April 24th Friday (四月　二十四日　金曜日)

How to make a program dispatch enough number of threads in order to adapt multiple CPUs? There is a simple function to
solve it about loop logic.

/** get the number of thread that iterator or loop.
From interator number and CPU number, and one thread's loop number,
compute out thread number, and make sure that maximum thread number
is not more than the number of CPUs.

@param int n interator / loop number
@param int min_n the minimum loop number in one thread
@return int thread number
*/
int dtn(int n, int min_n)
{
int max_tn = n / min_n;
int tn = max_tn > g_ncore ? g_ncore : max_tn; //tn - thread number we need.
if ( tn < 1 )
{
tn = 1;
}
return tn;
}

#pragma omp parallel for num_threads(dtn(n, MIN_ITERATOR_NUM))
for ( i = 0; i < n; i++ )
{
printf("Thread Id = %ld/n", omp_get_thread_num());
//Do some work here
}