tools/opt/opt.cpp

   1 //===- opt.cpp - The LLVM Modular 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 // Optimizations may be specified an arbitrary number of times on the command
  10 // line, They are run in the order specified.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 #include "BreakpointPrinter.h"
  15 #include "Debugify.h"
  16 #include "NewPMDriver.h"
  17 #include "PassPrinters.h"
  18 #include "llvm/ADT/Triple.h"
  19 #include "llvm/Analysis/CallGraph.h"
  20 #include "llvm/Analysis/CallGraphSCCPass.h"
  21 #include "llvm/Analysis/LoopPass.h"
  22 #include "llvm/Analysis/RegionPass.h"
  23 #include "llvm/Analysis/TargetLibraryInfo.h"
  24 #include "llvm/Analysis/TargetTransformInfo.h"
  25 #include "llvm/Bitcode/BitcodeWriterPass.h"
  26 #include "llvm/CodeGen/CommandFlags.inc"
  27 #include "llvm/CodeGen/TargetPassConfig.h"
  28 #include "llvm/Config/llvm-config.h"
  29 #include "llvm/IR/DataLayout.h"
  30 #include "llvm/IR/DebugInfo.h"
  31 #include "llvm/IR/IRPrintingPasses.h"
  32 #include "llvm/IR/LLVMContext.h"
  33 #include "llvm/IR/LegacyPassManager.h"
  34 #include "llvm/IR/LegacyPassNameParser.h"
  35 #include "llvm/IR/Module.h"
  36 #include "llvm/IR/RemarkStreamer.h"
  37 #include "llvm/IR/Verifier.h"
  38 #include "llvm/IRReader/IRReader.h"
  39 #include "llvm/InitializePasses.h"
  40 #include "llvm/LinkAllIR.h"
  41 #include "llvm/LinkAllPasses.h"
  42 #include "llvm/MC/SubtargetFeature.h"
  43 #include "llvm/Support/Debug.h"
  44 #include "llvm/Support/FileSystem.h"
  45 #include "llvm/Support/Host.h"
  46 #include "llvm/Support/InitLLVM.h"
  47 #include "llvm/Support/PluginLoader.h"
  48 #include "llvm/Support/SourceMgr.h"
  49 #include "llvm/Support/SystemUtils.h"
  50 #include "llvm/Support/TargetRegistry.h"
  51 #include "llvm/Support/TargetSelect.h"
  52 #include "llvm/Support/ToolOutputFile.h"
  53 #include "llvm/Support/YAMLTraits.h"
  54 #include "llvm/Target/TargetMachine.h"
  55 #include "llvm/Transforms/Coroutines.h"
  56 #include "llvm/Transforms/IPO/AlwaysInliner.h"
  57 #include "llvm/Transforms/IPO/PassManagerBuilder.h"
  58 #include "llvm/Transforms/Utils/Cloning.h"
  59 #include <algorithm>
  60 #include <memory>
  61 using namespace llvm;
  62 using namespace opt_tool;
  63
  64 // The OptimizationList is automatically populated with registered Passes by the
  65 // PassNameParser.
  66 //
  67 static cl::list<const PassInfo*, bool, PassNameParser>
  68 PassList(cl::desc("Optimizations available:"));
  69
  70 // This flag specifies a textual description of the optimization pass pipeline
  71 // to run over the module. This flag switches opt to use the new pass manager
  72 // infrastructure, completely disabling all of the flags specific to the old
  73 // pass management.
  74 static cl::opt<std::string> PassPipeline(
  75     "passes",
  76     cl::desc("A textual description of the pass pipeline for optimizing"),
  77     cl::Hidden);
  78
  79 // Other command line options...
