The introduction of floating-point pixel shaders has initiated a trend of moving algorithms from CPUs to graphics cards. The first algorithms were in the rendering domain, but recently we have witnessed increased interest in modeling algorithms as well. In this paper we present techniques for generating subdivision curves on a modern Graphics Processing Unit (GPU). We use an existing method for generating subdivision curves with L-systems, we extend these L-systems to implement adaptive subdivision, and we show how these L-systems can be implemented on a GPU. We chose L-systems because they can express many modeling algorithms in a compact way and are parallel in nature, making them an attractive paradigm for programming a GPU.
Download PDF here (400 kb)