         (SGV->hasLocalLinkage() || SGV->hasLinkOnceLinkage() ||
          SGV->hasAvailableExternallyLinkage())) {
       DoNotLinkFromSource.insert(SGV);
       return false;
     }
 
     // When we only want to link in unresolved dependencies, blacklist
     // the symbol unless unless DestM has a matching declaration (DGV).
     if (shouldLinkOnlyNeeded() && !(DGV && DGV->isDeclaration())) {
       DoNotLinkFromSource.insert(SGV);
       return false;
     }
 
-    NewGV = copyGlobalValueProto(TypeMap, *DstM, SGV);
+    NewGV = copyGlobalValueProto(TypeMap, SGV, DGV);
 
     if (DGV && isa<Function>(DGV))
       if (auto *NewF = dyn_cast<Function>(NewGV))
         OverridingFunctions.insert(NewF);
   }
 
   NewGV->setUnnamedAddr(HasUnnamedAddr);
-  NewGV->setVisibility(Visibility);
 
   if (auto *NewGO = dyn_cast<GlobalObject>(NewGV)) {
     if (C)
       NewGO->setComdat(C);
 
     if (DGV && DGV->hasCommonLinkage() && SGV->hasCommonLinkage())
       NewGO->setAlignment(std::max(DGV->getAlignment(), SGV->getAlignment()));
   }
for (Function &SF : *SrcM) {
     // Skip if no body (function is external).
     if (SF.isDeclaration())
       continue;
 
     // Skip if not linking from source.
     if (DoNotLinkFromSource.count(&SF))
       continue;
 
+    // When importing, only materialize the function requested for import.
+    if (isPerformingImport() && &SF != ImportFunction) continue;
+
     if (linkGlobalValueBody(SF))
       return true;
   }
 
   // Resolve all uses of aliases with aliasees.
   for (GlobalAlias &Src : SrcM->aliases()) {
     if (DoNotLinkFromSource.count(&Src))
       continue;
if (!Src.hasInitializer() || DoNotLinkFromSource.count(&Src))
       continue;
     linkGlobalValueBody(Src);
   }
 
   // Process vector of lazily linked in functions.
   while (!LazilyLinkGlobalValues.empty()) {
     GlobalValue *SGV = LazilyLinkGlobalValues.back();
     LazilyLinkGlobalValues.pop_back();
+    if (isPerformingImport() && !doImportAsDefinition(SGV)) continue;
 
     // Skip declarations that ValueMaterializer may have created in
     // case we link in only some of SrcM.
     if (shouldLinkOnlyNeeded() && SGV->isDeclaration())
       continue;
 
     assert(!SGV->isDeclaration() && "users should not pass down decls");
     if (linkGlobalValueBody(*SGV))
Linker::Linker(Module *M) {
   });
 }
 
 void Linker::deleteModule() {
   delete Composite;
   Composite = nullptr;
 }
 
-bool Linker::linkInModule(Module *Src, unsigned Flags) {
+bool Linker::linkInModule(Module *Src, unsigned Flags, FunctionInfoIndex *Index,
+                          Function *FuncToImport) {
   ModuleLinker TheLinker(Composite, IdentifiedStructTypes, Src,
-                         DiagnosticHandler, Flags);
+                         DiagnosticHandler, Flags, Index, FuncToImport);
   bool RetCode = TheLinker.run();
   Composite->dropTriviallyDeadConstantArrays();
   return RetCode;
 }
 
 void Linker::setModule(Module *Dst) {
   init(Dst, DiagnosticHandler);
 }

lib/Transforms/IPO/ElimAvailExtern.cpp

 char EliminateAvailableExternally::ID = 0;
 INITIALIZE_PASS(EliminateAvailableExternally, "elim-avail-extern",
                 "Eliminate Available Externally Globals", false, false)
 
 ModulePass *llvm::createEliminateAvailableExternallyPass() {
   return new EliminateAvailableExternally();
 }
 
