diff options
Diffstat (limited to 'utils/google-benchmark/src')
| -rw-r--r-- | utils/google-benchmark/src/CMakeLists.txt | 45 | ||||
| -rw-r--r-- | utils/google-benchmark/src/benchmark.cc | 37 | ||||
| -rw-r--r-- | utils/google-benchmark/src/benchmark_api_internal.h | 2 | ||||
| -rw-r--r-- | utils/google-benchmark/src/benchmark_register.cc | 35 | ||||
| -rw-r--r-- | utils/google-benchmark/src/complexity.cc | 40 | ||||
| -rw-r--r-- | utils/google-benchmark/src/console_reporter.cc | 31 | ||||
| -rw-r--r-- | utils/google-benchmark/src/counter.cc | 68 | ||||
| -rw-r--r-- | utils/google-benchmark/src/counter.h | 26 | ||||
| -rw-r--r-- | utils/google-benchmark/src/csv_reporter.cc | 56 | ||||
| -rw-r--r-- | utils/google-benchmark/src/cycleclock.h | 31 | ||||
| -rw-r--r-- | utils/google-benchmark/src/internal_macros.h | 19 | ||||
| -rw-r--r-- | utils/google-benchmark/src/json_reporter.cc | 7 | ||||
| -rw-r--r-- | utils/google-benchmark/src/re.h | 16 | ||||
| -rw-r--r-- | utils/google-benchmark/src/sleep.cc | 3 | ||||
| -rw-r--r-- | utils/google-benchmark/src/sleep.h | 12 | ||||
| -rw-r--r-- | utils/google-benchmark/src/string_util.cc | 20 | ||||
| -rw-r--r-- | utils/google-benchmark/src/sysinfo.cc | 9 | ||||
| -rw-r--r-- | utils/google-benchmark/src/timers.cc | 45 |
18 files changed, 424 insertions, 78 deletions
diff --git a/utils/google-benchmark/src/CMakeLists.txt b/utils/google-benchmark/src/CMakeLists.txt index 403887517797..7707773930c5 100644 --- a/utils/google-benchmark/src/CMakeLists.txt +++ b/utils/google-benchmark/src/CMakeLists.txt @@ -21,24 +21,55 @@ set_target_properties(benchmark PROPERTIES # Link threads. target_link_libraries(benchmark ${BENCHMARK_CXX_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT}) +find_library(LIBRT rt) +if(LIBRT) + target_link_libraries(benchmark ${LIBRT}) +endif() # We need extra libraries on Windows if(${CMAKE_SYSTEM_NAME} MATCHES "Windows") target_link_libraries(benchmark Shlwapi) endif() -# Expose public API -target_include_directories(benchmark PUBLIC ${PROJECT_SOURCE_DIR}/include) +set(include_install_dir "include") +set(lib_install_dir "lib/") +set(bin_install_dir "bin/") +set(config_install_dir "lib/cmake/${PROJECT_NAME}") + +set(generated_dir "${CMAKE_CURRENT_BINARY_DIR}/generated") + +set(version_config "${generated_dir}/${PROJECT_NAME}ConfigVersion.cmake") +set(project_config "${generated_dir}/${PROJECT_NAME}Config.cmake") +set(targets_export_name "${PROJECT_NAME}Targets") + +set(namespace "${PROJECT_NAME}::") + +include(CMakePackageConfigHelpers) +write_basic_package_version_file( + "${version_config}" VERSION ${GIT_VERSION} COMPATIBILITY SameMajorVersion +) + +configure_file("${PROJECT_SOURCE_DIR}/cmake/Config.cmake.in" "${project_config}" @ONLY) # Install target (will install the library to specified CMAKE_INSTALL_PREFIX variable) install( TARGETS benchmark - ARCHIVE DESTINATION lib - LIBRARY DESTINATION lib - RUNTIME DESTINATION bin - COMPONENT library) + EXPORT ${targets_export_name} + ARCHIVE DESTINATION ${lib_install_dir} + LIBRARY DESTINATION ${lib_install_dir} + RUNTIME DESTINATION ${bin_install_dir} + INCLUDES DESTINATION ${include_install_dir}) install( DIRECTORY "${PROJECT_SOURCE_DIR}/include/benchmark" - DESTINATION include + DESTINATION ${include_install_dir} FILES_MATCHING PATTERN "*.*h") + +install( + FILES "${project_config}" "${version_config}" + DESTINATION "${config_install_dir}") + +install( + EXPORT "${targets_export_name}" + NAMESPACE "${namespace}" + DESTINATION "${config_install_dir}") diff --git a/utils/google-benchmark/src/benchmark.cc b/utils/google-benchmark/src/benchmark.cc index 95f6a25b7563..00ffa07ff56e 100644 --- a/utils/google-benchmark/src/benchmark.cc +++ b/utils/google-benchmark/src/benchmark.cc @@ -37,6 +37,7 @@ #include "colorprint.h" #include "commandlineflags.h" #include "complexity.h" +#include "counter.h" #include "log.h" #include "mutex.h" #include "re.h" @@ -145,6 +146,7 @@ class ThreadManager { std::string report_label_; std::string error_message_; bool has_error_ = false; + UserCounters counters; }; GUARDED_BY(GetBenchmarkMutex()) Result results; @@ -249,6 +251,7 @@ BenchmarkReporter::Run CreateRunReport( report.complexity_n = results.complexity_n; report.complexity = b.complexity; report.complexity_lambda = b.complexity_lambda; + report.counters = results.counters; } return report; } @@ -272,6 +275,7 @@ void RunInThread(const benchmark::internal::Benchmark::Instance* b, results.bytes_processed += st.bytes_processed(); results.items_processed += st.items_processed(); results.complexity_n += st.complexity_length_n(); + internal::Increment(&results.counters, st.counters); } manager->NotifyThreadComplete(); } @@ -281,7 +285,8 @@ std::vector<BenchmarkReporter::Run> RunBenchmark( std::vector<BenchmarkReporter::Run>* complexity_reports) { std::vector<BenchmarkReporter::Run> reports; // return value - size_t iters = 1; + const bool has_explicit_iteration_count = b.iterations != 0; + size_t iters = has_explicit_iteration_count ? b.iterations : 1; std::unique_ptr<internal::ThreadManager> manager; std::vector<std::thread> pool(b.threads - 1); const int repeats = @@ -291,7 +296,7 @@ std::vector<BenchmarkReporter::Run> RunBenchmark( (b.report_mode == internal::RM_Unspecified ? FLAGS_benchmark_report_aggregates_only : b.report_mode == internal::RM_ReportAggregatesOnly); - for (int i = 0; i < repeats; i++) { + for (int repetition_num = 0; repetition_num < repeats; repetition_num++) { for (;;) { // Try benchmark VLOG(2) << "Running " << b.name << " for " << iters << "\n"; @@ -327,10 +332,20 @@ std::vector<BenchmarkReporter::Run> RunBenchmark( const double min_time = !IsZero(b.min_time) ? b.min_time : FLAGS_benchmark_min_time; - // If this was the first run, was elapsed time or cpu time large enough? - // If this is not the first run, go with the current value of iter. - if ((i > 0) || results.has_error_ || (iters >= kMaxIterations) || - (seconds >= min_time) || (results.real_time_used >= 5 * min_time)) { + + // Determine if this run should be reported; Either it has + // run for a sufficient amount of time or because an error was reported. + const bool should_report = repetition_num > 0 + || has_explicit_iteration_count // An exact iteration count was requested + || results.has_error_ + || iters >= kMaxIterations + || seconds >= min_time // the elapsed time is large enough + // CPU time is specified but the elapsed real time greatly exceeds the + // minimum time. Note that user provided timers are except from this + // sanity check. + || ((results.real_time_used >= 5 * min_time) && !b.use_manual_time); + + if (should_report) { BenchmarkReporter::Run report = CreateRunReport(b, results, iters, seconds); if (!report.error_occurred && b.complexity != oNone) @@ -386,6 +401,7 @@ State::State(size_t max_iters, const std::vector<int>& ranges, int thread_i, items_processed_(0), complexity_n_(0), error_occurred_(false), + counters(), thread_index(thread_i), threads(n_threads), max_iterations(max_iters), @@ -634,7 +650,7 @@ void ParseCommandLineFlags(int* argc, char** argv) { // TODO: Remove this. ParseStringFlag(argv[i], "color_print", &FLAGS_benchmark_color) || ParseInt32Flag(argv[i], "v", &FLAGS_v)) { - for (int j = i; j != *argc; ++j) argv[j] = argv[j + 1]; + for (int j = i; j != *argc - 1; ++j) argv[j] = argv[j + 1]; --(*argc); --i; @@ -664,4 +680,11 @@ void Initialize(int* argc, char** argv) { internal::LogLevel() = FLAGS_v; } +bool ReportUnrecognizedArguments(int argc, char** argv) { + for (int i = 1; i < argc; ++i) { + fprintf(stderr, "%s: error: unrecognized command-line flag: %s\n", argv[0], argv[i]); + } + return argc > 1; +} + } // end namespace benchmark diff --git a/utils/google-benchmark/src/benchmark_api_internal.h b/utils/google-benchmark/src/benchmark_api_internal.h index 8b97ce600873..828ed12165ff 100644 --- a/utils/google-benchmark/src/benchmark_api_internal.h +++ b/utils/google-benchmark/src/benchmark_api_internal.h @@ -24,9 +24,11 @@ struct Benchmark::Instance { bool use_manual_time; BigO complexity; BigOFunc* complexity_lambda; + UserCounters counters; bool last_benchmark_instance; int repetitions; double min_time; + size_t iterations; int threads; // Number of concurrent threads to us }; diff --git a/utils/google-benchmark/src/benchmark_register.cc b/utils/google-benchmark/src/benchmark_register.cc index 4e580d8ea383..fe373204189f 100644 --- a/utils/google-benchmark/src/benchmark_register.cc +++ b/utils/google-benchmark/src/benchmark_register.cc @@ -143,6 +143,7 @@ bool