immagine esempio

AMG4PSBLAS: a package of Algebraic MulriGrid preconditioners for solving sparse linear systems in HPC environments


AMG4PSBLAS: a package of scalable preconditioners to solve sparse linear systems at extreme scale on pre-exascale supercomputers for HPC applications selected by DG of EU Innovation Radar as recent excellent innovation

The AMG4PSBLAS development team gladly announce the release of version 1.0 (release candidate 1 --- rc1) of the package. 

AMG4PSBLAS (Algebraic MultiGrid Preconditioners Package based on PSBLAS) is a package of parallel algebraic multilevel preconditioners included in the PSCToolkit (Parallel Sparse Computation Toolkit) software framework; its development is supported by the EU-H2020 EoCoE (Energy Oriented Center of Excellence) project, and the package has been selected by EU Innovation Radar as recent excellent innovation.

AMG4PSBLAS is  designed to provide scalable and easy-to-use preconditioners in the context of the  PSBLAS (Parallel Sparse Basic Linear Algebra Subprograms) parallel computing framework, to  be used in conjunction with the PSBLAS Krylov solvers. 

The library uses a fully algebraic approach to generate a hierarchy of coarse-level matrices and operators; it includes a new parallel coupled aggregation algorithm exploiting maximum edge-weighted matchings.

The preconditioners in AMG4PSBLAS can combine different types of AMG cycles with many smoothers and coarsest-level solvers. AMG4PSBLAS runs on most parallel computers, requiring only PSBLAS, the BLAS and MPI.  A GPU plugin for PSBLAS (available separately from enables the execution of AMG4PSBLAS applications on clusters with hybrid CPU/GPU nodes.