llvm/lib/LTO/LTOBackend.cpp

   1 //===-LTOBackend.cpp - LLVM Link Time Optimizer Backend -------------------===//
   2 //
   3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
   4 // See https://llvm.org/LICENSE.txt for license information.
   5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
   6 //
   7 //===----------------------------------------------------------------------===//
   8 //
   9 // This file implements the "backend" phase of LTO, i.e. it performs
  10 // optimization and code generation on a loaded module. It is generally used
  11 // internally by the LTO class but can also be used independently, for example
  12 // to implement a standalone ThinLTO backend.
  13 //
  14 //===----------------------------------------------------------------------===//
  15
  16 #include "llvm/LTO/LTOBackend.h"
  17 #include "llvm/Analysis/AliasAnalysis.h"
  18 #include "llvm/Analysis/CGSCCPassManager.h"
  19 #include "llvm/Analysis/ModuleSummaryAnalysis.h"
  20 #include "llvm/Analysis/TargetLibraryInfo.h"
  21 #include "llvm/Bitcode/BitcodeReader.h"
  22 #include "llvm/Bitcode/BitcodeWriter.h"
  23 #include "llvm/IR/LLVMRemarkStreamer.h"
  24 #include "llvm/IR/LegacyPassManager.h"
  25 #include "llvm/IR/PassManager.h"
  26 #include "llvm/IR/Verifier.h"
  27 #include "llvm/LTO/LTO.h"
  28 #include "llvm/MC/SubtargetFeature.h"
  29 #include "llvm/MC/TargetRegistry.h"
  30 #include "llvm/Object/ModuleSymbolTable.h"
  31 #include "llvm/Passes/PassBuilder.h"
  32 #include "llvm/Passes/PassPlugin.h"
  33 #include "llvm/Passes/StandardInstrumentations.h"
  34 #include "llvm/Support/Error.h"
  35 #include "llvm/Support/FileSystem.h"
  36 #include "llvm/Support/MemoryBuffer.h"
  37 #include "llvm/Support/Path.h"
  38 #include "llvm/Support/Program.h"
  39 #include "llvm/Support/ThreadPool.h"
  40 #include "llvm/Support/ToolOutputFile.h"
  41 #include "llvm/Support/VirtualFileSystem.h"
  42 #include "llvm/Support/raw_ostream.h"
  43 #include "llvm/Target/TargetMachine.h"
  44 #include "llvm/Transforms/IPO/WholeProgramDevirt.h"
  45 #include "llvm/Transforms/Scalar/LoopPassManager.h"
  46 #include "llvm/Transforms/Utils/FunctionImportUtils.h"
  47 #include "llvm/Transforms/Utils/SplitModule.h"
  48 #include <optional>
  49
  50 using namespace llvm;
  51 using namespace lto;
  52
  53 #define DEBUG_TYPE "lto-backend"
  54
  55 enum class LTOBitcodeEmbedding {
  56   DoNotEmbed = 0,
  57   EmbedOptimized = 1,
  58   EmbedPostMergePreOptimized = 2
  59 };
  60
  61 static cl::opt<LTOBitcodeEmbedding> EmbedBitcode(
  62     "lto-embed-bitcode", cl::init(LTOBitcodeEmbedding::DoNotEmbed),
  63     cl::values(clEnumValN(LTOBitcodeEmbedding::DoNotEmbed, "none",
  64                           "Do not embed"),
  65                clEnumValN(LTOBitcodeEmbedding::EmbedOptimized, "optimized",
  66                           "Embed after all optimization passes"),
  67                clEnumValN(LTOBitcodeEmbedding::EmbedPostMergePreOptimized,
  68                           "post-merge-pre-opt",
  69                           "Embed post merge, but before optimizations")),
  70     cl::desc("Embed LLVM bitcode in object files produced by LTO"));
  71
  72 static cl::opt<bool> ThinLTOAssumeMerged(
  73     "thinlto-assume-merged", cl::init(false),
  74     cl::desc("Assume the input has already undergone ThinLTO function "
  75              "importing and the other pre-optimization pipeline changes."));
  76
  77 namespace llvm {
  78 extern cl::opt<bool> NoPGOWarnMismatch;
  79 }
  80
  81 [[noreturn]] static void reportOpenError(StringRef Path, Twine Msg) {
  82   errs() << "failed to open " << Path << ": " << Msg << '\n';
  83   errs().flush();
  84   exit(1);
  85 }
  86
  87 Error Config::addSaveTemps(std::string OutputFileName, bool UseInputModulePath,
  88                            const DenseSet<StringRef> &SaveTempsArgs) {
  89   ShouldDiscardValueNames = false;
  90
  91   std::error_code EC;
  92   if (SaveTempsArgs.empty() || SaveTempsArgs.contains("resolution")) {
  93     ResolutionFile =
  94         std::make_unique<raw_fd_ostream>(OutputFileName + "resolution.txt", EC,
  95                                          sys::fs::OpenFlags::OF_TextWithCRLF);
  96     if (EC) {
  97       ResolutionFile.reset();
  98       return errorCodeToError(EC);
  99     }
 100   }
 101
 102   auto setHook = [&](std::string PathSuffix, ModuleHookFn &Hook) {
 103     // Keep track of the hook provided by the linker, which also needs to run.
