Modify ex26 with anisotropic diffusion and ILU smoothing

Variable order transfer operator

.gitignore

@@ -259,6 +259,8 @@ miniapps/performance/sol.*
 
 miniapps/shifted/distance
 miniapps/shifted/ParaViewDistance
+miniapps/shifted/extrapolate
+miniapps/shifted/ParaViewExtrapolate
 miniapps/shifted/diffusion
 miniapps/shifted/diffusion.mesh
 miniapps/shifted/diffusion.gf

.gitlab/configs/lassen-config.yml

@@ -45,5 +45,5 @@ variables:
     - echo ${MFEM_DATA_DIR}
     - echo ${SPEC}
     # Next script uses 'THREADS': leaving it empty --> it uses 'make all -j'
-    - lalloc 1 -W 30 -q pdebug tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
+    - lalloc 1 -W 30 -q pdebug --atsdisable tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
   needs: [setup]

.gitlab/lassen-build-and-test.yml

@@ -19,6 +19,11 @@ opt_mpi_cuda_xl_16_1_1_8:
     SPEC: "%xl@16.1.1.8 +mpi +cuda cuda_arch=70"
   extends: .build_and_test_on_lassen
 
+opt_mpi_cuda_hypre_cuda_xl:
+  variables:
+    SPEC: "%xl@16.1.1.8 +mpi +cuda cuda_arch=70 ^hypre+cuda~shared cuda_arch=70"
+  extends: .build_and_test_on_lassen
+
 # Jobs report
 report_job_success:
   stage: report

.gitlab/quartz-build-and-test.yml

@@ -59,7 +59,7 @@ opt_par_gcc_6_1_0_sundials:
 
 opt_par_gcc_6_1_0_petsc:
   variables:
-    SPEC: "%gcc@6.1.0 +petsc ^petsc+mumps"
+    SPEC: "%gcc@6.1.0 +petsc ^petsc+mumps~superlu-dist"
   extends: .build_and_test_on_quartz
 
 opt_par_gcc_6_1_0_pumi:

.gitlab/scripts/baseline

@@ -29,9 +29,9 @@ mkdir _${BASELINE_TEST} && cd _${BASELINE_TEST}
 
 # run
 if [[ "${MACHINE_NAME}" == "quartz" || "${MACHINE_NAME}" == "ruby" ]]; then
-  srun --nodes=1 -p pdebug ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+  salloc --nodes=1 -p pdebug ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 elif [[ ${MACHINE_NAME} == "corona" ]]; then
-  srun --nodes=1 -t 60 -p mi60 ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+  salloc --nodes=1 -t 60 -p pbatch ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 elif [[ ${MACHINE_NAME} == "lassen" ]]; then
   lalloc 1 -q pdebug ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 else

CHANGELOG

@@ -10,10 +10,27 @@
 
 Version 4.3.1 (development)
 ===========================
-- Added support for automatic differentiation. Users can select between 
-  native implementation and external library implementation at the 
-  configuration phase. A parallel and two serial examples are implemented
-  in the autodiff miniapp directory.
+
+- Add hipSPARSE support for sparse mat-vec multiplications.
+
+- Added support for using the HYPRE library built with HIP support. Similar to
+  the HYPRE + CUDA support added earlier, most of the MFEM examples and miniapps
+  work transparently with HYPRE + HIP builds. This includes the BoomerAMG, AMS,
+  and ADS solvers.
+
+- More explicit and consistent formating of the output of iterative solvers
+  with the new IterativeSolver::PrintLevel options. See linalg/solvers.hpp.
+
+- Added a miniapp for PDE-based extrapolation of finite element functions. See
+  miniapps/shifted/extrapolate.cpp.
+
+- Added support for automatic differentiation. Users can select between native
+  implementation and external library implementation during configuration. One
+  parallel and two serial examples are implemented in the miniapps/autodiff/
+  directory.
+
+- GridFunctionCoefficient (and the related vector, gradient, divergence, and
+  curl classes) now work properly with LORDiscretization and LORSolver.
 
 - Added support for mesh preprocessing to resolve fine scale problem data
   before simulation. This feature uses adaptive mesh refinement to control the
@@ -71,13 +88,19 @@ Version 4.3.1 (development)
 
 - Added initial TMOP-based capabilities for surface fitting and tangential
   relaxation in the mesh-optimizer and pmesh-optimizer miniapps.
- 
+
 - Added ParMesh Adjaceny Set (adjset) creation support to the Conduit Mesh
   Blueprint MFEM wrapper functions in ConduitDataCollection.
 
 - `HypreParVector` and `Vector` now support move semantics, and the copy
   constructor for `HypreParVector` now copies the local vector data.
 
+- The HPC versions of ex1 and ex1p (in miniapps/performance) now support
+  runtime selection of either 2D or 3D meshes.
+
+- Added ParaView visualization of `QuadratureFunction` fields, through both
+  `QuadratureFunction::SaveVTU` and `ParaViewDataCollection::RegisterQField`.
+
 
 Version 4.3, released on July 29, 2021
 ======================================

CMakeLists.txt

@@ -16,9 +16,6 @@ set(USER_CONFIG "${CMAKE_CURRENT_SOURCE_DIR}/config/user.cmake" CACHE PATH
 
 # Require C++11 and disable compiler-specific extensions
 set(CMAKE_CXX_STANDARD 11)
-if (MFEM_USE_GINKGO)
-  set(CMAKE_CXX_STANDARD 14)
-endif()
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
 
@@ -84,6 +81,9 @@ if (MFEM_USE_STRUMPACK)
   # Just needed to find the MPI_Fortran libraries to link with
   set(XSDK_ENABLE_Fortran ON)
 endif()
+if (MFEM_USE_GINKGO AND ("${CMAKE_CXX_STANDARD}" LESS "14"))
+  set(CMAKE_CXX_STANDARD 14)
+endif()
 
 # Include xSDK default CMake file.
 include("${CMAKE_CURRENT_SOURCE_DIR}/config/XSDKDefaults.cmake")
@@ -180,6 +180,19 @@ else()
   set(MFEM_DEBUG OFF)
 endif()
 
+# AMD HIP
+if (MFEM_USE_HIP)
+  if (HIP_ARCH)
+    message(STATUS "Using HIP architecture: ${HIP_ARCH}")
+    set(GPU_TARGETS "${HIP_ARCH}" CACHE STRING "HIP targets to compile for")
+  endif()
+  if (ROCM_PATH)
+    list(INSERT CMAKE_PREFIX_PATH 0 ${ROCM_PATH})
+  endif()
+  find_package(HIP REQUIRED)
+  find_package(HIPSPARSE REQUIRED)
+endif()
+
 # MPI -> hypre; PETSc (optional)
 if (MFEM_USE_MPI)
   find_package(MPI REQUIRED)
@@ -354,7 +367,7 @@ endif()
 if (MFEM_USE_PUMI)
   # If PUMI_DIR was specified, only link to that directory,
   # i.e. don't link to another installation in /usr/lib by mistake
-  find_package(SCOREC 2.1.0 REQUIRED OPTIONAL_COMPONENTS gmi_sim
+  find_package(SCOREC 2.2.6 REQUIRED OPTIONAL_COMPONENTS gmi_sim
     CONFIG PATHS ${PUMI_DIR} NO_DEFAULT_PATH)
   if (SCOREC_FOUND)
     # Define a header file with the MFEM_USE_SIMMETRIX preprocessor variable
@@ -403,18 +416,6 @@ if (MFEM_USE_CALIPER)
   find_package(Caliper REQUIRED)
 endif()
 
-# AMD HIP
-if (MFEM_USE_HIP)
-  find_package(HIP REQUIRED)
-  if (HIP_ARCH)
-    message(STATUS "Using HIP architecture: ${HIP_ARCH}")
-    list(APPEND HIP_HIPCC_FLAGS "--amdgpu-target=${HIP_ARCH}")
-    if (MFEM_USE_GINKGO)
-      list(APPEND HIP_HIPCC_FLAGS "-std=c++14")
-    endif()
-  endif()
-endif()
-
 # ADIOS2 for parallel I/O
 if (MFEM_USE_ADIOS2)
    find_package(ADIOS2 REQUIRED)
@@ -454,10 +455,11 @@ endif()
 #    With newer versions of SuiteSparse which include METIS header using 64-bit
 #    integers, the METIS header (with 32-bit indices, as used by mfem) needs to
 #    be before SuiteSparse.
-set(MFEM_TPLS OPENMP HYPRE BLAS LAPACK SuperLUDist METIS SuiteSparse SUNDIALS PETSC
-    SLEPC MESQUITE MUMPS STRUMPACK AXOM FMS CONDUIT Ginkgo GNUTLS GSLIB NETCDF
-    MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE ADIOS2
-    CUSPARSE MKL_CPARDISO AMGX CALIPER CODIPACK BENCHMARK PARELAG MPI_CXX)
+set(MFEM_TPLS OPENMP HYPRE BLAS LAPACK SuperLUDist METIS SuiteSparse SUNDIALS
+    PETSC SLEPC MESQUITE MUMPS STRUMPACK AXOM FMS CONDUIT Ginkgo GNUTLS GSLIB
+    NETCDF MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE
+    ADIOS2 CUSPARSE MKL_CPARDISO AMGX CALIPER CODIPACK BENCHMARK PARELAG
+    MPI_CXX HIP HIPSPARSE)
 
 # Add all *_FOUND libraries in the variable TPL_LIBRARIES.
 set(TPL_LIBRARIES "")
@@ -500,8 +502,6 @@ endforeach()
 
 if (MFEM_USE_CUDA)
   set_source_files_properties(${SOURCES} PROPERTIES LANGUAGE CUDA)
-elseif(MFEM_USE_HIP)
-  set_source_files_properties(${SOURCES} PROPERTIES HIP_SOURCE_PROPERTY_FORMAT TRUE)
 endif()
 
 add_subdirectory(config)

CODE_OF_CONDUCT.md

@@ -0,0 +1,133 @@
+# MFEM Code of Conduct
+
+## Our Pledge
+
+We as members, contributors, and leaders pledge to make participation in our
+community a harassment-free experience for everyone, regardless of age, body
+size, visible or invisible disability, ethnicity, sex characteristics, gender
+identity and expression, level of experience, education, socio-economic status,
+nationality, personal appearance, race, caste, color, religion, or sexual
+identity and orientation.
+
+We pledge to act and interact in ways that contribute to an open, welcoming,
+diverse, inclusive, and healthy community.
+
+## Our Standards
+
+Examples of behavior that contributes to a positive environment for our
+community include:
+
+* Demonstrating empathy and kindness toward other people
+* Being respectful of differing opinions, viewpoints, and experiences
+* Giving and gracefully accepting constructive feedback
+* Accepting responsibility and apologizing to those affected by our mistakes,
+  and learning from the experience
+* Focusing on what is best not just for us as individuals, but for the overall
+  community
+
+Examples of unacceptable behavior include:
+
+* The use of sexualized language or imagery, and sexual attention or advances of
+  any kind
+* Trolling, insulting or derogatory comments, and personal or political attacks
+* Public or private harassment
+* Publishing others' private information, such as a physical or email address,
+  without their explicit permission
+* Other conduct which could reasonably be considered inappropriate in a
+  professional setting
+
+## Enforcement Responsibilities
+
+Community leaders are responsible for clarifying and enforcing our standards of
+acceptable behavior and will take appropriate and fair corrective action in
+response to any behavior that they deem inappropriate, threatening, offensive,
+or harmful.
+
+Community leaders have the right and responsibility to remove, edit, or reject
+comments, commits, code, wiki edits, issues, and other contributions that are
+not aligned to this Code of Conduct, and will communicate reasons for moderation
+decisions when appropriate.
+
+## Scope
+
+This Code of Conduct applies within all community spaces, and also applies when
+an individual is officially representing the community in public spaces.
+Examples of representing our community include using an official e-mail address,
+posting via an official social media account, or acting as an appointed
+representative at an online or offline event.
+
+## Enforcement
+
+Instances of abusive, harassing, or otherwise unacceptable behavior may be
+reported to the community leaders responsible for enforcement at mfem@llnl.gov.
+All complaints will be reviewed and investigated promptly and fairly.
+
+All community leaders are obligated to respect the privacy and security of the
+reporter of any incident. Anyone involved in the reported behavior will recuse
+themselves from the investigation and decision making about the resolution of
+the complaint.
+
+## Enforcement Guidelines
+
+Community leaders will follow these Community Impact Guidelines in determining
+the consequences for any action they deem in violation of this Code of Conduct:
+
+### 1. Correction
+
+**Community Impact**: Use of inappropriate language or other behavior deemed
+unprofessional or unwelcome in the community.
+
+**Consequence**: A private, written warning from community leaders, providing
+clarity around the nature of the violation and an explanation of why the
+behavior was inappropriate. A public apology may be requested.
+
+### 2. Warning
+
+**Community Impact**: A violation through a single incident or series of
+actions.
+
+**Consequence**: A warning with consequences for continued behavior. No
+interaction with the people involved, including unsolicited interaction with
+those enforcing the Code of Conduct, for a specified period of time. This
+includes avoiding interactions in community spaces as well as external channels
+like social media. Violating these terms may lead to a temporary or permanent
+ban.
+
+### 3. Temporary Ban
+
+**Community Impact**: A serious violation of community standards, including
+sustained inappropriate behavior.
+
+**Consequence**: A temporary ban from any sort of interaction or public
+communication with the community for a specified period of time. No public or
+private interaction with the people involved, including unsolicited interaction
+with those enforcing the Code of Conduct, is allowed during this period.
+Violating these terms may lead to a permanent ban.
+
+### 4. Permanent Ban
+
+**Community Impact**: Demonstrating a pattern of violation of community
+standards, including sustained inappropriate behavior, harassment of an
+individual, or aggression toward or disparagement of classes of individuals.
+
+**Consequence**: A permanent ban from any sort of public interaction within the
+community.
+
+## Attribution
+
+This Code of Conduct is adapted from the [Contributor Covenant][homepage],
+version 2.1, available at
+[https://www.contributor-covenant.org/version/2/1/code_of_conduct.html][v2.1].
+
+Community Impact Guidelines were inspired by
+[Mozilla's code of conduct enforcement ladder][Mozilla CoC].
+
+For answers to common questions about this code of conduct, see the FAQ at
+[https://www.contributor-covenant.org/faq][FAQ]. Translations are available at
+[https://www.contributor-covenant.org/translations][translations].
+
+[homepage]: https://www.contributor-covenant.org
+[v2.1]: https://www.contributor-covenant.org/version/2/1/code_of_conduct.html
+[Mozilla CoC]: https://github.com/mozilla/diversity
+[FAQ]: https://www.contributor-covenant.org/faq
+[translations]: https://www.contributor-covenant.org/translations

CONTRIBUTING.md

@@ -21,6 +21,9 @@ documentation; new examples and miniapps; HPC performance improvements; etc.
 MFEM is distributed under the terms of the BSD-3 license. All new contributions
 must be made under this license.
 
+Note also that MFEM has a [Code of Conduct](CODE_OF_CONDUCT.md). By participating
+in the MFEM community, you agree to abide by its rules.
+
 If you plan on contributing to MFEM, consider reviewing the
 [issue tracker](https://github.com/mfem/mfem/issues) first to check if a thread
 already exists for your desired feature or the bug you ran into. Use a pull

INSTALL

@@ -37,7 +37,7 @@ as CUDA, HIP, OCCA, OpenMP and RAJA.
   https://developer.nvidia.com/cuda-toolkit
 
 - HIP support requires an AMD GPU and an installation of the ROCm software stack
-  https://rocm.github.io/ROCmInstall.html#installing-from-amd-rocm-repositories
+  https://rocmdocs.amd.com
 
 - OCCA support requires the OCCA library
   https://libocca.org
@@ -58,7 +58,8 @@ following package managers:
 
 - Spack, https://github.com/spack/spack
 - OpenHPC, http://openhpc.community
-- Conda-forge, https://conda-forge.org (pre-built binaries linked with OpenMPI/MPICH, hypre, and METIS)
+- Conda-forge, https://conda-forge.org (pre-built binaries linked with
+  OpenMPI/MPICH, hypre, and METIS)
 - Homebrew/Science, https://github.com/Homebrew/homebrew-science (deprecated)
 
 We also recommend downloading and building the MFEM-based GLVis visualization
@@ -78,7 +79,7 @@ Parallel build:
 
 CUDA build:
    make cuda -j 4
-   (build for a specific compute capability: 'make cuda -j 4 CUDA_ARCH=sm_30')
+   (build for a specific compute capability: 'make cuda -j 4 CUDA_ARCH=sm_70')
 
 HIP build:
    make hip -j 4
@@ -566,7 +567,7 @@ The specific libraries and their options are:
   Options: HYPRE_OPT, HYPRE_LIB.
   Versions: HYPRE >= 2.10.0b  (HYPRE built without CUDA)
             HYPRE >= 2.20.0   (HYPRE built with '--enable-mixedint')
-            HYPRE >= 2.22.1   (HYPRE built with CUDA)
+            HYPRE >= 2.22.1   (HYPRE built with CUDA or HIP)
 
 - METIS, used when MFEM_USE_METIS = YES. If using METIS 5, set
   MFEM_USE_METIS_5 = YES (default is to use METIS 4).
@@ -639,7 +640,8 @@ The specific libraries and their options are:
   C++ compiler that supports the C++-14 standard. For additional requirements
   and dependencies of specific modules, see the Ginkgo webpage below.
   URL: https://ginkgo-project.github.io
-  Options: GINKGO_OPT, GINKGO_LIB, GINKGO_DIR, GINKGO_BUILD_TYPE (Release or Debug).
+  Options: GINKGO_OPT, GINKGO_LIB, GINKGO_DIR, GINKGO_BUILD_TYPE (Release or
+           Debug).
   Versions: Ginkgo >= 1.4.0.
 
 - AmgX (optional), used when MFEM_USE_AMGX = YES.
@@ -704,7 +706,7 @@ The specific libraries and their options are:
   URL: https://scorec.rpi.edu/pumi
        https://github.com/SCOREC/core
   Options: PUMI_OPT, PUMI_LIB.
-  Versions: PUMI == 2.2.3.
+  Versions: PUMI >= 2.2.6.
 
 - HiOp (optional), used when MFEM_USE_HIOP = YES.
   URL: https://github.com/LLNL/hiop
@@ -737,7 +739,7 @@ The specific libraries and their options are:
   Versions: CUDA >= 10.1.168.
 
 - HIP (optional), used when MFEM_USE_HIP = YES.
-  URL: https://rocm.github.io/ROCmInstall.html
+  URL: https://rocmdocs.amd.com
   Options: HIP_CXX, HIP_ARCH, HIP_OPT, HIP_LIB.
 
 - OCCA (optional), used when MFEM_USE_OCCA = YES.
@@ -924,6 +926,7 @@ MFEM_USE_HIOP
 MFEM_USE_CODIPACK
 MFEM_USE_ADFORWARD
 MFEM_USE_CUDA
+MFEM_USE_HIP
 MFEM_USE_OCCA
 MFEM_USE_CEED
 MFEM_USE_RAJA

README

@@ -12,6 +12,9 @@ to enable high-performance scalable finite element discretization research and
 application development on a wide variety of platforms, ranging from laptops to
 supercomputers.
 
+We welcome contributions and feedback from the community. Please see the file
+CONTRIBUTING.md for additional details about our development process.
+
 * For building instructions, see the file INSTALL, or type "make help".
 
 * Copyright and licensing information can be found in files LICENSE and NOTICE.
@@ -19,9 +22,6 @@ supercomputers.
 * The best starting point for new users interested in MFEM's features is to
   review the examples and miniapps at https://mfem.org/examples.
 
-* Developers interested in contributing to the library, should read the
-  instructions and documentation in the CONTRIBUTING.md file.
-
 Conceptually, MFEM can be viewed as a finite element toolbox that provides the
 building blocks for developing finite element algorithms in a manner similar to
 that of MATLAB for linear algebra methods. In particular, MFEM provides support

