基于GPU的BWA序列比对算法分析与加速

Analysis and accelerating of BWA sequeuce comparison algorithm based on GPU

文中实现了GPU平台加速的BWA-MEM算法,将BWA-MEM算法中的两个热点模块:SMEM查找和chain生成模块利用GPU平台进行加速,通过重构算法流程、精简需要向CUDA设备传输的数据结构,采用合理的任务划分方式来提升BWA-MEM在GPU平台的性能。论文对BWA-MEM算法的特点进行了深入分析,总结了BWA-MEM算法在GPU平台加速效果受到限制的原因。

This paper implements the BWA-MEM algorithm on CUDA platform. SMEM finding and chain generating, two of the hotspots,are accelerated with GPU devices. I t improves the performance of GPU-based BWA-MEM by reconstructing the algorithm, reducing the data that has to be transferred between host and device and developing a reasonable task partition method. I t analyzed the characteristics of the BWA-MEM algorithm, and summed up the reason why BWA-MEM algorithm is limited in GPU platform.