1 //===-ThinLTOCodeGenerator.cpp - LLVM Link Time Optimizer -----------------===// 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 Thin Link Time Optimization library. This library is 10 // intended to be used by linker to optimize code at link time. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/LTO/legacy/ThinLTOCodeGenerator.h" 15 #include "llvm/Support/CommandLine.h" 16 17 #include "llvm/ADT/Statistic.h" 18 #include "llvm/ADT/StringExtras.h" 19 #include "llvm/Analysis/ModuleSummaryAnalysis.h" 20 #include "llvm/Analysis/ProfileSummaryInfo.h" 21 #include "llvm/Analysis/TargetLibraryInfo.h" 22 #include "llvm/Analysis/TargetTransformInfo.h" 23 #include "llvm/Bitcode/BitcodeReader.h" 24 #include "llvm/Bitcode/BitcodeWriter.h" 25 #include "llvm/Bitcode/BitcodeWriterPass.h" 26 #include "llvm/Config/llvm-config.h" 27 #include "llvm/IR/DebugInfo.h" 28 #include "llvm/IR/DiagnosticPrinter.h" 29 #include "llvm/IR/LLVMContext.h" 30 #include "llvm/IR/LLVMRemarkStreamer.h" 31 #include "llvm/IR/LegacyPassManager.h" 32 #include "llvm/IR/Mangler.h" 33 #include "llvm/IR/PassTimingInfo.h" 34 #include "llvm/IR/Verifier.h" 35 #include "llvm/IRReader/IRReader.h" 36 #include "llvm/LTO/LTO.h" 37 #include "llvm/LTO/SummaryBasedOptimizations.h" 38 #include "llvm/MC/SubtargetFeature.h" 39 #include "llvm/Object/IRObjectFile.h" 40 #include "llvm/Support/CachePruning.h" 41 #include "llvm/Support/Debug.h" 42 #include "llvm/Support/Error.h" 43 #include "llvm/Support/FileUtilities.h" 44 #include "llvm/Support/Path.h" 45 #include "llvm/Support/SHA1.h" 46 #include "llvm/Support/SmallVectorMemoryBuffer.h" 47 #include "llvm/Support/TargetRegistry.h" 48 #include "llvm/Support/ThreadPool.h" 49 #include "llvm/Support/Threading.h" 50 #include "llvm/Support/ToolOutputFile.h" 51 #include "llvm/Target/TargetMachine.h" 52 #include "llvm/Transforms/IPO.h" 53 #include "llvm/Transforms/IPO/FunctionImport.h" 54 #include "llvm/Transforms/IPO/Internalize.h" 55 #include "llvm/Transforms/IPO/PassManagerBuilder.h" 56 #include "llvm/Transforms/IPO/WholeProgramDevirt.h" 57 #include "llvm/Transforms/ObjCARC.h" 58 #include "llvm/Transforms/Utils/FunctionImportUtils.h" 59 60 #include <numeric> 61 62 #if !defined(_MSC_VER) && !defined(__MINGW32__) 63 #include <unistd.h> 64 #else 65 #include <io.h> 66 #endif 67 68 using namespace llvm; 69 70 #define DEBUG_TYPE "thinlto" 71 72 namespace llvm { 73 // Flags -discard-value-names, defined in LTOCodeGenerator.cpp 74 extern cl::opt<bool> LTODiscardValueNames; 75 extern cl::opt<std::string> RemarksFilename; 76 extern cl::opt<std::string> RemarksPasses; 77 extern cl::opt<bool> RemarksWithHotness; 78 extern cl::opt<std::string> RemarksFormat; 79 } 80 81 namespace { 82 83 // Default to using all available threads in the system, but using only one 84 // thred per core, as indicated by the usage of 85 // heavyweight_hardware_concurrency() below. 86 static cl::opt<int> ThreadCount("threads", cl::init(0)); 87 88 // Simple helper to save temporary files for debug. 89 static void saveTempBitcode(const Module &TheModule, StringRef TempDir, 90 unsigned count, StringRef Suffix) { 91 if (TempDir.empty()) 92 return; 93 // User asked to save temps, let dump the bitcode file after import. 94 std::string SaveTempPath = (TempDir + llvm::Twine(count) + Suffix).str(); 95 std::error_code EC; 96 raw_fd_ostream OS(SaveTempPath, EC, sys::fs::OF_None); 97 if (EC) 98 report_fatal_error(Twine("Failed to open ") + SaveTempPath + 99 " to save optimized bitcode\n"); 100 WriteBitcodeToFile(TheModule, OS, /* ShouldPreserveUseListOrder */ true); 101 } 102 103 static const GlobalValueSummary * 104 getFirstDefinitionForLinker(const GlobalValueSummaryList &GVSummaryList) { 105 // If there is any strong definition anywhere, get it. 106 auto StrongDefForLinker = llvm::find_if( 107 GVSummaryList, [](const std::unique_ptr<GlobalValueSummary> &Summary) { 108 auto Linkage = Summary->linkage(); 109 return !GlobalValue::isAvailableExternallyLinkage(Linkage) && 110 !GlobalValue::isWeakForLinker(Linkage); 111 }); 112 if (StrongDefForLinker != GVSummaryList.end()) 113 return StrongDefForLinker->get(); 114 // Get the first *linker visible* definition for this global in the summary 115 // list. 116 auto FirstDefForLinker = llvm::find_if( 117 GVSummaryList, [](const std::unique_ptr<GlobalValueSummary> &Summary) { 118 auto Linkage = Summary->linkage(); 119 return !GlobalValue::isAvailableExternallyLinkage(Linkage); 120 }); 121 // Extern templates can be emitted as available_externally. 122 if (FirstDefForLinker == GVSummaryList.end()) 123 return nullptr; 124 return FirstDefForLinker->get(); 125 } 126 127 // Populate map of GUID to the prevailing copy for any multiply defined 128 // symbols. Currently assume first copy is prevailing, or any strong 129 // definition. Can be refined with Linker information in the future. 