config/cmake/modules/FindHIP.cmake

@@ -1,692 +0,0 @@
-###############################################################################
-# FindHIP.cmake
-###############################################################################
-include(CheckCXXCompilerFlag)
-###############################################################################
-# SET: Variable defaults
-###############################################################################
-# User defined flags
-set(HIP_HIPCC_FLAGS "" CACHE STRING "Semicolon delimited flags for HIPCC")
-set(HIP_HCC_FLAGS "" CACHE STRING "Semicolon delimited flags for HCC")
-set(HIP_CLANG_FLAGS "" CACHE STRING "Semicolon delimited flags for CLANG")
-set(HIP_NVCC_FLAGS "" CACHE STRING "Semicolon delimted flags for NVCC")
-mark_as_advanced(HIP_HIPCC_FLAGS HIP_HCC_FLAGS HIP_CLANG_FLAGS HIP_NVCC_FLAGS)
-
-set(_hip_configuration_types ${CMAKE_CONFIGURATION_TYPES} ${CMAKE_BUILD_TYPE} Debug MinSizeRel Release RelWithDebInfo)
-list(REMOVE_DUPLICATES _hip_configuration_types)
-foreach(config ${_hip_configuration_types})
-    string(TOUPPER ${config} config_upper)
-    set(HIP_HIPCC_FLAGS_${config_upper} "" CACHE STRING "Semicolon delimited flags for HIPCC")
-    set(HIP_HCC_FLAGS_${config_upper} "" CACHE STRING "Semicolon delimited flags for HCC")
-    set(HIP_CLANG_FLAGS_${config_upper} "" CACHE STRING "Semicolon delimited flags for CLANG")
-    set(HIP_NVCC_FLAGS_${config_upper} "" CACHE STRING "Semicolon delimited flags for NVCC")
-    mark_as_advanced(HIP_HIPCC_FLAGS_${config_upper} HIP_HCC_FLAGS_${config_upper} HIP_CLANG_FLAGS_${config_upper} HIP_NVCC_FLAGS_${config_upper})
-endforeach()
-option(HIP_HOST_COMPILATION_CPP "Host code compilation mode" ON)
-option(HIP_VERBOSE_BUILD "Print out the commands run while compiling the HIP source file.  With the Makefile generator this defaults to VERBOSE variable specified on the command line, but can be forced on with this option." OFF)
-mark_as_advanced(HIP_HOST_COMPILATION_CPP)
-
-###############################################################################
-# FIND: HIP and associated helper binaries
-###############################################################################
-
-get_filename_component(_IMPORT_PREFIX "${CMAKE_CURRENT_LIST_DIR}/../" REALPATH)
-
-# HIP is supported on Linux only
-if(UNIX AND NOT APPLE AND NOT CYGWIN)
-    # Search for HIP installation
-    if(NOT HIP_ROOT_DIR)
-        # Search in user specified path first
-        find_path(
-            HIP_ROOT_DIR
-            NAMES bin/hipconfig
-            PATHS
-            "$ENV{ROCM_PATH}/hip"
-            ENV HIP_PATH
-            ${_IMPORT_PREFIX}
-            /opt/rocm/hip
-            DOC "HIP installed location"
-            NO_DEFAULT_PATH
-            )
-        if(NOT EXISTS ${HIP_ROOT_DIR})
-            if(HIP_FIND_REQUIRED)
-                message(FATAL_ERROR "Specify HIP_ROOT_DIR")
-            elseif(NOT HIP_FIND_QUIETLY)
-                message("HIP_ROOT_DIR not found or specified")
-            endif()
-        endif()
-        # And push it back to the cache
-        set(HIP_ROOT_DIR ${HIP_ROOT_DIR} CACHE PATH "HIP installed location" FORCE)
-    endif()
-
-    # Find HIPCC executable
-    find_program(
-        HIP_HIPCC_EXECUTABLE
-        NAMES hipcc
-        PATHS
-        "${HIP_ROOT_DIR}"
-        ENV ROCM_PATH
-        ENV HIP_PATH
-        /opt/rocm
-        /opt/rocm/hip
-        PATH_SUFFIXES bin
-        NO_DEFAULT_PATH
-        )
-    if(NOT HIP_HIPCC_EXECUTABLE)
-        # Now search in default paths
-        find_program(HIP_HIPCC_EXECUTABLE hipcc)
-    endif()
-    mark_as_advanced(HIP_HIPCC_EXECUTABLE)
-
-    # Find HIPCONFIG executable
-    find_program(
-        HIP_HIPCONFIG_EXECUTABLE
-        NAMES hipconfig
-        PATHS
-        "${HIP_ROOT_DIR}"
-        ENV ROCM_PATH
-        ENV HIP_PATH
-        /opt/rocm
-        /opt/rocm/hip
-        PATH_SUFFIXES bin
-        NO_DEFAULT_PATH
-        )
-    if(NOT HIP_HIPCONFIG_EXECUTABLE)
-        # Now search in default paths
-        find_program(HIP_HIPCONFIG_EXECUTABLE hipconfig)
-    endif()
-    mark_as_advanced(HIP_HIPCONFIG_EXECUTABLE)
-
-    # Find HIPCC_CMAKE_LINKER_HELPER executable
-    find_program(
-        HIP_HIPCC_CMAKE_LINKER_HELPER
-        NAMES hipcc_cmake_linker_helper
-        PATHS
-        "${HIP_ROOT_DIR}"
-        ENV ROCM_PATH
-        ENV HIP_PATH
-        /opt/rocm
-        /opt/rocm/hip
-        PATH_SUFFIXES bin
-        NO_DEFAULT_PATH
-        )
-    if(NOT HIP_HIPCC_CMAKE_LINKER_HELPER)
-        # Now search in default paths
-        find_program(HIP_HIPCC_CMAKE_LINKER_HELPER hipcc_cmake_linker_helper)
-    endif()
-    mark_as_advanced(HIP_HIPCC_CMAKE_LINKER_HELPER)
-
-    if(HIP_HIPCONFIG_EXECUTABLE AND NOT HIP_VERSION)
-        # Compute the version
-        execute_process(
-            COMMAND ${HIP_HIPCONFIG_EXECUTABLE} --version
-            OUTPUT_VARIABLE _hip_version
-            ERROR_VARIABLE _hip_error
-            OUTPUT_STRIP_TRAILING_WHITESPACE
-            ERROR_STRIP_TRAILING_WHITESPACE
-            )
-        if(NOT _hip_error)
-            set(HIP_VERSION ${_hip_version} CACHE STRING "Version of HIP as computed from hipcc")
-        else()
-            set(HIP_VERSION "0.0.0" CACHE STRING "Version of HIP as computed by FindHIP()")
-        endif()
-        mark_as_advanced(HIP_VERSION)
-    endif()
-    if(HIP_VERSION)
-        string(REPLACE "." ";" _hip_version_list "${HIP_VERSION}")
-        list(GET _hip_version_list 0 HIP_VERSION_MAJOR)
-        list(GET _hip_version_list 1 HIP_VERSION_MINOR)
-        list(GET _hip_version_list 2 HIP_VERSION_PATCH)
-        set(HIP_VERSION_STRING "${HIP_VERSION}")
-    endif()
-
-    if(HIP_HIPCONFIG_EXECUTABLE AND NOT HIP_PLATFORM)
-        # Compute the platform
-        execute_process(
-            COMMAND ${HIP_HIPCONFIG_EXECUTABLE} --platform
-            OUTPUT_VARIABLE _hip_platform
-            OUTPUT_STRIP_TRAILING_WHITESPACE
-            )
-        set(HIP_PLATFORM ${_hip_platform} CACHE STRING "HIP platform as computed by hipconfig")
-        mark_as_advanced(HIP_PLATFORM)
-    endif()
-
-    if(HIP_HIPCONFIG_EXECUTABLE AND NOT HIP_COMPILER)
-        # Compute the compiler
-        execute_process(
-            COMMAND ${HIP_HIPCONFIG_EXECUTABLE} --compiler
-            OUTPUT_VARIABLE _hip_compiler
-            OUTPUT_STRIP_TRAILING_WHITESPACE
-            )
-        set(HIP_COMPILER ${_hip_compiler} CACHE STRING "HIP compiler as computed by hipconfig")
-        mark_as_advanced(HIP_COMPILER)
-    endif()
-
-    if(HIP_HIPCONFIG_EXECUTABLE AND NOT HIP_RUNTIME)
-        # Compute the runtime
-        execute_process(
-            COMMAND ${HIP_HIPCONFIG_EXECUTABLE} --runtime
-            OUTPUT_VARIABLE _hip_runtime
-            OUTPUT_STRIP_TRAILING_WHITESPACE
-            )
-        set(HIP_RUNTIME ${_hip_runtime} CACHE STRING "HIP runtime as computed by hipconfig")
-        mark_as_advanced(HIP_RUNTIME)
-    endif()
-endif()
-
-include(FindPackageHandleStandardArgs)
-find_package_handle_standard_args(
-    HIP
-    REQUIRED_VARS
-    HIP_ROOT_DIR
-    HIP_HIPCC_EXECUTABLE
-    HIP_HIPCONFIG_EXECUTABLE
-    HIP_PLATFORM
-    HIP_COMPILER
-    HIP_RUNTIME
-    VERSION_VAR HIP_VERSION
-    )
-
-###############################################################################
-# Set HIP CMAKE Flags
-###############################################################################
-# Copy the invocation styles from CXX to HIP
-set(CMAKE_HIP_ARCHIVE_CREATE ${CMAKE_CXX_ARCHIVE_CREATE})
-set(CMAKE_HIP_ARCHIVE_APPEND ${CMAKE_CXX_ARCHIVE_APPEND})
-set(CMAKE_HIP_ARCHIVE_FINISH ${CMAKE_CXX_ARCHIVE_FINISH})
-set(CMAKE_SHARED_LIBRARY_SONAME_HIP_FLAG ${CMAKE_SHARED_LIBRARY_SONAME_CXX_FLAG})
-set(CMAKE_SHARED_LIBRARY_CREATE_HIP_FLAGS ${CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS})
-set(CMAKE_SHARED_LIBRARY_HIP_FLAGS ${CMAKE_SHARED_LIBRARY_CXX_FLAGS})
-#set(CMAKE_SHARED_LIBRARY_LINK_HIP_FLAGS ${CMAKE_SHARED_LIBRARY_LINK_CXX_FLAGS})
-set(CMAKE_SHARED_LIBRARY_RUNTIME_HIP_FLAG ${CMAKE_SHARED_LIBRARY_RUNTIME_CXX_FLAG})
-set(CMAKE_SHARED_LIBRARY_RUNTIME_HIP_FLAG_SEP ${CMAKE_SHARED_LIBRARY_RUNTIME_CXX_FLAG_SEP})
-set(CMAKE_SHARED_LIBRARY_LINK_STATIC_HIP_FLAGS ${CMAKE_SHARED_LIBRARY_LINK_STATIC_CXX_FLAGS})
-set(CMAKE_SHARED_LIBRARY_LINK_DYNAMIC_HIP_FLAGS ${CMAKE_SHARED_LIBRARY_LINK_DYNAMIC_CXX_FLAGS})
-
-set(HIP_CLANG_PARALLEL_BUILD_COMPILE_OPTIONS "")
-set(HIP_CLANG_PARALLEL_BUILD_LINK_OPTIONS "")
-
-if("${HIP_COMPILER}" STREQUAL "nvcc")
-    # Set the CMake Flags to use the nvcc Compiler.
-    set(CMAKE_HIP_CREATE_SHARED_LIBRARY "${HIP_HIPCC_CMAKE_LINKER_HELPER} <CMAKE_SHARED_LIBRARY_CXX_FLAGS> <LANGUAGE_COMPILE_FLAGS> <LINK_FLAGS> <CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS> <SONAME_FLAG><TARGET_SONAME> -o <TARGET> <OBJECTS> <LINK_LIBRARIES>")
-    set(CMAKE_HIP_CREATE_SHARED_MODULE "${HIP_HIPCC_CMAKE_LINKER_HELPER} <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> <SONAME_FLAG><TARGET_SONAME> -o <TARGET> <LINK_LIBRARIES> -shared" )
-    set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_CMAKE_LINKER_HELPER} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>")
-elseif("${HIP_COMPILER}" STREQUAL "hcc")
-    # Set the CMake Flags to use the hcc Compiler.
-    set(CMAKE_HIP_CREATE_SHARED_LIBRARY "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HCC_HOME} <CMAKE_SHARED_LIBRARY_CXX_FLAGS> <LANGUAGE_COMPILE_FLAGS> <LINK_FLAGS> <CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS> <SONAME_FLAG><TARGET_SONAME> -o <TARGET> <OBJECTS> <LINK_LIBRARIES>")
-    set(CMAKE_HIP_CREATE_SHARED_MODULE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HCC_HOME} <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> <SONAME_FLAG><TARGET_SONAME> -o <TARGET> <LINK_LIBRARIES> -shared" )
-    set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HCC_HOME} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>")
-elseif("${HIP_COMPILER}" STREQUAL "clang")
-    #Number of parallel jobs by default is 1
-    if(NOT DEFINED HIP_CLANG_NUM_PARALLEL_JOBS)
-      set(HIP_CLANG_NUM_PARALLEL_JOBS 1)
-    endif()
-    #Add support for parallel build and link
-    if(${CMAKE_CXX_COMPILER_ID} STREQUAL "Clang")
-      check_cxx_compiler_flag("-parallel-jobs=1" HIP_CLANG_SUPPORTS_PARALLEL_JOBS)
-    endif()
-    if(HIP_CLANG_NUM_PARALLEL_JOBS GREATER 1)
-      if(${HIP_CLANG_SUPPORTS_PARALLEL_JOBS})
-        set(HIP_CLANG_PARALLEL_BUILD_COMPILE_OPTIONS "-Wno-format-nonliteral -parallel-jobs=${HIP_CLANG_NUM_PARALLEL_JOBS}")
-        set(HIP_CLANG_PARALLEL_BUILD_LINK_OPTIONS "-parallel-jobs=${HIP_CLANG_NUM_PARALLEL_JOBS}")
-      else()
-        message("clang compiler doesn't support parallel jobs")
-      endif()
-    endif()
-
-    # Set the CMake Flags to use the HIP-Clang Compiler.
-    set(CMAKE_HIP_CREATE_SHARED_LIBRARY "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HIP_CLANG_PATH} ${HIP_CLANG_PARALLEL_BUILD_LINK_OPTIONS} <CMAKE_SHARED_LIBRARY_CXX_FLAGS> <LANGUAGE_COMPILE_FLAGS> <LINK_FLAGS> <CMAKE_SHARED_LIBRARY_CREATE_CXX_FLAGS> <SONAME_FLAG><TARGET_SONAME> -o <TARGET> <OBJECTS> <LINK_LIBRARIES>")
-    set(CMAKE_HIP_CREATE_SHARED_MODULE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HIP_CLANG_PATH} ${HIP_CLANG_PARALLEL_BUILD_LINK_OPTIONS} <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> <SONAME_FLAG><TARGET_SONAME> -o <TARGET> <LINK_LIBRARIES> -shared" )
-    set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HIP_CLANG_PATH} ${HIP_CLANG_PARALLEL_BUILD_LINK_OPTIONS} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>")
-
-    if("${HIP_RUNTIME}" STREQUAL "rocclr")
-      if(TARGET host)
-        message(STATUS "host interface - found")
-        set(HIP_HOST_INTERFACE host)
-      endif()
-    endif()
-endif()
-
-###############################################################################
-# MACRO: Locate helper files
-###############################################################################
-macro(HIP_FIND_HELPER_FILE _name _extension)
-    set(_hip_full_name "${_name}.${_extension}")
-    get_filename_component(CMAKE_CURRENT_LIST_DIR "${CMAKE_CURRENT_LIST_FILE}" PATH)
-    set(HIP_${_name} "${CMAKE_CURRENT_LIST_DIR}/FindHIP/${_hip_full_name}")
-    if(NOT EXISTS "${HIP_${_name}}")
-        set(error_message "${_hip_full_name} not found in ${CMAKE_CURRENT_LIST_DIR}/FindHIP")
-        if(HIP_FIND_REQUIRED)
-            message(FATAL_ERROR "${error_message}")
-        else()
-            if(NOT HIP_FIND_QUIETLY)
-                message(STATUS "${error_message}")
-            endif()
-        endif()
-    endif()
-    # Set this variable as internal, so the user isn't bugged with it.
-    set(HIP_${_name} ${HIP_${_name}} CACHE INTERNAL "Location of ${_full_name}" FORCE)
-endmacro()
-
-###############################################################################
-hip_find_helper_file(run_make2cmake cmake)
-hip_find_helper_file(run_hipcc cmake)
-###############################################################################
-
-###############################################################################
-# MACRO: Reset compiler flags
-###############################################################################
-macro(HIP_RESET_FLAGS)
-    unset(HIP_HIPCC_FLAGS)
-    unset(HIP_HCC_FLAGS)
-    unset(HIP_CLANG_FLAGS)
-    unset(HIP_NVCC_FLAGS)
-    foreach(config ${_hip_configuration_types})
-        string(TOUPPER ${config} config_upper)
-        unset(HIP_HIPCC_FLAGS_${config_upper})
-        unset(HIP_HCC_FLAGS_${config_upper})
-        unset(HIP_CLANG_FLAGS_${config_upper})
-        unset(HIP_NVCC_FLAGS_${config_upper})
-    endforeach()
-endmacro()
-
-###############################################################################
-# MACRO: Separate the options from the sources
-###############################################################################
-macro(HIP_GET_SOURCES_AND_OPTIONS _sources _cmake_options _hipcc_options _hcc_options _clang_options _nvcc_options)
-    set(${_sources})
-    set(${_cmake_options})
-    set(${_hipcc_options})
-    set(${_hcc_options})
-    set(${_clang_options})
-    set(${_nvcc_options})
-    set(_hipcc_found_options FALSE)
-    set(_hcc_found_options FALSE)
-    set(_clang_found_options FALSE)
-    set(_nvcc_found_options FALSE)
-    foreach(arg ${ARGN})
-        if("x${arg}" STREQUAL "xHIPCC_OPTIONS")
-            set(_hipcc_found_options TRUE)
-            set(_hcc_found_options FALSE)
-            set(_clang_found_options FALSE)
-            set(_nvcc_found_options FALSE)
-        elseif("x${arg}" STREQUAL "xHCC_OPTIONS")
-            set(_hipcc_found_options FALSE)
-            set(_hcc_found_options TRUE)
-            set(_clang_found_options FALSE)
-            set(_nvcc_found_options FALSE)
-        elseif("x${arg}" STREQUAL "xCLANG_OPTIONS")
-            set(_hipcc_found_options FALSE)
-            set(_hcc_found_options FALSE)
-            set(_clang_found_options TRUE)
-            set(_nvcc_found_options FALSE)
-        elseif("x${arg}" STREQUAL "xNVCC_OPTIONS")
-            set(_hipcc_found_options FALSE)
-            set(_hcc_found_options FALSE)
-            set(_clang_found_options FALSE)
-            set(_nvcc_found_options TRUE)
-        elseif(
-                "x${arg}" STREQUAL "xEXCLUDE_FROM_ALL" OR
-                "x${arg}" STREQUAL "xSTATIC" OR
-                "x${arg}" STREQUAL "xSHARED" OR
-                "x${arg}" STREQUAL "xMODULE"
-                )
-            list(APPEND ${_cmake_options} ${arg})
-        else()
-            if(_hipcc_found_options)
-                list(APPEND ${_hipcc_options} ${arg})
-            elseif(_hcc_found_options)
-                list(APPEND ${_hcc_options} ${arg})
-            elseif(_clang_found_options)
-                list(APPEND ${_clang_options} ${arg})
-            elseif(_nvcc_found_options)
-                list(APPEND ${_nvcc_options} ${arg})
-            else()
-                # Assume this is a file
-                list(APPEND ${_sources} ${arg})
-            endif()
-        endif()
-    endforeach()
-endmacro()
-
-###############################################################################
-# MACRO: Add include directories to pass to the hipcc command
-###############################################################################
-set(HIP_HIPCC_INCLUDE_ARGS_USER "")
-macro(HIP_INCLUDE_DIRECTORIES)
-    foreach(dir ${ARGN})
-        list(APPEND HIP_HIPCC_INCLUDE_ARGS_USER $<$<BOOL:${dir}>:-I${dir}>)
-    endforeach()
-endmacro()
-
-###############################################################################
-# FUNCTION: Helper to avoid clashes of files with the same basename but different paths
-###############################################################################
-function(HIP_COMPUTE_BUILD_PATH path build_path)
-    # Convert to cmake style paths
-    file(TO_CMAKE_PATH "${path}" bpath)
-    if(IS_ABSOLUTE "${bpath}")
-        string(FIND "${bpath}" "${CMAKE_CURRENT_BINARY_DIR}" _binary_dir_pos)
-        if(_binary_dir_pos EQUAL 0)
-            file(RELATIVE_PATH bpath "${CMAKE_CURRENT_BINARY_DIR}" "${bpath}")
-        else()
-            file(RELATIVE_PATH bpath "${CMAKE_CURRENT_SOURCE_DIR}" "${bpath}")
-        endif()
-    endif()
-
-    # Remove leading /
-    string(REGEX REPLACE "^[/]+" "" bpath "${bpath}")
-    # Avoid absolute paths by removing ':'
-    string(REPLACE ":" "_" bpath "${bpath}")
-    # Avoid relative paths that go up the tree
-    string(REPLACE "../" "__/" bpath "${bpath}")
-    # Avoid spaces
-    string(REPLACE " " "_" bpath "${bpath}")
-    # Strip off the filename
-    get_filename_component(bpath "${bpath}" PATH)
-
-    set(${build_path} "${bpath}" PARENT_SCOPE)
-endfunction()
-
-###############################################################################
-# MACRO: Parse OPTIONS from ARGN & set variables prefixed by _option_prefix
-###############################################################################
-macro(HIP_PARSE_HIPCC_OPTIONS _option_prefix)
-    set(_hip_found_config)
-    foreach(arg ${ARGN})
-        # Determine if we are dealing with a per-configuration flag
-        foreach(config ${_hip_configuration_types})
-            string(TOUPPER ${config} config_upper)
-            if(arg STREQUAL "${config_upper}")
-                set(_hip_found_config _${arg})
-                # Clear arg to prevent it from being processed anymore
-                set(arg)
-            endif()
-        endforeach()
-        if(arg)
-            list(APPEND ${_option_prefix}${_hip_found_config} "${arg}")
-        endif()
-    endforeach()
-endmacro()
-
-###############################################################################
-# MACRO: Try and include dependency file if it exists
-###############################################################################
-macro(HIP_INCLUDE_HIPCC_DEPENDENCIES dependency_file)
-    set(HIP_HIPCC_DEPEND)
-    set(HIP_HIPCC_DEPEND_REGENERATE FALSE)
-
-    # Create the dependency file if it doesn't exist
-    if(NOT EXISTS ${dependency_file})
-        file(WRITE ${dependency_file} "# Generated by: FindHIP.cmake. Do not edit.\n")
-    endif()
-    # Include the dependency file
-    include(${dependency_file})
-
-    # Verify the existence of all the included files
-    if(HIP_HIPCC_DEPEND)
-        foreach(f ${HIP_HIPCC_DEPEND})
-            if(NOT EXISTS ${f})
-                # If they aren't there, regenerate the file again
-                set(HIP_HIPCC_DEPEND_REGENERATE TRUE)
-            endif()
-        endforeach()
-    else()
-        # No dependencies, so regenerate the file
-        set(HIP_HIPCC_DEPEND_REGENERATE TRUE)
-    endif()
-
-    # Regenerate the dependency file if needed
-    if(HIP_HIPCC_DEPEND_REGENERATE)
-        set(HIP_HIPCC_DEPEND ${dependency_file})
-        file(WRITE ${dependency_file} "# Generated by: FindHIP.cmake. Do not edit.\n")
-    endif()
-endmacro()
-
-###############################################################################
-# MACRO: Prepare cmake commands for the target
-###############################################################################
-macro(HIP_PREPARE_TARGET_COMMANDS _target _format _generated_files _source_files)
-    set(_hip_flags "")
-    string(TOUPPER "${CMAKE_BUILD_TYPE}" _hip_build_configuration)
-    if(HIP_HOST_COMPILATION_CPP)
-        set(HIP_C_OR_CXX CXX)
-    else()
-        set(HIP_C_OR_CXX C)
-    endif()
-    set(generated_extension ${CMAKE_${HIP_C_OR_CXX}_OUTPUT_EXTENSION})
-
-    # Initialize list of includes with those specified by the user. Append with
-    # ones specified to cmake directly.
-    set(HIP_HIPCC_INCLUDE_ARGS ${HIP_HIPCC_INCLUDE_ARGS_USER})
-
-    # Add the include directories
-    set(include_directories_generator "$<TARGET_PROPERTY:${_target},INCLUDE_DIRECTORIES>")
-    list(APPEND HIP_HIPCC_INCLUDE_ARGS "$<$<BOOL:${include_directories_generator}>:-I$<JOIN:${include_directories_generator}, -I>>")
-
-    get_directory_property(_hip_include_directories INCLUDE_DIRECTORIES)
-    list(REMOVE_DUPLICATES _hip_include_directories)
-    if(_hip_include_directories)
-        foreach(dir ${_hip_include_directories})
-            list(APPEND HIP_HIPCC_INCLUDE_ARGS $<$<BOOL:${dir}>:-I${dir}>)
-        endforeach()
-    endif()
-
-    HIP_GET_SOURCES_AND_OPTIONS(_hip_sources _hip_cmake_options _hipcc_options _hcc_options _clang_options _nvcc_options ${ARGN})
-    HIP_PARSE_HIPCC_OPTIONS(HIP_HIPCC_FLAGS ${_hipcc_options})
-    HIP_PARSE_HIPCC_OPTIONS(HIP_HCC_FLAGS ${_hcc_options})
-    HIP_PARSE_HIPCC_OPTIONS(HIP_CLANG_FLAGS ${_clang_options})
-    HIP_PARSE_HIPCC_OPTIONS(HIP_NVCC_FLAGS ${_nvcc_options})
-
-    # Add the compile definitions
-    set(compile_definition_generator "$<TARGET_PROPERTY:${_target},COMPILE_DEFINITIONS>")
-    list(APPEND HIP_HIPCC_FLAGS "$<$<BOOL:${compile_definition_generator}>:-D$<JOIN:${compile_definition_generator}, -D>>")
-
-    # Check if we are building shared library.
-    set(_hip_build_shared_libs FALSE)
-    list(FIND _hip_cmake_options SHARED _hip_found_SHARED)
-    list(FIND _hip_cmake_options MODULE _hip_found_MODULE)
-    if(_hip_found_SHARED GREATER -1 OR _hip_found_MODULE GREATER -1)
-        set(_hip_build_shared_libs TRUE)
-    endif()
-    list(FIND _hip_cmake_options STATIC _hip_found_STATIC)
-    if(_hip_found_STATIC GREATER -1)
-        set(_hip_build_shared_libs FALSE)
-    endif()
-
-    # If we are building a shared library, add extra flags to HIP_HIPCC_FLAGS
-    if(_hip_build_shared_libs)
-        list(APPEND HIP_HCC_FLAGS "-fPIC")
-        list(APPEND HIP_CLANG_FLAGS "-fPIC")
-        list(APPEND HIP_NVCC_FLAGS "--shared -Xcompiler '-fPIC'")
-    endif()
-
-    # Set host compiler
-    set(HIP_HOST_COMPILER "${CMAKE_${HIP_C_OR_CXX}_COMPILER}")
-
-    # Set compiler flags
-    set(_HIP_HOST_FLAGS "set(CMAKE_HOST_FLAGS ${CMAKE_${HIP_C_OR_CXX}_FLAGS})")
-    set(_HIP_HIPCC_FLAGS "set(HIP_HIPCC_FLAGS ${HIP_HIPCC_FLAGS})")
-    set(_HIP_HCC_FLAGS "set(HIP_HCC_FLAGS ${HIP_HCC_FLAGS})")
-    set(_HIP_CLANG_FLAGS "set(HIP_CLANG_FLAGS ${HIP_CLANG_FLAGS})")
-    set(_HIP_NVCC_FLAGS "set(HIP_NVCC_FLAGS ${HIP_NVCC_FLAGS})")
-    foreach(config ${_hip_configuration_types})
-        string(TOUPPER ${config} config_upper)
-        set(_HIP_HOST_FLAGS "${_HIP_HOST_FLAGS}\nset(CMAKE_HOST_FLAGS_${config_upper} ${CMAKE_${HIP_C_OR_CXX}_FLAGS_${config_upper}})")
-        set(_HIP_HIPCC_FLAGS "${_HIP_HIPCC_FLAGS}\nset(HIP_HIPCC_FLAGS_${config_upper} ${HIP_HIPCC_FLAGS_${config_upper}})")
-        set(_HIP_HCC_FLAGS "${_HIP_HCC_FLAGS}\nset(HIP_HCC_FLAGS_${config_upper} ${HIP_HCC_FLAGS_${config_upper}})")
-        set(_HIP_CLANG_FLAGS "${_HIP_CLANG_FLAGS}\nset(HIP_CLANG_FLAGS_${config_upper} ${HIP_CLANG_FLAGS_${config_upper}})")
-        set(_HIP_NVCC_FLAGS "${_HIP_NVCC_FLAGS}\nset(HIP_NVCC_FLAGS_${config_upper} ${HIP_NVCC_FLAGS_${config_upper}})")
-    endforeach()
-
-    # Reset the output variable
-    set(_hip_generated_files "")
-    set(_hip_source_files "")
-
-    # Iterate over all arguments and create custom commands for all source files
-    foreach(file ${ARGN})
-        # Ignore any file marked as a HEADER_FILE_ONLY
-        get_source_file_property(_is_header ${file} HEADER_FILE_ONLY)
-        # Allow per source file overrides of the format. Also allows compiling non .cu files.
-        get_source_file_property(_hip_source_format ${file} HIP_SOURCE_PROPERTY_FORMAT)
-        if((${file} MATCHES "\\.cu$" OR _hip_source_format) AND NOT _is_header)
-            set(host_flag FALSE)
-        else()
-            set(host_flag TRUE)
-        endif()
-
-        if(NOT host_flag)
-            # Determine output directory
-            HIP_COMPUTE_BUILD_PATH("${file}" hip_build_path)
-            set(hip_compile_output_dir "${CMAKE_CURRENT_BINARY_DIR}/CMakeFiles/${_target}.dir/${hip_build_path}")
-
-            get_filename_component(basename ${file} NAME)
-            set(generated_file_path "${hip_compile_output_dir}/${CMAKE_CFG_INTDIR}")
-            set(generated_file_basename "${_target}_generated_${basename}${generated_extension}")
-
-            # Set file names
-            set(generated_file "${generated_file_path}/${generated_file_basename}")
-            set(cmake_dependency_file "${hip_compile_output_dir}/${generated_file_basename}.depend")
-            set(custom_target_script_pregen "${hip_compile_output_dir}/${generated_file_basename}.cmake.pre-gen")
-            set(custom_target_script "${hip_compile_output_dir}/${generated_file_basename}.cmake")
-
-            # Set properties for object files
-            set_source_files_properties("${generated_file}"
-                PROPERTIES
-                EXTERNAL_OBJECT true # This is an object file not to be compiled, but only be linked
-                )
-
-            # Don't add CMAKE_CURRENT_SOURCE_DIR if the path is already an absolute path
-            get_filename_component(file_path "${file}" PATH)
-            if(IS_ABSOLUTE "${file_path}")
-                set(source_file "${file}")
-            else()
-                set(source_file "${CMAKE_CURRENT_SOURCE_DIR}/${file}")
-            endif()
-
-            # Bring in the dependencies
-            HIP_INCLUDE_HIPCC_DEPENDENCIES(${cmake_dependency_file})
-
-            # Configure the build script
-            configure_file("${HIP_run_hipcc}" "${custom_target_script_pregen}" @ONLY)
-            file(GENERATE
-                OUTPUT "${custom_target_script}"
-                INPUT "${custom_target_script_pregen}"
-                )
-            set(main_dep DEPENDS ${source_file})
-            if(CMAKE_GENERATOR MATCHES "Makefiles")
-                set(verbose_output "$(VERBOSE)")
-            elseif(HIP_VERBOSE_BUILD)
-                set(verbose_output ON)
-            else()
-                set(verbose_output OFF)
-            endif()
-
-            # Create up the comment string
-            file(RELATIVE_PATH generated_file_relative_path "${CMAKE_BINARY_DIR}" "${generated_file}")
-            set(hip_build_comment_string "Building HIPCC object ${generated_file_relative_path}")
-
-            # Build the generated file and dependency file
-            add_custom_command(
-                OUTPUT ${generated_file}
-                # These output files depend on the source_file and the contents of cmake_dependency_file
-                ${main_dep}
-                DEPENDS ${HIP_HIPCC_DEPEND}
-                DEPENDS ${custom_target_script}
-                # Make sure the output directory exists before trying to write to it.
-                COMMAND ${CMAKE_COMMAND} -E make_directory "${generated_file_path}"
-                COMMAND ${CMAKE_COMMAND} ARGS
-                -D verbose:BOOL=${verbose_output}
-                -D build_configuration:STRING=${_hip_build_configuration}
-                -D "generated_file:STRING=${generated_file}"
-                -P "${custom_target_script}"
-                WORKING_DIRECTORY "${hip_compile_output_dir}"
-                COMMENT "${hip_build_comment_string}"
-                )
-
-            # Make sure the build system knows the file is generated
-            set_source_files_properties(${generated_file} PROPERTIES GENERATED TRUE)
-            list(APPEND _hip_generated_files ${generated_file})
-            list(APPEND _hip_source_files ${file})
-        endif()
-    endforeach()
-
-    # Set the return parameter
-    set(${_generated_files} ${_hip_generated_files})
-    set(${_source_files} ${_hip_source_files})
-endmacro()
-
-###############################################################################
-# HIP_ADD_EXECUTABLE
-###############################################################################
-macro(HIP_ADD_EXECUTABLE hip_target)
-    # Separate the sources from the options
-    HIP_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _hipcc_options _hcc_options _clang_options _nvcc_options ${ARGN})
-    HIP_PREPARE_TARGET_COMMANDS(${hip_target} OBJ _generated_files _source_files ${_sources} HIPCC_OPTIONS ${_hipcc_options} HCC_OPTIONS ${_hcc_options} CLANG_OPTIONS ${_clang_options} NVCC_OPTIONS ${_nvcc_options})
-    if(_source_files)
-        list(REMOVE_ITEM _sources ${_source_files})
-    endif()
-    if("${HIP_COMPILER}" STREQUAL "hcc")
-        if("x${HCC_HOME}" STREQUAL "x")
-            if (DEFINED ENV{ROCM_PATH})
-                set(HCC_HOME "$ENV{ROCM_PATH}/hcc")
-            elseif(DEFINED ENV{HIP_PATH})
-                set(HCC_HOME "$ENV{HIP_PATH}/../hcc")
-            else()
-                set(HCC_HOME "/opt/rocm/hcc")
-            endif()
-        endif()
-        set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HCC_HOME} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>")
-    elseif("${HIP_COMPILER}" STREQUAL "clang")
-        if("x${HIP_CLANG_PATH}" STREQUAL "x")
-            if(DEFINED ENV{HIP_CLANG_PATH})
-                set(HIP_CLANG_PATH $ENV{HIP_CLANG_PATH})
-            elseif(DEFINED ENV{ROCM_PATH})
-                set(HIP_CLANG_PATH "$ENV{ROCM_PATH}/llvm/bin")
-            elseif(DEFINED ENV{HIP_PATH})
-                set(HIP_CLANG_PATH "$ENV{HIP_PATH}/../llvm/bin")
-            else()
-                set(HIP_CLANG_PATH "/opt/rocm/llvm/bin")
-            endif()
-        endif()
-        set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_CMAKE_LINKER_HELPER} ${HIP_CLANG_PATH} ${HIP_CLANG_PARALLEL_BUILD_LINK_OPTIONS} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>")
-    else()
-        set(CMAKE_HIP_LINK_EXECUTABLE "${HIP_HIPCC_CMAKE_LINKER_HELPER} <FLAGS> <CMAKE_CXX_LINK_FLAGS> <LINK_FLAGS> <OBJECTS> -o <TARGET> <LINK_LIBRARIES>")
-    endif()
-    if ("${_sources}" STREQUAL "")
-        add_executable(${hip_target} ${_cmake_options} ${_generated_files} "")
-    else()
-        add_executable(${hip_target} ${_cmake_options} ${_generated_files} ${_sources})
-    endif()
-    set_target_properties(${hip_target} PROPERTIES LINKER_LANGUAGE HIP)
-    # Link with host
-    if (HIP_HOST_INTERFACE)
-        # hip rt should be rocclr, compiler should be clang
-        target_link_libraries(${hip_target} ${HIP_HOST_INTERFACE})
-    endif()
-endmacro()
-
-###############################################################################
-# HIP_ADD_LIBRARY
-###############################################################################
-macro(HIP_ADD_LIBRARY hip_target)
-    # Separate the sources from the options
-    HIP_GET_SOURCES_AND_OPTIONS(_sources _cmake_options _hipcc_options _hcc_options _clang_options _nvcc_options ${ARGN})
-    HIP_PREPARE_TARGET_COMMANDS(${hip_target} OBJ _generated_files _source_files ${_sources} ${_cmake_options} HIPCC_OPTIONS ${_hipcc_options} HCC_OPTIONS ${_hcc_options} CLANG_OPTIONS ${_clang_options} NVCC_OPTIONS ${_nvcc_options})
-    if(_source_files)
-        list(REMOVE_ITEM _sources ${_source_files})
-    endif()
-    if ("${_sources}" STREQUAL "")
-        add_library(${hip_target} ${_cmake_options} ${_generated_files} "")
-    else()
-        add_library(${hip_target} ${_cmake_options} ${_generated_files} ${_sources})
-    endif()
-    set_target_properties(${hip_target} PROPERTIES LINKER_LANGUAGE ${HIP_C_OR_CXX})
-    # Link with host
-    if (HIP_HOST_INTERFACE)
-        # hip rt should be rocclr, compiler should be clang
-        target_link_libraries(${hip_target} ${HIP_HOST_INTERFACE})
-    endif()
-endmacro()
-
-# vim: ts=4:sw=4:expandtab:smartindent

config/cmake/modules/FindHIP/run_hipcc.cmake

@@ -1,182 +0,0 @@
-###############################################################################
-# Runs commands using HIPCC
-###############################################################################
-
-###############################################################################
-# This file runs the hipcc commands to produce the desired output file
-# along with the dependency file needed by CMake to compute dependencies.
-#
-# Input variables:
-#
-# verbose:BOOL=<>               OFF: Be as quiet as possible (default)
-#                               ON : Describe each step
-# build_configuration:STRING=<> Build configuration. Defaults to Debug.
-# generated_file:STRING=<>      File to generate. Mandatory argument.
-
-if(NOT build_configuration)
-    set(build_configuration Debug)
-endif()
-if(NOT generated_file)
-    message(FATAL_ERROR "You must specify generated_file on the command line")
-endif()
-
-# Set these up as variables to make reading the generated file easier
-set(HIP_HIPCC_EXECUTABLE "@HIP_HIPCC_EXECUTABLE@") # path
-set(HIP_HIPCONFIG_EXECUTABLE "@HIP_HIPCONFIG_EXECUTABLE@") #path
-set(HIP_HOST_COMPILER "@HIP_HOST_COMPILER@") # path
-set(CMAKE_COMMAND "@CMAKE_COMMAND@") # path
-set(HIP_run_make2cmake "@HIP_run_make2cmake@") # path
-set(HCC_HOME "@HCC_HOME@") #path
-set(HIP_CLANG_PATH "@HIP_CLANG_PATH@") #path
-set(HIP_CLANG_PARALLEL_BUILD_COMPILE_OPTIONS "@HIP_CLANG_PARALLEL_BUILD_COMPILE_OPTIONS@")
-
-@HIP_HOST_FLAGS@
-@_HIP_HIPCC_FLAGS@
-@_HIP_HCC_FLAGS@
-@_HIP_CLANG_FLAGS@
-@_HIP_NVCC_FLAGS@
-#Needed to bring the HIP_HIPCC_INCLUDE_ARGS variable in scope
-set(HIP_HIPCC_INCLUDE_ARGS @HIP_HIPCC_INCLUDE_ARGS@) # list
-
-set(cmake_dependency_file "@cmake_dependency_file@") # path
-set(source_file "@source_file@") # path
-set(host_flag "@host_flag@") # bool
-
-# Determine compiler and compiler flags
-execute_process(COMMAND ${HIP_HIPCONFIG_EXECUTABLE} --platform OUTPUT_VARIABLE HIP_PLATFORM OUTPUT_STRIP_TRAILING_WHITESPACE)
-execute_process(COMMAND ${HIP_HIPCONFIG_EXECUTABLE} --compiler OUTPUT_VARIABLE HIP_COMPILER OUTPUT_STRIP_TRAILING_WHITESPACE)
-execute_process(COMMAND ${HIP_HIPCONFIG_EXECUTABLE} --runtime OUTPUT_VARIABLE HIP_RUNTIME OUTPUT_STRIP_TRAILING_WHITESPACE)
-if(NOT host_flag)
-    set(__CC ${HIP_HIPCC_EXECUTABLE})
-    if("${HIP_PLATFORM}" STREQUAL "amd")
-        if("${HIP_COMPILER}" STREQUAL "hcc")
-            if(NOT "x${HCC_HOME}" STREQUAL "x")
-                set(ENV{HCC_HOME} ${HCC_HOME})
-            endif()
-            set(__CC_FLAGS ${HIP_HIPCC_FLAGS} ${HIP_HCC_FLAGS} ${HIP_HIPCC_FLAGS_${build_configuration}} ${HIP_HCC_FLAGS_${build_configuration}})
-        elseif("${HIP_COMPILER}" STREQUAL "clang")
-            if(NOT "x${HIP_CLANG_PATH}" STREQUAL "x")
-                set(ENV{HIP_CLANG_PATH} ${HIP_CLANG_PATH})
-            endif()
-            # Temporarily include HIP_HCC_FLAGS for HIP-Clang for PyTorch builds
-            set(__CC_FLAGS ${HIP_CLANG_PARALLEL_BUILD_COMPILE_OPTIONS} ${HIP_HIPCC_FLAGS} ${HIP_HCC_FLAGS} ${HIP_CLANG_FLAGS} ${HIP_HIPCC_FLAGS_${build_configuration}} ${HIP_HCC_FLAGS_${build_configuration}} ${HIP_CLANG_FLAGS_${build_configuration}})
-        endif()
-    else()
-        set(__CC_FLAGS ${HIP_HIPCC_FLAGS} ${HIP_NVCC_FLAGS} ${HIP_HIPCC_FLAGS_${build_configuration}} ${HIP_NVCC_FLAGS_${build_configuration}})
-    endif()
-else()
-    set(__CC ${HIP_HOST_COMPILER})
-    set(__CC_FLAGS ${CMAKE_HOST_FLAGS} ${CMAKE_HOST_FLAGS_${build_configuration}})
-endif()
-set(__CC_INCLUDES ${HIP_HIPCC_INCLUDE_ARGS})
-
-# hip_execute_process - Executes a command with optional command echo and status message.
-#   status     - Status message to print if verbose is true
-#   command    - COMMAND argument from the usual execute_process argument structure
-#   ARGN       - Remaining arguments are the command with arguments
-#   HIP_result - Return value from running the command
-macro(hip_execute_process status command)
-    set(_command ${command})
-    if(NOT "x${_command}" STREQUAL "xCOMMAND")
-        message(FATAL_ERROR "Malformed call to hip_execute_process.  Missing COMMAND as second argument. (command = ${command})")
-    endif()
-    if(verbose)
-        execute_process(COMMAND "${CMAKE_COMMAND}" -E echo -- ${status})
-        # Build command string to print
-        set(hip_execute_process_string)
-        foreach(arg ${ARGN})
-            # Escape quotes if any
-            string(REPLACE "\"" "\\\"" arg ${arg})
-            # Surround args with spaces with quotes
-            if(arg MATCHES " ")
-                list(APPEND hip_execute_process_string "\"${arg}\"")
-            else()
-                list(APPEND hip_execute_process_string ${arg})
-            endif()
-        endforeach()
-        # Echo the command
-        execute_process(COMMAND ${CMAKE_COMMAND} -E echo ${hip_execute_process_string})
-    endif()
-    # Run the command
-    execute_process(COMMAND ${ARGN} RESULT_VARIABLE HIP_result)
-endmacro()
-
-# Delete the target file
-hip_execute_process(
-    "Removing ${generated_file}"
-    COMMAND "${CMAKE_COMMAND}" -E remove "${generated_file}"
-    )
-
-# Generate the dependency file
-hip_execute_process(
-    "Generating dependency file: ${cmake_dependency_file}.pre"
-    COMMAND "${__CC}"
-    -M
-    "${source_file}"
-    -o "${cmake_dependency_file}.pre"
-    ${__CC_FLAGS}
-    ${__CC_INCLUDES}
-    )
-
-if(HIP_result)
-    message(FATAL_ERROR "Error generating ${generated_file}")
-endif()
-
-# Generate the cmake readable dependency file to a temp file
-hip_execute_process(
-    "Generating temporary cmake readable file: ${cmake_dependency_file}.tmp"
-    COMMAND "${CMAKE_COMMAND}"
-    -D "input_file:FILEPATH=${cmake_dependency_file}.pre"
-    -D "output_file:FILEPATH=${cmake_dependency_file}.tmp"
-    -D "verbose=${verbose}"
-    -P "${HIP_run_make2cmake}"
-    )
-
-if(HIP_result)
-    message(FATAL_ERROR "Error generating ${generated_file}")
-endif()
-
-# Copy the file if it is different
-hip_execute_process(
-    "Copy if different ${cmake_dependency_file}.tmp to ${cmake_dependency_file}"
-    COMMAND "${CMAKE_COMMAND}" -E copy_if_different "${cmake_dependency_file}.tmp" "${cmake_dependency_file}"
-    )
-
-if(HIP_result)
-    message(FATAL_ERROR "Error generating ${generated_file}")
-endif()
-
-# Delete the temporary file
-hip_execute_process(
-    "Removing ${cmake_dependency_file}.tmp and ${cmake_dependency_file}.pre"
-    COMMAND "${CMAKE_COMMAND}" -E remove "${cmake_dependency_file}.tmp" "${cmake_dependency_file}.pre"
-    )
-
-if(HIP_result)
-    message(FATAL_ERROR "Error generating ${generated_file}")
-endif()
-
-# Generate the output file
-hip_execute_process(
-    "Generating ${generated_file}"
-    COMMAND "${__CC}"
-    -c
-    "${source_file}"
-    -o "${generated_file}"
-    ${__CC_FLAGS}
-    ${__CC_INCLUDES}
-    )
-
-if(HIP_result)
-    # Make sure that we delete the output file
-    hip_execute_process(
-        "Removing ${generated_file}"
-        COMMAND "${CMAKE_COMMAND}" -E remove "${generated_file}"
-        )
-    message(FATAL_ERROR "Error generating file ${generated_file}")
-else()
-    if(verbose)
-        message("Generated ${generated_file} successfully.")
-    endif()
-endif()
-# vim: ts=4:sw=4:expandtab:smartindent