  80 //
  81 static cl::opt<std::string>
  82 InputFilename(cl::Positional, cl::desc("<input bitcode file>"),
  83     cl::init("-"), cl::value_desc("filename"));
  84
  85 static cl::opt<std::string>
  86 OutputFilename("o", cl::desc("Override output filename"),
  87                cl::value_desc("filename"));
  88
  89 static cl::opt<bool>
  90 Force("f", cl::desc("Enable binary output on terminals"));
  91
  92 static cl::opt<bool>
  93 PrintEachXForm("p", cl::desc("Print module after each transformation"));
  94
  95 static cl::opt<bool>
  96 NoOutput("disable-output",
  97          cl::desc("Do not write result bitcode file"), cl::Hidden);
  98
  99 static cl::opt<bool>
 100 OutputAssembly("S", cl::desc("Write output as LLVM assembly"));
 101
 102 static cl::opt<bool>
 103     OutputThinLTOBC("thinlto-bc",
 104                     cl::desc("Write output as ThinLTO-ready bitcode"));
 105
 106 static cl::opt<bool>
 107     SplitLTOUnit("thinlto-split-lto-unit",
 108                  cl::desc("Enable splitting of a ThinLTO LTOUnit"));
 109
 110 static cl::opt<std::string> ThinLinkBitcodeFile(
 111     "thin-link-bitcode-file", cl::value_desc("filename"),
 112     cl::desc(
 113         "A file in which to write minimized bitcode for the thin link only"));
 114
 115 static cl::opt<bool>
 116 NoVerify("disable-verify", cl::desc("Do not run the verifier"), cl::Hidden);
 117
 118 static cl::opt<bool>
 119 VerifyEach("verify-each", cl::desc("Verify after each transform"));
 120
 121 static cl::opt<bool>
 122     DisableDITypeMap("disable-debug-info-type-map",
 123                      cl::desc("Don't use a uniquing type map for debug info"));
 124
 125 static cl::opt<bool>
 126 StripDebug("strip-debug",
 127            cl::desc("Strip debugger symbol info from translation unit"));
 128
 129 static cl::opt<bool>
 130     StripNamedMetadata("strip-named-metadata",
 131                        cl::desc("Strip module-level named metadata"));
 132
 133 static cl::opt<bool> DisableInline("disable-inlining",
 134                                    cl::desc("Do not run the inliner pass"));
 135
 136 static cl::opt<bool>
 137 DisableOptimizations("disable-opt",
 138                      cl::desc("Do not run any optimization passes"));
 139
 140 static cl::opt<bool>
 141 StandardLinkOpts("std-link-opts",
 142                  cl::desc("Include the standard link time optimizations"));
 143
 144 static cl::opt<bool>
 145 OptLevelO0("O0",
 146   cl::desc("Optimization level 0. Similar to clang -O0"));
 147
 148 static cl::opt<bool>
 149 OptLevelO1("O1",
 150            cl::desc("Optimization level 1. Similar to clang -O1"));
 151
 152 static cl::opt<bool>
 153 OptLevelO2("O2",
 154            cl::desc("Optimization level 2. Similar to clang -O2"));
 155
 156 static cl::opt<bool>
 157 OptLevelOs("Os",
 158            cl::desc("Like -O2 with extra optimizations for size. Similar to clang -Os"));
 159
 160 static cl::opt<bool>
 161 OptLevelOz("Oz",
 162            cl::desc("Like -Os but reduces code size further. Similar to clang -Oz"));
 163
 164 static cl::opt<bool>
 165 OptLevelO3("O3",
 166            cl::desc("Optimization level 3. Similar to clang -O3"));
 167
 168 static cl::opt<unsigned>
 169 CodeGenOptLevel("codegen-opt-level",
 170                 cl::desc("Override optimization level for codegen hooks"));
 171
 172 static cl::opt<std::string>
 173 TargetTriple("mtriple", cl::desc("Override target triple for module"));
 174
 175 static cl::opt<bool>
 176 DisableLoopUnrolling("disable-loop-unrolling",
 177                      cl::desc("Disable loop unrolling in all relevant passes"),
 178                      cl::init(false));
 179
 180 static cl::opt<bool>
 181 DisableSLPVectorization("disable-slp-vectorization",
 182                         cl::desc("Disable the slp vectorization pass"),
 183                         cl::init(false));
 184
 185 static cl::opt<bool> EmitSummaryIndex("module-summary",
 186                                       cl::desc("Emit module summary index"),
 187                                       cl::init(false));
 188
 189 static cl::opt<bool> EmitModuleHash("module-hash", cl::desc("Emit module hash"),
 190                                     cl::init(false));
 191
 192 static cl::opt<bool>
 193 DisableSimplifyLibCalls("disable-simplify-libcalls",
 194                         cl::desc("Disable simplify-libcalls"));
 195
 196 static cl::opt<bool>
 197 Quiet("q", cl::desc("Obsolete option"), cl::Hidden);
 198
 199 static cl::alias
 200 QuietA("quiet", cl::desc("Alias for -q"), cl::aliasopt(Quiet));
 201
 202 static cl::opt<bool>
 203 AnalyzeOnly("analyze", cl::desc("Only perform analysis, no optimization"));
 204
 205 static cl::opt<bool> EnableDebugify(
 206     "enable-debugify",
 207     cl::desc(
 208         "Start the pipeline with debugify and end it with check-debugify"));
 209
 210 static cl::opt<bool> DebugifyEach(
 211     "debugify-each",
 212     cl::desc(
 213         "Start each pass with debugify and end it with check-debugify"));
 214
 215 static cl::opt<std::string>
 216     DebugifyExport("debugify-export",
 217                    cl::desc("Export per-pass debugify statistics to this file"),
