fix c++20 implementations

.github/workflows/continuousIntegration.yml

@@ -80,12 +80,14 @@ jobs:
           lfs: false
       - name: code_checks
         env:
-          DOCKER_REPOSITORY: geosx/ubuntu:22.04-${{ needs.get_docker_image_tag_hash.outputs.DOCKER_IMAGE_TAG_HASH }}
+          # Cppcheck 2.13 and LLVM 20 support the project's required C++20 syntax.
+          DOCKER_REPOSITORY: geosx/ubuntu:24.04-clang20-1c3c049b3f629d9d44838656fd306b2a0c04c9e8
           HOST_CONFIG: hostconfigs/environment.cmake
-          CMAKE_CXX_COMPILER: /usr/bin/clang++
-          CMAKE_C_COMPILER: /usr/bin/clang
+          CMAKE_CXX_COMPILER: /usr/lib/llvm-20/bin/clang++
+          CMAKE_C_COMPILER: /usr/lib/llvm-20/bin/clang
           CMAKE_BUILD_TYPE: Release
           BUILD_AND_TEST_ARGS: --code-checks
+          DOCKER_RUN_ARGS: --env PATH=/usr/lib/llvm-20/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
         run: ./scripts/ci_build_and_test.sh
 
   linux_builds:
@@ -154,51 +156,7 @@ jobs:
             CMAKE_C_COMPILER: /usr/bin/gcc
             CMAKE_BUILD_TYPE: Debug
 
-          - name: ubuntu22-gcc11-cuda11-rel (ubuntu-22, gcc 11.4.0, cuda-11.8.0, Release)
-            DOCKER_REPOSITORY: geosx/ubuntu:22.04-cuda11.8
-            RUNS_ON: streak2
-            CMAKE_CXX_COMPILER: /usr/bin/g++
-            CMAKE_C_COMPILER: /usr/bin/gcc
-            CMAKE_BUILD_TYPE: Release
-            ENABLE_CUDA: ON
-            CMAKE_CUDA_ARCHITECTURES: "86"
-            NPROC: 1
-            DOCKER_RUN_ARGS: "--cpus=1 --memory=64g --runtime=nvidia --gpus all"
-
-          - name: ubuntu22-gcc11-cuda11-dbg (ubuntu-22, gcc 11.4.0, cuda-11.8.0, Debug)
-            DOCKER_REPOSITORY: geosx/ubuntu:22.04-cuda11.8
-            RUNS_ON: streak2
-            CMAKE_CXX_COMPILER: /usr/bin/g++
-            CMAKE_C_COMPILER: /usr/bin/gcc
-            CMAKE_BUILD_TYPE: Debug
-            ENABLE_CUDA: ON
-            CMAKE_CUDA_ARCHITECTURES: "86"
-            NPROC: 1
-            DOCKER_RUN_ARGS: "--cpus=1 --memory=64g --runtime=nvidia --gpus all"
-
-          - name: ubuntu22-clang14-cuda11-rel (ubuntu-22, clang-14.0.0, cuda-11.8.0, Release)
-            DOCKER_REPOSITORY: geosx/ubuntu:22.04-cuda11.8
-            RUNS_ON: streak2
-            CMAKE_CXX_COMPILER: /usr/bin/clang++
-            CMAKE_C_COMPILER: /usr/bin/clang
-            CMAKE_BUILD_TYPE: Release
-            ENABLE_CUDA: ON
-            CMAKE_CUDA_ARCHITECTURES: "86"
-            NPROC: 1
-            DOCKER_RUN_ARGS: "--cpus=1 --memory=64g --runtime=nvidia --gpus all"
-
-          - name: ubuntu22-clang14-cuda11-dbg (ubuntu-22, clang-14.0.0, cuda-11.8.0, Debug)
-            DOCKER_REPOSITORY: geosx/ubuntu:22.04-cuda11.8
-            RUNS_ON: streak2
-            CMAKE_CXX_COMPILER: /usr/bin/clang++
-            CMAKE_C_COMPILER: /usr/bin/clang
-            CMAKE_BUILD_TYPE: Debug
-            ENABLE_CUDA: ON
-            CMAKE_CUDA_ARCHITECTURES: "86"
-            NPROC: 1
-            DOCKER_RUN_ARGS: "--cpus=1 --memory=64g --runtime=nvidia --gpus all"
-
-          - name: RHEL8.10-clang17-cuda12-rel (ubi8.10, clang-17.0.6, cuda-12.4.1, Release)
+          - name: RHEL8.10-clang17-cuda12-rel (ubi8.10, clang-17.0.6, cuda-12.8, Release)
             DOCKER_REPOSITORY: geosx/ubi:8.10-cuda12.8
             RUNS_ON: streak2
             CMAKE_CXX_COMPILER: clang++
@@ -209,7 +167,7 @@ jobs:
             NPROC: 1
             DOCKER_RUN_ARGS: "--cpus=1 --memory=64g --runtime=nvidia --gpus all"
 