config/cmake/modules/FindHIP/run_make2cmake.cmake

@@ -1,50 +0,0 @@
-###############################################################################
-# Computes dependencies using HIPCC
-###############################################################################
-
-###############################################################################
-# This file converts dependency files generated using hipcc to a format that
-# cmake can understand.
-
-# Input variables:
-#
-# input_file:STRING=<> Dependency file to parse. Required argument
-# output_file:STRING=<> Output file to generate. Required argument
-
-if(NOT input_file OR NOT output_file)
-    message(FATAL_ERROR "You must specify input_file and output_file on the command line")
-endif()
-
-file(READ ${input_file} depend_text)
-
-if (NOT "${depend_text}" STREQUAL "")
-    string(REPLACE " /" "\n/" depend_text ${depend_text})
-    string(REGEX REPLACE "^.*:" "" depend_text ${depend_text})
-    string(REGEX REPLACE "[ \\\\]*\n" ";" depend_text ${depend_text})
-
-    set(dependency_list "")
-
-    foreach(file ${depend_text})
-        string(REGEX REPLACE "^ +" "" file ${file})
-        if(NOT EXISTS "${file}")
-            message(WARNING " Removing non-existent dependency file: ${file}")
-            set(file "")
-        endif()
-
-        if(NOT IS_DIRECTORY "${file}")
-            get_filename_component(file_absolute "${file}" ABSOLUTE)
-            list(APPEND dependency_list "${file_absolute}")
-        endif()
-    endforeach()
-endif()
-
-# Remove the duplicate entries and sort them.
-list(REMOVE_DUPLICATES dependency_list)
-list(SORT dependency_list)
-
-foreach(file ${dependency_list})
-    set(hip_hipcc_depend "${hip_hipcc_depend} \"${file}\"\n")
-endforeach()
-
-file(WRITE ${output_file} "# Generated by: FindHIP.cmake. Do not edit.\nSET(HIP_HIPCC_DEPEND\n ${hip_hipcc_depend})\n\n")
-# vim: ts=4:sw=4:expandtab:smartindent

config/cmake/modules/FindHYPRE.cmake

@@ -14,10 +14,33 @@
 #   - HYPRE_LIBRARIES
 #   - HYPRE_INCLUDE_DIRS
 #   - HYPRE_VERSION
+#   - HYPRE_USING_HIP (internal)
+
+if (HYPRE_FOUND)
+  if (HYPRE_USING_HIP)
+    find_package(rocsparse REQUIRED)
+    find_package(rocrand REQUIRED)
+  endif()
+  return()
+endif()
 
 include(MfemCmakeUtilities)
 mfem_find_package(HYPRE HYPRE HYPRE_DIR "include" "HYPRE.h" "lib" "HYPRE"
-  "Paths to headers required by HYPRE." "Libraries required by HYPRE.")
+  "Paths to headers required by HYPRE." "Libraries required by HYPRE."
+  CHECK_BUILD HYPRE_USING_HIP FALSE
+  "
+#undef HYPRE_USING_HIP
+#include <HYPRE_config.h>
+
+#ifndef HYPRE_USING_HIP
+#error HYPRE is built without HIP.
+#endif
+
+int main()
+{
+   return 0;
+}
+")
 
 if (HYPRE_FOUND AND (NOT HYPRE_VERSION))
   try_run(HYPRE_VERSION_RUN_RESULT HYPRE_VERSION_COMPILE_RESULT
@@ -33,3 +56,12 @@ if (HYPRE_FOUND AND (NOT HYPRE_VERSION))
     message(FATAL_ERROR "Unable to determine HYPRE version.")
   endif()
 endif()
+
+if (HYPRE_FOUND AND HYPRE_USING_HIP)
+  find_package(rocsparse REQUIRED)
+  find_package(rocrand REQUIRED)
+  list(APPEND HYPRE_LIBRARIES ${rocsparse_LIBRARIES} ${rocrand_LIBRARIES})
+  set(HYPRE_LIBRARIES ${HYPRE_LIBRARIES} CACHE STRING
+      "HYPRE libraries + dependencies." FORCE)
+  message(STATUS "Updated HYPRE_LIBRARIES: ${HYPRE_LIBRARIES}")
+endif()

config/cmake/modules/MfemCmakeUtilities.cmake

@@ -46,8 +46,7 @@ endfunction()
 # Wrapper for add_executable that calls the HIP wrapper if applicable
 macro(mfem_add_executable NAME)
   if (MFEM_USE_HIP)
-    hip_add_executable(${NAME} ${ARGN})
-    set_target_properties(${NAME} PROPERTIES LINKER_LANGUAGE CXX)
+    add_executable(${NAME} ${ARGN})
   else()
     add_executable(${NAME} ${ARGN})
   endif()
@@ -56,7 +55,7 @@ endmacro()
 # Wrapper for add_library that calls the HIP wrapper if applicable
 macro(mfem_add_library NAME)
   if (MFEM_USE_HIP)
-    hip_add_library(${NAME} ${ARGN})
+    add_library(${NAME} ${ARGN})
   else()
     add_library(${NAME} ${ARGN})
   endif()
@@ -92,8 +91,6 @@ macro(add_mfem_examples EXE_SRCS)
     # If CUDA is enabled, tag source files to be compiled with nvcc.
     if (MFEM_USE_CUDA)
       set_source_files_properties(${SRC_FILE} PROPERTIES LANGUAGE CUDA)
-    elseif(MFEM_USE_HIP)
-      set_source_files_properties(${SRC_FILE} PROPERTIES HIP_SOURCE_PROPERTY_FORMAT TRUE)
     endif()
 
     get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
@@ -159,8 +156,6 @@ macro(add_mfem_miniapp MFEM_EXE_NAME)
       endforeach()
       set(EXTRA_OPTIONS_LIST ${LIST_})
     endif()
-  elseif(MFEM_USE_HIP)
-    set_source_files_properties(${MAIN_LIST} ${EXTRA_SOURCES_LIST} PROPERTIES HIP_SOURCE_PROPERTY_FORMAT TRUE)
   endif()
 
   # Actually add the executable
@@ -534,12 +529,15 @@ function(mfem_find_package Name Prefix DirVar IncSuffixes Header LibSuffixes
             if (NOT ImportConfig)
               set(ImportConfig RELEASE)
             endif()
+            set(ImportConfigSuffix "_${ImportConfig}")
             get_target_property(ImpConfigs ${TargetName} IMPORTED_CONFIGURATIONS)
             list(FIND ImpConfigs ${ImportConfig} _Index)
-            if (_Index EQUAL -1)
-              message(FATAL_ERROR " *** ${ReqPack}: configuration "
-                "${ImportConfig} not found. Set ${ReqPack}_IMPORT_CONFIG "
-                "from the list: ${ImpConfigs}.")
+            if ((_Index EQUAL -1) OR ("${ImportConfig}" STREQUAL "NO_CONFIG"))
+              set(ImportConfig "NO_CONFIG")
+              set(ImportConfigSuffix "")
+              # message(FATAL_ERROR " *** ${ReqPack}: configuration "
+              #   "${ImportConfig} not found. Set ${ReqPack}_IMPORT_CONFIG "
+              #   "from the list: ${ImpConfigs}.")
             endif()
           endif()
           # Set _Pack_LIBS
@@ -551,8 +549,8 @@ function(mfem_find_package Name Prefix DirVar IncSuffixes Header LibSuffixes
             endif()
           else()
             # Set _Pack_LIBS from the target properties for ImportConfig
-            foreach (_prop IMPORTED_LOCATION_${ImportConfig}
-                IMPORTED_LINK_INTERFACE_LIBRARIES_${ImportConfig})
+            foreach (_prop IMPORTED_LOCATION${ImportConfigSuffix}
+                IMPORTED_LINK_INTERFACE_LIBRARIES${ImportConfigSuffix})
               get_target_property(_value ${TargetName} ${_prop})
               if (_value)
                 list(APPEND _Pack_LIBS ${_value})
@@ -564,7 +562,7 @@ function(mfem_find_package Name Prefix DirVar IncSuffixes Header LibSuffixes
             endif()
           endif()
           # Set _Pack_INCS
-          foreach (_prop INCLUDE_DIRECTORIES)
+          foreach (_prop INCLUDE_DIRECTORIES INTERFACE_INCLUDE_DIRECTORIES)
             get_target_property(_value ${TargetName} ${_prop})
             if (_value)
               list(APPEND _Pack_INCS ${_value})
@@ -742,6 +740,133 @@ function(mfem_find_library Name Prefix Lib LibDoc CheckVar CheckSrc)
 endfunction(mfem_find_library)
 
 
+#
+# Extract compile and link options needed by the given target.
+#
+function(mfem_get_target_options Target CompileOptsVar LinkOptsVar)
+
+  if (NOT TARGET ${Target})
+    return()
+  endif()
+
+  # CMAKE_SHARED_LIBRARY_RUNTIME_C_FLAG -> '-Wl,-rpath,'
+  set(shared_link_flag ${CMAKE_SHARED_LIBRARY_RUNTIME_C_FLAG})
+  if (NOT shared_link_flag)
+    set(shared_link_flag "-Wl,-rpath,")
+  endif()
+
+  set(tgt "${Target}")
+  unset(CompileOpts)
+  unset(LinkOpts)
+  get_target_property(IsImported ${tgt} IMPORTED)
+  # message(STATUS "${tgt}[IMPORTED]: ${IsImported}")
+  # Generally, the possible target types are: STATIC_LIBRARY, MODULE_LIBRARY,
+  # SHARED_LIBRARY, INTERFACE_LIBRARY, EXECUTABLE.
+  get_target_property(type ${tgt} TYPE)
+  # message(STATUS "${tgt}[TYPE]: ${type}")
+  unset(ImportConfig)
+  get_target_property(ImportConfigs ${tgt} IMPORTED_CONFIGURATIONS)
+  if (ImportConfigs)
+    list(GET ImportConfigs 0 ImportConfig)
+  endif()
+  if (NOT ImportConfig)
+    set(ImportConfig RELEASE)
+  endif()
+  # message(STATUS "${tgt}[ImportConfig]: ${ImportConfig}")
+  # List all properties with: cmake --help-property-list
+  get_target_property(Defs ${tgt} INTERFACE_COMPILE_DEFINITIONS)
+  if (Defs)
+    list(REMOVE_DUPLICATES Defs)
+    foreach(Def ${Defs})
+      list(APPEND CompileOpts "-D${Def}")
+    endforeach()
+  endif()
+  get_target_property(Opts ${tgt} INTERFACE_COMPILE_OPTIONS)
+  if (Opts)
+    foreach(Opt ${Opts})
+      list(APPEND CompileOpts "${Opt}")
+    endforeach()
+  endif()
+  get_target_property(Dirs ${tgt} INTERFACE_INCLUDE_DIRECTORIES)
+  if (Dirs)
+    list(REMOVE_DUPLICATES Dirs)
+    foreach(Dir ${Dirs})
+      list(APPEND CompileOpts "-I\"${Dir}\"")
+    endforeach()
+  endif()
+  get_target_property(SysDirs ${tgt} INTERFACE_SYSTEM_INCLUDE_DIRECTORIES)
+  if (SysDirs)
+    list(REMOVE_DUPLICATES SysDirs)
+    foreach(SysDir ${SysDirs})
+      list(APPEND CompileOpts "-isystem \"${SysDir}\"")
+    endforeach()
+  endif()
+  if ("${type}" STREQUAL "STATIC_LIBRARY")
+    get_target_property(Location ${tgt} LOCATION)
+    if (Location)
+      list(APPEND LinkOpts "\"${Location}\"")
+    else()
+      message(STATUS " *** Warning: [${tgt}] LOCATION not defined!")
+    endif()
+  elseif ("${type}" STREQUAL "SHARED_LIBRARY")
+    get_target_property(Location ${tgt} LOCATION)
+    if (Location)
+      get_filename_component(Dir ${Location} DIRECTORY)
+      get_filename_component(NameWE ${Location} NAME_WE)
+      string(REGEX REPLACE "^lib" "" LibName ${NameWE})
+      list(APPEND LinkOpts
+        "-L\"${Dir}\""
+        "${shared_link_flag}\"${Dir}\""
+        "-l${LibName}")
+    else()
+      message(STATUS " *** Warning: [${tgt}] LOCATION not defined!")
+    endif()
+  elseif ("${type}" STREQUAL "INTERFACE_LIBRARY")
+    get_target_property(Libs ${tgt} INTERFACE_LINK_LIBRARIES)
+    if (Libs)
+      foreach(Lib ${Libs})
+        if (NOT (TARGET ${Lib}))
+          list(APPEND LinkOpts "${Lib}")
+        else()
+          mfem_get_target_options(${Lib} COpts LOpts)
+          list(APPEND CompileOpts ${COpts})
+          list(APPEND LinkOpts ${LOpts})
+        endif()
+      endforeach()
+    endif()
+    # Other properties we may need to handle:
+    # INTERFACE_LINK_DEPENDS
+    # INTERFACE_LINK_DIRECTORIES
+    # INTERFACE_LINK_OPTIONS
+  else()
+    message(STATUS " *** Warning: [${tgt}] uses target type '${type}'"
+      " which is not supported!")
+  endif()
+
+  # Other potentially relevant properties:
+  # - For all target types:
+  # IMPORTED_LIBNAME
+  # IMPORTED_LIBNAME_${ImportConfig}
+  # INTERFACE_AUTOUIC_OPTIONS
+  # INTERFACE_COMPILE_FEATURES
+  # INTERFACE_POSITION_INDEPENDENT_CODE
+  # INTERFACE_SOURCES
+  # INTERFACE_SYSTEM_INCLUDE_DIRECTORIES)
+  # - For non-"INTERFACE_LIBRARY" target types only:
+  # IMPORTED_LOCATION
+  # IMPORTED_LOCATION_${ImportConfig}
+  # IMPORTED_LINK_INTERFACE_LIBRARIES
+  # IMPORTED_LINK_INTERFACE_LIBRARIES_${ImportConfig}
+  # LINK_FLAGS
+  # LINK_FLAGS_${ImportConfig}
+  # LOCATION_${ImportConfig})
+
+  set(${CompileOptsVar} "${CompileOpts}" PARENT_SCOPE)
+  set(${LinkOptsVar} "${LinkOpts}" PARENT_SCOPE)
+
+endfunction(mfem_get_target_options)
+
+
 #
 #   Function that creates 'config.mk' from 'config.mk.in' for the both the
 #   build- and the install-locations and define install rules for 'config.mk'
@@ -760,13 +885,15 @@ function(mfem_export_mk_files)
   # Convert Boolean vars to YES/NO without writing the values to cache
   set(CONFIG_MK_BOOL_VARS MFEM_USE_MPI MFEM_USE_METIS MFEM_USE_METIS_5
       MFEM_DEBUG MFEM_USE_EXCEPTIONS MFEM_USE_ZLIB MFEM_USE_LIBUNWIND
-      MFEM_USE_LAPACK MFEM_THREAD_SAFE MFEM_USE_OPENMP MFEM_USE_LEGACY_OPENMP
+      MFEM_USE_LAPACK MFEM_THREAD_SAFE MFEM_USE_LEGACY_OPENMP MFEM_USE_OPENMP
       MFEM_USE_MEMALLOC MFEM_USE_SUNDIALS MFEM_USE_MESQUITE MFEM_USE_SUITESPARSE
-      MFEM_USE_SUPERLU MFEM_USE_STRUMPACK MFEM_USE_GINKGO MFEM_USE_AMGX
-      MFEM_USE_GNUTLS MFEM_USE_GSLIB MFEM_USE_NETCDF MFEM_USE_PETSC
-      MFEM_USE_SLEPC MFEM_USE_MPFR MFEM_USE_SIDRE MFEM_USE_CONDUIT MFEM_USE_PUMI
-      MFEM_USE_CUDA MFEM_USE_OCCA MFEM_USE_RAJA MFEM_USE_UMPIRE MFEM_USE_SIMD
-      MFEM_USE_ADIOS2 MFEM_USE_BENCHMARK MFEM_USE_PARELAG)
+      MFEM_USE_SUPERLU MFEM_USE_SUPERLU5 MFEM_USE_MUMPS MFEM_USE_STRUMPACK
+      MFEM_USE_GINKGO MFEM_USE_AMGX MFEM_USE_GNUTLS MFEM_USE_NETCDF
+      MFEM_USE_PETSC MFEM_USE_SLEPC MFEM_USE_MPFR MFEM_USE_SIDRE MFEM_USE_FMS
+      MFEM_USE_CONDUIT MFEM_USE_PUMI MFEM_USE_HIOP MFEM_USE_GSLIB MFEM_USE_CUDA
+      MFEM_USE_HIP MFEM_USE_RAJA MFEM_USE_OCCA MFEM_USE_CEED MFEM_USE_CALIPER
+      MFEM_USE_UMPIRE MFEM_USE_SIMD MFEM_USE_ADIOS2 MFEM_USE_MKL_CPARDISO
+      MFEM_USE_ADFORWARD MFEM_USE_CODIPACK MFEM_USE_BENCHMARK MFEM_USE_PARELAG)
   foreach(var ${CONFIG_MK_BOOL_VARS})
     if (${var})
       set(${var} YES)
@@ -868,8 +995,18 @@ function(mfem_export_mk_files)
     get_filename_component(suffix ${lib} EXT)
     # handle interfaces (e.g., SCOREC::apf)
     if ("${lib}" MATCHES "SCOREC::.*" OR "${lib}" MATCHES "Ginkgo::.*")
-    elseif (NOT "${lib}" MATCHES "SCOREC::.*" AND "${lib}" MATCHES ".*::.*")
-      message(FATAL_ERROR "***** interface lib found ... exiting *****")
+    elseif (TARGET "${lib}")
+      mfem_get_target_options(${lib} CompileOpts LinkOpts)
+      # Removing duplicates may lead to issues:
+      # list(REMOVE_DUPLICATES CompileOpts)
+      # list(REMOVE_DUPLICATES LinkOpts)
+      string(REPLACE ";" " " COpts "${CompileOpts}")
+      string(REPLACE ";" " " LOpts "${LinkOpts}")
+      # message(STATUS "${lib}[COpts]: '${COpts}'")
+      # message(STATUS "${lib}[LOpts]: '${LOpts}'")
+      set(MFEM_TPLFLAGS "${MFEM_TPLFLAGS} ${COpts}")
+      set(MFEM_EXT_LIBS "${MFEM_EXT_LIBS} ${LOpts}")
+      # message(FATAL_ERROR "***** interface lib found ... exiting *****")
       # handle static and shared libs
     elseif ("${suffix}" STREQUAL "${CMAKE_SHARED_LIBRARY_SUFFIX}")
       get_filename_component(dir ${lib} DIRECTORY)

config/config.mk.in

@@ -91,7 +91,7 @@ MFEM_MPIEXEC_NP = @MFEM_MPIEXEC_NP@
 MFEM_MPI_NP     = @MFEM_MPI_NP@
 
 # The NVCC compiler cannot link with -x=cu
-MFEM_LINK_FLAGS := $(filter-out -x=cu, $(MFEM_FLAGS))
+MFEM_LINK_FLAGS := $(filter-out -x=cu -xhip, $(MFEM_FLAGS))
 
 # Optional extra configuration
 @MFEM_CONFIG_EXTRA@

config/defaults.cmake

@@ -50,6 +50,7 @@ option(MFEM_USE_CONDUIT "Enable Conduit usage" OFF)
 option(MFEM_USE_PUMI "Enable PUMI" OFF)
 option(MFEM_USE_HIOP "Enable HiOp" OFF)
 option(MFEM_USE_CUDA "Enable CUDA" OFF)
+option(MFEM_USE_HIP "Enable HIP" OFF)
 option(MFEM_USE_OCCA "Enable OCCA" OFF)
 option(MFEM_USE_RAJA "Enable RAJA" OFF)
 option(MFEM_USE_CEED "Enable CEED" OFF)
@@ -107,6 +108,7 @@ if (MFEM_USE_CUDA)
    set(HYPRE_REQUIRED_LIBRARIES "-lcusparse" "-lcurand" CACHE STRING
        "Libraries that HYPRE depends on.")
 endif()
+# HIP dependency for HYPRE is handled in FindHYPRE.cmake.
 
 set(METIS_DIR "${MFEM_DIR}/../metis-4.0" CACHE PATH "Path to the METIS library.")
 
@@ -226,6 +228,11 @@ set(MKL_LIBRARY_DIR "" CACHE STRING "Custom library subdirectory")
 
 set(OCCA_DIR "${MFEM_DIR}/../occa" CACHE PATH "Path to OCCA")
 set(RAJA_DIR "${MFEM_DIR}/../raja" CACHE PATH "Path to RAJA")
+# If RAJA is built with external CAMP:
+# set(RAJA_REQUIRED_PACKAGES "camp"
+#     CACHE STRING "Packages that RAJA depends on.")
+# set(camp_DIR "${MFEM_DIR}/../camp/lib/cmake/camp"
+#     CACHE PATH "Path to CAMP CMake files.")
 set(CEED_DIR "${MFEM_DIR}/../libCEED" CACHE PATH "Path to libCEED")
 set(UMPIRE_DIR "${MFEM_DIR}/../umpire" CACHE PATH "Path to Umpire")
 set(CALIPER_DIR "${MFEM_DIR}/../caliper" CACHE PATH "Path to Caliper")
@@ -245,7 +252,8 @@ set(BLAS_LIBRARIES "" CACHE STRING "The BLAS library.")
 set(LAPACK_INCLUDE_DIRS "" CACHE STRING "Path to LAPACK headers.")
 set(LAPACK_LIBRARIES "" CACHE STRING "The LAPACK library.")
 
-set(CODIPACK_INCLUDE_DIRS "${MFEM_DIR}/../CoDiPack/inlude" CACHE STRING "Path to CoDiPack headers.")
+set(CODIPACK_INCLUDE_DIRS "${MFEM_DIR}/../CoDiPack/include" CACHE STRING
+    "Path to CoDiPack headers.")
 set(CODIPACK_LIBRARIES "")
 
 # Some useful variables:

config/defaults.mk

@@ -172,6 +172,20 @@ ifeq ($(MFEM_USE_MPI)$(MFEM_USE_HIP),YESYES)
    MPI_LIB = -L$(MPI_DIR)/lib $(XLINKER)-rpath,$(MPI_DIR)/lib -lmpi
 endif
 
+# ROCM/HIP directory such that ROCM/HIP libraries like rocsparse and rocrand are
+# found in $(HIP_DIR)/lib, usually as links. Typically, this directoory is of
+# the form /opt/rocm-X.Y.Z which is called ROCM_PATH by hipconfig.
+ifeq ($(MFEM_USE_HIP),YES)
+   HIP_DIR := $(patsubst %/,%,$(dir $(shell which $(HIP_CXX))))
+   HIP_DIR := $(patsubst %/,%,$(dir $(HIP_DIR)))
+   ifeq (,$(wildcard $(HIP_DIR)/lib/librocsparse.*))
+      HIP_DIR := $(shell hipconfig --rocmpath 2> /dev/null)
+      ifeq (,$(wildcard $(HIP_DIR)/lib/librocsparse.*))
+         $(error Unable to determine HIP_DIR. Please set it manually.)
+      endif
+   endif
+endif
+
 # Compile and link options for zlib.
 ZLIB_DIR =
 ZLIB_OPT = $(if $(ZLIB_DIR),-I$(ZLIB_DIR)/include)
@@ -189,6 +203,11 @@ ifeq (YES,$(MFEM_USE_CUDA))
    # This is only necessary when hypre is built with cuda:
    HYPRE_LIB += -lcusparse -lcurand
 endif
+ifeq (YES,$(MFEM_USE_HIP))
+   # This is only necessary when hypre is built with hip:
+   HYPRE_LIB += -L$(HIP_DIR)/lib $(XLINKER)-rpath,$(HIP_DIR)/lib\
+ -lrocsparse -lrocrand
+endif
 
 # METIS library configuration
 ifeq ($(MFEM_USE_SUPERLU)$(MFEM_USE_STRUMPACK)$(MFEM_USE_MUMPS),NONONO)
@@ -430,9 +449,9 @@ GSLIB_LIB = -L$(GSLIB_DIR)/lib -lgs
 CUDA_OPT =
 CUDA_LIB = -lcusparse
 
-# HIP library configuration (currently not needed)
+# HIP library configuration
 HIP_OPT =
-HIP_LIB =
+HIP_LIB = -L$(HIP_DIR)/lib $(XLINKER)-rpath,$(HIP_DIR)/lib -lhipsparse
 
 # OCCA library configuration
 OCCA_DIR = @MFEM_DIR@/../occa

config/test.mk

@@ -58,14 +58,14 @@ ifneq (,$(filter test%,$(MAKECMDGOALS)))
    MAKEFLAGS += -k
 endif
 # Test runs of the examples/miniapps with parameters - check exit code:
-# 0 means success, 255 means the test was skipped, anything else means error
+# 0 means success, 242 means the test was skipped, anything else means error
 mfem-test = \
    printf "   $(3) [$(2) $(1) ... ]: "; \
    $(call $(TIMEFUN),$(TIMECMD),$(2) ./$(1) $(if $(5),,-no-vis )$(4) \
      > $(1).stderr 2>&1); \
    err="$$3"; \
    if [ "$$3" = 0 ]; then $(PRINT_OK); \
-   else if [ "$$3" = 255 ]; then $(PRINT_SKIP); err=0; \
+   else if [ "$$3" = 242 ]; then $(PRINT_SKIP); err=0; \
    else $(PRINT_FAILED); cat $(1).stderr; fi; fi; \
    rm -f $(1).stderr; exit $$err
 
@@ -76,7 +76,7 @@ mfem-test-file = \
    $(call $(TIMEFUN),$(TIMECMD),$(2) ./$(1) -no-vis > $(1).stderr 2>&1); \
    err="$$3"; \
    if [ "$$3" = 0 ] && [ -e $(4) ]; then $(PRINT_OK); \
-   else if [ "$$3" = 255 ] && [ -e $(4) ]; then $(PRINT_SKIP); err=0; \
+   else if [ "$$3" = 242 ] && [ -e $(4) ]; then $(PRINT_SKIP); err=0; \
    else $(PRINT_FAILED); cat $(1).stderr; err=64; fi; fi; \
    rm -f $(1).stderr; exit $$err
 

examples/ex10.cpp

@@ -149,7 +149,7 @@ void InitialDeformation(const Vector &x, Vector &y);
 
 void InitialVelocity(const Vector &x, Vector &v);
 
-void visualize(ostream &out, Mesh *mesh, GridFunction *deformed_nodes,
+void visualize(ostream &os, Mesh *mesh, GridFunction *deformed_nodes,
                GridFunction *field, const char *field_name = NULL,
                bool init_vis = false);
 
@@ -376,10 +376,10 @@ int main(int argc, char *argv[])
 }
 
 
-void visualize(ostream &out, Mesh *mesh, GridFunction *deformed_nodes,
+void visualize(ostream &os, Mesh *mesh, GridFunction *deformed_nodes,
                GridFunction *field, const char *field_name, bool init_vis)
 {
-   if (!out)
+   if (!os)
    {
       return;
    }
@@ -389,24 +389,25 @@ void visualize(ostream &out, Mesh *mesh, GridFunction *deformed_nodes,
 
    mesh->SwapNodes(nodes, owns_nodes);
 
-   out << "solution\n" << *mesh << *field;
+   os << "solution\n" << *mesh << *field;
 
    mesh->SwapNodes(nodes, owns_nodes);
 
    if (init_vis)
    {
-      out << "window_size 800 800\n";
-      out << "window_title '" << field_name << "'\n";
+      os << "window_size 800 800\n";
+      os << "window_title '" << field_name << "'\n";
       if (mesh->SpaceDimension() == 2)
       {
-         out << "view 0 0\n"; // view from top
-         out << "keys jl\n";  // turn off perspective and light
+         os << "view 0 0\n"; // view from top
+         os << "keys jl\n";  // turn off perspective and light
       }
-      out << "keys cm\n";         // show colorbar and mesh
-      out << "autoscale value\n"; // update value-range; keep mesh-extents fixed
-      out << "pause\n";
+      os << "keys cm\n";         // show colorbar and mesh
+      // update value-range; keep mesh-extents fixed
+      os << "autoscale value\n";
+      os << "pause\n";
    }
-   out << flush;
+   os << flush;
 }
 
 

examples/ex10p.cpp

@@ -154,7 +154,8 @@ void InitialDeformation(const Vector &x, Vector &y);
 
 void InitialVelocity(const Vector &x, Vector &v);
 
-void visualize(ostream &out, ParMesh *mesh, ParGridFunction *deformed_nodes,
+void visualize(ostream &os, ParMesh *mesh,
+               ParGridFunction *deformed_nodes,
                ParGridFunction *field, const char *field_name = NULL,
                bool init_vis = false);
 
@@ -438,10 +439,11 @@ int main(int argc, char *argv[])
    return 0;
 }
 
-void visualize(ostream &out, ParMesh *mesh, ParGridFunction *deformed_nodes,
+void visualize(ostream &os, ParMesh *mesh,
+               ParGridFunction *deformed_nodes,
                ParGridFunction *field, const char *field_name, bool init_vis)
 {
-   if (!out)
+   if (!os)
    {
       return;
    }
@@ -451,25 +453,27 @@ void visualize(ostream &out, ParMesh *mesh, ParGridFunction *deformed_nodes,
 
    mesh->SwapNodes(nodes, owns_nodes);
 
-   out << "parallel " << mesh->GetNRanks() << " " << mesh->GetMyRank() << "\n";
-   out << "solution\n" << *mesh << *field;
+   os << "parallel " << mesh->GetNRanks()
+      << " " << mesh->GetMyRank() << "\n";
+   os << "solution\n" << *mesh << *field;
 
    mesh->SwapNodes(nodes, owns_nodes);
 
    if (init_vis)
    {
-      out << "window_size 800 800\n";
-      out << "window_title '" << field_name << "'\n";
+      os << "window_size 800 800\n";
+      os << "window_title '" << field_name << "'\n";
       if (mesh->SpaceDimension() == 2)
       {
-         out << "view 0 0\n"; // view from top
-         out << "keys jl\n";  // turn off perspective and light
+         os << "view 0 0\n"; // view from top
+         os << "keys jl\n";  // turn off perspective and light
       }
-      out << "keys cm\n";         // show colorbar and mesh
-      out << "autoscale value\n"; // update value-range; keep mesh-extents fixed
-      out << "pause\n";
+      os << "keys cm\n";         // show colorbar and mesh
+      // update value-range; keep mesh-extents fixed
+      os << "autoscale value\n";
+      os << "pause\n";
    }
-   out << flush;
+   os << flush;
 }
 
 

examples/ex18.hpp

@@ -32,7 +32,7 @@ private:
    mutable DenseTensor flux;
    mutable Vector z;
 
-   void GetFlux(const DenseMatrix &state, DenseTensor &flux) const;
+   void GetFlux(const DenseMatrix &state_, DenseTensor &flux_) const;
 
 public:
    FE_Evolution(FiniteElementSpace &vfes_,
@@ -256,26 +256,26 @@ inline double ComputeMaxCharSpeed(const Vector &state, const int dim)
 }
 
 // Compute the flux at solution nodes.
-void FE_Evolution::GetFlux(const DenseMatrix &x, DenseTensor &flux) const
+void FE_Evolution::GetFlux(const DenseMatrix &x_, DenseTensor &flux_) const
 {
-   const int dof = flux.SizeI();
-   const int dim = flux.SizeJ();
+   const int flux_dof = flux_.SizeI();
+   const int flux_dim = flux_.SizeJ();
 
-   for (int i = 0; i < dof; i++)
+   for (int i = 0; i < flux_dof; i++)
    {
-      for (int k = 0; k < num_equation; k++) { state(k) = x(i, k); }
-      ComputeFlux(state, dim, f);
+      for (int k = 0; k < num_equation; k++) { state(k) = x_(i, k); }
+      ComputeFlux(state, flux_dim, f);
 
-      for (int d = 0; d < dim; d++)
+      for (int d = 0; d < flux_dim; d++)
       {
          for (int k = 0; k < num_equation; k++)
          {
-            flux(i, d, k) = f(k, d);
+            flux_(i, d, k) = f(k, d);
          }
       }
 
       // Update max char speed
-      const double mcs = ComputeMaxCharSpeed(state, dim);
+      const double mcs = ComputeMaxCharSpeed(state, flux_dim);
       if (mcs > max_char_speed) { max_char_speed = mcs; }
    }
 }

examples/ex19.cpp

@@ -171,7 +171,7 @@ public:
 };
 
 // Visualization driver
-void visualize(ostream &out, Mesh *mesh, GridFunction *deformed_nodes,
+void visualize(ostream &os, Mesh *mesh, GridFunction *deformed_nodes,
                GridFunction *field, const char *field_name = NULL,
                bool init_vis = false);
 
@@ -542,10 +542,10 @@ RubberOperator::~RubberOperator()
 
 
 // Inline visualization
-void visualize(ostream &out, Mesh *mesh, GridFunction *deformed_nodes,
+void visualize(ostream &os, Mesh *mesh, GridFunction *deformed_nodes,
                GridFunction *field, const char *field_name, bool init_vis)
 {
-   if (!out)
+   if (!os)
    {
       return;
    }
@@ -555,23 +555,25 @@ void visualize(ostream &out, Mesh *mesh, GridFunction *deformed_nodes,
 
    mesh->SwapNodes(nodes, owns_nodes);
 
-   out << "solution\n" << *mesh << *field;
+   os << "solution\n" << *mesh << *field;
 
    mesh->SwapNodes(nodes, owns_nodes);
 
    if (init_vis)
    {
-      out << "window_size 800 800\n";
-      out << "window_title '" << field_name << "'\n";
+      os << "window_size 800 800\n";
+      os << "window_title '" << field_name << "'\n";
       if (mesh->SpaceDimension() == 2)
       {
-         out << "view 0 0\n"; // view from top
-         out << "keys jlA\n"; // turn off perspective and light, +anti-aliasing
+         os << "view 0 0\n"; // view from top
+         // turn off perspective and light, +anti-aliasing
+         os << "keys jlA\n";
       }
-      out << "keys cmA\n";        // show colorbar and mesh, +anti-aliasing
-      out << "autoscale value\n"; // update value-range; keep mesh-extents fixed
+      os << "keys cmA\n"; // show colorbar and mesh, +anti-aliasing
+      // update value-range; keep mesh-extents fixed
+      os << "autoscale value\n";
    }
-   out << flush;
+   os << flush;
 }
 
 void ReferenceConfiguration(const Vector &x, Vector &y)

examples/ex19p.cpp

@@ -185,7 +185,8 @@ public:
 };
 
