CUDA架构下的靶丸辐射能流并行计算  

CUDA Based Parallel Computation in Thermal Radiation on Implosion Pellet

作　　者：赵权[1] 黄运保[1] 孙宇航[1] 

机构地区：[1]华中科技大学国家CAD支撑软件工程技术研究中心,武汉430074

出　　处：《计算机辅助设计与图形学学报》2013年第7期937-945,共9页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(50975107;51175198;61173115;51075162);广东教育部产学研结合项目(2011A090200059)

摘　　要：针对惯性约束聚变(ICF)间接驱动对称性分析中靶自身遮挡判断困难、表面辐射能流计算效率低等问题,提出了基于CUDA的靶丸辐射能流并行计算方法.该方法以两区近似均匀模型为基础,利用四叉树结构对靶离散单元进行层次化表示,再根据包围盒的相交判断快速剔除遮挡单元;通过单元视角因子计算过程到GPU各运算线程的映射,实现了靶丸辐射能流计算的并行化.实验结果表明,在保证同样计算精度的同时,文中算法平均加速比在考虑遮挡时为15,不考虑遮挡时为400,具有较高的计算效率.The computation of radiation indirect drive from hohlraum to the capsule in the symmetry analysis of inertial confinement fusion （ICF） is difficult to deal with occlusion culling and low efficiency. With respect to these defects, a parallel algorithm based on computing unified device architecture （CUDA） is proposed for the computation in thermal radiation on implosion pellet. Firstly, based on the equivalent model of two-region uniformly, the interior wall of hohlraum and the surface of target are divided into grid cells. Then those cells are organized by a quad-tree structure to construct hierarchical bounding box. The obstructed cells can be excluded quickly by the intersection judgment among these boxes. Finally, to realize the parallel computation of energy flow distribution, the calculation processes of view factor are mapped to GPU threads. The experimental results show that the presented algorithm can achieve a speed-up ratio about 15 with considering occlusion, and 400 without considering occlusion.

关 键 词：惯性约束聚变 视角因子 四叉树 遮挡剔除 CUDA 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]