Integrate Nonius benchmark into Catch2

Changes done to Nonius:
* Moved things into "Catch::Benchmark" namespace
* Benchmarks were integrated with `TEST_CASE`/`SECTION`/`GENERATE` macros
* Removed Nonius's parameters for benchmarks, Generators should be used instead
* Added relevant methods to the reporter interface (default-implemented, to avoid
breaking existing 3rd party reporters)
* Async processing is guarded with `_REENTRANT` macro for GCC/Clang, used by default
on MSVC
* Added a macro `CATCH_CONFIG_DISABLE_BENCHMARKING` that removes all traces of
benchmarking from Catch
This commit is contained in:
Joachim Meyer 2019-04-23 23:41:13 +02:00 committed by Martin Hořeňovský
parent 00347f1e79
commit ce2560ca95
46 changed files with 2614 additions and 190 deletions

249
docs/benchmarks.md Normal file
View File

@ -0,0 +1,249 @@
# Authoring benchmarks
Writing benchmarks is not easy. Catch simplifies certain aspects but you'll
always need to take care about various aspects. Understanding a few things about
the way Catch runs your code will be very helpful when writing your benchmarks.
First off, let's go over some terminology that will be used throughout this
guide.
- *User code*: user code is the code that the user provides to be measured.
- *Run*: one run is one execution of the user code.
- *Sample*: one sample is one data point obtained by measuring the time it takes
to perform a certain number of runs. One sample can consist of more than one
run if the clock available does not have enough resolution to accurately
measure a single run. All samples for a given benchmark execution are obtained
with the same number of runs.
## Execution procedure
Now I can explain how a benchmark is executed in Catch. There are three main
steps, though the first does not need to be repeated for every benchmark.
1. *Environmental probe*: before any benchmarks can be executed, the clock's
resolution is estimated. A few other environmental artifacts are also estimated
at this point, like the cost of calling the clock function, but they almost
never have any impact in the results.
2. *Estimation*: the user code is executed a few times to obtain an estimate of
the amount of runs that should be in each sample. This also has the potential
effect of bringing relevant code and data into the caches before the actual
measurement starts.
3. *Measurement*: all the samples are collected sequentially by performing the
number of runs estimated in the previous step for each sample.
This already gives us one important rule for writing benchmarks for Catch: the
benchmarks must be repeatable. The user code will be executed several times, and
the number of times it will be executed during the estimation step cannot be
known beforehand since it depends on the time it takes to execute the code.
User code that cannot be executed repeatedly will lead to bogus results or
crashes.
## Benchmark specification
Benchmarks can be specified anywhere inside a Catch test case.
There is a simple and a slightly more advanced version of the `BENCHMARK` macro.
Let's have a look how a naive Fibonacci implementation could be benchmarked:
```c++
std::uint64_t Fibonacci(std::uint64_t number) {
return number < 2 ? 1 : Fibonacci(number - 1) + Fibonacci(number - 2);
}
```
Now the most straight forward way to benchmark this function, is just adding a `BENCHMARK` macro to our test case:
```c++
TEST_CASE("Fibonacci") {
CHECK(Fibonacci(0) == 1);
// some more asserts..
CHECK(Fibonacci(5) == 8);
// some more asserts..
// now let's benchmark:
BENCHMARK("Fibonacci 20") {
return Fibonacci(20);
};
BENCHMARK("Fibonacci 25") {
return Fibonacci(25);
};
BENCHMARK("Fibonacci 30") {
return Fibonacci(30);
};
BENCHMARK("Fibonacci 35") {
return Fibonacci(35);
};
}
```
There's a few things to note:
- As `BENCHMARK` expands to a lambda expression it is necessary to add a semicolon after
the closing brace (as opposed to the first experimental version).
- The `return` is a handy way to avoid the compiler optimizing away the benchmark code.
Running this already runs the benchmarks and outputs something similar to:
```
-------------------------------------------------------------------------------
Fibonacci
-------------------------------------------------------------------------------
C:\path\to\Catch2\Benchmark.tests.cpp(10)
...............................................................................
benchmark name samples iterations estimated
mean low mean high mean
std dev low std dev high std dev
-------------------------------------------------------------------------------
Fibonacci 20 100 416439 83.2878 ms
2 ns 2 ns 2 ns
0 ns 0 ns 0 ns
Fibonacci 25 100 400776 80.1552 ms
3 ns 3 ns 3 ns
0 ns 0 ns 0 ns
Fibonacci 30 100 396873 79.3746 ms
17 ns 17 ns 17 ns
0 ns 0 ns 0 ns
Fibonacci 35 100 145169 87.1014 ms
468 ns 464 ns 473 ns
21 ns 15 ns 34 ns
```
### Advanced benchmarking
The simplest use case shown above, takes no arguments and just runs the user code that needs to be measured.
However, if using the `BENCHMARK_ADVANCED` macro and adding a `Catch::Benchmark::Chronometer` argument after
the macro, some advanced features are available. The contents of the simple benchmarks are invoked once per run,
while the blocks of the advanced benchmarks are invoked exactly twice:
once during the estimation phase, and another time during the execution phase.
```c++
BENCHMARK("simple"){ return long_computation(); };
BENCHMARK_ADVANCED("advanced")(Catch::Benchmark::Chronometer meter) {
set_up();
meter.measure([] { return long_computation(); });
};
```
These advanced benchmarks no longer consist entirely of user code to be measured.
In these cases, the code to be measured is provided via the
`Catch::Benchmark::Chronometer::measure` member function. This allows you to set up any
kind of state that might be required for the benchmark but is not to be included
in the measurements, like making a vector of random integers to feed to a
sorting algorithm.
A single call to `Catch::Benchmark::Chronometer::measure` performs the actual measurements
by invoking the callable object passed in as many times as necessary. Anything
that needs to be done outside the measurement can be done outside the call to
`measure`.
The callable object passed in to `measure` can optionally accept an `int`
parameter.
```c++
meter.measure([](int i) { return long_computation(i); });
```
If it accepts an `int` parameter, the sequence number of each run will be passed
in, starting with 0. This is useful if you want to measure some mutating code,
for example. The number of runs can be known beforehand by calling
`Catch::Benchmark::Chronometer::runs`; with this one can set up a different instance to be
mutated by each run.
```c++
std::vector<std::string> v(meter.runs());
std::fill(v.begin(), v.end(), test_string());
meter.measure([&v](int i) { in_place_escape(v[i]); });
```
Note that it is not possible to simply use the same instance for different runs
and resetting it between each run since that would pollute the measurements with
the resetting code.
It is also possible to just provide an argument name to the simple `BENCHMARK` macro to get
the same semantics as providing a callable to `meter.measure` with `int` argument:
```c++
BENCHMARK("indexed", i){ return long_computation(i); };
```
### Constructors and destructors
All of these tools give you a lot mileage, but there are two things that still
need special handling: constructors and destructors. The problem is that if you
use automatic objects they get destroyed by the end of the scope, so you end up
measuring the time for construction and destruction together. And if you use
dynamic allocation instead, you end up including the time to allocate memory in
the measurements.
To solve this conundrum, Catch provides class templates that let you manually
construct and destroy objects without dynamic allocation and in a way that lets
you measure construction and destruction separately.
```c++
BENCHMARK_ADVANCED("construct")(Catch::Benchmark::Chronometer meter)
{
std::vector<Catch::Benchmark::storage_for<std::string>> storage(meter.runs());
meter.measure([&](int i) { storage[i].construct("thing"); });
})
BENCHMARK_ADVANCED("destroy", [](Catch::Benchmark::Chronometer meter)
{
std::vector<Catch::Benchmark::destructable_object<std::string>> storage(meter.runs());
for(auto&& o : storage)
o.construct("thing");
meter.measure([&](int i) { storage[i].destruct(); });
})
```
`Catch::Benchmark::storage_for<T>` objects are just pieces of raw storage suitable for `T`
objects. You can use the `Catch::Benchmark::storage_for::construct` member function to call a constructor and
create an object in that storage. So if you want to measure the time it takes
for a certain constructor to run, you can just measure the time it takes to run
this function.
When the lifetime of a `Catch::Benchmark::storage_for<T>` object ends, if an actual object was
constructed there it will be automatically destroyed, so nothing leaks.
If you want to measure a destructor, though, we need to use
`Catch::Benchmark::destructable_object<T>`. These objects are similar to
`Catch::Benchmark::storage_for<T>` in that construction of the `T` object is manual, but
it does not destroy anything automatically. Instead, you are required to call
the `Catch::Benchmark::destructable_object::destruct` member function, which is what you
can use to measure the destruction time.
### The optimizer
Sometimes the optimizer will optimize away the very code that you want to
measure. There are several ways to use results that will prevent the optimiser
from removing them. You can use the `volatile` keyword, or you can output the
value to standard output or to a file, both of which force the program to
actually generate the value somehow.
Catch adds a third option. The values returned by any function provided as user
code are guaranteed to be evaluated and not optimised out. This means that if
your user code consists of computing a certain value, you don't need to bother
with using `volatile` or forcing output. Just `return` it from the function.
That helps with keeping the code in a natural fashion.
Here's an example:
```c++
// may measure nothing at all by skipping the long calculation since its
// result is not used
BENCHMARK("no return"){ long_calculation(); };
// the result of long_calculation() is guaranteed to be computed somehow
BENCHMARK("with return"){ return long_calculation(); };
```
However, there's no other form of control over the optimizer whatsoever. It is
up to you to write a benchmark that actually measures what you want and doesn't
just measure the time to do a whole bunch of nothing.
To sum up, there are two simple rules: whatever you would do in handwritten code
to control optimization still works in Catch; and Catch makes return values
from user code into observable effects that can't be optimized away.
<i>Adapted from nonius' documentation.</i>

View File

@ -20,7 +20,10 @@
[Specify a seed for the Random Number Generator](#specify-a-seed-for-the-random-number-generator)<br>
[Identify framework and version according to the libIdentify standard](#identify-framework-and-version-according-to-the-libidentify-standard)<br>
[Wait for key before continuing](#wait-for-key-before-continuing)<br>
[Specify multiples of clock resolution to run benchmarks for](#specify-multiples-of-clock-resolution-to-run-benchmarks-for)<br>
[Specify the number of benchmark samples to collect](#specify-the-number-of-benchmark-samples-to-collect)<br>
[Specify the number of benchmark resamples for bootstrapping](#specify-the-number-of-resamples-for-bootstrapping)<br>
[Specify the confidence interval for bootstrapping](#specify-the-confidence-interval-for-bootstrapping)<br>
[Disable statistical analysis of collected benchmark samples](#disable-statistical-analysis-of-collected-benchmark-samples)<br>
[Usage](#usage)<br>
[Specify the section to run](#specify-the-section-to-run)<br>
[Filenames as tags](#filenames-as-tags)<br>
@ -57,7 +60,10 @@ Click one of the following links to take you straight to that option - or scroll
<a href="#rng-seed"> ` --rng-seed`</a><br />
<a href="#libidentify"> ` --libidentify`</a><br />
<a href="#wait-for-keypress"> ` --wait-for-keypress`</a><br />
<a href="#benchmark-resolution-multiple"> ` --benchmark-resolution-multiple`</a><br />
<a href="#benchmark-samples"> ` --benchmark-samples`</a><br />
<a href="#benchmark-resamples"> ` --benchmark-resamples`</a><br />
<a href="#benchmark-confidence-interval"> ` --benchmark-confidence-interval`</a><br />
<a href="#benchmark-no-analysis"> ` --benchmark-no-analysis`</a><br />
<a href="#use-colour"> ` --use-colour`</a><br />
</br>
@ -267,13 +273,40 @@ See [The LibIdentify repo for more information and examples](https://github.com/
Will cause the executable to print a message and wait until the return/ enter key is pressed before continuing -
either before running any tests, after running all tests - or both, depending on the argument.
<a id="benchmark-resolution-multiple"></a>
## Specify multiples of clock resolution to run benchmarks for
<pre>--benchmark-resolution-multiple &lt;multiplier&gt;</pre>
<a id="benchmark-samples"></a>
## Specify the number of benchmark samples to collect
<pre>--benchmark-samples &lt;# of samples&gt;</pre>
When running benchmarks the clock resolution is estimated. Benchmarks are then run for exponentially increasing
numbers of iterations until some multiple of the estimated resolution is exceed. By default that multiple is 100, but
it can be overridden here.
When running benchmarks a number of "samples" is collected. This is the base data for later statistical analysis.
Per sample a clock resolution dependent number of iterations of the user code is run, which is independent of the number of samples. Defaults to 100.
<a id="benchmark-resamples"></a>
## Specify the number of resamples for bootstrapping
<pre>--benchmark-resamples &lt;# of resamples&gt;</pre>
After the measurements are performed, statistical [bootstrapping] is performed
on the samples. The number of resamples for that bootstrapping is configurable
but defaults to 100000. Due to the bootstrapping it is possible to give
estimates for the mean and standard deviation. The estimates come with a lower
bound and an upper bound, and the confidence interval (which is configurable but
defaults to 95%).
[bootstrapping]: http://en.wikipedia.org/wiki/Bootstrapping_%28statistics%29
<a id="benchmark-confidence-interval"></a>
## Specify the confidence-interval for bootstrapping
<pre>--benchmark-confidence-interval &lt;confidence-interval&gt;</pre>
The confidence-interval is used for statistical bootstrapping on the samples to
calculate the upper and lower bounds of mean and standard deviation.
Must be between 0 and 1 and defaults to 0.95.
<a id="benchmark-no-analysis"></a>
## Disable statistical analysis of collected benchmark samples
<pre>--benchmark-no-analysis</pre>
When this flag is specified no bootstrapping or any other statistical analysis is performed.
Instead the user code is only measured and the plain mean from the samples is reported.
<a id="usage"></a>
## Usage

View File

@ -149,6 +149,7 @@ by using `_NO_` in the macro, e.g. `CATCH_CONFIG_NO_CPP17_UNCAUGHT_EXCEPTIONS`.
CATCH_CONFIG_DISABLE // Disables assertions and test case registration
CATCH_CONFIG_WCHAR // Enables use of wchart_t
CATCH_CONFIG_EXPERIMENTAL_REDIRECT // Enables the new (experimental) way of capturing stdout/stderr
CATCH_CONFIG_DISABLE_BENCHMARKING // Disables the compile-time heavy benchmarking features
Currently Catch enables `CATCH_CONFIG_WINDOWS_SEH` only when compiled with MSVC, because some versions of MinGW do not have the necessary Win32 API support.