 // Visualization driver
-void visualize(ostream &out, ParMesh *mesh, ParGridFunction *deformed_nodes,
+void visualize(ostream &os, ParMesh *mesh,
+               ParGridFunction *deformed_nodes,
                ParGridFunction *field, const char *field_name = NULL,
                bool init_vis = false);
 
@@ -196,10 +197,10 @@ void InitialDeformation(const Vector &x, Vector &y);
 
 int main(int argc, char *argv[])
 {
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    cout << "\nAs of mfem-4.3 and hypre-2.22.0 (July 2021) this example\n"
-        << "is NOT supported with the CUDA version of hypre.\n\n";
-   return 255;
+        << "is NOT supported with the GPU version of hypre.\n\n";
+   return 242;
 #endif
 
    // 1. Initialize MPI
@@ -486,8 +487,8 @@ void JacobianPreconditioner::SetOperator(const Operator &op)
 
       if (!spaces[0]->GetParMesh()->Nonconforming())
       {
-#ifndef HYPRE_USING_CUDA
-         // Not available yet when hypre is built with CUDA
+#if !defined(HYPRE_USING_GPU)
+         // Not available yet when hypre is built with GPU support
          stiff_prec_amg->SetElasticityOptions(spaces[0]);
 #endif
       }
@@ -617,10 +618,11 @@ RubberOperator::~RubberOperator()
 
 
 // Inline visualization
-void visualize(ostream &out, ParMesh *mesh, ParGridFunction *deformed_nodes,
+void visualize(ostream &os, ParMesh *mesh,
+               ParGridFunction *deformed_nodes,
                ParGridFunction *field, const char *field_name, bool init_vis)
 {
-   if (!out)
+   if (!os)
    {
       return;
    }
@@ -630,24 +632,27 @@ void visualize(ostream &out, ParMesh *mesh, ParGridFunction *deformed_nodes,
 
    mesh->SwapNodes(nodes, owns_nodes);
 
-   out << "parallel " << mesh->GetNRanks() << " " << mesh->GetMyRank() << "\n";
-   out << "solution\n" << *mesh << *field;
+   os << "parallel " << mesh->GetNRanks() << " " << mesh->GetMyRank() <<
+      "\n";
+   os << "solution\n" << *mesh << *field;
 
    mesh->SwapNodes(nodes, owns_nodes);
 
    if (init_vis)
    {
-      out << "window_size 800 800\n";
-      out << "window_title '" << field_name << "'\n";
+      os << "window_size 800 800\n";
+      os << "window_title '" << field_name << "'\n";
       if (mesh->SpaceDimension() == 2)
       {
-         out << "view 0 0\n"; // view from top
-         out << "keys jlA\n"; // turn off perspective and light, +anti-aliasing
+         os << "view 0 0\n"; // view from top
+         // turn off perspective and light, +anti-aliasing
+         os << "keys jlA\n";
       }
-      out << "keys cmA\n";        // show colorbar and mesh, +anti-aliasing
-      out << "autoscale value\n"; // update value-range; keep mesh-extents fixed
+      os << "keys cmA\n"; // show colorbar and mesh, +anti-aliasing
+      // update value-range; keep mesh-extents fixed
+      os << "autoscale value\n";
    }
-   out << flush;
+   os << flush;
 }
 
 void ReferenceConfiguration(const Vector &x, Vector &y)

examples/ex26.cpp

@@ -34,6 +34,43 @@
 using namespace std;
 using namespace mfem;
 
+MatrixConstantCoefficient AnisotropicCoefficient(int dim, double anisotropy)
+{
+   DenseMatrix coeff(dim, dim);
+   coeff = 0.0;
+   coeff(0,0) = anisotropy;
+   for (int d = 1; d < dim; ++d)
+   {
+      coeff(d, d) = 1.0;
+   }
+   return coeff;
+}
+
+class SymmetricILUSmoother : public Solver
+{
+   BlockILU ilu;
+   double alpha;
+
+public:
+   SymmetricILUSmoother(Operator &op, double alpha_)
+      : ilu(op), alpha(alpha_)
+   { }
+
+   void Mult(const Vector &b, Vector &x) const
+   {
+      ilu.Mult(b, x);
+      x *= alpha;
+   }
+
+   void MultTranspose(const Vector &b, Vector &x) const
+   {
+      ilu.Mult(b, x);
+      x *= alpha;
+   }
+
+   void SetOperator(const Operator &op) { }
+};
+
 // Class for constructing a multigrid preconditioner for the diffusion operator.
 // This example multigrid preconditioner class demonstrates the creation of the
 // diffusion bilinear forms and operators using partial assembly for all spaces
@@ -43,13 +80,18 @@ using namespace mfem;
 class DiffusionMultigrid : public GeometricMultigrid
 {
 private:
-   ConstantCoefficient one;
+   MatrixConstantCoefficient coeff;
+   bool use_ilu;
 
 public:
    // Constructs a diffusion multigrid for the given FiniteElementSpaceHierarchy
    // and the array of essential boundaries
-   DiffusionMultigrid(FiniteElementSpaceHierarchy& fespaces, Array<int>& ess_bdr)
-      : GeometricMultigrid(fespaces), one(1.0)
+   DiffusionMultigrid(FiniteElementSpaceHierarchy& fespaces, Array<int>& ess_bdr,
+                      double anisotropy, bool use_ilu_)
+      : GeometricMultigrid(fespaces),
+        coeff(AnisotropicCoefficient(fespaces.GetFinestFESpace().GetMesh()->Dimension(),
+                                     anisotropy)),
+        use_ilu(use_ilu_)
    {
       ConstructCoarseOperatorAndSolver(fespaces.GetFESpaceAtLevel(0), ess_bdr);
 
@@ -63,8 +105,7 @@ private:
    void ConstructBilinearForm(FiniteElementSpace& fespace, Array<int>& ess_bdr)
    {
       BilinearForm* form = new BilinearForm(&fespace);
-      form->SetAssemblyLevel(AssemblyLevel::PARTIAL);
-      form->AddDomainIntegrator(new DiffusionIntegrator(one));
+      form->AddDomainIntegrator(new DiffusionIntegrator(coeff));
       form->Assemble();
       bfs.Append(form);
 
@@ -78,18 +119,13 @@ private:
       ConstructBilinearForm(coarse_fespace, ess_bdr);
 
       OperatorPtr opr;
-      opr.SetType(Operator::ANY_TYPE);
+      opr.SetType(Operator::MFEM_SPARSEMAT);
       bfs.Last()->FormSystemMatrix(*essentialTrueDofs.Last(), opr);
       opr.SetOperatorOwner(false);
 
-      CGSolver* pcg = new CGSolver();
-      pcg->SetPrintLevel(-1);
-      pcg->SetMaxIter(200);
-      pcg->SetRelTol(sqrt(1e-4));
-      pcg->SetAbsTol(0.0);
-      pcg->SetOperator(*opr.Ptr());
+      UMFPackSolver *coarse_solver = new UMFPackSolver(*opr.As<SparseMatrix>());
 
-      AddLevel(opr.Ptr(), pcg, true, true);
+      AddLevel(opr.Ptr(), coarse_solver, false, true);
    }
 
    void ConstructOperatorAndSmoother(FiniteElementSpace& fespace,
@@ -98,16 +134,25 @@ private:
       ConstructBilinearForm(fespace, ess_bdr);
 
       OperatorPtr opr;
-      opr.SetType(Operator::ANY_TYPE);
+      opr.SetType(Operator::MFEM_SPARSEMAT);
       bfs.Last()->FormSystemMatrix(*essentialTrueDofs.Last(), opr);
       opr.SetOperatorOwner(false);
 
-      Vector diag(fespace.GetTrueVSize());
-      bfs.Last()->AssembleDiagonal(diag);
+      Solver *smoother;
 
-      Solver* smoother = new OperatorChebyshevSmoother(*opr, diag,
-                                                       *essentialTrueDofs.Last(), 2);
-      AddLevel(opr.Ptr(), smoother, true, true);
+      if (use_ilu)
+      {
+         smoother = new SymmetricILUSmoother(*opr, 0.5);
+      }
+      else
+      {
+         Vector diag(fespace.GetTrueVSize());
+         bfs.Last()->AssembleDiagonal(diag);
+         smoother = new OperatorChebyshevSmoother(
+            *opr, diag, *essentialTrueDofs.Last(), 2);
+      }
+
+      AddLevel(opr.Ptr(), smoother, false, true);
    }
 };
 
@@ -120,6 +165,8 @@ int main(int argc, char *argv[])
    int order_refinements = 2;
    const char *device_config = "cpu";
    bool visualization = true;
+   double anisotropy = 1.0;
+   bool use_ilu = false;
 
    OptionsParser args(argc, argv);
    args.AddOption(&mesh_file, "-m", "--mesh",
@@ -128,8 +175,11 @@ int main(int argc, char *argv[])
                   "Number of geometric refinements done prior to order refinements.");
    args.AddOption(&order_refinements, "-or", "--order-refinements",
                   "Number of order refinements. Finest level in the hierarchy has order 2^{or}.");
+   args.AddOption(&anisotropy, "-a", "--anisotropy", "Anisotropy coefficient.");
    args.AddOption(&device_config, "-d", "--device",
                   "Device configuration string, see Device::Configure().");
+   args.AddOption(&use_ilu, "-i", "--use-ilu", "-no-i", "--no-ilu",
+                  "Use ILU smoothing?");
    args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                   "--no-visualization",
                   "Enable or disable GLVis visualization.");
@@ -210,7 +260,7 @@ int main(int argc, char *argv[])
    Array<int> ess_bdr(mesh->bdr_attributes.Max());
    ess_bdr = 1;
 
-   DiffusionMultigrid M(fespaces, ess_bdr);
+   DiffusionMultigrid M(fespaces, ess_bdr, anisotropy, use_ilu);
    M.SetCycleType(Multigrid::CycleType::VCYCLE, 1, 1);
 
    OperatorPtr A;

examples/ex27.cpp

@@ -75,7 +75,7 @@ Mesh * GenerateSerialMesh(int ref);
 // alpha*n.Grad(sol) + beta*sol - gamma over the same boundary.
 double IntegrateBC(const GridFunction &sol, const Array<int> &bdr_marker,
                    double alpha, double beta, double gamma,
-                   double &err);
+                   double &error);
 
 int main(int argc, char *argv[])
 {
@@ -302,26 +302,26 @@ int main(int argc, char *argv[])
    {
       // Integrate the solution on the Dirichlet boundary and compare to the
       // expected value.
-      double err, avg = IntegrateBC(u, dbc_bdr, 0.0, 1.0, dbc_val, err);
+      double error, avg = IntegrateBC(u, dbc_bdr, 0.0, 1.0, dbc_val, error);
 
       bool hom_dbc = (dbc_val == 0.0);
-      err /=  hom_dbc ? 1.0 : fabs(dbc_val);
+      error /=  hom_dbc ? 1.0 : fabs(dbc_val);
       mfem::out << "Average of solution on Gamma_dbc:\t"
                 << avg << ", \t"
                 << (hom_dbc ? "absolute" : "relative")
-                << " error " << err << endl;
+                << " error " << error << endl;
    }
    {
       // Integrate n.Grad(u) on the inhomogeneous Neumann boundary and compare
       // to the expected value.
-      double err, avg = IntegrateBC(u, nbc_bdr, 1.0, 0.0, nbc_val, err);
+      double error, avg = IntegrateBC(u, nbc_bdr, 1.0, 0.0, nbc_val, error);
 
       bool hom_nbc = (nbc_val == 0.0);
-      err /=  hom_nbc ? 1.0 : fabs(nbc_val);
+      error /=  hom_nbc ? 1.0 : fabs(nbc_val);
       mfem::out << "Average of n.Grad(u) on Gamma_nbc:\t"
                 << avg << ", \t"
                 << (hom_nbc ? "absolute" : "relative")
-                << " error " << err << endl;
+                << " error " << error << endl;
    }
    {
       // Integrate n.Grad(u) on the homogeneous Neumann boundary and compare to
@@ -330,25 +330,26 @@ int main(int argc, char *argv[])
       nbc0_bdr = 0;
       nbc0_bdr[3] = 1;
 
-      double err, avg = IntegrateBC(u, nbc0_bdr, 1.0, 0.0, 0.0, err);
+      double error, avg = IntegrateBC(u, nbc0_bdr, 1.0, 0.0, 0.0, error);
 
       bool hom_nbc = true;
       mfem::out << "Average of n.Grad(u) on Gamma_nbc0:\t"
                 << avg << ", \t"
                 << (hom_nbc ? "absolute" : "relative")
-                << " error " << err << endl;
+                << " error " << error << endl;
    }
    {
       // Integrate n.Grad(u) + a * u on the Robin boundary and compare to the
       // expected value.
-      double err, avg = IntegrateBC(u, rbc_bdr, 1.0, rbc_a_val, rbc_b_val, err);
+      double error;
+      double avg = IntegrateBC(u, rbc_bdr, 1.0, rbc_a_val, rbc_b_val, error);
 
       bool hom_rbc = (rbc_b_val == 0.0);
-      err /=  hom_rbc ? 1.0 : fabs(rbc_b_val);
+      error /=  hom_rbc ? 1.0 : fabs(rbc_b_val);
       mfem::out << "Average of n.Grad(u)+a*u on Gamma_rbc:\t"
                 << avg << ", \t"
                 << (hom_rbc ? "absolute" : "relative")
-                << " error " << err << endl;
+                << " error " << error << endl;
    }
 
    // 14. Save the refined mesh and the solution. This output can be viewed
@@ -637,11 +638,11 @@ Mesh * GenerateSerialMesh(int ref)
 
 double IntegrateBC(const GridFunction &x, const Array<int> &bdr,
                    double alpha, double beta, double gamma,
-                   double &err)
+                   double &error)
 {
    double nrm = 0.0;
    double avg = 0.0;
-   err = 0.0;
+   error = 0.0;
 
    const bool a_is_zero = alpha == 0.0;
    const bool b_is_zero = beta == 0.0;
@@ -705,20 +706,20 @@ double IntegrateBC(const GridFunction &x, const Array<int> &bdr,
 
          // Integrate |alpha * n.Grad(x) + beta * x - gamma|^2
          val -= gamma;
-         err += (val*val) * ip.weight * face_weight;
+         error += (val*val) * ip.weight * face_weight;
       }
    }
 
    // Normalize by the length of the boundary
    if (std::abs(nrm) > 0.0)
    {
-      err /= nrm;
+      error /= nrm;
       avg /= nrm;
    }
 
    // Compute l2 norm of the error in the boundary condition (negative
-   // quadrature weights may produce negative 'err')
-   err = (err >= 0.0) ? sqrt(err) : -sqrt(-err);
+   // quadrature weights may produce negative 'error')
+   error = (error >= 0.0) ? sqrt(error) : -sqrt(-error);
 
    // Return the average value of alpha * n.Grad(x) + beta * x
    return avg;

examples/ex27p.cpp

@@ -75,7 +75,7 @@ Mesh * GenerateSerialMesh(int ref);
 // alpha*n.Grad(sol) + beta*sol - gamma over the same boundary.
 double IntegrateBC(const ParGridFunction &sol, const Array<int> &bdr_marker,
                    double alpha, double beta, double gamma,
-                   double &err);
+                   double &error);
 
 int main(int argc, char *argv[])
 {
@@ -321,26 +321,26 @@ int main(int argc, char *argv[])
    {
       // Integrate the solution on the Dirichlet boundary and compare to the
       // expected value.
-      double err, avg = IntegrateBC(u, dbc_bdr, 0.0, 1.0, dbc_val, err);
+      double error, avg = IntegrateBC(u, dbc_bdr, 0.0, 1.0, dbc_val, error);
 
       bool hom_dbc = (dbc_val == 0.0);
-      err /=  hom_dbc ? 1.0 : fabs(dbc_val);
+      error /=  hom_dbc ? 1.0 : fabs(dbc_val);
       mfem::out << "Average of solution on Gamma_dbc:\t"
                 << avg << ", \t"
                 << (hom_dbc ? "absolute" : "relative")
-                << " error " << err << endl;
+                << " error " << error << endl;
    }
    {
       // Integrate n.Grad(u) on the inhomogeneous Neumann boundary and compare
       // to the expected value.
-      double err, avg = IntegrateBC(u, nbc_bdr, 1.0, 0.0, nbc_val, err);
+      double error, avg = IntegrateBC(u, nbc_bdr, 1.0, 0.0, nbc_val, error);
 
       bool hom_nbc = (nbc_val == 0.0);
-      err /=  hom_nbc ? 1.0 : fabs(nbc_val);
+      error /=  hom_nbc ? 1.0 : fabs(nbc_val);
       mfem::out << "Average of n.Grad(u) on Gamma_nbc:\t"
                 << avg << ", \t"
                 << (hom_nbc ? "absolute" : "relative")
-                << " error " << err << endl;
+                << " error " << error << endl;
    }
    {
       // Integrate n.Grad(u) on the homogeneous Neumann boundary and compare to
@@ -349,25 +349,26 @@ int main(int argc, char *argv[])
       nbc0_bdr = 0;
       nbc0_bdr[3] = 1;
 
-      double err, avg = IntegrateBC(u, nbc0_bdr, 1.0, 0.0, 0.0, err);
+      double error, avg = IntegrateBC(u, nbc0_bdr, 1.0, 0.0, 0.0, error);
 
       bool hom_nbc = true;
       mfem::out << "Average of n.Grad(u) on Gamma_nbc0:\t"
                 << avg << ", \t"
                 << (hom_nbc ? "absolute" : "relative")
-                << " error " << err << endl;
+                << " error " << error << endl;
    }
    {
       // Integrate n.Grad(u) + a * u on the Robin boundary and compare to the
       // expected value.
-      double err, avg = IntegrateBC(u, rbc_bdr, 1.0, rbc_a_val, rbc_b_val, err);
+      double error, avg = IntegrateBC(u, rbc_bdr, 1.0, rbc_a_val, rbc_b_val,
+                                      error);
 
       bool hom_rbc = (rbc_b_val == 0.0);
-      err /=  hom_rbc ? 1.0 : fabs(rbc_b_val);
+      error /=  hom_rbc ? 1.0 : fabs(rbc_b_val);
       mfem::out << "Average of n.Grad(u)+a*u on Gamma_rbc:\t"
                 << avg << ", \t"
                 << (hom_rbc ? "absolute" : "relative")
-                << " error " << err << endl;
+                << " error " << error << endl;
    }
 
    // 15. Save the refined mesh and the solution in parallel. This output can be
@@ -667,11 +668,11 @@ double IntegrateBC(const ParGridFunction &x, const Array<int> &bdr,
    double loc_vals[3];
    double &nrm = loc_vals[0];
    double &avg = loc_vals[1];
-   double &err = loc_vals[2];
+   double &error = loc_vals[2];
 
    nrm = 0.0;
    avg = 0.0;
-   err = 0.0;
+   error = 0.0;
 
    const bool a_is_zero = alpha == 0.0;
    const bool b_is_zero = beta == 0.0;
@@ -735,7 +736,7 @@ double IntegrateBC(const ParGridFunction &x, const Array<int> &bdr,
 
          // Integrate |alpha * n.Grad(x) + beta * x - gamma|^2
          val -= gamma;
-         err += (val*val) * ip.weight * face_weight;
+         error += (val*val) * ip.weight * face_weight;
       }
    }
 
@@ -754,7 +755,7 @@ double IntegrateBC(const ParGridFunction &x, const Array<int> &bdr,
    }
 
    // Compute l2 norm of the error in the boundary condition (negative
-   // quadrature weights may produce negative 'err')
+   // quadrature weights may produce negative 'error')
    glb_err = (glb_err >= 0.0) ? sqrt(glb_err) : -sqrt(-glb_err);
 
    // Return the average value of alpha * n.Grad(x) + beta * x

examples/ex28p.cpp

@@ -81,10 +81,10 @@ Mesh * build_trapezoid_mesh(double offset)
 
 int main(int argc, char *argv[])
 {
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    cout << "\nAs of mfem-4.3 and hypre-2.22.0 (July 2021) this example\n"
-        << "is NOT supported with the CUDA version of hypre.\n\n";
-   return 255;
+        << "is NOT supported with the GPU version of hypre.\n\n";
+   return 242;
 #endif
 
    // 1. Initialize MPI.

examples/ex29.cpp

@@ -167,9 +167,9 @@ int main(int argc, char *argv[])
 
    // 13. Compute error in the solution and its flux
    FunctionCoefficient uCoef(uExact);
-   double err = x.ComputeL2Error(uCoef);
+   double error = x.ComputeL2Error(uCoef);
 
-   cout << "|u - u_h|_2 = " << err << endl;
+   cout << "|u - u_h|_2 = " << error << endl;
 
    FiniteElementSpace flux_fespace(mesh, &fec, 3);
    GridFunction flux(&flux_fespace);

examples/ex29p.cpp

@@ -197,9 +197,9 @@ int main(int argc, char *argv[])
 
    // 15. Compute error in the solution and its flux
    FunctionCoefficient uCoef(uExact);
-   double err = x.ComputeL2Error(uCoef);
+   double error = x.ComputeL2Error(uCoef);
 
-   if (myid == 0) { cout << "|u - u_h|_2 = " << err << endl; }
+   if (myid == 0) { cout << "|u - u_h|_2 = " << error << endl; }
 
    ParFiniteElementSpace flux_fespace(&pmesh, &fec, 3);
    ParGridFunction flux(&flux_fespace);

examples/ex3p.cpp

@@ -255,10 +255,10 @@ int main(int argc, char *argv[])
 
    // 15. Compute and print the L^2 norm of the error.
    {
-      double err = x.ComputeL2Error(E);
+      double error = x.ComputeL2Error(E);
       if (myid == 0)
       {
-         cout << "\n|| E_h - E ||_{L^2} = " << err << '\n' << endl;
+         cout << "\n|| E_h - E ||_{L^2} = " << error << '\n' << endl;
       }
    }
 

examples/ex4p.cpp

@@ -256,10 +256,10 @@ int main(int argc, char *argv[])
 
    // 15. Compute and print the L^2 norm of the error.
    {
-      double err = x.ComputeL2Error(F);
+      double error = x.ComputeL2Error(F);
       if (myid == 0)
       {
-         cout << "\n|| F_h - F ||_{L^2} = " << err << '\n' << endl;
+         cout << "\n|| F_h - F ||_{L^2} = " << error << '\n' << endl;
       }
    }
 

examples/ex7p.cpp

@@ -282,10 +282,10 @@ int main(int argc, char *argv[])
    delete b;
 
    // 12. Compute and print the L^2 norm of the error.
-   double err = x.ComputeL2Error(sol_coef);
+   double error = x.ComputeL2Error(sol_coef);
    if (myid == 0)
    {
-      cout << "\nL2 norm of error: " << err << endl;
+      cout << "\nL2 norm of error: " << error << endl;
    }
 
    // 13. Save the refined mesh and the solution. This output can be viewed

examples/pumi/ex1p.cpp

@@ -145,7 +145,7 @@ int main(int argc, char *argv[])
    // Perform Uniform refinement
    if (ref_levels > 1)
    {
-      ma::Input* uniInput = ma::configureUniformRefine(pumi_mesh, ref_levels);
+      auto uniInput = ma::configureUniformRefine(pumi_mesh, ref_levels);
 
       if (geom_order > 1)
       {

examples/pumi/ex6p.cpp

@@ -150,7 +150,7 @@ int main(int argc, char *argv[])
 
    if (ref_levels > 1)
    {
-      ma::Input* uniInput = ma::configureUniformRefine(pumi_mesh, ref_levels);
+      auto uniInput = ma::configureUniformRefine(pumi_mesh, ref_levels);
 
       if ( geom_order > 1)
       {
@@ -345,9 +345,7 @@ int main(int argc, char *argv[])
       apf::destroyField(ipfield);
 
       // 18. Perform MesAdapt.
-      ma::Input* erinput = ma::configure(pumi_mesh, sizefield);
-      erinput->shouldFixShape = true;
-      erinput->maximumIterations = 2;
+      auto erinput = ma::configure(pumi_mesh, sizefield);
       if ( geom_order > 1)
       {
          crv::adapt(erinput);

fem/bilinearform_ext.cpp

@@ -514,6 +514,10 @@ void EABilinearFormExtension::Assemble()
 
    Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
    const int integratorCount = integrators.Size();
+   if ( integratorCount == 0 )
+   {
+      ea_data = 0.0;
+   }
    for (int i = 0; i < integratorCount; ++i)
    {
       integrators[i]->AssembleEA(*a->FESpace(), ea_data, i);

fem/coefficient.cpp

@@ -21,6 +21,33 @@ namespace mfem
 
 using namespace std;
 
+// Given an ElementTransformation and IntegrationPoint in a refined mesh,
+// return the ElementTransformation of the parent coarse element, and set
+// coarse_ip to the location of the original ip within the coarse element.
+ElementTransformation *RefinedToCoarse(
+   Mesh &coarse_mesh, const ElementTransformation &T,
+   const IntegrationPoint &ip, IntegrationPoint &coarse_ip)
+{
+   Mesh &fine_mesh = *T.mesh;
+   // Get the element transformation of the coarse element containing the
+   // fine element.
+   int fine_element = T.ElementNo;
+   const CoarseFineTransformations &cf = fine_mesh.GetRefinementTransforms();
+   int coarse_element = cf.embeddings[fine_element].parent;
+   ElementTransformation *coarse_T = coarse_mesh.GetElementTransformation(
+                                        coarse_element);
+   // Transform the integration point from fine element coordinates to coarse
+   // element coordinates.
+   Geometry::Type geom = T.GetGeometryType();
+   IntegrationPointTransformation fine_to_coarse;
+   IsoparametricTransformation &emb_tr = fine_to_coarse.Transf;
+   emb_tr.SetIdentityTransformation(geom);
+   emb_tr.SetPointMat(cf.point_matrices[geom](cf.embeddings[fine_element].matrix));
+   fine_to_coarse.Transform(ip, coarse_ip);
+   coarse_T->SetIntPoint(&coarse_ip);
+   return coarse_T;
+}
+
 double PWConstCoefficient::Eval(ElementTransformation & T,
                                 const IntegrationPoint & ip)
 {
@@ -95,7 +122,17 @@ double FunctionCoefficient::Eval(ElementTransformation & T,
 double GridFunctionCoefficient::Eval (ElementTransformation &T,
                                       const IntegrationPoint &ip)
 {
-   return GridF -> GetValue (T, ip, Component);
+   Mesh *gf_mesh = GridF->FESpace()->GetMesh();
+   if (T.mesh == gf_mesh)
+   {
+      return GridF->GetValue(T, ip, Component);
+   }
+   else
+   {
+      IntegrationPoint coarse_ip;
+      ElementTransformation *coarse_T = RefinedToCoarse(*gf_mesh, T, ip, coarse_ip);
+      return GridF->GetValue(*coarse_T, coarse_ip, Component);
+   }
 }
 
 void TransformedCoefficient::SetTime(double t)
@@ -305,13 +342,30 @@ void VectorGridFunctionCoefficient::SetGridFunction(const GridFunction *gf)
 void VectorGridFunctionCoefficient::Eval(Vector &V, ElementTransformation &T,
                                          const IntegrationPoint &ip)
 {
-   GridFunc->GetVectorValue(T, ip, V);
+   Mesh *gf_mesh = GridFunc->FESpace()->GetMesh();
+   if (T.mesh == gf_mesh)
+   {
+      GridFunc->GetVectorValue(T, ip, V);
+   }
+   else
+   {
+      IntegrationPoint coarse_ip;
+      ElementTransformation *coarse_T = RefinedToCoarse(*gf_mesh, T, ip, coarse_ip);
+      GridFunc->GetVectorValue(*coarse_T, coarse_ip, V);
+   }
 }
 
 void VectorGridFunctionCoefficient::Eval(
    DenseMatrix &M, ElementTransformation &T, const IntegrationRule &ir)
 {
-   GridFunc->GetVectorValues(T, ir, M);
+   if (T.mesh == GridFunc->FESpace()->GetMesh())
+   {
+      GridFunc->GetVectorValues(T, ir, M);
+   }
+   else
+   {
+      VectorCoefficient::Eval(M, T, ir);
+   }
 }
 
 GradientGridFunctionCoefficient::GradientGridFunctionCoefficient (
@@ -331,13 +385,30 @@ void GradientGridFunctionCoefficient::SetGridFunction(const GridFunction *gf)
 void GradientGridFunctionCoefficient::Eval(Vector &V, ElementTransformation &T,
                                            const IntegrationPoint &ip)
 {
-   GridFunc->GetGradient(T, V);
+   Mesh *gf_mesh = GridFunc->FESpace()->GetMesh();
+   if (T.mesh == gf_mesh)
+   {
+      GridFunc->GetGradient(T, V);
+   }
+   else
+   {
+      IntegrationPoint coarse_ip;
+      ElementTransformation *coarse_T = RefinedToCoarse(*gf_mesh, T, ip, coarse_ip);
+      GridFunc->GetGradient(*coarse_T, V);
+   }
 }
 
 void GradientGridFunctionCoefficient::Eval(
    DenseMatrix &M, ElementTransformation &T, const IntegrationRule &ir)
 {
-   GridFunc->GetGradients(T, ir, M);
+   if (T.mesh == GridFunc->FESpace()->GetMesh())
+   {
+      GridFunc->GetGradients(T, ir, M);
+   }
+   else
+   {
+      VectorCoefficient::Eval(M, T, ir);
+   }
 }
 
 CurlGridFunctionCoefficient::CurlGridFunctionCoefficient(
@@ -363,7 +434,17 @@ void CurlGridFunctionCoefficient::SetGridFunction(const GridFunction *gf)
 void CurlGridFunctionCoefficient::Eval(Vector &V, ElementTransformation &T,
                                        const IntegrationPoint &ip)
 {
-   GridFunc->GetCurl(T, V);
+   Mesh *gf_mesh = GridFunc->FESpace()->GetMesh();
+   if (T.mesh == gf_mesh)
+   {
+      GridFunc->GetCurl(T, V);
+   }
+   else
+   {
+      IntegrationPoint coarse_ip;
+      ElementTransformation *coarse_T = RefinedToCoarse(*gf_mesh, T, ip, coarse_ip);
+      GridFunc->GetCurl(*coarse_T, V);
+   }
 }
 
 DivergenceGridFunctionCoefficient::DivergenceGridFunctionCoefficient (
@@ -375,7 +456,17 @@ DivergenceGridFunctionCoefficient::DivergenceGridFunctionCoefficient (
 double DivergenceGridFunctionCoefficient::Eval(ElementTransformation &T,
                                                const IntegrationPoint &ip)
 {
-   return GridFunc->GetDivergence(T);
+   Mesh *gf_mesh = GridFunc->FESpace()->GetMesh();
+   if (T.mesh == gf_mesh)
+   {
+      return GridFunc->GetDivergence(T);
+   }
+   else
+   {
+      IntegrationPoint coarse_ip;
+      ElementTransformation *coarse_T = RefinedToCoarse(*gf_mesh, T, ip, coarse_ip);
+      return GridFunc->GetDivergence(*coarse_T);
+   }
 }
 
 void VectorDeltaCoefficient::SetTime(double t)

fem/coefficient.hpp

@@ -574,6 +574,7 @@ public:
    /// Evaluate the coefficient.
    virtual void Eval(Vector &V, ElementTransformation &T,
                      const IntegrationPoint &ip);
+   using VectorCoefficient::Eval;
 };
 
 /// A general vector function coefficient

fem/complex_fem.cpp

@@ -1243,7 +1243,7 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
          HypreParMatrix * Ah;
          A_i.Get(Ah);
          hypre_ParCSRMatrix *Aih = *Ah;
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
          ess_tdof_list.HostRead();
          for (int k = 0; k < n; k++)
          {
@@ -1256,7 +1256,7 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
             ess_tdof_list.GetMemory().Read(MemoryClass::DEVICE, n);
          const int *d_diag_i = Aih->diag->i;
          double *d_diag_data = Aih->diag->data;
-         CuWrap1D(n, [=] MFEM_DEVICE (int k)
+         MFEM_GPU_FORALL(k, n,
          {
             const int j = d_ess_tdof_list[k];
             d_diag_data[d_diag_i[j]] = 0.0;

fem/conduitdatacollection.cpp

@@ -330,7 +330,7 @@ ConduitDataCollection::BlueprintMeshToMesh(const Node &n_mesh,
          }
          else
          {
-            Node &(n_bndry_conn_conv) =
+            Node &n_bndry_conn_conv =
                n_conv["topologies"][bndry_topo_name]["elements/connectivity"];
             n_bndry_conn.to_int_array(n_bndry_conn_conv);
             bndry_indices = (n_bndry_conn_conv).value();

fem/datacollection.cpp

@@ -11,6 +11,7 @@
 
 #include "fem.hpp"
 #include "../mesh/nurbs.hpp"
+#include "../mesh/vtk.hpp"
 #include "../general/binaryio.hpp"
 #include "../general/text.hpp"
 #include "picojson.h"
@@ -787,53 +788,44 @@ void ParaViewDataCollection::Load(int )
 
 std::string ParaViewDataCollection::GenerateCollectionPath()
 {
-   std::string out = "";
-   out = prefix_path + DataCollection::GetCollectionName();
-   return out;
+   return prefix_path + DataCollection::GetCollectionName();
 }
 
 std::string ParaViewDataCollection::GeneratePVTUPath()
 {
-   std::string out = "Cycle" + to_padded_string(cycle,pad_digits_cycle);
-   return out;
+   return "Cycle" + to_padded_string(cycle,pad_digits_cycle);
 }
 
 std::string ParaViewDataCollection::GenerateVTUPath()
 {
-   std::string out = GeneratePVTUPath();
-   return out;
+   return GeneratePVTUPath();
 }
 
 std::string ParaViewDataCollection::GeneratePVDFileName()
 {
-   std::string out = GetCollectionName()+".pvd";
-   return out;
+   return GetCollectionName() + ".pvd";
 }
 