 104     ModuleHookFn LinkerHook = Hook;
 105     Hook = [=](unsigned Task, const Module &M) {
 106       // If the linker's hook returned false, we need to pass that result
 107       // through.
 108       if (LinkerHook && !LinkerHook(Task, M))
 109         return false;
 110
 111       std::string PathPrefix;
 112       // If this is the combined module (not a ThinLTO backend compile) or the
 113       // user hasn't requested using the input module's path, emit to a file
 114       // named from the provided OutputFileName with the Task ID appended.
 115       if (M.getModuleIdentifier() == "ld-temp.o" || !UseInputModulePath) {
 116         PathPrefix = OutputFileName;
 117         if (Task != (unsigned)-1)
 118           PathPrefix += utostr(Task) + ".";
 119       } else
 120         PathPrefix = M.getModuleIdentifier() + ".";
 121       std::string Path = PathPrefix + PathSuffix + ".bc";
 122       std::error_code EC;
 123       raw_fd_ostream OS(Path, EC, sys::fs::OpenFlags::OF_None);
 124       // Because -save-temps is a debugging feature, we report the error
 125       // directly and exit.
 126       if (EC)
 127         reportOpenError(Path, EC.message());
 128       WriteBitcodeToFile(M, OS, /*ShouldPreserveUseListOrder=*/false);
 129       return true;
 130     };
 131   };
 132
 133   auto SaveCombinedIndex =
 134       [=](const ModuleSummaryIndex &Index,
 135           const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols) {
 136         std::string Path = OutputFileName + "index.bc";
 137         std::error_code EC;
 138         raw_fd_ostream OS(Path, EC, sys::fs::OpenFlags::OF_None);
 139         // Because -save-temps is a debugging feature, we report the error
 140         // directly and exit.
 141         if (EC)
 142           reportOpenError(Path, EC.message());
 143         writeIndexToFile(Index, OS);
 144
 145         Path = OutputFileName + "index.dot";
 146         raw_fd_ostream OSDot(Path, EC, sys::fs::OpenFlags::OF_None);
 147         if (EC)
 148           reportOpenError(Path, EC.message());
 149         Index.exportToDot(OSDot, GUIDPreservedSymbols);
 150         return true;
 151       };
 152
 153   if (SaveTempsArgs.empty()) {
 154     setHook("0.preopt", PreOptModuleHook);
 155     setHook("1.promote", PostPromoteModuleHook);
 156     setHook("2.internalize", PostInternalizeModuleHook);
 157     setHook("3.import", PostImportModuleHook);
 158     setHook("4.opt", PostOptModuleHook);
 159     setHook("5.precodegen", PreCodeGenModuleHook);
 160     CombinedIndexHook = SaveCombinedIndex;
 161   } else {
 162     if (SaveTempsArgs.contains("preopt"))
 163       setHook("0.preopt", PreOptModuleHook);
 164     if (SaveTempsArgs.contains("promote"))
 165       setHook("1.promote", PostPromoteModuleHook);
 166     if (SaveTempsArgs.contains("internalize"))
 167       setHook("2.internalize", PostInternalizeModuleHook);
 168     if (SaveTempsArgs.contains("import"))
 169       setHook("3.import", PostImportModuleHook);
 170     if (SaveTempsArgs.contains("opt"))
 171       setHook("4.opt", PostOptModuleHook);
 172     if (SaveTempsArgs.contains("precodegen"))
 173       setHook("5.precodegen", PreCodeGenModuleHook);
 174     if (SaveTempsArgs.contains("combinedindex"))
 175       CombinedIndexHook = SaveCombinedIndex;
 176   }
 177
 178   return Error::success();
 179 }
 180
 181 #define HANDLE_EXTENSION(Ext)                                                  \
 182   llvm::PassPluginLibraryInfo get##Ext##PluginInfo();
 183 #include "llvm/Support/Extension.def"
 184
 185 static void RegisterPassPlugins(ArrayRef<std::string> PassPlugins,
 186                                 PassBuilder &PB) {
