【摘要】 利用GPU的强大浮点数计算能力和并行处理能力,提出一种完全基于GPU的视点相关自适应细分内核进行快速细分计算的方法.在GPU中,依次实现视点相关的面片细分深度值计算、基于基函数表的细分表面顶点求值、细分表面绘制等核心步骤,无须与CPU端系统内存进行几何数据交换.视点相关的自适应细分准则在表面绘制精度保持不变的情况下,有效地降低了细分表面的细分深度和细分的计算量,在此基础上完全基于GPU的细分框架使得曲面细分具有快速高效的特点.该方法还可以在局部重要细节用较大深度值进行实时自适应细分,以逼近极限曲面.更多还原

【Abstract】 A novel method of view-dependent adaptive subdivision is proposed by using the floating-point computation power of modern programmable graphics hardware or GPU to accelerate subdivision computation. The whole subdivision process includes two rendering passes. During the first pass, the subdivision depth of each surface patch is computed in the fragment shaders using projection error metric with backface testing. The resultant buffer is taken as the input to the second pass to evaluate the new subdivision vertex using basis function table stored in texture form on GPU. Our approach makes the subdivision process be entirely implemented on GPU without any geometry data transmission on graphics bus. It improves the efficiency of subdivision while preserving the rendering quality of subdivision surface, and alleviates the computing load on CPU. The view-dependent subdivision can also attain a deep subdivision depth in local region if necessary.更多还原