Parallel Progressive Radiosity with Adaptive Meshing

Progressive radiosity is widely used for realistic image synthesis in computer graphics applications. High-quality image generation usually requires radiosity with adaptive patch refinement to account for global illumination effects from irregular objects whose proximity varies in a 3D space. Parallelizing such an algorithm is difficult since computation cost for each object varies from one iteration to another depending on the location of dynamically selected shooting patches. Dynamic load balancing is required but its overhead is high for distributed memory systems. This paper presents an efficient parallel algorithm for progressive radiosity, which adopts a static processor assignment strategy to take advantages of hierarchical computation structure in this problem, minimize communication and balance dynamic load. Our experiments on a Meiko CS-2 distributed memory machine show that this algorithm has achieved good speedup for the tested cases.