 218                    cl::value_desc("filename"), cl::init(""));
 219
 220 static cl::opt<bool>
 221 PrintBreakpoints("print-breakpoints-for-testing",
 222                  cl::desc("Print select breakpoints location for testing"));
 223
 224 static cl::opt<std::string> ClDataLayout("data-layout",
 225                                          cl::desc("data layout string to use"),
 226                                          cl::value_desc("layout-string"),
 227                                          cl::init(""));
 228
 229 static cl::opt<bool> PreserveBitcodeUseListOrder(
 230     "preserve-bc-uselistorder",
 231     cl::desc("Preserve use-list order when writing LLVM bitcode."),
 232     cl::init(true), cl::Hidden);
 233
 234 static cl::opt<bool> PreserveAssemblyUseListOrder(
 235     "preserve-ll-uselistorder",
 236     cl::desc("Preserve use-list order when writing LLVM assembly."),
 237     cl::init(false), cl::Hidden);
 238
 239 static cl::opt<bool>
 240     RunTwice("run-twice",
 241              cl::desc("Run all passes twice, re-using the same pass manager."),
 242              cl::init(false), cl::Hidden);
 243
 244 static cl::opt<bool> DiscardValueNames(
 245     "discard-value-names",
 246     cl::desc("Discard names from Value (other than GlobalValue)."),
 247     cl::init(false), cl::Hidden);
 248
 249 static cl::opt<bool> Coroutines(
 250   "enable-coroutines",
 251   cl::desc("Enable coroutine passes."),
 252   cl::init(false), cl::Hidden);
 253
 254 static cl::opt<bool> RemarksWithHotness(
 255     "pass-remarks-with-hotness",
 256     cl::desc("With PGO, include profile count in optimization remarks"),
 257     cl::Hidden);
 258
 259 static cl::opt<unsigned>
 260     RemarksHotnessThreshold("pass-remarks-hotness-threshold",
 261                             cl::desc("Minimum profile count required for "
 262                                      "an optimization remark to be output"),
 263                             cl::Hidden);
 264
 265 static cl::opt<std::string>
 266     RemarksFilename("pass-remarks-output",
 267                     cl::desc("Output filename for pass remarks"),
 268                     cl::value_desc("filename"));
 269
 270 static cl::opt<std::string>
 271     RemarksPasses("pass-remarks-filter",
 272                   cl::desc("Only record optimization remarks from passes whose "
 273                            "names match the given regular expression"),
 274                   cl::value_desc("regex"));
 275
 276 static cl::opt<std::string> RemarksFormat(
 277     "pass-remarks-format",
 278     cl::desc("The format used for serializing remarks (default: YAML)"),
 279     cl::value_desc("format"), cl::init("yaml"));
 280
 281 cl::opt<PGOKind>
 282     PGOKindFlag("pgo-kind", cl::init(NoPGO), cl::Hidden,
 283                 cl::desc("The kind of profile guided optimization"),
 284                 cl::values(clEnumValN(NoPGO, "nopgo", "Do not use PGO."),
 285                            clEnumValN(InstrGen, "pgo-instr-gen-pipeline",
 286                                       "Instrument the IR to generate profile."),
 287                            clEnumValN(InstrUse, "pgo-instr-use-pipeline",
 288                                       "Use instrumented profile to guide PGO."),
 289                            clEnumValN(SampleUse, "pgo-sample-use-pipeline",
 290                                       "Use sampled profile to guide PGO.")));
 291 cl::opt<std::string> ProfileFile("profile-file",
 292                                  cl::desc("Path to the profile."), cl::Hidden);
 293
 294 cl::opt<CSPGOKind> CSPGOKindFlag(
 295     "cspgo-kind", cl::init(NoCSPGO), cl::Hidden,
 296     cl::desc("The kind of context sensitive profile guided optimization"),
 297     cl::values(
 298         clEnumValN(NoCSPGO, "nocspgo", "Do not use CSPGO."),
 299         clEnumValN(
 300             CSInstrGen, "cspgo-instr-gen-pipeline",
 301             "Instrument (context sensitive) the IR to generate profile."),
 302         clEnumValN(
 303             CSInstrUse, "cspgo-instr-use-pipeline",
 304             "Use instrumented (context sensitive) profile to guide PGO.")));
 305 cl::opt<std::string> CSProfileGenFile(
 306     "cs-profilegen-file",
 307     cl::desc("Path to the instrumented context sensitive profile."),
 308     cl::Hidden);
 309
 310 class OptCustomPassManager : public legacy::PassManager {
 311   DebugifyStatsMap DIStatsMap;
 312
 313 public:
 314   using super = legacy::PassManager;
 315
 316   void add(Pass *P) override {
 317     // Wrap each pass with (-check)-debugify passes if requested, making
 318     // exceptions for passes which shouldn't see -debugify instrumentation.
 319     bool WrapWithDebugify = DebugifyEach && !P->getAsImmutablePass() &&
 320                             !isIRPrintingPass(P) && !isBitcodeWriterPass(P);
 321     if (!WrapWithDebugify) {
 322       super::add(P);
 323       return;
 324     }
 325
 326     // Apply -debugify/-check-debugify before/after each pass and collect
 327     // debug info loss statistics.
 328     PassKind Kind = P->getPassKind();
 329     StringRef Name = P->getPassName();
 330
 331     // TODO: Implement Debugify for BasicBlockPass, LoopPass.
