Merge 'trilinos/Trilinos:develop' (1aaaa05) into 'tcad-charon/Trilino…

…s:develop' (557c5ec).

* trilinos-develop:
  ATDM/cee-rhel6: gnu-7.2.0_openmpi-4.0.2 -> gnu-7.2.0_openmpi-4.0.3 (ATDV-353)
  Sacado:  The dreaded "const-lambda-capture" gcc compiler bug strikes again!
  tpetra:  fixed two picky compiler errors (signed vs unsigned)
  Sacado:  Add specializations for ScalarValue, Value for const types.
  Sacado:  rework mat-vec and advection performance tests.
  Sacado:  Fix performance bug with hierarchical-dfad enabled.
  Fixed typo
  tpetra:  Adding a test that doesn't use row-based matrix distribution

cmake/std/atdm/cee-rhel6/all_supported_builds.sh

@@ -5,7 +5,7 @@ export ATDM_CONFIG_CTEST_S_BUILD_NAME_PREFIX=Trilinos-atdm-
 export ATDM_CONFIG_ALL_SUPPORTED_BUILDS=(
   #cee-rhel6_clang-9.0.1_openmpi-4.0.3_serial_static_dbg    # SPARC has installs with this build
   cee-rhel6_clang-9.0.1_openmpi-4.0.3_serial_static_opt    # SPARC CI build
-  cee-rhel6_gnu-7.2.0_openmpi-4.0.2_serial_shared_opt      # SPARC CI build
+  cee-rhel6_gnu-7.2.0_openmpi-4.0.3_serial_shared_opt      # SPARC CI build
   cee-rhel6_intel-18.0.2_mpich2-3.2_openmp_static_opt      # SPARC CI build
   cee-rhel6_intel-19.0.3_intelmpi-2018.4_serial_static_opt # SPARC Nightly bulid
   )

cmake/std/atdm/cee-rhel6/custom_builds.sh

@@ -20,13 +20,13 @@ if atdm_match_any_buildname_keyword \
   # correct matching of of defaults
 
 elif atdm_match_any_buildname_keyword \
-    gnu-7.2.0-openmpi-4.0.2 \
-    gnu-7.2.0_openmpi-4.0.2 \
+    gnu-7.2.0-openmpi-4.0.3 \
+    gnu-7.2.0_openmpi-4.0.3 \
     gnu-7.2.0 \
     gnu-7 \
     gnu \
   ; then
-  export ATDM_CONFIG_COMPILER=GNU-7.2.0_OPENMPI-4.0.2
+  export ATDM_CONFIG_COMPILER=GNU-7.2.0_OPENMPI-4.0.3
   # List default "gnu"* build last of all the 'gnu' builds for correct
   # matching of defaults
 
@@ -57,7 +57,7 @@ else
   echo "*** Supported compilers include:"
   echo "***"
   echo "****  clang-9.0.1-openmpi-4.0.3      (default, default clang)"
-  echo "****  gnu-7.2.0-openmpi-4.0.2        (default gnu)"
+  echo "****  gnu-7.2.0-openmpi-4.0.3        (default gnu)"
   echo "****  intel-18.0.2-mpich2-3.2"
   echo "****  intel-19.0.3-intelmpi-2018.4   (default intel)"
   echo "***"  

cmake/std/atdm/cee-rhel6/custom_builds_unit_tests.sh

@@ -42,23 +42,23 @@ testAll() {
 
   ATDM_CONFIG_BUILD_NAME=before_gnu-7.2.0-openmpi-4.0.3-after
   . ${ATDM_CONFIG_SCRIPT_DIR}/utils/set_build_options.sh
-  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.2
+  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.3
 
   ATDM_CONFIG_BUILD_NAME=before_gnu-7.2.0_openmpi-4.0.3-after
   . ${ATDM_CONFIG_SCRIPT_DIR}/utils/set_build_options.sh
-  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.2
+  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.3
 
   ATDM_CONFIG_BUILD_NAME=before_gnu-7.2.0-after
   . ${ATDM_CONFIG_SCRIPT_DIR}/utils/set_build_options.sh
-  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.2
+  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.3
 
   ATDM_CONFIG_BUILD_NAME=before_gnu-7-after
   . ${ATDM_CONFIG_SCRIPT_DIR}/utils/set_build_options.sh
-  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.2
+  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.3
 
   ATDM_CONFIG_BUILD_NAME=before_gnu-after
   . ${ATDM_CONFIG_SCRIPT_DIR}/utils/set_build_options.sh
-  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.2
+  ${_ASSERT_EQUALS_} ${ATDM_CONFIG_COMPILER} GNU-7.2.0_OPENMPI-4.0.3
 
