lib/LTO/LTOBackend.cpp

   1 //===-LTOBackend.cpp - LLVM Link Time Optimizer Backend -------------------===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // This file implements the "backend" phase of LTO, i.e. it performs
  11 // optimization and code generation on a loaded module. It is generally used
  12 // internally by the LTO class but can also be used independently, for example
  13 // to implement a standalone ThinLTO backend.
  14 //
  15 //===----------------------------------------------------------------------===//
  16
  17 #include "llvm/LTO/LTOBackend.h"
  18 #include "llvm/Analysis/AliasAnalysis.h"
  19 #include "llvm/Analysis/CGSCCPassManager.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/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/Object/ModuleSymbolTable.h"
  30 #include "llvm/Passes/PassBuilder.h"
  31 #include "llvm/Support/Error.h"
  32 #include "llvm/Support/FileSystem.h"
  33 #include "llvm/Support/MemoryBuffer.h"
  34 #include "llvm/Support/Path.h"
  35 #include "llvm/Support/Program.h"
  36 #include "llvm/Support/raw_ostream.h"
  37 #include "llvm/Support/TargetRegistry.h"
  38 #include "llvm/Support/ThreadPool.h"
  39 #include "llvm/Target/TargetMachine.h"
  40 #include "llvm/Transforms/IPO.h"
  41 #include "llvm/Transforms/IPO/PassManagerBuilder.h"
  42 #include "llvm/Transforms/Scalar/LoopPassManager.h"
  43 #include "llvm/Transforms/Utils/FunctionImportUtils.h"
  44 #include "llvm/Transforms/Utils/SplitModule.h"
  45
  46 using namespace llvm;
  47 using namespace lto;
  48
  49 LLVM_ATTRIBUTE_NORETURN static void reportOpenError(StringRef Path, Twine Msg) {
  50   errs() << "failed to open " << Path << ": " << Msg << '\n';
  51   errs().flush();
  52   exit(1);
  53 }
  54
  55 Error Config::addSaveTemps(std::string OutputFileName,
  56                            bool UseInputModulePath) {
  57   ShouldDiscardValueNames = false;
  58
  59   std::error_code EC;
  60   ResolutionFile = llvm::make_unique<raw_fd_ostream>(
  61       OutputFileName + "resolution.txt", EC, sys::fs::OpenFlags::F_Text);
  62   if (EC)
  63     return errorCodeToError(EC);
  64
  65   auto setHook = [&](std::string PathSuffix, ModuleHookFn &Hook) {
  66     // Keep track of the hook provided by the linker, which also needs to run.
  67     ModuleHookFn LinkerHook = Hook;
  68     Hook = [=](unsigned Task, const Module &M) {
  69       // If the linker's hook returned false, we need to pass that result
  70       // through.
  71       if (LinkerHook && !LinkerHook(Task, M))
  72         return false;
  73
  74       std::string PathPrefix;
  75       // If this is the combined module (not a ThinLTO backend compile) or the
  76       // user hasn't requested using the input module's path, emit to a file
  77       // named from the provided OutputFileName with the Task ID appended.
  78       if (M.getModuleIdentifier() == "ld-temp.o" || !UseInputModulePath) {
  79         PathPrefix = OutputFileName;
  80         if (Task != (unsigned)-1)
  81           PathPrefix += utostr(Task) + ".";
  82       } else
  83         PathPrefix = M.getModuleIdentifier() + ".";
  84       std::string Path = PathPrefix + PathSuffix + ".bc";
  85       std::error_code EC;
  86       raw_fd_ostream OS(Path, EC, sys::fs::OpenFlags::F_None);
  87       // Because -save-temps is a debugging feature, we report the error
  88       // directly and exit.
