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/MCAsmBackend.h"
  36 #include "llvm/MC/MCAsmInfo.h"
  37 #include "llvm/MC/MCCodeEmitter.h"
  38 #include "llvm/MC/MCContext.h"
  39 #include "llvm/MC/MCObjectFileInfo.h"
  40 #include "llvm/MC/MCRegisterInfo.h"
  41 #include "llvm/MC/MCSubtargetInfo.h"
  42 #include "llvm/MC/MCTargetOptionsCommandFlags.inc"
  43 #include "llvm/MCA/CodeEmitter.h"
  44 #include "llvm/MCA/Context.h"
  45 #include "llvm/MCA/InstrBuilder.h"
  46 #include "llvm/MCA/Pipeline.h"
  47 #include "llvm/MCA/Stages/EntryStage.h"
  48 #include "llvm/MCA/Stages/InstructionTables.h"
  49 #include "llvm/MCA/Support.h"
  50 #include "llvm/Support/CommandLine.h"
  51 #include "llvm/Support/ErrorHandling.h"
  52 #include "llvm/Support/ErrorOr.h"
  53 #include "llvm/Support/FileSystem.h"
  54 #include "llvm/Support/Host.h"
  55 #include "llvm/Support/InitLLVM.h"
  56 #include "llvm/Support/MemoryBuffer.h"
  57 #include "llvm/Support/SourceMgr.h"
  58 #include "llvm/Support/TargetRegistry.h"
  59 #include "llvm/Support/TargetSelect.h"
  60 #include "llvm/Support/ToolOutputFile.h"
  61 #include "llvm/Support/WithColor.h"
  62
  63 using namespace llvm;
  64
  65 static cl::OptionCategory ToolOptions("Tool Options");
  66 static cl::OptionCategory ViewOptions("View Options");
  67
  68 static cl::opt<std::string> InputFilename(cl::Positional,
  69                                           cl::desc("<input file>"),
  70                                           cl::cat(ToolOptions), cl::init("-"));
  71
  72 static cl::opt<std::string> OutputFilename("o", cl::desc("Output filename"),
  73                                            cl::init("-"), cl::cat(ToolOptions),
  74                                            cl::value_desc("filename"));
  75
  76 static cl::opt<std::string>
  77     ArchName("march",
  78              cl::desc("Target architecture. "
  79                       "See -version for available targets"),
  80              cl::cat(ToolOptions));
  81
  82 static cl::opt<std::string>
  83     TripleName("mtriple",
  84                cl::desc("Target triple. See -version for available targets"),
  85                cl::cat(ToolOptions));
  86
  87 static cl::opt<std::string>
  88     MCPU("mcpu",
  89          cl::desc("Target a specific cpu type (-mcpu=help for details)"),
  90          cl::value_desc("cpu-name"), cl::cat(ToolOptions), cl::init("native"));
  91
  92 static cl::opt<std::string>
  93     MATTR("mattr",
  94           cl::desc("Additional target features."),
  95           cl::cat(ToolOptions));
  96
  97 static cl::opt<int>
  98     OutputAsmVariant("output-asm-variant",
  99                      cl::desc("Syntax variant to use for output printing"),
 100                      cl::cat(ToolOptions), cl::init(-1));
 101
 102 static cl::opt<bool>
 103     PrintImmHex("print-imm-hex", cl::cat(ToolOptions), cl::init(false),
 104                 cl::desc("Prefer hex format when printing immediate values"));
 105
 106 static cl::opt<unsigned> Iterations("iterations",
 107                                     cl::desc("Number of iterations to run"),
 108                                     cl::cat(ToolOptions), cl::init(0));
 109
 110 static cl::opt<unsigned>
 111     DispatchWidth("dispatch", cl::desc("Override the processor dispatch width"),
 112                   cl::cat(ToolOptions), cl::init(0));
 113
 114 static cl::opt<unsigned>
 115     RegisterFileSize("register-file-size",
 116                      cl::desc("Maximum number of physical registers which can "
 117                               "be used for register mappings"),
 118                      cl::cat(ToolOptions), cl::init(0));
 119
 120 static cl::opt<unsigned>
 121     MicroOpQueue("micro-op-queue-size", cl::Hidden,
 122                  cl::desc("Number of entries in the micro-op queue"),
 123                  cl::cat(ToolOptions), cl::init(0));
 124
 125 static cl::opt<unsigned>
 126     DecoderThroughput("decoder-throughput", cl::Hidden,
 127                       cl::desc("Maximum throughput from the decoders "
 128                                "(instructions per cycle)"),
 129                       cl::cat(ToolOptions), cl::init(0));
 130
