Use EvaluationFlags inside CUDAWrapper::FEEvaluation

diff --git a/doc/news/changes/incompatibilities/20231027Munch b/doc/news/changes/incompatibilities/20231027Munch
new file mode 100644 (file)
index 0000000..0b996f7
--- /dev/null
+++ b/doc/news/changes/incompatibilities/20231027Munch
@@ -0,0 +1,6 @@
+Removed: The functions evaluate() and integrate() of
+CUDAWrappers::FEEvaluation that take Bools have been
+deprecated and replaced by versions that take
+EvaluationFlags.
+<br>
+(Peter Munch, 2023/10/27)

diff --git a/examples/step-64/step-64.cc b/examples/step-64/step-64.cc
index 5b14e2b24c9de1c9ca64cbe05554db69fa5201ac..4678f23cea46bc68a9bcd3ac08418b46fe6b6878 100644 (file)
--- a/examples/step-64/step-64.cc
+++ b/examples/step-64/step-64.cc
@@ -221,10 +221,10 @@ namespace Step64
     CUDAWrappers::FEEvaluation<dim, fe_degree, fe_degree + 1, 1, double>
       fe_eval(gpu_data, shared_data);
     fe_eval.read_dof_values(src);
-    fe_eval.evaluate(true, true);
+    fe_eval.evaluate(EvaluationFlags::values | EvaluationFlags::gradients);
     fe_eval.apply_for_each_quad_point(
       HelmholtzOperatorQuad<dim, fe_degree>(gpu_data, coef, cell));
-    fe_eval.integrate(true, true);
+    fe_eval.integrate(EvaluationFlags::values | EvaluationFlags::gradients);
     fe_eval.distribute_local_to_global(dst);
   }
 

diff --git a/include/deal.II/matrix_free/cuda_fe_evaluation.h b/include/deal.II/matrix_free/cuda_fe_evaluation.h
index 060fba7c4dcc33a5c1fdd9fe6fb2711079c34494..f8779fce0b8c82b10526e1860bf9c52561e27ea9 100644 (file)
--- a/include/deal.II/matrix_free/cuda_fe_evaluation.h
+++ b/include/deal.II/matrix_free/cuda_fe_evaluation.h
@@ -29,6 +29,7 @@
 #include <deal.II/matrix_free/cuda_matrix_free.h>
 #include <deal.II/matrix_free/cuda_matrix_free.templates.h>
 #include <deal.II/matrix_free/cuda_tensor_product_kernels.h>
+#include <deal.II/matrix_free/evaluation_flags.h>
 
 #include <Kokkos_Core.hpp>
 
@@ -143,6 +144,19 @@ namespace CUDAWrappers
      * @p get_value() or @p get_gradient() give useful information.
      */
     DEAL_II_HOST_DEVICE void
+    evaluate(const EvaluationFlags::EvaluationFlags evaluate_flag);
+
+    /**
+     * Evaluate the function values and the gradients of the FE function given
+     * at the DoF values in the input vector at the quadrature points on the
+     * unit cell. The function arguments specify which parts shall actually be
+     * computed. This function needs to be called before the functions
+     * @p get_value() or @p get_gradient() give useful information.
+     */
+    DEAL_II_DEPRECATED_EARLY_WITH_COMMENT(
+      "Use the version taking EvaluationFlags.")
+    DEAL_II_HOST_DEVICE
+    void
     evaluate(const bool evaluate_val, const bool evaluate_grad);
 
     /**
@@ -153,6 +167,19 @@ namespace CUDAWrappers
      * of the values or the gradients.
      */
     DEAL_II_HOST_DEVICE void
+    integrate(const EvaluationFlags::EvaluationFlags integration_flag);
+
+    /**
+     * This function takes the values and/or gradients that are stored on
+     * quadrature points, tests them by all the basis functions/gradients on
+     * the cell and performs the cell integration. The two function arguments
+     * @p integrate_val and @p integrate_grad are used to enable/disable some
+     * of the values or the gradients.
+     */
+    DEAL_II_DEPRECATED_EARLY_WITH_COMMENT(
+      "Use the version taking EvaluationFlags.")
+    DEAL_II_HOST_DEVICE
+    void
     integrate(const bool integrate_val, const bool integrate_grad);
 
