摘要
基于Intel Many-Integerated-Core(MIC)协处理器平台实现了二维particle-in-cell(PIC)粒子模拟,并根据MIC的特性进行了指令集层面上的向量化优化。为体现MIC平台下的并行加速效果,采用英伟达的GPU计算卡和英特尔Ivy-Bridge架构CPU处理器进行针对性模拟,对3者的计算性能进行比较,当PIC模拟网格较小时,GPU架构具有更好的适用性和计算加速;随着模拟网格逐渐增大,MIC平台的计算加速比逐渐超过GPU平台。研究表明,针对PIC模拟应用,MIC平台具有良好的计算性能、较好的可扩展性和可编程性,利用MIC异构众核对大数据量并行计算程序优化将极大地促进高性能计算的应用和发展。
In this paper,we optimized and tested a general PIC simulation code based on the Intel MIC coprocessor platform.We have parallelized the fundamental PIC code on MIC platform and achieved SIMD optimization by using 512-bit vectorization instructs.In order to better present the advantage of MIC platform,we have carefully compared simulation results from MIC with those from NVIDIA graphics processors（GPU）and Intel Ivy-Bridge CPU processor.When the grid size of PIC simulation is small,GPU platform has better applicability and higher speedup than MIC,while when the grid size is increased,the MIC＇s speedup is gradually close to the GPU.Our simulation results indicate that the MIC has good computing performance,programmability and scalability.The optimization of parallel computing on MIC will enormously promote the development of high performance computing in scientific research and industrial applications.
作者
罗跃剑
陈哲
唐荣欣
刘志伟
钟志宏
周猛
LUO Yuejian;CHEN Zhe;TANG Rongxin;LIU Zhiwei;ZHONG Zhihong;ZHOU Meng(Institute of Space Science and Technology,Nanchang University,Nanchang 330031,China)
出处
《中国科技论文》
CAS
北大核心
2018年第8期961-966,共6页
China Sciencepaper
基金
国家自然科学基金资助项目(41464007
41674144)
高等学校博士学科点专项科研基金资助项目(20133601120005)
南昌大学研究生创新专项资金资助项目(cx2016278)
