1 //===- llvm-profdata.cpp - LLVM profile data tool -------------------------===//
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
7 //===----------------------------------------------------------------------===//
9 // llvm-profdata merges .profdata files.
11 //===----------------------------------------------------------------------===//
13 #include "llvm/ADT/SmallSet.h"
14 #include "llvm/ADT/SmallVector.h"
15 #include "llvm/ADT/StringRef.h"
16 #include "llvm/IR/LLVMContext.h"
17 #include "llvm/ProfileData/InstrProfReader.h"
18 #include "llvm/ProfileData/InstrProfWriter.h"
19 #include "llvm/ProfileData/ProfileCommon.h"
20 #include "llvm/ProfileData/SampleProfReader.h"
21 #include "llvm/ProfileData/SampleProfWriter.h"
22 #include "llvm/Support/CommandLine.h"
23 #include "llvm/Support/Errc.h"
24 #include "llvm/Support/FileSystem.h"
25 #include "llvm/Support/Format.h"
26 #include "llvm/Support/InitLLVM.h"
27 #include "llvm/Support/MemoryBuffer.h"
28 #include "llvm/Support/Path.h"
29 #include "llvm/Support/ThreadPool.h"
30 #include "llvm/Support/WithColor.h"
31 #include "llvm/Support/raw_ostream.h"
45 static void warn(Twine Message
, std::string Whence
= "",
46 std::string Hint
= "") {
49 errs() << Whence
<< ": ";
50 errs() << Message
<< "\n";
52 WithColor::note() << Hint
<< "\n";
55 static void exitWithError(Twine Message
, std::string Whence
= "",
56 std::string Hint
= "") {
59 errs() << Whence
<< ": ";
60 errs() << Message
<< "\n";
62 WithColor::note() << Hint
<< "\n";
66 static void exitWithError(Error E
, StringRef Whence
= "") {
67 if (E
.isA
<InstrProfError
>()) {
68 handleAllErrors(std::move(E
), [&](const InstrProfError
&IPE
) {
69 instrprof_error instrError
= IPE
.get();
71 if (instrError
== instrprof_error::unrecognized_format
) {
72 // Hint for common error of forgetting -sample for sample profiles.
73 Hint
= "Perhaps you forgot to use the -sample option?";
75 exitWithError(IPE
.message(), Whence
, Hint
);
79 exitWithError(toString(std::move(E
)), Whence
);
82 static void exitWithErrorCode(std::error_code EC
, StringRef Whence
= "") {
83 exitWithError(EC
.message(), Whence
);
87 enum ProfileKinds
{ instr
, sample
};
88 enum FailureMode
{ failIfAnyAreInvalid
, failIfAllAreInvalid
};
91 static void warnOrExitGivenError(FailureMode FailMode
, std::error_code EC
,
92 StringRef Whence
= "") {
93 if (FailMode
== failIfAnyAreInvalid
)
94 exitWithErrorCode(EC
, Whence
);
96 warn(EC
.message(), Whence
);
99 static void handleMergeWriterError(Error E
, StringRef WhenceFile
= "",
100 StringRef WhenceFunction
= "",
101 bool ShowHint
= true) {
102 if (!WhenceFile
.empty())
103 errs() << WhenceFile
<< ": ";
104 if (!WhenceFunction
.empty())
105 errs() << WhenceFunction
<< ": ";
107 auto IPE
= instrprof_error::success
;
108 E
= handleErrors(std::move(E
),
109 [&IPE
](std::unique_ptr
<InstrProfError
> E
) -> Error
{
111 return Error(std::move(E
));
113 errs() << toString(std::move(E
)) << "\n";
117 if (IPE
!= instrprof_error::success
) {
119 case instrprof_error::hash_mismatch
:
120 case instrprof_error::count_mismatch
:
121 case instrprof_error::value_site_count_mismatch
:
122 Hint
= "Make sure that all profile data to be merged is generated "
123 "from the same binary.";
131 errs() << Hint
<< "\n";
136 /// A remapper from original symbol names to new symbol names based on a file
137 /// containing a list of mappings from old name to new name.
138 class SymbolRemapper
{
139 std::unique_ptr
<MemoryBuffer
> File
;
140 DenseMap
<StringRef
, StringRef
> RemappingTable
;
143 /// Build a SymbolRemapper from a file containing a list of old/new symbols.
144 static std::unique_ptr
<SymbolRemapper
> create(StringRef InputFile
) {
145 auto BufOrError
= MemoryBuffer::getFileOrSTDIN(InputFile
);
147 exitWithErrorCode(BufOrError
.getError(), InputFile
);
149 auto Remapper
= std::make_unique
<SymbolRemapper
>();
150 Remapper
->File
= std::move(BufOrError
.get());
152 for (line_iterator
LineIt(*Remapper
->File
, /*SkipBlanks=*/true, '#');
153 !LineIt
.is_at_eof(); ++LineIt
) {
154 std::pair
<StringRef
, StringRef
> Parts
= LineIt
->split(' ');
155 if (Parts
.first
.empty() || Parts
.second
.empty() ||
156 Parts
.second
.count(' ')) {
157 exitWithError("unexpected line in remapping file",
158 (InputFile
+ ":" + Twine(LineIt
.line_number())).str(),
159 "expected 'old_symbol new_symbol'");
161 Remapper
->RemappingTable
.insert(Parts
);
166 /// Attempt to map the given old symbol into a new symbol.
168 /// \return The new symbol, or \p Name if no such symbol was found.
169 StringRef
operator()(StringRef Name
) {
170 StringRef New
= RemappingTable
.lookup(Name
);
171 return New
.empty() ? Name
: New
;
176 struct WeightedFile
{
177 std::string Filename
;
180 typedef SmallVector
<WeightedFile
, 5> WeightedFileVector
;
182 /// Keep track of merged data and reported errors.
