Austin Schuh | cbc1740 | 2019-01-21 21:00:30 -0800 | [diff] [blame] | 1 | // Copyright 2015 Google Inc. All rights reserved. |
| 2 | // |
| 3 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | // you may not use this file except in compliance with the License. |
| 5 | // You may obtain a copy of the License at |
| 6 | // |
| 7 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 8 | // |
| 9 | // Unless required by applicable law or agreed to in writing, software |
| 10 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | // See the License for the specific language governing permissions and |
| 13 | // limitations under the License. |
| 14 | |
| 15 | #include "benchmark/benchmark.h" |
| 16 | #include "benchmark_api_internal.h" |
| 17 | #include "benchmark_runner.h" |
| 18 | #include "internal_macros.h" |
| 19 | |
| 20 | #ifndef BENCHMARK_OS_WINDOWS |
| 21 | #ifndef BENCHMARK_OS_FUCHSIA |
| 22 | #include <sys/resource.h> |
| 23 | #endif |
| 24 | #include <sys/time.h> |
| 25 | #include <unistd.h> |
| 26 | #endif |
| 27 | |
| 28 | #include <algorithm> |
| 29 | #include <atomic> |
| 30 | #include <condition_variable> |
| 31 | #include <cstdio> |
| 32 | #include <cstdlib> |
| 33 | #include <fstream> |
| 34 | #include <iostream> |
| 35 | #include <memory> |
| 36 | #include <string> |
| 37 | #include <thread> |
| 38 | #include <utility> |
| 39 | |
| 40 | #include "check.h" |
| 41 | #include "colorprint.h" |
| 42 | #include "commandlineflags.h" |
| 43 | #include "complexity.h" |
| 44 | #include "counter.h" |
| 45 | #include "internal_macros.h" |
| 46 | #include "log.h" |
| 47 | #include "mutex.h" |
| 48 | #include "re.h" |
| 49 | #include "statistics.h" |
| 50 | #include "string_util.h" |
| 51 | #include "thread_manager.h" |
| 52 | #include "thread_timer.h" |
| 53 | |
| 54 | DEFINE_bool(benchmark_list_tests, false, |
| 55 | "Print a list of benchmarks. This option overrides all other " |
| 56 | "options."); |
| 57 | |
| 58 | DEFINE_string(benchmark_filter, ".", |
| 59 | "A regular expression that specifies the set of benchmarks " |
| 60 | "to execute. If this flag is empty, or if this flag is the " |
| 61 | "string \"all\", all benchmarks linked into the binary are " |
| 62 | "run."); |
| 63 | |
| 64 | DEFINE_double(benchmark_min_time, 0.5, |
| 65 | "Minimum number of seconds we should run benchmark before " |
| 66 | "results are considered significant. For cpu-time based " |
| 67 | "tests, this is the lower bound on the total cpu time " |
| 68 | "used by all threads that make up the test. For real-time " |
| 69 | "based tests, this is the lower bound on the elapsed time " |
| 70 | "of the benchmark execution, regardless of number of " |
| 71 | "threads."); |
| 72 | |
| 73 | DEFINE_int32(benchmark_repetitions, 1, |
| 74 | "The number of runs of each benchmark. If greater than 1, the " |
| 75 | "mean and standard deviation of the runs will be reported."); |
| 76 | |
| 77 | DEFINE_bool( |
| 78 | benchmark_report_aggregates_only, false, |
| 79 | "Report the result of each benchmark repetitions. When 'true' is specified " |
| 80 | "only the mean, standard deviation, and other statistics are reported for " |
| 81 | "repeated benchmarks. Affects all reporters."); |
| 82 | |
| 83 | DEFINE_bool( |
| 84 | benchmark_display_aggregates_only, false, |
