Optimization of the Tile Size for Parallelizing Nested Loops Using the Genetic Algorithm

Abstract

The method of dividing nested cycles into tiles and the problem of finding the optimal tile size using a genetic algorithm are considered. It is proposed to use the software package PLUTO - a tool for transforming nested loops. This package does not introduce mechanisms for finding the optimal tile size. During the development of the optimization system, the existing solutions to this problem were considered. The results of the system on. As a result of the system, measurements of the execution time of test programs before optimization and after were presented. The programs were taken on the basis of a test collection of widely used algorithms of different classes of PolyBench.