Skip to content

A simple timer class easy to use. Calcs Min, Max, Mean, STD deviation

License

Notifications You must be signed in to change notification settings

Jakobimatrix/timer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

44 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Summary

Header only library. Requires c++17.

Features a class to calculate with time and multiple timer classes for recording and profiling.

Please use clang-tidy if you want to contribute: easy installation

PreciseTime class:

  • adding/substracting while considering the unit: 1[s] + 1[s] -> ok; 1[s] + 1[s^2] -> error
  • division/multiplication with
    • a) int/double/etc
    • b) PreciseTime class while keeping track of the unit: 1[s] * 1[s] = 1[s^2]
  • calculateing the square root of a PreciseTime keeping track of the unit: sqrt(4[s^2]) = 2[s].
  • Comparisons: {==, !=, <=, >=, <, >}
  • Nice print output: std::cout << my_precise_time << std::endl;
  • Max Time: h: [9223372036854775807] m: [59] s: [59] ms: [999] us: [999] ns: [999]
  • Min Time: h: [-9223372036854775808] m: [-59] s: [-59] ms: [-999] us: [-999] ns:[-999]
  • Constexpr construction and arithmetic (appart from sqrt)

Todos

  • A dynamic PreciseTime where the user can specify the needed resolution and the max time span to optimize calculation
  • Optimize: get rid of the internal seconds

CollectingTimer class:

  • Record multiple times (e.g. in a loop) the execution time of e.g. a function.
  • As many (named) timers as you like, held in one instance of the CollectingTimer class.
  • On demand output of max, min, mean, median, standard deviation for all timers.
  • Print Histogram of measurements into console
  • Write multiple histograms on top of each other for better comparison in console.
  • Write measurements to file for further investigation in your favorite table calculation or MATLAB/Octave
  • Print histogram to file for further investigation in your favorite table calculation (choose X-Y-Plot) or MATLAB/Octave.

FrameTimer class:

  • To be used in a loop: For every loop/frame record the execution time of multiple functions (via named timers) called (multiple times) in that loop.
  • Print for every frame the total execution time of a (named) timer into a file for further investigation in your favorite table calculation or MATLAB/Octave
  • simple console debug output showing the 3 longest running functions in the last frame.

Todos

  • LiveStream every Frame via tcp/ip socet into a GUI to have a live graph
  • At the moment only the accumulated time per frame per timer is available in the output file. Maybe show every call and duration as a rectangle in a timeline for every function.

ScopedTimer class:

  • Starts the timer on creation and stops it on destruction. A callback function to report the result must be provided.

SimpleTimer class:

  • Start/Reset/getTime nothing more.

Build the project (Linux)

  • If you use CMake to build your project place this repro inside 'path' and use add_subdirectory(path/ timer) and target_link_libraries(your_lib timer_lib)
  • Or just include the header
  • To run the tests run scripts/debug_build.sh

Examples

PreciseTime:

#include "include/precise_time.hpp"

int main() {
  constexpr PreciseTime max_pt = PreciseTime::max();
  constexpr PreciseTime min_pt = PreciseTime::min();
  std::cout << "max: " << max_pt << "\n"
            << "min: " << min_pt << "\n";

  using ns = std::chrono::nanoseconds;
  using us = std::chrono::microseconds;
  using ms = std::chrono::milliseconds;
  using s = std::chrono::seconds;
  using m = std::chrono::minutes;
  using h = std::chrono::hours;

  // construction
  PreciseTime my_time1(ns(987654321));
  std::cout << "1. " << my_time1 << "\n";
  PreciseTime my_time2(ms(42));
  std::cout << "2. " << my_time2 << "\n";
  constexpr PreciseTime my_time3 = ns(22) + us(450) + ms(12);
  std::cout << "3. " << my_time3 << "\n";
  
  constexpr std::array<long, 6> seperated_times = my_time3.getSeperatedTimeComponents();
  const std::array<std::string, 6> info = {
      "nanoseconds.",  "microseconds.",
      "milliseconds.", "seconds.",
      "minutes.",      "hours."};
  for(size_t i = 0; i < 6; i++){
    std::cout << seperated_times[i] << " " << info[i] "\n";
  }

  // calculations: add/substract times
  my_time1 = my_time1 + my_time1;
  my_time1 -= my_time2;

  // calculations: multiply/divide by factor
  my_time2 *= 1.5;
  my_time2 = my_time2 / 3.3;

  // calculations: s*s = s^2
  PreciseTime timeSquared = my_time1 * my_time2;
  // This throws a runtime error if build in debug mode.
  // calculations: s^2 + s: can't mix units
  // PreciseTime corruptedTime = timeSquared + my_time1;

  // Takeing the square root only works with usints s^n where n is even.
  PreciseTime normalTime = PreciseTime::sqrt(timeSquared);
  std::cout << "4. " << normalTime << "\n";

  // overflow protection
  constexpr auto my_time4 = max_pt * 5;  // my_time4 is still max_pt
  std::cout << "5. " << my_time4 << "\n";

  // constexpr arithmetic
  constexpr auto zero_t = max_pt - my_time4;
  std::cout << "6. " << zero_t << "\n";

  // coversations
  typedef std::chrono::microseconds exampleType;

  // This gets the time in exampleType ! this might result in resolution loss
  // if the time is 22ns and 450us and 12ms, This returns exact 12450us
  constexpr exampleType us = my_time3.convert<exampleType>();
  std::cout << "7. " << us.count() << "\n";

