Iterative layer-based raytracing on CUDA

A raytracer consists in an application capable of tracing rays from a point into a scene in order to determine the closest sphere intersection along the ray´s direction. Because of their recursive nature, raytracing algorithms are hard to implement on architectures which do not support recursion, such as the NVIDIA CUDA architecture. Even if the recursive portion of a typical raytracing algorithm can be rewritten iteratively, naively doing so could tamper with the image generation process, preventing the parallel algorithm´s results from maintaining high fidelity to the sequential algorithm´s and resulting, in many cases, in lower quality images. In this paper we address both issues by presenting a novel approach for expressing the recursive structure of raytracer algorithms iteratively, while still maintaining high fidelity to the images generated by the sequential algorithm, and leveraging the processing power of the GPU for parallelizing the image generation process. Our work focuses on designing and implementing a raytracer that renders arbitrary scenes and the reflections among the objects contained in it. However, it can be easily extended to implement other natural phenomena, such as light refraction, and to aid the iterative implementation of recursive algorithms in architectures like CUDA, which do not support recursive function calls.