+static void convertAliasToDeclaration(GlobalAlias &GA, Module &M) {
+  GlobalValue *GVal = GA.getBaseObject();
+  GlobalValue *NewGV;
+  if (auto *GVar = dyn_cast<GlobalVariable>(GVal)) {
+    GlobalVariable *NewGVar = new GlobalVariable(
+        M, GVar->getType()->getElementType(), GVar->isConstant(),
+        GVar->getLinkage(), /*init*/ nullptr, GA.getName(), GVar,
+        GVar->getThreadLocalMode(), GVar->getType()->getAddressSpace());
+    NewGV = NewGVar;
+    NewGV->copyAttributesFrom(GVar);
+  } else {
+    auto *F = dyn_cast<Function>(GVal);
+    assert(F);
+    Function *NewF = Function::Create(F->getFunctionType(), F->getLinkage(),
+                                      GA.getName(), &M);
+    NewGV = NewF;
+    NewGV->copyAttributesFrom(F);
+  }
+  GA.replaceAllUsesWith(ConstantExpr::getBitCast(NewGV, GA.getType()));
+  GA.eraseFromParent();
+}
+
 bool EliminateAvailableExternally::runOnModule(Module &M) {
   bool Changed = false;
 
+  // Convert any aliases that alias with an available externally
+  // value (which will be turned into declarations later on in this routine)
+  // into declarations themselves. All aliases must be definitions, and
+  // must alias with a definition. So this involves creating a declaration
+  // equivalent to the alias's base object.
+  for (Module::alias_iterator I = M.alias_begin(), E = M.alias_end(); I != E;) {
+    // Increment the iterator first since we may delete the current alias.
+    GlobalAlias &GA = *(I++);
+    GlobalValue *GVal = GA.getBaseObject();
+    if (!GVal->hasAvailableExternallyLinkage()) continue;
+    convertAliasToDeclaration(GA, M);
+    Changed = true;
+  }
+
   // Drop initializers of available externally global variables.
   for (GlobalVariable &GV :M.globals()) {
     if (!GV.hasAvailableExternallyLinkage())
       continue;
     if (GV.hasInitializer()) {
       Constant *Init = GV.getInitializer();
       GV.setInitializer(nullptr);
       if (isSafeToDestroyConstant(Init))
         Init->destroyConstant();
     }
     GV.removeDeadConstantUsers();
     GV.setLinkage(GlobalValue::ExternalLinkage);
     NumVariables++;
+    Changed = true;
   }
 
   // Drop the bodies of available externally functions.
   for (Function &F : M) {
     if (!F.hasAvailableExternallyLinkage())
       continue;
     if (!F.isDeclaration())
       // This will set the linkage to external
       F.deleteBody();
     F.removeDeadConstantUsers();
     NumFunctions++;
+    Changed = true;
   }
 
   return Changed;
 }

test/Linker/Inputs/funcimport.ll

+define i32 @main() #0 {
+entry:
+  call void (...) @weakalias()
+  call void (...) @analias()
+  %call = call i32 (...) @referencestatics()
+  %call1 = call i32 (...) @referenceglobals()
+  %call2 = call i32 (...) @referencecommon()
+  call void (...) @setfuncptr()
+  call void (...) @callfuncptr()
+  call void (...) @callweakfunc()
+  ret i32 0
+}
+
+declare void @weakalias(...) #1
+
+declare void @analias(...) #1
+
+declare i32 @referencestatics(...) #1
+
+declare i32 @referenceglobals(...) #1
+
+declare i32 @referencecommon(...) #1
+
+declare void @setfuncptr(...) #1
+
+declare void @callfuncptr(...) #1
+
+declare void @callweakfunc(...) #1