View File

@ -33,6 +33,9 @@
# if defined(CATCH_CONFIG_DISABLE_MATCHERS)
# undef CATCH_CONFIG_DISABLE_MATCHERS
# endif
# if defined(CATCH_CONFIG_DISABLE_BENCHMARKING)
# undef CATCH_CONFIG_DISABLE_BENCHMARKING
# endif
# if !defined(CATCH_CONFIG_ENABLE_CHRONO_STRINGMAKER)
# define CATCH_CONFIG_ENABLE_CHRONO_STRINGMAKER
# endif
@ -53,7 +56,6 @@
#include "internal/catch_test_registry.h"
#include "internal/catch_capture.hpp"
#include "internal/catch_section.h"
#include "internal/catch_benchmark.h"
#include "internal/catch_interfaces_exception.h"
#include "internal/catch_approx.h"
#include "internal/catch_compiler_capabilities.h"
@ -75,6 +77,10 @@
#include "internal/catch_objc.hpp"
#endif
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
#include "internal/benchmark/catch_benchmark.hpp"
#endif
#ifdef CATCH_CONFIG_EXTERNAL_INTERFACES
#include "internal/catch_external_interfaces.h"
#endif
@ -89,6 +95,7 @@
#include "internal/catch_default_main.hpp"
#endif
#if !defined(CATCH_CONFIG_IMPL_ONLY)
#ifdef CLARA_CONFIG_MAIN_NOT_DEFINED
@ -188,6 +195,13 @@
#define CATCH_THEN( desc ) INTERNAL_CATCH_DYNAMIC_SECTION( " Then: " << desc )
#define CATCH_AND_THEN( desc ) INTERNAL_CATCH_DYNAMIC_SECTION( " And: " << desc )
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
#define CATCH_BENCHMARK(...) \
INTERNAL_CATCH_BENCHMARK(INTERNAL_CATCH_UNIQUE_NAME(____C_A_T_C_H____B_E_N_C_H____), INTERNAL_CATCH_GET_1_ARG(__VA_ARGS__,,), INTERNAL_CATCH_GET_2_ARG(__VA_ARGS__,,))
#define CATCH_BENCHMARK_ADVANCED(name) \
INTERNAL_CATCH_BENCHMARK_ADVANCED(INTERNAL_CATCH_UNIQUE_NAME(____C_A_T_C_H____B_E_N_C_H____), name)
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
// If CATCH_CONFIG_PREFIX_ALL is not defined then the CATCH_ prefix is not required
#else
@ -283,6 +297,13 @@
#define THEN( desc ) INTERNAL_CATCH_DYNAMIC_SECTION( " Then: " << desc )
#define AND_THEN( desc ) INTERNAL_CATCH_DYNAMIC_SECTION( " And: " << desc )
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
#define BENCHMARK(...) \
INTERNAL_CATCH_BENCHMARK(INTERNAL_CATCH_UNIQUE_NAME(____C_A_T_C_H____B_E_N_C_H____), INTERNAL_CATCH_GET_1_ARG(__VA_ARGS__,,), INTERNAL_CATCH_GET_2_ARG(__VA_ARGS__,,))
#define BENCHMARK_ADVANCED(name) \
INTERNAL_CATCH_BENCHMARK_ADVANCED(INTERNAL_CATCH_UNIQUE_NAME(____C_A_T_C_H____B_E_N_C_H____), name)
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
using Catch::Detail::Approx;
#else // CATCH_CONFIG_DISABLE

View File

@ -0,0 +1,122 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Benchmark
#ifndef TWOBLUECUBES_CATCH_BENCHMARK_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_BENCHMARK_HPP_INCLUDED
#include "../catch_config.hpp"
#include "../catch_context.h"
#include "../catch_interfaces_reporter.h"
#include "../catch_test_registry.h"
#include "catch_chronometer.hpp"
#include "catch_clock.hpp"
#include "catch_environment.hpp"
#include "catch_execution_plan.hpp"
#include "detail/catch_estimate_clock.hpp"
#include "detail/catch_complete_invoke.hpp"
#include "detail/catch_analyse.hpp"
#include "detail/catch_benchmark_function.hpp"
#include "detail/catch_run_for_at_least.hpp"
#include <algorithm>
#include <functional>
#include <string>
#include <vector>
#include <cmath>
namespace Catch {
namespace Benchmark {
struct Benchmark {
Benchmark(std::string &&name)
: name(std::move(name)) {}
template <class FUN>
Benchmark(std::string &&name, FUN &&func)
: fun(std::move(func)), name(std::move(name)) {}
template <typename Clock>
ExecutionPlan<FloatDuration<Clock>> prepare(const IConfig &cfg, Environment<FloatDuration<Clock>> env) const {
auto min_time = env.clock_resolution.mean * Detail::minimum_ticks;
auto run_time = std::max(min_time, std::chrono::duration_cast<decltype(min_time)>(Detail::warmup_time));
auto&& test = Detail::run_for_at_least<Clock>(std::chrono::duration_cast<ClockDuration<Clock>>(run_time), 1, fun);
int new_iters = static_cast<int>(std::ceil(min_time * test.iterations / test.elapsed));
return { new_iters, test.elapsed / test.iterations * new_iters * cfg.benchmarkSamples(), fun, std::chrono::duration_cast<FloatDuration<Clock>>(Detail::warmup_time), Detail::warmup_iterations };
}
template <typename Clock = default_clock>
void run() {
IConfigPtr cfg = getCurrentContext().getConfig();
auto env = Detail::measure_environment<Clock>();
getResultCapture().benchmarkPreparing(name);
CATCH_TRY{
auto plan = user_code([&] {
return prepare<Clock>(*cfg, env);
});
BenchmarkInfo info {
name,
plan.estimated_duration.count(),
plan.iterations_per_sample,
cfg->benchmarkSamples(),
cfg->benchmarkResamples(),
env.clock_resolution.mean.count(),
env.clock_cost.mean.count()
};
getResultCapture().benchmarkStarting(info);
auto samples = user_code([&] {
return plan.template run<Clock>(*cfg, env);
});
auto analysis = Detail::analyse(*cfg, env, samples.begin(), samples.end());
BenchmarkStats<std::chrono::duration<double, std::nano>> stats{ info, analysis.samples, analysis.mean, analysis.standard_deviation, analysis.outliers, analysis.outlier_variance };
getResultCapture().benchmarkEnded(stats);
} CATCH_CATCH_ALL{
if (translateActiveException() != Detail::benchmarkErrorMsg) // benchmark errors have been reported, otherwise rethrow.
std::rethrow_exception(std::current_exception());
}
}
// sets lambda to be used in fun *and* executes benchmark!
template <typename Fun,
typename std::enable_if<!Detail::is_related<Fun, Benchmark>::value, int>::type = 0>
Benchmark & operator=(Fun func) {
fun = Detail::BenchmarkFunction(func);
run();
return *this;
}
explicit operator bool() {
return true;
}
private:
Detail::BenchmarkFunction fun;
std::string name;
};
}
} // namespace Catch
#define INTERNAL_CATCH_GET_1_ARG(arg1, arg2, ...) arg1
#define INTERNAL_CATCH_GET_2_ARG(arg1, arg2, ...) arg2
#define INTERNAL_CATCH_BENCHMARK(BenchmarkName, name, benchmarkIndex)\
if( Catch::Benchmark::Benchmark BenchmarkName{name} ) \
BenchmarkName = [&](int benchmarkIndex)
#define INTERNAL_CATCH_BENCHMARK_ADVANCED(BenchmarkName, name)\
if( Catch::Benchmark::Benchmark BenchmarkName{name} ) \
BenchmarkName = [&]
#endif // TWOBLUECUBES_CATCH_BENCHMARK_HPP_INCLUDED

View File

@ -0,0 +1,71 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// User-facing chronometer
#ifndef TWOBLUECUBES_CATCH_CHRONOMETER_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_CHRONOMETER_HPP_INCLUDED
#include "catch_clock.hpp"
#include "catch_optimizer.hpp"
#include "detail/catch_complete_invoke.hpp"
#include "../catch_meta.hpp"
namespace Catch {
namespace Benchmark {
namespace Detail {
struct ChronometerConcept {
virtual void start() = 0;
virtual void finish() = 0;
virtual ~ChronometerConcept() = default;
};
template <typename Clock>
struct ChronometerModel final : public ChronometerConcept {
void start() override { started = Clock::now(); }
void finish() override { finished = Clock::now(); }
ClockDuration<Clock> elapsed() const { return finished - started; }
TimePoint<Clock> started;
TimePoint<Clock> finished;
};
} // namespace Detail
struct Chronometer {
public:
template <typename Fun>
void measure(Fun&& fun) { measure(std::forward<Fun>(fun), is_callable<Fun(int)>()); }
int runs() const { return k; }
Chronometer(Detail::ChronometerConcept& meter, int k)
: impl(&meter)
, k(k) {}
private:
template <typename Fun>
void measure(Fun&& fun, std::false_type) {
measure([&fun](int) { return fun(); }, std::true_type());
}
template <typename Fun>
void measure(Fun&& fun, std::true_type) {
Detail::optimizer_barrier();
impl->start();
for (int i = 0; i < k; ++i) invoke_deoptimized(fun, i);
impl->finish();
Detail::optimizer_barrier();
}
Detail::ChronometerConcept* impl;
int k;
};
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_CHRONOMETER_HPP_INCLUDED

View File

@ -0,0 +1,46 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Clocks
#ifndef TWOBLUECUBES_CATCH_CLOCK_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_CLOCK_HPP_INCLUDED
#include <chrono>
#include <ratio>
namespace Catch {
namespace Benchmark {
template <unsigned Num, unsigned Den = 1>
using ratio = std::ratio<Num, Den>;
using milli = ratio<1, 1000>;
using micro = ratio<1, 1000000>;
using nano = ratio<1, 1000000000>;
template <typename Clock>
using ClockDuration = typename Clock::duration;
template <typename Clock>
using FloatDuration = std::chrono::duration<double, typename Clock::period>;
template <typename Clock>
using TimePoint = typename Clock::time_point;
using default_clock = std::chrono::high_resolution_clock;
template <typename Clock>
struct now {
TimePoint<Clock> operator()() const {
return Clock::now();
}
};
using fp_seconds = std::chrono::duration<double, ratio<1>>;
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_CLOCK_HPP_INCLUDED

View File

@ -0,0 +1,73 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Constructor and destructor helpers
#ifndef TWOBLUECUBES_CATCH_CONSTRUCTOR_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_CONSTRUCTOR_HPP_INCLUDED
#include <type_traits>
namespace Catch {
namespace Detail {
template <typename T, bool Destruct>
struct ObjectStorage
{
using TStorage = typename std::aligned_storage<sizeof(T), std::alignment_of<T>::value>::type;
ObjectStorage() : data() {}
ObjectStorage(const ObjectStorage& other)
{
new(&data) T(other.stored_object());
}
ObjectStorage(ObjectStorage&& other)
{
new(&data) T(std::move(other.stored_object()));
}
~ObjectStorage() { destruct_on_exit<T>(); }
template <typename... Args>
void construct(Args&&... args)
{
new (&data) T(std::forward<Args>(args)...);
}
template <bool AllowManualDestruction = !Destruct>
typename std::enable_if<AllowManualDestruction>::type destruct()
{
stored_object().~T();
}
private:
// If this is a constructor benchmark, destruct the underlying object
template <typename U>
void destruct_on_exit(typename std::enable_if<Destruct, U>::type* = 0) { destruct<true>(); }
// Otherwise, don't
template <typename U>
void destruct_on_exit(typename std::enable_if<!Destruct, U>::type* = 0) { }
T& stored_object()
{
return *static_cast<T*>(static_cast<void*>(&data));
}
TStorage data;
};
}
template <typename T>
using storage_for = Detail::ObjectStorage<T, true>;
template <typename T>
using destructable_object = Detail::ObjectStorage<T, false>;
}
#endif // TWOBLUECUBES_CATCH_CONSTRUCTOR_HPP_INCLUDED

