timeit-cpp

A C++ version of IPython's timeit magic.

Usage

Measure time for anything "callable" for example a lambda or an struct with operator(). timeit::timeit prints results similar to IPython's timeit magic. It automatically guesses number of repeats and iterations.

      #include <iostream>
      #include <vector>
      #include "timeit.h"
      using namespace std;
      int main(){
        vector<float> A(10000,1.0);
        auto find_sum = [&](){
          float s=0.0;
          for(float x:A)
            s += x;
        };

        timeit::timeit(find_sum);
      }

      97.5 μs ± 93.3 ns per loop (mean ± std. dev. of 3 runs, 100000 loops each)

timeit::timeit can also take other arguments.

    long repeat=5, number=100;
    timeit::timeit(find_sum, repeat, number);

    0.101 ms ± 3.06 μs per loop (mean ± std. dev. of 5 runs, 100 loops each)

timeit::repeat returns a vector of results

    long repeat=5, number=100;
    vector<double> timings = timeit::repeat(find_sum, repeat, number);

timeit::autorange returns the number of loops and time taken so that total time >= 1.0.

    cout << timeit::autorange(find_sum) << endl;

    20000

Examples

Comparing different algorithms to solve the Sliding Window Maximum (Maximum of all subarrays of size k) problem. See complete source.

    cout<<"Approach 1(deque):\n";
    timeit::timeit(sliding_deque);
    cout<<"Approach 2(cache):\n";
    timeit::timeit(sliding_cache);
    cout<<"Approach 3(naive):\n";
    timeit::timeit(sliding_naive);

    Approach 1(deque):
    95.6 μs ± 53.5 ns per loop (mean ± std. dev. of 3 runs, 100000 loops each)
    Approach 2(cache):
    7.6 μs ± 5.2 ns per loop (mean ± std. dev. of 3 runs, 1000000 loops each)
    Approach 3(naive):
    0.246 ms ± 2.07 μs per loop (mean ± std. dev. of 7 runs, 10000 loops each)

Future Work

• Compare best and worst timing results and show the same.