摘要
在目前系统发育学研究中,多数系统发育分析工具不能在GPU架构上分析蛋白质序列。为此,提出一种大规模系统发育分析方法 tgpMC^3。以添加虚字符的形式重新构造条件似然概率矩阵,降低由于多线程分支发散导致的时间消耗。设计粒度适中的半任务间并行策略,增加流多处理器上活跃的线程块数量。通过简单的键值对应方法传输含有模糊状态的转移概率矩阵,实现数据访问速度的提升。实验结果表明,与MrBayes v3. 1. 2串行版本方法相比,该方法最高可实现117的加速比,与taMC^3方法相比,该方法的并行分析性能更好。
In current phylogeny studies,most phylogenetic analysis tools cannot analyze protein sequences on the GPU architecture.Therefore,a large scale phylogenetic analysis method tgpMC 3 is proposed.The Conditional Likelihood Probabilities(CLPs) matrix is reconstructed in the form of adding virtual characters to reduce the time consumption caused by the divergence of multi-thread branches.Parallel strategy between half tasks with moderate granularity is designed to increase the number of active thread blocks on Stream Multiprocessor (SM).The transfer probability matrix with fuzzy state is transmitted by a simple key-value correspondence method to improve the speed of data access.Experimental results show that compared with MrBayes v3.1.2 serial version method,this method can achieve a maximum speedup of 117,Compared with taMC 3 method,the parallel analysis performance of this method is better.
作者
李易禅
凌诚
LI Yichan;LING Cheng(Collage of Information Science and Technology,Beijing University of Chemical Technology,Beijing 100029,China)
出处
《计算机工程》
CAS
CSCD
北大核心
2019年第2期296-302,共7页
Computer Engineering
基金
国家自然科学基金(61602026)
关键词
系统发育分析
条件似然概率
CUDA编程
并行计算
MC3算法
phylogenetic analysis
Conditional Likelihood Probabilities(CLPs)
CUDA programming
parallel computation
MC 3 algorithm