183 struct WriterContext
{
185 InstrProfWriter Writer
;
186 std::vector
<std::pair
<Error
, std::string
>> Errors
;
188 SmallSet
<instrprof_error
, 4> &WriterErrorCodes
;
190 WriterContext(bool IsSparse
, std::mutex
&ErrLock
,
191 SmallSet
<instrprof_error
, 4> &WriterErrorCodes
)
192 : Lock(), Writer(IsSparse
), Errors(), ErrLock(ErrLock
),
193 WriterErrorCodes(WriterErrorCodes
) {}
196 /// Computer the overlap b/w profile BaseFilename and TestFileName,
197 /// and store the program level result to Overlap.
198 static void overlapInput(const std::string
&BaseFilename
,
199 const std::string
&TestFilename
, WriterContext
*WC
,
200 OverlapStats
&Overlap
,
201 const OverlapFuncFilters
&FuncFilter
,
202 raw_fd_ostream
&OS
, bool IsCS
) {
203 auto ReaderOrErr
= InstrProfReader::create(TestFilename
);
204 if (Error E
= ReaderOrErr
.takeError()) {
205 // Skip the empty profiles by returning sliently.
206 instrprof_error IPE
= InstrProfError::take(std::move(E
));
207 if (IPE
!= instrprof_error::empty_raw_profile
)
208 WC
->Errors
.emplace_back(make_error
<InstrProfError
>(IPE
), TestFilename
);
212 auto Reader
= std::move(ReaderOrErr
.get());
213 for (auto &I
: *Reader
) {
214 OverlapStats
FuncOverlap(OverlapStats::FunctionLevel
);
215 FuncOverlap
.setFuncInfo(I
.Name
, I
.Hash
);
217 WC
->Writer
.overlapRecord(std::move(I
), Overlap
, FuncOverlap
, FuncFilter
);
218 FuncOverlap
.dump(OS
);
222 /// Load an input into a writer context.
223 static void loadInput(const WeightedFile
&Input
, SymbolRemapper
*Remapper
,
225 std::unique_lock
<std::mutex
> CtxGuard
{WC
->Lock
};
227 // Copy the filename, because llvm::ThreadPool copied the input "const
228 // WeightedFile &" by value, making a reference to the filename within it
229 // invalid outside of this packaged task.
230 std::string Filename
= Input
.Filename
;
232 auto ReaderOrErr
= InstrProfReader::create(Input
.Filename
);
233 if (Error E
= ReaderOrErr
.takeError()) {
234 // Skip the empty profiles by returning sliently.
235 instrprof_error IPE
= InstrProfError::take(std::move(E
));
236 if (IPE
!= instrprof_error::empty_raw_profile
)
237 WC
->Errors
.emplace_back(make_error
<InstrProfError
>(IPE
), Filename
);
241 auto Reader
= std::move(ReaderOrErr
.get());
242 bool IsIRProfile
= Reader
->isIRLevelProfile();
243 bool HasCSIRProfile
= Reader
->hasCSIRLevelProfile();
244 if (WC
->Writer
.setIsIRLevelProfile(IsIRProfile
, HasCSIRProfile
)) {
245 WC
->Errors
.emplace_back(
246 make_error
<StringError
>(
247 "Merge IR generated profile with Clang generated profile.",
253 for (auto &I
: *Reader
) {
255 I
.Name
= (*Remapper
)(I
.Name
);
256 const StringRef FuncName
= I
.Name
;
257 bool Reported
= false;
258 WC
->Writer
.addRecord(std::move(I
), Input
.Weight
, [&](Error E
) {
260 consumeError(std::move(E
));
264 // Only show hint the first time an error occurs.
265 instrprof_error IPE
= InstrProfError::take(std::move(E
));
266 std::unique_lock
<std::mutex
> ErrGuard
{WC
->ErrLock
};
267 bool firstTime
= WC
->WriterErrorCodes
.insert(IPE
).second
;
268 handleMergeWriterError(make_error
<InstrProfError
>(IPE
), Input
.Filename
,
269 FuncName
, firstTime
);
272 if (Reader
->hasError())
273 if (Error E
= Reader
->getError())
274 WC
->Errors
.emplace_back(std::move(E
), Filename
);
277 /// Merge the \p Src writer context into \p Dst.
278 static void mergeWriterContexts(WriterContext
*Dst
, WriterContext
*Src
) {
279 for (auto &ErrorPair
: Src
->Errors
)
280 Dst
->Errors
.push_back(std::move(ErrorPair
));
283 Dst
->Writer
.mergeRecordsFromWriter(std::move(Src
->Writer
), [&](Error E
) {
284 instrprof_error IPE
= InstrProfError::take(std::move(E
));
285 std::unique_lock
<std::mutex
> ErrGuard
{Dst
->ErrLock
};
286 bool firstTime
= Dst
->WriterErrorCodes
.insert(IPE
).second
;
288 warn(toString(make_error
<InstrProfError
>(IPE
)));
292 static void mergeInstrProfile(const WeightedFileVector
&Inputs
,
293 SymbolRemapper
*Remapper
,
294 StringRef OutputFilename
,
295 ProfileFormat OutputFormat
, bool OutputSparse
,
296 unsigned NumThreads
, FailureMode FailMode
) {
297 if (OutputFilename
.compare("-") == 0)
298 exitWithError("Cannot write indexed profdata format to stdout.");
300 if (OutputFormat
!= PF_Binary
&& OutputFormat
!= PF_Compact_Binary
&&
301 OutputFormat
!= PF_Ext_Binary
&& OutputFormat
!= PF_Text
)
302 exitWithError("Unknown format is specified.");
304 std::mutex ErrorLock
;
305 SmallSet
<instrprof_error
, 4> WriterErrorCodes
;
307 // If NumThreads is not specified, auto-detect a good default.
310 std::min(hardware_concurrency(), unsigned((Inputs
.size() + 1) / 2));
312 // Initialize the writer contexts.
313 SmallVector
<std::unique_ptr
<WriterContext
>, 4> Contexts
;
314 for (unsigned I
= 0; I
< NumThreads
; ++I
)
315 Contexts
.emplace_back(std::make_unique
<WriterContext
>(
316 OutputSparse
, ErrorLock
, WriterErrorCodes
));
318 if (NumThreads
== 1) {
319 for (const auto &Input
: Inputs
)
320 loadInput(Input
, Remapper
, Contexts
[0].get());
322 ThreadPool
Pool(NumThreads
);
324 // Load the inputs in parallel (N/NumThreads serial steps).
