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/TargetRegistry.h"
  29 #include "llvm/Object/ModuleSymbolTable.h"
  30 #include "llvm/Passes/PassBuilder.h"
  31 #include "llvm/Passes/PassPlugin.h"
  32 #include "llvm/Passes/StandardInstrumentations.h"
  33 #include "llvm/Support/Error.h"
  34 #include "llvm/Support/FileSystem.h"
  35 #include "llvm/Support/MemoryBuffer.h"
  36 #include "llvm/Support/Path.h"
  37 #include "llvm/Support/Program.h"
  38 #include "llvm/Support/ThreadPool.h"
  39 #include "llvm/Support/ToolOutputFile.h"
  40 #include "llvm/Support/VirtualFileSystem.h"
  41 #include "llvm/Support/raw_ostream.h"
  42 #include "llvm/Target/TargetMachine.h"
  43 #include "llvm/TargetParser/SubtargetFeature.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
 229   assert(TM && "Failed to create target machine");
 230
 231   if (std::optional<uint64_t> LargeDataThreshold = M.getLargeDataThreshold())
 232     TM->setLargeDataThreshold(*LargeDataThreshold);
 233
 234   return TM;
 235 }
 236
 237 static void runNewPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
 238                            unsigned OptLevel, bool IsThinLTO,
 239                            ModuleSummaryIndex *ExportSummary,
 240                            const ModuleSummaryIndex *ImportSummary) {
 241   auto FS = vfs::getRealFileSystem();
 242   std::optional<PGOOptions> PGOOpt;
 243   if (!Conf.SampleProfile.empty())
 244     PGOOpt = PGOOptions(Conf.SampleProfile, "", Conf.ProfileRemapping,
 245                         /*MemoryProfile=*/"", FS, PGOOptions::SampleUse,
 246                         PGOOptions::NoCSAction, true);
 247   else if (Conf.RunCSIRInstr) {
 248     PGOOpt = PGOOptions("", Conf.CSIRProfile, Conf.ProfileRemapping,
 249                         /*MemoryProfile=*/"", FS, PGOOptions::IRUse,
 250                         PGOOptions::CSIRInstr, Conf.AddFSDiscriminator);
 251   } else if (!Conf.CSIRProfile.empty()) {
 252     PGOOpt = PGOOptions(Conf.CSIRProfile, "", Conf.ProfileRemapping,
 253                         /*MemoryProfile=*/"", FS, PGOOptions::IRUse,
 254                         PGOOptions::CSIRUse, Conf.AddFSDiscriminator);
 255     NoPGOWarnMismatch = !Conf.PGOWarnMismatch;
 256   } else if (Conf.AddFSDiscriminator) {
 257     PGOOpt = PGOOptions("", "", "", /*MemoryProfile=*/"", nullptr,
 258                         PGOOptions::NoAction, PGOOptions::NoCSAction, true);
 259   }
 260   TM->setPGOOption(PGOOpt);
 261
 262   LoopAnalysisManager LAM;
 263   FunctionAnalysisManager FAM;
 264   CGSCCAnalysisManager CGAM;
 265   ModuleAnalysisManager MAM;
 266
 267   PassInstrumentationCallbacks PIC;
 268   StandardInstrumentations SI(Mod.getContext(), Conf.DebugPassManager,
 269                               Conf.VerifyEach);
 270   SI.registerCallbacks(PIC, &MAM);
 271   PassBuilder PB(TM, Conf.PTO, PGOOpt, &PIC);
 272
 273   RegisterPassPlugins(Conf.PassPlugins, PB);
 274
 275   std::unique_ptr<TargetLibraryInfoImpl> TLII(
 276       new TargetLibraryInfoImpl(Triple(TM->getTargetTriple())));
 277   if (Conf.Freestanding)
 278     TLII->disableAllFunctions();
 279   FAM.registerPass([&] { return TargetLibraryAnalysis(*TLII); });
 280
 281   // Parse a custom AA pipeline if asked to.
 282   if (!Conf.AAPipeline.empty()) {
 283     AAManager AA;
 284     if (auto Err = PB.parseAAPipeline(AA, Conf.AAPipeline)) {
 285       report_fatal_error(Twine("unable to parse AA pipeline description '") +
 286                          Conf.AAPipeline + "': " + toString(std::move(Err)));
 287     }
 288     // Register the AA manager first so that our version is the one used.
