]> https://gitweb.dealii.org/ - dealii.git/commitdiff
Add FE_Nothing test for CUDA matrix-free 14272/head
authorBruno Turcksin <bruno.turcksin@gmail.com>
Thu, 15 Sep 2022 13:18:22 +0000 (13:18 +0000)
committerBruno Turcksin <bruno.turcksin@gmail.com>
Thu, 15 Sep 2022 13:31:38 +0000 (13:31 +0000)
tests/cuda/matrix_free_matrix_vector_26.cu [new file with mode: 0644]
tests/cuda/matrix_free_matrix_vector_26.with_trilinos=true.with_p4est=true.mpirun=1.output [new file with mode: 0644]
tests/cuda/matrix_free_matrix_vector_26.with_trilinos=true.with_p4est=true.mpirun=2.output [new file with mode: 0644]

diff --git a/tests/cuda/matrix_free_matrix_vector_26.cu b/tests/cuda/matrix_free_matrix_vector_26.cu
new file mode 100644 (file)
index 0000000..09dbd64
--- /dev/null
@@ -0,0 +1,258 @@
+// ---------------------------------------------------------------------
+//
+// Copyright (C) 2022 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.
+//
+// ---------------------------------------------------------------------
+
+// Same as matrix_free_matrix_vector_25 test but we only perform the operator
+// evaluation on the cells that have a fe_index equal to 0
+
+#include <deal.II/base/function.h>
+#include <deal.II/base/utilities.h>
+
+#include <deal.II/distributed/tria.h>
+
+#include <deal.II/dofs/dof_handler.h>
+#include <deal.II/dofs/dof_tools.h>
+
+#include <deal.II/fe/fe_nothing.h>
+#include <deal.II/fe/fe_q.h>
+#include <deal.II/fe/fe_values.h>
+
+#include "deal.II/grid/filtered_iterator.h"
+#include <deal.II/grid/grid_generator.h>
+#include <deal.II/grid/manifold_lib.h>
+
+#include <deal.II/lac/affine_constraints.h>
+#include <deal.II/lac/trilinos_sparse_matrix.h>
+#include <deal.II/lac/trilinos_sparsity_pattern.h>
+
+#include <deal.II/numerics/vector_tools.h>
+
+#include <iostream>
+
+#include "../tests.h"
+
+#include "matrix_vector_mf.h"
+
+
+
+template <int dim, int fe_degree>
+void
+test()
+{
+  using Number = double;
+
+  parallel::distributed::Triangulation<dim> tria(MPI_COMM_WORLD);
+  GridGenerator::hyper_cube(tria);
+  tria.refine_global(1);
+  typename Triangulation<dim>::active_cell_iterator cell = tria.begin_active(),
+                                                    endc = tria.end();
+  for (; cell != endc; ++cell)
+    if (cell->is_locally_owned())
+      if (cell->center().norm() < 0.2)
+        cell->set_refine_flag();
+  tria.execute_coarsening_and_refinement();
+  if (dim < 3 && fe_degree < 2)
+    tria.refine_global(2);
+  else
+    tria.refine_global(1);
+  if (tria.begin(tria.n_levels() - 1)->is_locally_owned())
+    tria.begin(tria.n_levels() - 1)->set_refine_flag();
+  if (tria.last()->is_locally_owned())
+    tria.last()->set_refine_flag();
+  tria.execute_coarsening_and_refinement();
+  cell = tria.begin_active();
+  for (unsigned int i = 0; i < 10 - 3 * dim; ++i)
+    {
+      cell                 = tria.begin_active();
+      unsigned int counter = 0;
+      for (; cell != endc; ++cell, ++counter)
+        if (cell->is_locally_owned())
+          if (counter % (7 - i) == 0)
+            cell->set_refine_flag();
+      tria.execute_coarsening_and_refinement();
+    }
+
+  hp::FECollection<dim> fe_collection;
+  fe_collection.push_back(FE_Q<dim>(fe_degree));
+  fe_collection.push_back(FE_Nothing<dim>());
+
+  DoFHandler<dim> dof(tria);
+  // Set FE_Index
+  for (auto &cell_ : dof.active_cell_iterators())
+    if (cell_->is_locally_owned())
+      {
+        if (cell_->center()[0] < 0.5)
+          cell_->set_active_fe_index(0);
+        else
+          cell_->set_active_fe_index(1);
