/* 
 * Written by Josh Dybnis and released to the public domain, as explained at
 * http://creativecommons.org/licenses/publicdomain
 */
#include <stdio.h>
#include <errno.h>
#include <pthread.h>
#include <sys/time.h>

//#define USE_MFENCE
#define NUM_ITERATIONS 10000000

static volatile int wait_;
static volatile int count_ = 0;
static volatile int guard0_ = 0;
static volatile int guard1_ = 0;

void *worker0 (void *arg) {
    __sync_fetch_and_add(&wait_, -1);
    do {} while (wait_); // Wait for all the worker threads to be ready.

    int i;
    for (i = 0; i < NUM_ITERATIONS; ++ i) {
        guard0_ = 1;
        // prevent re-ordering of the above store (guard0_ = 1) and subsequent loads
#ifdef USE_MFENCE
        __asm__ __volatile__("mfence");
#else
        { int dummy = guard0_; dummy = dummy; } 
#endif
        while (guard1_) {
            __asm__ __volatile__("rep;nop");
        }
        count_ = count_ + 1;
        guard0_ = 0;
    }

    return NULL;
}

void *worker1 (void *arg) {
    __sync_fetch_and_add(&wait_, -1);
    do {} while (wait_); // Wait for all the worker threads to be ready.

    int i;
    for (i = 0; i < NUM_ITERATIONS; ++ i) {
        guard1_ = 1;
        // prevent re-ordering of the above store (guard1_ = 1) and subsequent loads
#ifdef USE_MFENCE
            __asm__ __volatile__("mfence");
#else
        { int dummy = guard1_; dummy = dummy; }
#endif
        while (guard0_) {
            guard1_ = 0;
            // prevent re-ordering of the above store (guard1_ = 0) and subsequent loads
#ifdef USE_MFENCE
            __asm__ __volatile__("mfence");
#else
            { int dummy = guard1_; dummy = dummy; }
#endif
            while (guard0_) {
                __asm__ __volatile__("rep;nop");
            }
            guard1_ = 1;
            // prevent re-ordering of the above store (guard1_ = 1) and subsequent loads
#ifdef USE_MFENCE
            __asm__ __volatile__("mfence");
#else
            { int dummy = guard1_; dummy = dummy; }
#endif
        }
        count_ = count_ + 1;
        guard1_ = 0;
    }

    return NULL;
}

int main (int argc, char **argv) {

    struct timeval tv1, tv2;
    gettimeofday(&tv1, NULL);
    wait_ = 2;

    int rc;
    pthread_t thread0, thread1;
    rc = pthread_create(&thread0, NULL, worker0, NULL);
    if (rc != 0) { perror("pthread_create"); return rc; }

    rc = pthread_create(&thread1, NULL, worker1, NULL);
    if (rc != 0) { perror("pthread_create"); return rc; }

    pthread_join(thread0, NULL);
    pthread_join(thread1, NULL);

    gettimeofday(&tv2, NULL);
    int ms = (int)(1000000*(tv2.tv_sec - tv1.tv_sec) + tv2.tv_usec - tv1.tv_usec) / 1000;
    printf("Count: %d Time:%dms\n\n", count_, ms);
    fflush(stdout);

    return 0;
}