1 //===-LTO.cpp - LLVM Link Time Optimizer ----------------------------------===// 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 functions and classes used to support LTO. 11 // 12 //===----------------------------------------------------------------------===// 13 14 #include "llvm/LTO/LTO.h" 15 #include "llvm/Analysis/TargetLibraryInfo.h" 16 #include "llvm/Analysis/TargetTransformInfo.h" 17 #include "llvm/Bitcode/BitcodeReader.h" 18 #include "llvm/Bitcode/BitcodeWriter.h" 19 #include "llvm/CodeGen/Analysis.h" 20 #include "llvm/CodeGen/TargetLoweringObjectFileImpl.h" 21 #include "llvm/IR/AutoUpgrade.h" 22 #include "llvm/IR/DiagnosticPrinter.h" 23 #include "llvm/IR/LegacyPassManager.h" 24 #include "llvm/IR/Mangler.h" 25 #include "llvm/IR/Metadata.h" 26 #include "llvm/LTO/LTOBackend.h" 27 #include "llvm/Linker/IRMover.h" 28 #include "llvm/Object/ModuleSummaryIndexObjectFile.h" 29 #include "llvm/Support/Error.h" 30 #include "llvm/Support/ManagedStatic.h" 31 #include "llvm/Support/MemoryBuffer.h" 32 #include "llvm/Support/Path.h" 33 #include "llvm/Support/SHA1.h" 34 #include "llvm/Support/SourceMgr.h" 35 #include "llvm/Support/TargetRegistry.h" 36 #include "llvm/Support/ThreadPool.h" 37 #include "llvm/Support/Threading.h" 38 #include "llvm/Support/raw_ostream.h" 39 #include "llvm/Target/TargetMachine.h" 40 #include "llvm/Target/TargetOptions.h" 41 #include "llvm/Transforms/IPO.h" 42 #include "llvm/Transforms/IPO/PassManagerBuilder.h" 43 #include "llvm/Transforms/Utils/SplitModule.h" 44 45 #include <set> 46 47 using namespace llvm; 48 using namespace lto; 49 using namespace object; 50 51 #define DEBUG_TYPE "lto" 52 53 // Returns a unique hash for the Module considering the current list of 54 // export/import and other global analysis results. 55 // The hash is produced in \p Key. 56 static void computeCacheKey( 57 SmallString<40> &Key, const Config &Conf, const ModuleSummaryIndex &Index, 58 StringRef ModuleID, const FunctionImporter::ImportMapTy &ImportList, 59 const FunctionImporter::ExportSetTy &ExportList, 60 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 61 const GVSummaryMapTy &DefinedGlobals) { 62 // Compute the unique hash for this entry. 63 // This is based on the current compiler version, the module itself, the 64 // export list, the hash for every single module in the import list, the 65 // list of ResolvedODR for the module, and the list of preserved symbols. 66 SHA1 Hasher; 67 68 // Start with the compiler revision 69 Hasher.update(LLVM_VERSION_STRING); 70 #ifdef HAVE_LLVM_REVISION 71 Hasher.update(LLVM_REVISION); 72 #endif 73 74 // Include the parts of the LTO configuration that affect code generation. 75 auto AddString = [&](StringRef Str) { 76 Hasher.update(Str); 77 Hasher.update(ArrayRef<uint8_t>{0}); 78 }; 79 auto AddUnsigned = [&](unsigned I) { 80 uint8_t Data[4]; 81 Data[0] = I; 82 Data[1] = I >> 8; 83 Data[2] = I >> 16; 84 Data[3] = I >> 24; 85 Hasher.update(ArrayRef<uint8_t>{Data, 4}); 86 }; 87 AddString(Conf.CPU); 88 // FIXME: Hash more of Options. For now all clients initialize Options from 89 // command-line flags (which is unsupported in production), but may set 90 // RelaxELFRelocations. The clang driver can also pass FunctionSections, 91 // DataSections and DebuggerTuning via command line flags. 92 AddUnsigned(Conf.Options.RelaxELFRelocations); 93 AddUnsigned(Conf.Options.FunctionSections); 94 AddUnsigned(Conf.Options.DataSections); 95 AddUnsigned((unsigned)Conf.Options.DebuggerTuning); 96 for (auto &A : Conf.MAttrs) 97 AddString(A); 98 AddUnsigned(Conf.RelocModel); 99 AddUnsigned(Conf.CodeModel); 100 AddUnsigned(Conf.CGOptLevel); 101 AddUnsigned(Conf.CGFileType); 102 AddUnsigned(Conf.OptLevel); 103 AddString(Conf.OptPipeline); 104 AddString(Conf.AAPipeline); 105 AddString(Conf.OverrideTriple); 106 AddString(Conf.DefaultTriple); 107 108 // Include the hash for the current module 109 auto ModHash = Index.getModuleHash(ModuleID); 110 Hasher.update(ArrayRef<uint8_t>((uint8_t *)&ModHash[0], sizeof(ModHash))); 111 for (auto F : ExportList) 112 // The export list can impact the internalization, be conservative here 113 Hasher.update(ArrayRef<uint8_t>((uint8_t *)&F, sizeof(F))); 114 115 // Include the hash for every module we import functions from 116 for (auto &Entry : ImportList) { 117 auto ModHash = Index.getModuleHash(Entry.first()); 118 Hasher.update(ArrayRef<uint8_t>((uint8_t *)&ModHash[0], sizeof(ModHash))); 119 } 120 121 // Include the hash for the resolved ODR. 122 for (auto &Entry : ResolvedODR) { 123 Hasher.update(ArrayRef<uint8_t>((const uint8_t *)&Entry.first, 124 sizeof(GlobalValue::GUID))); 125 Hasher.update(ArrayRef<uint8_t>((const uint8_t *)&Entry.second, 126 sizeof(GlobalValue::LinkageTypes))); 127 } 128 129 // Include the hash for the linkage type to reflect internalization and weak 130 // resolution. 