Real-time solid voxelization using multi-core pipelining
Simon Berkovich, Duoduo Liao · 2009
This dissertation not only proposes a new approach for real-time solid voxelization by Graphics Processor Unit (GPU) acceleration, but also proposes a novel multi-core pipelined parallel architecture to further improve total performance. Especially, the new generic parallel architecture can also be successfully applied to other multi-core/many-core architectures or heterogeneous systems. The most important advantage of this new architecture is that most existing single-processor software programs can be directly used with nearly no change, unlike conventional parallel programming which has to take into account data distribution, load balancing, and scheduling. The algorithms and architectures introduced in this dissertation not only improve performance and quality, but also are convenient to implement and integrate into many applications. The experimental results, performance analysis, comparison, and case studies demonstrate the effectiveness, flexibility, and diversity of developed approaches and architecture systems. The GPU accelerated real-time solid voxelization approach utilizes an innovative dynamic slice function mechanism and masking techniques to significantly improve solid voxelization speed in real time as well as create various multi-valued solid volumetric models. In particular, by dynamically applying different slice functions, any surface-closed geometric model can be voxelized into a solid volumetric representation with various multi-valued interior features, such as rainbow, marble, wood, or translucent jade. The algorithms are easy to implement and convenient to integrate into diverse applications, such as volume modeling, collision detection, volume rendering, medical simulation, and computer art. An effective scheme for real-time parallel solid voxelization is presented using multi-core pipelining. A novel generic multi-core pipelined parallel architecture is first proposed based on program slicing by forced interrupts and a distinct memory/cache dynamic management technology. A new parallel solid voxelization system is discussed based on this novel architecture extended to multi-core GPUs. These architectures provide simple and effective solutions to on-the-fly computations by transferring the operating states from processor to processor. They require practically the same software as is currently used on a single-CPU system (i.e. a sequential computer), unlike conventional parallel programming costing extra time and effort. The parallel solid voxelization experimental results, performance analysis, and comparison with existing multi-core GPUs show the effectiveness of the new architectures. At last, the two important volume graphics applications based on the real-time GPU-accelerated solid voxelization, volumetric Constructive Solid Geometry (CSG) modeling and volumetric collision detection, are further studied in detail. First, a new algorithm of fast volumetric CSG modeling for volume scene by scene computation is proposed. It is capable of carrying out interactive modeling of heterogeneous volume scene. Second, a simple and fast collision detection method is presented as one of the successful applications using this core voxelization approach. It improves the collision detection performance as well as detection accuracy based on voxel level.