摘要
在集群与GPU组成的异构并行计算平台上,使用MPI+CUDA混合编程模型,实现基于ABEEMσπ模型的分子动力学模拟中电荷分布的计算。通过对电荷分布分布求解中的计算部分移植到GPU上进行,并针对算法中通信开销大和资源未充分利用的问题,通过异构平台的异步并发方法进行优化,提高了求解效率。性能测试结果表明,相比于单纯MPI并行算法,优化后GPU加速的异构并行算法,在化学大分子模型电荷分布计算上,有着明显的性能优势。
The calculation of electric charge distribution in molecular dynamics simulations based on ABEEM（rlr model is implemented using hYbrid MPI ＋ CUDA programming model on heterogeneous parallel computation platform constituted by clusters and GPUs. The solving efficiency has been improved by transplanting to GPU the computation part in solving the electric charge distribution and by optimising with asynchronous and concurrent approach on heterogeneous platform in light of the problems of excessive communication overhead as well as insufficient resource utilisation. The results of performance test shows that the optimised heterogeneous parallel algorithm using GPU boosting has manifest performance predominance in charge distribution calculation for chemical macromolecular model compared with pure MPI parallel algorithm
出处
《计算机应用与软件》
CSCD
北大核心
2012年第10期79-81,93,共4页
Computer Applications and Software
基金
国家自然科学基金项目(21133005
20703022)