131 for (auto &GS : DefinedGlobals) { 132 GlobalValue::LinkageTypes Linkage = GS.second->linkage(); 133 Hasher.update( 134 ArrayRef<uint8_t>((const uint8_t *)&Linkage, sizeof(Linkage))); 135 } 136 137 if (!Conf.SampleProfile.empty()) { 138 auto FileOrErr = MemoryBuffer::getFile(Conf.SampleProfile); 139 if (FileOrErr) 140 Hasher.update(FileOrErr.get()->getBuffer()); 141 } 142 143 Key = toHex(Hasher.result()); 144 } 145 146 static void thinLTOResolveWeakForLinkerGUID( 147 GlobalValueSummaryList &GVSummaryList, GlobalValue::GUID GUID, 148 DenseSet<GlobalValueSummary *> &GlobalInvolvedWithAlias, 149 function_ref<bool(GlobalValue::GUID, const GlobalValueSummary *)> 150 isPrevailing, 151 function_ref<void(StringRef, GlobalValue::GUID, GlobalValue::LinkageTypes)> 152 recordNewLinkage) { 153 for (auto &S : GVSummaryList) { 154 GlobalValue::LinkageTypes OriginalLinkage = S->linkage(); 155 if (!GlobalValue::isWeakForLinker(OriginalLinkage)) 156 continue; 157 // We need to emit only one of these. The prevailing module will keep it, 158 // but turned into a weak, while the others will drop it when possible. 159 // This is both a compile-time optimization and a correctness 160 // transformation. This is necessary for correctness when we have exported 161 // a reference - we need to convert the linkonce to weak to 162 // ensure a copy is kept to satisfy the exported reference. 163 // FIXME: We may want to split the compile time and correctness 164 // aspects into separate routines. 165 if (isPrevailing(GUID, S.get())) { 166 if (GlobalValue::isLinkOnceLinkage(OriginalLinkage)) 167 S->setLinkage(GlobalValue::getWeakLinkage( 168 GlobalValue::isLinkOnceODRLinkage(OriginalLinkage))); 169 } 170 // Alias and aliasee can't be turned into available_externally. 171 else if (!isa<AliasSummary>(S.get()) && 172 !GlobalInvolvedWithAlias.count(S.get())) 173 S->setLinkage(GlobalValue::AvailableExternallyLinkage); 174 if (S->linkage() != OriginalLinkage) 175 recordNewLinkage(S->modulePath(), GUID, S->linkage()); 176 } 177 } 178 179 // Resolve Weak and LinkOnce values in the \p Index. 180 // 181 // We'd like to drop these functions if they are no longer referenced in the 182 // current module. However there is a chance that another module is still 183 // referencing them because of the import. We make sure we always emit at least 184 // one copy. 185 void llvm::thinLTOResolveWeakForLinkerInIndex( 186 ModuleSummaryIndex &Index, 187 function_ref<bool(GlobalValue::GUID, const GlobalValueSummary *)> 188 isPrevailing, 189 function_ref<void(StringRef, GlobalValue::GUID, GlobalValue::LinkageTypes)> 190 recordNewLinkage) { 191 // We won't optimize the globals that are referenced by an alias for now 192 // Ideally we should turn the alias into a global and duplicate the definition 193 // when needed. 194 DenseSet<GlobalValueSummary *> GlobalInvolvedWithAlias; 195 for (auto &I : Index) 196 for (auto &S : I.second) 197 if (auto AS = dyn_cast<AliasSummary>(S.get())) 198 GlobalInvolvedWithAlias.insert(&AS->getAliasee()); 199 200 for (auto &I : Index) 201 thinLTOResolveWeakForLinkerGUID(I.second, I.first, GlobalInvolvedWithAlias, 202 isPrevailing, recordNewLinkage); 203 } 204 205 static void thinLTOInternalizeAndPromoteGUID( 206 GlobalValueSummaryList &GVSummaryList, GlobalValue::GUID GUID, 207 function_ref<bool(StringRef, GlobalValue::GUID)> isExported) { 208 for (auto &S : GVSummaryList) { 209 if (isExported(S->modulePath(), GUID)) { 210 if (GlobalValue::isLocalLinkage(S->linkage())) 211 S->setLinkage(GlobalValue::ExternalLinkage); 212 } else if (!GlobalValue::isLocalLinkage(S->linkage())) 213 S->setLinkage(GlobalValue::InternalLinkage); 214 } 215 } 216 217 // Update the linkages in the given \p Index to mark exported values 218 // as external and non-exported values as internal. 219 void llvm::thinLTOInternalizeAndPromoteInIndex( 220 ModuleSummaryIndex &Index, 221 function_ref<bool(StringRef, GlobalValue::GUID)> isExported) { 222 for (auto &I : Index) 223 thinLTOInternalizeAndPromoteGUID(I.second, I.first, isExported); 224 } 225 226 struct InputFile::InputModule { 227 BitcodeModule BM; 228 std::unique_ptr<Module> Mod; 229 230 // The range of ModuleSymbolTable entries for this input module. 231 size_t SymBegin, SymEnd; 232 }; 233 234 // Requires a destructor for std::vector<InputModule>. 235 InputFile::~InputFile() = default; 236 237 Expected<std::unique_ptr<InputFile>> InputFile::create(MemoryBufferRef Object) { 238 std::unique_ptr<InputFile> File(new InputFile); 239 240 ErrorOr<MemoryBufferRef> BCOrErr = 241 IRObjectFile::findBitcodeInMemBuffer(Object); 242 if (!BCOrErr) 243 return errorCodeToError(BCOrErr.getError()); 244 245 Expected<std::vector<BitcodeModule>> BMsOrErr = 246 getBitcodeModuleList(*BCOrErr); 247 if (!BMsOrErr) 248 return BMsOrErr.takeError(); 249 250 if (BMsOrErr->empty()) 251 return make_error<StringError>("Bitcode file does not contain any modules", 252 inconvertibleErrorCode()); 253 254 // Create an InputModule for each module in the InputFile, and add it to the 255 // ModuleSymbolTable. 