Distributed Block Compressed Sparse Row matrix library
libDBCSR (or DBCSR for short) is a sparse matrix library designed to efficiently perform sparse matrix matrix multiplication, among other operations. It is MPI and OpenMP parallel, and can exploit accelerators. It is developed as part of CP2K , where it provides core functionality for linear scaling electronic structure theory. A general overview of the library has been published. A discussion of recent developments, in particular GPU work, has appeared as a chapter in 'Electronic Structure Calculations on Graphics Processing Units', John Wiley and Sons, ISBN 9781118661789, and is available as a preprint. The use of one-sided MPI and a 2.5D algorithm to reduce communication is shown to be effective for sparse matrix matrix multiplication in this manuscript. libDBCSR is made available for integration in other projects, see the github webpage.
pkg install dbcsr