Merged with upstream

.gitignore

@@ -4,3 +4,4 @@ Makefile
 bin/
 cmake_install.cmake
 lib/
+tags

README.md

 benchmark
 =========
+[![Build Status](https://drone.io/github.com/google/benchmark/status.png)](https://drone.io/github.com/google/benchmark/latest)
+
 A library to support the benchmarking of functions, similar to unit-tests.
 
+Discussion group: https://groups.google.com/d/forum/benchmark-discuss
+
 Example usage:
 Define a function that executes the code to be measured a
 specified number of times:

include/benchmark/benchmark.h

@@ -145,24 +145,27 @@ BENCHMARK(BM_MultiThreaded)->Threads(4);
 #include "macros.h"
 
 namespace benchmark {
-// If the --benchmarks flag is empty, do nothing.
-//
-// Otherwise, run all benchmarks specified by the --benchmarks flag,
+class BenchmarkReporter;
+
+void Initialize(int* argc, const char** argv);
+
+// Otherwise, run all benchmarks specified by the --benchmark_filter flag,
 // and exit after running the benchmarks.
-extern void RunSpecifiedBenchmarks();
+void RunSpecifiedBenchmarks(const BenchmarkReporter* reporter = nullptr);
 
 // ------------------------------------------------------
 // Routines that can be called from within a benchmark
 
 //
 // REQUIRES: a benchmark is currently executing
-extern void SetLabel(const std::string& label);
+void SetLabel(const std::string& label);
 
 // If this routine is called, peak memory allocation past this point in the
 // benchmark is reported at the end of the benchmark report line. (It is
 // computed by running the benchmark once with a single iteration and a memory
 // tracer.)
-extern void MemoryUsage();
+// TODO(dominic)
+//void MemoryUsage();
 
 // If a particular benchmark is I/O bound, or if for some reason CPU
 // timings are not representative, call this method from within the
@@ -170,7 +173,7 @@ extern void MemoryUsage();
 // control how many iterations are run, and in the printing of
 // items/second or MB/seconds values.  If not called, the cpu time
 // used by the benchmark will be used.
-extern void UseRealTime();
+void UseRealTime();
 
 namespace internal {
 class Benchmark;
@@ -290,15 +293,73 @@ class State {
   DISALLOW_COPY_AND_ASSIGN(State);
 };
 
+// Interface for custom benchmark result printers.
+// By default, benchmark reports are printed to stdout. However an application
+// can control the destination of the reports by calling
+// RunSpecifiedBenchmarks and passing it a custom reporter object.
+// The reporter object must implement the following interface.
+class BenchmarkReporter {
+ public:
+  struct Context {
+    int num_cpus;
+    double mhz_per_cpu;
+    //std::string cpu_info;
+    bool cpu_scaling_enabled;
+
+    // The number of chars in the longest benchmark name.
+    int name_field_width;
+  };
+
+  struct Run {
+    Run() :
+        thread_index(-1),
+        iterations(1),
+        real_accumulated_time(0),
+        cpu_accumulated_time(0),
+        bytes_per_second(0),
+        items_per_second(0),
+        max_heapbytes_used(0) {}
+
+    std::string benchmark_name;
+    std::string report_label;
+    int thread_index;
+    int64_t iterations;
+    double real_accumulated_time;
+    double cpu_accumulated_time;
+
+    // Zero if not set by benchmark.
+    double bytes_per_second;
+    double items_per_second;
+
+    // This is set to 0.0 if memory tracing is not enabled.
+    double max_heapbytes_used;
+  };
+
+  // Called once for every suite of benchmarks run.
+  // The parameter "context" contains information that the
+  // reporter may wish to use when generating its report, for example the
+  // platform under which the benchmarks are running. The benchmark run is
+  // never started if this function returns false, allowing the reporter
+  // to skip runs based on the context information.
+  virtual bool ReportContext(const Context& context) const = 0;
+
+  // Called once for each group of benchmark runs, gives information about
+  // cpu-time and heap memory usage during the benchmark run.
+  // Note that all the grouped benchmark runs should refer to the same
+  // benchmark, thus have the same name.
+  virtual void ReportRuns(const std::vector<Run>& report) const = 0;
+
+  virtual ~BenchmarkReporter() {}
+};
+
 namespace internal {
-class BenchmarkReporter;
 
 typedef std::function<void(State&)> BenchmarkFunction;
 
