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/Analysis/TargetTransformInfo.h"
  22 #include "llvm/Bitcode/BitcodeReader.h"
  23 #include "llvm/Bitcode/BitcodeWriter.h"
  24 #include "llvm/IR/LLVMRemarkStreamer.h"
  25 #include "llvm/IR/LegacyPassManager.h"
  26 #include "llvm/IR/PassManager.h"
  27 #include "llvm/IR/Verifier.h"
  28 #include "llvm/LTO/LTO.h"
  29 #include "llvm/MC/SubtargetFeature.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/SmallVectorMemoryBuffer.h"
  40 #include "llvm/Support/TargetRegistry.h"
  41 #include "llvm/Support/ThreadPool.h"
  42 #include "llvm/Support/raw_ostream.h"
  43 #include "llvm/Target/TargetMachine.h"
  44 #include "llvm/Transforms/IPO.h"
  45 #include "llvm/Transforms/IPO/PassManagerBuilder.h"
  46 #include "llvm/Transforms/Scalar/LoopPassManager.h"
  47 #include "llvm/Transforms/Utils/FunctionImportUtils.h"
  48 #include "llvm/Transforms/Utils/SplitModule.h"
  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,
  88                            bool UseInputModulePath) {
  89   ShouldDiscardValueNames = false;
  90
  91   std::error_code EC;
  92   ResolutionFile =
  93       std::make_unique<raw_fd_ostream>(OutputFileName + "resolution.txt", EC,
  94                                        sys::fs::OpenFlags::OF_TextWithCRLF);
  95   if (EC) {
  96     ResolutionFile.reset();
  97     return errorCodeToError(EC);
  98   }
  99
 100   auto setHook = [&](std::string PathSuffix, ModuleHookFn &Hook) {
 101     // Keep track of the hook provided by the linker, which also needs to run.
 102     ModuleHookFn LinkerHook = Hook;
 103     Hook = [=](unsigned Task, const Module &M) {
 104       // If the linker's hook returned false, we need to pass that result
 105       // through.
 106       if (LinkerHook && !LinkerHook(Task, M))
 107         return false;
 108
 109       std::string PathPrefix;
 110       // If this is the combined module (not a ThinLTO backend compile) or the
 111       // user hasn't requested using the input module's path, emit to a file
 112       // named from the provided OutputFileName with the Task ID appended.
 113       if (M.getModuleIdentifier() == "ld-temp.o" || !UseInputModulePath) {
 114         PathPrefix = OutputFileName;
 115         if (Task != (unsigned)-1)
 116           PathPrefix += utostr(Task) + ".";
 117       } else
 118         PathPrefix = M.getModuleIdentifier() + ".";
 119       std::string Path = PathPrefix + PathSuffix + ".bc";
 120       std::error_code EC;
 121       raw_fd_ostream OS(Path, EC, sys::fs::OpenFlags::OF_None);
 122       // Because -save-temps is a debugging feature, we report the error
 123       // directly and exit.
 124       if (EC)
 125         reportOpenError(Path, EC.message());
 126       WriteBitcodeToFile(M, OS, /*ShouldPreserveUseListOrder=*/false);
 127       return true;
 128     };
 129   };
 130
 131   setHook("0.preopt", PreOptModuleHook);
 132   setHook("1.promote", PostPromoteModuleHook);
 133   setHook("2.internalize", PostInternalizeModuleHook);
 134   setHook("3.import", PostImportModuleHook);
 135   setHook("4.opt", PostOptModuleHook);
 136   setHook("5.precodegen", PreCodeGenModuleHook);
 137
 138   CombinedIndexHook =
 139       [=](const ModuleSummaryIndex &Index,
 140           const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols) {
 141         std::string Path = OutputFileName + "index.bc";
 142         std::error_code EC;
 143         raw_fd_ostream OS(Path, EC, sys::fs::OpenFlags::OF_None);
 144         // Because -save-temps is a debugging feature, we report the error
 145         // directly and exit.