 187 #define HANDLE_EXTENSION(Ext)                                                  \
 188   get##Ext##PluginInfo().RegisterPassBuilderCallbacks(PB);
 189 #include "llvm/Support/Extension.def"
 190
 191   // Load requested pass plugins and let them register pass builder callbacks
 192   for (auto &PluginFN : PassPlugins) {
 193     auto PassPlugin = PassPlugin::Load(PluginFN);
 194     if (!PassPlugin) {
 195       errs() << "Failed to load passes from '" << PluginFN
 196              << "'. Request ignored.\n";
 197       continue;
 198     }
 199
 200     PassPlugin->registerPassBuilderCallbacks(PB);
 201   }
 202 }
 203
 204 static std::unique_ptr<TargetMachine>
 205 createTargetMachine(const Config &Conf, const Target *TheTarget, Module &M) {
 206   StringRef TheTriple = M.getTargetTriple();
 207   SubtargetFeatures Features;
 208   Features.getDefaultSubtargetFeatures(Triple(TheTriple));
 209   for (const std::string &A : Conf.MAttrs)
 210     Features.AddFeature(A);
 211
 212   std::optional<Reloc::Model> RelocModel;
 213   if (Conf.RelocModel)
 214     RelocModel = *Conf.RelocModel;
 215   else if (M.getModuleFlag("PIC Level"))
 216     RelocModel =
 217         M.getPICLevel() == PICLevel::NotPIC ? Reloc::Static : Reloc::PIC_;
 218
 219   std::optional<CodeModel::Model> CodeModel;
 220   if (Conf.CodeModel)
 221     CodeModel = *Conf.CodeModel;
 222   else
 223     CodeModel = M.getCodeModel();
 224
 225   std::unique_ptr<TargetMachine> TM(TheTarget->createTargetMachine(
 226       TheTriple, Conf.CPU, Features.getString(), Conf.Options, RelocModel,
 227       CodeModel, Conf.CGOptLevel));
 228   assert(TM && "Failed to create target machine");
 229   return TM;
 230 }
 231
 232 static void runNewPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
 233                            unsigned OptLevel, bool IsThinLTO,
 234                            ModuleSummaryIndex *ExportSummary,
 235                            const ModuleSummaryIndex *ImportSummary) {
 236   auto FS = vfs::getRealFileSystem();
 237   std::optional<PGOOptions> PGOOpt;
 238   if (!Conf.SampleProfile.empty())
 239     PGOOpt = PGOOptions(Conf.SampleProfile, "", Conf.ProfileRemapping, FS,
 240                         PGOOptions::SampleUse, PGOOptions::NoCSAction, true);
 241   else if (Conf.RunCSIRInstr) {
 242     PGOOpt = PGOOptions("", Conf.CSIRProfile, Conf.ProfileRemapping, FS,
 243                         PGOOptions::IRUse, PGOOptions::CSIRInstr,
 244                         Conf.AddFSDiscriminator);
 245   } else if (!Conf.CSIRProfile.empty()) {
 246     PGOOpt = PGOOptions(Conf.CSIRProfile, "", Conf.ProfileRemapping, FS,
 247                         PGOOptions::IRUse, PGOOptions::CSIRUse,
 248                         Conf.AddFSDiscriminator);
 249     NoPGOWarnMismatch = !Conf.PGOWarnMismatch;
 250   } else if (Conf.AddFSDiscriminator) {
 251     PGOOpt = PGOOptions("", "", "", nullptr, PGOOptions::NoAction,
 252                         PGOOptions::NoCSAction, true);
 253   }
 254   TM->setPGOOption(PGOOpt);
 255
 256   LoopAnalysisManager LAM;
 257   FunctionAnalysisManager FAM;
 258   CGSCCAnalysisManager CGAM;
 259   ModuleAnalysisManager MAM;
 260
 261   PassInstrumentationCallbacks PIC;
 262   StandardInstrumentations SI(Mod.getContext(), Conf.DebugPassManager);
 263   SI.registerCallbacks(PIC, &MAM);
 264   PassBuilder PB(TM, Conf.PTO, PGOOpt, &PIC);
 265
 266   RegisterPassPlugins(Conf.PassPlugins, PB);
 267
 268   std::unique_ptr<TargetLibraryInfoImpl> TLII(
 269       new TargetLibraryInfoImpl(Triple(TM->getTargetTriple())));
 270   if (Conf.Freestanding)
 271     TLII->disableAllFunctions();
 272   FAM.registerPass([&] { return TargetLibraryAnalysis(*TLII); });
 273
 274   // Parse a custom AA pipeline if asked to.
