A New Space Partitioning for Mapping Computations of the Radiosity Method onto a Highly Pipelined Parallel Architecture

Abstract

Despite the fact that realistic images can be generated by ray-tracing and radiosity shading, these techniques are impractical for scenes of high complexity because of the extremely high time cost. Several attempts have been made to reduce image synthesis time by using parallel architectures, but they still suffer from communication problems. In this paper, we present a new space partitioning which is adaptive to the local environment seen by a bundle of rays. Two tracking mechanisms are embedded to guarantee adaptation. When using a shared memory parallel architecture, the communication load between the host and the PEs can be alleviated with this approach. Furthermore, the partitioning provides a better balancing between processing throughput and I/O bandwidth which will enhance the pipelinability of computations, especially when a high speed cache memory is allowed for each PE. Combining those factors, a highly pipelined parallel architecture can be used to accelerate computations in ray-tracing and radiosity methods. The technique has been tested on different scenes with randomly generated patches in a 2D setting. When compared with the conventional technique, promising results have been observed. This technique can be easily extended to 3D.