1 //===- llvm-profdata.cpp - LLVM profile data tool -------------------------===//
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 // llvm-profdata merges .profdata files.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "llvm/ADT/SmallSet.h"
15 #include "llvm/ADT/SmallVector.h"
16 #include "llvm/ADT/StringRef.h"
17 #include "llvm/IR/LLVMContext.h"
18 #include "llvm/ProfileData/InstrProfReader.h"
19 #include "llvm/ProfileData/InstrProfWriter.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/ManagedStatic.h"
27 #include "llvm/Support/MemoryBuffer.h"
28 #include "llvm/Support/Path.h"
29 #include "llvm/Support/PrettyStackTrace.h"
30 #include "llvm/Support/Signals.h"
31 #include "llvm/Support/raw_ostream.h"
32 #include <algorithm>
33 #include <tuple>
34 
35 using namespace llvm;
36 
37 enum ProfileFormat { PF_None = 0, PF_Text, PF_Binary, PF_GCC };
38 
39 static void exitWithError(const Twine &Message, StringRef Whence = "",
40                           StringRef Hint = "") {
41   errs() << "error: ";
42   if (!Whence.empty())
43     errs() << Whence << ": ";
44   errs() << Message << "\n";
45   if (!Hint.empty())
46     errs() << Hint << "\n";
47   ::exit(1);
48 }
49 
50 static void exitWithErrorCode(const std::error_code &Error,
51                               StringRef Whence = "") {
52   if (Error.category() == instrprof_category()) {
53     instrprof_error instrError = static_cast<instrprof_error>(Error.value());
54     if (instrError == instrprof_error::unrecognized_format) {
55       // Hint for common error of forgetting -sample for sample profiles.
56       exitWithError(Error.message(), Whence,
57                     "Perhaps you forgot to use the -sample option?");
58     }
59   }
60   exitWithError(Error.message(), Whence);
61 }
62 
63 namespace {
64 enum ProfileKinds { instr, sample };
65 }
66 
67 static void handleMergeWriterError(std::error_code &Error,
68                                    StringRef WhenceFile = "",
69                                    StringRef WhenceFunction = "",
70                                    bool ShowHint = true) {
71   if (!WhenceFile.empty())
72     errs() << WhenceFile << ": ";
73   if (!WhenceFunction.empty())
74     errs() << WhenceFunction << ": ";
75   errs() << Error.message() << "\n";
76 
77   if (ShowHint) {
78     StringRef Hint = "";
79     if (Error.category() == instrprof_category()) {
80       instrprof_error instrError = static_cast<instrprof_error>(Error.value());
81       switch (instrError) {
82       case instrprof_error::hash_mismatch:
83       case instrprof_error::count_mismatch:
84       case instrprof_error::value_site_count_mismatch:
85         Hint = "Make sure that all profile data to be merged is generated "
86                "from the same binary.";
87         break;
88       default:
89         break;
90       }
91     }
92 
93     if (!Hint.empty())
94       errs() << Hint << "\n";
95   }
96 }
97 
98 struct WeightedFile {
99   StringRef Filename;
100   uint64_t Weight;
101 
102   WeightedFile() {}
103 
104   WeightedFile(StringRef F, uint64_t W) : Filename{F}, Weight{W} {}
105 };
106 typedef SmallVector<WeightedFile, 5> WeightedFileVector;
107 
108 static void mergeInstrProfile(const WeightedFileVector &Inputs,
109                               StringRef OutputFilename,
110                               ProfileFormat OutputFormat) {
111   if (OutputFilename.compare("-") == 0)
112     exitWithError("Cannot write indexed profdata format to stdout.");
113 
114   if (OutputFormat != PF_Binary && OutputFormat != PF_Text)
115     exitWithError("Unknown format is specified.");
116 
117   std::error_code EC;
118   raw_fd_ostream Output(OutputFilename.data(), EC, sys::fs::F_None);
119   if (EC)
120     exitWithErrorCode(EC, OutputFilename);
121 
122   InstrProfWriter Writer;
123   SmallSet<std::error_code, 4> WriterErrorCodes;
124   for (const auto &Input : Inputs) {
125     auto ReaderOrErr = InstrProfReader::create(Input.Filename);
126     if (std::error_code ec = ReaderOrErr.getError())
127       exitWithErrorCode(ec, Input.Filename);
128 
129     auto Reader = std::move(ReaderOrErr.get());
130     for (auto &I : *Reader) {
131       if (std::error_code EC = Writer.addRecord(std::move(I), Input.Weight)) {
132         // Only show hint the first time an error occurs.