 131 static cl::opt<bool>
 132     PrintRegisterFileStats("register-file-stats",
 133                            cl::desc("Print register file statistics"),
 134                            cl::cat(ViewOptions), cl::init(false));
 135
 136 static cl::opt<bool> PrintDispatchStats("dispatch-stats",
 137                                         cl::desc("Print dispatch statistics"),
 138                                         cl::cat(ViewOptions), cl::init(false));
 139
 140 static cl::opt<bool>
 141     PrintSummaryView("summary-view", cl::Hidden,
 142                      cl::desc("Print summary view (enabled by default)"),
 143                      cl::cat(ViewOptions), cl::init(true));
 144
 145 static cl::opt<bool> PrintSchedulerStats("scheduler-stats",
 146                                          cl::desc("Print scheduler statistics"),
 147                                          cl::cat(ViewOptions), cl::init(false));
 148
 149 static cl::opt<bool>
 150     PrintRetireStats("retire-stats",
 151                      cl::desc("Print retire control unit statistics"),
 152                      cl::cat(ViewOptions), cl::init(false));
 153
 154 static cl::opt<bool> PrintResourcePressureView(
 155     "resource-pressure",
 156     cl::desc("Print the resource pressure view (enabled by default)"),
 157     cl::cat(ViewOptions), cl::init(true));
 158
 159 static cl::opt<bool> PrintTimelineView("timeline",
 160                                        cl::desc("Print the timeline view"),
 161                                        cl::cat(ViewOptions), cl::init(false));
 162
 163 static cl::opt<unsigned> TimelineMaxIterations(
 164     "timeline-max-iterations",
 165     cl::desc("Maximum number of iterations to print in timeline view"),
 166     cl::cat(ViewOptions), cl::init(0));
 167
 168 static cl::opt<unsigned> TimelineMaxCycles(
 169     "timeline-max-cycles",
 170     cl::desc(
 171         "Maximum number of cycles in the timeline view. Defaults to 80 cycles"),
 172     cl::cat(ViewOptions), cl::init(80));
 173
 174 static cl::opt<bool>
 175     AssumeNoAlias("noalias",
 176                   cl::desc("If set, assume that loads and stores do not alias"),
 177                   cl::cat(ToolOptions), cl::init(true));
 178
 179 static cl::opt<unsigned> LoadQueueSize("lqueue",
 180                                        cl::desc("Size of the load queue"),
 181                                        cl::cat(ToolOptions), cl::init(0));
 182
 183 static cl::opt<unsigned> StoreQueueSize("squeue",
 184                                         cl::desc("Size of the store queue"),
 185                                         cl::cat(ToolOptions), cl::init(0));
 186
 187 static cl::opt<bool>
 188     PrintInstructionTables("instruction-tables",
 189                            cl::desc("Print instruction tables"),
 190                            cl::cat(ToolOptions), cl::init(false));
 191
 192 static cl::opt<bool> PrintInstructionInfoView(
 193     "instruction-info",
 194     cl::desc("Print the instruction info view (enabled by default)"),
 195     cl::cat(ViewOptions), cl::init(true));
 196
 197 static cl::opt<bool> EnableAllStats("all-stats",
 198                                     cl::desc("Print all hardware statistics"),
 199                                     cl::cat(ViewOptions), cl::init(false));
 200
 201 static cl::opt<bool>
 202     EnableAllViews("all-views",
 203                    cl::desc("Print all views including hardware statistics"),
 204                    cl::cat(ViewOptions), cl::init(false));
 205
 206 static cl::opt<bool> EnableBottleneckAnalysis(
 207     "bottleneck-analysis",
 208     cl::desc("Enable bottleneck analysis (disabled by default)"),
 209     cl::cat(ViewOptions), cl::init(false));
 210
 211 static cl::opt<bool> ShowEncoding(
 212     "show-encoding",
 213     cl::desc("Print encoding information in the instruction info view"),
 214     cl::cat(ViewOptions), cl::init(false));
 215
 216 namespace {
 217
 218 const Target *getTarget(const char *ProgName) {
 219   if (TripleName.empty())
 220     TripleName = Triple::normalize(sys::getDefaultTargetTriple());
 221   Triple TheTriple(TripleName);
 222
 223   // Get the target specific parser.
