Fragment-Parallel Composite and Filter

• We identify a novel strategy to parallelize composite and filter across individual depth fragments, and describe an implementation on a modern GPU. We demonstrate that this technique is applicable to several areas with four examples: a particle system, a volume renderer, a polygon rasterizer and an interactive Reyes pipeline. • We compare the performance of the proposed fragment-parallel strategy against a pixel-parallel version, and study the behavior of both with varying depth complexity. We show that for the same number of fragments, a pixel-parallel technique loses performance with high as well as variable depth complexity, while the performance of a fragment-parallel version is much more consistent. We study the parallelization of composite/filter on a modern GPU architecture, and propose a scheme that parallelizes the operation not just across pixel or subpixel locations, but over all available fragments. We demonstrate the applicability of this technique for several applications, and compare the obtained performance of the proposed fragment-parallel composite (FPC) against a conventional pixel-parallel composite (PPC) scheme, i.e. one that parallelizes only across pixel/subpixel locations. Using a synthetic benchmark, we also study the variation in the relative performance of FPC and PPC across the continuum from scenes with a few depth layers to those with a large number of depth layers. Our implementation allows an arbitrary number of semi-transparent layers for each subpixel location. Our contributions are as follows: Geometry Processing