256 for (auto BM : *BMsOrErr) { 257 Expected<std::unique_ptr<Module>> MOrErr = 258 BM.getLazyModule(File->Ctx, /*ShouldLazyLoadMetadata*/ true, 259 /*IsImporting*/ false); 260 if (!MOrErr) 261 return MOrErr.takeError(); 262 263 size_t SymBegin = File->SymTab.symbols().size(); 264 File->SymTab.addModule(MOrErr->get()); 265 size_t SymEnd = File->SymTab.symbols().size(); 266 267 for (const auto &C : (*MOrErr)->getComdatSymbolTable()) { 268 auto P = File->ComdatMap.insert( 269 std::make_pair(&C.second, File->Comdats.size())); 270 assert(P.second); 271 (void)P; 272 File->Comdats.push_back(C.first()); 273 } 274 275 File->Mods.push_back({BM, std::move(*MOrErr), SymBegin, SymEnd}); 276 } 277 278 return std::move(File); 279 } 280 281 Expected<int> InputFile::Symbol::getComdatIndex() const { 282 if (!isGV()) 283 return -1; 284 const GlobalObject *GO = getGV()->getBaseObject(); 285 if (!GO) 286 return make_error<StringError>("Unable to determine comdat of alias!", 287 inconvertibleErrorCode()); 288 if (const Comdat *C = GO->getComdat()) { 289 auto I = File->ComdatMap.find(C); 290 assert(I != File->ComdatMap.end()); 291 return I->second; 292 } 293 return -1; 294 } 295 296 Expected<std::string> InputFile::getLinkerOpts() { 297 std::string LinkerOpts; 298 raw_string_ostream LOS(LinkerOpts); 299 // Extract linker options from module metadata. 300 for (InputModule &Mod : Mods) { 301 std::unique_ptr<Module> &M = Mod.Mod; 302 if (auto E = M->materializeMetadata()) 303 return std::move(E); 304 if (Metadata *Val = M->getModuleFlag("Linker Options")) { 305 MDNode *LinkerOptions = cast<MDNode>(Val); 306 for (const MDOperand &MDOptions : LinkerOptions->operands()) 307 for (const MDOperand &MDOption : cast<MDNode>(MDOptions)->operands()) 308 LOS << " " << cast<MDString>(MDOption)->getString(); 309 } 310 } 311 312 // Synthesize export flags for symbols with dllexport storage. 313 const Triple TT(Mods[0].Mod->getTargetTriple()); 314 Mangler M; 315 for (const ModuleSymbolTable::Symbol &Sym : SymTab.symbols()) 316 if (auto *GV = Sym.dyn_cast<GlobalValue*>()) 317 emitLinkerFlagsForGlobalCOFF(LOS, GV, TT, M); 318 LOS.flush(); 319 return LinkerOpts; 320 } 321 322 StringRef InputFile::getName() const { 323 return Mods[0].BM.getModuleIdentifier(); 324 } 325 326 StringRef InputFile::getSourceFileName() const { 327 return Mods[0].Mod->getSourceFileName(); 328 } 329 330 iterator_range<InputFile::symbol_iterator> 331 InputFile::module_symbols(InputModule &IM) { 332 return llvm::make_range( 333 symbol_iterator(SymTab.symbols().data() + IM.SymBegin, SymTab, this), 334 symbol_iterator(SymTab.symbols().data() + IM.SymEnd, SymTab, this)); 335 } 336 337 LTO::RegularLTOState::RegularLTOState(unsigned ParallelCodeGenParallelismLevel, 338 Config &Conf) 339 : ParallelCodeGenParallelismLevel(ParallelCodeGenParallelismLevel), 340 Ctx(Conf) {} 341 342 LTO::ThinLTOState::ThinLTOState(ThinBackend Backend) : Backend(Backend) { 343 if (!Backend) 344 this->Backend = 345 createInProcessThinBackend(llvm::heavyweight_hardware_concurrency()); 346 } 347 348 LTO::LTO(Config Conf, ThinBackend Backend, 349 unsigned ParallelCodeGenParallelismLevel) 350 : Conf(std::move(Conf)), 351 RegularLTO(ParallelCodeGenParallelismLevel, this->Conf), 352 ThinLTO(std::move(Backend)) {} 353 354 // Requires a destructor for MapVector<BitcodeModule>. 355 LTO::~LTO() = default; 356 357 // Add the given symbol to the GlobalResolutions map, and resolve its partition. 358 void LTO::addSymbolToGlobalRes(SmallPtrSet<GlobalValue *, 8> &Used, 359 const InputFile::Symbol &Sym, 360 SymbolResolution Res, unsigned Partition) { 361 GlobalValue *GV = Sym.isGV() ? Sym.getGV() : nullptr; 362 363 auto &GlobalRes = GlobalResolutions[Sym.getName()]; 364 if (GV) { 365 GlobalRes.UnnamedAddr &= GV->hasGlobalUnnamedAddr(); 366 if (Res.Prevailing) 367 GlobalRes.IRName = GV->getName(); 368 } 369 // Set the partition to external if we know it is used elsewhere, e.g. 370 // it is visible to a regular object, is referenced from llvm.compiler_used, 371 // or was already recorded as being referenced from a different partition. 372 if (Res.VisibleToRegularObj || (GV && Used.count(GV)) || 373 (GlobalRes.Partition != GlobalResolution::Unknown && 374 GlobalRes.Partition != Partition)) { 375 GlobalRes.Partition = GlobalResolution::External; 376 } else 377 // First recorded reference, save the current partition. 