trilinos_block_sparse_matrix.cc

Go to the documentation of this file.

// -----------------------------------------------------------------------------
//
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception OR LGPL-2.1-or-later
// Copyright (C) 2008 - 2026 by the deal.II authors
//
// This file is part of the deal.II library.
//
// Detailed license information governing the source code and contributions
// can be found in LICENSE.md and CONTRIBUTING.md at the top level directory.
//
// -----------------------------------------------------------------------------
 
#include <deal.II/lac/trilinos_block_sparse_matrix.h>
 
#ifdef DEAL_II_WITH_TRILINOS
 
#  include <deal.II/lac/block_sparse_matrix.h>
#  include <deal.II/lac/block_sparsity_pattern.h>
 
 
#endif
 
DEAL_II_NAMESPACE_OPEN
 
#ifdef DEAL_II_WITH_TRILINOS
 
namespace TrilinosWrappers
{
  BlockSparseMatrix::~BlockSparseMatrix()
  {
    // delete previous content of
    // the subobjects array
    try
      {
        clear();
      }
    catch (...)
      {}
  }
 
 
 
#  ifndef DOXYGEN
  void
  BlockSparseMatrix::reinit(const size_type n_block_rows,
                            const size_type n_block_columns)
  {
    // first delete previous content of
    // the subobjects array
    clear();
 
    // then resize. set sizes of blocks to
    // zero. user will later have to call
    // collect_sizes for this
    this->sub_objects.reinit(n_block_rows, n_block_columns);
    this->row_block_indices.reinit(n_block_rows, 0);
    this->column_block_indices.reinit(n_block_columns, 0);
 
    // and reinitialize the blocks
    for (size_type r = 0; r < this->n_block_rows(); ++r)
      for (size_type c = 0; c < this->n_block_cols(); ++c)
        {
          BlockType *p = new BlockType();
 
          Assert(this->sub_objects[r][c] == nullptr, ExcInternalError());
          this->sub_objects[r][c] = p;
        }
  }
#  endif
 
 
 
  template <typename BlockSparsityPatternType>
  void
  BlockSparseMatrix::reinit(
    const std::vector<IndexSet>    &parallel_partitioning,
    const BlockSparsityPatternType &block_sparsity_pattern,
    const MPI_Comm                  communicator,
    const bool                      exchange_data)
  {
    std::vector<Epetra_Map> epetra_maps;
    epetra_maps.reserve(block_sparsity_pattern.n_block_rows());
    for (size_type i = 0; i < block_sparsity_pattern.n_block_rows(); ++i)
      epetra_maps.push_back(
        parallel_partitioning[i].make_trilinos_map(communicator, false));
 
    Assert(epetra_maps.size() == block_sparsity_pattern.n_block_rows(),
           ExcDimensionMismatch(epetra_maps.size(),
                                block_sparsity_pattern.n_block_rows()));
    Assert(epetra_maps.size() == block_sparsity_pattern.n_block_cols(),
           ExcDimensionMismatch(epetra_maps.size(),
                                block_sparsity_pattern.n_block_cols()));
 
    const size_type n_block_rows = epetra_maps.size();
    Assert(n_block_rows == block_sparsity_pattern.n_block_rows(),
           ExcDimensionMismatch(n_block_rows,
                                block_sparsity_pattern.n_block_rows()));
    Assert(n_block_rows == block_sparsity_pattern.n_block_cols(),
           ExcDimensionMismatch(n_block_rows,
                                block_sparsity_pattern.n_block_cols()));
 
    // Call the other basic reinit function, ...
    reinit(block_sparsity_pattern.n_block_rows(),
           block_sparsity_pattern.n_block_cols());
 
    // ... set the correct sizes, ...
    this->row_block_indices    = block_sparsity_pattern.get_row_indices();
    this->column_block_indices = block_sparsity_pattern.get_column_indices();
 
    // ... and then assign the correct
    // data to the blocks.
    for (size_type r = 0; r < this->n_block_rows(); ++r)
      for (size_type c = 0; c < this->n_block_cols(); ++c)
        {
          this->sub_objects[r][c]->reinit(parallel_partitioning[r],
                                          parallel_partitioning[c],
                                          block_sparsity_pattern.block(r, c),
                                          communicator,
                                          exchange_data);
        }
  }
 
 
 
  template <typename BlockSparsityPatternType>
  void
  BlockSparseMatrix::reinit(
    const BlockSparsityPatternType &block_sparsity_pattern)
  {
    std::vector<IndexSet> parallel_partitioning;
    parallel_partitioning.reserve(block_sparsity_pattern.n_block_rows());
    for (size_type i = 0; i < block_sparsity_pattern.n_block_rows(); ++i)
      parallel_partitioning.emplace_back(
        complete_index_set(block_sparsity_pattern.block(i, 0).n_rows()));
 
    reinit(parallel_partitioning, block_sparsity_pattern);
  }
 
 
 
  template <>
  void
  BlockSparseMatrix::reinit(const BlockSparsityPattern &block_sparsity_pattern)
  {
    // Call the other basic reinit function, ...
    reinit(block_sparsity_pattern.n_block_rows(),
           block_sparsity_pattern.n_block_cols());
 
    // ... set the correct sizes, ...
    this->row_block_indices    = block_sparsity_pattern.get_row_indices();
    this->column_block_indices = block_sparsity_pattern.get_column_indices();
 
    // ... and then assign the correct
    // data to the blocks.
    for (size_type r = 0; r < this->n_block_rows(); ++r)
      for (size_type c = 0; c < this->n_block_cols(); ++c)
        {
          this->sub_objects[r][c]->reinit(block_sparsity_pattern.block(r, c));
        }
  }
 