133         bool firstTime = WriterErrorCodes.insert(EC).second;
134         handleMergeWriterError(EC, Input.Filename, I.Name, firstTime);
135       }
136     }
137     if (Reader->hasError())
138       exitWithErrorCode(Reader->getError(), Input.Filename);
139   }
140   if (OutputFormat == PF_Text)
141     Writer.writeText(Output);
142   else
143     Writer.write(Output);
144 }
145 
146 static sampleprof::SampleProfileFormat FormatMap[] = {
147     sampleprof::SPF_None, sampleprof::SPF_Text, sampleprof::SPF_Binary,
148     sampleprof::SPF_GCC};
149 
150 static void mergeSampleProfile(const WeightedFileVector &Inputs,
151                                StringRef OutputFilename,
152                                ProfileFormat OutputFormat) {
153   using namespace sampleprof;
154   auto WriterOrErr =
155       SampleProfileWriter::create(OutputFilename, FormatMap[OutputFormat]);
156   if (std::error_code EC = WriterOrErr.getError())
157     exitWithErrorCode(EC, OutputFilename);
158 
159   auto Writer = std::move(WriterOrErr.get());
160   StringMap<FunctionSamples> ProfileMap;
161   SmallVector<std::unique_ptr<sampleprof::SampleProfileReader>, 5> Readers;
162   for (const auto &Input : Inputs) {
163     auto ReaderOrErr =
164         SampleProfileReader::create(Input.Filename, getGlobalContext());
165     if (std::error_code EC = ReaderOrErr.getError())
166       exitWithErrorCode(EC, Input.Filename);
167 
168     // We need to keep the readers around until after all the files are
169     // read so that we do not lose the function names stored in each
170     // reader's memory. The function names are needed to write out the
171     // merged profile map.
172     Readers.push_back(std::move(ReaderOrErr.get()));
173     const auto Reader = Readers.back().get();
174     if (std::error_code EC = Reader->read())
175       exitWithErrorCode(EC, Input.Filename);
176 
177     StringMap<FunctionSamples> &Profiles = Reader->getProfiles();
178     for (StringMap<FunctionSamples>::iterator I = Profiles.begin(),
179                                               E = Profiles.end();
180          I != E; ++I) {
181       StringRef FName = I->first();
182       FunctionSamples &Samples = I->second;
183       sampleprof_error Result = ProfileMap[FName].merge(Samples, Input.Weight);
184       if (Result != sampleprof_error::success) {
185         std::error_code EC = make_error_code(Result);
186         handleMergeWriterError(EC, Input.Filename, FName);
187       }
188     }
189   }
190   Writer->write(ProfileMap);
191 }
192 
193 static WeightedFile parseWeightedFile(const StringRef &WeightedFilename) {
194   StringRef WeightStr, FileName;
195   std::tie(WeightStr, FileName) = WeightedFilename.split(',');
196 
197   uint64_t Weight;
198   if (WeightStr.getAsInteger(10, Weight) || Weight < 1)
199     exitWithError("Input weight must be a positive integer.");
200 
201   if (!sys::fs::exists(FileName))
202     exitWithErrorCode(make_error_code(errc::no_such_file_or_directory),
203                       FileName);
204 
205   return WeightedFile(FileName, Weight);
206 }
207 
208 static int merge_main(int argc, const char *argv[]) {
209   cl::list<std::string> InputFilenames(cl::Positional,
210                                        cl::desc("<filename...>"));
211   cl::list<std::string> WeightedInputFilenames("weighted-input",
212                                                cl::desc("<weight>,<filename>"));
213   cl::opt<std::string> OutputFilename("output", cl::value_desc("output"),
214                                       cl::init("-"), cl::Required,