-          - name: RHEL8.10-gcc13-cuda12-rel (ubi8.10, gcc 13.2.1, cuda-12.4.1, Release)
+          - name: RHEL8.10-gcc13-cuda12-rel (ubi8.10, gcc 13.2.1, cuda-12.8, Release)
             DOCKER_REPOSITORY: geosx/ubi:8.10-cuda12.8
             RUNS_ON: streak2
             CMAKE_CXX_COMPILER: /opt/rh/gcc-toolset-13/root/bin/g++
@@ -293,4 +251,4 @@ jobs:
       run: "true"
     - name: PR Failure
       if: ${{ contains(env.RETURN_VAL, 'false') }}
-      run: "false"
+      run: "false"

CMakeLists.txt

@@ -46,11 +46,11 @@ target_include_directories( Shiva INTERFACE
                             "$<BUILD_INTERFACE:${SHIVA_BINARY_DIR}/include>"   # build/include/shiva/ShivaConfig.hpp
                             "$<INSTALL_INTERFACE:${CMAKE_INSTALL_INCLUDEDIR}>" )
 
-target_compile_features(Shiva INTERFACE cxx_std_17)
+target_compile_features(Shiva INTERFACE cxx_std_20 cuda_std_20)
 
 # -------- BLT (optional; usable as top-level or submodule) ---------------------
 if( SHIVA_ENABLE_BLT )
-  set( BLT_CXX_STD "c++17" )
+  set( BLT_CXX_STD "c++20" )
   # Load BLT if Shiva is standalone
   if( SHIVA_IS_TOPLEVEL )
     set(BLT_SOURCE_DIR "${PROJECT_SOURCE_DIR}/cmake/blt" CACHE PATH "Path to BLT")

README.md

@@ -4,10 +4,20 @@
 # Shiva
 Shiva Discretization Library
 
+Requirements: a C++20 compiler. CUDA builds require CUDA 12 or newer.
+
+HIP consumers must explicitly require C++20 on each target that compiles HIP sources using Shiva. The Shiva interface propagates the C++ and CUDA standards; HIP standard propagation is unavailable at the project's minimum CMake version. Replace `my_target` with your consuming target:
+
+```cmake
+set_target_properties(my_target PROPERTIES
+  HIP_STANDARD 20
+  HIP_STANDARD_REQUIRED ON)
+```
+
 # Project Goals
 The goals are:
 
 - Provide portable device callable functionality to enable the application of discrete numerical methods such as the Finite Element Method, and Finite Difference Method, etc.
 - Provide compile time interface to functionality through use of templates and `constexpr` for the special case where compile time knowledge is available.
 - Specialize for common use cases to achieve maximum performance. For example a Hexahedral element with 1st order Lagrange shape functions and Gauss-Legendre quadrature.
-- Support wide range of base geometries (tet, prism, pyramid, hex, polyhedral, etc), and discretization methods ( FEM, FVM, MPM, etc. )
+- Support wide range of base geometries (tet, prism, pyramid, hex, polyhedral, etc), and discretization methods ( FEM, FVM, MPM, etc. )

cmake/Macros.cmake

@@ -32,19 +32,23 @@ macro(shiva_add_code_checks)
         endforeach()
     endif()
 
-    set( CPPCHECK_FLAGS --std=c++17 
-                        --enable=all 
+    set( CPPCHECK_FLAGS --std=c++20
+                        --enable=all
                         --quiet
+                        # Use the existing SHIVA_HAS_BUILTIN fallback during analysis.
+                        -U__has_builtin
+                        # Omit the informational summary: testCppCheck rejects any stderr.
+                        --suppress=checkersReport
                         --suppress=missingIncludeSystem
                         --suppress=unmatchedSuppression
-                        --suppress=missingInclude 
-                        --suppress=noConstructor 
+                        --suppress=missingInclude
+                        --suppress=noConstructor
                         --suppress=noExplicitConstructor
-                        --suppress=unusedFunction 
-                        --suppress=constStatement 
+                        --suppress=unusedFunction
+                        --suppress=constStatement
                         --suppress=unusedStructMember
                         --suppress=unknownMacro )
-                        
+
     blt_add_code_checks( PREFIX    ${arg_PREFIX}
                          SOURCES   ${_sources}
                          UNCRUSTIFY_CFG_FILE ${PROJECT_SOURCE_DIR}/src/uncrustify.cfg
@@ -71,4 +75,4 @@ macro(shiva_add_code_checks)
                                      RUNNER ctest -E 'blt_gtest_smoke|testCppCheck|testClangTidy|testUncrustifyCheck|testDoxygenCheck|testCppCheck|testClangTidy'
                                      SOURCE_DIRECTORIES ${PROJECT_SOURCE_DIR}/include )
     endif()
-endmacro(shiva_add_code_checks)
+endmacro(shiva_add_code_checks)

