Accelerating MATLAB Performance

378 Accelerating MATLAB

®

Performance

We can now rewrite the example code from §7.3.1 using OpenMP constructs. Notice how

compact and easy-to-understand the new algorithm looks:

#include <mex.h>

#include <omp.h>

#include <gmp.h>

/* parallel compute function */

void do_calculation(unsigned *g_in, mxArray *g_out, unsigned n_thids)

{

// start parallel section with n_thids threads

#pragma omp parallel num_threads(n_thids)

{

int tid = omp_get_thread_num();

long tid = (long)t; // get this thread's ID

unsigned *in = g_in + tid*2; // and the inputs 'n' and 'k'

mpz_t z; // initialize a GMP large integer

mpz_init(z);

// calculate binomial coefficient

mpz_bin_uiui(z, (unsigned long)in[0], (unsigned long)in[1]);

// estimate the ...