378 GlobalRes.Partition = Partition; 379 380 // Flag as visible outside of ThinLTO if visible from a regular object or 381 // if this is a reference in the regular LTO partition. 382 GlobalRes.VisibleOutsideThinLTO |= 383 (Res.VisibleToRegularObj || (Partition == GlobalResolution::RegularLTO)); 384 } 385 386 static void writeToResolutionFile(raw_ostream &OS, InputFile *Input, 387 ArrayRef<SymbolResolution> Res) { 388 StringRef Path = Input->getName(); 389 OS << Path << '\n'; 390 auto ResI = Res.begin(); 391 for (const InputFile::Symbol &Sym : Input->symbols()) { 392 assert(ResI != Res.end()); 393 SymbolResolution Res = *ResI++; 394 395 OS << "-r=" << Path << ',' << Sym.getName() << ','; 396 if (Res.Prevailing) 397 OS << 'p'; 398 if (Res.FinalDefinitionInLinkageUnit) 399 OS << 'l'; 400 if (Res.VisibleToRegularObj) 401 OS << 'x'; 402 OS << '\n'; 403 } 404 OS.flush(); 405 assert(ResI == Res.end()); 406 } 407 408 Error LTO::add(std::unique_ptr<InputFile> Input, 409 ArrayRef<SymbolResolution> Res) { 410 assert(!CalledGetMaxTasks); 411 412 if (Conf.ResolutionFile) 413 writeToResolutionFile(*Conf.ResolutionFile, Input.get(), Res); 414 415 const SymbolResolution *ResI = Res.begin(); 416 for (InputFile::InputModule &IM : Input->Mods) 417 if (Error Err = addModule(*Input, IM, ResI, Res.end())) 418 return Err; 419 420 assert(ResI == Res.end()); 421 return Error::success(); 422 } 423 424 Error LTO::addModule(InputFile &Input, InputFile::InputModule &IM, 425 const SymbolResolution *&ResI, 426 const SymbolResolution *ResE) { 427 // FIXME: move to backend 428 Module &M = *IM.Mod; 429 430 if (M.getDataLayoutStr().empty()) 431 return make_error<StringError>("input module has no datalayout", 432 inconvertibleErrorCode()); 433 434 if (!Conf.OverrideTriple.empty()) 435 M.setTargetTriple(Conf.OverrideTriple); 436 else if (M.getTargetTriple().empty()) 437 M.setTargetTriple(Conf.DefaultTriple); 438 439 Expected<bool> HasThinLTOSummary = IM.BM.hasSummary(); 440 if (!HasThinLTOSummary) 441 return HasThinLTOSummary.takeError(); 442 443 if (*HasThinLTOSummary) 444 return addThinLTO(IM.BM, M, Input.module_symbols(IM), ResI, ResE); 445 else 446 return addRegularLTO(IM.BM, ResI, ResE); 447 } 448 449 // Add a regular LTO object to the link. 450 Error LTO::addRegularLTO(BitcodeModule BM, const SymbolResolution *&ResI, 451 const SymbolResolution *ResE) { 452 if (!RegularLTO.CombinedModule) { 453 RegularLTO.CombinedModule = 454 llvm::make_unique<Module>("ld-temp.o", RegularLTO.Ctx); 455 RegularLTO.Mover = llvm::make_unique<IRMover>(*RegularLTO.CombinedModule); 456 } 457 Expected<std::unique_ptr<Module>> MOrErr = 458 BM.getLazyModule(RegularLTO.Ctx, /*ShouldLazyLoadMetadata*/ true, 459 /*IsImporting*/ false); 460 if (!MOrErr) 461 return MOrErr.takeError(); 462 463 Module &M = **MOrErr; 464 if (Error Err = M.materializeMetadata()) 465 return Err; 466 UpgradeDebugInfo(M); 467 468 ModuleSymbolTable SymTab; 469 SymTab.addModule(&M); 470 471 SmallPtrSet<GlobalValue *, 8> Used; 472 collectUsedGlobalVariables(M, Used, /*CompilerUsed*/ false); 473 474 std::vector<GlobalValue *> Keep; 475 476 for (GlobalVariable &GV : M.globals()) 477 if (GV.hasAppendingLinkage()) 478 Keep.push_back(&GV); 479 480 DenseSet<GlobalObject *> AliasedGlobals; 481 for (auto &GA : M.aliases()) 482 if (GlobalObject *GO = GA.getBaseObject()) 483 AliasedGlobals.insert(GO); 484 485 for (const InputFile::Symbol &Sym : 486 make_range(InputFile::symbol_iterator(SymTab.symbols().begin(), SymTab, 487 nullptr), 488 InputFile::symbol_iterator(SymTab.symbols().end(), SymTab, 489 nullptr))) { 490 assert(ResI != ResE); 491 SymbolResolution Res = *ResI++; 492 addSymbolToGlobalRes(Used, Sym, Res, 0); 493 494 if (Sym.isGV()) { 495 GlobalValue *GV = Sym.getGV(); 496 if (Res.Prevailing) { 497 if (Sym.getFlags() & object::BasicSymbolRef::SF_Undefined) 498 continue; 499 Keep.push_back(GV); 500 switch (GV->getLinkage()) { 501 default: 502 break; 503 case GlobalValue::LinkOnceAnyLinkage: 504 GV->setLinkage(GlobalValue::WeakAnyLinkage); 505 break; 506 case GlobalValue::LinkOnceODRLinkage: 507 GV->setLinkage(GlobalValue::WeakODRLinkage); 508 break; 509 } 510 } else if (isa<GlobalObject>(GV) && 511 (GV->hasLinkOnceODRLinkage() || GV->hasWeakODRLinkage() || 512 GV->hasAvailableExternallyLinkage()) && 513 !AliasedGlobals.count(cast<GlobalObject>(GV))) { 514 // Either of the above three types of linkage indicates that the 515 // chosen prevailing symbol will have the same semantics as this copy of 516 // the symbol, so we can link it with available_externally linkage. We 517 // only need to do this if the symbol is undefined. 518 GlobalValue *CombinedGV = 519 RegularLTO.CombinedModule->getNamedValue(GV->getName()); 520 if (!CombinedGV || CombinedGV->isDeclaration()) { 521 Keep.push_back(GV); 522 GV->setLinkage(GlobalValue::AvailableExternallyLinkage); 523 cast<GlobalObject>(GV)->setComdat(nullptr); 524 } 525 } 526 } 527 // Common resolution: collect the maximum size/alignment over all commons. 