hostconfigs/TTE/maple_rocky9.cmake

@@ -15,7 +15,7 @@ set(CUDA_TOOLKIT_ROOT_DIR /hrtc/apps/cuda/12.6.20/aarch64/rocky9 CACHE STRING ""
 set(CMAKE_CUDA_HOST_COMPILER ${CMAKE_CXX_COMPILER} CACHE STRING "")
 set(CMAKE_CUDA_COMPILER ${CUDA_TOOLKIT_ROOT_DIR}/bin/nvcc CACHE STRING "")
 set(CMAKE_CUDA_ARCHITECTURES 90 CACHE STRING "")
-set(CMAKE_CUDA_STANDARD 17 CACHE STRING "")
+set(CMAKE_CUDA_STANDARD 20 CACHE STRING "")
 set(CMAKE_CUDA_FLAGS "-restrict --expt-extended-lambda --expt-relaxed-constexpr -Werror cross-execution-space-call,reorder,deprecated-declarations" CACHE STRING "")
 #set(CMAKE_CUDA_FLAGS_RELEASE "-O3 -DNDEBUG -Xcompiler -DNDEBUG -Xcompiler -O3 -Xcompiler -mcpu=powerpc64le -Xcompiler -mtune=powerpc64le" CACHE STRING "")
 #set(CMAKE_CUDA_FLAGS_RELWITHDEBINFO "-g -lineinfo ${CMAKE_CUDA_FLAGS_RELEASE}" CACHE STRING "")

include/shiva/common/ShivaMacros.hpp

@@ -120,7 +120,7 @@ void i_g_n_o_r_e( ARGS const & ... ) {}
  * @brief Define SHIVA_IS_CONST_EVAL() depending on compiler/toolchain
  */
 #if defined(__CUDA_ARCH__)
-// Device code (nvcc, hipcc): no support in C++17
+// Keep CUDA device checks on the runtime path.
   #define SHIVA_IS_CONST_EVAL() (false)
 
 #elif SHIVA_HAS_BUILTIN( __builtin_is_constant_evaluated )

include/shiva/functions/bases/BasisProduct.hpp

@@ -70,6 +70,7 @@ struct BasisProduct
    * coordinate.
    * @tparam BASIS_FUNCTION_INDICES Pack of indices of the basis function to
    * evaluate in each dimension.
+   * @tparam COORD_TYPE Coordinate type supporting indexing in each dimension.
    * @param parentCoord The parent coordinate at which to evaluate the basis
    * function.
    * @return The value of the basis function at the specified parent coordinate.
@@ -79,7 +80,7 @@ struct BasisProduct
    * \Phi_{i_0 i_1 ... i_{(numDims-1)}}(\boldsymbol{\xi}) =
    * \prod_{k=0}^{(numDims-1)} \phi_{i_k}(\xi_k)\text{, where } \\
    * i_j \text{is index of the basis function in the jth dimension, and
-   * ranges from [0...(order+1)]}
+   * ranges from 0 to one less than the number of support points in that dimension.}
    * \f]
    *
    */
@@ -89,30 +90,19 @@ struct BasisProduct
   {
     static_assert( sizeof...(BASIS_FUNCTION_INDICES) == numDims, "Wrong number of basis function indicies specified" );
 
-
-#if __cplusplus >= 202002L
-    // expand pack over number of dimensions
-    return executeSequence< numDims >( [&]< int ... PRODUCT_TERM_INDEX > () constexpr
-    {
-      return ( BASIS_TYPE::template value< BASIS_FUNCTION_INDICES >( parentCoord[PRODUCT_TERM_INDEX] ) * ... );
-    } );
-#else
+    // Integral-constant arguments preserve distinct coordinate indices in NVCC device code.
     return executeSequence< numDims >( [&] ( auto ... PRODUCT_TERM_INDEX ) constexpr
     {
-      // fold expression to multiply the value of each BASIS_TYPE in each
-      // dimension. In other words the fold expands on BASIS_TYPE...,
-      // BASIS_FUNCTION_INDICES..., and PRODUCT_TERM_INDEX... together.
       return ( BASIS_TYPES::template value< BASIS_FUNCTION_INDICES >( parentCoord[decltype(PRODUCT_TERM_INDEX)::value] ) * ... );
     } );
-
-#endif
   }
 
