1 //===-LTOBackend.cpp - LLVM Link Time Optimizer Backend -------------------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // This file implements the "backend" phase of LTO, i.e. it performs
10 // optimization and code generation on a loaded module. It is generally used
11 // internally by the LTO class but can also be used independently, for example
12 // to implement a standalone ThinLTO backend.
13 //
14 //===----------------------------------------------------------------------===//
15 
16 #include "llvm/LTO/LTOBackend.h"
17 #include "llvm/Analysis/AliasAnalysis.h"
18 #include "llvm/Analysis/CGSCCPassManager.h"
19 #include "llvm/Analysis/ModuleSummaryAnalysis.h"
20 #include "llvm/Analysis/TargetLibraryInfo.h"
21 #include "llvm/Analysis/TargetTransformInfo.h"
22 #include "llvm/Bitcode/BitcodeReader.h"
23 #include "llvm/Bitcode/BitcodeWriter.h"
24 #include "llvm/IR/LLVMRemarkStreamer.h"
25 #include "llvm/IR/LegacyPassManager.h"
26 #include "llvm/IR/PassManager.h"
27 #include "llvm/IR/Verifier.h"
28 #include "llvm/LTO/LTO.h"
29 #include "llvm/MC/SubtargetFeature.h"
30 #include "llvm/MC/TargetRegistry.h"
31 #include "llvm/Object/ModuleSymbolTable.h"
32 #include "llvm/Passes/PassBuilder.h"
33 #include "llvm/Passes/PassPlugin.h"
34 #include "llvm/Passes/StandardInstrumentations.h"
35 #include "llvm/Support/Error.h"
36 #include "llvm/Support/FileSystem.h"
37 #include "llvm/Support/MemoryBuffer.h"
38 #include "llvm/Support/Path.h"
39 #include "llvm/Support/Program.h"
40 #include "llvm/Support/ThreadPool.h"
41 #include "llvm/Support/raw_ostream.h"
42 #include "llvm/Target/TargetMachine.h"
43 #include "llvm/Transforms/IPO.h"
44 #include "llvm/Transforms/IPO/PassManagerBuilder.h"
45 #include "llvm/Transforms/Scalar/LoopPassManager.h"
46 #include "llvm/Transforms/Utils/FunctionImportUtils.h"
47 #include "llvm/Transforms/Utils/SplitModule.h"
48 
49 using namespace llvm;
50 using namespace lto;
51 
52 #define DEBUG_TYPE "lto-backend"
53 
54 enum class LTOBitcodeEmbedding {
55   DoNotEmbed = 0,
56   EmbedOptimized = 1,
57   EmbedPostMergePreOptimized = 2
58 };
59 
60 static cl::opt<LTOBitcodeEmbedding> EmbedBitcode(
61     "lto-embed-bitcode", cl::init(LTOBitcodeEmbedding::DoNotEmbed),
62     cl::values(clEnumValN(LTOBitcodeEmbedding::DoNotEmbed, "none",
63                           "Do not embed"),
64                clEnumValN(LTOBitcodeEmbedding::EmbedOptimized, "optimized",
65                           "Embed after all optimization passes"),
66                clEnumValN(LTOBitcodeEmbedding::EmbedPostMergePreOptimized,
67                           "post-merge-pre-opt",
68                           "Embed post merge, but before optimizations")),
69     cl::desc("Embed LLVM bitcode in object files produced by LTO"));
70 
71 static cl::opt<bool> ThinLTOAssumeMerged(
72     "thinlto-assume-merged", cl::init(false),
73     cl::desc("Assume the input has already undergone ThinLTO function "
74              "importing and the other pre-optimization pipeline changes."));
75 
76 namespace llvm {
77 extern cl::opt<bool> NoPGOWarnMismatch;
78 }
79 
80 [[noreturn]] static void reportOpenError(StringRef Path, Twine Msg) {
81   errs() << "failed to open " << Path << ": " << Msg << '\n';
82   errs().flush();
83   exit(1);
84 }
85 
86 Error Config::addSaveTemps(std::string OutputFileName,
87                            bool UseInputModulePath) {
88   ShouldDiscardValueNames = false;
89 
90   std::error_code EC;
91   ResolutionFile =
92       std::make_unique<raw_fd_ostream>(OutputFileName + "resolution.txt", EC,
93                                        sys::fs::OpenFlags::OF_TextWithCRLF);
94   if (EC) {
95     ResolutionFile.reset();
96     return errorCodeToError(EC);
97   }
98 
99   auto setHook = [&](std::string PathSuffix, ModuleHookFn &Hook) {
100     // Keep track of the hook provided by the linker, which also needs to run.
