PME算法在神威太湖之光上的移植和优化

Porting and Optimizing PME Algorithm on Sunway TaihuLight System

分子动力学模拟(MD)是一套通过计算机模拟生物体系内分子、原子运动的多体模拟方法.GROMACS是著名的MD应用,能够快速模拟生物及非生物体系运动过程,广泛应用于各高性能平台.作为世界排名第3的超级计算机,神威太湖之光拥有40960块SW26010异构众核处理器,峰值性能达到125.4PFlops.目前太湖之光上已有对GROM ACS短程力优化的相关研究,但对于PM E(Particle Mesh Ewald)算法未有探索性工作.本文基于申威平台对PME算法展开研究,针对随机访存模式、网格点写写冲突等挑战,提出了基于局部网格序的分块策略、数据重组策略、非线性函数近似等方法进行优化.最终优化后的结果相较于初始版本性能提升了8.85倍,相较于Intel CPU版本提升了1.2倍.本文采用的优化技术也可以为神威太湖之光上其他分子动力学模拟软件和涉及散乱数据插值程序的优化提供借鉴.

Molecular dynamics simulation is a multi-body simulation method,which imitates the motion of molecules and atoms in biological systems by computer.GROM ACS is a well-known application in the field of molecular dynamics simulation.Because it can imitate motion of biotic and abiotic systems quickly,it is widely applied in various high-performance platforms.As the third fastest supercomputer in the world,Sunw ay TaihuLight contains 40960 SW26010 heterogeneous many-core processors with a peak performance of 125.4 PFLOPS.At present,there are too much researches about short-range interaction optimization,but not enough researches about PME algorithm for GROM ACS on Sunway.So we work on PME algorithm on Sunway platform.But there also are many challenges in algorithm parallelism,like random access mode,write conflict of the grid points and so on.To solve these problems,we optimize algorithm with blocking strategy based on local grid order,data reorganization solution,nonlinear function approximation and so on.In the end,optimized result is 8.85 times faster than initial version,and 1.2 times faster than Intel CPU version.M oreover,the optimization techniques used in this paper can also provide reference for optimization of other molecular dynamic simulation softw are and scattered data interpolation programs on Sunw ay TaihuLight.