Fix intermittent CI test failure

.gitignore

@@ -1,3 +1,5 @@
+.vscode
 TODO.md
 doc/benchmarks/**
+docker/dealii-master-focal-root/**
 docker/dealii-weak_forms-dev/**

cmake/checks/check_02_compiler_features.cmake

@@ -14,10 +14,25 @@
 ## ---------------------------------------------------------------------
 
 
+# Save the current flags
+SET(CMAKE_REQUIRED_LIBRARIES_SAVED ${CMAKE_REQUIRED_LIBRARIES})
+SET(CMAKE_REQUIRED_INCLUDES_SAVED  ${CMAKE_REQUIRED_INCLUDES})
+SET(CMAKE_REQUIRED_FLAGS_SAVED     ${CMAKE_REQUIRED_FLAGS})
+
+# Add deal.II's flags to the current project's
+# (which should be empty unless the user specifies otherwise)
 LIST(APPEND CMAKE_REQUIRED_LIBRARIES ${DEAL_II_LIBRARIES})
 LIST(APPEND CMAKE_REQUIRED_INCLUDES  ${DEAL_II_INCLUDE_DIRS})
 LIST(APPEND CMAKE_REQUIRED_FLAGS     ${DEAL_II_CXX_FLAGS})
 
+IF("${CMAKE_BUILD_TYPE}" STREQUAL "Release")
+  LIST(APPEND CMAKE_REQUIRED_FLAGS   ${DEAL_II_CXX_FLAGS_RELEASE})
+ELSEIF("${CMAKE_BUILD_TYPE}" STREQUAL "Debug")
+  LIST(APPEND CMAKE_REQUIRED_FLAGS   ${DEAL_II_CXX_FLAGS_DEBUG})
+ELSE()
+  MESSAGE(FATAL_ERROR "CMAKE_BUILD_TYPE doesn't match either Release or Debug!")
+ENDIF()
+
 #
 # Check whether division by zero in a vectorized array results in
 # a floating point exception.
@@ -29,6 +44,16 @@ WF_CHECK_CXX_SOURCE_RUNS(
   "
   #include <deal.II/base/vectorization.h>
   using namespace dealii;
+  class BinaryOp
+  {
+  public:
+    template <typename T>
+    T
+    operator()(const T &num, const T &den) const
+    {
+      return num/den;
+    }
+  };
   template <typename Number, std::size_t width>
   void
   do_test()
@@ -38,7 +63,8 @@ WF_CHECK_CXX_SOURCE_RUNS(
     {
       const Vec_t num = 1.0;
       const Vec_t den = 0.0;
-      const auto result = num / den;
+      auto result = num / den;
+      result = BinaryOp()(num, den);
       (void)result;
     }
   }
@@ -73,13 +99,25 @@ WF_CHECK_CXX_SOURCE_RUNS(
 # results in a floating point exception.
 #
 # The test is multiple times to ensure that the failure is not in
-# any way sporadic.
+# any way sporadic. The tested value 6.916...e-310 comes from some
+# backtraced output of a test failure on a Docker image. 
 #
 WF_CHECK_CXX_SOURCE_RUNS(
   "
   #include <deal.II/base/vectorization.h>
   #include <limits>
   using namespace dealii;
+  class UnaryOp
+  {
+  public:
+    template <typename T>
+    T
+    operator()(const T &value) const
+    {
+      using namespace std;
+      return sqrt(value);
+    }
+  };
   template <typename Number, std::size_t width>
   void
   do_test()
@@ -90,8 +128,13 @@ WF_CHECK_CXX_SOURCE_RUNS(
     {
       Vec_t val = 0.0;
       auto result = sqrt(val);
+      result = UnaryOp()(val);
       val = std::numeric_limits<Number>::epsilon();
       result = sqrt(val);
+      result = UnaryOp()(val);
+      val = 6.9161116785120245e-310;
+      result = sqrt(val);
+      result = UnaryOp()(val);
       (void)result;
     }
   }
@@ -111,11 +154,22 @@ WF_CHECK_CXX_SOURCE_RUNS(
     do_test<double, 2>();
     do_test<float, 4>();
   #endif
+
+    // All indications are that _mm_sqrt_pd() sporadically segfaults with zero input vector.
+  #if DEAL_II_VECTORIZATION_WIDTH_IN_BITS == 128
+    static_assert(false, 'Problematic vectorization width detected.');
+  #endif
 