   /**
    * @brief Calculates the gradient of the basis function at the specified
    * parent coordinate.
    * @tparam BASIS_FUNCTION_INDICES Pack of indices of the basis function to
    * evaluate in each dimension.
+   * @tparam COORD_TYPE Coordinate type supporting indexing in each dimension.
    * @param parentCoord The parent coordinate at which to evaluate the basis
    * function gradient.
    * @return The gradient of the basis function at the specified parent
@@ -131,14 +121,12 @@ struct BasisProduct
   gradient( COORD_TYPE const & parentCoord )
   {
     static_assert( sizeof...(BASIS_FUNCTION_INDICES) == numDims, "Wrong number of basis function indicies specified" );
-
-#if __cplusplus >= 202002L
     return executeSequence< numDims >( [&]< int ... i > () constexpr->CArrayNd< RealType, numDims >
     {
       auto gradientComponent = [&] ( auto const iGrad,
                                      auto const  ... PRODUCT_TERM_INDICES ) constexpr
       {
-        // Ca
+        // Multiply the directional gradient by the values in every other dimension.
         return ( gradientComponentHelper< BASIS_TYPES,
                                           decltype(iGrad)::value,
                                           BASIS_FUNCTION_INDICES,
@@ -147,46 +135,26 @@ struct BasisProduct
 
       return { (executeSequence< numDims >( gradientComponent, std::integral_constant< int, i >{} ) )...  };
     } );
-#else
-    // Expand over the dimensions.
-    return executeSequence< numDims >( [&] ( auto ... a ) constexpr->CArrayNd< RealType, numDims >
-    {
-      // define a lambda that calculates the gradient of the basis function in
-      // a single dimension/direction.
-      auto gradientComponent = [&] ( auto GRADIENT_COMPONENT, auto ... PRODUCT_TERM_INDICES ) constexpr
-      {
-        // fold expression calling gradientComponentHelper using expanding on
-        // BASIS_TYPE, BASIS_FUNCTION_INDICES, and PRODUCT_TERM_INDICES.
-        return ( gradientComponentHelper< BASIS_TYPES,
-                                          decltype(GRADIENT_COMPONENT)::value,
-                                          BASIS_FUNCTION_INDICES,
-                                          decltype(PRODUCT_TERM_INDICES)::value >( parentCoord ) * ... );
-      };
-
-      // execute the gradientComponent lambda on each direction, expand the
-      // pack on "i" corresponding to each direction of the gradient.
-      return { (executeSequence< numDims >( gradientComponent, a ) )...  };
-    } );
-#endif
   }
 
 
 private:
   /**
    * @brief Helper function to return the gradient of a basis function, or the
-   * value of the basis function depending on whether or not the index of the
-   * basis function (@p BASIS_FUNCTION) matches the index of the direction
-   * component index (@p GRADIENT_COMPONENT). This filter is illustrated in the
+   * value of the basis function depending on whether the coordinate dimension
+   * (@p COORD_INDEX) matches the gradient direction (@p GRADIENT_COMPONENT).
+   * This filter is illustrated in the
    * documentation for the gradient function.
    * @tparam BASIS_FUNCTION The basis function type
    * @tparam GRADIENT_COMPONENT The dimension component of the gradient.
    * @tparam BASIS_FUNCTION_INDEX The index of the basis function that is being
    * evaluated.
    * @tparam COORD_INDEX The dimension component of the coordinate.
+   * @tparam COORD_TYPE Coordinate type supporting indexing in each dimension.
    * @param parentCoord The parent coordinate at which to evaluate the basis
    * function gradient.
-   * @return The gradient component of the basis function at the specified
-   * parent coordinate.
+   * @return The one-dimensional basis derivative when the dimensions match,
+   * otherwise the one-dimensional basis value.
    */
   template< typename BASIS_FUNCTION, int GRADIENT_COMPONENT, int BASIS_FUNCTION_INDEX, int COORD_INDEX, typename COORD_TYPE >
   SHIVA_STATIC_CONSTEXPR_HOSTDEVICE_FORCEINLINE RealType