-std::string ParaViewDataCollection::GeneratePVTUFileName()
+std::string ParaViewDataCollection::GeneratePVTUFileName(
+   const std::string &prefix)
 {
-   std::string out = "data.pvtu";
-   return out;
+   return prefix + ".pvtu";
 }
 
-std::string ParaViewDataCollection::GenerateVTUFileName()
+std::string ParaViewDataCollection::GenerateVTUFileName(
+   const std::string &prefix, int rank)
 {
-   std::string out = "proc" + to_padded_string(myid,pad_digits_rank)+".vtu";
-   return out;
-}
-std::string ParaViewDataCollection::GenerateVTUFileName(int crank)
-{
-   std::string out = "proc" + to_padded_string(crank,pad_digits_rank)+".vtu";
-   return out;
+   return prefix + to_padded_string(rank, pad_digits_rank) + ".vtu";
 }
 
 void ParaViewDataCollection::Save()
 {
    // add a new collection to the PDV file
 
+   std::string col_path = GenerateCollectionPath();
    // check if the directories are created
    {
-      std::string path = GenerateCollectionPath()+"/"+GenerateVTUPath();
+      std::string path = col_path + "/" + GenerateVTUPath();
       int err = create_directory(path, mesh, myid);
       if (err)
       {
@@ -850,8 +842,7 @@ void ParaViewDataCollection::Save()
 
    if (myid == 0 && !pvd_stream.is_open())
    {
-      std::string dpath=GenerateCollectionPath();
-      std::string pvdname=dpath+"/"+GeneratePVDFileName();
+      std::string pvdname = col_path + "/" + GeneratePVDFileName();
 
       bool write_header = true;
       std::ifstream pvd_in;
@@ -915,80 +906,87 @@ void ParaViewDataCollection::Save()
       }
    }
 
-   // define the vtu file
+   std::string vtu_prefix = col_path + "/" + GenerateVTUPath() + "/";
+
+   // Save the local part of the mesh and grid functions fields to the local
+   // VTU file
    {
-      std::string fname = GenerateCollectionPath()+"/"+GenerateVTUPath()+"/"
-                          +GenerateVTUFileName();
-      std::fstream out(fname, std::ios::out);
+      std::ofstream out(vtu_prefix + GenerateVTUFileName("proc", myid));
       out.precision(precision);
-      SaveDataVTU(out,levels_of_detail);
-      out.close();
+      SaveDataVTU(out, levels_of_detail);
    }
 
-   // define the pvtu file only on process 0
-   if (myid==0)
+   // Save the local part of the quadrature function fields
+   for (const auto &qfield : q_field_map)
    {
-      std::string fname = GenerateCollectionPath()+"/"+GeneratePVTUPath()+"/"
-                          +GeneratePVTUFileName();
-      std::fstream out(fname, std::ios::out);
+      const std::string &field_name = qfield.first;
+      std::ofstream out(vtu_prefix + GenerateVTUFileName(field_name, myid));
+      qfield.second->SaveVTU(out, pv_data_format, compression);
+   }
 
-      out << "<?xml version=\"1.0\"?>\n";
-      out << "<VTKFile type=\"PUnstructuredGrid\"";
-      out << " version =\"0.1\" byte_order=\"" << VTKByteOrder() << "\">\n";
-      out << "<PUnstructuredGrid GhostLevel=\"0\">\n";
-
-      out << "<PPoints>\n";
-      out << "\t<PDataArray type=\"" << GetDataTypeString() << "\" ";
-      out << " Name=\"Points\" NumberOfComponents=\"3\""
-          << " format=\"" << GetDataFormatString() << "\"/>\n";
-      out << "</PPoints>\n";
-
-      out << "<PCells>\n";
-      out << "\t<PDataArray type=\"Int32\" ";
-      out << " Name=\"connectivity\" NumberOfComponents=\"1\""
-          << " format=\"" << GetDataFormatString() << "\"/>\n";
-      out << "\t<PDataArray type=\"Int32\" ";
-      out << " Name=\"offsets\"      NumberOfComponents=\"1\""
-          << " format=\"" << GetDataFormatString() << "\"/>\n";
-      out << "\t<PDataArray type=\"UInt8\" ";
-      out << " Name=\"types\"        NumberOfComponents=\"1\""
-          << " format=\"" << GetDataFormatString() << "\"/>\n";
-      out << "</PCells>\n";
-
-      out << "<PPointData>\n";
-      for (FieldMapIterator it=field_map.begin(); it!=field_map.end(); ++it)
+   // MPI rank 0 also creates a "PVTU" file that points to all of the separately
+   // written VTU files.
+   // This file path is then appended to the PVD file.
+   if (myid == 0)
+   {
+      // Create the main PVTU file
       {
-         int vec_dim=it->second->VectorDim();
-         out << "<PDataArray type=\"" << GetDataTypeString()
-             << "\" Name=\"" << it->first
-             << "\" NumberOfComponents=\"" << vec_dim << "\" "
-             << "format=\"" << GetDataFormatString() << "\" />\n";
+         std::ofstream pvtu_out(vtu_prefix + GeneratePVTUFileName("data"));
+         WritePVTUHeader(pvtu_out);
+
+         // Grid function fields
+         pvtu_out << "<PPointData>\n";
+         for (auto &field_it : field_map)
+         {
+            int vec_dim = field_it.second->VectorDim();
+            pvtu_out << "<PDataArray type=\"" << GetDataTypeString()
+                     << "\" Name=\"" << field_it.first
+                     << "\" NumberOfComponents=\"" << vec_dim << "\" "
+                     << "format=\"" << GetDataFormatString() << "\" />\n";
+         }
+         pvtu_out << "</PPointData>\n";
+         // Element attributes
+         pvtu_out << "<PCellData>\n";
+         pvtu_out << "\t<PDataArray type=\"Int32\" Name=\"" << "attribute"
+                  << "\" NumberOfComponents=\"1\""
+                  << " format=\"" << GetDataFormatString() << "\"/>\n";
+         pvtu_out << "</PCellData>\n";
+
+         WritePVTUFooter(pvtu_out, "proc");
       }
-      out << "</PPointData>\n";
 
-      // CELL DATA
-      out << "<PCellData>\n";
-      out << "\t<PDataArray type=\"Int32\" Name=\"" << "attribute"
-          << "\" NumberOfComponents=\"1\""
-          << " format=\"" << GetDataFormatString() << "\"/>\n";
-      out << "</PCellData>\n";
+      // Add the latest PVTU to the PVD
+      pvd_stream << "<DataSet timestep=\"" << GetTime()
+                 << "\" group=\"\" part=\"" << 0 << "\" file=\""
+                 << GeneratePVTUPath() + "/" + GeneratePVTUFileName("data")
+                 << "\" name=\"mesh\"/>\n";
 
-      for (int ii=0; ii<num_procs; ii++)
+      // Create PVTU files for each quadrature field and add them to the PVD
+      // file
+      for (auto &q_field : q_field_map)
       {
-         // this one is generated without the path
-         std::string nfname=GenerateVTUFileName(ii);
-         out << "<Piece Source=\"" << nfname << "\"/>\n";
-      }
-      out << "</PUnstructuredGrid>\n";
-      out << "</VTKFile>\n";
-      out.close();
+         const std::string &q_field_name = q_field.first;
+         std::string q_fname = GeneratePVTUPath() + "/"
+                               + GeneratePVTUFileName(q_field_name);
 
-      fname = GeneratePVTUPath()+"/"+GeneratePVTUFileName();
-      // add the pvtu file to the pvd_stream
-      pvd_stream << "<DataSet timestep=\"" << GetTime();  // GetCycle();
-      pvd_stream << "\" group=\"\" part=\"" << 0 << "\" file=\"";
-      pvd_stream << fname << "\"/>\n";
+         std::ofstream pvtu_out(col_path + "/" + q_fname);
+         WritePVTUHeader(pvtu_out);
+         int vec_dim = q_field.second->GetVDim();
+         pvtu_out << "<PPointData>\n";
+         pvtu_out << "<PDataArray type=\"" << GetDataTypeString()
+                  << "\" Name=\"" << q_field_name
+                  << "\" NumberOfComponents=\"" << vec_dim << "\" "
+                  << "format=\"" << GetDataFormatString() << "\" />\n";
+         pvtu_out << "</PPointData>\n";
+         WritePVTUFooter(pvtu_out, q_field_name);
+
+         pvd_stream << "<DataSet timestep=\"" << GetTime()
+                    << "\" group=\"\" part=\"" << 0 << "\" file=\""
+                    << q_fname << "\" name=\"" << q_field_name << "\"/>\n";
+      }
       pvd_stream.flush();
+      // Move the insertion point before the closing collection tag, so that
+      // the PVD file is valid even when writing incrementally.
       std::fstream::pos_type pos = pvd_stream.tellp();
       pvd_stream << "</Collection>\n";
       pvd_stream << "</VTKFile>" << std::endl;
@@ -996,6 +994,44 @@ void ParaViewDataCollection::Save()
    }
 }
 
+void ParaViewDataCollection::WritePVTUHeader(std::ostream &out)
+{
+   out << "<?xml version=\"1.0\"?>\n";
+   out << "<VTKFile type=\"PUnstructuredGrid\"";
+   out << " version =\"0.1\" byte_order=\"" << VTKByteOrder() << "\">\n";
+   out << "<PUnstructuredGrid GhostLevel=\"0\">\n";
+
+   out << "<PPoints>\n";
+   out << "\t<PDataArray type=\"" << GetDataTypeString() << "\" ";
+   out << " Name=\"Points\" NumberOfComponents=\"3\""
+       << " format=\"" << GetDataFormatString() << "\"/>\n";
+   out << "</PPoints>\n";
+
+   out << "<PCells>\n";
+   out << "\t<PDataArray type=\"Int32\" ";
+   out << " Name=\"connectivity\" NumberOfComponents=\"1\""
+       << " format=\"" << GetDataFormatString() << "\"/>\n";
+   out << "\t<PDataArray type=\"Int32\" ";
+   out << " Name=\"offsets\"      NumberOfComponents=\"1\""
+       << " format=\"" << GetDataFormatString() << "\"/>\n";
+   out << "\t<PDataArray type=\"UInt8\" ";
+   out << " Name=\"types\"        NumberOfComponents=\"1\""
+       << " format=\"" << GetDataFormatString() << "\"/>\n";
+   out << "</PCells>\n";
+}
+
+void ParaViewDataCollection::WritePVTUFooter(std::ostream &out,
+                                             const std::string &vtu_prefix)
+{
+   for (int ii=0; ii<num_procs; ii++)
+   {
+      std::string vtu_filename = GenerateVTUFileName(vtu_prefix, ii);
+      out << "<Piece Source=\"" << vtu_filename << "\"/>\n";
+   }
+   out << "</PUnstructuredGrid>\n";
+   out << "</VTKFile>\n";
+}
+
 void ParaViewDataCollection::SaveDataVTU(std::ostream &out, int ref)
 {
    out << "<VTKFile type=\"UnstructuredGrid\"";
@@ -1015,16 +1051,6 @@ void ParaViewDataCollection::SaveDataVTU(std::ostream &out, int ref)
    {
       SaveGFieldVTU(out,ref,it);
    }
-   // iterate over all quadrature functions
-   // if the Quadrature functions are dumped as cell data
-   // the cycle should be moved before the grid functions
-   // and the PrintVTU CellData section should be open in the mesh dump
-   for (QFieldMapIterator it=q_field_map.begin(); it!=q_field_map.end(); ++it)
-   {
-      // save the quadrature functions
-      // this one is not implemented yet
-      SaveQFieldVTU(out,ref,it);
-   }
    out << "</PointData>\n";
    // close the mesh
    out << "</Piece>\n"; // close the piece open in the PrintVTU method
@@ -1032,27 +1058,21 @@ void ParaViewDataCollection::SaveDataVTU(std::ostream &out, int ref)
    out << "</VTKFile>" << std::endl;
 }
 
-void ParaViewDataCollection::SaveQFieldVTU(std::ostream &out, int ref,
-                                           const QFieldMapIterator& it )
-{
-   MFEM_WARNING("SaveQFieldVTU is not currently implemented - field name:"<<it->second);
-}
-
 void ParaViewDataCollection::SaveGFieldVTU(std::ostream &out, int ref_,
-                                           const FieldMapIterator& it)
+                                           const FieldMapIterator &it)
 {
    RefinedGeometry *RefG;
    Vector val;
    DenseMatrix vval, pmat;
    std::vector<char> buf;
    int vec_dim = it->second->VectorDim();
+   out << "<DataArray type=\"" << GetDataTypeString()
+       << "\" Name=\"" << it->first;
+   out << "\" NumberOfComponents=\"" << vec_dim << "\""
+       << " format=\"" << GetDataFormatString() << "\" >" << '\n';
    if (vec_dim == 1)
    {
       // scalar data
-      out << "<DataArray type=\"" << GetDataTypeString()
-          << "\" Name=\"" << it->first;
-      out << "\" NumberOfComponents=\"1\" format=\""
-          << GetDataFormatString() << "\" >\n";
       for (int i = 0; i < mesh->GetNE(); i++)
       {
          RefG = GlobGeometryRefiner.Refine(
@@ -1060,51 +1080,23 @@ void ParaViewDataCollection::SaveGFieldVTU(std::ostream &out, int ref_,
          it->second->GetValues(i, RefG->RefPts, val, pmat);
          for (int j = 0; j < val.Size(); j++)
          {
-            if (pv_data_format == VTKFormat::ASCII)
-            {
-               out << ZeroSubnormal(val(j)) << '\n';
-            }
-            else if (pv_data_format == VTKFormat::BINARY)
-            {
-               bin_io::AppendBytes(buf, val(j));
-            }
-            else
-            {
-               bin_io::AppendBytes<float>(buf, float(val(j)));
-            }
+            WriteBinaryOrASCII(out, buf, val(j), "\n", pv_data_format);
          }
       }
    }
    else
    {
       // vector data
-      out << "<DataArray type=\"" << GetDataTypeString()
-          << "\" Name=\"" << it->first;
-      out << "\" NumberOfComponents=\"" << vec_dim << "\""
-          << " format=\"" << GetDataFormatString() << "\" >" << '\n';
       for (int i = 0; i < mesh->GetNE(); i++)
       {
          RefG = GlobGeometryRefiner.Refine(
                    mesh->GetElementBaseGeometry(i), ref_, 1);
-
          it->second->GetVectorValues(i, RefG->RefPts, vval, pmat);
-
          for (int jj = 0; jj < vval.Width(); jj++)
          {
             for (int ii = 0; ii < vval.Height(); ii++)
             {
-               if (pv_data_format == VTKFormat::ASCII)
-               {
-                  out << ZeroSubnormal(vval(ii,jj)) << ' ';
-               }
-               else if (pv_data_format == VTKFormat::BINARY)
-               {
-                  bin_io::AppendBytes(buf, vval(ii,jj));
-               }
-               else
-               {
-                  bin_io::AppendBytes<float>(buf, float(vval(ii,jj)));
-               }
+               WriteBinaryOrASCII(out, buf, vval(ii,jj), " ", pv_data_format);
             }
             if (pv_data_format == VTKFormat::ASCII) { out << '\n'; }
          }

fem/datacollection.hpp

@@ -491,19 +491,20 @@ private:
    bool restart_mode;
 
 protected:
+   void WritePVTUHeader(std::ostream &out);
+   void WritePVTUFooter(std::ostream &out, const std::string &vtu_prefix);
    void SaveDataVTU(std::ostream &out, int ref);
    void SaveGFieldVTU(std::ostream& out, int ref_, const FieldMapIterator& it);
-   void SaveQFieldVTU(std::ostream &out, int ref, const QFieldMapIterator& it);
    const char *GetDataFormatString() const;
    const char *GetDataTypeString() const;
 
-   std::string  GenerateCollectionPath();
-   std::string  GenerateVTUFileName();
-   std::string  GenerateVTUFileName(int rank);
-   std::string  GenerateVTUPath();
-   std::string  GeneratePVDFileName();
-   std::string  GeneratePVTUFileName();
-   std::string  GeneratePVTUPath();
+   std::string GenerateCollectionPath();
+   std::string GenerateVTUFileName(const std::string &prefix, int rank);
+   std::string GenerateVTUPath();
+   std::string GeneratePVDFileName();
+   std::string GeneratePVTUFileName(const std::string &prefix);
+   std::string GeneratePVTUPath();
+
 
 public:
    /// Constructor. The collection name is used when saving the data.

fem/doftrans.hpp

@@ -243,6 +243,7 @@ public:
    void TransformDual(double *v) const;
 
    void InvTransformDual(double *v) const;
+   using DofTransformation::InvTransformDual;
 };
 
 /// DoF transformation implementation for the Nedelec basis on tetrahedra

fem/eltrans.cpp

@@ -21,7 +21,8 @@ ElementTransformation::ElementTransformation()
      EvalState(0),
      geom(Geometry::INVALID),
      Attribute(-1),
-     ElementNo(-1)
+     ElementNo(-1),
+     mesh(nullptr)
 { }
 
 double ElementTransformation::EvalWeight()

fem/eltrans.hpp

@@ -75,6 +75,12 @@ public:
 
    int Attribute, ElementNo, ElementType;
 
+   /// The Mesh object containing the element.
+   /** If the element transformation belongs to a mesh, this will point to the
+       containing Mesh object. ElementNo will be the number of the element in
+       this Mesh. This will be NULL if the element does not belong to a mesh. */
+   class Mesh *mesh;
+
    ElementTransformation();
 
    /** @brief Force the reevaluation of the Jacobian in the next call. */

fem/fespace.cpp

@@ -894,9 +894,10 @@ int FiniteElementSpace::GetDegenerateFaceDofs(int index, Array<int> &dofs,
 int FiniteElementSpace::GetNumBorderDofs(Geometry::Type geom, int order) const
 {
    // return the number of vertex and edge DOFs that precede inner DOFs
-   int nv = fec->GetNumDof(Geometry::POINT, order);
-   int ne = fec->GetNumDof(Geometry::SEGMENT, order);
-   return Geometry::NumVerts[geom] * (nv + ne);
+   const int nv = fec->GetNumDof(Geometry::POINT, order);
+   const int ne = fec->GetNumDof(Geometry::SEGMENT, order);
+
+   return Geometry::NumVerts[geom] * (geom == Geometry::SEGMENT ? nv : (nv + ne));
 }
 
 int FiniteElementSpace::GetEntityDofs(int entity, int index, Array<int> &dofs,

fem/fmsconvert.cpp

@@ -112,7 +112,7 @@ FmsFieldToGridFunction(FmsMesh fms_mesh, FmsField f, Mesh *mesh,
    // NOTE: transplanted from the FmsMeshToMesh function
    //       We should do this work once and save it.
    //--------------------------------------------------
-   FmsInt dim, n_vert, n_elem, space_dim;
+   FmsInt dim, n_elem, space_dim;
 
    // Find the first component that has coordinates - that will be the new mfem
    // mesh.
@@ -144,7 +144,6 @@ FmsFieldToGridFunction(FmsMesh fms_mesh, FmsField f, Mesh *mesh,
          n_ents[et] += num_ents;
       }
    }
-   n_vert = n_ents[FMS_VERTEX];
    //--------------------------------------------------
 
    // Interrogate the field.

fem/gridfunc.cpp

@@ -3948,6 +3948,129 @@ std::ostream &operator<<(std::ostream &out, const QuadratureFunction &qf)
    return out;
 }
 
+void QuadratureFunction::SaveVTU(std::ostream &out, VTKFormat format,
+                                 int compression_level) const
+{
+   out << R"(<VTKFile type="UnstructuredGrid" version="0.1")";
+   if (compression_level != 0)
+   {
+      out << R"( compressor="vtkZLibDataCompressor")";
+   }
+   out << " byte_order=\"" << VTKByteOrder() << "\">\n";
+   out << "<UnstructuredGrid>\n";
+
+   const char *fmt_str = (format == VTKFormat::ASCII) ? "ascii" : "binary";
+   const char *type_str = (format != VTKFormat::BINARY32) ? "Float64" : "Float32";
+   std::vector<char> buf;
+
+   int np = qspace->GetSize();
+   int ne = qspace->GetNE();
+   int sdim = qspace->GetMesh()->SpaceDimension();
+
+   // For quadrature functions, each point is a vertex cell, so number of cells
+   // is equal to number of points
+   out << "<Piece NumberOfPoints=\"" << np
+       << "\" NumberOfCells=\"" << np << "\">\n";
+
+   // print out the points
+   out << "<Points>\n";
+   out << "<DataArray type=\"" << type_str
+       << "\" NumberOfComponents=\"3\" format=\"" << fmt_str << "\">\n";
+
+   Vector pt(sdim);
+   for (int i = 0; i < ne; i++)
+   {
+      ElementTransformation &T = *qspace->GetMesh()->GetElementTransformation(i);
+      const IntegrationRule &ir = GetElementIntRule(i);
+      for (int j = 0; j < ir.Size(); j++)
+      {
+         T.Transform(ir[j], pt);
+         WriteBinaryOrASCII(out, buf, pt[0], " ", format);
+         if (sdim > 1) { WriteBinaryOrASCII(out, buf, pt[1], " ", format); }
+         else { WriteBinaryOrASCII(out, buf, 0.0, " ", format); }
+         if (sdim > 2) { WriteBinaryOrASCII(out, buf, pt[2], "", format); }
+         else { WriteBinaryOrASCII(out, buf, 0.0, "", format); }
+         if (format == VTKFormat::ASCII) { out << '\n'; }
+      }
+   }
+   if (format != VTKFormat::ASCII)
+   {
+      WriteBase64WithSizeAndClear(out, buf, compression_level);
+   }
+   out << "</DataArray>\n";
+   out << "</Points>\n";
+
+   // Write cells (each cell is just a vertex)
+   out << "<Cells>\n";
+   // Connectivity
+   out << R"(<DataArray type="Int32" Name="connectivity" format=")"
+       << fmt_str << "\">\n";
+
+   for (int i=0; i<np; ++i) { WriteBinaryOrASCII(out, buf, i, "\n", format); }
+   if (format != VTKFormat::ASCII)
+   {
+      WriteBase64WithSizeAndClear(out, buf, compression_level);
+   }
+   out << "</DataArray>\n";
+   // Offsets
+   out << R"(<DataArray type="Int32" Name="offsets" format=")"
+       << fmt_str << "\">\n";
+   for (int i=0; i<np; ++i) { WriteBinaryOrASCII(out, buf, i, "\n", format); }
+   if (format != VTKFormat::ASCII)
+   {
+      WriteBase64WithSizeAndClear(out, buf, compression_level);
+   }
+   out << "</DataArray>\n";
+   // Types
+   out << R"(<DataArray type="UInt8" Name="types" format=")"
+       << fmt_str << "\">\n";
+   for (int i = 0; i < np; i++)
+   {
+      uint8_t vtk_cell_type = VTKGeometry::POINT;
+      WriteBinaryOrASCII(out, buf, vtk_cell_type, "\n", format);
+   }
+   if (format != VTKFormat::ASCII)
+   {
+      WriteBase64WithSizeAndClear(out, buf, compression_level);
+   }
+   out << "</DataArray>\n";
+   out << "</Cells>\n";
+
+   out << "<PointData>\n";
+   out << "<DataArray type=\"" << type_str << "\" Name=\"u\" format=\""
+       << fmt_str << "\" NumberOfComponents=\"" << vdim << "\">\n";
+   for (int i = 0; i < ne; i++)
+   {
+      DenseMatrix vals;
+      GetElementValues(i, vals);
+      for (int j = 0; j < vals.Size(); ++j)
+      {
+         for (int vd = 0; vd < vdim; ++vd)
+         {
+            WriteBinaryOrASCII(out, buf, vals(vd, j), " ", format);
+         }
+         if (format == VTKFormat::ASCII) { out << '\n'; }
+      }
+   }
+   if (format != VTKFormat::ASCII)
+   {
+      WriteBase64WithSizeAndClear(out, buf, compression_level);
+   }
+   out << "</DataArray>\n";
+   out << "</PointData>\n";
+
+   out << "</Piece>\n";
+   out << "</UnstructuredGrid>\n";
+   out << "</VTKFile>" << std::endl;
+}
+
+void QuadratureFunction::SaveVTU(const std::string &filename, VTKFormat format,
+                                 int compression_level) const
+{
+   std::ofstream f(filename + ".vtu");
+   SaveVTU(f, format, compression_level);
+}
+
 
 double ZZErrorEstimator(BilinearFormIntegrator &blfi,
                         GridFunction &u,

fem/gridfunc.hpp

@@ -902,6 +902,22 @@ public:
 
    /// Write the QuadratureFunction to the stream @a out.
    void Save(std::ostream &out) const;
+
+   /// @brief Write the QuadratureFunction to @a out in VTU (ParaView) format.
+   ///
+   /// The data will be uncompressed if @a compression_level is zero, or if the
+   /// format is VTKFormat::ASCII. Otherwise, zlib compression will be used for
+   /// binary data.
+   void SaveVTU(std::ostream &out, VTKFormat format=VTKFormat::ASCII,
+                int compression_level=0) const;
+
+   /// @brief Save the QuadratureFunction to a VTU (ParaView) file.
+   ///
+   /// The extension ".vtu" will be appended to @a filename.
+   /// @sa SaveVTU(std::ostream &out, VTKFormat format=VTKFormat::ASCII,
+   ///             int compression_level=0)
+   void SaveVTU(const std::string &filename, VTKFormat format=VTKFormat::ASCII,
+                int compression_level=0) const;
 };
 
 /// Overload operator<< for std::ostream and QuadratureFunction.

fem/gslib.cpp

@@ -78,6 +78,8 @@ void FindPointsGSLIB::Setup(Mesh &m, const double bb_t, const double newt_tol,
    MFEM_VERIFY(m.GetNodes() != NULL, "Mesh nodes are required.");
    MFEM_VERIFY(m.GetNumGeometries(m.Dimension()) == 1,
                "Mixed meshes are not currently supported in FindPointsGSLIB.");
+   MFEM_VERIFY(!(m.GetNodes()->FESpace()->IsVariableOrder()),
+               "Variable order mesh is not currently supported.");
 
    // call FreeData if FindPointsGSLIB::Setup has been called already
    if (setupflag) { FreeData(); }
@@ -590,7 +592,8 @@ void FindPointsGSLIB::Interpolate(const GridFunction &field_in,
    const L2_FECollection *fec_l2 = dynamic_cast<const L2_FECollection *>(fec_in);
 
    if (fec_h1 && gf_order == mesh_order &&
-       fec_h1->GetBasisType() == BasisType::GaussLobatto)
+       fec_h1->GetBasisType() == BasisType::GaussLobatto &&
+       !field_in.FESpace()->IsVariableOrder())
    {
       InterpolateH1(field_in, field_out);
       return;
@@ -857,6 +860,8 @@ void OversetFindPointsGSLIB::Setup(Mesh &m, const int meshid,
    MFEM_VERIFY(m.GetNodes() != NULL, "Mesh nodes are required.");
    MFEM_VERIFY(m.GetNumGeometries(m.Dimension()) == 1,
                "Mixed meshes are not currently supported in FindPointsGSLIB.");
+   MFEM_VERIFY(!(m.GetNodes()->FESpace()->IsVariableOrder()),
+               "Variable order mesh is not currently supported.");
 
    // FreeData if OversetFindPointsGSLIB::Setup has been called already
    if (setupflag) { FreeData(); }

fem/pfespace.cpp

@@ -1010,12 +1010,9 @@ void ParFiniteElementSpace::GetEssentialVDofs(const Array<int> &bdr_attr_is_ess,
 {
    FiniteElementSpace::GetEssentialVDofs(bdr_attr_is_ess, ess_dofs, component);
 
-   if (Conforming())
-   {
-      // Make sure that processors without boundary elements mark
-      // their boundary dofs (if they have any).
-      Synchronize(ess_dofs);
-   }
+   // Make sure that processors without boundary elements mark
+   // their boundary dofs (if they have any).
+   Synchronize(ess_dofs);
 }
 
 void ParFiniteElementSpace::GetEssentialTrueDofs(const Array<int>
@@ -1041,7 +1038,8 @@ void ParFiniteElementSpace::GetEssentialTrueDofs(const Array<int>
    {
       if (bool(ted[i]) != bool(true_ess_dofs2[i])) { counter++; }
    }
-   MFEM_VERIFY(counter == 0, "internal MFEM error: counter = " << counter);
+   MFEM_VERIFY(counter == 0, "internal MFEM error: counter = " << counter
+               << ", rank = " << MyRank);
 #endif
 
    MarkerToList(true_ess_dofs, ess_tdof_list);

fem/qinterp/eval.hpp

@@ -65,13 +65,13 @@ static void Values2D(const int NE,
       MFEM_SHARED double sm0[NBZ][MDQ*MDQ];
       MFEM_SHARED double sm1[NBZ][MDQ*MDQ];
 
+      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,sB);
+
       ConstDeviceMatrix B(sB, D1D,Q1D);
       DeviceMatrix DD(sm0[tidz], MD1, MD1);
       DeviceMatrix DQ(sm1[tidz], MD1, MQ1);
       DeviceMatrix QQ(sm0[tidz], MQ1, MQ1);
 
-      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,sB);
-
       for (int c = 0; c < VDIM; c++)
       {
          kernels::internal::LoadX(e,D1D,c,x,DD);
@@ -126,14 +126,14 @@ static void Values3D(const int NE,
       MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
       MFEM_SHARED double sm1[MDQ*MDQ*MDQ];
 
+      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,sB);
+
       ConstDeviceMatrix B(sB, D1D,Q1D);
       DeviceCube DDD(sm0, MD1,MD1,MD1);
       DeviceCube DDQ(sm1, MD1,MD1,MQ1);
       DeviceCube DQQ(sm0, MD1,MQ1,MQ1);
       DeviceCube QQQ(sm1, MQ1,MQ1,MQ1);
 
-      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,sB);
-
       for (int c = 0; c < VDIM; c++)
       {
          kernels::internal::LoadX(e,D1D,c,x,DDD);

fem/tmop.hpp

@@ -1133,7 +1133,7 @@ protected:
 
    // Evaluation of the discrete target specification on different meshes.
    // Owned.
-   AdaptivityEvaluator *adapt_eval;
+   AdaptivityEvaluator *adapt_lim_eval;
 
    void SetDiscreteTargetBase(const GridFunction &tspec_);
    void SetTspecAtIndex(int idx, const GridFunction &tspec_);
@@ -1156,7 +1156,7 @@ public:
 #endif
         amr_el(-1), lim_min_size(-0.1),
         good_tspec(false), good_tspec_grad(false), good_tspec_hess(false),
-        adapt_eval(NULL) { }
+        adapt_lim_eval(NULL) { }
 
    virtual ~DiscreteAdaptTC();
 
@@ -1232,8 +1232,8 @@ public:
 
    void SetAdaptivityEvaluator(AdaptivityEvaluator *ae)
    {
-      if (adapt_eval) { delete adapt_eval; }
-      adapt_eval = ae;
+      if (adapt_lim_eval) { delete adapt_lim_eval; }
+      adapt_lim_eval = ae;
    }
 
    const Vector &GetTspecPert1H()   { return tspec_pert1h; }
@@ -1315,15 +1315,15 @@ protected:
    int integ_order;
 
    // Weight Coefficient multiplying the quality metric term.
-   Coefficient *coeff1; // not owned, if NULL -> coeff1 is 1.
+   Coefficient *metric_coeff; // not owned, if NULL -> metric_coeff is 1.
    // Normalization factor for the metric term.
    double metric_normal;
 
    // Nodes and weight Coefficient used for "limiting" the TMOP_Integrator.
    // These are both NULL when there is no limiting.
-   // The class doesn't own nodes0 and coeff0.
-   const GridFunction *nodes0;
-   Coefficient *coeff0;
+   // The class doesn't own lim_nodes0 and lim_coeff.
+   const GridFunction *lim_nodes0;
+   Coefficient *lim_coeff;
    // Limiting reference distance. Not owned.
    const GridFunction *lim_dist;
    // Limiting function. Owned.
@@ -1332,20 +1332,21 @@ protected:
    double lim_normal;
 
    // Adaptive limiting.
-   const GridFunction *zeta_0;       // Not owned.
+   const GridFunction *adapt_lim_gf0;    // Not owned.
 #ifdef MFEM_USE_MPI
-   const ParGridFunction *pzeta_0;
+   const ParGridFunction *adapt_lim_pgf0;
 #endif
-   GridFunction *zeta;               // Owned. Updated by adapt_eval.
-   Coefficient *coeff_zeta;          // Not owned.
-   AdaptivityEvaluator *adapt_eval;  // Not owned.
+   GridFunction *adapt_lim_gf;           // Owned. Updated by adapt_lim_eval.
+   Coefficient *adapt_lim_coeff;         // Not owned.
+   AdaptivityEvaluator *adapt_lim_eval;  // Not owned.
 
