1 //===-LTO.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 functions and classes used to support LTO. 10 // 11 //===----------------------------------------------------------------------===// 12 13 #include "llvm/LTO/LTO.h" 14 #include "llvm/ADT/Statistic.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/Config/llvm-config.h" 21 #include "llvm/IR/AutoUpgrade.h" 22 #include "llvm/IR/DiagnosticPrinter.h" 23 #include "llvm/IR/Intrinsics.h" 24 #include "llvm/IR/LegacyPassManager.h" 25 #include "llvm/IR/Mangler.h" 26 #include "llvm/IR/Metadata.h" 27 #include "llvm/IR/RemarkStreamer.h" 28 #include "llvm/LTO/LTOBackend.h" 29 #include "llvm/LTO/SummaryBasedOptimizations.h" 30 #include "llvm/Linker/IRMover.h" 31 #include "llvm/Object/IRObjectFile.h" 32 #include "llvm/Support/Error.h" 33 #include "llvm/Support/ManagedStatic.h" 34 #include "llvm/Support/MemoryBuffer.h" 35 #include "llvm/Support/Path.h" 36 #include "llvm/Support/SHA1.h" 37 #include "llvm/Support/SourceMgr.h" 38 #include "llvm/Support/TargetRegistry.h" 39 #include "llvm/Support/ThreadPool.h" 40 #include "llvm/Support/Threading.h" 41 #include "llvm/Support/VCSRevision.h" 42 #include "llvm/Support/raw_ostream.h" 43 #include "llvm/Target/TargetMachine.h" 44 #include "llvm/Target/TargetOptions.h" 45 #include "llvm/Transforms/IPO.h" 46 #include "llvm/Transforms/IPO/PassManagerBuilder.h" 47 #include "llvm/Transforms/IPO/WholeProgramDevirt.h" 48 #include "llvm/Transforms/Utils/FunctionImportUtils.h" 49 #include "llvm/Transforms/Utils/SplitModule.h" 50 51 #include <set> 52 53 using namespace llvm; 54 using namespace lto; 55 using namespace object; 56 57 #define DEBUG_TYPE "lto" 58 59 static cl::opt<bool> 60 DumpThinCGSCCs("dump-thin-cg-sccs", cl::init(false), cl::Hidden, 61 cl::desc("Dump the SCCs in the ThinLTO index's callgraph")); 62 63 /// Enable global value internalization in LTO. 64 cl::opt<bool> EnableLTOInternalization( 65 "enable-lto-internalization", cl::init(true), cl::Hidden, 66 cl::desc("Enable global value internalization in LTO")); 67 68 // Computes a unique hash for the Module considering the current list of 69 // export/import and other global analysis results. 70 // The hash is produced in \p Key. 71 void llvm::computeLTOCacheKey( 72 SmallString<40> &Key, const Config &Conf, const ModuleSummaryIndex &Index, 73 StringRef ModuleID, const FunctionImporter::ImportMapTy &ImportList, 74 const FunctionImporter::ExportSetTy &ExportList, 75 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 76 const GVSummaryMapTy &DefinedGlobals, 77 const std::set<GlobalValue::GUID> &CfiFunctionDefs, 78 const std::set<GlobalValue::GUID> &CfiFunctionDecls) { 79 // Compute the unique hash for this entry. 80 // This is based on the current compiler version, the module itself, the 81 // export list, the hash for every single module in the import list, the 82 // list of ResolvedODR for the module, and the list of preserved symbols. 83 SHA1 Hasher; 84 85 // Start with the compiler revision 86 Hasher.update(LLVM_VERSION_STRING); 87 #ifdef LLVM_REVISION 88 Hasher.update(LLVM_REVISION); 89 #endif 90 91 // Include the parts of the LTO configuration that affect code generation. 92 auto AddString = [&](StringRef Str) { 93 Hasher.update(Str); 94 Hasher.update(ArrayRef<uint8_t>{0}); 95 }; 96 auto AddUnsigned = [&](unsigned I) { 97 uint8_t Data[4]; 98 Data[0] = I; 99 Data[1] = I >> 8; 100 Data[2] = I >> 16; 101 Data[3] = I >> 24; 102 Hasher.update(ArrayRef<uint8_t>{Data, 4}); 103 }; 104 auto AddUint64 = [&](uint64_t I) { 105 uint8_t Data[8]; 106 Data[0] = I; 107 Data[1] = I >> 8; 108 Data[2] = I >> 16; 109 Data[3] = I >> 24; 110 Data[4] = I >> 32; 111 Data[5] = I >> 40; 112 Data[6] = I >> 48; 113 Data[7] = I >> 56; 114 Hasher.update(ArrayRef<uint8_t>{Data, 8}); 115 }; 116 AddString(Conf.CPU); 117 // FIXME: Hash more of Options. For now all clients initialize Options from 118 // command-line flags (which is unsupported in production), but may set 119 // RelaxELFRelocations. The clang driver can also pass FunctionSections, 120 // DataSections and DebuggerTuning via command line flags. 121 AddUnsigned(Conf.Options.RelaxELFRelocations); 122 AddUnsigned(Conf.Options.FunctionSections); 123 AddUnsigned(Conf.Options.DataSections); 124 AddUnsigned((unsigned)Conf.Options.DebuggerTuning); 125 for (auto &A : Conf.MAttrs) 126 AddString(A); 127 if (Conf.RelocModel) 128 AddUnsigned(*Conf.RelocModel); 129 else 130 AddUnsigned(-1); 131 if (Conf.CodeModel) 132 AddUnsigned(*Conf.CodeModel); 133 else 134 AddUnsigned(-1); 135 AddUnsigned(Conf.CGOptLevel); 136 AddUnsigned(Conf.CGFileType); 137 AddUnsigned(Conf.OptLevel); 138 AddUnsigned(Conf.UseNewPM); 139 AddUnsigned(Conf.Freestanding); 140 AddString(Conf.OptPipeline); 141 AddString(Conf.AAPipeline); 142 AddString(Conf.OverrideTriple); 143 AddString(Conf.DefaultTriple); 144 AddString(Conf.DwoDir); 145 146 // Include the hash for the current module 147 auto ModHash = Index.getModuleHash(ModuleID); 148 Hasher.update(ArrayRef<uint8_t>((uint8_t *)&ModHash[0], sizeof(ModHash))); 149 for (auto F : ExportList) 150 // The export list can impact the internalization, be conservative here 151 Hasher.update(ArrayRef<uint8_t>((uint8_t *)&F, sizeof(F))); 152 153 // Include the hash for every module we import functions from. The set of 154 // imported symbols for each module may affect code generation and is 155 // sensitive to link order, so include that as well. 156 for (auto &Entry : ImportList) { 157 auto ModHash = Index.getModuleHash(Entry.first()); 158 Hasher.update(ArrayRef<uint8_t>((uint8_t *)&ModHash[0], sizeof(ModHash))); 159 160 AddUint64(Entry.second.size()); 161 for (auto &Fn : Entry.second) 162 AddUint64(Fn); 163 } 164 165 // Include the hash for the resolved ODR. 166 for (auto &Entry : ResolvedODR) { 167 Hasher.update(ArrayRef<uint8_t>((const uint8_t *)&Entry.first, 168 sizeof(GlobalValue::GUID))); 169 Hasher.update(ArrayRef<uint8_t>((const uint8_t *)&Entry.second, 170 sizeof(GlobalValue::LinkageTypes))); 171 } 172 173 // Members of CfiFunctionDefs and CfiFunctionDecls that are referenced or 174 // defined in this module. 175 std::set<GlobalValue::GUID> UsedCfiDefs; 176 std::set<GlobalValue::GUID> UsedCfiDecls; 177 178 // Typeids used in this module. 