[ThinLTO][Bitcode] Generate import type in bitcode

llvm/include/llvm/Bitcode/BitcodeWriter.h

@@ -102,7 +102,8 @@ class raw_ostream;
 
     void writeIndex(
         const ModuleSummaryIndex *Index,
-        const std::map<std::string, GVSummaryMapTy> *ModuleToSummariesForIndex);
+        const std::map<std::string, GVSummaryMapTy> *ModuleToSummariesForIndex,
+        const GVSummaryPtrSet *DecSummaries);
   };
 
   /// Write the specified module to the specified raw output stream.
@@ -147,10 +148,12 @@ class raw_ostream;
   /// where it will be written in a new bitcode block. This is used when
   /// writing the combined index file for ThinLTO. When writing a subset of the
   /// index for a distributed backend, provide the \p ModuleToSummariesForIndex
-  /// map.
+  /// map. \p DecSummaries specifies the set of summaries for which the
+  /// corresponding value should be imported as a declaration (prototype).
   void writeIndexToFile(const ModuleSummaryIndex &Index, raw_ostream &Out,
                         const std::map<std::string, GVSummaryMapTy>
-                            *ModuleToSummariesForIndex = nullptr);
+                            *ModuleToSummariesForIndex = nullptr,
+                        const GVSummaryPtrSet *DecSummaries = nullptr);
 
   /// If EmbedBitcode is set, save a copy of the llvm IR as data in the
   ///  __LLVM,__bitcode section (.llvmbc on non-MacOS).

llvm/include/llvm/IR/ModuleSummaryIndex.h

@@ -587,6 +587,10 @@ class GlobalValueSummary {
 
   void setImportKind(ImportKind IK) { Flags.ImportType = IK; }
 
+  GlobalValueSummary::ImportKind importType() const {
+    return static_cast<ImportKind>(Flags.ImportType);
+  }
+
   GlobalValue::VisibilityTypes getVisibility() const {
     return (GlobalValue::VisibilityTypes)Flags.Visibility;
   }
@@ -1272,6 +1276,9 @@ using ModulePathStringTableTy = StringMap<ModuleHash>;
 /// a particular module, and provide efficient access to their summary.
 using GVSummaryMapTy = DenseMap<GlobalValue::GUID, GlobalValueSummary *>;
 
+/// A set of global value summary pointers.
+using GVSummaryPtrSet = SmallPtrSet<GlobalValueSummary *, 4>;
+
 /// Map of a type GUID to type id string and summary (multimap used
 /// in case of GUID conflicts).
 using TypeIdSummaryMapTy =

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

@@ -271,12 +271,13 @@ class ThinLTOCodeGenerator {
                          const lto::InputFile &File);
 
   /**
-   * Compute the list of summaries needed for importing into module.
+   * Compute the list of summaries and the subset of declaration summaries
+   * needed for importing into module.
    */
   void gatherImportedSummariesForModule(
       Module &Module, ModuleSummaryIndex &Index,
       std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex,
-      const lto::InputFile &File);
+      GVSummaryPtrSet &DecSummaries, const lto::InputFile &File);
 
   /**
    * Perform internalization. Index is updated to reflect linkage changes.

llvm/include/llvm/Transforms/IPO/FunctionImport.h

@@ -31,9 +31,9 @@ class Module;
 /// based on the provided summary informations.
 class FunctionImporter {
 public:
-  /// Set of functions to import from a source module. Each entry is a set
-  /// containing all the GUIDs of all functions to import for a source module.
-  using FunctionsToImportTy = std::unordered_set<GlobalValue::GUID>;
+  /// The functions to import from a source module and their import type.
+  using FunctionsToImportTy =
+      DenseMap<GlobalValue::GUID, GlobalValueSummary::ImportKind>;
 
   /// The different reasons selectCallee will chose not to import a
   /// candidate.
@@ -99,8 +99,13 @@ class FunctionImporter {
   /// index's module path string table).
   using ImportMapTy = DenseMap<StringRef, FunctionsToImportTy>;
 