215                                       cl::desc("Output file"));
216   cl::alias OutputFilenameA("o", cl::desc("Alias for --output"),
217                             cl::aliasopt(OutputFilename));
218   cl::opt<ProfileKinds> ProfileKind(
219       cl::desc("Profile kind:"), cl::init(instr),
220       cl::values(clEnumVal(instr, "Instrumentation profile (default)"),
221                  clEnumVal(sample, "Sample profile"), clEnumValEnd));
222 
223   cl::opt<ProfileFormat> OutputFormat(
224       cl::desc("Format of output profile"), cl::init(PF_Binary),
225       cl::values(clEnumValN(PF_Binary, "binary", "Binary encoding (default)"),
226                  clEnumValN(PF_Text, "text", "Text encoding"),
227                  clEnumValN(PF_GCC, "gcc",
228                             "GCC encoding (only meaningful for -sample)"),
229                  clEnumValEnd));
230 
231   cl::ParseCommandLineOptions(argc, argv, "LLVM profile data merger\n");
232 
233   if (InputFilenames.empty() && WeightedInputFilenames.empty())
234     exitWithError("No input files specified. See " +
235                   sys::path::filename(argv[0]) + " -help");
236 
237   WeightedFileVector WeightedInputs;
238   for (StringRef Filename : InputFilenames)
239     WeightedInputs.push_back(WeightedFile(Filename, 1));
240   for (StringRef WeightedFilename : WeightedInputFilenames)
241     WeightedInputs.push_back(parseWeightedFile(WeightedFilename));
242 
243   if (ProfileKind == instr)
244     mergeInstrProfile(WeightedInputs, OutputFilename, OutputFormat);
245   else
246     mergeSampleProfile(WeightedInputs, OutputFilename, OutputFormat);
247 
248   return 0;
249 }
250 
251 static int showInstrProfile(std::string Filename, bool ShowCounts,
252                             bool ShowIndirectCallTargets,
253                             bool ShowDetailedSummary,
254                             std::vector<uint32_t> DetailedSummaryCutoffs,
255                             bool ShowAllFunctions, std::string ShowFunction,
256                             bool TextFormat, raw_fd_ostream &OS) {
257   auto ReaderOrErr = InstrProfReader::create(Filename);
258   std::vector<uint32_t> Cutoffs(DetailedSummaryCutoffs);
259   if (ShowDetailedSummary && DetailedSummaryCutoffs.empty()) {
260     Cutoffs = {800000, 900000, 950000, 990000, 999000, 999900, 999990};
261   }
262   ProfileSummary PS(Cutoffs);
263   if (std::error_code EC = ReaderOrErr.getError())
264     exitWithErrorCode(EC, Filename);
265 
266   auto Reader = std::move(ReaderOrErr.get());
267   size_t ShownFunctions = 0;
268   for (const auto &Func : *Reader) {
269     bool Show =
270         ShowAllFunctions || (!ShowFunction.empty() &&
271                              Func.Name.find(ShowFunction) != Func.Name.npos);
272 
273     bool doTextFormatDump = (Show && ShowCounts && TextFormat);
274 
275     if (doTextFormatDump) {
276       InstrProfSymtab &Symtab = Reader->getSymtab();
277       InstrProfWriter::writeRecordInText(Func, Symtab, OS);
278       continue;
279     }
280 
281     assert(Func.Counts.size() > 0 && "function missing entry counter");
282     PS.addRecord(Func);
283 
284     if (Show) {
285 
286       if (!ShownFunctions)
287         OS << "Counters:\n";
288 
289       ++ShownFunctions;
290 
291       OS << "  " << Func.Name << ":\n"
292          << "    Hash: " << format("0x%016" PRIx64, Func.Hash) << "\n"
293          << "    Counters: " << Func.Counts.size() << "\n"
294          << "    Function count: " << Func.Counts[0] << "\n";
295 
296       if (ShowIndirectCallTargets)