 224   std::string Error;
 225   const Target *TheTarget =
 226       TargetRegistry::lookupTarget(ArchName, TheTriple, Error);
 227   if (!TheTarget) {
 228     errs() << ProgName << ": " << Error;
 229     return nullptr;
 230   }
 231
 232   // Return the found target.
 233   return TheTarget;
 234 }
 235
 236 ErrorOr<std::unique_ptr<ToolOutputFile>> getOutputStream() {
 237   if (OutputFilename == "")
 238     OutputFilename = "-";
 239   std::error_code EC;
 240   auto Out =
 241       std::make_unique<ToolOutputFile>(OutputFilename, EC, sys::fs::OF_Text);
 242   if (!EC)
 243     return std::move(Out);
 244   return EC;
 245 }
 246 } // end of anonymous namespace
 247
 248 static void processOptionImpl(cl::opt<bool> &O, const cl::opt<bool> &Default) {
 249   if (!O.getNumOccurrences() || O.getPosition() < Default.getPosition())
 250     O = Default.getValue();
 251 }
 252
 253 static void processViewOptions() {
 254   if (!EnableAllViews.getNumOccurrences() &&
 255       !EnableAllStats.getNumOccurrences())
 256     return;
 257
 258   if (EnableAllViews.getNumOccurrences()) {
 259     processOptionImpl(PrintSummaryView, EnableAllViews);
 260     processOptionImpl(EnableBottleneckAnalysis, EnableAllViews);
 261     processOptionImpl(PrintResourcePressureView, EnableAllViews);
 262     processOptionImpl(PrintTimelineView, EnableAllViews);
 263     processOptionImpl(PrintInstructionInfoView, EnableAllViews);
 264   }
 265
 266   const cl::opt<bool> &Default =
 267       EnableAllViews.getPosition() < EnableAllStats.getPosition()
 268           ? EnableAllStats
 269           : EnableAllViews;
 270   processOptionImpl(PrintRegisterFileStats, Default);
 271   processOptionImpl(PrintDispatchStats, Default);
 272   processOptionImpl(PrintSchedulerStats, Default);
 273   processOptionImpl(PrintRetireStats, Default);
 274 }
 275
 276 // Returns true on success.
 277 static bool runPipeline(mca::Pipeline &P) {
 278   // Handle pipeline errors here.
 279   Expected<unsigned> Cycles = P.run();
 280   if (!Cycles) {
 281     WithColor::error() << toString(Cycles.takeError());
 282     return false;
 283   }
 284   return true;
 285 }
 286
 287 int main(int argc, char **argv) {
 288   InitLLVM X(argc, argv);
 289
 290   // Initialize targets and assembly parsers.
 291   InitializeAllTargetInfos();
 292   InitializeAllTargetMCs();
 293   InitializeAllAsmParsers();
 294
 295   // Enable printing of available targets when flag --version is specified.