| 85 | "Display the result of each benchmark repetitions. When 'true' is " |
| 86 | "specified only the mean, standard deviation, and other statistics are " |
| 87 | "displayed for repeated benchmarks. Unlike " |
| 88 | "benchmark_report_aggregates_only, only affects the display reporter, but " |
| 89 | "*NOT* file reporter, which will still contain all the output."); |
| 90 | |
| 91 | DEFINE_string(benchmark_format, "console", |
| 92 | "The format to use for console output. Valid values are " |
| 93 | "'console', 'json', or 'csv'."); |
| 94 | |
| 95 | DEFINE_string(benchmark_out_format, "json", |
| 96 | "The format to use for file output. Valid values are " |
| 97 | "'console', 'json', or 'csv'."); |
| 98 | |
| 99 | DEFINE_string(benchmark_out, "", "The file to write additional output to"); |
| 100 | |
| 101 | DEFINE_string(benchmark_color, "auto", |
| 102 | "Whether to use colors in the output. Valid values: " |
| 103 | "'true'/'yes'/1, 'false'/'no'/0, and 'auto'. 'auto' means to use " |
| 104 | "colors if the output is being sent to a terminal and the TERM " |
| 105 | "environment variable is set to a terminal type that supports " |
| 106 | "colors."); |
| 107 | |
| 108 | DEFINE_bool(benchmark_counters_tabular, false, |
| 109 | "Whether to use tabular format when printing user counters to " |
| 110 | "the console. Valid values: 'true'/'yes'/1, 'false'/'no'/0." |
| 111 | "Defaults to false."); |
| 112 | |
| 113 | DEFINE_int32(v, 0, "The level of verbose logging to output"); |
| 114 | |
| 115 | namespace benchmark { |
| 116 | |
| 117 | namespace internal { |
| 118 | |
| 119 | // FIXME: wouldn't LTO mess this up? |
| 120 | void UseCharPointer(char const volatile*) {} |
| 121 | |
| 122 | } // namespace internal |
| 123 | |
| 124 | State::State(size_t max_iters, const std::vector<int64_t>& ranges, int thread_i, |
| 125 | int n_threads, internal::ThreadTimer* timer, |
| 126 | internal::ThreadManager* manager) |
| 127 | : total_iterations_(0), |
| 128 | batch_leftover_(0), |
| 129 | max_iterations(max_iters), |
| 130 | started_(false), |
| 131 | finished_(false), |
| 132 | error_occurred_(false), |
| 133 | range_(ranges), |
| 134 | complexity_n_(0), |
| 135 | counters(), |
| 136 | thread_index(thread_i), |
| 137 | threads(n_threads), |
| 138 | timer_(timer), |
| 139 | manager_(manager) { |
| 140 | CHECK(max_iterations != 0) << "At least one iteration must be run"; |
| 141 | CHECK_LT(thread_index, threads) << "thread_index must be less than threads"; |
| 142 | |
| 143 | // Note: The use of offsetof below is technically undefined until C++17 |
| 144 | // because State is not a standard layout type. However, all compilers |
| 145 | // currently provide well-defined behavior as an extension (which is |
| 146 | // demonstrated since constexpr evaluation must diagnose all undefined |
| 147 | // behavior). However, GCC and Clang also warn about this use of offsetof, |
| 148 | // which must be suppressed. |
| 149 | #if defined(__INTEL_COMPILER) |
| 150 | #pragma warning push |
| 151 | #pragma warning(disable:1875) |
| 152 | #elif defined(__GNUC__) |
| 153 | #pragma GCC diagnostic push |
| 154 | #pragma GCC diagnostic ignored "-Winvalid-offsetof" |
| 155 | #endif |
| 156 | // Offset tests to ensure commonly accessed data is on the first cache line. |
| 157 | const int cache_line_size = 64; |