528 // We also record if we see an instance of a common as prevailing, so that 529 // if none is prevailing we can ignore it later. 530 if (Sym.getFlags() & object::BasicSymbolRef::SF_Common) { 531 // FIXME: We should figure out what to do about commons defined by asm. 532 // For now they aren't reported correctly by ModuleSymbolTable. 533 auto &CommonRes = RegularLTO.Commons[Sym.getGV()->getName()]; 534 CommonRes.Size = std::max(CommonRes.Size, Sym.getCommonSize()); 535 CommonRes.Align = std::max(CommonRes.Align, Sym.getCommonAlignment()); 536 CommonRes.Prevailing |= Res.Prevailing; 537 } 538 539 // FIXME: use proposed local attribute for FinalDefinitionInLinkageUnit. 540 } 541 542 return RegularLTO.Mover->move(std::move(*MOrErr), Keep, 543 [](GlobalValue &, IRMover::ValueAdder) {}, 544 /* IsPerformingImport */ false); 545 } 546 547 // Add a ThinLTO object to the link. 548 // FIXME: This function should not need to take as many parameters once we have 549 // a bitcode symbol table. 550 Error LTO::addThinLTO(BitcodeModule BM, Module &M, 551 iterator_range<InputFile::symbol_iterator> Syms, 552 const SymbolResolution *&ResI, 553 const SymbolResolution *ResE) { 554 SmallPtrSet<GlobalValue *, 8> Used; 555 collectUsedGlobalVariables(M, Used, /*CompilerUsed*/ false); 556 557 Expected<std::unique_ptr<ModuleSummaryIndex>> SummaryOrErr = BM.getSummary(); 558 if (!SummaryOrErr) 559 return SummaryOrErr.takeError(); 560 ThinLTO.CombinedIndex.mergeFrom(std::move(*SummaryOrErr), 561 ThinLTO.ModuleMap.size()); 562 563 for (const InputFile::Symbol &Sym : Syms) { 564 assert(ResI != ResE); 565 SymbolResolution Res = *ResI++; 566 addSymbolToGlobalRes(Used, Sym, Res, ThinLTO.ModuleMap.size() + 1); 567 568 if (Res.Prevailing && Sym.isGV()) 569 ThinLTO.PrevailingModuleForGUID[Sym.getGV()->getGUID()] = 570 BM.getModuleIdentifier(); 571 } 572 573 if (!ThinLTO.ModuleMap.insert({BM.getModuleIdentifier(), BM}).second) 574 return make_error<StringError>( 575 "Expected at most one ThinLTO module per bitcode file", 576 inconvertibleErrorCode()); 577 578 return Error::success(); 579 } 580 581 unsigned LTO::getMaxTasks() const { 582 CalledGetMaxTasks = true; 583 return RegularLTO.ParallelCodeGenParallelismLevel + ThinLTO.ModuleMap.size(); 584 } 585 586 Error LTO::run(AddStreamFn AddStream, NativeObjectCache Cache) { 587 // Save the status of having a regularLTO combined module, as 588 // this is needed for generating the ThinLTO Task ID, and 589 // the CombinedModule will be moved at the end of runRegularLTO. 590 bool HasRegularLTO = RegularLTO.CombinedModule != nullptr; 591 // Invoke regular LTO if there was a regular LTO module to start with. 592 if (HasRegularLTO) 593 if (auto E = runRegularLTO(AddStream)) 594 return E; 595 return runThinLTO(AddStream, Cache, HasRegularLTO); 596 } 597 598 Error LTO::runRegularLTO(AddStreamFn AddStream) { 599 // Make sure commons have the right size/alignment: we kept the largest from 600 // all the prevailing when adding the inputs, and we apply it here. 601 const DataLayout &DL = RegularLTO.CombinedModule->getDataLayout(); 602 for (auto &I : RegularLTO.Commons) { 603 if (!I.second.Prevailing) 604 // Don't do anything if no instance of this common was prevailing. 605 continue; 606 GlobalVariable *OldGV = RegularLTO.CombinedModule->getNamedGlobal(I.first); 607 if (OldGV && DL.getTypeAllocSize(OldGV->getValueType()) == I.second.Size) { 608 // Don't create a new global if the type is already correct, just make 609 // sure the alignment is correct. 610 OldGV->setAlignment(I.second.Align); 611 continue; 612 } 613 ArrayType *Ty = 614 ArrayType::get(Type::getInt8Ty(RegularLTO.Ctx), I.second.Size); 615 auto *GV = new GlobalVariable(*RegularLTO.CombinedModule, Ty, false, 616 GlobalValue::CommonLinkage, 617 ConstantAggregateZero::get(Ty), ""); 618 GV->setAlignment(I.second.Align); 619 if (OldGV) { 620 OldGV->replaceAllUsesWith(ConstantExpr::getBitCast(GV, OldGV->getType())); 621 GV->takeName(OldGV); 622 OldGV->eraseFromParent(); 623 } else { 624 GV->setName(I.first); 625 } 626 } 627 628 if (Conf.PreOptModuleHook && 629 !Conf.PreOptModuleHook(0, *RegularLTO.CombinedModule)) 630 return Error::success(); 631 632 if (!Conf.CodeGenOnly) { 633 for (const auto &R : GlobalResolutions) { 634 if (R.second.IRName.empty()) 635 continue; 636 if (R.second.Partition != 0 && 637 R.second.Partition != GlobalResolution::External) 638 continue; 639 640 GlobalValue *GV = 641 RegularLTO.CombinedModule->getNamedValue(R.second.IRName); 642 // Ignore symbols defined in other partitions. 