Uso de operações SIMD em uma biblioteca de algoritmos bio-inspirados

Abstract

Optimization algorithms seeks to resolve complex problems. A strategy to algorithmically model a problem is to use concepts of Natural Computing, also known as Bio-inspired Computing. Bio-inspired algorithms simulate the ability of learn or adapt a smart group of individuals subjected to a complex environment and thus optimize initial solution in order to find optimal solutions in a polynomial time. In this way, this work proposes an open source, bio-inspired, and parallel library. This library is composed by Artificial Bee Colony (ABC), Ant Colony Optimization (ACO) and Particle Swarm Optimization (PSO) algorithms developed in C++ programming language. The OpenMP API was used for the parallelization of algorithms. The work tests parallel and vectorization directives execution by comparing the performance of individual and combined directives. The vectorization directive execution obtained the highest number of tests with performance gain. This directive was applied to the selected algorithms of the library. After the parallel implementation, the algorithms were submitted to the classic test functions Alpine, Booth, Easom, Griewank, Rastrigin, Rosenbrock and Sphere, being applied to a population of 10 to 100 agents. The results showed that the sequential and parallel versions obtained the same average solution and that, for at least one population test, all functions obtained performance gains. With this, we prove that the parallel version of the developed algorithms maintains the quality of the solution and reduces the execution time.