Implementation of computational geometry algorithms on CUDA

Anže Škerjanc (2013) Implementation of computational geometry algorithms on CUDA. EngD thesis.

Preview

PDF Download (1557Kb)

Abstract

The aim of the thesis is implementation of certain algorithms in computational geometry on the CUDA architecture. We present an overview of CUDA architecture, libraries and approaches of solving the algorithms and their time complexity. The main focus of the thesis is on searching (finding) the nearest pair and convex hull. The benefits and disadvantages of the implementation of algorithms on the graphics card against the central processing unit are also introduced. The algorithm search of the closest pair finds two points in the set, that have the nearest value of the Euclidean distance. The algorithm of searching the convex hull finds the shortest possible point connection, so that all points are on the inside of the convex hull or on it. The algorithms were implemented on processor as well as on graphic card. Our set goal in this thesis was to provide concrete answers to the speedup factor in algorithm implementation on CUDA architecture. The answers are solely based on practical performance tests.

Actions (login required)