[ThinLTO] Fix ThinLTOCodegenerator to export llvm.used symbols

[steven_wu]

Summary

ThinLTOCodeGenerator currently does not preserve llvm.used symbols and
it can internalize them. In order to pass the necessary information to the
legacy ThinLTOCodeGenerator, the input to the code generator is
rewritten to be based on lto::InputFile.

This fixes: PR41236
rdar://problem/49293439

[steven_wu]

Also do note, I did a quick benchmark on thin-link time for libLTO.dylib. thin-link was 2.6s before the change vs. 4.0s after the change. Building IRSymtab during thin-link does introduce non-negligible overhead. It might be possible to build them async in threads to reduce the overhead.

[tejohnson]

In D60226#1453856, @steven_wu wrote:

Also do note, I did a quick benchmark on thin-link time for libLTO.dylib. thin-link was 2.6s before the change vs. 4.0s after the change. Building IRSymtab during thin-link does introduce non-negligible overhead. It might be possible to build them async in threads to reduce the overhead.

@pcc is this surprising? Typically we should just be loading it out of the bitcode file. There are cases where it has to be rebuilt, but theoretically those should be exceptions.

[pcc]

In D60226#1455477, @tejohnson wrote:

In D60226#1453856, @steven_wu wrote:

Also do note, I did a quick benchmark on thin-link time for libLTO.dylib. thin-link was 2.6s before the change vs. 4.0s after the change. Building IRSymtab during thin-link does introduce non-negligible overhead. It might be possible to build them async in threads to reduce the overhead.

@pcc is this surprising? Typically we should just be loading it out of the bitcode file. There are cases where it has to be rebuilt, but theoretically those should be exceptions.

Yes, we normally just load it out of the bitcode file. The irsymtab only needs to be rebuilt when the version of the producer and the consumer do not match. You might be hitting the "upgrade" path here: http://llvm-cs.pcc.me.uk/lib/Object/IRSymtab.cpp#330

[steven_wu]

In D60226#1455485, @pcc wrote:

In D60226#1455477, @tejohnson wrote:

In D60226#1453856, @steven_wu wrote:

Also do note, I did a quick benchmark on thin-link time for libLTO.dylib. thin-link was 2.6s before the change vs. 4.0s after the change. Building IRSymtab during thin-link does introduce non-negligible overhead. It might be possible to build them async in threads to reduce the overhead.

@pcc is this surprising? Typically we should just be loading it out of the bitcode file. There are cases where it has to be rebuilt, but theoretically those should be exceptions.

Yes, we normally just load it out of the bitcode file. The irsymtab only needs to be rebuilt when the version of the producer and the consumer do not match. You might be hitting the "upgrade" path here: http://llvm-cs.pcc.me.uk/lib/Object/IRSymtab.cpp#330

This is likely. I didn't bootstrap the build. I will try again with a matching version.

[steven_wu]

In D60226#1455688, @steven_wu wrote:

In D60226#1455485, @pcc wrote:

In D60226#1455477, @tejohnson wrote:

In D60226#1453856, @steven_wu wrote:

Also do note, I did a quick benchmark on thin-link time for libLTO.dylib. thin-link was 2.6s before the change vs. 4.0s after the change. Building IRSymtab during thin-link does introduce non-negligible overhead. It might be possible to build them async in threads to reduce the overhead.

@pcc is this surprising? Typically we should just be loading it out of the bitcode file. There are cases where it has to be rebuilt, but theoretically those should be exceptions.

Yes, we normally just load it out of the bitcode file. The irsymtab only needs to be rebuilt when the version of the producer and the consumer do not match. You might be hitting the "upgrade" path here: http://llvm-cs.pcc.me.uk/lib/Object/IRSymtab.cpp#330

This is likely. I didn't bootstrap the build. I will try again with a matching version.

Without hitting the upgrade path, the time is almost negligible. Thanks for pointing that out.

[tejohnson]

I have mostly small suggestions below. Overall, it looks pretty good and straightforward.

llvm/lib/LTO/LTO.cpp

423

Suggest renaming to something more generic like "getSingleBitcodeModule", and just change the assert text to "Expect only one module". In case we want to use this interface for something other than that client in the future.

llvm/tools/llvm-lto/llvm-lto.cpp

462

The new patch loses this option, which seems to be used by a couple tests under test/ThinLTO/X86.

480

Maybe combine this with loadFile, since they are always used together and the Buffer from loadFile doesn't seem to be used elsewhere?

[steven_wu]

llvm/tools/llvm-lto/llvm-lto.cpp

462

I will add this back.

480

This is just to extend the lifetime of buffer till module is created from InputFile.

[steven_wu]

Address review feedback

[tejohnson]

Just noticed something else looking through it again, question below.

llvm/include/llvm/LTO/legacy/ThinLTOCodeGenerator.h

254–255

This and the functions below take an InputFile pointer with a nullptr default - why the default (and null check in these routines), since it looks like it is always being passed (other than emitImports, note on that further down). Can you remove the default parameter value, and also make this take a reference and remove the null checks?

llvm/tools/llvm-lto/llvm-lto.cpp

624–626

This should now pass Input.get()

[steven_wu]

Review feedback.

