tools/llvm-mca/llvm-mca.cpp

   1 //===-- llvm-mca.cpp - Machine Code Analyzer -------------------*- C++ -* -===//
   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 utility is a simple driver that allows static performance analysis on
  10 // machine code similarly to how IACA (Intel Architecture Code Analyzer) works.
  11 //
  12 //   llvm-mca [options] <file-name>
  13 //      -march <type>
  14 //      -mcpu <cpu>
  15 //      -o <file>
  16 //
  17 // The target defaults to the host target.
  18 // The cpu defaults to the 'native' host cpu.
  19 // The output defaults to standard output.
  20 //
  21 //===----------------------------------------------------------------------===//
  22
  23 #include "CodeRegion.h"
  24 #include "CodeRegionGenerator.h"
  25 #include "PipelinePrinter.h"
  26 #include "Views/BottleneckAnalysis.h"
  27 #include "Views/DispatchStatistics.h"
  28 #include "Views/InstructionInfoView.h"
  29 #include "Views/RegisterFileStatistics.h"
  30 #include "Views/ResourcePressureView.h"
  31 #include "Views/RetireControlUnitStatistics.h"
  32 #include "Views/SchedulerStatistics.h"
  33 #include "Views/SummaryView.h"
  34 #include "Views/TimelineView.h"
  35 #include "llvm/MC/MCAsmInfo.h"
  36 #include "llvm/MC/MCContext.h"
  37 #include "llvm/MC/MCCodeEmitter.h"
  38 #include "llvm/MC/MCObjectFileInfo.h"
  39 #include "llvm/MC/MCRegisterInfo.h"
  40 #include "llvm/MC/MCSubtargetInfo.h"
  41 #include "llvm/MCA/Context.h"
  42 #include "llvm/MCA/InstrBuilder.h"
  43 #include "llvm/MCA/Pipeline.h"
  44 #include "llvm/MCA/Stages/EntryStage.h"
  45 #include "llvm/MCA/Stages/InstructionTables.h"
  46 #include "llvm/MCA/Support.h"
  47 #include "llvm/Support/CommandLine.h"
  48 #include "llvm/Support/ErrorHandling.h"
  49 #include "llvm/Support/ErrorOr.h"
  50 #include "llvm/Support/FileSystem.h"
  51 #include "llvm/Support/Host.h"
  52 #include "llvm/Support/InitLLVM.h"
  53 #include "llvm/Support/MemoryBuffer.h"
  54 #include "llvm/Support/SourceMgr.h"
  55 #include "llvm/Support/TargetRegistry.h"
  56 #include "llvm/Support/TargetSelect.h"
  57 #include "llvm/Support/ToolOutputFile.h"
  58 #include "llvm/Support/WithColor.h"
  59
  60 using namespace llvm;
  61
  62 static cl::OptionCategory ToolOptions("Tool Options");
  63 static cl::OptionCategory ViewOptions("View Options");
  64
  65 static cl::opt<std::string> InputFilename(cl::Positional,
  66                                           cl::desc("<input file>"),
  67                                           cl::cat(ToolOptions), cl::init("-"));
  68
  69 static cl::opt<std::string> OutputFilename("o", cl::desc("Output filename"),
  70                                            cl::init("-"), cl::cat(ToolOptions),
  71                                            cl::value_desc("filename"));
  72
  73 static cl::opt<std::string>
  74     ArchName("march",
  75              cl::desc("Target architecture. "
  76                       "See -version for available targets"),
  77              cl::cat(ToolOptions));
  78
  79 static cl::opt<std::string>
  80     TripleName("mtriple",
  81                cl::desc("Target triple. See -version for available targets"),
  82                cl::cat(ToolOptions));
  83
  84 static cl::opt<std::string>
  85     MCPU("mcpu",
  86          cl::desc("Target a specific cpu type (-mcpu=help for details)"),
  87          cl::value_desc("cpu-name"), cl::cat(ToolOptions), cl::init("native"));
  88
  89 static cl::opt<int>
  90     OutputAsmVariant("output-asm-variant",
  91                      cl::desc("Syntax variant to use for output printing"),
  92                      cl::cat(ToolOptions), cl::init(-1));
  93
  94 static cl::opt<bool>