View File

@ -0,0 +1,38 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Environment information
#ifndef TWOBLUECUBES_CATCH_ENVIRONMENT_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_ENVIRONMENT_HPP_INCLUDED
#include "catch_clock.hpp"
#include "catch_outlier_classification.hpp"
namespace Catch {
namespace Benchmark {
template <typename Duration>
struct EnvironmentEstimate {
Duration mean;
OutlierClassification outliers;
template <typename Duration2>
operator EnvironmentEstimate<Duration2>() const {
return { mean, outliers };
}
};
template <typename Clock>
struct Environment {
using clock_type = Clock;
EnvironmentEstimate<FloatDuration<Clock>> clock_resolution;
EnvironmentEstimate<FloatDuration<Clock>> clock_cost;
};
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_ENVIRONMENT_HPP_INCLUDED

View File

@ -0,0 +1,31 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Statistics estimates
#ifndef TWOBLUECUBES_CATCH_ESTIMATE_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_ESTIMATE_HPP_INCLUDED
namespace Catch {
namespace Benchmark {
template <typename Duration>
struct Estimate {
Duration point;
Duration lower_bound;
Duration upper_bound;
double confidence_interval;
template <typename Duration2>
operator Estimate<Duration2>() const {
return { point, lower_bound, upper_bound, confidence_interval };
}
};
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_ESTIMATE_HPP_INCLUDED

View File

@ -0,0 +1,58 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Execution plan
#ifndef TWOBLUECUBES_CATCH_EXECUTION_PLAN_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_EXECUTION_PLAN_HPP_INCLUDED
#include "../catch_config.hpp"
#include "catch_clock.hpp"
#include "catch_environment.hpp"
#include "detail/catch_benchmark_function.hpp"
#include "detail/catch_repeat.hpp"
#include "detail/catch_run_for_at_least.hpp"
#include <algorithm>
namespace Catch {
namespace Benchmark {
template <typename Duration>
struct ExecutionPlan {
int iterations_per_sample;
Duration estimated_duration;
Detail::BenchmarkFunction benchmark;
Duration warmup_time;
int warmup_iterations;
template <typename Duration2>
operator ExecutionPlan<Duration2>() const {
return { iterations_per_sample, estimated_duration, benchmark, warmup_time, warmup_iterations };
}
template <typename Clock>
std::vector<FloatDuration<Clock>> run(const IConfig &cfg, Environment<FloatDuration<Clock>> env) const {
// warmup a bit
Detail::run_for_at_least<Clock>(std::chrono::duration_cast<ClockDuration<Clock>>(warmup_time), warmup_iterations, Detail::repeat(now<Clock>{}));
std::vector<FloatDuration<Clock>> times;
times.reserve(cfg.benchmarkSamples());
std::generate_n(std::back_inserter(times), cfg.benchmarkSamples(), [this, env] {
Detail::ChronometerModel<Clock> model;
this->benchmark(Chronometer(model, iterations_per_sample));
auto sample_time = model.elapsed() - env.clock_cost.mean;
if (sample_time < FloatDuration<Clock>::zero()) sample_time = FloatDuration<Clock>::zero();
return sample_time / iterations_per_sample;
});
return times;
}
};
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_EXECUTION_PLAN_HPP_INCLUDED

View File

@ -0,0 +1,68 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Hinting the optimizer
#ifndef TWOBLUECUBES_CATCH_OPTIMIZER_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_OPTIMIZER_HPP_INCLUDED
#if defined(_MSC_VER)
# include <atomic> // atomic_thread_fence
#endif
namespace Catch {
namespace Benchmark {
#if defined(__GNUC__) || defined(__clang__)
template <typename T>
inline void keep_memory(T* p) {
asm volatile("" : : "g"(p) : "memory");
}
inline void keep_memory() {
asm volatile("" : : : "memory");
}
namespace Detail {
inline void optimizer_barrier() { keep_memory(); }
} // namespace Detail
#elif defined(_MSC_VER)
#pragma optimize("", off)
template <typename T>
inline void keep_memory(T* p) {
// thanks @milleniumbug
*reinterpret_cast<char volatile*>(p) = *reinterpret_cast<char const volatile*>(p);
}
// TODO equivalent keep_memory()
#pragma optimize("", on)
namespace Detail {
inline void optimizer_barrier() {
std::atomic_thread_fence(std::memory_order_seq_cst);
}
} // namespace Detail
#endif
template <typename T>
inline void deoptimize_value(T&& x) {
keep_memory(&x);
}
template <typename Fn, typename... Args>
inline auto invoke_deoptimized(Fn&& fn, Args&&... args) -> typename std::enable_if<!std::is_same<void, decltype(fn(args...))>::value>::type {
deoptimize_value(std::forward<Fn>(fn) (std::forward<Args...>(args...)));
}
template <typename Fn, typename... Args>
inline auto invoke_deoptimized(Fn&& fn, Args&&... args) -> typename std::enable_if<std::is_same<void, decltype(fn(args...))>::value>::type {
std::forward<Fn>(fn) (std::forward<Args...>(args...));
}
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_OPTIMIZER_HPP_INCLUDED

View File

@ -0,0 +1,29 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Outlier information
#ifndef TWOBLUECUBES_CATCH_OUTLIERS_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_OUTLIERS_HPP_INCLUDED
namespace Catch {
namespace Benchmark {
struct OutlierClassification {
int samples_seen = 0;
int low_severe = 0; // more than 3 times IQR below Q1
int low_mild = 0; // 1.5 to 3 times IQR below Q1
int high_mild = 0; // 1.5 to 3 times IQR above Q3
int high_severe = 0; // more than 3 times IQR above Q3
int total() const {
return low_severe + low_mild + high_mild + high_severe;
}
};
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_OUTLIERS_HPP_INCLUDED

View File

@ -0,0 +1,50 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Benchmark results
#ifndef TWOBLUECUBES_CATCH_BENCHMARK_RESULTS_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_BENCHMARK_RESULTS_HPP_INCLUDED
#include "catch_clock.hpp"
#include "catch_estimate.hpp"
#include "catch_outlier_classification.hpp"
#include <algorithm>
#include <vector>
#include <string>
#include <iterator>
namespace Catch {
namespace Benchmark {
template <typename Duration>
struct SampleAnalysis {
std::vector<Duration> samples;
Estimate<Duration> mean;
Estimate<Duration> standard_deviation;
OutlierClassification outliers;
double outlier_variance;
template <typename Duration2>
operator SampleAnalysis<Duration2>() const {
std::vector<Duration2> samples2;
samples2.reserve(samples.size());
std::transform(samples.begin(), samples.end(), std::back_inserter(samples2), [](Duration d) { return Duration2(d); });
return {
std::move(samples2),
mean,
standard_deviation,
outliers,
outlier_variance,
};
}
};
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_BENCHMARK_RESULTS_HPP_INCLUDED

View File

@ -0,0 +1,78 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Run and analyse one benchmark
#ifndef TWOBLUECUBES_CATCH_DETAIL_ANALYSE_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_DETAIL_ANALYSE_HPP_INCLUDED
#include "../catch_clock.hpp"
#include "../catch_sample_analysis.hpp"
#include "catch_stats.hpp"
#include <algorithm>
#include <iterator>
#include <vector>
namespace Catch {
namespace Benchmark {
namespace Detail {
template <typename Duration, typename Iterator>
SampleAnalysis<Duration> analyse(const IConfig &cfg, Environment<Duration>, Iterator first, Iterator last) {
if (!cfg.benchmarkNoAnalysis()) {
std::vector<double> samples;
samples.reserve(last - first);
std::transform(first, last, std::back_inserter(samples), [](Duration d) { return d.count(); });
auto analysis = Catch::Benchmark::Detail::analyse_samples(cfg.benchmarkConfidenceInterval(), cfg.benchmarkResamples(), samples.begin(), samples.end());
auto outliers = Catch::Benchmark::Detail::classify_outliers(samples.begin(), samples.end());
auto wrap_estimate = [](Estimate<double> e) {
return Estimate<Duration> {
Duration(e.point),
Duration(e.lower_bound),
Duration(e.upper_bound),
e.confidence_interval,
};
};
std::vector<Duration> samples2;
samples2.reserve(samples.size());
std::transform(samples.begin(), samples.end(), std::back_inserter(samples2), [](double d) { return Duration(d); });
return {
std::move(samples2),
wrap_estimate(analysis.mean),
wrap_estimate(analysis.standard_deviation),
outliers,
analysis.outlier_variance,
};
} else {
std::vector<Duration> samples;
samples.reserve(last - first);
Duration mean = Duration(0);
int i = 0;
for (auto it = first; it < last; ++it, ++i) {
samples.push_back(Duration(*it));
mean += Duration(*it);
}
mean /= i;
return {
std::move(samples),
Estimate<Duration>{mean, mean, mean, 0.0},
Estimate<Duration>{Duration(0), Duration(0), Duration(0), 0.0},
OutlierClassification{},
0.0
};
}
}
} // namespace Detail
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_DETAIL_ANALYSE_HPP_INCLUDED

View File

@ -0,0 +1,105 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Dumb std::function implementation for consistent call overhead
#ifndef TWOBLUECUBES_CATCH_DETAIL_BENCHMARK_FUNCTION_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_DETAIL_BENCHMARK_FUNCTION_HPP_INCLUDED
#include "../catch_chronometer.hpp"
#include "catch_complete_invoke.hpp"
#include "../../catch_meta.hpp"
#include <cassert>
#include <type_traits>
#include <utility>
#include <memory>
namespace Catch {
namespace Benchmark {
namespace Detail {
template <typename T>
using Decay = typename std::decay<T>::type;
template <typename T, typename U>
struct is_related
: std::is_same<Decay<T>, Decay<U>> {};
/// We need to reinvent std::function because every piece of code that might add overhead
/// in a measurement context needs to have consistent performance characteristics so that we
/// can account for it in the measurement.
/// Implementations of std::function with optimizations that aren't always applicable, like
/// small buffer optimizations, are not uncommon.
/// This is effectively an implementation of std::function without any such optimizations;
/// it may be slow, but it is consistently slow.
struct BenchmarkFunction {
private:
struct callable {
virtual void call(Chronometer meter) const = 0;
virtual callable* clone() const = 0;
virtual ~callable() = default;
};
template <typename Fun>
struct model : public callable {
model(Fun&& fun) : fun(std::move(fun)) {}
model(Fun const& fun) : fun(fun) {}
model<Fun>* clone() const override { return new model<Fun>(*this); }
void call(Chronometer meter) const override {
call(meter, is_callable<Fun(Chronometer)>());
}
void call(Chronometer meter, std::true_type) const {
fun(meter);
}
void call(Chronometer meter, std::false_type) const {
meter.measure(fun);
}
Fun fun;
};
struct do_nothing { void operator()() const {} };
template <typename T>
BenchmarkFunction(model<T>* c) : f(c) {}
public:
BenchmarkFunction()
: f(new model<do_nothing>{ {} }) {}
template <typename Fun,
typename std::enable_if<!is_related<Fun, BenchmarkFunction>::value, int>::type = 0>
BenchmarkFunction(Fun&& fun)
: f(new model<typename std::decay<Fun>::type>(std::forward<Fun>(fun))) {}
BenchmarkFunction(BenchmarkFunction&& that)
: f(std::move(that.f)) {}
BenchmarkFunction(BenchmarkFunction const& that)
: f(that.f->clone()) {}
BenchmarkFunction& operator=(BenchmarkFunction&& that) {
f = std::move(that.f);
return *this;
}
BenchmarkFunction& operator=(BenchmarkFunction const& that) {
f.reset(that.f->clone());
return *this;
}
void operator()(Chronometer meter) const { f->call(meter); }
private:
std::unique_ptr<callable> f;
};
} // namespace Detail
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_DETAIL_BENCHMARK_FUNCTION_HPP_INCLUDED

View File

@ -0,0 +1,69 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Invoke with a special case for void
#ifndef TWOBLUECUBES_CATCH_DETAIL_COMPLETE_INVOKE_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_DETAIL_COMPLETE_INVOKE_HPP_INCLUDED
#include "../../catch_enforce.h"
#include <type_traits>
#include <utility>
namespace Catch {
namespace Benchmark {
namespace Detail {
template <typename T>
struct CompleteType { using type = T; };
template <>
struct CompleteType<void> { struct type {}; };
template <typename T>
using CompleteType_t = typename CompleteType<T>::type;
template <typename Result>
struct CompleteInvoker {
template <typename Fun, typename... Args>
static Result invoke(Fun&& fun, Args&&... args) {
return std::forward<Fun>(fun)(std::forward<Args>(args)...);
}
};
template <>
struct CompleteInvoker<void> {
template <typename Fun, typename... Args>
static CompleteType_t<void> invoke(Fun&& fun, Args&&... args) {
std::forward<Fun>(fun)(std::forward<Args>(args)...);
return {};
}
};
template <typename Sig>
using ResultOf_t = typename std::result_of<Sig>::type;
// invoke and not return void :(
template <typename Fun, typename... Args>
CompleteType_t<ResultOf_t<Fun(Args...)>> complete_invoke(Fun&& fun, Args&&... args) {
return CompleteInvoker<ResultOf_t<Fun(Args...)>>::invoke(std::forward<Fun>(fun), std::forward<Args>(args)...);
}
const std::string benchmarkErrorMsg = "a benchmark failed to run successfully";
} // namespace Detail
template <typename Fun>
Detail::CompleteType_t<Detail::ResultOf_t<Fun()>> user_code(Fun&& fun) {
CATCH_TRY{
return Detail::complete_invoke(std::forward<Fun>(fun));
} CATCH_CATCH_ALL{
getResultCapture().benchmarkFailed(translateActiveException());
CATCH_RUNTIME_ERROR(Detail::benchmarkErrorMsg);
}
}
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_DETAIL_COMPLETE_INVOKE_HPP_INCLUDED