179 std::set<GlobalValue::GUID> UsedTypeIds; 180 181 auto AddUsedCfiGlobal = [&](GlobalValue::GUID ValueGUID) { 182 if (CfiFunctionDefs.count(ValueGUID)) 183 UsedCfiDefs.insert(ValueGUID); 184 if (CfiFunctionDecls.count(ValueGUID)) 185 UsedCfiDecls.insert(ValueGUID); 186 }; 187 188 auto AddUsedThings = [&](GlobalValueSummary *GS) { 189 if (!GS) return; 190 AddUnsigned(GS->isLive()); 191 AddUnsigned(GS->canAutoHide()); 192 for (const ValueInfo &VI : GS->refs()) { 193 AddUnsigned(VI.isDSOLocal()); 194 AddUsedCfiGlobal(VI.getGUID()); 195 } 196 if (auto *GVS = dyn_cast<GlobalVarSummary>(GS)) { 197 AddUnsigned(GVS->maybeReadOnly()); 198 AddUnsigned(GVS->maybeWriteOnly()); 199 } 200 if (auto *FS = dyn_cast<FunctionSummary>(GS)) { 201 for (auto &TT : FS->type_tests()) 202 UsedTypeIds.insert(TT); 203 for (auto &TT : FS->type_test_assume_vcalls()) 204 UsedTypeIds.insert(TT.GUID); 205 for (auto &TT : FS->type_checked_load_vcalls()) 206 UsedTypeIds.insert(TT.GUID); 207 for (auto &TT : FS->type_test_assume_const_vcalls()) 208 UsedTypeIds.insert(TT.VFunc.GUID); 209 for (auto &TT : FS->type_checked_load_const_vcalls()) 210 UsedTypeIds.insert(TT.VFunc.GUID); 211 for (auto &ET : FS->calls()) { 212 AddUnsigned(ET.first.isDSOLocal()); 213 AddUsedCfiGlobal(ET.first.getGUID()); 214 } 215 } 216 }; 217 218 // Include the hash for the linkage type to reflect internalization and weak 219 // resolution, and collect any used type identifier resolutions. 220 for (auto &GS : DefinedGlobals) { 221 GlobalValue::LinkageTypes Linkage = GS.second->linkage(); 222 Hasher.update( 223 ArrayRef<uint8_t>((const uint8_t *)&Linkage, sizeof(Linkage))); 224 AddUsedCfiGlobal(GS.first); 225 AddUsedThings(GS.second); 226 } 227 228 // Imported functions may introduce new uses of type identifier resolutions, 229 // so we need to collect their used resolutions as well. 230 for (auto &ImpM : ImportList) 231 for (auto &ImpF : ImpM.second) { 232 GlobalValueSummary *S = Index.findSummaryInModule(ImpF, ImpM.first()); 233 AddUsedThings(S); 234 // If this is an alias, we also care about any types/etc. that the aliasee 235 // may reference. 236 if (auto *AS = dyn_cast_or_null<AliasSummary>(S)) 237 AddUsedThings(AS->getBaseObject()); 238 } 239 240 auto AddTypeIdSummary = [&](StringRef TId, const TypeIdSummary &S) { 241 AddString(TId); 242 243 AddUnsigned(S.TTRes.TheKind); 244 AddUnsigned(S.TTRes.SizeM1BitWidth); 245 246 AddUint64(S.TTRes.AlignLog2); 247 AddUint64(S.TTRes.SizeM1); 248 AddUint64(S.TTRes.BitMask); 249 AddUint64(S.TTRes.InlineBits); 250 251 AddUint64(S.WPDRes.size()); 252 for (auto &WPD : S.WPDRes) { 253 AddUnsigned(WPD.first); 254 AddUnsigned(WPD.second.TheKind); 255 AddString(WPD.second.SingleImplName); 256 257 AddUint64(WPD.second.ResByArg.size()); 258 for (auto &ByArg : WPD.second.ResByArg) { 259 AddUint64(ByArg.first.size()); 260 for (uint64_t Arg : ByArg.first) 261 AddUint64(Arg); 262 AddUnsigned(ByArg.second.TheKind); 263 AddUint64(ByArg.second.Info); 264 AddUnsigned(ByArg.second.Byte); 265 AddUnsigned(ByArg.second.Bit); 266 } 267 } 268 }; 269 270 // Include the hash for all type identifiers used by this module. 271 for (GlobalValue::GUID TId : UsedTypeIds) { 272 auto TidIter = Index.typeIds().equal_range(TId); 273 for (auto It = TidIter.first; It != TidIter.second; ++It) 274 AddTypeIdSummary(It->second.first, It->second.second); 275 } 276 277 AddUnsigned(UsedCfiDefs.size()); 278 for (auto &V : UsedCfiDefs) 279 AddUint64(V); 280 281 AddUnsigned(UsedCfiDecls.size()); 282 for (auto &V : UsedCfiDecls) 283 AddUint64(V); 284 285 if (!Conf.SampleProfile.empty()) { 286 auto FileOrErr = MemoryBuffer::getFile(Conf.SampleProfile); 287 if (FileOrErr) { 288 Hasher.update(FileOrErr.get()->getBuffer()); 289 290 if (!Conf.ProfileRemapping.empty()) { 291 FileOrErr = MemoryBuffer::getFile(Conf.ProfileRemapping); 292 if (FileOrErr) 293 Hasher.update(FileOrErr.get()->getBuffer()); 294 } 295 } 296 } 297 298 Key = toHex(Hasher.result()); 299 } 300 301 static void thinLTOResolvePrevailingGUID( 302 ValueInfo VI, DenseSet<GlobalValueSummary *> &GlobalInvolvedWithAlias, 303 function_ref<bool(GlobalValue::GUID, const GlobalValueSummary *)> 304 isPrevailing, 305 function_ref<void(StringRef, GlobalValue::GUID, GlobalValue::LinkageTypes)> 306 recordNewLinkage, 307 const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols) { 308 for (auto &S : VI.getSummaryList()) { 309 GlobalValue::LinkageTypes OriginalLinkage = S->linkage(); 310 // Ignore local and appending linkage values since the linker 311 // doesn't resolve them. 312 if (GlobalValue::isLocalLinkage(OriginalLinkage) || 313 GlobalValue::isAppendingLinkage(S->linkage())) 314 continue; 315 // We need to emit only one of these. The prevailing module will keep it, 316 // but turned into a weak, while the others will drop it when possible. 317 // This is both a compile-time optimization and a correctness 318 // transformation. This is necessary for correctness when we have exported 319 // a reference - we need to convert the linkonce to weak to 320 // ensure a copy is kept to satisfy the exported reference. 321 // FIXME: We may want to split the compile time and correctness 322 // aspects into separate routines. 323 if (isPrevailing(VI.getGUID(), S.get())) { 324 if (GlobalValue::isLinkOnceLinkage(OriginalLinkage)) { 325 S->setLinkage(GlobalValue::getWeakLinkage( 326 GlobalValue::isLinkOnceODRLinkage(OriginalLinkage))); 327 // The kept copy is eligible for auto-hiding (hidden visibility) if all 328 // copies were (i.e. they were all linkonce_odr global unnamed addr). 329 // If any copy is not (e.g. it was originally weak_odr), then the symbol 330 // must remain externally available (e.g. a weak_odr from an explicitly 331 // instantiated template). Additionally, if it is in the 332 // GUIDPreservedSymbols set, that means that it is visibile outside 333 // the summary (e.g. in a native object or a bitcode file without 334 // summary), and in that case we cannot hide it as it isn't possible to 335 // check all copies. 336 S->setCanAutoHide(VI.canAutoHide() && 337 !GUIDPreservedSymbols.count(VI.getGUID())); 338 } 339 } 340 // Alias and aliasee can't be turned into available_externally. 341 else if (!isa<AliasSummary>(S.get()) && 342 !GlobalInvolvedWithAlias.count(S.get())) 343 S->setLinkage(GlobalValue::AvailableExternallyLinkage); 344 if (S->linkage() != OriginalLinkage) 345 recordNewLinkage(S->modulePath(), VI.getGUID(), S->linkage()); 346 } 347 } 348 349 /// Resolve linkage for prevailing symbols in the \p Index. 350 // 351 // We'd like to drop these functions if they are no longer referenced in the 352 // current module. However there is a chance that another module is still 353 // referencing them because of the import. We make sure we always emit at least 354 // one copy. 355 void llvm::thinLTOResolvePrevailingInIndex( 356 ModuleSummaryIndex &Index, 357 function_ref<bool(GlobalValue::GUID, const GlobalValueSummary *)> 358 isPrevailing, 359 function_ref<void(StringRef, GlobalValue::GUID, GlobalValue::LinkageTypes)> 360 recordNewLinkage, 361 const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols) { 362 // We won't optimize the globals that are referenced by an alias for now 363 // Ideally we should turn the alias into a global and duplicate the definition 364 // when needed. 