Find us on…

GitHub The Web Download the Source


DBCSR stands for Distributed Blocked Compressed Sparse Row.

DBCSR is a library designed to efficiently perform sparse matrix-matrix multiplication, among other operations.

It is MPI and OpenMP parallel and can exploit Nvidia and AMD GPUs via CUDA and HIP.

To get started with DBCSR, go to


DBCSR's source code and related files and documentation are distributed under GPL. See the LICENSE file for more details.

How to cite

To cite DBCSR, use the following paper

    title = {{Sparse Matrix Multiplication: The Distributed Block-Compressed Sparse Row Library}},
    journal = {Parallel Computing},
    volume = {40},
    number = {5-6},
    year = {2014},
    issn = {0167-8191},
    author = {Urban Borstnik and Joost VandeVondele and Valery Weber and Juerg Hutter}

To cite the DBCSR software library, use:

    author = {The CP2K Developers Group},
    title = {{DBCSR: Distributed Block Compressed Sparse Row matrix library}},
    publisher = {GitHub},
    journal = {GitHub repository},
    year = {2020},
    url = {https://github.com/cp2k/dbcsr}


Your contribution to the project is welcome! Please see DBCSR's contribution guidelines and this wiki page.

Developer Info

DBCSR Authors