• JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
  • JoomlaWorks Simple Image Rotator
 
  Bookmark and Share
 
 
Master's Dissertation
DOI
https://doi.org/10.11606/D.45.2016.tde-20230727-113441
Document
Author
Full name
Nils Urmersbach
E-mail
Institute/School/College
Knowledge Area
Date of Defense
Published
São Paulo, 2016
Supervisor
Title in Portuguese
A Classic Linear System Solver on Modern Hardware Architecture for Sparse Systems
Keywords in Portuguese
Métodos Numéricos
Sistemas Lineares
Abstract in Portuguese
Nesse trabalho apresentamos as nossas implementações do Método de Jacobi para sistemas lineares esparsos gerais no formato de Compressed Sparse Row (CSR) usando OpenMP, OpenACC e CUDA. Aplicamos essas implementações no sistema linear derivado da discretização de diferenças finitas centrais da Equação de Poisson em duas dimensões em domínios retangulares e comparamos o desempenho das implementações de CSR com o desempenho de um solver direto da Equação de Poisson usando o estêncil de cinco pontos. Para nosso estudo de caso nós consideramos cinco tamanhos diferentes de malhas (com até 223C67.1 milhões desconhecidos), ambos precisão simples e dupla, e uma variedade de números de threads para a implementação de OpenMP, resultando em 300 configurações diferentes executadas para esse trabalho. Nós discutimos o comportamento de escalagem das implementações diferentes e apresentamos alguns resultados de perfilamento dos nossos programas paralelizados.
Title in English
Um solver de sistemas lineares clássico na arquitetura moderna de hardware para sistemas esparsos
Abstract in English
In this work we present our implementations for the Jacobi Method for general sparse linear systems in the Compressed Sparse Row (CSR) format using OpenMP, OpenACC and CUDA. We apply these implementations to the linear system derived from the central finite difference discretization of the two- dimensional Poisson Equation on rectangular domains, and compare the performance of the CSR imple- mentations to the performance of a direct Poisson Equation solver using the five-point stencil. For our case study, we consider five different grid size (with up to 223C67.1 million unknowns), both in single precision and double precision, and a variety of thread numbers for the OpenMP implementation, resulting in 300 different configurations in total that were executed for this work. We discuss the scaling behaviour of the different implementations and present some profiling results of our parallelized programs.
 
WARNING - Viewing this document is conditioned on your acceptance of the following terms of use:
This document is only for private use for research and teaching activities. Reproduction for commercial use is forbidden. This rights cover the whole data about this document as well as its contents. Any uses or copies of this document in whole or in part must include the author's name.
UrmersbachNils.pdf (2.11 Mbytes)
Publishing Date
2023-07-27
 
WARNING: Learn what derived works are clicking here.
All rights of the thesis/dissertation are from the authors
CeTI-SC/STI
Digital Library of Theses and Dissertations of USP. Copyright © 2001-2024. All rights reserved.