BenchmarkFamilies::FindBenchmarks( instance.time_unit = family->time_unit_; instance.range_multiplier = family->range_multiplier_; instance.min_time = family->min_time_; + instance.iterations = family->iterations_; instance.repetitions = family->repetitions_; instance.use_real_time = family->use_real_time_; instance.use_manual_time = family->use_manual_time_; @@ -163,16 +164,17 @@ bool BenchmarkFamilies::FindBenchmarks( } } - AppendHumanReadable(arg, &instance.name); + instance.name += std::to_string(arg); ++arg_i; } - if (!IsZero(family->min_time_)) { + if (!IsZero(family->min_time_)) instance.name += StringPrintF("/min_time:%0.3f", family->min_time_); - } - if (family->repetitions_ != 0) { + if (family->iterations_ != 0) + instance.name += StringPrintF("/iterations:%d", family->iterations_); + if (family->repetitions_ != 0) instance.name += StringPrintF("/repeats:%d", family->repetitions_); - } + if (family->use_manual_time_) { instance.name += "/manual_time"; } else if (family->use_real_time_) { @@ -219,6 +221,7 @@ Benchmark::Benchmark(const char* name) time_unit_(kNanosecond), range_multiplier_(kRangeMultiplier), min_time_(0), + iterations_(0), repetitions_(0), use_real_time_(false), use_manual_time_(false), @@ -344,6 +347,22 @@ Benchmark* Benchmark::RangeMultiplier(int multiplier) { return this; } + +Benchmark* Benchmark::MinTime(double t) { + CHECK(t > 0.0); + CHECK(iterations_ == 0); + min_time_ = t; + return this; +} + + +Benchmark* Benchmark::Iterations(size_t n) { + CHECK(n > 0); + CHECK(IsZero(min_time_)); + iterations_ = n; + return this; +} + Benchmark* Benchmark::Repetitions(int n) { CHECK(n > 0); repetitions_ = n; @@ -355,12 +374,6 @@ Benchmark* Benchmark::ReportAggregatesOnly(bool value) { return this; } -Benchmark* Benchmark::MinTime(double t) { - CHECK(t > 0.0); - min_time_ = t; - return this; -} - Benchmark* Benchmark::UseRealTime() { CHECK(!use_manual_time_) << "Cannot set UseRealTime and UseManualTime simultaneously."; diff --git a/utils/google-benchmark/src/complexity.cc b/utils/google-benchmark/src/complexity.cc index dfab791a3293..02adbef62926 100644 --- a/utils/google-benchmark/src/complexity.cc +++ b/utils/google-benchmark/src/complexity.cc @@ -171,6 +171,22 @@ std::vector<BenchmarkReporter::Run> ComputeStats( // All repetitions should be run with the same number of iterations so we // can take this information from the first benchmark. int64_t const run_iterations = reports.front().iterations; + // create stats for user counters + struct CounterStat { + Counter c; + Stat1_d s; + }; + std::map< std::string, CounterStat > counter_stats; + for(Run const& r : reports) { + for(auto const& cnt : r.counters) { + auto it = counter_stats.find(cnt.first); + if(it == counter_stats.end()) { + counter_stats.insert({cnt.first, {cnt.second, Stat1_d{}}}); + } else { + CHECK_EQ(counter_stats[cnt.first].c.flags, cnt.second.flags); + } + } + } // Populate the accumulators. for (Run const& run : reports) { @@ -183,6 +199,12 @@ std::vector<BenchmarkReporter::Run> ComputeStats( Stat1_d(run.cpu_accumulated_time / run.iterations, run.iterations); items_per_second_stat += Stat1_d(run.items_per_second, run.iterations); bytes_per_second_stat += Stat1_d(run.bytes_per_second, run.iterations); + // user counters + for(auto const& cnt : run.counters) { + auto it = counter_stats.find(cnt.first); + CHECK_NE(it, counter_stats.end()); + it->second.s += Stat1_d(cnt.second, run.iterations); + } } // Get the data from the accumulator to BenchmarkReporter::Run's. @@ -196,6 +218,11 @@ std::vector<BenchmarkReporter::Run> ComputeStats( mean_data.bytes_per_second = bytes_per_second_stat.Mean(); mean_data.items_per_second = items_per_second_stat.Mean(); mean_data.time_unit = reports[0].time_unit; + // user counters + for(auto const& kv : counter_stats) { + auto c = Counter(kv.second.s.Mean(), counter_stats[kv.first].c.flags); + mean_data.counters[kv.first] = c; + } // Only add label to mean/stddev if it is same for all runs mean_data.report_label = reports[0].report_label; @@ -215,6 +242,11 @@ std::vector<BenchmarkReporter::Run> ComputeStats( stddev_data.bytes_per_second = bytes_per_second_stat.StdDev(); stddev_data.items_per_second = items_per_second_stat.StdDev(); stddev_data.time_unit = reports[0].time_unit; + // user counters + for(auto const& kv : counter_stats) { + auto c = Counter(kv.second.s.StdDev(), counter_stats[kv.first].c.flags); + stddev_data.counters[kv.first] = c; + } results.push_back(mean_data); results.push_back(stddev_data); @@ -263,6 +295,11 @@ std::vector<BenchmarkReporter::Run> ComputeBigO( big_o.report_big_o = true; big_o.complexity = result_cpu.complexity; + // All the time results are reported after being multiplied by the + // time unit multiplier. But since RMS is a relative quantity it + // should not be multiplied at all. So, here, we _divide_ it by the + // multiplier so that when it is multiplied later the result is the + // correct one. double multiplier = GetTimeUnitMultiplier(reports[0].time_unit); // Only add label to mean/stddev if it is same for all runs @@ -275,6 +312,9 @@ std::vector<BenchmarkReporter::Run> ComputeBigO( rms.cpu_accumulated_time = result_cpu.rms / multiplier; rms.report_rms = true; rms.complexity = result_cpu.complexity; + // don't forget to keep the time unit, or we won't be able to + // recover the correct value. + rms.time_unit = reports[0].time_unit; results.push_back(big_o); results.push_back(rms); diff --git a/utils/google-benchmark/src/console_reporter.cc b/utils/google-benchmark/src/console_reporter.cc index 7e0cca3e3a0b..3f3de02945a3 100644 --- a/utils/google-benchmark/src/console_reporter.cc +++ b/utils/google-benchmark/src/console_reporter.cc @@ -14,6 +14,7 @@ #include "benchmark/reporter.h" #include "complexity.h" +#include "counter.h" #include <algorithm> #include <cstdint> @@ -34,6 +35,7 @@ namespace benchmark { bool ConsoleReporter::ReportContext(const Context& context) { name_field_width_ = context.name_field_width; + printed_header_ = false; PrintBasicContext(&GetErrorStream(), context); @@ -45,16 +47,32 @@ bool ConsoleReporter::ReportContext(const Context& context) { color_output_ = false; } #endif + + return true; +} + +void ConsoleReporter::PrintHeader(const Run& run) { std::string str = FormatString("%-*s %13s %13s %10s\n", static_cast<int>(name_field_width_), "Benchmark", "Time", "CPU", "Iterations"); - GetOutputStream() << str << std::string(str.length() - 1, '-') << "\n"; - - return true; + if(!run.counters.empty()) { + str += " UserCounters..."; + } + std::string line = std::string(str.length(), '-'); + GetOutputStream() << line << "\n" << str << line << "\n"; } void ConsoleReporter::ReportRuns(const std::vector<Run>& reports) { - for (const auto& run : reports) PrintRunData(run); + for (const auto& run : reports) { + // print the header if none was printed yet + if (!printed_header_) { + printed_header_ = true; + PrintHeader(run); + } + // As an alternative to printing the headers like this, we could sort + // the benchmarks by header and then print like that. + PrintRunData(run); + } } static void IgnoreColorPrint(std::ostream& out, LogColor, const char* fmt, @@ -114,6 +132,11 @@ void ConsoleReporter::PrintRunData(const Run& result) { printer(Out, COLOR_CYAN, "%10lld", result.iterations); } + for (auto& c : result.counters) { + auto const& s = HumanReadableNumber(c.second.value); + printer(Out, COLOR_DEFAULT, " %s=%s", c.first.c_str(), s.c_str()); + } + if (!rate.empty()) { printer(Out, COLOR_DEFAULT, " %*s", 13, rate.c_str()); } diff --git a/utils/google-benchmark/src/counter.cc b/utils/google-benchmark/src/counter.cc new file mode 100644 index 000000000000..307863d3c4d6 --- /dev/null +++ b/utils/google-benchmark/src/counter.cc @@ -0,0 +1,68 @@ +// Copyright 2015 Google Inc. All rights reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "counter.h" + +namespace benchmark { +namespace internal { + +double Finish(Counter const& c, double cpu_time, double num_threads) { + double v = c.value; + if (c.flags & Counter::kIsRate) { + v /= cpu_time; + } + if (c.flags & Counter::kAvgThreads) { + v /= num_threads; + } + return v; +} + +void Finish(UserCounters *l, double cpu_time, double num_threads) { + for (auto &c : *l) { + c.second = Finish(c.second, cpu_time, num_threads); + } +} + +void Increment(UserCounters *l, UserCounters const& r) { + // add counters present in both or just in *l + for (auto &c : *l) { + auto it = r.find(c.first); + if (it != r.end()) { + c.second = c.second + it->second; + } + } + // add counters present in r, but