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 "NewPMDriver.h" 16 #include "llvm/ADT/Triple.h" 17 #include "llvm/Analysis/CallGraph.h" 18 #include "llvm/Analysis/CallGraphSCCPass.h" 19 #include "llvm/Analysis/LoopPass.h" 20 #include "llvm/Analysis/RegionPass.h" 21 #include "llvm/Analysis/TargetLibraryInfo.h" 22 #include "llvm/Analysis/TargetTransformInfo.h" 23 #include "llvm/AsmParser/Parser.h" 24 #include "llvm/CodeGen/CommandFlags.h" 25 #include "llvm/CodeGen/TargetPassConfig.h" 26 #include "llvm/Config/llvm-config.h" 27 #include "llvm/IR/DataLayout.h" 28 #include "llvm/IR/DebugInfo.h" 29 #include "llvm/IR/LLVMContext.h" 30 #include "llvm/IR/LLVMRemarkStreamer.h" 31 #include "llvm/IR/LegacyPassManager.h" 32 #include "llvm/IR/LegacyPassNameParser.h" 33 #include "llvm/IR/Module.h" 34 #include "llvm/IR/ModuleSummaryIndex.h" 35 #include "llvm/IR/Verifier.h" 36 #include "llvm/IRReader/IRReader.h" 37 #include "llvm/InitializePasses.h" 38 #include "llvm/LinkAllIR.h" 39 #include "llvm/LinkAllPasses.h" 40 #include "llvm/MC/SubtargetFeature.h" 41 #include "llvm/MC/TargetRegistry.h" 42 #include "llvm/Passes/PassPlugin.h" 43 #include "llvm/Remarks/HotnessThresholdParser.h" 44 #include "llvm/Support/Debug.h" 45 #include "llvm/Support/FileSystem.h" 46 #include "llvm/Support/Host.h" 47 #include "llvm/Support/InitLLVM.h" 48 #include "llvm/Support/PluginLoader.h" 49 #include "llvm/Support/SourceMgr.h" 50 #include "llvm/Support/SystemUtils.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/IPO/WholeProgramDevirt.h" 56 #include "llvm/Transforms/Utils/Cloning.h" 57 #include "llvm/Transforms/Utils/Debugify.h" 58 #include <algorithm> 59 #include <memory> 60 #include <optional> 61 using namespace llvm; 62 using namespace opt_tool; 63 64 static codegen::RegisterCodeGenFlags CFG; 65 66 // The OptimizationList is automatically populated with registered Passes by the 67 // PassNameParser. 68 static cl::list<const PassInfo *, bool, PassNameParser> PassList(cl::desc( 69 "Optimizations available (use '-passes=' for the new pass manager)")); 70 71 static cl::opt<bool> EnableNewPassManager( 72 "enable-new-pm", 73 cl::desc("Enable the new pass manager, translating " 74 "'opt -foo' to 'opt -passes=foo'. This is strictly for the new PM " 75 "migration, use '-passes=' when possible."), 76 cl::init(true)); 77 78 // This flag specifies a textual description of the optimization pass pipeline 79 // to run over the module. This flag switches opt to use the new pass manager 80 // infrastructure, completely disabling all of the flags specific to the old 81 // pass management. 82 static cl::opt<std::string> PassPipeline( 83 "passes", 84 cl::desc( 85 "A textual description of the pass pipeline. To have analysis passes " 86 "available before a certain pass, add 'require<foo-analysis>'.")); 87 static cl::alias PassPipeline2("p", cl::aliasopt(PassPipeline), 88 cl::desc("Alias for -passes")); 89 90 static cl::opt<bool> PrintPasses("print-passes", 91 cl::desc("Print available passes that can be " 92 "specified in -passes=foo and exit")); 93 94 static cl::opt<std::string> 95 InputFilename(cl::Positional, cl::desc("<input bitcode file>"), 96 cl::init("-"), cl::value_desc("filename")); 97 98 static cl::opt<std::string> 99 OutputFilename("o", cl::desc("Override output filename"), 100 cl::value_desc("filename")); 101 102 static cl::opt<bool> 103 Force("f", cl::desc("Enable binary output on terminals")); 104 105 static cl::opt<bool> 106 NoOutput("disable-output", 107 cl::desc("Do not write result bitcode file"), cl::Hidden); 108 109 static cl::opt<bool> 110 OutputAssembly("S", cl::desc("Write output as LLVM assembly")); 111 112 static cl::opt<bool> 113 OutputThinLTOBC("thinlto-bc", 114 cl::desc("Write output as ThinLTO-ready bitcode")); 115 116 static cl::opt<bool> 117 SplitLTOUnit("thinlto-split-lto-unit", 118 cl::desc("Enable splitting of a ThinLTO LTOUnit")); 