面向OpenCL架构的大规模生物序列比对  

Mega-base Biological Sequence Alignment Targeting OpenCL Architecture

作　　者：陈钢[1] 韦刚[1] 李国波[1] 裴颂文[2] 吴百锋[1] 

机构地区：[1]复旦大学计算机科学技术学院,上海200433 [2]上海理工大学计算机科学工程系,上海200933

出　　处：《小型微型计算机系统》2012年第2期392-398,共7页Journal of Chinese Computer Systems

基　　金：上海市重点学科建设基金项目(B114)资助;AMD大学合作计划基金项目资助

摘　　要：为提高生物序列比对算法的性能和效率,提出一种异构处理平台下可移植的大规模生物序列比对算法及其优化方法.通过改变原有Smith-Waterman算法的计算流程和数据依赖关系,增加序列比对的并行性;通过改变存储器布局后使用向量数据类型,提高全局存储器的带宽利用率;通过增加偏移量改变存储器模块的映射方式,避免模块访问冲突,提高局部存储器的使用效率.实验结果表明,优化后的生物序列比对性能提升了近100倍.In order to improve the performance and efficiency of biological sequence alignment algorithm, we propose a portable mega-base sequence alignment approach and its optimization method based on heterogeneous parallel processing platform. Our ap- proach enhances the parallelism by altering the computation flows and data dependencies of the original Smith-Waterman algorithm. Meanwhile, we change memory layout ~ use vector data type for the sake of improving the global memory bandwidth. Finally, we employ offset inserting with the purpose of optimizing the local memory bank conflict, making best usage of the local-memory. Ex- perimental results show that our proposed optimization methods can acttieve almost 100 limes speedup.

关 键 词：OPENCL GPU 生物序列比对 SMITH-WATERMAN算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]