test/Linker/funcimport.ll

+; Do setup work for all below tests: generate bitcode and combined index
+; RUN: llvm-as -function-summary %s -o %t.o
+; RUN: llvm-as -function-summary %p/Inputs/funcimport.ll -o %t2.o
+; RUN: %gold -plugin %llvmshlibdir/LLVMgold.so \
+; RUN:    --plugin-opt=thinlto \
+; RUN:    -shared %t.o %t2.o -o %t3
+
+; Ensure statics are promoted/renamed correctly from this file (all but
+; constant variable need promotion).
+; RUN: llvm-link %t.o -functionindex=%t3.thinlto.bc -S | FileCheck %s --check-prefix=EXPORTSTATIC
+; EXPORTSTATIC: @staticvar.llvm.1 = hidden global
+; EXPORTSTATIC: @staticconstvar = internal constant
+; EXPORTSTATIC: @P.llvm.1 = hidden global void ()* null
+; EXPORTSTATIC: define hidden i32 @staticfunc.llvm.1
+; EXPORTSTATIC: define hidden void @staticfunc2.llvm.1
+
+; Ensure that both weak alias to an imported function and strong alias to a
+; non-imported function are correctly turned into declarations.
+; RUN: llvm-link %t2.o -functionindex=%t3.thinlto.bc -import=globalfunc1:%t.o -S | FileCheck %s --check-prefix=IMPORTGLOB1
+; IMPORTGLOB1: define available_externally void @globalfunc1
+; IMPORTGLOB1: declare void @globalfunc2
+; IMPORTGLOB1: declare extern_weak void @weakalias
+; IMPORTGLOB1: declare void @analias
+
+; Ensure that both weak alias to a non-imported function is correctly
+; turned into a declaration, but that strong alias to an imported function
+; is imported as alias.
+; RUN: llvm-link %t2.o -functionindex=%t3.thinlto.bc -import=globalfunc2:%t.o -S | FileCheck %s --check-prefix=IMPORTGLOB2
+; IMPORTGLOB2: @analias = alias void (...), bitcast (void ()* @globalfunc2
+; IMPORTGLOB2: declare void @globalfunc1
+; IMPORTGLOB2: define available_externally void @globalfunc2
+; IMPORTGLOB2: declare extern_weak void @weakalias
+
+; Ensure that strong alias imported in second pass of importing ends up
+; as an alias.
+; RUN: llvm-link %t2.o -functionindex=%t3.thinlto.bc -import=globalfunc1:%t.o -import=globalfunc2:%t.o -S | FileCheck %s --check-prefix=IMPORTGLOB3
+; IMPORTGLOB3: @analias = alias void (...), bitcast (void ()* @globalfunc2
+; IMPORTGLOB3: define available_externally void @globalfunc1
+; IMPORTGLOB3: define available_externally void @globalfunc2
+; IMPORTGLOB3: declare extern_weak void @weakalias
+
+; Ensure that imported static variable and function references are correctly
+; promoted and renamed (except for static constant variable).
+; RUN: llvm-link %t2.o -functionindex=%t3.thinlto.bc -import=referencestatics:%t.o -S | FileCheck %s --check-prefix=IMPORTSTATIC
+; IMPORTSTATIC: @staticvar.llvm.1 = available_externally hidden global
+; IMPORTSTATIC: @staticconstvar = internal constant
+; IMPORTSTATIC: define available_externally i32 @referencestatics
+; IMPORTSTATIC: %call = call i32 @staticfunc.llvm.1
+; IMPORTSTATIC: %0 = load i32, i32* @staticvar.llvm.1
+; IMPORTSTATIC: declare hidden i32 @staticfunc.llvm.1
+
+; Ensure that imported global (external) function and variable references
+; are handled correctly (including referenced variable imported as
+; available_externally definition)
+; RUN: llvm-link %t2.o -functionindex=%t3.thinlto.bc -import=referenceglobals:%t.o -S | FileCheck %s --check-prefix=IMPORTGLOBALS
+; IMPORTGLOBALS: @globalvar = available_externally global
