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.h"
43#include "llvm/MC/TargetRegistry.h"
44#include "llvm/MCA/CodeEmitter.h"
45#include "llvm/MCA/Context.h"
46#include "llvm/MCA/CustomBehaviour.h"
47#include "llvm/MCA/InstrBuilder.h"
48#include "llvm/MCA/Pipeline.h"
49#include "llvm/MCA/Stages/EntryStage.h"
50#include "llvm/MCA/Stages/InstructionTables.h"
51#include "llvm/MCA/Support.h"
52#include "llvm/Support/CommandLine.h"
53#include "llvm/Support/ErrorHandling.h"
54#include "llvm/Support/ErrorOr.h"
55#include "llvm/Support/FileSystem.h"
56#include "llvm/Support/InitLLVM.h"
57#include "llvm/Support/MemoryBuffer.h"
58#include "llvm/Support/SourceMgr.h"
59#include "llvm/Support/TargetSelect.h"
60#include "llvm/Support/ToolOutputFile.h"
61#include "llvm/Support/WithColor.h"
62#include "llvm/TargetParser/Host.h"
63
64using namespace llvm;
65
66static mc::RegisterMCTargetOptionsFlags MOF;
67
68static cl::OptionCategory ToolOptions("Tool Options");
69static cl::OptionCategory ViewOptions("View Options");
70
71static cl::opt<std::string> InputFilename(cl::Positional,
72 cl::desc("<input file>"),
73 cl::cat(ToolOptions), cl::init(Val: "-"));
74
75static cl::opt<std::string> OutputFilename("o", cl::desc("Output filename"),
76 cl::init(Val: "-"), cl::cat(ToolOptions),
77 cl::value_desc("filename"));
78
79static cl::opt<std::string>
80 ArchName("march",
81 cl::desc("Target architecture. "
82 "See -version for available targets"),
83 cl::cat(ToolOptions));
84
85static cl::opt<std::string>
86 TripleName("mtriple",
87 cl::desc("Target triple. See -version for available targets"),
88 cl::cat(ToolOptions));
89
90static cl::opt<std::string>
91 MCPU("mcpu",
92 cl::desc("Target a specific cpu type (-mcpu=help for details)"),
93 cl::value_desc("cpu-name"), cl::cat(ToolOptions), cl::init(Val: "native"));
94
95static cl::list<std::string>
96 MATTRS("mattr", cl::CommaSeparated,
97 cl::desc("Target specific attributes (-mattr=help for details)"),
98 cl::value_desc("a1,+a2,-a3,..."), cl::cat(ToolOptions));
99
100static cl::opt<bool> PrintJson("json",
101 cl::desc("Print the output in json format"),
102 cl::cat(ToolOptions), cl::init(Val: false));
103
104static cl::opt<int>
105 OutputAsmVariant("output-asm-variant",
106 cl::desc("Syntax variant to use for output printing"),
107 cl::cat(ToolOptions), cl::init(Val: -1));
108
109static cl::opt<bool>
110 PrintImmHex("print-imm-hex", cl::cat(ToolOptions), cl::init(Val: false),
111 cl::desc("Prefer hex format when printing immediate values"));
112
113static cl::opt<unsigned> Iterations("iterations",
114 cl::desc("Number of iterations to run"),
115 cl::cat(ToolOptions), cl::init(Val: 0));
116
117static cl::opt<unsigned>
118 DispatchWidth("dispatch", cl::desc("Override the processor dispatch width"),
119 cl::cat(ToolOptions), cl::init(Val: 0));
120
121static cl::opt<unsigned>
122 RegisterFileSize("register-file-size",
123 cl::desc("Maximum number of physical registers which can "
124 "be used for register mappings"),
125 cl::cat(ToolOptions), cl::init(Val: 0));
126
127static cl::opt<unsigned>
128 MicroOpQueue("micro-op-queue-size", cl::Hidden,
129 cl::desc("Number of entries in the micro-op queue"),
130 cl::cat(ToolOptions), cl::init(Val: 0));
131
132static cl::opt<unsigned>
133 DecoderThroughput("decoder-throughput", cl::Hidden,
134 cl::desc("Maximum throughput from the decoders "
135 "(instructions per cycle)"),
136 cl::cat(ToolOptions), cl::init(Val: 0));
137
138static cl::opt<unsigned>
139 CallLatency("call-latency", cl::Hidden,
140 cl::desc("Number of cycles to assume for a call instruction"),
141 cl::cat(ToolOptions), cl::init(Val: 100U));
142
143enum class SkipType { NONE, LACK_SCHED, PARSE_FAILURE, ANY_FAILURE };
144
145static cl::opt<enum SkipType> SkipUnsupportedInstructions(
146 "skip-unsupported-instructions",
147 cl::desc("Force analysis to continue in the presence of unsupported "
