tools/llvm-mca/llvm-mca.cpp

   1 //===-- llvm-mca.cpp - Machine Code Analyzer -------------------*- C++ -* -===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // This utility is a simple driver that allows static performance analysis on
  11 // machine code similarly to how IACA (Intel Architecture Code Analyzer) works.
  12 //
  13 //   llvm-mca [options] <file-name>
  14 //      -march <type>
  15 //      -mcpu <cpu>
  16 //      -o <file>
  17 //
  18 // The target defaults to the host target.
  19 // The cpu defaults to the 'native' host cpu.
  20 // The output defaults to standard output.
  21 //
  22 //===----------------------------------------------------------------------===//
  23
  24 #include "CodeRegion.h"
  25 #include "PipelinePrinter.h"
  26 #include "Stages/FetchStage.h"
  27 #include "Stages/InstructionTables.h"
  28 #include "Views/DispatchStatistics.h"
  29 #include "Views/InstructionInfoView.h"
  30 #include "Views/RegisterFileStatistics.h"
  31 #include "Views/ResourcePressureView.h"
  32 #include "Views/RetireControlUnitStatistics.h"
  33 #include "Views/SchedulerStatistics.h"
  34 #include "Views/SummaryView.h"
  35 #include "Views/TimelineView.h"
  36 #include "include/Context.h"
  37 #include "include/Pipeline.h"
  38 #include "llvm/MC/MCAsmInfo.h"
  39 #include "llvm/MC/MCContext.h"
  40 #include "llvm/MC/MCObjectFileInfo.h"
  41 #include "llvm/MC/MCParser/MCTargetAsmParser.h"
  42 #include "llvm/MC/MCRegisterInfo.h"
  43 #include "llvm/MC/MCStreamer.h"
  44 #include "llvm/Support/CommandLine.h"
  45 #include "llvm/Support/ErrorHandling.h"
  46 #include "llvm/Support/ErrorOr.h"
  47 #include "llvm/Support/FileSystem.h"
  48 #include "llvm/Support/Host.h"
  49 #include "llvm/Support/InitLLVM.h"
  50 #include "llvm/Support/MemoryBuffer.h"
  51 #include "llvm/Support/SourceMgr.h"
  52 #include "llvm/Support/TargetRegistry.h"
  53 #include "llvm/Support/TargetSelect.h"
  54 #include "llvm/Support/ToolOutputFile.h"
  55 #include "llvm/Support/WithColor.h"
  56
  57 using namespace llvm;
  58
  59 static cl::OptionCategory ToolOptions("Tool Options");
  60 static cl::OptionCategory ViewOptions("View Options");
  61
  62 static cl::opt<std::string> InputFilename(cl::Positional,
  63                                           cl::desc("<input file>"),
  64                                           cl::cat(ToolOptions), cl::init("-"));
  65
  66 static cl::opt<std::string> OutputFilename("o", cl::desc("Output filename"),
  67                                            cl::init("-"), cl::cat(ToolOptions),
  68                                            cl::value_desc("filename"));
  69
  70 static cl::opt<std::string>
  71     ArchName("march", cl::desc("Target arch to assemble for, "
  72                                "see -version for available targets"),
  73              cl::cat(ToolOptions));
  74
  75 static cl::opt<std::string>
  76     TripleName("mtriple", cl::desc("Target triple to assemble for, "
  77                                    "see -version for available targets"),
  78                cl::cat(ToolOptions));
  79
  80 static cl::opt<std::string>
  81     MCPU("mcpu",
  82          cl::desc("Target a specific cpu type (-mcpu=help for details)"),
  83          cl::value_desc("cpu-name"), cl::cat(ToolOptions), cl::init("native"));
  84
  85 static cl::opt<int>
  86     OutputAsmVariant("output-asm-variant",
  87                      cl::desc("Syntax variant to use for output printing"),
  88                      cl::cat(ToolOptions), cl::init(-1));
  89
  90 static cl::opt<unsigned> Iterations("iterations",
  91                                     cl::desc("Number of iterations to run"),
  92                                     cl::cat(ToolOptions), cl::init(0));
  93
  94 static cl::opt<unsigned>
  95     DispatchWidth("dispatch", cl::desc("Override the processor dispatch width"),
  96                   cl::cat(ToolOptions), cl::init(0));
  97
  98 static cl::opt<unsigned>
  99     RegisterFileSize("register-file-size",