  89       if (EC)
  90         reportOpenError(Path, EC.message());
  91       WriteBitcodeToFile(M, OS, /*ShouldPreserveUseListOrder=*/false);
  92       return true;
  93     };
  94   };
  95
  96   setHook("0.preopt", PreOptModuleHook);
  97   setHook("1.promote", PostPromoteModuleHook);
  98   setHook("2.internalize", PostInternalizeModuleHook);
  99   setHook("3.import", PostImportModuleHook);
 100   setHook("4.opt", PostOptModuleHook);
 101   setHook("5.precodegen", PreCodeGenModuleHook);
 102
 103   CombinedIndexHook = [=](const ModuleSummaryIndex &Index) {
 104     std::string Path = OutputFileName + "index.bc";
 105     std::error_code EC;
 106     raw_fd_ostream OS(Path, EC, sys::fs::OpenFlags::F_None);
 107     // Because -save-temps is a debugging feature, we report the error
 108     // directly and exit.
 109     if (EC)
 110       reportOpenError(Path, EC.message());
 111     WriteIndexToFile(Index, OS);
 112
 113     Path = OutputFileName + "index.dot";
 114     raw_fd_ostream OSDot(Path, EC, sys::fs::OpenFlags::F_None);
 115     if (EC)
 116       reportOpenError(Path, EC.message());
 117     Index.exportToDot(OSDot);
 118     return true;
 119   };
 120
 121   return Error::success();
 122 }
 123
 124 namespace {
 125
 126 std::unique_ptr<TargetMachine>
 127 createTargetMachine(Config &Conf, const Target *TheTarget, Module &M) {
 128   StringRef TheTriple = M.getTargetTriple();
 129   SubtargetFeatures Features;
 130   Features.getDefaultSubtargetFeatures(Triple(TheTriple));
 131   for (const std::string &A : Conf.MAttrs)
 132     Features.AddFeature(A);
 133
 134   Reloc::Model RelocModel;
 135   if (Conf.RelocModel)
 136     RelocModel = *Conf.RelocModel;
 137   else
 138     RelocModel =
 139         M.getPICLevel() == PICLevel::NotPIC ? Reloc::Static : Reloc::PIC_;
 140
 141   Optional<CodeModel::Model> CodeModel;
 142   if (Conf.CodeModel)
 143     CodeModel = *Conf.CodeModel;
 144   else
 145     CodeModel = M.getCodeModel();
 146
 147   return std::unique_ptr<TargetMachine>(TheTarget->createTargetMachine(
 148       TheTriple, Conf.CPU, Features.getString(), Conf.Options, RelocModel,
 149       CodeModel, Conf.CGOptLevel));
 150 }
 151
 152 static void runNewPMPasses(Config &Conf, Module &Mod, TargetMachine *TM,
 153                            unsigned OptLevel, bool IsThinLTO,
 154                            ModuleSummaryIndex *ExportSummary,
 155                            const ModuleSummaryIndex *ImportSummary) {
 156   Optional<PGOOptions> PGOOpt;
 157   if (!Conf.SampleProfile.empty())
 158     PGOOpt = PGOOptions("", "", Conf.SampleProfile, Conf.ProfileRemapping,
 159                         false, true);
 160
 161   PassBuilder PB(TM, PGOOpt);
 162   AAManager AA;
 163
 164   // Parse a custom AA pipeline if asked to.
 165   if (!PB.parseAAPipeline(AA, "default"))
 166     report_fatal_error("Error parsing default AA pipeline");
 167
 168   LoopAnalysisManager LAM(Conf.DebugPassManager);
 169   FunctionAnalysisManager FAM(Conf.DebugPassManager);
 170   CGSCCAnalysisManager CGAM(Conf.DebugPassManager);
 171   ModuleAnalysisManager MAM(Conf.DebugPassManager);
 172
 173   // Register the AA manager first so that our version is the one used.
 174   FAM.registerPass([&] { return std::move(AA); });
 175
 176   // Register all the basic analyses with the managers.
 177   PB.registerModuleAnalyses(MAM);
 178   PB.registerCGSCCAnalyses(CGAM);
 179   PB.registerFunctionAnalyses(FAM);
 180   PB.registerLoopAnalyses(LAM);
 181   PB.crossRegisterProxies(LAM, FAM, CGAM, MAM);
 182
 183   ModulePassManager MPM(Conf.DebugPassManager);
 184   // FIXME (davide): verify the input.
