make style

update linear form move ctor/assign to correctly move recently added LinearFormExtension member ext
Merge branch 'master' into move-nlf-dev
2022-06-30 12:33:11 -06:00 · 2022-06-30 12:30:45 -06:00 · 2022-06-30 12:27:48 -06:00 · 2022-03-24 10:11:05 -06:00 · 2022-01-19 10:06:55 -07:00 · 2022-01-17 15:14:17 -07:00
357 changed files with 5938 additions and 26757 deletions
@@ -23,8 +23,9 @@ install:
 - set MSMPI_LIB64=C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x64
 - set MSMPI_INC=C:\Program Files (x86)\Microsoft SDKs\MPI\Include

-# Install METIS, use MFEM's mirror because the original source server is often
-# down and we don't support yet the new repo https://github.com/KarypisLab/METIS
+# Install METIS, use a mirror because the original source server is not always
+# up. Original url:
+#    http://glaros.dtc.umn.edu/gkhome/fetch/sw/metis/metis-5.1.0.tar.gz
 - ps: Start-FileDownload 'https://mfem.github.io/tpls/metis-5.1.0.tar.gz'
 - 7z x metis-5.1.0.tar.gz -so | 7z x -si -ttar > nul
 - cd metis-5.1.0
@@ -1,4 +1,4 @@
-name: "Docker"
+name: Build Deploy Container

 on:

@@ -28,8 +28,7 @@ jobs:
      matrix:

        # Dockerfiles to build, a matrix supports future expanded builds
-        container: [["config/docker/Dockerfile.base", "ghcr.io/mfem/mfem-ubuntu-base"],
-                    ["config/docker/Dockerfile", "ghcr.io/mfem/mfem-ubuntu"]]
+        container: [["config/docker/Dockerfile", "ghcr.io/mfem/mfem-ubuntu-base"]]

    runs-on: ubuntu-latest
    name: Build
@@ -10,7 +10,7 @@
 # CONTRIBUTING.md for details.

 # In this CI section, we build different variants of mfem and run test on them.
-name: "Tests"
+name: builds-and-tests

 # Github actions can use the default "GITHUB_TOKEN". By default, this token
 # is set to have permissive access. However, this is not a good practice
@@ -47,17 +47,17 @@ jobs:
  builds-and-tests:
    strategy:
      matrix:
-        os: [ubuntu-latest, macos-latest, windows-latest]
+        os: [ubuntu-20.04, macos-10.15, windows-2022]
        target: [dbg, opt]
        mpi: [seq, par]
        build-system: [make, cmake]
        hypre-target: [int32]
        exclude:
-          - os: ubuntu-latest
+          - os: ubuntu-20.04
            build-system: cmake
-          - os: macos-latest
+          - os: macos-10.15
            build-system: cmake
-          - os: windows-latest
+          - os: windows-2022
            build-system: make
        # 'include' allows us to:
        # - Add a variable to all jobs without creating a new matrix dimension.
@@ -72,15 +72,15 @@ jobs:
            codecov: NO
          - target: opt
            codecov: YES
-          - os: windows-latest
+          - os: windows-2022
            codecov: NO
-          - os: ubuntu-latest
+          - os: ubuntu-20.04
            target: opt
            codecov: NO
            mpi: par
            build-system: cmake
            hypre-target: int32
-          - os: ubuntu-latest
+          - os: ubuntu-20.04
            target: opt
            codecov: NO
            mpi: par
@@ -112,35 +112,35 @@ jobs:
    # TODO: It would be nice to have only one step, e.g. with a dedicated
    # action, but I (@adrienbernede) don't see how at the moment.
    - name: get MPI (Linux)
-      if: matrix.mpi == 'par' && matrix.os == 'ubuntu-latest'
+      if: matrix.mpi == 'par' && matrix.os == 'ubuntu-20.04'
      run: |
        sudo apt-get install mpich libmpich-dev
        export MAKE_CXX_FLAG="MPICXX=mpic++"

    - name: get lcov (Linux)
-      if: matrix.codecov == 'YES' && matrix.os == 'ubuntu-latest'
+      if: matrix.codecov == 'YES' && matrix.os == 'ubuntu-20.04'
      run: |
        sudo apt-get install lcov

    - name: Set up Homebrew
-      if: ( matrix.mpi == 'par' || matrix.codecov == 'YES' ) && matrix.os == 'macos-latest'
+      if: ( matrix.mpi == 'par' || matrix.codecov == 'YES' ) && matrix.os == 'macos-10.15'
      uses: Homebrew/actions/setup-homebrew@c4aafe8c4620bf08883dd4679c374f11e73329d3

    - name: get MPI (MacOS)
-      if: matrix.mpi == 'par' && matrix.os == 'macos-latest'
+      if: matrix.mpi == 'par' && matrix.os == 'macos-10.15'
      run: |
        export HOMEBREW_NO_INSTALL_CLEANUP=1
        brew install openmpi
        export MAKE_CXX_FLAG="MPICXX=mpic++"

    - name: get MPI (MacOS)
-      if: matrix.codecov == 'YES' && matrix.os == 'macos-latest'
+      if: matrix.codecov == 'YES' && matrix.os == 'macos-10.15'
      run: |
        export HOMEBREW_NO_INSTALL_CLEANUP=1
        brew install lcov

    - name: get MPI (Windows)
-      if: matrix.mpi == 'par' && matrix.os == 'windows-latest'
+      if: matrix.mpi == 'par' && matrix.os == 'windows-2022'
      uses: mpi4py/setup-mpi@v1.0.3

    # Get Hypre through cache, or build it.
@@ -154,7 +154,7 @@ jobs:
        key: ${{ runner.os }}-build-${{ env.HYPRE_TOP_DIR }}-${{ matrix.hypre-target }}-v2.2