130 static void computePrevailingCopies( 131 const ModuleSummaryIndex &Index, 132 DenseMap<GlobalValue::GUID, const GlobalValueSummary *> &PrevailingCopy) { 133 auto HasMultipleCopies = [&](const GlobalValueSummaryList &GVSummaryList) { 134 return GVSummaryList.size() > 1; 135 }; 136 137 for (auto &I : Index) { 138 if (HasMultipleCopies(I.second.SummaryList)) 139 PrevailingCopy[I.first] = 140 getFirstDefinitionForLinker(I.second.SummaryList); 141 } 142 } 143 144 static StringMap<lto::InputFile *> 145 generateModuleMap(std::vector<std::unique_ptr<lto::InputFile>> &Modules) { 146 StringMap<lto::InputFile *> ModuleMap; 147 for (auto &M : Modules) { 148 assert(ModuleMap.find(M->getName()) == ModuleMap.end() && 149 "Expect unique Buffer Identifier"); 150 ModuleMap[M->getName()] = M.get(); 151 } 152 return ModuleMap; 153 } 154 155 static void promoteModule(Module &TheModule, const ModuleSummaryIndex &Index, 156 bool ClearDSOLocalOnDeclarations) { 157 if (renameModuleForThinLTO(TheModule, Index, ClearDSOLocalOnDeclarations)) 158 report_fatal_error("renameModuleForThinLTO failed"); 159 } 160 161 namespace { 162 class ThinLTODiagnosticInfo : public DiagnosticInfo { 163 const Twine &Msg; 164 public: 165 ThinLTODiagnosticInfo(const Twine &DiagMsg, 166 DiagnosticSeverity Severity = DS_Error) 167 : DiagnosticInfo(DK_Linker, Severity), Msg(DiagMsg) {} 168 void print(DiagnosticPrinter &DP) const override { DP << Msg; } 169 }; 170 } 171 172 /// Verify the module and strip broken debug info. 173 static void verifyLoadedModule(Module &TheModule) { 174 bool BrokenDebugInfo = false; 175 if (verifyModule(TheModule, &dbgs(), &BrokenDebugInfo)) 176 report_fatal_error("Broken module found, compilation aborted!"); 177 if (BrokenDebugInfo) { 178 TheModule.getContext().diagnose(ThinLTODiagnosticInfo( 179 "Invalid debug info found, debug info will be stripped", DS_Warning)); 180 StripDebugInfo(TheModule); 181 } 182 } 183 184 static std::unique_ptr<Module> loadModuleFromInput(lto::InputFile *Input, 185 LLVMContext &Context, 186 bool Lazy, 187 bool IsImporting) { 188 auto &Mod = Input->getSingleBitcodeModule(); 189 SMDiagnostic Err; 190 Expected<std::unique_ptr<Module>> ModuleOrErr = 191 Lazy ? Mod.getLazyModule(Context, 192 /* ShouldLazyLoadMetadata */ true, IsImporting) 193 : Mod.parseModule(Context); 194 if (!ModuleOrErr) { 195 handleAllErrors(ModuleOrErr.takeError(), [&](ErrorInfoBase &EIB) { 196 SMDiagnostic Err = SMDiagnostic(Mod.getModuleIdentifier(), 197 SourceMgr::DK_Error, EIB.message()); 198 Err.print("ThinLTO", errs()); 199 }); 200 report_fatal_error("Can't load module, abort."); 201 } 202 if (!Lazy) 203 verifyLoadedModule(*ModuleOrErr.get()); 204 return std::move(*ModuleOrErr); 205 } 206 207 static void 208 crossImportIntoModule(Module &TheModule, const ModuleSummaryIndex &Index, 209 StringMap<lto::InputFile *> &ModuleMap, 210 const FunctionImporter::ImportMapTy &ImportList, 211 bool ClearDSOLocalOnDeclarations) { 212 auto Loader = [&](StringRef Identifier) { 213 auto &Input = ModuleMap[Identifier]; 214 return loadModuleFromInput(Input, TheModule.getContext(), 215 /*Lazy=*/true, /*IsImporting*/ true); 216 }; 217 218 FunctionImporter Importer(Index, Loader, ClearDSOLocalOnDeclarations); 219 Expected<bool> Result = Importer.importFunctions(TheModule, ImportList); 220 if (!Result) { 221 handleAllErrors(Result.takeError(), [&](ErrorInfoBase &EIB) { 222 SMDiagnostic Err = SMDiagnostic(TheModule.getModuleIdentifier(), 223 SourceMgr::DK_Error, EIB.message()); 224 Err.print("ThinLTO", errs()); 225 }); 226 report_fatal_error("importFunctions failed"); 227 } 228 // Verify again after cross-importing. 229 verifyLoadedModule(TheModule); 230 } 231 232 static void optimizeModule(Module &TheModule, TargetMachine &TM, 233 unsigned OptLevel, bool Freestanding, 234 ModuleSummaryIndex *Index) { 235 // Populate the PassManager 236 PassManagerBuilder PMB; 237 PMB.LibraryInfo = new TargetLibraryInfoImpl(TM.getTargetTriple()); 238 if (Freestanding) 239 PMB.LibraryInfo->disableAllFunctions(); 240 PMB.Inliner = createFunctionInliningPass(); 241 // FIXME: should get it from the bitcode? 242 PMB.OptLevel = OptLevel; 243 PMB.LoopVectorize = true; 244 PMB.SLPVectorize = true; 245 // Already did this in verifyLoadedModule(). 246 PMB.VerifyInput = false; 247 PMB.VerifyOutput = false; 248 PMB.ImportSummary = Index; 249 250 legacy::PassManager PM; 251 252 // Add the TTI (required to inform the vectorizer about register size for 253 // instance) 254 PM.add(createTargetTransformInfoWrapperPass(TM.getTargetIRAnalysis())); 255 256 // Add optimizations 257 PMB.populateThinLTOPassManager(PM); 258 259 PM.run(TheModule); 260 } 261 262 static void 263 addUsedSymbolToPreservedGUID(const lto::InputFile &File, 264 DenseSet<GlobalValue::GUID> &PreservedGUID) { 265 for (const auto &Sym : File.symbols()) { 266 if (Sym.isUsed()) 267 PreservedGUID.insert(GlobalValue::getGUID(Sym.getIRName())); 268 } 269 } 270 271 // Convert the PreservedSymbols map from "Name" based to "GUID" based. 272 static void computeGUIDPreservedSymbols(const lto::InputFile &File, 273 const StringSet<> &PreservedSymbols, 274 const Triple &TheTriple, 275 DenseSet<GlobalValue::GUID> &GUIDs) { 276 // Iterate the symbols in the input file and if the input has preserved symbol 277 // compute the GUID for the symbol. 278 for (const auto &Sym : File.symbols()) { 279 if (PreservedSymbols.count(Sym.getName()) && !Sym.getIRName().empty()) 280 GUIDs.insert(GlobalValue::getGUID(GlobalValue::getGlobalIdentifier( 281 Sym.getIRName(), GlobalValue::ExternalLinkage, ""))); 282 } 283 } 284 285 static DenseSet<GlobalValue::GUID> 286 computeGUIDPreservedSymbols(const lto::InputFile &File, 287 const StringSet<> &PreservedSymbols, 288 const Triple &TheTriple) { 289 DenseSet<GlobalValue::GUID> GUIDPreservedSymbols(PreservedSymbols.size()); 290 computeGUIDPreservedSymbols(File, PreservedSymbols, TheTriple, 291 GUIDPreservedSymbols); 292 return GUIDPreservedSymbols; 293 } 294 295 std::unique_ptr<MemoryBuffer> codegenModule(Module &TheModule, 296 TargetMachine &TM) { 297 SmallVector<char, 128> OutputBuffer; 298 299 // CodeGen 300 { 301 raw_svector_ostream OS(OutputBuffer); 302 legacy::PassManager PM; 303 304 // If the bitcode files contain ARC code and were compiled with optimization, 305 // the ObjCARCContractPass must be run, so do it unconditionally here. 306 PM.add(createObjCARCContractPass()); 307 308 // Setup the codegen now. 309 if (TM.addPassesToEmitFile(PM, OS, nullptr, CGFT_ObjectFile, 310 /* DisableVerify */ true)) 311 report_fatal_error("Failed to setup codegen"); 312 313 // Run codegen now. resulting binary is in OutputBuffer. 