 289     FAM.registerPass([&] { return std::move(AA); });
 290   }
 291
 292   // Register all the basic analyses with the managers.
 293   PB.registerModuleAnalyses(MAM);
 294   PB.registerCGSCCAnalyses(CGAM);
 295   PB.registerFunctionAnalyses(FAM);
 296   PB.registerLoopAnalyses(LAM);
 297   PB.crossRegisterProxies(LAM, FAM, CGAM, MAM);
 298
 299   ModulePassManager MPM;
 300
 301   if (!Conf.DisableVerify)
 302     MPM.addPass(VerifierPass());
 303
 304   OptimizationLevel OL;
 305
 306   switch (OptLevel) {
 307   default:
 308     llvm_unreachable("Invalid optimization level");
 309   case 0:
 310     OL = OptimizationLevel::O0;
 311     break;
 312   case 1:
 313     OL = OptimizationLevel::O1;
 314     break;
 315   case 2:
 316     OL = OptimizationLevel::O2;
 317     break;
 318   case 3:
 319     OL = OptimizationLevel::O3;
 320     break;
 321   }
 322
 323   // Parse a custom pipeline if asked to.
 324   if (!Conf.OptPipeline.empty()) {
 325     if (auto Err = PB.parsePassPipeline(MPM, Conf.OptPipeline)) {
 326       report_fatal_error(Twine("unable to parse pass pipeline description '") +
 327                          Conf.OptPipeline + "': " + toString(std::move(Err)));
 328     }
 329   } else if (Conf.UseDefaultPipeline) {
 330     MPM.addPass(PB.buildPerModuleDefaultPipeline(OL));
 331   } else if (IsThinLTO) {
 332     MPM.addPass(PB.buildThinLTODefaultPipeline(OL, ImportSummary));
 333   } else {
 334     MPM.addPass(PB.buildLTODefaultPipeline(OL, ExportSummary));
 335   }
 336
 337   if (!Conf.DisableVerify)
 338     MPM.addPass(VerifierPass());
 339
 340   MPM.run(Mod, MAM);
 341 }
 342
 343 bool lto::opt(const Config &Conf, TargetMachine *TM, unsigned Task, Module &Mod,
 344               bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
 345               const ModuleSummaryIndex *ImportSummary,
 346               const std::vector<uint8_t> &CmdArgs) {
 347   if (EmbedBitcode == LTOBitcodeEmbedding::EmbedPostMergePreOptimized) {
 348     // FIXME: the motivation for capturing post-merge bitcode and command line
 349     // is replicating the compilation environment from bitcode, without needing
 350     // to understand the dependencies (the functions to be imported). This
 351     // assumes a clang - based invocation, case in which we have the command
 352     // line.
 353     // It's not very clear how the above motivation would map in the
 354     // linker-based case, so we currently don't plumb the command line args in
 355     // that case.
 356     if (CmdArgs.empty())
 357       LLVM_DEBUG(
 358           dbgs() << "Post-(Thin)LTO merge bitcode embedding was requested, but "
 359                     "command line arguments are not available");
 360     llvm::embedBitcodeInModule(Mod, llvm::MemoryBufferRef(),
 361                                /*EmbedBitcode*/ true, /*EmbedCmdline*/ true,
 362                                /*Cmdline*/ CmdArgs);
 363   }
 364   // FIXME: Plumb the combined index into the new pass manager.