not in *l + for (auto const &tc : r) { + auto it = l->find(tc.first); + if (it == l->end()) { + (*l)[tc.first] = tc.second; + } + } +} + +bool SameNames(UserCounters const& l, UserCounters const& r) { + if (&l == &r) return true; + if (l.size() != r.size()) { + return false; + } + for (auto const& c : l) { + if ( r.find(c.first) == r.end()) { + return false; + } + } + return true; +} + +} // end namespace internal +} // end namespace benchmark diff --git a/utils/google-benchmark/src/counter.h b/utils/google-benchmark/src/counter.h new file mode 100644 index 000000000000..bbb92d9a2983 --- /dev/null +++ b/utils/google-benchmark/src/counter.h @@ -0,0 +1,26 @@ +// Copyright 2015 Google Inc. All rights reserved. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include "benchmark/benchmark_api.h" + +namespace benchmark { + +// these counter-related functions are hidden to reduce API surface. +namespace internal { +void Finish(UserCounters *l, double time, double num_threads); +void Increment(UserCounters *l, UserCounters const& r); +bool SameNames(UserCounters const& l, UserCounters const& r); +} // end namespace internal + +} //end namespace benchmark diff --git a/utils/google-benchmark/src/csv_reporter.cc b/utils/google-benchmark/src/csv_reporter.cc index 18ab3b668951..6779815b3c37 100644 --- a/utils/google-benchmark/src/csv_reporter.cc +++ b/utils/google-benchmark/src/csv_reporter.cc @@ -24,6 +24,7 @@ #include "string_util.h" #include "timers.h" +#include "check.h" // File format reference: http://edoceo.com/utilitas/csv-file-format. @@ -38,21 +39,51 @@ std::vector<std::string> elements = { bool CSVReporter::ReportContext(const Context& context) { PrintBasicContext(&GetErrorStream(), context); + return true; +} +void CSVReporter::ReportRuns(const std::vector<Run> & reports) { std::ostream& Out = GetOutputStream(); - for (auto B = elements.begin(); B != elements.end();) { - Out << *B++; - if (B != elements.end()) Out << ","; + + if (!printed_header_) { + // save the names of all the user counters + for (const auto& run : reports) { + for (const auto& cnt : run.counters) { + user_counter_names_.insert(cnt.first); + } + } + + // print the header + for (auto B = elements.begin(); B != elements.end();) { + Out << *B++; + if (B != elements.end()) Out << ","; + } + for (auto B = user_counter_names_.begin(); B != user_counter_names_.end();) { + Out << ",\"" << *B++ << "\""; + } + Out << "\n"; + + printed_header_ = true; + } else { + // check that all the current counters are saved in the name set + for (const auto& run : reports) { + for (const auto& cnt : run.counters) { + CHECK(user_counter_names_.find(cnt.first) != user_counter_names_.end()) + << "All counters must be present in each run. " + << "Counter named \"" << cnt.first + << "\" was not in a run after being added to the header"; + } + } + } + + // print results for each run + for (const auto& run : reports) { + PrintRunData(run); } - Out << "\n"; - return true; -} -void CSVReporter::ReportRuns(const std::vector<Run>& reports) { - for (const auto& run : reports) PrintRunData(run); } -void CSVReporter::PrintRunData(const Run& run) { +void CSVReporter::PrintRunData(const Run & run) { std::ostream& Out = GetOutputStream(); // Field with embedded double-quote characters must be doubled and the field @@ -102,6 +133,13 @@ void CSVReporter::PrintRunData(const Run& run) { Out << "\"" << label << "\""; } Out << ",,"; // for error_occurred and error_message + + // Print user counters + for (const auto &ucn : user_counter_names_) { + auto it = run.counters.find(ucn); + CHECK(it != run.counters.end()); + Out << "," << it->second; + } Out << '\n'; } diff --git a/utils/google-benchmark/src/cycleclock.h b/utils/google-benchmark/src/cycleclock.h index e4825d4ba244..e0f9b01f9d39 100644 --- a/utils/google-benchmark/src/cycleclock.h +++ b/utils/google-benchmark/src/cycleclock.h @@ -43,6 +43,11 @@ extern "C" uint64_t __rdtsc(); #ifndef BENCHMARK_OS_WINDOWS #include <sys/time.h> +#include <time.h> +#endif + +#ifdef BENCHMARK_OS_EMSCRIPTEN +#include <emscripten.h> #endif namespace benchmark { @@ -65,6 +70,10 @@ inline BENCHMARK_ALWAYS_INLINE int64_t Now() { // counter pauses; it does not continue counting, nor does it // reset to zero. return mach_absolute_time(); +#elif