/* author: Guochun Shi <gshi@ncsa.uiuc.edu>
 */
#include <ctype.h>
#include <sys/time.h>
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
#include <libspe2.h>
#include <pthread.h>
#include <cbe_mfc.h>

extern spe_program_handle_t simple_spu;

double gettime(void)
{
    struct timeval t;
    double result;
    
    if (gettimeofday(&t, NULL) < 0){
	fprintf(stderr, "ERROR: gettimeofday() failed\n");
	return -1;
    }
    
    result = t.tv_sec +  t.tv_usec*0.000001;
    return result;
}

#define MAX_THREADS 16
pthread_t spe_pthread_ids[MAX_THREADS];
spe_context_ptr_t spe[MAX_THREADS];
spe_spu_control_area_t* ps_area[MAX_THREADS];


void*
my_spe_thread(void* _arg){

    spe_context_ptr_t specontext = (spe_context_ptr_t)_arg;
    unsigned int runflags = 0;
    unsigned int entry = SPE_DEFAULT_ENTRY;

    spe_context_run(specontext, &entry, runflags, NULL, NULL, NULL);
    pthread_exit(NULL);

    return NULL;

}


int main(int argc, char** argv)
{
    int i;
    double t0, t1;
    int ite;
    int iterations = 1000000;
    int rc;
    int n = 8;

    if (argc >=2 ){
 	n = atoi(argv[1]);
	if (n <= 0){
		printf("invalid spe numbers\n");
		exit(1);
	}	
    }
    for ( i=0;i <n; i++){
    	spe[i] = spe_context_create( SPE_MAP_PS, NULL);
    	spe_program_load(spe[i], &simple_spu);
   	rc =  pthread_create(&spe_pthread_ids[i], NULL, &my_spe_thread, spe[i]);
    	if (rc != 0){
         	perror("Unable to create pthread for spe");
         	return -1;
    	}
   	ps_area[i] = (spe_spu_control_area_t *)spe_ps_area_get(spe[i], SPE_CONTROL_AREA);

    }

    t0 = gettime();

    	/* now start the latency test*/
    for (ite =0; ite < iterations ; ite++){
    	for (i =0;i < n; i++){
   	 	_spe_in_mbox_write(ps_area[i], 0);
    	}
        for (i =0;i < n; i++){
                _spe_out_mbox_read(ps_area[i]);
        }
	
    }

     t1 = gettime();
     printf("latency=%.2f us\n", 1000000*(t1 - t0)/iterations);	


    return (0);
}