+      }
+  dof.distribute_dofs(fe_collection);
+
+  IndexSet owned_set = dof.locally_owned_dofs();
+  IndexSet relevant_set;
+  DoFTools::extract_locally_relevant_dofs(dof, relevant_set);
+
+  AffineConstraints<double> constraints(relevant_set);
+  DoFTools::make_hanging_node_constraints(dof, constraints);
+  VectorTools::interpolate_boundary_values(dof,
+                                           0,
+                                           Functions::ZeroFunction<dim>(),
+                                           constraints);
+  constraints.close();
+
+  deallog << "Testing " << dof.get_fe().get_name() << std::endl;
+
+  MappingQ<dim>                         mapping(fe_degree);
+  CUDAWrappers::MatrixFree<dim, Number> mf_data;
+  const QGauss<1>                       quad(fe_degree + 1);
+  typename CUDAWrappers::MatrixFree<dim, Number>::AdditionalData
+    additional_data;
+  additional_data.mapping_update_flags = update_values | update_gradients |
+                                         update_JxW_values |
+                                         update_quadrature_points;
+  IteratorFilters::ActiveFEIndexEqualTo filter(0, true);
+  mf_data.reinit(mapping, dof, constraints, quad, filter, additional_data);
+
+  const unsigned int coef_size =
+    tria.n_locally_owned_active_cells() * std::pow(fe_degree + 1, dim);
+  MatrixFreeTest<dim,
+                 fe_degree,
+                 Number,
+                 LinearAlgebra::distributed::Vector<Number, MemorySpace::CUDA>>
+    mf(mf_data, coef_size);
+  LinearAlgebra::distributed::Vector<Number, MemorySpace::CUDA> in_dev;
+  LinearAlgebra::distributed::Vector<Number, MemorySpace::CUDA> out_dev;
+  mf_data.initialize_dof_vector(in_dev);
+  mf_data.initialize_dof_vector(out_dev);
+
+  LinearAlgebra::ReadWriteVector<Number> rw_in(owned_set);
+  for (unsigned int i = 0; i < in_dev.local_size(); ++i)
+    {
+      const unsigned int glob_index = owned_set.nth_index_in_set(i);
+      if (constraints.is_constrained(glob_index))
+        continue;
+      rw_in.local_element(i) = random_value<double>();
+    }
+  in_dev.import(rw_in, VectorOperation::insert);
+  mf.vmult(out_dev, in_dev);
+
+  LinearAlgebra::distributed::Vector<Number, MemorySpace::Host> out_host(
+    owned_set, MPI_COMM_WORLD);
+  LinearAlgebra::ReadWriteVector<Number> rw_out(owned_set);
+  rw_out.import(out_dev, VectorOperation::insert);
+  out_host.import(rw_out, VectorOperation::insert);
+
+  // assemble trilinos sparse matrix with
+  // (\nabla v, \nabla u) + (v, 10 * u) for
+  // reference
+  LinearAlgebra::distributed::Vector<Number, MemorySpace::Host> in_host(
+    owned_set, MPI_COMM_WORLD);
+  in_host.import(rw_in, VectorOperation::insert);
+  LinearAlgebra::distributed::Vector<Number, MemorySpace::Host> ref(
+    owned_set, MPI_COMM_WORLD);
+  TrilinosWrappers::SparseMatrix sparse_matrix;
+  {
+    TrilinosWrappers::SparsityPattern csp(owned_set, MPI_COMM_WORLD);
+    DoFTools::make_sparsity_pattern(dof,
+                                    csp,
+                                    constraints,
+                                    true,
+                                    Utilities::MPI::this_mpi_process(
+                                      MPI_COMM_WORLD));
+    csp.compress();
+    sparse_matrix.reinit(csp);
+  }
+  {
+    QGauss<dim> quadrature_formula(fe_degree + 1);
+
+    FEValues<dim> fe_values(dof.get_fe(),
+                            quadrature_formula,
+                            update_values | update_gradients |
+                              update_JxW_values);
