timing your functions

Applications Programming
1

hi everyone. If you have a function created in your code and you want to find out how long it takes for it to run you can use a struct called gettimeofday().

so lets say we have a function like this
int myfunction (int r)
{
/some math calculations/
return answer;
}

How do i set up gettimeofday to find out how long it takes for the function to run. thanks

2

Suggestion: use a profiler. Most UNIX systems compilers come with one.

cc myfile.c -o myfile -g -p
./myfile # run your code #
prof ./myfile

You will get a display of how long your code spent in every function, plus the number of times it was called. The reason for the suggestion - it doesn't require instrumenting your code with timing calls.

3

Simplest way to obtain timing info if to run:
time ./a.out

For info within program Id suggest using clock() instead of gettimeofday().

#include <time.h>
clock_t t0, t1;
t0=clock();
fun();
t1=clock();
printf("%f",(t1-t0)/CLOCKS_PER_SEC);

Regards.

4

To use gettimeofday you can do this:

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

static float getTimeDiff(struct timeval *s, struct timeval *e);

void main()
{
    struct timeval  sTv, eTv;

    gettimeofday(&sTv, NULL);

    /* do work */

    gettimeofday(&eTv, NULL);

    printf("Time difference is %.2f seconds\n", getTimeDiff(&sTv, &eTv));
}

The getTimeDiff function works on the two timeval structures to compute the seconds elapsed. The source is simple, and as follows:

static float getTimeDiff(struct timeval *s, struct timeval *e)
{
	struct timeval	diff_tv;
		
	diff_tv.tv_usec = e->tv_usec - s->tv_usec;
	diff_tv.tv_sec = e->tv_sec - s->tv_sec;
		
	if (s->tv_usec > e->tv_usec)
	{
		diff_tv.tv_usec += 1000000;
		diff_tv.tv_sec--;
	}
		
	return (float) diff_tv.tv_sec + ((float) diff_tv.tv_usec / 1000000.0);
}

Furthermore, it should be noted that clock() does *NOT* return values based upon real time, rather it returns CPU time. If you want to know how much REAL TIME something takes then clock() is not the solution. For example:

#include <stdio.h>
#include <stdlib.h>
#include <time.h>

void main()
{
	int	i, j, t;
	
	clock_t	clockSample[3];

	clockSample[0] = clock();
	
	sleep(2);
		
	clockSample[1] = clock();
	
	for (i = 0; i < 1024; i++)
		for (j = 0; j < 1024; j++)
			t = i * i + j;
	
	clockSample[2] = clock();
	
	for (i = 0; i < 3; i++)
	{
		printf("Clock sample %d is %lu\n", i, clockSample);
	}
}

Upon running this code, you'll notice that the first two samples for clock are positioned squarely at 0. The sleep call, which will take around 2 seconds *real* time, has no effect on the second clock sample. Since sleep takes no CPU time, it is not calculated into the real time usage of the application by clock(). Similar behavior will be seen for other non-CPU (blocking) tasks the application incurs, such as waiting in a select call for socket I/O. The third clock sample, however, does incur CPU time as it is a CPU bound loop. Thus, the third sample will reflect the CPU time incurred by the loop. If, however, you had the crazy idea to insert (for example) a usleep in the loop, you would never calculate the additional real time incurred by the usleep.

So, beware that both methods, while equally useful, answer two different questions. clock() tells you CPU time used, the gettimeofday method tells you actual time elapsed.