defined(BENCHMARK_OS_EMSCRIPTEN) + // this goes above x86-specific code because old versions of Emscripten + // define __x86_64__, although they have nothing to do with it. + return static_cast<int64_t>(emscripten_get_now() * 1e+6); #elif defined(__i386__) int64_t ret; __asm__ volatile("rdtsc" : "=A"(ret)); @@ -79,7 +88,7 @@ inline BENCHMARK_ALWAYS_INLINE int64_t Now() { asm("mftbu %0" : "=r"(tbu0)); asm("mftb %0" : "=r"(tbl)); asm("mftbu %0" : "=r"(tbu1)); - tbl &= -static_cast<int64>(tbu0 == tbu1); + tbl &= -static_cast<int64_t>(tbu0 == tbu1); // high 32 bits in tbu1; low 32 bits in tbl (tbu0 is garbage) return (tbu1 << 32) | tbl; #elif defined(__sparc__) @@ -99,6 +108,22 @@ inline BENCHMARK_ALWAYS_INLINE int64_t Now() { _asm rdtsc #elif defined(COMPILER_MSVC) return __rdtsc(); +#elif defined(BENCHMARK_OS_NACL) + // Native Client validator on x86/x86-64 allows RDTSC instructions, + // and this case is handled above. Native Client validator on ARM + // rejects MRC instructions (used in the ARM-specific sequence below), + // so we handle it here. Portable Native Client compiles to + // architecture-agnostic bytecode, which doesn't provide any + // cycle counter access mnemonics. + + // Native Client does not provide any API to access cycle counter. + // Use clock_gettime(CLOCK_MONOTONIC, ...) instead of gettimeofday + // because is provides nanosecond resolution (which is noticable at + // least for PNaCl modules running on x86 Mac & Linux). + // Initialize to always return 0 if clock_gettime fails. + struct timespec ts = { 0, 0 }; + clock_gettime(CLOCK_MONOTONIC, &ts); + return static_cast<int64_t>(ts.tv_sec) * 1000000000 + ts.tv_nsec; #elif defined(__aarch64__) // System timer of ARMv8 runs at a different frequency than the CPU's. // The frequency is fixed, typically in the range 1-50MHz. It can be @@ -108,7 +133,9 @@ inline BENCHMARK_ALWAYS_INLINE int64_t Now() { asm volatile("mrs %0, cntvct_el0" : "=r"(virtual_timer_value)); return virtual_timer_value; #elif defined(__ARM_ARCH) -#if (__ARM_ARCH >= 6) // V6 is the earliest arch that has a standard cyclecount + // V6 is the earliest arch that has a standard cyclecount + // Native Client validator doesn't allow MRC instructions. +#if (__ARM_ARCH >= 6) uint32_t pmccntr; uint32_t pmuseren; uint32_t pmcntenset; diff --git a/utils/google-benchmark/src/internal_macros.h b/utils/google-benchmark/src/internal_macros.h index e8efcbb39c18..ab9dd85c1023 100644 --- a/utils/google-benchmark/src/internal_macros.h +++ b/utils/google-benchmark/src/internal_macros.h @@ -30,13 +30,26 @@ #elif defined(_WIN32) #define BENCHMARK_OS_WINDOWS 1 #elif defined(__APPLE__) -// TODO(ericwf) This doesn't actually check that it is a Mac OSX system. Just -// that it is an apple system. -#define BENCHMARK_OS_MACOSX 1 +#include "TargetConditionals.h" + #if defined(TARGET_OS_MAC) + #define BENCHMARK_OS_MACOSX 1 + #if defined(TARGET_OS_IPHONE) + #define BENCHMARK_OS_IOS 1 + #endif + #endif #elif defined(__FreeBSD__) #define BENCHMARK_OS_FREEBSD 1 #elif defined(__linux__) #define BENCHMARK_OS_LINUX 1 +#elif defined(__native_client__) +#define BENCHMARK_OS_NACL 1 +#elif defined(EMSCRIPTEN) +#define BENCHMARK_OS_EMSCRIPTEN 1 +#endif + +#if !__has_feature(cxx_exceptions) && !defined(__cpp_exceptions) \ + && !defined(__EXCEPTIONS) +#define BENCHMARK_HAS_NO_EXCEPTIONS #endif #endif // BENCHMARK_INTERNAL_MACROS_H_ diff --git a/utils/google-benchmark/src/json_reporter.cc b/utils/google-benchmark/src/json_reporter.cc index cea5f9bfa1c9..5a653088e5bf 100644 --- a/utils/google-benchmark/src/json_reporter.cc +++ b/utils/google-benchmark/src/json_reporter.cc @@ -154,10 +154,15 @@ void JSONReporter::PrintRunData(Run const& run) { << indent << FormatKV("items_per_second", RoundDouble(run.items_per_second)); } + for(auto &c : run.counters) { + out << ",\n" + << indent + << FormatKV(c.first, RoundDouble(c.second)); + } if (!run.report_label.empty()) { out << ",\n" << indent << FormatKV("label", run.report_label); } out << '\n'; } -} // end namespace benchmark +} // end namespace benchmark diff --git a/utils/google-benchmark/src/re.h b/utils/google-benchmark/src/re.h index af4a498c9f7a..01e9736505e5 100644 --- a/utils/google-benchmark/src/re.h +++ b/utils/google-benchmark/src/re.h @@ -15,6 +15,15 @@ #ifndef BENCHMARK_RE_H_ #define BENCHMARK_RE_H_ +#include "internal_macros.h" + +// Prefer C regex libraries when compiling w/o exceptions so that we can +// correctly report errors. +#if defined(BENCHMARK_HAS_NO_EXCEPTIONS) && defined(HAVE_STD_REGEX) && \ + (defined(HAVE_GNU_POSIX_REGEX) || defined(HAVE_POSIX_REGEX)) +#undef HAVE_STD_REGEX +#endif + #if defined(HAVE_STD_REGEX) #include <regex> #elif defined(HAVE_GNU_POSIX_REGEX) @@ -62,15 +71,20 @@ class Regex { #if defined(HAVE_STD_REGEX) inline bool Regex::Init(const std::string& spec, std::string* error) { +#ifdef BENCHMARK_HAS_NO_EXCEPTIONS + ((void)error); // suppress unused warning +#else try { +#endif re_ = std::regex(spec, std::regex_constants::extended); - init_ = true; +#ifndef BENCHMARK_HAS_NO_EXCEPTIONS } catch (const std::regex_error& e) { if (error) { *error = e.what(); } } +#endif return init_; } diff --git a/utils/google-benchmark/src/sleep.cc b/utils/google-benchmark/src/sleep.cc index 918abc485dc3..54aa04a42247 100644 --- a/utils/google-benchmark/src/sleep.cc +++ b/utils/google-benchmark/src/sleep.cc @@ -15,6 +15,7 @@ #include "sleep.h" #include <cerrno> +#include <cstdlib> #include <ctime> #include "internal_macros.h" @@ -40,7 +41,7 @@ void SleepForMicroseconds(int microseconds) { } void SleepForMilliseconds(int milliseconds) { - SleepForMicroseconds(static_cast<int>(milliseconds) * kNumMicrosPerMilli); + SleepForMicroseconds(milliseconds * kNumMicrosPerMilli); } void SleepForSeconds(double seconds) { diff --git a/utils/google-benchmark/src/sleep.h b/utils/google-benchmark/src/sleep.h index f1e515ca4f3a..f98551afe284 100644 --- a/utils/google-benchmark/src/sleep.h +++ b/utils/google-benchmark/src/sleep.h @@ -1,14 +1,12 @@ #ifndef BENCHMARK_SLEEP_H_ #define BENCHMARK_SLEEP_H_ -#include <cstdint> - namespace benchmark { -const int64_t kNumMillisPerSecond = 1000LL; -const int64_t kNumMicrosPerMilli = 1000LL; -const int64_t kNumMicrosPerSecond = kNumMillisPerSecond * 1000LL; -const int64_t kNumNanosPerMicro = 1000LL; -const int64_t kNumNanosPerSecond = kNumNanosPerMicro * kNumMicrosPerSecond; +const int kNumMillisPerSecond = 1000; +const int kNumMicrosPerMilli = 1000; +const int kNumMicrosPerSecond = kNumMillisPerSecond * 1000; +const int kNumNanosPerMicro = 1000; +const int kNumNanosPerSecond = kNumNanosPerMicro * kNumMicrosPerSecond; void SleepForMilliseconds(int milliseconds); void SleepForSeconds(double seconds); diff --git a/utils/google-benchmark/src/string_util.cc b/utils/google-benchmark/src/string_util.cc index 4cefbfba6c27..cd4e7cfde57c 100644 --- a/utils/google-benchmark/src/string_util.cc +++ b/utils/google-benchmark/src/string_util.cc @@ -45,6 +45,8 @@ void ToExponentAndMantissa(double val, double thresh, int precision, std::max(thresh, 1.0 / std::pow(10.0, precision)); const double big_threshold = adjusted_threshold * one_k; const double small_threshold = adjusted_threshold; + // Values in ]simple_threshold,small_threshold[ will be printed as-is + const double simple_threshold = 0.01; if (val > big_threshold) { // Positive powers @@ -62,14 +64,16 @@ void ToExponentAndMantissa(double val, double thresh, int precision, *exponent = 0; } else if (val < small_threshold) { // Negative powers - double scaled = val; - for (size_t i = 0; i < arraysize(kSmallSIUnits); ++i) { - scaled *= one_k; - if (scaled >= small_threshold) { - mantissa_stream << scaled; - *exponent = -static_cast<int64_t>(i + 1); - *mantissa = mantissa_stream.str(); - return; + if (val < simple_threshold) { + double scaled = val; + for (size_t i = 0; i < arraysize(kSmallSIUnits); ++i) { + scaled *= one_k; + if (scaled >= small_threshold) { + mantissa_stream << scaled; + *exponent = -static_cast<int64_t>(i + 1); + *mantissa = mantissa_stream.str(); + return; + } } } mantissa_stream << val; diff --git a/utils/google-benchmark/src/sysinfo.cc b/utils/google-benchmark/src/sysinfo.cc index dd1e66306e18..7feb79e65f20 100644 --- a/utils/google-benchmark/src/sysinfo.cc +++ b/utils/google-benchmark/src/sysinfo.cc @@ -75,7 +75,9 @@ bool ReadIntFromFile(const char* file, long* value) { char line[1024]; char* err; memset(line, '\0', sizeof(line)); - CHECK(read(fd, line, sizeof(line) - 1)); + ssize_t read_err = read(fd, line, sizeof(line) - 