     /**
@@ -312,8 +339,7 @@ namespace CUDAWrappers
             typename Number>
   DEAL_II_HOST_DEVICE void
   FEEvaluation<dim, fe_degree, n_q_points_1d, n_components_, Number>::evaluate(
-    const bool evaluate_val,
-    const bool evaluate_grad)
+    const EvaluationFlags::EvaluationFlags evaluate_flag)
   {
     // First evaluate the gradients because it requires values that will be
     // changed if evaluate_val is true
@@ -327,19 +353,21 @@ namespace CUDAWrappers
                                data->shape_values,
                                data->shape_gradients,
                                data->co_shape_gradients);
-    if (evaluate_val == true && evaluate_grad == true)
+
+    if ((evaluate_flag & EvaluationFlags::values) &&
+        (evaluate_flag & EvaluationFlags::gradients))
       {
         evaluator_tensor_product.value_and_gradient_at_quad_pts(
           shared_data->values, shared_data->gradients);
         shared_data->team_member.team_barrier();
       }
-    else if (evaluate_grad == true)
+    else if (evaluate_flag & EvaluationFlags::gradients)
       {
         evaluator_tensor_product.gradient_at_quad_pts(shared_data->values,
                                                       shared_data->gradients);
         shared_data->team_member.team_barrier();
       }
-    else if (evaluate_val == true)
+    else if (evaluate_flag & EvaluationFlags::values)
       {
         evaluator_tensor_product.value_at_quad_pts(shared_data->values);
         shared_data->team_member.team_barrier();
@@ -348,6 +376,23 @@ namespace CUDAWrappers
 
 
 
+  template <int dim,
+            int fe_degree,
+            int n_q_points_1d,
+            int n_components_,
+            typename Number>
+  DEAL_II_HOST_DEVICE void
+  FEEvaluation<dim, fe_degree, n_q_points_1d, n_components_, Number>::evaluate(
+    const bool evaluate_val,
+    const bool evaluate_grad)
+  {
+    evaluate(
+      (evaluate_val ? EvaluationFlags::values : EvaluationFlags::nothing) |
+      (evaluate_grad ? EvaluationFlags::gradients : EvaluationFlags::nothing));
+  }
+
+
+
   template <int dim,
             int fe_degree,
             int n_q_points_1d,
@@ -355,8 +400,7 @@ namespace CUDAWrappers
             typename Number>
   DEAL_II_HOST_DEVICE void
   FEEvaluation<dim, fe_degree, n_q_points_1d, n_components_, Number>::integrate(
-    const bool integrate_val,
-    const bool integrate_grad)
+    const EvaluationFlags::EvaluationFlags integration_flag)
   {
     internal::EvaluatorTensorProduct<
       internal::EvaluatorVariant::evaluate_general,
@@ -368,17 +412,19 @@ namespace CUDAWrappers
                                data->shape_values,
                                data->shape_gradients,
                                data->co_shape_gradients);
-    if (integrate_val == true && integrate_grad == true)
+
+    if ((integration_flag & EvaluationFlags::values) &&
+        (integration_flag & EvaluationFlags::gradients))
       {
         evaluator_tensor_product.integrate_value_and_gradient(
           shared_data->values, shared_data->gradients);
       }
-    else if (integrate_val == true)
+    else if (integration_flag & EvaluationFlags::gradients)
       {
         evaluator_tensor_product.integrate_value(shared_data->values);
         shared_data->team_member.team_barrier();
       }
-    else if (integrate_grad == true)
+    else if (integration_flag & EvaluationFlags::values)
       {
         evaluator_tensor_product.template integrate_gradient<false>(
           shared_data->values, shared_data->gradients);
@@ -388,6 +434,23 @@ namespace CUDAWrappers
 
 
 
+  template <int dim,
+            int fe_degree,
+            int n_q_points_1d,
+            int n_components_,
+            typename Number>
+  DEAL_II_HOST_DEVICE void
+  FEEvaluation<dim, fe_degree, n_q_points_1d, n_components_, Number>::integrate(
+    const bool integrate_val,
+    const bool integrate_grad)
+  {
+    integrate(
+      (integrate_val ? EvaluationFlags::values : EvaluationFlags::nothing) |
+      (integrate_grad ? EvaluationFlags::gradients : EvaluationFlags::nothing));
+  }
+
+
+
   template <int dim,
             int fe_degree,
             int n_q_points_1d,

