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