119 120 static cl::opt<std::string> ThinLinkBitcodeFile( 121 "thin-link-bitcode-file", cl::value_desc("filename"), 122 cl::desc( 123 "A file in which to write minimized bitcode for the thin link only")); 124 125 static cl::opt<bool> 126 NoVerify("disable-verify", cl::desc("Do not run the verifier"), cl::Hidden); 127 128 static cl::opt<bool> NoUpgradeDebugInfo("disable-upgrade-debug-info", 129 cl::desc("Generate invalid output"), 130 cl::ReallyHidden); 131 132 static cl::opt<bool> VerifyEach("verify-each", 133 cl::desc("Verify after each transform")); 134 135 static cl::opt<bool> 136 DisableDITypeMap("disable-debug-info-type-map", 137 cl::desc("Don't use a uniquing type map for debug info")); 138 139 static cl::opt<bool> 140 StripDebug("strip-debug", 141 cl::desc("Strip debugger symbol info from translation unit")); 142 143 static cl::opt<bool> 144 StripNamedMetadata("strip-named-metadata", 145 cl::desc("Strip module-level named metadata")); 146 147 148 149 static cl::opt<bool> 150 OptLevelO0("O0", cl::desc("Optimization level 0. Similar to clang -O0. " 151 "Use -passes='default<O0>' for the new PM")); 152 153 static cl::opt<bool> 154 OptLevelO1("O1", cl::desc("Optimization level 1. Similar to clang -O1. " 155 "Use -passes='default<O1>' for the new PM")); 156 157 static cl::opt<bool> 158 OptLevelO2("O2", cl::desc("Optimization level 2. Similar to clang -O2. " 159 "Use -passes='default<O2>' for the new PM")); 160 161 static cl::opt<bool> 162 OptLevelOs("Os", cl::desc("Like -O2 but size-conscious. Similar to clang " 163 "-Os. Use -passes='default<Os>' for the new PM")); 164 165 static cl::opt<bool> OptLevelOz( 166 "Oz", 167 cl::desc("Like -O2 but optimize for code size above all else. Similar to " 168 "clang -Oz. Use -passes='default<Oz>' for the new PM")); 169 170 static cl::opt<bool> 171 OptLevelO3("O3", cl::desc("Optimization level 3. Similar to clang -O3. " 172 "Use -passes='default<O3>' for the new PM")); 173 174 static cl::opt<unsigned> CodeGenOptLevel( 175 "codegen-opt-level", 176 cl::desc("Override optimization level for codegen hooks, legacy PM only")); 177 178 static cl::opt<std::string> 179 TargetTriple("mtriple", cl::desc("Override target triple for module")); 180 181 static cl::opt<bool> EmitSummaryIndex("module-summary", 182 cl::desc("Emit module summary index"), 183 cl::init(false)); 184 185 static cl::opt<bool> EmitModuleHash("module-hash", cl::desc("Emit module hash"), 186 cl::init(false)); 187 188 static cl::opt<bool> 189 DisableSimplifyLibCalls("disable-simplify-libcalls", 190 cl::desc("Disable simplify-libcalls")); 191 192 static cl::list<std::string> DisableBuiltins( 193 "disable-builtin", 194 cl::desc("Disable specific target library builtin function")); 195 196 static cl::opt<bool> EnableDebugify( 197 "enable-debugify", 198 cl::desc( 199 "Start the pipeline with debugify and end it with check-debugify")); 200 201 static cl::opt<bool> VerifyDebugInfoPreserve( 202 "verify-debuginfo-preserve", 203 cl::desc("Start the pipeline with collecting and end it with checking of " 204 "debug info preservation.")); 205 206 static cl::opt<bool> 207 PrintBreakpoints("print-breakpoints-for-testing", 208 cl::desc("Print select breakpoints location for testing")); 209 210 static cl::opt<std::string> ClDataLayout("data-layout", 211 cl::desc("data layout string to use"), 212 cl::value_desc("layout-string"), 213 cl::init("")); 214 215 static cl::opt<bool> PreserveBitcodeUseListOrder( 216 "preserve-bc-uselistorder", 217 cl::desc("Preserve use-list order when writing LLVM bitcode."), 218 cl::init(true), cl::Hidden); 219 220 static cl::opt<bool> PreserveAssemblyUseListOrder( 221 "preserve-ll-uselistorder", 222 cl::desc("Preserve use-list order when writing LLVM assembly."), 223 cl::init(false), cl::Hidden); 224 225 static cl::opt<bool> RunTwice("run-twice", 226 cl::desc("Run all passes twice, re-using the " 227 "same pass