Space Partitioning for Mapping RadiosityComputations onto a Pipelined Parallel Architecture (II)

Abstract

A new space partitioning technique is elaborated. In part I of the paper [3], we proposed a shell-like structure which is to be superimposed on a uniform grid data structure and is adaptive to the local environment seen by a bundle of rays. Here we extend this segmentation by embedding it in a static partitioning which is determined by low resolution ray casting. This partitioning is useful in achieving a balanced computation while mapping it onto a pipelined parallel architecture. Moreover, a run-time control of workloads is applied during a subsequent high resolution ray casting so as to adjust low resolution partitioning. The technique has been tested on practical and randomly generated scenes. The performance evaluation of a pipelined parallel architecture has been done by queueing network simulation. Promising results have been obtained.