101     ModuleHookFn LinkerHook = Hook;
102     Hook = [=](unsigned Task, const Module &M) {
103       // If the linker's hook returned false, we need to pass that result
104       // through.
105       if (LinkerHook && !LinkerHook(Task, M))
106         return false;
107 
108       std::string PathPrefix;
109       // If this is the combined module (not a ThinLTO backend compile) or the
110       // user hasn't requested using the input module's path, emit to a file
111       // named from the provided OutputFileName with the Task ID appended.
112       if (M.getModuleIdentifier() == "ld-temp.o" || !UseInputModulePath) {
113         PathPrefix = OutputFileName;
114         if (Task != (unsigned)-1)
115           PathPrefix += utostr(Task) + ".";
116       } else
117         PathPrefix = M.getModuleIdentifier() + ".";
118       std::string Path = PathPrefix + PathSuffix + ".bc";
119       std::error_code EC;
120       raw_fd_ostream OS(Path, EC, sys::fs::OpenFlags::OF_None);
121       // Because -save-temps is a debugging feature, we report the error
122       // directly and exit.
123       if (EC)
124         reportOpenError(Path, EC.message());
125       WriteBitcodeToFile(M, OS, /*ShouldPreserveUseListOrder=*/false);
126       return true;
127     };
128   };
129 
130   setHook("0.preopt", PreOptModuleHook);
131   setHook("1.promote", PostPromoteModuleHook);
132   setHook("2.internalize", PostInternalizeModuleHook);
133   setHook("3.import", PostImportModuleHook);
134   setHook("4.opt", PostOptModuleHook);
135   setHook("5.precodegen", PreCodeGenModuleHook);
136 
137   CombinedIndexHook =
138       [=](const ModuleSummaryIndex &Index,
139           const DenseSet<GlobalValue::GUID> &GUIDPreservedSymbols) {
140         std::string Path = OutputFileName + "index.bc";
141         std::error_code EC;
142         raw_fd_ostream OS(Path, EC, sys::fs::OpenFlags::OF_None);
143         // Because -save-temps is a debugging feature, we report the error
144         // directly and exit.
145         if (EC)
146           reportOpenError(Path, EC.message());
147         WriteIndexToFile(Index, OS);
148 
149         Path = OutputFileName + "index.dot";
150         raw_fd_ostream OSDot(Path, EC, sys::fs::OpenFlags::OF_None);
151         if (EC)
152           reportOpenError(Path, EC.message());
153         Index.exportToDot(OSDot, GUIDPreservedSymbols);
154         return true;
155       };
156 
157   return Error::success();
158 }
159 
160 #define HANDLE_EXTENSION(Ext)                                                  \
161   llvm::PassPluginLibraryInfo get##Ext##PluginInfo();
162 #include "llvm/Support/Extension.def"
163 
164 static void RegisterPassPlugins(ArrayRef<std::string> PassPlugins,
165                                 PassBuilder &PB) {
166 #define HANDLE_EXTENSION(Ext)                                                  \
167   get##Ext##PluginInfo().RegisterPassBuilderCallbacks(PB);
168 #include "llvm/Support/Extension.def"
169 
170   // Load requested pass plugins and let them register pass builder callbacks
171   for (auto &PluginFN : PassPlugins) {
172     auto PassPlugin = PassPlugin::Load(PluginFN);
173     if (!PassPlugin) {
174       errs() << "Failed to load passes from '" << PluginFN
175              << "'. Request ignored.\n";
176       continue;
177     }
178 
179     PassPlugin->registerPassBuilderCallbacks(PB);
180   }
181 }
182 
183 static std::unique_ptr<TargetMachine>