 // Run all benchmarks whose name is a partial match for the regular
 // expression in "spec". The results of benchmark runs are fed to "reporter".
 void RunMatchingBenchmarks(const std::string& spec,
-                           BenchmarkReporter* reporter);
+                           const BenchmarkReporter* reporter);
 
 // Extract the list of benchmark names that match the specified regular
 // expression.
@@ -376,8 +437,6 @@ class Benchmark {
   // Equivalent to ThreadRange(NumCPUs(), NumCPUs())
   Benchmark* ThreadPerCpu();
 
-  // TODO(dominic): Control whether or not real-time is used for this benchmark
-
   // -------------------------------
   // Following methods are not useful for clients
 
@@ -411,100 +470,31 @@ class Benchmark {
 
   static void AddRange(std::vector<int>* dst, int lo, int hi, int mult);
   static double MeasurePeakHeapMemory(const Instance& b);
-  static void RunInstance(const Instance& b, BenchmarkReporter* br);
+  static void RunInstance(const Instance& b, const BenchmarkReporter* br);
   friend class ::benchmark::State;
   friend struct ::benchmark::internal::Benchmark::Instance;
-  friend void ::benchmark::internal::RunMatchingBenchmarks(const std::string&,
-                                                           BenchmarkReporter*);
+  friend void ::benchmark::internal::RunMatchingBenchmarks(
+      const std::string&, const BenchmarkReporter*);
   DISALLOW_COPY_AND_ASSIGN(Benchmark);
 };
 
 // ------------------------------------------------------
-// Benchmarks reporter interface + data containers.
-
-struct BenchmarkContextData {
-  int num_cpus;
-  double mhz_per_cpu;
-  // std::string cpu_info;
-  bool cpu_scaling_enabled;
-
-  // The number of chars in the longest benchmark name.
-  int name_field_width;
-};
-
-struct BenchmarkRunData {
-  BenchmarkRunData()
-      : thread_index(-1),
-        iterations(1),
-        real_accumulated_time(0),
-        cpu_accumulated_time(0),
-        bytes_per_second(0),
-        items_per_second(0),
-        max_heapbytes_used(0) {}
-
-  std::string benchmark_name;
-  std::string report_label;
-  int thread_index;
-  int64_t iterations;
-  double real_accumulated_time;
-  double cpu_accumulated_time;
-
-  // Zero if not set by benchmark.
-  double bytes_per_second;
-  double items_per_second;
-
-  // This is set to 0.0 if memory tracing is not enabled.
-  double max_heapbytes_used;
-};
-
-// Interface for custom benchmark result printers.
-// By default, benchmark reports are printed to stdout. However an application
-// can control the destination of the reports by calling
-// RunMatchingBenchmarks and passing it a custom reporter object.
-// The reporter object must implement the following interface.
-class BenchmarkReporter {
- public:
-  // Called once for every suite of benchmarks run.
-  // The parameter "context" contains information that the
-  // reporter may wish to use when generating its report, for example the
-  // platform under which the benchmarks are running. The benchmark run is
-  // never started if this function returns false, allowing the reporter
-  // to skip runs based on the context information.
-  virtual bool ReportContext(const BenchmarkContextData& context) = 0;
-
-  // Called once for each group of benchmark runs, gives information about
-  // cpu-time and heap memory usage during the benchmark run.
-  // Note that all the grouped benchmark runs should refer to the same
-  // benchmark, thus have the same name.
-  virtual void ReportRuns(const std::vector<BenchmarkRunData>& report) = 0;
-
-  virtual ~BenchmarkReporter();
-};
-
-// ------------------------------------------------------
 // Internal implementation details follow; please ignore
 
-// Given a collection of reports, computes their mean and stddev.
-// REQUIRES: all runs in "reports" must be from the same benchmark.
-void ComputeStats(const std::vector<BenchmarkRunData>& reports,
-                  BenchmarkRunData* mean_data, BenchmarkRunData* stddev_data);
-
 // Simple reporter that outputs benchmark data to the console. This is the
 // default reporter used by RunSpecifiedBenchmarks().
 class ConsoleReporter : public BenchmarkReporter {
  public:
-  virtual bool ReportContext(const BenchmarkContextData& context);
-  virtual void ReportRuns(const std::vector<BenchmarkRunData>& reports);
+  virtual bool ReportContext(const Context& context) const;
+  virtual void ReportRuns(const std::vector<Run>& reports) const;
 
  private:
-  std::string PrintMemoryUsage(double bytes);
-  virtual void PrintRunData(const BenchmarkRunData& report);
-  int name_field_width_;
+  std::string PrintMemoryUsage(double bytes) const;
+  virtual void PrintRunData(const Run& report) const;
+  mutable int name_field_width_;
 };
 