297         OS << "    Indirect Call Site Count: "
298            << Func.getNumValueSites(IPVK_IndirectCallTarget) << "\n";
299 
300       if (ShowCounts) {
301         OS << "    Block counts: [";
302         for (size_t I = 1, E = Func.Counts.size(); I < E; ++I) {
303           OS << (I == 1 ? "" : ", ") << Func.Counts[I];
304         }
305         OS << "]\n";
306       }
307 
308       if (ShowIndirectCallTargets) {
309         InstrProfSymtab &Symtab = Reader->getSymtab();
310         uint32_t NS = Func.getNumValueSites(IPVK_IndirectCallTarget);
311         OS << "    Indirect Target Results: \n";
312         for (size_t I = 0; I < NS; ++I) {
313           uint32_t NV = Func.getNumValueDataForSite(IPVK_IndirectCallTarget, I);
314           std::unique_ptr<InstrProfValueData[]> VD =
315               Func.getValueForSite(IPVK_IndirectCallTarget, I);
316           for (uint32_t V = 0; V < NV; V++) {
317             OS << "\t[ " << I << ", ";
318             OS << Symtab.getFuncName(VD[V].Value) << ", " << VD[V].Count
319                << " ]\n";
320           }
321         }
322       }
323     }
324   }
325 
326   if (Reader->hasError())
327     exitWithErrorCode(Reader->getError(), Filename);
328 
329   if (ShowCounts && TextFormat)
330     return 0;
331 
332   if (ShowAllFunctions || !ShowFunction.empty())
333     OS << "Functions shown: " << ShownFunctions << "\n";
334   OS << "Total functions: " << PS.getNumFunctions() << "\n";
335   OS << "Maximum function count: " << PS.getMaxFunctionCount() << "\n";
336   OS << "Maximum internal block count: " << PS.getMaxInternalBlockCount() << "\n";
337 
338   if (ShowDetailedSummary) {
339     OS << "Detailed summary:\n";
340     OS << "Total number of blocks: " << PS.getNumBlocks() << "\n";
341     OS << "Total count: " << PS.getTotalCount() << "\n";
342     for (auto Entry : PS.getDetailedSummary()) {
343       OS << Entry.NumBlocks << " blocks with count >= " << Entry.MinBlockCount
344          << " account for "
345          << format("%0.6g", (float)Entry.Cutoff / ProfileSummary::Scale * 100)
346          << " percentage of the total counts.\n";
347     }
348   }
349   return 0;
350 }
351 
352 static int showSampleProfile(std::string Filename, bool ShowCounts,
353                              bool ShowAllFunctions, std::string ShowFunction,
354                              raw_fd_ostream &OS) {
355   using namespace sampleprof;
356   auto ReaderOrErr = SampleProfileReader::create(Filename, getGlobalContext());
357   if (std::error_code EC = ReaderOrErr.getError())
358     exitWithErrorCode(EC, Filename);
359 
360   auto Reader = std::move(ReaderOrErr.get());
361   if (std::error_code EC = Reader->read())
362     exitWithErrorCode(EC, Filename);
363 
364   if (ShowAllFunctions || ShowFunction.empty())
365     Reader->dump(OS);
366   else
367     Reader->dumpFunctionProfile(ShowFunction, OS);
368 
369   return 0;
370 }
371 
372 static int show_main(int argc, const char *argv[]) {
373   cl::opt<std::string> Filename(cl::Positional, cl::Required,
374                                 cl::desc("<profdata-file>"));
375 
376   cl::opt<bool> ShowCounts("counts", cl::init(false),
377                            cl::desc("Show counter values for shown functions"));
378   cl::opt<bool> TextFormat(
379       "text", cl::init(false),
380       cl::desc("Show instr profile data in text dump format"));
381   cl::opt<bool> ShowIndirectCallTargets(
382       "ic-targets", cl::init(false),
383       cl::desc("Show indirect call site target values for shown functions"));
384   cl::opt<bool> ShowDetailedSummary("detailed-summary", cl::init(false),
