STRUMPACK - High-Performance Scalable Software Library Based on Low-Rank Approximations

Workshop:Women in HPC: Diversifying the HPC Community and Engaging Male Allies

Authors: Lisa Claus (Lawrence Berkeley National Laboratory)

Abstract: During this lightning talk, I am excited to share my HPC work on the high-performance scalable software library STRUMPACK. This open-source software, developed by researchers at Lawrence Berkeley National Laboratory, leverages hierarchical and non-hierarchical matrices, using low-rank approximations, to get low complexity solvers and efficient preconditioners for large sparse systems. The benefit of this software is that it is effective for solving a variety of discretized partial differential equations. The software package provides a parallel and fully algebraic preconditioner based on an approximate sparse factorization using rankstructured matrix compression.

My work is based on improving the package even further by expanding the efficiency to a variety of applications. We recently integrated the so-called Hierarchical Off-Diagonal Butterfly compression method to compress some of the larger matrix sub-blocks of the system. Block low-rank (BLR) is another matrix format that has been useful for the compression of sub-blocks for a variety of large sparse systems. The BLR compression method is especially efficient for medium-sized matrix sub-blocks. Therefore, we are currently working on an implementation to combine both methods, Hierarchical Off-Diagonal compression for large sub-blocks and BLR compression for medium-sized sub-block. The compression of small sub-blocks doesn’t pay off. The STRUMPACK software is constantly tested on several supercomputers to benefit a variety of users with access to different resources.

Website:

Back to Women in HPC: Diversifying the HPC Community and Engaging Male Allies Archive Listing

Back to Full Workshop Archive Listing