[tejohnson]

lgtm

[steven_wu]

I reverted this because the old thinLTO test cases need to be updated to have data layout as a requirement to use IRSymtab. I will start a separate review to address all those issues.

llvm/include/llvm/LTO/LTO.h

public:
   public:
     Symbol(const irsymtab::Symbol &S) : irsymtab::Symbol(S) {}
 
     using irsymtab::Symbol::isUndefined;
     using irsymtab::Symbol::isCommon;
     using irsymtab::Symbol::isWeak;
     using irsymtab::Symbol::isIndirect;
     using irsymtab::Symbol::getName;
+    using irsymtab::Symbol::getIRName;
     using irsymtab::Symbol::getVisibility;
     using irsymtab::Symbol::canBeOmittedFromSymbolTable;
     using irsymtab::Symbol::isTLS;
     using irsymtab::Symbol::getComdatIndex;
     using irsymtab::Symbol::getCommonSize;
     using irsymtab::Symbol::getCommonAlignment;
     using irsymtab::Symbol::getCOFFWeakExternalFallback;
     using irsymtab::Symbol::getSectionName;
public:
   StringRef getTargetTriple() const { return TargetTriple; }
 
   /// Returns the source file path specified at compile time.
   StringRef getSourceFileName() const { return SourceFileName; }
 
   // Returns a table with all the comdats used by this file.
   ArrayRef<StringRef> getComdatTable() const { return ComdatTable; }
 
+  // Returns the only BitcodeModule from InputFile.
+  BitcodeModule &getSingleBitcodeModule();
+
 private:
   ArrayRef<Symbol> module_symbols(unsigned I) const {
     const auto &Indices = ModuleSymIndices[I];
     return {Symbols.data() + Indices.first, Symbols.data() + Indices.second};
   }
 };
 
 /// This class wraps an output stream for a native object. Most clients should

llvm/include/llvm/LTO/legacy/ThinLTOCodeGenerator.h

 
 #ifndef LLVM_LTO_THINLTOCODEGENERATOR_H
 #define LLVM_LTO_THINLTOCODEGENERATOR_H
 
 #include "llvm-c/lto.h"
 #include "llvm/ADT/StringSet.h"
 #include "llvm/ADT/Triple.h"
 #include "llvm/IR/ModuleSummaryIndex.h"
+#include "llvm/LTO/LTO.h"
 #include "llvm/Support/CachePruning.h"
 #include "llvm/Support/CodeGen.h"
 #include "llvm/Support/MemoryBuffer.h"
 #include "llvm/Target/TargetOptions.h"
 
 #include <string>
 
 namespace llvm {
 class StringRef;
 class LLVMContext;
 class TargetMachine;
 
-/// Wrapper around MemoryBufferRef, owning the identifier
-class ThinLTOBuffer {
-  std::string OwnedIdentifier;
-  StringRef Buffer;
-
-public:
-  ThinLTOBuffer(StringRef Buffer, StringRef Identifier)
-      : OwnedIdentifier(Identifier), Buffer(Buffer) {}
-
-  MemoryBufferRef getMemBuffer() const {
-    return MemoryBufferRef(Buffer,
-                           {OwnedIdentifier.c_str(), OwnedIdentifier.size()});
-  }
-  StringRef getBuffer() const { return Buffer; }
-  StringRef getBufferIdentifier() const { return OwnedIdentifier; }
-};
-
 /// Helper to gather options relevant to the target machine creation
 struct TargetMachineBuilder {
   Triple TheTriple;
   std::string MCpu;
   std::string MAttr;
   TargetOptions Options;
   Optional<Reloc::Model> RelocModel;
   CodeGenOpt::Level CGOptLevel = CodeGenOpt::Aggressive;
public:
    */
   std::unique_ptr<ModuleSummaryIndex> linkCombinedIndex();
 
   /**
    * Perform promotion and renaming of exported internal functions,
    * and additionally resolve weak and linkonce symbols.
    * Index is updated to reflect linkage changes from weak resolution.
    */
-  void promote(Module &Module, ModuleSummaryIndex &Index);
+  void promote(Module &Module, ModuleSummaryIndex &Index,
+               const lto::InputFile &File);
 
   /**
    * Compute and emit the imported files for module at \p ModulePath.
    */
   void emitImports(Module &Module, StringRef OutputName,
-                   ModuleSummaryIndex &Index);
+                   ModuleSummaryIndex &Index,
+                   const lto::InputFile &File);
 
   /**
    * Perform cross-module importing for the module identified by
    * ModuleIdentifier.
    */
-  void crossModuleImport(Module &Module, ModuleSummaryIndex &Index);
+  void crossModuleImport(Module &Module, ModuleSummaryIndex &Index,
+                         const lto::InputFile &File);
 
   /**
    * Compute the list of summaries needed for importing into module.
    */
   void gatherImportedSummariesForModule(
       Module &Module, ModuleSummaryIndex &Index,
-      std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex);
+      std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex,
+      const lto::InputFile &File);
 
   /**
    * Perform internalization. Index is updated to reflect linkage changes.
    */
-  void internalize(Module &Module, ModuleSummaryIndex &Index);
+  void internalize(Module &Module, ModuleSummaryIndex &Index,
+                   const lto::InputFile &File);
 
   /**
    * Perform post-importing ThinLTO optimizations.
    */
   void optimize(Module &Module);
 
   /**@}*/
 
 private:
   /// Helper factory to build a TargetMachine
   TargetMachineBuilder TMBuilder;
 
   /// Vector holding the in-memory buffer containing the produced binaries, when
   /// SavedObjectsDirectoryPath isn't set.
   std::vector<std::unique_ptr<MemoryBuffer>> ProducedBinaries;
 
   /// Path to generated files in the supplied SavedObjectsDirectoryPath if any.
   std::vector<std::string> ProducedBinaryFiles;
 
   /// Vector holding the input buffers containing the bitcode modules to
   /// process.
-  std::vector<ThinLTOBuffer> Modules;
+  std::vector<std::unique_ptr<lto::InputFile>> Modules;
 
