From: Peter Munch Date: Sat, 5 Mar 2022 20:59:25 +0000 (+0100) Subject: GC: enable inplace operation for coarse side X-Git-Tag: v9.4.0-rc1~397^2 X-Git-Url: https://gitweb.dealii.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=refs%2Fpull%2F13494%2Fhead;p=dealii.git GC: enable inplace operation for coarse side --- diff --git a/include/deal.II/multigrid/mg_transfer_global_coarsening.templates.h b/include/deal.II/multigrid/mg_transfer_global_coarsening.templates.h index 4e4e88db04..6f251fee8b 100644 --- a/include/deal.II/multigrid/mg_transfer_global_coarsening.templates.h +++ b/include/deal.II/multigrid/mg_transfer_global_coarsening.templates.h @@ -969,11 +969,58 @@ namespace internal class MGTwoLevelTransferImplementation { + template + static std::shared_ptr + create_coarse_partitioner( + const DoFHandler & dof_handler_coarse, + const dealii::AffineConstraints &constraints_coarse, + const unsigned int mg_level_coarse) + { + IndexSet locally_relevant_dofs = + (mg_level_coarse == numbers::invalid_unsigned_int) ? + DoFTools::extract_locally_active_dofs(dof_handler_coarse) : + DoFTools::extract_locally_active_level_dofs(dof_handler_coarse, + mg_level_coarse); + + std::vector locally_relevant_dofs_temp; + + for (const auto i : locally_relevant_dofs) + { + if (locally_relevant_dofs.is_element(i) == false) + locally_relevant_dofs_temp.emplace_back(i); + + const auto constraints = constraints_coarse.get_constraint_entries(i); + + if (constraints) + for (const auto &p : *constraints) + if (locally_relevant_dofs.is_element(p.first) == false) + locally_relevant_dofs_temp.emplace_back(p.first); + } + + std::sort(locally_relevant_dofs_temp.begin(), + locally_relevant_dofs_temp.end()); + + locally_relevant_dofs_temp.erase( + std::unique(locally_relevant_dofs_temp.begin(), + locally_relevant_dofs_temp.end()), + locally_relevant_dofs_temp.end()); + + locally_relevant_dofs.add_indices(locally_relevant_dofs_temp.begin(), + locally_relevant_dofs_temp.end()); + + return std::make_shared( + mg_level_coarse == numbers::invalid_unsigned_int ? + dof_handler_coarse.locally_owned_dofs() : + dof_handler_coarse.locally_owned_mg_dofs(mg_level_coarse), + locally_relevant_dofs, + dof_handler_coarse.get_communicator()); + } + template static void precompute_restriction_constraints( const std::shared_ptr &partitioner, - const dealii::AffineConstraints &constraint_coarse, + const dealii::AffineConstraints &constraints_coarse, MGTwoLevelTransfer> &transfer) { @@ -984,22 +1031,22 @@ namespace internal const auto fu = [&](const auto &index_set) { for (const auto i : index_set) { - Assert(constraint_coarse.is_inhomogeneously_constrained(i) == false, + Assert(constraints_coarse.is_inhomogeneously_constrained(i) == + false, ExcNotImplemented()); - if (constraint_coarse.is_constrained(i)) - { - const auto constraints = - constraint_coarse.get_constraint_entries(i); + const auto constraints = + constraints_coarse.get_constraint_entries(i); - if (constraints) - for (const auto &p : *constraints) - { - transfer.distribute_local_to_global_indices.emplace_back( - partitioner->global_to_local(p.first)); - transfer.distribute_local_to_global_values.emplace_back( - p.second); - } + if (constraints) + { + for (const auto &p : *constraints) + { + transfer.distribute_local_to_global_indices.emplace_back( + partitioner->global_to_local(p.first)); + transfer.distribute_local_to_global_values.emplace_back( + p.second); + } // add a dummy entry for homogeneous constraints if (transfer.distribute_local_to_global_indices.size() == @@ -1028,7 +1075,7 @@ namespace internal compute_weights( const DoFHandler & dof_handler_fine, const unsigned int mg_level_fine, - const dealii::AffineConstraints &constraint_fine, + const dealii::AffineConstraints &constraints_fine, const MGTwoLevelTransfer> & transfer, LinearAlgebra::distributed::Vector &touch_count) @@ -1064,7 +1111,7 @@ namespace internal for (unsigned int i = 0; i < touch_count_.local_size(); ++i) touch_count_.local_element(i) = - constraint_fine.is_constrained( + constraints_fine.is_constrained( touch_count_.get_partitioner()->local_to_global(i)) ? Number(0.) : Number(1.) / touch_count_.local_element(i); @@ -1200,8 +1247,8 @@ namespace internal reinit_geometric_transfer( const DoFHandler & dof_handler_fine, const DoFHandler & dof_handler_coarse, - const dealii::AffineConstraints &constraint_fine, - const dealii::AffineConstraints &constraint_coarse, + const dealii::AffineConstraints &constraints_fine, + const dealii::AffineConstraints &constraints_coarse, const unsigned int mg_level_fine, const unsigned int mg_level_coarse, MGTwoLevelTransfer> @@ -1402,25 +1449,13 @@ namespace internal // ... coarse mesh (needed since user vector might be const) { - IndexSet locally_relevant_dofs; - - if (mg_level_coarse == numbers::invalid_unsigned_int) - DoFTools::extract_locally_relevant_dofs(dof_handler_coarse, - locally_relevant_dofs); - else - DoFTools::extract_locally_relevant_level_dofs( - dof_handler_coarse, mg_level_coarse, locally_relevant_dofs); - transfer.partitioner_coarse = - std::make_shared( - mg_level_coarse == numbers::invalid_unsigned_int ? - dof_handler_coarse.locally_owned_dofs() : - dof_handler_coarse.locally_owned_mg_dofs(mg_level_coarse), - locally_relevant_dofs, - dof_handler_coarse.get_communicator()); + create_coarse_partitioner(dof_handler_coarse, + constraints_coarse, + mg_level_coarse); transfer.vec_coarse.reinit(transfer.partitioner_coarse); precompute_restriction_constraints(transfer.partitioner_coarse, - constraint_coarse, + constraints_coarse, transfer); } } @@ -1874,7 +1909,7 @@ namespace internal LinearAlgebra::distributed::Vector weight_vector; compute_weights(dof_handler_fine, mg_level_fine, - constraint_fine, + constraints_fine, transfer, weight_vector); @@ -1928,8 +1963,8 @@ namespace internal reinit_polynomial_transfer( const DoFHandler & dof_handler_fine, const DoFHandler & dof_handler_coarse, - const dealii::AffineConstraints &constraint_fine, - const dealii::AffineConstraints &constraint_coarse, + const dealii::AffineConstraints &constraints_fine, + const dealii::AffineConstraints &constraints_coarse, const unsigned int mg_level_fine, const unsigned int mg_level_coarse, MGTwoLevelTransfer> @@ -2066,26 +2101,13 @@ namespace internal transfer.vec_fine.reinit(transfer.partitioner_fine); } { - IndexSet locally_relevant_dofs; - - if (mg_level_coarse == numbers::invalid_unsigned_int) - DoFTools::extract_locally_relevant_dofs(dof_handler_coarse, - locally_relevant_dofs); - else - DoFTools::extract_locally_relevant_level_dofs(dof_handler_coarse, - mg_level_coarse, - locally_relevant_dofs); - transfer.partitioner_coarse = - std::make_shared( - mg_level_coarse == numbers::invalid_unsigned_int ? - dof_handler_coarse.locally_owned_dofs() : - dof_handler_coarse.locally_owned_mg_dofs(mg_level_coarse), - locally_relevant_dofs, - comm); + create_coarse_partitioner(dof_handler_coarse, + constraints_coarse, + mg_level_coarse); transfer.vec_coarse.reinit(transfer.partitioner_coarse); precompute_restriction_constraints(transfer.partitioner_coarse, - constraint_coarse, + constraints_coarse, transfer); } @@ -2399,7 +2421,7 @@ namespace internal LinearAlgebra::distributed::Vector weight_vector; compute_weights(dof_handler_fine, mg_level_fine, - constraint_fine, + constraints_fine, transfer, weight_vector); @@ -2660,11 +2682,15 @@ MGTwoLevelTransfer>:: const unsigned int n_lanes = VectorizedArrayType::size(); const bool use_dst_inplace = this->vec_fine.size() == 0; + const auto vec_fine_ptr = use_dst_inplace ? &dst : &this->vec_fine; - const auto vec_fine_ptr = use_dst_inplace ? &dst : &this->vec_fine; + const bool use_src_inplace = this->vec_coarse.size() == 0; + const auto vec_coarse_ptr = use_src_inplace ? &src : &this->vec_coarse; - this->vec_coarse.copy_locally_owned_data_from(src); - this->vec_coarse.update_ghost_values(); + if (use_src_inplace == false) + vec_coarse.copy_locally_owned_data_from(src); + + vec_coarse_ptr->update_ghost_values(); // a helper function similar to FEEvaluation::read_dof_values() const auto read_dof_values = [&](const auto &index, @@ -2719,6 +2745,9 @@ MGTwoLevelTransfer>:: for (const auto &scheme : schemes) { + if (scheme.n_coarse_cells == 0) + continue; + const bool needs_interpolation = (scheme.prolongation_matrix.size() == 0 && scheme.prolongation_matrix_1d.size() == 0) == false; @@ -2746,7 +2775,7 @@ MGTwoLevelTransfer>:: { for (unsigned int i = 0; i < scheme.n_dofs_per_cell_coarse; ++i) evaluation_data_coarse[i][v] = - read_dof_values(indices_coarse[i], this->vec_coarse); + read_dof_values(indices_coarse[i], *vec_coarse_ptr); indices_coarse += scheme.n_dofs_per_cell_coarse; indices_coarse_plain += scheme.n_dofs_per_cell_coarse; } @@ -2818,6 +2847,9 @@ MGTwoLevelTransfer>:: if (use_dst_inplace == false) dst += this->vec_fine; + + if (use_src_inplace) + vec_coarse_ptr->zero_out_ghost_values(); } @@ -2835,15 +2867,20 @@ MGTwoLevelTransfer>:: const bool use_src_inplace = this->vec_fine.size() == 0; const auto vec_fine_ptr = use_src_inplace ? &src : &this->vec_fine; + const bool use_dst_inplace = this->vec_coarse.size() == 0; + const auto vec_coarse_ptr = use_dst_inplace ? &dst : &this->vec_coarse; + if (use_src_inplace == false) this->vec_fine.copy_locally_owned_data_from(src); if (fine_element_is_continuous || use_src_inplace == false) vec_fine_ptr->update_ghost_values(); - this->vec_coarse.copy_locally_owned_data_from(dst); - this->vec_coarse.zero_out_ghost_values(); // since we might add into the - // ghost values and call compress + if (use_dst_inplace == false) + *vec_coarse_ptr = 0.0; + + vec_coarse_ptr->zero_out_ghost_values(); // since we might add into the + // ghost values and call compress AlignedVector evaluation_data_fine; AlignedVector evaluation_data_coarse; @@ -2888,6 +2925,9 @@ MGTwoLevelTransfer>:: for (const auto &scheme : schemes) { + if (scheme.n_coarse_cells == 0) + continue; + const bool needs_interpolation = (scheme.prolongation_matrix.size() == 0 && scheme.prolongation_matrix_1d.size() == 0) == false; @@ -2977,7 +3017,7 @@ MGTwoLevelTransfer>:: for (unsigned int i = 0; i < scheme.n_dofs_per_cell_coarse; ++i) distribute_local_to_global(indices_coarse[i], evaluation_data_coarse[i][v], - this->vec_coarse); + *vec_coarse_ptr); indices_coarse += scheme.n_dofs_per_cell_coarse; indices_coarse_plain += scheme.n_dofs_per_cell_coarse; } @@ -2992,9 +3032,10 @@ MGTwoLevelTransfer>:: else if (fine_element_is_continuous) vec_fine_ptr->zero_out_ghost_values(); // external vector - this->vec_coarse.compress(VectorOperation::add); + vec_coarse_ptr->compress(VectorOperation::add); - dst.copy_locally_owned_data_from(this->vec_coarse); + if (use_dst_inplace == false) + dst += this->vec_coarse; } @@ -3012,13 +3053,16 @@ MGTwoLevelTransfer>:: const bool use_src_inplace = this->vec_fine.size() == 0; const auto vec_fine_ptr = use_src_inplace ? &src : &this->vec_fine; + const bool use_dst_inplace = this->vec_coarse.size() == 0; + const auto vec_coarse_ptr = use_dst_inplace ? &dst : &this->vec_coarse; + if (use_src_inplace == false) this->vec_fine.copy_locally_owned_data_from(src); if (fine_element_is_continuous || use_src_inplace == false) vec_fine_ptr->update_ghost_values(); - this->vec_coarse = 0.0; + *vec_coarse_ptr = 0.0; AlignedVector evaluation_data_fine; AlignedVector evaluation_data_coarse; @@ -3038,7 +3082,7 @@ MGTwoLevelTransfer>:: if ((scheme.n_dofs_per_cell_fine != 0) && (scheme.n_dofs_per_cell_coarse != 0)) for (unsigned int i = 0; i < scheme.n_dofs_per_cell_fine; ++i) - this->vec_coarse.local_element(indices_coarse_plain[i]) = + vec_coarse_ptr->local_element(indices_coarse_plain[i]) = vec_fine_ptr->local_element(indices_fine[i]); indices_fine += scheme.n_dofs_per_cell_fine; @@ -3098,7 +3142,7 @@ MGTwoLevelTransfer>:: for (unsigned int v = 0; v < n_lanes_filled; ++v) { for (unsigned int i = 0; i < scheme.n_dofs_per_cell_coarse; ++i) - this->vec_coarse.local_element(indices_coarse_plain[i]) = + vec_coarse_ptr->local_element(indices_coarse_plain[i]) = evaluation_data_coarse[i][v]; indices_coarse_plain += scheme.n_dofs_per_cell_coarse; } @@ -3111,7 +3155,8 @@ MGTwoLevelTransfer>:: else if (fine_element_is_continuous) vec_fine_ptr->zero_out_ghost_values(); // external vector - dst.copy_locally_owned_data_from(this->vec_coarse); + if (use_dst_inplace == false) + dst.copy_locally_owned_data_from(this->vec_coarse); } @@ -3173,16 +3218,16 @@ void MGTwoLevelTransfer>:: reinit_geometric_transfer(const DoFHandler & dof_handler_fine, const DoFHandler & dof_handler_coarse, - const AffineConstraints &constraint_fine, - const AffineConstraints &constraint_coarse, + const AffineConstraints &constraints_fine, + const AffineConstraints &constraints_coarse, const unsigned int mg_level_fine, const unsigned int mg_level_coarse) { internal::MGTwoLevelTransferImplementation::reinit_geometric_transfer( dof_handler_fine, dof_handler_coarse, - constraint_fine, - constraint_coarse, + constraints_fine, + constraints_coarse, mg_level_fine, mg_level_coarse, *this); @@ -3193,18 +3238,19 @@ MGTwoLevelTransfer>:: template void MGTwoLevelTransfer>:: - reinit_polynomial_transfer(const DoFHandler &dof_handler_fine, - const DoFHandler &dof_handler_coarse, - const AffineConstraints &constraint_fine, - const AffineConstraints &constraint_coarse, - const unsigned int mg_level_fine, - const unsigned int mg_level_coarse) + reinit_polynomial_transfer( + const DoFHandler & dof_handler_fine, + const DoFHandler & dof_handler_coarse, + const AffineConstraints &constraints_fine, + const AffineConstraints &constraints_coarse, + const unsigned int mg_level_fine, + const unsigned int mg_level_coarse) { internal::MGTwoLevelTransferImplementation::reinit_polynomial_transfer( dof_handler_fine, dof_handler_coarse, - constraint_fine, - constraint_coarse, + constraints_fine, + constraints_coarse, mg_level_fine, mg_level_coarse, *this); @@ -3217,8 +3263,8 @@ void MGTwoLevelTransfer>::reinit( const DoFHandler & dof_handler_fine, const DoFHandler & dof_handler_coarse, - const AffineConstraints &constraint_fine, - const AffineConstraints &constraint_coarse, + const AffineConstraints &constraints_fine, + const AffineConstraints &constraints_coarse, const unsigned int mg_level_fine, const unsigned int mg_level_coarse) { @@ -3286,8 +3332,8 @@ MGTwoLevelTransfer>::reinit( internal::MGTwoLevelTransferImplementation::reinit_polynomial_transfer( dof_handler_fine, dof_handler_coarse, - constraint_fine, - constraint_coarse, + constraints_fine, + constraints_coarse, mg_level_fine, mg_level_coarse, *this); @@ -3295,8 +3341,8 @@ MGTwoLevelTransfer>::reinit( internal::MGTwoLevelTransferImplementation::reinit_geometric_transfer( dof_handler_fine, dof_handler_coarse, - constraint_fine, - constraint_coarse, + constraints_fine, + constraints_coarse, mg_level_fine, mg_level_coarse, *this);