184 createTargetMachine(const Config &Conf, const Target *TheTarget, Module &M) {
185   StringRef TheTriple = M.getTargetTriple();
186   SubtargetFeatures Features;
187   Features.getDefaultSubtargetFeatures(Triple(TheTriple));
188   for (const std::string &A : Conf.MAttrs)
189     Features.AddFeature(A);
190 
191   Optional<Reloc::Model> RelocModel = None;
192   if (Conf.RelocModel)
193     RelocModel = *Conf.RelocModel;
194   else if (M.getModuleFlag("PIC Level"))
195     RelocModel =
196         M.getPICLevel() == PICLevel::NotPIC ? Reloc::Static : Reloc::PIC_;
197 
198   Optional<CodeModel::Model> CodeModel;
199   if (Conf.CodeModel)
200     CodeModel = *Conf.CodeModel;
201   else
202     CodeModel = M.getCodeModel();
203 
204   std::unique_ptr<TargetMachine> TM(TheTarget->createTargetMachine(
205       TheTriple, Conf.CPU, Features.getString(), Conf.Options, RelocModel,
206       CodeModel, Conf.CGOptLevel));
207   assert(TM && "Failed to create target machine");
208   return TM;
209 }
210 
211 static void runNewPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
212                            unsigned OptLevel, bool IsThinLTO,
213                            ModuleSummaryIndex *ExportSummary,
214                            const ModuleSummaryIndex *ImportSummary) {
215   Optional<PGOOptions> PGOOpt;
216   if (!Conf.SampleProfile.empty())
217     PGOOpt = PGOOptions(Conf.SampleProfile, "", Conf.ProfileRemapping,
218                         PGOOptions::SampleUse, PGOOptions::NoCSAction, true);
219   else if (Conf.RunCSIRInstr) {
220     PGOOpt = PGOOptions("", Conf.CSIRProfile, Conf.ProfileRemapping,
221                         PGOOptions::IRUse, PGOOptions::CSIRInstr,
222                         Conf.AddFSDiscriminator);
223   } else if (!Conf.CSIRProfile.empty()) {
224     PGOOpt = PGOOptions(Conf.CSIRProfile, "", Conf.ProfileRemapping,
225                         PGOOptions::IRUse, PGOOptions::CSIRUse,
226                         Conf.AddFSDiscriminator);
227     NoPGOWarnMismatch = !Conf.PGOWarnMismatch;
228   } else if (Conf.AddFSDiscriminator) {
229     PGOOpt = PGOOptions("", "", "", PGOOptions::NoAction,
230                         PGOOptions::NoCSAction, true);
231   }
232   if (TM)
233     TM->setPGOOption(PGOOpt);
234 
235   LoopAnalysisManager LAM;
236   FunctionAnalysisManager FAM;
237   CGSCCAnalysisManager CGAM;
238   ModuleAnalysisManager MAM;
239 
240   PassInstrumentationCallbacks PIC;
241   StandardInstrumentations SI(Conf.DebugPassManager);
242   SI.registerCallbacks(PIC, &FAM);
243   PassBuilder PB(TM, Conf.PTO, PGOOpt, &PIC);
244 
245   RegisterPassPlugins(Conf.PassPlugins, PB);
246 
247   std::unique_ptr<TargetLibraryInfoImpl> TLII(
248       new TargetLibraryInfoImpl(Triple(TM->getTargetTriple())));
249   if (Conf.Freestanding)
250     TLII->disableAllFunctions();
251   FAM.registerPass([&] { return TargetLibraryAnalysis(*TLII); });
252 
253   // Parse a custom AA pipeline if asked to.
254   if (!Conf.AAPipeline.empty()) {
255     AAManager AA;
256     if (auto Err = PB.parseAAPipeline(AA, Conf.AAPipeline)) {
257       report_fatal_error(Twine("unable to parse AA pipeline description '") +
258                          Conf.AAPipeline + "': " + toString(std::move(Err)));
259     }
260     // Register the AA manager first so that our version is the one used.
261     FAM.registerPass([&] { return std::move(AA); });
262   }
263 
264   // Register all the basic analyses with the managers.