   /// Set of symbols that need to be preserved outside of the set of bitcode
   /// files.
   StringSet<> PreservedSymbols;
 
   /// Set of symbols that are cross-referenced between bitcode files.
   StringSet<> CrossReferencedSymbols;
 

llvm/lib/LTO/LTO.cpp

Expected<std::unique_ptr<InputFile>> InputFile::create(MemoryBufferRef Object) {
   File->Strtab = std::move(FOrErr->Strtab);
   return std::move(File);
 }
 
 StringRef InputFile::getName() const {
   return Mods[0].getModuleIdentifier();
 }
 
+BitcodeModule &InputFile::getSingleBitcodeModule() {
+  assert(Mods.size() == 1 && "Expect only one bitcode module");
+  return Mods[0];
+}
+
 LTO::RegularLTOState::RegularLTOState(unsigned ParallelCodeGenParallelismLevel,
                                       Config &Conf)
     : ParallelCodeGenParallelismLevel(ParallelCodeGenParallelismLevel),
       Ctx(Conf), CombinedModule(llvm::make_unique<Module>("ld-temp.o", Ctx)),
       Mover(llvm::make_unique<IRMover>(*CombinedModule)) {}
 
 LTO::ThinLTOState::ThinLTOState(ThinBackend Backend)
     : Backend(Backend), CombinedIndex(/*HaveGVs*/ false) {

llvm/lib/LTO/ThinLTOCodeGenerator.cpp

static void computePrevailingCopies(
 
   for (auto &I : Index) {
     if (HasMultipleCopies(I.second.SummaryList))
       PrevailingCopy[I.first] =
           getFirstDefinitionForLinker(I.second.SummaryList);
   }
 }
 
-static StringMap<MemoryBufferRef>
-generateModuleMap(const std::vector<ThinLTOBuffer> &Modules) {
-  StringMap<MemoryBufferRef> ModuleMap;
-  for (auto &ModuleBuffer : Modules) {
-    assert(ModuleMap.find(ModuleBuffer.getBufferIdentifier()) ==
-               ModuleMap.end() &&
+static StringMap<lto::InputFile *>
+generateModuleMap(std::vector<std::unique_ptr<lto::InputFile>> &Modules) {
+  StringMap<lto::InputFile *> ModuleMap;
+  for (auto &M : Modules) {
+    assert(ModuleMap.find(M->getName()) == ModuleMap.end() &&
            "Expect unique Buffer Identifier");
-    ModuleMap[ModuleBuffer.getBufferIdentifier()] = ModuleBuffer.getMemBuffer();
+    ModuleMap[M->getName()] = M.get();
   }
   return ModuleMap;
 }
 
 static void promoteModule(Module &TheModule, const ModuleSummaryIndex &Index) {
   if (renameModuleForThinLTO(TheModule, Index))
     report_fatal_error("renameModuleForThinLTO failed");
 }
if (verifyModule(TheModule, &dbgs(), &BrokenDebugInfo))
     report_fatal_error("Broken module found, compilation aborted!");
   if (BrokenDebugInfo) {
     TheModule.getContext().diagnose(ThinLTODiagnosticInfo(
         "Invalid debug info found, debug info will be stripped", DS_Warning));
     StripDebugInfo(TheModule);
   }
 }
 
-static std::unique_ptr<Module>
-loadModuleFromBuffer(const MemoryBufferRef &Buffer, LLVMContext &Context,
-                     bool Lazy, bool IsImporting) {
+static std::unique_ptr<Module> loadModuleFromInput(lto::InputFile *Input,
+                                                   LLVMContext &Context,
+                                                   bool Lazy,
+                                                   bool IsImporting) {
+  auto &Mod = Input->getSingleBitcodeModule();
   SMDiagnostic Err;
   Expected<std::unique_ptr<Module>> ModuleOrErr =
-      Lazy
-          ? getLazyBitcodeModule(Buffer, Context,
-                                 /* ShouldLazyLoadMetadata */ true, IsImporting)
-          : parseBitcodeFile(Buffer, Context);
+      Lazy ? Mod.getLazyModule(Context,
+           : Mod.parseModule(Context);
   if (!ModuleOrErr) {
     handleAllErrors(ModuleOrErr.takeError(), [&](ErrorInfoBase &EIB) {
-      SMDiagnostic Err = SMDiagnostic(Buffer.getBufferIdentifier(),
+      SMDiagnostic Err = SMDiagnostic(Mod.getModuleIdentifier(),
                                       SourceMgr::DK_Error, EIB.message());
       Err.print("ThinLTO", errs());
     });
     report_fatal_error("Can't load module, abort.");
   }
   if (!Lazy)
     verifyLoadedModule(*ModuleOrErr.get());
-  return std::move(ModuleOrErr.get());
+  return std::move(*ModuleOrErr);
 }
 
 static void
 crossImportIntoModule(Module &TheModule, const ModuleSummaryIndex &Index,
-                      StringMap<MemoryBufferRef> &ModuleMap,
+                      StringMap<lto::InputFile*> &ModuleMap,
                       const FunctionImporter::ImportMapTy &ImportList) {
   auto Loader = [&](StringRef Identifier) {
-    return loadModuleFromBuffer(ModuleMap[Identifier], TheModule.getContext(),
-                                /*Lazy=*/true, /*IsImporting*/ true);
+    auto &Input = ModuleMap[Identifier];
+    return loadModuleFromInput(Input, TheModule.getContext(),
   };
 