+; IMPORTGLOBALS: declare void @globalfunc1()
+; IMPORTGLOBALS: define available_externally i32 @referenceglobals
+
+; Ensure that common variable correctly imported as common defition.
+; RUN: llvm-link %t2.o -functionindex=%t3.thinlto.bc -import=referencecommon:%t.o -S | FileCheck %s --check-prefix=IMPORTCOMMON
+; IMPORTCOMMON: @commonvar = common global
+; IMPORTCOMMON: define available_externally i32 @referencecommon
+
+; Ensure that imported static function pointer correctly promoted and renamed.
+; RUN: llvm-link %t2.o -functionindex=%t3.thinlto.bc -import=callfuncptr:%t.o -S | FileCheck %s --check-prefix=IMPORTFUNCPTR
+; IMPORTFUNCPTR: @P.llvm.1 = available_externally hidden global void ()* null
+; IMPORTFUNCPTR: define available_externally void @callfuncptr
+; IMPORTFUNCPTR: %0 = load void ()*, void ()** @P.llvm.1
+
+; Ensure that imported weak function reference/definition handled properly.
+; Imported weak_any definition should be skipped with warning, and imported
+; reference should turned into an external_weak declaration.
+; RUN: llvm-link %t2.o -functionindex=%t3.thinlto.bc -import=callweakfunc:%t.o -import=weakfunc:%t.o -S 2>&1 | FileCheck %s --check-prefix=IMPORTWEAKFUNC
+; IMPORTWEAKFUNC: Ignoring import request for weak-any function weakfunc
+; IMPORTWEAKFUNC: declare extern_weak void @weakfunc
+; IMPORTWEAKFUNC: define available_externally void @callweakfunc
+
+@globalvar = global i32 1, align 4
+@staticvar = internal global i32 1, align 4
+@staticconstvar = internal constant [2 x i32] [i32 10, i32 20], align 4
+@commonvar = common global i32 0, align 4
+@P = internal global void ()* null, align 8
+
+@weakalias = weak alias void (...), bitcast (void ()* @globalfunc1 to void (...)*)
+@analias = alias void (...), bitcast (void ()* @globalfunc2 to void (...)*)
+
+define void @globalfunc1() #0 {
+entry:
+  ret void
+}
+
+define void @globalfunc2() #0 {
+entry:
+  ret void
+}
+
+define i32 @referencestatics(i32 %i) #0 {
+entry:
+  %i.addr = alloca i32, align 4
+  store i32 %i, i32* %i.addr, align 4
+  %call = call i32 @staticfunc()
+  %0 = load i32, i32* @staticvar, align 4
+  %add = add nsw i32 %call, %0
+  %1 = load i32, i32* %i.addr, align 4
+  %idxprom = sext i32 %1 to i64
+  %arrayidx = getelementptr inbounds [2 x i32], [2 x i32]* @staticconstvar, i64 0, i64 %idxprom
+  %2 = load i32, i32* %arrayidx, align 4
+  %add1 = add nsw i32 %add, %2
+  ret i32 %add1
+}
+
+define i32 @referenceglobals(i32 %i) #0 {
+entry:
+  %i.addr = alloca i32, align 4
+  store i32 %i, i32* %i.addr, align 4
+  call void @globalfunc1()
+  %0 = load i32, i32* @globalvar, align 4
+  ret i32 %0
+}
+
+define i32 @referencecommon(i32 %i) #0 {
+entry:
+  %i.addr = alloca i32, align 4
+  store i32 %i, i32* %i.addr, align 4
+  %0 = load i32, i32* @commonvar, align 4
+  ret i32 %0
+}
+
+define void @setfuncptr() #0 {
+entry:
+  store void ()* @staticfunc2, void ()** @P, align 8
+  ret void
+}
+
+define void @callfuncptr() #0 {
+entry:
+  %0 = load void ()*, void ()** @P, align 8
+  call void %0()
+  ret void
+}
+
+define weak void @weakfunc() #0 {
+entry:
+  ret void
+}
+
+define void @callweakfunc() #0 {
+entry:
+  call void @weakfunc()
+  ret void
+}
+
+define internal i32 @staticfunc() #0 {
+entry:
+  ret i32 1
+}
+
+define internal void @staticfunc2() #0 {
+entry:
+  ret void
+}

