Journal of Parallel and Distributed Computing

Data used to generate the graphs and figures in the paper

HLAM brings hybrid parallelism to several linear algebra iterative methods. While distributed-memory parallelism relies on MPI, shared-memory parallelism can be based on either the fork-join paradigm (via OpenMP) or the task paradigm (via OpenMP or OmpSs-2). The list of supported algorithms includes: · CG: conjugate gradient (classical implementation) · CGNB: nonblocking conjugate gradient (SpMV on matrix Ar) · BiCGStab: biconjugate gradient stabilished (classical implementation) · BiCGStabB1: one blocking biconjugate gradient stabilished · Jacobi · GaussSeidelColour: Gauss-Seidel (coloured scheme implementation) · GaussSeidelRelax: Gauss-Seidel (relaxed implementation)