Avaliação do desempenho de bibliotecas para paralelização em arquiteturas multicore: pThread e OpenMP

Abstract

This monograph aims to highlight and to evaluate two of the most popular parallelization libraries nowadays, pThreads and OpenMP. Both of them were used to parallelize a digital image processing algorithm. With the advances in area of computational technology, accessibility to information on the internet and be able to use multicore machines in the comfort of your home or wherever you are, it becomes possible to apply parallel computing as an ally for improvement in processing time of certain applications. Inserted in this context, this work evaluate and analyse the performance of both libraries previously cited. The algorithm uses a pixel convolution process to enhance borders and textures patterns, to analyse the image fragmentation index of the processed image. The execution time of the serial version of the algorithm tooked about 43 minutes of processing in a example using a input mask value of 15X15 pixels over and 8460x9530 pixels image. Ina more complex context, it could be used dozens or hundreds of images like these, and the time spent to process them could take days. Were implemented seven parallel versions, which one of them reached the result that was 6times faster compared to the serial version processing time.