365 DenseSet<GlobalValueSummary *> GlobalInvolvedWithAlias; 366 for (auto &I : Index) 367 for (auto &S : I.second.SummaryList) 368 if (auto AS = dyn_cast<AliasSummary>(S.get())) 369 GlobalInvolvedWithAlias.insert(&AS->getAliasee()); 370 371 for (auto &I : Index) 372 thinLTOResolvePrevailingGUID(Index.getValueInfo(I), GlobalInvolvedWithAlias, 373 isPrevailing, recordNewLinkage, 374 GUIDPreservedSymbols); 375 } 376 377 static bool isWeakObjectWithRWAccess(GlobalValueSummary *GVS) { 378 if (auto *VarSummary = dyn_cast<GlobalVarSummary>(GVS->getBaseObject())) 379 return !VarSummary->maybeReadOnly() && !VarSummary->maybeWriteOnly() && 380 (VarSummary->linkage() == GlobalValue::WeakODRLinkage || 381 VarSummary->linkage() == GlobalValue::LinkOnceODRLinkage); 382 return false; 383 } 384 385 static void thinLTOInternalizeAndPromoteGUID( 386 GlobalValueSummaryList &GVSummaryList, GlobalValue::GUID GUID, 387 function_ref<bool(StringRef, GlobalValue::GUID)> isExported, 388 function_ref<bool(GlobalValue::GUID, const GlobalValueSummary *)> 389 isPrevailing) { 390 for (auto &S : GVSummaryList) { 391 if (isExported(S->modulePath(), GUID)) { 392 if (GlobalValue::isLocalLinkage(S->linkage())) 393 S->setLinkage(GlobalValue::ExternalLinkage); 394 } else if (EnableLTOInternalization && 395 // Ignore local and appending linkage values since the linker 396 // doesn't resolve them. 397 !GlobalValue::isLocalLinkage(S->linkage()) && 398 (!GlobalValue::isInterposableLinkage(S->linkage()) || 399 isPrevailing(GUID, S.get())) && 400 S->linkage() != GlobalValue::AppendingLinkage && 401 // We can't internalize available_externally globals because this 402 // can break function pointer equality. 403 S->linkage() != GlobalValue::AvailableExternallyLinkage && 404 // Functions and read-only variables with linkonce_odr and 405 // weak_odr linkage can be internalized. We can't internalize 406 // linkonce_odr and weak_odr variables which are both modified 407 // and read somewhere in the program because reads and writes 408 // will become inconsistent. 409 !isWeakObjectWithRWAccess(S.get())) 410 S->setLinkage(GlobalValue::InternalLinkage); 411 } 412 } 413 414 // Update the linkages in the given \p Index to mark exported values 415 // as external and non-exported values as internal. 416 void llvm::thinLTOInternalizeAndPromoteInIndex( 417 ModuleSummaryIndex &Index, 418 function_ref<bool(StringRef, GlobalValue::GUID)> isExported, 419 function_ref<bool(GlobalValue::GUID, const GlobalValueSummary *)> 420 isPrevailing) { 421 for (auto &I : Index) 422 thinLTOInternalizeAndPromoteGUID(I.second.SummaryList, I.first, isExported, 423 isPrevailing); 424 } 425 426 // Requires a destructor for std::vector<InputModule>. 427 InputFile::~InputFile() = default; 428 429 Expected<std::unique_ptr<InputFile>> InputFile::create(MemoryBufferRef Object) { 430 std::unique_ptr<InputFile> File(new InputFile); 431 432 Expected<IRSymtabFile> FOrErr = readIRSymtab(Object); 433 if (!FOrErr) 434 return FOrErr.takeError(); 435 436 File->TargetTriple = FOrErr->TheReader.getTargetTriple(); 437 File->SourceFileName = FOrErr->TheReader.getSourceFileName(); 438 File->COFFLinkerOpts = FOrErr->TheReader.getCOFFLinkerOpts(); 439 File->DependentLibraries = FOrErr->TheReader.getDependentLibraries(); 440 File->ComdatTable = FOrErr->TheReader.getComdatTable(); 441 442 for (unsigned I = 0; I != FOrErr->Mods.size(); ++I) { 443 size_t Begin = File->Symbols.size(); 444 for (const irsymtab::Reader::SymbolRef &Sym : 445 FOrErr->TheReader.module_symbols(I)) 446 // Skip symbols that are irrelevant to LTO. Note that this condition needs 447 // to match the one in Skip() in LTO::addRegularLTO(). 448 if (Sym.isGlobal() && !Sym.isFormatSpecific()) 449 File->Symbols.push_back(Sym); 450 File->ModuleSymIndices.push_back({Begin, File->Symbols.size()}); 451 } 452 453 File->Mods = FOrErr->Mods; 454 File->Strtab = std::move(FOrErr->Strtab); 455 return std::move(File); 456 } 457 458 StringRef InputFile::getName() const { 459 return Mods[0].getModuleIdentifier(); 460 } 461 462 BitcodeModule &InputFile::getSingleBitcodeModule() { 463 assert(Mods.size() == 1 && "Expect only one bitcode module"); 464 return Mods[0]; 465 } 466 467 LTO::RegularLTOState::RegularLTOState(unsigned ParallelCodeGenParallelismLevel, 468 Config &Conf) 469 : ParallelCodeGenParallelismLevel(ParallelCodeGenParallelismLevel), 470 Ctx(Conf), CombinedModule(std::make_unique<Module>("ld-temp.o", Ctx)), 471 Mover(std::make_unique<IRMover>(*CombinedModule)) {} 472 473 LTO::ThinLTOState::ThinLTOState(ThinBackend Backend) 474 : Backend(Backend), CombinedIndex(/*HaveGVs*/ false) { 475 if (!Backend) 476 this->Backend = 477 createInProcessThinBackend(llvm::heavyweight_hardware_concurrency()); 478 } 479 480 LTO::LTO(Config Conf, ThinBackend Backend, 481 unsigned ParallelCodeGenParallelismLevel) 482 : Conf(std::move(Conf)), 483 RegularLTO(ParallelCodeGenParallelismLevel, this->Conf), 484 ThinLTO(std::move(Backend)) {} 485 486 // Requires a destructor for MapVector<BitcodeModule>. 487 LTO::~LTO() = default; 488 489 // Add the symbols in the given module to the GlobalResolutions map, and resolve 490 // their partitions. 491 void LTO::addModuleToGlobalRes(ArrayRef<InputFile::Symbol> Syms, 492 ArrayRef<SymbolResolution> Res, 493 unsigned Partition, bool InSummary) { 494 auto *ResI = Res.begin(); 495 auto *ResE = Res.end(); 496 (void)ResE; 497 for (const InputFile::Symbol &Sym : Syms) { 498 assert(ResI != ResE); 499 SymbolResolution Res = *ResI++; 500 501 StringRef Name = Sym.getName(); 502 Triple TT(RegularLTO.CombinedModule->getTargetTriple()); 503 // Strip the __imp_ prefix from COFF dllimport symbols (similar to the 504 // way they are handled by lld), otherwise we can end up with two 505 // global resolutions (one with and one for a copy of the symbol without). 506 if (TT.isOSBinFormatCOFF() && Name.startswith("__imp_")) 507 Name = Name.substr(strlen("__imp_")); 508 auto &GlobalRes = GlobalResolutions[Name]; 509 GlobalRes.UnnamedAddr &= Sym.isUnnamedAddr(); 510 if (Res.Prevailing) { 511 assert(!GlobalRes.Prevailing && 512 "Multiple prevailing defs are not allowed"); 513 GlobalRes.Prevailing = true; 514 GlobalRes.IRName = Sym.getIRName(); 515 } else if (!GlobalRes.Prevailing && GlobalRes.IRName.empty()) { 516 // Sometimes it can be two copies of symbol in a module and prevailing 517 // symbol can have no IR name. That might happen if symbol is defined in 518 // module level inline asm block. In case we have multiple modules with 519 // the same symbol we want to use IR name of the prevailing symbol. 520 // Otherwise, if we haven't seen a prevailing symbol, set the name so that 521 // we can later use it to check if there is any prevailing copy in IR. 522 GlobalRes.IRName = Sym.getIRName(); 523 } 524 525 // Set the partition to external if we know it is re-defined by the linker 526 // with -defsym or -wrap options, used elsewhere, e.g. it is visible to a 527 // regular object, is referenced from llvm.compiler_used, or was already 528 // recorded as being referenced from a different partition. 