high performance computing on graphics processing units: hgpu.org

This paper presents a hybrid photon-mapping approach for global illumination. It represents a significant improvement over a previously described approach, both with respect to speed and accuracy. Using OptiX for ray tracing provides a considerable improvement in the speed of ray tracing and would keep synchronization to a minimum by using texture memory to cache access to the photon information. [Parker et al. 2010]. Recent developments in processor design towards multi-core systems do not favor the tree based photon map approach. So, instead we use the Spatial Hashing Method [Fleisz 2009] to store and retrieve a photon map. Also, the density of photon maps is reduced by storing photons selectively on a local required density criterion, while preserving the correct illumination. At some selected locations with high photon density, we pre-compute the irradiance to speed-up the final gather stage. We use the method to simulate some global illumination scenes and objects. Comparision with existing photon mapping techniques indicates that our method gives significant improvement in speed with the same or better accuracy of the scene.