Skip to content

Latest commit

 

History

History
1312 lines (1038 loc) · 44.3 KB

user_guide.md

File metadata and controls

1312 lines (1038 loc) · 44.3 KB

User Guide

Command Line

Output Formats

Output Files

Running Benchmarks

Running a Subset of Benchmarks

Result Comparison

Extra Context

Library

Runtime and Reporting Considerations

Setup/Teardown

Passing Arguments

Custom Benchmark Name

Calculating Asymptotic Complexity

Templated Benchmarks

Templated Benchmarks that take arguments

Fixtures

Custom Counters

Multithreaded Benchmarks

CPU Timers

Manual Timing

Setting the Time Unit

Random Interleaving

User-Requested Performance Counters

Preventing Optimization

Reporting Statistics

Custom Statistics

Memory Usage

Using RegisterBenchmark

Exiting with an Error

A Faster KeepRunning Loop

Benchmarking Tips

Disabling CPU Frequency Scaling

Reducing Variance in Benchmarks

The library supports multiple output formats. Use the --benchmark_format=<console|json|csv> flag (or set the BENCHMARK_FORMAT=<console|json|csv> environment variable) to set the format type. console is the default format.

The Console format is intended to be a human readable format. By default the format generates color output. Context is output on stderr and the tabular data on stdout. Example tabular output looks like:

Benchmark                               Time(ns)    CPU(ns) Iterations
----------------------------------------------------------------------
BM_SetInsert/1024/1                        28928      29349      23853  133.097kB/s   33.2742k items/s
BM_SetInsert/1024/8                        32065      32913      21375  949.487kB/s   237.372k items/s
BM_SetInsert/1024/10                       33157      33648      21431  1.13369MB/s   290.225k items/s

The JSON format outputs human readable json split into two top level attributes. The context attribute contains information about the run in general, including information about the CPU and the date. The benchmarks attribute contains a list of every benchmark run. Example json output looks like:

{
  "context": {
    "date": "2015/03/17-18:40:25",
    "num_cpus": 40,
    "mhz_per_cpu": 2801,
    "cpu_scaling_enabled": false,
    "build_type": "debug"
  },
  "benchmarks": [
    {
      "name": "BM_SetInsert/1024/1",
      "iterations": 94877,
      "real_time": 29275,
      "cpu_time": 29836,
      "bytes_per_second": 134066,
      "items_per_second": 33516
    },
    {
      "name": "BM_SetInsert/1024/8",
      "iterations": 21609,
      "real_time": 32317,
      "cpu_time": 32429,
      "bytes_per_second": 986770,
      "items_per_second": 246693
    },
    {
      "name": "BM_SetInsert/1024/10",
      "iterations": 21393,
      "real_time": 32724,
      "cpu_time": 33355,
      "bytes_per_second": 1199226,
      "items_per_second": 299807
    }
  ]
}

The CSV format outputs comma-separated values. The context is output on stderr and the CSV itself on stdout. Example CSV output looks like:

name,iterations,real_time,cpu_time,bytes_per_second,items_per_second,label
"BM_SetInsert/1024/1",65465,17890.7,8407.45,475768,118942,
"BM_SetInsert/1024/8",116606,18810.1,9766.64,3.27646e+06,819115,
"BM_SetInsert/1024/10",106365,17238.4,8421.53,4.74973e+06,1.18743e+06,

Write benchmark results to a file with the --benchmark_out=<filename> option (or set BENCHMARK_OUT). Specify the output format with --benchmark_out_format={json|console|csv} (or set BENCHMARK_OUT_FORMAT={json|console|csv}). Note that the 'csv' reporter is deprecated and the saved .csv file is not parsable by csv parsers.

Specifying --benchmark_out does not suppress the console output.

Benchmarks are executed by running the produced binaries. Benchmarks binaries, by default, accept options that may be specified either through their command line interface or by setting environment variables before execution. For every --option_flag=<value> CLI switch, a corresponding environment variable OPTION_FLAG=<value> exist and is used as default if set (CLI switches always prevails). A complete list of CLI options is available running benchmarks with the --help switch.

The --benchmark_filter=<regex> option (or BENCHMARK_FILTER=<regex> environment variable) can be used to only run the benchmarks that match the specified <regex>. For example:

$ ./run_benchmarks.x --benchmark_filter=BM_memcpy/32
Run on (1 X 2300 MHz CPU )
2016-06-25 19:34:24
Benchmark              Time           CPU Iterations
----------------------------------------------------
BM_memcpy/32          11 ns         11 ns   79545455
BM_memcpy/32k       2181 ns       2185 ns     324074
BM_memcpy/32          12 ns         12 ns   54687500
BM_memcpy/32k       1834 ns       1837 ns     357143

Disabling Benchmarks

It is possible to temporarily disable benchmarks by renaming the benchmark function to have the prefix "DISABLED_". This will cause the benchmark to be skipped at runtime.