    - name: get hypre
-      if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os != 'windows-latest'
+      if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os != 'windows-2022'
      uses: mfem/github-actions/build-hypre@v2.2
      with:
        archive: ${{ env.HYPRE_ARCHIVE }}
@@ -163,7 +163,7 @@ jobs:
        build-system: make

    - name: get hypre (Windows)
-      if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os == 'windows-latest'
+      if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os == 'windows-2022'
      uses: mfem/github-actions/build-hypre@v2.2
      with:
        archive: ${{ env.HYPRE_ARCHIVE }}
@@ -175,14 +175,14 @@ jobs:
    # Install will only run on cache miss.
    - name: cache metis
      id: metis-cache
-      if: matrix.mpi == 'par' && matrix.os != 'windows-latest'
+      if: matrix.mpi == 'par' && matrix.os != 'windows-2022'
      uses: actions/cache@v2
      with:
        path: ${{ env.METIS_TOP_DIR }}
        key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.2

    - name: install metis
-      if: matrix.mpi == 'par' && matrix.os != 'windows-latest' && steps.metis-cache.outputs.cache-hit != 'true'
+      if: matrix.mpi == 'par' && matrix.os != 'windows-2022' && steps.metis-cache.outputs.cache-hit != 'true'
      uses: mfem/github-actions/build-metis@v2.2
      with:
        archive: ${{ env.METIS_ARCHIVE }}
@@ -196,16 +196,16 @@ jobs:
        key: ${{ runner.os }}-${{ matrix.mpi }}-vcpkg-v1

    - name: prepare binary cache location
-      if: matrix.os == 'windows-latest' && steps.vcpkg-cache.outputs.cache-hit != 'true'
+      if: matrix.os == 'windows-2022' && steps.vcpkg-cache.outputs.cache-hit != 'true'
      run: |
        mkdir -p vcpkg_cache

    - name: install metis (Windows)
-      if: matrix.mpi == 'par' && matrix.os == 'windows-latest'
+      if: matrix.mpi == 'par' && matrix.os == 'windows-2022'
      env:
        VCPKG_DEFAULT_BINARY_CACHE: ${{ github.workspace }}/vcpkg_cache
      run: |
-        vcpkg install metis-mfem --triplet=x64-windows-static --overlay-ports=${{ env.MFEM_TOP_DIR }}/config/vcpkg/ports
+        vcpkg install metis --triplet=x64-windows-static

    # MFEM build and test
    - name: build
@@ -248,7 +248,7 @@ jobs:
      shell: bash

    - name: cmake unit tests (Ubuntu 20.04)
-      if: matrix.build-system == 'cmake' && matrix.target == 'opt' && matrix.os == 'ubuntu-latest'
+      if: matrix.build-system == 'cmake' && matrix.target == 'opt' && matrix.os == 'ubuntu-20.04'
      run: |
        CTEST_CONFIG="Release"
        [[ ${{ matrix.target }} == 'dbg' ]] && CTEST_CONFIG="Debug"
@@ -256,7 +256,7 @@ jobs:
      shell: bash

    - name: cmake tests
-      if: matrix.build-system == 'cmake' && matrix.target == 'opt' && matrix.os != 'ubuntu-latest'
+      if: matrix.build-system == 'cmake' && matrix.target == 'opt' && matrix.os != 'ubuntu-20.04'
      run: |
        CTEST_CONFIG="Release"
        cd ${{ env.MFEM_TOP_DIR }}/build && ctest --output-on-failure -C ${CTEST_CONFIG}
@@ -1,70 +0,0 @@
-# For most projects, this workflow file will not need changing; you simply need
-# to commit it to your repository.
-#
-# You may wish to alter this file to override the set of languages analyzed,
-# or to provide custom queries or build logic.
-#
-# ******** NOTE ********
-# We have attempted to detect the languages in your repository. Please check
-# the `language` matrix defined below to confirm you have the correct set of
-# supported CodeQL languages.
-#
-name: "Static Analysis"
-
-on:
-  push:
-    branches: [ "master", "next"]
-  pull_request:
-    # The branches below must be a subset of the branches above
-    branches: [ "master" ]
-
-jobs:
-  analyze:
-    name: Analyze
-    runs-on: ubuntu-latest
-    permissions:
-      actions: read
-      contents: read
-      security-events: write
-
-    strategy:
-      fail-fast: false
-      matrix:
-        language: [ 'cpp' ]
-        # CodeQL supports [ 'cpp', 'csharp', 'go', 'java', 'javascript', 'python', 'ruby' ]
-        # Learn more about CodeQL language support at https://aka.ms/codeql-docs/language-support
-
-    steps:
-    - name: Checkout repository
-      uses: actions/checkout@v3
-
-    # Initializes the CodeQL tools for scanning.
-    - name: Initialize CodeQL
-      uses: github/codeql-action/init@v2
-      with:
-        languages: ${{ matrix.language }}
-        # If you wish to specify custom queries, you can do so here or in a config file.
-        # By default, queries listed here will override any specified in a config file.
-        # Prefix the list here with "+" to use these queries and those in the config file.
-
-        # Details on CodeQL's query packs refer to : https://docs.github.com/en/code-security/code-scanning/automatically-scanning-your-code-for-vulnerabilities-and-errors/configuring-code-scanning#using-queries-in-ql-packs
-        # queries: security-extended,security-and-quality
-
-
-    # Autobuild attempts to build any compiled languages  (C/C++, C#, or Java).
-    # If this step fails, then you should remove it and run the build manually (see below)
-    - name: Autobuild
-      uses: github/codeql-action/autobuild@v2
-
-    # ℹ️ Command-line programs to run using the OS shell.
-    # 📚 See https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#jobsjob_idstepsrun
-
-    #   If the Autobuild fails above, remove it and uncomment the following three lines.
-    #   modify them (or add more) to build your code if your project, please refer to the EXAMPLE below for guidance.
-
-    # - run: |
-    #   echo "Run, Build Application using script"
-    #   ./location_of_script_within_repo/buildscript.sh
-
-    - name: Perform CodeQL Analysis
-      uses: github/codeql-action/analyze@v2
@@ -9,7 +9,7 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.