314 PM.run(TheModule); 315 } 316 return std::make_unique<SmallVectorMemoryBuffer>(std::move(OutputBuffer)); 317 } 318 319 /// Manage caching for a single Module. 320 class ModuleCacheEntry { 321 SmallString<128> EntryPath; 322 323 public: 324 // Create a cache entry. This compute a unique hash for the Module considering 325 // the current list of export/import, and offer an interface to query to 326 // access the content in the cache. 327 ModuleCacheEntry( 328 StringRef CachePath, const ModuleSummaryIndex &Index, StringRef ModuleID, 329 const FunctionImporter::ImportMapTy &ImportList, 330 const FunctionImporter::ExportSetTy &ExportList, 331 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 332 const GVSummaryMapTy &DefinedGVSummaries, unsigned OptLevel, 333 bool Freestanding, const TargetMachineBuilder &TMBuilder) { 334 if (CachePath.empty()) 335 return; 336 337 if (!Index.modulePaths().count(ModuleID)) 338 // The module does not have an entry, it can't have a hash at all 339 return; 340 341 if (all_of(Index.getModuleHash(ModuleID), 342 [](uint32_t V) { return V == 0; })) 343 // No hash entry, no caching! 344 return; 345 346 llvm::lto::Config Conf; 347 Conf.OptLevel = OptLevel; 348 Conf.Options = TMBuilder.Options; 349 Conf.CPU = TMBuilder.MCpu; 350 Conf.MAttrs.push_back(TMBuilder.MAttr); 351 Conf.RelocModel = TMBuilder.RelocModel; 352 Conf.CGOptLevel = TMBuilder.CGOptLevel; 353 Conf.Freestanding = Freestanding; 354 SmallString<40> Key; 355 computeLTOCacheKey(Key, Conf, Index, ModuleID, ImportList, ExportList, 356 ResolvedODR, DefinedGVSummaries); 357 358 // This choice of file name allows the cache to be pruned (see pruneCache() 359 // in include/llvm/Support/CachePruning.h). 360 sys::path::append(EntryPath, CachePath, "llvmcache-" + Key); 361 } 362 363 // Access the path to this entry in the cache. 364 StringRef getEntryPath() { return EntryPath; } 365 366 // Try loading the buffer for this cache entry. 367 ErrorOr<std::unique_ptr<MemoryBuffer>> tryLoadingBuffer() { 368 if (EntryPath.empty()) 369 return std::error_code(); 370 SmallString<64> ResultPath; 371 Expected<sys::fs::file_t> FDOrErr = sys::fs::openNativeFileForRead( 372 Twine(EntryPath), sys::fs::OF_UpdateAtime, &ResultPath); 373 if (!FDOrErr) 374 return errorToErrorCode(FDOrErr.takeError()); 375 ErrorOr<std::unique_ptr<MemoryBuffer>> MBOrErr = MemoryBuffer::getOpenFile( 376 *FDOrErr, EntryPath, /*FileSize=*/-1, /*RequiresNullTerminator=*/false); 377 sys::fs::closeFile(*FDOrErr); 378 return MBOrErr; 379 } 380 381 // Cache the Produced object file 382 void write(const MemoryBuffer &OutputBuffer) { 383 if (EntryPath.empty()) 384 return; 385 386 // Write to a temporary to avoid race condition 387 SmallString<128> TempFilename; 388 SmallString<128> CachePath(EntryPath); 389 llvm::sys::path::remove_filename(CachePath); 390 sys::path::append(TempFilename, CachePath, "Thin-%%%%%%.tmp.o"); 391 392 if (auto Err = handleErrors( 393 llvm::writeFileAtomically(TempFilename, EntryPath, 394 OutputBuffer.getBuffer()), 395 [](const llvm::AtomicFileWriteError &E) { 396 std::string ErrorMsgBuffer; 397 llvm::raw_string_ostream S(ErrorMsgBuffer); 398 E.log(S); 399 400 if (E.Error == 401 llvm::atomic_write_error::failed_to_create_uniq_file) { 402 errs() << "Error: " << ErrorMsgBuffer << "\n"; 403 report_fatal_error("ThinLTO: Can't get a temporary file"); 404 } 405 })) { 406 // FIXME 407 consumeError(std::move(Err)); 408 } 409 } 410 }; 411 412 static std::unique_ptr<MemoryBuffer> 413 ProcessThinLTOModule(Module &TheModule, ModuleSummaryIndex &Index, 414 StringMap<lto::InputFile *> &ModuleMap, TargetMachine &TM, 415 const FunctionImporter::ImportMapTy &ImportList, 416 const FunctionImporter::ExportSetTy &ExportList, 417 const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols, 418 const GVSummaryMapTy &DefinedGlobals, 419 const ThinLTOCodeGenerator::CachingOptions &CacheOptions, 420 bool DisableCodeGen, StringRef SaveTempsDir, 421 bool Freestanding, unsigned OptLevel, unsigned count) { 422 423 // "Benchmark"-like optimization: single-source case 424 bool SingleModule = (ModuleMap.size() == 1); 425 426 // When linking an ELF shared object, dso_local should be dropped. We 427 // conservatively do this for -fpic. 428 bool ClearDSOLocalOnDeclarations = 429 TM.getTargetTriple().isOSBinFormatELF() && 430 TM.getRelocationModel() != Reloc::Static && 431 TheModule.getPIELevel() == PIELevel::Default; 432 433 if (!SingleModule) { 434 promoteModule(TheModule, Index, ClearDSOLocalOnDeclarations); 435 436 // Apply summary-based prevailing-symbol resolution decisions. 437 thinLTOResolvePrevailingInModule(TheModule, DefinedGlobals); 438 439 // Save temps: after promotion. 440 saveTempBitcode(TheModule, SaveTempsDir, count, ".1.promoted.bc"); 441 } 442 443 // Be friendly and don't nuke totally the module when the client didn't 444 // supply anything to preserve. 445 if (!ExportList.empty() || !GUIDPreservedSymbols.empty()) { 446 // Apply summary-based internalization decisions. 447 thinLTOInternalizeModule(TheModule, DefinedGlobals); 448 } 449 450 // Save internalized bitcode 451 saveTempBitcode(TheModule, SaveTempsDir, count, ".2.internalized.bc"); 452 453 if (!SingleModule) { 454 crossImportIntoModule(TheModule, Index, ModuleMap, ImportList, 455 ClearDSOLocalOnDeclarations); 456 457 // Save temps: after cross-module import. 