326 for (const auto &Input
: Inputs
) {
327 Pool
.async(loadInput
, Input
, Remapper
, Contexts
[Ctx
].get());
328 Ctx
= (Ctx
+ 1) % NumThreads
;
332 // Merge the writer contexts together (~ lg(NumThreads) serial steps).
333 unsigned Mid
= Contexts
.size() / 2;
334 unsigned End
= Contexts
.size();
335 assert(Mid
> 0 && "Expected more than one context");
337 for (unsigned I
= 0; I
< Mid
; ++I
)
338 Pool
.async(mergeWriterContexts
, Contexts
[I
].get(),
339 Contexts
[I
+ Mid
].get());
342 Pool
.async(mergeWriterContexts
, Contexts
[0].get(),
343 Contexts
[End
- 1].get());
351 // Handle deferred errors encountered during merging. If the number of errors
352 // is equal to the number of inputs the merge failed.
353 unsigned NumErrors
= 0;
354 for (std::unique_ptr
<WriterContext
> &WC
: Contexts
) {
355 for (auto &ErrorPair
: WC
->Errors
) {
357 warn(toString(std::move(ErrorPair
.first
)), ErrorPair
.second
);
360 if (NumErrors
== Inputs
.size() ||
361 (NumErrors
> 0 && FailMode
== failIfAnyAreInvalid
))
362 exitWithError("No profiles could be merged.");
365 raw_fd_ostream
Output(OutputFilename
.data(), EC
, sys::fs::OF_None
);
367 exitWithErrorCode(EC
, OutputFilename
);
369 InstrProfWriter
&Writer
= Contexts
[0]->Writer
;
370 if (OutputFormat
== PF_Text
) {
371 if (Error E
= Writer
.writeText(Output
))
372 exitWithError(std::move(E
));
374 Writer
.write(Output
);
378 /// Make a copy of the given function samples with all symbol names remapped
379 /// by the provided symbol remapper.
380 static sampleprof::FunctionSamples
381 remapSamples(const sampleprof::FunctionSamples
&Samples
,
382 SymbolRemapper
&Remapper
, sampleprof_error
&Error
) {
383 sampleprof::FunctionSamples Result
;
384 Result
.setName(Remapper(Samples
.getName()));
385 Result
.addTotalSamples(Samples
.getTotalSamples());
386 Result
.addHeadSamples(Samples
.getHeadSamples());
387 for (const auto &BodySample
: Samples
.getBodySamples()) {
388 Result
.addBodySamples(BodySample
.first
.LineOffset
,
389 BodySample
.first
.Discriminator
,
390 BodySample
.second
.getSamples());
391 for (const auto &Target
: BodySample
.second
.getCallTargets()) {
392 Result
.addCalledTargetSamples(BodySample
.first
.LineOffset
,
393 BodySample
.first
.Discriminator
,
394 Remapper(Target
.first()), Target
.second
);
397 for (const auto &CallsiteSamples
: Samples
.getCallsiteSamples()) {
398 sampleprof::FunctionSamplesMap
&Target
=
399 Result
.functionSamplesAt(CallsiteSamples
.first
);
400 for (const auto &Callsite
: CallsiteSamples
.second
) {
401 sampleprof::FunctionSamples Remapped
=
402 remapSamples(Callsite
.second
, Remapper
, Error
);
403 MergeResult(Error
, Target
[Remapped
.getName()].merge(Remapped
));
409 static sampleprof::SampleProfileFormat FormatMap
[] = {
410 sampleprof::SPF_None
,
411 sampleprof::SPF_Text
,
412 sampleprof::SPF_Compact_Binary
,
413 sampleprof::SPF_Ext_Binary
,
415 sampleprof::SPF_Binary
};
417 static std::unique_ptr
<MemoryBuffer
>
418 getInputFileBuf(const StringRef
&InputFile
) {
422 auto BufOrError
= MemoryBuffer::getFileOrSTDIN(InputFile
);
424 exitWithErrorCode(BufOrError
.getError(), InputFile
);
426 return std::move(*BufOrError
);
429 static void populateProfileSymbolList(MemoryBuffer
*Buffer
,
430 sampleprof::ProfileSymbolList
&PSL
) {
434 SmallVector
<StringRef
, 32> SymbolVec
;
435 StringRef Data
= Buffer
->getBuffer();
436 Data
.split(SymbolVec
, '\n', /*MaxSplit=*/-1, /*KeepEmpty=*/false);
438 for (StringRef symbol
: SymbolVec
)
442 static void mergeSampleProfile(const WeightedFileVector
&Inputs
,
443 SymbolRemapper
*Remapper
,
444 StringRef OutputFilename
,
445 ProfileFormat OutputFormat
,
446 StringRef ProfileSymbolListFile
,
447 bool CompressProfSymList
, FailureMode FailMode
) {
448 using namespace sampleprof
;
449 StringMap
<FunctionSamples
> ProfileMap
;
450 SmallVector
<std::unique_ptr
<sampleprof::SampleProfileReader
>, 5> Readers
;
452 sampleprof::ProfileSymbolList WriterList
;
453 for (const auto &Input
: Inputs
) {
454 auto ReaderOrErr
= SampleProfileReader::create(Input
.Filename
, Context
);
455 if (std::error_code EC
= ReaderOrErr
.getError()) {
456 warnOrExitGivenError(FailMode
, EC
, Input
.Filename
);
460 // We need to keep the readers around until after all the files are
461 // read so that we do not lose the function names stored in each
462 // reader's memory. The function names are needed to write out the
463 // merged profile map.
