Parallel occlusion culling on GPUs cluster

Abstract

Occlusion culling, level-of-detail and parallel rendering are key techniques to accelerate interactively rendering of large dataset. Although each of these techniques has been independently researched to an extensive degree and some systems have been developed to combine them together. Unfortunately, parallel occlusion culling, i. e. distributing the computation of occlusion culling onto multiple computing nodes such as a CPUs cluster or GPUs cluster, has rarely been touched partly because most of existing occlusion culling algorithms do not scale well when parallelized.To address this issue, we first introduce a novel occlusion culling algorithm which uses occlusion query function provided by current GPUs and adopts a visibility predictor based on temporal coherence to reduce the count of occlusion queries. Furthermore, different schemes to parallelize this occlusion culling algorithm on a GPUs cluster are proposed including data parallelism schemes and functionality parallelism schemes. Data parallelism scheme decomposes data for occlusion culling into separate parts and maps the computation of them onto different cluster nodes for parallel processing. Functionality parallelism scheme assembles multiple cluster nodes as an occlusion culling pipeline which outputs steady image stream. We show a number of viable solutions to the special problems on parallelizing this occlusion culling algorithm, such as the transferring of data dependency and the load-balancing of occlusion culling pipeline. We report on the experimental results to demonstrate the efficiency of the proposed parallelism schemes based on the novel occlusion culling algorithm.