tools/llvm-link/CMakeLists.txt

 set(LLVM_LINK_COMPONENTS
   BitWriter
   Core
   IRReader
   Linker
+  Object
   Support
   )
 
 add_llvm_tool(llvm-link
   llvm-link.cpp
   )

tools/llvm-link/LLVMBuild.txt

 ;   http://llvm.org/docs/LLVMBuild.html
 ;
 ;===------------------------------------------------------------------------===;
 
 [component_0]
 type = Tool
 name = llvm-link
 parent = Tools
-required_libraries = AsmParser BitReader BitWriter IRReader Linker
+required_libraries = AsmParser BitReader BitWriter IRReader Linker Object

tools/llvm-link/llvm-link.cpp

 //===----------------------------------------------------------------------===//
 
 #include "llvm/Linker/Linker.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/Bitcode/ReaderWriter.h"
 #include "llvm/IR/AutoUpgrade.h"
 #include "llvm/IR/DiagnosticInfo.h"
 #include "llvm/IR/DiagnosticPrinter.h"
+#include "llvm/IR/FunctionInfo.h"
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Verifier.h"
 #include "llvm/IRReader/IRReader.h"
+#include "llvm/Object/FunctionIndexObjectFile.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/ManagedStatic.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/PrettyStackTrace.h"
 #include "llvm/Support/Signals.h"
 #include "llvm/Support/SourceMgr.h"
 #include "llvm/Support/SystemUtils.h"
 #include "llvm/Support/ToolOutputFile.h"
 #include <memory>
 using namespace llvm;
 
 static cl::list<std::string>
 InputFilenames(cl::Positional, cl::OneOrMore,
                cl::desc("<input bitcode files>"));
 
 static cl::list<std::string> OverridingInputs(
     "override", cl::ZeroOrMore, cl::value_desc("filename"),
     cl::desc(
         "input bitcode file which can override previously defined symbol(s)"));
 
+// Option to simulate function importing for testing. This enables using
+// llvm-link to simulate ThinLTO backend processes.
+static cl::list<std::string> Imports(
+    "import", cl::ZeroOrMore, cl::value_desc("function:filename"),
+    cl::desc("Pair of function name and filename, where function should be "
+             "imported from bitcode in filename"));
+
+// Option to support testing of function importing. The function index
+// must be specified in the case were we request imports via the -import
+// option, as well as when compiling any module with functions that may be
+// exported (imported by a different llvm-link -import invocation), to ensure
+// consistent promotion and renaming of locals.
+static cl::opt<std::string> FunctionIndex("functionindex",
+                                          cl::desc("Function index filename"),
+                                          cl::init(""),
+                                          cl::value_desc("filename"));
+
 static cl::opt<std::string>
 OutputFilename("o", cl::desc("Override output filename"), cl::init("-"),
                cl::value_desc("filename"));
 
 static cl::opt<bool>
 Internalize("internalize", cl::desc("Internalize linked symbols"));
 
 static cl::opt<bool>
case DS_Note:
     llvm_unreachable("Only expecting warnings and errors");
   }
 
