Advanced Light Transport Algorithms on Heterogeneous Platforms: Evaluation of the DICE Framework

Student: César Perdigão
Supervisor: Luís Paulo Santos

Abstract

Efficient physically based rendering is still an actual challenge due to both the wide variety of available algorithms and their heavy computational demands. The performance of such techniques can be enhanced by combining two complementary approaches: using the most accurate and robust light transport algorithms and an efficient use of all available computing resources. Although most modern computers have both powerful multicore CPU devices coupled to manycore GPU devices, most current implementations of light transport algorithms only address one of these types of computing units and thus not using all available resources efficiently. This dissertation work addresses these efficiency issues by evaluating the most recent global illumination algorithms based on path space integration, and evaluating a framework that automatically takes full advantage of an heterogeneous computing system, based on CPU and GPU devices: the DICE framework. The outcomes from our scalability test on such heterogeneous computing platforms show promising results and a path worth pursuing.

Thesis Download (PDF)