 100                      cl::desc("Maximum number of physical registers which can "
 101                               "be used for register mappings"),
 102                      cl::cat(ToolOptions), cl::init(0));
 103
 104 static cl::opt<bool>
 105     PrintRegisterFileStats("register-file-stats",
 106                            cl::desc("Print register file statistics"),
 107                            cl::cat(ViewOptions), cl::init(false));
 108
 109 static cl::opt<bool> PrintDispatchStats("dispatch-stats",
 110                                         cl::desc("Print dispatch statistics"),
 111                                         cl::cat(ViewOptions), cl::init(false));
 112
 113 static cl::opt<bool>
 114     PrintSummaryView("summary-view", cl::Hidden,
 115                      cl::desc("Print summary view (enabled by default)"),
 116                      cl::cat(ViewOptions), cl::init(true));
 117
 118 static cl::opt<bool> PrintSchedulerStats("scheduler-stats",
 119                                          cl::desc("Print scheduler statistics"),
 120                                          cl::cat(ViewOptions), cl::init(false));
 121
 122 static cl::opt<bool>
 123     PrintRetireStats("retire-stats",
 124                      cl::desc("Print retire control unit statistics"),
 125                      cl::cat(ViewOptions), cl::init(false));
 126
 127 static cl::opt<bool> PrintResourcePressureView(
 128     "resource-pressure",
 129     cl::desc("Print the resource pressure view (enabled by default)"),
 130     cl::cat(ViewOptions), cl::init(true));
 131
 132 static cl::opt<bool> PrintTimelineView("timeline",
 133                                        cl::desc("Print the timeline view"),
 134                                        cl::cat(ViewOptions), cl::init(false));
 135
 136 static cl::opt<unsigned> TimelineMaxIterations(
 137     "timeline-max-iterations",
 138     cl::desc("Maximum number of iterations to print in timeline view"),
 139     cl::cat(ViewOptions), cl::init(0));
 140
 141 static cl::opt<unsigned> TimelineMaxCycles(
 142     "timeline-max-cycles",
 143     cl::desc(
 144         "Maximum number of cycles in the timeline view. Defaults to 80 cycles"),
 145     cl::cat(ViewOptions), cl::init(80));
 146
 147 static cl::opt<bool>
 148     AssumeNoAlias("noalias",
 149                   cl::desc("If set, assume that loads and stores do not alias"),
 150                   cl::cat(ToolOptions), cl::init(true));
 151
 152 static cl::opt<unsigned>
 153     LoadQueueSize("lqueue",
 154                   cl::desc("Size of the load queue (unbound by default)"),
 155                   cl::cat(ToolOptions), cl::init(0));
 156
 157 static cl::opt<unsigned>
 158     StoreQueueSize("squeue",
 159                    cl::desc("Size of the store queue (unbound by default)"),
 160                    cl::cat(ToolOptions), cl::init(0));
 161
 162 static cl::opt<bool>
 163     PrintInstructionTables("instruction-tables",
 164                            cl::desc("Print instruction tables"),
 165                            cl::cat(ToolOptions), cl::init(false));
 166
 167 static cl::opt<bool> PrintInstructionInfoView(
 168     "instruction-info",
 169     cl::desc("Print the instruction info view (enabled by default)"),
 170     cl::cat(ViewOptions), cl::init(true));
 171
 172 static cl::opt<bool> EnableAllStats("all-stats",
 173                                     cl::desc("Print all hardware statistics"),
 174                                     cl::cat(ViewOptions), cl::init(false));
 175
 176 static cl::opt<bool>
 177     EnableAllViews("all-views",
 178                    cl::desc("Print all views including hardware statistics"),
 179                    cl::cat(ViewOptions), cl::init(false));
 180
 181 namespace {
 182
 183 const Target *getTarget(const char *ProgName) {
 184   if (TripleName.empty())
 185     TripleName = Triple::normalize(sys::getDefaultTargetTriple());
 186   Triple TheTriple(TripleName);
 187
 188   // Get the target specific parser.