265   PB.registerModuleAnalyses(MAM);
266   PB.registerCGSCCAnalyses(CGAM);
267   PB.registerFunctionAnalyses(FAM);
268   PB.registerLoopAnalyses(LAM);
269   PB.crossRegisterProxies(LAM, FAM, CGAM, MAM);
270 
271   ModulePassManager MPM;
272 
273   if (!Conf.DisableVerify)
274     MPM.addPass(VerifierPass());
275 
276   OptimizationLevel OL;
277 
278   switch (OptLevel) {
279   default:
280     llvm_unreachable("Invalid optimization level");
281   case 0:
282     OL = OptimizationLevel::O0;
283     break;
284   case 1:
285     OL = OptimizationLevel::O1;
286     break;
287   case 2:
288     OL = OptimizationLevel::O2;
289     break;
290   case 3:
291     OL = OptimizationLevel::O3;
292     break;
293   }
294 
295   // Parse a custom pipeline if asked to.
296   if (!Conf.OptPipeline.empty()) {
297     if (auto Err = PB.parsePassPipeline(MPM, Conf.OptPipeline)) {
298       report_fatal_error(Twine("unable to parse pass pipeline description '") +
299                          Conf.OptPipeline + "': " + toString(std::move(Err)));
300     }
301   } else if (IsThinLTO) {
302     MPM.addPass(PB.buildThinLTODefaultPipeline(OL, ImportSummary));
303   } else {
304     MPM.addPass(PB.buildLTODefaultPipeline(OL, ExportSummary));
305   }
306 
307   if (!Conf.DisableVerify)
308     MPM.addPass(VerifierPass());
309 
310   MPM.run(Mod, MAM);
311 }
312 
313 static void runOldPMPasses(const Config &Conf, Module &Mod, TargetMachine *TM,
314                            bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
315                            const ModuleSummaryIndex *ImportSummary) {
316   legacy::PassManager passes;
317   passes.add(createTargetTransformInfoWrapperPass(TM->getTargetIRAnalysis()));
318 
319   PassManagerBuilder PMB;
320   PMB.LibraryInfo = new TargetLibraryInfoImpl(Triple(TM->getTargetTriple()));
321   if (Conf.Freestanding)
322     PMB.LibraryInfo->disableAllFunctions();
323   PMB.Inliner = createFunctionInliningPass();
324   PMB.ExportSummary = ExportSummary;
325   PMB.ImportSummary = ImportSummary;
326   // Unconditionally verify input since it is not verified before this
327   // point and has unknown origin.
328   PMB.VerifyInput = true;
329   PMB.VerifyOutput = !Conf.DisableVerify;
330   PMB.LoopVectorize = true;
331   PMB.SLPVectorize = true;
332   PMB.OptLevel = Conf.OptLevel;
333   PMB.PGOSampleUse = Conf.SampleProfile;
334   PMB.EnablePGOCSInstrGen = Conf.RunCSIRInstr;
335   if (!Conf.RunCSIRInstr && !Conf.CSIRProfile.empty()) {
336     PMB.EnablePGOCSInstrUse = true;
337     PMB.PGOInstrUse = Conf.CSIRProfile;
338   }
339   if (IsThinLTO)
340     PMB.populateThinLTOPassManager(passes);
341   else
342     PMB.populateLTOPassManager(passes);
343   passes.run(Mod);
344 }
345 
346 bool lto::opt(const Config &Conf, TargetMachine *TM, unsigned Task, Module &Mod,
347               bool IsThinLTO, ModuleSummaryIndex *ExportSummary,
348               const ModuleSummaryIndex *ImportSummary,
349               const std::vector<uint8_t> &CmdArgs) {
350   if (EmbedBitcode == LTOBitcodeEmbedding::EmbedPostMergePreOptimized) {
351     // FIXME: the motivation for capturing post-merge bitcode and command line
352     // is replicating the compilation environment from bitcode, without needing
353     // to understand the dependencies (the functions to be imported). This
354     // assumes a clang - based invocation, case in which we have the command
355     // line.
356     // It's not very clear how the above motivation would map in the
357     // linker-based case, so we currently don't plumb the command line args in
358     // that case.
359     if (CmdArgs.empty())
360       LLVM_DEBUG(
361           dbgs() << "Post-(Thin)LTO merge bitcode embedding was requested, but "
362                     "command line arguments are not available");
363     llvm::EmbedBitcodeInModule(Mod, llvm::MemoryBufferRef(),
364                                /*EmbedBitcode*/ true, /*EmbedCmdline*/ true,
365                                /*Cmdline*/ CmdArgs);
366   }
367   // FIXME: Plumb the combined index into the new pass manager.