     do_test<double, 1>();
     do_test<float, 1>();
 
     return 0;
   }
   "
-  WEAK_FORMS_VECTORIZATION_FPE_SQRT_OF_ZERO)
+  WEAK_FORMS_VECTORIZATION_FPE_SQRT_OF_ZERO)
+
+
+# Restore all flags
+SET(CMAKE_REQUIRED_LIBRARIES ${CMAKE_REQUIRED_LIBRARIES_SAVED})
+SET(CMAKE_REQUIRED_INCLUDES  ${CMAKE_REQUIRED_INCLUDES_SAVED})
+SET(CMAKE_REQUIRED_FLAGS     ${CMAKE_REQUIRED_FLAGS_SAVED})

include/weak_forms/integrator.h

@@ -319,7 +319,7 @@ namespace WeakForms
                             MeshWorker::assemble_own_cells);
 
       if (mpi_communicator)
-        dealii::Utilities::MPI::sum(integral, *mpi_communicator);
+        integral = dealii::Utilities::MPI::sum(integral, *mpi_communicator);
 
       return integral;
     }
@@ -443,7 +443,7 @@ namespace WeakForms
                             boundary_worker);
 
       if (mpi_communicator)
-        dealii::Utilities::MPI::sum(integral, *mpi_communicator);
+        integral = dealii::Utilities::MPI::sum(integral, *mpi_communicator);
 
       return integral;
     }
@@ -869,7 +869,7 @@ namespace WeakForms
                             MeshWorker::assemble_own_cells);
 
       if (mpi_communicator)
-        dealii::Utilities::MPI::sum(integral, *mpi_communicator);
+        integral = dealii::Utilities::MPI::sum(integral, *mpi_communicator);
 
       return integral;
     }
@@ -1015,7 +1015,7 @@ namespace WeakForms
                             boundary_worker);
 
       if (mpi_communicator)
-        dealii::Utilities::MPI::sum(integral, *mpi_communicator);
+        integral = dealii::Utilities::MPI::sum(integral, *mpi_communicator);
 
       return integral;
     }

include/weak_forms/unary_operators.h

@@ -1608,25 +1608,19 @@ private:                                                                 \
         using namespace std;
 
 #ifdef WEAK_FORMS_VECTORIZATION_FPE_SQRT_OF_ZERO
-        // We need to check if the vector is a zero vector. If so, simply
-        // avoid the operation that would result in an FPE as we know the
-        // result already.
-        bool has_non_zero_value = false;
+        // Avoid calls to  _mm_sqrt_pd():
+        // All indications are that it segfaults, typically with a zero
+        // input vector.
+        vectorized_value_type<ScalarType, width> result;
+
         DEAL_II_OPENMP_SIMD_PRAGMA
         for (unsigned int v = 0; v < width; v++)
-          {
-            if (std::abs(value[v]) >
-                dealii::internal::NumberType<ScalarType>::value(
-                  10.0 * std::numeric_limits<ScalarType>::epsilon()))
-              {
-                has_non_zero_value = true;
-              }
-          }
-        if (!has_non_zero_value)
-          return vectorized_value_type<ScalarType, width>{};
-#endif
+          result[v] = sqrt(value[v]);
 
+        return result;
+#else
         return sqrt(value);
+#endif
       }
     };
 

tests/weak_forms/prm/parameters-step-curl_curl.prm

@@ -114,5 +114,7 @@ subsection Nonlinear solver
   # Nonlinear solver residual
   set Residual                   = 1e-9
 
+  set Min iterations             = 3
+
   set Max iterations             = 10
 end

tests/weak_forms/step-transient_curl_curl-variant_02a.cc

@@ -163,6 +163,45 @@ namespace StepTransientCurlCurl
                                 this->qf_cell_esp,
                                 update_gradients);
 