   FunctionImporter Importer(Index, Loader);
   Expected<bool> Result = Importer.importFunctions(TheModule, ImportList);
   if (!Result) {
     handleAllErrors(Result.takeError(), [&](ErrorInfoBase &EIB) {
       SMDiagnostic Err = SMDiagnostic(TheModule.getModuleIdentifier(),
                                       SourceMgr::DK_Error, EIB.message());
static void optimizeModule(Module &TheModule, TargetMachine &TM,
   PM.add(createTargetTransformInfoWrapperPass(TM.getTargetIRAnalysis()));
 
   // Add optimizations
   PMB.populateThinLTOPassManager(PM);
 
   PM.run(TheModule);
 }
 
+static void
+addUsedSymbolToPreservedGUID(const lto::InputFile &File,
+                             DenseSet<GlobalValue::GUID> &PreservedGUID) {
+  for (const auto &Sym : File.symbols()) {
+    if (Sym.isUsed())
+      PreservedGUID.insert(GlobalValue::getGUID(Sym.getIRName()));
+  }
+}
+
 // Convert the PreservedSymbols map from "Name" based to "GUID" based.
 static DenseSet<GlobalValue::GUID>
 computeGUIDPreservedSymbols(const StringSet<> &PreservedSymbols,
                             const Triple &TheTriple) {
   DenseSet<GlobalValue::GUID> GUIDPreservedSymbols(PreservedSymbols.size());
   for (auto &Entry : PreservedSymbols) {
     StringRef Name = Entry.first();
     if (TheTriple.isOSBinFormatMachO() && Name.size() > 0 && Name[0] == '_')
void write(const MemoryBuffer &OutputBuffer) {
     EC = sys::fs::rename(TempFilename, EntryPath);
     if (EC)
       sys::fs::remove(TempFilename);
   }
 };
 
 static std::unique_ptr<MemoryBuffer>
 ProcessThinLTOModule(Module &TheModule, ModuleSummaryIndex &Index,
-                     StringMap<MemoryBufferRef> &ModuleMap, TargetMachine &TM,
+                     StringMap<lto::InputFile *> &ModuleMap, TargetMachine &TM,
                      const FunctionImporter::ImportMapTy &ImportList,
                      const FunctionImporter::ExportSetTy &ExportList,
                      const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols,
                      const GVSummaryMapTy &DefinedGlobals,
                      const ThinLTOCodeGenerator::CachingOptions &CacheOptions,
                      bool DisableCodeGen, StringRef SaveTempsDir,
                      bool Freestanding, unsigned OptLevel, unsigned count) {
 
else if (TheTriple.getArch() == llvm::Triple::aarch64)
       TMBuilder.MCpu = "cyclone";
   }
   TMBuilder.TheTriple = std::move(TheTriple);
 }
 
 } // end anonymous namespace
 
 void ThinLTOCodeGenerator::addModule(StringRef Identifier, StringRef Data) {
-  ThinLTOBuffer Buffer(Data, Identifier);
-  LLVMContext Context;
-  StringRef TripleStr;
-  ErrorOr<std::string> TripleOrErr = expectedToErrorOrAndEmitErrors(
-      Context, getBitcodeTargetTriple(Buffer.getMemBuffer()));
+  MemoryBufferRef Buffer(Data, Identifier);
 
-  if (TripleOrErr)
-    TripleStr = *TripleOrErr;
+  auto InputOrError = lto::InputFile::create(Buffer);
+  if (!InputOrError)
+    report_fatal_error("ThinLTO cannot create input file");
 
+  auto TripleStr = (*InputOrError)->getTargetTriple();
   Triple TheTriple(TripleStr);
 
   if (Modules.empty())
     initTMBuilder(TMBuilder, Triple(TheTriple));
   else if (TMBuilder.TheTriple != TheTriple) {
     if (!TMBuilder.TheTriple.isCompatibleWith(TheTriple))
       report_fatal_error("ThinLTO modules with incompatible triples not "
                          "supported");
     initTMBuilder(TMBuilder, Triple(TMBuilder.TheTriple.merge(TheTriple)));
   }
 
-  Modules.push_back(Buffer);
+  Modules.emplace_back(std::move(*InputOrError));
 }
 
 void ThinLTOCodeGenerator::preserveSymbol(StringRef Name) {
   PreservedSymbols.insert(Name);
 }
 
 void ThinLTOCodeGenerator::crossReferenceSymbol(StringRef Name) {
   // FIXME: At the moment, we don't take advantage of this extra information,

 /**
  * Produce the combined summary index from all the bitcode files:
  * "thin-link".
  */
 std::unique_ptr<ModuleSummaryIndex> ThinLTOCodeGenerator::linkCombinedIndex() {
   std::unique_ptr<ModuleSummaryIndex> CombinedIndex =
       llvm::make_unique<ModuleSummaryIndex>(/*HaveGVs=*/false);
   uint64_t NextModuleId = 0;
-  for (auto &ModuleBuffer : Modules) {
-    if (Error Err = readModuleSummaryIndex(ModuleBuffer.getMemBuffer(),
-                                           *CombinedIndex, NextModuleId++)) {
+  for (auto &Mod : Modules) {
+    auto &M = Mod->getSingleBitcodeModule();
+    if (Error Err =
+            M.readSummary(*CombinedIndex, Mod->getName(), NextModuleId++)) {
       // FIXME diagnose
       logAllUnhandledErrors(
           std::move(Err), errs(),
           "error: can't create module summary index for buffer: ");
       return nullptr;
     }
   }
   return CombinedIndex;
static void computeDeadSymbolsInIndex(
   computeDeadSymbolsWithConstProp(Index, GUIDPreservedSymbols, isPrevailing,
                                   /* ImportEnabled = */ true);
 }
 