 275   if (!Conf.AAPipeline.empty()) {
 276     AAManager AA;
 277     if (auto Err = PB.parseAAPipeline(AA, Conf.AAPipeline)) {
 278       report_fatal_error(Twine("unable to parse AA pipeline description '") +
 279                          Conf.AAPipeline + "': " + toString(std::move(Err)));
 280     }
 281     // Register the AA manager first so that our version is the one used.
 282     FAM.registerPass([&] { return std::move(AA); });
 283   }
 284
 285   // Register all the basic analyses with the managers.
 286   PB.registerModuleAnalyses(MAM);
 287   PB.registerCGSCCAnalyses(CGAM);
 288   PB.registerFunctionAnalyses(FAM);
 289   PB.registerLoopAnalyses(LAM);
 290   PB.crossRegisterProxies(LAM, FAM, CGAM, MAM);
 291
 292   ModulePassManager MPM;
 293
 294   if (!Conf.DisableVerify)
 295     MPM.addPass(VerifierPass());
 296
 297   OptimizationLevel OL;
 298
 299   switch (OptLevel) {
 300   default:
 301     llvm_unreachable("Invalid optimization level");
 302   case 0:
 303     OL = OptimizationLevel::O0;
 304     break;
 305   case 1:
 306     OL = OptimizationLevel::O1;
 307     break;
 308   case 2:
 309     OL = OptimizationLevel::O2;
 310     break;
 311   case 3:
 312     OL = OptimizationLevel::O3;
 313     break;
 314   }
 315
 316   // Parse a custom pipeline if asked to.
 317   if (!Conf.OptPipeline.empty()) {
 318     if (auto Err = PB.parsePassPipeline(MPM, Conf.OptPipeline)) {
 319       report_fatal_error(Twine("unable to parse pass pipeline description '") +
 320                          Conf.OptPipeline + "': " + toString(std::move(Err)));
 321     }
 322   } else if (Conf.UseDefaultPipeline) {
 323     MPM.addPass(PB.buildPerModuleDefaultPipeline(OL));
 324   } else if (IsThinLTO) {
 325     MPM.addPass(PB.buildThinLTODefaultPipeline(OL, ImportSummary));
 326   } else {
 327     MPM.addPass(PB.buildLTODefaultPipeline(OL, ExportSummary));
 328   }
 329
 330   if (!Conf.DisableVerify)
 331     MPM.addPass(VerifierPass());
 332
 333   MPM.run(Mod, MAM);
 334 }
 335
 336 bool lto::opt(const Config &Conf, TargetMachine *TM, unsigned Task, Module &Mod,
 337               bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
 338               const ModuleSummaryIndex *ImportSummary,
 339               const std::vector<uint8_t> &CmdArgs) {
 340   if (EmbedBitcode == LTOBitcodeEmbedding::EmbedPostMergePreOptimized) {
 341     // FIXME: the motivation for capturing post-merge bitcode and command line
 342     // is replicating the compilation environment from bitcode, without needing
 343     // to understand the dependencies (the functions to be imported). This
 344     // assumes a clang - based invocation, case in which we have the command
 345     // line.
 346     // It's not very clear how the above motivation would map in the
 347     // linker-based case, so we currently don't plumb the command line args in
 348     // that case.
 349     if (CmdArgs.empty())
 350       LLVM_DEBUG(
 351           dbgs() << "Post-(Thin)LTO merge bitcode embedding was requested, but "
 352                     "command line arguments are not available");
 353     llvm::embedBitcodeInModule(Mod, llvm::MemoryBufferRef(),
 354                                /*EmbedBitcode*/ true, /*EmbedCmdline*/ true,
 355                                /*Cmdline*/ CmdArgs);
 356   }
 357   // FIXME: Plumb the combined index into the new pass manager.
