面向定制结构的稀疏矩阵分块方法

稀疏矩阵向量乘是科学计算的核心问题,采用定制结构来加速稀疏矩阵向量乘的执行对提升科学计算性能具有重要意义。针对目前面向定制结构的稀疏矩阵分块方法和表示方法的缺点,提出了稀疏矩阵二维均匀分块方法和相应的表示方法嵌套分块CSR。实验结果表明,提出的稀疏矩阵分块方法和表示方法能够有效减少填零个数。

Sparse Matrix Blocking Method for Custom Architecture

Sparse matrix vector multiplication is one of the most important applications in scientific computing.Using custom architectures to implement sparse matrix vector multiplication is introduced to improve the performance of scientific computing.To address the problem in existing sparse matrix blocking method and representation,a two-dimension uniform blocking method and its according representation were proposed in this paper.The experimental results show that the proposed sparse matrix blocking method and representation can reduce the padding zero significantly.