  95     PrintImmHex("print-imm-hex", cl::cat(ToolOptions), cl::init(false),
  96                 cl::desc("Prefer hex format when printing immediate values"));
  97
  98 static cl::opt<unsigned> Iterations("iterations",
  99                                     cl::desc("Number of iterations to run"),
 100                                     cl::cat(ToolOptions), cl::init(0));
 101
 102 static cl::opt<unsigned>
 103     DispatchWidth("dispatch", cl::desc("Override the processor dispatch width"),
 104                   cl::cat(ToolOptions), cl::init(0));
 105
 106 static cl::opt<unsigned>
 107     RegisterFileSize("register-file-size",
 108                      cl::desc("Maximum number of physical registers which can "
 109                               "be used for register mappings"),
 110                      cl::cat(ToolOptions), cl::init(0));
 111
 112 static cl::opt<unsigned>
 113     MicroOpQueue("micro-op-queue-size", cl::Hidden,
 114                  cl::desc("Number of entries in the micro-op queue"),
 115                  cl::cat(ToolOptions), cl::init(0));
 116
 117 static cl::opt<unsigned>
 118     DecoderThroughput("decoder-throughput", cl::Hidden,
 119                       cl::desc("Maximum throughput from the decoders "
 120                                "(instructions per cycle)"),
 121                       cl::cat(ToolOptions), cl::init(0));
 122
 123 static cl::opt<bool>
 124     PrintRegisterFileStats("register-file-stats",
 125                            cl::desc("Print register file statistics"),
 126                            cl::cat(ViewOptions), cl::init(false));
 127
 128 static cl::opt<bool> PrintDispatchStats("dispatch-stats",
 129                                         cl::desc("Print dispatch statistics"),
 130                                         cl::cat(ViewOptions), cl::init(false));
 131
 132 static cl::opt<bool>
 133     PrintSummaryView("summary-view", cl::Hidden,
 134                      cl::desc("Print summary view (enabled by default)"),
 135                      cl::cat(ViewOptions), cl::init(true));
 136
 137 static cl::opt<bool> PrintSchedulerStats("scheduler-stats",
 138                                          cl::desc("Print scheduler statistics"),
 139                                          cl::cat(ViewOptions), cl::init(false));
 140
 141 static cl::opt<bool>
 142     PrintRetireStats("retire-stats",
 143                      cl::desc("Print retire control unit statistics"),
 144                      cl::cat(ViewOptions), cl::init(false));
 145
 146 static cl::opt<bool> PrintResourcePressureView(
 147     "resource-pressure",
 148     cl::desc("Print the resource pressure view (enabled by default)"),
 149     cl::cat(ViewOptions), cl::init(true));
 150
 151 static cl::opt<bool> PrintTimelineView("timeline",
 152                                        cl::desc("Print the timeline view"),
 153                                        cl::cat(ViewOptions), cl::init(false));
 154
 155 static cl::opt<unsigned> TimelineMaxIterations(
 156     "timeline-max-iterations",
 157     cl::desc("Maximum number of iterations to print in timeline view"),
 158     cl::cat(ViewOptions), cl::init(0));
 159
 160 static cl::opt<unsigned> TimelineMaxCycles(
 161     "timeline-max-cycles",
 162     cl::desc(
 163         "Maximum number of cycles in the timeline view. Defaults to 80 cycles"),
 164     cl::cat(ViewOptions), cl::init(80));
 165
 166 static cl::opt<bool>
 167     AssumeNoAlias("noalias",
 168                   cl::desc("If set, assume that loads and stores do not alias"),
 169                   cl::cat(ToolOptions), cl::init(true));
 170
 171 static cl::opt<unsigned> LoadQueueSize("lqueue",