 365   runNewPMPasses(Conf, Mod, TM, Conf.OptLevel, IsThinLTO, ExportSummary,
 366                  ImportSummary);
 367   return !Conf.PostOptModuleHook || Conf.PostOptModuleHook(Task, Mod);
 368 }
 369
 370 static void codegen(const Config &Conf, TargetMachine *TM,
 371                     AddStreamFn AddStream, unsigned Task, Module &Mod,
 372                     const ModuleSummaryIndex &CombinedIndex) {
 373   if (Conf.PreCodeGenModuleHook && !Conf.PreCodeGenModuleHook(Task, Mod))
 374     return;
 375
 376   if (EmbedBitcode == LTOBitcodeEmbedding::EmbedOptimized)
 377     llvm::embedBitcodeInModule(Mod, llvm::MemoryBufferRef(),
 378                                /*EmbedBitcode*/ true,
 379                                /*EmbedCmdline*/ false,
 380                                /*CmdArgs*/ std::vector<uint8_t>());
 381
 382   std::unique_ptr<ToolOutputFile> DwoOut;
 383   SmallString<1024> DwoFile(Conf.SplitDwarfOutput);
 384   if (!Conf.DwoDir.empty()) {
 385     std::error_code EC;
 386     if (auto EC = llvm::sys::fs::create_directories(Conf.DwoDir))
 387       report_fatal_error(Twine("Failed to create directory ") + Conf.DwoDir +
 388                          ": " + EC.message());
 389
 390     DwoFile = Conf.DwoDir;
 391     sys::path::append(DwoFile, std::to_string(Task) + ".dwo");
 392     TM->Options.MCOptions.SplitDwarfFile = std::string(DwoFile);
 393   } else
 394     TM->Options.MCOptions.SplitDwarfFile = Conf.SplitDwarfFile;
 395
 396   if (!DwoFile.empty()) {
 397     std::error_code EC;
 398     DwoOut = std::make_unique<ToolOutputFile>(DwoFile, EC, sys::fs::OF_None);
 399     if (EC)
 400       report_fatal_error(Twine("Failed to open ") + DwoFile + ": " +
 401                          EC.message());
 402   }
 403
 404   Expected<std::unique_ptr<CachedFileStream>> StreamOrErr =
 405       AddStream(Task, Mod.getModuleIdentifier());
 406   if (Error Err = StreamOrErr.takeError())
 407     report_fatal_error(std::move(Err));
 408   std::unique_ptr<CachedFileStream> &Stream = *StreamOrErr;
 409   TM->Options.ObjectFilenameForDebug = Stream->ObjectPathName;
 410
 411   legacy::PassManager CodeGenPasses;
 412   TargetLibraryInfoImpl TLII(Triple(Mod.getTargetTriple()));
 413   CodeGenPasses.add(new TargetLibraryInfoWrapperPass(TLII));
 414   CodeGenPasses.add(
 415       createImmutableModuleSummaryIndexWrapperPass(&CombinedIndex));
 416   if (Conf.PreCodeGenPassesHook)
 417     Conf.PreCodeGenPassesHook(CodeGenPasses);
 418   if (TM->addPassesToEmitFile(CodeGenPasses, *Stream->OS,
 419                               DwoOut ? &DwoOut->os() : nullptr,
 420                               Conf.CGFileType))
 421     report_fatal_error("Failed to setup codegen");
 422   CodeGenPasses.run(Mod);
 423
 424   if (DwoOut)
 425     DwoOut->keep();
 426 }
 427
 428 static void splitCodeGen(const Config &C, TargetMachine *TM,
 429                          AddStreamFn AddStream,
 430                          unsigned ParallelCodeGenParallelismLevel, Module &Mod,
 431                          const ModuleSummaryIndex &CombinedIndex) {
 432   ThreadPool CodegenThreadPool(
 433       heavyweight_hardware_concurrency(ParallelCodeGenParallelismLevel));
 434   unsigned ThreadCount = 0;
 435   const Target *T = &TM->getTarget();
 436
 437   SplitModule(
 438       Mod, ParallelCodeGenParallelismLevel,
 439       [&](std::unique_ptr<Module> MPart) {
 440         // We want to clone the module in a new context to multi-thread the
 441         // codegen. We do it by serializing partition modules to bitcode
 442         // (while still on the main thread, in order to avoid data races) and
 443         // spinning up new threads which deserialize the partitions into
 444         // separate contexts.
 445         // FIXME: Provide a more direct way to do this in LLVM.