 358   runNewPMPasses(Conf, Mod, TM, Conf.OptLevel, IsThinLTO, ExportSummary,
 359                  ImportSummary);
 360   return !Conf.PostOptModuleHook || Conf.PostOptModuleHook(Task, Mod);
 361 }
 362
 363 static void codegen(const Config &Conf, TargetMachine *TM,
 364                     AddStreamFn AddStream, unsigned Task, Module &Mod,
 365                     const ModuleSummaryIndex &CombinedIndex) {
 366   if (Conf.PreCodeGenModuleHook && !Conf.PreCodeGenModuleHook(Task, Mod))
 367     return;
 368
 369   if (EmbedBitcode == LTOBitcodeEmbedding::EmbedOptimized)
 370     llvm::embedBitcodeInModule(Mod, llvm::MemoryBufferRef(),
 371                                /*EmbedBitcode*/ true,
 372                                /*EmbedCmdline*/ false,
 373                                /*CmdArgs*/ std::vector<uint8_t>());
 374
 375   std::unique_ptr<ToolOutputFile> DwoOut;
 376   SmallString<1024> DwoFile(Conf.SplitDwarfOutput);
 377   if (!Conf.DwoDir.empty()) {
 378     std::error_code EC;
 379     if (auto EC = llvm::sys::fs::create_directories(Conf.DwoDir))
 380       report_fatal_error(Twine("Failed to create directory ") + Conf.DwoDir +
 381                          ": " + EC.message());
 382
 383     DwoFile = Conf.DwoDir;
 384     sys::path::append(DwoFile, std::to_string(Task) + ".dwo");
 385     TM->Options.MCOptions.SplitDwarfFile = std::string(DwoFile);
 386   } else
 387     TM->Options.MCOptions.SplitDwarfFile = Conf.SplitDwarfFile;
 388
 389   if (!DwoFile.empty()) {
 390     std::error_code EC;
 391     DwoOut = std::make_unique<ToolOutputFile>(DwoFile, EC, sys::fs::OF_None);
 392     if (EC)
 393       report_fatal_error(Twine("Failed to open ") + DwoFile + ": " +
 394                          EC.message());
 395   }
 396
 397   Expected<std::unique_ptr<CachedFileStream>> StreamOrErr =
 398       AddStream(Task, Mod.getModuleIdentifier());
 399   if (Error Err = StreamOrErr.takeError())
 400     report_fatal_error(std::move(Err));
 401   std::unique_ptr<CachedFileStream> &Stream = *StreamOrErr;
 402   TM->Options.ObjectFilenameForDebug = Stream->ObjectPathName;
 403
 404   legacy::PassManager CodeGenPasses;
 405   TargetLibraryInfoImpl TLII(Triple(Mod.getTargetTriple()));
 406   CodeGenPasses.add(new TargetLibraryInfoWrapperPass(TLII));
 407   CodeGenPasses.add(
 408       createImmutableModuleSummaryIndexWrapperPass(&CombinedIndex));
 409   if (Conf.PreCodeGenPassesHook)
 410     Conf.PreCodeGenPassesHook(CodeGenPasses);
 411   if (TM->addPassesToEmitFile(CodeGenPasses, *Stream->OS,
 412                               DwoOut ? &DwoOut->os() : nullptr,
 413                               Conf.CGFileType))
 414     report_fatal_error("Failed to setup codegen");
 415   CodeGenPasses.run(Mod);
 416
 417   if (DwoOut)
 418     DwoOut->keep();
 419 }
 420
 421 static void splitCodeGen(const Config &C, TargetMachine *TM,
 422                          AddStreamFn AddStream,
 423                          unsigned ParallelCodeGenParallelismLevel, Module &Mod,
 424                          const ModuleSummaryIndex &CombinedIndex) {
 425   ThreadPool CodegenThreadPool(
 426       heavyweight_hardware_concurrency(ParallelCodeGenParallelismLevel));
 427   unsigned ThreadCount = 0;
 428   const Target *T = &TM->getTarget();
 429
 430   SplitModule(
 431       Mod, ParallelCodeGenParallelismLevel,
 432       [&](std::unique_ptr<Module> MPart) {
 433         // We want to clone the module in a new context to multi-thread the
 434         // codegen. We do it by serializing partition modules to bitcode
 435         // (while still on the main thread, in order to avoid data races) and
 436         // spinning up new threads which deserialize the partitions into
 437         // separate contexts.
