source: CIVL/examples/opencl/2.15/reduce_int_kernel.cvl

#ifndef GROUP_SIZE
#define GROUP_SIZE (64)
#endif

#ifndef OPERATIONS
#define OPERATIONS (1)
#endif

////////////////////////////////////////////////////////////////////////////////////////////////////

#define LOAD_GLOBAL_I1(s, i) \
    ((  const int*)(s))[(size_t)(i)]

#define STORE_GLOBAL_I1(s, i, v) \
    ((  int*)(s))[(size_t)(i)] = (v)

////////////////////////////////////////////////////////////////////////////////////////////////////

#define LOAD_LOCAL_I1(s, i) \
    ((  int*)(s))[(size_t)(i)]

#define STORE_LOCAL_I1(s, i, v) \
    ((  int*)(s))[(size_t)(i)] = (v)

#define ACCUM_LOCAL_I1(s, i, j) \
{ \
    int x = ((  int*)(s))[(size_t)(i)]; \
    int y = ((  int*)(s))[(size_t)(j)]; \
    ((  int*)(s))[(size_t)(i)] = (x + y); \
}
////////////////////////////////////////////////////////////////////////////////////////////////////
/*
    __global int *output, 
    __global const int *input, 
    __local int *shared,
    const unsigned int n
*/
void reduce(
  int workgroup, 
  int cl_global_id, 
  int cl_local_id,
  int *output, 
  const int *input, 
  int *shared,
  const unsigned int n)
{
  const int zero = 0.0f;
  const unsigned int group_id = workgroup;
  const unsigned int group_size = GROUP_SIZE;
  const unsigned int group_stride = 2 * group_size;
  const size_t local_stride = group_stride * group_size; 
    
  unsigned int op = 0;
  unsigned int last = OPERATIONS - 1;
  for(op = 0; op < OPERATIONS; op++)
  {
    const unsigned int offset = (last - op);
    const size_t local_id = cl_local_id + offset;

    STORE_LOCAL_I1(shared, local_id, zero);
        
    size_t i = group_id * group_stride + local_id; 
    while (i < n)
    {
      int a = LOAD_GLOBAL_I1(input, i);
      int b = LOAD_GLOBAL_I1(input, i + group_size);
      int s = LOAD_LOCAL_I1(shared, local_id);
      STORE_LOCAL_I1(shared, local_id, (a + b + s));
      i += local_stride;
    }
        
  #if (GROUP_SIZE >= 2) 
    if (local_id <   1) { ACCUM_LOCAL_I1(shared, local_id, local_id +   1); }
  #endif
  }

        
        
  if (cl_local_id == 0)
  {
    int v = LOAD_LOCAL_I1(shared, 0);
    STORE_GLOBAL_I1(output, group_id, v);
  }        
}