论文部分内容阅读
稀疏矩阵与向量相乘(Sp MV)是科学计算和工程应用中一个重要问题,而且非常适宜进行并行计算,目前在GPU对Sp M V的实现和优化是一个研究热点.针对准对角矩阵存在的一些不规则性,采用CSR+DLA混合存储格式来进行Sp M V计算,能够提高压缩的效果.为了发挥CPU多核的并行计算能力,采用一种CPU+GPU混合计算模式,这样可以把混合存储格式不同格式的数据分割到CPU和GPU上,从而提高了资源的利用效能.本文另外还在分析CPU+GPU异构计算模式的特征基础上,提出一些优化策略,能够改进准对角矩阵与向量相乘在异构计算环境中的计算性能.
Sp MV is an important issue in scientific computing and engineering applications, and it is very suitable for parallel computing.At present GPU is a hotspot in the research of Sp MV.Aiming at the existence of quasi-diagonal matrix Some irregularities, the use of CSR + DLA hybrid storage format for Sp MV calculation, can improve the compression effect.In order to play CPU multicore parallel computing power, using a CPU + GPU hybrid computing mode, which can be mixed storage format is different Format data is divided into CPU and GPU, so as to improve resource utilization efficiency.This paper also analyzes the characteristics of CPU + GPU heterogeneous computing mode and puts forward some optimization strategies to improve the quasi-diagonal matrix and vector multiplication Computational performance in heterogeneous computing environments.