 /**
  * Perform promotion and renaming of exported internal functions.
  * Index is updated to reflect linkage changes from weak resolution.
  */
-void ThinLTOCodeGenerator::promote(Module &TheModule,
-                                   ModuleSummaryIndex &Index) {
+void ThinLTOCodeGenerator::promote(Module &TheModule, ModuleSummaryIndex &Index,
+                                   const lto::InputFile &File) {
   auto ModuleCount = Index.modulePaths().size();
   auto ModuleIdentifier = TheModule.getModuleIdentifier();
 
   // Collect for each module the list of function it defines (GUID -> Summary).
   StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries;
   Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries);
 
   // Convert the preserved symbols set from string to GUID
   auto GUIDPreservedSymbols = computeGUIDPreservedSymbols(
       PreservedSymbols, Triple(TheModule.getTargetTriple()));
 
+  // Add used symbol to the preserved symbols.
+  addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols);
+
   // Compute "dead" symbols, we don't want to import/export these!
   computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols);
 
   // Generate import/export list
   StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount);
   StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount);
   ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists,
                            ExportLists);
 
   // Resolve prevailing symbols
   StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> ResolvedODR;
   resolvePrevailingInIndex(Index, ResolvedODR);
 
-  thinLTOResolvePrevailingInModule(
-      TheModule, ModuleToDefinedGVSummaries[ModuleIdentifier]);
-
   // Promote the exported values in the index, so that they are promoted
   // in the module.
   internalizeAndPromoteInIndex(ExportLists, GUIDPreservedSymbols, Index);
 
   promoteModule(TheModule, Index);
+
+  thinLTOResolvePrevailingInModule(
+      TheModule, ModuleToDefinedGVSummaries[ModuleIdentifier]);
 }
 
 /**
  * Perform cross-module importing for the module identified by ModuleIdentifier.
  */
 void ThinLTOCodeGenerator::crossModuleImport(Module &TheModule,
-                                             ModuleSummaryIndex &Index) {
+                                             ModuleSummaryIndex &Index,
+                                             const lto::InputFile &File) {
   auto ModuleMap = generateModuleMap(Modules);
   auto ModuleCount = Index.modulePaths().size();
 
   // Collect for each module the list of function it defines (GUID -> Summary).
   StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount);
   Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries);
 
   // Convert the preserved symbols set from string to GUID
   auto GUIDPreservedSymbols = computeGUIDPreservedSymbols(
       PreservedSymbols, Triple(TheModule.getTargetTriple()));
 
+  addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols);
+
   // Compute "dead" symbols, we don't want to import/export these!
   computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols);
 
   // Generate import/export list
   StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount);
   StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount);
   ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists,
                            ExportLists);
   auto &ImportList = ImportLists[TheModule.getModuleIdentifier()];
 
   crossImportIntoModule(TheModule, Index, ModuleMap, ImportList);
 }
 
 /**
  * Compute the list of summaries needed for importing into module.
  */
 void ThinLTOCodeGenerator::gatherImportedSummariesForModule(
     Module &TheModule, ModuleSummaryIndex &Index,
-    std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex) {
+    std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex,
+    const lto::InputFile &File) {
   auto ModuleCount = Index.modulePaths().size();
   auto ModuleIdentifier = TheModule.getModuleIdentifier();
 
   // Collect for each module the list of function it defines (GUID -> Summary).
   StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount);
   Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries);
 
   // Convert the preserved symbols set from string to GUID
   auto GUIDPreservedSymbols = computeGUIDPreservedSymbols(
       PreservedSymbols, Triple(TheModule.getTargetTriple()));
 
+  addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols);
+
   // Compute "dead" symbols, we don't want to import/export these!
   computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols);
 
   // Generate import/export list
   StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount);
   StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount);
   ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists,
                            ExportLists);
 
   llvm::gatherImportedSummariesForModule(
       ModuleIdentifier, ModuleToDefinedGVSummaries,
       ImportLists[ModuleIdentifier], ModuleToSummariesForIndex);
 }
 
 /**
  * Emit the list of files needed for importing into module.
  */
 void ThinLTOCodeGenerator::emitImports(Module &TheModule, StringRef OutputName,
-                                       ModuleSummaryIndex &Index) {
+                                       ModuleSummaryIndex &Index,
+                                       const lto::InputFile &File) {
   auto ModuleCount = Index.modulePaths().size();
   auto ModuleIdentifier = TheModule.getModuleIdentifier();
 
   // Collect for each module the list of function it defines (GUID -> Summary).
   StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount);
   Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries);
 
   // Convert the preserved symbols set from string to GUID
   auto GUIDPreservedSymbols = computeGUIDPreservedSymbols(
       PreservedSymbols, Triple(TheModule.getTargetTriple()));
 
+  addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols);
+
   // Compute "dead" symbols, we don't want to import/export these!
   computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols);
 
   // Generate import/export list
   StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount);
   StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount);
   ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists,
                            ExportLists);
if ((EC = EmitImportsFiles(ModuleIdentifier, OutputName,
     report_fatal_error(Twine("Failed to open ") + OutputName +
                        " to save imports lists\n");
 }
 
 /**
  * Perform internalization. Index is updated to reflect linkage changes.
  */
 void ThinLTOCodeGenerator::internalize(Module &TheModule,
-                                       ModuleSummaryIndex &Index) {
+                                       ModuleSummaryIndex &Index,
+                                       const lto::InputFile &File) {
   initTMBuilder(TMBuilder, Triple(TheModule.getTargetTriple()));
   auto ModuleCount = Index.modulePaths().size();
   auto ModuleIdentifier = TheModule.getModuleIdentifier();
 