458 saveTempBitcode(TheModule, SaveTempsDir, count, ".3.imported.bc"); 459 } 460 461 optimizeModule(TheModule, TM, OptLevel, Freestanding, &Index); 462 463 saveTempBitcode(TheModule, SaveTempsDir, count, ".4.opt.bc"); 464 465 if (DisableCodeGen) { 466 // Configured to stop before CodeGen, serialize the bitcode and return. 467 SmallVector<char, 128> OutputBuffer; 468 { 469 raw_svector_ostream OS(OutputBuffer); 470 ProfileSummaryInfo PSI(TheModule); 471 auto Index = buildModuleSummaryIndex(TheModule, nullptr, &PSI); 472 WriteBitcodeToFile(TheModule, OS, true, &Index); 473 } 474 return std::make_unique<SmallVectorMemoryBuffer>(std::move(OutputBuffer)); 475 } 476 477 return codegenModule(TheModule, TM); 478 } 479 480 /// Resolve prevailing symbols. Record resolutions in the \p ResolvedODR map 481 /// for caching, and in the \p Index for application during the ThinLTO 482 /// backends. This is needed for correctness for exported symbols (ensure 483 /// at least one copy kept) and a compile-time optimization (to drop duplicate 484 /// copies when possible). 485 static void resolvePrevailingInIndex( 486 ModuleSummaryIndex &Index, 487 StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> 488 &ResolvedODR, 489 const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols, 490 const DenseMap<GlobalValue::GUID, const GlobalValueSummary *> 491 &PrevailingCopy) { 492 493 auto isPrevailing = [&](GlobalValue::GUID GUID, const GlobalValueSummary *S) { 494 const auto &Prevailing = PrevailingCopy.find(GUID); 495 // Not in map means that there was only one copy, which must be prevailing. 496 if (Prevailing == PrevailingCopy.end()) 497 return true; 498 return Prevailing->second == S; 499 }; 500 501 auto recordNewLinkage = [&](StringRef ModuleIdentifier, 502 GlobalValue::GUID GUID, 503 GlobalValue::LinkageTypes NewLinkage) { 504 ResolvedODR[ModuleIdentifier][GUID] = NewLinkage; 505 }; 506 507 thinLTOResolvePrevailingInIndex(Index, isPrevailing, recordNewLinkage, 508 GUIDPreservedSymbols); 509 } 510 511 // Initialize the TargetMachine builder for a given Triple 512 static void initTMBuilder(TargetMachineBuilder &TMBuilder, 513 const Triple &TheTriple) { 514 // Set a default CPU for Darwin triples (copied from LTOCodeGenerator). 515 // FIXME this looks pretty terrible... 516 if (TMBuilder.MCpu.empty() && TheTriple.isOSDarwin()) { 517 if (TheTriple.getArch() == llvm::Triple::x86_64) 518 TMBuilder.MCpu = "core2"; 519 else if (TheTriple.getArch() == llvm::Triple::x86) 520 TMBuilder.MCpu = "yonah"; 521 else if (TheTriple.getArch() == llvm::Triple::aarch64 || 522 TheTriple.getArch() == llvm::Triple::aarch64_32) 523 TMBuilder.MCpu = "cyclone"; 524 } 525 TMBuilder.TheTriple = std::move(TheTriple); 526 } 527 528 } // end anonymous namespace 529 530 void ThinLTOCodeGenerator::addModule(StringRef Identifier, StringRef Data) { 531 MemoryBufferRef Buffer(Data, Identifier); 532 533 auto InputOrError = lto::InputFile::create(Buffer); 534 if (!InputOrError) 535 report_fatal_error("ThinLTO cannot create input file: " + 536 toString(InputOrError.takeError())); 537 538 auto TripleStr = (*InputOrError)->getTargetTriple(); 539 Triple TheTriple(TripleStr); 540 541 if (Modules.empty()) 542 initTMBuilder(TMBuilder, Triple(TheTriple)); 543 else if (TMBuilder.TheTriple != TheTriple) { 544 if (!TMBuilder.TheTriple.isCompatibleWith(TheTriple)) 545 report_fatal_error("ThinLTO modules with incompatible triples not " 546 "supported"); 547 initTMBuilder(TMBuilder, Triple(TMBuilder.TheTriple.merge(TheTriple))); 548 } 549 550 Modules.emplace_back(std::move(*InputOrError)); 551 } 552 553 void ThinLTOCodeGenerator::preserveSymbol(StringRef Name) { 554 PreservedSymbols.insert(Name); 555 } 556 557 void ThinLTOCodeGenerator::crossReferenceSymbol(StringRef Name) { 558 // FIXME: At the moment, we don't take advantage of this extra information, 559 // we're conservatively considering cross-references as preserved. 560 // CrossReferencedSymbols.insert(Name); 561 PreservedSymbols.insert(Name); 562 } 563 564 // TargetMachine factory 565 std::unique_ptr<TargetMachine> TargetMachineBuilder::create() const { 566 std::string ErrMsg; 567 const Target *TheTarget = 568 TargetRegistry::lookupTarget(TheTriple.str(), ErrMsg); 569 if (!TheTarget) { 570 report_fatal_error("Can't load target for this Triple: " + ErrMsg); 571 } 572 573 // Use MAttr as the default set of features. 574 SubtargetFeatures Features(MAttr); 575 Features.getDefaultSubtargetFeatures(TheTriple); 576 std::string FeatureStr = Features.getString(); 577 578 std::unique_ptr<TargetMachine> TM( 579 TheTarget->createTargetMachine(TheTriple.str(), MCpu, FeatureStr, Options, 580 RelocModel, None, CGOptLevel)); 581 assert(TM && "Cannot create target machine"); 582 583 return TM; 584 } 585 586 /** 587 * Produce the combined summary index from all the bitcode files: 588 * "thin-link". 589 */ 590 std::unique_ptr<ModuleSummaryIndex> ThinLTOCodeGenerator::linkCombinedIndex() { 591 std::unique_ptr<ModuleSummaryIndex> CombinedIndex = 592 std::make_unique<ModuleSummaryIndex>(/*HaveGVs=*/false); 593 uint64_t NextModuleId = 0; 594 for (auto &Mod : Modules) { 595 auto &M = Mod->getSingleBitcodeModule(); 596 if (Error Err = 597 M.readSummary(*CombinedIndex, Mod->getName(), NextModuleId++)) { 598 // FIXME diagnose 599 logAllUnhandledErrors( 600 std::move(Err), errs(), 601 "error: can't create module summary index for buffer: "); 602 return nullptr; 603 } 604 } 605 return CombinedIndex; 606 } 607 608 namespace { 609 struct IsExported { 610 const StringMap<FunctionImporter::ExportSetTy> &ExportLists; 611 const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols; 612 613 IsExported(const StringMap<FunctionImporter::ExportSetTy> &ExportLists, 614 