529 if (Res.LinkerRedefined || Res.VisibleToRegularObj || Sym.isUsed() || 530 (GlobalRes.Partition != GlobalResolution::Unknown && 531 GlobalRes.Partition != Partition)) { 532 GlobalRes.Partition = GlobalResolution::External; 533 } else 534 // First recorded reference, save the current partition. 535 GlobalRes.Partition = Partition; 536 537 // Flag as visible outside of summary if visible from a regular object or 538 // from a module that does not have a summary. 539 GlobalRes.VisibleOutsideSummary |= 540 (Res.VisibleToRegularObj || Sym.isUsed() || !InSummary); 541 } 542 } 543 544 static void writeToResolutionFile(raw_ostream &OS, InputFile *Input, 545 ArrayRef<SymbolResolution> Res) { 546 StringRef Path = Input->getName(); 547 OS << Path << '\n'; 548 auto ResI = Res.begin(); 549 for (const InputFile::Symbol &Sym : Input->symbols()) { 550 assert(ResI != Res.end()); 551 SymbolResolution Res = *ResI++; 552 553 OS << "-r=" << Path << ',' << Sym.getName() << ','; 554 if (Res.Prevailing) 555 OS << 'p'; 556 if (Res.FinalDefinitionInLinkageUnit) 557 OS << 'l'; 558 if (Res.VisibleToRegularObj) 559 OS << 'x'; 560 if (Res.LinkerRedefined) 561 OS << 'r'; 562 OS << '\n'; 563 } 564 OS.flush(); 565 assert(ResI == Res.end()); 566 } 567 568 Error LTO::add(std::unique_ptr<InputFile> Input, 569 ArrayRef<SymbolResolution> Res) { 570 assert(!CalledGetMaxTasks); 571 572 if (Conf.ResolutionFile) 573 writeToResolutionFile(*Conf.ResolutionFile, Input.get(), Res); 574 575 if (RegularLTO.CombinedModule->getTargetTriple().empty()) 576 RegularLTO.CombinedModule->setTargetTriple(Input->getTargetTriple()); 577 578 const SymbolResolution *ResI = Res.begin(); 579 for (unsigned I = 0; I != Input->Mods.size(); ++I) 580 if (Error Err = addModule(*Input, I, ResI, Res.end())) 581 return Err; 582 583 assert(ResI == Res.end()); 584 return Error::success(); 585 } 586 587 Error LTO::addModule(InputFile &Input, unsigned ModI, 588 const SymbolResolution *&ResI, 589 const SymbolResolution *ResE) { 590 Expected<BitcodeLTOInfo> LTOInfo = Input.Mods[ModI].getLTOInfo(); 591 if (!LTOInfo) 592 return LTOInfo.takeError(); 593 594 if (EnableSplitLTOUnit.hasValue()) { 595 // If only some modules were split, flag this in the index so that 596 // we can skip or error on optimizations that need consistently split 597 // modules (whole program devirt and lower type tests). 598 if (EnableSplitLTOUnit.getValue() != LTOInfo->EnableSplitLTOUnit) 599 ThinLTO.CombinedIndex.setPartiallySplitLTOUnits(); 600 } else 601 EnableSplitLTOUnit = LTOInfo->EnableSplitLTOUnit; 602 603 BitcodeModule BM = Input.Mods[ModI]; 604 auto ModSyms = Input.module_symbols(ModI); 605 addModuleToGlobalRes(ModSyms, {ResI, ResE}, 606 LTOInfo->IsThinLTO ? ThinLTO.ModuleMap.size() + 1 : 0, 607 LTOInfo->HasSummary); 608 609 if (LTOInfo->IsThinLTO) 610 return addThinLTO(BM, ModSyms, ResI, ResE); 611 612 Expected<RegularLTOState::AddedModule> ModOrErr = 613 addRegularLTO(BM, ModSyms, ResI, ResE); 614 if (!ModOrErr) 615 return ModOrErr.takeError(); 616 617 if (!LTOInfo->HasSummary) 618 return linkRegularLTO(std::move(*ModOrErr), /*LivenessFromIndex=*/false); 619 620 // Regular LTO module summaries are added to a dummy module that represents 621 // the combined regular LTO module. 622 if (Error Err = BM.readSummary(ThinLTO.CombinedIndex, "", -1ull)) 623 return Err; 624 RegularLTO.ModsWithSummaries.push_back(std::move(*ModOrErr)); 625 return Error::success(); 626 } 627 628 // Checks whether the given global value is in a non-prevailing comdat 629 // (comdat containing values the linker indicated were not prevailing, 630 // which we then dropped to available_externally), and if so, removes 631 // it from the comdat. This is called for all global values to ensure the 632 // comdat is empty rather than leaving an incomplete comdat. It is needed for 633 // regular LTO modules, in case we are in a mixed-LTO mode (both regular 634 // and thin LTO modules) compilation. Since the regular LTO module will be 635 // linked first in the final native link, we want to make sure the linker 636 // doesn't select any of these incomplete comdats that would be left 637 // in the regular LTO module without this cleanup. 638 static void 639 handleNonPrevailingComdat(GlobalValue &GV, 640 std::set<const Comdat *> &NonPrevailingComdats) { 641 Comdat *C = GV.getComdat(); 642 if (!C) 643 return; 644 645 if (!NonPrevailingComdats.count(C)) 646 return; 647 648 // Additionally need to drop externally visible global values from the comdat 649 // to available_externally, so that there aren't multiply defined linker 650 // errors. 651 if (!GV.hasLocalLinkage()) 652 GV.setLinkage(GlobalValue::AvailableExternallyLinkage); 653 654 if (auto GO = dyn_cast<GlobalObject>(&GV)) 655 GO->setComdat(nullptr); 656 } 657 658 // Add a regular LTO object to the link. 659 // The resulting module needs to be linked into the combined LTO module with 660 // linkRegularLTO. 661 Expected<LTO::RegularLTOState::AddedModule> 662 LTO::addRegularLTO(BitcodeModule BM, ArrayRef<InputFile::Symbol> Syms, 663 const SymbolResolution *&ResI, 664 const SymbolResolution *ResE) { 665 RegularLTOState::AddedModule Mod; 666 Expected<std::unique_ptr<Module>> MOrErr = 667 BM.getLazyModule(RegularLTO.Ctx, /*ShouldLazyLoadMetadata*/ true, 668 /*IsImporting*/ false); 669 if (!MOrErr) 670 return MOrErr.takeError(); 671 Module &M = **MOrErr; 672 Mod.M = std::move(*MOrErr); 673 674 if (Error Err = M.materializeMetadata()) 675 return std::move(Err); 676 UpgradeDebugInfo(M); 677 678 ModuleSymbolTable SymTab; 679 SymTab.addModule(&M); 680 681 for (GlobalVariable &GV : M.globals()) 682 if (GV.hasAppendingLinkage()) 683 Mod.Keep.push_back(&GV); 684 685 DenseSet<GlobalObject *> AliasedGlobals; 686 for (auto &GA : M.aliases()) 687 if (GlobalObject *GO = GA.getBaseObject()) 688 AliasedGlobals.insert(GO); 689 690 // In this function we need IR GlobalValues matching the symbols in Syms 691 // (which is not backed by a module), so we need to enumerate them in the same 692 // order. The symbol enumeration order of a ModuleSymbolTable intentionally 693 // matches the order of an irsymtab, but when we read the irsymtab in 694 // InputFile::create we omit some symbols that are irrelevant to LTO. The 695 // Skip() function skips the same symbols from the module as InputFile does 696 // from the symbol table. 697 auto MsymI = SymTab.symbols().begin(), MsymE = SymTab.symbols().end(); 698 auto Skip = [&]() { 699 while (MsymI != MsymE) { 700 auto Flags = SymTab.getSymbolFlags(*MsymI); 701 if ((Flags & object::BasicSymbolRef::SF_Global) && 702 !