   // Convert the preserved symbols set from string to GUID
   auto GUIDPreservedSymbols =
       computeGUIDPreservedSymbols(PreservedSymbols, TMBuilder.TheTriple);
 
+  addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols);
+
   // Collect for each module the list of function it defines (GUID -> Summary).
   StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount);
   Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries);
 
   // Compute "dead" symbols, we don't want to import/export these!
   computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols);
 
   // Generate import/export list
if (!IsDir)
       report_fatal_error("Unexistent dir: '" + SavedObjectsDirectoryPath + "'");
     ProducedBinaryFiles.resize(Modules.size());
   }
 
   if (CodeGenOnly) {
     // Perform only parallel codegen and return.
     ThreadPool Pool;
     int count = 0;
-    for (auto &ModuleBuffer : Modules) {
+    for (auto &Mod : Modules) {
       Pool.async([&](int count) {
         LLVMContext Context;
         Context.setDiscardValueNames(LTODiscardValueNames);
 
         // Parse module now
-        auto TheModule =
-            loadModuleFromBuffer(ModuleBuffer.getMemBuffer(), Context, false,
+        auto TheModule = loadModuleFromInput(Mod.get(), Context, false,
                                              /*IsImporting*/ false);
 
         // CodeGen
         auto OutputBuffer = codegenModule(*TheModule, *TMBuilder.create());
         if (SavedObjectsDirectoryPath.empty())
           ProducedBinaries[count] = std::move(OutputBuffer);
         else
           ProducedBinaryFiles[count] = writeGeneratedObject(
               count, "", SavedObjectsDirectoryPath, *OutputBuffer);
void ThinLTOCodeGenerator::run() {
   StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount);
   Index->collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries);
 
   // Convert the preserved symbols set from string to GUID, this is needed for
   // computing the caching hash and the internalization.
   auto GUIDPreservedSymbols =
       computeGUIDPreservedSymbols(PreservedSymbols, TMBuilder.TheTriple);
 
+  // Add used symbol from inputs to the preserved symbols.
+  for (const auto &M : Modules)
+    addUsedSymbolToPreservedGUID(*M, GUIDPreservedSymbols);
+
   // Compute "dead" symbols, we don't want to import/export these!
   computeDeadSymbolsInIndex(*Index, GUIDPreservedSymbols);
 
   // Synthesize entry counts for functions in the combined index.
   computeSyntheticCounts(*Index);
 
   // Collect the import/export lists for all modules from the call-graph in the
   // combined index.
void ThinLTOCodeGenerator::run() {
   // internalized (because they aren't exported or preserved as per callback).
   // Changes are made in the index, consumed in the ThinLTO backends.
   internalizeAndPromoteInIndex(ExportLists, GUIDPreservedSymbols, *Index);
 
   // Make sure that every module has an entry in the ExportLists, ImportList,
   // GVSummary and ResolvedODR maps to enable threaded access to these maps
   // below.
   for (auto &Module : Modules) {
-    auto ModuleIdentifier = Module.getBufferIdentifier();
+    auto ModuleIdentifier = Module->getName();
     ExportLists[ModuleIdentifier];
     ImportLists[ModuleIdentifier];
     ResolvedODR[ModuleIdentifier];
     ModuleToDefinedGVSummaries[ModuleIdentifier];
   }
 
   // Compute the ordering we will process the inputs: the rough heuristic here
   // is to sort them per size so that the largest module get schedule as soon as
   // possible. This is purely a compile-time optimization.
   std::vector<int> ModulesOrdering;
   ModulesOrdering.resize(Modules.size());
   std::iota(ModulesOrdering.begin(), ModulesOrdering.end(), 0);
   llvm::sort(ModulesOrdering, [&](int LeftIndex, int RightIndex) {
-    auto LSize = Modules[LeftIndex].getBuffer().size();
-    auto RSize = Modules[RightIndex].getBuffer().size();
+    auto LSize =
+        Modules[LeftIndex]->getSingleBitcodeModule().getBuffer().size();
+    auto RSize =
+        Modules[RightIndex]->getSingleBitcodeModule().getBuffer().size();
     return LSize > RSize;
   });
 
   // Parallel optimizer + codegen
   {
     ThreadPool Pool(ThreadCount);
     for (auto IndexCount : ModulesOrdering) {
-      auto &ModuleBuffer = Modules[IndexCount];
+      auto &Mod = Modules[IndexCount];
       Pool.async([&](int count) {
-        auto ModuleIdentifier = ModuleBuffer.getBufferIdentifier();
+        auto ModuleIdentifier = Mod->getName();
         auto &ExportList = ExportLists[ModuleIdentifier];
 
         auto &DefinedGVSummaries = ModuleToDefinedGVSummaries[ModuleIdentifier];
 
         // The module may be cached, this helps handling it.
         ModuleCacheEntry CacheEntry(CacheOptions.Path, *Index, ModuleIdentifier,
                                     ImportLists[ModuleIdentifier], ExportList,
                                     ResolvedODR[ModuleIdentifier],
for (auto IndexCount : ModulesOrdering) {
             LTOPassRemarksWithHotness, count);
         if (!DiagFileOrErr) {
           errs() << "Error: " << toString(DiagFileOrErr.takeError()) << "\n";
           report_fatal_error("ThinLTO: Can't get an output file for the "
                              "remarks");
         }
 
         // Parse module now
-        auto TheModule =
-            loadModuleFromBuffer(ModuleBuffer.getMemBuffer(), Context, false,
+        auto TheModule = loadModuleFromInput(Mod.get(), Context, false,
                                              /*IsImporting*/ false);
 
