基于SSE2的Smith-Waterman算法.pdfVIP

戴正华t，z张庆丹1·z徐琳·，z谭光明t，2冯圣中1 1(中国科学院计算技术研究所，北京100080) 2(中国科学院研究生院，北京100039) E—mail：dzh@ncic．ac．cn 摘 要 法的细粒度数据并行性开发受到了很大限制。文章从简化数据依赖关系出发，采用前驱计算思想，提出了基于X86处理 5倍，且与测试集无关。一般相似性不显著的情形下，同目前最好的动态规划细粒度并行算法SWMMX相比可以获得1．5 倍的加速比。 关键词 Smith—Waterman算法细粒度并行算法 SIMDSSE2 文章编号1002—8331一(2006)11-0085—03文献标识码A 中图分类号TP301 Smith-WatermanBasedonSSE2 Algorithm Linl2 Dai Xu Tan Zhenghual’2ZhangQingdanl2 Guangmin912FengShengzhon91 of of 1(Institute ComputingTechnology，ChineseAcademySciences，Beijing100080) 2(GraduateSchooloftheChinese of 100039) AcademySciences，Beijing Abstract：The Smith—Watermanis themostcommon inthefieldof dynamicprogramming algorithm alignalgorithm bioinformation．Becauseofseriousdata fine data ofthat is dependence，thegranularityparallelismalgorithmpoor．We thedata thethinkof we thefine Smith—Waterman simplify dependence，andusing prefixcomputationpresent granularity calledSWSSE2basedonmultimediainstructionsetSSE2ofX86 thecaseof algorithm processor．In highsimilarity，the ofSWSSE2is 5timesas asthatof SW the has todo ordinary performance high algorithm，andperformancenothing withthesetoftest．Inthecaseof ofSWSSE2is 1．5 timesas as inconspicuoussimilarity，theperformance high SWMMXwhichisthebest inthecaseof algorithm conspic