memory optimization group organization optimization

Discussion created by hayabusaxps on Aug 9, 2011
Latest reply on Aug 20, 2011 by hayabusaxps
which is faster, between organization of the local into workgroups vs the global given there is a large program

lets say there is a global index of 128 by 128 by 128 cube matrix, there will be about 20 calculations on each of the 128^3 =2097152 values then 1 stored value, and this process will repeated 10,000 times.

because im working with a huge number of calculations what i need to know is, is it benifietial to organize the cube matrix into workgroups and work-items in local memory


should i just allow opencl to handle it all in global, which losses the fast local speeds but doesnt waste time with thousands of data copies?