 446         SmallString<0> BC;
 447         raw_svector_ostream BCOS(BC);
 448         WriteBitcodeToFile(*MPart, BCOS);
 449
 450         // Enqueue the task
 451         CodegenThreadPool.async(
 452             [&](const SmallString<0> &BC, unsigned ThreadId) {
 453               LTOLLVMContext Ctx(C);
 454               Expected<std::unique_ptr<Module>> MOrErr = parseBitcodeFile(
 455                   MemoryBufferRef(StringRef(BC.data(), BC.size()), "ld-temp.o"),
 456                   Ctx);
 457               if (!MOrErr)
 458                 report_fatal_error("Failed to read bitcode");
 459               std::unique_ptr<Module> MPartInCtx = std::move(MOrErr.get());
 460
 461               std::unique_ptr<TargetMachine> TM =
 462                   createTargetMachine(C, T, *MPartInCtx);
 463
 464               codegen(C, TM.get(), AddStream, ThreadId, *MPartInCtx,
 465                       CombinedIndex);
 466             },
 467             // Pass BC using std::move to ensure that it get moved rather than
 468             // copied into the thread's context.
 469             std::move(BC), ThreadCount++);
 470       },
 471       false);
 472
 473   // Because the inner lambda (which runs in a worker thread) captures our local
 474   // variables, we need to wait for the worker threads to terminate before we
 475   // can leave the function scope.
 476   CodegenThreadPool.wait();
 477 }
 478
 479 static Expected<const Target *> initAndLookupTarget(const Config &C,
 480                                                     Module &Mod) {
 481   if (!C.OverrideTriple.empty())
 482     Mod.setTargetTriple(C.OverrideTriple);
 483   else if (Mod.getTargetTriple().empty())
 484     Mod.setTargetTriple(C.DefaultTriple);
 485
 486   std::string Msg;
 487   const Target *T = TargetRegistry::lookupTarget(Mod.getTargetTriple(), Msg);
 488   if (!T)
 489     return make_error<StringError>(Msg, inconvertibleErrorCode());
 490   return T;
 491 }
 492
 493 Error lto::finalizeOptimizationRemarks(
 494     std::unique_ptr<ToolOutputFile> DiagOutputFile) {
 495   // Make sure we flush the diagnostic remarks file in case the linker doesn't
 496   // call the global destructors before exiting.
 497   if (!DiagOutputFile)
 498     return Error::success();
 499   DiagOutputFile->keep();
 500   DiagOutputFile->os().flush();
 501   return Error::success();
 502 }
 503
 504 Error lto::backend(const Config &C, AddStreamFn AddStream,
 505                    unsigned ParallelCodeGenParallelismLevel, Module &Mod,
 506                    ModuleSummaryIndex &CombinedIndex) {
 507   Expected<const Target *> TOrErr = initAndLookupTarget(C, Mod);
 508   if (!TOrErr)
 509     return TOrErr.takeError();
 510
 511   std::unique_ptr<TargetMachine> TM = createTargetMachine(C, *TOrErr, Mod);
 512
 513   LLVM_DEBUG(dbgs() << "Running regular LTO\n");
 514   if (!C.CodeGenOnly) {
 515     if (!opt(C, TM.get(), 0, Mod, /*IsThinLTO=*/false,
 516              /*ExportSummary=*/&CombinedIndex, /*ImportSummary=*/nullptr,
 517              /*CmdArgs*/ std::vector<uint8_t>()))
 518       return Error::success();
 519   }
 520
 521   if (ParallelCodeGenParallelismLevel == 1) {
 522     codegen(C, TM.get(), AddStream, 0, Mod, CombinedIndex);
 523   } else {
 524     splitCodeGen(C, TM.get(), AddStream, ParallelCodeGenParallelismLevel, Mod,
 525                  CombinedIndex);
 526   }
 527   return Error::success();
 528 }
 529
 530 static void dropDeadSymbols(Module &Mod, const GVSummaryMapTy &DefinedGlobals,
 531                             const ModuleSummaryIndex &Index) {
 532   std::vector<GlobalValue*> DeadGVs;
 533   for (auto &GV : Mod.global_values())
 534     if (GlobalValueSummary *GVS = DefinedGlobals.lookup(GV.getGUID()))
 535       if (!Index.isGlobalValueLive(GVS)) {
 536         DeadGVs.push_back(&GV);
 537         convertToDeclaration(GV);
 538       }
 539
 540   // Now that all dead bodies have been dropped, delete the actual objects
 541   // themselves when possible.
 542   for (GlobalValue *GV : DeadGVs) {
 543     GV->removeDeadConstantUsers();
 544     // Might reference something defined in native object (i.e. dropped a
 545     // non-prevailing IR def, but we need to keep the declaration).
