ABSTRACT
This chapter presents a new implementation suitable for the Graphics processing units (GPU), which completely relieves the CPU without sacrificing scalability. It shows that T-junctions can be removed completely using a simple tessellation shader. The chapter provides the procedures to retrieve the column major numbering from this representation. It explores a complete OpenGL pipeline to extract a scalable grid from this representation, which in turn can be used to render terrains, as well as parametric surfaces. The chapter outlines a novel implementation for quadtrees running completely on the GPU. Linear quadtrees were introduced by Gargantini as a nonrecursive alternative to regular quadtrees. When relying on quadtrees to generate progressive meshes, efficient updates are critical. Naturally, the average GPU rendering time depends on how the terrain is shaded. Managing a linear quadtree on the GPU requires two buffers and a geometry shader. Using the same distance-based criterion, the Transvoxel algorithm could be employed to produce a crack-free volume extractor, running entirely on the GPU.