(Flags & object::BasicSymbolRef::SF_FormatSpecific)) 703 return; 704 ++MsymI; 705 } 706 }; 707 Skip(); 708 709 std::set<const Comdat *> NonPrevailingComdats; 710 for (const InputFile::Symbol &Sym : Syms) { 711 assert(ResI != ResE); 712 SymbolResolution Res = *ResI++; 713 714 assert(MsymI != MsymE); 715 ModuleSymbolTable::Symbol Msym = *MsymI++; 716 Skip(); 717 718 if (GlobalValue *GV = Msym.dyn_cast<GlobalValue *>()) { 719 if (Res.Prevailing) { 720 if (Sym.isUndefined()) 721 continue; 722 Mod.Keep.push_back(GV); 723 // For symbols re-defined with linker -wrap and -defsym options, 724 // set the linkage to weak to inhibit IPO. The linkage will be 725 // restored by the linker. 726 if (Res.LinkerRedefined) 727 GV->setLinkage(GlobalValue::WeakAnyLinkage); 728 729 GlobalValue::LinkageTypes OriginalLinkage = GV->getLinkage(); 730 if (GlobalValue::isLinkOnceLinkage(OriginalLinkage)) 731 GV->setLinkage(GlobalValue::getWeakLinkage( 732 GlobalValue::isLinkOnceODRLinkage(OriginalLinkage))); 733 } else if (isa<GlobalObject>(GV) && 734 (GV->hasLinkOnceODRLinkage() || GV->hasWeakODRLinkage() || 735 GV->hasAvailableExternallyLinkage()) && 736 !AliasedGlobals.count(cast<GlobalObject>(GV))) { 737 // Any of the above three types of linkage indicates that the 738 // chosen prevailing symbol will have the same semantics as this copy of 739 // the symbol, so we may be able to link it with available_externally 740 // linkage. We will decide later whether to do that when we link this 741 // module (in linkRegularLTO), based on whether it is undefined. 742 Mod.Keep.push_back(GV); 743 GV->setLinkage(GlobalValue::AvailableExternallyLinkage); 744 if (GV->hasComdat()) 745 NonPrevailingComdats.insert(GV->getComdat()); 746 cast<GlobalObject>(GV)->setComdat(nullptr); 747 } 748 749 // Set the 'local' flag based on the linker resolution for this symbol. 750 if (Res.FinalDefinitionInLinkageUnit) { 751 GV->setDSOLocal(true); 752 if (GV->hasDLLImportStorageClass()) 753 GV->setDLLStorageClass(GlobalValue::DLLStorageClassTypes:: 754 DefaultStorageClass); 755 } 756 } 757 // Common resolution: collect the maximum size/alignment over all commons. 758 // We also record if we see an instance of a common as prevailing, so that 759 // if none is prevailing we can ignore it later. 760 if (Sym.isCommon()) { 761 // FIXME: We should figure out what to do about commons defined by asm. 762 // For now they aren't reported correctly by ModuleSymbolTable. 763 auto &CommonRes = RegularLTO.Commons[Sym.getIRName()]; 764 CommonRes.Size = std::max(CommonRes.Size, Sym.getCommonSize()); 765 CommonRes.Align = 766 std::max(CommonRes.Align, MaybeAlign(Sym.getCommonAlignment())); 767 CommonRes.Prevailing |= Res.Prevailing; 768 } 769 770 } 771 if (!M.getComdatSymbolTable().empty()) 772 for (GlobalValue &GV : M.global_values()) 773 handleNonPrevailingComdat(GV, NonPrevailingComdats); 774 assert(MsymI == MsymE); 775 return std::move(Mod); 776 } 777 778 Error LTO::linkRegularLTO(RegularLTOState::AddedModule Mod, 779 bool LivenessFromIndex) { 780 std::vector<GlobalValue *> Keep; 781 for (GlobalValue *GV : Mod.Keep) { 782 if (LivenessFromIndex && !ThinLTO.CombinedIndex.isGUIDLive(GV->getGUID())) 783 continue; 784 785 if (!GV->hasAvailableExternallyLinkage()) { 786 Keep.push_back(GV); 787 continue; 788 } 789 790 // Only link available_externally definitions if we don't already have a 791 // definition. 792 GlobalValue *CombinedGV = 793 RegularLTO.CombinedModule->getNamedValue(GV->getName()); 794 if (CombinedGV && !CombinedGV->isDeclaration()) 795 continue; 796 797 Keep.push_back(GV); 798 } 799 800 return RegularLTO.Mover->move(std::move(Mod.M), Keep, 801 [](GlobalValue &, IRMover::ValueAdder) {}, 802 /* IsPerformingImport */ false); 803 } 804 805 // Add a ThinLTO module to the link. 806 Error LTO::addThinLTO(BitcodeModule BM, ArrayRef<InputFile::Symbol> Syms, 807 const SymbolResolution *&ResI, 808 const SymbolResolution *ResE) { 809 if (Error Err = 810 BM.readSummary(ThinLTO.CombinedIndex, BM.getModuleIdentifier(), 811 ThinLTO.ModuleMap.size())) 812 return Err; 813 814 for (const InputFile::Symbol &Sym : Syms) { 815 assert(ResI != ResE); 816 SymbolResolution Res = *ResI++; 817 818 if (!Sym.getIRName().empty()) { 819 auto GUID = GlobalValue::getGUID(GlobalValue::getGlobalIdentifier( 820 Sym.getIRName(), GlobalValue::ExternalLinkage, "")); 821 if (Res.Prevailing) { 822 ThinLTO.PrevailingModuleForGUID[GUID] = BM.getModuleIdentifier(); 823 824 // For linker redefined symbols (via --wrap or --defsym) we want to 825 // switch the linkage to `weak` to prevent IPOs from happening. 826 // Find the summary in the module for this very GV and record the new 827 // linkage so that we can switch it when we import the GV. 828 if (Res.LinkerRedefined) 829 if (auto S = ThinLTO.CombinedIndex.findSummaryInModule( 830 GUID, BM.getModuleIdentifier())) 831 S->setLinkage(GlobalValue::WeakAnyLinkage); 832 } 833 834 // If the linker resolved the symbol to a local definition then mark it 835 // as local in the summary for the module we are adding. 836 if (Res.FinalDefinitionInLinkageUnit) { 837 if (auto S = ThinLTO.CombinedIndex.findSummaryInModule( 838 GUID, BM.getModuleIdentifier())) { 839 S->setDSOLocal(true); 840 } 841 } 842 } 843 } 844 845 if (!ThinLTO.ModuleMap.insert({BM.getModuleIdentifier(), BM}).second) 846 return make_error<StringError>( 847 "Expected at most one ThinLTO module per bitcode file", 848 inconvertibleErrorCode()); 849 850 return Error::success(); 851 } 852 853 unsigned LTO::getMaxTasks() const { 854 CalledGetMaxTasks = true; 855 return RegularLTO.ParallelCodeGenParallelismLevel + ThinLTO.ModuleMap.size(); 856 } 857 858 // If only some of the modules were split, we cannot correctly handle 859 // code that contains type tests or type checked loads. 860 Error LTO::checkPartiallySplit() { 861 if (!ThinLTO.CombinedIndex.partiallySplitLTOUnits()) 862 return Error::success(); 863 864 Function *TypeTestFunc = RegularLTO.CombinedModule->getFunction( 865 Intrinsic::getName(Intrinsic::type_test)); 866 Function *TypeCheckedLoadFunc = RegularLTO.CombinedModule->getFunction( 867 Intrinsic::getName(Intrinsic::type_checked_load)); 868 869 // First check if there are type tests / type checked loads in the 870 // merged regular LTO module IR. 871 if ((TypeTestFunc && !TypeTestFunc->use_empty()) || 872 (TypeCheckedLoadFunc && !TypeCheckedLoadFunc->use_empty())) 873 return make_error<StringError>( 874 "inconsistent LTO Unit splitting (recompile with -fsplit-lto-unit)", 875 inconvertibleErrorCode()); 876 877 // Otherwise check if there are any recorded in the combined summary from the 878 // ThinLTO modules. 879 for (auto &P : ThinLTO.CombinedIndex) { 880 for (auto &S : P.second.SummaryList) { 881 auto *FS = dyn_cast<FunctionSummary>(S.get()); 882 if (!FS) 883 continue; 884 if (!FS->type_test_assume_vcalls().empty() || 885 !FS->type_checked_load_vcalls().empty() || 886 !FS->type_test_assume_const_vcalls().empty() || 887 !FS->type_checked_load_const_vcalls().empty() || 888 !FS->type_tests().empty()) 889 return make_error<StringError>( 890 "inconsistent LTO Unit splitting (recompile with -fsplit-lto-unit)", 891 inconvertibleErrorCode()); 892 } 893 } 894 return Error::success(); 895 } 896 897 Error LTO::run(AddStreamFn AddStream, NativeObjectCache Cache) { 898 // Compute "dead" symbols, we don't want to import/export these! 