148 "instructions"),
149 cl::values(
150 clEnumValN(SkipType::NONE, "none",
151 "Exit with an error when an instruction is unsupported for "
152 "any reason (default)"),
153 clEnumValN(
154 SkipType::LACK_SCHED, "lack-sched",
155 "Skip instructions on input which lack scheduling information"),
156 clEnumValN(
157 SkipType::PARSE_FAILURE, "parse-failure",
158 "Skip lines on the input which fail to parse for any reason"),
159 clEnumValN(SkipType::ANY_FAILURE, "any",
160 "Skip instructions or lines on input which are unsupported "
161 "for any reason")),
162 cl::init(Val: SkipType::NONE), cl::cat(ViewOptions));
163
164bool shouldSkip(enum SkipType skipType) {
165 if (SkipUnsupportedInstructions == SkipType::NONE)
166 return false;
167 if (SkipUnsupportedInstructions == SkipType::ANY_FAILURE)
168 return true;
169 return skipType == SkipUnsupportedInstructions;
170}
171
172static cl::opt<bool>
173 PrintRegisterFileStats("register-file-stats",
174 cl::desc("Print register file statistics"),
175 cl::cat(ViewOptions), cl::init(Val: false));
176
177static cl::opt<bool> PrintDispatchStats("dispatch-stats",
178 cl::desc("Print dispatch statistics"),
179 cl::cat(ViewOptions), cl::init(Val: false));
180
181static cl::opt<bool>
182 PrintSummaryView("summary-view", cl::Hidden,
183 cl::desc("Print summary view (enabled by default)"),
184 cl::cat(ViewOptions), cl::init(Val: true));
185
186static cl::opt<bool> PrintSchedulerStats("scheduler-stats",
187 cl::desc("Print scheduler statistics"),
188 cl::cat(ViewOptions), cl::init(Val: false));
189
190static cl::opt<bool>
191 PrintRetireStats("retire-stats",
192 cl::desc("Print retire control unit statistics"),
193 cl::cat(ViewOptions), cl::init(Val: false));
194
195static cl::opt<bool> PrintResourcePressureView(
196 "resource-pressure",
197 cl::desc("Print the resource pressure view (enabled by default)"),
198 cl::cat(ViewOptions), cl::init(Val: true));
199
200static cl::opt<bool> PrintTimelineView("timeline",
201 cl::desc("Print the timeline view"),
202 cl::cat(ViewOptions), cl::init(Val: false));
203
204static cl::opt<unsigned> TimelineMaxIterations(
205 "timeline-max-iterations",
206 cl::desc("Maximum number of iterations to print in timeline view"),
207 cl::cat(ViewOptions), cl::init(Val: 0));
208
209static cl::opt<unsigned>
210 TimelineMaxCycles("timeline-max-cycles",
211 cl::desc("Maximum number of cycles in the timeline view, "
212 "or 0 for unlimited. Defaults to 80 cycles"),
213 cl::cat(ViewOptions), cl::init(Val: 80));
214
215static cl::opt<bool>
216 AssumeNoAlias("noalias",
217 cl::desc("If set, assume that loads and stores do not alias"),
218 cl::cat(ToolOptions), cl::init(Val: true));
219
220static cl::opt<unsigned> LoadQueueSize("lqueue",
221 cl::desc("Size of the load queue"),
222 cl::cat(ToolOptions), cl::init(Val: 0));
223
224static cl::opt<unsigned> StoreQueueSize("squeue",
225 cl::desc("Size of the store queue"),
226 cl::cat(ToolOptions), cl::init(Val: 0));
227
228enum class InstructionTablesType { NONE, NORMAL, FULL };
229
230static cl::opt<enum InstructionTablesType> InstructionTablesOption(
231 "instruction-tables", cl::desc("Print instruction tables"),
232 cl::values(clEnumValN(InstructionTablesType::NONE, "none",
233 "Do not print instruction tables"),
234 clEnumValN(InstructionTablesType::NORMAL, "normal",
235 "Print instruction tables"),
236 clEnumValN(InstructionTablesType::NORMAL, "", ""),
237 clEnumValN(InstructionTablesType::FULL, "full",
238 "Print instruction tables with additional"
239 " information: bypass latency, LLVM opcode,"
240 " used resources")),
241 cl::cat(ToolOptions), cl::init(Val: InstructionTablesType::NONE),
242 cl::ValueOptional);
243
244static bool shouldPrintInstructionTables(enum InstructionTablesType ITType) {
245 return InstructionTablesOption == ITType;
246}
247
248static bool shouldPrintInstructionTables() {