 }  // end namespace internal
-
-void Initialize(int* argc, const char** argv);
 }  // end namespace benchmark
 
 // ------------------------------------------------------

src/benchmark.cc

+// Copyright 2014 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.h"
 #include "benchmark/macros.h"
 #include "colorprint.h"
@@ -224,14 +238,11 @@ bool CpuScalingEnabled() {
   return false;
 }
 
-}  // namespace
-
-namespace internal {
-
-BenchmarkReporter::~BenchmarkReporter() {}
-
-void ComputeStats(const std::vector<BenchmarkRunData>& reports,
-                  BenchmarkRunData* mean_data, BenchmarkRunData* stddev_data) {
+// Given a collection of reports, computes their mean and stddev.
+// REQUIRES: all runs in "reports" must be from the same benchmark.
+void ComputeStats(const std::vector<BenchmarkReporter::Run>& reports,
+                  BenchmarkReporter::Run* mean_data,
+                  BenchmarkReporter::Run* stddev_data) {
   // Accumulators.
   Stat1_d real_accumulated_time_stat;
   Stat1_d cpu_accumulated_time_stat;
@@ -241,7 +252,7 @@ void ComputeStats(const std::vector<BenchmarkRunData>& reports,
   Stat1MinMax_d max_heapbytes_used_stat;
 
   // Populate the accumulators.
-  for (std::vector<BenchmarkRunData>::const_iterator it = reports.begin();
+  for (std::vector<BenchmarkReporter::Run>::const_iterator it = reports.begin();
        it != reports.end(); ++it) {
     CHECK_EQ(reports[0].benchmark_name, it->benchmark_name);
     real_accumulated_time_stat +=
@@ -282,16 +293,21 @@ void ComputeStats(const std::vector<BenchmarkRunData>& reports,
   stddev_data->items_per_second = items_per_second_stat.StdDev();
   stddev_data->max_heapbytes_used = max_heapbytes_used_stat.StdDev();
 }
+}  // namespace
 
-std::string ConsoleReporter::PrintMemoryUsage(double bytes) {
-  if (!get_memory_usage || bytes < 0.0) return "";
+namespace internal {
+
+std::string ConsoleReporter::PrintMemoryUsage(double bytes) const {
+  if (!get_memory_usage || bytes < 0.0)
+    return "";
 
   std::stringstream ss;
   ss << " " << HumanReadableNumber(bytes) << "B peak-mem";
   return ss.str();
 }
 
-bool ConsoleReporter::ReportContext(const BenchmarkContextData& context) {
+bool ConsoleReporter::ReportContext(
+    const BenchmarkReporter::Context& context) const {
   name_field_width_ = context.name_field_width;
 
   std::cout << "Benchmarking on " << context.num_cpus << " X "
@@ -319,8 +335,9 @@ bool ConsoleReporter::ReportContext(const BenchmarkContextData& context) {
   return true;
 }
 
-void ConsoleReporter::ReportRuns(const std::vector<BenchmarkRunData>& reports) {
-  for (std::vector<BenchmarkRunData>::const_iterator it = reports.begin();
+void ConsoleReporter::ReportRuns(
+    const std::vector<BenchmarkReporter::Run>& reports) const {
+  for (std::vector<BenchmarkReporter::Run>::const_iterator it = reports.begin();
        it != reports.end(); ++it) {
     CHECK_EQ(reports[0].benchmark_name, it->benchmark_name);
     PrintRunData(*it);
@@ -329,15 +346,15 @@ void ConsoleReporter::ReportRuns(const std::vector<BenchmarkRunData>& reports) {
   // We don't report aggregated data if there was a single run.
   if (reports.size() < 2) return;
 
-  BenchmarkRunData mean_data;
-  BenchmarkRunData stddev_data;
-  internal::ComputeStats(reports, &mean_data, &stddev_data);
+  BenchmarkReporter::Run mean_data;
+  BenchmarkReporter::Run stddev_data;
+  ComputeStats(reports, &mean_data, &stddev_data);
 