 146         if (EC)
 147           reportOpenError(Path, EC.message());
 148         WriteIndexToFile(Index, OS);
 149
 150         Path = OutputFileName + "index.dot";
 151         raw_fd_ostream OSDot(Path, EC, sys::fs::OpenFlags::OF_None);
 152         if (EC)
 153           reportOpenError(Path, EC.message());
 154         Index.exportToDot(OSDot, GUIDPreservedSymbols);
 155         return true;
 156       };
 157
 158   return Error::success();
 159 }
 160
 161 #define HANDLE_EXTENSION(Ext)                                                  \
 162   llvm::PassPluginLibraryInfo get##Ext##PluginInfo();
 163 #include "llvm/Support/Extension.def"
 164
 165 static void RegisterPassPlugins(ArrayRef<std::string> PassPlugins,
 166                                 PassBuilder &PB) {
 167 #define HANDLE_EXTENSION(Ext)                                                  \
 168   get##Ext##PluginInfo().RegisterPassBuilderCallbacks(PB);
 169 #include "llvm/Support/Extension.def"
 170
 171   // Load requested pass plugins and let them register pass builder callbacks
 172   for (auto &PluginFN : PassPlugins) {
 173     auto PassPlugin = PassPlugin::Load(PluginFN);
 174     if (!PassPlugin) {
 175       errs() << "Failed to load passes from '" << PluginFN
 176              << "'. Request ignored.\n";
 177       continue;
 178     }
 179
 180     PassPlugin->registerPassBuilderCallbacks(PB);
 181   }
 182 }
 183
 184 static std::unique_ptr<TargetMachine>
 185 createTargetMachine(const Config &Conf, const Target *TheTarget, Module &M) {
 186   StringRef TheTriple = M.getTargetTriple();
 187   SubtargetFeatures Features;
 188   Features.getDefaultSubtargetFeatures(Triple(TheTriple));
 189   for (const std::string &A : Conf.MAttrs)
 190     Features.AddFeature(A);
 191
 192   Optional<Reloc::Model> RelocModel = None;
 193   if (Conf.RelocModel)
 194     RelocModel = *Conf.RelocModel;
 195   else if (M.getModuleFlag("PIC Level"))
 196     RelocModel =
 197         M.getPICLevel() == PICLevel::NotPIC ? Reloc::Static : Reloc::PIC_;
 198
 199   Optional<CodeModel::Model> CodeModel;
 200   if (Conf.CodeModel)
 201     CodeModel = *Conf.CodeModel;
 202   else
 203     CodeModel = M.getCodeModel();
 204
 205   std::unique_ptr<TargetMachine> TM(TheTarget->createTargetMachine(
 206       TheTriple, Conf.CPU, Features.getString(), Conf.Options, RelocModel,
 207       CodeModel, Conf.CGOptLevel));
 208   assert(TM && "Failed to create target machine");
 209   return TM;
 210 }
 211
 212 static void runNewPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
 213                            unsigned OptLevel, bool IsThinLTO,
 214                            ModuleSummaryIndex *ExportSummary,
 215                            const ModuleSummaryIndex *ImportSummary) {
 216   Optional<PGOOptions> PGOOpt;
 217   if (!Conf.SampleProfile.empty())
 218     PGOOpt = PGOOptions(Conf.SampleProfile, "", Conf.ProfileRemapping,
 219                         PGOOptions::SampleUse, PGOOptions::NoCSAction, true);
 220   else if (Conf.RunCSIRInstr) {
 221     PGOOpt = PGOOptions("", Conf.CSIRProfile, Conf.ProfileRemapping,
 222                         PGOOptions::IRUse, PGOOptions::CSIRInstr,
 223                         Conf.AddFSDiscriminator);
 224   } else if (!Conf.CSIRProfile.empty()) {
 225     PGOOpt = PGOOptions(Conf.CSIRProfile, "", Conf.ProfileRemapping,
 226                         PGOOptions::IRUse, PGOOptions::CSIRUse,
 227                         Conf.AddFSDiscriminator);
 228     NoPGOWarnMismatch = !Conf.PGOWarnMismatch;
 229   } else if (Conf.AddFSDiscriminator) {