-name: "Build Analysis"
+name: build-analysis

 permissions:
  actions: write
@@ -31,7 +31,7 @@ env:

 jobs:
  gitignore:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-18.04

    steps:
    - name: Cancel Previous Runs
@@ -9,7 +9,7 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.

-name: "Checks"
+name: repo-check

 permissions:
  actions: write
@@ -28,7 +28,7 @@ on:

 jobs:
  file-headers-check:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-18.04
    if: |
      (github.event_name == 'push' ||
       github.event.pull_request.head.repo.full_name != github.repository)
@@ -78,7 +78,7 @@ jobs:
        exit 1

  code-style:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-18.04
    if: |
      (github.event_name == 'push' ||
       github.event.pull_request.head.repo.full_name != github.repository)
@@ -88,14 +88,14 @@ jobs:

    - name: get astyle
      run: |
-        sudo apt-get install astyle
+        sudo apt-get install astyle=3.1-1ubuntu2

    - name: style check
      run: |
        ./config/githooks/pre-push --style

  documentation:
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-18.04
    if: |
      (github.event_name == 'push' ||
       github.event.pull_request.head.repo.full_name != github.repository)
@@ -106,8 +106,6 @@ jobs:
    - name: get doxygen and graphviz
      run: |
        sudo apt-get install doxygen graphviz
-        cd doc
-        doxygen -u CodeDocumentation.conf.in

    - name: build documentation
      run: |
@@ -120,7 +118,7 @@ jobs:
      github.ref != 'refs/heads/master' &&
      (github.event_name == 'push' ||
       github.event.pull_request.head.repo.full_name != github.repository)
-    runs-on: ubuntu-latest
+    runs-on: ubuntu-18.04
    steps:
    - name: checkout mfem
      uses: actions/checkout@v2
@@ -18,11 +18,6 @@ CMakeFiles/
 # Backup files
 *~

-*.sqlite
-*.nsys-rep
-*.qdstrm
-*.csv
-
 # Default install location
 /mfem/

@@ -280,7 +275,6 @@ miniapps/tools/load-dc
 miniapps/tools/convert-dc
 miniapps/tools/lor-transfer
 miniapps/tools/get-values
-miniapps/tools/check-tmop-metric

 miniapps/toys/automata
 miniapps/toys/life
@@ -313,9 +307,6 @@ miniapps/solvers/sol.*
 miniapps/parelag/MultilevelHcurlHdivSolver
 miniapps/parelag/*.mesh

-miniapps/multidomain/multidomain
-miniapps/hooke/hooke
-
 # Unit test binary and outputs
 tests/unit/output_meshes
 tests/unit/unit_tests
@@ -334,7 +325,6 @@ tests/benchmarks/bench_ceed
 tests/benchmarks/bench_tmop
 tests/benchmarks/bench_vector
 tests/benchmarks/bench_virtuals
-tests/benchmarks/bench_lor

 # Test script output
 tests/scripts/*.err
@@ -8,48 +8,10 @@
                               https://mfem.org


-Version 4.5, released on October 22, 2022
-=========================================
-
-Meshing improvements
--------------------
- Added new SubMesh and ParSubMesh classes that can be used to extract a subset
-  of a given Mesh. These classes have the same functionality as Mesh and ParMesh
-  and work with all existing MFEM interfaces like finite element spaces etc.
-
- Added a method, ParMesh::GetSerialMesh(), that reconstructs a partitioned
-  parallel mesh on a given single rank. Also, added ParMesh::PrintAsSerial(),
-  which saves the reconstructed serial mesh to a C++ stream on rank 0.
-
- Added more 3D TMOP metrics, as well as specialized metrics for mesh
-  untangling and worst-case quality improvement.
-
- Added a new method, Mesh::NodesUpdated, which should be called after the mesh
-  node coordinates have changed, e.g. after the mesh has moved. This is
-  necessary, for example, with device assembly of linear and bilinear forms.
-
- Added support for mixed meshes and pyramids in GSLIB-FindPoints.
-
-Discretization improvements
---------------------------
- Added full assembly and device support for several LinearForm integrators:
-  * DomainLF: (f, v)
-  * VectorDomainLF: ((f1,...,fn), (v1,...,vn))
-  * DomainLFGrad: (f, grad(v))
-  * VectorDomainLFGrad: ((f1x,f1y,f1z,...,fnx,fny,fnz), grad(v1,...,vn))
-  The device assembly of linear forms has to be explicitly enabled by calling
-  LinearForm::UseFastAssembly(true), otherwise the legacy linear form assembly
-  is used by default.
-
- Added support for assembling low-order-refined matrices using a GPU-enabled
-  "batched" algorithm. The lor_solvers and plor_solvers now fully support GPU
-  acceleration with arbitrary user-supplied coefficients.
-
- Added a new class FaceQuadratureSpace that allows for the construction of
-  QuadratureFunctions on the interior or boundary faces of a mesh.
-
- Added a class CoefficientVector for efficient access of variable coefficient
-  values at quadrature points (in particular for GPU/device kernels).
+Version 4.4.1 (development)
+===========================
+- Added example for body-fitted volumetric and shape integration using the 
+  Algoim library.

 - Added WhiteGaussianNoiseDomainLFIntegrator: a LinearFormIntegrator class for
  spatial Gaussian white noise.
@@ -57,97 +19,30 @@ Discretization improvements
 - Added a new Zienkiewicz-Zhu patch recovery-based a posteriori error estimator.
  See fem/estimators.hpp.

