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