#include <pthread.h>
#include <xmmintrin.h>
float *x, *y;
#define PROC (8)
float e_vec[PROC];
int n_thread0, n_thread1;
void *thread1(void *arg) {
int p = (int)arg;
int n0 = n_thread0 + (p * (n_thread1-n_thread0)) / PROC;
int n1 = n_thread0 + ((p+1) * (n_thread1-n_thread0)) / PROC;
// do computation
float ve[4] = {0, 0, 0, 0};
float e = 0;
for (int i=n0; i<n1; i+=4) {
float half =.5;
_mm_store_ps(&x[i],
_mm_mul_ps(_mm_load1_ps(&half),
_mm_add_ps(_mm_loadu_ps(&y[i+1]),
_mm_loadu_ps(&y[i-1]))));
_mm_store_ps(&ve[0],
_mm_add_ps(_mm_load_ps(&ve[0]),
_mm_mul_ps(_mm_load_ps(&y[i]),
_mm_load_ps(&y[i]))));
}
e += ve[0] + ve[1] + ve[2] + ve[3];
e_vec[p] = e;
return (void*) 0;
}
int main(int argc, char *argv[]) {
int n = ...;
x = new float[n+1];
y = new float[n+1];
... // fill x, y
pthread_t threads[PROC];
pthread_attr_t attr;
pthread_attr_init(&attr);
n_thread0 = 1;
n_thread1 = n;
float e = 0;
// start threads and wait for termination
for (int p=0; p<PROC; ++p)
pthread_create(&threads[p], &attr, thread1, (void *)p);
for (int p=0; p<PROC; ++p) {
pthread_join(threads[p], NULL);
e += e_vec[p];
}
... // output x, e
delete[] x, y;
return 0;
}
|