 296   cl::AddExtraVersionPrinter(TargetRegistry::printRegisteredTargetsForVersion);
 297
 298   cl::HideUnrelatedOptions({&ToolOptions, &ViewOptions});
 299
 300   // Parse flags and initialize target options.
 301   cl::ParseCommandLineOptions(argc, argv,
 302                               "llvm machine code performance analyzer.\n");
 303
 304   // Get the target from the triple. If a triple is not specified, then select
 305   // the default triple for the host. If the triple doesn't correspond to any
 306   // registered target, then exit with an error message.
 307   const char *ProgName = argv[0];
 308   const Target *TheTarget = getTarget(ProgName);
 309   if (!TheTarget)
 310     return 1;
 311
 312   // GetTarget() may replaced TripleName with a default triple.
 313   // For safety, reconstruct the Triple object.
 314   Triple TheTriple(TripleName);
 315
 316   ErrorOr<std::unique_ptr<MemoryBuffer>> BufferPtr =
 317       MemoryBuffer::getFileOrSTDIN(InputFilename);
 318   if (std::error_code EC = BufferPtr.getError()) {
 319     WithColor::error() << InputFilename << ": " << EC.message() << '\n';
 320     return 1;
 321   }
 322
 323   // Apply overrides to llvm-mca specific options.
 324   processViewOptions();
 325
 326   if (!MCPU.compare("native"))
 327     MCPU = llvm::sys::getHostCPUName();
 328
 329   std::unique_ptr<MCSubtargetInfo> STI(
 330       TheTarget->createMCSubtargetInfo(TripleName, MCPU, MATTR));
 331   if (!STI->isCPUStringValid(MCPU))
 332     return 1;
 333
 334   if (!PrintInstructionTables && !STI->getSchedModel().isOutOfOrder()) {
 335     WithColor::error() << "please specify an out-of-order cpu. '" << MCPU
 336                        << "' is an in-order cpu.\n";
 337     return 1;
 338   }
 339
 340   if (!STI->getSchedModel().hasInstrSchedModel()) {
 341     WithColor::error()
 342         << "unable to find instruction-level scheduling information for"
 343         << " target triple '" << TheTriple.normalize() << "' and cpu '" << MCPU
 344         << "'.\n";
 345
 346     if (STI->getSchedModel().InstrItineraries)
 347       WithColor::note()
 348           << "cpu '" << MCPU << "' provides itineraries. However, "
 349           << "instruction itineraries are currently unsupported.\n";
 350     return 1;
 351   }
 352
 353   std::unique_ptr<MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TripleName));
 354   assert(MRI && "Unable to create target register info!");
 355
 356   std::unique_ptr<MCAsmInfo> MAI(TheTarget->createMCAsmInfo(*MRI, TripleName));
 357   assert(MAI && "Unable to create target asm info!");
 358
 359   MCObjectFileInfo MOFI;
 360   SourceMgr SrcMgr;
 361
 362   // Tell SrcMgr about this buffer, which is what the parser will pick up.
 363   SrcMgr.AddNewSourceBuffer(std::move(*BufferPtr), SMLoc());
 364
 365   MCContext Ctx(MAI.get(), MRI.get(), &MOFI, &SrcMgr);
 366
 367   MOFI.InitMCObjectFileInfo(TheTriple, /* PIC= */ false, Ctx);
 368
 369   std::unique_ptr<buffer_ostream> BOS;
 370
 371   std::unique_ptr<MCInstrInfo> MCII(TheTarget->createMCInstrInfo());
 372
 373   std::unique_ptr<MCInstrAnalysis> MCIA(
 374       TheTarget->createMCInstrAnalysis(MCII.get()));
 375
 376   // Parse the input and create CodeRegions that llvm-mca can analyze.
 377   mca::AsmCodeRegionGenerator CRG(*TheTarget, SrcMgr, Ctx, *MAI, *STI, *MCII);
 378   Expected<const mca::CodeRegions &> RegionsOrErr = CRG.parseCodeRegions();
 379   if (!RegionsOrErr) {
 380     if (auto Err =
 381             handleErrors(RegionsOrErr.takeError(), [](const StringError &E) {
 382               WithColor::error() << E.getMessage() << '\n';
 383             })) {
 384       // Default case.