| 158 | static_assert(offsetof(State, error_occurred_) <= |
| 159 | (cache_line_size - sizeof(error_occurred_)), |
| 160 | ""); |
| 161 | #if defined(__INTEL_COMPILER) |
| 162 | #pragma warning pop |
| 163 | #elif defined(__GNUC__) |
| 164 | #pragma GCC diagnostic pop |
| 165 | #endif |
| 166 | } |
| 167 | |
| 168 | void State::PauseTiming() { |
| 169 | // Add in time accumulated so far |
| 170 | CHECK(started_ && !finished_ && !error_occurred_); |
| 171 | timer_->StopTimer(); |
| 172 | } |
| 173 | |
| 174 | void State::ResumeTiming() { |
| 175 | CHECK(started_ && !finished_ && !error_occurred_); |
| 176 | timer_->StartTimer(); |
| 177 | } |
| 178 | |
| 179 | void State::SkipWithError(const char* msg) { |
| 180 | CHECK(msg); |
| 181 | error_occurred_ = true; |
| 182 | { |
| 183 | MutexLock l(manager_->GetBenchmarkMutex()); |
| 184 | if (manager_->results.has_error_ == false) { |
| 185 | manager_->results.error_message_ = msg; |
| 186 | manager_->results.has_error_ = true; |
| 187 | } |
| 188 | } |
| 189 | total_iterations_ = 0; |
| 190 | if (timer_->running()) timer_->StopTimer(); |
| 191 | } |
| 192 | |
| 193 | void State::SetIterationTime(double seconds) { |
| 194 | timer_->SetIterationTime(seconds); |
| 195 | } |
| 196 | |
| 197 | void State::SetLabel(const char* label) { |
| 198 | MutexLock l(manager_->GetBenchmarkMutex()); |
| 199 | manager_->results.report_label_ = label; |
| 200 | } |
| 201 | |
| 202 | void State::StartKeepRunning() { |
| 203 | CHECK(!started_ && !finished_); |
| 204 | started_ = true; |
| 205 | total_iterations_ = error_occurred_ ? 0 : max_iterations; |
| 206 | manager_->StartStopBarrier(); |
| 207 | if (!error_occurred_) ResumeTiming(); |
| 208 | } |
| 209 | |
| 210 | void State::FinishKeepRunning() { |
| 211 | CHECK(started_ && (!finished_ || error_occurred_)); |
| 212 | if (!error_occurred_) { |
| 213 | PauseTiming(); |
| 214 | } |
| 215 | // Total iterations has now wrapped around past 0. Fix this. |
| 216 | total_iterations_ = 0; |
| 217 | finished_ = true; |
| 218 | manager_->StartStopBarrier(); |
| 219 | } |
| 220 | |
| 221 | namespace internal { |
| 222 | namespace { |
| 223 | |
| 224 | void RunBenchmarks(const std::vector<BenchmarkInstance>& benchmarks, |
| 225 | BenchmarkReporter* display_reporter, |
| 226 | BenchmarkReporter* file_reporter) { |
| 227 | // Note the file_reporter can be null. |
| 228 | CHECK(display_reporter != nullptr); |
| 229 | |
| 230 | // Determine the width of the name field using a minimum width of 10. |
| 231 | bool might_have_aggregates = FLAGS_benchmark_repetitions > 1; |
| 232 | size_t name_field_width = 10; |
| 233 | size_t stat_field_width = 0; |
| 234 | for (const BenchmarkInstance& benchmark : benchmarks) { |
| 235 | name_field_width = |
| 236 | std::max<size_t>(name_field_width, benchmark.name.size()); |
| 237 | might_have_aggregates |= benchmark.repetitions > 1; |
| 238 | |
| 239 | for (const auto& Stat : *benchmark.statistics) |
| 240 | stat_field_width = std::max<size_t>(stat_field_width, Stat.name_.size()); |
| 241 | } |
| 242 | if (might_have_aggregates) name_field_width += 1 + stat_field_width; |
| 243 | |
| 244 | // Print header here |
| 245 | BenchmarkReporter::Context context; |
| 246 | context.name_field_width = name_field_width; |
| 247 | |