   # Check that 'gnus' does not match 'gnu'! (Shows true keyword matching is
   # working)

cmake/std/atdm/cee-rhel6/environment.sh

@@ -71,8 +71,8 @@ if  [[ "$ATDM_CONFIG_COMPILER" == "CLANG-9.0.1_OPENMPI-4.0.3" ]]; then
   fi
   export ATDM_CONFIG_MKL_ROOT=${CBLAS_ROOT}
 
-elif [[ "$ATDM_CONFIG_COMPILER" == "GNU-7.2.0_OPENMPI-4.0.2" ]] ; then
-  module load sparc-dev/gcc-7.2.0_openmpi-4.0.2
+elif [[ "$ATDM_CONFIG_COMPILER" == "GNU-7.2.0_OPENMPI-4.0.3" ]] ; then
+  module load sparc-dev/gcc-7.2.0_openmpi-4.0.3
   unset OMP_NUM_THREADS  # SPARC module sets these and we must unset!
   unset OMP_PROC_BIND
   unset OMP_PLACES

packages/sacado/src/Sacado_Fad_ViewStorage.hpp

@@ -56,6 +56,14 @@ namespace Sacado {
 
   namespace Fad {
 
+#ifndef SACADO_FAD_DERIV_LOOP
+#if defined(SACADO_VIEW_CUDA_HIERARCHICAL_DFAD) && !defined(SACADO_DISABLE_CUDA_IN_KOKKOS) && defined(__CUDA_ARCH__)
+#define SACADO_FAD_DERIV_LOOP(I,SZ) for (int I=threadIdx.x; I<SZ; I+=blockDim.x)
+#else
+#define SACADO_FAD_DERIV_LOOP(I,SZ) for (int I=0; I<SZ; ++I)
+#endif
+#endif
+
     /*!
      * \brief Derivative array storage class that is a view into a contiguous
      * memory allocation.  It does not provide proper value semantics and
@@ -110,10 +118,12 @@ namespace Sacado {
         if (this != &x) {
           *val_ = *x.val_;
           if (stride_one)
-            for (int i=0; i<sz_.value; ++i)
+            //for (int i=0; i<sz_.value; ++i)
+            SACADO_FAD_DERIV_LOOP(i,sz_.value)
               dx_[i] = x.dx_[i];
           else
-            for (int i=0; i<sz_.value; ++i)
+            //for (int i=0; i<sz_.value; ++i)
+            SACADO_FAD_DERIV_LOOP(i,sz_.value)
               dx_[i*stride_.value] = x.dx_[i*x.stride_.value];
         }
         return *this;

packages/sacado/src/Sacado_Traits.hpp

@@ -365,6 +365,15 @@ namespace Sacado {
     static const T& eval(const T& x) { return x; }
   };
 
+  //! Specialization of Value for const types
+  template <typename T> struct Value<const T> {
+    typedef typename ValueType<T>::type value_type;
+    KOKKOS_INLINE_FUNCTION
+    static const value_type& eval(const T& x) {
+      return Value<T>::eval(x);
+    }
+  };
+
   //! Base template specification for %ScalarValue
   /*!
    * The %ScalarValue functor returns the base scalar value of an AD type,
@@ -375,6 +384,15 @@ namespace Sacado {
     static const T& eval(const T& x) { return x; }
   };
 
+  //! Specialization of ScalarValue for const types
+  template <typename T> struct ScalarValue<const T> {
+    typedef typename ScalarType<T>::type scalar_type;
+    KOKKOS_INLINE_FUNCTION
+    static const scalar_type& eval(const T& x) {
+      return ScalarValue<T>::eval(x);
+    }
+  };
+
   //! Base template specification for marking constants
   template <typename T> struct MarkConstant {
     KOKKOS_INLINE_FUNCTION