View File

@ -0,0 +1,113 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Environment measurement
#ifndef TWOBLUECUBES_CATCH_DETAIL_ESTIMATE_CLOCK_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_DETAIL_ESTIMATE_CLOCK_HPP_INCLUDED
#include "../catch_clock.hpp"
#include "../catch_environment.hpp"
#include "catch_stats.hpp"
#include "catch_measure.hpp"
#include "catch_run_for_at_least.hpp"
#include "../catch_clock.hpp"
#include <algorithm>
#include <iterator>
#include <tuple>
#include <vector>
#include <cmath>
namespace Catch {
namespace Benchmark {
namespace Detail {
template <typename Clock>
std::vector<double> resolution(int k) {
std::vector<TimePoint<Clock>> times;
times.reserve(k + 1);
std::generate_n(std::back_inserter(times), k + 1, now<Clock>{});
std::vector<double> deltas;
deltas.reserve(k);
std::transform(std::next(times.begin()), times.end(), times.begin(),
std::back_inserter(deltas),
[](TimePoint<Clock> a, TimePoint<Clock> b) { return static_cast<double>((a - b).count()); });
return deltas;
}
const auto warmup_iterations = 10000;
const auto warmup_time = std::chrono::milliseconds(100);
const auto minimum_ticks = 1000;
const auto warmup_seed = 10000;
const auto clock_resolution_estimation_time = std::chrono::milliseconds(500);
const auto clock_cost_estimation_time_limit = std::chrono::seconds(1);
const auto clock_cost_estimation_tick_limit = 100000;
const auto clock_cost_estimation_time = std::chrono::milliseconds(10);
const auto clock_cost_estimation_iterations = 10000;
template <typename Clock>
int warmup() {
return run_for_at_least<Clock>(std::chrono::duration_cast<ClockDuration<Clock>>(warmup_time), warmup_seed, &resolution<Clock>)
.iterations;
}
template <typename Clock>
EnvironmentEstimate<FloatDuration<Clock>> estimate_clock_resolution(int iterations) {
auto r = run_for_at_least<Clock>(std::chrono::duration_cast<ClockDuration<Clock>>(clock_resolution_estimation_time), iterations, &resolution<Clock>)
.result;
return {
FloatDuration<Clock>(mean(r.begin(), r.end())),
classify_outliers(r.begin(), r.end()),
};
}
template <typename Clock>
EnvironmentEstimate<FloatDuration<Clock>> estimate_clock_cost(FloatDuration<Clock> resolution) {
auto time_limit = std::min(resolution * clock_cost_estimation_tick_limit, FloatDuration<Clock>(clock_cost_estimation_time_limit));
auto time_clock = [](int k) {
return Detail::measure<Clock>([k] {
for (int i = 0; i < k; ++i) {
volatile auto ignored = Clock::now();
(void)ignored;
}
}).elapsed;
};
time_clock(1);
int iters = clock_cost_estimation_iterations;
auto&& r = run_for_at_least<Clock>(std::chrono::duration_cast<ClockDuration<Clock>>(clock_cost_estimation_time), iters, time_clock);
std::vector<double> times;
int nsamples = static_cast<int>(std::ceil(time_limit / r.elapsed));
times.reserve(nsamples);
std::generate_n(std::back_inserter(times), nsamples, [time_clock, &r] {
return static_cast<double>((time_clock(r.iterations) / r.iterations).count());
});
return {
FloatDuration<Clock>(mean(times.begin(), times.end())),
classify_outliers(times.begin(), times.end()),
};
}
template <typename Clock>
Environment<FloatDuration<Clock>> measure_environment() {
static Environment<FloatDuration<Clock>>* env = nullptr;
if (env) {
return *env;
}
auto iters = Detail::warmup<Clock>();
auto resolution = Detail::estimate_clock_resolution<Clock>(iters);
auto cost = Detail::estimate_clock_cost<Clock>(resolution.mean);
env = new Environment<FloatDuration<Clock>>{ resolution, cost };
return *env;
}
} // namespace Detail
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_DETAIL_ESTIMATE_CLOCK_HPP_INCLUDED

View File

@ -0,0 +1,35 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Measure
#ifndef TWOBLUECUBES_CATCH_DETAIL_MEASURE_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_DETAIL_MEASURE_HPP_INCLUDED
#include "../catch_clock.hpp"
#include "catch_complete_invoke.hpp"
#include "catch_timing.hpp"
#include <utility>
namespace Catch {
namespace Benchmark {
namespace Detail {
template <typename Clock, typename Fun, typename... Args>
TimingOf<Clock, Fun(Args...)> measure(Fun&& fun, Args&&... args) {
auto start = Clock::now();
auto&& r = Detail::complete_invoke(fun, std::forward<Args>(args)...);
auto end = Clock::now();
auto delta = end - start;
return { delta, std::forward<decltype(r)>(r), 1 };
}
} // namespace Detail
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_DETAIL_MEASURE_HPP_INCLUDED

View File

@ -0,0 +1,37 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// repeat algorithm
#ifndef TWOBLUECUBES_CATCH_DETAIL_REPEAT_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_DETAIL_REPEAT_HPP_INCLUDED
#include <type_traits>
#include <utility>
namespace Catch {
namespace Benchmark {
namespace Detail {
template <typename Fun>
struct repeater {
void operator()(int k) const {
for (int i = 0; i < k; ++i) {
fun();
}
}
Fun fun;
};
template <typename Fun>
repeater<typename std::decay<Fun>::type> repeat(Fun&& fun) {
return { std::forward<Fun>(fun) };
}
} // namespace Detail
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_DETAIL_REPEAT_HPP_INCLUDED

View File

@ -0,0 +1,65 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Run a function for a minimum amount of time
#ifndef TWOBLUECUBES_CATCH_RUN_FOR_AT_LEAST_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_RUN_FOR_AT_LEAST_HPP_INCLUDED
#include "../catch_clock.hpp"
#include "../catch_chronometer.hpp"
#include "catch_measure.hpp"
#include "catch_complete_invoke.hpp"
#include "catch_timing.hpp"
#include "../../catch_meta.hpp"
#include <utility>
#include <type_traits>
namespace Catch {
namespace Benchmark {
namespace Detail {
template <typename Clock, typename Fun>
TimingOf<Clock, Fun(int)> measure_one(Fun&& fun, int iters, std::false_type) {
return Detail::measure<Clock>(fun, iters);
}
template <typename Clock, typename Fun>
TimingOf<Clock, Fun(Chronometer)> measure_one(Fun&& fun, int iters, std::true_type) {
Detail::ChronometerModel<Clock> meter;
auto&& result = Detail::complete_invoke(fun, Chronometer(meter, iters));
return { meter.elapsed(), std::move(result), iters };
}
template <typename Clock, typename Fun>
using run_for_at_least_argument_t = typename std::conditional<is_callable<Fun(Chronometer)>::value, Chronometer, int>::type;
struct optimized_away_error : std::exception {
const char* what() const noexcept override {
return "could not measure benchmark, maybe it was optimized away";
}
};
template <typename Clock, typename Fun>
TimingOf<Clock, Fun(run_for_at_least_argument_t<Clock, Fun>)> run_for_at_least(ClockDuration<Clock> how_long, int seed, Fun&& fun) {
auto iters = seed;
while (iters < (1 << 30)) {
auto&& Timing = measure_one<Clock>(fun, iters, is_callable<Fun(Chronometer)>());
if (Timing.elapsed >= how_long) {
return { Timing.elapsed, std::move(Timing.result), iters };
}
iters *= 2;
}
throw optimized_away_error{};
}
} // namespace Detail
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_RUN_FOR_AT_LEAST_HPP_INCLUDED