464 Readers
.push_back(std::move(ReaderOrErr
.get()));
465 const auto Reader
= Readers
.back().get();
466 if (std::error_code EC
= Reader
->read()) {
467 warnOrExitGivenError(FailMode
, EC
, Input
.Filename
);
472 StringMap
<FunctionSamples
> &Profiles
= Reader
->getProfiles();
473 for (StringMap
<FunctionSamples
>::iterator I
= Profiles
.begin(),
476 sampleprof_error Result
= sampleprof_error::success
;
477 FunctionSamples Remapped
=
478 Remapper
? remapSamples(I
->second
, *Remapper
, Result
)
480 FunctionSamples
&Samples
= Remapper
? Remapped
: I
->second
;
481 StringRef FName
= Samples
.getName();
482 MergeResult(Result
, ProfileMap
[FName
].merge(Samples
, Input
.Weight
));
483 if (Result
!= sampleprof_error::success
) {
484 std::error_code EC
= make_error_code(Result
);
485 handleMergeWriterError(errorCodeToError(EC
), Input
.Filename
, FName
);
489 std::unique_ptr
<sampleprof::ProfileSymbolList
> ReaderList
=
490 Reader
->getProfileSymbolList();
492 WriterList
.merge(*ReaderList
);
495 SampleProfileWriter::create(OutputFilename
, FormatMap
[OutputFormat
]);
496 if (std::error_code EC
= WriterOrErr
.getError())
497 exitWithErrorCode(EC
, OutputFilename
);
499 // WriterList will have StringRef refering to string in Buffer.
500 // Make sure Buffer lives as long as WriterList.
501 auto Buffer
= getInputFileBuf(ProfileSymbolListFile
);
502 populateProfileSymbolList(Buffer
.get(), WriterList
);
503 WriterList
.setToCompress(CompressProfSymList
);
504 if (WriterList
.size() > 0 && OutputFormat
!= PF_Ext_Binary
)
505 warn("Profile Symbol list is not empty but the output format is not "
506 "ExtBinary format. The list will be lost in the output. ");
508 auto Writer
= std::move(WriterOrErr
.get());
509 Writer
->setProfileSymbolList(&WriterList
);
510 Writer
->write(ProfileMap
);
513 static WeightedFile
parseWeightedFile(const StringRef
&WeightedFilename
) {
514 StringRef WeightStr
, FileName
;
515 std::tie(WeightStr
, FileName
) = WeightedFilename
.split(',');
518 if (WeightStr
.getAsInteger(10, Weight
) || Weight
< 1)
519 exitWithError("Input weight must be a positive integer.");
521 return {FileName
, Weight
};
524 static void addWeightedInput(WeightedFileVector
&WNI
, const WeightedFile
&WF
) {
525 StringRef Filename
= WF
.Filename
;
526 uint64_t Weight
= WF
.Weight
;
528 // If it's STDIN just pass it on.
529 if (Filename
== "-") {
530 WNI
.push_back({Filename
, Weight
});
534 llvm::sys::fs::file_status Status
;
535 llvm::sys::fs::status(Filename
, Status
);
536 if (!llvm::sys::fs::exists(Status
))
537 exitWithErrorCode(make_error_code(errc::no_such_file_or_directory
),
539 // If it's a source file, collect it.
540 if (llvm::sys::fs::is_regular_file(Status
)) {
541 WNI
.push_back({Filename
, Weight
});
545 if (llvm::sys::fs::is_directory(Status
)) {
547 for (llvm::sys::fs::recursive_directory_iterator
F(Filename
, EC
), E
;
548 F
!= E
&& !EC
; F
.increment(EC
)) {
549 if (llvm::sys::fs::is_regular_file(F
->path())) {
550 addWeightedInput(WNI
, {F
->path(), Weight
});
554 exitWithErrorCode(EC
, Filename
);
558 static void parseInputFilenamesFile(MemoryBuffer
*Buffer
,
559 WeightedFileVector
&WFV
) {
563 SmallVector
<StringRef
, 8> Entries
;
564 StringRef Data
= Buffer
->getBuffer();
565 Data
.split(Entries
, '\n', /*MaxSplit=*/-1, /*KeepEmpty=*/false);
566 for (const StringRef
&FileWeightEntry
: Entries
) {
567 StringRef SanitizedEntry
= FileWeightEntry
.trim(" \t\v\f\r");
569 if (SanitizedEntry
.startswith("#"))
571 // If there's no comma, it's an unweighted profile.
572 else if (SanitizedEntry
.find(',') == StringRef::npos
)
573 addWeightedInput(WFV
, {SanitizedEntry
, 1});
575 addWeightedInput(WFV
, parseWeightedFile(SanitizedEntry
));
579 static int merge_main(int argc
, const char *argv
[]) {
580 cl::list
<std::string
> InputFilenames(cl::Positional
,
581 cl::desc("<filename...>"));
582 cl::list
<std::string
> WeightedInputFilenames("weighted-input",
583 cl::desc("<weight>,<filename>"));
584 cl::opt
<std::string
> InputFilenamesFile(
585 "input-files", cl::init(""),
586 cl::desc("Path to file containing newline-separated "
587 "[<weight>,]<filename> entries"));
588 cl::alias
InputFilenamesFileA("f", cl::desc("Alias for --input-files"),
589 cl::aliasopt(InputFilenamesFile
));
590 cl::opt
<bool> DumpInputFileList(
591 "dump-input-file-list", cl::init(false), cl::Hidden
,
592 cl::desc("Dump the list of input files and their weights, then exit"));
593 cl::opt
<std::string
> RemappingFile("remapping-file", cl::value_desc("file"),
594 cl::desc("Symbol remapping file"));
595 cl::alias