899 DenseSet<GlobalValue::GUID> GUIDPreservedSymbols; 900 DenseMap<GlobalValue::GUID, PrevailingType> GUIDPrevailingResolutions; 901 for (auto &Res : GlobalResolutions) { 902 // Normally resolution have IR name of symbol. We can do nothing here 903 // otherwise. See comments in GlobalResolution struct for more details. 904 if (Res.second.IRName.empty()) 905 continue; 906 907 GlobalValue::GUID GUID = GlobalValue::getGUID( 908 GlobalValue::dropLLVMManglingEscape(Res.second.IRName)); 909 910 if (Res.second.VisibleOutsideSummary && Res.second.Prevailing) 911 GUIDPreservedSymbols.insert(GUID); 912 913 GUIDPrevailingResolutions[GUID] = 914 Res.second.Prevailing ? PrevailingType::Yes : PrevailingType::No; 915 } 916 917 auto isPrevailing = [&](GlobalValue::GUID G) { 918 auto It = GUIDPrevailingResolutions.find(G); 919 if (It == GUIDPrevailingResolutions.end()) 920 return PrevailingType::Unknown; 921 return It->second; 922 }; 923 computeDeadSymbolsWithConstProp(ThinLTO.CombinedIndex, GUIDPreservedSymbols, 924 isPrevailing, Conf.OptLevel > 0); 925 926 // Setup output file to emit statistics. 927 auto StatsFileOrErr = setupStatsFile(Conf.StatsFile); 928 if (!StatsFileOrErr) 929 return StatsFileOrErr.takeError(); 930 std::unique_ptr<ToolOutputFile> StatsFile = std::move(StatsFileOrErr.get()); 931 932 // Finalize linking of regular LTO modules containing summaries now that 933 // we have computed liveness information. 934 for (auto &M : RegularLTO.ModsWithSummaries) 935 if (Error Err = linkRegularLTO(std::move(M), 936 /*LivenessFromIndex=*/true)) 937 return Err; 938 939 // Ensure we don't have inconsistently split LTO units with type tests. 940 if (Error Err = checkPartiallySplit()) 941 return Err; 942 943 Error Result = runRegularLTO(AddStream); 944 if (!Result) 945 Result = runThinLTO(AddStream, Cache, GUIDPreservedSymbols); 946 947 if (StatsFile) 948 PrintStatisticsJSON(StatsFile->os()); 949 950 return Result; 951 } 952 953 Error LTO::runRegularLTO(AddStreamFn AddStream) { 954 // Make sure commons have the right size/alignment: we kept the largest from 955 // all the prevailing when adding the inputs, and we apply it here. 956 const DataLayout &DL = RegularLTO.CombinedModule->getDataLayout(); 957 for (auto &I : RegularLTO.Commons) { 958 if (!I.second.Prevailing) 959 // Don't do anything if no instance of this common was prevailing. 960 continue; 961 GlobalVariable *OldGV = RegularLTO.CombinedModule->getNamedGlobal(I.first); 962 if (OldGV && DL.getTypeAllocSize(OldGV->getValueType()) == I.second.Size) { 963 // Don't create a new global if the type is already correct, just make 964 // sure the alignment is correct. 965 OldGV->setAlignment(I.second.Align); 966 continue; 967 } 968 ArrayType *Ty = 969 ArrayType::get(Type::getInt8Ty(RegularLTO.Ctx), I.second.Size); 970 auto *GV = new GlobalVariable(*RegularLTO.CombinedModule, Ty, false, 971 GlobalValue::CommonLinkage, 972 ConstantAggregateZero::get(Ty), ""); 973 GV->setAlignment(I.second.Align); 974 if (OldGV) { 975 OldGV->replaceAllUsesWith(ConstantExpr::getBitCast(GV, OldGV->getType())); 976 GV->takeName(OldGV); 977 OldGV->eraseFromParent(); 978 } else { 979 GV->setName(I.first); 980 } 981 } 982 983 if (Conf.PreOptModuleHook && 984 !Conf.PreOptModuleHook(0, *RegularLTO.CombinedModule)) 985 return Error::success(); 986 987 if (!Conf.CodeGenOnly) { 988 for (const auto &R : GlobalResolutions) { 989 if (!R.second.isPrevailingIRSymbol()) 990 continue; 991 if (R.second.Partition != 0 && 992 R.second.Partition != GlobalResolution::External) 993 continue; 994 995 GlobalValue *GV = 996 RegularLTO.CombinedModule->getNamedValue(R.second.IRName); 997 // Ignore symbols defined in other partitions. 998 // Also skip declarations, which are not allowed to have internal linkage. 999 if (!GV || GV->hasLocalLinkage() || GV->isDeclaration()) 1000 continue; 1001 GV->setUnnamedAddr(R.second.UnnamedAddr ? GlobalValue::UnnamedAddr::Global 1002 : GlobalValue::UnnamedAddr::None); 1003 if (EnableLTOInternalization && R.second.Partition == 0) 1004 GV->setLinkage(GlobalValue::InternalLinkage); 1005 } 1006 1007 RegularLTO.CombinedModule->addModuleFlag(Module::Error, "LTOPostLink", 1); 1008 1009 if (Conf.PostInternalizeModuleHook && 1010 !Conf.PostInternalizeModuleHook(0, *RegularLTO.CombinedModule)) 1011 return Error::success(); 1012 } 1013 return backend(Conf, AddStream, RegularLTO.ParallelCodeGenParallelismLevel, 1014 std::move(RegularLTO.CombinedModule), ThinLTO.CombinedIndex); 1015 } 1016 1017 static const char *libcallRoutineNames[] = { 1018 #define HANDLE_LIBCALL(code, name) name, 1019 #include "llvm/IR/RuntimeLibcalls.def" 1020 #undef HANDLE_LIBCALL 1021 }; 1022 1023 ArrayRef<const char*> LTO::getRuntimeLibcallSymbols() { 1024 return makeArrayRef(libcallRoutineNames); 1025 } 1026 1027 /// This class defines the interface to the ThinLTO backend. 1028 class lto::ThinBackendProc { 1029 protected: 1030 Config &Conf; 1031 ModuleSummaryIndex &CombinedIndex; 1032 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries; 1033 1034 public: 1035 ThinBackendProc(Config &Conf, ModuleSummaryIndex &CombinedIndex, 1036 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries) 1037 : Conf(Conf), CombinedIndex(CombinedIndex), 1038 ModuleToDefinedGVSummaries(ModuleToDefinedGVSummaries) {} 1039 1040 virtual ~ThinBackendProc() {} 1041 virtual Error start( 1042 unsigned Task, BitcodeModule BM, 1043 const FunctionImporter::ImportMapTy &ImportList, 1044 const FunctionImporter::ExportSetTy &ExportList, 1045 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 1046 MapVector<StringRef, BitcodeModule> &ModuleMap) = 0; 1047 virtual Error wait() = 0; 1048 }; 1049 1050 namespace { 1051 class InProcessThinBackend : public ThinBackendProc { 1052 ThreadPool BackendThreadPool; 1053 AddStreamFn AddStream; 1054 NativeObjectCache Cache; 1055 std::set<GlobalValue::GUID> CfiFunctionDefs; 1056 std::set<GlobalValue::GUID> CfiFunctionDecls; 1057 1058 Optional<Error> Err; 1059 std::mutex ErrMu; 1060 1061 public: 1062 InProcessThinBackend( 1063 Config &Conf, ModuleSummaryIndex &CombinedIndex, 1064 unsigned ThinLTOParallelismLevel, 1065 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries, 1066 AddStreamFn AddStream, NativeObjectCache Cache) 1067 : ThinBackendProc(Conf, CombinedIndex, ModuleToDefinedGVSummaries), 1068 BackendThreadPool(ThinLTOParallelismLevel), 1069 AddStream(std::move(AddStream)), Cache(std::move(Cache)) { 1070 for (auto &Name : CombinedIndex.cfiFunctionDefs()) 1071 CfiFunctionDefs.insert( 1072 GlobalValue::getGUID(GlobalValue::dropLLVMManglingEscape(Name))); 1073 for (auto &Name : CombinedIndex.cfiFunctionDecls()) 1074 CfiFunctionDecls.insert( 