Methods for performance estimation of the block algorithm implementations on GPU-based open heterogeneous systems

 

E. G. Andrianova, D. I. Mirzoyan, A. B. Petrov
 

Moscow State Technical University of Radioengineering, Electronics and Automation (MSTU MIREA)

 

The paper is received on March 20, 2014

 

Abstract: Much attention is paid to optimizing the implementation when building high-performance systems. Open heterogeneous systems are providing high levels of efficiency in a certain range of applications. Some tasks from that range do not have any approved methods for optimization and modification, i.e. tasks of block symbolic data processing.  Proofing the procedure of implementation’s efficiency estimation looks like a very significant step to build such methods.  This procedure is the most important thing when building automated or automatic optimization instruments implementing such methods.

Key words: open heterogeneous systems, mathematical evaluation of the effectiveness of implementation, automated or automatic optimization instruments implement, block symbolic data processing.