diff --git a/include/deal.II/matrix_free/evaluation_flags.h b/include/deal.II/matrix_free/evaluation_flags.h
index 05a0d0656371b0ce1034983cf198f09c75891a77..290c7a34f3c45476ef24513c79e1b6203050de88 100644 (file)
--- a/include/deal.II/matrix_free/evaluation_flags.h
+++ b/include/deal.II/matrix_free/evaluation_flags.h
@@ -69,7 +69,7 @@ namespace EvaluationFlags
    *
    * @ref EvaluationFlags
    */
-  inline EvaluationFlags
+  DEAL_II_HOST_DEVICE inline EvaluationFlags
   operator|(const EvaluationFlags f1, const EvaluationFlags f2)
   {
     return static_cast<EvaluationFlags>(static_cast<unsigned int>(f1) |
@@ -84,7 +84,7 @@ namespace EvaluationFlags
    *
    * @ref EvaluationFlags
    */
-  inline EvaluationFlags &
+  DEAL_II_HOST_DEVICE inline EvaluationFlags &
   operator|=(EvaluationFlags &f1, const EvaluationFlags f2)
   {
     f1 = f1 | f2;
@@ -101,7 +101,7 @@ namespace EvaluationFlags
    *
    * @ref EvaluationFlags
    */
-  inline EvaluationFlags
+  DEAL_II_HOST_DEVICE inline EvaluationFlags
   operator&(const EvaluationFlags f1, const EvaluationFlags f2)
   {
     return static_cast<EvaluationFlags>(static_cast<unsigned int>(f1) &
@@ -115,7 +115,7 @@ namespace EvaluationFlags
    *
    * @ref EvaluationFlags
    */
-  inline EvaluationFlags &
+  DEAL_II_HOST_DEVICE inline EvaluationFlags &
   operator&=(EvaluationFlags &f1, const EvaluationFlags f2)
   {
     f1 = f1 & f2;

diff --git a/tests/matrix_free/matrix_free_device_no_index_initialize.cc b/tests/matrix_free/matrix_free_device_no_index_initialize.cc
index f741a6a4b8b8456f71cefa09e8a0c9ca4f1a9826..e2ce81d585466fed33a31bbd0341482932c93a04 100644 (file)
--- a/tests/matrix_free/matrix_free_device_no_index_initialize.cc
+++ b/tests/matrix_free/matrix_free_device_no_index_initialize.cc
@@ -57,7 +57,7 @@ public:
                                                  n_local_dofs),
                          [&](int i) { fe_eval_ptr->submit_dof_value(1., i); });
     shared_data->team_member.team_barrier();
-    fe_eval.evaluate(/*evaluate_values =*/true, /*evaluate_gradients=*/true);
+    fe_eval.evaluate(EvaluationFlags::values | EvaluationFlags::gradients);
 
 #ifndef __APPLE__
     Kokkos::parallel_for(Kokkos::TeamThreadRange(shared_data->team_member,
@@ -69,8 +69,7 @@ public:
                              assert(fe_eval_ptr->get_gradient(i)[e] == 0.);
                          });
 
-    fe_eval.integrate(/*integrate_values = */ true,
-                      /*integrate_gradients=*/true);
+    fe_eval.integrate(EvaluationFlags::values | EvaluationFlags::gradients);
 
     Kokkos::parallel_for(
       Kokkos::TeamThreadRange(shared_data->team_member, n_local_dofs),

diff --git a/tests/matrix_free/matrix_vector_device_mf.h b/tests/matrix_free/matrix_vector_device_mf.h
index ce84faa0071541fec08c25938edc0a9fae080fc2..a67705df33fcab81a9b704183ac886a3a38c24aa 100644 (file)
--- a/tests/matrix_free/matrix_vector_device_mf.h
+++ b/tests/matrix_free/matrix_vector_device_mf.h
@@ -109,12 +109,12 @@ HelmholtzOperator<dim, fe_degree, Number, n_q_points_1d>::operator()(
   CUDAWrappers::FEEvaluation<dim, fe_degree, n_q_points_1d, 1, Number> fe_eval(
     gpu_data, shared_data);
   fe_eval.read_dof_values(src);
-  fe_eval.evaluate(true, true);
+  fe_eval.evaluate(EvaluationFlags::values | EvaluationFlags::gradients);
   fe_eval.apply_for_each_quad_point(
     HelmholtzOperatorQuad<dim, fe_degree, Number, n_q_points_1d>(gpu_data,
                                                                  coef,
                                                                  cell));
-  fe_eval.integrate(true, true);
+  fe_eval.integrate(EvaluationFlags::values | EvaluationFlags::gradients);
   fe_eval.distribute_local_to_global(dst);
 }