  // if the time is 22ns and 450us and 12ms, This returns exact 450us
  constexpr exampleType part_us = my_time3.get<exampleType>();
  std::cout << "8. " << part_us.count() << "\n";

  // if the time is 22ns and 450us and 12ms, This returns exact* 12450.022
  // double precision exact
  constexpr double floatingPoint = my_time3.toDouble<exampleType>();
  std::cout << "9. " << std::fixed << floatingPoint << "\n\n";

  return 0;
}

Output:

PreciseTime Console output

1. CollectingTimer:

#include "include/collecting_timer.hpp"
#include <string>
#include <cmath>
#include <random>
#include <unistd.h>

int main() {
  CollectingTimer timer;

  std::random_device rd;
  std::default_random_engine generator(rd());
  const double mean = 1.;
  const double deviation = 0.15;
  std::normal_distribution<double> distribution(mean, deviation);
  
  const std::string timer_cos("std::cos");
  const std::string timer_sleep("sleep"); 

  for (int i = 0; i < 100; i++) {
    const double sleep = distribution(generator);
    timer.start(timer_sleep);
    usleep(sleep * 100);
    timer.stop(timer_sleep);
    timer.start(timer_cos);
    std::cos(static_cast<double>(i) / 100.);
    timer.stop(timer_cos);
  }
  std::cout << timer << "\n";
  
  return 0;
}

Collectin Timer Console output 2

2. CollectingTimer:

#include "include/collecting_timer.hpp"
#include <string>
#include <cmath>
#include <random>
#include <unistd.h>

// from https://github.com/google/benchmark/blob/main/include/benchmark/benchmark.h
#define BENCHMARK_ALWAYS_INLINE __attribute__((always_inline))

template <class Tp>
inline BENCHMARK_ALWAYS_INLINE void DoNotOptimize(Tp const& value) {
  asm volatile("" : : "r,m"(value) : "memory");
}

template <class Tp>
inline BENCHMARK_ALWAYS_INLINE void DoNotOptimize(Tp& value) {
#if defined(__clang__)
  asm volatile("" : "+r,m"(value) : : "memory");
#else
  asm volatile("" : "+m,r"(value) : : "memory");
#endif
}

constexpr double fastModF(double x, double& full) noexcept {
    full = static_cast<double>(static_cast<long long>(x));
    return x - full;
}

int main() {
  CollectingTimer timer;

  std::random_device rd;
  std::default_random_engine generator(rd());
  
  std::uniform_real_distribution<double> distr(-1000000000000000000., 100000000000000000000.);
  
  constexpr int num_tests = 1000000;
  const std::string timer_std_modf("std::modf");
  const std::string timer_my_mod_f("my::modf");

  for(int i = 0; i < num_tests; i++){
    const double d = distr(rd);
    timer.start(timer_std_modf);
    double full = 0.;
    const double decimals = std::modf(d, &full);
    timer.stop(timer_std_modf);
    DoNotOptimize(full);
    DoNotOptimize(decimals);

    timer.start(timer_my_mod_f);
    double full_mine = 0.;
    const double decimals_mine = fastModF(d, full_mine);
    timer.stop(timer_my_mod_f);
    DoNotOptimize(full_mine);
    DoNotOptimize(decimals_mine);
  }
  
  // save in file
  timer.histogramToFile<std::chrono::nanoseconds>("/tmp/histo.csv", ';');
  timer.measurementsToFile<std::chrono::nanoseconds>("/tmp/measure.csv", ';');
  
  // print to console
  std::vector<CollectingTimer::Result>results(2);
  timer.getResult(timer_std_modf, results[0]);
  timer.getResult(timer_my_mod_f, results[1]);
  
  std::cout << results;
  
  return 0;
}

Collectin Timer Console output 2 Collectin Timer Histogram Collectin Timer RawValues

FrameTimer:

#include <timer/frame_timer.hpp>

FrameTimer ft;
std::string t1("innerLoop");
std::string t2("outerLoop");
std::string t3("cos");
std::string t4("sin");

int f1(int i) {
  const auto t = ft.startScopedTimer(t1);
  int a = 0;
  for (int j = i; j > 0; --j) {
    a += j;
  }
  return a;
}

int f2(int i) {
  const auto t = ft.startScopedTimer(t2);
  int a = 0;
  for (int j = i; j > 0; --j) {
    a += f1(j);
  }
  return a;
}

int f3(int i) {
  const auto t = ft.startScopedTimer(t3);
  int a = 0;
  for (int j = i; j > 0; --j) {
    a += std::cos(j);
  }
  return a;
}

int f4(int i) {
  const auto t = ft.startScopedTimer(t4);
  int a = 0;
  for (int j = i; j > 0; --j) {
    a += std::sin(j);
  }
  return a;
}

int main() {

  std::vector<int> erg;
  for (int i = 1; i < 1000; ++i) {
    ft.frameStart<true>();
    erg.push_back(f2(i));
    erg.push_back(f3(i));
    erg.push_back(f4(i));
  }

  ft.measurementsToFile<std::chrono::microseconds>("/tmp/frames.csv", ';');
  return erg[0];
}

Frame Timer console output Frame Timer console output

About

A simple timer class easy to use. Calcs Min, Max, Mean, STD deviation

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published