368   if (Conf.UseNewPM || !Conf.OptPipeline.empty()) {
369     runNewPMPasses(Conf, Mod, TM, Conf.OptLevel, IsThinLTO, ExportSummary,
370                    ImportSummary);
371   } else {
372     runOldPMPasses(Conf, Mod, TM, IsThinLTO, ExportSummary, ImportSummary);
373   }
374   return !Conf.PostOptModuleHook || Conf.PostOptModuleHook(Task, Mod);
375 }
376 
377 static void codegen(const Config &Conf, TargetMachine *TM,
378                     AddStreamFn AddStream, unsigned Task, Module &Mod,
379                     const ModuleSummaryIndex &CombinedIndex) {
380   if (Conf.PreCodeGenModuleHook && !Conf.PreCodeGenModuleHook(Task, Mod))
381     return;
382 
383   if (EmbedBitcode == LTOBitcodeEmbedding::EmbedOptimized)
384     llvm::EmbedBitcodeInModule(Mod, llvm::MemoryBufferRef(),
385                                /*EmbedBitcode*/ true,
386                                /*EmbedCmdline*/ false,
387                                /*CmdArgs*/ std::vector<uint8_t>());
388 
389   std::unique_ptr<ToolOutputFile> DwoOut;
390   SmallString<1024> DwoFile(Conf.SplitDwarfOutput);
391   if (!Conf.DwoDir.empty()) {
392     std::error_code EC;
393     if (auto EC = llvm::sys::fs::create_directories(Conf.DwoDir))
394       report_fatal_error(Twine("Failed to create directory ") + Conf.DwoDir +
395                          ": " + EC.message());
396 
397     DwoFile = Conf.DwoDir;
398     sys::path::append(DwoFile, std::to_string(Task) + ".dwo");
399     TM->Options.MCOptions.SplitDwarfFile = std::string(DwoFile);
400   } else
401     TM->Options.MCOptions.SplitDwarfFile = Conf.SplitDwarfFile;
402 
403   if (!DwoFile.empty()) {
404     std::error_code EC;
405     DwoOut = std::make_unique<ToolOutputFile>(DwoFile, EC, sys::fs::OF_None);
406     if (EC)
407       report_fatal_error(Twine("Failed to open ") + DwoFile + ": " +
408                          EC.message());
409   }
410 
411   Expected<std::unique_ptr<CachedFileStream>> StreamOrErr = AddStream(Task);
412   if (Error Err = StreamOrErr.takeError())
413     report_fatal_error(std::move(Err));
414   std::unique_ptr<CachedFileStream> &Stream = *StreamOrErr;
415   TM->Options.ObjectFilenameForDebug = Stream->ObjectPathName;
416 
417   legacy::PassManager CodeGenPasses;
418   TargetLibraryInfoImpl TLII(Triple(Mod.getTargetTriple()));
419   CodeGenPasses.add(new TargetLibraryInfoWrapperPass(TLII));
420   CodeGenPasses.add(
421       createImmutableModuleSummaryIndexWrapperPass(&CombinedIndex));
422   if (Conf.PreCodeGenPassesHook)
423     Conf.PreCodeGenPassesHook(CodeGenPasses);
424   if (TM->addPassesToEmitFile(CodeGenPasses, *Stream->OS,
425                               DwoOut ? &DwoOut->os() : nullptr,
426                               Conf.CGFileType))
427     report_fatal_error("Failed to setup codegen");
428   CodeGenPasses.run(Mod);
429 
430   if (DwoOut)
431     DwoOut->keep();
432 }
433 
434 static void splitCodeGen(const Config &C, TargetMachine *TM,
435                          AddStreamFn AddStream,
436                          unsigned ParallelCodeGenParallelismLevel, Module &Mod,
437                          const ModuleSummaryIndex &CombinedIndex) {
438   ThreadPool CodegenThreadPool(
439       heavyweight_hardware_concurrency(ParallelCodeGenParallelismLevel));
440   unsigned ThreadCount = 0;
441   const Target *T = &TM->getTarget();
442 
443   SplitModule(
444       Mod, ParallelCodeGenParallelismLevel,
445       [&](std::unique_ptr<Module> MPart) {
446         // We want to clone the module in a new context to multi-thread the
447         // codegen. We do it by serializing partition modules to bitcode
448         // (while still on the main thread, in order to avoid data races) and
449         // spinning up new threads which deserialize the partitions into
450         // separate contexts.
