Abstract:

Focusing on the issues of creating accelerationstructures of 3D scenes for ray tracing, a GPU-based parallel construction algorithm for threaded-Biovision Hierarchy (BVH) was proposed. The algorithm builds the Bounding-Volume Nodes (BVNs) completely in a breadth-first search order and includes two BVN construction processes. The first process is for upper large nodes, which parallelizes the computation over geometric primitives in each node, and the second is for lower small nodes, which parallelizes the computation over nodes at each level. These parallelization strategies can solve the problem of lacking parallelism in the construction process of upper large nodes. The algorithm performs operations of the BVH storage-structure transformation from the breadthfirst to the depth-first order and threading computation of BVNs in three individual steps. Experiment results show that the construction speed of the proposed algorithm is faster than the existing unthreaded-BVH construction algorithm.

Key words: computer application, bounding volume hierarchy(BVH), threading, graphic processing unit(GPU), ray tracing