 185
 186   PassBuilder::OptimizationLevel OL;
 187
 188   switch (OptLevel) {
 189   default:
 190     llvm_unreachable("Invalid optimization level");
 191   case 0:
 192     OL = PassBuilder::O0;
 193     break;
 194   case 1:
 195     OL = PassBuilder::O1;
 196     break;
 197   case 2:
 198     OL = PassBuilder::O2;
 199     break;
 200   case 3:
 201     OL = PassBuilder::O3;
 202     break;
 203   }
 204
 205   if (IsThinLTO)
 206     MPM = PB.buildThinLTODefaultPipeline(OL, Conf.DebugPassManager,
 207                                          ImportSummary);
 208   else
 209     MPM = PB.buildLTODefaultPipeline(OL, Conf.DebugPassManager, ExportSummary);
 210   MPM.run(Mod, MAM);
 211
 212   // FIXME (davide): verify the output.
 213 }
 214
 215 static void runNewPMCustomPasses(Module &Mod, TargetMachine *TM,
 216                                  std::string PipelineDesc,
 217                                  std::string AAPipelineDesc,
 218                                  bool DisableVerify) {
 219   PassBuilder PB(TM);
 220   AAManager AA;
 221
 222   // Parse a custom AA pipeline if asked to.
 223   if (!AAPipelineDesc.empty())
 224     if (!PB.parseAAPipeline(AA, AAPipelineDesc))
 225       report_fatal_error("unable to parse AA pipeline description: " +
 226                          AAPipelineDesc);
 227
 228   LoopAnalysisManager LAM;
 229   FunctionAnalysisManager FAM;
 230   CGSCCAnalysisManager CGAM;
 231   ModuleAnalysisManager MAM;
 232
 233   // Register the AA manager first so that our version is the one used.
 234   FAM.registerPass([&] { return std::move(AA); });
 235
 236   // Register all the basic analyses with the managers.
 237   PB.registerModuleAnalyses(MAM);
 238   PB.registerCGSCCAnalyses(CGAM);
 239   PB.registerFunctionAnalyses(FAM);
 240   PB.registerLoopAnalyses(LAM);
 241   PB.crossRegisterProxies(LAM, FAM, CGAM, MAM);
 242
 243   ModulePassManager MPM;
 244
 245   // Always verify the input.
 246   MPM.addPass(VerifierPass());
 247
 248   // Now, add all the passes we've been requested to.
 249   if (!PB.parsePassPipeline(MPM, PipelineDesc))
 250     report_fatal_error("unable to parse pass pipeline description: " +
 251                        PipelineDesc);
 252
 253   if (!DisableVerify)
 254     MPM.addPass(VerifierPass());
 255   MPM.run(Mod, MAM);
 256 }
 257
 258 static void runOldPMPasses(Config &Conf, Module &Mod, TargetMachine *TM,
 259                            bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
 260                            const ModuleSummaryIndex *ImportSummary) {
 261   legacy::PassManager passes;
 262   passes.add(createTargetTransformInfoWrapperPass(TM->getTargetIRAnalysis()));
 263
 264   PassManagerBuilder PMB;
 265   PMB.LibraryInfo = new TargetLibraryInfoImpl(Triple(TM->getTargetTriple()));
 266   PMB.Inliner = createFunctionInliningPass();
 267   PMB.ExportSummary = ExportSummary;
 268   PMB.ImportSummary = ImportSummary;
 269   // Unconditionally verify input since it is not verified before this
 270   // point and has unknown origin.