 
 
  void
  BlockSparseMatrix::reinit(
    const std::vector<IndexSet>               &parallel_partitioning,
    const ::BlockSparseMatrix<double> &dealii_block_sparse_matrix,
    const MPI_Comm                             communicator,
    const double                               drop_tolerance)
  {
    const size_type n_block_rows = parallel_partitioning.size();
 
    Assert(n_block_rows == dealii_block_sparse_matrix.n_block_rows(),
           ExcDimensionMismatch(n_block_rows,
                                dealii_block_sparse_matrix.n_block_rows()));
    Assert(n_block_rows == dealii_block_sparse_matrix.n_block_cols(),
           ExcDimensionMismatch(n_block_rows,
                                dealii_block_sparse_matrix.n_block_cols()));
 
    // Call the other basic reinit function ...
    reinit(n_block_rows, n_block_rows);
 
    // ... and then assign the correct
    // data to the blocks.
    for (size_type r = 0; r < this->n_block_rows(); ++r)
      for (size_type c = 0; c < this->n_block_cols(); ++c)
        {
          this->sub_objects[r][c]->reinit(parallel_partitioning[r],
                                          parallel_partitioning[c],
                                          dealii_block_sparse_matrix.block(r,
                                                                           c),
                                          communicator,
                                          drop_tolerance);
        }
 
    collect_sizes();
  }
 
 
 
  void
  BlockSparseMatrix::reinit(
    const ::BlockSparseMatrix<double> &dealii_block_sparse_matrix,
    const double                               drop_tolerance)
  {
    Assert(dealii_block_sparse_matrix.n_block_rows() ==
             dealii_block_sparse_matrix.n_block_cols(),
           ExcDimensionMismatch(dealii_block_sparse_matrix.n_block_rows(),
                                dealii_block_sparse_matrix.n_block_cols()));
    Assert(dealii_block_sparse_matrix.m() == dealii_block_sparse_matrix.n(),
           ExcDimensionMismatch(dealii_block_sparse_matrix.m(),
                                dealii_block_sparse_matrix.n()));
 
    std::vector<IndexSet> parallel_partitioning;
    parallel_partitioning.reserve(dealii_block_sparse_matrix.n_block_rows());
    for (size_type i = 0; i < dealii_block_sparse_matrix.n_block_rows(); ++i)
      parallel_partitioning.emplace_back(
        complete_index_set(dealii_block_sparse_matrix.block(i, 0).m()));
 
    reinit(parallel_partitioning,
           dealii_block_sparse_matrix,
           MPI_COMM_SELF,
           drop_tolerance);
  }
 
 
 
  void
  BlockSparseMatrix::collect_sizes()
  {
    // simply forward to the (non-public) function of the base class
    BaseClass::collect_sizes();
  }
 
 
 
  std::uint64_t
  BlockSparseMatrix::n_nonzero_elements() const
  {
    std::uint64_t n_nonzero = 0;
    for (size_type rows = 0; rows < this->n_block_rows(); ++rows)
      for (size_type cols = 0; cols < this->n_block_cols(); ++cols)
        n_nonzero += this->block(rows, cols).n_nonzero_elements();
 
    return n_nonzero;
  }
 
 
 
  TrilinosScalar
  BlockSparseMatrix::residual(MPI::BlockVector       &dst,
                              const MPI::BlockVector &x,
                              const MPI::BlockVector &b) const
  {
    vmult(dst, x);
    dst -= b;
    dst *= -1.;
 
    return dst.l2_norm();
  }
 
 
 
  // TODO: In the following we
  // use the same code as just
  // above three more times. Use
  // templates.
  TrilinosScalar
  BlockSparseMatrix::residual(MPI::BlockVector       &dst,
                              const MPI::Vector      &x,
                              const MPI::BlockVector &b) const
  {
    vmult(dst, x);
    dst -= b;
    dst *= -1.;
 
    return dst.l2_norm();
  }
 
 
 
  TrilinosScalar
  BlockSparseMatrix::residual(MPI::Vector            &dst,
                              const MPI::BlockVector &x,
                              const MPI::Vector      &b) const
  {
    vmult(dst, x);
    dst -= b;
    dst *= -1.;
 
    return dst.l2_norm();
  }
 
 
 
  TrilinosScalar
  BlockSparseMatrix::residual(MPI::Vector       &dst,
                              const MPI::Vector &x,
                              const MPI::Vector &b) const
  {
    vmult(dst, x);
    dst -= b;
    dst *= -1.;
 
    return dst.l2_norm();
  }
 
 
 
  MPI_Comm
  BlockSparseMatrix::get_mpi_communicator() const
  {
    Assert(this->n_block_cols() != 0, ExcNotInitialized());
    Assert(this->n_block_rows() != 0, ExcNotInitialized());
    return this->sub_objects[0][0]->get_mpi_communicator();
  }
 
 
 
#  ifndef DOXYGEN
  // -------------------- explicit instantiations -----------------------
  //
  template void
  BlockSparseMatrix::reinit(const ::BlockSparsityPattern &);
  template void
  BlockSparseMatrix::reinit(const ::BlockDynamicSparsityPattern &);
 
  template void
  BlockSparseMatrix::reinit(const std::vector<IndexSet> &,
                            const ::BlockDynamicSparsityPattern &,
                            const MPI_Comm,
                            const bool);
#  endif // DOXYGEN
 
} // namespace TrilinosWrappers
 
 
 
#endif
DEAL_II_NAMESPACE_CLOSE