 230     PGOOpt = PGOOptions("", "", "", PGOOptions::NoAction,
 231                         PGOOptions::NoCSAction, true);
 232   }
 233   if (TM)
 234     TM->setPGOOption(PGOOpt);
 235
 236   LoopAnalysisManager LAM;
 237   FunctionAnalysisManager FAM;
 238   CGSCCAnalysisManager CGAM;
 239   ModuleAnalysisManager MAM;
 240
 241   PassInstrumentationCallbacks PIC;
 242   StandardInstrumentations SI(Conf.DebugPassManager);
 243   SI.registerCallbacks(PIC, &FAM);
 244   PassBuilder PB(TM, Conf.PTO, PGOOpt, &PIC);
 245
 246   RegisterPassPlugins(Conf.PassPlugins, PB);
 247
 248   std::unique_ptr<TargetLibraryInfoImpl> TLII(
 249       new TargetLibraryInfoImpl(Triple(TM->getTargetTriple())));
 250   if (Conf.Freestanding)
 251     TLII->disableAllFunctions();
 252   FAM.registerPass([&] { return TargetLibraryAnalysis(*TLII); });
 253
 254   AAManager AA;
 255   // Parse a custom AA pipeline if asked to.
 256   if (!Conf.AAPipeline.empty()) {
 257     if (auto Err = PB.parseAAPipeline(AA, Conf.AAPipeline)) {
 258       report_fatal_error("unable to parse AA pipeline description '" +
 259                          Conf.AAPipeline + "': " + toString(std::move(Err)));
 260     }
 261   } else {
 262     AA = PB.buildDefaultAAPipeline();
 263   }
 264   // Register the AA manager first so that our version is the one used.
 265   FAM.registerPass([&] { return std::move(AA); });
 266
 267   // Register all the basic analyses with the managers.
 268   PB.registerModuleAnalyses(MAM);
 269   PB.registerCGSCCAnalyses(CGAM);
 270   PB.registerFunctionAnalyses(FAM);
 271   PB.registerLoopAnalyses(LAM);
 272   PB.crossRegisterProxies(LAM, FAM, CGAM, MAM);
 273
 274   ModulePassManager MPM;
 275
 276   if (!Conf.DisableVerify)
 277     MPM.addPass(VerifierPass());
 278
 279   OptimizationLevel OL;
 280
 281   switch (OptLevel) {
 282   default:
 283     llvm_unreachable("Invalid optimization level");
 284   case 0:
 285     OL = OptimizationLevel::O0;
 286     break;
 287   case 1:
 288     OL = OptimizationLevel::O1;
 289     break;
 290   case 2:
 291     OL = OptimizationLevel::O2;
 292     break;
 293   case 3:
 294     OL = OptimizationLevel::O3;
 295     break;
 296   }
 297
 298   // Parse a custom pipeline if asked to.
 299   if (!Conf.OptPipeline.empty()) {
 300     if (auto Err = PB.parsePassPipeline(MPM, Conf.OptPipeline)) {
 301       report_fatal_error("unable to parse pass pipeline description '" +
 302                          Conf.OptPipeline + "': " + toString(std::move(Err)));
 303     }
 304   } else if (IsThinLTO) {
 305     MPM.addPass(PB.buildThinLTODefaultPipeline(OL, ImportSummary));
 306   } else {
 307     MPM.addPass(PB.buildLTODefaultPipeline(OL, ExportSummary));
 308   }
 309
 310   if (!Conf.DisableVerify)
 311     MPM.addPass(VerifierPass());
 312
 313   MPM.run(Mod, MAM);
 314 }
 315
 316 static void runOldPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
 317                            bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
 318                            const ModuleSummaryIndex *ImportSummary) {
 319   legacy::PassManager passes;
 320   passes.add(createTargetTransformInfoWrapperPass(TM->getTargetIRAnalysis()));
 321
 322   PassManagerBuilder PMB;
 323   PMB.LibraryInfo = new TargetLibraryInfoImpl(Triple(TM->getTargetTriple()));
 324   if (Conf.Freestanding)
 325     PMB.LibraryInfo->disableAllFunctions();
 326   PMB.Inliner = createFunctionInliningPass();
 327   PMB.ExportSummary = ExportSummary;
 328   PMB.ImportSummary = ImportSummary;
 329   // Unconditionally verify input since it is not verified before this
 330   // point and has unknown origin.