manager (legacy PM only)."), 228 cl::init(false), cl::Hidden); 229 230 static cl::opt<bool> DiscardValueNames( 231 "discard-value-names", 232 cl::desc("Discard names from Value (other than GlobalValue)."), 233 cl::init(false), cl::Hidden); 234 235 static cl::opt<bool> TimeTrace( 236 "time-trace", 237 cl::desc("Record time trace")); 238 239 static cl::opt<unsigned> TimeTraceGranularity( 240 "time-trace-granularity", 241 cl::desc("Minimum time granularity (in microseconds) traced by time profiler"), 242 cl::init(500), cl::Hidden); 243 244 static cl::opt<std::string> 245 TimeTraceFile("time-trace-file", 246 cl::desc("Specify time trace file destination"), 247 cl::value_desc("filename")); 248 249 static cl::opt<bool> RemarksWithHotness( 250 "pass-remarks-with-hotness", 251 cl::desc("With PGO, include profile count in optimization remarks"), 252 cl::Hidden); 253 254 static cl::opt<std::optional<uint64_t>, false, remarks::HotnessThresholdParser> 255 RemarksHotnessThreshold( 256 "pass-remarks-hotness-threshold", 257 cl::desc("Minimum profile count required for " 258 "an optimization remark to be output. " 259 "Use 'auto' to apply the threshold from profile summary"), 260 cl::value_desc("N or 'auto'"), cl::init(0), cl::Hidden); 261 262 static cl::opt<std::string> 263 RemarksFilename("pass-remarks-output", 264 cl::desc("Output filename for pass remarks"), 265 cl::value_desc("filename")); 266 267 static cl::opt<std::string> 268 RemarksPasses("pass-remarks-filter", 269 cl::desc("Only record optimization remarks from passes whose " 270 "names match the given regular expression"), 271 cl::value_desc("regex")); 272 273 static cl::opt<std::string> RemarksFormat( 274 "pass-remarks-format", 275 cl::desc("The format used for serializing remarks (default: YAML)"), 276 cl::value_desc("format"), cl::init("yaml")); 277 278 static cl::list<std::string> 279 PassPlugins("load-pass-plugin", 280 cl::desc("Load passes from plugin library")); 281 282 static inline void addPass(legacy::PassManagerBase &PM, Pass *P) { 283 // Add the pass to the pass manager... 284 PM.add(P); 285 286 // If we are verifying all of the intermediate steps, add the verifier... 287 if (VerifyEach) 288 PM.add(createVerifierPass()); 289 } 290 291 //===----------------------------------------------------------------------===// 292 // CodeGen-related helper functions. 293 // 294 295 static CodeGenOpt::Level GetCodeGenOptLevel() { 296 return static_cast<CodeGenOpt::Level>(unsigned(CodeGenOptLevel)); 297 } 298 299 // Returns the TargetMachine instance or zero if no triple is provided. 300 static TargetMachine* GetTargetMachine(Triple TheTriple, StringRef CPUStr, 301 StringRef FeaturesStr, 302 const TargetOptions &Options) { 303 std::string Error; 304 const Target *TheTarget = 305 TargetRegistry::lookupTarget(codegen::getMArch(), TheTriple, Error); 306 // Some modules don't specify a triple, and this is okay. 307 if (!TheTarget) { 308 return nullptr; 309 } 310 311 return TheTarget->createTargetMachine( 312 TheTriple.getTriple(), codegen::getCPUStr(), codegen::getFeaturesStr(), 313 Options, codegen::getExplicitRelocModel(), 314 codegen::getExplicitCodeModel(), GetCodeGenOptLevel()); 315 } 316 317 struct TimeTracerRAII { 318 TimeTracerRAII(StringRef ProgramName) { 319 if (TimeTrace) 320 timeTraceProfilerInitialize(TimeTraceGranularity, ProgramName); 321 } 322 ~TimeTracerRAII() { 323 if (TimeTrace) { 324 if (auto E = timeTraceProfilerWrite(TimeTraceFile, OutputFilename)) { 325 handleAllErrors(std::move(E), [&](const StringError &SE) { 326 errs() << SE.getMessage() << "\n"; 327 }); 328 return; 329 } 330 timeTraceProfilerCleanup(); 331 } 332 } 333 }; 334 335 // For use in NPM transition. Currently this contains most codegen-specific 336 // passes. Remove passes from here when porting to the NPM. 337 // TODO: use a codegen version of PassRegistry.def/PassBuilder::is*Pass() once 338 // it exists. 