249 return !shouldPrintInstructionTables(ITType: InstructionTablesType::NONE);
250}
251
252static cl::opt<bool> PrintInstructionInfoView(
253 "instruction-info",
254 cl::desc("Print the instruction info view (enabled by default)"),
255 cl::cat(ViewOptions), cl::init(Val: true));
256
257static cl::opt<bool> EnableAllStats("all-stats",
258 cl::desc("Print all hardware statistics"),
259 cl::cat(ViewOptions), cl::init(Val: false));
260
261static cl::opt<bool>
262 EnableAllViews("all-views",
263 cl::desc("Print all views including hardware statistics"),
264 cl::cat(ViewOptions), cl::init(Val: false));
265
266static cl::opt<bool> EnableBottleneckAnalysis(
267 "bottleneck-analysis",
268 cl::desc("Enable bottleneck analysis (disabled by default)"),
269 cl::cat(ViewOptions), cl::init(Val: false));
270
271static cl::opt<bool> ShowEncoding(
272 "show-encoding",
273 cl::desc("Print encoding information in the instruction info view"),
274 cl::cat(ViewOptions), cl::init(Val: false));
275
276static cl::opt<bool> ShowBarriers(
277 "show-barriers",
278 cl::desc("Print memory barrier information in the instruction info view"),
279 cl::cat(ViewOptions), cl::init(Val: false));
280
281static cl::opt<bool> DisableCustomBehaviour(
282 "disable-cb",
283 cl::desc(
284 "Disable custom behaviour (use the default class which does nothing)."),
285 cl::cat(ViewOptions), cl::init(Val: false));
286
287static cl::opt<bool> DisableInstrumentManager(
288 "disable-im",
289 cl::desc("Disable instrumentation manager (use the default class which "
290 "ignores instruments.)."),
291 cl::cat(ViewOptions), cl::init(Val: false));
292
293namespace {
294
295const Target *getTarget(const char *ProgName) {
296 if (TripleName.empty())
297 TripleName = Triple::normalize(Str: sys::getDefaultTargetTriple());
298 Triple TheTriple(TripleName);
299
300 // Get the target specific parser.
301 std::string Error;
302 const Target *TheTarget =
303 TargetRegistry::lookupTarget(ArchName, TheTriple, Error);
304 if (!TheTarget) {
305 errs() << ProgName << ": " << Error;
306 return nullptr;
307 }
308
309 // Update TripleName with the updated triple from the target lookup.
310 TripleName = TheTriple.str();
311
312 // Return the found target.
313 return TheTarget;
314}
315
316ErrorOr<std::unique_ptr<ToolOutputFile>> getOutputStream() {
317 if (OutputFilename == "")
318 OutputFilename = "-";
319 std::error_code EC;
320 auto Out = std::make_unique<ToolOutputFile>(args&: OutputFilename, args&: EC,
321 args: sys::fs::OF_TextWithCRLF);
322 if (!EC)
323 return std::move(Out);
324 return EC;
325}
326} // end of anonymous namespace
327
328static void processOptionImpl(cl::opt<bool> &O, const cl::opt<bool> &Default) {
329 if (!O.getNumOccurrences() || O.getPosition() < Default.getPosition())
330 O = Default.getValue();
331}
332
333static void processViewOptions(bool IsOutOfOrder) {
334 if (!EnableAllViews.getNumOccurrences() &&
335 !EnableAllStats.getNumOccurrences())
336 return;
337
338 if (EnableAllViews.getNumOccurrences()) {
339 processOptionImpl(O&: PrintSummaryView, Default: EnableAllViews);
340 if (IsOutOfOrder)
341 processOptionImpl(O&: EnableBottleneckAnalysis, Default: EnableAllViews);
342 processOptionImpl(O&: PrintResourcePressureView, Default: EnableAllViews);
343 processOptionImpl(O&: PrintTimelineView, Default: EnableAllViews);
344 processOptionImpl(O&: PrintInstructionInfoView, Default: EnableAllViews);
345 }
346
347 const cl::opt<bool> &Default =
348 EnableAllViews.getPosition() < EnableAllStats.getPosition()
349 ? EnableAllStats
350 : EnableAllViews;
351 processOptionImpl(O&: PrintRegisterFileStats, Default);
352 processOptionImpl(O&: PrintDispatchStats, Default);
353 processOptionImpl(O&: PrintSchedulerStats, Default);
354 if (IsOutOfOrder)
355 processOptionImpl(O&: PrintRetireStats, Default);
356}
357
358// Returns true on success.