 438         // FIXME: Provide a more direct way to do this in LLVM.
 439         SmallString<0> BC;
 440         raw_svector_ostream BCOS(BC);
 441         WriteBitcodeToFile(*MPart, BCOS);
 442
 443         // Enqueue the task
 444         CodegenThreadPool.async(
 445             [&](const SmallString<0> &BC, unsigned ThreadId) {
 446               LTOLLVMContext Ctx(C);
 447               Expected<std::unique_ptr<Module>> MOrErr = parseBitcodeFile(
 448                   MemoryBufferRef(StringRef(BC.data(), BC.size()), "ld-temp.o"),
 449                   Ctx);
 450               if (!MOrErr)
 451                 report_fatal_error("Failed to read bitcode");
 452               std::unique_ptr<Module> MPartInCtx = std::move(MOrErr.get());
 453
 454               std::unique_ptr<TargetMachine> TM =
 455                   createTargetMachine(C, T, *MPartInCtx);
 456
 457               codegen(C, TM.get(), AddStream, ThreadId, *MPartInCtx,
 458                       CombinedIndex);
 459             },
 460             // Pass BC using std::move to ensure that it get moved rather than
 461             // copied into the thread's context.
 462             std::move(BC), ThreadCount++);
 463       },
 464       false);
 465
 466   // Because the inner lambda (which runs in a worker thread) captures our local
 467   // variables, we need to wait for the worker threads to terminate before we
 468   // can leave the function scope.
 469   CodegenThreadPool.wait();
 470 }
 471
 472 static Expected<const Target *> initAndLookupTarget(const Config &C,
 473                                                     Module &Mod) {
 474   if (!C.OverrideTriple.empty())
 475     Mod.setTargetTriple(C.OverrideTriple);
 476   else if (Mod.getTargetTriple().empty())
 477     Mod.setTargetTriple(C.DefaultTriple);
 478
 479   std::string Msg;
 480   const Target *T = TargetRegistry::lookupTarget(Mod.getTargetTriple(), Msg);
 481   if (!T)
 482     return make_error<StringError>(Msg, inconvertibleErrorCode());
 483   return T;
 484 }
 485
 486 Error lto::finalizeOptimizationRemarks(
 487     std::unique_ptr<ToolOutputFile> DiagOutputFile) {
 488   // Make sure we flush the diagnostic remarks file in case the linker doesn't
 489   // call the global destructors before exiting.
 490   if (!DiagOutputFile)
 491     return Error::success();
 492   DiagOutputFile->keep();
 493   DiagOutputFile->os().flush();
 494   return Error::success();
 495 }
 496
 497 Error lto::backend(const Config &C, AddStreamFn AddStream,
 498                    unsigned ParallelCodeGenParallelismLevel, Module &Mod,
 499                    ModuleSummaryIndex &CombinedIndex) {
 500   Expected<const Target *> TOrErr = initAndLookupTarget(C, Mod);
 501   if (!TOrErr)
 502     return TOrErr.takeError();
 503
 504   std::unique_ptr<TargetMachine> TM = createTargetMachine(C, *TOrErr, Mod);
 505
 506   if (!C.CodeGenOnly) {
 507     if (!opt(C, TM.get(), 0, Mod, /*IsThinLTO=*/false,
 508              /*ExportSummary=*/&CombinedIndex, /*ImportSummary=*/nullptr,
 509              /*CmdArgs*/ std::vector<uint8_t>()))
 510       return Error::success();
 511   }
 512
 513   if (ParallelCodeGenParallelismLevel == 1) {
 514     codegen(C, TM.get(), AddStream, 0, Mod, CombinedIndex);
 515   } else {
 516     splitCodeGen(C, TM.get(), AddStream, ParallelCodeGenParallelismLevel, Mod,
 517                  CombinedIndex);
 518   }
 519   return Error::success();
 520 }
 521
 522 static void dropDeadSymbols(Module &Mod, const GVSummaryMapTy &DefinedGlobals,
 523                             const ModuleSummaryIndex &Index) {
 524   std::vector<GlobalValue*> DeadGVs;
 525   for (auto &GV : Mod.global_values())
 526     if (GlobalValueSummary *GVS = DefinedGlobals.lookup(GV.getGUID()))
 527       if (!Index.isGlobalValueLive(GVS)) {
 528         DeadGVs.push_back(&GV);
 529         convertToDeclaration(GV);
 530       }
 531
 532   // Now that all dead bodies have been dropped, delete the actual objects
 533   // themselves when possible.