   PrintRunData(mean_data);
   PrintRunData(stddev_data);
 }
 
-void ConsoleReporter::PrintRunData(const BenchmarkRunData& result) {
+void ConsoleReporter::PrintRunData(const BenchmarkReporter::Run& result) const {
   // Format bytes per second
   std::string rate;
   if (result.bytes_per_second > 0) {
@@ -363,13 +380,14 @@ void ConsoleReporter::PrintRunData(const BenchmarkRunData& result) {
               (result.cpu_accumulated_time * 1e9) /
                   (static_cast<double>(result.iterations)));
   ColorPrintf(COLOR_CYAN, "%10lld", result.iterations);
-  ColorPrintf(COLOR_DEFAULT, "%*s %*s %s%s\n",
-              16, rate.c_str(),
+  ColorPrintf(COLOR_DEFAULT, "%*s %*s %s %s\n",
+              13, rate.c_str(),
               18, items.c_str(),
               result.report_label.c_str(),
               PrintMemoryUsage(result.max_heapbytes_used).c_str());
 }
 
+/* TODO(dominic)
 void MemoryUsage() {
   // if (benchmark_mc) {
   //  benchmark_mc->Reset();
@@ -377,6 +395,7 @@ void MemoryUsage() {
   get_memory_usage = true;
   //}
 }
+*/
 
 void UseRealTime() { use_real_time = true; }
 
@@ -542,7 +561,7 @@ struct State::SharedState {
   int stopping;  // Number of threads that have entered STOPPING state
   int threads;   // Number of total threads that are running concurrently
   ThreadStats stats;
-  std::vector<internal::BenchmarkRunData> runs;  // accumulated runs
+  std::vector<BenchmarkReporter::Run> runs;  // accumulated runs
   std::string label;
 
   explicit SharedState(const internal::Benchmark::Instance* b)
@@ -766,7 +785,7 @@ void Benchmark::MeasureOverhead() {
 #endif
 }
 
-void Benchmark::RunInstance(const Instance& b, BenchmarkReporter* br) {
+void Benchmark::RunInstance(const Instance& b, const BenchmarkReporter* br) {
   use_real_time = false;
   running_benchmark = true;
   // get_memory_usage = FLAGS_gbenchmark_memory_usage;
@@ -810,7 +829,7 @@ void Benchmark::RunInstance(const Instance& b, BenchmarkReporter* br) {
   */
   running_benchmark = false;
 
-  for (internal::BenchmarkRunData& report : state.runs) {
+  for (BenchmarkReporter::Run& report : state.runs) {
     double seconds = (use_real_time ? report.real_accumulated_time
                                     : report.cpu_accumulated_time);
     report.benchmark_name = b.name;
@@ -1006,8 +1025,7 @@ bool State::FinishInterval() {
     return true;
   }
 
-  internal::BenchmarkRunData data;
-  data.thread_index = thread_index;
+  BenchmarkReporter::Run data;
   data.iterations = iterations_;
   data.thread_index = thread_index;
 
@@ -1094,8 +1112,7 @@ void* State::RunWrapper(void* arg) {
 namespace internal {
 
 void RunMatchingBenchmarks(const std::string& spec,
-                           BenchmarkReporter* reporter) {
-  CHECK(reporter != NULL);
+                           const BenchmarkReporter* reporter) {
   if (spec.empty()) return;
 
   std::vector<internal::Benchmark::Instance> benchmarks;
@@ -1121,7 +1138,7 @@ void RunMatchingBenchmarks(const std::string& spec,
   }
 
   // Print header here
-  BenchmarkContextData context;
+  BenchmarkReporter::Context context;
   context.num_cpus = NumCPUs();
   context.mhz_per_cpu = CyclesPerSecond() / 1000000.0f;
   //  context.cpu_info = base::CompactCPUIDInfoString();
@@ -1145,12 +1162,12 @@ void FindMatchingBenchmarkNames(const std::string& spec,
 
 }  // end namespace internal
 
-void RunSpecifiedBenchmarks() {
+void RunSpecifiedBenchmarks(const BenchmarkReporter* reporter /*= nullptr*/) {
   std::string spec = FLAGS_benchmark_filter;
   if (spec.empty() || spec == "all")
     spec = ".";  // Regexp that matches all benchmarks
   internal::ConsoleReporter default_reporter;
-  internal::RunMatchingBenchmarks(spec, &default_reporter);
+  internal::RunMatchingBenchmarks(spec, reporter == nullptr ? &default_reporter : reporter);
   pthread_cond_destroy(&starting_cv);
   pthread_mutex_destroy(&starting_mutex);
   pthread_mutex_destroy(&benchmark_mutex);

src/colorprint.cc

+// Copyright 2014 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 "colorprint.h"
 
