looptest.c

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/time.h>

/* compile with gcc -m32 -O2 -o looptest */

unsigned int looptest1() {
    unsigned int rv = 0;
    asm volatile (
        "mov $0x80000000, %%ecx\n"
        "1:\n"
        "    add %%ecx, %0\n"
        "    loop 1b\n"
        : "+r" (rv)
        :
        : "ecx", "cc"
    );
    return rv;
}

unsigned int looptest2() {
    unsigned int rv = 0;
    asm volatile (
        "mov $0x80000000, %%ecx\n"
        "1:\n"
        "    add %%ecx, %0\n"
        "    dec %%ecx\n"
        "    jnz 1b\n"
        : "+r" (rv)
        :
        : "ecx", "cc"
    );
    return rv;
}

unsigned int looptest3() {
    unsigned int rv = 0;
    asm volatile (
        "mov $0x80000000, %%ecx\n"
        "1:\n"
        "    add %%ecx, %0\n"
        "    sub $1, %%ecx\n"
        "    cmp $0, %%ecx\n"
        "    jnz 1b\n"
        : "+r" (rv)
        :
        : "ecx", "cc"
    );
    return rv;
}

long long timefunc(unsigned int (*func)()) {
    struct timeval tv_start, tv_end;
    gettimeofday(&tv_start, NULL);
    func();
    gettimeofday(&tv_end, NULL);
    tv_end.tv_sec -= tv_start.tv_sec;
    tv_end.tv_usec -= tv_start.tv_usec;
    return ((long long)tv_end.tv_sec * 1000000LL) + (long long)tv_end.tv_usec;
}

int main(int argc, char** argv) {
    int i, j;
    long long times[12];
    for (i = 0; i < 4; i++) {
        /* Unroll the loop a few times so that any effect printf might
         * have (branch prediction, trace cache, etc) are minimized */
        times[0] = timefunc(looptest1);
        times[1] = timefunc(looptest2);
        times[2] = timefunc(looptest3);
        times[3] = timefunc(looptest1);
        times[4] = timefunc(looptest2);
        times[5] = timefunc(looptest3);
        times[6] = timefunc(looptest1);
        times[7] = timefunc(looptest2);
        times[8] = timefunc(looptest3);
        times[9] = timefunc(looptest1);
        times[10] = timefunc(looptest2);
        times[11] = timefunc(looptest3);
        for (j = 0; j < 12; j += 3) {
            /* Print microseconds in CSV-compatible format */
            printf("%lld,%lld,%lld\n", times[j + 0], times[j + 1], times[j + 2]);
        }
    }
    return 0;
}