   DiagnosticPrinterRawOStream DP(errs());
   DI.print(DP);
   errs() << '\n';
 }
 
+/// Load a function index if requested by the -functionindex option.
+static ErrorOr<std::unique_ptr<FunctionInfoIndex>> loadIndex(
+    LLVMContext &Context) {
+  assert(!FunctionIndex.empty());
+  ErrorOr<std::unique_ptr<MemoryBuffer>> FileOrErr =
+      MemoryBuffer::getFileOrSTDIN(FunctionIndex);
+  std::error_code EC = FileOrErr.getError();
+  if (EC) return EC;
+  MemoryBufferRef BufferRef = (FileOrErr.get())->getMemBufferRef();
+  ErrorOr<std::unique_ptr<object::FunctionIndexObjectFile>> ObjOrErr =
+      object::FunctionIndexObjectFile::create(BufferRef, Context);
+  EC = ObjOrErr.getError();
+  if (EC) return EC;
+
+  object::FunctionIndexObjectFile &Obj = **ObjOrErr;
+  return Obj.takeIndex();
+}
+
+/// Import any functions requested via the -import option.
+static bool importFunctions(const char *argv0, LLVMContext &Context,
+                            Linker &L) {
+  for (const auto &Import : Imports) {
+    // Identify the requested function and its bitcode source file.
+    size_t Idx = Import.find(':');
+    if (Idx == std::string::npos) {
+      errs() << "Import parameter bad format: " << Import << "\n";
+      return false;
+    }
+    std::string FunctionName = Import.substr(0, Idx);
+    std::string FileName = Import.substr(Idx + 1, std::string::npos);
+
+    // Load the specified source module.
+    std::unique_ptr<Module> M = loadFile(argv0, FileName, Context);
+    if (!M.get()) {
+      errs() << argv0 << ": error loading file '" << FileName << "'\n";
+      return false;
+    }
+
+    if (verifyModule(*M, &errs())) {
+      errs() << argv0 << ": " << FileName
+             << ": error: input module is broken!\n";
+      return false;
+    }
+
+    Function *F = M->getFunction(FunctionName);
+    if (!F) {
+      errs() << "Ignoring import request for non-existent function "
+             << FunctionName << " from " << FileName << "\n";
+      continue;
+    }
+    // We cannot import weak_any functions without possibly affecting the
+    // order they are seen and selected by the linker, changing program
+    // semantics.
+    if (F->hasWeakAnyLinkage()) {
+      errs() << "Ignoring import request for weak-any function " << FunctionName
+             << " from " << FileName << "\n";
+      continue;
+    }
+
+    if (Verbose)
+      errs() << "Importing " << FunctionName << " from " << FileName << "\n";
+
+    std::unique_ptr<FunctionInfoIndex> Index;
+    if (!FunctionIndex.empty()) {
+      ErrorOr<std::unique_ptr<FunctionInfoIndex>> IndexOrErr =
+          loadIndex(Context);
+      std::error_code EC = IndexOrErr.getError();
+      if (EC) {
+        errs() << EC.message() << '\n';
+        return false;
+      }
+      Index = std::move(IndexOrErr.get());
+    }
+
+    // Link in the specified function.
+    if (L.linkInModule(M.get(), false, Index.get(), F)) return false;
+  }
+  return true;
+}
+
 static bool linkFiles(const char *argv0, LLVMContext &Context, Linker &L,
                       const cl::list<std::string> &Files,
                       unsigned Flags) {
   // Filter out flags that don't apply to the first file we load.
   unsigned ApplicableFlags = Flags & Linker::Flags::OverrideFromSrc;
   for (const auto &File : Files) {
     std::unique_ptr<Module> M = loadFile(argv0, File, Context);
     if (!M.get()) {
       errs() << argv0 << ": error loading file '" << File << "'\n";
       return false;
     }
 
     if (verifyModule(*M, &errs())) {
       errs() << argv0 << ": " << File << ": error: input module is broken!\n";
       return false;
     }
 
+    // If a function index is supplied, load it so linkInModule can treat
+    // local functions/variables as exported and promote if necessary.
+    std::unique_ptr<FunctionInfoIndex> Index;
+    if (!FunctionIndex.empty()) {
+      ErrorOr<std::unique_ptr<FunctionInfoIndex>> IndexOrErr =
+          loadIndex(Context);
+      std::error_code EC = IndexOrErr.getError();
+      if (EC) {
+        errs() << EC.message() << '\n';
+        return false;
+      }
+      Index = std::move(IndexOrErr.get());
+    }
+
     if (Verbose)
       errs() << "Linking in '" << File << "'\n";
 
-    if (L.linkInModule(M.get(), ApplicableFlags))
-      return false;
+    if (L.linkInModule(M.get(), ApplicableFlags, Index.get())) return false;
     // All linker flags apply to linking of subsequent files.
     ApplicableFlags = Flags;
   }
 
   return true;
 }
 
 int main(int argc, char **argv) {
int main(int argc, char **argv) {
   if (!linkFiles(argv[0], Context, L, InputFilenames, Flags))
     return 1;
 
   // Next the -override ones.
   if (!linkFiles(argv[0], Context, L, OverridingInputs,
                  Flags | Linker::Flags::OverrideFromSrc))
     return 1;
 
+  // Import any functions requested via -import
+  if (!importFunctions(argv[0], Context, L)) return 1;
+
   if (DumpAsm) errs() << "Here's the assembly:\n" << *Composite;
 
   std::error_code EC;
   tool_output_file Out(OutputFilename, EC, sys::fs::F_None);
   if (EC) {
     errs() << EC.message() << '\n';
     return 1;
   }