 331   PMB.VerifyInput = true;
 332   PMB.VerifyOutput = !Conf.DisableVerify;
 333   PMB.LoopVectorize = true;
 334   PMB.SLPVectorize = true;
 335   PMB.OptLevel = Conf.OptLevel;
 336   PMB.PGOSampleUse = Conf.SampleProfile;
 337   PMB.EnablePGOCSInstrGen = Conf.RunCSIRInstr;
 338   if (!Conf.RunCSIRInstr && !Conf.CSIRProfile.empty()) {
 339     PMB.EnablePGOCSInstrUse = true;
 340     PMB.PGOInstrUse = Conf.CSIRProfile;
 341   }
 342   if (IsThinLTO)
 343     PMB.populateThinLTOPassManager(passes);
 344   else
 345     PMB.populateLTOPassManager(passes);
 346   passes.run(Mod);
 347 }
 348
 349 bool lto::opt(const Config &Conf, TargetMachine *TM, unsigned Task, Module &Mod,
 350               bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
 351               const ModuleSummaryIndex *ImportSummary,
 352               const std::vector<uint8_t> &CmdArgs) {
 353   if (EmbedBitcode == LTOBitcodeEmbedding::EmbedPostMergePreOptimized) {
 354     // FIXME: the motivation for capturing post-merge bitcode and command line
 355     // is replicating the compilation environment from bitcode, without needing
 356     // to understand the dependencies (the functions to be imported). This
 357     // assumes a clang - based invocation, case in which we have the command
 358     // line.
 359     // It's not very clear how the above motivation would map in the
 360     // linker-based case, so we currently don't plumb the command line args in
 361     // that case.
 362     if (CmdArgs.empty())
 363       LLVM_DEBUG(
 364           dbgs() << "Post-(Thin)LTO merge bitcode embedding was requested, but "
 365                     "command line arguments are not available");
 366     llvm::EmbedBitcodeInModule(Mod, llvm::MemoryBufferRef(),
 367                                /*EmbedBitcode*/ true, /*EmbedCmdline*/ true,
 368                                /*Cmdline*/ CmdArgs);
 369   }
 370   // FIXME: Plumb the combined index into the new pass manager.
 371   if (Conf.UseNewPM || !Conf.OptPipeline.empty()) {
 372     runNewPMPasses(Conf, Mod, TM, Conf.OptLevel, IsThinLTO, ExportSummary,
 373                    ImportSummary);
 374   } else {
 375     runOldPMPasses(Conf, Mod, TM, IsThinLTO, ExportSummary, ImportSummary);
 376   }
 377   return !Conf.PostOptModuleHook || Conf.PostOptModuleHook(Task, Mod);
 378 }
 379
 380 static void codegen(const Config &Conf, TargetMachine *TM,
 381                     AddStreamFn AddStream, unsigned Task, Module &Mod,
 382                     const ModuleSummaryIndex &CombinedIndex) {
 383   if (Conf.PreCodeGenModuleHook && !Conf.PreCodeGenModuleHook(Task, Mod))
 384     return;
 385
 386   if (EmbedBitcode == LTOBitcodeEmbedding::EmbedOptimized)
 387     llvm::EmbedBitcodeInModule(Mod, llvm::MemoryBufferRef(),
 388                                /*EmbedBitcode*/ true,
 389                                /*EmbedCmdline*/ false,
 390                                /*CmdArgs*/ std::vector<uint8_t>());
 391
 392   std::unique_ptr<ToolOutputFile> DwoOut;
 393   SmallString<1024> DwoFile(Conf.SplitDwarfOutput);
 394   if (!Conf.DwoDir.empty()) {
 395     std::error_code EC;
 396     if (auto EC = llvm::sys::fs::create_directories(Conf.DwoDir))
 397       report_fatal_error("Failed to create directory " + Conf.DwoDir + ": " +
 398                          EC.message());
 399
 400     DwoFile = Conf.DwoDir;
 401     sys::path::append(DwoFile, std::to_string(Task) + ".dwo");