339 static bool shouldPinPassToLegacyPM(StringRef Pass) { 340 std::vector<StringRef> PassNameExactToIgnore = { 341 "nvvm-reflect", 342 "nvvm-intr-range", 343 "amdgpu-simplifylib", 344 "amdgpu-usenative", 345 "amdgpu-promote-alloca", 346 "amdgpu-promote-alloca-to-vector", 347 "amdgpu-lower-kernel-attributes", 348 "amdgpu-propagate-attributes-early", 349 "amdgpu-propagate-attributes-late", 350 "amdgpu-unify-metadata", 351 "amdgpu-printf-runtime-binding", 352 "amdgpu-always-inline"}; 353 if (llvm::is_contained(PassNameExactToIgnore, Pass)) 354 return false; 355 356 std::vector<StringRef> PassNamePrefix = { 357 "x86-", "xcore-", "wasm-", "systemz-", "ppc-", "nvvm-", 358 "nvptx-", "mips-", "lanai-", "hexagon-", "bpf-", "avr-", 359 "thumb2-", "arm-", "si-", "gcn-", "amdgpu-", "aarch64-", 360 "amdgcn-", "polly-", "riscv-", "dxil-"}; 361 std::vector<StringRef> PassNameContain = {"ehprepare"}; 362 std::vector<StringRef> PassNameExact = { 363 "safe-stack", 364 "cost-model", 365 "codegenprepare", 366 "interleaved-load-combine", 367 "unreachableblockelim", 368 "verify-safepoint-ir", 369 "atomic-expand", 370 "expandvp", 371 "hardware-loops", 372 "mve-tail-predication", 373 "interleaved-access", 374 "global-merge", 375 "pre-isel-intrinsic-lowering", 376 "expand-reductions", 377 "indirectbr-expand", 378 "generic-to-nvvm", 379 "expandmemcmp", 380 "loop-reduce", 381 "lower-amx-type", 382 "pre-amx-config", 383 "lower-amx-intrinsics", 384 "polyhedral-info", 385 "print-polyhedral-info", 386 "replace-with-veclib", 387 "jmc-instrument", 388 "dot-regions", 389 "dot-regions-only", 390 "view-regions", 391 "view-regions-only", 392 "select-optimize", 393 "expand-large-div-rem", 394 "structurizecfg", 395 "fix-irreducible", 396 "expand-large-fp-convert" 397 }; 398 for (const auto &P : PassNamePrefix) 399 if (Pass.startswith(P)) 400 return true; 401 for (const auto &P : PassNameContain) 402 if (Pass.contains(P)) 403 return true; 404 return llvm::is_contained(PassNameExact, Pass); 405 } 406 407 // For use in NPM transition. 408 static bool shouldForceLegacyPM() { 409 for (const auto &P : PassList) { 410 StringRef Arg = P->getPassArgument(); 411 if (shouldPinPassToLegacyPM(Arg)) 412 return true; 413 } 414 return false; 415 } 416 417 //===----------------------------------------------------------------------===// 418 // main for opt 419 // 420 int main(int argc, char **argv) { 421 InitLLVM X(argc, argv); 422 423 // Enable debug stream buffering. 424 EnableDebugBuffering = true; 425 426 InitializeAllTargets(); 427 InitializeAllTargetMCs(); 428 InitializeAllAsmPrinters(); 429 InitializeAllAsmParsers(); 430 431 // Initialize passes 432 PassRegistry &Registry = *PassRegistry::getPassRegistry(); 433 initializeCore(Registry); 434 initializeScalarOpts(Registry); 435 initializeVectorization(Registry); 436 initializeIPO(Registry); 437 initializeAnalysis(Registry); 438 initializeTransformUtils(Registry); 439 initializeInstCombine(Registry); 440 initializeTarget(Registry); 441 // For codegen passes, only passes that do IR to IR transformation are 442 // supported. 443 initializeExpandLargeDivRemLegacyPassPass(Registry); 444 initializeExpandLargeFpConvertLegacyPassPass(Registry); 445 initializeExpandMemCmpPassPass(Registry); 446 initializeScalarizeMaskedMemIntrinLegacyPassPass(Registry); 447 initializeSelectOptimizePass(Registry); 448 initializeCodeGenPreparePass(Registry); 449 initializeAtomicExpandPass(Registry); 450 initializeRewriteSymbolsLegacyPassPass(Registry); 451 initializeWinEHPreparePass(Registry); 452 initializeDwarfEHPrepareLegacyPassPass(Registry); 453 initializeSafeStackLegacyPassPass(Registry); 454 initializeSjLjEHPreparePass(Registry); 455 initializePreISelIntrinsicLoweringLegacyPassPass(Registry); 456 initializeGlobalMergePass(Registry); 457 initializeIndirectBrExpandPassPass(Registry); 458 initializeInterleavedLoadCombinePass(Registry); 