HI, could you tell me when do the GPU programming, to write a KERNEL according to their characteristics, how to understand the data partition or is close to the optimal task parallel of performance?


Example:

 
Const int ARRAY_SIZE=1000; 
Size_t globalWorkSize [1]={ARRAY_SIZE}; 
Size_t localWorkSize [1]={1}; 

//the Queue the kernel up for execution across indicates the array 
ErrNum=clEnqueueNDRangeKernel (commandQueue, the kernel, 1, NULL, 
GlobalWorkSize localWorkSize, 
0, NULL, NULL); 
If (errNum!=CL_SUCCESS) 
{
STD: : cerr & lt; <"Error queuing kernel for execution." & lt; Cleanup (context, commandQueue, program, kernel, memObjects); 
return 1; 
} 

Size_t globalWorkSize [1]={ARRAY_SIZE};
Size_t localWorkSize [1]={1};

How to set up in order to achieve optimal?