多核CPU和GPU系统上Motif发现并行算法研究-计算机软件与理论专业论文.docxVIP

II II 根据GPU的结构特点以及存储器特点，将求解(l，d)-Motif发现问题过 程中产生的大量不同位置组合的数据和候选序列集合绑定到纹理存储器 中，通过缓存加速访问，减少访存时间；对不同的(l，d)-Motif发现问题规 模，依据组合数据大小设置GPU端线程块的尺寸，设计实现了多核CPU与 单GPU混合机器上(l，d)-Motif发现并行算法，为提高处理较大规模(l， d)-Motif发现问题的加速效果，同时设计实现了多核CPU与多GPU混合机器 上(l，d)-Motif发现并行算法。实验测试综合考虑了不同(l，d)-Motif发现问 题规模、不同线程块尺寸、GPU数目等多种因素对多核CPU与GPU混合机 器上(l，d)-Motif发现并行算法性能的影响。实验结果表明，与多核计算机 上(l，d)-Motif发现并行算法相比，多核CPU与单GPU混合机器上的(l， d)-Motif发现并行算法具有更高的加速比，当求解更长l的(l，d)-Motif发现问 题时，多核CPU与多GPU混合机器上(l，d)-Motif发现并行算法的运行时间 更短，效果更优。 关键词：(l,d)-Motif 发现 多核计算机 多核 CPU 与 GPU 系统 多级缓存 线程级并行 纹理缓存 访存高效 线程绑定 III III STUDY ON PARALLEL ALGORITHMS FOR MOTIF FINDING ON MULTI-CORE CPU AND GPU SYSTEMS ABSTRACT Motif is a fragment of DNA sequence with shorter length and has conserved function.Motif finding problem is the process of looking Motif from the DNA sequence and it is a fundamental and important research topic in bioinformatics.The goal of using computer technology to solve Motif finding problem is find a longer l of Motif.Therefore,studying a cache-efficient,high speed ratio and good scalability parallel (l,d)-Motif finding algorithm on multi-core CPU and GPU systems has practical significance. In this paper,combine the feature of multi-level cache memory structure on multi-core computer,a cache-efficient,high speed ratio and good scalability thread-level parallel (l,d)-Motif finding algorithm is designed.To greatly reducing the number of data exchange between main memory and cache when each core parallel to solve (l，d)-Motif finding problem, the produced large-scale data with the different combination of positions are distributed to the multi-level caches by fully using of the shared L3 Cache, private L2 Cache and L1 Cache on the multi-core architecture,and achieve cache-efficient.Simultaneously,to reducing the cost of thread migration,bind the threads to the cores to balance their computational loads.The experimental tests consider the affect of different PAGE PAGE VI size of (l，d)-Motif finding problem,number of p