| 248 | // Keep track of running times of all instances of current benchmark |
| 249 | std::vector<BenchmarkReporter::Run> complexity_reports; |
| 250 | |
| 251 | // We flush streams after invoking reporter methods that write to them. This |
| 252 | // ensures users get timely updates even when streams are not line-buffered. |
| 253 | auto flushStreams = [](BenchmarkReporter* reporter) { |
| 254 | if (!reporter) return; |
| 255 | std::flush(reporter->GetOutputStream()); |
| 256 | std::flush(reporter->GetErrorStream()); |
| 257 | }; |
| 258 | |
| 259 | if (display_reporter->ReportContext(context) && |
| 260 | (!file_reporter || file_reporter->ReportContext(context))) { |
| 261 | flushStreams(display_reporter); |
| 262 | flushStreams(file_reporter); |
| 263 | |
| 264 | for (const auto& benchmark : benchmarks) { |
| 265 | RunResults run_results = RunBenchmark(benchmark, &complexity_reports); |
| 266 | |
| 267 | auto report = [&run_results](BenchmarkReporter* reporter, |
| 268 | bool report_aggregates_only) { |
| 269 | assert(reporter); |
| 270 | // If there are no aggregates, do output non-aggregates. |
| 271 | report_aggregates_only &= !run_results.aggregates_only.empty(); |
| 272 | if (!report_aggregates_only) |
| 273 | reporter->ReportRuns(run_results.non_aggregates); |
| 274 | if (!run_results.aggregates_only.empty()) |
| 275 | reporter->ReportRuns(run_results.aggregates_only); |
| 276 | }; |
| 277 | |
| 278 | report(display_reporter, run_results.display_report_aggregates_only); |
| 279 | if (file_reporter) |
| 280 | report(file_reporter, run_results.file_report_aggregates_only); |
| 281 | |
| 282 | flushStreams(display_reporter); |
| 283 | flushStreams(file_reporter); |
| 284 | } |
| 285 | } |
| 286 | display_reporter->Finalize(); |
| 287 | if (file_reporter) file_reporter->Finalize(); |
| 288 | flushStreams(display_reporter); |
| 289 | flushStreams(file_reporter); |
| 290 | } |
| 291 | |
| 292 | std::unique_ptr<BenchmarkReporter> CreateReporter( |
| 293 | std::string const& name, ConsoleReporter::OutputOptions output_opts) { |
| 294 | typedef std::unique_ptr<BenchmarkReporter> PtrType; |
| 295 | if (name == "console") { |
| 296 | return PtrType(new ConsoleReporter(output_opts)); |
| 297 | } else if (name == "json") { |
| 298 | return PtrType(new JSONReporter); |
| 299 | } else if (name == "csv") { |
| 300 | return PtrType(new CSVReporter); |
| 301 | } else { |
| 302 | std::cerr << "Unexpected format: '" << name << "'\n"; |
| 303 | std::exit(1); |
| 304 | } |
| 305 | } |
| 306 | |
| 307 | } // end namespace |
| 308 | |
| 309 | bool IsZero(double n) { |
| 310 | return std::abs(n) < std::numeric_limits<double>::epsilon(); |
| 311 | } |
| 312 | |
| 313 | ConsoleReporter::OutputOptions GetOutputOptions(bool force_no_color) { |
| 314 | int output_opts = ConsoleReporter::OO_Defaults; |
| 315 | auto is_benchmark_color = [force_no_color] () -> bool { |
| 316 | if (force_no_color) { |
| 317 | return false; |
| 318 | } |
| 319 | if (FLAGS_benchmark_color == "auto") { |
| 320 | return IsColorTerminal(); |
| 321 | } |
| 322 | return IsTruthyFlagValue(FLAGS_benchmark_color); |
| 323 | }; |
| 324 | if (is_benchmark_color()) { |
| 325 | output_opts |= ConsoleReporter::OO_Color; |
| 326 | } else { |
| 327 | output_opts &= ~ConsoleReporter::OO_Color; |
| 328 | } |
| 329 | if (FLAGS_benchmark_counters_tabular) { |