const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols) 615 : ExportLists(ExportLists), GUIDPreservedSymbols(GUIDPreservedSymbols) {} 616 617 bool operator()(StringRef ModuleIdentifier, ValueInfo VI) const { 618 const auto &ExportList = ExportLists.find(ModuleIdentifier); 619 return (ExportList != ExportLists.end() && ExportList->second.count(VI)) || 620 GUIDPreservedSymbols.count(VI.getGUID()); 621 } 622 }; 623 624 struct IsPrevailing { 625 const DenseMap<GlobalValue::GUID, const GlobalValueSummary *> &PrevailingCopy; 626 IsPrevailing(const DenseMap<GlobalValue::GUID, const GlobalValueSummary *> 627 &PrevailingCopy) 628 : PrevailingCopy(PrevailingCopy) {} 629 630 bool operator()(GlobalValue::GUID GUID, const GlobalValueSummary *S) const { 631 const auto &Prevailing = PrevailingCopy.find(GUID); 632 // Not in map means that there was only one copy, which must be prevailing. 633 if (Prevailing == PrevailingCopy.end()) 634 return true; 635 return Prevailing->second == S; 636 }; 637 }; 638 } // namespace 639 640 static void computeDeadSymbolsInIndex( 641 ModuleSummaryIndex &Index, 642 const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols) { 643 // We have no symbols resolution available. And can't do any better now in the 644 // case where the prevailing symbol is in a native object. It can be refined 645 // with linker information in the future. 646 auto isPrevailing = [&](GlobalValue::GUID G) { 647 return PrevailingType::Unknown; 648 }; 649 computeDeadSymbolsWithConstProp(Index, GUIDPreservedSymbols, isPrevailing, 650 /* ImportEnabled = */ true); 651 } 652 653 /** 654 * Perform promotion and renaming of exported internal functions. 655 * Index is updated to reflect linkage changes from weak resolution. 656 */ 657 void ThinLTOCodeGenerator::promote(Module &TheModule, ModuleSummaryIndex &Index, 658 const lto::InputFile &File) { 659 auto ModuleCount = Index.modulePaths().size(); 660 auto ModuleIdentifier = TheModule.getModuleIdentifier(); 661 662 // Collect for each module the list of function it defines (GUID -> Summary). 663 StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries; 664 Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries); 665 666 // Convert the preserved symbols set from string to GUID 667 auto GUIDPreservedSymbols = computeGUIDPreservedSymbols( 668 File, PreservedSymbols, Triple(TheModule.getTargetTriple())); 669 670 // Add used symbol to the preserved symbols. 671 addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols); 672 673 // Compute "dead" symbols, we don't want to import/export these! 674 computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols); 675 676 // Generate import/export list 677 StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount); 678 StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount); 679 ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists, 680 ExportLists); 681 682 DenseMap<GlobalValue::GUID, const GlobalValueSummary *> PrevailingCopy; 683 computePrevailingCopies(Index, PrevailingCopy); 684 685 // Resolve prevailing symbols 686 StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> ResolvedODR; 687 resolvePrevailingInIndex(Index, ResolvedODR, GUIDPreservedSymbols, 688 PrevailingCopy); 689 690 thinLTOResolvePrevailingInModule( 691 TheModule, ModuleToDefinedGVSummaries[ModuleIdentifier]); 692 693 // Promote the exported values in the index, so that they are promoted 694 // in the module. 695 thinLTOInternalizeAndPromoteInIndex( 696 Index, IsExported(ExportLists, GUIDPreservedSymbols), 697 IsPrevailing(PrevailingCopy)); 698 699 // FIXME Set ClearDSOLocalOnDeclarations. 700 promoteModule(TheModule, Index, /*ClearDSOLocalOnDeclarations=*/false); 701 } 702 703 /** 704 * Perform cross-module importing for the module identified by ModuleIdentifier. 705 */ 706 void ThinLTOCodeGenerator::crossModuleImport(Module &TheModule, 707 ModuleSummaryIndex &Index, 708 const lto::InputFile &File) { 709 auto ModuleMap = generateModuleMap(Modules); 710 auto ModuleCount = Index.modulePaths().size(); 711 712 // Collect for each module the list of function it defines (GUID -> Summary). 713 StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount); 714 Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries); 715 716 // Convert the preserved symbols set from string to GUID 717 auto GUIDPreservedSymbols = computeGUIDPreservedSymbols( 718 File, PreservedSymbols, Triple(TheModule.getTargetTriple())); 719 720 addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols); 721 722 // Compute "dead" symbols, we don't want to import/export these! 723 computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols); 724 725 // Generate import/export list 726 StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount); 727 StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount); 728 ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists, 729 ExportLists); 730 auto &ImportList = ImportLists[TheModule.getModuleIdentifier()]; 731 732 // FIXME Set ClearDSOLocalOnDeclarations. 733 crossImportIntoModule(TheModule, Index, ModuleMap, ImportList, 734 /*ClearDSOLocalOnDeclarations=*/false); 735 } 736 737 /** 738 * Compute the list of summaries needed for importing into module. 739 */ 740 void ThinLTOCodeGenerator::gatherImportedSummariesForModule( 741 Module &TheModule, ModuleSummaryIndex &Index, 742 std::map<std::string, GVSummaryMapTy> &ModuleToSummariesForIndex, 743 const lto::InputFile &File) { 744 auto ModuleCount = Index.modulePaths().size(); 745 auto ModuleIdentifier = TheModule.getModuleIdentifier(); 746 747 // Collect for each module the list of function it defines (GUID -> Summary). 