643 if (!GV || GV->hasLocalLinkage()) 644 continue; 645 GV->setUnnamedAddr(R.second.UnnamedAddr ? GlobalValue::UnnamedAddr::Global 646 : GlobalValue::UnnamedAddr::None); 647 if (R.second.Partition == 0) 648 GV->setLinkage(GlobalValue::InternalLinkage); 649 } 650 651 if (Conf.PostInternalizeModuleHook && 652 !Conf.PostInternalizeModuleHook(0, *RegularLTO.CombinedModule)) 653 return Error::success(); 654 } 655 return backend(Conf, AddStream, RegularLTO.ParallelCodeGenParallelismLevel, 656 std::move(RegularLTO.CombinedModule), ThinLTO.CombinedIndex); 657 } 658 659 /// This class defines the interface to the ThinLTO backend. 660 class lto::ThinBackendProc { 661 protected: 662 Config &Conf; 663 ModuleSummaryIndex &CombinedIndex; 664 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries; 665 666 public: 667 ThinBackendProc(Config &Conf, ModuleSummaryIndex &CombinedIndex, 668 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries) 669 : Conf(Conf), CombinedIndex(CombinedIndex), 670 ModuleToDefinedGVSummaries(ModuleToDefinedGVSummaries) {} 671 672 virtual ~ThinBackendProc() {} 673 virtual Error start( 674 unsigned Task, BitcodeModule BM, 675 const FunctionImporter::ImportMapTy &ImportList, 676 const FunctionImporter::ExportSetTy &ExportList, 677 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 678 MapVector<StringRef, BitcodeModule> &ModuleMap) = 0; 679 virtual Error wait() = 0; 680 }; 681 682 namespace { 683 class InProcessThinBackend : public ThinBackendProc { 684 ThreadPool BackendThreadPool; 685 AddStreamFn AddStream; 686 NativeObjectCache Cache; 687 688 Optional<Error> Err; 689 std::mutex ErrMu; 690 691 public: 692 InProcessThinBackend( 693 Config &Conf, ModuleSummaryIndex &CombinedIndex, 694 unsigned ThinLTOParallelismLevel, 695 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries, 696 AddStreamFn AddStream, NativeObjectCache Cache) 697 : ThinBackendProc(Conf, CombinedIndex, ModuleToDefinedGVSummaries), 698 BackendThreadPool(ThinLTOParallelismLevel), 699 AddStream(std::move(AddStream)), Cache(std::move(Cache)) {} 700 701 Error runThinLTOBackendThread( 702 AddStreamFn AddStream, NativeObjectCache Cache, unsigned Task, 703 BitcodeModule BM, ModuleSummaryIndex &CombinedIndex, 704 const FunctionImporter::ImportMapTy &ImportList, 705 const FunctionImporter::ExportSetTy &ExportList, 706 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 707 const GVSummaryMapTy &DefinedGlobals, 708 MapVector<StringRef, BitcodeModule> &ModuleMap) { 709 auto RunThinBackend = [&](AddStreamFn AddStream) { 710 LTOLLVMContext BackendContext(Conf); 711 Expected<std::unique_ptr<Module>> MOrErr = BM.parseModule(BackendContext); 712 if (!MOrErr) 713 return MOrErr.takeError(); 714 715 return thinBackend(Conf, Task, AddStream, **MOrErr, CombinedIndex, 716 ImportList, DefinedGlobals, ModuleMap); 717 }; 718 719 auto ModuleID = BM.getModuleIdentifier(); 720 721 if (!Cache || !CombinedIndex.modulePaths().count(ModuleID) || 722 all_of(CombinedIndex.getModuleHash(ModuleID), 723 [](uint32_t V) { return V == 0; })) 724 // Cache disabled or no entry for this module in the combined index or 725 // no module hash. 726 return RunThinBackend(AddStream); 727 728 SmallString<40> Key; 729 // The module may be cached, this helps handling it. 730 computeCacheKey(Key, Conf, CombinedIndex, ModuleID, ImportList, ExportList, 731 ResolvedODR, DefinedGlobals); 732 if (AddStreamFn CacheAddStream = Cache(Task, Key)) 733 return RunThinBackend(CacheAddStream); 734 735 return Error::success(); 736 } 737 738 Error start( 739 unsigned Task, BitcodeModule BM, 740 const FunctionImporter::ImportMapTy &ImportList, 741 const FunctionImporter::ExportSetTy &ExportList, 742 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 743 MapVector<StringRef, BitcodeModule> &ModuleMap) override { 744 StringRef ModulePath = BM.getModuleIdentifier(); 745 assert(ModuleToDefinedGVSummaries.count(ModulePath)); 746 const GVSummaryMapTy &DefinedGlobals = 747 ModuleToDefinedGVSummaries.find(ModulePath)->second; 748 BackendThreadPool.async( 749 [=](BitcodeModule BM, ModuleSummaryIndex &CombinedIndex, 750 const FunctionImporter::ImportMapTy &ImportList, 751 const FunctionImporter::ExportSetTy &ExportList, 752 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> 753 &ResolvedODR, 754 const GVSummaryMapTy &DefinedGlobals, 755 MapVector<StringRef, BitcodeModule> &ModuleMap) { 756 Error E = runThinLTOBackendThread( 757 AddStream, Cache, Task, BM, CombinedIndex, ImportList, 758 ExportList, ResolvedODR, DefinedGlobals, ModuleMap); 759 if (E) { 760 std::unique_lock<std::mutex> L(ErrMu); 761 if (Err) 762 Err = joinErrors(std::move(*Err), std::move(E)); 763 else 764 Err = std::move(E); 765 } 766 }, 767 BM, std::ref(CombinedIndex), std::ref(ImportList), 768 std::ref(ExportList), std::ref(ResolvedODR), std::ref(DefinedGlobals), 769 std::ref(ModuleMap)); 