RemappingFileA("r", cl::desc("Alias for --remapping-file"),
596 cl::aliasopt(RemappingFile
));
597 cl::opt
<std::string
> OutputFilename("output", cl::value_desc("output"),
598 cl::init("-"), cl::Required
,
599 cl::desc("Output file"));
600 cl::alias
OutputFilenameA("o", cl::desc("Alias for --output"),
601 cl::aliasopt(OutputFilename
));
602 cl::opt
<ProfileKinds
> ProfileKind(
603 cl::desc("Profile kind:"), cl::init(instr
),
604 cl::values(clEnumVal(instr
, "Instrumentation profile (default)"),
605 clEnumVal(sample
, "Sample profile")));
606 cl::opt
<ProfileFormat
> OutputFormat(
607 cl::desc("Format of output profile"), cl::init(PF_Binary
),
609 clEnumValN(PF_Binary
, "binary", "Binary encoding (default)"),
610 clEnumValN(PF_Compact_Binary
, "compbinary",
611 "Compact binary encoding"),
612 clEnumValN(PF_Ext_Binary
, "extbinary", "Extensible binary encoding"),
613 clEnumValN(PF_Text
, "text", "Text encoding"),
614 clEnumValN(PF_GCC
, "gcc",
615 "GCC encoding (only meaningful for -sample)")));
616 cl::opt
<FailureMode
> FailureMode(
617 "failure-mode", cl::init(failIfAnyAreInvalid
), cl::desc("Failure mode:"),
618 cl::values(clEnumValN(failIfAnyAreInvalid
, "any",
619 "Fail if any profile is invalid."),
620 clEnumValN(failIfAllAreInvalid
, "all",
621 "Fail only if all profiles are invalid.")));
622 cl::opt
<bool> OutputSparse("sparse", cl::init(false),
623 cl::desc("Generate a sparse profile (only meaningful for -instr)"));
624 cl::opt
<unsigned> NumThreads(
625 "num-threads", cl::init(0),
626 cl::desc("Number of merge threads to use (default: autodetect)"));
627 cl::alias
NumThreadsA("j", cl::desc("Alias for --num-threads"),
628 cl::aliasopt(NumThreads
));
629 cl::opt
<std::string
> ProfileSymbolListFile(
630 "prof-sym-list", cl::init(""),
631 cl::desc("Path to file containing the list of function symbols "
632 "used to populate profile symbol list"));
633 cl::opt
<bool> CompressProfSymList(
634 "compress-prof-sym-list", cl::init(false), cl::Hidden
,
635 cl::desc("Compress profile symbol list before write it into profile. "));
637 cl::ParseCommandLineOptions(argc
, argv
, "LLVM profile data merger\n");
639 WeightedFileVector WeightedInputs
;
640 for (StringRef Filename
: InputFilenames
)
641 addWeightedInput(WeightedInputs
, {Filename
, 1});
642 for (StringRef WeightedFilename
: WeightedInputFilenames
)
643 addWeightedInput(WeightedInputs
, parseWeightedFile(WeightedFilename
));
645 // Make sure that the file buffer stays alive for the duration of the
646 // weighted input vector's lifetime.
647 auto Buffer
= getInputFileBuf(InputFilenamesFile
);
648 parseInputFilenamesFile(Buffer
.get(), WeightedInputs
);
650 if (WeightedInputs
.empty())
651 exitWithError("No input files specified. See " +
652 sys::path::filename(argv
[0]) + " -help");
654 if (DumpInputFileList
) {
655 for (auto &WF
: WeightedInputs
)
656 outs() << WF
.Weight
<< "," << WF
.Filename
<< "\n";
660 std::unique_ptr
<SymbolRemapper
> Remapper
;
661 if (!RemappingFile
.empty())
662 Remapper
= SymbolRemapper::create(RemappingFile
);
664 if (ProfileKind
== instr
)
665 mergeInstrProfile(WeightedInputs
, Remapper
.get(), OutputFilename
,
666 OutputFormat
, OutputSparse
, NumThreads
, FailureMode
);
668 mergeSampleProfile(WeightedInputs
, Remapper
.get(), OutputFilename
,
669 OutputFormat
, ProfileSymbolListFile
,
670 CompressProfSymList
, FailureMode
);
675 /// Computer the overlap b/w profile BaseFilename and profile TestFilename.
676 static void overlapInstrProfile(const std::string
&BaseFilename
,
677 const std::string
&TestFilename
,
678 const OverlapFuncFilters
&FuncFilter
,
679 raw_fd_ostream
&OS
, bool IsCS
) {
680 std::mutex ErrorLock
;
681 SmallSet
<instrprof_error
, 4> WriterErrorCodes
;
682 WriterContext
Context(false, ErrorLock
, WriterErrorCodes
);
683 WeightedFile WeightedInput
{BaseFilename
, 1};
684 OverlapStats Overlap
;
685 Error E
= Overlap
.accumuateCounts(BaseFilename
, TestFilename
, IsCS
);
687 exitWithError(std::move(E
), "Error in getting profile count sums");
688 if (Overlap
.Base
.CountSum
< 1.0f
) {
689 OS
<< "Sum of edge counts for profile " << BaseFilename
<< " is 0.\n";
692 if (Overlap
.Test
.CountSum
< 1.0f
) {
693 OS
<< "Sum of edge counts for profile " << TestFilename
<< " is 0.\n";
696 loadInput(WeightedInput
, nullptr, &Context
);
697 overlapInput(BaseFilename
, TestFilename
, &Context
, Overlap
, FuncFilter
, OS
,
702 static int overlap_main(int argc
, const char *argv
[]) {
703 cl::opt
<std::string
> BaseFilename(cl::Positional
, cl::Required
,
704 cl::desc("<base profile file>"));
705 cl::opt
<std::string