1); + ((void)read_err); // prevent unused warning + CHECK(read_err >= 0); const long temp_value = strtol(line, &err, 10); if (line[0] != '\0' && (*err == '\n' || *err == '\0')) { *value = temp_value; @@ -295,8 +297,13 @@ void InitializeSystemInfo() { (size == sizeof(cpu_freq))) { cpuinfo_cycles_per_second = cpu_freq; } else { + #if defined BENCHMARK_OS_IOS + fprintf(stderr, "CPU frequency cannot be detected. \n"); + cpuinfo_cycles_per_second = 0; + #else fprintf(stderr, "%s\n", strerror(errno)); std::exit(EXIT_FAILURE); + #endif } #else // Generic cycles per second counter diff --git a/utils/google-benchmark/src/timers.cc b/utils/google-benchmark/src/timers.cc index fadc08f791e7..8d56e8adf8d4 100644 --- a/utils/google-benchmark/src/timers.cc +++ b/utils/google-benchmark/src/timers.cc @@ -35,6 +35,10 @@ #endif #endif +#ifdef BENCHMARK_OS_EMSCRIPTEN +#include <emscripten.h> +#endif + #include <cerrno> #include <cstdint> #include <cstdio> @@ -100,14 +104,7 @@ BENCHMARK_NORETURN static void DiagnoseAndExit(const char* msg) { } // end namespace double ProcessCPUUsage() { -// FIXME We want to use clock_gettime, but its not available in MacOS 10.11. See -// https://github.com/google/benchmark/pull/292 -#if defined(CLOCK_PROCESS_CPUTIME_ID) && !defined(BENCHMARK_OS_MACOSX) - struct timespec spec; - if (clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &spec) == 0) - return MakeTime(spec); - DiagnoseAndExit("clock_gettime(CLOCK_PROCESS_CPUTIME_ID, ...) failed"); -#elif defined(BENCHMARK_OS_WINDOWS) +#if defined(BENCHMARK_OS_WINDOWS) HANDLE proc = GetCurrentProcess(); FILETIME creation_time; FILETIME exit_time; @@ -117,21 +114,28 @@ double ProcessCPUUsage() { &user_time)) return MakeTime(kernel_time, user_time); DiagnoseAndExit("GetProccessTimes() failed"); +#elif defined(BENCHMARK_OS_EMSCRIPTEN) + // clock_gettime(CLOCK_PROCESS_CPUTIME_ID, ...) returns 0 on Emscripten. + // Use Emscripten-specific API. Reported CPU time would be exactly the + // same as total time, but this is ok because there aren't long-latency + // syncronous system calls in Emscripten. + return emscripten_get_now() * 1e-3; +#elif defined(CLOCK_PROCESS_CPUTIME_ID) && !defined(BENCHMARK_OS_MACOSX) + // FIXME We want to use clock_gettime, but its not available in MacOS 10.11. See + // https://github.com/google/benchmark/pull/292 + struct timespec spec; + if (clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &spec) == 0) + return MakeTime(spec); + DiagnoseAndExit("clock_gettime(CLOCK_PROCESS_CPUTIME_ID, ...) failed"); #else struct rusage ru; if (getrusage(RUSAGE_SELF, &ru) == 0) return MakeTime(ru); - DiagnoseAndExit("clock_gettime(CLOCK_PROCESS_CPUTIME_ID, ...) failed"); + DiagnoseAndExit("getrusage(RUSAGE_SELF, ...) failed"); #endif } double ThreadCPUUsage() { -// FIXME We want to use clock_gettime, but its not available in MacOS 10.11. See -// https://github.com/google/benchmark/pull/292 -#if defined(CLOCK_THREAD_CPUTIME_ID) && !defined(BENCHMARK_OS_MACOSX) - struct timespec ts; - if (clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts) == 0) return MakeTime(ts); - DiagnoseAndExit("clock_gettime(CLOCK_THREAD_CPUTIME_ID, ...) failed"); -#elif defined(BENCHMARK_OS_WINDOWS) +#if defined(BENCHMARK_OS_WINDOWS) HANDLE this_thread = GetCurrentThread(); FILETIME creation_time; FILETIME exit_time; @@ -141,6 +145,8 @@ double ThreadCPUUsage() { &user_time); return MakeTime(kernel_time, user_time); #elif defined(BENCHMARK_OS_MACOSX) + // FIXME We want to use clock_gettime, but its not available in MacOS 10.11. See + // https://github.com/google/benchmark/pull/292 mach_msg_type_number_t count = THREAD_BASIC_INFO_COUNT; thread_basic_info_data_t info; mach_port_t thread = pthread_mach_thread_np(pthread_self()); @@ -149,6 +155,13 @@ double ThreadCPUUsage() { return MakeTime(info); } DiagnoseAndExit("ThreadCPUUsage() failed when evaluating thread_info"); +#elif defined(BENCHMARK_OS_EMSCRIPTEN) + // Emscripten doesn't support traditional threads + return ProcessCPUUsage(); +#elif defined(CLOCK_THREAD_CPUTIME_ID) + struct timespec ts; + if (clock_gettime(CLOCK_THREAD_CPUTIME_ID, &ts) == 0) return MakeTime(ts); + DiagnoseAndExit("clock_gettime(CLOCK_THREAD_CPUTIME_ID, ...) failed"); #else #error Per-thread timing is not available on your system. #endif |