Primerjava paralelizacij algoritmov za procesiranje slik v OpenCL

KENDA, BOJANA (author), Bulić, Patricio (mentor)

V pričujoči nalogi smo raziskovali vpliv različnih nastavitev pri poganjanju programa za procesiranje slik z OpenCL ogrodjem. Vzeli smo tri konkretne metode iz programskega orodja GEGL, ki se tipično pojavljajo pri obdelavi slik, in primerjali njihovo hitrost izvajanja tako pri slikah različnih velikosti kot tudi na različni strojni opremi. Eno izmed metod smo poleg tega predelali za izrabo lokalnih pomnilnikov na OpenCL napravi in ugotavljali morebitne prednosti in slabosti takšne implementacije. V teoretičnem delu smo se v ta namen posvetili pregledovanju OpenCL arhitekture in možnostih, ki jih ponuja paralelno oziroma vzporedno računanje.

• Language: Slovenian
• Keywords: OpenCL, paralelnost, delovne skupine, lokalni pomnilnik
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2015
• PID: 20.500.12556/RUL-30578
• Publication date in RUL: 18.02.2015

Secondary language

• Language: English
• Title: Comparison of parallelizations of the algorithms for image processing in OpenCL

Abstract

The thesis explores the effects of running an image processing program with the OpenCL framework at different settings. Three methods, common in image processing, have been selected from the programming library GEGL and their execution times compared when using images of different sizes as well as on different hardware. In addition, one of the methods has been modified to use local memory on an OpenCL device, allowing us to identify the potential advantages and disadvantages of such an implementation. The theoretical part has therefore been dedicated to the overview of OpenCL architecture and the possibilities opened by parallel, or rather concurrent, computing.

• Keywords: OpenCL, parallelism, work groups, local memory