| 330 | output_opts |= ConsoleReporter::OO_Tabular; |
| 331 | } else { |
| 332 | output_opts &= ~ConsoleReporter::OO_Tabular; |
| 333 | } |
| 334 | return static_cast<ConsoleReporter::OutputOptions>(output_opts); |
| 335 | } |
| 336 | |
| 337 | } // end namespace internal |
| 338 | |
| 339 | size_t RunSpecifiedBenchmarks() { |
| 340 | return RunSpecifiedBenchmarks(nullptr, nullptr); |
| 341 | } |
| 342 | |
| 343 | size_t RunSpecifiedBenchmarks(BenchmarkReporter* display_reporter) { |
| 344 | return RunSpecifiedBenchmarks(display_reporter, nullptr); |
| 345 | } |
| 346 | |
| 347 | size_t RunSpecifiedBenchmarks(BenchmarkReporter* display_reporter, |
| 348 | BenchmarkReporter* file_reporter) { |
| 349 | std::string spec = FLAGS_benchmark_filter; |
| 350 | if (spec.empty() || spec == "all") |
| 351 | spec = "."; // Regexp that matches all benchmarks |
| 352 | |
| 353 | // Setup the reporters |
| 354 | std::ofstream output_file; |
| 355 | std::unique_ptr<BenchmarkReporter> default_display_reporter; |
| 356 | std::unique_ptr<BenchmarkReporter> default_file_reporter; |
| 357 | if (!display_reporter) { |
| 358 | default_display_reporter = internal::CreateReporter( |
| 359 | FLAGS_benchmark_format, internal::GetOutputOptions()); |
| 360 | display_reporter = default_display_reporter.get(); |
| 361 | } |
| 362 | auto& Out = display_reporter->GetOutputStream(); |
| 363 | auto& Err = display_reporter->GetErrorStream(); |
| 364 | |
| 365 | std::string const& fname = FLAGS_benchmark_out; |
| 366 | if (fname.empty() && file_reporter) { |
| 367 | Err << "A custom file reporter was provided but " |
| 368 | "--benchmark_out=<file> was not specified." |
| 369 | << std::endl; |
| 370 | std::exit(1); |
| 371 | } |
| 372 | if (!fname.empty()) { |
| 373 | output_file.open(fname); |
| 374 | if (!output_file.is_open()) { |
| 375 | Err << "invalid file name: '" << fname << std::endl; |
| 376 | std::exit(1); |
| 377 | } |
| 378 | if (!file_reporter) { |
| 379 | default_file_reporter = internal::CreateReporter( |
| 380 | FLAGS_benchmark_out_format, ConsoleReporter::OO_None); |
| 381 | file_reporter = default_file_reporter.get(); |
| 382 | } |
| 383 | file_reporter->SetOutputStream(&output_file); |
| 384 | file_reporter->SetErrorStream(&output_file); |
| 385 | } |
| 386 | |
| 387 | std::vector<internal::BenchmarkInstance> benchmarks; |
| 388 | if (!FindBenchmarksInternal(spec, &benchmarks, &Err)) return 0; |
| 389 | |
| 390 | if (benchmarks.empty()) { |
| 391 | Err << "Failed to match any benchmarks against regex: " << spec << "\n"; |
| 392 | return 0; |
| 393 | } |
| 394 | |
| 395 | if (FLAGS_benchmark_list_tests) { |
| 396 | for (auto const& benchmark : benchmarks) Out << benchmark.name << "\n"; |
| 397 | } else { |
| 398 | internal::RunBenchmarks(benchmarks, display_reporter, file_reporter); |
| 399 | } |
| 400 | |
| 401 | return benchmarks.size(); |
| 402 | } |
| 403 | |
| 404 | void RegisterMemoryManager(MemoryManager* manager) { |
| 405 | internal::memory_manager = manager; |
| 406 | } |
| 407 | |
| 408 | namespace internal { |
| 409 | |
| 410 | void PrintUsageAndExit() { |
| 411 | fprintf(stdout, |
| 412 | "benchmark" |
| 413 | " [--benchmark_list_tests={true|false}]\n" |
| 414 | " [--benchmark_filter=<regex>]\n" |
| 415 | " [--benchmark_min_time=<min_time>]\n" |