451         // FIXME: Provide a more direct way to do this in LLVM.
452         SmallString<0> BC;
453         raw_svector_ostream BCOS(BC);
454         WriteBitcodeToFile(*MPart, BCOS);
455 
456         // Enqueue the task
457         CodegenThreadPool.async(
458             [&](const SmallString<0> &BC, unsigned ThreadId) {
459               LTOLLVMContext Ctx(C);
460               Expected<std::unique_ptr<Module>> MOrErr = parseBitcodeFile(
461                   MemoryBufferRef(StringRef(BC.data(), BC.size()), "ld-temp.o"),
462                   Ctx);
463               if (!MOrErr)
464                 report_fatal_error("Failed to read bitcode");
465               std::unique_ptr<Module> MPartInCtx = std::move(MOrErr.get());
466 
467               std::unique_ptr<TargetMachine> TM =
468                   createTargetMachine(C, T, *MPartInCtx);
469 
470               codegen(C, TM.get(), AddStream, ThreadId, *MPartInCtx,
471                       CombinedIndex);
472             },
473             // Pass BC using std::move to ensure that it get moved rather than
474             // copied into the thread's context.
475             std::move(BC), ThreadCount++);
476       },
477       false);
478 
479   // Because the inner lambda (which runs in a worker thread) captures our local
480   // variables, we need to wait for the worker threads to terminate before we
481   // can leave the function scope.
482   CodegenThreadPool.wait();
483 }
484 
485 static Expected<const Target *> initAndLookupTarget(const Config &C,
486                                                     Module &Mod) {
487   if (!C.OverrideTriple.empty())
488     Mod.setTargetTriple(C.OverrideTriple);
489   else if (Mod.getTargetTriple().empty())
490     Mod.setTargetTriple(C.DefaultTriple);
491 
492   std::string Msg;
493   const Target *T = TargetRegistry::lookupTarget(Mod.getTargetTriple(), Msg);
494   if (!T)
495     return make_error<StringError>(Msg, inconvertibleErrorCode());
496   return T;
497 }
498 
499 Error lto::finalizeOptimizationRemarks(
500     std::unique_ptr<ToolOutputFile> DiagOutputFile) {
501   // Make sure we flush the diagnostic remarks file in case the linker doesn't
502   // call the global destructors before exiting.
503   if (!DiagOutputFile)
504     return Error::success();
505   DiagOutputFile->keep();
506   DiagOutputFile->os().flush();
507   return Error::success();
508 }
509 
510 Error lto::backend(const Config &C, AddStreamFn AddStream,
511                    unsigned ParallelCodeGenParallelismLevel, Module &Mod,
512                    ModuleSummaryIndex &CombinedIndex) {
513   Expected<const Target *> TOrErr = initAndLookupTarget(C, Mod);
514   if (!TOrErr)
515     return TOrErr.takeError();
516 
517   std::unique_ptr<TargetMachine> TM = createTargetMachine(C, *TOrErr, Mod);
518 
519   if (!C.CodeGenOnly) {
520     if (!opt(C, TM.get(), 0, Mod, /*IsThinLTO=*/false,
521              /*ExportSummary=*/&CombinedIndex, /*ImportSummary=*/nullptr,
522              /*CmdArgs*/ std::vector<uint8_t>()))
523       return Error::success();
524   }
525 
526   if (ParallelCodeGenParallelismLevel == 1) {
527     codegen(C, TM.get(), AddStream, 0, Mod, CombinedIndex);
528   } else {
529     splitCodeGen(C, TM.get(), AddStream, ParallelCodeGenParallelismLevel, Mod,
530                  CombinedIndex);
531   }
532   return Error::success();
533 }
534 
535 static void dropDeadSymbols(Module &Mod, const GVSummaryMapTy &DefinedGlobals,
536                             const ModuleSummaryIndex &Index) {
537   std::vector<GlobalValue*> DeadGVs;
538   for (auto &GV : Mod.global_values())
539     if (GlobalValueSummary *GVS = DefinedGlobals.lookup(GV.getGUID()))
540       if (!Index.isGlobalValueLive(GVS)) {
541         DeadGVs.push_back(&GV);
542         convertToDeclaration(GV);
543       }
544 
545   // Now that all dead bodies have been dropped, delete the actual objects
546   // themselves when possible.