748 StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount); 749 Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries); 750 751 // Convert the preserved symbols set from string to GUID 752 auto GUIDPreservedSymbols = computeGUIDPreservedSymbols( 753 File, PreservedSymbols, Triple(TheModule.getTargetTriple())); 754 755 addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols); 756 757 // Compute "dead" symbols, we don't want to import/export these! 758 computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols); 759 760 // Generate import/export list 761 StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount); 762 StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount); 763 ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists, 764 ExportLists); 765 766 llvm::gatherImportedSummariesForModule( 767 ModuleIdentifier, ModuleToDefinedGVSummaries, 768 ImportLists[ModuleIdentifier], ModuleToSummariesForIndex); 769 } 770 771 /** 772 * Emit the list of files needed for importing into module. 773 */ 774 void ThinLTOCodeGenerator::emitImports(Module &TheModule, StringRef OutputName, 775 ModuleSummaryIndex &Index, 776 const lto::InputFile &File) { 777 auto ModuleCount = Index.modulePaths().size(); 778 auto ModuleIdentifier = TheModule.getModuleIdentifier(); 779 780 // Collect for each module the list of function it defines (GUID -> Summary). 781 StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount); 782 Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries); 783 784 // Convert the preserved symbols set from string to GUID 785 auto GUIDPreservedSymbols = computeGUIDPreservedSymbols( 786 File, PreservedSymbols, Triple(TheModule.getTargetTriple())); 787 788 addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols); 789 790 // Compute "dead" symbols, we don't want to import/export these! 791 computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols); 792 793 // Generate import/export list 794 StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount); 795 StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount); 796 ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists, 797 ExportLists); 798 799 std::map<std::string, GVSummaryMapTy> ModuleToSummariesForIndex; 800 llvm::gatherImportedSummariesForModule( 801 ModuleIdentifier, ModuleToDefinedGVSummaries, 802 ImportLists[ModuleIdentifier], ModuleToSummariesForIndex); 803 804 std::error_code EC; 805 if ((EC = EmitImportsFiles(ModuleIdentifier, OutputName, 806 ModuleToSummariesForIndex))) 807 report_fatal_error(Twine("Failed to open ") + OutputName + 808 " to save imports lists\n"); 809 } 810 811 /** 812 * Perform internalization. Runs promote and internalization together. 813 * Index is updated to reflect linkage changes. 814 */ 815 void ThinLTOCodeGenerator::internalize(Module &TheModule, 816 ModuleSummaryIndex &Index, 817 const lto::InputFile &File) { 818 initTMBuilder(TMBuilder, Triple(TheModule.getTargetTriple())); 819 auto ModuleCount = Index.modulePaths().size(); 820 auto ModuleIdentifier = TheModule.getModuleIdentifier(); 821 822 // Convert the preserved symbols set from string to GUID 823 auto GUIDPreservedSymbols = 824 computeGUIDPreservedSymbols(File, PreservedSymbols, TMBuilder.TheTriple); 825 826 addUsedSymbolToPreservedGUID(File, GUIDPreservedSymbols); 827 828 // Collect for each module the list of function it defines (GUID -> Summary). 829 StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount); 830 Index.collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries); 831 832 // Compute "dead" symbols, we don't want to import/export these! 833 computeDeadSymbolsInIndex(Index, GUIDPreservedSymbols); 834 835 // Generate import/export list 836 StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount); 837 StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount); 838 ComputeCrossModuleImport(Index, ModuleToDefinedGVSummaries, ImportLists, 839 ExportLists); 840 auto &ExportList = ExportLists[ModuleIdentifier]; 841 842 // Be friendly and don't nuke totally the module when the client didn't 843 // supply anything to preserve. 844 if (ExportList.empty() && GUIDPreservedSymbols.empty()) 845 return; 846 847 DenseMap<GlobalValue::GUID, const GlobalValueSummary *> PrevailingCopy; 848 computePrevailingCopies(Index, PrevailingCopy); 849 850 // Resolve prevailing symbols 851 StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> ResolvedODR; 852 resolvePrevailingInIndex(Index, ResolvedODR, GUIDPreservedSymbols, 853 PrevailingCopy); 854 855 // Promote the exported values in the index, so that they are promoted 856 // in the module. 857 thinLTOInternalizeAndPromoteInIndex( 858 Index, IsExported(ExportLists, GUIDPreservedSymbols), 859 IsPrevailing(PrevailingCopy)); 860 861 // FIXME Set ClearDSOLocalOnDeclarations. 862 promoteModule(TheModule, Index, /*ClearDSOLocalOnDeclarations=*/false); 863 864 // Internalization 865 thinLTOResolvePrevailingInModule( 866 TheModule, ModuleToDefinedGVSummaries[ModuleIdentifier]); 867 868 thinLTOInternalizeModule(TheModule, 869 ModuleToDefinedGVSummaries[ModuleIdentifier]); 870 } 871 872 /** 873 * Perform post-importing ThinLTO optimizations. 874 */ 875 void ThinLTOCodeGenerator::optimize(Module &TheModule) { 876 initTMBuilder(TMBuilder, Triple(TheModule.getTargetTriple())); 877 878 // Optimize now 879 optimizeModule(TheModule, *TMBuilder.create(), OptLevel, Freestanding, 880 nullptr); 881 } 882 883 /// Write out the generated object file, either from CacheEntryPath or from 884 /// OutputBuffer, preferring hard-link when possible. 885 /// Returns the path to the generated file in SavedObjectsDirectoryPath. 