    // Surface fitting.
-   GridFunction *sigma, *sigma_bar;  // Owned. Updated by sigma_eval.
-   const Array<bool> *sigma_marker;  // Not owned.
-   Coefficient *coeff_sigma;         // Not owned.
-   AdaptivityEvaluator *sigma_eval;  // Not owned.
-   double sigma_normal;
+   GridFunction *surf_fit_gf,
+                *surf_fit_gf_bar;       // Owned, Updated by surf_fit_eval.
+   const Array<bool> *surf_fit_marker;  // Not owned.
+   Coefficient *surf_fit_coeff;         // Not owned.
+   AdaptivityEvaluator *surf_fit_eval;  // Not owned.
+   double surf_fit_normal;
 
    DiscreteAdaptTC *discr_tc;
 
@@ -1416,7 +1417,7 @@ protected:
 
    void ComputeNormalizationEnergies(const GridFunction &x,
                                      double &metric_energy, double &lim_energy,
-                                     double &sigma_energy);
+                                     double &surf_fit_gf_energy);
 
    void AssembleElementVectorExact(const FiniteElement &el,
                                    ElementTransformation &T,
@@ -1471,7 +1472,7 @@ protected:
 
    void DisableLimiting()
    {
-      nodes0 = NULL; coeff0 = NULL; lim_dist = NULL;
+      lim_nodes0 = NULL; lim_coeff = NULL; lim_dist = NULL;
       delete lim_func; lim_func = NULL;
    }
 
@@ -1531,12 +1532,14 @@ public:
    TMOP_Integrator(TMOP_QualityMetric *m, TargetConstructor *tc,
                    TMOP_QualityMetric *hm)
       : h_metric(hm), metric(m), targetC(tc), IntegRules(NULL),
-        integ_order(-1), coeff1(NULL), metric_normal(1.0),
-        nodes0(NULL), coeff0(NULL),
+        integ_order(-1), metric_coeff(NULL), metric_normal(1.0),
+        lim_nodes0(NULL), lim_coeff(NULL),
         lim_dist(NULL), lim_func(NULL), lim_normal(1.0),
-        zeta_0(NULL), zeta(NULL), coeff_zeta(NULL), adapt_eval(NULL),
-        sigma(NULL), sigma_bar(NULL), sigma_marker(NULL), coeff_sigma(NULL),
-        sigma_eval(NULL), sigma_normal(1.0),
+        adapt_lim_gf0(NULL), adapt_lim_gf(NULL), adapt_lim_coeff(NULL),
+        adapt_lim_eval(NULL),
+        surf_fit_gf(NULL), surf_fit_gf_bar(NULL), surf_fit_marker(NULL),
+        surf_fit_coeff(NULL),
+        surf_fit_eval(NULL), surf_fit_normal(1.0),
         discr_tc(dynamic_cast<DiscreteAdaptTC *>(tc)),
         fdflag(false), dxscale(1.0e3), fd_call_flag(false), exact_action(false)
    { PA.enabled = false; }
@@ -1564,7 +1567,7 @@ public:
 
        Note that the Coefficient is evaluated in the physical configuration and
        not in the target configuration which may be undefined. */
-   void SetCoefficient(Coefficient &w1) { coeff1 = &w1; }
+   void SetCoefficient(Coefficient &w1) { metric_coeff = &w1; }
 
    /** @brief Limiting of the mesh displacements (general version).
 
@@ -1631,7 +1634,7 @@ public:
    void GetSurfaceFittingErrors(double &err_avg, double &err_max);
 
    /// Update the original/reference nodes used for limiting.
-   void SetLimitingNodes(const GridFunction &n0) { nodes0 = &n0; }
+   void SetLimitingNodes(const GridFunction &n0) { lim_nodes0 = &n0; }
 
    /** @brief Computes the integral of W(Jacobian(Trt)) over a target zone.
        @param[in] el     Type of FiniteElement.

fem/tmop/tmop_pa.cpp

@@ -38,13 +38,13 @@ void TMOP_Integrator::AssembleGradPA(const Vector &xe,
    if (PA.dim == 2)
    {
       AssembleGradPA_2D(xe);
-      if (coeff0) { AssembleGradPA_C0_2D(xe); }
+      if (lim_coeff) { AssembleGradPA_C0_2D(xe); }
    }
 
    if (PA.dim == 3)
    {
       AssembleGradPA_3D(xe);
-      if (coeff0) { AssembleGradPA_C0_3D(xe); }
+      if (lim_coeff) { AssembleGradPA_C0_3D(xe); }
    }
 }
 
@@ -53,8 +53,8 @@ void TMOP_Integrator::AssemblePA_Limiting()
    const MemoryType mt = (pa_mt == MemoryType::DEFAULT) ?
                          Device::GetDeviceMemoryType() : pa_mt;
    // Return immediately if limiting is not enabled
-   if (coeff0 == nullptr) { return; }
-   MFEM_VERIFY(nodes0, "internal error");
+   if (lim_coeff == nullptr) { return; }
+   MFEM_VERIFY(lim_nodes0, "internal error");
 
    MFEM_VERIFY(PA.enabled, "AssemblePA_Limiting but PA is not enabled!");
    MFEM_VERIFY(lim_func, "No TMOP_LimiterFunction specification!")
@@ -68,14 +68,14 @@ void TMOP_Integrator::AssemblePA_Limiting()
 
    const ElementDofOrdering ordering = ElementDofOrdering::LEXICOGRAPHIC;
 
-   // H0 for coeff0, (dim x dim) Q-vector
+   // H0 for lim_coeff, (dim x dim) Q-vector
    PA.H0.UseDevice(true);
    PA.H0.SetSize(PA.dim * PA.dim * PA.nq * NE, mt);
 
-   // coeff0 -> PA.C0 (Q-vector)
+   // lim_coeff -> PA.C0 (Q-vector)
    PA.C0.UseDevice(true);
    if (ConstantCoefficient* cQ =
-          dynamic_cast<ConstantCoefficient*>(coeff0))
+          dynamic_cast<ConstantCoefficient*>(lim_coeff))
    {
       PA.C0.SetSize(1, Device::GetMemoryType());
       PA.C0.HostWrite();
@@ -90,17 +90,17 @@ void TMOP_Integrator::AssemblePA_Limiting()
          ElementTransformation& T = *fes->GetElementTransformation(e);
          for (int q = 0; q < ir.GetNPoints(); ++q)
          {
-            C0(q,e) = coeff0->Eval(T, ir.IntPoint(q));
+            C0(q,e) = lim_coeff->Eval(T, ir.IntPoint(q));
          }
       }
    }
 
-   // nodes0 -> PA.X0 (E-vector)
-   MFEM_VERIFY(nodes0->FESpace() == fes, "");
+   // lim_nodes0 -> PA.X0 (E-vector)
+   MFEM_VERIFY(lim_nodes0->FESpace() == fes, "");
    const Operator *n0_R = fes->GetElementRestriction(ordering);
    PA.X0.SetSize(n0_R->Height(), Device::GetMemoryType());
    PA.X0.UseDevice(true);
-   n0_R->Mult(*nodes0, PA.X0);
+   n0_R->Mult(*lim_nodes0, PA.X0);
 
    // Limiting distances: lim_dist -> PA.LD (E-vector)
    // TODO: remove the hack for the case lim_dist == NULL.
@@ -217,8 +217,8 @@ void TMOP_Integrator::AssemblePA(const FiniteElementSpace &fes)
    PA.Jtr_needs_update = true;
    PA.Jtr_debug_grad = false;
 
-   // Limiting: coeff0 -> PA.C0, nodes0 -> PA.X0, lim_dist -> PA.LD, PA.H0
-   if (coeff0) { AssemblePA_Limiting(); }
+   // Limiting: lim_coeff -> PA.C0, lim_nodes0 -> PA.X0, lim_dist -> PA.LD, PA.H0
+   if (lim_coeff) { AssemblePA_Limiting(); }
 }
 
 void TMOP_Integrator::AssembleGradDiagonalPA(Vector &de) const
@@ -236,13 +236,13 @@ void TMOP_Integrator::AssembleGradDiagonalPA(Vector &de) const
    if (PA.dim == 2)
    {
       AssembleDiagonalPA_2D(de);
-      if (coeff0) { AssembleDiagonalPA_C0_2D(de); }
+      if (lim_coeff) { AssembleDiagonalPA_C0_2D(de); }
    }
 
    if (PA.dim == 3)
    {
       AssembleDiagonalPA_3D(de);
-      if (coeff0) { AssembleDiagonalPA_C0_3D(de); }
+      if (lim_coeff) { AssembleDiagonalPA_C0_3D(de); }
    }
 }
 
@@ -258,13 +258,13 @@ void TMOP_Integrator::AddMultPA(const Vector &xe, Vector &ye) const
    if (PA.dim == 2)
    {
       AddMultPA_2D(xe,ye);
-      if (coeff0) { AddMultPA_C0_2D(xe,ye); }
+      if (lim_coeff) { AddMultPA_C0_2D(xe,ye); }
    }
 
    if (PA.dim == 3)
    {
       AddMultPA_3D(xe,ye);
-      if (coeff0) { AddMultPA_C0_3D(xe,ye); }
+      if (lim_coeff) { AddMultPA_C0_3D(xe,ye); }
    }
 }
 
@@ -283,13 +283,13 @@ void TMOP_Integrator::AddMultGradPA(const Vector &re, Vector &ce) const
    if (PA.dim == 2)
    {
       AddMultGradPA_2D(re,ce);
-      if (coeff0) { AddMultGradPA_C0_2D(re,ce); }
+      if (lim_coeff) { AddMultGradPA_C0_2D(re,ce); }
    }
 
    if (PA.dim == 3)
    {
       AddMultGradPA_3D(re,ce);
-      if (coeff0) { AddMultGradPA_C0_3D(re,ce); }
+      if (lim_coeff) { AddMultGradPA_C0_3D(re,ce); }
    }
 }
 
@@ -307,13 +307,13 @@ double TMOP_Integrator::GetLocalStateEnergyPA(const Vector &xe) const
    if (PA.dim == 2)
    {
       energy = GetLocalStateEnergyPA_2D(xe);
-      if (coeff0) { energy += GetLocalStateEnergyPA_C0_2D(xe); }
+      if (lim_coeff) { energy += GetLocalStateEnergyPA_C0_2D(xe); }
    }
 
    if (PA.dim == 3)
    {
       energy = GetLocalStateEnergyPA_3D(xe);
-      if (coeff0) { energy += GetLocalStateEnergyPA_C0_3D(xe); }
+      if (lim_coeff) { energy += GetLocalStateEnergyPA_C0_3D(xe); }
    }
 
    return energy;

fem/tmop/tmop_pa.hpp

@@ -181,9 +181,9 @@ template<int T_D1D, int T_Q1D, int T_MAX> return_t kernel(__VA_ARGS__)
 if (K##kernel.Find(id)) { return K##kernel.At(id)(__VA_ARGS__,0,0); }\
 else {\
    constexpr int T_MAX = 4;\
-   const int D1D = (id>>4)&0xF, Q1D = id&0xF;\
-   MFEM_VERIFY(D1D <= MAX_D1D && Q1D <= MAX_Q1D, "Max size error!");\
-   return kernel<0,0,T_MAX>(__VA_ARGS__,D1D,Q1D); }
+   const int d1d = (id>>4)&0xF, q1d = id&0xF;\
+   MFEM_VERIFY(d1d <= MAX_D1D && q1d <= MAX_Q1D, "Max size error!");\
+   return kernel<0,0,T_MAX>(__VA_ARGS__,d1d,q1d); }
 
 } // namespace kernels
 

fem/tmop/tmop_pa_da3.cpp

@@ -55,6 +55,8 @@ MFEM_REGISTER_TMOP_KERNELS(void, DatcSize,
       MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
       MFEM_SHARED double sm1[MDQ*MDQ*MDQ];
 
+      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,sB);
+
       ConstDeviceMatrix B(sB, D1D, Q1D);
       DeviceCube DDD(sm0, MD1,MD1,MD1);
       DeviceCube DDQ(sm1, MD1,MD1,MQ1);
@@ -88,7 +90,6 @@ MFEM_REGISTER_TMOP_KERNELS(void, DatcSize,
       }
       min = min_size[0];
 
-      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,sB);
       kernels::internal::EvalX(D1D,Q1D,B,DDD,DDQ);
       kernels::internal::EvalY(D1D,Q1D,B,DDQ,DQQ);
       kernels::internal::EvalZ(D1D,Q1D,B,DQQ,QQQ);

fem/tmop/tmop_pa_h2d.cpp

@@ -97,9 +97,9 @@ MFEM_REGISTER_TMOP_KERNELS(void, AssembleDiagonalPA_Kernel_2D,
                   const double *Jtr = &J(0,0,qx,qy,e);
 
                   // Jrt = Jtr^{-1}
-                  double j[4];
-                  ConstDeviceMatrix Jrt(j,2,2);
-                  kernels::CalcInverse<2>(Jtr, j);
+                  double jrt_data[4];
+                  ConstDeviceMatrix Jrt(jrt_data,2,2);
+                  kernels::CalcInverse<2>(Jtr, jrt_data);
 
                   const double gg = G(qy,dy) * G(qy,dy);
                   const double gb = G(qy,dy) * B(qy,dy);

fem/tmop/tmop_pa_h2m_c0.cpp

@@ -68,8 +68,8 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultGradPA_Kernel_C0_2D,
             double Xh[2];
             kernels::internal::PullEval<MQ1,NBZ>(Q1D,qx,qy,QQ,Xh);
 
-            double B[4];
-            DeviceMatrix H(B,2,2);
+            double H_data[4];
+            DeviceMatrix H(H_data,2,2);
             for (int i = 0; i < DIM; i++)
             {
                for (int j = 0; j < DIM; j++)
@@ -78,9 +78,9 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultGradPA_Kernel_C0_2D,
                }
             }
 
-            // p2 = B . Xh
+            // p2 = H . Xh
             double p2[2];
-            kernels::Mult(2,2,B,Xh,p2);
+            kernels::Mult(2,2,H_data,Xh,p2);
             kernels::internal::PushEval<MQ1,NBZ>(Q1D,qx,qy,p2,QQ);
          }
       }

fem/tmop/tmop_pa_h3m_c0.cpp

@@ -70,8 +70,8 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultGradPA_Kernel_C0_3D,
                double Xh[3];
                kernels::internal::PullEval<MQ1>(Q1D,qx,qy,qz,QQQ,Xh);
 
-               double B[9];
-               DeviceMatrix H(B,3,3);
+               double H_data[9];
+               DeviceMatrix H(H_data,3,3);
                for (int i = 0; i < DIM; i++)
                {
                   for (int j = 0; j < DIM; j++)
@@ -80,9 +80,9 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultGradPA_Kernel_C0_3D,
                   }
                }
 
-               // p2 = B . Xh
+               // p2 = H . Xh
                double p2[3];
-               kernels::Mult(3,3,B,Xh,p2);
+               kernels::Mult(3,3,H_data,Xh,p2);
                kernels::internal::PushEval<MQ1>(Q1D,qx,qy,qz,p2,QQQ);
             }
          }

fem/tmop/tmop_pa_h3s_c0.cpp

@@ -55,6 +55,7 @@ MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_Kernel_C0_3D,
       constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
 
       MFEM_SHARED double sBLD[MQ1*MD1];
+      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,bld,sBLD);
       ConstDeviceMatrix BLD(sBLD, D1D, Q1D);
 
       MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
@@ -66,8 +67,6 @@ MFEM_REGISTER_TMOP_KERNELS(void, SetupGradPA_Kernel_C0_3D,
 
       kernels::internal::LoadX(e,D1D,LD,DDD);
 
-      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,bld,sBLD);
-
       kernels::internal::EvalX(D1D,Q1D,BLD,DDD,DDQ);
       kernels::internal::EvalY(D1D,Q1D,BLD,DDQ,DQQ);
       kernels::internal::EvalZ(D1D,Q1D,BLD,DQQ,QQQ);

fem/tmop/tmop_pa_p3_c0.cpp

@@ -62,6 +62,7 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultPA_Kernel_C0_3D,
 
       MFEM_SHARED double B[MQ1*MD1];
       MFEM_SHARED double sBLD[MQ1*MD1];
+      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,bld,sBLD);
       ConstDeviceMatrix BLD(sBLD, D1D, Q1D);
 
       MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
@@ -86,7 +87,6 @@ MFEM_REGISTER_TMOP_KERNELS(void, AddMultPA_Kernel_C0_3D,
       kernels::internal::LoadX<MD1>(e,D1D,X1,DDD1);
 
       kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,B);
-      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,bld,sBLD);
 
       kernels::internal::EvalX(D1D,Q1D,BLD,DDD,DDQ);
       kernels::internal::EvalY(D1D,Q1D,BLD,DDQ,DQQ);

fem/tmop/tmop_pa_w3_c0.cpp

@@ -63,6 +63,7 @@ MFEM_REGISTER_TMOP_KERNELS(double, EnergyPA_C0_3D,
 
       MFEM_SHARED double B[MQ1*MD1];
       MFEM_SHARED double sBLD[MQ1*MD1];
+      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,bld,sBLD);
       ConstDeviceMatrix BLD(sBLD, D1D, Q1D);
 
       MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
@@ -87,7 +88,6 @@ MFEM_REGISTER_TMOP_KERNELS(double, EnergyPA_C0_3D,
       kernels::internal::LoadX<MD1>(e,D1D,X1,DDD1);
 
       kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,b,B);
-      kernels::internal::LoadB<MD1,MQ1>(D1D,Q1D,bld,sBLD);
 
       kernels::internal::EvalX(D1D,Q1D,BLD,DDD,DDQ);
       kernels::internal::EvalY(D1D,Q1D,BLD,DDQ,DQQ);

fem/tmop_amr.cpp

@@ -837,26 +837,26 @@ void TMOPHRSolver::ParUpdate()
 }
 #endif
 
-void TMOPHRSolver::UpdateNonlinearFormAndBC(Mesh *mesh, NonlinearForm *nlf)
+void TMOPHRSolver::UpdateNonlinearFormAndBC(Mesh *mesh_, NonlinearForm *nlf_)
 {
-   const FiniteElementSpace &fes = *mesh->GetNodalFESpace();
+   const FiniteElementSpace &fes = *mesh_->GetNodalFESpace();
 
    // Update Nonlinear form and Set Essential BC
-   nlf->Update();
+   nlf_->Update();
    const int dim = fes.GetFE(0)->GetDim();
    if (move_bnd == false)
    {
-      Array<int> ess_bdr(mesh->bdr_attributes.Max());
+      Array<int> ess_bdr(mesh_->bdr_attributes.Max());
       ess_bdr = 1;
-      nlf->SetEssentialBC(ess_bdr);
+      nlf_->SetEssentialBC(ess_bdr);
    }
    else
    {
       const int nd  = fes.GetBE(0)->GetDof();
       int n = 0;
-      for (int i = 0; i < mesh->GetNBE(); i++)
+      for (int i = 0; i < mesh_->GetNBE(); i++)
       {
-         const int attr = mesh->GetBdrElement(i)->GetAttribute();
+         const int attr = mesh_->GetBdrElement(i)->GetAttribute();
          MFEM_VERIFY(!(dim == 2 && attr == 3),
                      "Boundary attribute 3 must be used only for 3D meshes. "
                      "Adjust the attributes (1/2/3/4 for fixed x/y/z/all "
@@ -866,9 +866,9 @@ void TMOPHRSolver::UpdateNonlinearFormAndBC(Mesh *mesh, NonlinearForm *nlf)
       }
       Array<int> ess_vdofs(n), vdofs;
       n = 0;
-      for (int i = 0; i < mesh->GetNBE(); i++)
+      for (int i = 0; i < mesh_->GetNBE(); i++)
       {
-         const int attr = mesh->GetBdrElement(i)->GetAttribute();
+         const int attr = mesh_->GetBdrElement(i)->GetAttribute();
          fes.GetBdrElementVDofs(i, vdofs);
          if (attr == 1) // Fix x components.
          {
@@ -891,7 +891,7 @@ void TMOPHRSolver::UpdateNonlinearFormAndBC(Mesh *mesh, NonlinearForm *nlf)
             { ess_vdofs[n++] = vdofs[j]; }
          }
       }
-      nlf->SetEssentialVDofs(ess_vdofs);
+      nlf_->SetEssentialVDofs(ess_vdofs);
    }
 }
 

fem/tmop_tools.cpp

@@ -444,15 +444,19 @@ double TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
       if (untangling == false && min_detT_out < 0.0)
       {
          // No untangling, and detJ got negative -- no good.
-         if (print_level >= 0)
-         { mfem::out << "Scale = " << scale << " Neg det(J) found.\n"; }
+         if (print_options.iterations)
+         {
+            mfem::out << "Scale = " << scale << " Neg det(J) found.\n";
+         }
          scale *= detJ_factor; continue;
       }
       if (untangling == true && min_detT_out < *min_det_ptr)
       {
          // Untangling, and detJ got even more negative -- no good.
-         if (print_level >= 0)
-         { mfem::out << "Scale = " << scale << " Neg det(J) decreased.\n"; }
+         if (print_options.iterations)
+         {
+            mfem::out << "Scale = " << scale << " Neg det(J) decreased.\n";
+         }
          scale *= detJ_factor; continue;
       }
 
@@ -477,7 +481,7 @@ double TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
       if (energy_out > energy_in + 0.2*fabs(energy_in) ||
           std::isnan(energy_out) != 0)
       {
-         if (print_level >= 0)
+         if (print_options.iterations)
          {
             mfem::out << "Scale = " << scale << " Increasing energy: "
                       << energy_in << " --> " << energy_out << '\n';
@@ -492,7 +496,7 @@ double TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
 
       if (norm_out > 1.2*norm_in)
       {
-         if (print_level >= 0)
+         if (print_options.iterations)
          {
             mfem::out << "Scale = " << scale << " Norm increased: "
                       << norm_in << " --> " << norm_out << '\n';
@@ -508,13 +512,15 @@ double TMOPNewtonSolver::ComputeScalingFactor(const Vector &x,
       if (min_detT_out > 0.0)
       {
          *min_det_ptr = 0.0;
-         if (print_level >= 0)
+         if (print_options.summary || print_options.iterations ||
+             print_options.first_and_last)
          { mfem::out << "The mesh has been untangled at the used points!\n"; }
       }
       else { *min_det_ptr = untangle_factor * min_detT_out; }
    }
 
-   if (print_level >= 0)
+   if (print_options.summary || print_options.iterations ||
+       print_options.first_and_last)
    {
       if (untangling)
       {
@@ -568,9 +574,9 @@ void TMOPNewtonSolver::ProcessNewState(const Vector &x) const
    if (parallel)
    {
 #ifdef MFEM_USE_MPI
-      const ParNonlinearForm *nlf =
+      const ParNonlinearForm *pnlf =
          dynamic_cast<const ParNonlinearForm *>(oper);
-      const ParFiniteElementSpace *pfesc = nlf->ParFESpace();
+      const ParFiniteElementSpace *pfesc = pnlf->ParFESpace();
       Vector x_loc(pfesc->GetVSize());
       pfesc->GetProlongationMatrix()->Mult(x, x_loc);
       for (int i = 0; i < integs.Size(); i++)

fem/transfer.cpp

@@ -903,7 +903,8 @@ TransferOperator::TransferOperator(const FiniteElementSpace& lFESpace_,
                                    const FiniteElementSpace& hFESpace_)
    : Operator(hFESpace_.GetVSize(), lFESpace_.GetVSize())
 {
-   if (lFESpace_.FEColl() == hFESpace_.FEColl())
+   bool isvar_order = lFESpace_.IsVariableOrder() || hFESpace_.IsVariableOrder();
+   if (lFESpace_.FEColl() == hFESpace_.FEColl() && !isvar_order)
    {
       OperatorPtr P(Operator::ANY_TYPE);
       hFESpace_.GetTransferOperator(lFESpace_, P);
@@ -912,8 +913,11 @@ TransferOperator::TransferOperator(const FiniteElementSpace& lFESpace_,
    }
    else if (lFESpace_.GetMesh()->GetNE() > 0
             && hFESpace_.GetMesh()->GetNE() > 0
+            && lFESpace_.GetVDim() == 1
+            && hFESpace_.GetVDim() == 1
             && dynamic_cast<const TensorBasisElement*>(lFESpace_.GetFE(0))
             && dynamic_cast<const TensorBasisElement*>(hFESpace_.GetFE(0))
+            && !isvar_order
             && (hFESpace_.FEColl()->GetContType() ==
                 mfem::FiniteElementCollection::CONTINUOUS ||
                 hFESpace_.FEColl()->GetContType() ==
@@ -945,6 +949,7 @@ PRefinementTransferOperator::PRefinementTransferOperator(
    : Operator(hFESpace_.GetVSize(), lFESpace_.GetVSize()), lFESpace(lFESpace_),
      hFESpace(hFESpace_)
 {
+   isvar_order = lFESpace_.IsVariableOrder() || hFESpace_.IsVariableOrder();
 }
 
 PRefinementTransferOperator::~PRefinementTransferOperator() {}
@@ -969,7 +974,7 @@ void PRefinementTransferOperator::Mult(const Vector& x, Vector& y) const
       DofTransformation * doftrans_l = lFESpace.GetElementDofs(i, l_dofs);
 
       const Geometry::Type geom = mesh->GetElementBaseGeometry(i);
-      if (geom != cached_geom)
+      if (geom != cached_geom || isvar_order)
       {
          h_fe = hFESpace.GetFE(i);
          l_fe = lFESpace.GetFE(i);
@@ -1026,7 +1031,7 @@ void PRefinementTransferOperator::MultTranspose(const Vector& x,
       DofTransformation * doftrans_l = lFESpace.GetElementDofs(i, l_dofs);
 
       const Geometry::Type geom = mesh->GetElementBaseGeometry(i);
-      if (geom != cached_geom)
+      if (geom != cached_geom || isvar_order)
       {
          h_fe = hFESpace.GetFE(i);
          l_fe = lFESpace.GetFE(i);
@@ -1424,20 +1429,36 @@ void TensorProductPRefinementTransferOperator::MultTranspose(const Vector& x,
    elem_restrict_lex_l->MultTranspose(localL, y);
 }
 
-#ifdef MFEM_USE_MPI
-TrueTransferOperator::TrueTransferOperator(const
-                                           ParFiniteElementSpace& lFESpace_,
-                                           const ParFiniteElementSpace& hFESpace_)
+
+TrueTransferOperator::TrueTransferOperator(const FiniteElementSpace& lFESpace_,
+                                           const FiniteElementSpace& hFESpace_)
    : Operator(hFESpace_.GetTrueVSize(), lFESpace_.GetTrueVSize()),
      lFESpace(lFESpace_),
      hFESpace(hFESpace_)
 {
    localTransferOperator = new TransferOperator(lFESpace_, hFESpace_);
 
-   tmpL.SetSize(lFESpace_.GetVSize());
-   tmpH.SetSize(hFESpace_.GetVSize());
+   P = lFESpace.GetProlongationMatrix();
+   R = hFESpace.IsVariableOrder() ? hFESpace.GetHpRestrictionMatrix() :
+       hFESpace.GetRestrictionMatrix();
 
-   hFESpace.GetRestrictionMatrix()->BuildTranspose();
+   // P and R can be both null
+   // P can be null and R not null
+   // If P is not null it is assumed that R is not null as well
+   if (P) { MFEM_VERIFY(R, "Both P and R have to be not NULL") }
+
+   if (P)
+   {
+      tmpL.SetSize(lFESpace_.GetVSize());
+      tmpH.SetSize(hFESpace_.GetVSize());
+      R->BuildTranspose();
+   }
+   // P can be null and R not null
+   else if (R)
+   {
+      tmpH.SetSize(hFESpace_.GetVSize());
+      R->BuildTranspose();
+   }
 }
 
 TrueTransferOperator::~TrueTransferOperator()
@@ -1447,17 +1468,40 @@ TrueTransferOperator::~TrueTransferOperator()
 
 void TrueTransferOperator::Mult(const Vector& x, Vector& y) const
 {
-   lFESpace.GetProlongationMatrix()->Mult(x, tmpL);
-   localTransferOperator->Mult(tmpL, tmpH);
-   hFESpace.GetRestrictionMatrix()->Mult(tmpH, y);
+   if (P)
+   {
+      P->Mult(x, tmpL);
+      localTransferOperator->Mult(tmpL, tmpH);
+      R->Mult(tmpH, y);
+   }
+   else if (R)
+   {
+      localTransferOperator->Mult(x, tmpH);
+      R->Mult(tmpH, y);
+   }
+   else
+   {
+      localTransferOperator->Mult(x, y);
+   }
 }
 
 void TrueTransferOperator::MultTranspose(const Vector& x, Vector& y) const
 {
-   hFESpace.GetRestrictionMatrix()->MultTranspose(x, tmpH);
-   localTransferOperator->MultTranspose(tmpH, tmpL);
-   lFESpace.GetProlongationMatrix()->MultTranspose(tmpL, y);
+   if (P)
+   {
+      R->MultTranspose(x, tmpH);
+      localTransferOperator->MultTranspose(tmpH, tmpL);
+      P->MultTranspose(tmpL, y);
+   }
+   else if (R)
+   {
+      R->MultTranspose(x, tmpH);
+      localTransferOperator->MultTranspose(tmpH, y);
+   }
+   else
+   {
+      localTransferOperator->MultTranspose(x, y);
+   }
 }
-#endif
 
 } // namespace mfem

fem/transfer.hpp

@@ -387,6 +387,7 @@ class PRefinementTransferOperator : public Operator
 private:
    const FiniteElementSpace& lFESpace;
    const FiniteElementSpace& hFESpace;
+   bool isvar_order;
 
 public:
    /// @brief Constructs a transfer operator from \p lFESpace to \p hFESpace
@@ -452,14 +453,15 @@ public:
    virtual void MultTranspose(const Vector& x, Vector& y) const override;
 };
 
-#ifdef MFEM_USE_MPI
 /// @brief Matrix-free transfer operator between finite element spaces working
 /// on true degrees of freedom
 class TrueTransferOperator : public Operator
 {
 private:
-   const ParFiniteElementSpace& lFESpace;
-   const ParFiniteElementSpace& hFESpace;
+   const FiniteElementSpace& lFESpace;
+   const FiniteElementSpace& hFESpace;
+   const Operator * P = nullptr;
+   const SparseMatrix * R = nullptr;
    TransferOperator* localTransferOperator;
    mutable Vector tmpL;
    mutable Vector tmpH;
@@ -467,8 +469,8 @@ private:
 public:
    /// @brief Constructs a transfer operator working on true degrees of freedom
    /// from \p lFESpace to \p hFESpace
-   TrueTransferOperator(const ParFiniteElementSpace& lFESpace_,
-                        const ParFiniteElementSpace& hFESpace_);
+   TrueTransferOperator(const FiniteElementSpace& lFESpace_,
+                        const FiniteElementSpace& hFESpace_);
 
    /// Destructor
    ~TrueTransferOperator();
@@ -484,7 +486,6 @@ public:
        the true dof vector \p y corresponding to the coarse space. */
    virtual void MultTranspose(const Vector& x, Vector& y) const override;
 };
-#endif
 
 } // namespace mfem
 

general/binaryio.hpp

@@ -58,16 +58,19 @@ void AppendBytes(std::vector<char> &vec, const T &val)
    vec.insert(vec.end(), ptr, ptr + sizeof(T));
 }
 
-/// Given a buffer @a buf of length @a nbytes, encode the data in base-64
-/// format, and write the encoded data to the output stream @a out.
+/// @brief Given a buffer @a bytes of length @a nbytes, encode the data in
+/// base-64 format, and write the encoded data to the output stream @a out.
 void WriteBase64(std::ostream &out, const void *bytes, size_t nbytes);
 
-/// Decode @a len base-64 encoded characters in the buffer @a src, and store the
-/// resulting decoded data in @a buf. @a buf will be resized as needed.
+/// @brief Decode @a len base-64 encoded characters in the buffer @a src, and
+/// store the resulting decoded data in @a buf. @a buf will be resized as
+/// needed.
 void DecodeBase64(const char *src, size_t len, std::vector<char> &buf);
 
-/// Return the number of characters needed to encode @a nbytes in base-64. This
-/// is equal to 4*nbytes/3, rounded up to the nearest multiple of 4.
+/// @brief Return the number of characters needed to encode @a nbytes in
+/// base-64.
+///
+/// This is equal to 4*nbytes/3, rounded up to the nearest multiple of 4.
 size_t NumBase64Chars(size_t nbytes);
 
 } // namespace mfem::bin_io

general/cuda.hpp

@@ -19,6 +19,7 @@
 #define MFEM_CUDA_BLOCKS 256
 
 #ifdef MFEM_USE_CUDA
+#define MFEM_USE_CUDA_OR_HIP
 #define MFEM_DEVICE __device__
 #define MFEM_LAMBDA __host__
 #define MFEM_HOST_DEVICE __host__ __device__

general/device.cpp

@@ -177,7 +177,7 @@ Device::~Device()
    Get().device_mem_class = MemoryClass::HOST;
 }
 
-void Device::Configure(const std::string &device, const int dev)
+void Device::Configure(const std::string &device, const int device_id)
 {
    // If a device was configured via the environment, skip the configuration,
    // and avoid the 'singleton_device' to destroy the mm.
@@ -240,7 +240,7 @@ void Device::Configure(const std::string &device, const int dev)
 #endif
 
    // Perform setup.
-   Get().Setup(dev);
+   Get().Setup(device_id);
 
    // Enable the device
    Enable();
@@ -276,35 +276,35 @@ void Device::SetMemoryTypes(MemoryType h_mt, MemoryType d_mt)
    // the call mm.Configure(...) in UpdateMemoryTypeAndClass()
 }
 
-void Device::Print(std::ostream &out)
+void Device::Print(std::ostream &os)
 {
-   out << "Device configuration: ";
+   os << "Device configuration: ";
    bool add_comma = false;
    for (int i = 0; i < Backend::NUM_BACKENDS; i++)
    {
       if (backends & internal::backend_list[i])
       {
-         if (add_comma) { out << ','; }
+         if (add_comma) { os << ','; }
          add_comma = true;
-         out << internal::backend_name[i];
+         os << internal::backend_name[i];
       }
    }
-   out << '\n';
+   os << '\n';
 #ifdef MFEM_USE_CEED
    if (Allows(Backend::CEED_MASK))
    {
       const char *ceed_backend;
       CeedGetResource(internal::ceed, &ceed_backend);
-      out << "libCEED backend: " << ceed_backend << '\n';
+      os << "libCEED backend: " << ceed_backend << '\n';
    }
 #endif
-   out << "Memory configuration: "
-       << MemoryTypeName[static_cast<int>(host_mem_type)];
+   os << "Memory configuration: "
+      << MemoryTypeName[static_cast<int>(host_mem_type)];
    if (Device::Allows(Backend::DEVICE_MASK))
    {
-      out << ',' << MemoryTypeName[static_cast<int>(device_mem_type)];
+      os << ',' << MemoryTypeName[static_cast<int>(device_mem_type)];
    }
-   out << std::endl;
+   os << std::endl;
 }
 
 void Device::UpdateMemoryTypeAndClass()
@@ -502,12 +502,12 @@ static void CeedDeviceSetup(const char* ceed_spec)
 #endif
 }
 
