Current GPU computational power enables the execution of complex and parallel algorithms, such as ray tracing techniques supported by kD-trees for 3D scene rendering in real time. This work describes in detail the study and implementation of eight different kD-tree traversal algorithms using the parallel framework NVIDIA Compute Unified Device Architecture, in order to point their pros and cons regarding performance, memory consumption, branch divergencies and scalability on multiple GPUs. In addition, two new algorithms are proposed by the authors based on this analysis, aiming to performance improvement. Both of them are capable of reaching speedup gains up to 3× when compared to recent and optimized parallel traversal implementations. As a consequence, interactive frame rates are possible for scenes with 1,408 × 768 pixels of resolution and 3.6 million primitives.

No abstract

No abstract

We introduce a GPU grid-based data structure for massively parallel nearest neighbor searches for dynamic point clouds. The implementation provides real-time performance and it is executed on GPU, both grid construction and nearest neighbors (approximate or exact) searches. This minimizes the memory transfer between device and system memories, improving overall performance. The proposed algorithm may be used across different applications with static and dynamic scenarios. Moreover, our data structure supports three-dimensional point clouds and given its dynamic nature, the user can change the data structure's parameters at runtime. The same applies to the number of neighbors to be found. Performance comparisons were made against previous works, endorsing the benefits of our solution. Finally, we were able to develop a real-time Point-Based Rendering application for validation of the data structure. Its drawbacks and data distribution's impact on performance were analysed and some directions for further investigation are given.

No abstract