 #include <stdarg.h>

src/commandlineflags.cc

+// Copyright 2014 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 "commandlineflags.h"
 
 #include <string.h>

src/sleep.cc

+// Copyright 2014 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 "sleep.h"
 
 #include <time.h>

src/sysinfo.cc

+// Copyright 2014 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 "sysinfo.h"
 
 #include <errno.h>
@@ -20,21 +34,18 @@
 
 namespace benchmark {
 namespace {
+const int64_t estimate_time_ms = 1000;
 pthread_once_t cpuinfo_init = PTHREAD_ONCE_INIT;
 double cpuinfo_cycles_per_second = 1.0;
 int cpuinfo_num_cpus = 1;  // Conservative guess
-static pthread_mutex_t cputimens_mutex;
+pthread_mutex_t cputimens_mutex;
 
 // Helper function estimates cycles/sec by observing cycles elapsed during
 // sleep(). Using small sleep time decreases accuracy significantly.
-int64_t EstimateCyclesPerSecond(const int estimate_time_ms) {
-  CHECK(estimate_time_ms > 0);
-  double multiplier = 1000.0 / (double)estimate_time_ms;  // scale by this much
-
+int64_t EstimateCyclesPerSecond() {
   const int64_t start_ticks = cycleclock::Now();
   SleepForMilliseconds(estimate_time_ms);
-  const int64_t guess = int64_t(multiplier * (cycleclock::Now() - start_ticks));
-  return guess;
+  return cycleclock::Now() - start_ticks;
 }
 
 // Helper function for reading an int from a file. Returns true if successful
@@ -99,9 +110,9 @@ void InitializeSystemInfo() {
   if (fd == -1) {
     perror(pname);
     if (!saw_mhz) {
-      cpuinfo_cycles_per_second = EstimateCyclesPerSecond(1000);
+      cpuinfo_cycles_per_second = EstimateCyclesPerSecond();
     }
-    return;  // TODO: use generic tester instead?
+    return;
   }
 
   double bogo_clock = 1.0;
@@ -165,7 +176,7 @@ void InitializeSystemInfo() {
       cpuinfo_cycles_per_second = bogo_clock;
     } else {
       // If we don't even have bogomips, we'll use the slow estimation.
-      cpuinfo_cycles_per_second = EstimateCyclesPerSecond(1000);
+      cpuinfo_cycles_per_second = EstimateCyclesPerSecond();
     }
   }
   if (num_cpus == 0) {
@@ -201,7 +212,7 @@ void InitializeSystemInfo() {
   if (sysctlbyname(sysctl_path, &hz, &sz, NULL, 0) != 0) {
     fprintf(stderr, "Unable to determine clock rate from sysctl: %s: %s\n",
             sysctl_path, strerror(errno));
-    cpuinfo_cycles_per_second = EstimateCyclesPerSecond(1000);
+    cpuinfo_cycles_per_second = EstimateCyclesPerSecond();
   } else {
     cpuinfo_cycles_per_second = hz;
   }
@@ -221,8 +232,8 @@ void InitializeSystemInfo() {
                       "~MHz", NULL, &data, &data_size)))
     cpuinfo_cycles_per_second = (int64)data * (int64)(1000 * 1000);  // was mhz
   else
-    cpuinfo_cycles_per_second = EstimateCyclesPerSecond(500);  // TODO <500?
-                                                               // TODO: also figure out cpuinfo_num_cpus
+    cpuinfo_cycles_per_second = EstimateCyclesPerSecond();
+  // TODO: also figure out cpuinfo_num_cpus
 
 #elif defined OS_MACOSX
   // returning "mach time units" per second. the current number of elapsed
@@ -250,7 +261,7 @@ void InitializeSystemInfo() {
 
 #else
   // Generic cycles per second counter
-  cpuinfo_cycles_per_second = EstimateCyclesPerSecond(1000);
+  cpuinfo_cycles_per_second = EstimateCyclesPerSecond();
 #endif
 }
 }  // end namespace

src/walltime.cc

+// Copyright 2014 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 "walltime.h"
 
 #include <stdio.h>