Serialization of work items inside a kernel function - OpenCL

Question

I'm working on generate hadware modules (Descriptions on Verilog) from kernels of openCL. I was reading an article about this, and the authors do some pre processing to the kernels, the first step of the pre processing is : Serialization of work items inside a kernel function, they explained like this:

"Work-items inside a work-group can be executed in any sequence, provided that no synchronization operation is present inside a kernel function. Based on this observation, we serialize the execution of work-items by enclosing the instructions in the body of a kernel function into a triple nested loop, given that the maximum number of dimensions in the abstract index space within a workgroup is three. Each loop nest enumerates the work-items in the corresponding dimension, thus serializing their execution."

And they provide an example like this:

int __kernel_indices[3];

__kernel_indices[2] = 0;
while (__kernel_indices[2]< __local_size[2]){
   __kernel_indices[1] = 0;
   while (__kernel_indices[1]< __local_size[1]){
      __kernel_indices[0] = 0;
      while (__kernel_indices[0]< __local_size[0]){

               //Kernel Body

      }
      __kernel_indices[0]++;
   }
   __kernel_indices[]++;
}
__kernel_indices[2]++;

I don't understand the meaning of serialization, does one work item is executed at the time? or the work items from only 1 work gruop will be in execution at the time?

Thanks for reading the post and sorry for my bad english.

Ani Ani · Accepted Answer · 2012-07-25T22:06:48

Yes, the kernel above makes different work items execute in sequence rather than run together. To me it looks like an increasing number of work items begin executing as the while loops run. To understand this, think of each work-item as having a three-dimensional index.

Serialization of work items inside a kernel function - OpenCL

1 Answers