It is possible to compare the benchmarking results. See Additional Tooling Documentation

Sometimes it's useful to add extra context to the content printed before the results. By default this section includes information about the CPU on which the benchmarks are running. If you do want to add more context, you can use the benchmark_context command line flag:

$ ./run_benchmarks --benchmark_context=pwd=`pwd`
Run on (1 x 2300 MHz CPU)
pwd: /home/user/benchmark/
Benchmark              Time           CPU Iterations
----------------------------------------------------
BM_memcpy/32          11 ns         11 ns   79545455
BM_memcpy/32k       2181 ns       2185 ns     324074

You can get the same effect with the API:

  benchmark::AddCustomContext("foo", "bar");

Note that attempts to add a second value with the same key will fail with an error message.

When the benchmark binary is executed, each benchmark function is run serially. The number of iterations to run is determined dynamically by running the benchmark a few times and measuring the time taken and ensuring that the ultimate result will be statistically stable. As such, faster benchmark functions will be run for more iterations than slower benchmark functions, and the number of iterations is thus reported.

In all cases, the number of iterations for which the benchmark is run is governed by the amount of time the benchmark takes. Concretely, the number of iterations is at least one, not more than 1e9, until CPU time is greater than the minimum time, or the wallclock time is 5x minimum time. The minimum time is set per benchmark by calling MinTime on the registered benchmark object.

Furthermore warming up a benchmark might be necessary in order to get stable results because of e.g caching effects of the code under benchmark. Warming up means running the benchmark a given amount of time, before results are actually taken into account. The amount of time for which the warmup should be run can be set per benchmark by calling MinWarmUpTime on the registered benchmark object or for all benchmarks using the --benchmark_min_warmup_time command-line option. Note that MinWarmUpTime will overwrite the value of --benchmark_min_warmup_time for the single benchmark. How many iterations the warmup run of each benchmark takes is determined the same way as described in the paragraph above. Per default the warmup phase is set to 0 seconds and is therefore disabled.

Average timings are then reported over the iterations run. If multiple repetitions are requested using the --benchmark_repetitions command-line option, or at registration time, the benchmark function will be run several times and statistical results across these repetitions will also be reported.

As well as the per-benchmark entries, a preamble in the report will include information about the machine on which the benchmarks are run.

Global setup/teardown specific to each benchmark can be done by passing a callback to Setup/Teardown:

The setup/teardown callbacks will be invoked once for each benchmark. If the benchmark is multi-threaded (will run in k threads), they will be invoked exactly once before each run with k threads.

If the benchmark uses different size groups of threads, the above will be true for each size group.

Eg.,

static void DoSetup(const benchmark::State& state) {
}

static void DoTeardown(const benchmark::State& state) {
}

static void BM_func(benchmark::State& state) {...}

BENCHMARK(BM_func)->Arg(1)->Arg(3)->Threads(16)->Threads(32)->Setup(DoSetup)->Teardown(DoTeardown);

In this example, DoSetup and DoTearDown will be invoked 4 times each, specifically, once for each of this family:

  • BM_func_Arg_1_Threads_16, BM_func_Arg_1_Threads_32
  • BM_func_Arg_3_Threads_16, BM_func_Arg_3_Threads_32

Sometimes a family of benchmarks can be implemented with just one routine that takes an extra argument to specify which one of the family of benchmarks to run. For example, the following code defines a family of benchmarks for measuring the speed of memcpy() calls of different lengths:

static void BM_memcpy(benchmark::State& state) {
  char* src = new char[state.range(0)];
  char* dst = new char[state.range(0)];
  memset(src, 'x', state.range(0));
  for (auto _ : state)
    memcpy(dst, src, state.range(0));
  state.SetBytesProcessed(int64_t(state.iterations()) *
                          int64_t(state.range(0)));
  delete[] src;
  delete[] dst;
}
BENCHMARK(BM_memcpy)->Arg(8)->Arg(64)->Arg(512)->Arg(4<<10)->Arg(8<<10);

The preceding code is quite repetitive, and can be replaced with the following short-hand. The following invocation will pick a few appropriate arguments in the specified range and will generate a benchmark for each such argument.

BENCHMARK(BM_memcpy)->Range(8, 8<<10);

By default the arguments in the range are generated in multiples of eight and the command above selects [ 8, 64, 512, 4k, 8k ]. In the following code the range multiplier is changed to multiples of two.

BENCHMARK(BM_memcpy)->RangeMultiplier(2)->Range(8, 8<<10);

Now arguments generated are [ 8, 16, 32, 64, 128, 256, 512, 1024, 2k, 4k, 8k ].

The preceding code shows a method of defining a sparse range. The following example shows a method of defining a dense range. It is then used to benchmark the performance of std::vector initialization for uniformly increasing sizes.

static void BM_DenseRange(benchmark::State& state) {
  for(auto _ : state) {
    std::vector<int> v(state.range(0), state.range(0));
    auto data = v.data();
    benchmark::DoNotOptimize(data);
    benchmark::ClobberMemory();
  }
}
BENCHMARK(BM_DenseRange)->DenseRange(0, 1024, 128);

Now arguments generated are [ 0, 128, 256, 384, 512, 640, 768, 896, 1024 ].