-  /// The set contains an entry for every global value the module exports.
-  using ExportSetTy = DenseSet<ValueInfo>;
+  /// The map contains an entry for every global value the module exports.
+  /// The key is ValueInfo, and the value indicates whether the definition
+  /// or declaration is visible to another module. If a function's definition is
+  /// visible to other modules, the global values this function referenced are
+  /// visible and shouldn't be internalized.
+  /// TODO: Rename to `ExportMapTy`.
+  using ExportSetTy = DenseMap<ValueInfo, GlobalValueSummary::ImportKind>;
 
   /// A function of this type is used to load modules referenced by the index.
   using ModuleLoaderTy =
@@ -207,11 +212,15 @@ bool convertToDeclaration(GlobalValue &GV);
 /// \p ModuleToSummariesForIndex will be populated with the needed summaries
 /// from each required module path. Use a std::map instead of StringMap to get
 /// stable order for bitcode emission.
+///
+/// \p DecSummaries will be popluated with the subset of of summary pointers
+/// that have 'declaration' import type among all summaries the module need.
 void gatherImportedSummariesForModule(
     StringRef ModulePath,
     const DenseMap<StringRef, GVSummaryMapTy> &ModuleToDefinedGVSummaries,
     const FunctionImporter::ImportMapTy &ImportList,
-    std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex);
+    std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex,
+    GVSummaryPtrSet &DecSummaries);
 
 /// Emit into \p OutputFilename the files module \p ModulePath will import from.
 std::error_code EmitImportsFiles(

llvm/lib/Bitcode/Writer/BitcodeWriter.cpp

@@ -428,6 +428,11 @@ class IndexBitcodeWriter : public BitcodeWriterBase {
   /// The combined index to write to bitcode.
   const ModuleSummaryIndex &Index;
 
+  /// When writing combined summaries, provides the set of global value
+  /// summaries for which the value (function, function alias, etc) should be
+  /// imported as a declaration.
+  const GVSummaryPtrSet *DecSummaries = nullptr;
+
   /// When writing a subset of the index for distributed backends, client
   /// provides a map of modules to the corresponding GUIDs/summaries to write.
   const std::map<std::string, GVSummaryMapTy> *ModuleToSummariesForIndex;
@@ -452,11 +457,16 @@ class IndexBitcodeWriter : public BitcodeWriterBase {
   /// Constructs a IndexBitcodeWriter object for the given combined index,
   /// writing to the provided \p Buffer. When writing a subset of the index
   /// for a distributed backend, provide a \p ModuleToSummariesForIndex map.
+  /// If provided, \p ModuleToDecSummaries specifies the set of summaries for
+  /// which the corresponding functions or aliased functions should be imported
+  /// as a declaration (but not definition) for each module.
   IndexBitcodeWriter(BitstreamWriter &Stream, StringTableBuilder &StrtabBuilder,
                      const ModuleSummaryIndex &Index,
+                     const GVSummaryPtrSet *DecSummaries = nullptr,
                      const std::map<std::string, GVSummaryMapTy>
                          *ModuleToSummariesForIndex = nullptr)
       : BitcodeWriterBase(Stream, StrtabBuilder), Index(Index),
+        DecSummaries(DecSummaries),
         ModuleToSummariesForIndex(ModuleToSummariesForIndex) {
     // Assign unique value ids to all summaries to be written, for use
     // in writing out the call graph edges. Save the mapping from GUID
@@ -1202,7 +1212,8 @@ static uint64_t getEncodedFFlags(FunctionSummary::FFlags Flags) {
 
 // Decode the flags for GlobalValue in the summary. See getDecodedGVSummaryFlags
 // in BitcodeReader.cpp.
-static uint64_t getEncodedGVSummaryFlags(GlobalValueSummary::GVFlags Flags) {
+static uint64_t getEncodedGVSummaryFlags(GlobalValueSummary::GVFlags Flags,
+                                         bool ImportAsDecl = false) {
   uint64_t RawFlags = 0;
 
