Análise comparativa de métodos de carregamento de dados em GPU para multiplicação de matriz-vetor

• Main Author: Bossoni, Giovani
• Publication Date: 2023
• Other Authors: Silva Junior, José Moscardi da
• Format: Bachelor thesis
• Language: por
• Source: Repositório Institucional da UTFPR (da Universidade Tecnológica Federal do Paraná (RIUT))
• Download full: http://repositorio.utfpr.edu.br/jspui/handle/1/36581
• topic: Programação paralela (Computação); Unidades de processamento gráfico; Algorítmos paralelos; Gerenciamento de memória (Computação); Parallel programming (Computer science); Graphics processing units; Parallel algorithms; Memory Management (Computer science); CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO

With the introdution of multicores processors, there is a new need to develop new algorithms capable of taking advantage of the parallelism from the new Hardware. This project porpose a comparative analysis between methods for data transfer between CPU and GPU and the impact on perfomance on computation of matrix-vector multiplication on CUDA plataform, Using three approaches: common, Unifed Memory and Streams, where the last one has shown gains from 10% to 35% in scenarios where there is a large amount of data.