You might have a benchmark that depends on two or more inputs. For example, the following code defines a family of benchmarks for measuring the speed of set insertion.

static void BM_SetInsert(benchmark::State& state) {
  std::set<int> data;
  for (auto _ : state) {
    state.PauseTiming();
    data = ConstructRandomSet(state.range(0));
    state.ResumeTiming();
    for (int j = 0; j < state.range(1); ++j)
      data.insert(RandomNumber());
  }
}
BENCHMARK(BM_SetInsert)
    ->Args({1<<10, 128})
    ->Args({2<<10, 128})
    ->Args({4<<10, 128})
    ->Args({8<<10, 128})
    ->Args({1<<10, 512})
    ->Args({2<<10, 512})
    ->Args({4<<10, 512})
    ->Args({8<<10, 512});

The preceding code is quite repetitive, and can be replaced with the following short-hand. The following macro will pick a few appropriate arguments in the product of the two specified ranges and will generate a benchmark for each such pair.

BENCHMARK(BM_SetInsert)->Ranges({{1<<10, 8<<10}, {128, 512}});

Some benchmarks may require specific argument values that cannot be expressed with Ranges. In this case, ArgsProduct offers the ability to generate a benchmark input for each combination in the product of the supplied vectors.

BENCHMARK(BM_SetInsert)
    ->ArgsProduct({{1<<10, 3<<10, 8<<10}, {20, 40, 60, 80}})
// would generate the same benchmark arguments as
BENCHMARK(BM_SetInsert)
    ->Args({1<<10, 20})
    ->Args({3<<10, 20})
    ->Args({8<<10, 20})
    ->Args({3<<10, 40})
    ->Args({8<<10, 40})
    ->Args({1<<10, 40})
    ->Args({1<<10, 60})
    ->Args({3<<10, 60})
    ->Args({8<<10, 60})
    ->Args({1<<10, 80})
    ->Args({3<<10, 80})
    ->Args({8<<10, 80});

For the most common scenarios, helper methods for creating a list of integers for a given sparse or dense range are provided.

BENCHMARK(BM_SetInsert)
    ->ArgsProduct({
      benchmark::CreateRange(8, 128, /*multi=*/2),
      benchmark::CreateDenseRange(1, 4, /*step=*/1)
    })
// would generate the same benchmark arguments as
BENCHMARK(BM_SetInsert)
    ->ArgsProduct({
      {8, 16, 32, 64, 128},
      {1, 2, 3, 4}
    });

For more complex patterns of inputs, passing a custom function to Apply allows programmatic specification of an arbitrary set of arguments on which to run the benchmark. The following example enumerates a dense range on one parameter, and a sparse range on the second.

static void CustomArguments(benchmark::internal::Benchmark* b) {
  for (int i = 0; i <= 10; ++i)
    for (int j = 32; j <= 1024*1024; j *= 8)
      b->Args({i, j});
}
BENCHMARK(BM_SetInsert)->Apply(CustomArguments);

Passing Arbitrary Arguments to a Benchmark

In C++11 it is possible to define a benchmark that takes an arbitrary number of extra arguments. The BENCHMARK_CAPTURE(func, test_case_name, ...args) macro creates a benchmark that invokes func with the benchmark::State as the first argument followed by the specified args.... The test_case_name is appended to the name of the benchmark and should describe the values passed.

template <class ...Args>
void BM_takes_args(benchmark::State& state, Args&&... args) {
  auto args_tuple = std::make_tuple(std::move(args)...);
  for (auto _ : state) {
    std::cout << std::get<0>(args_tuple) << ": " << std::get<1>(args_tuple)
              << '\n';
    [...]
  }
}
// Registers a benchmark named "BM_takes_args/int_string_test" that passes
// the specified values to `args`.
BENCHMARK_CAPTURE(BM_takes_args, int_string_test, 42, std::string("abc"));

// Registers the same benchmark "BM_takes_args/int_test" that passes
// the specified values to `args`.
BENCHMARK_CAPTURE(BM_takes_args, int_test, 42, 43);

Note that elements of ...args may refer to global variables. Users should avoid modifying global state inside of a benchmark.

Asymptotic complexity might be calculated for a family of benchmarks. The following code will calculate the coefficient for the high-order term in the running time and the normalized root-mean square error of string comparison.

static void BM_StringCompare(benchmark::State& state) {
  std::string s1(state.range(0), '-');
  std::string s2(state.range(0), '-');
  for (auto _ : state) {
    auto comparison_result = s1.compare(s2);
    benchmark::DoNotOptimize(comparison_result);
  }
  state.SetComplexityN(state.range(0));
}
BENCHMARK(BM_StringCompare)
    ->RangeMultiplier(2)->Range(1<<10, 1<<18)->Complexity(benchmark::oN);

As shown in the following invocation, asymptotic complexity might also be calculated automatically.

BENCHMARK(BM_StringCompare)
    ->RangeMultiplier(2)->Range(1<<10, 1<<18)->Complexity();

The following code will specify asymptotic complexity with a lambda function, that might be used to customize high-order term calculation.

BENCHMARK(BM_StringCompare)->RangeMultiplier(2)
    ->Range(1<<10, 1<<18)->Complexity([](benchmark::IterationCount n)->double{return n; });

You can change the benchmark's name as follows:

BENCHMARK(BM_memcpy)->Name("memcpy")->RangeMultiplier(2)->Range(8, 8<<10);

The invocation will execute the benchmark as before using BM_memcpy but changes the prefix in the report to memcpy.