> TestFilename(cl::Positional
, cl::Required
,
706 cl::desc("<test profile file>"));
707 cl::opt
<std::string
> Output("output", cl::value_desc("output"), cl::init("-"),
708 cl::desc("Output file"));
709 cl::alias
OutputA("o", cl::desc("Alias for --output"), cl::aliasopt(Output
));
710 cl::opt
<bool> IsCS("cs", cl::init(false),
711 cl::desc("For context sensitive counts"));
712 cl::opt
<unsigned long long> ValueCutoff(
713 "value-cutoff", cl::init(-1),
715 "Function level overlap information for every function in test "
716 "profile with max count value greater then the parameter value"));
717 cl::opt
<std::string
> FuncNameFilter(
719 cl::desc("Function level overlap information for matching functions"));
720 cl::ParseCommandLineOptions(argc
, argv
, "LLVM profile data overlap tool\n");
723 raw_fd_ostream
OS(Output
.data(), EC
, sys::fs::OF_Text
);
725 exitWithErrorCode(EC
, Output
);
727 overlapInstrProfile(BaseFilename
, TestFilename
,
728 OverlapFuncFilters
{ValueCutoff
, FuncNameFilter
}, OS
,
734 typedef struct ValueSitesStats
{
736 : TotalNumValueSites(0), TotalNumValueSitesWithValueProfile(0),
738 uint64_t TotalNumValueSites
;
739 uint64_t TotalNumValueSitesWithValueProfile
;
740 uint64_t TotalNumValues
;
741 std::vector
<unsigned> ValueSitesHistogram
;
744 static void traverseAllValueSites(const InstrProfRecord
&Func
, uint32_t VK
,
745 ValueSitesStats
&Stats
, raw_fd_ostream
&OS
,
746 InstrProfSymtab
*Symtab
) {
747 uint32_t NS
= Func
.getNumValueSites(VK
);
748 Stats
.TotalNumValueSites
+= NS
;
749 for (size_t I
= 0; I
< NS
; ++I
) {
750 uint32_t NV
= Func
.getNumValueDataForSite(VK
, I
);
751 std::unique_ptr
<InstrProfValueData
[]> VD
= Func
.getValueForSite(VK
, I
);
752 Stats
.TotalNumValues
+= NV
;
754 Stats
.TotalNumValueSitesWithValueProfile
++;
755 if (NV
> Stats
.ValueSitesHistogram
.size())
756 Stats
.ValueSitesHistogram
.resize(NV
, 0);
757 Stats
.ValueSitesHistogram
[NV
- 1]++;
760 uint64_t SiteSum
= 0;
761 for (uint32_t V
= 0; V
< NV
; V
++)
762 SiteSum
+= VD
[V
].Count
;
766 for (uint32_t V
= 0; V
< NV
; V
++) {
767 OS
<< "\t[ " << format("%2u", I
) << ", ";
768 if (Symtab
== nullptr)
769 OS
<< format("%4" PRIu64
, VD
[V
].Value
);
771 OS
<< Symtab
->getFuncName(VD
[V
].Value
);
772 OS
<< ", " << format("%10" PRId64
, VD
[V
].Count
) << " ] ("
773 << format("%.2f%%", (VD
[V
].Count
* 100.0 / SiteSum
)) << ")\n";
778 static void showValueSitesStats(raw_fd_ostream
&OS
, uint32_t VK
,
779 ValueSitesStats
&Stats
) {
780 OS
<< " Total number of sites: " << Stats
.TotalNumValueSites
<< "\n";
781 OS
<< " Total number of sites with values: "
782 << Stats
.TotalNumValueSitesWithValueProfile
<< "\n";
783 OS
<< " Total number of profiled values: " << Stats
.TotalNumValues
<< "\n";
785 OS
<< " Value sites histogram:\n\tNumTargets, SiteCount\n";
786 for (unsigned I
= 0; I
< Stats
.ValueSitesHistogram
.size(); I
++) {
787 if (Stats
.ValueSitesHistogram
[I
] > 0)
788 OS
<< "\t" << I
+ 1 << ", " << Stats
.ValueSitesHistogram
[I
] << "\n";
792 static int showInstrProfile(const std::string
&Filename
, bool ShowCounts
,
793 uint32_t TopN
, bool ShowIndirectCallTargets
,
794 bool ShowMemOPSizes
, bool ShowDetailedSummary
,
795 std::vector
<uint32_t> DetailedSummaryCutoffs
,
796 bool ShowAllFunctions
, bool ShowCS
,
797 uint64_t ValueCutoff
, bool OnlyListBelow
,
798 const std::string
&ShowFunction
, bool TextFormat
,
799 raw_fd_ostream
&OS
) {
800 auto ReaderOrErr
= InstrProfReader::create(Filename
);
801 std::vector
<uint32_t> Cutoffs
= std::move(DetailedSummaryCutoffs
);
802 if (ShowDetailedSummary
&& Cutoffs
.empty()) {
803 Cutoffs
= {800000, 900000, 950000, 990000, 999000, 999900, 999990};
805 InstrProfSummaryBuilder
Builder(std::move(Cutoffs
));
806 if (Error E
= ReaderOrErr
.takeError())
807 exitWithError(std::move(E
), Filename
);
809 auto Reader
= std::move(ReaderOrErr
.get());
810 bool IsIRInstr
= Reader
->isIRLevelProfile();
811 size_t ShownFunctions
= 0;
812 size_t BelowCutoffFunctions
= 0;
813 int NumVPKind
= IPVK_Last
- IPVK_First
+ 1;
814 std::vector
<ValueSitesStats
> VPStats(NumVPKind
);
816 auto MinCmp
= [](const std::pair
<std::string
, uint64_t> &v1
,
817 const std::pair
<std::string
, uint64_t> &v2
) {
818 return v1
.second
> v2
.second
;
821 std::priority_queue
<std::pair
<std::string
, uint64_t>,
822 std::vector
<std::pair
<std::string
, uint64_t>>,
824 HottestFuncs(MinCmp
);
826 if (!TextFormat
&& OnlyListBelow
) {
827 OS
<< "The list of functions with the maximum counter less than "
828 << ValueCutoff
<< ":\n";
831 // Add marker so that IR-level instrumentation round-trips properly.