 402     TM->Options.MCOptions.SplitDwarfFile = std::string(DwoFile);
 403   } else
 404     TM->Options.MCOptions.SplitDwarfFile = Conf.SplitDwarfFile;
 405
 406   if (!DwoFile.empty()) {
 407     std::error_code EC;
 408     DwoOut = std::make_unique<ToolOutputFile>(DwoFile, EC, sys::fs::OF_None);
 409     if (EC)
 410       report_fatal_error("Failed to open " + DwoFile + ": " + EC.message());
 411   }
 412
 413   auto Stream = AddStream(Task);
 414   legacy::PassManager CodeGenPasses;
 415   CodeGenPasses.add(
 416       createImmutableModuleSummaryIndexWrapperPass(&CombinedIndex));
 417   if (Conf.PreCodeGenPassesHook)
 418     Conf.PreCodeGenPassesHook(CodeGenPasses);
 419   if (TM->addPassesToEmitFile(CodeGenPasses, *Stream->OS,
 420                               DwoOut ? &DwoOut->os() : nullptr,
 421                               Conf.CGFileType))
 422     report_fatal_error("Failed to setup codegen");
 423   CodeGenPasses.run(Mod);
 424
 425   if (DwoOut)
 426     DwoOut->keep();
 427 }
 428
 429 static void splitCodeGen(const Config &C, TargetMachine *TM,
 430                          AddStreamFn AddStream,
 431                          unsigned ParallelCodeGenParallelismLevel, Module &Mod,
 432                          const ModuleSummaryIndex &CombinedIndex) {
 433   ThreadPool CodegenThreadPool(
 434       heavyweight_hardware_concurrency(ParallelCodeGenParallelismLevel));
 435   unsigned ThreadCount = 0;
 436   const Target *T = &TM->getTarget();
 437
 438   SplitModule(
 439       Mod, ParallelCodeGenParallelismLevel,
 440       [&](std::unique_ptr<Module> MPart) {
 441         // We want to clone the module in a new context to multi-thread the
 442         // codegen. We do it by serializing partition modules to bitcode
 443         // (while still on the main thread, in order to avoid data races) and
 444         // spinning up new threads which deserialize the partitions into
 445         // separate contexts.
 446         // FIXME: Provide a more direct way to do this in LLVM.
 447         SmallString<0> BC;
 448         raw_svector_ostream BCOS(BC);
 449         WriteBitcodeToFile(*MPart, BCOS);
 450
 451         // Enqueue the task
 452         CodegenThreadPool.async(
 453             [&](const SmallString<0> &BC, unsigned ThreadId) {
 454               LTOLLVMContext Ctx(C);
 455               Expected<std::unique_ptr<Module>> MOrErr = parseBitcodeFile(
 456                   MemoryBufferRef(StringRef(BC.data(), BC.size()), "ld-temp.o"),
 457                   Ctx);
 458               if (!MOrErr)
 459                 report_fatal_error("Failed to read bitcode");
 460               std::unique_ptr<Module> MPartInCtx = std::move(MOrErr.get());
 461
 462               std::unique_ptr<TargetMachine> TM =
 463                   createTargetMachine(C, T, *MPartInCtx);
 464
 465               codegen(C, TM.get(), AddStream, ThreadId, *MPartInCtx,
 466                       CombinedIndex);
 467             },
 468             // Pass BC using std::move to ensure that it get moved rather than
 469             // copied into the thread's context.
 470             std::move(BC), ThreadCount++);
 471       },
 472       false);
 473
 474   // Because the inner lambda (which runs in a worker thread) captures our local
 475   // variables, we need to wait for the worker threads to terminate before we
 476   // can leave the function scope.