- Various fixes and improvements in LinearFormExtension.
-
-Linear and nonlinear solvers
----------------------------
- Added a new class DGMassInverse that performs a local element-wise CG
-  iteration to solve systems involving the discontinuous Galerkin mass matrix,
-  including support for device/GPU acceleration.
-
- Added more flexibility to the constrained solver classes:
-  * PenaltyConstrainedSolver now allows for a vector of penalty parameters
-    (necessary for penalty contact)
-  * PenaltyConstrainedSolver and EliminationSolver can use GMRES or PCG
-  * All constraint solver classes can take a user-defined preconditioner
-
- - Added functions to toggle additional options for the SuperLU_Dist and Hypre
-   preconditioners (ParaSails, Euclid, ILU).
-
- Added boundary elimination with device support for `SparseMatrix` and
-  `HypreParMatrix`.
-
-New and updated examples and miniapps
-------------------------------------
- Added a new elasticity miniapp, Hooke, that showcases a low-level approach of
-  using MFEM to solve a nonlinear elasticity problem based on the fundamental
-  finite element operator decomposition. The miniapp also integrates with
-  automatic differentiation tools like a native dual number implementation or a
-  third party library such as Enzyme. See miniapps/elasticity for more details.
-
- Added example for body-fitted volumetric and shape integration using the
-  Algoim library in miniapps/shifted.
-
- Add a new example code, Example 33/33p, to demonstrate the solution of
-  spectral fractional PDEs with MFEM.
-
-Integrations, testing and documentation
---------------------------------------
- Added a Dockerfile for a simple MFEM container, see config/docker/README.md.
-  More sophisticated developer containers are available in the new repo
-  https://github.com/mfem/containers.
-
- Added support for the LLVM-based automatic differentiation tool Enzyme, see
-  https://github.com/EnzymeAD/Enzyme. Build system flags and a convenience
-  header are provided. The functionality and interaction are demonstrated in a
-  new miniapp in miniapps/elasticity.
-
- Added support for partial assembly and fully matrix-free operators on mixed
-  meshes (different element types and p-adaptivity) through libCEED, including
-  device acceleration, e.g. with NVIDIA and AMD GPUs. The p-adaptivity is
-  currently limited to 2D serial meshes. All mixed element topologies are
-  supported in both serial and parallel.
-
 - Added support for ParMoonolith, https://bitbucket.org/zulianp/par_moonolith,
  which provides parallel non-conforming, non-matching, variational, volumetric
  mesh information transfer. With ParMortarAssember, fields can be exchanged
  between arbitrarily distributed and unrelated finite element meshes in a
  variationally consistent way.

- Fully encapsulated SUNDIALS `N_Vector` object within the `SundialsNVector`
-  class by removing deprecated (e.g. `HypreParVector::ToNVector`) and
-  non-deprecated (e.g. `Vector::ToNVector`) functions in other classes.
+- Added full assembly and device support for several LinearForm integrators:
+  * DomainLF: (f, v)
+  * VectorDomainLF: ((f1,...,fn), (v1,...,vn))
+  * DomainLFGrad: (f, grad(v))
+  * VectorDomainLFGrad: ((f1x,f1y,f1z,...,fnx,fny,fnz), grad(v1,...,vn))

- New benchmark for the different assembly levels inspired by the CEED
-  Bake-Off Problems, see tests/benchmarks/bench_assembly_levels.cpp.
+- Add a new example code, Example 33/33p, to demonstrate the solution of
+  spectral fractional PDEs with MFEM.
+
+- Added a Dockerfile for a simple MFEM container, see config/docker/README.md.
+
+- Added support for assembling low-order-refined matrices using a GPU-enabled
+  "batched" algorithm. The lor_solvers and plor_solvers now fully support GPU
+  acceleration.

 - Added Windows 2022 CI testing with GitHub actions.

-Miscellaneous
-------------
- The method SparseMatrix::EnsureMultTranspose() is now automatically called
-  by the methods AddMultTranspose(), MultTranspose(), and AbsMultTranspose().
-  Added a method with the same name to class HypreParMatrix which is also called
-  automatically by the HypreParMatrix::MultTranspose() methods.
-
- Updated various MemoryUsage methods to return 'std::size_t' instead of 'long'
-  since the latter is 32-bit in Win64 builds.
-
- When using `AssemblyLevel::FULL`, `FABilinearFormExtension::FormSystemMatrix`
-  outputs an `OperatorHandle` containing a `SparseMatrix` in serial, and an
-  `HypreParMatrix` in parallel (instead of a `ConstrainedOperator`).
-
- In various places in the library, replace the use of 'long' with 'long long'
-  to better support Win64 builds where 'long' is 32-bit and 'long long' is
-  64-bit. On Linux and MacOS, both types are typically 64-bit.
-
- The behavior of GridFunction::GetTrueVector() has been changed to not return
-  an empty true vector.
-
- Added support for ordering search points byVDIM in FindPointsGSLIB.
-
- Various other simplifications, extensions, and bugfixes in the code.
-
+- Added support for mixed meshes and pyramids in GSLIB-FindPoints.

 Version 4.4, released on March 21, 2022
 =======================================
@@ -269,6 +164,9 @@ Integrations, testing and documentation
 - Switched from Artistic Style (astyle) version 2.05.1 to version 3.1 for code
  formatting. See the "make style" target.