 332     switch (Kind) {
 333       case PT_Function:
 334         super::add(createDebugifyFunctionPass());
 335         super::add(P);
 336         super::add(createCheckDebugifyFunctionPass(true, Name, &DIStatsMap));
 337         break;
 338       case PT_Module:
 339         super::add(createDebugifyModulePass());
 340         super::add(P);
 341         super::add(createCheckDebugifyModulePass(true, Name, &DIStatsMap));
 342         break;
 343       default:
 344         super::add(P);
 345         break;
 346     }
 347   }
 348
 349   const DebugifyStatsMap &getDebugifyStatsMap() const { return DIStatsMap; }
 350 };
 351
 352 static inline void addPass(legacy::PassManagerBase &PM, Pass *P) {
 353   // Add the pass to the pass manager...
 354   PM.add(P);
 355
 356   // If we are verifying all of the intermediate steps, add the verifier...
 357   if (VerifyEach)
 358     PM.add(createVerifierPass());
 359 }
 360
 361 /// This routine adds optimization passes based on selected optimization level,
 362 /// OptLevel.
 363 ///
 364 /// OptLevel - Optimization Level
 365 static void AddOptimizationPasses(legacy::PassManagerBase &MPM,
 366                                   legacy::FunctionPassManager &FPM,
 367                                   TargetMachine *TM, unsigned OptLevel,
 368                                   unsigned SizeLevel) {
 369   if (!NoVerify || VerifyEach)
 370     FPM.add(createVerifierPass()); // Verify that input is correct
 371
 372   PassManagerBuilder Builder;
 373   Builder.OptLevel = OptLevel;
 374   Builder.SizeLevel = SizeLevel;
 375
 376   if (DisableInline) {
 377     // No inlining pass
 378   } else if (OptLevel > 1) {
 379     Builder.Inliner = createFunctionInliningPass(OptLevel, SizeLevel, false);
 380   } else {
 381     Builder.Inliner = createAlwaysInlinerLegacyPass();
 382   }
 383   Builder.DisableUnrollLoops = (DisableLoopUnrolling.getNumOccurrences() > 0) ?
 384                                DisableLoopUnrolling : OptLevel == 0;
 385
 386   // Check if vectorization is explicitly disabled via -vectorize-loops=false.
 387   // The flag enables vectorization in the LoopVectorize pass, it is on by
 388   // default, and if it was disabled, leave it disabled here.
 389   // Another flag that exists: -loop-vectorize, controls adding the pass to the
 390   // pass manager. If set, the pass is added, and there is no additional check
 391   // here for it.
 392   if (Builder.LoopVectorize)
 393     Builder.LoopVectorize = OptLevel > 1 && SizeLevel < 2;
 394
 395   // When #pragma vectorize is on for SLP, do the same as above
 396   Builder.SLPVectorize =
 397       DisableSLPVectorization ? false : OptLevel > 1 && SizeLevel < 2;
 398
 399   if (TM)
 400     TM->adjustPassManager(Builder);
 401
 402   if (Coroutines)
 403     addCoroutinePassesToExtensionPoints(Builder);
 404
 405   switch (PGOKindFlag) {
 406   case InstrGen:
 407     Builder.EnablePGOInstrGen = true;
 408     Builder.PGOInstrGen = ProfileFile;
 409     break;
 410   case InstrUse:
 411     Builder.PGOInstrUse = ProfileFile;
 412     break;
 413   case SampleUse:
 414     Builder.PGOSampleUse = ProfileFile;
 415     break;
 416   default:
 417     break;
 418   }
 419
 420   switch (CSPGOKindFlag) {
 421   case CSInstrGen:
 422     Builder.EnablePGOCSInstrGen = true;
 423     break;
 424   case CSInstrUse:
 425     Builder.EnablePGOCSInstrUse = true;
 426     break;
 427   default:
 428     break;
 429   }
 430
 431   Builder.populateFunctionPassManager(FPM);
 432   Builder.populateModulePassManager(MPM);
 433 }
 434
 435 static void AddStandardLinkPasses(legacy::PassManagerBase &PM) {
 436   PassManagerBuilder Builder;
 437   Builder.VerifyInput = true;
 438   if (DisableOptimizations)
 439     Builder.OptLevel = 0;
 440
 441   if (!DisableInline)
 442     Builder.Inliner = createFunctionInliningPass();
 443   Builder.populateLTOPassManager(PM);
 444 }
 445
 446 //===----------------------------------------------------------------------===//
 447 // CodeGen-related helper functions.
 448 //
 449
 450 static CodeGenOpt::Level GetCodeGenOptLevel() {
 451   if (CodeGenOptLevel.getNumOccurrences())
 452     return static_cast<CodeGenOpt::Level>(unsigned(CodeGenOptLevel));
 453   if (OptLevelO1)
 454     return CodeGenOpt::Less;
 455   if (OptLevelO2)
 456     return CodeGenOpt::Default;
 457   if (OptLevelO3)
 458     return CodeGenOpt::Aggressive;
 459   return CodeGenOpt::None;
 460 }
 461
 462 // Returns the TargetMachine instance or zero if no triple is provided.
