nemo_main()
{
int n=getiparam("n");
int iter=getiparam("iter");
int m=getiparam("m");
int seed = init_xrandom(getparam("seed"));
int i,j,k,l;
real *x, sum;
real t0,t1,t2;
init_timers(100);
stamp_timers(0);
x = (real *) allocate(n*sizeof(real));
for (i=0; i<n; i++) /* init the whole array */
x[i] = xrandom(0.0,1.0);
for (i=0; i<m; i++) /* cache it in again ? */
x[i] = xrandom(0.0,1.0);
sum = 0.0;
t0 = cputime();
stamp_timers(1);
if (m==0) { /* do it in one sweep, the N^2 algorithm */
stamp_timers(2);
for (l=0; l<iter; l++)
for (j=0; j<n; j++)
for (i=0; i<n; i++)
sum += FUN(x[i],x[j]);
stamp_timers(3);
} else { /* N/M times a small M*M patch that may be in cache */
stamp_timers(2);
for (l=0; l<iter; l++)
for (k=0; k<n-m; k++)
for (j=k; j<k+m; j++)
for (i=k; i<k+m; i++)
sum += FUN(x[i],x[j]);
stamp_timers(3);
}
stamp_timers(4);
t1 = cputime();
if (m)
printf("%d %d %d sum=%lg Mops=%lg\n",
n,iter,m,sum,iter*m*m*n/(t1-t0)/60.0/1e6);
else
printf("%d %d %d sum=%lg Mops=%lg\n",
n,iter,m,sum,iter*n*n/(t1-t0)/60.0/1e6);
stamp_timers(5);
printf("%Ld %Ld %Ld %Ld %Ld\n",
diff_timers(0,1),
diff_timers(1,2),
diff_timers(2,3),
diff_timers(3,4),
diff_timers(4,5));
}
double run_experiment_ji ( mtype_t * matrix , mtype_t scalar , int buffer_size ) {
int i, j, iter;
hptimer_t start , end;
double final_time ;
start = get_time ();
for ( iter = 0; iter < MAX_ITERS ; ++ iter ) {
for (j = 0; j < buffer_size ; ++j ) {
for (i = 0; i < buffer_size ; ++i ) {
matrix [ GET_INDEX (i, j, buffer_size ) ] = scalar * matrix [ GET_INDEX (i, j, buffer_size ) ];
}
}
}
end = get_time ();
return final_time = diff_timers (start , end) / MAX_ITERS ;
}