385                                     cl::desc("Show detailed profile summary"));
386   cl::list<uint32_t> DetailedSummaryCutoffs(
387       cl::CommaSeparated, "detailed-summary-cutoffs",
388       cl::desc(
389           "Cutoff percentages (times 10000) for generating detailed summary"),
390       cl::value_desc("800000,901000,999999"));
391   cl::opt<bool> ShowAllFunctions("all-functions", cl::init(false),
392                                  cl::desc("Details for every function"));
393   cl::opt<std::string> ShowFunction("function",
394                                     cl::desc("Details for matching functions"));
395 
396   cl::opt<std::string> OutputFilename("output", cl::value_desc("output"),
397                                       cl::init("-"), cl::desc("Output file"));
398   cl::alias OutputFilenameA("o", cl::desc("Alias for --output"),
399                             cl::aliasopt(OutputFilename));
400   cl::opt<ProfileKinds> ProfileKind(
401       cl::desc("Profile kind:"), cl::init(instr),
402       cl::values(clEnumVal(instr, "Instrumentation profile (default)"),
403                  clEnumVal(sample, "Sample profile"), clEnumValEnd));
404 
405   cl::ParseCommandLineOptions(argc, argv, "LLVM profile data summary\n");
406 
407   if (OutputFilename.empty())
408     OutputFilename = "-";
409 
410   std::error_code EC;
411   raw_fd_ostream OS(OutputFilename.data(), EC, sys::fs::F_Text);
412   if (EC)
413     exitWithErrorCode(EC, OutputFilename);
414 
415   if (ShowAllFunctions && !ShowFunction.empty())
416     errs() << "warning: -function argument ignored: showing all functions\n";
417 
418   std::vector<uint32_t> Cutoffs(DetailedSummaryCutoffs.begin(),
419                                 DetailedSummaryCutoffs.end());
420   if (ProfileKind == instr)
421     return showInstrProfile(Filename, ShowCounts, ShowIndirectCallTargets,
422                             ShowDetailedSummary, DetailedSummaryCutoffs,
423                             ShowAllFunctions, ShowFunction, TextFormat, OS);
424   else
425     return showSampleProfile(Filename, ShowCounts, ShowAllFunctions,
426                              ShowFunction, OS);
427 }
428 
429 int main(int argc, const char *argv[]) {
430   // Print a stack trace if we signal out.
431   sys::PrintStackTraceOnErrorSignal();
432   PrettyStackTraceProgram X(argc, argv);
433   llvm_shutdown_obj Y; // Call llvm_shutdown() on exit.
434 
435   StringRef ProgName(sys::path::filename(argv[0]));
436   if (argc > 1) {
437     int (*func)(int, const char *[]) = nullptr;
438 
439     if (strcmp(argv[1], "merge") == 0)
440       func = merge_main;
441     else if (strcmp(argv[1], "show") == 0)
442       func = show_main;
443 
444     if (func) {
445       std::string Invocation(ProgName.str() + " " + argv[1]);
446       argv[1] = Invocation.c_str();
447       return func(argc - 1, argv + 1);
448     }
449 
450     if (strcmp(argv[1], "-h") == 0 || strcmp(argv[1], "-help") == 0 ||
451         strcmp(argv[1], "--help") == 0) {
452 
453       errs() << "OVERVIEW: LLVM profile data tools\n\n"
454              << "USAGE: " << ProgName << " <command> [args...]\n"
455              << "USAGE: " << ProgName << " <command> -help\n\n"
456              << "Available commands: merge, show\n";
457       return 0;
458     }
459   }
460 
461   if (argc < 2)
462     errs() << ProgName << ": No command specified!\n";
463   else
464     errs() << ProgName << ": Unknown command!\n";
465 
466   errs() << "USAGE: " << ProgName << " <merge|show> [args...]\n";
467   return 1;
468 }
469