packages/sacado/src/new_design/Sacado_Fad_Exp_ViewStorage.hpp

@@ -42,6 +42,14 @@ namespace Sacado {
   namespace Fad {
   namespace Exp {
 
+#ifndef SACADO_FAD_DERIV_LOOP
+#if defined(SACADO_VIEW_CUDA_HIERARCHICAL_DFAD) && !defined(SACADO_DISABLE_CUDA_IN_KOKKOS) && defined(__CUDA_ARCH__)
+#define SACADO_FAD_DERIV_LOOP(I,SZ) for (int I=threadIdx.x; I<SZ; I+=blockDim.x)
+#else
+#define SACADO_FAD_DERIV_LOOP(I,SZ) for (int I=0; I<SZ; ++I)
+#endif
+#endif
+
     // Class representing a pointer to ViewFad so that &ViewFad is supported
     template <typename T, unsigned sl, unsigned ss, typename U>
     class ViewFadPtr;
@@ -118,10 +126,12 @@ namespace Sacado {
                 reinterpret_cast<const volatile char&>(x)))) {
           *val_ = *x.val_;
           if (stride_one)
-            for (int i=0; i<sz_.value; ++i)
+            //for (int i=0; i<sz_.value; ++i)
+            SACADO_FAD_DERIV_LOOP(i,sz_.value)
               dx_[i] = x.dx_[i];
           else
-            for (int i=0; i<sz_.value; ++i)
+            //for (int i=0; i<sz_.value; ++i)
+            SACADO_FAD_DERIV_LOOP(i,sz_.value)
               dx_[i*stride_.value] = x.dx_[i*x.stride_.value];
         }
         return *this;

packages/sacado/test/performance/CMakeLists.txt

@@ -76,29 +76,34 @@ ENDIF()
 
 IF (Sacado_ENABLE_TeuchosCore AND Sacado_ENABLE_KokkosCore)
 
-  TRIBITS_ADD_EXECUTABLE(
-    fad_kokkos_view
-    SOURCES fad_kokkos_view.cpp
-    COMM serial mpi
-    )
-
-  # These tests do not compile with gcc 4.7.x because it doesn't properly
-  # support lambdas.  See github issue #854
-  IF(NOT ((CMAKE_CXX_COMPILER_ID STREQUAL "GNU") AND (CMAKE_CXX_COMPILER_VERSION VERSION_LESS "4.8")))
-    TRIBITS_ADD_EXECUTABLE(
-      fad_kokkos_hierarchical
-      SOURCES fad_kokkos_hierarchical.cpp
-      COMM serial mpi
-      )
-
-#    TRIBITS_ADD_EXECUTABLE(
-#      fad_kokkos_mat_vec_perf
-#      SOURCES fad_kokkos_mat_vec_perf.cpp
-#      COMM serial mpi
-#      )
-  ENDIF()
+  # Disable these tests as they have been replaced by mat_vec, advection* below
+
+  # TRIBITS_ADD_EXECUTABLE(
+  #   fad_kokkos_view
+  #   SOURCES fad_kokkos_view.cpp
+  #   COMM serial mpi
+  #   )
+
+  # # These tests do not compile with gcc 4.7.x because it doesn't properly
+  # # support lambdas.  See github issue #854
+  # IF(NOT ((CMAKE_CXX_COMPILER_ID STREQUAL "GNU") AND (CMAKE_CXX_COMPILER_VERSION VERSION_LESS "4.8")))
+  #   TRIBITS_ADD_EXECUTABLE(
+  #     fad_kokkos_hierarchical
+  #     SOURCES fad_kokkos_hierarchical.cpp
+  #     COMM serial mpi
+  #     )
+
+  #  TRIBITS_ADD_EXECUTABLE(
+  #    fad_kokkos_mat_vec_perf
+  #    SOURCES fad_kokkos_mat_vec_perf.cpp
+  #    COMM serial mpi
+  #    )
+  # ENDIF()
 
 ENDIF()
 
 ADD_SUBDIRECTORY(fenl_assembly)
 ADD_SUBDIRECTORY(fenl_assembly_view)
+ADD_SUBDIRECTORY(mat_vec)
+ADD_SUBDIRECTORY(advection)
+ADD_SUBDIRECTORY(advection_const_basis)

packages/sacado/test/performance/advection/CMakeLists.txt

@@ -0,0 +1,22 @@
+ASSERT_DEFINED(PACKAGE_SOURCE_DIR CMAKE_CURRENT_SOURCE_DIR)
+
+INCLUDE_DIRECTORIES(REQUIRED_DURING_INSTALLATION_TESTING ${CMAKE_CURRENT_SOURCE_DIR})
+INCLUDE_DIRECTORIES(${CMAKE_CURRENT_BINARY_DIR})
+
+IF (Sacado_ENABLE_KokkosCore AND Sacado_ENABLE_TeuchosCore)
+
+  IF(NOT ((CMAKE_CXX_COMPILER_ID STREQUAL "GNU") AND (CMAKE_CXX_COMPILER_VERSION VERSION_LESS "4.8")))
+
+    TRIBITS_ADD_EXECUTABLE(
+      FadAdvectionHierarchicalTest
+      SOURCES common.hpp
+              advection.hpp advection.cpp
+              advection_hierarchical.hpp advection_hierarchical.cpp
+              advection_hierarchical_dfad.hpp advection_hierarchical_dfad.cpp
+              driver.cpp
+      COMM serial mpi
+      )
+
+  ENDIF()
+
+ENDIF()