770 return Error::success(); 771 } 772 773 Error wait() override { 774 BackendThreadPool.wait(); 775 if (Err) 776 return std::move(*Err); 777 else 778 return Error::success(); 779 } 780 }; 781 } // end anonymous namespace 782 783 ThinBackend lto::createInProcessThinBackend(unsigned ParallelismLevel) { 784 return [=](Config &Conf, ModuleSummaryIndex &CombinedIndex, 785 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries, 786 AddStreamFn AddStream, NativeObjectCache Cache) { 787 return llvm::make_unique<InProcessThinBackend>( 788 Conf, CombinedIndex, ParallelismLevel, ModuleToDefinedGVSummaries, 789 AddStream, Cache); 790 }; 791 } 792 793 // Given the original \p Path to an output file, replace any path 794 // prefix matching \p OldPrefix with \p NewPrefix. Also, create the 795 // resulting directory if it does not yet exist. 796 std::string lto::getThinLTOOutputFile(const std::string &Path, 797 const std::string &OldPrefix, 798 const std::string &NewPrefix) { 799 if (OldPrefix.empty() && NewPrefix.empty()) 800 return Path; 801 SmallString<128> NewPath(Path); 802 llvm::sys::path::replace_path_prefix(NewPath, OldPrefix, NewPrefix); 803 StringRef ParentPath = llvm::sys::path::parent_path(NewPath.str()); 804 if (!ParentPath.empty()) { 805 // Make sure the new directory exists, creating it if necessary. 806 if (std::error_code EC = llvm::sys::fs::create_directories(ParentPath)) 807 llvm::errs() << "warning: could not create directory '" << ParentPath 808 << "': " << EC.message() << '\n'; 809 } 810 return NewPath.str(); 811 } 812 813 namespace { 814 class WriteIndexesThinBackend : public ThinBackendProc { 815 std::string OldPrefix, NewPrefix; 816 bool ShouldEmitImportsFiles; 817 818 std::string LinkedObjectsFileName; 819 std::unique_ptr<llvm::raw_fd_ostream> LinkedObjectsFile; 820 821 public: 822 WriteIndexesThinBackend( 823 Config &Conf, ModuleSummaryIndex &CombinedIndex, 824 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries, 825 std::string OldPrefix, std::string NewPrefix, bool ShouldEmitImportsFiles, 826 std::string LinkedObjectsFileName) 827 : ThinBackendProc(Conf, CombinedIndex, ModuleToDefinedGVSummaries), 828 OldPrefix(OldPrefix), NewPrefix(NewPrefix), 829 ShouldEmitImportsFiles(ShouldEmitImportsFiles), 830 LinkedObjectsFileName(LinkedObjectsFileName) {} 831 832 Error start( 833 unsigned Task, BitcodeModule BM, 834 const FunctionImporter::ImportMapTy &ImportList, 835 const FunctionImporter::ExportSetTy &ExportList, 836 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 837 MapVector<StringRef, BitcodeModule> &ModuleMap) override { 838 StringRef ModulePath = BM.getModuleIdentifier(); 839 std::string NewModulePath = 840 getThinLTOOutputFile(ModulePath, OldPrefix, NewPrefix); 841 842 std::error_code EC; 843 if (!LinkedObjectsFileName.empty()) { 844 if (!LinkedObjectsFile) { 845 LinkedObjectsFile = llvm::make_unique<raw_fd_ostream>( 846 LinkedObjectsFileName, EC, sys::fs::OpenFlags::F_None); 847 if (EC) 848 return errorCodeToError(EC); 849 } 850 *LinkedObjectsFile << NewModulePath << '\n'; 851 } 852 853 std::map<std::string, GVSummaryMapTy> ModuleToSummariesForIndex; 854 gatherImportedSummariesForModule(ModulePath, ModuleToDefinedGVSummaries, 855 ImportList, ModuleToSummariesForIndex); 856 857 raw_fd_ostream OS(NewModulePath + ".thinlto.bc", EC, 858 sys::fs::OpenFlags::F_None); 859 if (EC) 860 return errorCodeToError(EC); 861 WriteIndexToFile(CombinedIndex, OS, &ModuleToSummariesForIndex); 862 863 if (ShouldEmitImportsFiles) 864 return errorCodeToError( 865 EmitImportsFiles(ModulePath, NewModulePath + ".imports", ImportList)); 866 return Error::success(); 867 } 868 869 Error wait() override { return Error::success(); } 870 }; 871 } // end anonymous namespace 872 873 ThinBackend lto::createWriteIndexesThinBackend(std::string OldPrefix, 874 std::string NewPrefix, 875 bool ShouldEmitImportsFiles, 876 std::string LinkedObjectsFile) { 877 return [=](Config &Conf, ModuleSummaryIndex &CombinedIndex, 878 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries, 879 AddStreamFn AddStream, NativeObjectCache Cache) { 880 return llvm::make_unique<WriteIndexesThinBackend>( 881 Conf, CombinedIndex, ModuleToDefinedGVSummaries, OldPrefix, NewPrefix, 882 ShouldEmitImportsFiles, LinkedObjectsFile); 883 }; 884 } 885 886 Error LTO::runThinLTO(AddStreamFn AddStream, NativeObjectCache Cache, 887 bool HasRegularLTO) { 888 if (ThinLTO.ModuleMap.empty()) 889 return Error::success(); 890 891 if (Conf.CombinedIndexHook && !Conf.CombinedIndexHook(ThinLTO.CombinedIndex)) 892 return Error::success(); 893 894 // Collect for each module the list of function it defines (GUID -> 895 // Summary). 