This example produces and consumes messages of size sizeof(v) range_x times. It also outputs throughput in the absence of multiprogramming.

template <class Q> void BM_Sequential(benchmark::State& state) {
  Q q;
  typename Q::value_type v;
  for (auto _ : state) {
    for (int i = state.range(0); i--; )
      q.push(v);
    for (int e = state.range(0); e--; )
      q.Wait(&v);
  }
  // actually messages, not bytes:
  state.SetBytesProcessed(
      static_cast<int64_t>(state.iterations())*state.range(0));
}
// C++03
BENCHMARK_TEMPLATE(BM_Sequential, WaitQueue<int>)->Range(1<<0, 1<<10);

// C++11 or newer, you can use the BENCHMARK macro with template parameters:
BENCHMARK(BM_Sequential<WaitQueue<int>>)->Range(1<<0, 1<<10);

Three macros are provided for adding benchmark templates.

#ifdef BENCHMARK_HAS_CXX11
#define BENCHMARK(func<...>) // Takes any number of parameters.
#else // C++ < C++11
#define BENCHMARK_TEMPLATE(func, arg1)
#endif
#define BENCHMARK_TEMPLATE1(func, arg1)
#define BENCHMARK_TEMPLATE2(func, arg1, arg2)

Sometimes there is a need to template benchmarks, and provide arguments to them.

template <class Q> void BM_Sequential_With_Step(benchmark::State& state, int step) {
  Q q;
  typename Q::value_type v;
  for (auto _ : state) {
    for (int i = state.range(0); i-=step; )
      q.push(v);
    for (int e = state.range(0); e-=step; )
      q.Wait(&v);
  }
  // actually messages, not bytes:
  state.SetBytesProcessed(
      static_cast<int64_t>(state.iterations())*state.range(0));
}

BENCHMARK_TEMPLATE1_CAPTURE(BM_Sequential, WaitQueue<int>, Step1, 1)->Range(1<<0, 1<<10);

Fixture tests are created by first defining a type that derives from ::benchmark::Fixture and then creating/registering the tests using the following macros:

  • BENCHMARK_F(ClassName, Method)
  • BENCHMARK_DEFINE_F(ClassName, Method)
  • BENCHMARK_REGISTER_F(ClassName, Method)

For Example:

class MyFixture : public benchmark::Fixture {
public:
  void SetUp(::benchmark::State& state) {
  }

  void TearDown(::benchmark::State& state) {
  }
};

// Defines and registers `FooTest` using the class `MyFixture`.
BENCHMARK_F(MyFixture, FooTest)(benchmark::State& st) {
   for (auto _ : st) {
     ...
  }
}

// Only defines `BarTest` using the class `MyFixture`.
BENCHMARK_DEFINE_F(MyFixture, BarTest)(benchmark::State& st) {
   for (auto _ : st) {
     ...
  }
}
// `BarTest` is NOT registered.
BENCHMARK_REGISTER_F(MyFixture, BarTest)->Threads(2);
// `BarTest` is now registered.

Templated Fixtures

Also you can create templated fixture by using the following macros:

  • BENCHMARK_TEMPLATE_F(ClassName, Method, ...)
  • BENCHMARK_TEMPLATE_DEFINE_F(ClassName, Method, ...)

For example:

template<typename T>
class MyFixture : public benchmark::Fixture {};

// Defines and registers `IntTest` using the class template `MyFixture<int>`.
BENCHMARK_TEMPLATE_F(MyFixture, IntTest, int)(benchmark::State& st) {
   for (auto _ : st) {
     ...
  }
}

// Only defines `DoubleTest` using the class template `MyFixture<double>`.
BENCHMARK_TEMPLATE_DEFINE_F(MyFixture, DoubleTest, double)(benchmark::State& st) {
   for (auto _ : st) {
     ...
  }
}
// `DoubleTest` is NOT registered.
BENCHMARK_REGISTER_F(MyFixture, DoubleTest)->Threads(2);
// `DoubleTest` is now registered.

You can add your own counters with user-defined names. The example below will add columns "Foo", "Bar" and "Baz" in its output:

static void UserCountersExample1(benchmark::State& state) {
  double numFoos = 0, numBars = 0, numBazs = 0;
  for (auto _ : state) {
    // ... count Foo,Bar,Baz events
  }
  state.counters["Foo"] = numFoos;
  state.counters["Bar"] = numBars;
  state.counters["Baz"] = numBazs;
}

The state.counters object is a std::map with std::string keys and Counter values. The latter is a double-like class, via an implicit conversion to double&. Thus you can use all of the standard arithmetic assignment operators (=,+=,-=,*=,/=) to change the value of each counter.

In multithreaded benchmarks, each counter is set on the calling thread only. When the benchmark finishes, the counters from each thread will be summed; the resulting sum is the value which will be shown for the benchmark.

The Counter constructor accepts three parameters: the value as a double ; a bit flag which allows you to show counters as rates, and/or as per-thread iteration, and/or as per-thread averages, and/or iteration invariants, and/or finally inverting the result; and a flag specifying the 'unit' - i.e. is 1k a 1000 (default, benchmark::Counter::OneK::kIs1000), or 1024 (benchmark::Counter::OneK::kIs1024)?

