trilinos_precondition_ml.cc

// ---------------------------------------------------------------------
//
// Copyright (C) 2008 - 2018 by the deal.II authors
//
// This file is part of the deal.II library.
//
// The deal.II library is free software; you can use it, redistribute
// it, and/or modify it under the terms of the GNU Lesser General
// Public License as published by the Free Software Foundation; either
// version 2.1 of the License, or (at your option) any later version.
// The full text of the license can be found in the file LICENSE.md at
// the top level directory of deal.II.
//
// ---------------------------------------------------------------------

#include <deal.II/lac/trilinos_index_access.h>
#include <deal.II/lac/trilinos_precondition.h>

#ifdef DEAL_II_WITH_TRILINOS

#  include <deal.II/lac/sparse_matrix.h>
#  include <deal.II/lac/trilinos_index_access.h>
#  include <deal.II/lac/trilinos_sparse_matrix.h>
#  include <deal.II/lac/vector.h>

#  include <Epetra_MultiVector.h>
#  include <Ifpack.h>
#  include <Ifpack_Chebyshev.h>
#  include <Teuchos_ParameterList.hpp>
#  include <Teuchos_RCP.hpp>
#  include <ml_MultiLevelPreconditioner.h>
#  include <ml_include.h>

DEAL_II_NAMESPACE_OPEN

namespace TrilinosWrappers
{
  /* -------------------------- PreconditionAMG -------------------------- */

  PreconditionAMG::AdditionalData::AdditionalData(
    const bool                            elliptic,
    const bool                            higher_order_elements,
    const unsigned int                    n_cycles,
    const bool                            w_cycle,
    const double                          aggregation_threshold,
    const std::vector<std::vector<bool>> &constant_modes,
    const unsigned int                    smoother_sweeps,
    const unsigned int                    smoother_overlap,
    const bool                            output_details,
    const char *                          smoother_type,
    const char *                          coarse_type)
    : elliptic(elliptic)
    , higher_order_elements(higher_order_elements)
    , n_cycles(n_cycles)
    , w_cycle(w_cycle)
    , aggregation_threshold(aggregation_threshold)
    , constant_modes(constant_modes)
    , smoother_sweeps(smoother_sweeps)
    , smoother_overlap(smoother_overlap)
    , output_details(output_details)
    , smoother_type(smoother_type)
    , coarse_type(coarse_type)
  {}


  PreconditionAMG::~PreconditionAMG()
  {
    preconditioner.reset();
    trilinos_matrix.reset();
  }



  void
  PreconditionAMG::initialize(const SparseMatrix &  matrix,
                              const AdditionalData &additional_data)
  {
    initialize(matrix.trilinos_matrix(), additional_data);
  }



  void
  PreconditionAMG::initialize(const Epetra_RowMatrix &matrix,
                              const AdditionalData &  additional_data)
  {
    // Build the AMG preconditioner.
    Teuchos::ParameterList parameter_list;

    if (additional_data.elliptic == true)
      {
        ML_Epetra::SetDefaults("SA", parameter_list);

        // uncoupled mode can give a lot of warnings or even fail when there
        // are too many entries per row and aggreggation gets complicated, but
        // MIS does not work if too few elements are located on one
        // processor. work around these warnings by choosing the different
        // strategies in different situations: for low order, always use the
        // standard choice uncoupled. if higher order, right now we also just
        // use Uncoupled, but we should be aware that maybe MIS might be
        // needed
        if (additional_data.higher_order_elements)
          parameter_list.set("aggregation: type", "Uncoupled");
      }
    else
      {
        ML_Epetra::SetDefaults("NSSA", parameter_list);
        parameter_list.set("aggregation: type", "Uncoupled");
        parameter_list.set("aggregation: block scaling", true);
      }

    parameter_list.set("smoother: type", additional_data.smoother_type);
    parameter_list.set("coarse: type", additional_data.coarse_type);

    // Force re-initialization of the random seed to make ML deterministic
    // (only supported in trilinos >12.2):
#  if DEAL_II_TRILINOS_VERSION_GTE(12, 4, 0)
    parameter_list.set("initialize random seed", true);
#  endif

    parameter_list.set("smoother: sweeps",
                       static_cast<int>(additional_data.smoother_sweeps));
    parameter_list.set("cycle applications",
                       static_cast<int>(additional_data.n_cycles));
    if (additional_data.w_cycle == true)
      parameter_list.set("prec type", "MGW");
    else
      parameter_list.set("prec type", "MGV");

    parameter_list.set("smoother: Chebyshev alpha", 10.);
    parameter_list.set("smoother: ifpack overlap",
                       static_cast<int>(additional_data.smoother_overlap));
    parameter_list.set("aggregation: threshold",
                       additional_data.aggregation_threshold);
    parameter_list.set("coarse: max size", 2000);

    if (additional_data.output_details)
      parameter_list.set("ML output", 10);
    else
      parameter_list.set("ML output", 0);

    const Epetra_Map &domain_map = matrix.OperatorDomainMap();

    const size_type constant_modes_dimension =
      additional_data.constant_modes.size();
    Epetra_MultiVector distributed_constant_modes(
      domain_map, constant_modes_dimension > 0 ? constant_modes_dimension : 1);
    std::vector<double> dummy(constant_modes_dimension);

    if (constant_modes_dimension > 0)
      {
        const size_type global_size = TrilinosWrappers::n_global_rows(matrix);
        (void)global_length; // work around compiler warning about unused
                             // function in release mode
        Assert(global_size ==
                 static_cast<size_type>(
                   TrilinosWrappers::global_length(distributed_constant_modes)),
               ExcDimensionMismatch(global_size,
                                    TrilinosWrappers::global_length(
                                      distributed_constant_modes)));
        const bool constant_modes_are_global =
          additional_data.constant_modes[0].size() == global_size;
        const size_type my_size = domain_map.NumMyElements();

