cl_out_of_resources error in atom_inc/atomic_inc

Question

Hi everyone,

i am using atom_inc in one of my kernels. More or less like this:

kernel exampleKernel(
  global volatile int * x, 
  global const int   maxX,
  global const int * buf1, 
  global const int * buf2
  )
{
  if(x < maxX)
  {
    int y = atom_inc(x);
    buf2[y] = buf1[get_global_id(0)];
  }
}

But it gives me an CL_OUT_OF_RESOURCES error when i call it. Replacing atom_inc(x) with 0 resolves this problem ... but makes the kernel useless.

Any ideas what can cause that error?

Edit: i have included these extensions in the program:

#pragma OPENCL EXTENSION cl_khr_global_int32_base_atomics : enable
#pragma OPENCL EXTENSION cl_khr_global_int32_extended_atomics : enable

Edit2: actually i initialize x with:

kernel.setArgs(0,0);

Do i have to use a memory object in this place?

Kind Regards, Florian

Answer 1

Ah ... ok ... it looks like you have to use memory objects in this case.

The solution here was to create a one element OpenCL buffer and set it as the first argument.

I'll leave this question and answer here so that others might not make the same fault.