| We look at the broader field of geometric subdivision and the emerging field of parallel computing for the purpose of creating higher visual fidelity at an efficient pace. Primarily, we present a parallel algorithm for ✓3-Subdivision. When considering animation, we find that it is possible to do subdivision by providing only one variable input, with the rest being considered static. This reduces the amount of data transfer required to continually update a subdividing mesh. We can support recursive subdivision by applying the technique in passes. As a basis for analysis, we look at performance in an OpenCL implementation that utilizes a local graphics processing unit (GPU) and a parallel CPU. By overcoming current hardware limitations, we present an environment where general GPU computation of ✓3-Subdivision can be practical. |