+
+    const unsigned int dofs_per_cell = dof.get_fe().dofs_per_cell;
+    const unsigned int n_q_points    = quadrature_formula.size();
+
+    FullMatrix<double> cell_matrix(dofs_per_cell, dofs_per_cell);
+    std::vector<types::global_dof_index> local_dof_indices(dofs_per_cell);
+
+    typename DoFHandler<dim>::active_cell_iterator cell = dof.begin_active(),
+                                                   endc = dof.end();
+    for (; cell != endc; ++cell)
+      if (cell->is_locally_owned() && cell->active_fe_index() == 0)
+        {
+          cell_matrix = 0;
+          fe_values.reinit(cell);
+
+          for (unsigned int q_point = 0; q_point < n_q_points; ++q_point)
+            for (unsigned int i = 0; i < dofs_per_cell; ++i)
+              {
+                for (unsigned int j = 0; j < dofs_per_cell; ++j)
+                  cell_matrix(i, j) +=
+                    ((fe_values.shape_grad(i, q_point) *
+                        fe_values.shape_grad(j, q_point) +
+                      10. * fe_values.shape_value(i, q_point) *
+                        fe_values.shape_value(j, q_point)) *
+                     fe_values.JxW(q_point));
+              }
+
+          cell->get_dof_indices(local_dof_indices);
+          constraints.distribute_local_to_global(cell_matrix,
+                                                 local_dof_indices,
+                                                 sparse_matrix);
+        }
+  }
+  sparse_matrix.compress(VectorOperation::add);
+
+  sparse_matrix.vmult(ref, in_host);
+  out_host -= ref;
+
+  const double diff_norm = out_host.linfty_norm();
+
+  deallog << "Norm of difference: " << diff_norm << std::endl << std::endl;
+}
+
+
+int
+main(int argc, char **argv)
+{
+  Utilities::MPI::MPI_InitFinalize mpi_initialization(
+    argc, argv, testing_max_num_threads());
+
+  unsigned int myid = Utilities::MPI::this_mpi_process(MPI_COMM_WORLD);
+  deallog.push(Utilities::int_to_string(myid));
+
+  init_cuda(true);
+
+  if (myid == 0)
+    {
+      initlog();
+      deallog << std::setprecision(4);
+
+      deallog.push("2d");
+      test<2, 1>();
+      deallog.pop();
+
+      deallog.push("3d");
+      test<3, 1>();
+      test<3, 2>();
+      deallog.pop();
+    }
+  else
+    {
+      test<2, 1>();
+      test<3, 1>();
+      test<3, 2>();
+    }
+}
diff --git a/tests/cuda/matrix_free_matrix_vector_26.with_trilinos=true.with_p4est=true.mpirun=1.output b/tests/cuda/matrix_free_matrix_vector_26.with_trilinos=true.with_p4est=true.mpirun=1.output
new file mode 100644 (file)
index 0000000..3b90478
--- /dev/null
@@ -0,0 +1,10 @@
+
+DEAL:0:2d::Testing FE_Q<2>(1)
+DEAL:0:2d::Norm of difference: 0
+DEAL:0:2d::
+DEAL:0:3d::Testing FE_Q<3>(1)
+DEAL:0:3d::Norm of difference: 0
+DEAL:0:3d::
+DEAL:0:3d::Testing FE_Q<3>(2)
+DEAL:0:3d::Norm of difference: 0
+DEAL:0:3d::
diff --git a/tests/cuda/matrix_free_matrix_vector_26.with_trilinos=true.with_p4est=true.mpirun=2.output b/tests/cuda/matrix_free_matrix_vector_26.with_trilinos=true.with_p4est=true.mpirun=2.output
new file mode 100644 (file)
index 0000000..3b90478
--- /dev/null
@@ -0,0 +1,10 @@
+
+DEAL:0:2d::Testing FE_Q<2>(1)
+DEAL:0:2d::Norm of difference: 0
+DEAL:0:2d::
+DEAL:0:3d::Testing FE_Q<3>(1)
+DEAL:0:3d::Norm of difference: 0
+DEAL:0:3d::
+DEAL:0:3d::Testing FE_Q<3>(2)
+DEAL:0:3d::Norm of difference: 0
+DEAL:0:3d::

In the beginning the Universe was created. This has made a lot of people very angry and has been widely regarded as a bad move.

Douglas Adams


Typeset in Trocchi and Trocchi Bold Sans Serif.