 189   std::string Error;
 190   const Target *TheTarget =
 191       TargetRegistry::lookupTarget(ArchName, TheTriple, Error);
 192   if (!TheTarget) {
 193     errs() << ProgName << ": " << Error;
 194     return nullptr;
 195   }
 196
 197   // Return the found target.
 198   return TheTarget;
 199 }
 200
 201 // A comment consumer that parses strings.
 202 // The only valid tokens are strings.
 203 class MCACommentConsumer : public AsmCommentConsumer {
 204 public:
 205   mca::CodeRegions &Regions;
 206
 207   MCACommentConsumer(mca::CodeRegions &R) : Regions(R) {}
 208   void HandleComment(SMLoc Loc, StringRef CommentText) override {
 209     // Skip empty comments.
 210     StringRef Comment(CommentText);
 211     if (Comment.empty())
 212       return;
 213
 214     // Skip spaces and tabs
 215     unsigned Position = Comment.find_first_not_of(" \t");
 216     if (Position >= Comment.size())
 217       // we reached the end of the comment. Bail out.
 218       return;
 219
 220     Comment = Comment.drop_front(Position);
 221     if (Comment.consume_front("LLVM-MCA-END")) {
 222       Regions.endRegion(Loc);
 223       return;
 224     }
 225
 226     // Now try to parse string LLVM-MCA-BEGIN
 227     if (!Comment.consume_front("LLVM-MCA-BEGIN"))
 228       return;
 229
 230     // Skip spaces and tabs
 231     Position = Comment.find_first_not_of(" \t");
 232     if (Position < Comment.size())
 233       Comment = Comment.drop_front(Position);
 234     // Use the rest of the string as a descriptor for this code snippet.
 235     Regions.beginRegion(Comment, Loc);
 236   }
 237 };
 238
 239 int AssembleInput(MCAsmParser &Parser, const Target *TheTarget,
 240                   MCSubtargetInfo &STI, MCInstrInfo &MCII,
 241                   MCTargetOptions &MCOptions) {
 242   std::unique_ptr<MCTargetAsmParser> TAP(
 243       TheTarget->createMCAsmParser(STI, Parser, MCII, MCOptions));
 244
 245   if (!TAP) {
 246     WithColor::error() << "this target does not support assembly parsing.\n";
 247     return 1;
 248   }
 249
 250   Parser.setTargetParser(*TAP);
 251   return Parser.Run(false);
 252 }
 253
 254 ErrorOr<std::unique_ptr<ToolOutputFile>> getOutputStream() {
 255   if (OutputFilename == "")
 256     OutputFilename = "-";
 257   std::error_code EC;
 258   auto Out =
 259       llvm::make_unique<ToolOutputFile>(OutputFilename, EC, sys::fs::F_None);
 260   if (!EC)
 261     return std::move(Out);
 262   return EC;
 263 }
 264
 265 class MCStreamerWrapper final : public MCStreamer {
 266   mca::CodeRegions &Regions;
 267
 268 public:
 269   MCStreamerWrapper(MCContext &Context, mca::CodeRegions &R)
 270       : MCStreamer(Context), Regions(R) {}
 271
 272   // We only want to intercept the emission of new instructions.