459 initializeInterleavedAccessPass(Registry); 460 initializeUnreachableBlockElimLegacyPassPass(Registry); 461 initializeExpandReductionsPass(Registry); 462 initializeExpandVectorPredicationPass(Registry); 463 initializeWasmEHPreparePass(Registry); 464 initializeWriteBitcodePassPass(Registry); 465 initializeHardwareLoopsPass(Registry); 466 initializeReplaceWithVeclibLegacyPass(Registry); 467 initializeJMCInstrumenterPass(Registry); 468 469 SmallVector<PassPlugin, 1> PluginList; 470 PassPlugins.setCallback([&](const std::string &PluginPath) { 471 auto Plugin = PassPlugin::Load(PluginPath); 472 if (!Plugin) { 473 errs() << "Failed to load passes from '" << PluginPath 474 << "'. Request ignored.\n"; 475 return; 476 } 477 PluginList.emplace_back(Plugin.get()); 478 }); 479 480 // Register the Target and CPU printer for --version. 481 cl::AddExtraVersionPrinter(sys::printDefaultTargetAndDetectedCPU); 482 483 cl::ParseCommandLineOptions(argc, argv, 484 "llvm .bc -> .bc modular optimizer and analysis printer\n"); 485 486 LLVMContext Context; 487 488 // If `-passes=` is specified, use NPM. 489 // If `-enable-new-pm` is specified and there are no codegen passes, use NPM. 490 // e.g. `-enable-new-pm -sroa` will use NPM. 491 // but `-enable-new-pm -codegenprepare` will still revert to legacy PM. 492 const bool UseNPM = (EnableNewPassManager && !shouldForceLegacyPM()) || 493 PassPipeline.getNumOccurrences() > 0; 494 495 if (UseNPM && !PassList.empty()) { 496 errs() << "The `opt -passname` syntax for the new pass manager is " 497 "not supported, please use `opt -passes=<pipeline>` (or the `-p` " 498 "alias for a more concise version).\n"; 499 errs() << "See https://llvm.org/docs/NewPassManager.html#invoking-opt " 500 "for more details on the pass pipeline syntax.\n\n"; 501 return 1; 502 } 503 504 if (!UseNPM && PluginList.size()) { 505 errs() << argv[0] << ": " << PassPlugins.ArgStr 506 << " specified with legacy PM.\n"; 507 return 1; 508 } 509 510 // FIXME: once the legacy PM code is deleted, move runPassPipeline() here and 511 // construct the PassBuilder before parsing IR so we can reuse the same 512 // PassBuilder for print passes. 513 if (PrintPasses) { 514 printPasses(outs()); 515 return 0; 516 } 517 518 TimeTracerRAII TimeTracer(argv[0]); 519 520 SMDiagnostic Err; 521 522 Context.setDiscardValueNames(DiscardValueNames); 523 if (!DisableDITypeMap) 524 Context.enableDebugTypeODRUniquing(); 525 526 Expected<std::unique_ptr<ToolOutputFile>> RemarksFileOrErr = 527 setupLLVMOptimizationRemarks(Context, RemarksFilename, RemarksPasses, 528 RemarksFormat, RemarksWithHotness, 529 RemarksHotnessThreshold); 530 if (Error E = RemarksFileOrErr.takeError()) { 531 errs() << toString(std::move(E)) << '\n'; 532 return 1; 533 } 534 std::unique_ptr<ToolOutputFile> RemarksFile = std::move(*RemarksFileOrErr); 535 536 // Load the input module... 537 auto SetDataLayout = [](StringRef, StringRef) -> std::optional<std::string> { 538 if (ClDataLayout.empty()) 539 return std::nullopt; 540 return ClDataLayout; 541 }; 542 std::unique_ptr<Module> M; 543 if (NoUpgradeDebugInfo) 544 M = parseAssemblyFileWithIndexNoUpgradeDebugInfo( 545 InputFilename, Err, Context, nullptr, SetDataLayout) 546 .Mod; 547 else 548 M = parseIRFile(InputFilename, Err, Context, 549 ParserCallbacks(SetDataLayout)); 550 551 if (!M) { 552 Err.print(argv[0], errs()); 553 return 1; 554 } 555 556 // Strip debug info before running the verifier. 557 if (StripDebug) 558 StripDebugInfo(*M); 559 560 // Erase module-level named metadata, if requested. 561 if (StripNamedMetadata) { 562 while (!M->named_metadata_empty()) { 563 NamedMDNode *NMD = &*M->named_metadata_begin(); 564 M->eraseNamedMetadata(NMD); 565 } 566 } 567 568 // If we are supposed to override the target triple or data layout, do so now. 569 if (!TargetTriple.empty()) 570 M->setTargetTriple(Triple::normalize(TargetTriple)); 571 572 // Immediately run the verifier to catch any problems before starting up the 573 // pass pipelines. Otherwise we can crash on broken code during 574 // doInitialization(). 