 172                                        cl::desc("Size of the load queue"),
 173                                        cl::cat(ToolOptions), cl::init(0));
 174
 175 static cl::opt<unsigned> StoreQueueSize("squeue",
 176                                         cl::desc("Size of the store queue"),
 177                                         cl::cat(ToolOptions), cl::init(0));
 178
 179 static cl::opt<bool>
 180     PrintInstructionTables("instruction-tables",
 181                            cl::desc("Print instruction tables"),
 182                            cl::cat(ToolOptions), cl::init(false));
 183
 184 static cl::opt<bool> PrintInstructionInfoView(
 185     "instruction-info",
 186     cl::desc("Print the instruction info view (enabled by default)"),
 187     cl::cat(ViewOptions), cl::init(true));
 188
 189 static cl::opt<bool> EnableAllStats("all-stats",
 190                                     cl::desc("Print all hardware statistics"),
 191                                     cl::cat(ViewOptions), cl::init(false));
 192
 193 static cl::opt<bool>
 194     EnableAllViews("all-views",
 195                    cl::desc("Print all views including hardware statistics"),
 196                    cl::cat(ViewOptions), cl::init(false));
 197
 198 static cl::opt<bool> EnableBottleneckAnalysis(
 199     "bottleneck-analysis",
 200     cl::desc("Enable bottleneck analysis (disabled by default)"),
 201     cl::cat(ViewOptions), cl::init(false));
 202
 203 namespace {
 204
 205 const Target *getTarget(const char *ProgName) {
 206   if (TripleName.empty())
 207     TripleName = Triple::normalize(sys::getDefaultTargetTriple());
 208   Triple TheTriple(TripleName);
 209
 210   // Get the target specific parser.
 211   std::string Error;
 212   const Target *TheTarget =
 213       TargetRegistry::lookupTarget(ArchName, TheTriple, Error);
 214   if (!TheTarget) {
 215     errs() << ProgName << ": " << Error;
 216     return nullptr;
 217   }
 218
 219   // Return the found target.
 220   return TheTarget;
 221 }
 222
 223 ErrorOr<std::unique_ptr<ToolOutputFile>> getOutputStream() {
 224   if (OutputFilename == "")
 225     OutputFilename = "-";
 226   std::error_code EC;
 227   auto Out =
 228       llvm::make_unique<ToolOutputFile>(OutputFilename, EC, sys::fs::OF_None);
 229   if (!EC)
 230     return std::move(Out);
 231   return EC;
 232 }
 233 } // end of anonymous namespace
 234
 235 static void processOptionImpl(cl::opt<bool> &O, const cl::opt<bool> &Default) {
 236   if (!O.getNumOccurrences() || O.getPosition() < Default.getPosition())
 237     O = Default.getValue();
 238 }
 239
 240 static void processViewOptions() {
 241   if (!EnableAllViews.getNumOccurrences() &&
 242       !EnableAllStats.getNumOccurrences())
 243     return;
 244
 245   if (EnableAllViews.getNumOccurrences()) {
 246     processOptionImpl(PrintSummaryView, EnableAllViews);
 247     processOptionImpl(EnableBottleneckAnalysis, EnableAllViews);
 248     processOptionImpl(PrintResourcePressureView, EnableAllViews);
 249     processOptionImpl(PrintTimelineView, EnableAllViews);
 250     processOptionImpl(PrintInstructionInfoView, EnableAllViews);
 251   }
 252
 253   const cl::opt<bool> &Default =
 254       EnableAllViews.getPosition() < EnableAllStats.getPosition()
 255           ? EnableAllStats
 256           : EnableAllViews;
 257   processOptionImpl(PrintRegisterFileStats, Default);
 258   processOptionImpl(PrintDispatchStats, Default);
 259   processOptionImpl(PrintSchedulerStats, Default);
 260   processOptionImpl(PrintRetireStats, Default);
 261 }
 262
 263 // Returns true on success.
 264 static bool runPipeline(mca::Pipeline &P) {
 265   // Handle pipeline errors here.