 463 static TargetMachine* GetTargetMachine(Triple TheTriple, StringRef CPUStr,
 464                                        StringRef FeaturesStr,
 465                                        const TargetOptions &Options) {
 466   std::string Error;
 467   const Target *TheTarget = TargetRegistry::lookupTarget(MArch, TheTriple,
 468                                                          Error);
 469   // Some modules don't specify a triple, and this is okay.
 470   if (!TheTarget) {
 471     return nullptr;
 472   }
 473
 474   return TheTarget->createTargetMachine(TheTriple.getTriple(), CPUStr,
 475                                         FeaturesStr, Options, getRelocModel(),
 476                                         getCodeModel(), GetCodeGenOptLevel());
 477 }
 478
 479 #ifdef LINK_POLLY_INTO_TOOLS
 480 namespace polly {
 481 void initializePollyPasses(llvm::PassRegistry &Registry);
 482 }
 483 #endif
 484
 485 //===----------------------------------------------------------------------===//
 486 // main for opt
 487 //
 488 int main(int argc, char **argv) {
 489   InitLLVM X(argc, argv);
 490
 491   // Enable debug stream buffering.
 492   EnableDebugBuffering = true;
 493
 494   LLVMContext Context;
 495
 496   InitializeAllTargets();
 497   InitializeAllTargetMCs();
 498   InitializeAllAsmPrinters();
 499   InitializeAllAsmParsers();
 500
 501   // Initialize passes
 502   PassRegistry &Registry = *PassRegistry::getPassRegistry();
 503   initializeCore(Registry);
 504   initializeCoroutines(Registry);
 505   initializeScalarOpts(Registry);
 506   initializeObjCARCOpts(Registry);
 507   initializeVectorization(Registry);
 508   initializeIPO(Registry);
 509   initializeAnalysis(Registry);
 510   initializeTransformUtils(Registry);
 511   initializeInstCombine(Registry);
 512   initializeAggressiveInstCombine(Registry);
 513   initializeInstrumentation(Registry);
 514   initializeTarget(Registry);
 515   // For codegen passes, only passes that do IR to IR transformation are
 516   // supported.
 517   initializeExpandMemCmpPassPass(Registry);
 518   initializeScalarizeMaskedMemIntrinPass(Registry);
 519   initializeCodeGenPreparePass(Registry);
 520   initializeAtomicExpandPass(Registry);
 521   initializeRewriteSymbolsLegacyPassPass(Registry);
 522   initializeWinEHPreparePass(Registry);
 523   initializeDwarfEHPreparePass(Registry);
 524   initializeSafeStackLegacyPassPass(Registry);
 525   initializeSjLjEHPreparePass(Registry);
 526   initializeStackProtectorPass(Registry);
 527   initializePreISelIntrinsicLoweringLegacyPassPass(Registry);
 528   initializeGlobalMergePass(Registry);
 529   initializeIndirectBrExpandPassPass(Registry);
 530   initializeInterleavedLoadCombinePass(Registry);
 531   initializeInterleavedAccessPass(Registry);
 532   initializeEntryExitInstrumenterPass(Registry);
 533   initializePostInlineEntryExitInstrumenterPass(Registry);
 534   initializeUnreachableBlockElimLegacyPassPass(Registry);
 535   initializeExpandReductionsPass(Registry);
 536   initializeWasmEHPreparePass(Registry);
 537   initializeWriteBitcodePassPass(Registry);
 538   initializeHardwareLoopsPass(Registry);
 539
 540 #ifdef LINK_POLLY_INTO_TOOLS
 541   polly::initializePollyPasses(Registry);
 542 #endif
 543
 544   cl::ParseCommandLineOptions(argc, argv,
 545     "llvm .bc -> .bc modular optimizer and analysis printer\n");
 546
 547   if (AnalyzeOnly && NoOutput) {
 548     errs() << argv[0] << ": analyze mode conflicts with no-output mode.\n";
 549     return 1;
 550   }
 551
 552   SMDiagnostic Err;
 553
 554   Context.setDiscardValueNames(DiscardValueNames);
 555   if (!DisableDITypeMap)
 556     Context.enableDebugTypeODRUniquing();
 557
 558   Expected<std::unique_ptr<ToolOutputFile>> RemarksFileOrErr =
 559       setupOptimizationRemarks(Context, RemarksFilename, RemarksPasses,
 560                                RemarksFormat, RemarksWithHotness,
 561                                RemarksHotnessThreshold);
 562   if (Error E = RemarksFileOrErr.takeError()) {
 563     errs() << toString(std::move(E)) << '\n';
 564     return 1;
 565   }
 566   std::unique_ptr<ToolOutputFile> RemarksFile = std::move(*RemarksFileOrErr);
 567
 568   // Load the input module...
 569   std::unique_ptr<Module> M =
 570       parseIRFile(InputFilename, Err, Context, !NoVerify, ClDataLayout);
 571
 572   if (!M) {
 573     Err.print(argv[0], errs());
 574     return 1;
 575   }
 576
 577   // Strip debug info before running the verifier.
 578   if (StripDebug)
 579     StripDebugInfo(*M);
 580
 581   // Erase module-level named metadata, if requested.