 385       WithColor::error() << toString(std::move(Err)) << '\n';
 386     }
 387     return 1;
 388   }
 389   const mca::CodeRegions &Regions = *RegionsOrErr;
 390
 391   // Early exit if errors were found by the code region parsing logic.
 392   if (!Regions.isValid())
 393     return 1;
 394
 395   if (Regions.empty()) {
 396     WithColor::error() << "no assembly instructions found.\n";
 397     return 1;
 398   }
 399
 400   // Now initialize the output file.
 401   auto OF = getOutputStream();
 402   if (std::error_code EC = OF.getError()) {
 403     WithColor::error() << EC.message() << '\n';
 404     return 1;
 405   }
 406
 407   unsigned AssemblerDialect = CRG.getAssemblerDialect();
 408   if (OutputAsmVariant >= 0)
 409     AssemblerDialect = static_cast<unsigned>(OutputAsmVariant);
 410   std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter(
 411       Triple(TripleName), AssemblerDialect, *MAI, *MCII, *MRI));
 412   if (!IP) {
 413     WithColor::error()
 414         << "unable to create instruction printer for target triple '"
 415         << TheTriple.normalize() << "' with assembly variant "
 416         << AssemblerDialect << ".\n";
 417     return 1;
 418   }
 419
 420   // Set the display preference for hex vs. decimal immediates.
 421   IP->setPrintImmHex(PrintImmHex);
 422
 423   std::unique_ptr<ToolOutputFile> TOF = std::move(*OF);
 424
 425   const MCSchedModel &SM = STI->getSchedModel();
 426
 427   // Create an instruction builder.
 428   mca::InstrBuilder IB(*STI, *MCII, *MRI, MCIA.get());
 429
 430   // Create a context to control ownership of the pipeline hardware.
 431   mca::Context MCA(*MRI, *STI);
 432
 433   mca::PipelineOptions PO(MicroOpQueue, DecoderThroughput, DispatchWidth,
 434                           RegisterFileSize, LoadQueueSize, StoreQueueSize,
 435                           AssumeNoAlias, EnableBottleneckAnalysis);
 436
 437   // Number each region in the sequence.
 438   unsigned RegionIdx = 0;
 439
 440   std::unique_ptr<MCCodeEmitter> MCE(
 441       TheTarget->createMCCodeEmitter(*MCII, *MRI, Ctx));
 442
 443   std::unique_ptr<MCAsmBackend> MAB(TheTarget->createMCAsmBackend(
 444       *STI, *MRI, InitMCTargetOptionsFromFlags()));
 445
 446   for (const std::unique_ptr<mca::CodeRegion> &Region : Regions) {
 447     // Skip empty code regions.
 448     if (Region->empty())
 449       continue;
 450
 451     // Don't print the header of this region if it is the default region, and
 452     // it doesn't have an end location.
 453     if (Region->startLoc().isValid() || Region->endLoc().isValid()) {
 454       TOF->os() << "\n[" << RegionIdx++ << "] Code Region";
 455       StringRef Desc = Region->getDescription();
 456       if (!Desc.empty())
 457         TOF->os() << " - " << Desc;
 458       TOF->os() << "\n\n";
 459     }
 460
 461     // Lower the MCInst sequence into an mca::Instruction sequence.