547   for (GlobalValue *GV : DeadGVs) {
548     GV->removeDeadConstantUsers();
549     // Might reference something defined in native object (i.e. dropped a
550     // non-prevailing IR def, but we need to keep the declaration).
551     if (GV->use_empty())
552       GV->eraseFromParent();
553   }
554 }
555 
556 Error lto::thinBackend(const Config &Conf, unsigned Task, AddStreamFn AddStream,
557                        Module &Mod, const ModuleSummaryIndex &CombinedIndex,
558                        const FunctionImporter::ImportMapTy &ImportList,
559                        const GVSummaryMapTy &DefinedGlobals,
560                        MapVector<StringRef, BitcodeModule> *ModuleMap,
561                        const std::vector<uint8_t> &CmdArgs) {
562   Expected<const Target *> TOrErr = initAndLookupTarget(Conf, Mod);
563   if (!TOrErr)
564     return TOrErr.takeError();
565 
566   std::unique_ptr<TargetMachine> TM = createTargetMachine(Conf, *TOrErr, Mod);
567 
568   // Setup optimization remarks.
569   auto DiagFileOrErr = lto::setupLLVMOptimizationRemarks(
570       Mod.getContext(), Conf.RemarksFilename, Conf.RemarksPasses,
571       Conf.RemarksFormat, Conf.RemarksWithHotness, Conf.RemarksHotnessThreshold,
572       Task);
573   if (!DiagFileOrErr)
574     return DiagFileOrErr.takeError();
575   auto DiagnosticOutputFile = std::move(*DiagFileOrErr);
576 
577   // Set the partial sample profile ratio in the profile summary module flag of
578   // the module, if applicable.
579   Mod.setPartialSampleProfileRatio(CombinedIndex);
580 
581   if (Conf.CodeGenOnly) {
582     codegen(Conf, TM.get(), AddStream, Task, Mod, CombinedIndex);
583     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
584   }
585 
586   if (Conf.PreOptModuleHook && !Conf.PreOptModuleHook(Task, Mod))
587     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
588 
589   auto OptimizeAndCodegen =
590       [&](Module &Mod, TargetMachine *TM,
591           std::unique_ptr<ToolOutputFile> DiagnosticOutputFile) {
592         if (!opt(Conf, TM, Task, Mod, /*IsThinLTO=*/true,
593                  /*ExportSummary=*/nullptr, /*ImportSummary=*/&CombinedIndex,
594                  CmdArgs))
595           return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
596 
597         codegen(Conf, TM, AddStream, Task, Mod, CombinedIndex);
598         return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
599       };
600 
601   if (ThinLTOAssumeMerged)
602     return OptimizeAndCodegen(Mod, TM.get(), std::move(DiagnosticOutputFile));
603 
604   // When linking an ELF shared object, dso_local should be dropped. We
605   // conservatively do this for -fpic.
