Execute, Time and Plot C++ Programs with Python

February 02, 2011

Hate excel? Why not have python take all the pain away! This should appeal to everyone out there who needs a quick way to analyze the runtime time of a bunch of commands to a program given an arbitrary list of arguments. Hating excel/openoffice/spreadsheets in general is optional.

Just like every programmer, I am always trying to find new and interesting ways of automating routine tasks. To that end, I wrote a pretty cool program last night to help automate my first algorithms lab which consisted of many repeated calls to the same program on the command line.

Consistent with the recent trend of mine, I wrote this in python and makes use of the subprocess module to spawn and time instances of my C++ program and pylab to make plotting the times super easy. The comments in my code explain things in more detail, but here is the summary: Python is being used to execute two programs written in C++ with the same set of arguments and spits out a plot of their times. In this case the two C++ programs are recursive (rfib) and iterative (ifib) implementations of the Fibonacci algorithm. The fun stuff is in the python:

#!/usr/bin/env python

import subprocess
import string
import sys
import pylab  # matplotlib

def timerun(program, args) :
    print 'Starting timed execution of ' + program + ' with ' + str(len(args)) + ' arguments.'
    i = 1

    # Execute program, once for each n argument
    for n in args :

        # This was really annoying. Build the arguments to the time system call to the time command.
        # First of all, for whatever reason 'time' didn't work correctly with any arguments other than -p,
        # so I used /usr/bin/time instead. Since I could not figure out why the output of 'time' was not
        # coming back to stdin, I use the -o (output file) and -a (append) option to just output the real
        # execution time ( thats where '-f' and '%e' comes from ) to the file.
        p = subprocess.Popen(['/usr/bin/time', '-o', 'runtimes.txt', '-a', '-f', '%e', './' + program, str(n)], stdout=subprocess.PIPE)

        # Read back from stdin, print where we are (not required, but its nice)
        output = p.communicate()[0]
        sys.stdout.write( str(i) + ':\tfib('+ str(n) + ') = ' + output)
        i += 1
    print 'done'

    # Open up, read and return the times in the output file
    f = open('runtimes.txt', 'r')
    times = f.read().splitlines()

    # Clean up old runtimes
    subprocess.Popen(['rm', 'runtimes.txt'],stdout=subprocess.PIPE)

    return times

def main():

    # Arguments we are interested in testing runtimes for
    # args = 1, 5, 10, 15, 20, 25, 30, 35, 40, 41, 42, 43, 44, 45, 46, 47, 48
    args = range(1, 56)

    # Compute the runtimes of the recursive algorithm, then the iterative one
    rtimes = timerun('rfib', args)
    itimes = timerun('ifib', args)

    # Plot it with pylab
    pylab.xlabel('N')
    pylab.ylabel('Time to compute fib(N) (seconds)')
    pylab.title('Recursive vs. Iterative Execution Time for Fibionacci Sequence')
    pylab.plot(args, rtimes, 'ro-', label='Recursive')
    pylab.plot(args, itimes, 'bo-', label='Iterative')
    pylab.legend()

    # other drawing styles for plots in pylab:
    # 'r' red line, 'g' green line, 'y' yellow line
    # 'ro' red dots as markers, 'r.' smaller red dots, 'r+' red pluses
    # 'r--' red dashed line, 'g^' green triangles, 'bs' blue squares
    # 'rp' red pentagons, 'r1', 'r2', 'r3', 'r4' well, check out the markers

    # save the plot as a PNG image
    pylab.savefig('Fig.png')

    # show the pylab plot window
    pylab.show()

if __name__ == "__main__":
    main()