shared_tria.cc

Go to the documentation of this file.

// ------------------------------------------------------------------------
//
// SPDX-License-Identifier: LGPL-2.1-or-later
// Copyright (C) 2015 - 2024 by the deal.II authors
//
// This file is part of the deal.II library.
//
// Part of the source code is dual licensed under Apache-2.0 WITH
// LLVM-exception OR LGPL-2.1-or-later. Detailed license information
// governing the source code and code contributions can be found in
// LICENSE.md and CONTRIBUTING.md at the top level directory of deal.II.
//
// ------------------------------------------------------------------------
 
#include <deal.II/base/mpi.h>
#include <deal.II/base/mpi.templates.h>
#include <deal.II/base/utilities.h>
 
#include <deal.II/distributed/shared_tria.h>
#include <deal.II/distributed/tria.h>
 
#include <deal.II/grid/filtered_iterator.h>
#include <deal.II/grid/grid_tools.h>
#include <deal.II/grid/tria.h>
#include <deal.II/grid/tria_accessor.h>
#include <deal.II/grid/tria_iterator.h>
 
#include <deal.II/lac/sparsity_tools.h>
 
#include <type_traits>
 
 
DEAL_II_NAMESPACE_OPEN
 
#ifdef DEAL_II_WITH_MPI
namespace parallel
{
  namespace shared
  {
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    Triangulation<dim, spacedim>::Triangulation(
      const MPI_Comm mpi_communicator,
      const typename ::Triangulation<dim, spacedim>::MeshSmoothing
                     smooth_grid,
      const bool     allow_artificial_cells,
      const Settings settings)
      : ::parallel::TriangulationBase<dim, spacedim>(mpi_communicator,
                                                           smooth_grid,
                                                           false)
      , settings(settings)
      , allow_artificial_cells(allow_artificial_cells)
    {
      const auto partition_settings =
        (partition_zoltan | partition_metis | partition_zorder |
         partition_custom_signal) &
        settings;
      (void)partition_settings;
      Assert(partition_settings == partition_auto ||
               partition_settings == partition_metis ||
               partition_settings == partition_zoltan ||
               partition_settings == partition_zorder ||
               partition_settings == partition_custom_signal,
             ExcMessage("Settings must contain exactly one type of the active "
                        "cell partitioning scheme."));
 
      if (settings & construct_multigrid_hierarchy)
        Assert(allow_artificial_cells,
               ExcMessage("construct_multigrid_hierarchy requires "
                          "allow_artificial_cells to be set to true."));
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    bool Triangulation<dim, spacedim>::is_multilevel_hierarchy_constructed()
      const
    {
      return (settings & construct_multigrid_hierarchy);
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    void Triangulation<dim, spacedim>::partition()
    {
#  ifdef DEBUG
      // Check that all meshes are the same (or at least have the same
      // total number of active cells):
      const unsigned int max_active_cells =
        Utilities::MPI::max(this->n_active_cells(),
                            this->get_mpi_communicator());
      Assert(
        max_active_cells == this->n_active_cells(),
        ExcMessage(
          "A parallel::shared::Triangulation needs to be refined in the same "
          "way on all processors, but the participating processors don't "
          "agree on the number of active cells."));
#  endif
 
      auto partition_settings = (partition_zoltan | partition_metis |
                                 partition_zorder | partition_custom_signal) &
                                settings;
      if (partition_settings == partition_auto)
#  ifdef DEAL_II_TRILINOS_WITH_ZOLTAN
        partition_settings = partition_zoltan;
#  elif defined DEAL_II_WITH_METIS
        partition_settings = partition_metis;
#  else
        partition_settings = partition_zorder;
#  endif
 
      if (partition_settings == partition_zoltan)
        {
#  ifndef DEAL_II_TRILINOS_WITH_ZOLTAN
          AssertThrow(false,
                      ExcMessage(
                        "Choosing 'partition_zoltan' requires the library "
                        "to be compiled with support for Zoltan! "
                        "Instead, you might use 'partition_auto' to select "
                        "a partitioning algorithm that is supported "
                        "by your current configuration."));
#  else
          GridTools::partition_triangulation(
            this->n_subdomains, *this, SparsityTools::Partitioner::zoltan);
#  endif
        }
      else if (partition_settings == partition_metis)
        {
#  ifndef DEAL_II_WITH_METIS
          AssertThrow(false,
                      ExcMessage(
                        "Choosing 'partition_metis' requires the library "
                        "to be compiled with support for METIS! "
                        "Instead, you might use 'partition_auto' to select "
                        "a partitioning algorithm that is supported "
                        "by your current configuration."));
#  else
          GridTools::partition_triangulation(this->n_subdomains,
                                             *this,
                                             SparsityTools::Partitioner::metis);
#  endif
        }
      else if (partition_settings == partition_zorder)
        {
          GridTools::partition_triangulation_zorder(this->n_subdomains, *this);
        }
      else if (partition_settings == partition_custom_signal)
        {
          // User partitions mesh manually
        }
      else
        {
          AssertThrow(false, ExcInternalError());
        }
 