 266   Expected<unsigned> Cycles = P.run();
 267   if (!Cycles) {
 268     WithColor::error() << toString(Cycles.takeError());
 269     return false;
 270   }
 271   return true;
 272 }
 273
 274 int main(int argc, char **argv) {
 275   InitLLVM X(argc, argv);
 276
 277   // Initialize targets and assembly parsers.
 278   InitializeAllTargetInfos();
 279   InitializeAllTargetMCs();
 280   InitializeAllAsmParsers();
 281
 282   // Enable printing of available targets when flag --version is specified.
 283   cl::AddExtraVersionPrinter(TargetRegistry::printRegisteredTargetsForVersion);
 284
 285   cl::HideUnrelatedOptions({&ToolOptions, &ViewOptions});
 286
 287   // Parse flags and initialize target options.
 288   cl::ParseCommandLineOptions(argc, argv,
 289                               "llvm machine code performance analyzer.\n");
 290
 291   // Get the target from the triple. If a triple is not specified, then select
 292   // the default triple for the host. If the triple doesn't correspond to any
 293   // registered target, then exit with an error message.
 294   const char *ProgName = argv[0];
 295   const Target *TheTarget = getTarget(ProgName);
 296   if (!TheTarget)
 297     return 1;
 298
 299   // GetTarget() may replaced TripleName with a default triple.
 300   // For safety, reconstruct the Triple object.
 301   Triple TheTriple(TripleName);
 302
 303   ErrorOr<std::unique_ptr<MemoryBuffer>> BufferPtr =
 304       MemoryBuffer::getFileOrSTDIN(InputFilename);
 305   if (std::error_code EC = BufferPtr.getError()) {
 306     WithColor::error() << InputFilename << ": " << EC.message() << '\n';
 307     return 1;
 308   }
 309
 310   // Apply overrides to llvm-mca specific options.
 311   processViewOptions();
 312
 313   if (!MCPU.compare("native"))
 314     MCPU = llvm::sys::getHostCPUName();
 315
 316   std::unique_ptr<MCSubtargetInfo> STI(
 317       TheTarget->createMCSubtargetInfo(TripleName, MCPU, /* FeaturesStr */ ""));
 318   if (!STI->isCPUStringValid(MCPU))
 319     return 1;
 320
 321   if (!PrintInstructionTables && !STI->getSchedModel().isOutOfOrder()) {
 322     WithColor::error() << "please specify an out-of-order cpu. '" << MCPU
 323                        << "' is an in-order cpu.\n";
 324     return 1;
 325   }
 326
 327   if (!STI->getSchedModel().hasInstrSchedModel()) {
 328     WithColor::error()
 329         << "unable to find instruction-level scheduling information for"
 330         << " target triple '" << TheTriple.normalize() << "' and cpu '" << MCPU
 331         << "'.\n";
 332
 333     if (STI->getSchedModel().InstrItineraries)
 334       WithColor::note()
 335           << "cpu '" << MCPU << "' provides itineraries. However, "
 336           << "instruction itineraries are currently unsupported.\n";
 337     return 1;
 338   }
 339
 340   std::unique_ptr<MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TripleName));
 341   assert(MRI && "Unable to create target register info!");
 342
 343   std::unique_ptr<MCAsmInfo> MAI(TheTarget->createMCAsmInfo(*MRI, TripleName));
 344   assert(MAI && "Unable to create target asm info!");
 345
 346   MCObjectFileInfo MOFI;
 347   SourceMgr SrcMgr;
 348
 349   // Tell SrcMgr about this buffer, which is what the parser will pick up.
 350   SrcMgr.AddNewSourceBuffer(std::move(*BufferPtr), SMLoc());
 351
 352   MCContext Ctx(MAI.get(), MRI.get(), &MOFI, &SrcMgr);
 353
 354   MOFI.InitMCObjectFileInfo(TheTriple, /* PIC= */ false, Ctx);
 355
 356   std::unique_ptr<buffer_ostream> BOS;
 357
 358   std::unique_ptr<MCInstrInfo> MCII(TheTarget->createMCInstrInfo());
 359
 360   std::unique_ptr<MCInstrAnalysis> MCIA(
 361       TheTarget->createMCInstrAnalysis(MCII.get()));
 362
 363   // Parse the input and create CodeRegions that llvm-mca can analyze.
