I did some simple speed measurement and can not understand the results. I measured just 1 million additions, then I measured 1 million subtractions and results are way different. It shows that I don't quite understand what is going on 'inside'. I would really appreciate if someone explain why:

my code:


#include "stdio.h" /* i changed brackets into double quotes */
#include "fcntl.h" /* on these include statements just for forum */
#include "sys/time.h" /* as the brackets are treated as html */

#define MAX 1000000

main(int argc, char *argv[])
{
struct timeval t1, t2;

printf("Additions:\n");
gettimeofday(&t1, NULL);
for(i = 0; i < MAX; i++);
gettimeofday(&t2, NULL);
showdiff(t1, t2);

printf("Subtractions:\n");
gettimeofday(&t1, NULL);
for(i = MAX; i < 0; i--);
gettimeofday(&t2, NULL);
showdiff(t1, t2);
}

#define MICROX 1000000.0
showdiff(t1, t2)
struct timeval t1;
struct timeval t2;
{
double secs;

if(t1.tv_usec > t2.tv_usec)
{
secs = ((MICROX + (t2.tv_usec - t1.tv_usec)) / MICROX);
}
else
{
secs = (t2.tv_usec-t1.tv_usec) / MICROX;
}

secs += t2.tv_sec - t1.tv_sec;
printf("Interval: %.6f secs\n", secs);

}

Compiled and run
$ cc -gc measure_time.c && cc -o measure_time measure_time.o
(I specifically added -g to turn off all optimization)

$ measure_time
Additions:
Interval: 0.001876 secs
Subtractions:
Interval: 0.000001 secs

This is SCO OpenServer (3.2 MHz Xeon) system.
On the same hardware under Linux kernel 2.6.9 results are proportionally same.

Why my i-- code takes no time, where i++ does take some time?