      // do not partition multigrid levels if user is
      // defining a custom partition
      if ((settings & construct_multigrid_hierarchy) &&
          !(settings & partition_custom_signal))
        ::GridTools::partition_multigrid_levels(*this);
 
      true_subdomain_ids_of_cells.resize(this->n_active_cells());
 
      // loop over all cells and mark artificial:
      typename parallel::shared::Triangulation<dim,
                                               spacedim>::active_cell_iterator
        cell = this->begin_active(),
        endc = this->end();
 
      if (allow_artificial_cells)
        {
          // get active halo layer of (ghost) cells
          // parallel::shared::Triangulation<dim>::
          std::function<bool(
            const typename parallel::shared::Triangulation<dim, spacedim>::
              active_cell_iterator &)>
            predicate = IteratorFilters::SubdomainEqualTo(this->my_subdomain);
 
          const std::vector<typename parallel::shared::Triangulation<
            dim,
            spacedim>::active_cell_iterator>
            active_halo_layer_vector =
              ::GridTools::compute_active_cell_halo_layer(*this,
                                                                predicate);
          std::set<typename parallel::shared::Triangulation<dim, spacedim>::
                     active_cell_iterator>
            active_halo_layer(active_halo_layer_vector.begin(),
                              active_halo_layer_vector.end());
 
          for (unsigned int index = 0; cell != endc; cell++, index++)
            {
              // store original/true subdomain ids:
              true_subdomain_ids_of_cells[index] = cell->subdomain_id();
 
              if (cell->is_locally_owned() == false &&
                  active_halo_layer.find(cell) == active_halo_layer.end())
                cell->set_subdomain_id(numbers::artificial_subdomain_id);
            }
 
          // loop over all cells in multigrid hierarchy and mark artificial:
          if (settings & construct_multigrid_hierarchy)
            {
              true_level_subdomain_ids_of_cells.resize(this->n_levels());
 
              std::function<bool(
                const typename parallel::shared::Triangulation<dim, spacedim>::
                  cell_iterator &)>
                predicate = IteratorFilters::LocallyOwnedLevelCell();
              for (unsigned int lvl = 0; lvl < this->n_levels(); ++lvl)
                {
                  true_level_subdomain_ids_of_cells[lvl].resize(
                    this->n_cells(lvl));
 
                  const std::vector<typename parallel::shared::Triangulation<
                    dim,
                    spacedim>::cell_iterator>
                    level_halo_layer_vector =
                      ::GridTools::compute_cell_halo_layer_on_level(
                        *this, predicate, lvl);
                  std::set<typename parallel::shared::
                             Triangulation<dim, spacedim>::cell_iterator>
                    level_halo_layer(level_halo_layer_vector.begin(),
                                     level_halo_layer_vector.end());
 
                  typename parallel::shared::Triangulation<dim, spacedim>::
                    cell_iterator cell = this->begin(lvl),
                                  endc = this->end(lvl);
                  for (unsigned int index = 0; cell != endc; cell++, index++)
                    {
                      // Store true level subdomain IDs before setting
                      // artificial
                      true_level_subdomain_ids_of_cells[lvl][index] =
                        cell->level_subdomain_id();
 
                      // for active cells, we must have knowledge of level
                      // subdomain ids of all neighbors to our subdomain, not
                      // just neighbors on the same level. if the cells
                      // subdomain id was not set to artitficial above, we will
                      // also keep its level subdomain id since it is either
                      // owned by this processor or in the ghost layer of the
                      // active mesh.
                      if (cell->is_active() &&
                          cell->subdomain_id() !=
                            numbers::artificial_subdomain_id)
                        continue;
 
                      // we must have knowledge of our parent in the hierarchy
                      if (cell->has_children())
                        {
                          bool keep_cell = false;
                          for (unsigned int c = 0;
                               c < GeometryInfo<dim>::max_children_per_cell;
                               ++c)
                            if (cell->child(c)->level_subdomain_id() ==
                                this->my_subdomain)
                              {
                                keep_cell = true;
                                break;
                              }
                          if (keep_cell)
                            continue;
                        }
 