359static bool runPipeline(mca::Pipeline &P) {
360 // Handle pipeline errors here.
361 Expected<unsigned> Cycles = P.run();
362 if (!Cycles) {
363 WithColor::error() << toString(E: Cycles.takeError());
364 return false;
365 }
366 return true;
367}
368
369int main(int argc, char **argv) {
370 InitLLVM X(argc, argv);
371
372 // Initialize targets and assembly parsers.
373 InitializeAllTargetInfos();
374 InitializeAllTargetMCs();
375 InitializeAllAsmParsers();
376 InitializeAllTargetMCAs();
377
378 // Register the Target and CPU printer for --version.
379 cl::AddExtraVersionPrinter(func: sys::printDefaultTargetAndDetectedCPU);
380
381 // Enable printing of available targets when flag --version is specified.
382 cl::AddExtraVersionPrinter(func: TargetRegistry::printRegisteredTargetsForVersion);
383
384 cl::HideUnrelatedOptions(Categories: {&ToolOptions, &ViewOptions});
385
386 // Parse flags and initialize target options.
387 cl::ParseCommandLineOptions(argc, argv,
388 Overview: "llvm machine code performance analyzer.\n");
389
390 // Get the target from the triple. If a triple is not specified, then select
391 // the default triple for the host. If the triple doesn't correspond to any
392 // registered target, then exit with an error message.
393 const char *ProgName = argv[0];
394 const Target *TheTarget = getTarget(ProgName);
395 if (!TheTarget)
396 return 1;
397
398 // GetTarget() may replaced TripleName with a default triple.
399 // For safety, reconstruct the Triple object.
400 Triple TheTriple(TripleName);
401
402 ErrorOr<std::unique_ptr<MemoryBuffer>> BufferPtr =
403 MemoryBuffer::getFileOrSTDIN(Filename: InputFilename);
404 if (std::error_code EC = BufferPtr.getError()) {
405 WithColor::error() << InputFilename << ": " << EC.message() << '\n';
406 return 1;
407 }
408
409 if (MCPU == "native")
410 MCPU = std::string(llvm::sys::getHostCPUName());
411
412 // Package up features to be passed to target/subtarget
413 std::string FeaturesStr;
414 if (MATTRS.size()) {
415 SubtargetFeatures Features;
416 for (std::string &MAttr : MATTRS)
417 Features.AddFeature(String: MAttr);
418 FeaturesStr = Features.getString();
419 }
420
421 std::unique_ptr<MCSubtargetInfo> STI(
422 TheTarget->createMCSubtargetInfo(TheTriple: TripleName, CPU: MCPU, Features: FeaturesStr));
423 assert(STI && "Unable to create subtarget info!");
424 if (!STI->isCPUStringValid(CPU: MCPU))
425 return 1;
426
427 if (!STI->getSchedModel().hasInstrSchedModel()) {
428 WithColor::error()
429 << "unable to find instruction-level scheduling information for"
430 << " target triple '" << TheTriple.normalize() << "' and cpu '" << MCPU
431 << "'.\n";
432
433 if (STI->getSchedModel().InstrItineraries)
434 WithColor::note()
435 << "cpu '" << MCPU << "' provides itineraries. However, "
436 << "instruction itineraries are currently unsupported.\n";
437 return 1;
438 }
439
440 // Apply overrides to llvm-mca specific options.
441 bool IsOutOfOrder = STI->getSchedModel().isOutOfOrder();
442 processViewOptions(IsOutOfOrder);
443
444 std::unique_ptr<MCRegisterInfo> MRI(TheTarget->createMCRegInfo(TT: TripleName));
445 assert(MRI && "Unable to create target register info!");
446
447 MCTargetOptions MCOptions = mc::InitMCTargetOptionsFromFlags();
448 std::unique_ptr<MCAsmInfo> MAI(
449 TheTarget->createMCAsmInfo(MRI: *MRI, TheTriple: TripleName, Options: MCOptions));
450 assert(MAI && "Unable to create target asm info!");
451
452 SourceMgr SrcMgr;
453
454 // Tell SrcMgr about this buffer, which is what the parser will pick up.
455 SrcMgr.AddNewSourceBuffer(F: std::move(*BufferPtr), IncludeLoc: SMLoc());
456
457 std::unique_ptr<MCInstrInfo> MCII(TheTarget->createMCInstrInfo());
458 assert(MCII && "Unable to create instruction info!");
459
460 std::unique_ptr<MCInstrAnalysis> MCIA(
461 TheTarget->createMCInstrAnalysis(Info: MCII.get()));
462
463 // Need to initialize an MCInstPrinter as it is
464 // required for initializing the MCTargetStreamer
465 // which needs to happen within the CRG.parseAnalysisRegions() call below.
