deal.II version GIT relicensing-2206-gaa53ff9447 2024-12-02 09:10:00+00:00
|
A namespace for TriangulationDescription::Description utility functions.
Description< dim, spacedim > TriangulationDescription::Utilities::create_description_from_triangulation | ( | const ::Triangulation< dim, spacedim > & | tria, |
const MPI_Comm | comm, | ||
const TriangulationDescription::Settings | settings = TriangulationDescription::Settings::default_setting , |
||
const unsigned int | my_rank_in = numbers::invalid_unsigned_int |
||
) |
Construct a TriangulationDescription::Description object that for the current process, using a given triangulation tria
describes how one can re-create a partitioned triangulation (such as a parallel::fullydistributed::Triangulation). The input triangulation can be either a serial triangulation of type Triangulation which has been colored (i.e., for which the subdomain_id and/or level_subdomain_id have been set to create partitions), or a distributed triangulation of type parallel::distributed::Triangulation, where the output of this function takes over the existing partitioning unaltered.
An example for a serial Triangulation looks like this:
In this example, all processes in a parallel universe read the same mesh, and partition it in the same way based on the subdomain ids set by GridTools::partition_triangulation(). Then, on each process, the current function creates building blocks that extract information about those cells whose subdomain ids match the current process's rank (plus perhaps other necessary information), and in a last step, we create a fully distributed triangulation out of these building blocks.
In contrast, when starting with a parallel::distributed::Triangulation object, partitioning can be skipped, since the triangulation has already been partitioned and the corresponding code looks as follows:
This code therefore takes a parallel::distributed::Triangulation object (such as the ones used, for example, in step-40 and many other tutorial programs) and converts it into a parallel::fullydistributed::Triangulation.
[in] | tria | The input triangulation. This function uses the partitioning provided by either explicitly setting subdomain_ids (if the triangulation is sequential) or based on how the existing triangulation is already distributed to different processes. |
[in] | comm | The MPI communicator to be used. If the input triangulation is parallel, then the communicators have to match. |
[in] | settings | See the description of the Settings enumerator. |
[in] | my_rank_in | Construct Description for the specified rank. This parameter can only be set for serial input triangulations that have been partitioned by functions like GridTools::partition_triangulation(). For parallel input triangulations, my_rank_in needs to equal to the default value, or to the rank of the current process within the given communicator. |
construct_multigrid_hierarchy
is set in the settings
, then the input triangulation has to be set up with limit_level_difference_at_vertices
among the triangulation's smoothing flags provided to the triangulation constructor. Definition at line 717 of file tria_description.cc.
Description< dim, spacedim > TriangulationDescription::Utilities::create_description_from_triangulation | ( | const Triangulation< dim, spacedim > & | tria, |
const LinearAlgebra::distributed::Vector< double, MemorySpace::Host > & | partition, | ||
const TriangulationDescription::Settings | settings = TriangulationDescription::Settings::default_setting |
||
) |
Similar to the above function but the desired owners of active cells are not provided via the subdomain_ids of cells, but rather by the corresponding values of a cell vector (see also parallel::TriangulationBase::global_active_cell_index_partitioner() and CellAccessor::global_active_cell_index()). This function allows repartitioning Triangulation objects that are already distributed and that may be partitioned differently than the desired partitioning.
If the setup of multigrid levels is requested by the settings
argument, they are partitioned according to a first-child policy. In other words, a process owns a (non-active) cell if it owns its first child. If that first child is not active itself, the policy is applied recursively.
The partitioning of cells is determined based on the elements of the partition
vector. While that vector stores elements of type double
, actual values must be integers and be within the range of process ranks of the relevant communicator.
partition
.tria
can be set up on a subcommunicator of the communicator of partitioner
. All processes that are not part of that subcommunicator need to set up the local triangulation with the special-purpose communicator MPI_COMM_NULL.partition
only describes the partitioning of the active level. Description< dim, spacedim > TriangulationDescription::Utilities::create_description_from_triangulation | ( | const Triangulation< dim, spacedim > & | tria, |
const LinearAlgebra::distributed::Vector< double, MemorySpace::Host > & | partition, | ||
const std::vector< LinearAlgebra::distributed::Vector< double, MemorySpace::Host > > & | mg_partitions, | ||
const TriangulationDescription::Settings | settings = TriangulationDescription::Settings::default_setting |
||
) |
Similar to the above function but allowing the user to prescribe the partitioning of the multigrid levels. As with the other function, while both the global and level-wise partition vectors store elements with data type double
, actual elements must be integers and equal possible process ranks within the relevant communicator.
Description< dim, spacedim > TriangulationDescription::Utilities::create_description_from_triangulation_in_groups | ( | const std::function< void(::Triangulation< dim, spacedim > &)> & | serial_grid_generator, |
const std::function< void(::Triangulation< dim, spacedim > &, const MPI_Comm, const unsigned int)> & | serial_grid_partitioner, | ||
const MPI_Comm | comm, | ||
const int | group_size = 1 , |
||
const typename Triangulation< dim, spacedim >::MeshSmoothing | smoothing = ::Triangulation<dim, spacedim>::none , |
||
const TriangulationDescription::Settings | setting = TriangulationDescription::Settings::default_setting |
||
) |
Construct a TriangulationDescription::Description. In contrast to the function above, this function is also responsible for creating a serial triangulation and for its partitioning (by calling the provided std::function
objects). Internally only selected processes ( every n-th/each root of a group of size group_size) create a serial triangulation and the TriangulationDescription::Description for all processes in its group, which is communicated.
This function can also be used to read an external mesh only once (by the root process and a group consisting all processes). The following code snippet shows a modified version of the example provided in the documentation of create_description_from_triangulation(). Function calls that only need to be performed by the root process (read and partition mesh) have been moved into std::function
objects.
serial_grid_generator | A function which creates a serial triangulation. |
serial_grid_partitioner | A function which can partition a serial triangulation, i.e., sets the sudomain_ids of the active cells. The function takes as the first argument a serial triangulation, as the second argument the MPI communicator, and as the third argument the group size. |
comm | MPI communicator. |
group_size | The size of each group. |
smoothing | Mesh smoothing type. |
setting | See the description of the Settings enumerator. |
smoothing
parameter is extended with the limit_level_difference_at_vertices flag. Definition at line 794 of file tria_description.cc.
Description< dim, spacedim > TriangulationDescription::Utilities::create_description_from_triangulation | ( | const Triangulation< dim, spacedim > & | tria, |
const LinearAlgebra::distributed::Vector< double > & | partition, | ||
const TriangulationDescription::Settings | settings | ||
) |
Definition at line 924 of file tria_description.cc.
Description< dim, spacedim > TriangulationDescription::Utilities::create_description_from_triangulation | ( | const Triangulation< dim, spacedim > & | tria, |
const LinearAlgebra::distributed::Vector< double > & | partition, | ||
const std::vector< LinearAlgebra::distributed::Vector< double > > & | partitions_mg, | ||
const TriangulationDescription::Settings | settings_in | ||
) |
Definition at line 1008 of file tria_description.cc.