1075 GlobalValue::getGUID(GlobalValue::dropLLVMManglingEscape(Name))); 1076 } 1077 1078 Error runThinLTOBackendThread( 1079 AddStreamFn AddStream, NativeObjectCache Cache, unsigned Task, 1080 BitcodeModule BM, ModuleSummaryIndex &CombinedIndex, 1081 const FunctionImporter::ImportMapTy &ImportList, 1082 const FunctionImporter::ExportSetTy &ExportList, 1083 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 1084 const GVSummaryMapTy &DefinedGlobals, 1085 MapVector<StringRef, BitcodeModule> &ModuleMap) { 1086 auto RunThinBackend = [&](AddStreamFn AddStream) { 1087 LTOLLVMContext BackendContext(Conf); 1088 Expected<std::unique_ptr<Module>> MOrErr = BM.parseModule(BackendContext); 1089 if (!MOrErr) 1090 return MOrErr.takeError(); 1091 1092 return thinBackend(Conf, Task, AddStream, **MOrErr, CombinedIndex, 1093 ImportList, DefinedGlobals, ModuleMap); 1094 }; 1095 1096 auto ModuleID = BM.getModuleIdentifier(); 1097 1098 if (!Cache || !CombinedIndex.modulePaths().count(ModuleID) || 1099 all_of(CombinedIndex.getModuleHash(ModuleID), 1100 [](uint32_t V) { return V == 0; })) 1101 // Cache disabled or no entry for this module in the combined index or 1102 // no module hash. 1103 return RunThinBackend(AddStream); 1104 1105 SmallString<40> Key; 1106 // The module may be cached, this helps handling it. 1107 computeLTOCacheKey(Key, Conf, CombinedIndex, ModuleID, ImportList, 1108 ExportList, ResolvedODR, DefinedGlobals, CfiFunctionDefs, 1109 CfiFunctionDecls); 1110 if (AddStreamFn CacheAddStream = Cache(Task, Key)) 1111 return RunThinBackend(CacheAddStream); 1112 1113 return Error::success(); 1114 } 1115 1116 Error start( 1117 unsigned Task, BitcodeModule BM, 1118 const FunctionImporter::ImportMapTy &ImportList, 1119 const FunctionImporter::ExportSetTy &ExportList, 1120 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 1121 MapVector<StringRef, BitcodeModule> &ModuleMap) override { 1122 StringRef ModulePath = BM.getModuleIdentifier(); 1123 assert(ModuleToDefinedGVSummaries.count(ModulePath)); 1124 const GVSummaryMapTy &DefinedGlobals = 1125 ModuleToDefinedGVSummaries.find(ModulePath)->second; 1126 BackendThreadPool.async( 1127 [=](BitcodeModule BM, ModuleSummaryIndex &CombinedIndex, 1128 const FunctionImporter::ImportMapTy &ImportList, 1129 const FunctionImporter::ExportSetTy &ExportList, 1130 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> 1131 &ResolvedODR, 1132 const GVSummaryMapTy &DefinedGlobals, 1133 MapVector<StringRef, BitcodeModule> &ModuleMap) { 1134 Error E = runThinLTOBackendThread( 1135 AddStream, Cache, Task, BM, CombinedIndex, ImportList, ExportList, 1136 ResolvedODR, DefinedGlobals, ModuleMap); 1137 if (E) { 1138 std::unique_lock<std::mutex> L(ErrMu); 1139 if (Err) 1140 Err = joinErrors(std::move(*Err), std::move(E)); 1141 else 1142 Err = std::move(E); 1143 } 1144 }, 1145 BM, std::ref(CombinedIndex), std::ref(ImportList), std::ref(ExportList), 1146 std::ref(ResolvedODR), std::ref(DefinedGlobals), std::ref(ModuleMap)); 1147 return Error::success(); 1148 } 1149 1150 Error wait() override { 1151 BackendThreadPool.wait(); 1152 if (Err) 1153 return std::move(*Err); 1154 else 1155 return Error::success(); 1156 } 1157 }; 1158 } // end anonymous namespace 1159 1160 ThinBackend lto::createInProcessThinBackend(unsigned ParallelismLevel) { 1161 return [=](Config &Conf, ModuleSummaryIndex &CombinedIndex, 1162 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries, 1163 AddStreamFn AddStream, NativeObjectCache Cache) { 1164 return std::make_unique<InProcessThinBackend>( 1165 Conf, CombinedIndex, ParallelismLevel, ModuleToDefinedGVSummaries, 1166 AddStream, Cache); 1167 }; 1168 } 1169 1170 // Given the original \p Path to an output file, replace any path 1171 // prefix matching \p OldPrefix with \p NewPrefix. Also, create the 1172 // resulting directory if it does not yet exist. 1173 std::string lto::getThinLTOOutputFile(const std::string &Path, 1174 const std::string &OldPrefix, 1175 const std::string &NewPrefix) { 1176 if (OldPrefix.empty() && NewPrefix.empty()) 1177 return Path; 1178 SmallString<128> NewPath(Path); 1179 llvm::sys::path::replace_path_prefix(NewPath, OldPrefix, NewPrefix); 1180 StringRef ParentPath = llvm::sys::path::parent_path(NewPath.str()); 1181 if (!ParentPath.empty()) { 1182 // Make sure the new directory exists, creating it if necessary. 1183 if (std::error_code EC = llvm::sys::fs::create_directories(ParentPath)) 1184 llvm::errs() << "warning: could not create directory '" << ParentPath 1185 << "': " << EC.message() << '\n'; 1186 } 1187 return NewPath.str(); 1188 } 1189 1190 namespace { 1191 class WriteIndexesThinBackend : public ThinBackendProc { 1192 std::string OldPrefix, NewPrefix; 1193 bool ShouldEmitImportsFiles; 1194 raw_fd_ostream *LinkedObjectsFile; 1195 lto::IndexWriteCallback OnWrite; 1196 1197 public: 1198 WriteIndexesThinBackend( 1199 Config &Conf, ModuleSummaryIndex &CombinedIndex, 1200 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries, 1201 std::string OldPrefix, std::string NewPrefix, bool ShouldEmitImportsFiles, 1202 raw_fd_ostream *LinkedObjectsFile, lto::IndexWriteCallback OnWrite) 1203 : ThinBackendProc(Conf, CombinedIndex, ModuleToDefinedGVSummaries), 1204 OldPrefix(OldPrefix), NewPrefix(NewPrefix), 1205 ShouldEmitImportsFiles(ShouldEmitImportsFiles), 1206 LinkedObjectsFile(LinkedObjectsFile), OnWrite(OnWrite) {} 1207 1208 Error start( 1209 unsigned Task, BitcodeModule BM, 1210 const FunctionImporter::ImportMapTy &ImportList, 1211 const FunctionImporter::ExportSetTy &ExportList, 1212 const std::map<GlobalValue::GUID, GlobalValue::LinkageTypes> &ResolvedODR, 1213 MapVector<StringRef, BitcodeModule> &ModuleMap) override { 1214 StringRef ModulePath = BM.getModuleIdentifier(); 1215 std::string NewModulePath = 1216 getThinLTOOutputFile(ModulePath, OldPrefix, NewPrefix); 1217 1218 if (LinkedObjectsFile) 1219 *LinkedObjectsFile << NewModulePath << '\n'; 1220 1221 std::map<std::string, GVSummaryMapTy> ModuleToSummariesForIndex; 1222 gatherImportedSummariesForModule(ModulePath, ModuleToDefinedGVSummaries, 1223 ImportList, ModuleToSummariesForIndex); 1224 1225 std::error_code EC; 1226 raw_fd_ostream OS(NewModulePath + ".thinlto.bc", EC, 1227 sys::fs::OpenFlags::OF_None); 1228 if (EC) 1229 return errorCodeToError(EC); 1230 WriteIndexToFile(CombinedIndex, OS, &ModuleToSummariesForIndex); 1231 1232 if (ShouldEmitImportsFiles) { 1233 EC = EmitImportsFiles(ModulePath, NewModulePath + ".imports", 1234 ModuleToSummariesForIndex); 1235 if (EC) 1236 return errorCodeToError(EC); 1237 } 1238 1239 if (OnWrite) 1240 OnWrite(ModulePath); 1241 return Error::success(); 1242 } 1243 1244 Error wait() override { return Error::success(); } 1245 }; 1246 } // end anonymous namespace 1247 1248 ThinBackend lto::createWriteIndexesThinBackend( 1249 std::string OldPrefix, std::string NewPrefix, bool ShouldEmitImportsFiles, 1250 raw_fd_ostream *LinkedObjectsFile, IndexWriteCallback OnWrite) { 1251 return [=](Config &Conf, ModuleSummaryIndex &CombinedIndex, 1252 const StringMap<GVSummaryMapTy> &ModuleToDefinedGVSummaries, 1253 AddStreamFn AddStream, NativeObjectCache Cache) { 1254 return std::make_unique<WriteIndexesThinBackend>( 1255 Conf, CombinedIndex, ModuleToDefinedGVSummaries, OldPrefix, NewPrefix, 1256 ShouldEmitImportsFiles, LinkedObjectsFile, OnWrite); 1257 }; 1258 } 1259 1260 Error LTO::runThinLTO(AddStreamFn AddStream, NativeObjectCache Cache, 1261 const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols) { 1262 if (ThinLTO.ModuleMap.empty()) 1263 return Error::success(); 1264 1265 if (Conf.CombinedIndexHook && !Conf.CombinedIndexHook(ThinLTO.CombinedIndex)) 1266 return Error::success(); 1267 1268 // Collect for each module the list of function it defines (GUID -> 1269 // Summary). 