+- New benchmark for the different assembly levels inspired by the CEED
+  Bake-Off Problems, see tests/benchmarks/bench_assembly_levels.cpp.
+
 Miscellaneous
 -------------
 - Added a simple singleton class, Mpi, as a replacement for MPI_Session. New
@@ -51,7 +51,7 @@ project(mfem NONE)
 # Current version of MFEM, see also `makefile`.
 #   mfem_VERSION = (string)
 #   MFEM_VERSION = (int)   [automatically derived from mfem_VERSION]
-set(${PROJECT_NAME}_VERSION 4.5.0)
+set(${PROJECT_NAME}_VERSION 4.4.1)

 # Prohibit in-source build
 if (${PROJECT_SOURCE_DIR} STREQUAL ${PROJECT_BINARY_DIR})
@@ -81,10 +81,6 @@ if (MFEM_USE_STRUMPACK)
  # Just needed to find the MPI_Fortran libraries to link with
  set(XSDK_ENABLE_Fortran ON)
 endif()
-# SUNDIALS >= 6.4.0 requires C++14:
-if (MFEM_USE_SUNDIALS AND ("${CMAKE_CXX_STANDARD}" LESS "14"))
-  set(CMAKE_CXX_STANDARD 14)
-endif()
 if (MFEM_USE_GINKGO AND ("${CMAKE_CXX_STANDARD}" LESS "14"))
  set(CMAKE_CXX_STANDARD 14)
 endif()
@@ -140,8 +136,6 @@ if (MFEM_USE_CUDA)
       "CUDA flags set for MFEM" FORCE)
   set(CUSPARSE_FOUND TRUE)
   set(CUSPARSE_LIBRARIES "cusparse")
-   set(CUBLAS_FOUND TRUE)
-   set(CUBLAS_LIBRARIES "cublas")
 endif()

 if (XSDK_ENABLE_C)
@@ -204,10 +198,10 @@ if (MFEM_USE_MPI)
  find_package(MPI REQUIRED)
  set(MPI_CXX_INCLUDE_DIRS ${MPI_CXX_INCLUDE_PATH})
  if (MFEM_MPIEXEC)
-    string(REPLACE " " ";" MPIEXEC ${MFEM_MPIEXEC})
+    set(MPIEXEC ${MFEM_MPIEXEC})
  endif()
  if (MFEM_MPIEXEC_NP)
-    string(REPLACE " " ";" MPIEXEC_NUMPROC_FLAG ${MFEM_MPIEXEC_NP})
+    set(MPIEXEC_NUMPROC_FLAG ${MFEM_MPIEXEC_NP})
  endif()
  # Parallel MFEM depends on hypre
  find_package(HYPRE REQUIRED)
@@ -458,11 +452,6 @@ if (MFEM_USE_PARELAG)
   find_package(PARELAG REQUIRED)
 endif()

-# Enzyme
-if (MFEM_USE_ENZYME)
-   find_package(ENZYME REQUIRED)
-endif()
-
 # MFEM_TIMER_TYPE
 if (NOT DEFINED MFEM_TIMER_TYPE)
  if (APPLE)
@@ -481,25 +470,16 @@ if (NOT DEFINED MFEM_TIMER_TYPE)
  endif()
 endif()

-# Without this, CMake 3.21.1 (and 3.20.2) run into CMake Errors like the following:
-# CMake Error at config/cmake/modules/MfemCmakeUtilities.cmake:60 (add_library):
-#   Target "mfem" links to target "Threads::Threads" but the target was not
-#   found.  Perhaps a find_package() call is missing for an IMPORTED target, or
-#   an ALIAS target is missing?
-# Call Stack (most recent call first):
-#   CMakeLists.txt:474 (mfem_add_library)
-find_package(Threads REQUIRED)
-
 # List all possible libraries in order of dependencies.
 # [METIS < SuiteSparse]:
 #    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 LAPACK BLAS SuperLUDist METIS SuiteSparse SUNDIALS
+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 CUBLAS CUSPARSE MKL_CPARDISO AMGX CALIPER CODIPACK BENCHMARK PARELAG
-    MPI_CXX HIP HIPSPARSE MOONOLITH BLITZ ALGOIM ENZYME)
+    ADIOS2 CUSPARSE MKL_CPARDISO AMGX CALIPER CODIPACK BENCHMARK PARELAG
+    MPI_CXX HIP HIPSPARSE MOONOLITH BLITZ ALGOIM)

 # Add all *_FOUND libraries in the variable TPL_LIBRARIES.
 set(TPL_LIBRARIES "")
@@ -102,9 +102,7 @@ The MFEM source code has the following structure:
  .
  ├── config
  │   ├── cmake
-  │   ├── docker
-  │   ├── githooks
-  │   └── vcpkg
+  │   └── githooks
  ├── data
  ├── doc
  ├── examples
@@ -113,7 +111,6 @@ The MFEM source code has the following structure:
  │   ├── ginkgo
  │   ├── hiop
  │   ├── jupyter
-  │   ├── moonolith
  │   ├── petsc
  │   ├── pumi
  │   ├── sundials
@@ -121,25 +118,21 @@ The MFEM source code has the following structure:
  ├── fem
  │   ├── ceed
  │   ├── fe
-  │   ├── lor
-  │   ├── moonolith
  │   ├── qinterp
+  │   ├── moonolith
  │   └── tmop
  ├── general
  ├── linalg
  │   └── simd
  ├── mesh
-  │   └── submesh
  ├── miniapps
  │   ├── adjoint
  │   ├── autodiff
  │   ├── common
  │   ├── electromagnetics
  │   ├── gslib
-  │   ├── hooke
  │   ├── meshing
  │   ├── mtop
-  │   ├── multidomain
  │   ├── navier
  │   ├── nurbs
  │   ├── parelag
@@ -7,10 +7,6 @@

                               https://mfem.org

-This file provides a detailed description of how to build and install the MFEM
-library. For a simple build, see the step-by-step instructions on the website
-at https://mfem.org/building.
-
 The MFEM library has a serial and an MPI-based parallel version, which largely
 share the same code base. The only prerequisite for building the serial version
 of MFEM is a (modern) C++ compiler, such as g++. The parallel version of MFEM
@@ -20,11 +16,7 @@ requires an MPI C++ compiler, as well as the following external libraries:
  https://github.com/hypre-space/hypre