  // sets a simple counter
  state.counters["Foo"] = numFoos;

  // Set the counter as a rate. It will be presented divided
  // by the duration of the benchmark.
  // Meaning: per one second, how many 'foo's are processed?
  state.counters["FooRate"] = Counter(numFoos, benchmark::Counter::kIsRate);

  // Set the counter as a rate. It will be presented divided
  // by the duration of the benchmark, and the result inverted.
  // Meaning: how many seconds it takes to process one 'foo'?
  state.counters["FooInvRate"] = Counter(numFoos, benchmark::Counter::kIsRate | benchmark::Counter::kInvert);

  // Set the counter as a thread-average quantity. It will
  // be presented divided by the number of threads.
  state.counters["FooAvg"] = Counter(numFoos, benchmark::Counter::kAvgThreads);

  // There's also a combined flag:
  state.counters["FooAvgRate"] = Counter(numFoos,benchmark::Counter::kAvgThreadsRate);

  // This says that we process with the rate of state.range(0) bytes every iteration:
  state.counters["BytesProcessed"] = Counter(state.range(0), benchmark::Counter::kIsIterationInvariantRate, benchmark::Counter::OneK::kIs1024);

When you're compiling in C++11 mode or later you can use insert() with std::initializer_list:

  // With C++11, this can be done:
  state.counters.insert({{"Foo", numFoos}, {"Bar", numBars}, {"Baz", numBazs}});
  // ... instead of:
  state.counters["Foo"] = numFoos;
  state.counters["Bar"] = numBars;
  state.counters["Baz"] = numBazs;

Counter Reporting

When using the console reporter, by default, user counters are printed at the end after the table, the same way as bytes_processed and items_processed. This is best for cases in which there are few counters, or where there are only a couple of lines per benchmark. Here's an example of the default output:

------------------------------------------------------------------------------
Benchmark                        Time           CPU Iterations UserCounters...
------------------------------------------------------------------------------
BM_UserCounter/threads:8      2248 ns      10277 ns      68808 Bar=16 Bat=40 Baz=24 Foo=8
BM_UserCounter/threads:1      9797 ns       9788 ns      71523 Bar=2 Bat=5 Baz=3 Foo=1024m
BM_UserCounter/threads:2      4924 ns       9842 ns      71036 Bar=4 Bat=10 Baz=6 Foo=2
BM_UserCounter/threads:4      2589 ns      10284 ns      68012 Bar=8 Bat=20 Baz=12 Foo=4
BM_UserCounter/threads:8      2212 ns      10287 ns      68040 Bar=16 Bat=40 Baz=24 Foo=8
BM_UserCounter/threads:16     1782 ns      10278 ns      68144 Bar=32 Bat=80 Baz=48 Foo=16
BM_UserCounter/threads:32     1291 ns      10296 ns      68256 Bar=64 Bat=160 Baz=96 Foo=32
BM_UserCounter/threads:4      2615 ns      10307 ns      68040 Bar=8 Bat=20 Baz=12 Foo=4
BM_Factorial                    26 ns         26 ns   26608979 40320
BM_Factorial/real_time          26 ns         26 ns   26587936 40320
BM_CalculatePiRange/1           16 ns         16 ns   45704255 0
BM_CalculatePiRange/8           73 ns         73 ns    9520927 3.28374
BM_CalculatePiRange/64         609 ns        609 ns    1140647 3.15746
BM_CalculatePiRange/512       4900 ns       4901 ns     142696 3.14355

If this doesn't suit you, you can print each counter as a table column by passing the flag --benchmark_counters_tabular=true to the benchmark application. This is best for cases in which there are a lot of counters, or a lot of lines per individual benchmark. Note that this will trigger a reprinting of the table header any time the counter set changes between individual benchmarks. Here's an example of corresponding output when --benchmark_counters_tabular=true is passed:

---------------------------------------------------------------------------------------
Benchmark                        Time           CPU Iterations    Bar   Bat   Baz   Foo
---------------------------------------------------------------------------------------
BM_UserCounter/threads:8      2198 ns       9953 ns      70688     16    40    24     8
BM_UserCounter/threads:1      9504 ns       9504 ns      73787      2     5     3     1
BM_UserCounter/threads:2      4775 ns       9550 ns      72606      4    10     6     2
BM_UserCounter/threads:4      2508 ns       9951 ns      70332      8    20    12     4
BM_UserCounter/threads:8      2055 ns       9933 ns      70344     16    40    24     8
BM_UserCounter/threads:16     1610 ns       9946 ns      70720     32    80    48    16
BM_UserCounter/threads:32     1192 ns       9948 ns      70496     64   160    96    32
BM_UserCounter/threads:4      2506 ns       9949 ns      70332      8    20    12     4
--------------------------------------------------------------
Benchmark                        Time           CPU Iterations
--------------------------------------------------------------
BM_Factorial                    26 ns         26 ns   26392245 40320
BM_Factorial/real_time          26 ns         26 ns   26494107 40320
BM_CalculatePiRange/1           15 ns         15 ns   45571597 0
BM_CalculatePiRange/8           74 ns         74 ns    9450212 3.28374
BM_CalculatePiRange/64         595 ns        595 ns    1173901 3.15746
BM_CalculatePiRange/512       4752 ns       4752 ns     147380 3.14355
BM_CalculatePiRange/4k       37970 ns      37972 ns      18453 3.14184
BM_CalculatePiRange/32k     303733 ns     303744 ns       2305 3.14162
BM_CalculatePiRange/256k   2434095 ns    2434186 ns        288 3.1416
BM_CalculatePiRange/1024k  9721140 ns    9721413 ns         71 3.14159
BM_CalculatePi/threads:8      2255 ns       9943 ns      70936

Note above the additional header printed when the benchmark changes from BM_UserCounter to BM_Factorial. This is because BM_Factorial does not have the same counter set as BM_UserCounter.