 271   PMB.VerifyInput = true;
 272   PMB.VerifyOutput = !Conf.DisableVerify;
 273   PMB.LoopVectorize = true;
 274   PMB.SLPVectorize = true;
 275   PMB.OptLevel = Conf.OptLevel;
 276   PMB.PGOSampleUse = Conf.SampleProfile;
 277   if (IsThinLTO)
 278     PMB.populateThinLTOPassManager(passes);
 279   else
 280     PMB.populateLTOPassManager(passes);
 281   passes.run(Mod);
 282 }
 283
 284 bool opt(Config &Conf, TargetMachine *TM, unsigned Task, Module &Mod,
 285          bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
 286          const ModuleSummaryIndex *ImportSummary) {
 287   // FIXME: Plumb the combined index into the new pass manager.
 288   if (!Conf.OptPipeline.empty())
 289     runNewPMCustomPasses(Mod, TM, Conf.OptPipeline, Conf.AAPipeline,
 290                          Conf.DisableVerify);
 291   else if (Conf.UseNewPM)
 292     runNewPMPasses(Conf, Mod, TM, Conf.OptLevel, IsThinLTO, ExportSummary,
 293                    ImportSummary);
 294   else
 295     runOldPMPasses(Conf, Mod, TM, IsThinLTO, ExportSummary, ImportSummary);
 296   return !Conf.PostOptModuleHook || Conf.PostOptModuleHook(Task, Mod);
 297 }
 298
 299 void codegen(Config &Conf, TargetMachine *TM, AddStreamFn AddStream,
 300              unsigned Task, Module &Mod) {
 301   if (Conf.PreCodeGenModuleHook && !Conf.PreCodeGenModuleHook(Task, Mod))
 302     return;
 303
 304   std::unique_ptr<ToolOutputFile> DwoOut;
 305   SmallString<1024> DwoFile(Conf.DwoPath);
 306   if (!Conf.DwoDir.empty()) {
 307     std::error_code EC;
 308     if (auto EC = llvm::sys::fs::create_directories(Conf.DwoDir))
 309       report_fatal_error("Failed to create directory " + Conf.DwoDir + ": " +
 310                          EC.message());
 311
 312     DwoFile = Conf.DwoDir;
 313     sys::path::append(DwoFile, std::to_string(Task) + ".dwo");
 314   }
 315
 316   if (!DwoFile.empty()) {
 317     std::error_code EC;
 318     TM->Options.MCOptions.SplitDwarfFile = DwoFile.str().str();
 319     DwoOut = llvm::make_unique<ToolOutputFile>(DwoFile, EC, sys::fs::F_None);
 320     if (EC)
 321       report_fatal_error("Failed to open " + DwoFile + ": " + EC.message());
 322   }
 323
 324   auto Stream = AddStream(Task);
 325   legacy::PassManager CodeGenPasses;
 326   if (TM->addPassesToEmitFile(CodeGenPasses, *Stream->OS,
 327                               DwoOut ? &DwoOut->os() : nullptr,
 328                               Conf.CGFileType))
 329     report_fatal_error("Failed to setup codegen");
 330   CodeGenPasses.run(Mod);
 331
 332   if (DwoOut)
 333     DwoOut->keep();
 334 }
 335
 336 void splitCodeGen(Config &C, TargetMachine *TM, AddStreamFn AddStream,
 337                   unsigned ParallelCodeGenParallelismLevel,
 338                   std::unique_ptr<Module> Mod) {
 339   ThreadPool CodegenThreadPool(ParallelCodeGenParallelismLevel);
 340   unsigned ThreadCount = 0;
 341   const Target *T = &TM->getTarget();
 342
 343   SplitModule(
 344       std::move(Mod), ParallelCodeGenParallelismLevel,
 345       [&](std::unique_ptr<Module> MPart) {
 346         // We want to clone the module in a new context to multi-thread the
 347         // codegen. We do it by serializing partition modules to bitcode
 348         // (while still on the main thread, in order to avoid data races) and
 349         // spinning up new threads which deserialize the partitions into
 350         // separate contexts.
 351         // FIXME: Provide a more direct way to do this in LLVM.