575 if (!NoVerify && verifyModule(*M, &errs())) { 576 errs() << argv[0] << ": " << InputFilename 577 << ": error: input module is broken!\n"; 578 return 1; 579 } 580 581 // Enable testing of whole program devirtualization on this module by invoking 582 // the facility for updating public visibility to linkage unit visibility when 583 // specified by an internal option. This is normally done during LTO which is 584 // not performed via opt. 585 updateVCallVisibilityInModule(*M, 586 /* WholeProgramVisibilityEnabledInLTO */ false, 587 /* DynamicExportSymbols */ {}); 588 589 // Figure out what stream we are supposed to write to... 590 std::unique_ptr<ToolOutputFile> Out; 591 std::unique_ptr<ToolOutputFile> ThinLinkOut; 592 if (NoOutput) { 593 if (!OutputFilename.empty()) 594 errs() << "WARNING: The -o (output filename) option is ignored when\n" 595 "the --disable-output option is used.\n"; 596 } else { 597 // Default to standard output. 598 if (OutputFilename.empty()) 599 OutputFilename = "-"; 600 601 std::error_code EC; 602 sys::fs::OpenFlags Flags = 603 OutputAssembly ? sys::fs::OF_TextWithCRLF : sys::fs::OF_None; 604 Out.reset(new ToolOutputFile(OutputFilename, EC, Flags)); 605 if (EC) { 606 errs() << EC.message() << '\n'; 607 return 1; 608 } 609 610 if (!ThinLinkBitcodeFile.empty()) { 611 ThinLinkOut.reset( 612 new ToolOutputFile(ThinLinkBitcodeFile, EC, sys::fs::OF_None)); 613 if (EC) { 614 errs() << EC.message() << '\n'; 615 return 1; 616 } 617 } 618 } 619 620 Triple ModuleTriple(M->getTargetTriple()); 621 std::string CPUStr, FeaturesStr; 622 TargetMachine *Machine = nullptr; 623 const TargetOptions Options = 624 codegen::InitTargetOptionsFromCodeGenFlags(ModuleTriple); 625 626 if (ModuleTriple.getArch()) { 627 CPUStr = codegen::getCPUStr(); 628 FeaturesStr = codegen::getFeaturesStr(); 629 Machine = GetTargetMachine(ModuleTriple, CPUStr, FeaturesStr, Options); 630 } else if (ModuleTriple.getArchName() != "unknown" && 631 ModuleTriple.getArchName() != "") { 632 errs() << argv[0] << ": unrecognized architecture '" 633 << ModuleTriple.getArchName() << "' provided.\n"; 634 return 1; 635 } 636 637 std::unique_ptr<TargetMachine> TM(Machine); 638 639 // Override function attributes based on CPUStr, FeaturesStr, and command line 640 // flags. 641 codegen::setFunctionAttributes(CPUStr, FeaturesStr, *M); 642 643 // If the output is set to be emitted to standard out, and standard out is a 644 // console, print out a warning message and refuse to do it. We don't 645 // impress anyone by spewing tons of binary goo to a terminal. 646 if (!Force && !NoOutput && !OutputAssembly) 647 if (CheckBitcodeOutputToConsole(Out->os())) 648 NoOutput = true; 649 650 if (OutputThinLTOBC) 651 M->addModuleFlag(Module::Error, "EnableSplitLTOUnit", SplitLTOUnit); 652 653 // Add an appropriate TargetLibraryInfo pass for the module's triple. 654 TargetLibraryInfoImpl TLII(ModuleTriple); 655 656 // The -disable-simplify-libcalls flag actually disables all builtin optzns. 657 if (DisableSimplifyLibCalls) 658 TLII.disableAllFunctions(); 659 else { 660 // Disable individual builtin functions in TargetLibraryInfo. 