Krylov子空间算法的并行化策略研究

Study on parallel strategy for Krylov subspace algorithms

针对单机计算已不能满足求解大规模稀疏线性方程组的问题,提出一种Krylov子空间方法的并行化设计方案,改进之处在于用一次同步通信完成所有内积计算,改进后的算法降低了内积计算开销,提高了计算效率.

In this paper,aimed at the problem that a stand-alone calculation cannot meet the needs of solving large sparse linear equations,the author proposed a Krylov subspace method of parallel design.Using only one synchronous communication can complete all the inner product calculation.The improved algorithm reduced the computational overhead of the inner product and improved the computational efficiency.