 534   for (GlobalValue *GV : DeadGVs) {
 535     GV->removeDeadConstantUsers();
 536     // Might reference something defined in native object (i.e. dropped a
 537     // non-prevailing IR def, but we need to keep the declaration).
 538     if (GV->use_empty())
 539       GV->eraseFromParent();
 540   }
 541 }
 542
 543 Error lto::thinBackend(const Config &Conf, unsigned Task, AddStreamFn AddStream,
 544                        Module &Mod, const ModuleSummaryIndex &CombinedIndex,
 545                        const FunctionImporter::ImportMapTy &ImportList,
 546                        const GVSummaryMapTy &DefinedGlobals,
 547                        MapVector<StringRef, BitcodeModule> *ModuleMap,
 548                        const std::vector<uint8_t> &CmdArgs) {
 549   Expected<const Target *> TOrErr = initAndLookupTarget(Conf, Mod);
 550   if (!TOrErr)
 551     return TOrErr.takeError();
 552
 553   std::unique_ptr<TargetMachine> TM = createTargetMachine(Conf, *TOrErr, Mod);
 554
 555   // Setup optimization remarks.
 556   auto DiagFileOrErr = lto::setupLLVMOptimizationRemarks(
 557       Mod.getContext(), Conf.RemarksFilename, Conf.RemarksPasses,
 558       Conf.RemarksFormat, Conf.RemarksWithHotness, Conf.RemarksHotnessThreshold,
 559       Task);
 560   if (!DiagFileOrErr)
 561     return DiagFileOrErr.takeError();
 562   auto DiagnosticOutputFile = std::move(*DiagFileOrErr);
 563
 564   // Set the partial sample profile ratio in the profile summary module flag of
 565   // the module, if applicable.
 566   Mod.setPartialSampleProfileRatio(CombinedIndex);
 567
 568   if (Conf.CodeGenOnly) {
 569     codegen(Conf, TM.get(), AddStream, Task, Mod, CombinedIndex);
 570     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 571   }
 572
 573   if (Conf.PreOptModuleHook && !Conf.PreOptModuleHook(Task, Mod))
 574     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 575
 576   auto OptimizeAndCodegen =
 577       [&](Module &Mod, TargetMachine *TM,
 578           std::unique_ptr<ToolOutputFile> DiagnosticOutputFile) {
 579         if (!opt(Conf, TM, Task, Mod, /*IsThinLTO=*/true,
 580                  /*ExportSummary=*/nullptr, /*ImportSummary=*/&CombinedIndex,
 581                  CmdArgs))
 582           return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 583
 584         codegen(Conf, TM, AddStream, Task, Mod, CombinedIndex);
 585         return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 586       };
 587
 588   if (ThinLTOAssumeMerged)
 589     return OptimizeAndCodegen(Mod, TM.get(), std::move(DiagnosticOutputFile));
 590
 591   // When linking an ELF shared object, dso_local should be dropped. We
 592   // conservatively do this for -fpic.