View File

@ -0,0 +1,342 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Statistical analysis tools
#ifndef TWOBLUECUBES_CATCH_DETAIL_ANALYSIS_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_DETAIL_ANALYSIS_HPP_INCLUDED
#include "../catch_clock.hpp"
#include "../catch_estimate.hpp"
#include "../catch_outlier_classification.hpp"
#include <algorithm>
#include <cassert>
#include <functional>
#include <iterator>
#include <vector>
#include <array>
#include <random>
#include <numeric>
#include <tuple>
#include <cmath>
#include <utility>
#include <cstddef>
#ifdef CATCH_USE_ASYNC
#include <future>
#endif
namespace Catch {
namespace Benchmark {
namespace Detail {
using sample = std::vector<double>;
template <typename Iterator>
double weighted_average_quantile(int k, int q, Iterator first, Iterator last) {
auto count = last - first;
double idx = (count - 1) * k / static_cast<double>(q);
int j = static_cast<int>(idx);
double g = idx - j;
std::nth_element(first, first + j, last);
auto xj = first[j];
if (g == 0) return xj;
auto xj1 = *std::min_element(first + (j + 1), last);
return xj + g * (xj1 - xj);
}
template <typename Iterator>
OutlierClassification classify_outliers(Iterator first, Iterator last) {
std::vector<double> copy(first, last);
auto q1 = weighted_average_quantile(1, 4, copy.begin(), copy.end());
auto q3 = weighted_average_quantile(3, 4, copy.begin(), copy.end());
auto iqr = q3 - q1;
auto los = q1 - (iqr * 3.);
auto lom = q1 - (iqr * 1.5);
auto him = q3 + (iqr * 1.5);
auto his = q3 + (iqr * 3.);
OutlierClassification o;
for (; first != last; ++first) {
auto&& t = *first;
if (t < los) ++o.low_severe;
else if (t < lom) ++o.low_mild;
else if (t > his) ++o.high_severe;
else if (t > him) ++o.high_mild;
++o.samples_seen;
}
return o;
}
template <typename Iterator>
double mean(Iterator first, Iterator last) {
auto count = last - first;
double sum = std::accumulate(first, last, 0.);
return sum / count;
}
template <typename Iterator>
double standard_deviation(Iterator first, Iterator last) {
auto m = mean(first, last);
double variance = std::accumulate(first, last, 0., [m](double a, double b) {
double diff = b - m;
return a + diff * diff;
}) / (last - first);
return std::sqrt(variance);
}
template <typename URng, typename Iterator, typename Estimator>
sample resample(URng& rng, int resamples, Iterator first, Iterator last, Estimator& estimator) {
auto n = last - first;
std::uniform_int_distribution<decltype(n)> dist(0, n - 1);
sample out;
out.reserve(resamples);
std::generate_n(std::back_inserter(out), resamples, [n, first, &estimator, &dist, &rng] {
std::vector<double> resampled;
resampled.reserve(n);
std::generate_n(std::back_inserter(resampled), n, [first, &dist, &rng] { return first[dist(rng)]; });
return estimator(resampled.begin(), resampled.end());
});
std::sort(out.begin(), out.end());
return out;
}
template <typename Estimator, typename Iterator>
sample jackknife(Estimator&& estimator, Iterator first, Iterator last) {
auto n = last - first;
auto second = std::next(first);
sample results;
results.reserve(n);
for (auto it = first; it != last; ++it) {
std::iter_swap(it, first);
results.push_back(estimator(second, last));
}
return results;
}
inline double normal_cdf(double x) {
return std::erfc(-x / std::sqrt(2.0)) / 2.0;
}
inline double erf_inv(double x) {
// Code accompanying the article "Approximating the erfinv function" in GPU Computing Gems, Volume 2
double w, p;
w = -log((1.0 - x)*(1.0 + x));
if (w < 6.250000) {
w = w - 3.125000;
p = -3.6444120640178196996e-21;
p = -1.685059138182016589e-19 + p * w;
p = 1.2858480715256400167e-18 + p * w;
p = 1.115787767802518096e-17 + p * w;
p = -1.333171662854620906e-16 + p * w;
p = 2.0972767875968561637e-17 + p * w;
p = 6.6376381343583238325e-15 + p * w;
p = -4.0545662729752068639e-14 + p * w;
p = -8.1519341976054721522e-14 + p * w;
p = 2.6335093153082322977e-12 + p * w;
p = -1.2975133253453532498e-11 + p * w;
p = -5.4154120542946279317e-11 + p * w;
p = 1.051212273321532285e-09 + p * w;
p = -4.1126339803469836976e-09 + p * w;
p = -2.9070369957882005086e-08 + p * w;
p = 4.2347877827932403518e-07 + p * w;
p = -1.3654692000834678645e-06 + p * w;
p = -1.3882523362786468719e-05 + p * w;
p = 0.0001867342080340571352 + p * w;
p = -0.00074070253416626697512 + p * w;
p = -0.0060336708714301490533 + p * w;
p = 0.24015818242558961693 + p * w;
p = 1.6536545626831027356 + p * w;
} else if (w < 16.000000) {
w = sqrt(w) - 3.250000;
p = 2.2137376921775787049e-09;
p = 9.0756561938885390979e-08 + p * w;
p = -2.7517406297064545428e-07 + p * w;
p = 1.8239629214389227755e-08 + p * w;
p = 1.5027403968909827627e-06 + p * w;
p = -4.013867526981545969e-06 + p * w;
p = 2.9234449089955446044e-06 + p * w;
p = 1.2475304481671778723e-05 + p * w;
p = -4.7318229009055733981e-05 + p * w;
p = 6.8284851459573175448e-05 + p * w;
p = 2.4031110387097893999e-05 + p * w;
p = -0.0003550375203628474796 + p * w;
p = 0.00095328937973738049703 + p * w;
p = -0.0016882755560235047313 + p * w;
p = 0.0024914420961078508066 + p * w;
p = -0.0037512085075692412107 + p * w;
p = 0.005370914553590063617 + p * w;
p = 1.0052589676941592334 + p * w;
p = 3.0838856104922207635 + p * w;
} else {
w = sqrt(w) - 5.000000;
p = -2.7109920616438573243e-11;
p = -2.5556418169965252055e-10 + p * w;
p = 1.5076572693500548083e-09 + p * w;
p = -3.7894654401267369937e-09 + p * w;
p = 7.6157012080783393804e-09 + p * w;
p = -1.4960026627149240478e-08 + p * w;
p = 2.9147953450901080826e-08 + p * w;
p = -6.7711997758452339498e-08 + p * w;
p = 2.2900482228026654717e-07 + p * w;
p = -9.9298272942317002539e-07 + p * w;
p = 4.5260625972231537039e-06 + p * w;
p = -1.9681778105531670567e-05 + p * w;
p = 7.5995277030017761139e-05 + p * w;
p = -0.00021503011930044477347 + p * w;
p = -0.00013871931833623122026 + p * w;
p = 1.0103004648645343977 + p * w;
p = 4.8499064014085844221 + p * w;
}
return p * x;
}
inline double erfc_inv(double x) {
return erf_inv(1.0 - x);
}
inline double normal_quantile(double p) {
static const double ROOT_TWO = std::sqrt(2.0);
double result = 0.0;
assert(p >= 0 && p <= 1);
if (p < 0 || p > 1) {
return result;
}
result = -erfc_inv(2.0 * p);
// result *= normal distribution standard deviation (1.0) * sqrt(2)
result *= /*sd * */ ROOT_TWO;
// result += normal disttribution mean (0)
return result;
}
template <typename Iterator, typename Estimator>
Estimate<double> bootstrap(double confidence_level, Iterator first, Iterator last, sample const& resample, Estimator&& estimator) {
auto n_samples = last - first;
double point = estimator(first, last);
// Degenerate case with a single sample
if (n_samples == 1) return { point, point, point, confidence_level };
sample jack = jackknife(estimator, first, last);
double jack_mean = mean(jack.begin(), jack.end());
double sum_squares, sum_cubes;
std::tie(sum_squares, sum_cubes) = std::accumulate(jack.begin(), jack.end(), std::make_pair(0., 0.), [jack_mean](std::pair<double, double> sqcb, double x) -> std::pair<double, double> {
auto d = jack_mean - x;
auto d2 = d * d;
auto d3 = d2 * d;
return { sqcb.first + d2, sqcb.second + d3 };
});
double accel = sum_cubes / (6 * std::pow(sum_squares, 1.5));
int n = static_cast<int>(resample.size());
double prob_n = std::count_if(resample.begin(), resample.end(), [point](double x) { return x < point; }) / (double)n;
// degenerate case with uniform samples
if (prob_n == 0) return { point, point, point, confidence_level };
double bias = normal_quantile(prob_n);
double z1 = normal_quantile((1. - confidence_level) / 2.);
auto cumn = [n](double x) -> int {
return std::lround(normal_cdf(x) * n); };
auto a = [bias, accel](double b) { return bias + b / (1. - accel * b); };
double b1 = bias + z1;
double b2 = bias - z1;
double a1 = a(b1);
double a2 = a(b2);
auto lo = std::max(cumn(a1), 0);
auto hi = std::min(cumn(a2), n - 1);
return { point, resample[lo], resample[hi], confidence_level };
}
inline double outlier_variance(Estimate<double> mean, Estimate<double> stddev, int n) {
double sb = stddev.point;
double mn = mean.point / n;
double mg_min = mn / 2.;
double sg = std::min(mg_min / 4., sb / std::sqrt(n));
double sg2 = sg * sg;
double sb2 = sb * sb;
auto c_max = [n, mn, sb2, sg2](double x) -> double {
double k = mn - x;
double d = k * k;
double nd = n * d;
double k0 = -n * nd;
double k1 = sb2 - n * sg2 + nd;
double det = k1 * k1 - 4 * sg2 * k0;
return (int)(-2. * k0 / (k1 + std::sqrt(det)));
};
auto var_out = [n, sb2, sg2](double c) {
double nc = n - c;
return (nc / n) * (sb2 - nc * sg2);
};
return std::min(var_out(1), var_out(std::min(c_max(0.), c_max(mg_min)))) / sb2;
}
struct bootstrap_analysis {
Estimate<double> mean;
Estimate<double> standard_deviation;
double outlier_variance;
};
template <typename Iterator>
bootstrap_analysis analyse_samples(double confidence_level, int n_resamples, Iterator first, Iterator last) {
static std::random_device entropy;
auto n = static_cast<int>(last - first); // seriously, one can't use integral types without hell in C++
auto mean = &Detail::mean<Iterator>;
auto stddev = &Detail::standard_deviation<Iterator>;
#ifdef CATCH_USE_ASYNC
auto Estimate = [=](double(*f)(Iterator, Iterator)) {
auto seed = entropy();
return std::async(std::launch::async, [=] {
std::mt19937 rng(seed);
auto resampled = resample(rng, n_resamples, first, last, f);
return bootstrap(confidence_level, first, last, resampled, f);
});
};
auto mean_future = Estimate(mean);
auto stddev_future = Estimate(stddev);
auto mean_estimate = mean_future.get();
auto stddev_estimate = stddev_future.get();
#else
auto Estimate = [=](double(*f)(Iterator, Iterator)) {
auto seed = entropy();
std::mt19937 rng(seed);
auto resampled = resample(rng, n_resamples, first, last, f);
return bootstrap(confidence_level, first, last, resampled, f);
};
auto mean_estimate = Estimate(mean);
auto stddev_estimate = Estimate(stddev);
#endif // CATCH_USE_ASYNC
double outlier_variance = Detail::outlier_variance(mean_estimate, stddev_estimate, n);
return { mean_estimate, stddev_estimate, outlier_variance };
}
} // namespace Detail
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_DETAIL_ANALYSIS_HPP_INCLUDED

View File

@ -0,0 +1,33 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
// Timing
#ifndef TWOBLUECUBES_CATCH_DETAIL_TIMING_HPP_INCLUDED
#define TWOBLUECUBES_CATCH_DETAIL_TIMING_HPP_INCLUDED
#include "../catch_clock.hpp"
#include "catch_complete_invoke.hpp"
#include <tuple>
#include <type_traits>
namespace Catch {
namespace Benchmark {
template <typename Duration, typename Result>
struct Timing {
Duration elapsed;
Result result;
int iterations;
};
template <typename Clock, typename Sig>
using TimingOf = Timing<ClockDuration<Clock>, Detail::CompleteType_t<Detail::ResultOf_t<Sig>>>;
} // namespace Benchmark
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_DETAIL_TIMING_HPP_INCLUDED

View File

@ -1,36 +0,0 @@
/*
* Created by Phil on 04/07/2017.
* Copyright 2017 Two Blue Cubes Ltd. All rights reserved.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
#include "catch_benchmark.h"
#include "catch_capture.hpp"
#include "catch_interfaces_reporter.h"
#include "catch_context.h"
namespace Catch {
auto BenchmarkLooper::getResolution() -> uint64_t {
return getEstimatedClockResolution() * getCurrentContext().getConfig()->benchmarkResolutionMultiple();
}
void BenchmarkLooper::reportStart() {
getResultCapture().benchmarkStarting( { m_name } );
}
auto BenchmarkLooper::needsMoreIterations() -> bool {
auto elapsed = m_timer.getElapsedNanoseconds();
// Exponentially increasing iterations until we're confident in our timer resolution
if( elapsed < m_resolution ) {
m_iterationsToRun *= 10;
return true;
}
getResultCapture().benchmarkEnded( { { m_name }, m_count, elapsed } );
return false;
}
} // end namespace Catch

View File

@ -1,57 +0,0 @@
/*
* Created by Phil on 04/07/2017.
* Copyright 2017 Two Blue Cubes Ltd. All rights reserved.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
#ifndef TWOBLUECUBES_CATCH_BENCHMARK_H_INCLUDED
#define TWOBLUECUBES_CATCH_BENCHMARK_H_INCLUDED
#include "catch_stringref.h"
#include "catch_timer.h"
#include <cstdint>
#include <string>
namespace Catch {
class BenchmarkLooper {
std::string m_name;
std::size_t m_count = 0;
std::size_t m_iterationsToRun = 1;
uint64_t m_resolution;
Timer m_timer;
static auto getResolution() -> uint64_t;
public:
// Keep most of this inline as it's on the code path that is being timed
BenchmarkLooper( StringRef name )
: m_name( name ),
m_resolution( getResolution() )
{
reportStart();
m_timer.start();
}
explicit operator bool() {
if( m_count < m_iterationsToRun )
return true;
return needsMoreIterations();
}
void increment() {
++m_count;
}
void reportStart();
auto needsMoreIterations() -> bool;
};
} // end namespace Catch
#define BENCHMARK( name ) \
for( Catch::BenchmarkLooper looper( name ); looper; looper.increment() )
#endif // TWOBLUECUBES_CATCH_BENCHMARK_H_INCLUDED

View File

@ -196,10 +196,20 @@ namespace Catch {
| Opt( setWaitForKeypress, "start|exit|both" )
["--wait-for-keypress"]
( "waits for a keypress before exiting" )
| Opt( config.benchmarkResolutionMultiple, "multiplier" )
["--benchmark-resolution-multiple"]
( "multiple of clock resolution to run benchmarks" )
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
| Opt( config.benchmarkSamples, "samples" )
["--benchmark-samples"]
( "number of samples to collect (default: 100)" )
| Opt( config.benchmarkResamples, "resamples" )
["--benchmark-resamples"]
( "number of resamples for the bootstrap (default: 100000)" )
| Opt( config.benchmarkConfidenceInterval, "confidence interval" )
["--benchmark-confidence-interval"]
( "confidence interval for the bootstrap (between 0 and 1, default: 0.95)" )
| Opt( config.benchmarkNoAnalysis )
["--benchmark-no-analysis"]
( "perform only measurements; do not perform any analysis" )
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
| Arg( config.testsOrTags, "test name|pattern|tags" )
( "which test or tests to use" );

View File

@ -148,7 +148,11 @@
# if !defined(_MSVC_TRADITIONAL) || (defined(_MSVC_TRADITIONAL) && _MSVC_TRADITIONAL)
# define CATCH_INTERNAL_CONFIG_TRADITIONAL_MSVC_PREPROCESSOR
# endif
#endif // _MSC_VER
#if defined(_REENTRANT) || defined(_MSC_VER)
// Enable async processing, as -pthread is specified or no additional linking is required
# define CATCH_USE_ASYNC
#endif // _MSC_VER
////////////////////////////////////////////////////////////////////////////////

View File

@ -54,13 +54,19 @@ namespace Catch {
ShowDurations::OrNot Config::showDurations() const { return m_data.showDurations; }
RunTests::InWhatOrder Config::runOrder() const { return m_data.runOrder; }
unsigned int Config::rngSeed() const { return m_data.rngSeed; }
int Config::benchmarkResolutionMultiple() const { return m_data.benchmarkResolutionMultiple; }
UseColour::YesOrNo Config::useColour() const { return m_data.useColour; }
bool Config::shouldDebugBreak() const { return m_data.shouldDebugBreak; }
int Config::abortAfter() const { return m_data.abortAfter; }
bool Config::showInvisibles() const { return m_data.showInvisibles; }
Verbosity Config::verbosity() const { return m_data.verbosity; }
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
bool Config::benchmarkNoAnalysis() const { return m_data.benchmarkNoAnalysis; }
int Config::benchmarkSamples() const { return m_data.benchmarkSamples; }
double Config::benchmarkConfidenceInterval() const { return m_data.benchmarkConfidenceInterval; }
unsigned int Config::benchmarkResamples() const { return m_data.benchmarkResamples; }
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
IStream const* Config::openStream() {
return Catch::makeStream(m_data.outputFilename);
}

View File

@ -42,7 +42,13 @@ namespace Catch {
int abortAfter = -1;
unsigned int rngSeed = 0;
int benchmarkResolutionMultiple = 100;
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
bool benchmarkNoAnalysis = false;
unsigned int benchmarkSamples = 100;
double benchmarkConfidenceInterval = 0.95;
unsigned int benchmarkResamples = 100000;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
Verbosity verbosity = Verbosity::Normal;
WarnAbout::What warnings = WarnAbout::Nothing;
@ -100,12 +106,17 @@ namespace Catch {
ShowDurations::OrNot showDurations() const override;
RunTests::InWhatOrder runOrder() const override;
unsigned int rngSeed() const override;
int benchmarkResolutionMultiple() const override;
UseColour::YesOrNo useColour() const override;
bool shouldDebugBreak() const override;
int abortAfter() const override;
bool showInvisibles() const override;
Verbosity verbosity() const override;
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
bool benchmarkNoAnalysis() const override;
int benchmarkSamples() const override;
double benchmarkConfidenceInterval() const override;
unsigned int benchmarkResamples() const override;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
private:

View File

@ -9,6 +9,7 @@
#define TWOBLUECUBES_CATCH_INTERFACES_CAPTURE_H_INCLUDED
#include <string>
#include <chrono>
#include "catch_stringref.h"
#include "catch_result_type.h"
@ -22,14 +23,18 @@ namespace Catch {
struct MessageInfo;
struct MessageBuilder;
struct Counts;
struct BenchmarkInfo;
struct BenchmarkStats;
struct AssertionReaction;
struct SourceLineInfo;
struct ITransientExpression;
struct IGeneratorTracker;
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
struct BenchmarkInfo;
template <typename Duration = std::chrono::duration<double, std::nano>>
struct BenchmarkStats;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
struct IResultCapture {
virtual ~IResultCapture();
@ -41,8 +46,12 @@ namespace Catch {
virtual auto acquireGeneratorTracker( SourceLineInfo const& lineInfo ) -> IGeneratorTracker& = 0;
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
virtual void benchmarkPreparing( std::string const& name ) = 0;
virtual void benchmarkStarting( BenchmarkInfo const& info ) = 0;
virtual void benchmarkEnded( BenchmarkStats const& stats ) = 0;
virtual void benchmarkEnded( BenchmarkStats<> const& stats ) = 0;
virtual void benchmarkFailed( std::string const& error ) = 0;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
virtual void pushScopedMessage( MessageInfo const& message ) = 0;
virtual void popScopedMessage( MessageInfo const& message ) = 0;

View File

@ -9,6 +9,7 @@
#define TWOBLUECUBES_CATCH_INTERFACES_CONFIG_H_INCLUDED
#include "catch_common.h"
#include "catch_option.hpp"
#include <iosfwd>
#include <string>
@ -72,10 +73,16 @@ namespace Catch {
virtual std::vector<std::string> const& getTestsOrTags() const = 0;
virtual RunTests::InWhatOrder runOrder() const = 0;
virtual unsigned int rngSeed() const = 0;
virtual int benchmarkResolutionMultiple() const = 0;
virtual UseColour::YesOrNo useColour() const = 0;
virtual std::vector<std::string> const& getSectionsToRun() const = 0;
virtual Verbosity verbosity() const = 0;
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
virtual bool benchmarkNoAnalysis() const = 0;
virtual int benchmarkSamples() const = 0;
virtual double benchmarkConfidenceInterval() const = 0;
virtual unsigned int benchmarkResamples() const = 0;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
};
using IConfigPtr = std::shared_ptr<IConfig const>;

View File

@ -18,12 +18,18 @@
#include "catch_option.hpp"
#include "catch_stringref.h"
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
#include "benchmark/catch_estimate.hpp"
#include "benchmark/catch_outlier_classification.hpp"
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
#include <string>
#include <iosfwd>
#include <map>
#include <set>
#include <memory>
#include <algorithm>
namespace Catch {
@ -159,14 +165,43 @@ namespace Catch {
bool aborting;
};
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
struct BenchmarkInfo {
std::string name;
double estimatedDuration;
int iterations;
int samples;
unsigned int resamples;
double clockResolution;
double clockCost;
};
template <class Duration>
struct BenchmarkStats {
BenchmarkInfo info;
std::size_t iterations;
uint64_t elapsedTimeInNanoseconds;
std::vector<Duration> samples;
Benchmark::Estimate<Duration> mean;
Benchmark::Estimate<Duration> standardDeviation;
Benchmark::OutlierClassification outliers;
double outlierVariance;
template <typename Duration2>
operator BenchmarkStats<Duration2>() const {
std::vector<Duration2> samples2;
samples2.reserve(samples.size());
std::transform(samples.begin(), samples.end(), std::back_inserter(samples2), [](Duration d) { return Duration2(d); });
return {
info,
std::move(samples2),
mean,
standardDeviation,
outliers,
outlierVariance,
};
}
};
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
struct IStreamingReporter {
virtual ~IStreamingReporter() = default;
@ -185,17 +220,18 @@ namespace Catch {
virtual void testCaseStarting( TestCaseInfo const& testInfo ) = 0;
virtual void sectionStarting( SectionInfo const& sectionInfo ) = 0;
// *** experimental ***
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
virtual void benchmarkPreparing( std::string const& ) {}
virtual void benchmarkStarting( BenchmarkInfo const& ) {}
virtual void benchmarkEnded( BenchmarkStats<> const& ) {}
virtual void benchmarkFailed( std::string const& ) {}
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
virtual void assertionStarting( AssertionInfo const& assertionInfo ) = 0;
// The return value indicates if the messages buffer should be cleared:
virtual bool assertionEnded( AssertionStats const& assertionStats ) = 0;
// *** experimental ***
virtual void benchmarkEnded( BenchmarkStats const& ) {}
virtual void sectionEnded( SectionStats const& sectionStats ) = 0;
virtual void testCaseEnded( TestCaseStats const& testCaseStats ) = 0;
virtual void testGroupEnded( TestGroupStats const& testGroupStats ) = 0;

View File

@ -14,6 +14,21 @@
namespace Catch {
template<typename T>
struct always_false : std::false_type {};
template <typename> struct true_given : std::true_type {};
struct is_callable_tester {
template <typename Fun, typename... Args>
true_given<decltype(std::declval<Fun>()(std::declval<Args>()...))> static test(int);
template <typename...>
std::false_type static test(...);
};
template <typename T>
struct is_callable;
template <typename Fun, typename... Args>
struct is_callable<Fun(Args...)> : decltype(is_callable_tester::test<Fun, Args...>(0)) {};
} // namespace Catch
#endif // TWOBLUECUBES_CATCH_META_HPP_INCLUDED

View File

@ -230,12 +230,21 @@ namespace Catch {
m_unfinishedSections.push_back(endInfo);
}
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
void RunContext::benchmarkPreparing(std::string const& name) {
m_reporter->benchmarkPreparing(name);
}
void RunContext::benchmarkStarting( BenchmarkInfo const& info ) {
m_reporter->benchmarkStarting( info );
}
void RunContext::benchmarkEnded( BenchmarkStats const& stats ) {
void RunContext::benchmarkEnded( BenchmarkStats<> const& stats ) {
m_reporter->benchmarkEnded( stats );
}
void RunContext::benchmarkFailed(std::string const & error) {
m_reporter->benchmarkFailed(error);
}
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
void RunContext::pushScopedMessage(MessageInfo const & message) {
m_messages.push_back(message);

View File

@ -82,8 +82,12 @@ namespace Catch {
auto acquireGeneratorTracker( SourceLineInfo const& lineInfo ) -> IGeneratorTracker& override;
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
void benchmarkPreparing( std::string const& name ) override;
void benchmarkStarting( BenchmarkInfo const& info ) override;
void benchmarkEnded( BenchmarkStats const& stats ) override;
void benchmarkEnded( BenchmarkStats<> const& stats ) override;
void benchmarkFailed( std::string const& error ) override;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
void pushScopedMessage( MessageInfo const& message ) override;
void popScopedMessage( MessageInfo const& message ) override;

View File

@ -25,7 +25,7 @@ namespace Catch {
Catch::IStream::~IStream() = default;
namespace detail { namespace {
namespace Detail { namespace {
template<typename WriterF, std::size_t bufferSize=256>
class StreamBufImpl : public std::streambuf {
char data[bufferSize];
@ -124,15 +124,15 @@ namespace Catch {
auto makeStream( StringRef const &filename ) -> IStream const* {
if( filename.empty() )
return new detail::CoutStream();
return new Detail::CoutStream();
else if( filename[0] == '%' ) {
if( filename == "%debug" )
return new detail::DebugOutStream();
return new Detail::DebugOutStream();
else
CATCH_ERROR( "Unrecognised stream: '" << filename << "'" );
}
else
return new detail::FileStream( filename );
return new Detail::FileStream( filename );
}

View File

@ -208,6 +208,10 @@ class Duration {
Unit m_units;
public:
explicit Duration(double inNanoseconds, Unit units = Unit::Auto)
: Duration(static_cast<uint64_t>(inNanoseconds), units) {
}
explicit Duration(uint64_t inNanoseconds, Unit units = Unit::Auto)
: m_inNanoseconds(inNanoseconds),
m_units(units) {
@ -283,9 +287,15 @@ public:
if (!m_isOpen) {
m_isOpen = true;
*this << RowBreak();
for (auto const& info : m_columnInfos)
*this << info.name << ColumnBreak();
*this << RowBreak();
Columns headerCols;
Spacer spacer(2);
for (auto const& info : m_columnInfos) {
headerCols += Column(info.name).width(static_cast<std::size_t>(info.width - 2));
headerCols += spacer;
}
m_os << headerCols << "\n";
m_os << Catch::getLineOfChars<'-'>() << "\n";
}
}
@ -340,9 +350,9 @@ ConsoleReporter::ConsoleReporter(ReporterConfig const& config)
m_tablePrinter(new TablePrinter(config.stream(),
{
{ "benchmark name", CATCH_CONFIG_CONSOLE_WIDTH - 32, ColumnInfo::Left },
{ "iters", 8, ColumnInfo::Right },
{ "elapsed ns", 14, ColumnInfo::Right },
{ "average", 14, ColumnInfo::Right }
{ "samples mean std dev", 14, ColumnInfo::Right },
{ "iterations low mean low std dev", 14, ColumnInfo::Right },
{ "estimated high mean high std dev", 14, ColumnInfo::Right }
})) {}
ConsoleReporter::~ConsoleReporter() = default;
@ -374,6 +384,7 @@ bool ConsoleReporter::assertionEnded(AssertionStats const& _assertionStats) {
}
void ConsoleReporter::sectionStarting(SectionInfo const& _sectionInfo) {
m_tablePrinter->close();
m_headerPrinted = false;
StreamingReporterBase::sectionStarting(_sectionInfo);
}
@ -397,11 +408,11 @@ void ConsoleReporter::sectionEnded(SectionStats const& _sectionStats) {
StreamingReporterBase::sectionEnded(_sectionStats);
}
void ConsoleReporter::benchmarkStarting(BenchmarkInfo const& info) {
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
void ConsoleReporter::benchmarkPreparing(std::string const& name) {
lazyPrintWithoutClosingBenchmarkTable();
auto nameCol = Column( info.name ).width( static_cast<std::size_t>( m_tablePrinter->columnInfos()[0].width - 2 ) );
auto nameCol = Column(name).width(static_cast<std::size_t>(m_tablePrinter->columnInfos()[0].width - 2));
bool firstLine = true;
for (auto line : nameCol) {
@ -413,13 +424,29 @@ void ConsoleReporter::benchmarkStarting(BenchmarkInfo const& info) {
(*m_tablePrinter) << line << ColumnBreak();
}
}
void ConsoleReporter::benchmarkEnded(BenchmarkStats const& stats) {
Duration average(stats.elapsedTimeInNanoseconds / stats.iterations);
(*m_tablePrinter)
<< stats.iterations << ColumnBreak()
<< stats.elapsedTimeInNanoseconds << ColumnBreak()
<< average << ColumnBreak();
void ConsoleReporter::benchmarkStarting(BenchmarkInfo const& info) {
(*m_tablePrinter) << info.samples << ColumnBreak()
<< info.iterations << ColumnBreak()
<< Duration(info.estimatedDuration) << ColumnBreak();
}
void ConsoleReporter::benchmarkEnded(BenchmarkStats<> const& stats) {
(*m_tablePrinter) << ColumnBreak()
<< Duration(stats.mean.point.count()) << ColumnBreak()
<< Duration(stats.mean.lower_bound.count()) << ColumnBreak()
<< Duration(stats.mean.upper_bound.count()) << ColumnBreak() << ColumnBreak()
<< Duration(stats.standardDeviation.point.count()) << ColumnBreak()
<< Duration(stats.standardDeviation.lower_bound.count()) << ColumnBreak()
<< Duration(stats.standardDeviation.upper_bound.count()) << ColumnBreak() << ColumnBreak() << ColumnBreak() << ColumnBreak() << ColumnBreak();
}
void ConsoleReporter::benchmarkFailed(std::string const& error) {
Colour colour(Colour::Red);
(*m_tablePrinter)
<< "Benchmark failed (" << error << ")"
<< ColumnBreak() << RowBreak();
}
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
void ConsoleReporter::testCaseEnded(TestCaseStats const& _testCaseStats) {
m_tablePrinter->close();

View File

@ -39,9 +39,12 @@ namespace Catch {
void sectionStarting(SectionInfo const& _sectionInfo) override;
void sectionEnded(SectionStats const& _sectionStats) override;
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
void benchmarkPreparing(std::string const& name) override;
void benchmarkStarting(BenchmarkInfo const& info) override;
void benchmarkEnded(BenchmarkStats const& stats) override;
void benchmarkEnded(BenchmarkStats<> const& stats) override;
void benchmarkFailed(std::string const& error) override;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
void testCaseEnded(TestCaseStats const& _testCaseStats) override;
void testGroupEnded(TestGroupStats const& _testGroupStats) override;