   RawFlags |= Flags.NotEligibleToImport; // bool
@@ -1217,7 +1228,8 @@ static uint64_t getEncodedGVSummaryFlags(GlobalValueSummary::GVFlags Flags) {
 
   RawFlags |= (Flags.Visibility << 8); // 2 bits
 
-  RawFlags |= (Flags.ImportType << 10); // 1 bit
+  unsigned ImportType = Flags.ImportType | ImportAsDecl;
+  RawFlags |= (ImportType << 10); // 1 bit
 
   return RawFlags;
 }
@@ -4543,6 +4555,12 @@ void IndexBitcodeWriter::writeCombinedGlobalValueSummary() {
   Abbv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 8));
   unsigned AllocAbbrev = Stream.EmitAbbrev(std::move(Abbv));
 
+  auto shouldImportValueAsDecl = [&](GlobalValueSummary *GVS) -> bool {
+    if (DecSummaries == nullptr)
+      return false;
+    return DecSummaries->contains(GVS);
+  };
+
   // The aliases are emitted as a post-pass, and will point to the value
   // id of the aliasee. Save them in a vector for post-processing.
   SmallVector<AliasSummary *, 64> Aliases;
@@ -4653,7 +4671,8 @@ void IndexBitcodeWriter::writeCombinedGlobalValueSummary() {
     NameVals.push_back(*ValueId);
     assert(ModuleIdMap.count(FS->modulePath()));
     NameVals.push_back(ModuleIdMap[FS->modulePath()]);
-    NameVals.push_back(getEncodedGVSummaryFlags(FS->flags()));
+    NameVals.push_back(
+        getEncodedGVSummaryFlags(FS->flags(), shouldImportValueAsDecl(FS)));
     NameVals.push_back(FS->instCount());
     NameVals.push_back(getEncodedFFlags(FS->fflags()));
     NameVals.push_back(FS->entryCount());
@@ -4702,7 +4721,8 @@ void IndexBitcodeWriter::writeCombinedGlobalValueSummary() {
     NameVals.push_back(AliasValueId);
     assert(ModuleIdMap.count(AS->modulePath()));
     NameVals.push_back(ModuleIdMap[AS->modulePath()]);
-    NameVals.push_back(getEncodedGVSummaryFlags(AS->flags()));
+    NameVals.push_back(
+        getEncodedGVSummaryFlags(AS->flags(), shouldImportValueAsDecl(AS)));
     auto AliaseeValueId = SummaryToValueIdMap[&AS->getAliasee()];
     assert(AliaseeValueId);
     NameVals.push_back(AliaseeValueId);
@@ -5036,8 +5056,9 @@ void BitcodeWriter::writeModule(const Module &M,
 
 void BitcodeWriter::writeIndex(
     const ModuleSummaryIndex *Index,
-    const std::map<std::string, GVSummaryMapTy> *ModuleToSummariesForIndex) {
-  IndexBitcodeWriter IndexWriter(*Stream, StrtabBuilder, *Index,
+    const std::map<std::string, GVSummaryMapTy> *ModuleToSummariesForIndex,
+    const GVSummaryPtrSet *DecSummaries) {
+  IndexBitcodeWriter IndexWriter(*Stream, StrtabBuilder, *Index, DecSummaries,
                                  ModuleToSummariesForIndex);
   IndexWriter.write();
 }
@@ -5090,12 +5111,13 @@ void IndexBitcodeWriter::write() {
 // index for a distributed backend, provide a \p ModuleToSummariesForIndex map.
 void llvm::writeIndexToFile(
     const ModuleSummaryIndex &Index, raw_ostream &Out,
-    const std::map<std::string, GVSummaryMapTy> *ModuleToSummariesForIndex) {
+    const std::map<std::string, GVSummaryMapTy> *ModuleToSummariesForIndex,
+    const GVSummaryPtrSet *DecSummaries) {
   SmallVector<char, 0> Buffer;
   Buffer.reserve(256 * 1024);
 