-void Device::Setup(const int device)
+void Device::Setup(const int device_id)
 {
    MFEM_VERIFY(ngpu == -1, "the mfem::Device is already configured!");
 
    ngpu = 0;
-   dev = device;
+   dev = device_id;
 #ifndef MFEM_USE_CUDA
    MFEM_VERIFY(!Allows(Backend::CUDA_MASK),
                "the CUDA backends require MFEM built with MFEM_USE_CUDA=YES");

general/device.hpp

@@ -150,7 +150,7 @@ private:
    static Device& Get() { return device_singleton; }
 
    /// Setup switcher based on configuration settings
-   void Setup(const int dev = 0);
+   void Setup(const int device_id = 0);
 
    void MarkBackend(Backend::Id b) { backends |= b; }
 

general/forall.hpp

@@ -42,6 +42,19 @@ const int MAX_Q1D = 14;
 #define MFEM_UNROLL(N)
 #endif
 
+// MFEM_GPU_FORALL: "parallel for" executed with CUDA or HIP based on the MFEM
+// build-time configuration (MFEM_USE_CUDA or MFEM_USE_HIP). If neither CUDA nor
+// HIP is enabled, this macro is a no-op.
+#if defined(MFEM_USE_CUDA)
+#define MFEM_GPU_FORALL(i, N,...) CuWrap1D(N, [=] MFEM_DEVICE      \
+                                       (int i) {__VA_ARGS__})
+#elif defined(MFEM_USE_HIP)
+#define MFEM_GPU_FORALL(i, N,...) HipWrap1D(N, [=] MFEM_DEVICE     \
+                                        (int i) {__VA_ARGS__})
+#else
+#define MFEM_GPU_FORALL(i, N,...) do { } while (false)
+#endif
+
 // Implementation of MFEM's "parallel for" (forall) device/host kernel
 // interfaces supporting RAJA, CUDA, OpenMP, and sequential backends.
 

general/gecko.cpp

@@ -1228,12 +1228,12 @@ Graph::reweight(uint k)
 
 // Linearly order graph.
 void
-Graph::order(Functional* functional, uint iterations, uint window, uint period,
-             uint seed, Progress* progress)
+Graph::order(Functional* functional_, uint iterations, uint window, uint period,
+             uint seed, Progress* progress_)
 {
    // Initialize graph.
-   this->functional = functional;
-   progress = this->progress = progress ? progress : new Progress;
+   this->functional = functional_;
+   progress_ = this->progress = progress_ ? progress_ : new Progress;
    for (level = 0; (1u << level) < nodes(); level++);
    place();
    Float mincost = cost();

general/hip.hpp

@@ -19,6 +19,7 @@
 #define MFEM_HIP_BLOCKS 256
 
 #ifdef MFEM_USE_HIP
+#define MFEM_USE_CUDA_OR_HIP
 #define MFEM_DEVICE __device__
 #define MFEM_LAMBDA __host__ __device__
 #define MFEM_HOST_DEVICE __host__ __device__

general/mem_manager.cpp

@@ -470,7 +470,10 @@ public:
    void *HtoD(void *dst, const void *src, size_t bytes)
    { return HipMemcpyHtoD(dst, src, bytes); }
    void *DtoD(void* dst, const void* src, size_t bytes)
-   { return HipMemcpyDtoD(dst, src, bytes); }
+   // Unlike cudaMemcpy(DtoD), hipMemcpy(DtoD) causes a host-side synchronization so
+   // instead we use hipMemcpyAsync to get similar behavior.
+   // for more info see: https://github.com/mfem/mfem/pull/2780
+   { return HipMemcpyDtoDAsync(dst, src, bytes); }
    void *DtoH(void *dst, const void *src, size_t bytes)
    { return HipMemcpyDtoH(dst, src, bytes); }
 };
@@ -593,7 +596,10 @@ public:
       return CuMemcpyDtoD(dst, src, bytes);
 #endif
 #ifdef MFEM_USE_HIP
-      return HipMemcpyDtoD(dst, src, bytes);
+      // Unlike cudaMemcpy(DtoD), hipMemcpy(DtoD) causes a host-side synchronization so
+      // instead we use hipMemcpyAsync to get similar behavior.
+      // for more info see: https://github.com/mfem/mfem/pull/2780
+      return HipMemcpyDtoDAsync(dst, src, bytes);
 #endif
       // rm.copy(dst, const_cast<void*>(src), bytes); return dst;
    }

general/mem_manager.hpp

@@ -18,7 +18,7 @@
 #include <type_traits> // std::is_const
 #include <cstddef> // std::max_align_t
 #ifdef MFEM_USE_MPI
-#include <HYPRE_config.h> // HYPRE_USING_CUDA
+#include <HYPRE_config.h> // HYPRE_USING_GPU
 #endif
 
 namespace mfem
@@ -849,13 +849,14 @@ inline void Memory<T>::New(int size, MemoryType mt)
 }
 
 template <typename T>
-inline void Memory<T>::New(int size, MemoryType h_mt, MemoryType d_mt)
+inline void Memory<T>::New(int size, MemoryType host_mt, MemoryType device_mt)
 {
    capacity = size;
    const size_t bytes = size*sizeof(T);
-   this->h_mt = h_mt;
-   T *h_tmp = (h_mt == MemoryType::HOST) ? NewHOST(size) : nullptr;
-   h_ptr = (T*)MemoryManager::New_(h_tmp, bytes, h_mt, d_mt, VALID_HOST, flags);
+   this->h_mt = host_mt;
+   T *h_tmp = (host_mt == MemoryType::HOST) ? NewHOST(size) : nullptr;
+   h_ptr = (T*)MemoryManager::New_(h_tmp, bytes, host_mt, device_mt,
+                                   VALID_HOST, flags);
 }
 
 template <typename T>
@@ -933,12 +934,12 @@ inline void Memory<T>::MakeAlias(const Memory &base, int offset, int size)
    if (!(base.flags & REGISTERED))
    {
       if (
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
          // If the following condition is true then MemoryManager::Exists()
          // should also be true:
          IsDeviceMemory(MemoryManager::GetDeviceMemoryType())
 #else
-         // When HYPRE_USING_CUDA is defined we always register the 'base' if
+         // When HYPRE_USING_GPU is defined we always register the 'base' if
          // the MemoryManager::Exists():
          MemoryManager::Exists()
 #endif

general/tic_toc.hpp

@@ -42,7 +42,7 @@ public:
    /// Clear the elapsed time on the stopwatch and restart it if it's running.
    void Clear();
 
-   /// Clear the elapsed time and start the stopwatch.
+   /// Start the stopwatch. The elapsed time is @b not cleared.
    void Start();
 
    /// Stop the stopwatch.

linalg/constraints.cpp

@@ -327,7 +327,7 @@ void EliminationSolver::Mult(const Vector& rhs, Vector& sol) const
    krylov->SetMaxIter(max_iter);
    krylov->SetRelTol(rel_tol);
    krylov->SetAbsTol(abs_tol);
-   krylov->SetPrintLevel(print_level);
+   krylov->SetPrintLevel(print_options);
 
    Vector rtilde(rhs.Size());
    if (constraint_rhs.Size() > 0)
@@ -447,7 +447,7 @@ void PenaltyConstrainedSolver::Mult(const Vector& b, Vector& x) const
    krylov->SetRelTol(rel_tol);
    krylov->SetAbsTol(abs_tol);
    krylov->SetMaxIter(max_iter);
-   krylov->SetPrintLevel(print_level);
+   krylov->SetPrintLevel(print_options);
    krylov->Mult(penalized_rhs, x);
    final_iter = krylov->GetNumIterations();
    final_norm = krylov->GetFinalNorm();
@@ -574,7 +574,7 @@ void SchurConstrainedSolver::LagrangeSystemMult(const Vector& x,
    gmres->SetRelTol(rel_tol);
    gmres->SetAbsTol(abs_tol);
    gmres->SetMaxIter(max_iter);
-   gmres->SetPrintLevel(print_level);
+   gmres->SetPrintLevel(print_options);
    gmres->SetPreconditioner(
       const_cast<BlockDiagonalPreconditioner&>(*block_pc));
 

linalg/hypre.cpp

@@ -65,7 +65,7 @@ bool CanShallowCopy(const Memory<T> &src, MemoryClass mc)
 inline void HypreParVector::_SetDataAndSize_()
 {
    hypre_Vector *x_loc = hypre_ParVectorLocalVector(x);
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
    SetDataAndSize(hypre_VectorData(x_loc),
                   internal::to_int(hypre_VectorSize(x_loc)));
 #else
@@ -112,7 +112,7 @@ HypreParVector::HypreParVector(MPI_Comm comm, HYPRE_BigInt glob_size,
 #endif
    double tmp = 0.0;
    hypre_VectorData(x_loc) = &tmp;
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_VectorMemoryLocation(x_loc) =
       is_device_ptr ? HYPRE_MEMORY_DEVICE : HYPRE_MEMORY_HOST;
 #else
@@ -257,7 +257,7 @@ void HypreParVector::HypreRead() const
    hypre_Vector *x_loc = hypre_ParVectorLocalVector(x);
    hypre_VectorData(x_loc) =
       const_cast<double*>(data.Read(GetHypreMemoryClass(), size));
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_VectorMemoryLocation(x_loc) = HYPRE_MEMORY_DEVICE;
 #endif
 }
@@ -266,7 +266,7 @@ void HypreParVector::HypreReadWrite()
 {
    hypre_Vector *x_loc = hypre_ParVectorLocalVector(x);
    hypre_VectorData(x_loc) = data.ReadWrite(GetHypreMemoryClass(), size);
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_VectorMemoryLocation(x_loc) = HYPRE_MEMORY_DEVICE;
 #endif
 }
@@ -275,7 +275,7 @@ void HypreParVector::HypreWrite()
 {
    hypre_Vector *x_loc = hypre_ParVectorLocalVector(x);
    hypre_VectorData(x_loc) = data.Write(GetHypreMemoryClass(), size);
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_VectorMemoryLocation(x_loc) = HYPRE_MEMORY_DEVICE;
 #endif
 }
@@ -289,7 +289,7 @@ void HypreParVector::WrapMemoryRead(const Memory<double> &mem)
    hypre_Vector *x_loc = hypre_ParVectorLocalVector(x);
    hypre_VectorData(x_loc) =
       const_cast<double*>(mem.Read(GetHypreMemoryClass(), size));
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_VectorMemoryLocation(x_loc) = HYPRE_MEMORY_DEVICE;
 #endif
    data.MakeAlias(mem, 0, size);
@@ -303,7 +303,7 @@ void HypreParVector::WrapMemoryReadWrite(Memory<double> &mem)
    data.Delete();
    hypre_Vector *x_loc = hypre_ParVectorLocalVector(x);
    hypre_VectorData(x_loc) = mem.ReadWrite(GetHypreMemoryClass(), size);
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_VectorMemoryLocation(x_loc) = HYPRE_MEMORY_DEVICE;
 #endif
    data.MakeAlias(mem, 0, size);
@@ -317,7 +317,7 @@ void HypreParVector::WrapMemoryWrite(Memory<double> &mem)
    data.Delete();
    hypre_Vector *x_loc = hypre_ParVectorLocalVector(x);
    hypre_VectorData(x_loc) = mem.Write(GetHypreMemoryClass(), size);
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_VectorMemoryLocation(x_loc) = HYPRE_MEMORY_DEVICE;
 #endif
    data.MakeAlias(mem, 0, size);
@@ -885,7 +885,7 @@ HypreParMatrix::HypreParMatrix(
    hypre_CSRMatrixJ(A->diag) = diag_j;
    hypre_CSRMatrixData(A->diag) = diag_data;
    hypre_CSRMatrixNumNonzeros(A->diag) = diag_i[local_num_rows];
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_CSRMatrixMemoryLocation(A->diag) = HYPRE_MEMORY_HOST;
 #endif
    hypre_CSRMatrixSetRownnz(A->diag);
@@ -895,7 +895,7 @@ HypreParMatrix::HypreParMatrix(
    hypre_CSRMatrixJ(A->offd) = offd_j;
    hypre_CSRMatrixData(A->offd) = offd_data;
    hypre_CSRMatrixNumNonzeros(A->offd) = offd_i[local_num_rows];
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_CSRMatrixMemoryLocation(A->offd) = HYPRE_MEMORY_HOST;
 #endif
    hypre_CSRMatrixSetRownnz(A->offd);
@@ -1070,7 +1070,7 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm, int id, int np,
    hypre_CSRMatrixI(A->diag)    = i_diag;
    hypre_CSRMatrixJ(A->diag)    = j_diag;
    hypre_CSRMatrixData(A->diag) = mem_diag.data;
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_CSRMatrixMemoryLocation(A->diag) = HYPRE_MEMORY_HOST;
 #endif
    hypre_CSRMatrixSetRownnz(A->diag);
@@ -1079,7 +1079,7 @@ HypreParMatrix::HypreParMatrix(MPI_Comm comm, int id, int np,
    hypre_CSRMatrixI(A->offd)    = i_offd;
    hypre_CSRMatrixJ(A->offd)    = j_offd;
    hypre_CSRMatrixData(A->offd) = mem_offd.data;
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    hypre_CSRMatrixMemoryLocation(A->offd) = HYPRE_MEMORY_HOST;
 #endif
    hypre_CSRMatrixSetRownnz(A->offd);
@@ -1306,7 +1306,7 @@ hypre_ParCSRMatrix* HypreParMatrix::StealData()
    MFEM_ASSERT(diagOwner == offdOwner, "");
    MFEM_ASSERT(ParCSROwner, "");
    hypre_ParCSRMatrix *R = A;
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    if (diagOwner == -1) { HostReadWrite(); }
    else { HypreReadWrite(); }
 #endif
@@ -1418,8 +1418,8 @@ void HypreParMatrix::GetDiag(Vector &diag) const
 {
    const int size = Height();
    diag.SetSize(size);
-#ifdef HYPRE_USING_CUDA
-   if (Device::Allows(Backend::CUDA_MASK))
+#ifdef HYPRE_USING_GPU
+   if (Device::Allows(Backend::CUDA_MASK | Backend::HIP_MASK))
    {
       MFEM_ASSERT(A->diag->memory_location == HYPRE_MEMORY_DEVICE, "");
       double *d_diag = diag.Write();
@@ -2447,7 +2447,7 @@ void HypreParMatrix::Destroy()
 
    if (A == NULL) { return; }
 
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    if (ParCSROwner && (diagOwner < 0 || offdOwner < 0))
    {
       // Put the "host" or "hypre" pointers in {i,j,data} of A->{diag,offd}, so
@@ -2590,7 +2590,7 @@ HypreParMatrix * ParMult(const HypreParMatrix *A, const HypreParMatrix *B,
                          bool own_matrix)
 {
    hypre_ParCSRMatrix * ab;
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    ab = hypre_ParCSRMatMat(*A, *B);
 #else
    ab = hypre_ParMatmul(*A,*B);
@@ -2611,7 +2611,7 @@ HypreParMatrix * RAP(const HypreParMatrix *A, const HypreParMatrix *P)
 {
    hypre_ParCSRMatrix * rap;
 
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    // FIXME: this way of computing Pt A P can completely eliminate zero rows
    //        from the sparsity pattern of the product which prevents
    //        EliminateZeroRows() from working correctly. This issue is observed
@@ -2658,7 +2658,7 @@ HypreParMatrix * RAP(const HypreParMatrix * Rt, const HypreParMatrix *A,
 {
    hypre_ParCSRMatrix * rap;
 
-#ifdef HYPRE_USING_CUDA
+#ifdef HYPRE_USING_GPU
    {
       hypre_ParCSRMatrix *Q = hypre_ParCSRMatMat(*A,*P);
       rap = hypre_ParCSRTMatMat(*Rt,Q);
@@ -3253,9 +3253,9 @@ void HypreSmoother::SetOperator(const Operator &op)
    }
    if (l1_norms && pos_l1_norms)
    {
-#ifdef HYPRE_USING_CUDA
+#if defined(HYPRE_USING_GPU)
       double *d_l1_norms = l1_norms;  // avoid *this capture
-      CuWrap1D(height, [=] MFEM_DEVICE (int i)
+      MFEM_GPU_FORALL(i, height,
       {
          d_l1_norms[i] = std::abs(d_l1_norms[i]);
       });
@@ -4458,7 +4458,7 @@ HypreBoomerAMG::HypreBoomerAMG(const HypreParMatrix &A) : HypreSolver(&A)
 
 void HypreBoomerAMG::SetDefaultOptions()
 {
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
    // AMG coarsening options:
    int coarsen_type = 10;   // 10 = HMIS, 8 = PMIS, 6 = Falgout, 0 = CLJP
    int agg_levels   = 1;    // number of aggressive coarsening levels
@@ -4727,8 +4727,8 @@ void HypreBoomerAMG::RecomputeRBMs()
 
 void HypreBoomerAMG::SetElasticityOptions(ParFiniteElementSpace *fespace)
 {
-#ifdef HYPRE_USING_CUDA
-   MFEM_ABORT("this method is not supported in hypre built with CUDA");
+#ifdef HYPRE_USING_GPU
+   MFEM_ABORT("this method is not supported in hypre built with GPU support");
 #endif
 
    // Save the finite element space to support multiple calls to SetOperator()
@@ -4908,7 +4908,7 @@ void HypreAMS::Init(ParFiniteElementSpace *edge_fespace)
    int rlx_sweeps       = 1;
    double rlx_weight    = 1.0;
    double rlx_omega     = 1.0;
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
    int amg_coarsen_type = 10;
    int amg_agg_levels   = 1;
    int amg_rlx_type     = 8;
@@ -5159,7 +5159,7 @@ void HypreADS::Init(ParFiniteElementSpace *face_fespace)
    int rlx_sweeps       = 1;
    double rlx_weight    = 1.0;
    double rlx_omega     = 1.0;
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
    int rlx_type         = 2;
    int amg_coarsen_type = 10;
    int amg_agg_levels   = 1;

linalg/hypre.hpp

@@ -32,9 +32,16 @@
 #error "MFEM does not work with HYPRE's complex numbers support"
 #endif
 
+#if defined(HYPRE_USING_GPU) && \
+    !(defined(HYPRE_USING_CUDA) || defined(HYPRE_USING_HIP))
+#error "Unsupported GPU build of HYPRE! Only CUDA and HIP builds are supported."
+#endif
 #if defined(HYPRE_USING_CUDA) && !defined(MFEM_USE_CUDA)
 #error "MFEM_USE_CUDA=YES is required when HYPRE is built with CUDA!"
 #endif
+#if defined(HYPRE_USING_HIP) && !defined(MFEM_USE_HIP)
+#error "MFEM_USE_HIP=YES is required when HYPRE is built with HIP!"
+#endif
 
 #include "sparsemat.hpp"
 #include "hypre_parcsr.hpp"
@@ -74,7 +81,7 @@ inline int to_int(HYPRE_Int i)
 /// The MemoryClass used by Hypre objects.
 inline constexpr MemoryClass GetHypreMemoryClass()
 {
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
    return MemoryClass::HOST;
 #elif defined(HYPRE_USING_UNIFIED_MEMORY)
    return MemoryClass::MANAGED;
@@ -86,7 +93,7 @@ inline constexpr MemoryClass GetHypreMemoryClass()
 /// The MemoryType used by MFEM when allocating arrays for Hypre objects.
 inline MemoryType GetHypreMemoryType()
 {
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
    return Device::GetHostMemoryType();
 #elif defined(HYPRE_USING_UNIFIED_MEMORY)
    return MemoryType::MANAGED;
@@ -919,7 +926,7 @@ public:
    enum Type { Jacobi = 0, l1Jacobi = 1, l1GS = 2, l1GStr = 4, lumpedJacobi = 5,
                GS = 6, OPFS = 10, Chebyshev = 16, Taubin = 1001, FIR = 1002
              };
-#ifndef HYPRE_USING_CUDA
+#if !defined(HYPRE_USING_GPU)
    static constexpr Type default_type = l1GS;
 #else
    static constexpr Type default_type = l1Jacobi;

linalg/operator.hpp

@@ -701,11 +701,15 @@ private:
 public:
    /// Create an operator which is a scalar multiple of A.
    explicit ScaledOperator(const Operator *A, double a)
-      : Operator(A->Width(), A->Height()), A_(*A), a_(a) { }
+      : Operator(A->Height(), A->Width()), A_(*A), a_(a) { }
 
    /// Operator application
    virtual void Mult(const Vector &x, Vector &y) const
    { A_.Mult(x, y); y *= a_; }
+
+   /// Application of the transpose.
+   virtual void MultTranspose(const Vector &x, Vector &y) const
+   { A_.MultTranspose(x, y); y *= a_; }
 };
 
 

linalg/simd/auto.hpp

@@ -29,6 +29,10 @@ struct MFEM_AUTOSIMD_ALIGN__ AutoSIMD
 
    scalar_t vec[size];
 
+   AutoSIMD() = default;
+
+   AutoSIMD(const AutoSIMD &) = default;
+
    inline MFEM_ALWAYS_INLINE scalar_t &operator[](int i)
    {
       return vec[i];
@@ -117,6 +121,11 @@ struct MFEM_AUTOSIMD_ALIGN__ AutoSIMD
       return r;
    }
 
+   inline MFEM_ALWAYS_INLINE AutoSIMD operator+() const
+   {
+      return *this;
+   }
+
    inline MFEM_ALWAYS_INLINE AutoSIMD operator+(const AutoSIMD &v) const
    {
       AutoSIMD r;

linalg/simd/m128.hpp

@@ -38,6 +38,10 @@ template <> struct AutoSIMD<double,2,16>
       double vec[size];
    };
 
+   AutoSIMD() = default;
+
+   AutoSIMD(const AutoSIMD &) = default;
+
    inline MFEM_ALWAYS_INLINE double &operator[](int i)
    {
       return vec[i];
@@ -115,6 +119,11 @@ template <> struct AutoSIMD<double,2,16>
       return r;
    }
 
+   inline MFEM_ALWAYS_INLINE AutoSIMD operator+() const
+   {
+      return *this;
+   }
+
    inline MFEM_ALWAYS_INLINE AutoSIMD operator+(const AutoSIMD &v) const
    {
       AutoSIMD r;
@@ -251,4 +260,3 @@ AutoSIMD<double,2,16> operator/(const double &e,
 #endif // __SSE2__
 
 #endif // MFEM_SIMD_M128_HPP
-

linalg/simd/m256.hpp

@@ -38,6 +38,10 @@ template <> struct AutoSIMD<double,4,32>
       double vec[size];
    };
 
+   AutoSIMD() = default;
+
+   AutoSIMD(const AutoSIMD &) = default;
+
    inline MFEM_ALWAYS_INLINE double &operator[](int i)
    {
       return vec[i];
@@ -115,6 +119,11 @@ template <> struct AutoSIMD<double,4,32>
       return r;
    }
 
+   inline MFEM_ALWAYS_INLINE AutoSIMD operator+() const
+   {
+      return *this;
+   }
+
    inline MFEM_ALWAYS_INLINE AutoSIMD operator+(const AutoSIMD &v) const
    {
       AutoSIMD r;

linalg/simd/m512.hpp

@@ -39,6 +39,10 @@ template <> struct AutoSIMD<double,8,64>
       double vec[size];
    };
 
+   AutoSIMD() = default;
+
+   AutoSIMD(const AutoSIMD &) = default;
+
    inline MFEM_ALWAYS_INLINE double &operator[](int i)
    {
       return vec[i];
@@ -120,6 +124,11 @@ template <> struct AutoSIMD<double,8,64>
       return r;
    }
 
+   inline MFEM_ALWAYS_INLINE AutoSIMD operator+() const
+   {
+      return *this;
+   }
+
    inline MFEM_ALWAYS_INLINE AutoSIMD operator+(const AutoSIMD &v) const
    {
       AutoSIMD r;

linalg/simd/qpx256.hpp

@@ -34,6 +34,10 @@ template <> struct AutoSIMD<double,4,32>
       double vec[size];
    };
 
+   AutoSIMD() = default;
+
+   AutoSIMD(const AutoSIMD &) = default;
+
    inline __ATTRS_ai double &operator[](int i) { return vec[i]; }
 
    inline __ATTRS_ai const double &operator[](int i) const { return vec[i]; }
@@ -105,6 +109,11 @@ template <> struct AutoSIMD<double,4,32>
       return r;
    }
 
+   inline __ATTRS_ai AutoSIMD operator+() const
+   {
+      return *this;
+   }
+
    inline __ATTRS_ai AutoSIMD operator+(const AutoSIMD &v) const
    {
       AutoSIMD r;

linalg/simd/sve.hpp

@@ -33,6 +33,10 @@ template <> struct MFEM_AUTOSIMD_ALIGN_SVE AutoSIMD<double,8,64>
 
    double vec[size];
 
+   AutoSIMD() = default;
+
+   AutoSIMD(const AutoSIMD &) = default;
+
    inline MFEM_ALWAYS_INLINE double &operator[](int i)
    {
       return vec[i];
@@ -123,6 +127,11 @@ template <> struct MFEM_AUTOSIMD_ALIGN_SVE AutoSIMD<double,8,64>
       return r;
    }
 
+   inline MFEM_ALWAYS_INLINE AutoSIMD operator+() const
+   {
+      return *this;
+   }
+
    inline MFEM_ALWAYS_INLINE AutoSIMD operator+(const AutoSIMD &v) const
    {
       AutoSIMD r;

linalg/simd/vsx128.hpp

@@ -17,6 +17,10 @@
 #include "../../config/tconfig.hpp"
 #include <altivec.h>
 
+#ifdef __GNUC__
+#undef bool
+#endif
+
 namespace mfem
 {
 
@@ -34,6 +38,10 @@ template <> struct AutoSIMD<double,2,16>
       double vec[size];
    };
 
+   AutoSIMD() = default;
+
+   AutoSIMD(const AutoSIMD &) = default;
+
    inline MFEM_ALWAYS_INLINE double &operator[](int i)
    {
       return vec[i];
@@ -107,10 +115,19 @@ template <> struct AutoSIMD<double,2,16>
    inline MFEM_ALWAYS_INLINE AutoSIMD operator-() const
    {
       AutoSIMD r;
+#ifndef __GNUC__
       r.vd = vec_neg(vd);
+#else
+      r.vd = vec_splats(0.0) - vd;
+#endif
       return r;
    }
 
+   inline MFEM_ALWAYS_INLINE AutoSIMD operator+() const
+   {
+      return *this;
+   }
+
    inline MFEM_ALWAYS_INLINE AutoSIMD operator+(const AutoSIMD &v) const
    {
       AutoSIMD r;

linalg/solvers.hpp

@@ -56,7 +56,7 @@ public:
    {
    }
 
-   /** @brief This method is invoked by ItertiveSolver::SetMonitor, informing
+   /** @brief This method is invoked by IterativeSolver::SetMonitor, informing
        the monitor which IterativeSolver is using it. */
    void SetIterativeSolver(const IterativeSolver &solver)
    { iter_solver = &solver; }
@@ -65,10 +65,55 @@ public:
 /// Abstract base class for iterative solver
 class IterativeSolver : public Solver
 {
+public:
+   /** @brief Settings for the output behavior of the IterativeSolver.
+
+       By default, all output is suppressed. The construction of the desired
+       print level can be achieved through a builder pattern, for example
+
+           PrintLevel().Errors().Warnings()
+
+       constructs the print level with only errors and warnings enabled.
+     */
+   struct PrintLevel
+   {
+      /** @brief If a fatal problem has been detected the failure will be
+          reported to @ref mfem::err. */
+      bool errors = false;
+      /** @brief If a non-fatal problem has been detected some context-specific
+          information will be reported to @ref mfem::out */
+      bool warnings = false;
+      /** @brief Detailed information about each iteration will be reported to
+          @ref mfem::out */
+      bool iterations = false;
+      /** @brief A summary of the solver process will be reported after the last
+          iteration to @ref mfem::out */
+      bool summary = false;
+      /** @brief Information about the first and last iteration will be printed
+          to @ref mfem::out */
+      bool first_and_last = false;
+
+      /// Initializes the print level to suppress
+      PrintLevel() = default;
+
+      /** @name Builder
+         These methods are utilized to construct PrintLevel objects through a
+         builder approach by chaining the function calls in this group. */
+      ///@{
+      PrintLevel &None() { *this = PrintLevel(); return *this; }
+      PrintLevel &Warnings() { warnings=true; return *this; }
+      PrintLevel &Errors() { errors=true; return *this; }
+      PrintLevel &Iterations() { iterations=true; return *this; }
+      PrintLevel &FirstAndLast() { first_and_last=true; return *this; }
+      PrintLevel &Summary() { summary=true; return *this; }
+      PrintLevel &All() { return Warnings().Errors().Iterations().FirstAndLast().Summary(); }
+      ///@}
+   };
+
 #ifdef MFEM_USE_MPI
 private:
    int dot_prod_type; // 0 - local, 1 - global over 'comm'
-   MPI_Comm comm;
+   MPI_Comm comm = MPI_COMM_NULL;
 #endif
 
 protected:
@@ -76,13 +121,52 @@ protected:
    Solver *prec;
    IterativeSolverMonitor *monitor = nullptr;
 
-   int max_iter, print_level;
-   double rel_tol, abs_tol;
+   /// @name Reporting (protected attributes and member functions)
+   ///@{
 
-   // stats
-   mutable int final_iter, converged;
+   /** @brief (DEPRECATED) Legacy print level definition, which is left for
+       compatibility with custom iterative solvers.
+       @deprecated #print_options should be used instead. */
+   int print_level = -1;
+
+   /** @brief Output behavior for the iterative solver.
+
+       This primarily controls the output behavior of the iterative solvers
+       provided by this library. This member must be synchronized with
+       #print_level to ensure compatibility with custom iterative solvers. */
+   PrintLevel print_options;
+
+   /// Convert a legacy print level integer to a PrintLevel object
+   PrintLevel FromLegacyPrintLevel(int);
+
+   /// @brief Use some heuristics to guess a legacy print level corresponding to
+   /// the given PrintLevel.
+   static int GuessLegacyPrintLevel(PrintLevel);
+   ///@}
+
+   /// @name Convergence (protected attributes)
+   ///@{
+
+   /// Limit for the number of iterations the solver is allowed to do
+   int max_iter;
+
+   /// Relative tolerance.
+   double rel_tol;
+
+   /// Absolute tolerance.
+   double abs_tol;
+
+   ///@}
+
+   /// @name Solver statistics (protected attributes)
+   ///@{
+
+   mutable int final_iter;
+   mutable bool converged;
    mutable double final_norm;
 
+   ///@}
+
    double Dot(const Vector &x, const Vector &y) const;
    double Norm(const Vector &x) const { return sqrt(Dot(x, x)); }
    void Monitor(int it, double norm, const Vector& r, const Vector& x,
@@ -95,20 +179,80 @@ public:
    IterativeSolver(MPI_Comm comm_);
 #endif
 
+   /** @name Convergence
+       @brief Termination criteria for the iterative solvers.
+
+       @details While the convergence criterion is solver specific, most of the
+       provided iterative solvers use one of the following criteria
+
+       \f$ ||r||_X \leq tol_{rel}||r_0||_X \f$,
+
+       \f$ ||r||_X \leq tol_{abs} \f$,
+
+       \f$ ||r||_X \leq \max\{ tol_{abs}, tol_{rel} ||r_0||_X \} \f$,
+
+       where X denotes the space in which the norm is measured. The choice of
+       X depends on the specific iterative solver.
+      */
+   ///@{
    void SetRelTol(double rtol) { rel_tol = rtol; }
    void SetAbsTol(double atol) { abs_tol = atol; }
    void SetMaxIter(int max_it) { max_iter = max_it; }
-   void SetPrintLevel(int print_lvl);
+   ///@}
 
+   /** @name Reporting
+       These options control the internal reporting behavior into ::mfem::out
+       and ::mfem::err of the iterative solvers.
+    */
+   ///@{
+
+   /// @brief Legacy method to set the level of verbosity of the solver output.
+   /** This is the old way to control what information will be printed to
+       ::mfem::out and ::mfem::err. The behavior for the print level for all
+       iterative solvers is:
+
+       - -1: Suppress all outputs.
+       -  0: Print information about all detected issues (e.g. no convergence).
+       -  1: Same as level 0, but with detailed information about each
+             iteration.
+       -  2: Print detected issues and a summary when the solver terminates.
+       -  3: Same as 2, but print also the first and last iterations.
+       - >3: Custom print options which are dependent on the specific solver.
+
+       In parallel, only rank 0 produces output.
+
+       @note It is recommended to use @ref SetPrintLevel(PrintLevel) instead.
+
+       @note Some derived classes, like KINSolver, redefine this method and use
+       their own set of print level constants. */
+   virtual void SetPrintLevel(int print_lvl);
+
+   /// @brief Set the level of verbosity of the solver output.
+   /** In parallel, only rank 0 produces outputs. Errors are output to
+       ::mfem::err and all other information to ::mfem::out.
+
+       @note Not all subclasses of IterativeSolver support all possible options.
+
+       @note Some derived classes, like KINSolver, disable this method in favor
+       of SetPrintLevel(int).
+
+       @sa PrintLevel for possible options.
+   */
+   virtual void SetPrintLevel(PrintLevel);
+   ///@}
+
+   /// @name Solver statistics
+   ///@{
    int GetNumIterations() const { return final_iter; }
-   int GetConverged() const { return converged; }
+   bool GetConverged() const { return converged; }
    double GetFinalNorm() const { return final_norm; }
+   ///@}
 
    /// This should be called before SetOperator
    virtual void SetPreconditioner(Solver &pr);
 
    /// Also calls SetOperator for the preconditioner if there is one
-   virtual void SetOperator(const Operator &op);
+   virtual void SetOperator(const Operator &op) override;
 