 582   if (StripNamedMetadata) {
 583     while (!M->named_metadata_empty()) {
 584       NamedMDNode *NMD = &*M->named_metadata_begin();
 585       M->eraseNamedMetadata(NMD);
 586     }
 587   }
 588
 589   // If we are supposed to override the target triple or data layout, do so now.
 590   if (!TargetTriple.empty())
 591     M->setTargetTriple(Triple::normalize(TargetTriple));
 592
 593   // Immediately run the verifier to catch any problems before starting up the
 594   // pass pipelines.  Otherwise we can crash on broken code during
 595   // doInitialization().
 596   if (!NoVerify && verifyModule(*M, &errs())) {
 597     errs() << argv[0] << ": " << InputFilename
 598            << ": error: input module is broken!\n";
 599     return 1;
 600   }
 601
 602   // Figure out what stream we are supposed to write to...
 603   std::unique_ptr<ToolOutputFile> Out;
 604   std::unique_ptr<ToolOutputFile> ThinLinkOut;
 605   if (NoOutput) {
 606     if (!OutputFilename.empty())
 607       errs() << "WARNING: The -o (output filename) option is ignored when\n"
 608                 "the --disable-output option is used.\n";
 609   } else {
 610     // Default to standard output.
 611     if (OutputFilename.empty())
 612       OutputFilename = "-";
 613
 614     std::error_code EC;
 615     Out.reset(new ToolOutputFile(OutputFilename, EC, sys::fs::OF_None));
 616     if (EC) {
 617       errs() << EC.message() << '\n';
 618       return 1;
 619     }
 620
 621     if (!ThinLinkBitcodeFile.empty()) {
 622       ThinLinkOut.reset(
 623           new ToolOutputFile(ThinLinkBitcodeFile, EC, sys::fs::OF_None));
 624       if (EC) {
 625         errs() << EC.message() << '\n';
 626         return 1;
 627       }
 628     }
 629   }
 630
 631   Triple ModuleTriple(M->getTargetTriple());
 632   std::string CPUStr, FeaturesStr;
 633   TargetMachine *Machine = nullptr;
 634   const TargetOptions Options = InitTargetOptionsFromCodeGenFlags();
 635
 636   if (ModuleTriple.getArch()) {
 637     CPUStr = getCPUStr();
 638     FeaturesStr = getFeaturesStr();
 639     Machine = GetTargetMachine(ModuleTriple, CPUStr, FeaturesStr, Options);
 640   } else if (ModuleTriple.getArchName() != "unknown" &&
 641              ModuleTriple.getArchName() != "") {
 642     errs() << argv[0] << ": unrecognized architecture '"
 643            << ModuleTriple.getArchName() << "' provided.\n";
 644     return 1;
 645   }
 646
 647   std::unique_ptr<TargetMachine> TM(Machine);
 648
 649   // Override function attributes based on CPUStr, FeaturesStr, and command line
 650   // flags.
 651   setFunctionAttributes(CPUStr, FeaturesStr, *M);
 652
 653   // If the output is set to be emitted to standard out, and standard out is a
 654   // console, print out a warning message and refuse to do it.  We don't
 655   // impress anyone by spewing tons of binary goo to a terminal.
 656   if (!Force && !NoOutput && !AnalyzeOnly && !OutputAssembly)
 657     if (CheckBitcodeOutputToConsole(Out->os(), !Quiet))
 658       NoOutput = true;
 659
 660   if (OutputThinLTOBC)
 661     M->addModuleFlag(Module::Error, "EnableSplitLTOUnit", SplitLTOUnit);
 662
 663   if (PassPipeline.getNumOccurrences() > 0) {
 664     OutputKind OK = OK_NoOutput;
 665     if (!NoOutput)
 666       OK = OutputAssembly
 667                ? OK_OutputAssembly
 668                : (OutputThinLTOBC ? OK_OutputThinLTOBitcode : OK_OutputBitcode);
 669
 670     VerifierKind VK = VK_VerifyInAndOut;
 671     if (NoVerify)
 672       VK = VK_NoVerifier;
 673     else if (VerifyEach)
 674       VK = VK_VerifyEachPass;
 675
 676     // The user has asked to use the new pass manager and provided a pipeline
 677     // string. Hand off the rest of the functionality to the new code for that
 678     // layer.
 679     return runPassPipeline(argv[0], *M, TM.get(), Out.get(), ThinLinkOut.get(),
 680                            RemarksFile.get(), PassPipeline, OK, VK,
 681                            PreserveAssemblyUseListOrder,
 682                            PreserveBitcodeUseListOrder, EmitSummaryIndex,
 683                            EmitModuleHash, EnableDebugify)
 684                ? 0
 685                : 1;
 686   }
 687
 688   // Create a PassManager to hold and optimize the collection of passes we are
 689   // about to build.
 690   OptCustomPassManager Passes;
 691   bool AddOneTimeDebugifyPasses = EnableDebugify && !DebugifyEach;
 692
 693   // Add an appropriate TargetLibraryInfo pass for the module's triple.