886 std::string 887 ThinLTOCodeGenerator::writeGeneratedObject(int count, StringRef CacheEntryPath, 888 const MemoryBuffer &OutputBuffer) { 889 auto ArchName = TMBuilder.TheTriple.getArchName(); 890 SmallString<128> OutputPath(SavedObjectsDirectoryPath); 891 llvm::sys::path::append(OutputPath, 892 Twine(count) + "." + ArchName + ".thinlto.o"); 893 OutputPath.c_str(); // Ensure the string is null terminated. 894 if (sys::fs::exists(OutputPath)) 895 sys::fs::remove(OutputPath); 896 897 // We don't return a memory buffer to the linker, just a list of files. 898 if (!CacheEntryPath.empty()) { 899 // Cache is enabled, hard-link the entry (or copy if hard-link fails). 900 auto Err = sys::fs::create_hard_link(CacheEntryPath, OutputPath); 901 if (!Err) 902 return std::string(OutputPath.str()); 903 // Hard linking failed, try to copy. 904 Err = sys::fs::copy_file(CacheEntryPath, OutputPath); 905 if (!Err) 906 return std::string(OutputPath.str()); 907 // Copy failed (could be because the CacheEntry was removed from the cache 908 // in the meantime by another process), fall back and try to write down the 909 // buffer to the output. 910 errs() << "remark: can't link or copy from cached entry '" << CacheEntryPath 911 << "' to '" << OutputPath << "'\n"; 912 } 913 // No cache entry, just write out the buffer. 914 std::error_code Err; 915 raw_fd_ostream OS(OutputPath, Err, sys::fs::OF_None); 916 if (Err) 917 report_fatal_error("Can't open output '" + OutputPath + "'\n"); 918 OS << OutputBuffer.getBuffer(); 919 return std::string(OutputPath.str()); 920 } 921 922 // Main entry point for the ThinLTO processing 923 void ThinLTOCodeGenerator::run() { 924 // Prepare the resulting object vector 925 assert(ProducedBinaries.empty() && "The generator should not be reused"); 926 if (SavedObjectsDirectoryPath.empty()) 927 ProducedBinaries.resize(Modules.size()); 928 else { 929 sys::fs::create_directories(SavedObjectsDirectoryPath); 930 bool IsDir; 931 sys::fs::is_directory(SavedObjectsDirectoryPath, IsDir); 932 if (!IsDir) 933 report_fatal_error("Unexistent dir: '" + SavedObjectsDirectoryPath + "'"); 934 ProducedBinaryFiles.resize(Modules.size()); 935 } 936 937 if (CodeGenOnly) { 938 // Perform only parallel codegen and return. 939 ThreadPool Pool; 940 int count = 0; 941 for (auto &Mod : Modules) { 942 Pool.async([&](int count) { 943 LLVMContext Context; 944 Context.setDiscardValueNames(LTODiscardValueNames); 945 946 // Parse module now 947 auto TheModule = loadModuleFromInput(Mod.get(), Context, false, 948 /*IsImporting*/ false); 949 950 // CodeGen 951 auto OutputBuffer = codegenModule(*TheModule, *TMBuilder.create()); 952 if (SavedObjectsDirectoryPath.empty()) 953 ProducedBinaries[count] = std::move(OutputBuffer); 954 else 955 ProducedBinaryFiles[count] = 956 writeGeneratedObject(count, "", *OutputBuffer); 957 }, count++); 958 } 959 960 return; 961 } 962 963 // Sequential linking phase 964 auto Index = linkCombinedIndex(); 965 966 // Save temps: index. 967 if (!SaveTempsDir.empty()) { 968 auto SaveTempPath = SaveTempsDir + "index.bc"; 969 std::error_code EC; 970 raw_fd_ostream OS(SaveTempPath, EC, sys::fs::OF_None); 971 if (EC) 972 report_fatal_error(Twine("Failed to open ") + SaveTempPath + 973 " to save optimized bitcode\n"); 974 WriteIndexToFile(*Index, OS); 975 } 976 977 978 // Prepare the module map. 979 auto ModuleMap = generateModuleMap(Modules); 980 auto ModuleCount = Modules.size(); 981 982 // Collect for each module the list of function it defines (GUID -> Summary). 983 StringMap<GVSummaryMapTy> ModuleToDefinedGVSummaries(ModuleCount); 984 Index->collectDefinedGVSummariesPerModule(ModuleToDefinedGVSummaries); 985 986 // Convert the preserved symbols set from string to GUID, this is needed for 987 // computing the caching hash and the internalization. 988 DenseSet<GlobalValue::GUID> GUIDPreservedSymbols; 989 for (const auto &M : Modules) 990 computeGUIDPreservedSymbols(*M, PreservedSymbols, TMBuilder.TheTriple, 991 GUIDPreservedSymbols); 992 993 // Add used symbol from inputs to the preserved symbols. 994 for (const auto &M : Modules) 995 addUsedSymbolToPreservedGUID(*M, GUIDPreservedSymbols); 996 997 // Compute "dead" symbols, we don't want to import/export these! 998 computeDeadSymbolsInIndex(*Index, GUIDPreservedSymbols); 999 1000 // Synthesize entry counts for functions in the combined index. 1001 computeSyntheticCounts(*Index); 1002 1003 // Currently there is no support for enabling whole program visibility via a 1004 // linker option in the old LTO API, but this call allows it to be specified 1005 // via the internal option. Must be done before WPD below. 1006 updateVCallVisibilityInIndex(*Index, 1007 /* WholeProgramVisibilityEnabledInLTO */ false); 1008 1009 // Perform index-based WPD. This will return immediately if there are 1010 // no index entries in the typeIdMetadata map (e.g. if we are instead 1011 // performing IR-based WPD in hybrid regular/thin LTO mode). 1012 std::map<ValueInfo, std::vector<VTableSlotSummary>> LocalWPDTargetsMap; 1013 std::set<GlobalValue::GUID> ExportedGUIDs; 1014 runWholeProgramDevirtOnIndex(*Index, ExportedGUIDs, LocalWPDTargetsMap); 1015 for (auto GUID : ExportedGUIDs) 1016 GUIDPreservedSymbols.insert(GUID); 1017 1018 // Collect the import/export lists for all modules from the call-graph in the 1019 // combined index. 1020 StringMap<FunctionImporter::ImportMapTy> ImportLists(ModuleCount); 1021 StringMap<FunctionImporter::ExportSetTy> ExportLists(ModuleCount); 1022 ComputeCrossModuleImport(*Index, ModuleToDefinedGVSummaries, ImportLists, 1023 ExportLists); 1024 1025 // We use a std::map here to be able to have a defined ordering when 1026 // producing a hash for the cache entry. 