 546     if (GV->use_empty())
 547       GV->eraseFromParent();
 548   }
 549 }
 550
 551 Error lto::thinBackend(const Config &Conf, unsigned Task, AddStreamFn AddStream,
 552                        Module &Mod, const ModuleSummaryIndex &CombinedIndex,
 553                        const FunctionImporter::ImportMapTy &ImportList,
 554                        const GVSummaryMapTy &DefinedGlobals,
 555                        MapVector<StringRef, BitcodeModule> *ModuleMap,
 556                        const std::vector<uint8_t> &CmdArgs) {
 557   Expected<const Target *> TOrErr = initAndLookupTarget(Conf, Mod);
 558   if (!TOrErr)
 559     return TOrErr.takeError();
 560
 561   std::unique_ptr<TargetMachine> TM = createTargetMachine(Conf, *TOrErr, Mod);
 562
 563   // Setup optimization remarks.
 564   auto DiagFileOrErr = lto::setupLLVMOptimizationRemarks(
 565       Mod.getContext(), Conf.RemarksFilename, Conf.RemarksPasses,
 566       Conf.RemarksFormat, Conf.RemarksWithHotness, Conf.RemarksHotnessThreshold,
 567       Task);
 568   if (!DiagFileOrErr)
 569     return DiagFileOrErr.takeError();
 570   auto DiagnosticOutputFile = std::move(*DiagFileOrErr);
 571
 572   // Set the partial sample profile ratio in the profile summary module flag of
 573   // the module, if applicable.
 574   Mod.setPartialSampleProfileRatio(CombinedIndex);
 575
 576   LLVM_DEBUG(dbgs() << "Running ThinLTO\n");
 577   if (Conf.CodeGenOnly) {
 578     codegen(Conf, TM.get(), AddStream, Task, Mod, CombinedIndex);
 579     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 580   }
 581
 582   if (Conf.PreOptModuleHook && !Conf.PreOptModuleHook(Task, Mod))
 583     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 584
 585   auto OptimizeAndCodegen =
 586       [&](Module &Mod, TargetMachine *TM,
 587           std::unique_ptr<ToolOutputFile> DiagnosticOutputFile) {
 588         if (!opt(Conf, TM, Task, Mod, /*IsThinLTO=*/true,
 589                  /*ExportSummary=*/nullptr, /*ImportSummary=*/&CombinedIndex,
 590                  CmdArgs))
 591           return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 592
 593         codegen(Conf, TM, AddStream, Task, Mod, CombinedIndex);
 594         return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 595       };
 596
 597   if (ThinLTOAssumeMerged)
 598     return OptimizeAndCodegen(Mod, TM.get(), std::move(DiagnosticOutputFile));
 599
 600   // When linking an ELF shared object, dso_local should be dropped. We
 601   // conservatively do this for -fpic.
 602   bool ClearDSOLocalOnDeclarations =
 603       TM->getTargetTriple().isOSBinFormatELF() &&
 604       TM->getRelocationModel() != Reloc::Static &&
 605       Mod.getPIELevel() == PIELevel::Default;
 606   renameModuleForThinLTO(Mod, CombinedIndex, ClearDSOLocalOnDeclarations);
 607
 608   dropDeadSymbols(Mod, DefinedGlobals, CombinedIndex);
 609
 610   thinLTOFinalizeInModule(Mod, DefinedGlobals, /*PropagateAttrs=*/true);
 611
 612   if (Conf.PostPromoteModuleHook && !Conf.PostPromoteModuleHook(Task, Mod))
 613     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 614
 615   if (!DefinedGlobals.empty())
 616     thinLTOInternalizeModule(Mod, DefinedGlobals);
 617
 618   if (Conf.PostInternalizeModuleHook &&
 619       !Conf.PostInternalizeModuleHook(Task, Mod))
 620     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 621
 622   auto ModuleLoader = [&](StringRef Identifier) {
 623     assert(Mod.getContext().isODRUniquingDebugTypes() &&
 624            "ODR Type uniquing should be enabled on the context");
 625     if (ModuleMap) {
 626       auto I = ModuleMap->find(Identifier);
 627       assert(I != ModuleMap->end());
 628       return I->second.getLazyModule(Mod.getContext(),
 629                                      /*ShouldLazyLoadMetadata=*/true,
 630                                      /*IsImporting*/ true);