   BitcodeWriter Writer(Buffer);
-  Writer.writeIndex(&Index, ModuleToSummariesForIndex);
+  Writer.writeIndex(&Index, ModuleToSummariesForIndex, DecSummaries);
   Writer.writeStrtab();
 
   Out.write((char *)&Buffer.front(), Buffer.size());

llvm/lib/LTO/LTO.cpp

@@ -121,6 +121,9 @@ void llvm::computeLTOCacheKey(
     support::endian::write64le(Data, I);
     Hasher.update(Data);
   };
+  auto AddUint8 = [&](const uint8_t &I) {
+    Hasher.update(ArrayRef<uint8_t>((const uint8_t *)&I, 1));
+  };
   AddString(Conf.CPU);
   // FIXME: Hash more of Options. For now all clients initialize Options from
   // command-line flags (which is unsupported in production), but may set
@@ -156,18 +159,23 @@ void llvm::computeLTOCacheKey(
   auto ModHash = Index.getModuleHash(ModuleID);
   Hasher.update(ArrayRef<uint8_t>((uint8_t *)&ModHash[0], sizeof(ModHash)));
 
-  std::vector<uint64_t> ExportsGUID;
+  std::vector<std::pair<uint64_t, uint8_t>> ExportsGUID;
   ExportsGUID.reserve(ExportList.size());
-  for (const auto &VI : ExportList) {
-    auto GUID = VI.getGUID();
-    ExportsGUID.push_back(GUID);
-  }
+  for (const auto &[VI, ExportType] : ExportList)
+    ExportsGUID.push_back(
+        std::make_pair(VI.getGUID(), static_cast<uint8_t>(ExportType)));
 
   // Sort the export list elements GUIDs.
-  llvm::sort(ExportsGUID);
-  for (uint64_t GUID : ExportsGUID) {
+  llvm::sort(ExportsGUID, [](const std::pair<uint64_t, uint8_t> &LHS,
+                             const std::pair<uint64_t, uint8_t> &RHS) {
+    if (LHS.first != RHS.first)
+      return LHS.first < RHS.first;
+    return LHS.second < RHS.second;
+  });
+  for (auto [GUID, ExportType] : ExportsGUID) {
     // The export list can impact the internalization, be conservative here
     Hasher.update(ArrayRef<uint8_t>((uint8_t *)&GUID, sizeof(GUID)));
+    AddUint8(ExportType);
   }
 
   // Include the hash for every module we import functions from. The set of
@@ -204,8 +212,10 @@ void llvm::computeLTOCacheKey(
     Hasher.update(ArrayRef<uint8_t>((uint8_t *)&ModHash[0], sizeof(ModHash)));
 
     AddUint64(Entry.getFunctions().size());
-    for (auto &Fn : Entry.getFunctions())
-      AddUint64(Fn);
+    for (auto &[GUID, ImportType] : Entry.getFunctions()) {
+      AddUint64(GUID);
+      AddUint8(ImportType);
+    }
   }
 
   // Include the hash for the resolved ODR.
@@ -275,9 +285,9 @@ void llvm::computeLTOCacheKey(
   // Imported functions may introduce new uses of type identifier resolutions,
   // so we need to collect their used resolutions as well.
   for (const ImportModule &ImpM : ImportModulesVector)
-    for (auto &ImpF : ImpM.getFunctions()) {
+    for (auto &[GUID, UnusedImportType] : ImpM.getFunctions()) {
       GlobalValueSummary *S =
-          Index.findSummaryInModule(ImpF, ImpM.getIdentifier());
+          Index.findSummaryInModule(GUID, ImpM.getIdentifier());
       AddUsedThings(S);
       // If this is an alias, we also care about any types/etc. that the aliasee
       // may reference.
@@ -1389,15 +1399,21 @@ class lto::ThinBackendProc {
                   llvm::StringRef ModulePath,
                   const std::string &NewModulePath) {
     std::map<std::string, GVSummaryMapTy> ModuleToSummariesForIndex;
+    GVSummaryPtrSet DeclarationSummaries;
+
     std::error_code EC;
     gatherImportedSummariesForModule(ModulePath, ModuleToDefinedGVSummaries,
-                                     ImportList, ModuleToSummariesForIndex);
+                                     ImportList, ModuleToSummariesForIndex,
+                                     DeclarationSummaries);
 