                      // we must have knowledge of our neighbors on the same
                      // level
                      if (!cell->is_locally_owned_on_level() &&
                          level_halo_layer.find(cell) != level_halo_layer.end())
                        continue;
 
                      // mark all other cells to artificial
                      cell->set_level_subdomain_id(
                        numbers::artificial_subdomain_id);
                    }
                }
            }
        }
      else
        {
          // just store true subdomain ids
          for (unsigned int index = 0; cell != endc; cell++, index++)
            true_subdomain_ids_of_cells[index] = cell->subdomain_id();
        }
 
#  ifdef DEBUG
      {
        // Assert that each cell is owned by a processor
        const unsigned int n_my_cells = std::count_if(
          this->begin_active(),
          typename Triangulation<dim, spacedim>::active_cell_iterator(
            this->end()),
          [](const auto &i) { return (i.is_locally_owned()); });
 
        const unsigned int total_cells =
          Utilities::MPI::sum(n_my_cells, this->get_mpi_communicator());
        Assert(total_cells == this->n_active_cells(),
               ExcMessage("Not all cells are assigned to a processor."));
      }
 
      // If running with multigrid, assert that each level
      // cell is owned by a processor
      if (settings & construct_multigrid_hierarchy)
        {
          const unsigned int n_my_cells =
            std::count_if(this->begin(), this->end(), [](const auto &i) {
              return (i.is_locally_owned_on_level());
            });
 
 
          const unsigned int total_cells =
            Utilities::MPI::sum(n_my_cells, this->get_mpi_communicator());
          Assert(total_cells == this->n_cells(),
                 ExcMessage("Not all cells are assigned to a processor."));
        }
#  endif
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    bool Triangulation<dim, spacedim>::with_artificial_cells() const
    {
      return allow_artificial_cells;
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    const std::vector<types::subdomain_id>
      &Triangulation<dim, spacedim>::get_true_subdomain_ids_of_cells() const
    {
      return true_subdomain_ids_of_cells;
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    const std::vector<types::subdomain_id>
      &Triangulation<dim, spacedim>::get_true_level_subdomain_ids_of_cells(
        const unsigned int level) const
    {
      Assert(level < true_level_subdomain_ids_of_cells.size(),
             ExcInternalError());
      Assert(true_level_subdomain_ids_of_cells[level].size() ==
               this->n_cells(level),
             ExcInternalError());
      return true_level_subdomain_ids_of_cells[level];
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    void Triangulation<dim, spacedim>::execute_coarsening_and_refinement()
    {
      // make sure that all refinement/coarsening flags are the same on all
      // processes
      {
        // Obtain the type used to store the different possibilities
        // a cell can be refined. This is a bit awkward because
        // what `cell->refine_flag_set()` returns is a struct
        // type, RefinementCase, which internally stores a
        // std::uint8_t, which actually holds integers of
        // enum type RefinementPossibilities<dim>::Possibilities.
        // In the following, use the actual name of the enum, but
        // make sure that it is in fact a `std::uint8_t` or
        // equally sized type.
        using int_type = std::underlying_type_t<
          typename RefinementPossibilities<dim>::Possibilities>;
        static_assert(sizeof(int_type) == sizeof(std::uint8_t),
                      "Internal type mismatch.");
 
        std::vector<int_type> refinement_configurations(this->n_active_cells() *
                                                          2,
                                                        int_type(0));
        for (const auto &cell : this->active_cell_iterators())
          if (cell->is_locally_owned())
            {
              refinement_configurations[cell->active_cell_index() * 2 + 0] =
                static_cast<int_type>(cell->refine_flag_set());
              refinement_configurations[cell->active_cell_index() * 2 + 1] =
                static_cast<int_type>(cell->coarsen_flag_set() ? 1 : 0);
            }
 
        Utilities::MPI::max(refinement_configurations,
                            this->get_mpi_communicator(),
                            refinement_configurations);
 
        for (const auto &cell : this->active_cell_iterators())
          {
            cell->clear_refine_flag();
            cell->clear_coarsen_flag();
 
            Assert(
              (refinement_configurations[cell->active_cell_index() * 2 + 0] >
                   0 ?
                 1 :
                 0) +
                  refinement_configurations[cell->active_cell_index() * 2 +
                                            1] <=
                1,
              ExcMessage(
                "Refinement/coarsening flags of cells are not consistent in parallel!"));
 
            if (refinement_configurations[cell->active_cell_index() * 2 + 0] !=
                0)
              cell->set_refine_flag(RefinementCase<dim>(
                refinement_configurations[cell->active_cell_index() * 2 + 0]));
 
            if (refinement_configurations[cell->active_cell_index() * 2 + 1] >
                0)
              cell->set_coarsen_flag();
          }
      }
 