661 LibFunc F; 662 for (auto &FuncName : DisableBuiltins) 663 if (TLII.getLibFunc(FuncName, F)) 664 TLII.setUnavailable(F); 665 else { 666 errs() << argv[0] << ": cannot disable nonexistent builtin function " 667 << FuncName << '\n'; 668 return 1; 669 } 670 } 671 672 if (UseNPM) { 673 if (legacy::debugPassSpecified()) { 674 errs() 675 << "-debug-pass does not work with the new PM, either use " 676 "-debug-pass-manager, or use the legacy PM (-enable-new-pm=0)\n"; 677 return 1; 678 } 679 auto NumOLevel = OptLevelO0 + OptLevelO1 + OptLevelO2 + OptLevelO3 + 680 OptLevelOs + OptLevelOz; 681 if (NumOLevel > 1) { 682 errs() << "Cannot specify multiple -O#\n"; 683 return 1; 684 } 685 if (NumOLevel > 0 && (PassPipeline.getNumOccurrences() > 0)) { 686 errs() << "Cannot specify -O# and --passes=/--foo-pass, use " 687 "-passes='default<O#>,other-pass'\n"; 688 return 1; 689 } 690 std::string Pipeline = PassPipeline; 691 692 if (OptLevelO0) 693 Pipeline = "default<O0>"; 694 if (OptLevelO1) 695 Pipeline = "default<O1>"; 696 if (OptLevelO2) 697 Pipeline = "default<O2>"; 698 if (OptLevelO3) 699 Pipeline = "default<O3>"; 700 if (OptLevelOs) 701 Pipeline = "default<Os>"; 702 if (OptLevelOz) 703 Pipeline = "default<Oz>"; 704 OutputKind OK = OK_NoOutput; 705 if (!NoOutput) 706 OK = OutputAssembly 707 ? OK_OutputAssembly 708 : (OutputThinLTOBC ? OK_OutputThinLTOBitcode : OK_OutputBitcode); 709 710 VerifierKind VK = VK_VerifyOut; 711 if (NoVerify) 712 VK = VK_NoVerifier; 713 else if (VerifyEach) 714 VK = VK_VerifyEachPass; 715 716 // The user has asked to use the new pass manager and provided a pipeline 717 // string. Hand off the rest of the functionality to the new code for that 718 // layer. 719 return runPassPipeline(argv[0], *M, TM.get(), &TLII, Out.get(), 720 ThinLinkOut.get(), RemarksFile.get(), Pipeline, 721 PluginList, OK, VK, PreserveAssemblyUseListOrder, 722 PreserveBitcodeUseListOrder, EmitSummaryIndex, 723 EmitModuleHash, EnableDebugify, 724 VerifyDebugInfoPreserve) 725 ? 0 726 : 1; 727 } 728 729 if (OptLevelO0 || OptLevelO1 || OptLevelO2 || OptLevelOs || OptLevelOz || 730 OptLevelO3) { 731 errs() << "Cannot use -O# with legacy PM.\n"; 732 return 1; 733 } 734 if (EmitSummaryIndex) { 735 errs() << "Cannot use -module-summary with legacy PM.\n"; 736 return 1; 737 } 738 if (EmitModuleHash) { 739 errs() << "Cannot use -module-hash with legacy PM.\n"; 740 return 1; 741 } 742 if (OutputThinLTOBC) { 743 errs() << "Cannot use -thinlto-bc with legacy PM.\n"; 744 return 1; 745 } 746 // Create a PassManager to hold and optimize the collection of passes we are 747 // about to build. If the -debugify-each option is set, wrap each pass with 748 // the (-check)-debugify passes. 749 DebugifyCustomPassManager Passes; 750 DebugifyStatsMap DIStatsMap; 751 DebugInfoPerPass DebugInfoBeforePass; 752 if (DebugifyEach) { 753 Passes.setDebugifyMode(DebugifyMode::SyntheticDebugInfo); 754 Passes.setDIStatsMap(DIStatsMap); 755 } else if (VerifyEachDebugInfoPreserve) { 756 Passes.setDebugifyMode(DebugifyMode::OriginalDebugInfo); 757 Passes.setDebugInfoBeforePass(DebugInfoBeforePass); 758 if (!VerifyDIPreserveExport.empty()) 759 Passes.setOrigDIVerifyBugsReportFilePath(VerifyDIPreserveExport); 760 } 761 762 bool AddOneTimeDebugifyPasses = 763 (EnableDebugify && !DebugifyEach) || 764 (VerifyDebugInfoPreserve && !VerifyEachDebugInfoPreserve); 765 766 Passes.add(new TargetLibraryInfoWrapperPass(TLII)); 767 768 // Add internal analysis passes from the target machine. 769 Passes.add(createTargetTransformInfoWrapperPass(TM ? TM->getTargetIRAnalysis() 770 : TargetIRAnalysis())); 771 772 if (AddOneTimeDebugifyPasses) { 773 if (EnableDebugify) { 774 Passes.setDIStatsMap(DIStatsMap); 775 Passes.add(createDebugifyModulePass()); 776 } else if (VerifyDebugInfoPreserve) { 777 Passes.setDebugInfoBeforePass(DebugInfoBeforePass); 778 Passes.add(createDebugifyModulePass( 779 DebugifyMode::OriginalDebugInfo, "", 780 &(Passes.getDebugInfoPerPass()))); 781 } 782 } 783 784 std::unique_ptr<legacy::FunctionPassManager> FPasses; 785 786 if (PrintBreakpoints) { 787 // Default to standard output. 788 if (!Out) { 789 if (OutputFilename.empty()) 790 OutputFilename = "-"; 791 792 std::error_code EC; 793 Out = std::make_unique<ToolOutputFile>(OutputFilename, EC, 794 sys::fs::OF_None); 795 if (EC) { 796 errs() << EC.message() << '\n'; 797 return 1; 798 } 799 } 800 Passes.add(createBreakpointPrinter(Out->os())); 801 NoOutput = true; 802 } 803 804 if (TM) { 805 // FIXME: We should dyn_cast this when supported. 