 273   virtual void EmitInstruction(const MCInst &Inst, const MCSubtargetInfo &STI,
 274                                bool /* unused */) override {
 275     Regions.addInstruction(llvm::make_unique<const MCInst>(Inst));
 276   }
 277
 278   bool EmitSymbolAttribute(MCSymbol *Symbol, MCSymbolAttr Attribute) override {
 279     return true;
 280   }
 281
 282   void EmitCommonSymbol(MCSymbol *Symbol, uint64_t Size,
 283                         unsigned ByteAlignment) override {}
 284   void EmitZerofill(MCSection *Section, MCSymbol *Symbol = nullptr,
 285                     uint64_t Size = 0, unsigned ByteAlignment = 0,
 286                     SMLoc Loc = SMLoc()) override {}
 287   void EmitGPRel32Value(const MCExpr *Value) override {}
 288   void BeginCOFFSymbolDef(const MCSymbol *Symbol) override {}
 289   void EmitCOFFSymbolStorageClass(int StorageClass) override {}
 290   void EmitCOFFSymbolType(int Type) override {}
 291   void EndCOFFSymbolDef() override {}
 292
 293   const std::vector<std::unique_ptr<const MCInst>> &
 294   GetInstructionSequence(unsigned Index) const {
 295     return Regions.getInstructionSequence(Index);
 296   }
 297 };
 298 } // end of anonymous namespace
 299
 300 static void processOptionImpl(cl::opt<bool> &O, const cl::opt<bool> &Default) {
 301   if (!O.getNumOccurrences() || O.getPosition() < Default.getPosition())
 302     O = Default.getValue();
 303 }
 304
 305 static void processViewOptions() {
 306   if (!EnableAllViews.getNumOccurrences() &&
 307       !EnableAllStats.getNumOccurrences())
 308     return;
 309
 310   if (EnableAllViews.getNumOccurrences()) {
 311     processOptionImpl(PrintSummaryView, EnableAllViews);
 312     processOptionImpl(PrintResourcePressureView, EnableAllViews);
 313     processOptionImpl(PrintTimelineView, EnableAllViews);
 314     processOptionImpl(PrintInstructionInfoView, EnableAllViews);
 315   }
 316
 317   const cl::opt<bool> &Default =
 318       EnableAllViews.getPosition() < EnableAllStats.getPosition()
 319           ? EnableAllStats
 320           : EnableAllViews;
 321   processOptionImpl(PrintRegisterFileStats, Default);
 322   processOptionImpl(PrintDispatchStats, Default);
 323   processOptionImpl(PrintSchedulerStats, Default);
 324   processOptionImpl(PrintRetireStats, Default);
 325 }
 326
 327 int main(int argc, char **argv) {
 328   InitLLVM X(argc, argv);
 329
 330   // Initialize targets and assembly parsers.
 331   llvm::InitializeAllTargetInfos();
 332   llvm::InitializeAllTargetMCs();
 333   llvm::InitializeAllAsmParsers();
 334
 335   // Enable printing of available targets when flag --version is specified.
 336   cl::AddExtraVersionPrinter(TargetRegistry::printRegisteredTargetsForVersion);
 337
 338   cl::HideUnrelatedOptions({&ToolOptions, &ViewOptions});
 339
 340   // Parse flags and initialize target options.
 341   cl::ParseCommandLineOptions(argc, argv,
 342                               "llvm machine code performance analyzer.\n");
 343
 344   MCTargetOptions MCOptions;
 345   MCOptions.PreserveAsmComments = false;
 346
 347   // Get the target from the triple. If a triple is not specified, then select
 348   // the default triple for the host. If the triple doesn't correspond to any
 349   // registered target, then exit with an error message.
 350   const char *ProgName = argv[0];
 351   const Target *TheTarget = getTarget(ProgName);
 352   if (!TheTarget)
 353     return 1;
 354
 355   // GetTarget() may replaced TripleName with a default triple.
 356   // For safety, reconstruct the Triple object.
 357   Triple TheTriple(TripleName);
 358
 359   ErrorOr<std::unique_ptr<MemoryBuffer>> BufferPtr =
 360       MemoryBuffer::getFileOrSTDIN(InputFilename);
 361   if (std::error_code EC = BufferPtr.getError()) {
 362     WithColor::error() << InputFilename << ": " << EC.message() << '\n';
 363     return 1;
 364   }
 365
 366   // Apply overrides to llvm-mca specific options.
 367   processViewOptions();
 368
 369   SourceMgr SrcMgr;
 370
 371   // Tell SrcMgr about this buffer, which is what the parser will pick up.