 694   TargetLibraryInfoImpl TLII(ModuleTriple);
 695
 696   // The -disable-simplify-libcalls flag actually disables all builtin optzns.
 697   if (DisableSimplifyLibCalls)
 698     TLII.disableAllFunctions();
 699   Passes.add(new TargetLibraryInfoWrapperPass(TLII));
 700
 701   // Add internal analysis passes from the target machine.
 702   Passes.add(createTargetTransformInfoWrapperPass(TM ? TM->getTargetIRAnalysis()
 703                                                      : TargetIRAnalysis()));
 704
 705   if (AddOneTimeDebugifyPasses)
 706     Passes.add(createDebugifyModulePass());
 707
 708   std::unique_ptr<legacy::FunctionPassManager> FPasses;
 709   if (OptLevelO0 || OptLevelO1 || OptLevelO2 || OptLevelOs || OptLevelOz ||
 710       OptLevelO3) {
 711     FPasses.reset(new legacy::FunctionPassManager(M.get()));
 712     FPasses->add(createTargetTransformInfoWrapperPass(
 713         TM ? TM->getTargetIRAnalysis() : TargetIRAnalysis()));
 714   }
 715
 716   if (PrintBreakpoints) {
 717     // Default to standard output.
 718     if (!Out) {
 719       if (OutputFilename.empty())
 720         OutputFilename = "-";
 721
 722       std::error_code EC;
 723       Out = std::make_unique<ToolOutputFile>(OutputFilename, EC,
 724                                               sys::fs::OF_None);
 725       if (EC) {
 726         errs() << EC.message() << '\n';
 727         return 1;
 728       }
 729     }
 730     Passes.add(createBreakpointPrinter(Out->os()));
 731     NoOutput = true;
 732   }
 733
 734   if (TM) {
 735     // FIXME: We should dyn_cast this when supported.
 736     auto &LTM = static_cast<LLVMTargetMachine &>(*TM);
 737     Pass *TPC = LTM.createPassConfig(Passes);
 738     Passes.add(TPC);
 739   }
 740
 741   // Create a new optimization pass for each one specified on the command line
 742   for (unsigned i = 0; i < PassList.size(); ++i) {
 743     if (StandardLinkOpts &&
 744         StandardLinkOpts.getPosition() < PassList.getPosition(i)) {
 745       AddStandardLinkPasses(Passes);
 746       StandardLinkOpts = false;
 747     }
 748
 749     if (OptLevelO0 && OptLevelO0.getPosition() < PassList.getPosition(i)) {
 750       AddOptimizationPasses(Passes, *FPasses, TM.get(), 0, 0);
 751       OptLevelO0 = false;
 752     }
 753
 754     if (OptLevelO1 && OptLevelO1.getPosition() < PassList.getPosition(i)) {
 755       AddOptimizationPasses(Passes, *FPasses, TM.get(), 1, 0);
 756       OptLevelO1 = false;
 757     }
 758
 759     if (OptLevelO2 && OptLevelO2.getPosition() < PassList.getPosition(i)) {
 760       AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 0);
 761       OptLevelO2 = false;
 762     }
 763
 764     if (OptLevelOs && OptLevelOs.getPosition() < PassList.getPosition(i)) {
 765       AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 1);
 766       OptLevelOs = false;
 767     }
 768
 769     if (OptLevelOz && OptLevelOz.getPosition() < PassList.getPosition(i)) {
 770       AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 2);
 771       OptLevelOz = false;
 772     }
 773
 774     if (OptLevelO3 && OptLevelO3.getPosition() < PassList.getPosition(i)) {
 775       AddOptimizationPasses(Passes, *FPasses, TM.get(), 3, 0);
 776       OptLevelO3 = false;
 777     }
 778
 779     const PassInfo *PassInf = PassList[i];
 780     Pass *P = nullptr;
 781     if (PassInf->getNormalCtor())
 782       P = PassInf->getNormalCtor()();
 783     else
 784       errs() << argv[0] << ": cannot create pass: "
 785              << PassInf->getPassName() << "\n";
 786     if (P) {
 787       PassKind Kind = P->getPassKind();
 788       addPass(Passes, P);
 789
 790       if (AnalyzeOnly) {
 791         switch (Kind) {
 792         case PT_BasicBlock:
 793           Passes.add(createBasicBlockPassPrinter(PassInf, Out->os(), Quiet));
 794           break;
 795         case PT_Region:
 796           Passes.add(createRegionPassPrinter(PassInf, Out->os(), Quiet));
 797           break;
 798         case PT_Loop:
 799           Passes.add(createLoopPassPrinter(PassInf, Out->os(), Quiet));
 800           break;
 801         case PT_Function:
 802           Passes.add(createFunctionPassPrinter(PassInf, Out->os(), Quiet));
 803           break;
 804         case PT_CallGraphSCC:
 805           Passes.add(createCallGraphPassPrinter(PassInf, Out->os(), Quiet));
 806           break;
 807         default:
 808           Passes.add(createModulePassPrinter(PassInf, Out->os(), Quiet));
 809           break;
 810         }