In a multithreaded test (benchmark invoked by multiple threads simultaneously), it is guaranteed that none of the threads will start until all have reached the start of the benchmark loop, and all will have finished before any thread exits the benchmark loop. (This behavior is also provided by the KeepRunning() API) As such, any global setup or teardown can be wrapped in a check against the thread index:

static void BM_MultiThreaded(benchmark::State& state) {
  if (state.thread_index() == 0) {
    // Setup code here.
  }
  for (auto _ : state) {
    // Run the test as normal.
  }
  if (state.thread_index() == 0) {
    // Teardown code here.
  }
}
BENCHMARK(BM_MultiThreaded)->Threads(2);

To run the benchmark across a range of thread counts, instead of Threads, use ThreadRange. This takes two parameters (min_threads and max_threads) and runs the benchmark once for values in the inclusive range. For example:

BENCHMARK(BM_MultiThreaded)->ThreadRange(1, 8);

will run BM_MultiThreaded with thread counts 1, 2, 4, and 8.

If the benchmarked code itself uses threads and you want to compare it to single-threaded code, you may want to use real-time ("wallclock") measurements for latency comparisons:

BENCHMARK(BM_test)->Range(8, 8<<10)->UseRealTime();

Without UseRealTime, CPU time is used by default.

By default, the CPU timer only measures the time spent by the main thread. If the benchmark itself uses threads internally, this measurement may not be what you are looking for. Instead, there is a way to measure the total CPU usage of the process, by all the threads.

void callee(int i);

static void MyMain(int size) {
#pragma omp parallel for
  for(int i = 0; i < size; i++)
    callee(i);
}

static void BM_OpenMP(benchmark::State& state) {
  for (auto _ : state)
    MyMain(state.range(0));
}

// Measure the time spent by the main thread, use it to decide for how long to
// run the benchmark loop. Depending on the internal implementation detail may
// measure to anywhere from near-zero (the overhead spent before/after work
// handoff to worker thread[s]) to the whole single-thread time.
BENCHMARK(BM_OpenMP)->Range(8, 8<<10);

// Measure the user-visible time, the wall clock (literally, the time that
// has passed on the clock on the wall), use it to decide for how long to
// run the benchmark loop. This will always be meaningful, and will match the
// time spent by the main thread in single-threaded case, in general decreasing
// with the number of internal threads doing the work.
BENCHMARK(BM_OpenMP)->Range(8, 8<<10)->UseRealTime();

// Measure the total CPU consumption, use it to decide for how long to
// run the benchmark loop. This will always measure to no less than the
// time spent by the main thread in single-threaded case.
BENCHMARK(BM_OpenMP)->Range(8, 8<<10)->MeasureProcessCPUTime();

// A mixture of the last two. Measure the total CPU consumption, but use the
// wall clock to decide for how long to run the benchmark loop.
BENCHMARK(BM_OpenMP)->Range(8, 8<<10)->MeasureProcessCPUTime()->UseRealTime();

Controlling Timers

Normally, the entire duration of the work loop (for (auto _ : state) {}) is measured. But sometimes, it is necessary to do some work inside of that loop, every iteration, but without counting that time to the benchmark time. That is possible, although it is not recommended, since it has high overhead.

static void BM_SetInsert_With_Timer_Control(benchmark::State& state) {
  std::set<int> data;
  for (auto _ : state) {
    state.PauseTiming(); // Stop timers. They will not count until they are resumed.
    data = ConstructRandomSet(state.range(0)); // Do something that should not be measured
    state.ResumeTiming(); // And resume timers. They are now counting again.
    // The rest will be measured.
    for (int j = 0; j < state.range(1); ++j)
      data.insert(RandomNumber());
  }
}
BENCHMARK(BM_SetInsert_With_Timer_Control)->Ranges({{1<<10, 8<<10}, {128, 512}});

For benchmarking something for which neither CPU time nor real-time are correct or accurate enough, completely manual timing is supported using the UseManualTime function.

When UseManualTime is used, the benchmarked code must call SetIterationTime once per iteration of the benchmark loop to report the manually measured time.

