• Version: 0.2-0
• Status:
• License: BSL-1.0
• Project home: https://github.com/wrathematics/spar
• Bug reports: https://github.com/wrathematics/spar/issues
• Documentation: http://librestats.com/spar/html/index.html

spar is the Sparse Allreduce library, a header-only C++14 framework. Its purpose is to enable the addition of many sparse matrices in CSC format, with the matrices spread across multiple processors.

A sparse allreduce is different from the addition of two distributed, sparse matrices. In that case, every process contains a piece of each of the two distributed matrices. After the sum each process contains a piece of the new distributed matrix, whose entries are the sum of the two. In a sparse allreduce, each of the, say, p processes has a sparse matrix, all of the p matrices are added, and afterwards, one or all processes contain the sum.

The library is header-only so no installation is strictly necessary. You can just include a copy/submodule in your project. However, if you want some analogue of make install, then you could do something like:

ln -s ./src/spar /usr/include/

Stable releases are posted on GitHub. Additionally, you can download the development version via:

git clone --recurse-submodules https://github.com/wrathematics/spar.git

To use the reducers, you will need an installation of MPI.

Tests use catch2, a copy of which is included under tests/catch. To build the tests, modify tests/make.inc as appropriate and type make.

To use the library with Eigen, you will need a copy of the Eigen headers. You may find the eigen example helpful.