466 // Without an MCTargetStreamer, certain assembly directives can trigger a
467 // segfault. (For example, the .cv_fpo_proc directive on x86 will segfault if
468 // we don't initialize the MCTargetStreamer.)
469 unsigned IPtempOutputAsmVariant =
470 OutputAsmVariant == -1 ? 0 : OutputAsmVariant;
471 std::unique_ptr<MCInstPrinter> IPtemp(TheTarget->createMCInstPrinter(
472 T: Triple(TripleName), SyntaxVariant: IPtempOutputAsmVariant, MAI: *MAI, MII: *MCII, MRI: *MRI));
473 if (!IPtemp) {
474 WithColor::error()
475 << "unable to create instruction printer for target triple '"
476 << TheTriple.normalize() << "' with assembly variant "
477 << IPtempOutputAsmVariant << ".\n";
478 return 1;
479 }
480
481 // Parse the input and create CodeRegions that llvm-mca can analyze.
482 MCContext ACtx(TheTriple, MAI.get(), MRI.get(), STI.get(), &SrcMgr);
483 std::unique_ptr<MCObjectFileInfo> AMOFI(
484 TheTarget->createMCObjectFileInfo(Ctx&: ACtx, /*PIC=*/false));
485 ACtx.setObjectFileInfo(AMOFI.get());
486 mca::AsmAnalysisRegionGenerator CRG(*TheTarget, SrcMgr, ACtx, *MAI, *STI,
487 *MCII);
488 Expected<const mca::AnalysisRegions &> RegionsOrErr =
489 CRG.parseAnalysisRegions(IP: std::move(IPtemp),
490 SkipFailures: shouldSkip(skipType: SkipType::PARSE_FAILURE));
491 if (!RegionsOrErr) {
492 if (auto Err =
493 handleErrors(E: RegionsOrErr.takeError(), Hs: [](const StringError &E) {
494 WithColor::error() << E.getMessage() << '\n';
495 })) {
496 // Default case.
497 WithColor::error() << toString(E: std::move(Err)) << '\n';
498 }
499 return 1;
500 }
501 const mca::AnalysisRegions &Regions = *RegionsOrErr;
502
503 // Early exit if errors were found by the code region parsing logic.
504 if (!Regions.isValid())
505 return 1;
506
507 if (Regions.empty()) {
508 WithColor::error() << "no assembly instructions found.\n";
509 return 1;
510 }
511
512 std::unique_ptr<mca::InstrumentManager> IM;
513 if (!DisableInstrumentManager) {
514 IM = std::unique_ptr<mca::InstrumentManager>(
515 TheTarget->createInstrumentManager(STI: *STI, MCII: *MCII));
516 }
517 if (!IM) {
518 // If the target doesn't have its own IM implemented (or the -disable-cb
519 // flag is set) then we use the base class (which does nothing).
520 IM = std::make_unique<mca::InstrumentManager>(args&: *STI, args&: *MCII);
521 }
522
523 // Parse the input and create InstrumentRegion that llvm-mca
524 // can use to improve analysis.
525 MCContext ICtx(TheTriple, MAI.get(), MRI.get(), STI.get(), &SrcMgr);
526 std::unique_ptr<MCObjectFileInfo> IMOFI(
527 TheTarget->createMCObjectFileInfo(Ctx&: ICtx, /*PIC=*/false));
528 ICtx.setObjectFileInfo(IMOFI.get());
529 mca::AsmInstrumentRegionGenerator IRG(*TheTarget, SrcMgr, ICtx, *MAI, *STI,
530 *MCII, *IM);
531 Expected<const mca::InstrumentRegions &> InstrumentRegionsOrErr =
532 IRG.parseInstrumentRegions(IP: std::move(IPtemp),
533 SkipFailures: shouldSkip(skipType: SkipType::PARSE_FAILURE));
534 if (!InstrumentRegionsOrErr) {
535 if (auto Err = handleErrors(E: InstrumentRegionsOrErr.takeError(),
536 Hs: [](const StringError &E) {
537 WithColor::error() << E.getMessage() << '\n';
538 })) {
539 // Default case.
540 WithColor::error() << toString(E: std::move(Err)) << '\n';
541 }
542 return 1;
543 }
544 const mca::InstrumentRegions &InstrumentRegions = *InstrumentRegionsOrErr;
545
546 // Early exit if errors were found by the instrumentation parsing logic.