 462     ArrayRef<MCInst> Insts = Region->getInstructions();
 463     mca::CodeEmitter CE(*STI, *MAB, *MCE, Insts);
 464     std::vector<std::unique_ptr<mca::Instruction>> LoweredSequence;
 465     for (const MCInst &MCI : Insts) {
 466       Expected<std::unique_ptr<mca::Instruction>> Inst =
 467           IB.createInstruction(MCI);
 468       if (!Inst) {
 469         if (auto NewE = handleErrors(
 470                 Inst.takeError(),
 471                 [&IP, &STI](const mca::InstructionError<MCInst> &IE) {
 472                   std::string InstructionStr;
 473                   raw_string_ostream SS(InstructionStr);
 474                   WithColor::error() << IE.Message << '\n';
 475                   IP->printInst(&IE.Inst, SS, "", *STI);
 476                   SS.flush();
 477                   WithColor::note()
 478                       << "instruction: " << InstructionStr << '\n';
 479                 })) {
 480           // Default case.
 481           WithColor::error() << toString(std::move(NewE));
 482         }
 483         return 1;
 484       }
 485
 486       LoweredSequence.emplace_back(std::move(Inst.get()));
 487     }
 488
 489     mca::SourceMgr S(LoweredSequence, PrintInstructionTables ? 1 : Iterations);
 490
 491     if (PrintInstructionTables) {
 492       //  Create a pipeline, stages, and a printer.
 493       auto P = std::make_unique<mca::Pipeline>();
 494       P->appendStage(std::make_unique<mca::EntryStage>(S));
 495       P->appendStage(std::make_unique<mca::InstructionTables>(SM));
 496       mca::PipelinePrinter Printer(*P);
 497
 498       // Create the views for this pipeline, execute, and emit a report.
 499       if (PrintInstructionInfoView) {
 500         Printer.addView(std::make_unique<mca::InstructionInfoView>(
 501             *STI, *MCII, CE, ShowEncoding, Insts, *IP));
 502       }
 503       Printer.addView(
 504           std::make_unique<mca::ResourcePressureView>(*STI, *IP, Insts));
 505
 506       if (!runPipeline(*P))
 507         return 1;
 508
 509       Printer.printReport(TOF->os());
 510       continue;
 511     }
 512
 513     // Create a basic pipeline simulating an out-of-order backend.
 514     auto P = MCA.createDefaultPipeline(PO, S);
 515     mca::PipelinePrinter Printer(*P);
 516
 517     if (PrintSummaryView)
 518       Printer.addView(
 519           std::make_unique<mca::SummaryView>(SM, Insts, DispatchWidth));
 520
 521     if (EnableBottleneckAnalysis) {
 522       Printer.addView(std::make_unique<mca::BottleneckAnalysis>(
 523           *STI, *IP, Insts, S.getNumIterations()));
 524     }
 525
 526     if (PrintInstructionInfoView)
 527       Printer.addView(std::make_unique<mca::InstructionInfoView>(
 528           *STI, *MCII, CE, ShowEncoding, Insts, *IP));
 529
 530     if (PrintDispatchStats)
 531       Printer.addView(std::make_unique<mca::DispatchStatistics>());
 532
 533     if (PrintSchedulerStats)
 534       Printer.addView(std::make_unique<mca::SchedulerStatistics>(*STI));
 535
 536     if (PrintRetireStats)
 537       Printer.addView(std::make_unique<mca::RetireControlUnitStatistics>(SM));
 538
 539     if (PrintRegisterFileStats)
 540       Printer.addView(std::make_unique<mca::RegisterFileStatistics>(*STI));
 541
 542     if (PrintResourcePressureView)
 543       Printer.addView(
 544           std::make_unique<mca::ResourcePressureView>(*STI, *IP, Insts));
 545
 546     if (PrintTimelineView) {
 547       unsigned TimelineIterations =
 548           TimelineMaxIterations ? TimelineMaxIterations : 10;
 549       Printer.addView(std::make_unique<mca::TimelineView>(
 550           *STI, *IP, Insts, std::min(TimelineIterations, S.getNumIterations()),
 551           TimelineMaxCycles));
 552     }
 553
 554     if (!runPipeline(*P))
 555       return 1;
 556
 557     Printer.printReport(TOF->os());
 558
 559     // Clear the InstrBuilder internal state in preparation for another round.
 560     IB.clear();
 561   }
 562
 563   TOF->keep();
 564   return 0;
 565 }