832 if (TextFormat
&& IsIRInstr
)
835 for (const auto &Func
: *Reader
) {
836 if (Reader
->isIRLevelProfile()) {
837 bool FuncIsCS
= NamedInstrProfRecord::hasCSFlagInHash(Func
.Hash
);
838 if (FuncIsCS
!= ShowCS
)
842 ShowAllFunctions
|| (!ShowFunction
.empty() &&
843 Func
.Name
.find(ShowFunction
) != Func
.Name
.npos
);
845 bool doTextFormatDump
= (Show
&& TextFormat
);
847 if (doTextFormatDump
) {
848 InstrProfSymtab
&Symtab
= Reader
->getSymtab();
849 InstrProfWriter::writeRecordInText(Func
.Name
, Func
.Hash
, Func
, Symtab
,
854 assert(Func
.Counts
.size() > 0 && "function missing entry counter");
855 Builder
.addRecord(Func
);
857 uint64_t FuncMax
= 0;
858 uint64_t FuncSum
= 0;
859 for (size_t I
= 0, E
= Func
.Counts
.size(); I
< E
; ++I
) {
860 FuncMax
= std::max(FuncMax
, Func
.Counts
[I
]);
861 FuncSum
+= Func
.Counts
[I
];
864 if (FuncMax
< ValueCutoff
) {
865 ++BelowCutoffFunctions
;
867 OS
<< " " << Func
.Name
<< ": (Max = " << FuncMax
868 << " Sum = " << FuncSum
<< ")\n";
871 } else if (OnlyListBelow
)
875 if (HottestFuncs
.size() == TopN
) {
876 if (HottestFuncs
.top().second
< FuncMax
) {
878 HottestFuncs
.emplace(std::make_pair(std::string(Func
.Name
), FuncMax
));
881 HottestFuncs
.emplace(std::make_pair(std::string(Func
.Name
), FuncMax
));
890 OS
<< " " << Func
.Name
<< ":\n"
891 << " Hash: " << format("0x%016" PRIx64
, Func
.Hash
) << "\n"
892 << " Counters: " << Func
.Counts
.size() << "\n";
894 OS
<< " Function count: " << Func
.Counts
[0] << "\n";
896 if (ShowIndirectCallTargets
)
897 OS
<< " Indirect Call Site Count: "
898 << Func
.getNumValueSites(IPVK_IndirectCallTarget
) << "\n";
900 uint32_t NumMemOPCalls
= Func
.getNumValueSites(IPVK_MemOPSize
);
901 if (ShowMemOPSizes
&& NumMemOPCalls
> 0)
902 OS
<< " Number of Memory Intrinsics Calls: " << NumMemOPCalls
906 OS
<< " Block counts: [";
907 size_t Start
= (IsIRInstr
? 0 : 1);
908 for (size_t I
= Start
, E
= Func
.Counts
.size(); I
< E
; ++I
) {
909 OS
<< (I
== Start
? "" : ", ") << Func
.Counts
[I
];
914 if (ShowIndirectCallTargets
) {
915 OS
<< " Indirect Target Results:\n";
916 traverseAllValueSites(Func
, IPVK_IndirectCallTarget
,
917 VPStats
[IPVK_IndirectCallTarget
], OS
,
918 &(Reader
->getSymtab()));
921 if (ShowMemOPSizes
&& NumMemOPCalls
> 0) {
922 OS
<< " Memory Intrinsic Size Results:\n";
923 traverseAllValueSites(Func
, IPVK_MemOPSize
, VPStats
[IPVK_MemOPSize
], OS
,
928 if (Reader
->hasError())
929 exitWithError(Reader
->getError(), Filename
);
933 std::unique_ptr
<ProfileSummary
> PS(Builder
.getSummary());
934 OS
<< "Instrumentation level: "
935 << (Reader
->isIRLevelProfile() ? "IR" : "Front-end") << "\n";
936 if (ShowAllFunctions
|| !ShowFunction
.empty())
937 OS
<< "Functions shown: " << ShownFunctions
<< "\n";
938 OS
<< "Total functions: " << PS
->getNumFunctions() << "\n";
939 if (ValueCutoff
> 0) {
940 OS
<< "Number of functions with maximum count (< " << ValueCutoff
941 << "): " << BelowCutoffFunctions
<< "\n";
942 OS
<< "Number of functions with maximum count (>= " << ValueCutoff
943 << "): " << PS
->getNumFunctions() - BelowCutoffFunctions
<< "\n";
945 OS
<< "Maximum function count: " << PS
->getMaxFunctionCount() << "\n";
946 OS
<< "Maximum internal block count: " << PS
->getMaxInternalCount() << "\n";
949 std::vector
<std::pair
<std::string
, uint64_t>> SortedHottestFuncs
;
950 while (!HottestFuncs
.empty()) {
951 SortedHottestFuncs
.emplace_back(HottestFuncs
.top());
955 << " functions with the largest internal block counts: \n";
956 for (auto &hotfunc
: llvm::reverse(SortedHottestFuncs
))
957 OS
<< " " << hotfunc
.first
<< ", max count = " << hotfunc
.second
<< "\n";
960 if (ShownFunctions
&& ShowIndirectCallTargets
) {
961 OS
<< "Statistics for indirect call sites profile:\n";
962 showValueSitesStats(OS
, IPVK_IndirectCallTarget
,
963 VPStats
[IPVK_IndirectCallTarget
]);
966 if (ShownFunctions
&& ShowMemOPSizes
) {
967 OS
<< "Statistics for memory intrinsic calls sizes profile:\n";
968 showValueSitesStats(OS
, IPVK_MemOPSize
, VPStats
[IPVK_MemOPSize
]);
971 if (ShowDetailedSummary
) {
972 OS
<< "Detailed summary:\n";
973 OS
<< "Total number of blocks: " << PS
->getNumCounts() << "\n";
974 OS
<< "Total count: " << PS
->getTotalCount() << "\n";
975 for (auto Entry
: PS
->getDetailedSummary()) {
976 OS
<< Entry
.NumCounts
<< " blocks with count >= " << Entry
.MinCount
978 << format("%0.6g", (float)Entry
.Cutoff
/ ProfileSummary::Scale
* 100)
979 << " percentage of the total counts.\n";
985 static int showSampleProfile(const std::string
&Filename
, bool ShowCounts
,
986 bool ShowAllFunctions
,
987 const std::string
&ShowFunction
,
988 bool ShowProfileSymbolList
, raw_fd_ostream
&OS
) {
989 using namespace sampleprof
;
991 auto ReaderOrErr
= SampleProfileReader::create(Filename
, Context
);
992 if (std::error_code EC
= ReaderOrErr
.getError())
993 exitWithErrorCode(EC
, Filename
);
995 auto Reader
= std::move(ReaderOrErr
.get());
996 if (std::error_code EC