806 auto <M = static_cast<LLVMTargetMachine &>(*TM); 807 Pass *TPC = LTM.createPassConfig(Passes); 808 Passes.add(TPC); 809 } 810 811 // Create a new optimization pass for each one specified on the command line 812 for (unsigned i = 0; i < PassList.size(); ++i) { 813 const PassInfo *PassInf = PassList[i]; 814 Pass *P = nullptr; 815 if (PassInf->getNormalCtor()) 816 P = PassInf->getNormalCtor()(); 817 else 818 errs() << argv[0] << ": cannot create pass: " 819 << PassInf->getPassName() << "\n"; 820 if (P) 821 addPass(Passes, P); 822 } 823 824 if (FPasses) { 825 FPasses->doInitialization(); 826 for (Function &F : *M) 827 FPasses->run(F); 828 FPasses->doFinalization(); 829 } 830 831 // Check that the module is well formed on completion of optimization 832 if (!NoVerify && !VerifyEach) 833 Passes.add(createVerifierPass()); 834 835 if (AddOneTimeDebugifyPasses) { 836 if (EnableDebugify) 837 Passes.add(createCheckDebugifyModulePass(false)); 838 else if (VerifyDebugInfoPreserve) { 839 if (!VerifyDIPreserveExport.empty()) 840 Passes.setOrigDIVerifyBugsReportFilePath(VerifyDIPreserveExport); 841 Passes.add(createCheckDebugifyModulePass( 842 false, "", nullptr, DebugifyMode::OriginalDebugInfo, 843 &(Passes.getDebugInfoPerPass()), VerifyDIPreserveExport)); 844 } 845 } 846 847 // In run twice mode, we want to make sure the output is bit-by-bit 848 // equivalent if we run the pass manager again, so setup two buffers and 849 // a stream to write to them. Note that llc does something similar and it 850 // may be worth to abstract this out in the future. 851 SmallVector<char, 0> Buffer; 852 SmallVector<char, 0> FirstRunBuffer; 853 std::unique_ptr<raw_svector_ostream> BOS; 854 raw_ostream *OS = nullptr; 855 856 const bool ShouldEmitOutput = !NoOutput; 857 858 // Write bitcode or assembly to the output as the last step... 859 if (ShouldEmitOutput || RunTwice) { 860 assert(Out); 861 OS = &Out->os(); 862 if (RunTwice) { 863 BOS = std::make_unique<raw_svector_ostream>(Buffer); 864 OS = BOS.get(); 865 } 866 if (OutputAssembly) 867 Passes.add(createPrintModulePass(*OS, "", PreserveAssemblyUseListOrder)); 868 else 869 Passes.add(createBitcodeWriterPass(*OS, PreserveBitcodeUseListOrder)); 870 } 871 872 // Before executing passes, print the final values of the LLVM options. 873 cl::PrintOptionValues(); 874 875 if (!RunTwice) { 876 // Now that we have all of the passes ready, run them. 877 Passes.run(*M); 878 } else { 879 // If requested, run all passes twice with the same pass manager to catch 880 // bugs caused by persistent state in the passes. 881 std::unique_ptr<Module> M2(CloneModule(*M)); 882 // Run all passes on the original module first, so the second run processes 883 // the clone to catch CloneModule bugs. 884 Passes.run(*M); 885 FirstRunBuffer = Buffer; 886 Buffer.clear(); 887 888 Passes.run(*M2); 889 890 // Compare the two outputs and make sure they're the same 891 assert(Out); 892 if (Buffer.size() != FirstRunBuffer.size() || 893 (memcmp(Buffer.data(), FirstRunBuffer.data(), Buffer.size()) != 0)) { 894 errs() 895 << "Running the pass manager twice changed the output.\n" 896 "Writing the result of the second run to the specified output.\n" 897 "To generate the one-run comparison binary, just run without\n" 898 "the compile-twice option\n"; 899 if (ShouldEmitOutput) { 900 Out->os() << BOS->str(); 901 Out->keep(); 902 } 903 if (RemarksFile) 904 RemarksFile->keep(); 905 return 1; 906 } 907 if (ShouldEmitOutput) 908 Out->os() << BOS->str(); 909 } 910 911 if (DebugifyEach && !DebugifyExport.empty()) 912 exportDebugifyStats(DebugifyExport, Passes.getDebugifyStatsMap()); 913 914 // Declare success. 915 if (!NoOutput || PrintBreakpoints) 916 Out->keep(); 917 918 if (RemarksFile) 919 RemarksFile->keep(); 920 921 if (ThinLinkOut) 922 ThinLinkOut->keep(); 923 924 return 0; 925 } 926