细分曲面的GPU完全实现

Full GPU Implementation of Subdivision Surfaces

本文提出一种完全在GPU上实现的细分曲面绘制策略。该方法以扇区为基本单元,利用三种新的枚举坐标对网格顶点进行编码。初始控制网格按扇区分割后,其顶点被编码到一张纹理上,控制顶点不规则性分离到扇区的同时扇区之间又具有松耦合联系,使GPU像素管线的并行性从片段网格内扩展到整个网格。另外,通过三张查找表来降低实时计算量,并且避免了在CPU上预先细分一次。经过实验对比说明,本文方法具有更高的实时性能。

A novel strategy is proposed for the full GPU Implementation of subdivision surfaces, using multi-pass general-purpose computation on GPU to accomplish refinement. ＂Full implementation＂ includes two meanings.. The first one is fully using the capability of GPU, and the second one is fully processing data on GPU. A sector-based idea is introduced and three novel enumeration systems are proposed to encode vertices. Vertices at the same depth can be encoded to a texture after beirN partitioned by sectors,and the irregularity of control vertices can be partitioned into sectors while sectors are loosely coupling with each other so that concurrent processing can expand from fragment meshes to the entire mesh. In addition, three lookup tables are used in order to reduee the computing cost and avoid the first refinement on CPU. According to the experiment, the proposed strategy has better realtime performance and the real-time deformation and editing can be done better.