 372   SrcMgr.AddNewSourceBuffer(std::move(*BufferPtr), SMLoc());
 373
 374   std::unique_ptr<MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TripleName));
 375   assert(MRI && "Unable to create target register info!");
 376
 377   std::unique_ptr<MCAsmInfo> MAI(TheTarget->createMCAsmInfo(*MRI, TripleName));
 378   assert(MAI && "Unable to create target asm info!");
 379
 380   MCObjectFileInfo MOFI;
 381   MCContext Ctx(MAI.get(), MRI.get(), &MOFI, &SrcMgr);
 382   MOFI.InitMCObjectFileInfo(TheTriple, /* PIC= */ false, Ctx);
 383
 384   std::unique_ptr<buffer_ostream> BOS;
 385
 386   mca::CodeRegions Regions(SrcMgr);
 387   MCStreamerWrapper Str(Ctx, Regions);
 388
 389   std::unique_ptr<MCInstrInfo> MCII(TheTarget->createMCInstrInfo());
 390
 391   std::unique_ptr<MCInstrAnalysis> MCIA(
 392       TheTarget->createMCInstrAnalysis(MCII.get()));
 393
 394   if (!MCPU.compare("native"))
 395     MCPU = llvm::sys::getHostCPUName();
 396
 397   std::unique_ptr<MCSubtargetInfo> STI(
 398       TheTarget->createMCSubtargetInfo(TripleName, MCPU, /* FeaturesStr */ ""));
 399   if (!STI->isCPUStringValid(MCPU))
 400     return 1;
 401
 402   if (!PrintInstructionTables && !STI->getSchedModel().isOutOfOrder()) {
 403     WithColor::error() << "please specify an out-of-order cpu. '" << MCPU
 404                        << "' is an in-order cpu.\n";
 405     return 1;
 406   }
 407
 408   if (!STI->getSchedModel().hasInstrSchedModel()) {
 409     WithColor::error()
 410         << "unable to find instruction-level scheduling information for"
 411         << " target triple '" << TheTriple.normalize() << "' and cpu '" << MCPU
 412         << "'.\n";
 413
 414     if (STI->getSchedModel().InstrItineraries)
 415       WithColor::note()
 416           << "cpu '" << MCPU << "' provides itineraries. However, "
 417           << "instruction itineraries are currently unsupported.\n";
 418     return 1;
 419   }
 420
 421   std::unique_ptr<MCAsmParser> P(createMCAsmParser(SrcMgr, Ctx, Str, *MAI));
 422   MCAsmLexer &Lexer = P->getLexer();
 423   MCACommentConsumer CC(Regions);
 424   Lexer.setCommentConsumer(&CC);
 425
 426   if (AssembleInput(*P, TheTarget, *STI, *MCII, MCOptions))
 427     return 1;
 428
 429   if (Regions.empty()) {
 430     WithColor::error() << "no assembly instructions found.\n";
 431     return 1;
 432   }
 433
 434   // Now initialize the output file.
 435   auto OF = getOutputStream();
 436   if (std::error_code EC = OF.getError()) {
 437     WithColor::error() << EC.message() << '\n';
 438     return 1;
 439   }
 440
 441   unsigned AssemblerDialect = P->getAssemblerDialect();
 442   if (OutputAsmVariant >= 0)
 443     AssemblerDialect = static_cast<unsigned>(OutputAsmVariant);
 444   std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter(
 445       Triple(TripleName), AssemblerDialect, *MAI, *MCII, *MRI));
 446   if (!IP) {
 447     WithColor::error()
 448         << "unable to create instruction printer for target triple '"
 449         << TheTriple.normalize() << "' with assembly variant "
 450         << AssemblerDialect << ".\n";
 451     return 1;
 452   }
 453
 454   std::unique_ptr<llvm::ToolOutputFile> TOF = std::move(*OF);
 455
 456   const MCSchedModel &SM = STI->getSchedModel();
 457
 458   unsigned Width = SM.IssueWidth;
 459   if (DispatchWidth)
 460     Width = DispatchWidth;
 461
 462   // Create an instruction builder.
