动态模式识别算法的GPU平台实现

研究动态模式识别算法在GPU并行计算平台的实现。随着GPGPU(通用计算图形处理器)硬件的发展,基于GPU的大规模并行计算技术将有效地处理动态模式识别算法带来的海量计算问题。文中通过介绍动态模式识别算法,对算法中涉及的巨大计算量进行分析,并针对性地对其中密集计算部分进行并行化分解,移除原算法中在执行中存在的依赖关系,最终得到算法在特定的GPU平台———Jacket上的并行计算实现。实例验证表明,相比于原CPU串行程序,在GPU上运行的并行化程序能实现明显加速,因而具有很好的工程应用价值。

Implementing Dynamical Pattern Recognition Algorithm on GPU

This paper presents an implementation of the rapid dynamical pattern recognition algorithm on the GPU platform. GPGPU has emerged as a powerful computation device for its particular parallel architecture, and the massive parallel computing technology based on GPU is considered to be a suitable method dealing with the large scale computation the algorithm brings .In this paper, the rapid dynamical pattern recognition algorithm is briefly introduced, the main computing problems are discussed, and the detailed procedure of parallelizing the algorithm to the Jacket platform is presented. Specific experiments are made and the results indicate that the parallelized algorithm running on GPU has huge speed-ups compared with the old serial versions on CPU and the efficiency of the parallelized algorithm has practical engineering value.