385 / 2022-11-13 23:40:01

Browser agile rendering to accelerate triangulation based on WebGPU technology

WebGPU, accelerate triangulation, Browser agile rendering.

In recent years, GPU has been widely used in various general computing fields. In the field of computational geometry, many papers have proposed approaches to solve traditional problems using GPU, such as discrete Voronoi graphs and Delaunay triangulation. And WebGPU, WebGPU's brand new web API, makes it possible to call the GPU more closely in the browser. Thanks to the development of these modern technologies, the efficient algorithms of GPU implementation have gradually increased, so we have studied the implementation of the complex triangulation of pure GPU calculation based on WebGPU technology. 

The latest GPU Delaunay triangulation algorithm uses the GPU to help calculate the discrete Voronoi graphs of a set of points S, from which the triangulation T'(S) can be constructed. Due to the discrete nature of the discrete Voronoi graph, several transformations of T'(S) are required to obtain the correct Delaunay triangulation of T(S), and they were initially performed in the CPU. When the number of triangles and the number of reflection times increases and the resolution of the discrete Voronoi graph is fixed, the necessary conversion work greatly affects the running time of the triangulation. In this paper, we propose a new method of converting T'(S) in parallel to T(S), as described in the browser using the WGSL language, by running entirely in the GPU, using NVIDIA's latest Massively Parallel Programming Model (CUDA). Using a well-designed multi-channel approach, our algorithm effectively improves the performance of the triangulation by twice, by 140% compared to the well-known fastest CPU Delaunay Triangulator algorithm.