The Chunks and Tasks Matrix Library

Abstract We present a C++ header-only parallel sparse matrix library, based on quadtree representation of matrices using the Chunks and Tasks programming model. The library implements number algorithms for distributed memory parallelization that are able to dynamically exploit data locality avoid movement data. This is demonstrated example block-sparse matrix–matrix multiplication applied three sequences with different nonzero structure, CHT-MPI 2.0 runtime implementation succeeds load balance calculation regardless sparsity structure.