基于GPU的高度并行Marching Cubes改进算法  

Highly Parallel Marching Cubes Algorithm Based on GPU

作　　者：刘伟峰 杨权一 曹邦功 孟凡密[1] 周洁[1] 

机构地区：[1]中国石油化工股份有限公司石油勘探开发研究院信息技术研究所,北京100083

出　　处：《微电子学与计算机》2008年第9期151-154,157,共5页Microelectronics & Computer

摘　　要：提出一种完全基于GPU(graphics processing unit)的高度并行Marching Cubes改进算法.针对Marching Cubes算法流程中数据处理阶段进行优化.首先并行遍历每个体元,以序列形式得到每体元的非空状态,再利用高效流式缩减操作获得非空体元序列以便仅针对非空体元高度并行地生成等值面三角形.实验表明,与同市场价格CPU实现相比,该算法效率最高提高到了前者的9倍以上.We present a fully GPU-based highly parallel Marching Cubes algorithm.The method mainly focuses on the data-processing stage to optimize the algorithm.Firstly,GPU use its highly parallel characters to traverse all voxels,and form non-empty state sequence.Secondly,we realize high-efficient stream reduction to achieve the sequence of the non-empty elements for triangles generation.The results show the speed of our algorithm can be improved by up to more than 9 times compared to same price CPU-based implementations.

关 键 词：图形处理器 MARCHING CUBES 流式缩减 图形处理器的通用计算 统一计算设备架构 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]