 477   CodegenThreadPool.wait();
 478 }
 479
 480 static Expected<const Target *> initAndLookupTarget(const Config &C,
 481                                                     Module &Mod) {
 482   if (!C.OverrideTriple.empty())
 483     Mod.setTargetTriple(C.OverrideTriple);
 484   else if (Mod.getTargetTriple().empty())
 485     Mod.setTargetTriple(C.DefaultTriple);
 486
 487   std::string Msg;
 488   const Target *T = TargetRegistry::lookupTarget(Mod.getTargetTriple(), Msg);
 489   if (!T)
 490     return make_error<StringError>(Msg, inconvertibleErrorCode());
 491   return T;
 492 }
 493
 494 Error lto::finalizeOptimizationRemarks(
 495     std::unique_ptr<ToolOutputFile> DiagOutputFile) {
 496   // Make sure we flush the diagnostic remarks file in case the linker doesn't
 497   // call the global destructors before exiting.
 498   if (!DiagOutputFile)
 499     return Error::success();
 500   DiagOutputFile->keep();
 501   DiagOutputFile->os().flush();
 502   return Error::success();
 503 }
 504
 505 Error lto::backend(const Config &C, AddStreamFn AddStream,
 506                    unsigned ParallelCodeGenParallelismLevel, Module &Mod,
 507                    ModuleSummaryIndex &CombinedIndex) {
 508   Expected<const Target *> TOrErr = initAndLookupTarget(C, Mod);
 509   if (!TOrErr)
 510     return TOrErr.takeError();
 511
 512   std::unique_ptr<TargetMachine> TM = createTargetMachine(C, *TOrErr, Mod);
 513
 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   if (Conf.CodeGenOnly) {
 577     codegen(Conf, TM.get(), AddStream, Task, Mod, CombinedIndex);
 578     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 579   }
 580
 581   if (Conf.PreOptModuleHook && !Conf.PreOptModuleHook(Task, Mod))
 582     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 583
 584   auto OptimizeAndCodegen =
 585       [&](Module &Mod, TargetMachine *TM,
 586           std::unique_ptr<ToolOutputFile> DiagnosticOutputFile) {
 587         if (!opt(Conf, TM, Task, Mod, /*IsThinLTO=*/true,
 588                  /*ExportSummary=*/nullptr, /*ImportSummary=*/&CombinedIndex,
 589                  CmdArgs))
 590           return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 591
 592         codegen(Conf, TM, AddStream, Task, Mod, CombinedIndex);
 593         return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 594       };
 595
 596   if (ThinLTOAssumeMerged)
 597     return OptimizeAndCodegen(Mod, TM.get(), std::move(DiagnosticOutputFile));
 598
 599   // When linking an ELF shared object, dso_local should be dropped. We
 600   // conservatively do this for -fpic.
 601   bool ClearDSOLocalOnDeclarations =
 602       TM->getTargetTriple().isOSBinFormatELF() &&
 603       TM->getRelocationModel() != Reloc::Static &&
 604       Mod.getPIELevel() == PIELevel::Default;
 605   renameModuleForThinLTO(Mod, CombinedIndex, ClearDSOLocalOnDeclarations);
 606
 607   dropDeadSymbols(Mod, DefinedGlobals, CombinedIndex);
 608
 609   thinLTOResolvePrevailingInModule(Mod, DefinedGlobals);
 610
 611   if (Conf.PostPromoteModuleHook && !Conf.PostPromoteModuleHook(Task, Mod))
 612     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 613
 614   if (!DefinedGlobals.empty())
 615     thinLTOInternalizeModule(Mod, DefinedGlobals);
 616
 617   if (Conf.PostInternalizeModuleHook &&
 618       !Conf.PostInternalizeModuleHook(Task, Mod))
 619     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 620
 621   auto ModuleLoader = [&](StringRef Identifier) {
 622     assert(Mod.getContext().isODRUniquingDebugTypes() &&
 623            "ODR Type uniquing should be enabled on the context");
 624     if (ModuleMap) {
 625       auto I = ModuleMap->find(Identifier);
 626       assert(I != ModuleMap->end());
 627       return I->second.getLazyModule(Mod.getContext(),