+    FEFaceValues<dim> fe_face_values_esp(this->mapping,
+                                this->fe_esp,
+                                this->qf_face_esp,
+                                update_gradients | update_normal_vectors | update_JxW_values);
+
+    // Compute total current flowing through inlet boundary
+    double I_total = 0;
+    {
+      typename DoFHandler<dim>::active_cell_iterator
+        cell = this->dof_handler_esp.begin_active(),
+        endc = this->dof_handler_esp.end();
+      for (; cell != endc; ++cell)
+      {
+        if (cell->is_locally_owned() == false)
+          continue;
+        for (const auto &face : cell->face_iterators())
+        {
+          if (face->at_boundary() && face->boundary_id() == this->parameters.bid_wire_inlet)
+          {
+            fe_face_values_esp.reinit(cell, face);
+            const std::vector<Tensor<1, dim>> & normals = fe_face_values_esp.get_normal_vectors();
+            std::vector<Tensor<1, dim>> source_values(fe_face_values_esp.n_quadrature_points);
+            fe_face_values_esp[this->esp_extractor].get_function_gradients(
+              this->solution_esp, source_values);
+
+            for (const unsigned int q_point : fe_face_values_esp.quadrature_point_indices())
+            {
+              source_values[q_point] *= -1.0;
+              const Tensor<1, dim> &J_f = source_values[q_point];
+              const Tensor<1, dim> &N = normals[q_point];
+              I_total += (J_f*N) * fe_face_values_esp.JxW(q_point);
+            }
+          }
+        }
+      }
+    }
+  I_total = dealii::Utilities::MPI::sum(I_total, this->mpi_communicator);
+   deallog << "I_total: " << I_total << std::endl;
+
     typename DoFHandler<dim>::active_cell_iterator
       cell = this->dof_handler_mvp.begin_active(),
       endc = this->dof_handler_mvp.end();

tests/weak_forms/step-transient_curl_curl-variant_02a.with_trilinos=on.mpirun=1.output