         // Save temps: original file.
         saveTempBitcode(*TheModule, SaveTempsDir, count, ".0.original.bc");
 
         auto &ImportList = ImportLists[ModuleIdentifier];
         // Run the main process now, and generates a binary
         auto OutputBuffer = ProcessThinLTOModule(
             *TheModule, *Index, ModuleMap, *TMBuilder.create(), ImportList,

llvm/test/LTO/X86/Inputs/thinlto-internalize-used2.ll

+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.15.0"
+
+define i32 @main() {
+entry:
+  %call = call i32 @bar()
+  ret i32 0
+}
+
+declare i32 @bar()

llvm/test/LTO/X86/thinlto-internalize-used.ll

+; RUN: opt -module-summary -o %t.bc %s
+; RUN: opt -module-summary -o %t-main.bc %S/Inputs/thinlto-internalize-used2.ll
+; RUN: llvm-lto -thinlto-action=thinlink %t.bc %t-main.bc -o %t-index.bc
+; RUN: llvm-lto -thinlto-action=promote -thinlto-index %t-index.bc %t.bc -o %t.promote.bc
+; RUN: llvm-dis %t.promote.bc -o - | FileCheck %s
+
+target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-apple-macosx10.15.0"
+
[e-mail] = appending global [1 x i8*] [i8* bitcast (i32 ()* @foo to i8*)], section "llvm.metadata"
+
+; Make sure foo is not internalized.
+; CHECK: define i32 @foo()
+define i32 @foo() {
+  ret i32 0
+}
+
+define hidden i32 @bar() {
+  ret i32 0
+}
+

llvm/tools/llvm-lto/llvm-lto.cpp

 std::unique_ptr<ModuleSummaryIndex> loadCombinedIndex() {
   if (ThinLTOIndex.empty())
     report_fatal_error("Missing -thinlto-index for ThinLTO promotion stage");
   ExitOnError ExitOnErr("llvm-lto: error loading file '" + ThinLTOIndex +
                         "': ");
   return ExitOnErr(getModuleSummaryIndexForFile(ThinLTOIndex));
 }
 
-static std::unique_ptr<Module> loadModule(StringRef Filename,
-                                          LLVMContext &Ctx) {
-  SMDiagnostic Err;
-  std::unique_ptr<Module> M(parseIRFile(Filename, Err, Ctx));
-  if (!M) {
-    Err.print("llvm-lto", errs());
-    report_fatal_error("Can't load module for file " + Filename);
-  }
-  maybeVerifyModule(*M);
+static std::unique_ptr<MemoryBuffer> loadFile(StringRef Filename) {
+  ExitOnError ExitOnErr("llvm-lto: error loading file '" + Filename.str() +
+                        "': ");
+  return ExitOnErr(errorOrToExpected(MemoryBuffer::getFileOrSTDIN(Filename)));
 
+static std::unique_ptr<lto::InputFile> loadInputFile(MemoryBufferRef Buffer) {
+  ExitOnError ExitOnErr("llvm-lto: error loading input '" +
+                        Buffer.getBufferIdentifier().str() + "': ");
+  return ExitOnErr(lto::InputFile::create(Buffer));
+}
+
+static std::unique_ptr<Module> loadModuleFromInput(lto::InputFile &File,
+                                                   LLVMContext &CTX) {
+  auto &Mod = File.getSingleBitcodeModule();
+  auto ModuleOrErr = Mod.parseModule(CTX);
+  if (!ModuleOrErr) {
+    handleAllErrors(ModuleOrErr.takeError(), [&](ErrorInfoBase &EIB) {
+      SMDiagnostic Err = SMDiagnostic(Mod.getModuleIdentifier(),
+                                      SourceMgr::DK_Error, EIB.message());
+      Err.print("llvm-lto", errs());
+    });
+    report_fatal_error("Can't load module, abort.");
+  }
+  maybeVerifyModule(**ModuleOrErr);
   if (ThinLTOModuleId.getNumOccurrences()) {
     if (InputFilenames.size() != 1)
       report_fatal_error("Can't override the module id for multiple files");
-    M->setModuleIdentifier(ThinLTOModuleId);
+    (*ModuleOrErr)->setModuleIdentifier(ThinLTOModuleId);
   }
-  return M;
+  return std::move(*ModuleOrErr);
 }
 
 static void writeModuleToFile(Module &TheModule, StringRef Filename) {
   std::error_code EC;
   raw_fd_ostream OS(Filename, EC, sys::fs::OpenFlags::F_None);
   error(EC, "error opening the file '" + Filename + "'");
   maybeVerifyModule(TheModule);
   WriteBitcodeToFile(TheModule, OS, /* ShouldPreserveUseListOrder */ true);
if (InputFilenames.size() != 1 && !OutputFilename.empty())
                          "ones.");
 
     std::string OldPrefix, NewPrefix;
     getThinLTOOldAndNewPrefix(OldPrefix, NewPrefix);
 
     auto Index = loadCombinedIndex();
     for (auto &Filename : InputFilenames) {
       LLVMContext Ctx;
-      auto TheModule = loadModule(Filename, Ctx);
+      auto Buffer = loadFile(Filename);
+      auto Input = loadInputFile(Buffer->getMemBufferRef());
+      auto TheModule = loadModuleFromInput(*Input, Ctx);
 
