基于分层存储理论模型的近似字符串匹配 并行算法研究

CUDA (Compute Unified Device Architecture)是一种重要的并行处理架构,但其具有相对复杂的线程管理机制和多重存储模块,从而使得基于CUDA的算法时间复杂度很难量化.针对这一问题,提出了一种分层存储理论模型—HMM (Hierarchical Memory Machine)模型,该模型所具有的分层存储结构可以有效地描述图形处理单元设备不同存储模块的物理特性,因此非常适用于对CUDA算法时间复杂度的量化评估.作为HMM模型的应用实例,文章提出了一种基于HMM模型的并行近似字符串匹配算法,并给出了相应算法时间复杂度的计算过程.与串行算法相比,该算法可以获得60倍以上的加速比.

A Parallel Algorithm for Approximate String Matching Based on Hierarchical Memory Machine

CUDA(Compute Unified Device Architecture) has a complex thread organization and multilevel memory modules, which makes it difficult to quantitatively evaluate time complexity of CUDA-based algorithms. In this paper, a Hierarchical Memory Machine (HMM) Model was investigated to solve this problem. HMM is a theoretical parallel computing model, which is capable of representing the essence of computing and memory structures on the GPU(Graphics Processing Units) devices. Based on the proposed HMM model, a parallel algorithm was presented for the approximate string matching problem. The proposed algorithm is evaluated and compared with existing approaches, to show a speedup ratio of more than 60.