 352         SmallString<0> BC;
 353         raw_svector_ostream BCOS(BC);
 354         WriteBitcodeToFile(*MPart, BCOS);
 355
 356         // Enqueue the task
 357         CodegenThreadPool.async(
 358             [&](const SmallString<0> &BC, unsigned ThreadId) {
 359               LTOLLVMContext Ctx(C);
 360               Expected<std::unique_ptr<Module>> MOrErr = parseBitcodeFile(
 361                   MemoryBufferRef(StringRef(BC.data(), BC.size()), "ld-temp.o"),
 362                   Ctx);
 363               if (!MOrErr)
 364                 report_fatal_error("Failed to read bitcode");
 365               std::unique_ptr<Module> MPartInCtx = std::move(MOrErr.get());
 366
 367               std::unique_ptr<TargetMachine> TM =
 368                   createTargetMachine(C, T, *MPartInCtx);
 369
 370               codegen(C, TM.get(), AddStream, ThreadId, *MPartInCtx);
 371             },
 372             // Pass BC using std::move to ensure that it get moved rather than
 373             // copied into the thread's context.
 374             std::move(BC), ThreadCount++);
 375       },
 376       false);
 377
 378   // Because the inner lambda (which runs in a worker thread) captures our local
 379   // variables, we need to wait for the worker threads to terminate before we
 380   // can leave the function scope.
 381   CodegenThreadPool.wait();
 382 }
 383
 384 Expected<const Target *> initAndLookupTarget(Config &C, Module &Mod) {
 385   if (!C.OverrideTriple.empty())
 386     Mod.setTargetTriple(C.OverrideTriple);
 387   else if (Mod.getTargetTriple().empty())
 388     Mod.setTargetTriple(C.DefaultTriple);
 389
 390   std::string Msg;
 391   const Target *T = TargetRegistry::lookupTarget(Mod.getTargetTriple(), Msg);
 392   if (!T)
 393     return make_error<StringError>(Msg, inconvertibleErrorCode());
 394   return T;
 395 }
 396
 397 }
 398
 399 static Error
 400 finalizeOptimizationRemarks(std::unique_ptr<ToolOutputFile> DiagOutputFile) {
 401   // Make sure we flush the diagnostic remarks file in case the linker doesn't
 402   // call the global destructors before exiting.
 403   if (!DiagOutputFile)
 404     return Error::success();
 405   DiagOutputFile->keep();
 406   DiagOutputFile->os().flush();
 407   return Error::success();
 408 }
 409
 410 Error lto::backend(Config &C, AddStreamFn AddStream,
 411                    unsigned ParallelCodeGenParallelismLevel,
 412                    std::unique_ptr<Module> Mod,
 413                    ModuleSummaryIndex &CombinedIndex) {
 414   Expected<const Target *> TOrErr = initAndLookupTarget(C, *Mod);
 415   if (!TOrErr)
 416     return TOrErr.takeError();
 417
 418   std::unique_ptr<TargetMachine> TM = createTargetMachine(C, *TOrErr, *Mod);
 419
 420   // Setup optimization remarks.
 421   auto DiagFileOrErr = lto::setupOptimizationRemarks(
 422       Mod->getContext(), C.RemarksFilename, C.RemarksWithHotness);
 423   if (!DiagFileOrErr)
 424     return DiagFileOrErr.takeError();
 425   auto DiagnosticOutputFile = std::move(*DiagFileOrErr);
 426
 427   if (!C.CodeGenOnly) {
 428     if (!opt(C, TM.get(), 0, *Mod, /*IsThinLTO=*/false,
 429              /*ExportSummary=*/&CombinedIndex, /*ImportSummary=*/nullptr))
 430       return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 431   }
 432
 433   if (ParallelCodeGenParallelismLevel == 1) {
 434     codegen(C, TM.get(), AddStream, 0, *Mod);
 435   } else {
 436     splitCodeGen(C, TM.get(), AddStream, ParallelCodeGenParallelismLevel,
 437                  std::move(Mod));
 438   }
 439   return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 440 }
 441
 442 static void dropDeadSymbols(Module &Mod, const GVSummaryMapTy &DefinedGlobals,
 443                             const ModuleSummaryIndex &Index) {
 444   std::vector<GlobalValue*> DeadGVs;
 445   for (auto &GV : Mod.global_values())
 446     if (GlobalValueSummary *GVS = DefinedGlobals.lookup(GV.getGUID()))
 447       if (!Index.isGlobalValueLive(GVS)) {
 448         DeadGVs.push_back(&GV);
 449         convertToDeclaration(GV);
 450       }
 451
 452   // Now that all dead bodies have been dropped, delete the actual objects
 453   // themselves when possible.
