#include <pthread.h>
float *x, *y;
#define PROC (8)
float e_vec[PROC];
int n_thread0, n_thread1;
// kernel
kernel void sub1(float x[], float y[],
out float xout<>, out float eout<>)
{
// do computation
float i = 1 + indexof(xout).x;
xout = x[i] + (y[i+1] + y[i-1])*.5f;
eout = y[i] * y[i];
}
reduce void sub2(float y<>, reduce float e<>)
{
e += y;
}
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;
#define BLOCK (8190)
float xStream<BLOCK>;
float xoutStream<BLOCK>;
float yStream<BLOCK+2>;
float einStream<BLOCK+2>;
float eStream<1>;
float e = 0;
float ee = 0;
for (int i=n0; i<n1; i+=BLOCK) {
streamRead(xStream, x+i);
streamRead(yStream, y+i-1);
sub1(xStream, yStream, xoutStream, einStream);
streamWrite(xoutStream, x+i);
sub2(einStream, eStream);
float e_local;
streamWrite(eStream, &e_local);
ee += e_local;
}
e += ee;
e_vec[p] = e;
return (void*) 0;
}
int main(int argc, char *argv[]) {
int n = ...;
float *x, *y;
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;
}
|