1270 StringMap<GVSummaryMapTy> 1271 ModuleToDefinedGVSummaries(ThinLTO.ModuleMap.size()); 1272 ThinLTO.CombinedIndex.collectDefinedGVSummariesPerModule( 1273 ModuleToDefinedGVSummaries); 1274 // Create entries for any modules that didn't have any GV summaries 1275 // (either they didn't have any GVs to start with, or we suppressed 1276 // generation of the summaries because they e.g. had inline assembly 1277 // uses that couldn't be promoted/renamed on export). This is so 1278 // InProcessThinBackend::start can still launch a backend thread, which 1279 // is passed the map of summaries for the module, without any special 1280 // handling for this case. 1281 for (auto &Mod : ThinLTO.ModuleMap) 1282 if (!ModuleToDefinedGVSummaries.count(Mod.first)) 1283 ModuleToDefinedGVSummaries.try_emplace(Mod.first); 1284 1285 // Synthesize entry counts for functions in the CombinedIndex. 1286 computeSyntheticCounts(ThinLTO.CombinedIndex); 1287 1288 StringMap<FunctionImporter::ImportMapTy> ImportLists( 1289 ThinLTO.ModuleMap.size()); 1290 StringMap<FunctionImporter::ExportSetTy> ExportLists( 1291 ThinLTO.ModuleMap.size()); 1292 StringMap<std::map<GlobalValue::GUID, GlobalValue::LinkageTypes>> ResolvedODR; 1293 1294 if (DumpThinCGSCCs) 1295 ThinLTO.CombinedIndex.dumpSCCs(outs()); 1296 1297 std::set<GlobalValue::GUID> ExportedGUIDs; 1298 1299 // Perform index-based WPD. This will return immediately if there are 1300 // no index entries in the typeIdMetadata map (e.g. if we are instead 1301 // performing IR-based WPD in hybrid regular/thin LTO mode). 1302 std::map<ValueInfo, std::vector<VTableSlotSummary>> LocalWPDTargetsMap; 1303 runWholeProgramDevirtOnIndex(ThinLTO.CombinedIndex, ExportedGUIDs, 1304 LocalWPDTargetsMap); 1305 1306 if (Conf.OptLevel > 0) 1307 ComputeCrossModuleImport(ThinLTO.CombinedIndex, ModuleToDefinedGVSummaries, 1308 ImportLists, ExportLists); 1309 1310 // Figure out which symbols need to be internalized. This also needs to happen 1311 // at -O0 because summary-based DCE is implemented using internalization, and 1312 // we must apply DCE consistently with the full LTO module in order to avoid 1313 // undefined references during the final link. 1314 for (auto &Res : GlobalResolutions) { 1315 // If the symbol does not have external references or it is not prevailing, 1316 // then not need to mark it as exported from a ThinLTO partition. 1317 if (Res.second.Partition != GlobalResolution::External || 1318 !Res.second.isPrevailingIRSymbol()) 1319 continue; 1320 auto GUID = GlobalValue::getGUID( 1321 GlobalValue::dropLLVMManglingEscape(Res.second.IRName)); 1322 // Mark exported unless index-based analysis determined it to be dead. 1323 if (ThinLTO.CombinedIndex.isGUIDLive(GUID)) 1324 ExportedGUIDs.insert(GUID); 1325 } 1326 1327 // Any functions referenced by the jump table in the regular LTO object must 1328 // be exported. 1329 for (auto &Def : ThinLTO.CombinedIndex.cfiFunctionDefs()) 1330 ExportedGUIDs.insert( 1331 GlobalValue::getGUID(GlobalValue::dropLLVMManglingEscape(Def))); 1332 1333 auto isExported = [&](StringRef ModuleIdentifier, GlobalValue::GUID GUID) { 1334 const auto &ExportList = ExportLists.find(ModuleIdentifier); 1335 return (ExportList != ExportLists.end() && 1336 ExportList->second.count(GUID)) || 1337 ExportedGUIDs.count(GUID); 1338 }; 1339 1340 // Update local devirtualized targets that were exported by cross-module 1341 // importing or by other devirtualizations marked in the ExportedGUIDs set. 1342 updateIndexWPDForExports(ThinLTO.CombinedIndex, isExported, 1343 LocalWPDTargetsMap); 1344 1345 auto isPrevailing = [&](GlobalValue::GUID GUID, 1346 const GlobalValueSummary *S) { 1347 return ThinLTO.PrevailingModuleForGUID[GUID] == S->modulePath(); 1348 }; 1349 thinLTOInternalizeAndPromoteInIndex(ThinLTO.CombinedIndex, isExported, 1350 isPrevailing); 1351 1352 auto recordNewLinkage = [&](StringRef ModuleIdentifier, 1353 GlobalValue::GUID GUID, 1354 GlobalValue::LinkageTypes NewLinkage) { 1355 ResolvedODR[ModuleIdentifier][GUID] = NewLinkage; 1356 }; 1357 thinLTOResolvePrevailingInIndex(ThinLTO.CombinedIndex, isPrevailing, 1358 recordNewLinkage, GUIDPreservedSymbols); 1359 1360 std::unique_ptr<ThinBackendProc> BackendProc = 1361 ThinLTO.Backend(Conf, ThinLTO.CombinedIndex, ModuleToDefinedGVSummaries, 1362 AddStream, Cache); 1363 1364 // Tasks 0 through ParallelCodeGenParallelismLevel-1 are reserved for combined 1365 // module and parallel code generation partitions. 1366 unsigned Task = RegularLTO.ParallelCodeGenParallelismLevel; 1367 for (auto &Mod : ThinLTO.ModuleMap) { 1368 if (Error E = BackendProc->start(Task, Mod.second, ImportLists[Mod.first], 1369 ExportLists[Mod.first], 1370 ResolvedODR[Mod.first], ThinLTO.ModuleMap)) 1371 return E; 1372 ++Task; 1373 } 1374 1375 return BackendProc->wait(); 1376 } 1377 1378 Expected<std::unique_ptr<ToolOutputFile>> 1379 lto::setupOptimizationRemarks(LLVMContext &Context, StringRef RemarksFilename, 1380 StringRef RemarksPasses, StringRef RemarksFormat, 1381 bool RemarksWithHotness, int Count) { 1382 std::string Filename = RemarksFilename; 1383 if (!Filename.empty() && Count != -1) 1384 Filename += ".thin." + llvm::utostr(Count) + ".yaml"; 1385 1386 auto ResultOrErr = llvm::setupOptimizationRemarks( 1387 Context, Filename, RemarksPasses, RemarksFormat, RemarksWithHotness); 1388 if (Error E = ResultOrErr.takeError()) 1389 return std::move(E); 1390 1391 if (*ResultOrErr) 1392 (*ResultOrErr)->keep(); 1393 1394 return ResultOrErr; 1395 } 1396 1397 Expected<std::unique_ptr<ToolOutputFile>> 1398 lto::setupStatsFile(StringRef StatsFilename) { 1399 // Setup output file to emit statistics. 1400 if (StatsFilename.empty()) 1401 return nullptr; 1402 1403 llvm::EnableStatistics(false); 1404 std::error_code EC; 1405 auto StatsFile = 1406 std::make_unique<ToolOutputFile>(StatsFilename, EC, sys::fs::OF_None); 1407 if (EC) 1408 return errorCodeToError(EC); 1409 1410 StatsFile->keep(); 1411 return std::move(StatsFile); 1412 } 1413