Archives Discussions

[mushucf]

I need to compute a function f(x,y, z, c1, c2) where x, y, z are vectors and c1 and c2 are constants and I need to repeat this calculus for many values of c1 and c2 (which are already known and with the same dimension). My question is: what is the most efficient way to write down this into an OpenCL kernel?

Sequential algorithm would be as follows:

for (int i = 0; i < c1.length;i++)

{

  for(int j = 0;j < x.length;j++)

  {

     f = x * cos(y + z) / (c1 + c2);

  }

}

Thanks in advance.

[nou]

something like this

__kernel void my_kernel(__global int *x,__global int *y,__global int *z,__global int *c1, __global int *c2, __global int *f)
{
     int i = get_global_id(0);
     int j = get_global_id(1);
     int w = get_gloval_size(0);
     f[i*w+j] = x * cos(y + z) / (c1 + c2);
}

just look for some tutorials on OpenCL.