        // Reshape null space as a contiguous vector of doubles so that
        // Trilinos can read from it.
        const size_type expected_mode_size =
          constant_modes_are_global ? global_size : my_size;
        for (size_type d = 0; d < constant_modes_dimension; ++d)
          {
            Assert(
              additional_data.constant_modes[d].size() == expected_mode_size,
              ExcDimensionMismatch(additional_data.constant_modes[d].size(),
                                   expected_mode_size));
            for (size_type row = 0; row < my_size; ++row)
              {
                const TrilinosWrappers::types::int_type mode_index =
                  constant_modes_are_global ?
                    TrilinosWrappers::global_index(domain_map, row) :
                    row;
                distributed_constant_modes[d][row] =
                  additional_data.constant_modes[d][mode_index];
              }
          }
        (void)expected_mode_size;

        parameter_list.set("null space: type", "pre-computed");
        parameter_list.set("null space: dimension",
                           distributed_constant_modes.NumVectors());
        if (my_size > 0)
          parameter_list.set("null space: vectors",
                             distributed_constant_modes.Values());
        // We need to set a valid pointer to data even if there is no data on
        // the current processor. Therefore, pass a dummy in that case
        else
          parameter_list.set("null space: vectors", dummy.data());
      }

    initialize(matrix, parameter_list);

    if (additional_data.output_details)
      {
        ML_Epetra::MultiLevelPreconditioner *multilevel_operator =
          dynamic_cast<ML_Epetra::MultiLevelPreconditioner *>(
            preconditioner.get());
        Assert(multilevel_operator != nullptr,
               ExcMessage("Preconditioner setup failed."));
        multilevel_operator->PrintUnused(0);
      }
  }



  void
  PreconditionAMG::initialize(const SparseMatrix &          matrix,
                              const Teuchos::ParameterList &ml_parameters)
  {
    initialize(matrix.trilinos_matrix(), ml_parameters);
  }



  void
  PreconditionAMG::initialize(const Epetra_RowMatrix &      matrix,
                              const Teuchos::ParameterList &ml_parameters)
  {
    preconditioner.reset();
    preconditioner =
      std::make_shared<ML_Epetra::MultiLevelPreconditioner>(matrix,
                                                            ml_parameters);
  }



  template <typename number>
  void
  PreconditionAMG::initialize(
    const ::SparseMatrix<number> &deal_ii_sparse_matrix,
    const AdditionalData &                additional_data,
    const double                          drop_tolerance,
    const ::SparsityPattern *     use_this_sparsity)
  {
    preconditioner.reset();
    const size_type n_rows = deal_ii_sparse_matrix.m();

    // Init Epetra Matrix using an
    // equidistributed map; avoid
    // storing the nonzero
    // elements.
    vector_distributor = std::make_shared<Epetra_Map>(
      static_cast<TrilinosWrappers::types::int_type>(n_rows), 0, communicator);

    if (trilinos_matrix.get() == nullptr)
      trilinos_matrix = std::make_shared<SparseMatrix>();

    trilinos_matrix->reinit(*vector_distributor,
                            *vector_distributor,
                            deal_ii_sparse_matrix,
                            drop_tolerance,
                            true,
                            use_this_sparsity);

    initialize(*trilinos_matrix, additional_data);
  }



  void
  PreconditionAMG::reinit()
  {
    ML_Epetra::MultiLevelPreconditioner *multilevel_operator =
      dynamic_cast<ML_Epetra::MultiLevelPreconditioner *>(preconditioner.get());
    multilevel_operator->ReComputePreconditioner();
  }



  void
  PreconditionAMG::clear()
  {
    PreconditionBase::clear();
    trilinos_matrix.reset();
  }



  PreconditionAMG::size_type
  PreconditionAMG::memory_consumption() const
  {
    unsigned int memory = sizeof(*this);

    // todo: find a way to read out ML's data
    // sizes
    if (trilinos_matrix.get() != nullptr)
      memory += trilinos_matrix->memory_consumption();
    return memory;
  }



  // explicit instantiations
  template void
  PreconditionAMG::initialize(const ::SparseMatrix<double> &,
                              const AdditionalData &,
                              const double,
                              const ::SparsityPattern *);
  template void
  PreconditionAMG::initialize(const ::SparseMatrix<float> &,
                              const AdditionalData &,
                              const double,
                              const ::SparsityPattern *);



} // namespace TrilinosWrappers

DEAL_II_NAMESPACE_CLOSE

#endif // DEAL_II_WITH_TRILINOS