Dynamic acceleration structures for the visualization of time-dependent volume data on the GPU

Abstract

Volume rendering extensive 3D data is requiring much computing power and doing this efficiently has been a challenge. To build a data structure of which rendering can take advantage is a common solution. Among others these data structures are built with the purpose of empty space skipping and adaptive sampling. Empty space skipping is the main strategy of acceleration in this work. Octree and Kd-tree is constructed and analyzed in reference to construction and rendering performance. Time-dependent data often come from numerical simulation and show similarities between two successive time instances. These similarities can be exploited to facilitate construction. The method and consequences of this exploitation will be discussed. Naive volume rendering is conceptually suitable for parallel computing. On the other side, both constructing and traversal of hiearchical data structures do not seem to agree with parallel nature of GPUs. Still implementation techniques to bypass GPU-specific traps of performance is applied in order to utilize the enormous computing power of GPUs.