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