547 if (!InstrumentRegions.isValid())
548 return 1;
549
550 // Now initialize the output file.
551 auto OF = getOutputStream();
552 if (std::error_code EC = OF.getError()) {
553 WithColor::error() << EC.message() << '\n';
554 return 1;
555 }
556
557 unsigned AssemblerDialect = CRG.getAssemblerDialect();
558 if (OutputAsmVariant >= 0)
559 AssemblerDialect = static_cast<unsigned>(OutputAsmVariant);
560 std::unique_ptr<MCInstPrinter> IP(TheTarget->createMCInstPrinter(
561 T: Triple(TripleName), SyntaxVariant: AssemblerDialect, MAI: *MAI, MII: *MCII, MRI: *MRI));
562 if (!IP) {
563 WithColor::error()
564 << "unable to create instruction printer for target triple '"
565 << TheTriple.normalize() << "' with assembly variant "
566 << AssemblerDialect << ".\n";
567 return 1;
568 }
569
570 // Set the display preference for hex vs. decimal immediates.
571 IP->setPrintImmHex(PrintImmHex);
572
573 std::unique_ptr<ToolOutputFile> TOF = std::move(*OF);
574
575 const MCSchedModel &SM = STI->getSchedModel();
576
577 std::unique_ptr<mca::InstrPostProcess> IPP;
578 if (!DisableCustomBehaviour) {
579 // TODO: It may be a good idea to separate CB and IPP so that they can
580 // be used independently of each other. What I mean by this is to add
581 // an extra command-line arg --disable-ipp so that CB and IPP can be
582 // toggled without needing to toggle both of them together.
583 IPP = std::unique_ptr<mca::InstrPostProcess>(
584 TheTarget->createInstrPostProcess(STI: *STI, MCII: *MCII));
585 }
586 if (!IPP) {
587 // If the target doesn't have its own IPP implemented (or the -disable-cb
588 // flag is set) then we use the base class (which does nothing).
589 IPP = std::make_unique<mca::InstrPostProcess>(args&: *STI, args&: *MCII);
590 }
591
592 // Create an instruction builder.
593 mca::InstrBuilder IB(*STI, *MCII, *MRI, MCIA.get(), *IM, CallLatency);
594
595 // Create a context to control ownership of the pipeline hardware.
596 mca::Context MCA(*MRI, *STI);
597
598 mca::PipelineOptions PO(MicroOpQueue, DecoderThroughput, DispatchWidth,
599 RegisterFileSize, LoadQueueSize, StoreQueueSize,
600 AssumeNoAlias, EnableBottleneckAnalysis);
601
602 // Number each region in the sequence.
603 unsigned RegionIdx = 0;
604
605 std::unique_ptr<MCCodeEmitter> MCE(
606 TheTarget->createMCCodeEmitter(II: *MCII, Ctx&: ACtx));
607 assert(MCE && "Unable to create code emitter!");
608
609 std::unique_ptr<MCAsmBackend> MAB(TheTarget->createMCAsmBackend(
610 STI: *STI, MRI: *MRI, Options: mc::InitMCTargetOptionsFromFlags()));
611 assert(MAB && "Unable to create asm backend!");
612
613 json::Object JSONOutput;
614 int NonEmptyRegions = 0;
615 for (const std::unique_ptr<mca::AnalysisRegion> &Region : Regions) {
616 // Skip empty code regions.
617 if (Region->empty())
618 continue;
619
620 IB.clear();
621
622 // Lower the MCInst sequence into an mca::Instruction sequence.
623 ArrayRef<MCInst> Insts = Region->getInstructions();
624 mca::CodeEmitter CE(*STI, *MAB, *MCE, Insts);
625
626 IPP->resetState();
627
628 DenseMap<const MCInst *, SmallVector<mca::Instrument *>> InstToInstruments;
629 SmallVector<std::unique_ptr<mca::Instruction>> LoweredSequence;
630 SmallPtrSet<const MCInst *, 16> DroppedInsts;
631 for (const MCInst &MCI : Insts) {
632 SMLoc Loc = MCI.getLoc();
633 const SmallVector<mca::Instrument *> Instruments =
634 InstrumentRegions.getActiveInstruments(Loc);
635
636 Expected<std::unique_ptr<mca::Instruction>> Inst =
637 IB.createInstruction(MCI, IVec: Instruments);
638 if (!Inst) {
639 if (auto NewE = handleErrors(
640 E: Inst.takeError(),
641 Hs: [&IP, &STI](const mca::InstructionError<MCInst> &IE) {
642 std::string InstructionStr;
643 raw_string_ostream SS(InstructionStr);
644 if (shouldSkip(skipType: SkipType::LACK_SCHED))
645 WithColor::warning()
646 << IE.Message
647 << ", skipping with -skip-unsupported-instructions, "
648 "note accuracy will be impacted:\n";
649 else
650 WithColor::error()
651 << IE.Message
652 << ", use -skip-unsupported-instructions=lack-sched to "
653 "ignore these on the input.\n";
654 IP->printInst(MI: &IE.Inst, Address: 0, Annot: "", STI: *STI, OS&: SS);
655 SS.flush();
656 WithColor::note()
657 << "instruction: " << InstructionStr << '\n';
658 })) {
659 // Default case.