View File

@ -42,19 +42,34 @@ namespace Catch {
m_reporter->noMatchingTestCases( spec );
}
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
void ListeningReporter::benchmarkPreparing( std::string const& name ) {
for (auto const& listener : m_listeners) {
listener->benchmarkPreparing(name);
}
m_reporter->benchmarkPreparing(name);
}
void ListeningReporter::benchmarkStarting( BenchmarkInfo const& benchmarkInfo ) {
for ( auto const& listener : m_listeners ) {
listener->benchmarkStarting( benchmarkInfo );
}
m_reporter->benchmarkStarting( benchmarkInfo );
}
void ListeningReporter::benchmarkEnded( BenchmarkStats const& benchmarkStats ) {
void ListeningReporter::benchmarkEnded( BenchmarkStats<> const& benchmarkStats ) {
for ( auto const& listener : m_listeners ) {
listener->benchmarkEnded( benchmarkStats );
}
m_reporter->benchmarkEnded( benchmarkStats );
}
void ListeningReporter::benchmarkFailed( std::string const& error ) {
for (auto const& listener : m_listeners) {
listener->benchmarkFailed(error);
}
m_reporter->benchmarkFailed(error);
}
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
void ListeningReporter::testRunStarting( TestRunInfo const& testRunInfo ) {
for ( auto const& listener : m_listeners ) {
listener->testRunStarting( testRunInfo );

View File

@ -31,8 +31,12 @@ namespace Catch {
static std::set<Verbosity> getSupportedVerbosities();
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
void benchmarkPreparing(std::string const& name) override;
void benchmarkStarting( BenchmarkInfo const& benchmarkInfo ) override;
void benchmarkEnded( BenchmarkStats const& benchmarkStats ) override;
void benchmarkEnded( BenchmarkStats<> const& benchmarkStats ) override;
void benchmarkFailed(std::string const&) override;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
void testRunStarting( TestRunInfo const& testRunInfo ) override;
void testGroupStarting( GroupInfo const& groupInfo ) override;

View File

@ -219,6 +219,48 @@ namespace Catch {
m_xml.endElement();
}
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
void XmlReporter::benchmarkStarting(BenchmarkInfo const &info) {
m_xml.startElement("BenchmarkResults")
.writeAttribute("name", info.name)
.writeAttribute("samples", info.samples)
.writeAttribute("resamples", info.resamples)
.writeAttribute("iterations", info.iterations)
.writeAttribute("clockResolution", static_cast<uint64_t>(info.clockResolution))
.writeAttribute("estimatedDuration", static_cast<uint64_t>(info.estimatedDuration))
.writeComment("All values in nano seconds");
}
void XmlReporter::benchmarkEnded(BenchmarkStats<> const& benchmarkStats) {
m_xml.startElement("mean")
.writeAttribute("value", static_cast<uint64_t>(benchmarkStats.mean.point.count()))
.writeAttribute("lowerBound", static_cast<uint64_t>(benchmarkStats.mean.lower_bound.count()))
.writeAttribute("upperBound", static_cast<uint64_t>(benchmarkStats.mean.upper_bound.count()))
.writeAttribute("ci", benchmarkStats.mean.confidence_interval);
m_xml.endElement();
m_xml.startElement("standardDeviation")
.writeAttribute("value", benchmarkStats.standardDeviation.point.count())
.writeAttribute("lowerBound", benchmarkStats.standardDeviation.lower_bound.count())
.writeAttribute("upperBound", benchmarkStats.standardDeviation.upper_bound.count())
.writeAttribute("ci", benchmarkStats.standardDeviation.confidence_interval);
m_xml.endElement();
m_xml.startElement("outliers")
.writeAttribute("variance", benchmarkStats.outlierVariance)
.writeAttribute("lowMild", benchmarkStats.outliers.low_mild)
.writeAttribute("lowSevere", benchmarkStats.outliers.low_severe)
.writeAttribute("highMild", benchmarkStats.outliers.high_mild)
.writeAttribute("highSevere", benchmarkStats.outliers.high_severe);
m_xml.endElement();
m_xml.endElement();
}
void XmlReporter::benchmarkFailed(std::string const &error) {
m_xml.scopedElement("failed").
writeAttribute("message", error);
m_xml.endElement();
}
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
CATCH_REGISTER_REPORTER( "xml", XmlReporter )
} // end namespace Catch

View File

@ -50,6 +50,12 @@ namespace Catch {
void testRunEnded(TestRunStats const& testRunStats) override;
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
void benchmarkStarting(BenchmarkInfo const&) override;
void benchmarkEnded(BenchmarkStats<> const&) override;
void benchmarkFailed(std::string const&) override;
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING
private:
Timer m_testCaseTimer;
XmlWriter m_xml;

View File

@ -18,6 +18,7 @@ set(TEST_SOURCES
${SELF_TEST_DIR}/TestMain.cpp
${SELF_TEST_DIR}/IntrospectiveTests/CmdLine.tests.cpp
${SELF_TEST_DIR}/IntrospectiveTests/GeneratorsImpl.tests.cpp
${SELF_TEST_DIR}/IntrospectiveTests/InternalBenchmark.tests.cpp
${SELF_TEST_DIR}/IntrospectiveTests/PartTracker.tests.cpp
${SELF_TEST_DIR}/IntrospectiveTests/Tag.tests.cpp
${SELF_TEST_DIR}/IntrospectiveTests/String.tests.cpp
@ -79,6 +80,28 @@ CheckFileList(EXTERNAL_HEADERS ${HEADER_DIR}/external)
# Please keep these ordered alphabetically
set(BENCHMARK_HEADERS
${HEADER_DIR}/internal/benchmark/catch_benchmark.hpp
${HEADER_DIR}/internal/benchmark/catch_chronometer.hpp
${HEADER_DIR}/internal/benchmark/catch_clock.hpp
${HEADER_DIR}/internal/benchmark/catch_constructor.hpp
${HEADER_DIR}/internal/benchmark/catch_environment.hpp
${HEADER_DIR}/internal/benchmark/catch_estimate.hpp
${HEADER_DIR}/internal/benchmark/catch_execution_plan.hpp
${HEADER_DIR}/internal/benchmark/catch_optimizer.hpp
${HEADER_DIR}/internal/benchmark/catch_outlier_classification.hpp
${HEADER_DIR}/internal/benchmark/catch_sample_analysis.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_analyse.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_benchmark_function.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_complete_invoke.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_estimate_clock.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_measure.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_repeat.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_run_for_at_least.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_stats.hpp
${HEADER_DIR}/internal/benchmark/detail/catch_timing.hpp
)
SOURCE_GROUP("benchmark" FILES ${BENCHMARK_HEADERS})
set(INTERNAL_HEADERS
${HEADER_DIR}/internal/catch_approx.h
${HEADER_DIR}/internal/catch_assertionhandler.h
@ -138,7 +161,6 @@ set(INTERNAL_HEADERS
${HEADER_DIR}/internal/catch_reporter_registry.h
${HEADER_DIR}/internal/catch_result_type.h
${HEADER_DIR}/internal/catch_run_context.h
${HEADER_DIR}/internal/catch_benchmark.h
${HEADER_DIR}/internal/catch_section.h
${HEADER_DIR}/internal/catch_section_info.h
${HEADER_DIR}/internal/catch_session.h
@ -174,7 +196,6 @@ set(IMPL_SOURCES
${HEADER_DIR}/internal/catch_approx.cpp
${HEADER_DIR}/internal/catch_assertionhandler.cpp
${HEADER_DIR}/internal/catch_assertionresult.cpp
${HEADER_DIR}/internal/catch_benchmark.cpp
${HEADER_DIR}/internal/catch_capture_matchers.cpp
${HEADER_DIR}/internal/catch_commandline.cpp
${HEADER_DIR}/internal/catch_common.cpp
@ -269,6 +290,7 @@ set(HEADERS
${EXTERNAL_HEADERS}
${INTERNAL_HEADERS}
${REPORTER_HEADERS}
${BENCHMARK_HEADERS}
)
# Provide some groupings for IDEs

View File

@ -462,4 +462,32 @@ TEST_CASE( "Process can be configured on command line", "[config][command-line]"
#endif
}
}
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
SECTION("Benchmark options") {
SECTION("samples") {
CHECK(cli.parse({ "test", "--benchmark-samples=200" }));
REQUIRE(config.benchmarkSamples == 200);
}
SECTION("resamples") {
CHECK(cli.parse({ "test", "--benchmark-resamples=20000" }));
REQUIRE(config.benchmarkResamples == 20000);
}
SECTION("resamples") {
CHECK(cli.parse({ "test", "--benchmark-confidence-interval=0.99" }));
REQUIRE(config.benchmarkConfidenceInterval == Catch::Detail::Approx(0.99));
}
SECTION("resamples") {
CHECK(cli.parse({ "test", "--benchmark-no-analysis" }));
REQUIRE(config.benchmarkNoAnalysis);
}
}
#endif
}

