#include <omp.h>
float e;
// 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;
}
int main(int argc, char *argv[]) {
int n = ...;
float *x, *y;
x = new float[n+1];
y = new float[n+1];
... // fill x, y
e = 0;
#pragma omp parallel
{
int p = omp_get_thread_num();
int num = omp_get_num_threads();
#define BLOCK (8190)
float xStream<BLOCK>;
float xoutStream<BLOCK>;
float yStream<BLOCK+2>;
float einStream<BLOCK+2>;
float eStream<1>;
int n0 = 1+((n-1)*p)/num;
int n1 = 1+((n-1)*(p+1))/num;
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;
}
#pragma omp atomic
e += ee;
}
... // output x, e
delete[] x, y;
return 0;
}
|