 454   for (GlobalValue *GV : DeadGVs) {
 455     GV->removeDeadConstantUsers();
 456     // Might reference something defined in native object (i.e. dropped a
 457     // non-prevailing IR def, but we need to keep the declaration).
 458     if (GV->use_empty())
 459       GV->eraseFromParent();
 460   }
 461 }
 462
 463 Error lto::thinBackend(Config &Conf, unsigned Task, AddStreamFn AddStream,
 464                        Module &Mod, const ModuleSummaryIndex &CombinedIndex,
 465                        const FunctionImporter::ImportMapTy &ImportList,
 466                        const GVSummaryMapTy &DefinedGlobals,
 467                        MapVector<StringRef, BitcodeModule> &ModuleMap) {
 468   Expected<const Target *> TOrErr = initAndLookupTarget(Conf, Mod);
 469   if (!TOrErr)
 470     return TOrErr.takeError();
 471
 472   std::unique_ptr<TargetMachine> TM = createTargetMachine(Conf, *TOrErr, Mod);
 473
 474   // Setup optimization remarks.
 475   auto DiagFileOrErr = lto::setupOptimizationRemarks(
 476       Mod.getContext(), Conf.RemarksFilename, Conf.RemarksWithHotness, Task);
 477   if (!DiagFileOrErr)
 478     return DiagFileOrErr.takeError();
 479   auto DiagnosticOutputFile = std::move(*DiagFileOrErr);
 480
 481   if (Conf.CodeGenOnly) {
 482     codegen(Conf, TM.get(), AddStream, Task, Mod);
 483     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 484   }
 485
 486   if (Conf.PreOptModuleHook && !Conf.PreOptModuleHook(Task, Mod))
 487     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 488
 489   renameModuleForThinLTO(Mod, CombinedIndex);
 490
 491   dropDeadSymbols(Mod, DefinedGlobals, CombinedIndex);
 492
 493   thinLTOResolveWeakForLinkerModule(Mod, DefinedGlobals);
 494
 495   if (Conf.PostPromoteModuleHook && !Conf.PostPromoteModuleHook(Task, Mod))
 496     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 497
 498   if (!DefinedGlobals.empty())
 499     thinLTOInternalizeModule(Mod, DefinedGlobals);
 500
 501   if (Conf.PostInternalizeModuleHook &&
 502       !Conf.PostInternalizeModuleHook(Task, Mod))
 503     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 504
 505   auto ModuleLoader = [&](StringRef Identifier) {
 506     assert(Mod.getContext().isODRUniquingDebugTypes() &&
 507            "ODR Type uniquing should be enabled on the context");
 508     auto I = ModuleMap.find(Identifier);
 509     assert(I != ModuleMap.end());
 510     return I->second.getLazyModule(Mod.getContext(),
 511                                    /*ShouldLazyLoadMetadata=*/true,
 512                                    /*IsImporting*/ true);
 513   };
 514
 515   FunctionImporter Importer(CombinedIndex, ModuleLoader);
 516   if (Error Err = Importer.importFunctions(Mod, ImportList).takeError())
 517     return Err;
 518
 519   if (Conf.PostImportModuleHook && !Conf.PostImportModuleHook(Task, Mod))
 520     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 521
 522   if (!opt(Conf, TM.get(), Task, Mod, /*IsThinLTO=*/true,
 523            /*ExportSummary=*/nullptr, /*ImportSummary=*/&CombinedIndex))
 524     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 525
 526   codegen(Conf, TM.get(), AddStream, Task, Mod);
 527   return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
 528 }