 811       }
 812     }
 813
 814     if (PrintEachXForm)
 815       Passes.add(
 816           createPrintModulePass(errs(), "", PreserveAssemblyUseListOrder));
 817   }
 818
 819   if (StandardLinkOpts) {
 820     AddStandardLinkPasses(Passes);
 821     StandardLinkOpts = false;
 822   }
 823
 824   if (OptLevelO0)
 825     AddOptimizationPasses(Passes, *FPasses, TM.get(), 0, 0);
 826
 827   if (OptLevelO1)
 828     AddOptimizationPasses(Passes, *FPasses, TM.get(), 1, 0);
 829
 830   if (OptLevelO2)
 831     AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 0);
 832
 833   if (OptLevelOs)
 834     AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 1);
 835
 836   if (OptLevelOz)
 837     AddOptimizationPasses(Passes, *FPasses, TM.get(), 2, 2);
 838
 839   if (OptLevelO3)
 840     AddOptimizationPasses(Passes, *FPasses, TM.get(), 3, 0);
 841
 842   if (FPasses) {
 843     FPasses->doInitialization();
 844     for (Function &F : *M)
 845       FPasses->run(F);
 846     FPasses->doFinalization();
 847   }
 848
 849   // Check that the module is well formed on completion of optimization
 850   if (!NoVerify && !VerifyEach)
 851     Passes.add(createVerifierPass());
 852
 853   if (AddOneTimeDebugifyPasses)
 854     Passes.add(createCheckDebugifyModulePass(false));
 855
 856   // In run twice mode, we want to make sure the output is bit-by-bit
 857   // equivalent if we run the pass manager again, so setup two buffers and
 858   // a stream to write to them. Note that llc does something similar and it
 859   // may be worth to abstract this out in the future.
 860   SmallVector<char, 0> Buffer;
 861   SmallVector<char, 0> FirstRunBuffer;
 862   std::unique_ptr<raw_svector_ostream> BOS;
 863   raw_ostream *OS = nullptr;
 864
 865   // Write bitcode or assembly to the output as the last step...
 866   if (!NoOutput && !AnalyzeOnly) {
 867     assert(Out);
 868     OS = &Out->os();
 869     if (RunTwice) {
 870       BOS = std::make_unique<raw_svector_ostream>(Buffer);
 871       OS = BOS.get();
 872     }
 873     if (OutputAssembly) {
 874       if (EmitSummaryIndex)
 875         report_fatal_error("Text output is incompatible with -module-summary");
 876       if (EmitModuleHash)
 877         report_fatal_error("Text output is incompatible with -module-hash");
 878       Passes.add(createPrintModulePass(*OS, "", PreserveAssemblyUseListOrder));
 879     } else if (OutputThinLTOBC)
 880       Passes.add(createWriteThinLTOBitcodePass(
 881           *OS, ThinLinkOut ? &ThinLinkOut->os() : nullptr));
 882     else
 883       Passes.add(createBitcodeWriterPass(*OS, PreserveBitcodeUseListOrder,
 884                                          EmitSummaryIndex, EmitModuleHash));
 885   }
 886
 887   // Before executing passes, print the final values of the LLVM options.
 888   cl::PrintOptionValues();
 889
 890   if (!RunTwice) {
 891     // Now that we have all of the passes ready, run them.
 892     Passes.run(*M);
 893   } else {
 894     // If requested, run all passes twice with the same pass manager to catch
 895     // bugs caused by persistent state in the passes.
 896     std::unique_ptr<Module> M2(CloneModule(*M));
 897     // Run all passes on the original module first, so the second run processes
 898     // the clone to catch CloneModule bugs.
 899     Passes.run(*M);
 900     FirstRunBuffer = Buffer;
 901     Buffer.clear();
 902
 903     Passes.run(*M2);
 904
 905     // Compare the two outputs and make sure they're the same
 906     assert(Out);
 907     if (Buffer.size() != FirstRunBuffer.size() ||
 908         (memcmp(Buffer.data(), FirstRunBuffer.data(), Buffer.size()) != 0)) {
 909       errs()
 910           << "Running the pass manager twice changed the output.\n"
 911              "Writing the result of the second run to the specified output.\n"
 912              "To generate the one-run comparison binary, just run without\n"
 913              "the compile-twice option\n";
 914       Out->os() << BOS->str();
 915       Out->keep();
 916       if (RemarksFile)
 917         RemarksFile->keep();
 918       return 1;
 919     }
 920     Out->os() << BOS->str();
 921   }
 922
 923   if (DebugifyEach && !DebugifyExport.empty())
 924     exportDebugifyStats(DebugifyExport, Passes.getDebugifyStatsMap());
 925
 926   // Declare success.
 927   if (!NoOutput || PrintBreakpoints)
 928     Out->keep();
 929
 930   if (RemarksFile)
 931     RemarksFile->keep();
 932
 933   if (ThinLinkOut)
 934     ThinLinkOut->keep();
 935
 936   return 0;
 937 }