606   bool ClearDSOLocalOnDeclarations =
607       TM->getTargetTriple().isOSBinFormatELF() &&
608       TM->getRelocationModel() != Reloc::Static &&
609       Mod.getPIELevel() == PIELevel::Default;
610   renameModuleForThinLTO(Mod, CombinedIndex, ClearDSOLocalOnDeclarations);
611 
612   dropDeadSymbols(Mod, DefinedGlobals, CombinedIndex);
613 
614   thinLTOFinalizeInModule(Mod, DefinedGlobals, /*PropagateAttrs=*/true);
615 
616   if (Conf.PostPromoteModuleHook && !Conf.PostPromoteModuleHook(Task, Mod))
617     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
618 
619   if (!DefinedGlobals.empty())
620     thinLTOInternalizeModule(Mod, DefinedGlobals);
621 
622   if (Conf.PostInternalizeModuleHook &&
623       !Conf.PostInternalizeModuleHook(Task, Mod))
624     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
625 
626   auto ModuleLoader = [&](StringRef Identifier) {
627     assert(Mod.getContext().isODRUniquingDebugTypes() &&
628            "ODR Type uniquing should be enabled on the context");
629     if (ModuleMap) {
630       auto I = ModuleMap->find(Identifier);
631       assert(I != ModuleMap->end());
632       return I->second.getLazyModule(Mod.getContext(),
633                                      /*ShouldLazyLoadMetadata=*/true,
634                                      /*IsImporting*/ true);
635     }
636 
637     ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> MBOrErr =
638         llvm::MemoryBuffer::getFile(Identifier);
639     if (!MBOrErr)
640       return Expected<std::unique_ptr<llvm::Module>>(make_error<StringError>(
641           Twine("Error loading imported file ") + Identifier + " : ",
642           MBOrErr.getError()));
643 
644     Expected<BitcodeModule> BMOrErr = findThinLTOModule(**MBOrErr);
645     if (!BMOrErr)
646       return Expected<std::unique_ptr<llvm::Module>>(make_error<StringError>(
647           Twine("Error loading imported file ") + Identifier + " : " +
648               toString(BMOrErr.takeError()),
649           inconvertibleErrorCode()));
650 
651     Expected<std::unique_ptr<Module>> MOrErr =
652         BMOrErr->getLazyModule(Mod.getContext(),
653                                /*ShouldLazyLoadMetadata=*/true,
654                                /*IsImporting*/ true);
655     if (MOrErr)
656       (*MOrErr)->setOwnedMemoryBuffer(std::move(*MBOrErr));
657     return MOrErr;
658   };
659 
660   FunctionImporter Importer(CombinedIndex, ModuleLoader,
661                             ClearDSOLocalOnDeclarations);
662   if (Error Err = Importer.importFunctions(Mod, ImportList).takeError())
663     return Err;
664 
665   if (Conf.PostImportModuleHook && !Conf.PostImportModuleHook(Task, Mod))
666     return finalizeOptimizationRemarks(std::move(DiagnosticOutputFile));
667 
668   return OptimizeAndCodegen(Mod, TM.get(), std::move(DiagnosticOutputFile));
669 }
670 
671 BitcodeModule *lto::findThinLTOModule(MutableArrayRef<BitcodeModule> BMs) {
672   if (ThinLTOAssumeMerged && BMs.size() == 1)
673     return BMs.begin();
674 
675   for (BitcodeModule &BM : BMs) {
676     Expected<BitcodeLTOInfo> LTOInfo = BM.getLTOInfo();
677     if (LTOInfo && LTOInfo->IsThinLTO)
678       return &BM;
679   }
680   return nullptr;
681 }
682 
683 Expected<BitcodeModule> lto::findThinLTOModule(MemoryBufferRef MBRef) {
684   Expected<std::vector<BitcodeModule>> BMsOrErr = getBitcodeModuleList(MBRef);
685   if (!BMsOrErr)
686     return BMsOrErr.takeError();
687 
688   // The bitcode file may contain multiple modules, we want the one that is
689   // marked as being the ThinLTO module.
690   if (const BitcodeModule *Bm = lto::findThinLTOModule(*BMsOrErr))
691     return *Bm;
692 
693   return make_error<StringError>("Could not find module summary",
694                                  inconvertibleErrorCode());
695 }
696 
697 bool lto::initImportList(const Module &M,
698                          const ModuleSummaryIndex &CombinedIndex,
699                          FunctionImporter::ImportMapTy &ImportList) {
700   if (ThinLTOAssumeMerged)
701     return true;
702   // We can simply import the values mentioned in the combined index, since
703   // we should only invoke this using the individual indexes written out
704   // via a WriteIndexesThinBackend.
705   for (const auto &GlobalList : CombinedIndex) {
706     // Ignore entries for undefined references.
707     if (GlobalList.second.SummaryList.empty())
708       continue;
709 
710     auto GUID = GlobalList.first;
711     for (const auto &Summary : GlobalList.second.SummaryList) {
712       // Skip the summaries for the importing module. These are included to
713       // e.g. record required linkage changes.
714       if (Summary->modulePath() == M.getModuleIdentifier())
715         continue;
716       // Add an entry to provoke importing by thinBackend.
717       ImportList[Summary->modulePath()].insert(GUID);
718     }
719   }
720   return true;
721 }
722