View File

@ -0,0 +1,405 @@
/*
* Created by Joachim on 16/04/2019.
* Adapted from donated nonius code.
*
* Distributed under the Boost Software License, Version 1.0. (See accompanying
* file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
*/
#include "catch.hpp"
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
namespace {
struct manual_clock {
public:
using duration = std::chrono::nanoseconds;
using time_point = std::chrono::time_point<manual_clock, duration>;
using rep = duration::rep;
using period = duration::period;
enum { is_steady = true };
static time_point now() {
return time_point(duration(tick()));
}
static void advance(int ticks = 1) {
tick() += ticks;
}
private:
static rep& tick() {
static rep the_tick = 0;
return the_tick;
}
};
struct counting_clock {
public:
using duration = std::chrono::nanoseconds;
using time_point = std::chrono::time_point<counting_clock, duration>;
using rep = duration::rep;
using period = duration::period;
enum { is_steady = true };
static time_point now() {
static rep ticks = 0;
return time_point(duration(ticks += rate()));
}
static void set_rate(rep new_rate) { rate() = new_rate; }
private:
static rep& rate() {
static rep the_rate = 1;
return the_rate;
}
};
struct TestChronometerModel : Catch::Benchmark::Detail::ChronometerConcept {
int started = 0;
int finished = 0;
void start() override { ++started; }
void finish() override { ++finished; }
};
} // namespace
TEST_CASE("warmup", "[benchmark]") {
auto rate = 1000;
counting_clock::set_rate(rate);
auto start = counting_clock::now();
auto iterations = Catch::Benchmark::Detail::warmup<counting_clock>();
auto end = counting_clock::now();
REQUIRE((iterations * rate) > Catch::Benchmark::Detail::warmup_time.count());
REQUIRE((end - start) > Catch::Benchmark::Detail::warmup_time);
}
TEST_CASE("resolution", "[benchmark]") {
auto rate = 1000;
counting_clock::set_rate(rate);
size_t count = 10;
auto res = Catch::Benchmark::Detail::resolution<counting_clock>(static_cast<int>(count));
REQUIRE(res.size() == count);
for (size_t i = 1; i < count; ++i) {
REQUIRE(res[i] == rate);
}
}
TEST_CASE("estimate_clock_resolution", "[benchmark]") {
auto rate = 1000;
counting_clock::set_rate(rate);
int iters = 160000;
auto res = Catch::Benchmark::Detail::estimate_clock_resolution<counting_clock>(iters);
REQUIRE(res.mean.count() == rate);
REQUIRE(res.outliers.total() == 0);
}
TEST_CASE("benchmark function call", "[benchmark]") {
SECTION("without chronometer") {
auto called = 0;
auto model = TestChronometerModel{};
auto meter = Catch::Benchmark::Chronometer{ model, 1 };
auto fn = Catch::Benchmark::Detail::BenchmarkFunction{ [&] {
CHECK(model.started == 1);
CHECK(model.finished == 0);
++called;
} };
fn(meter);
CHECK(model.started == 1);
CHECK(model.finished == 1);
CHECK(called == 1);
}
SECTION("with chronometer") {
auto called = 0;
auto model = TestChronometerModel{};
auto meter = Catch::Benchmark::Chronometer{ model, 1 };
auto fn = Catch::Benchmark::Detail::BenchmarkFunction{ [&](Catch::Benchmark::Chronometer) {
CHECK(model.started == 0);
CHECK(model.finished == 0);
++called;
} };
fn(meter);
CHECK(model.started == 0);
CHECK(model.finished == 0);
CHECK(called == 1);
}
}
TEST_CASE("uniform samples", "[benchmark]") {
std::vector<double> samples(100);
std::fill(samples.begin(), samples.end(), 23);
using it = std::vector<double>::iterator;
auto e = Catch::Benchmark::Detail::bootstrap(0.95, samples.begin(), samples.end(), samples, [](it a, it b) {
auto sum = std::accumulate(a, b, 0.);
return sum / (b - a);
});
CHECK(e.point == 23);
CHECK(e.upper_bound == 23);
CHECK(e.lower_bound == 23);
CHECK(e.confidence_interval == 0.95);
}
TEST_CASE("normal_cdf", "[benchmark]") {
using Catch::Benchmark::Detail::normal_cdf;
CHECK(normal_cdf(0.000000) == Approx(0.50000000000000000));
CHECK(normal_cdf(1.000000) == Approx(0.84134474606854293));
CHECK(normal_cdf(-1.000000) == Approx(0.15865525393145705));
CHECK(normal_cdf(2.809729) == Approx(0.99752083845315409));
CHECK(normal_cdf(-1.352570) == Approx(0.08809652095066035));
}
TEST_CASE("erfc_inv", "[benchmark]") {
using Catch::Benchmark::Detail::erfc_inv;
CHECK(erfc_inv(1.103560) == Approx(-0.09203687623843015));
CHECK(erfc_inv(1.067400) == Approx(-0.05980291115763361));
CHECK(erfc_inv(0.050000) == Approx(1.38590382434967796));
}
TEST_CASE("normal_quantile", "[benchmark]") {
using Catch::Benchmark::Detail::normal_quantile;
CHECK(normal_quantile(0.551780) == Approx(0.13015979861484198));
CHECK(normal_quantile(0.533700) == Approx(0.08457408802851875));
CHECK(normal_quantile(0.025000) == Approx(-1.95996398454005449));
}
TEST_CASE("mean", "[benchmark]") {
std::vector<double> x{ 10., 20., 14., 16., 30., 24. };
auto m = Catch::Benchmark::Detail::mean(x.begin(), x.end());
REQUIRE(m == 19.);
}
TEST_CASE("weighted_average_quantile", "[benchmark]") {
std::vector<double> x{ 10., 20., 14., 16., 30., 24. };
auto q1 = Catch::Benchmark::Detail::weighted_average_quantile(1, 4, x.begin(), x.end());
auto med = Catch::Benchmark::Detail::weighted_average_quantile(1, 2, x.begin(), x.end());
auto q3 = Catch::Benchmark::Detail::weighted_average_quantile(3, 4, x.begin(), x.end());
REQUIRE(q1 == 14.5);
REQUIRE(med == 18.);
REQUIRE(q3 == 23.);
}
TEST_CASE("classify_outliers", "[benchmark]") {
auto require_outliers = [](Catch::Benchmark::OutlierClassification o, int los, int lom, int him, int his) {
REQUIRE(o.low_severe == los);
REQUIRE(o.low_mild == lom);
REQUIRE(o.high_mild == him);
REQUIRE(o.high_severe == his);
REQUIRE(o.total() == los + lom + him + his);
};
SECTION("none") {
std::vector<double> x{ 10., 20., 14., 16., 30., 24. };
auto o = Catch::Benchmark::Detail::classify_outliers(x.begin(), x.end());
REQUIRE(o.samples_seen == static_cast<int>(x.size()));
require_outliers(o, 0, 0, 0, 0);
}
SECTION("low severe") {
std::vector<double> x{ -12., 20., 14., 16., 30., 24. };
auto o = Catch::Benchmark::Detail::classify_outliers(x.begin(), x.end());
REQUIRE(o.samples_seen == static_cast<int>(x.size()));
require_outliers(o, 1, 0, 0, 0);
}
SECTION("low mild") {
std::vector<double> x{ 1., 20., 14., 16., 30., 24. };
auto o = Catch::Benchmark::Detail::classify_outliers(x.begin(), x.end());
REQUIRE(o.samples_seen == static_cast<int>(x.size()));
require_outliers(o, 0, 1, 0, 0);
}
SECTION("high mild") {
std::vector<double> x{ 10., 20., 14., 16., 36., 24. };
auto o = Catch::Benchmark::Detail::classify_outliers(x.begin(), x.end());
REQUIRE(o.samples_seen == static_cast<int>(x.size()));
require_outliers(o, 0, 0, 1, 0);
}
SECTION("high severe") {
std::vector<double> x{ 10., 20., 14., 16., 49., 24. };
auto o = Catch::Benchmark::Detail::classify_outliers(x.begin(), x.end());
REQUIRE(o.samples_seen == static_cast<int>(x.size()));
require_outliers(o, 0, 0, 0, 1);
}
SECTION("mixed") {
std::vector<double> x{ -20., 20., 14., 16., 39., 24. };
auto o = Catch::Benchmark::Detail::classify_outliers(x.begin(), x.end());
REQUIRE(o.samples_seen == static_cast<int>(x.size()));
require_outliers(o, 1, 0, 1, 0);
}
}
TEST_CASE("analyse", "[benchmark]") {
Catch::ConfigData data{};
data.benchmarkConfidenceInterval = 0.95;
data.benchmarkNoAnalysis = false;
data.benchmarkResamples = 1000;
data.benchmarkSamples = 99;
Catch::Config config{data};
using Duration = Catch::Benchmark::FloatDuration<Catch::Benchmark::default_clock>;
Catch::Benchmark::Environment<Duration> env;
std::vector<Duration> samples(99);
for (size_t i = 0; i < samples.size(); ++i) {
samples[i] = Duration(23 + (i % 3 - 1));
}
auto analysis = Catch::Benchmark::Detail::analyse(config, env, samples.begin(), samples.end());
CHECK(analysis.mean.point.count() == 23);
CHECK(analysis.mean.lower_bound.count() < 23);
CHECK(analysis.mean.lower_bound.count() > 22);
CHECK(analysis.mean.upper_bound.count() > 23);
CHECK(analysis.mean.upper_bound.count() < 24);
CHECK(analysis.standard_deviation.point.count() > 0.5);
CHECK(analysis.standard_deviation.point.count() < 1);
CHECK(analysis.standard_deviation.lower_bound.count() > 0.5);
CHECK(analysis.standard_deviation.lower_bound.count() < 1);
CHECK(analysis.standard_deviation.upper_bound.count() > 0.5);
CHECK(analysis.standard_deviation.upper_bound.count() < 1);
CHECK(analysis.outliers.total() == 0);
CHECK(analysis.outliers.low_mild == 0);
CHECK(analysis.outliers.low_severe == 0);
CHECK(analysis.outliers.high_mild == 0);
CHECK(analysis.outliers.high_severe == 0);
CHECK(analysis.outliers.samples_seen == samples.size());
CHECK(analysis.outlier_variance < 0.5);
CHECK(analysis.outlier_variance > 0);
}
TEST_CASE("analyse no analysis", "[benchmark]") {
Catch::ConfigData data{};
data.benchmarkConfidenceInterval = 0.95;
data.benchmarkNoAnalysis = true;
data.benchmarkResamples = 1000;
data.benchmarkSamples = 99;
Catch::Config config{ data };
using Duration = Catch::Benchmark::FloatDuration<Catch::Benchmark::default_clock>;
Catch::Benchmark::Environment<Duration> env;
std::vector<Duration> samples(99);
for (size_t i = 0; i < samples.size(); ++i) {
samples[i] = Duration(23 + (i % 3 - 1));
}
auto analysis = Catch::Benchmark::Detail::analyse(config, env, samples.begin(), samples.end());
CHECK(analysis.mean.point.count() == 23);
CHECK(analysis.mean.lower_bound.count() == 23);
CHECK(analysis.mean.upper_bound.count() == 23);
CHECK(analysis.standard_deviation.point.count() == 0);
CHECK(analysis.standard_deviation.lower_bound.count() == 0);
CHECK(analysis.standard_deviation.upper_bound.count() == 0);
CHECK(analysis.outliers.total() == 0);
CHECK(analysis.outliers.low_mild == 0);
CHECK(analysis.outliers.low_severe == 0);
CHECK(analysis.outliers.high_mild == 0);
CHECK(analysis.outliers.high_severe == 0);
CHECK(analysis.outliers.samples_seen == 0);
CHECK(analysis.outlier_variance == 0);
}
TEST_CASE("run_for_at_least, int", "[benchmark]") {
manual_clock::duration time(100);
int old_x = 1;
auto Timing = Catch::Benchmark::Detail::run_for_at_least<manual_clock>(time, 1, [&old_x](int x) -> int {
CHECK(x >= old_x);
manual_clock::advance(x);
old_x = x;
return x + 17;
});
REQUIRE(Timing.elapsed >= time);
REQUIRE(Timing.result == Timing.iterations + 17);
REQUIRE(Timing.iterations >= time.count());
}
TEST_CASE("run_for_at_least, chronometer", "[benchmark]") {
manual_clock::duration time(100);
int old_runs = 1;
auto Timing = Catch::Benchmark::Detail::run_for_at_least<manual_clock>(time, 1, [&old_runs](Catch::Benchmark::Chronometer meter) -> int {
CHECK(meter.runs() >= old_runs);
manual_clock::advance(100);
meter.measure([] {
manual_clock::advance(1);
});
old_runs = meter.runs();
return meter.runs() + 17;
});
REQUIRE(Timing.elapsed >= time);
REQUIRE(Timing.result == Timing.iterations + 17);
REQUIRE(Timing.iterations >= time.count());
}
TEST_CASE("measure", "[benchmark]") {
auto r = Catch::Benchmark::Detail::measure<manual_clock>([](int x) -> int {
CHECK(x == 17);
manual_clock::advance(42);
return 23;
}, 17);
auto s = Catch::Benchmark::Detail::measure<manual_clock>([](int x) -> int {
CHECK(x == 23);
manual_clock::advance(69);
return 17;
}, 23);
CHECK(r.elapsed.count() == 42);
CHECK(r.result == 23);
CHECK(r.iterations == 1);
CHECK(s.elapsed.count() == 69);
CHECK(s.result == 17);
CHECK(s.iterations == 1);
}
TEST_CASE("run benchmark", "[benchmark]") {
counting_clock::set_rate(1000);
auto start = counting_clock::now();
Catch::Benchmark::Benchmark bench{ "Test Benchmark", [](Catch::Benchmark::Chronometer meter) {
counting_clock::set_rate(100000);
meter.measure([] { return counting_clock::now(); });
} };
bench.run<counting_clock>();
auto end = counting_clock::now();
CHECK((end - start).count() == 2867251000);
}
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING

View File

@ -2,25 +2,63 @@
#include <map>
TEST_CASE( "benchmarked", "[!benchmark]" ) {
#ifndef CATCH_CONFIG_DISABLE_BENCHMARKING
std::uint64_t Fibonacci(std::uint64_t number);
std::uint64_t Fibonacci(std::uint64_t number) {
return number < 2 ? 1 : Fibonacci(number - 1) + Fibonacci(number - 2);
}
TEST_CASE("Benchmark Fibonacci", "[!benchmark]") {
CHECK(Fibonacci(0) == 1);
// some more asserts..
CHECK(Fibonacci(5) == 8);
// some more asserts..
BENCHMARK("Fibonacci 20") {
return Fibonacci(20);
};
BENCHMARK("Fibonacci 25") {
return Fibonacci(25);
};
BENCHMARK("Fibonacci 30") {
return Fibonacci(30);
};
BENCHMARK("Fibonacci 35") {
return Fibonacci(35);
};
}
TEST_CASE("Benchmark containers", "[!benchmark]") {
static const int size = 100;
std::vector<int> v;
std::map<int, int> m;
SECTION("without generator") {
BENCHMARK("Load up a vector") {
v = std::vector<int>();
for (int i = 0; i < size; ++i)
v.push_back(i);
}
};
REQUIRE(v.size() == size);
// test optimizer control
BENCHMARK("Add up a vector's content") {
uint64_t add = 0;
for (int i = 0; i < size; ++i)
add += v[i];
return add;
};
BENCHMARK("Load up a map") {
m = std::map<int, int>();
for (int i = 0; i < size; ++i)
m.insert({ i, i + 1 });
}
};
REQUIRE(m.size() == size);
BENCHMARK("Reserved vector") {
@ -28,16 +66,65 @@ TEST_CASE( "benchmarked", "[!benchmark]" ) {
v.reserve(size);
for (int i = 0; i < size; ++i)
v.push_back(i);
}
};
REQUIRE(v.size() == size);
BENCHMARK("Resized vector") {
v = std::vector<int>();
v.resize(size);
for (int i = 0; i < size; ++i)
v[i] = i;
};
REQUIRE(v.size() == size);
int array[size];
BENCHMARK("A fixed size array that should require no allocations") {
for (int i = 0; i < size; ++i)
array[i] = i;
}
};
int sum = 0;
for (int i = 0; i < size; ++i)
sum += array[i];
REQUIRE(sum > size);
SECTION("XYZ") {
BENCHMARK_ADVANCED("Load up vector with chronometer")(Catch::Benchmark::Chronometer meter) {
std::vector<int> k;
meter.measure([&](int idx) {
k = std::vector<int>();
for (int i = 0; i < size; ++i)
k.push_back(idx);
});
REQUIRE(k.size() == size);
};
int runs = 0;
BENCHMARK("Fill vector indexed", benchmarkIndex) {
v = std::vector<int>();
v.resize(size);
for (int i = 0; i < size; ++i)
v[i] = benchmarkIndex;
runs = benchmarkIndex;
};
for (size_t i = 0; i < v.size(); ++i) {
REQUIRE(v[i] == runs);
}
}
}
SECTION("with generator") {
auto generated = GENERATE(range(0, 10));
BENCHMARK("Fill vector generated") {
v = std::vector<int>();
v.resize(size);
for (int i = 0; i < size; ++i)
v[i] = generated;
};
for (size_t i = 0; i < v.size(); ++i) {
REQUIRE(v[i] == generated);
}
}
}
#endif // CATCH_CONFIG_DISABLE_BENCHMARKING