An example use case for this is benchmarking GPU execution (e.g. OpenCL or CUDA kernels, OpenGL or Vulkan or Direct3D draw calls), which cannot be accurately measured using CPU time or real-time. Instead, they can be measured accurately using a dedicated API, and these measurement results can be reported back with SetIterationTime.

static void BM_ManualTiming(benchmark::State& state) {
  int microseconds = state.range(0);
  std::chrono::duration<double, std::micro> sleep_duration {
    static_cast<double>(microseconds)
  };

  for (auto _ : state) {
    auto start = std::chrono::high_resolution_clock::now();
    // Simulate some useful workload with a sleep
    std::this_thread::sleep_for(sleep_duration);
    auto end = std::chrono::high_resolution_clock::now();

    auto elapsed_seconds =
      std::chrono::duration_cast<std::chrono::duration<double>>(
        end - start);

    state.SetIterationTime(elapsed_seconds.count());
  }
}
BENCHMARK(BM_ManualTiming)->Range(1, 1<<17)->UseManualTime();

If a benchmark runs a few milliseconds it may be hard to visually compare the measured times, since the output data is given in nanoseconds per default. In order to manually set the time unit, you can specify it manually:

BENCHMARK(BM_test)->Unit(benchmark::kMillisecond);

Additionally the default time unit can be set globally with the --benchmark_time_unit={ns|us|ms|s} command line argument. The argument only affects benchmarks where the time unit is not set explicitly.

To prevent a value or expression from being optimized away by the compiler the benchmark::DoNotOptimize(...) and benchmark::ClobberMemory() functions can be used.

static void BM_test(benchmark::State& state) {
  for (auto _ : state) {
      int x = 0;
      for (int i=0; i < 64; ++i) {
        benchmark::DoNotOptimize(x += i);
      }
  }
}

DoNotOptimize(<expr>) forces the result of <expr> to be stored in either memory or a register. For GNU based compilers it acts as read/write barrier for global memory. More specifically it forces the compiler to flush pending writes to memory and reload any other values as necessary.

Note that DoNotOptimize(<expr>) does not prevent optimizations on <expr> in any way. <expr> may even be removed entirely when the result is already known. For example:

  // Example 1: `<expr>` is removed entirely.
  int foo(int x) { return x + 42; }
  while (...) DoNotOptimize(foo(0)); // Optimized to DoNotOptimize(42);

  // Example 2: Result of '<expr>' is only reused.
  int bar(int) __attribute__((const));
  while (...) DoNotOptimize(bar(0)); // Optimized to:
  // int __result__ = bar(0);
  // while (...) DoNotOptimize(__result__);

The second tool for preventing optimizations is ClobberMemory(). In essence ClobberMemory() forces the compiler to perform all pending writes to global memory. Memory managed by block scope objects must be "escaped" using DoNotOptimize(...) before it can be clobbered. In the below example ClobberMemory() prevents the call to v.push_back(42) from being optimized away.

static void BM_vector_push_back(benchmark::State& state) {
  for (auto _ : state) {
    std::vector<int> v;
    v.reserve(1);
    auto data = v.data();           // Allow v.data() to be clobbered. Pass as non-const
    benchmark::DoNotOptimize(data); // lvalue to avoid undesired compiler optimizations
    v.push_back(42);
    benchmark::ClobberMemory(); // Force 42 to be written to memory.
  }
}

Note that ClobberMemory() is only available for GNU or MSVC based compilers.

By default each benchmark is run once and that single result is reported. However benchmarks are often noisy and a single result may not be representative of the overall behavior. For this reason it's possible to repeatedly rerun the benchmark.

The number of runs of each benchmark is specified globally by the --benchmark_repetitions flag or on a per benchmark basis by calling Repetitions on the registered benchmark object. When a benchmark is run more than once the mean, median, standard deviation and coefficient of variation of the runs will be reported.

Additionally the --benchmark_report_aggregates_only={true|false}, --benchmark_display_aggregates_only={true|false} flags or ReportAggregatesOnly(bool), DisplayAggregatesOnly(bool) functions can be used to change how repeated tests are reported. By default the result of each repeated run is reported. When report aggregates only option is true, only the aggregates (i.e. mean, median, standard deviation and coefficient of variation, maybe complexity measurements if they were requested) of the runs is reported, to both the reporters - standard output (console), and the file. However when only the display aggregates only option is true, only the aggregates are displayed in the standard output, while the file output still contains everything. Calling ReportAggregatesOnly(bool) / DisplayAggregatesOnly(bool) on a registered benchmark object overrides the value of the appropriate flag for that benchmark.

While having these aggregates is nice, this may not be enough for everyone. For example you may want to know what the largest observation is, e.g. because you have some real-time constraints. This is easy. The following code will specify a custom statistic to be calculated, defined by a lambda function.

void BM_spin_empty(benchmark::State& state) {
  for (auto _ : state) {
    for (int x = 0; x < state.range(0); ++x) {
      benchmark::DoNotOptimize(x);
    }
  }
}

BENCHMARK(BM_spin_empty)
  ->ComputeStatistics("max", [](const std::vector<double>& v) -> double {
    return *(std::max_element(std::begin(v), std::end(v)));
  })
  ->Arg(512);

While usually the statistics produce values in time units, you can also produce percentages:

void BM_spin_empty(benchmark::State& state) {
  for (auto _ : state) {
    for (int x = 0; x < state.range(0); ++x) {
      benchmark::DoNotOptimize(x);
    }
  }
}

BENCHMARK(BM_spin_empty)
  ->ComputeStatistics("ratio", [](const std::vector<double>& v) -> double {
    return std::begin(v) / std::end(v);
  }, benchmark::StatisticUnit::kPercentage)
  ->Arg(512);

It's often useful to also track memory usage for benchmarks, alongside CPU performance. For this reason, benchmark offers the RegisterMemoryManager method that allows a custom MemoryManager to be injected.

