Consider using Nvidia RTX hardware when available

[aprokop]

Some Nvidia cards provide hardware implements of ray tracing called RTX. There has also been some recent research that demonstrates how one can use RTX hardware for spatial ([1, 2]) and nearest ([5]) search. There's also been research [3] that shows that these ideas can then be used to implement DBSCAN.

The main idea around the spatial searches is to use infinitesimal length rays, and construct a hierarchy using spheres around points.

I think it would be interesting to explore this capability and allow ArborX' users to use it if available. The way to do that is to use the Nvidia's OptiX interface. [4] has a publicly available source code here, which could jump start our understanding of how to use OptiX.

Potential areas benefiting from this capability:

• Monte-Carlo ray tracing
• Low density DBSCAN
• Regular ray tracing
• XGC locating points in tokamak mesh

[1] S. Zellmann, M. Weier, and I. Wald, “Accelerating force-directed graph drawing with rt cores,” in 2020 IEEE Visualization Conference (VIS), 2020, pp. 96–100. (pdf)

[2] I. Evangelou, G. Papaioannou, K. Vardis, and A. A. Vasilakis, “Fast radius search exploiting ray tracing frameworks,” Journal of Computer Graphics Techniques (JCGT), vol. 10, February 2021. (pdf)

[3] V. Nagarajan, M. Kulkarni, "RT-DBSCAN: Accelerating DBSCAN using Ray Tracing Hardware", preprint, 2023. (pdf)

[4] Y. Zhu. "RTNN: accelerating neighbor search using hardware ray tracing." In Proceedings of the 27th ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, pp. 76-89. 2022. (pdf)

[5] V. Nagarajan, D. Mandarapu, and M. Kulkarni. "RT-kNNS Unbound: Using RT Cores to Accelerate Unrestricted Neighbor Search." In Proceedings of the 37th International Conference on Supercomputing. 2023. (pdf)