激波与火焰面相互作用数值模拟的GPU加速

发布时间：2017-12-30 23:51

本文关键词：激波与火焰面相互作用数值模拟的GPU加速　出处：《计算物理》2016年01期 　论文类型：期刊论文

更多相关文章：激波 火焰界面 GPU 异构系统 加速比

【摘要】：为考察计算机图形处理器(GPU)在计算流体力学中的计算能力,采用基于CPU/GPU异构并行模式的方法对激波与火焰界面相互作用的典型可压缩反应流进行数值模拟,优化并行方案,考察不同网格精度对计算结果和计算加速性能的影响.结果表明,和传统的基于信息传递的MPI 8线程并行计算相比,GPU并行模拟结果与MPI并行模拟结果相同;两种计算方法的计算时间均随网格数量的增加呈线性增长趋势,但GPU的计算时间比MPI明显降低.当网格数量较小时(1.6×104),GPU计算得到的单个时间步长平均时间的加速比为8.6;随着网格数量的增加,GPU的加速比有所下降,但对较大规模的网格数量(4.2×106),GPU的加速比仍可达到5.9.基于GPU的异构并行加速算法为可压缩反应流的高分辨率大规模计算提供了较好的解决途径.
[Abstract]:In order to investigate the computational power of GPU in computational fluid dynamics (CFD). The typical compressible reaction flow between shock wave and flame interface is numerically simulated based on CPU/GPU heterogeneous parallel mode, and the parallel scheme is optimized. The effects of different grid precision on the computational results and computational acceleration performance are investigated. The results show that the parallel computation of MPI 8 threads based on information transfer is compared with that of traditional parallel computing. GPU parallel simulation results are the same as MPI parallel simulation results. The computational time of the two methods increases linearly with the increase of the number of meshes, but the computing time of GPU is obviously lower than that of MPI. When the number of meshes is small, the calculation time is 1.6 脳 10 ~ 4). The speedup ratio of the average time of single time step calculated by GPU is 8.6; The speedup ratio of GPU decreases with the increase of grid number, but 4.2 脳 10 ~ 6 for larger grid. The acceleration ratio of GPU can still reach 5.9. The heterogeneous parallel acceleration algorithm based on GPU provides a good solution to the high resolution large-scale computation of compressible reaction flow.
【作者单位】：南京理工大学瞬态物理重点实验室;
【基金】：国家自然科学基金(11372140) 江苏省研究生科研创新计划(KYLX_0328)资助项目
【分类号】：O354.5;TP391.41
【正文快照】： 0引言GPU(Graphic Processing Unit,即“图形处理器”)在并行数据计算上具有的强大浮点运算能力,近年来吸引了越来越多的计算流体力学研究者的注意.CUDA(Compute Unified Device Architecture,2008年NVIDA公司发布)是一种基于并行编程模型和指令集的通用计算构架,支持使用C语

【参考文献】