 631     }
 632
 633     ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> MBOrErr =
 634         llvm::MemoryBuffer::getFile(Identifier);
 635     if (!MBOrErr)
 636       return Expected<std::unique_ptr<llvm::Module>>(make_error<StringError>(
 637           Twine("Error loading imported file ") + Identifier + " : ",
 638           MBOrErr.getError()));
 639
 640     Expected<BitcodeModule> BMOrErr = findThinLTOModule(**MBOrErr);
 641     if (!BMOrErr)
 642       return Expected<std::unique_ptr<llvm::Module>>(make_error<StringError>(
 643           Twine("Error loading imported file ") + Identifier + " : " +
 644               toString(BMOrErr.takeError()),
 645           inconvertibleErrorCode()));
 646
 647     Expected<std::unique_ptr<Module>> MOrErr =
 648         BMOrErr->getLazyModule(Mod.getContext(),
 649                                /*ShouldLazyLoadMetadata=*/true,
 650                                /*IsImporting*/ true);
 651     if (MOrErr)
 652       (*MOrErr)->setOwnedMemoryBuffer(std::move(*MBOrErr));
 653     return MOrErr;
 654   };
 655
 656   FunctionImporter Importer(CombinedIndex, ModuleLoader,
 657                             ClearDSOLocalOnDeclarations);
 658   if (Error Err = Importer.importFunctions(Mod, ImportList).takeError())
 659     return Err;
 660
 661   // Do this after any importing so that imported code is updated.
 662   updateMemProfAttributes(Mod, CombinedIndex);
 663   updatePublicTypeTestCalls(Mod, CombinedIndex.withWholeProgramVisibility());
 664
 665   if (Conf.PostImportModuleHook && !Conf.PostImportModuleHook(Task, Mod))
 666     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 667
 668   return OptimizeAndCodegen(Mod, TM.get(), std::move(DiagnosticOutputFile));
 669 }
 670
 671 BitcodeModule *lto::findThinLTOModule(MutableArrayRef<BitcodeModule> BMs) {
 672   if (ThinLTOAssumeMerged && BMs.size() == 1)
 673     return BMs.begin();
 674
 675   for (BitcodeModule &BM : BMs) {
 676     Expected<BitcodeLTOInfo> LTOInfo = BM.getLTOInfo();
 677     if (LTOInfo && LTOInfo->IsThinLTO)
 678       return &BM;
 679   }
 680   return nullptr;
 681 }
 682
 683 Expected<BitcodeModule> lto::findThinLTOModule(MemoryBufferRef MBRef) {
 684   Expected<std::vector<BitcodeModule>> BMsOrErr = getBitcodeModuleList(MBRef);
 685   if (!BMsOrErr)
 686     return BMsOrErr.takeError();
 687
 688   // The bitcode file may contain multiple modules, we want the one that is
 689   // marked as being the ThinLTO module.
 690   if (const BitcodeModule *Bm = lto::findThinLTOModule(*BMsOrErr))
 691     return *Bm;
 692
 693   return make_error<StringError>("Could not find module summary",
 694                                  inconvertibleErrorCode());
 695 }
 696
 697 bool lto::initImportList(const Module &M,
 698                          const ModuleSummaryIndex &CombinedIndex,
 699                          FunctionImporter::ImportMapTy &ImportList) {
 700   if (ThinLTOAssumeMerged)
 701     return true;
 702   // We can simply import the values mentioned in the combined index, since
 703   // we should only invoke this using the individual indexes written out
 704   // via a WriteIndexesThinBackend.
 705   for (const auto &GlobalList : CombinedIndex) {
 706     // Ignore entries for undefined references.
 707     if (GlobalList.second.SummaryList.empty())
 708       continue;
 709
 710     auto GUID = GlobalList.first;
 711     for (const auto &Summary : GlobalList.second.SummaryList) {
 712       // Skip the summaries for the importing module. These are included to
 713       // e.g. record required linkage changes.
 714       if (Summary->modulePath() == M.getModuleIdentifier())
 715         continue;
 716       // Add an entry to provoke importing by thinBackend.
 717       ImportList[Summary->modulePath()].insert(GUID);
 718     }
 719   }
 720   return true;
 721 }