| 416 | " [--benchmark_repetitions=<num_repetitions>]\n" |
| 417 | " [--benchmark_report_aggregates_only={true|false}]\n" |
| 418 | " [--benchmark_display_aggregates_only={true|false}]\n" |
| 419 | " [--benchmark_format=<console|json|csv>]\n" |
| 420 | " [--benchmark_out=<filename>]\n" |
| 421 | " [--benchmark_out_format=<json|console|csv>]\n" |
| 422 | " [--benchmark_color={auto|true|false}]\n" |
| 423 | " [--benchmark_counters_tabular={true|false}]\n" |
| 424 | " [--v=<verbosity>]\n"); |
| 425 | exit(0); |
| 426 | } |
| 427 | |
| 428 | void ParseCommandLineFlags(int* argc, char** argv) { |
| 429 | using namespace benchmark; |
| 430 | BenchmarkReporter::Context::executable_name = |
| 431 | (argc && *argc > 0) ? argv[0] : "unknown"; |
| 432 | for (int i = 1; i < *argc; ++i) { |
| 433 | if (ParseBoolFlag(argv[i], "benchmark_list_tests", |
| 434 | &FLAGS_benchmark_list_tests) || |
| 435 | ParseStringFlag(argv[i], "benchmark_filter", &FLAGS_benchmark_filter) || |
| 436 | ParseDoubleFlag(argv[i], "benchmark_min_time", |
| 437 | &FLAGS_benchmark_min_time) || |
| 438 | ParseInt32Flag(argv[i], "benchmark_repetitions", |
| 439 | &FLAGS_benchmark_repetitions) || |
| 440 | ParseBoolFlag(argv[i], "benchmark_report_aggregates_only", |
| 441 | &FLAGS_benchmark_report_aggregates_only) || |
| 442 | ParseBoolFlag(argv[i], "benchmark_display_aggregates_only", |
| 443 | &FLAGS_benchmark_display_aggregates_only) || |
| 444 | ParseStringFlag(argv[i], "benchmark_format", &FLAGS_benchmark_format) || |
| 445 | ParseStringFlag(argv[i], "benchmark_out", &FLAGS_benchmark_out) || |
| 446 | ParseStringFlag(argv[i], "benchmark_out_format", |
| 447 | &FLAGS_benchmark_out_format) || |
| 448 | ParseStringFlag(argv[i], "benchmark_color", &FLAGS_benchmark_color) || |
| 449 | // "color_print" is the deprecated name for "benchmark_color". |
| 450 | // TODO: Remove this. |
| 451 | ParseStringFlag(argv[i], "color_print", &FLAGS_benchmark_color) || |
| 452 | ParseBoolFlag(argv[i], "benchmark_counters_tabular", |
| 453 | &FLAGS_benchmark_counters_tabular) || |
| 454 | ParseInt32Flag(argv[i], "v", &FLAGS_v)) { |
| 455 | for (int j = i; j != *argc - 1; ++j) argv[j] = argv[j + 1]; |
| 456 | |
| 457 | --(*argc); |
| 458 | --i; |
| 459 | } else if (IsFlag(argv[i], "help")) { |
| 460 | PrintUsageAndExit(); |
| 461 | } |
| 462 | } |
| 463 | for (auto const* flag : |
| 464 | {&FLAGS_benchmark_format, &FLAGS_benchmark_out_format}) |
| 465 | if (*flag != "console" && *flag != "json" && *flag != "csv") { |
| 466 | PrintUsageAndExit(); |
| 467 | } |
| 468 | if (FLAGS_benchmark_color.empty()) { |
| 469 | PrintUsageAndExit(); |
| 470 | } |
| 471 | } |
| 472 | |
| 473 | int InitializeStreams() { |
| 474 | static std::ios_base::Init init; |
| 475 | return 0; |
| 476 | } |
| 477 | |
| 478 | } // end namespace internal |
| 479 | |
| 480 | void Initialize(int* argc, char** argv) { |
| 481 | internal::ParseCommandLineFlags(argc, argv); |
| 482 | internal::LogLevel() = FLAGS_v; |
| 483 | } |
| 484 | |
| 485 | bool ReportUnrecognizedArguments(int argc, char** argv) { |
| 486 | for (int i = 1; i < argc; ++i) { |
| 487 | fprintf(stderr, "%s: error: unrecognized command-line flag: %s\n", argv[0], |
| 488 | argv[i]); |
| 489 | } |
| 490 | return argc > 1; |
| 491 | } |
| 492 | |
| 493 | } // end namespace benchmark |