660 WithColor::error() << toString(E: std::move(NewE));
661 }
662 if (shouldSkip(skipType: SkipType::LACK_SCHED)) {
663 DroppedInsts.insert(Ptr: &MCI);
664 continue;
665 }
666 return 1;
667 }
668
669 IPP->postProcessInstruction(Inst&: Inst.get(), MCI);
670 InstToInstruments.insert(KV: {&MCI, Instruments});
671 LoweredSequence.emplace_back(Args: std::move(Inst.get()));
672 }
673
674 Insts = Region->dropInstructions(Insts: DroppedInsts);
675
676 // Skip empty regions.
677 if (Insts.empty())
678 continue;
679 NonEmptyRegions++;
680
681 mca::CircularSourceMgr S(LoweredSequence,
682 shouldPrintInstructionTables() ? 1 : Iterations);
683
684 if (shouldPrintInstructionTables()) {
685 // Create a pipeline, stages, and a printer.
686 auto P = std::make_unique<mca::Pipeline>();
687 P->appendStage(S: std::make_unique<mca::EntryStage>(args&: S));
688 P->appendStage(S: std::make_unique<mca::InstructionTables>(args: SM));
689
690 mca::PipelinePrinter Printer(*P, *Region, RegionIdx, *STI, PO);
691 if (PrintJson) {
692 Printer.addView(
693 V: std::make_unique<mca::InstructionView>(args&: *STI, args&: *IP, args&: Insts));
694 }
695
696 // Create the views for this pipeline, execute, and emit a report.
697 if (PrintInstructionInfoView) {
698 Printer.addView(V: std::make_unique<mca::InstructionInfoView>(
699 args&: *STI, args&: *MCII, args&: CE, args&: ShowEncoding, args&: Insts, args&: *IP, args&: LoweredSequence,
700 args&: ShowBarriers,
701 args: shouldPrintInstructionTables(ITType: InstructionTablesType::FULL), args&: *IM,
702 args&: InstToInstruments));
703 }
704
705 if (PrintResourcePressureView)
706 Printer.addView(
707 V: std::make_unique<mca::ResourcePressureView>(args&: *STI, args&: *IP, args&: Insts));
708
709 if (!runPipeline(P&: *P))
710 return 1;
711
712 if (PrintJson) {
713 Printer.printReport(JO&: JSONOutput);
714 } else {
715 Printer.printReport(OS&: TOF->os());
716 }
717
718 ++RegionIdx;
719 continue;
720 }
721
722 // Create the CustomBehaviour object for enforcing Target Specific
723 // behaviours and dependencies that aren't expressed well enough
724 // in the tablegen. CB cannot depend on the list of MCInst or
725 // the source code (but it can depend on the list of
726 // mca::Instruction or any objects that can be reconstructed
727 // from the target information).
728 std::unique_ptr<mca::CustomBehaviour> CB;
729 if (!DisableCustomBehaviour)
730 CB = std::unique_ptr<mca::CustomBehaviour>(
731 TheTarget->createCustomBehaviour(STI: *STI, SrcMgr: S, MCII: *MCII));
732 if (!CB)
733 // If the target doesn't have its own CB implemented (or the -disable-cb
734 // flag is set) then we use the base class (which does nothing).
735 CB = std::make_unique<mca::CustomBehaviour>(args&: *STI, args&: S, args&: *MCII);
736
737 // Create a basic pipeline simulating an out-of-order backend.
738 auto P = MCA.createDefaultPipeline(Opts: PO, SrcMgr&: S, CB&: *CB);
739
740 mca::PipelinePrinter Printer(*P, *Region, RegionIdx, *STI, PO);
741
742 // Targets can define their own custom Views that exist within their
743 // /lib/Target/ directory so that the View can utilize their CustomBehaviour
744 // or other backend symbols / functionality that are not already exposed
745 // through one of the MC-layer classes. These Views will be initialized
746 // using the CustomBehaviour::getViews() variants.