    /// Set the iterative solver monitor
    void SetMonitor(IterativeSolverMonitor &m)
@@ -163,6 +307,9 @@ public:
 
    ~OperatorJacobiSmoother() {}
 
+   /// Replace diagonal entries with their absolute values.
+   void SetPositiveDiagonal(bool pos_diag = true) { use_abs_diag = pos_diag; }
+
    void Mult(const Vector &x, Vector &y) const;
    void MultTranspose(const Vector &x, Vector &y) const { Mult(x, y); }
 
@@ -184,6 +331,8 @@ private:
    const double damping;
    const Array<int> *ess_tdof_list; // not owned; may be NULL
    mutable Vector residual;
+   /// Uses absolute values of the diagonal entries.
+   bool use_abs_diag = false;
 
    const Operator *oper; // not owned
 

linalg/sparsemat.cpp

@@ -23,55 +23,75 @@
 #include <limits>
 #include <cstring>
 
+#if defined(MFEM_USE_CUDA)
+#define MFEM_cu_or_hip(stub) cu##stub
+#define MFEM_Cu_or_Hip(stub) Cu##stub
+#define MFEM_CU_or_HIP(stub) CU##stub
+#define MFEM_CUDA_or_HIP(stub) CUDA##stub
+
+#if CUSPARSE_VERSION >=  11400
+#define MFEM_GPUSPARSE_ALG CUSPARSE_SPMV_CSR_ALG1
+#else // CUSPARSE_VERSION >= 11400
+#define MFEM_GPUSPARSE_ALG CUSPARSE_CSRMV_ALG1
+#endif // CUSPARSE_VERSION >= 11400
+
+#elif defined(MFEM_USE_HIP)
+#define MFEM_cu_or_hip(stub) hip##stub
+#define MFEM_Cu_or_Hip(stub) Hip##stub
+#define MFEM_CU_or_HIP(stub) HIP##stub
+#define MFEM_CUDA_or_HIP(stub) HIP##stub
+
+// https://hipsparse.readthedocs.io/en/latest/usermanual.html#hipsparsespmvalg-t
+#define MFEM_GPUSPARSE_ALG HIPSPARSE_CSRMV_ALG1
+#endif // defined(MFEM_USE_CUDA)
+
 namespace mfem
 {
 
 using namespace std;
 
-#ifdef MFEM_USE_CUDA
+#ifdef MFEM_USE_CUDA_OR_HIP
 int SparseMatrix::SparseMatrixCount = 0;
-cusparseHandle_t SparseMatrix::handle = nullptr;
+// doxygen doesn't like the macro-assisted typename so let's skip parsing it:
+// \cond false
+MFEM_cu_or_hip(sparseHandle_t) SparseMatrix::handle = nullptr;
+// \endcond
 size_t SparseMatrix::bufferSize = 0;
 void * SparseMatrix::dBuffer = nullptr;
-#  if CUSPARSE_VERSION >=  11400
-#    define MFEM_CUSPARSE_ALG CUSPARSE_SPMV_CSR_ALG1
-#  else
-#    define MFEM_CUSPARSE_ALG CUSPARSE_CSRMV_ALG1
-#  endif // CUSPARSE_VERSION >= 11400
-#endif // MFEM_USE_CUDA
+#endif // MFEM_USE_CUDA_OR_HIP
 
-void SparseMatrix::InitCuSparse()
+void SparseMatrix::InitGPUSparse()
 {
-   // Initialize cuSPARSE library
-#ifdef MFEM_USE_CUDA
-   if (Device::Allows(Backend::CUDA_MASK))
+   // Initialize cuSPARSE/hipSPARSE library
+#ifdef MFEM_USE_CUDA_OR_HIP
+   if (Device::Allows(Backend::CUDA_MASK | Backend::HIP_MASK))
    {
-      if (!handle) { cusparseCreate(&handle); }
-      useCuSparse=true;
+      if (!handle) { MFEM_cu_or_hip(sparseCreate)(&handle); }
+      useGPUSparse=true;
       SparseMatrixCount++;
    }
    else
    {
-      useCuSparse=false;
+      useGPUSparse=false;
    }
-#endif
+#endif // MFEM_USE_CUDA_OR_HIP
 }
 
-void SparseMatrix::ClearCuSparse()
+void SparseMatrix::ClearGPUSparse()
 {
-#ifdef MFEM_USE_CUDA
+#ifdef MFEM_USE_CUDA_OR_HIP
    if (initBuffers)
    {
-#if CUDA_VERSION >= 10010
-      cusparseDestroySpMat(matA_descr);
-      cusparseDestroyDnVec(vecX_descr);
-      cusparseDestroyDnVec(vecY_descr);
+#if CUDA_VERSION >= 10010 || defined(MFEM_USE_HIP)
+      MFEM_cu_or_hip(sparseDestroySpMat)(matA_descr);
+      MFEM_cu_or_hip(sparseDestroyDnVec)(vecX_descr);
+      MFEM_cu_or_hip(sparseDestroyDnVec)(vecY_descr);
 #else
       cusparseDestroyMatDescr(matA_descr);
-#endif
+#endif // CUDA_VERSION >= 10010 || defined(MFEM_USE_HIP)
       initBuffers = false;
    }
-#endif
+#endif // MFEM_USE_CUDA_OR_HIP
 }
 
 SparseMatrix::SparseMatrix(int nrows, int ncols)
@@ -97,7 +117,7 @@ SparseMatrix::SparseMatrix(int nrows, int ncols)
    NodesMem = new RowNodeAlloc;
 #endif
 
-   InitCuSparse();
+   InitGPUSparse();
 }
 
 SparseMatrix::SparseMatrix(int *i, int *j, double *data, int m, int n)
@@ -116,7 +136,7 @@ SparseMatrix::SparseMatrix(int *i, int *j, double *data, int m, int n)
    NodesMem = NULL;
 #endif
 
-   InitCuSparse();
+   InitGPUSparse();
 }
 
 SparseMatrix::SparseMatrix(int *i, int *j, double *data, int m, int n,
@@ -149,7 +169,7 @@ SparseMatrix::SparseMatrix(int *i, int *j, double *data, int m, int n,
       }
    }
 
-   InitCuSparse();
+   InitGPUSparse();
 }
 
 SparseMatrix::SparseMatrix(int nrows, int ncols, int rowsize)
@@ -172,7 +192,7 @@ SparseMatrix::SparseMatrix(int nrows, int ncols, int rowsize)
       I[i] = i * rowsize;
    }
 
-   InitCuSparse();
+   InitGPUSparse();
 }
 
 SparseMatrix::SparseMatrix(const SparseMatrix &mat, bool copy_graph,
@@ -240,7 +260,7 @@ SparseMatrix::SparseMatrix(const SparseMatrix &mat, bool copy_graph,
    At = NULL;
    isSorted = mat.isSorted;
 
-   InitCuSparse();
+   InitGPUSparse();
 }
 
 SparseMatrix::SparseMatrix(const Vector &v)
@@ -269,7 +289,7 @@ SparseMatrix::SparseMatrix(const Vector &v)
       A[r] = v[r];
    }
 
-   InitCuSparse();
+   InitGPUSparse();
 }
 
 SparseMatrix& SparseMatrix::operator=(const SparseMatrix &rhs)
@@ -310,7 +330,7 @@ void SparseMatrix::SetEmpty()
 #endif
    isSorted = false;
 
-   ClearCuSparse();
+   ClearGPUSparse();
 }
 
 int SparseMatrix::RowSize(const int i) const
@@ -660,65 +680,100 @@ void SparseMatrix::AddMult(const Vector &x, Vector &y, const double a) const
 
    // Skip if matrix has no non-zeros
    if (nnz == 0) {return;}
-   if (Device::Allows(Backend::CUDA_MASK) && useCuSparse)
+   if ((Device::Allows(Backend::CUDA_MASK | Backend::HIP_MASK)) && useGPUSparse)
    {
-#ifdef MFEM_USE_CUDA
+#ifdef MFEM_USE_CUDA_OR_HIP
       const double alpha = a;
       const double beta  = 1.0;
 
       // Setup descriptors
       if (!initBuffers)
       {
-#if CUDA_VERSION >= 10010
+#if CUDA_VERSION >= 10010 || defined(MFEM_USE_HIP)
          // Setup matrix descriptor
-         cusparseCreateCsr(&matA_descr,Height(), Width(), J.Capacity(),
-                           const_cast<int *>(d_I),
-                           const_cast<int *>(d_J), const_cast<double *>(d_A), CUSPARSE_INDEX_32I,
-                           CUSPARSE_INDEX_32I, CUSPARSE_INDEX_BASE_ZERO, CUDA_R_64F);
+         MFEM_cu_or_hip(sparseCreateCsr)(
+            &matA_descr,Height(),
+            Width(),
+            J.Capacity(),
+            const_cast<int *>(d_I),
+            const_cast<int *>(d_J),
+            const_cast<double *>(d_A),
+            MFEM_CU_or_HIP(SPARSE_INDEX_32I),
+            MFEM_CU_or_HIP(SPARSE_INDEX_32I),
+            MFEM_CU_or_HIP(SPARSE_INDEX_BASE_ZERO),
+            MFEM_CUDA_or_HIP(_R_64F));
 
          // Create handles for input/output vectors
-         cusparseCreateDnVec(&vecX_descr, x.Size(), const_cast<double *>(d_x),
-                             CUDA_R_64F);
-         cusparseCreateDnVec(&vecY_descr, y.Size(), d_y, CUDA_R_64F);
+         MFEM_cu_or_hip(sparseCreateDnVec)(&vecX_descr,
+                                           x.Size(),
+                                           const_cast<double *>(d_x),
+                                           MFEM_CUDA_or_HIP(_R_64F));
+         MFEM_cu_or_hip(sparseCreateDnVec)(&vecY_descr, y.Size(), d_y,
+                                           MFEM_CUDA_or_HIP(_R_64F));
 #else
          cusparseCreateMatDescr(&matA_descr);
          cusparseSetMatIndexBase(matA_descr, CUSPARSE_INDEX_BASE_ZERO);
          cusparseSetMatType(matA_descr, CUSPARSE_MATRIX_TYPE_GENERAL);
-#endif
+#endif // CUDA_VERSION >= 10010 || defined(MFEM_USE_HIP)
          initBuffers = true;
       }
       // Allocate kernel space. Buffer is shared between different sparsemats
       size_t newBufferSize = 0;
 
-      cusparseSpMV_bufferSize(handle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha,
-                              matA_descr,
-                              vecX_descr, &beta, vecY_descr, CUDA_R_64F,
-                              MFEM_CUSPARSE_ALG, &newBufferSize);
+      MFEM_cu_or_hip(sparseSpMV_bufferSize)(
+         handle,
+         MFEM_CU_or_HIP(SPARSE_OPERATION_NON_TRANSPOSE),
+         &alpha,
+         matA_descr,
+         vecX_descr,
+         &beta,
+         vecY_descr,
+         MFEM_CUDA_or_HIP(_R_64F),
+         MFEM_GPUSPARSE_ALG,
+         &newBufferSize);
 
       // Check if we need to resize
       if (newBufferSize > bufferSize)
       {
          bufferSize = newBufferSize;
-         if (dBuffer != nullptr) { CuMemFree(dBuffer); }
-         CuMemAlloc(&dBuffer, bufferSize);
+         if (dBuffer != nullptr) { MFEM_Cu_or_Hip(MemFree)(dBuffer); }
+         MFEM_Cu_or_Hip(MemAlloc)(&dBuffer, bufferSize);
       }
 
-#if CUDA_VERSION >= 10010
+#if CUDA_VERSION >= 10010 || defined(MFEM_USE_HIP)
       // Update input/output vectors
-      cusparseDnVecSetValues(vecX_descr, const_cast<double *>(d_x));
-      cusparseDnVecSetValues(vecY_descr, d_y);
+      MFEM_cu_or_hip(sparseDnVecSetValues)(vecX_descr,
+                                           const_cast<double *>(d_x));
+      MFEM_cu_or_hip(sparseDnVecSetValues)(vecY_descr, d_y);
 
       // Y = alpha A * X + beta * Y
-      cusparseSpMV(handle, CUSPARSE_OPERATION_NON_TRANSPOSE, &alpha, matA_descr,
-                   vecX_descr, &beta, vecY_descr, CUDA_R_64F, MFEM_CUSPARSE_ALG, dBuffer);
+      MFEM_cu_or_hip(sparseSpMV)(
+         handle,
+         MFEM_CU_or_HIP(SPARSE_OPERATION_NON_TRANSPOSE),
+         &alpha,
+         matA_descr,
+         vecX_descr,
+         &beta,
+         vecY_descr,
+         MFEM_CUDA_or_HIP(_R_64F),
+         MFEM_GPUSPARSE_ALG,
+         dBuffer);
 #else
-      cusparseDcsrmv(handle, CUSPARSE_OPERATION_NON_TRANSPOSE,
-                     Height(), Width(), J.Capacity(),
-                     &alpha, matA_descr,
-                     const_cast<double *>(d_A), const_cast<int *>(d_I), const_cast<int *>(d_J),
-                     const_cast<double *>(d_x), &beta, d_y);
-#endif // CUDA_VERSION >= 10010
-#endif // MFEM_USE_CUDA
+      cusparseDcsrmv(handle,
+                     CUSPARSE_OPERATION_NON_TRANSPOSE,
+                     Height(),
+                     Width(),
+                     J.Capacity(),
+                     &alpha,
+                     matA_descr,
+                     const_cast<double *>(d_A),
+                     const_cast<int *>(d_I),
+                     const_cast<int *>(d_J),
+                     const_cast<double *>(d_x),
+                     &beta,
+                     d_y);
+#endif // CUDA_VERSION >= 10010 || defined(MFEM_USE_HIP)
+#endif // MFEM_USE_CUDA_OR_HIP
    }
    else
    {
@@ -736,7 +791,7 @@ void SparseMatrix::AddMult(const Vector &x, Vector &y, const double a) const
 
    }
 
-#else
+#else // MFEM_USE_LEGACY_OPENMP
    const double *Ap = A, *xp = x.GetData();
    double *yp = y.GetData();
    const int *Jp = J, *Ip = I;
@@ -752,7 +807,7 @@ void SparseMatrix::AddMult(const Vector &x, Vector &y, const double a) const
       }
       yp[i] += a * d;
    }
-#endif
+#endif // MFEM_USE_LEGACY_OPENMP
 }
 
 void SparseMatrix::MultTranspose(const Vector &x, Vector &y) const
@@ -772,12 +827,13 @@ void SparseMatrix::AddMultTranspose(const Vector &x, Vector &y,
 
    if (!Finalized())
    {
-      double *yp = y.GetData();
+      double *yp = y.HostReadWrite();
+      const double *xp = x.HostRead();
       // The matrix is not finalized, but multiplication is still possible
       for (int i = 0; i < height; i++)
       {
          RowNode *row = Rows[i];
-         double b = a * x(i);
+         double b = a * xp[i];
          for ( ; row != NULL; row = row->Prev)
          {
             yp[row->Column] += row->Value * b;
@@ -2382,7 +2438,7 @@ double SparseMatrix::GetJacobiScaling() const
 }
 
 void SparseMatrix::Jacobi(const Vector &b, const Vector &x0, Vector &x1,
-                          double sc) const
+                          double sc, bool use_abs_diag) const
 {
    MFEM_VERIFY(Finalized(), "Matrix must be finalized.");
 
@@ -2403,7 +2459,8 @@ void SparseMatrix::Jacobi(const Vector &b, const Vector &x0, Vector &x1,
       }
       if (d >= 0 && A[d] != 0.0)
       {
-         x1(i) = sc * (sum / A[d]) + (1.0 - sc) * x0(i);
+         const double diag = (use_abs_diag) ? fabs(A[d]) : A[d];
+         x1(i) = sc * (sum / diag) + (1.0 - sc) * x0(i);
       }
       else
       {
@@ -2412,7 +2469,8 @@ void SparseMatrix::Jacobi(const Vector &b, const Vector &x0, Vector &x1,
    }
 }
 
-void SparseMatrix::DiagScale(const Vector &b, Vector &x, double sc) const
+void SparseMatrix::DiagScale(const Vector &b, Vector &x,
+                             double sc, bool use_abs_diag) const
 {
    MFEM_VERIFY(Finalized(), "Matrix must be finalized.");
 
@@ -2438,11 +2496,12 @@ void SparseMatrix::DiagScale(const Vector &b, Vector &x, double sc) const
          }
          if (Jp[j] == i)
          {
-            if (!(std::abs(Ap[j]) > 0.0))
+            const double diag = (use_abs_diag) ? fabs(Ap[j]) : Ap[j];
+            if (diag == 0.0)
             {
                MFEM_ABORT_KERNEL("Zero diagonal in SparseMatrix::DiagScale");
             }
-            xp[i] = sc * bp[i] / Ap[j];
+            xp[i] = sc * bp[i] / diag;
             break;
          }
       }
@@ -3256,7 +3315,7 @@ void SparseMatrix::Destroy()
 #endif
    delete At;
 
-   ClearCuSparse();
+   ClearGPUSparse();
 }
 
 int SparseMatrix::ActualWidth() const
@@ -3992,4 +4051,29 @@ void SparseMatrix::Swap(SparseMatrix &other)
    mfem::Swap(isSorted, other.isSorted);
 }
 
+SparseMatrix::~SparseMatrix()
+{
+   Destroy();
+#ifdef MFEM_USE_CUDA_OR_HIP
+   if (Device::Allows(Backend::CUDA_MASK | Backend::HIP_MASK))
+   {
+      if (SparseMatrixCount==1)
+      {
+         if (handle)
+         {
+            MFEM_cu_or_hip(sparseDestroy)(handle);
+            handle = nullptr;
+         }
+         if (dBuffer)
+         {
+            MFEM_Cu_or_Hip(MemFree)(dBuffer);
+            dBuffer = nullptr;
+            bufferSize = 0;
+         }
+      }
+      SparseMatrixCount--;
+   }
+#endif // MFEM_USE_CUDA_OR_HIP
+}
+
 }

linalg/sparsemat.hpp

@@ -22,6 +22,11 @@
 #include "../general/globals.hpp"
 #include "densemat.hpp"
 
+#if defined(MFEM_USE_HIP)
+#include <hipsparse.h>
+#endif
+
+
 namespace mfem
 {
 
@@ -81,27 +86,41 @@ protected:
    void Destroy();   // Delete all owned data
    void SetEmpty();  // Init all entries with empty values
 
-   bool useCuSparse{true}; // Use cuSPARSE if available
+   bool useGPUSparse = true; // Use cuSPARSE or hipSPARSE if available
 
-   // Initialize cuSPARSE
-   void InitCuSparse();
+   // Initialize cuSPARSE/hipSPARSE
+   void InitGPUSparse();
 
-#ifdef MFEM_USE_CUDA
-   cusparseStatus_t status;
-   static cusparseHandle_t handle;
-   cusparseMatDescr_t descr=0;
+#ifdef MFEM_USE_CUDA_OR_HIP
+   // common for hipSPARSE and cuSPARSE
    static int SparseMatrixCount;
    static size_t bufferSize;
    static void *dBuffer;
-   mutable bool initBuffers{false};
+   mutable bool initBuffers = false;
+
+#if defined(MFEM_USE_CUDA)
+   cusparseStatus_t status;
+   static cusparseHandle_t handle;
+   cusparseMatDescr_t descr = 0;
+
 #if CUDA_VERSION >= 10010
    mutable cusparseSpMatDescr_t matA_descr;
    mutable cusparseDnVecDescr_t vecX_descr;
    mutable cusparseDnVecDescr_t vecY_descr;
-#else
+#else // CUDA_VERSION >= 10010
    mutable cusparseMatDescr_t matA_descr;
-#endif
-#endif
+#endif // CUDA_VERSION >= 10010
+
+#else // defined(MFEM_USE_CUDA)
+   hipsparseStatus_t status;
+   static hipsparseHandle_t handle;
+   hipsparseMatDescr_t descr = 0;
+
+   mutable hipsparseSpMatDescr_t matA_descr;
+   mutable hipsparseDnVecDescr_t vecX_descr;
+   mutable hipsparseDnVecDescr_t vecY_descr;
+#endif // defined(MFEM_USE_CUDA)
+#endif // MFEM_USE_CUDA_OR_HIP
 
 public:
    /// Create an empty SparseMatrix.
@@ -109,7 +128,7 @@ public:
    {
       SetEmpty();
 
-      InitCuSparse();
+      InitGPUSparse();
    }
 
    /** @brief Create a sparse matrix with flexible sparsity structure using a
@@ -151,8 +170,15 @@ public:
    /// Create a SparseMatrix with diagonal @a v, i.e. A = Diag(v)
    SparseMatrix(const Vector & v);
 
-   // Runtime option to use cuSPARSE. Only valid when using a CUDA backend.
-   void UseCuSparse(bool useCuSparse_ = true) { useCuSparse = useCuSparse_;}
+   /** @brief Runtime option to use cuSPARSE or hipSPARSE. Only valid when using
+       a CUDA or HIP backend.
+
+       @note This option is enabled by default, so typically one would use this
+       method to disable the use of cuSPARSE/hipSPARSE. */
+   void UseGPUSparse(bool useGPUSparse_ = true) { useGPUSparse = useGPUSparse_;}
+   /// Deprecated equivalent of UseGPUSparse().
+   MFEM_DEPRECATED
+   void UseCuSparse(bool useCuSparse_ = true) { UseGPUSparse(useCuSparse_); }
 
    /// Assignment operator: deep copy
    SparseMatrix& operator=(const SparseMatrix &rhs);
@@ -169,9 +195,12 @@ public:
    /// Clear the contents of the SparseMatrix.
    void Clear() { Destroy(); SetEmpty(); }
 
-   /** @brief Clear the CuSparse descriptors.
+   /** @brief Clear the cuSPARSE/hipSPARSE descriptors.
        This must be called after releasing the device memory of A. */
-   void ClearCuSparse();
+   void ClearGPUSparse();
+   /// Deprecated equivalent of ClearGPUSparse().
+   MFEM_DEPRECATED
+   void ClearCuSparse() { ClearGPUSparse(); }
 
    /// Check if the SparseMatrix is empty.
    bool Empty() const { return (A == NULL) && (Rows == NULL); }
@@ -446,10 +475,14 @@ public:
    /// Determine appropriate scaling for Jacobi iteration
    double GetJacobiScaling() const;
    /** One scaled Jacobi iteration for the system A x = b.
-       x1 = x0 + sc D^{-1} (b - A x0)  where D is the diag of A. */
-   void Jacobi(const Vector &b, const Vector &x0, Vector &x1, double sc) const;
+       x1 = x0 + sc D^{-1} (b - A x0)  where D is the diag of A.
+       Absolute values of D are used when use_abs_diag = true. */
+   void Jacobi(const Vector &b, const Vector &x0, Vector &x1,
+               double sc, bool use_abs_diag = false) const;
 
-   void DiagScale(const Vector &b, Vector &x, double sc = 1.0) const;
+   /// x = sc b / A_ii. When use_abs_diag = true, |A_ii| is used.
+   void DiagScale(const Vector &b, Vector &x,
+                  double sc = 1.0, bool use_abs_diag = false) const;
 
    /** x1 = x0 + sc D^{-1} (b - A x0) where \f$ D_{ii} = \sum_j |A_{ij}| \f$. */
    void Jacobi2(const Vector &b, const Vector &x0, Vector &x1,
@@ -624,30 +657,7 @@ public:
    void Swap(SparseMatrix &other);
 
    /// Destroys sparse matrix.
-   virtual ~SparseMatrix()
-   {
-      Destroy();
-#ifdef MFEM_USE_CUDA
-      if (useCuSparse)
-      {
-         if (SparseMatrixCount==1)
-         {
-            if (handle)
-            {
-               cusparseDestroy(handle);
-               handle = nullptr;
-            }
-            if (dBuffer)
-            {
-               CuMemFree(dBuffer);
-               dBuffer = nullptr;
-               bufferSize = 0;
-            }
-         }
-         SparseMatrixCount--;
-      }
-#endif
-   }
+   virtual ~SparseMatrix();
 
    Type GetType() const { return MFEM_SPARSEMAT; }
 };

linalg/sparsesmoothers.cpp

@@ -65,7 +65,7 @@ void DSmoother::Mult(const Vector &x, Vector &y) const
 {
    if (!iterative_mode && type == 0 && iterations == 1)
    {
-      oper->DiagScale(x, y, scale);
+      oper->DiagScale(x, y, scale, use_abs_diag);
       return;
    }
 
@@ -90,7 +90,7 @@ void DSmoother::Mult(const Vector &x, Vector &y) const
    {
       if (type == 0)
       {
-         oper->Jacobi(x, *p, *r, scale);
+         oper->Jacobi(x, *p, *r, scale, use_abs_diag);
       }
       else if (type == 1)
       {

linalg/sparsesmoothers.hpp

@@ -58,6 +58,8 @@ protected:
    int type; // 0, 1, 2 - scaled Jacobi, scaled l1-Jacobi, scaled lumped-Jacobi
    double scale;
    int iterations;
+   /// Uses abs values of the diagonal entries. Relevant only when type = 0.
+   bool use_abs_diag = false;
 
    mutable Vector z;
 
@@ -69,6 +71,9 @@ public:
    /// Create Jacobi smoother.
    DSmoother(const SparseMatrix &a, int t = 0, double s = 1., int it = 1);
 
+   /// Replace diag entries with their abs values. Relevant only when type = 0.
+   void SetPositiveDiagonal(bool pos_diag = true) { use_abs_diag = pos_diag; }
+
    /// Matrix vector multiplication with Jacobi smoother.
    virtual void Mult(const Vector &x, Vector &y) const;
 };

linalg/sundials.cpp

@@ -1976,6 +1976,11 @@ void KINSolver::SetMAA(int m_aa)
    }
 }
 
+void KINSolver::SetPrintLevel(PrintLevel)
+{
+   MFEM_ABORT("this method is not supported! Use SetPrintLevel(int) instead.");
+}
+
 // Compute the scaling vectors and solve nonlinear system
 void KINSolver::Mult(const Vector&, Vector &x) const
 {

linalg/sundials.hpp

@@ -833,6 +833,12 @@ public:
    /** @note Only valid in combination with JFNK */
    void SetLSMaxRestarts(int m) { maxlrs = m; }
 
+   /// Set the print level for the KINSetPrintLevel function.
+   virtual void SetPrintLevel(int print_lvl) { print_level = print_lvl; }
+
+   /// This method is not supported and will throw an error.
+   virtual void SetPrintLevel(PrintLevel);
+
    /// Solve the nonlinear system \f$ F(x) = 0 \f$.
    /** This method computes the x_scale and fx_scale vectors and calls the
        other Mult(Vector&, Vector&, Vector&) const method. The x_scale vector

mesh/mesh.cpp

@@ -350,6 +350,7 @@ void Mesh::GetElementTransformation(int i, IsoparametricTransformation *ElTr)
    ElTr->Attribute = GetAttribute(i);
    ElTr->ElementNo = i;
    ElTr->ElementType = ElementTransformation::ELEMENT;
+   ElTr->mesh = this;
    ElTr->Reset();
    if (Nodes == NULL)
    {
@@ -382,6 +383,7 @@ void Mesh::GetElementTransformation(int i, const Vector &nodes,
    ElTr->Attribute = GetAttribute(i);
    ElTr->ElementNo = i;
    ElTr->ElementType = ElementTransformation::ELEMENT;
+   ElTr->mesh = this;
    DenseMatrix &pm = ElTr->GetPointMat();
    ElTr->Reset();
    nodes.HostRead();
@@ -437,6 +439,7 @@ void Mesh::GetBdrElementTransformation(int i, IsoparametricTransformation* ElTr)
    ElTr->Attribute = GetBdrAttribute(i);
    ElTr->ElementNo = i; // boundary element number
    ElTr->ElementType = ElementTransformation::BDR_ELEMENT;
+   ElTr->mesh = this;
    DenseMatrix &pm = ElTr->GetPointMat();
    ElTr->Reset();
    if (Nodes == NULL)
@@ -481,6 +484,7 @@ void Mesh::GetBdrElementTransformation(int i, IsoparametricTransformation* ElTr)
                      "Mesh requires nodal Finite Element.");
          IntegrationRule eir(face_el->GetDof());
          FaceElemTr.Loc1.Transf.ElementNo = elem_id;
+         FaceElemTr.Loc1.Transf.mesh = this;
          FaceElemTr.Loc1.Transf.ElementType = ElementTransformation::ELEMENT;
          FaceElemTr.Loc1.Transform(face_el->GetNodes(), eir);
          Nodes->GetVectorValues(FaceElemTr.Loc1.Transf, eir, pm);
@@ -495,6 +499,7 @@ void Mesh::GetFaceTransformation(int FaceNo, IsoparametricTransformation *FTr)
    FTr->Attribute = (Dim == 1) ? 1 : faces[FaceNo]->GetAttribute();
    FTr->ElementNo = FaceNo;
    FTr->ElementType = ElementTransformation::FACE;
+   FTr->mesh = this;
    DenseMatrix &pm = FTr->GetPointMat();
    FTr->Reset();
    if (Nodes == NULL)
@@ -551,6 +556,7 @@ void Mesh::GetFaceTransformation(int FaceNo, IsoparametricTransformation *FTr)
          IntegrationRule eir(face_el->GetDof());
          FaceElemTr.Loc1.Transf.ElementNo = face_info.Elem1No;
          FaceElemTr.Loc1.Transf.ElementType = ElementTransformation::ELEMENT;
+         FaceElemTr.Loc1.Transf.mesh = this;
          FaceElemTr.Loc1.Transform(face_el->GetNodes(), eir);
          Nodes->GetVectorValues(FaceElemTr.Loc1.Transf, eir, pm);
 
@@ -580,6 +586,7 @@ void Mesh::GetEdgeTransformation(int EdgeNo, IsoparametricTransformation *EdTr)
    EdTr->Attribute = 1;
    EdTr->ElementNo = EdgeNo;
    EdTr->ElementType = ElementTransformation::EDGE;
+   EdTr->mesh = this;
    DenseMatrix &pm = EdTr->GetPointMat();
    EdTr->Reset();
    if (Nodes == NULL)
@@ -1098,6 +1105,7 @@ FaceElementTransformations *Mesh::GetBdrFaceTransformations(int BdrElemNo)
    tr->Attribute = boundary[BdrElemNo]->GetAttribute();
    tr->ElementNo = BdrElemNo;
    tr->ElementType = ElementTransformation::BDR_FACE;
+   tr->mesh = this;
    return tr;
 }
 
@@ -10144,61 +10152,6 @@ void Mesh::PrintBdrVTU(std::string fname,
    PrintVTU(fname, format, high_order_output, compression_level, true);
 }
 
-template <typename T>
-void WriteBinaryOrASCII(std::ostream &out, std::vector<char> &buf, const T &val,
-                        const char *suffix, VTKFormat format)
-{
-   if (format == VTKFormat::ASCII) { out << val << suffix; }
-   else { bin_io::AppendBytes(buf, val); }
-}
-
-// Ensure ASCII output of uint8_t to stream is integer rather than character
-template <>
-void WriteBinaryOrASCII<uint8_t>(std::ostream &out, std::vector<char> &buf,
-                                 const uint8_t &val, const char *suffix,
-                                 VTKFormat format)
-{
-   if (format == VTKFormat::ASCII) { out << static_cast<int>(val) << suffix; }
-   else { bin_io::AppendBytes(buf, val); }
-}
-
-template <>
-void WriteBinaryOrASCII<double>(std::ostream &out, std::vector<char> &buf,
-                                const double &val, const char *suffix,
-                                VTKFormat format)
-{
-   if (format == VTKFormat::BINARY32)
-   {
-      bin_io::AppendBytes<float>(buf, float(val));
-   }
-   else if (format == VTKFormat::BINARY)
-   {
-      bin_io::AppendBytes(buf, val);
-   }
-   else
-   {
-      out << val << suffix;
-   }
-}
-
-template <>
-void WriteBinaryOrASCII<float>(std::ostream &out, std::vector<char> &buf,
-                               const float &val, const char *suffix,
-                               VTKFormat format)
-{
-   if (format == VTKFormat::BINARY) { bin_io::AppendBytes<double>(buf, val); }
-   else if (format == VTKFormat::BINARY32) { bin_io::AppendBytes(buf, val); }
-   else { out << val << suffix; }
-}
-
-void WriteBase64WithSizeAndClear(std::ostream &out, std::vector<char> &buf,
-                                 int compression_level)
-{
-   WriteVTKEncodedCompressed(out, buf.data(), buf.size(), compression_level);
-   out << '\n';
-   buf.clear();
-}
-
 void Mesh::PrintVTU(std::ostream &out, int ref, VTKFormat format,
                     bool high_order_output, int compression_level,
                     bool bdr_elements)

mesh/ncmesh.hpp

@@ -737,6 +737,8 @@ protected: // implementation
 
       Point() { dim = 0; }
 
+      Point(const Point &) = default;
+
       Point(double x)
       { dim = 1; coord[0] = x; }
 

mesh/pmesh.cpp

@@ -1890,6 +1890,7 @@ void ParMesh::GetFaceNbrElementTransformation(
    ElTr->Attribute = elem->GetAttribute();
    ElTr->ElementNo = NumOfElements + i;
    ElTr->ElementType = ElementTransformation::ELEMENT;
+   ElTr->mesh = this;
    ElTr->Reset();
 
    if (Nodes == NULL)
@@ -2711,6 +2712,7 @@ STable3D *ParMesh::GetFaceNbrElementToFaceTable(int ret_ftbl)
    }
    face_nbr_el_to_face->Finalize();
 
+   delete sfaces_tbl;
    if (ret_ftbl)
    {
       return faces_tbl;
@@ -6178,6 +6180,9 @@ void ParMesh::Destroy()
       FreeElement(shared_edges[i]);
    }
    shared_edges.DeleteAll();
+
+   delete face_nbr_el_to_face;
+   face_nbr_el_to_face = NULL;
 }
 
 ParMesh::~ParMesh()