= Reader
->read())
997 exitWithErrorCode(EC
, Filename
);
999 if (ShowAllFunctions
|| ShowFunction
.empty())
1002 Reader
->dumpFunctionProfile(ShowFunction
, OS
);
1004 if (ShowProfileSymbolList
) {
1005 std::unique_ptr
<sampleprof::ProfileSymbolList
> ReaderList
=
1006 Reader
->getProfileSymbolList();
1007 ReaderList
->dump(OS
);
1013 static int show_main(int argc
, const char *argv
[]) {
1014 cl::opt
<std::string
> Filename(cl::Positional
, cl::Required
,
1015 cl::desc("<profdata-file>"));
1017 cl::opt
<bool> ShowCounts("counts", cl::init(false),
1018 cl::desc("Show counter values for shown functions"));
1019 cl::opt
<bool> TextFormat(
1020 "text", cl::init(false),
1021 cl::desc("Show instr profile data in text dump format"));
1022 cl::opt
<bool> ShowIndirectCallTargets(
1023 "ic-targets", cl::init(false),
1024 cl::desc("Show indirect call site target values for shown functions"));
1025 cl::opt
<bool> ShowMemOPSizes(
1026 "memop-sizes", cl::init(false),
1027 cl::desc("Show the profiled sizes of the memory intrinsic calls "
1028 "for shown functions"));
1029 cl::opt
<bool> ShowDetailedSummary("detailed-summary", cl::init(false),
1030 cl::desc("Show detailed profile summary"));
1031 cl::list
<uint32_t> DetailedSummaryCutoffs(
1032 cl::CommaSeparated
, "detailed-summary-cutoffs",
1034 "Cutoff percentages (times 10000) for generating detailed summary"),
1035 cl::value_desc("800000,901000,999999"));
1036 cl::opt
<bool> ShowAllFunctions("all-functions", cl::init(false),
1037 cl::desc("Details for every function"));
1038 cl::opt
<bool> ShowCS("showcs", cl::init(false),
1039 cl::desc("Show context sensitive counts"));
1040 cl::opt
<std::string
> ShowFunction("function",
1041 cl::desc("Details for matching functions"));
1043 cl::opt
<std::string
> OutputFilename("output", cl::value_desc("output"),
1044 cl::init("-"), cl::desc("Output file"));
1045 cl::alias
OutputFilenameA("o", cl::desc("Alias for --output"),
1046 cl::aliasopt(OutputFilename
));
1047 cl::opt
<ProfileKinds
> ProfileKind(
1048 cl::desc("Profile kind:"), cl::init(instr
),
1049 cl::values(clEnumVal(instr
, "Instrumentation profile (default)"),
1050 clEnumVal(sample
, "Sample profile")));
1051 cl::opt
<uint32_t> TopNFunctions(
1052 "topn", cl::init(0),
1053 cl::desc("Show the list of functions with the largest internal counts"));
1054 cl::opt
<uint32_t> ValueCutoff(
1055 "value-cutoff", cl::init(0),
1056 cl::desc("Set the count value cutoff. Functions with the maximum count "
1057 "less than this value will not be printed out. (Default is 0)"));
1058 cl::opt
<bool> OnlyListBelow(
1059 "list-below-cutoff", cl::init(false),
1060 cl::desc("Only output names of functions whose max count values are "
1061 "below the cutoff value"));
1062 cl::opt
<bool> ShowProfileSymbolList(
1063 "show-prof-sym-list", cl::init(false),
1064 cl::desc("Show profile symbol list if it exists in the profile. "));
1066 cl::ParseCommandLineOptions(argc
, argv
, "LLVM profile data summary\n");
1068 if (OutputFilename
.empty())
1069 OutputFilename
= "-";
1071 if (!Filename
.compare(OutputFilename
)) {
1072 errs() << sys::path::filename(argv
[0])
1073 << ": Input file name cannot be the same as the output file name!\n";
1078 raw_fd_ostream
OS(OutputFilename
.data(), EC
, sys::fs::OF_Text
);
1080 exitWithErrorCode(EC
, OutputFilename
);
1082 if (ShowAllFunctions
&& !ShowFunction
.empty())
1083 WithColor::warning() << "-function argument ignored: showing all functions\n";
1085 if (ProfileKind
== instr
)
1086 return showInstrProfile(Filename
, ShowCounts
, TopNFunctions
,
1087 ShowIndirectCallTargets
, ShowMemOPSizes
,
1088 ShowDetailedSummary
, DetailedSummaryCutoffs
,
1089 ShowAllFunctions
, ShowCS
, ValueCutoff
,
1090 OnlyListBelow
, ShowFunction
, TextFormat
, OS
);
1092 return showSampleProfile(Filename
, ShowCounts
, ShowAllFunctions
,
1093 ShowFunction
, ShowProfileSymbolList
, OS
);
1096 int main(int argc
, const char *argv
[]) {
1097 InitLLVM
X(argc
, argv
);
1099 StringRef
ProgName(sys::path::filename(argv
[0]));
1101 int (*func
)(int, const char *[]) = nullptr;
1103 if (strcmp(argv
[1], "merge") == 0)
1105 else if (strcmp(argv
[1], "show") == 0)
1107 else if (strcmp(argv
[1], "overlap") == 0)
1108 func
= overlap_main
;
1111 std::string
Invocation(ProgName
.str() + " " + argv
[1]);
1112 argv
[1] = Invocation
.c_str();
1113 return func(argc
- 1, argv
+ 1);
1116 if (strcmp(argv
[1], "-h") == 0 || strcmp(argv
[1], "-help") == 0 ||
1117 strcmp(argv
[1], "--help") == 0) {
1119 errs() << "OVERVIEW: LLVM profile data tools\n\n"
1120 << "USAGE: " << ProgName
<< " <command> [args...]\n"
1121 << "USAGE: " << ProgName
<< " <command> -help\n\n"
1122 << "See each individual command --help for more details.\n"
1123 << "Available commands: merge, show, overlap\n";
1129 errs() << ProgName
<< ": No command specified!\n";
1131 errs() << ProgName
<< ": Unknown command!\n";
1133 errs() << "USAGE: " << ProgName
<< " <merge|show|overlap> [args...]\n";