747 // If a target makes a custom View that does not depend on their target
748 // CB or their backend, they should put the View within
749 // /tools/llvm-mca/Views/ instead.
750 if (!DisableCustomBehaviour) {
751 std::vector<std::unique_ptr<mca::View>> CBViews =
752 CB->getStartViews(IP&: *IP, Insts);
753 for (auto &CBView : CBViews)
754 Printer.addView(V: std::move(CBView));
755 }
756
757 // When we output JSON, we add a view that contains the instructions
758 // and CPU resource information.
759 if (PrintJson) {
760 auto IV = std::make_unique<mca::InstructionView>(args&: *STI, args&: *IP, args&: Insts);
761 Printer.addView(V: std::move(IV));
762 }
763
764 if (PrintSummaryView)
765 Printer.addView(
766 V: std::make_unique<mca::SummaryView>(args: SM, args&: Insts, args&: DispatchWidth));
767
768 if (EnableBottleneckAnalysis) {
769 if (!IsOutOfOrder) {
770 WithColor::warning()
771 << "bottleneck analysis is not supported for in-order CPU '" << MCPU
772 << "'.\n";
773 }
774 Printer.addView(V: std::make_unique<mca::BottleneckAnalysis>(
775 args&: *STI, args&: *IP, args&: Insts, args: S.getNumIterations()));
776 }
777
778 if (PrintInstructionInfoView)
779 Printer.addView(V: std::make_unique<mca::InstructionInfoView>(
780 args&: *STI, args&: *MCII, args&: CE, args&: ShowEncoding, args&: Insts, args&: *IP, args&: LoweredSequence,
781 args&: ShowBarriers, /*ShouldPrintFullInfo=*/args: false, args&: *IM, args&: InstToInstruments));
782
783 // Fetch custom Views that are to be placed after the InstructionInfoView.
784 // Refer to the comment paired with the CB->getStartViews(*IP, Insts); line
785 // for more info.
786 if (!DisableCustomBehaviour) {
787 std::vector<std::unique_ptr<mca::View>> CBViews =
788 CB->getPostInstrInfoViews(IP&: *IP, Insts);
789 for (auto &CBView : CBViews)
790 Printer.addView(V: std::move(CBView));
791 }
792
793 if (PrintDispatchStats)
794 Printer.addView(V: std::make_unique<mca::DispatchStatistics>());
795
796 if (PrintSchedulerStats)
797 Printer.addView(V: std::make_unique<mca::SchedulerStatistics>(args&: *STI));
798
799 if (PrintRetireStats)
800 Printer.addView(V: std::make_unique<mca::RetireControlUnitStatistics>(args: SM));
801
802 if (PrintRegisterFileStats)
803 Printer.addView(V: std::make_unique<mca::RegisterFileStatistics>(args&: *STI));
804
805 if (PrintResourcePressureView)
806 Printer.addView(
807 V: std::make_unique<mca::ResourcePressureView>(args&: *STI, args&: *IP, args&: Insts));
808
809 if (PrintTimelineView) {
810 unsigned TimelineIterations =
811 TimelineMaxIterations ? TimelineMaxIterations : 10;
812 Printer.addView(V: std::make_unique<mca::TimelineView>(
813 args&: *STI, args&: *IP, args&: Insts, args: std::min(a: TimelineIterations, b: S.getNumIterations()),
814 args&: TimelineMaxCycles));
815 }
816
817 // Fetch custom Views that are to be placed after all other Views.
818 // Refer to the comment paired with the CB->getStartViews(*IP, Insts); line
819 // for more info.
820 if (!DisableCustomBehaviour) {
821 std::vector<std::unique_ptr<mca::View>> CBViews =
822 CB->getEndViews(IP&: *IP, Insts);
823 for (auto &CBView : CBViews)
824 Printer.addView(V: std::move(CBView));
825 }
826
827 if (!runPipeline(P&: *P))
828 return 1;
829
830 if (PrintJson) {
831 Printer.printReport(JO&: JSONOutput);
832 } else {
833 Printer.printReport(OS&: TOF->os());
834 }
835
836 ++RegionIdx;
837 }
838
839 if (NonEmptyRegions == 0) {
840 WithColor::error() << "no assembly instructions found.\n";
841 return 1;
842 }
843
844 if (PrintJson)
845 TOF->os() << formatv(Fmt: "{0:2}", Vals: json::Value(std::move(JSONOutput))) << "\n";
846
847 TOF->keep();
848 return 0;
849}
850