Towards adaptive rendering of smooth primitives on GPUs

Higher order surface primitives enable artists to create smooth, complex objects by manipulating only a few control points and allow for the generation of smooth surfaces from a very compact representation. The implementation of higher order primitives on Graphics Processing Units (GPUs) has the potential to significantly reduce the bandwidth requirements across the graphics bus. Unfortunately, the GPU support for higher order primitives is still rudimentary. We present an adaptive, depth-first tessellation algorithm for smooth surfaces. The algorithm takes a set of Bezier control points and tessellates them according to criteria such as screen-space edge length. Other representations, such as subdivision surfaces, can be handled through preprocessing. The algorithm is designed to provide consistent, hole-free tessellations of adjacent patches. In addition, the polygons generated by the tessellator reside on a space filling curve on the 2D manifold of the surface. This guarantees the good memory coherence for both framebuffer and texture memory access. While the current implementation of the method is purely CPU-based, we believe it is suitable for hardware implementation on future generations of GPUs.Science, Faculty ofComputer Science, Department ofGraduat