 364   mca::AsmCodeRegionGenerator CRG(*TheTarget, SrcMgr, Ctx, *MAI, *STI, *MCII);
 365   Expected<const mca::CodeRegions &> RegionsOrErr = CRG.parseCodeRegions();
 366   if (!RegionsOrErr) {
 367     if (auto Err =
 368             handleErrors(RegionsOrErr.takeError(), [](const StringError &E) {
 369               WithColor::error() << E.getMessage() << '\n';
 370             })) {
 371       // Default case.
 372       WithColor::error() << toString(std::move(Err)) << '\n';
 373     }
 374     return 1;
 375   }
 376   const mca::CodeRegions &Regions = *RegionsOrErr;
 377
 378   // Early exit if errors were found by the code region parsing logic.
 379   if (!Regions.isValid())
 380     return 1;
 381
 382   if (Regions.empty()) {
 383     WithColor::error() << "no assembly instructions found.\n";
 384     return 1;
 385   }
 386
 387   // Now initialize the output file.
 388   auto OF = getOutputStream();
 389   if (std::error_code EC = OF.getError()) {
 390     WithColor::error() << EC.message() << '\n';
 391     return 1;
 392   }
 393
 394   unsigned AssemblerDialect = CRG.getAssemblerDialect();
 395   if (OutputAsmVariant >= 0)
 396     AssemblerDialect = static_cast<unsigned>(OutputAsmVariant);
 397   std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter(
 398       Triple(TripleName), AssemblerDialect, *MAI, *MCII, *MRI));
 399   if (!IP) {
 400     WithColor::error()
 401         << "unable to create instruction printer for target triple '"
 402         << TheTriple.normalize() << "' with assembly variant "
 403         << AssemblerDialect << ".\n";
 404     return 1;
 405   }
 406
 407   // Set the display preference for hex vs. decimal immediates.
 408   IP->setPrintImmHex(PrintImmHex);
 409
 410   std::unique_ptr<ToolOutputFile> TOF = std::move(*OF);
 411
 412   const MCSchedModel &SM = STI->getSchedModel();
 413
 414   // Create an instruction builder.
 415   mca::InstrBuilder IB(*STI, *MCII, *MRI, MCIA.get());
 416
 417   // Create a context to control ownership of the pipeline hardware.
 418   mca::Context MCA(*MRI, *STI);
 419
 420   mca::PipelineOptions PO(MicroOpQueue, DecoderThroughput, DispatchWidth,
 421                           RegisterFileSize, LoadQueueSize, StoreQueueSize,
 422                           AssumeNoAlias, EnableBottleneckAnalysis);
 423
 424   // Number each region in the sequence.
 425   unsigned RegionIdx = 0;
 426
 427   for (const std::unique_ptr<mca::CodeRegion> &Region : Regions) {
 428     // Skip empty code regions.
 429     if (Region->empty())
 430       continue;
 431
 432     // Don't print the header of this region if it is the default region, and
 433     // it doesn't have an end location.
 434     if (Region->startLoc().isValid() || Region->endLoc().isValid()) {
 435       TOF->os() << "\n[" << RegionIdx++ << "] Code Region";
 436       StringRef Desc = Region->getDescription();
 437       if (!Desc.empty())
 438         TOF->os() << " - " << Desc;
 439       TOF->os() << "\n\n";
 440     }
 441
 442     // Lower the MCInst sequence into an mca::Instruction sequence.
