摘要
本文以数论中的Lucas-Lehmer检验法为基础,提出了梅森素数并行求解算法在FT64流处理器上的流式实现,并通过重设流记录的大小对程序进行了优化。评测数据表明,在FT64上运行该应用的时间平均比1.5GHz Itanium2快2.5倍。本文为梅森素数求解问题寻找了一条可行的加速方法,同时证实了流体系结构在高性能计算领域的极大潜力。本文提出的流式算法以及各种优化手段,对于其他科学计算领域中的计算密集型问题在流体系结构上的映射有极大的借鉴意义。
This paper presents a detailed study of porting the Mersenne Prime Search problem with the Lucas-Lehmertest method to a 64-bit stream processor named FT64. This paper begins with a study of the FT64 stream architecture; followed by a discussion of the stream algorithm of the Lucas-Lehmer method on FT64, together with experiments to analyze the performance of the implementation and to measure the application run-time. The paper presents an efficient approach to accelerate the Mersenne Prime Search problem with 2, 5 times speedup compared to Itanium2. And the comparison with the traditional, general-purpose processors code confirms to FT64＇s potential to deliver high performance.
出处
《计算机工程与科学》
CSCD
2007年第11期53-55,59,共4页
Computer Engineering & Science
基金
国家自然科学基金资助项目(60673148)
博士点基金资助项目(20069998025)
