A simple program using futex to mimic mutex, but much slower than mutex

futex.c

#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdatomic.h>
#include <errno.h>
#include <string.h>

#include <unistd.h>
#include <pthread.h>
#include <linux/futex.h>
#include <sys/syscall.h>

#define likely(x)       __builtin_expect((x),1)
#define unlikely(x)     __builtin_expect((x),0)

static int sum = 0;

static uint32_t futex_var = 1;
const static uint32_t off = 0;
const static uint32_t on  = 1;

static void futex_lock(uint32_t *futexp)
{
   long s;

    while (1) {
        if (__sync_bool_compare_and_swap(futexp, on, off))
            break;

        s = syscall(SYS_futex, futexp, FUTEX_WAIT, 0, NULL, NULL, 0);
        if (unlikely(s == -1 && errno != EAGAIN)) {
            printf("unexpected error...\n");
            pthread_exit(NULL);
        }
    }
}

static void futex_unlock(uint32_t *futexp)
{
   long s;

    while (1) {
        if (__sync_bool_compare_and_swap(futexp, off, on)) {
            // wakeup 1 thread now
            s= syscall(SYS_futex, futexp, FUTEX_WAKE, 1, NULL, NULL, 0);
            if (unlikely(s == -1)) {
                printf("unexpected error...\n");
                pthread_exit(NULL);
            }
            break;
        }
    }
}

static void* routine(void *arg)
{
    printf("start routine...\n");

    for (int i = 0; i < 1000000; i++) {
        futex_lock(&futex_var);
        sum++;
        futex_unlock(&futex_var);
    }

    return NULL;
}

int main ()
{
    long nthreads = sysconf(_SC_NPROCESSORS_ONLN);

    printf("number of threads: %ld\n", nthreads);

    pthread_t threads[nthreads];
    for (int i = 0; i < nthreads; i++) {
        if (pthread_create(&threads[i], NULL, &routine, NULL)) {
            printf("%s\n", strerror(errno));
            exit(-1);
        }
    }

    for (int i = 0; i < nthreads; i++) {
        pthread_join(threads[i], NULL);
    }

    printf("%d\n", sum);
}

All testing under 2 threads:

futex:

real	0m0.663s
user	0m0.277s
sys	0m0.876s

---

mutex:

real	0m0.149s
user	0m0.161s
sys	0m0.085s

---

atomic:

real	0m0.034s
user	0m0.026s
sys	0m0.001s