混合模型下FMM算法中近程计算的优化研究

摘要

为提高计算多体问题的效率,通过分析多体问题的典型算法FMM(fast multiple method)的计算特点,提出运用CPU和加速部件FPGA构成混合部件计算其近程作用的方案.重点研究混合计算模型上的近程计算特性和优化策略,从计算、通信和存储多方面分析近程计算,提出分层按块的数据准备策略,及在该策略中修改FMM空间编码方式,使近程计算更好地适应于混合模型,从而提高整个FMM算法的执行效率.实验结果表明了该数据准备策略和采用的空间编码方式的可行性和高效性.%To enhance the efficiency of calculating N-body problems, the short-range calculation of FMM is analyzed, which is a typical algorithm for calculating N-body problems, on the hybrid model of CPU and acceleration components. A strategy of data preparing and its FMM encoding model is presented. It makes the short-range calculation more adapt to the hybrid model, which thereby enhances the whole efficiency of the FMM algorithm. The experiment shows the feasibility and efficiency of the proposed scheme.