      ::Triangulation<dim, spacedim>::execute_coarsening_and_refinement();
      partition();
      this->update_number_cache();
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    void Triangulation<dim, spacedim>::create_triangulation(
      const std::vector<Point<spacedim>> &vertices,
      const std::vector<CellData<dim>>   &cells,
      const SubCellData                  &subcelldata)
    {
      try
        {
          ::Triangulation<dim, spacedim>::create_triangulation(
            vertices, cells, subcelldata);
        }
      catch (
        const typename ::Triangulation<dim, spacedim>::DistortedCellList
          &)
        {
          // the underlying triangulation should not be checking for distorted
          // cells
          DEAL_II_ASSERT_UNREACHABLE();
        }
      partition();
      this->update_number_cache();
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    void Triangulation<dim, spacedim>::create_triangulation(
      const TriangulationDescription::Description<dim, spacedim>
        &construction_data)
    {
      (void)construction_data;
 
      DEAL_II_ASSERT_UNREACHABLE();
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    void Triangulation<dim, spacedim>::copy_triangulation(
      const ::Triangulation<dim, spacedim> &other_tria)
    {
      Assert(
        (dynamic_cast<
           const ::parallel::DistributedTriangulationBase<dim, spacedim>
             *>(&other_tria) == nullptr),
        ExcMessage(
          "Cannot use this function on parallel::distributed::Triangulation."));
 
      ::parallel::TriangulationBase<dim, spacedim>::copy_triangulation(
        other_tria);
      partition();
      this->update_number_cache();
    }
  } // namespace shared
} // namespace parallel
 
#else
 
namespace parallel
{
  namespace shared
  {
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    bool Triangulation<dim, spacedim>::with_artificial_cells() const
    {
      DEAL_II_NOT_IMPLEMENTED();
      return true;
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    bool Triangulation<dim, spacedim>::is_multilevel_hierarchy_constructed()
      const
    {
      return false;
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    const std::vector<unsigned int>
      &Triangulation<dim, spacedim>::get_true_subdomain_ids_of_cells() const
    {
      DEAL_II_NOT_IMPLEMENTED();
      return true_subdomain_ids_of_cells;
    }
 
 
 
    template <int dim, int spacedim>
    DEAL_II_CXX20_REQUIRES((concepts::is_valid_dim_spacedim<dim, spacedim>))
    const std::vector<unsigned int>
      &Triangulation<dim, spacedim>::get_true_level_subdomain_ids_of_cells(
        const unsigned int) const
    {
      DEAL_II_NOT_IMPLEMENTED();
      return true_level_subdomain_ids_of_cells;
    }
  } // namespace shared
} // namespace parallel
 
 
#endif
 
 
 
namespace internal
{
  namespace parallel
  {
    namespace shared
    {
      template <int dim, int spacedim>
      TemporarilyRestoreSubdomainIds<dim, spacedim>::
        TemporarilyRestoreSubdomainIds(const Triangulation<dim, spacedim> &tria)
        : shared_tria(
            dynamic_cast<
              const ::parallel::shared::Triangulation<dim, spacedim> *>(
              &tria))
      {
        if (shared_tria && shared_tria->with_artificial_cells())
          {
            // Save the current set of subdomain IDs, and set subdomain IDs
            // to the "true" owner of each cell.
            const std::vector<types::subdomain_id> &true_subdomain_ids =
              shared_tria->get_true_subdomain_ids_of_cells();
 
            saved_subdomain_ids.resize(shared_tria->n_active_cells());
            for (const auto &cell : shared_tria->active_cell_iterators())
              {
                const unsigned int index   = cell->active_cell_index();
                saved_subdomain_ids[index] = cell->subdomain_id();
                cell->set_subdomain_id(true_subdomain_ids[index]);
              }
          }
      }
 
 
 
      template <int dim, int spacedim>
      TemporarilyRestoreSubdomainIds<dim, spacedim>::
        ~TemporarilyRestoreSubdomainIds()
      {
        if (shared_tria && shared_tria->with_artificial_cells())
          {
            // Undo the subdomain modification.
            for (const auto &cell : shared_tria->active_cell_iterators())
              {
                const unsigned int index = cell->active_cell_index();
                cell->set_subdomain_id(saved_subdomain_ids[index]);
              }
          }
      }
    } // namespace shared
  }   // namespace parallel
} // namespace internal
 
 
/*-------------- Explicit Instantiations -------------------------------*/
#include "shared_tria.inst"
 
DEAL_II_NAMESPACE_CLOSE