 628                                      /*ShouldLazyLoadMetadata=*/true,
 629                                      /*IsImporting*/ true);
 630     }
 631
 632     ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> MBOrErr =
 633         llvm::MemoryBuffer::getFile(Identifier);
 634     if (!MBOrErr)
 635       return Expected<std::unique_ptr<llvm::Module>>(make_error<StringError>(
 636           Twine("Error loading imported file ") + Identifier + " : ",
 637           MBOrErr.getError()));
 638
 639     Expected<BitcodeModule> BMOrErr = findThinLTOModule(**MBOrErr);
 640     if (!BMOrErr)
 641       return Expected<std::unique_ptr<llvm::Module>>(make_error<StringError>(
 642           Twine("Error loading imported file ") + Identifier + " : " +
 643               toString(BMOrErr.takeError()),
 644           inconvertibleErrorCode()));
 645
 646     Expected<std::unique_ptr<Module>> MOrErr =
 647         BMOrErr->getLazyModule(Mod.getContext(),
 648                                /*ShouldLazyLoadMetadata=*/true,
 649                                /*IsImporting*/ true);
 650     if (MOrErr)
 651       (*MOrErr)->setOwnedMemoryBuffer(std::move(*MBOrErr));
 652     return MOrErr;
 653   };
 654
 655   FunctionImporter Importer(CombinedIndex, ModuleLoader,
 656                             ClearDSOLocalOnDeclarations);
 657   if (Error Err = Importer.importFunctions(Mod, ImportList).takeError())
 658     return Err;
 659
 660   if (Conf.PostImportModuleHook && !Conf.PostImportModuleHook(Task, Mod))
 661     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 662
 663   return OptimizeAndCodegen(Mod, TM.get(), std::move(DiagnosticOutputFile));
 664 }
 665
 666 BitcodeModule *lto::findThinLTOModule(MutableArrayRef<BitcodeModule> BMs) {
 667   if (ThinLTOAssumeMerged && BMs.size() == 1)
 668     return BMs.begin();
 669
 670   for (BitcodeModule &BM : BMs) {
 671     Expected<BitcodeLTOInfo> LTOInfo = BM.getLTOInfo();
 672     if (LTOInfo && LTOInfo->IsThinLTO)
 673       return &BM;
 674   }
 675   return nullptr;
 676 }
 677
 678 Expected<BitcodeModule> lto::findThinLTOModule(MemoryBufferRef MBRef) {
 679   Expected<std::vector<BitcodeModule>> BMsOrErr = getBitcodeModuleList(MBRef);
 680   if (!BMsOrErr)
 681     return BMsOrErr.takeError();
 682
 683   // The bitcode file may contain multiple modules, we want the one that is
 684   // marked as being the ThinLTO module.
 685   if (const BitcodeModule *Bm = lto::findThinLTOModule(*BMsOrErr))
 686     return *Bm;
 687
 688   return make_error<StringError>("Could not find module summary",
 689                                  inconvertibleErrorCode());
 690 }
 691
 692 bool lto::initImportList(const Module &M,
 693                          const ModuleSummaryIndex &CombinedIndex,
 694                          FunctionImporter::ImportMapTy &ImportList) {
 695   if (ThinLTOAssumeMerged)
 696     return true;
 697   // We can simply import the values mentioned in the combined index, since
 698   // we should only invoke this using the individual indexes written out
 699   // via a WriteIndexesThinBackend.
 700   for (const auto &GlobalList : CombinedIndex) {
 701     // Ignore entries for undefined references.
 702     if (GlobalList.second.SummaryList.empty())
 703       continue;
 704
 705     auto GUID = GlobalList.first;
 706     for (const auto &Summary : GlobalList.second.SummaryList) {
 707       // Skip the summaries for the importing module. These are included to
 708       // e.g. record required linkage changes.
 709       if (Summary->modulePath() == M.getModuleIdentifier())
 710         continue;
 711       // Add an entry to provoke importing by thinBackend.
 712       ImportList[Summary->modulePath()].insert(GUID);
 713     }
 714   }
 715   return true;
 716 }