     raw_fd_ostream OS(NewModulePath + ".thinlto.bc", EC,
                       sys::fs::OpenFlags::OF_None);
     if (EC)
       return errorCodeToError(EC);
-    writeIndexToFile(CombinedIndex, OS, &ModuleToSummariesForIndex);
+
+    // TODO: Serialize declaration bits to bitcode.
+    writeIndexToFile(CombinedIndex, OS, &ModuleToSummariesForIndex,
+                     &DeclarationSummaries);
 
     if (ShouldEmitImportsFiles) {
       EC = EmitImportsFiles(ModulePath, NewModulePath + ".imports",

llvm/lib/LTO/LTOBackend.cpp

@@ -721,7 +721,14 @@ bool lto::initImportList(const Module &M,
       if (Summary->modulePath() == M.getModuleIdentifier())
         continue;
       // Add an entry to provoke importing by thinBackend.
-      ImportList[Summary->modulePath()].insert(GUID);
+      // Try emplace the entry first. If an entry with the same key already
+      // exists, set the value to 'std::min(existing-value, new-value)' to make
+      // sure a definition takes precedence over a declaration.
+      auto [Iter, Inserted] = ImportList[Summary->modulePath()].try_emplace(
+          GUID, Summary->importType());
+
+      if (!Inserted)
+        Iter->second = std::min(Iter->second, Summary->importType());
     }
   }
   return true;

llvm/lib/LTO/ThinLTOCodeGenerator.cpp

@@ -766,7 +766,7 @@ void ThinLTOCodeGenerator::crossModuleImport(Module &TheModule,
 void ThinLTOCodeGenerator::gatherImportedSummariesForModule(
     Module &TheModule, ModuleSummaryIndex &Index,
     std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex,
-    const lto::InputFile &File) {
+    GVSummaryPtrSet &DecSummaries, const lto::InputFile &File) {
   auto ModuleCount = Index.modulePaths().size();
   auto ModuleIdentifier = TheModule.getModuleIdentifier();
 
@@ -796,7 +796,7 @@ void ThinLTOCodeGenerator::gatherImportedSummariesForModule(
 
   llvm::gatherImportedSummariesForModule(
       ModuleIdentifier, ModuleToDefinedGVSummaries,
-      ImportLists[ModuleIdentifier], ModuleToSummariesForIndex);
+      ImportLists[ModuleIdentifier], ModuleToSummariesForIndex, DecSummaries);
 }
 
 /**
@@ -832,10 +832,14 @@ void ThinLTOCodeGenerator::emitImports(Module &TheModule, StringRef OutputName,
                            IsPrevailing(PrevailingCopy), ImportLists,
                            ExportLists);
 
+  // 'EmitImportsFiles' emits the list of modules from which to import from, and
+  // the set of keys in `ModuleToSummariesForIndex` should be a superset of keys
+  // in `DecSummaries`, so no need to use `DecSummaries` in `EmitImportFiles`.
+  GVSummaryPtrSet DecSummaries;
   std::map<std::string, GVSummaryMapTy> ModuleToSummariesForIndex;
   llvm::gatherImportedSummariesForModule(
       ModuleIdentifier, ModuleToDefinedGVSummaries,
-      ImportLists[ModuleIdentifier], ModuleToSummariesForIndex);
+      ImportLists[ModuleIdentifier], ModuleToSummariesForIndex, DecSummaries);
 
   std::error_code EC;
   if ((EC = EmitImportsFiles(ModuleIdentifier, OutputName,