 - METIS (a family of multilevel partitioning algorithms)
-  https://github.com/mfem/tpls
-
-  Note: We recommend our mirror of metis-4.0.3/5.1.0 above because the METIS
-  webpage, http://glaros.dtc.umn.edu/gkhome/metis/metis/overview, is often down
-  and we don't support yet the new repo https://github.com/KarypisLab/METIS.
+  http://glaros.dtc.umn.edu/gkhome/metis/metis/overview

 The hypre dependency can be downloaded as a tarball from GitHub or from the
 project webpage https://www.llnl.gov/casc/hypre. For example, the 2.24.0 release
@@ -480,10 +472,10 @@ MFEM_USE_CODIPACK = YES/NO
  www.scicomp.uni-kl.de/codi/

 MFEM_USE_ALGOIM = YES/NO
-  Enable the usage of Algoim - a collection of high-order accurate numerical
-  methods and C++ algorithms for working with implicitly-defined geometry and
-  level set methods. The Algoim library requires the Blitz++ library. The MFEM
-  provides interface to Algoim v1. Thus, to check out the specific state use:
+  Enable the usage of Algoim - a collection of high-order accurate numerical 
+  methods and C++ algorithms for working with implicitly-defined geometry and 
+  level set methods. The Algoim library requires the Blitz++ library. The MFEM 
+  provides interface to Algoim v1. Thus, to check out the specific state use: 
  git checkout 9c9ca0ef094d8ab0390ed36367a1151b459bbe0a
  https://algoim.github.io

@@ -558,7 +550,7 @@ MFEM_USE_FMS = YES/NO
   Enables support for the FMS library which consists of the DataCollection
   sub-class mfem::FMSDataCollection for I/O in FMS formats, see the header file
   fem/fmsdatacollection.hpp. In addition, this option enables in-memory
-   conversion routines between FMS's FmsDataCollection structure and MFEM's
+   convetion routines between FMS's FmsDataCollection structure and MFEM's
   DataCollection class, see the header file fem/fmsconvert.hpp.

 MFEM_USE_PARELAG = YES/NO
@@ -566,14 +558,6 @@ MFEM_USE_PARELAG = YES/NO
   use ParELAG. In fact, ParELAG is dependent on MFEM. Therefore, this option
   currently only concerns the miniapps.

-MFEM_USE_ENZYME = YES/NO
-   Enables automatic differentiation support through the LLVM plugin Enzyme.
-   This requires the compiler to be set to clang (>=14.0.0). We also advise to
-   use the link time optimization (LTO) plugin, to enable functions that you
-   define over multiple files (compilation units) and want to be differentiated
-   automatically, to work. This requires to also use LLVM/LLD for linking.
-   Recommended options are in config/defaults.mk.
-
 MFEM_BUILD_TAG = (any value)
   An optional tag to characterize the build. Exported to config/config.mk.
   Can be used to identify the MFEM build from other makefiles.
@@ -605,7 +589,7 @@ The specific libraries and their options are:

 - METIS, used when MFEM_USE_METIS = YES. If using METIS 5, set
  MFEM_USE_METIS_5 = YES (default is to use METIS 4).
-  URL: https://github.com/mfem/tpls (MFEM mirror, see above)
+  URL: http://glaros.dtc.umn.edu/gkhome/metis/metis/overview
  Options: METIS_OPT, METIS_LIB.
  Versions: METIS 4.0.3 or 5.1.0.

@@ -762,12 +746,12 @@ The specific libraries and their options are:
  Options: GSLIB_OPT, GSLIB_LIB.
  Versions: GSLIB >= 1.0.7.

- ALGOIM (optional), used when MFEM_USE_ALGOIM=YES. The library provides only
+- ALGOIM (optional), used when MFE_USE_ALGOIM=YES. The library provides only
  headers so it just needs to be downloaded at the same level as MFEM. Download
  the specific version we use as:
  "git clone https://github.com/algoim/algoim.git;
   git checkout 9c9ca0ef094d8ab0390ed36367a1151b459bbe0a"
-  ALGOIM depends on BLITZ and the library must be built prior to the MFEM build.
+  ALGOIM depends on BLITZ and rhe library must be built prior to the MFEM build.
  Download v1.0.2, untar it at the same level as MFEM and create a symbolic link:
  "ln -s blitz-1.0.2 blitz".
  Build Blitz using CMake as:
@@ -776,6 +760,8 @@ The specific libraries and their options are:
  Options: BLITZ_OPT, BLITZ_LIB
  Versions: BLITZ = 1.0.2

+
+
 - MKL CPardiso (optional), used when MFEM_USE_MKL_CPARDISO = YES.
  URL: https://software.intel.com/content/www/us/en/develop/tools/math-kernel-library.html
  Options: MKL_CPARDISO_OPT, MKL_CPARDISO_LIB.
@@ -852,12 +838,6 @@ The specific libraries and their options are:
  URL: https://github.com/LLNL/parelag
  Options: PARELAG_DIR, PARELAG_OPT, PARELAG_LIB.

- Enzyme, used when MFEM_USE_ENZYME = YES. Requires LLVM/Clang >= 14.0.0.
-  URL: https://github.com/EnzymeAD/Enzyme
-  Options: ENZYME_DIR, ENZYME_OPT, ENZYME_LIB.
-  Versions: Enzyme >= v0.0.33.
-
-
 Building with CMake
 ===================
 The MFEM build system consists of two steps: configuration and compilation.
@@ -996,7 +976,6 @@ MFEM_USE_CALIPER
 MFEM_USE_FMS
 MFEM_USE_BENCHMARK
 MFEM_USE_PARELAG
-MFEM_USE_ENZYME

 The following options are CMake specific:

@@ -1056,7 +1035,6 @@ The CMake build system adds auto-detection for the following packages/libraries:
 - FMS
 - BENCHMARK
 - ParELAG
- - Enzyme

 The following built-in CMake packages are also used:

@@ -61,7 +61,6 @@ set(MFEM_USE_CALIPER @MFEM_USE_CALIPER@)
 set(MFEM_USE_ALGOIM @MFEM_USE_ALGOIM@)
 set(MFEM_USE_BENCHMARK @MFEM_USE_BENCHMARK@)
 set(MFEM_USE_PARELAG @MFEM_USE_PARELAG@)
-set(MFEM_USE_ENZYME @MFEM_USE_ENZYME@)

 set(MFEM_CXX_COMPILER "@CMAKE_CXX_COMPILER@")
 set(MFEM_CXX_FLAGS "@CMAKE_CXX_FLAGS@")
@@ -190,7 +190,4 @@
 // Enable MFEM functionality based on the Google Benchmark library.
 #cmakedefine MFEM_USE_BENCHMARK

-// Enable Enzyme for AD
-#cmakedefine MFEM_USE_ENZYME
-
 #endif // MFEM_CONFIG_HEADER
@@ -1,27 +0,0 @@
-# Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
-# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-# LICENSE and NOTICE for details. LLNL-CODE-806117.
-#
-# This file is part of the MFEM library. For more information and source code
-# availability visit https://mfem.org.
-#
-# MFEM is free software; you can redistribute it and/or modify it under the
-# terms of the BSD-3 license. We welcome feedback and contributions, see file
-# CONTRIBUTING.md for details.
-
-message(STATUS "Looking for ENZYME ...")
-message(STATUS "   in ENZYME_DIR = ${ENZYME_DIR}")
-
-# Make sure the directory and version combination works. Do nothing otherwise.
-if(EXISTS "${ENZYME_DIR}/ClangEnzyme-${ENZYME_VERSION}.so")
-  message(STATUS "Found ENZYME: ${ENZYME_DIR}/ClangEnzyme-${ENZYME_VERSION}.so")
-
-  # Set ENZYME_FOUND
-  set(ENZYME_FOUND TRUE CACHE BOOL "ENZYME was found." FORCE)
-
-  # Set CXX flags to accommodate the Enzyme Clang plugin
-  set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Xclang -load -Xclang ${ENZYME_DIR}/ClangEnzyme-${ENZYME_VERSION}.so -mllvm -enzyme-loose-types=1")
-  set(MFEM_USE_ENZYME YES)
-else()
-
-endif()
@@ -1,57 +0,0 @@
-# Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
-# at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-# LICENSE and NOTICE for details. LLNL-CODE-806117.
-#
-# This file is part of the MFEM library. For more information and source code
-# availability visit https://mfem.org.
-#
-# MFEM is free software; you can redistribute it and/or modify it under the
-# terms of the BSD-3 license. We welcome feedback and contributions, see file
-# CONTRIBUTING.md for details.
-
-# Defines the following variables:
-#   - HDF5_FOUND - If HDF5 was found
-#   - HDF5_LIBRARIES - The HDF5 libraries
-#   - HDF5_INCLUDE_DIRS - The HDF5 include directories
-
-# First Check for HDF5_DIR
-if(NOT HDF5_DIR)
-    MESSAGE(FATAL_ERROR "Could not find HDF5. HDF5 support needs explicit HDF5_DIR")
-endif()
-
-# Find includes
-find_path( HDF5_INCLUDE_DIRS hdf5.h
-           PATHS  ${HDF5_DIR}/include/
-           NO_DEFAULT_PATH
-           NO_CMAKE_ENVIRONMENT_PATH
-           NO_CMAKE_PATH
-           NO_SYSTEM_ENVIRONMENT_PATH
-           NO_CMAKE_SYSTEM_PATH)
-
-find_library( __HDF5_LIBRARY NAMES hdf5 libhdf5 libhdf5_D libhdf5_debug
-              PATHS ${HDF5_DIR}/lib
-              NO_DEFAULT_PATH
-              NO_CMAKE_ENVIRONMENT_PATH
-              NO_CMAKE_PATH
-              NO_SYSTEM_ENVIRONMENT_PATH
-              NO_CMAKE_SYSTEM_PATH)
-
-find_library( __HDF5_HL_LIBRARY NAMES hdf5_hl libhdf5_hl libhdf5_hl_D libhdf5_hl_debug
-              PATHS ${HDF5_DIR}/lib
-              NO_DEFAULT_PATH
-              NO_CMAKE_ENVIRONMENT_PATH
-              NO_CMAKE_PATH
-              NO_SYSTEM_ENVIRONMENT_PATH
-              NO_CMAKE_SYSTEM_PATH)
-
-set(HDF5_LIBRARIES ${__HDF5_HL_LIBRARY} ${__HDF5_LIBRARY})
-
-include(FindPackageHandleStandardArgs)
-
-# Handle the QUIETLY and REQUIRED arguments and set HDF5_FOUND to TRUE if all
-# listed variables are TRUE
-find_package_handle_standard_args(HDF5  DEFAULT_MSG
-                                  HDF5_INCLUDE_DIRS
-                                  __HDF5_LIBRARY
-                                  __HDF5_HL_LIBRARY
-                                  HDF5_LIBRARIES )
@@ -14,6 +14,6 @@
 #   - UMPIRE_LIBRARIES
 #   - UMPIRE_INCLUDE_DIRS

-find_package(umpire REQUIRED CONFIG)
-set(UMPIRE_FOUND ${umpire_FOUND})
-set(UMPIRE_LIBRARIES "umpire")
+include(MfemCmakeUtilities)
+mfem_find_package(UMPIRE UMPIRE UMPIRE_DIR "include" "umpire/Umpire.hpp" "lib" "umpire"
+  "Paths to headers required by UMPIRE." "Libraries required by UMPIRE.")
@@ -43,14 +43,22 @@ function(convert_filenames_to_full_paths NAMES)
  set(${NAMES} ${tmp_names} PARENT_SCOPE)
 endfunction()