1027 // FIXME: we should be able to compute the caching hash for the entry based 1028 // on the index, and nuke this map. 1029 StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> ResolvedODR; 1030 1031 DenseMap<GlobalValue::GUID, const GlobalValueSummary *> PrevailingCopy; 1032 computePrevailingCopies(*Index, PrevailingCopy); 1033 1034 // Resolve prevailing symbols, this has to be computed early because it 1035 // impacts the caching. 1036 resolvePrevailingInIndex(*Index, ResolvedODR, GUIDPreservedSymbols, 1037 PrevailingCopy); 1038 1039 // Use global summary-based analysis to identify symbols that can be 1040 // internalized (because they aren't exported or preserved as per callback). 1041 // Changes are made in the index, consumed in the ThinLTO backends. 1042 updateIndexWPDForExports(*Index, 1043 IsExported(ExportLists, GUIDPreservedSymbols), 1044 LocalWPDTargetsMap); 1045 thinLTOInternalizeAndPromoteInIndex( 1046 *Index, IsExported(ExportLists, GUIDPreservedSymbols), 1047 IsPrevailing(PrevailingCopy)); 1048 1049 // Make sure that every module has an entry in the ExportLists, ImportList, 1050 // GVSummary and ResolvedODR maps to enable threaded access to these maps 1051 // below. 1052 for (auto &Module : Modules) { 1053 auto ModuleIdentifier = Module->getName(); 1054 ExportLists[ModuleIdentifier]; 1055 ImportLists[ModuleIdentifier]; 1056 ResolvedODR[ModuleIdentifier]; 1057 ModuleToDefinedGVSummaries[ModuleIdentifier]; 1058 } 1059 1060 std::vector<BitcodeModule *> ModulesVec; 1061 ModulesVec.reserve(Modules.size()); 1062 for (auto &Mod : Modules) 1063 ModulesVec.push_back(&Mod->getSingleBitcodeModule()); 1064 std::vector<int> ModulesOrdering = lto::generateModulesOrdering(ModulesVec); 1065 1066 // Parallel optimizer + codegen 1067 { 1068 ThreadPool Pool(heavyweight_hardware_concurrency(ThreadCount)); 1069 for (auto IndexCount : ModulesOrdering) { 1070 auto &Mod = Modules[IndexCount]; 1071 Pool.async([&](int count) { 1072 auto ModuleIdentifier = Mod->getName(); 1073 auto &ExportList = ExportLists[ModuleIdentifier]; 1074 1075 auto &DefinedGVSummaries = ModuleToDefinedGVSummaries[ModuleIdentifier]; 1076 1077 // The module may be cached, this helps handling it. 1078 ModuleCacheEntry CacheEntry(CacheOptions.Path, *Index, ModuleIdentifier, 1079 ImportLists[ModuleIdentifier], ExportList, 1080 ResolvedODR[ModuleIdentifier], 1081 DefinedGVSummaries, OptLevel, Freestanding, 1082 TMBuilder); 1083 auto CacheEntryPath = CacheEntry.getEntryPath(); 1084 1085 { 1086 auto ErrOrBuffer = CacheEntry.tryLoadingBuffer(); 1087 LLVM_DEBUG(dbgs() << "Cache " << (ErrOrBuffer ? "hit" : "miss") 1088 << " '" << CacheEntryPath << "' for buffer " 1089 << count << " " << ModuleIdentifier << "\n"); 1090 1091 if (ErrOrBuffer) { 1092 // Cache Hit! 1093 if (SavedObjectsDirectoryPath.empty()) 1094 ProducedBinaries[count] = std::move(ErrOrBuffer.get()); 1095 else 1096 ProducedBinaryFiles[count] = writeGeneratedObject( 1097 count, CacheEntryPath, *ErrOrBuffer.get()); 1098 return; 1099 } 1100 } 1101 1102 LLVMContext Context; 1103 Context.setDiscardValueNames(LTODiscardValueNames); 1104 Context.enableDebugTypeODRUniquing(); 1105 auto DiagFileOrErr = lto::setupLLVMOptimizationRemarks( 1106 Context, RemarksFilename, RemarksPasses, RemarksFormat, 1107 RemarksWithHotness, count); 1108 if (!DiagFileOrErr) { 1109 errs() << "Error: " << toString(DiagFileOrErr.takeError()) << "\n"; 1110 report_fatal_error("ThinLTO: Can't get an output file for the " 1111 "remarks"); 1112 } 1113 1114 // Parse module now 1115 auto TheModule = loadModuleFromInput(Mod.get(), Context, false, 1116 /*IsImporting*/ false); 1117 1118 // Save temps: original file. 1119 saveTempBitcode(*TheModule, SaveTempsDir, count, ".0.original.bc"); 1120 1121 auto &ImportList = ImportLists[ModuleIdentifier]; 1122 // Run the main process now, and generates a binary 1123 auto OutputBuffer = ProcessThinLTOModule( 1124 *TheModule, *Index, ModuleMap, *TMBuilder.create(), ImportList, 1125 ExportList, GUIDPreservedSymbols, 1126 ModuleToDefinedGVSummaries[ModuleIdentifier], CacheOptions, 1127 DisableCodeGen, SaveTempsDir, Freestanding, OptLevel, count); 1128 1129 // Commit to the cache (if enabled) 1130 CacheEntry.write(*OutputBuffer); 1131 1132 if (SavedObjectsDirectoryPath.empty()) { 1133 // We need to generated a memory buffer for the linker. 1134 if (!CacheEntryPath.empty()) { 1135 // When cache is enabled, reload from the cache if possible. 1136 // Releasing the buffer from the heap and reloading it from the 1137 // cache file with mmap helps us to lower memory pressure. 1138 // The freed memory can be used for the next input file. 1139 // The final binary link will read from the VFS cache (hopefully!) 1140 // or from disk (if the memory pressure was too high). 1141 auto ReloadedBufferOrErr = CacheEntry.tryLoadingBuffer(); 1142 if (auto EC = ReloadedBufferOrErr.getError()) { 1143 // On error, keep the preexisting buffer and print a diagnostic. 1144 errs() << "remark: can't reload cached file '" << CacheEntryPath 1145 << "': " << EC.message() << "\n"; 1146 } else { 1147 OutputBuffer = std::move(*ReloadedBufferOrErr); 1148 } 1149 } 1150 ProducedBinaries[count] = std::move(OutputBuffer); 1151 return; 1152 } 1153 ProducedBinaryFiles[count] = writeGeneratedObject( 1154 count, CacheEntryPath, *OutputBuffer); 1155 }, IndexCount); 1156 } 1157 } 1158 1159 pruneCache(CacheOptions.Path, CacheOptions.Policy); 1160 1161 // If statistics were requested, print them out now. 1162 if (llvm::AreStatisticsEnabled()) 1163 llvm::PrintStatistics(); 1164 reportAndResetTimings(); 1165 } 1166