Acmion / ComparisonCythonPythonJuliaCSharpC

Comparing the performance of Cython, Python, Julia, C# and C with two specific numerical micro benchmarks.

Comparison Cython Python Julia C# C

Comparing the performance of Cython, Python, Julia, C# and C with two specific numerical micro benchmarks.

Results

The results of this benchmark. Note that the startup times of each implementation were not accounted for. See the benchmark code, in the section "Benchmark Code". Suprisingly, Cython performed best, with C very closely behind. The compiler commands may be the cause of the slightly slower C code. Julia came in at about 2x slower, C# at about 10x slower and Python at about 200x - 1000x slower. All implementations calculate used the same input values and got the same output.

Language	Mean Execution Time (s)	Standard Deviation of Execution Time (s)	Result
C	0.005000	0.000471	78093734373750.0
Cython	0.005000	0.000737	78093734373750.0
Julia	0.012500	0.006603	78093734373750.0
C#	0.061540	0.033642	78093734373750.0
Python	5.209375	0.105768	78093734373750.0

Table 1: Statistics for calc_partial_result performance.

Language	Mean Execution Time (s)	Standard Deviation of Execution Time (s)	Result
C	0.032200	0.001229	39698235518841.0
Cython	0.028999	0.000666	39698235518841.0
Julia	0.053200	0.008230	39698235518841.0
C#	0.299488	0.101629	39698235518841.0
Python	6.353125	0.026760	39698235518841.0

Table 2: Statistics for calc_partial_conditional_result performance.

Benchmark Code

The performance of these two functions are measured (this code is in C, which has also been translated to the other languages):

double calc_partial_result(double *values, int num_of_values_to_process)
{
    int i, j;
    double result = 0.0;
    
    for(i = 0; i < num_of_values_to_process; i++)
    {
        for(j = 0; j <= i; j++)
        {
            result += values[i] * (j - 1);
        }
    }
    
    return result;
}

double calc_partial_conditional_result(double *values, double *values_conditional, int num_of_values_to_process, double min_conditional_value)
{
    int i, j;
    double result = 0.0;
    
    for(i = 0; i < num_of_values_to_process; i++)
    {
        for(j = 0; j <= i; j++)
        {
            if(values_conditional[j] >= min_conditional_value)
            {
                result += values[i] * (j - 1);
            }
        }
    }
    
    return result;
}

Requirements

Windows:

• Win10
• Python 3.9
• MinGW 64bit
• Julia 1.6
• Bash

Linux:

• ??? (the code should work for the most part, but the scripts under scripts may have to be tuned)

Execution

From the root directory run:

sh init.sh
sh execute.sh