Finding eigenvalues and vectors using Lantsosha method on cluster systems
Hrybenko D.V., Stirenko S.G.
This article discusses questions that arise during implementation of Lanczos method for clusters: choice of sparse matrix storage format and code optimization with SIMD instructions of modern CPUs. Discussed algorithms of sparse matrix-vector multiplication have been implemented and an experimental comparison was carried out. A parallel cluster eigenvalue and eigenvector solver cluster for real symmetric sparse matrices was developed. It uses MPI for communication and LAPACK for post-processing. It supports the de-facto standard Matrix Market format.