 463   mca::InstrBuilder IB(*STI, *MCII, *MRI, *MCIA, *IP);
 464
 465   // Create a context to control ownership of the pipeline hardware.
 466   mca::Context MCA(*MRI, *STI);
 467
 468   mca::PipelineOptions PO(Width, RegisterFileSize, LoadQueueSize,
 469                           StoreQueueSize, AssumeNoAlias);
 470
 471   // Number each region in the sequence.
 472   unsigned RegionIdx = 0;
 473   for (const std::unique_ptr<mca::CodeRegion> &Region : Regions) {
 474     // Skip empty code regions.
 475     if (Region->empty())
 476       continue;
 477
 478     // Don't print the header of this region if it is the default region, and
 479     // it doesn't have an end location.
 480     if (Region->startLoc().isValid() || Region->endLoc().isValid()) {
 481       TOF->os() << "\n[" << RegionIdx++ << "] Code Region";
 482       StringRef Desc = Region->getDescription();
 483       if (!Desc.empty())
 484         TOF->os() << " - " << Desc;
 485       TOF->os() << "\n\n";
 486     }
 487
 488     mca::SourceMgr S(Region->getInstructions(),
 489                      PrintInstructionTables ? 1 : Iterations);
 490
 491     if (PrintInstructionTables) {
 492       //  Create a pipeline, stages, and a printer.
 493       auto P = llvm::make_unique<mca::Pipeline>();
 494       P->appendStage(llvm::make_unique<mca::FetchStage>(IB, S));
 495       P->appendStage(llvm::make_unique<mca::InstructionTables>(SM, IB));
 496       mca::PipelinePrinter Printer(*P);
 497
 498       // Create the views for this pipeline, execute, and emit a report.
 499       if (PrintInstructionInfoView) {
 500         Printer.addView(
 501             llvm::make_unique<mca::InstructionInfoView>(*STI, *MCII, S, *IP));
 502       }
 503       Printer.addView(
 504           llvm::make_unique<mca::ResourcePressureView>(*STI, *IP, S));
 505       auto Err = P->run();
 506       if (Err)
 507         report_fatal_error(toString(std::move(Err)));
 508       Printer.printReport(TOF->os());
 509       continue;
 510     }
 511
 512     // Create a basic pipeline simulating an out-of-order backend.
 513     auto P = MCA.createDefaultPipeline(PO, IB, S);
 514     mca::PipelinePrinter Printer(*P);
 515
 516     if (PrintSummaryView)
 517       Printer.addView(llvm::make_unique<mca::SummaryView>(SM, S, Width));
 518
 519     if (PrintInstructionInfoView)
 520       Printer.addView(
 521           llvm::make_unique<mca::InstructionInfoView>(*STI, *MCII, S, *IP));
 522
 523     if (PrintDispatchStats)
 524       Printer.addView(llvm::make_unique<mca::DispatchStatistics>());
 525
 526     if (PrintSchedulerStats)
 527       Printer.addView(llvm::make_unique<mca::SchedulerStatistics>(*STI));
 528
 529     if (PrintRetireStats)
 530       Printer.addView(llvm::make_unique<mca::RetireControlUnitStatistics>());
 531
 532     if (PrintRegisterFileStats)
 533       Printer.addView(llvm::make_unique<mca::RegisterFileStatistics>(*STI));
 534
 535     if (PrintResourcePressureView)
 536       Printer.addView(
 537           llvm::make_unique<mca::ResourcePressureView>(*STI, *IP, S));
 538
 539     if (PrintTimelineView) {
 540       Printer.addView(llvm::make_unique<mca::TimelineView>(
 541           *STI, *IP, S, TimelineMaxIterations, TimelineMaxCycles));
 542     }
 543
 544     auto Err = P->run();
 545     if (Err)
 546       report_fatal_error(toString(std::move(Err)));
 547     Printer.printReport(TOF->os());
 548
 549     // Clear the InstrBuilder internal state in preparation for another round.
 550     IB.clear();
 551   }
 552
 553   TOF->keep();
 554   return 0;
 555 }