[ThinLTO][Bitcode] Generate import type in bitcode (llvm#87600)

mingmingl-llvm · web-flow · commit 6262763341fc · 2024-05-22T09:52:54.000-07:00
For distributed ThinLTO, the LTO indexing step generates combined
summary for each module, and postlink pipeline reads the combined
summary which stores the information for link-time optimization.

This patch populates the 'import type' of a summary in bitcode, and
updates bitcode reader to parse the bit correctly.
diff --git a/llvm/include/llvm/Bitcode/BitcodeWriter.h b/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).
diff --git a/llvm/include/llvm/LTO/legacy/ThinLTOCodeGenerator.h b/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.
diff --git a/llvm/include/llvm/Transforms/IPO/FunctionImport.h b/llvm/include/llvm/Transforms/IPO/FunctionImport.h
@@ -212,11 +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(
diff --git a/llvm/lib/Bitcode/Writer/BitcodeWriter.cpp b/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());
diff --git a/llvm/lib/LTO/LTO.cpp b/llvm/lib/LTO/LTO.cpp
@@ -1400,18 +1400,20 @@ 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);
 
-    // TODO: Serialize declaration bits to bitcode.
-    writeIndexToFile(CombinedIndex, OS, &ModuleToSummariesForIndex);
+    writeIndexToFile(CombinedIndex, OS, &ModuleToSummariesForIndex,
+                     &DeclarationSummaries);
 
     if (ShouldEmitImportsFiles) {
       EC = EmitImportsFiles(ModulePath, NewModulePath + ".imports",
diff --git a/llvm/lib/LTO/ThinLTOCodeGenerator.cpp b/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,
diff --git a/llvm/lib/Transforms/IPO/FunctionImport.cpp b/llvm/lib/Transforms/IPO/FunctionImport.cpp
@@ -1435,7 +1435,8 @@ void llvm::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) {
   // Include all summaries from the importing module.
   ModuleToSummariesForIndex[std::string(ModulePath)] =
       ModuleToDefinedGVSummaries.lookup(ModulePath);
@@ -1450,7 +1451,7 @@ void llvm::gatherImportedSummariesForModule(
       assert(DS != DefinedGVSummaries.end() &&
              "Expected a defined summary for imported global value");
       if (Type == GlobalValueSummary::Declaration)
-        continue;
+        DecSummaries.insert(DS->second);
 
       SummariesForIndex[GUID] = DS->second;
     }
diff --git a/llvm/test/ThinLTO/X86/import_callee_declaration.ll b/llvm/test/ThinLTO/X86/import_callee_declaration.ll
diff --git a/llvm/tools/llvm-lto/llvm-lto.cpp b/llvm/tools/llvm-lto/llvm-lto.cpp
