All tetrahedra must be sorted before visualization with projected tetrahedra (PT) algorithm, but it is difficult to be parallel due to irregularity and dependency of tetradedra. We present a new parallel scheme based on accurate sorting, which peels the entire dataset into unobstructed layers on sort phase, and then reduces the amount of processed tetradedra with early tetrahedron termination technique by leveraging the summed-area table for accumulating and comparing opacity of individual tetrahedron on rendering phase. Tetrahedra datasets have been projected orderly to get a rendering result. Experimental results identify that GPU implementation of the parallelized accurate sorting scheme achieves 91% acceleration compared with the CPU implementation, and the early tetrahedron termination scheme has a gain of more than 10% in terms of rendering performance for datasets with more than a million tetrahedra.%采用投影四面体法进行可视化时需要对所有四面体进行排序,而四面体的不规则性和较强的依赖性导致并行排序的难度很大,为此提出一种精确排序的并行化算法.该算法在排序阶段逐层并行提取互不遮挡的四面体,并在绘制阶段采用区域求和表、提前终止等技术直接减少处理的四面体个数,再将四面体数据集进行有序投影得到最终的绘制结果.实验结果表明,采用文中算法的GPU实现比基于CPU的精确排序快91％;对于大尺度数据集(大于百万个四面体),提前终止的算法使绘制效率提高10％以上.
展开▼