       // Build a map of module to the GUIDs and summary objects that should
       // be written to its index.
       std::map<std::string, GVSummaryMapTy> ModuleToSummariesForIndex;
-      ThinGenerator.gatherImportedSummariesForModule(*TheModule, *Index,
-                                                     ModuleToSummariesForIndex);
+      ThinGenerator.gatherImportedSummariesForModule(
+          *TheModule, *Index, ModuleToSummariesForIndex, *Input);
 
       std::string OutputName = OutputFilename;
       if (OutputName.empty()) {
         OutputName = Filename + ".thinlto.bc";
       }
       OutputName = getThinLTOOutputFile(OutputName, OldPrefix, NewPrefix);
       std::error_code EC;
       raw_fd_ostream OS(OutputName, EC, sys::fs::OpenFlags::F_None);
if (InputFilenames.size() != 1 && !OutputFilename.empty())
                          "ones.");
 
     std::string OldPrefix, NewPrefix;
     getThinLTOOldAndNewPrefix(OldPrefix, NewPrefix);
 
     auto Index = loadCombinedIndex();
     for (auto &Filename : InputFilenames) {
       LLVMContext Ctx;
-      auto TheModule = loadModule(Filename, Ctx);
+      auto Buffer = loadFile(Filename);
+      auto Input = loadInputFile(Buffer->getMemBufferRef());
+      auto TheModule = loadModuleFromInput(*Input, Ctx);
       std::string OutputName = OutputFilename;
       if (OutputName.empty()) {
         OutputName = Filename + ".imports";
       }
-      OutputName = getThinLTOOutputFile(OutputName, OldPrefix, NewPrefix);
-      ThinGenerator.emitImports(*TheModule, OutputName, *Index);
+      OutputName =
+          getThinLTOOutputFile(OutputName, OldPrefix, NewPrefix);
+      ThinGenerator.emitImports(*TheModule, OutputName, *Index, *Input);
     }
   }
 
   /// Load the combined index from disk, then load every file referenced by
   /// the index and add them to the generator, finally perform the promotion
   /// on the files mentioned on the command line (these must match the index
   /// content).
   void promote() {
     if (InputFilenames.size() != 1 && !OutputFilename.empty())
       report_fatal_error("Can't handle a single output filename and multiple "
                          "input files, do not provide an output filename and "
                          "the output files will be suffixed from the input "
                          "ones.");
 
     auto Index = loadCombinedIndex();
     for (auto &Filename : InputFilenames) {
       LLVMContext Ctx;
-      auto TheModule = loadModule(Filename, Ctx);
+      auto Buffer = loadFile(Filename);
+      auto Input = loadInputFile(Buffer->getMemBufferRef());
+      auto TheModule = loadModuleFromInput(*Input, Ctx);
 
-      ThinGenerator.promote(*TheModule, *Index);
+      ThinGenerator.promote(*TheModule, *Index, *Input);
 
       std::string OutputName = OutputFilename;
       if (OutputName.empty()) {
         OutputName = Filename + ".thinlto.promoted.bc";
       }
       writeModuleToFile(*TheModule, OutputName);
     }
   }
void import() {
     auto Index = loadCombinedIndex();
     auto InputBuffers = loadAllFilesForIndex(*Index);
     for (auto &MemBuffer : InputBuffers)
       ThinGenerator.addModule(MemBuffer->getBufferIdentifier(),
                               MemBuffer->getBuffer());
 
     for (auto &Filename : InputFilenames) {
       LLVMContext Ctx;
-      auto TheModule = loadModule(Filename, Ctx);
+      auto Buffer = loadFile(Filename);
+      auto Input = loadInputFile(Buffer->getMemBufferRef());
+      auto TheModule = loadModuleFromInput(*Input, Ctx);
 
-      ThinGenerator.crossModuleImport(*TheModule, *Index);
+      ThinGenerator.crossModuleImport(*TheModule, *Index, *Input);
 
       std::string OutputName = OutputFilename;
       if (OutputName.empty()) {
         OutputName = Filename + ".thinlto.imported.bc";
       }
       writeModuleToFile(*TheModule, OutputName);
     }
   }
void internalize() {
     auto Index = loadCombinedIndex();
     auto InputBuffers = loadAllFilesForIndex(*Index);
     for (auto &MemBuffer : InputBuffers)
       ThinGenerator.addModule(MemBuffer->getBufferIdentifier(),
                               MemBuffer->getBuffer());
 
     for (auto &Filename : InputFilenames) {
       LLVMContext Ctx;
-      auto TheModule = loadModule(Filename, Ctx);
+      auto Buffer = loadFile(Filename);
+      auto Input = loadInputFile(Buffer->getMemBufferRef());
+      auto TheModule = loadModuleFromInput(*Input, Ctx);
 
-      ThinGenerator.internalize(*TheModule, *Index);
+      ThinGenerator.internalize(*TheModule, *Index, *Input);
 
       std::string OutputName = OutputFilename;
       if (OutputName.empty()) {
         OutputName = Filename + ".thinlto.internalized.bc";
       }
       writeModuleToFile(*TheModule, OutputName);
     }
   }
 
   void optimize() {
     if (InputFilenames.size() != 1 && !OutputFilename.empty())
       report_fatal_error("Can't handle a single output filename and multiple "
                          "input files, do not provide an output filename and "
                          "the output files will be suffixed from the input "
                          "ones.");
     if (!ThinLTOIndex.empty())
       errs() << "Warning: -thinlto-index ignored for optimize stage";
 
     for (auto &Filename : InputFilenames) {
       LLVMContext Ctx;
-      auto TheModule = loadModule(Filename, Ctx);
+      auto Buffer = loadFile(Filename);
+      auto Input = loadInputFile(Buffer->getMemBufferRef());
+      auto TheModule = loadModuleFromInput(*Input, Ctx);
 
       ThinGenerator.optimize(*TheModule);
 
       std::string OutputName = OutputFilename;
       if (OutputName.empty()) {
         OutputName = Filename + ".thinlto.imported.bc";
       }
       writeModuleToFile(*TheModule, OutputName);