Fast Rendering Of Free-Form Surface On GPU

The rendering of free-form surface, especially subdivision surface, has been a research focus in computer graphics recently. In the past years, a great achievement has been made in GPU which is designed for the computation of computer graphics. GPU has a better performance than CPU in parallel computation, and also provide a considerable degree of programmability.To fully utilize the powerful parallel computation ability of modern GPU, a GPU based algorithm is proposed to render the free-form surface fast and efficiently. We break up the original control mesh into smaller meshes(patch), and then render them one by one. First, compute the Subdivision Pattern for each patch; then, transfer the patch data and subdivision pattern to GPU; finally, compute the limit point of surface in vertex processor within the GPU.In this paper, we first review the rendering algorithms about free-form surface, including parametric surface rendering algorithms and subdivision surface rendering algorithms. We also make a brief introduction about the elution of GPU. Then some basic theories about the four kinds of surface which our algorithm has been applied to are presented. Those four kinds of surface are: bi-cubic Bezier, bi-cubic B-Spline, Catmull-Clark surface and Loop surface.Finally our algorithm is presented. At first, we make an overview of our algorithm, then, provide a detailed description of each step, including patching the original control mesh, computing subdivision pattern, and subdividing in GPU, at last, we make a fully analyses to our experiment.Based on our algorithm, we have developed a prototype system, which can support the rendering of bi-cubic Bezier, bi-cubic B-Spline, Catmull-Clark surface and Loop surface. Experiment results prove that our algorithm has following strongpoints: 1) Efficience, our algorithm can be greatly more efficient than the algorithm based on CPU under condition of excellent GPU (like GeForce6 series) and also more fast than the algorithm based on fragment processor within the GPU. 2) Balance the workload between CPU and GPU. 3) Universality, our algorithm can support all kinds of parametric surface and subdivision surface, whose subdivision on the regular part of the mesh coincides with a known parametric representation.