 443     ArrayRef<MCInst> Insts = Region->getInstructions();
 444     std::vector<std::unique_ptr<mca::Instruction>> LoweredSequence;
 445     for (const MCInst &MCI : Insts) {
 446       Expected<std::unique_ptr<mca::Instruction>> Inst =
 447           IB.createInstruction(MCI);
 448       if (!Inst) {
 449         if (auto NewE = handleErrors(
 450                 Inst.takeError(),
 451                 [&IP, &STI](const mca::InstructionError<MCInst> &IE) {
 452                   std::string InstructionStr;
 453                   raw_string_ostream SS(InstructionStr);
 454                   WithColor::error() << IE.Message << '\n';
 455                   IP->printInst(&IE.Inst, SS, "", *STI);
 456                   SS.flush();
 457                   WithColor::note()
 458                       << "instruction: " << InstructionStr << '\n';
 459                 })) {
 460           // Default case.
 461           WithColor::error() << toString(std::move(NewE));
 462         }
 463         return 1;
 464       }
 465
 466       LoweredSequence.emplace_back(std::move(Inst.get()));
 467     }
 468
 469     mca::SourceMgr S(LoweredSequence, PrintInstructionTables ? 1 : Iterations);
 470
 471     if (PrintInstructionTables) {
 472       //  Create a pipeline, stages, and a printer.
 473       auto P = llvm::make_unique<mca::Pipeline>();
 474       P->appendStage(llvm::make_unique<mca::EntryStage>(S));
 475       P->appendStage(llvm::make_unique<mca::InstructionTables>(SM));
 476       mca::PipelinePrinter Printer(*P);
 477
 478       // Create the views for this pipeline, execute, and emit a report.
 479       if (PrintInstructionInfoView) {
 480         Printer.addView(llvm::make_unique<mca::InstructionInfoView>(
 481             *STI, *MCII, Insts, *IP));
 482       }
 483       Printer.addView(
 484           llvm::make_unique<mca::ResourcePressureView>(*STI, *IP, Insts));
 485
 486       if (!runPipeline(*P))
 487         return 1;
 488
 489       Printer.printReport(TOF->os());
 490       continue;
 491     }
 492
 493     // Create a basic pipeline simulating an out-of-order backend.
 494     auto P = MCA.createDefaultPipeline(PO, S);
 495     mca::PipelinePrinter Printer(*P);
 496
 497     if (PrintSummaryView)
 498       Printer.addView(
 499           llvm::make_unique<mca::SummaryView>(SM, Insts, DispatchWidth));
 500
 501     if (EnableBottleneckAnalysis) {
 502       Printer.addView(llvm::make_unique<mca::BottleneckAnalysis>(
 503           *STI, *IP, Insts, S.getNumIterations()));
 504     }
 505
 506     if (PrintInstructionInfoView)
 507       Printer.addView(
 508           llvm::make_unique<mca::InstructionInfoView>(*STI, *MCII, Insts, *IP));
 509
 510     if (PrintDispatchStats)
 511       Printer.addView(llvm::make_unique<mca::DispatchStatistics>());
 512
 513     if (PrintSchedulerStats)
 514       Printer.addView(llvm::make_unique<mca::SchedulerStatistics>(*STI));
 515
 516     if (PrintRetireStats)
 517       Printer.addView(llvm::make_unique<mca::RetireControlUnitStatistics>(SM));
 518
 519     if (PrintRegisterFileStats)
 520       Printer.addView(llvm::make_unique<mca::RegisterFileStatistics>(*STI));
 521
 522     if (PrintResourcePressureView)
 523       Printer.addView(
 524           llvm::make_unique<mca::ResourcePressureView>(*STI, *IP, Insts));
 525
 526     if (PrintTimelineView) {
 527       unsigned TimelineIterations =
 528           TimelineMaxIterations ? TimelineMaxIterations : 10;
 529       Printer.addView(llvm::make_unique<mca::TimelineView>(
 530           *STI, *IP, Insts, std::min(TimelineIterations, S.getNumIterations()),
 531           TimelineMaxCycles));
 532     }
 533
 534     if (!runPipeline(*P))
 535       return 1;
 536
 537     Printer.printReport(TOF->os());
 538
 539     // Clear the InstrBuilder internal state in preparation for another round.
 540     IB.clear();
 541   }
 542
 543   TOF->keep();
 544   return 0;
 545 }