Parallel Operations

  • Vector reductions

    • Norms

    • Dot product

    • Result replicated on all processors

    • Same syntax as non-distributed

  • Vector expressions

    • Full spectrum of non-distributed vectors

    • Same syntax as the latter

  • Matrix mapping with distributed views

    • Scaling from left and right with a scalar value

    • Division of a matrix by a scalar

    • Conjugate view

    • As implemented by means of local mapping, will extend naturally to new local mappings (including user-defined)

  • Transposition with distributed views

  • Hermitian of a distributed matrix

    • Algorithms can use adjunct linear projection

  • Distributed matrix vector product

    • Source and target vector can be distributed differently

    • Special implementation for transposed matrix

    • Works also with mapped matrices and Hermitians

  • Can be used as term in vector expressions

  • Distributed preconditioning, see above

  • Serialization of non-distributed containers

    • Vector

    • Dense matrix

    • Compressed matrix

    • Can be used with all boost::mpi commands

    • Fast serialization for intrinsic MPI types

    • Initialization of distributed matrices from files in matrix market format

 

© 2017 SimuNova UG