摘要
基于KKSO三维多元合金共晶多相场模型,研究了OpenCL+GPU软硬件体系结构的高性能计算方法,以CBr_4-C_2Cl_6为例,分别在AMD和NVIDIA 2种异构平台上通过多进程和多线程的并发执行实现了强迫对流下三维共晶组织的演化过程.结果表明:当计算规模相同时,分别在不同的平台上取得了一定的加速比.对并行算法在一定程度上进行优化,与CPU平台上的串行算法相比优化后的并行算法在异构平台上分别达到了20.2倍和23.6倍的加速比,大大提高了计算效率.同时,以其强大的浮点计算能力获取较为准确的模拟结果,达到计算效率和可移植性的双重需求,解决了传统求解相场模型存在的计算量大、效率低、限于定性研究等问题.
Based on KKSO three-dimensional multiphase-field model of multiple alloy eutectic,a high-performance computing method for OpenCL+GPU software and hardware architecture is studied.Taking CBr4-C2Cl6 as an example,the evolution process of three-dimensional eutectic microstructure is implemented on condition of forced convection by means of multi-process and multi-threaded concurrent execution on two heterogeneous platforms of AMD and NVIDIA,respectively.The result shows that,on the same calculation scale,a certain acceleration ratio will be obtained respectively on the two platforms.The parallel algorithm optimized to a certain extent will be compared with the serial algorithm on CPU platform and the acceleration ratio of the former on heterogeneous platforms will be able to amount to 20.2 times and 23.6 times of the latter,respectively,improving greatly the calculation efficiency.At the same time,with powerful floating-point computation ability,to the more accurate simulation result will be acquired,achieving the double demand of computational efficiency and transplanting ability.The problems of traditional solution of phase-field model such as large computation,low efficiency and limited qualitative research will be resolved.
作者
朱昶胜
金显
冯力
肖荣振
ZHU Chang-sheng;JIN Xian;FENG Li;XIAO Rong-zhen(College of Computer and Communication,Lanzhou Univ. of Tech.,Lanzhou 730050,China;State Key Laboratory of Advanced Processing and Recycling of Nonferrous Metals,Lanzhou Univ. of Tech.,Lanzhou 730050,China)
出处
《兰州理工大学学报》
CAS
北大核心
2019年第1期11-17,共7页
Journal of Lanzhou University of Technology
基金
国家自然科学基金(51661020
11504149
11364024)
