Extending Illumina PRT with SIMD ray packet intersections

Abstract

Ray tracing is a rendering technique used to synthesize high quality, realistic images of virtual scenes. Rays of light are traced from the perspective of the viewer, through the scene, until the closest intersection is found, and a radiance value is computed for that point. This value is then used to determine the colour of the respective pixel in the final image. This computation may require tracing other rays into the scene. One way of speeding up these computations is by tracing packets of rays in parallel, instead of single rays, allowing us to exploit the spatial coherence which exists between rays, and amortize certain operations over a packet of rays. Primary rays fired from the camera are highly coherent and provide an obvious starting point for optimization. Secondary rays traced into the scene after the first intersection tend to be less coherent and require more intelligent grouping to obtain any benefits from the use of packets. This FYP extends Illumina PRT, a physically-based rendering engine, to support the use of ray packets for rendering. The framework has been extended with a renderer that uses ray packets to render the scene, as well as support for ray packet intersections for both spatial acceleration structures (namely, kd-trees) as well as geometric primitives (namely, triangles). We also apply ray packet techniques to secondary rays, specifically to perform occlusion tests in an integrator that attempts to render global illumination effects. Our work is then evaluated and compared to single ray techniques to highlight any benefits obtained.