896 StringMap<std::map<GlobalValue::GUID, GlobalValueSummary *>> 897 ModuleToDefinedGVSummaries(ThinLTO.ModuleMap.size()); 898 ThinLTO.CombinedIndex.collectDefinedGVSummariesPerModule( 899 ModuleToDefinedGVSummaries); 900 // Create entries for any modules that didn't have any GV summaries 901 // (either they didn't have any GVs to start with, or we suppressed 902 // generation of the summaries because they e.g. had inline assembly 903 // uses that couldn't be promoted/renamed on export). This is so 904 // InProcessThinBackend::start can still launch a backend thread, which 905 // is passed the map of summaries for the module, without any special 906 // handling for this case. 907 for (auto &Mod : ThinLTO.ModuleMap) 908 if (!ModuleToDefinedGVSummaries.count(Mod.first)) 909 ModuleToDefinedGVSummaries.try_emplace(Mod.first); 910 911 StringMap<FunctionImporter::ImportMapTy> ImportLists( 912 ThinLTO.ModuleMap.size()); 913 StringMap<FunctionImporter::ExportSetTy> ExportLists( 914 ThinLTO.ModuleMap.size()); 915 StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> ResolvedODR; 916 917 if (Conf.OptLevel > 0) { 918 // Compute "dead" symbols, we don't want to import/export these! 919 DenseSet<GlobalValue::GUID> GUIDPreservedSymbols; 920 for (auto &Res : GlobalResolutions) { 921 if (Res.second.VisibleOutsideThinLTO && 922 // IRName will be defined if we have seen the prevailing copy of 923 // this value. If not, no need to preserve any ThinLTO copies. 924 !Res.second.IRName.empty()) 925 GUIDPreservedSymbols.insert(GlobalValue::getGUID(Res.second.IRName)); 926 } 927 928 auto DeadSymbols = 929 computeDeadSymbols(ThinLTO.CombinedIndex, GUIDPreservedSymbols); 930 931 ComputeCrossModuleImport(ThinLTO.CombinedIndex, ModuleToDefinedGVSummaries, 932 ImportLists, ExportLists, &DeadSymbols); 933 934 std::set<GlobalValue::GUID> ExportedGUIDs; 935 for (auto &Res : GlobalResolutions) { 936 // First check if the symbol was flagged as having external references. 937 if (Res.second.Partition != GlobalResolution::External) 938 continue; 939 // IRName will be defined if we have seen the prevailing copy of 940 // this value. If not, no need to mark as exported from a ThinLTO 941 // partition (and we can't get the GUID). 942 if (Res.second.IRName.empty()) 943 continue; 944 auto GUID = GlobalValue::getGUID(Res.second.IRName); 945 // Mark exported unless index-based analysis determined it to be dead. 946 if (!DeadSymbols.count(GUID)) 947 ExportedGUIDs.insert(GlobalValue::getGUID(Res.second.IRName)); 948 } 949 950 auto isPrevailing = [&](GlobalValue::GUID GUID, 951 const GlobalValueSummary *S) { 952 return ThinLTO.PrevailingModuleForGUID[GUID] == S->modulePath(); 953 }; 954 auto isExported = [&](StringRef ModuleIdentifier, GlobalValue::GUID GUID) { 955 const auto &ExportList = ExportLists.find(ModuleIdentifier); 956 return (ExportList != ExportLists.end() && 957 ExportList->second.count(GUID)) || 958 ExportedGUIDs.count(GUID); 959 }; 960 thinLTOInternalizeAndPromoteInIndex(ThinLTO.CombinedIndex, isExported); 961 962 auto recordNewLinkage = [&](StringRef ModuleIdentifier, 963 GlobalValue::GUID GUID, 964 GlobalValue::LinkageTypes NewLinkage) { 965 ResolvedODR[ModuleIdentifier][GUID] = NewLinkage; 966 }; 967 968 thinLTOResolveWeakForLinkerInIndex(ThinLTO.CombinedIndex, isPrevailing, 969 recordNewLinkage); 970 } 971 972 std::unique_ptr<ThinBackendProc> BackendProc = 973 ThinLTO.Backend(Conf, ThinLTO.CombinedIndex, ModuleToDefinedGVSummaries, 974 AddStream, Cache); 975 976 // Task numbers start at ParallelCodeGenParallelismLevel if an LTO 977 // module is present, as tasks 0 through ParallelCodeGenParallelismLevel-1 978 // are reserved for parallel code generation partitions. 979 unsigned Task = 980 HasRegularLTO ? RegularLTO.ParallelCodeGenParallelismLevel : 0; 981 for (auto &Mod : ThinLTO.ModuleMap) { 982 if (Error E = BackendProc->start(Task, Mod.second, ImportLists[Mod.first], 983 ExportLists[Mod.first], 984 ResolvedODR[Mod.first], ThinLTO.ModuleMap)) 985 return E; 986 ++Task; 987 } 988 989 return BackendProc->wait(); 990 } 991 992 Expected<std::unique_ptr<tool_output_file>> 993 lto::setupOptimizationRemarks(LLVMContext &Context, 994 StringRef LTORemarksFilename, 995 bool LTOPassRemarksWithHotness, int Count) { 996 if (LTORemarksFilename.empty()) 997 return nullptr; 998 999 std::string Filename = LTORemarksFilename; 1000 if (Count != -1) 1001 Filename += ".thin." + llvm::utostr(Count) + ".yaml"; 1002 1003 std::error_code EC; 1004 auto DiagnosticFile = 1005 llvm::make_unique<tool_output_file>(Filename, EC, sys::fs::F_None); 1006 if (EC) 1007 return errorCodeToError(EC); 1008 Context.setDiagnosticsOutputFile( 1009 llvm::make_unique<yaml::Output>(DiagnosticFile->os())); 1010 if (LTOPassRemarksWithHotness) 1011 Context.setDiagnosticHotnessRequested(true); 1012 DiagnosticFile->keep(); 1013 return std::move(DiagnosticFile); 1014 } 1015