 593   bool ClearDSOLocalOnDeclarations =
 594       TM->getTargetTriple().isOSBinFormatELF() &&
 595       TM->getRelocationModel() != Reloc::Static &&
 596       Mod.getPIELevel() == PIELevel::Default;
 597   renameModuleForThinLTO(Mod, CombinedIndex, ClearDSOLocalOnDeclarations);
 598
 599   dropDeadSymbols(Mod, DefinedGlobals, CombinedIndex);
 600
 601   thinLTOFinalizeInModule(Mod, DefinedGlobals, /*PropagateAttrs=*/true);
 602
 603   if (Conf.PostPromoteModuleHook && !Conf.PostPromoteModuleHook(Task, Mod))
 604     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 605
 606   if (!DefinedGlobals.empty())
 607     thinLTOInternalizeModule(Mod, DefinedGlobals);
 608
 609   if (Conf.PostInternalizeModuleHook &&
 610       !Conf.PostInternalizeModuleHook(Task, Mod))
 611     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 612
 613   auto ModuleLoader = [&](StringRef Identifier) {
 614     assert(Mod.getContext().isODRUniquingDebugTypes() &&
 615            "ODR Type uniquing should be enabled on the context");
 616     if (ModuleMap) {
 617       auto I = ModuleMap->find(Identifier);
 618       assert(I != ModuleMap->end());
 619       return I->second.getLazyModule(Mod.getContext(),
 620                                      /*ShouldLazyLoadMetadata=*/true,
 621                                      /*IsImporting*/ true);
 622     }
 623
 624     ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> MBOrErr =
 625         llvm::MemoryBuffer::getFile(Identifier);
 626     if (!MBOrErr)
 627       return Expected<std::unique_ptr<llvm::Module>>(make_error<StringError>(
 628           Twine("Error loading imported file ") + Identifier + " : ",
 629           MBOrErr.getError()));
 630
 631     Expected<BitcodeModule> BMOrErr = findThinLTOModule(**MBOrErr);
 632     if (!BMOrErr)
 633       return Expected<std::unique_ptr<llvm::Module>>(make_error<StringError>(
 634           Twine("Error loading imported file ") + Identifier + " : " +
 635               toString(BMOrErr.takeError()),
 636           inconvertibleErrorCode()));
 637
 638     Expected<std::unique_ptr<Module>> MOrErr =
 639         BMOrErr->getLazyModule(Mod.getContext(),
 640                                /*ShouldLazyLoadMetadata=*/true,
 641                                /*IsImporting*/ true);
 642     if (MOrErr)
 643       (*MOrErr)->setOwnedMemoryBuffer(std::move(*MBOrErr));
 644     return MOrErr;
 645   };
 646
 647   FunctionImporter Importer(CombinedIndex, ModuleLoader,
 648                             ClearDSOLocalOnDeclarations);
 649   if (Error Err = Importer.importFunctions(Mod, ImportList).takeError())
 650     return Err;
 651
 652   // Do this after any importing so that imported code is updated.
 653   updateMemProfAttributes(Mod, CombinedIndex);
 654   updatePublicTypeTestCalls(Mod, CombinedIndex.withWholeProgramVisibility());
 655
 656   if (Conf.PostImportModuleHook && !Conf.PostImportModuleHook(Task, Mod))
 657     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 658
 659   return OptimizeAndCodegen(Mod, TM.get(), std::move(DiagnosticOutputFile));
 660 }
 661
 662 BitcodeModule *lto::findThinLTOModule(MutableArrayRef<BitcodeModule> BMs) {
 663   if (ThinLTOAssumeMerged && BMs.size() == 1)
 664     return BMs.begin();
 665
 666   for (BitcodeModule &BM : BMs) {
 667     Expected<BitcodeLTOInfo> LTOInfo = BM.getLTOInfo();
 668     if (LTOInfo && LTOInfo->IsThinLTO)
 669       return &BM;
 670   }
 671   return nullptr;
 672 }
 673
 674 Expected<BitcodeModule> lto::findThinLTOModule(MemoryBufferRef MBRef) {
 675   Expected<std::vector<BitcodeModule>> BMsOrErr = getBitcodeModuleList(MBRef);
 676   if (!BMsOrErr)
 677     return BMsOrErr.takeError();
 678
 679   // The bitcode file may contain multiple modules, we want the one that is
 680   // marked as being the ThinLTO module.
 681   if (const BitcodeModule *Bm = lto::findThinLTOModule(*BMsOrErr))
 682     return *Bm;
 683
 684   return make_error<StringError>("Could not find module summary",
 685                                  inconvertibleErrorCode());
 686 }
 687
 688 bool lto::initImportList(const Module &M,
 689                          const ModuleSummaryIndex &CombinedIndex,
 690                          FunctionImporter::ImportMapTy &ImportList) {
 691   if (ThinLTOAssumeMerged)
 692     return true;
 693   // We can simply import the values mentioned in the combined index, since
 694   // we should only invoke this using the individual indexes written out
 695   // via a WriteIndexesThinBackend.
 696   for (const auto &GlobalList : CombinedIndex) {
 697     // Ignore entries for undefined references.
 698     if (GlobalList.second.SummaryList.empty())
 699       continue;
 700
 701     auto GUID = GlobalList.first;
 702     for (const auto &Summary : GlobalList.second.SummaryList) {
 703       // Skip the summaries for the importing module. These are included to
 704       // e.g. record required linkage changes.
 705       if (Summary->modulePath() == M.getModuleIdentifier())
 706         continue;
 707       // Add an entry to provoke importing by thinBackend.
 708       ImportList[Summary->modulePath()].insert(GUID);
 709     }
 710   }
 711   return true;
 712 }