@@ -1,38 +1,86 @@
 
+DEAL::I_total: -1.98873782
+DEAL::I_total: -1.98873782
+DEAL::I_total: -1.98873782
+DEAL::I_total: -1.98873782
 DEAL::Timestep: 0 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 4.61581794e-06 -9.56896981e-06 -2.13298753e-21 ; J_eddy (axial): 236.178640 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.63722304e-06 -6.79177661e-06 -9.29446723e-22 ; J_eddy (axial): 3.14211193e-06 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 4.61581794e-06 -9.56896981e-06 1.16231973e-22 ; J_eddy (axial): 236.178640 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.63722304e-06 -6.79177661e-06 -1.14263864e-22 ; J_eddy (axial): 3.14211189e-06 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 0 ; Cycle: 1
-DEAL:: ; Within wire: 1 ; B: 2.53135698e-06 -1.11267136e-05 -1.55306726e-20 ; J_eddy (axial): 237.827569 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.03564110e-06 -9.55446258e-06 6.61313418e-21 ; J_eddy (axial): 3.84595299e-06 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 2.53135698e-06 -1.11267136e-05 -3.15452678e-22 ; J_eddy (axial): 237.827569 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.03564110e-06 -9.55446258e-06 -2.40714949e-22 ; J_eddy (axial): 3.84595100e-06 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 1 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.13416573e-06 -1.37930118e-05 -5.44117145e-21 ; J_eddy (axial): 55.4961666 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.50748484e-06 -1.17820055e-05 1.42871597e-21 ; J_eddy (axial): 8.96208088e-07 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.13416573e-06 -1.37930118e-05 -4.65440152e-22 ; J_eddy (axial): 55.4961666 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.50748484e-06 -1.17820055e-05 -2.23651199e-22 ; J_eddy (axial): 8.96209655e-07 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 2 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.27600221e-06 -1.44207039e-05 -9.62728373e-22 ; J_eddy (axial): 13.0436231 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.61833934e-06 -1.23055350e-05 2.05985495e-22 ; J_eddy (axial): 2.10624092e-07 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.27600221e-06 -1.44207039e-05 -1.44215386e-22 ; J_eddy (axial): 13.0436231 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.61833934e-06 -1.23055350e-05 -1.12869605e-22 ; J_eddy (axial): 2.10625098e-07 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 3 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.30936216e-06 -1.45683440e-05 -4.55362927e-22 ; J_eddy (axial): 3.06743096 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.64440923e-06 -1.24286584e-05 1.88523346e-23 ; J_eddy (axial): 4.95318284e-08 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.30936216e-06 -1.45683440e-05 1.19427432e-22 ; J_eddy (axial): 3.06743096 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.64440923e-06 -1.24286584e-05 3.90628414e-22 ; J_eddy (axial): 4.95312749e-08 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 4 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.31720824e-06 -1.46030683e-05 -3.96641182e-22 ; J_eddy (axial): 0.721261992 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.65054066e-06 -1.24576161e-05 3.26313768e-23 ; J_eddy (axial): 1.16470979e-08 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.31720824e-06 -1.46030683e-05 2.28718671e-23 ; J_eddy (axial): 0.721261992 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.65054066e-06 -1.24576161e-05 1.42527483e-22 ; J_eddy (axial): 1.16471478e-08 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 5 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.31905359e-06 -1.46112353e-05 -6.48357735e-22 ; J_eddy (axial): 0.169460725 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.65198274e-06 -1.24644268e-05 -8.13720193e-23 ; J_eddy (axial): 2.73635533e-09 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.31905359e-06 -1.46112353e-05 7.15698638e-22 ; J_eddy (axial): 0.169460725 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.65198274e-06 -1.24644268e-05 1.59931874e-22 ; J_eddy (axial): 2.73634142e-09 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 6 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.31948760e-06 -1.46131561e-05 1.38743392e-22 ; J_eddy (axial): 0.0396804632 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.65232191e-06 -1.24660286e-05 -1.66358956e-22 ; J_eddy (axial): 6.40906458e-10 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.31948760e-06 -1.46131561e-05 3.31470012e-22 ; J_eddy (axial): 0.0396804632 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.65232191e-06 -1.24660286e-05 -4.49528122e-23 ; J_eddy (axial): 6.40790458e-10 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 7 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.31958968e-06 -1.46136078e-05 3.29709022e-22 ; J_eddy (axial): 0.00915692866 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.65240168e-06 -1.24664053e-05 6.22040570e-23 ; J_eddy (axial): 1.47915425e-10 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.31958968e-06 -1.46136078e-05 4.74045577e-22 ; J_eddy (axial): 0.00915692867 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.65240168e-06 -1.24664053e-05 -3.66948419e-23 ; J_eddy (axial): 1.47629544e-10 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 8 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.31961369e-06 -1.46137141e-05 -4.10359132e-24 ; J_eddy (axial): 0.00197797717 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.65242044e-06 -1.24664940e-05 2.74721941e-22 ; J_eddy (axial): 3.18269209e-11 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.31961369e-06 -1.46137141e-05 1.07200669e-22 ; J_eddy (axial): 0.00197797717 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.65242044e-06 -1.24664940e-05 -1.73290472e-22 ; J_eddy (axial): 3.24535248e-11 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 9 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.31961934e-06 -1.46137391e-05 4.16130009e-22 ; J_eddy (axial): 0.000289531007 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.65242485e-06 -1.24665148e-05 -2.21322621e-22 ; J_eddy (axial): 4.81064685e-12 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.31961934e-06 -1.46137391e-05 6.16688996e-22 ; J_eddy (axial): 0.000289531007 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.65242485e-06 -1.24665148e-05 -1.20243444e-22 ; J_eddy (axial): 4.56045290e-12 ; J_free (axial): 0.00000000
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
+DEAL::I_total: -1.98873821
 DEAL::Timestep: 10 ; Cycle: 0
-DEAL:: ; Within wire: 1 ; B: 3.31962066e-06 -1.46137450e-05 1.08717059e-21 ; J_eddy (axial): -0.000107581363 ; J_free (axial): 5092.95818
-DEAL:: ; Within wire: 0 ; B: 2.65242589e-06 -1.24665197e-05 3.40969838e-22 ; J_eddy (axial): -1.68748702e-12 ; J_free (axial): 0.00000000
+DEAL:: ; Within wire: 1 ; B: 3.31962066e-06 -1.46137450e-05 4.43843791e-22 ; J_eddy (axial): -0.000107581362 ; J_free (axial): 5092.95818
+DEAL:: ; Within wire: 0 ; B: 2.65242589e-06 -1.24665197e-05 -1.84550430e-24 ; J_eddy (axial): -1.50795949e-12 ; J_free (axial): 0.00000000
 DEAL::OK