If set, the MemoryManager::Start and MemoryManager::Stop methods will be called at the start and end of benchmark runs to allow user code to fill out a report on the number of allocations, bytes used, etc.

This data will then be reported alongside other performance data, currently only when using JSON output.

It's often useful to also profile benchmarks in particular ways, in addition to CPU performance. For this reason, benchmark offers the RegisterProfilerManager method that allows a custom ProfilerManager to be injected.

If set, the ProfilerManager::AfterSetupStart and ProfilerManager::BeforeTeardownStop methods will be called at the start and end of a separate benchmark run to allow user code to collect and report user-provided profile metrics.

Output collected from this profiling run must be reported separately.

The RegisterBenchmark(name, func, args...) function provides an alternative way to create and register benchmarks. RegisterBenchmark(name, func, args...) creates, registers, and returns a pointer to a new benchmark with the specified name that invokes func(st, args...) where st is a benchmark::State object.

Unlike the BENCHMARK registration macros, which can only be used at the global scope, the RegisterBenchmark can be called anywhere. This allows for benchmark tests to be registered programmatically.

Additionally RegisterBenchmark allows any callable object to be registered as a benchmark. Including capturing lambdas and function objects.

For Example:

auto BM_test = [](benchmark::State& st, auto Inputs) { /* ... */ };

int main(int argc, char** argv) {
  for (auto& test_input : { /* ... */ })
      benchmark::RegisterBenchmark(test_input.name(), BM_test, test_input);
  benchmark::Initialize(&argc, argv);
  benchmark::RunSpecifiedBenchmarks();
  benchmark::Shutdown();
}

When errors caused by external influences, such as file I/O and network communication, occur within a benchmark the State::SkipWithError(const std::string& msg) function can be used to skip that run of benchmark and report the error. Note that only future iterations of the KeepRunning() are skipped. For the ranged-for version of the benchmark loop Users must explicitly exit the loop, otherwise all iterations will be performed. Users may explicitly return to exit the benchmark immediately.

The SkipWithError(...) function may be used at any point within the benchmark, including before and after the benchmark loop. Moreover, if SkipWithError(...) has been used, it is not required to reach the benchmark loop and one may return from the benchmark function early.

For example:

static void BM_test(benchmark::State& state) {
  auto resource = GetResource();
  if (!resource.good()) {
    state.SkipWithError("Resource is not good!");
    // KeepRunning() loop will not be entered.
  }
  while (state.KeepRunning()) {
    auto data = resource.read_data();
    if (!resource.good()) {
      state.SkipWithError("Failed to read data!");
      break; // Needed to skip the rest of the iteration.
    }
    do_stuff(data);
  }
}

static void BM_test_ranged_fo(benchmark::State & state) {
  auto resource = GetResource();
  if (!resource.good()) {
    state.SkipWithError("Resource is not good!");
    return; // Early return is allowed when SkipWithError() has been used.
  }
  for (auto _ : state) {
    auto data = resource.read_data();
    if (!resource.good()) {
      state.SkipWithError("Failed to read data!");
      break; // REQUIRED to prevent all further iterations.
    }
    do_stuff(data);
  }
}

In C++11 mode, a ranged-based for loop should be used in preference to the KeepRunning loop for running the benchmarks. For example:

static void BM_Fast(benchmark::State &state) {
  for (auto _ : state) {
    FastOperation();
  }
}
BENCHMARK(BM_Fast);

The reason the ranged-for loop is faster than using KeepRunning, is because KeepRunning requires a memory load and store of the iteration count ever iteration, whereas the ranged-for variant is able to keep the iteration count in a register.

For example, an empty inner loop of using the ranged-based for method looks like:

# Loop Init
  mov rbx, qword ptr [r14 + 104]
  call benchmark::State::StartKeepRunning()
  test rbx, rbx
  je .LoopEnd
.LoopHeader: # =>This Inner Loop Header: Depth=1
  add rbx, -1
  jne .LoopHeader
.LoopEnd:

Compared to an empty KeepRunning loop, which looks like:

.LoopHeader: # in Loop: Header=BB0_3 Depth=1
  cmp byte ptr [rbx], 1
  jne .LoopInit
.LoopBody: # =>This Inner Loop Header: Depth=1
  mov rax, qword ptr [rbx + 8]
  lea rcx, [rax + 1]
  mov qword ptr [rbx + 8], rcx
  cmp rax, qword ptr [rbx + 104]
  jb .LoopHeader
  jmp .LoopEnd
.LoopInit:
  mov rdi, rbx
  call benchmark::State::StartKeepRunning()
  jmp .LoopBody
.LoopEnd:

Unless C++03 compatibility is required, the ranged-for variant of writing the benchmark loop should be preferred.

If you see this error:

***WARNING*** CPU scaling is enabled, the benchmark real time measurements may
be noisy and will incur extra overhead.

you might want to disable the CPU frequency scaling while running the benchmark, as well as consider other ways to stabilize the performance of your system while benchmarking.

See Reducing Variance for more information.