Deriving Shape Grammars on the GPU
Mark Dokter, Michael Kenzel · 2014
Due to growing demand for computer generated graphi-cal content, procedural modeling has become an important topic in the gaming and movie industry. Creating vast amounts of content by hand requires excessive amounts of manual labor. Using a procedural rule set, entire worlds can be generated by a computer. However, the traditional CPU-based derivation of a large city can take multiple hours, making rapid design iterations impossible. In this paper, we investigate different strategies to execute procedural modeling on graphics processors using CUDA. We com-pare a persistent threads megakernel approach to simple kernel calls and different rule queuing strategies. Along these lines, we explore the trade-off between precompiling an entire rule set and interpreting a rule set online.