-# Wrapper for add_executable
+# Wrapper for add_executable that calls the HIP wrapper if applicable
 macro(mfem_add_executable NAME)
-  add_executable(${NAME} ${ARGN})
+  if (MFEM_USE_HIP)
+    add_executable(${NAME} ${ARGN})
+  else()
+    add_executable(${NAME} ${ARGN})
+  endif()
 endmacro()

-# Wrapper for add_library
+# Wrapper for add_library that calls the HIP wrapper if applicable
 macro(mfem_add_library NAME)
-  add_library(${NAME} ${ARGN})
+  if (MFEM_USE_HIP)
+    add_library(${NAME} ${ARGN})
+  else()
+    add_library(${NAME} ${ARGN})
+  endif()
 endmacro()

 # Simple shortcut to add_custom_target() with option to add the target to the
@@ -886,7 +894,7 @@ function(mfem_export_mk_files)
      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
-      MFEM_USE_MOONOLITH MFEM_USE_ALGOIM MFEM_USE_ENZYME)
+      MFEM_USE_MOONOLITH MFEM_USE_ALGOIM)
  foreach(var ${CONFIG_MK_BOOL_VARS})
    if (${var})
      set(${var} YES)
@@ -31,11 +31,9 @@

 // Windows specific options
 #ifdef _WIN32
-#ifndef _USE_MATH_DEFINES
 // Macro needed to get defines like M_PI from <cmath>. (Visual Studio C++ only?)
 #define _USE_MATH_DEFINES
 #endif
-#endif
 // On Cygwin the option -std=c++11 prevents the definition of M_PI. Defining
 // the following macro allows us to get M_PI and some needed functions, e.g.
 // posix_memalign(), strdup(), strerror_r().
@@ -195,7 +195,4 @@
 // Enable functionality based on the Google Benchmark library.
 // #define MFEM_USE_BENCHMARK

-// Enable the Enzyme LLVM plugin
-// #define MFEM_USE_ENZYME
-
 #endif // MFEM_CONFIG_HEADER
@@ -63,7 +63,6 @@ MFEM_USE_ADFORWARD     = @MFEM_USE_ADFORWARD@
 MFEM_USE_CODIPACK      = @MFEM_USE_CODIPACK@
 MFEM_USE_BENCHMARK     = @MFEM_USE_BENCHMARK@
 MFEM_USE_PARELAG       = @MFEM_USE_PARELAG@
-MFEM_USE_ENZYME        = @MFEM_USE_ENZYME@

 # Compiler, compile options, and link options
 MFEM_CXX       = @MFEM_CXX@
@@ -64,7 +64,6 @@ option(MFEM_USE_ADFORWARD "Enable forward mode for AD" OFF)
 option(MFEM_USE_CODIPACK "Enable automatic differentiation (AD) using CoDiPack" OFF)
 option(MFEM_USE_BENCHMARK "Enable Google Benchmark" OFF)
 option(MFEM_USE_PARELAG "Enable ParELAG" OFF)
-option(MFEM_USE_ENZYME "Enable Enzyme" OFF)

 # Optional overrides for autodetected MPIEXEC and MPIEXEC_NUMPROC_FLAG
 # set(MFEM_MPIEXEC "mpirun" CACHE STRING "Command for running MPI tests")
@@ -42,9 +42,6 @@ STATIC = YES
 SHARED = NO

 # CUDA configuration options
-#
-# If you set MFEM_USE_ENZYME=YES, CUDA_CXX has to be configured to use cuda with
-# clang as its host compiler.
 CUDA_CXX = nvcc
 CUDA_ARCH = sm_60
 CUDA_FLAGS = -x=cu --expt-extended-lambda -arch=$(CUDA_ARCH)
@@ -166,7 +163,6 @@ MFEM_USE_ADFORWARD     = NO
 MFEM_USE_CODIPACK      = NO
 MFEM_USE_BENCHMARK     = NO
 MFEM_USE_PARELAG       = NO
-MFEM_USE_ENZYME        = NO

 # MPI library compile and link flags
 # These settings are used only when building MFEM with MPI + HIP
@@ -179,7 +175,7 @@ ifeq ($(MFEM_USE_MPI)$(MFEM_USE_HIP),YESYES)
 endif

 # ROCM/HIP directory such that ROCM/HIP libraries like rocsparse and rocrand are
-# found in $(HIP_DIR)/lib, usually as links. Typically, this directory is of
+# 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 $(HIP_CXX))))
@@ -207,7 +203,7 @@ HYPRE_OPT = -I$(HYPRE_DIR)/include
 HYPRE_LIB = -L$(HYPRE_DIR)/lib -lHYPRE
 ifeq (YES,$(MFEM_USE_CUDA))
   # This is only necessary when hypre is built with cuda:
-   HYPRE_LIB += -lcusparse -lcurand -lcublas
+   HYPRE_LIB += -lcusparse -lcurand
 endif
 ifeq (YES,$(MFEM_USE_HIP))
   # This is only necessary when hypre is built with hip:
@@ -251,16 +247,12 @@ POSIX_CLOCKS_LIB = -lrt
 # SUNDIALS library configuration
 # For sundials_nvecmpiplusx and nvecparallel remember to build with MPI_ENABLE=ON
 # and modify cmake variables for hypre for sundials
-SUNDIALS_DIR = @MFEM_DIR@/../sundials-5.0.0/instdir
-# SUNDIALS >= 6.4.0 requires C++14:
-ifeq ($(MFEM_USE_SUNDIALS),YES)
-   BASE_FLAGS = -std=c++14
-endif
-SUNDIALS_OPT = -I$(SUNDIALS_DIR)/include
-SUNDIALS_LIB = $(XLINKER)-rpath,$(SUNDIALS_DIR)/lib64\
- $(XLINKER)-rpath,$(SUNDIALS_DIR)/lib\
- -L$(SUNDIALS_DIR)/lib64 -L$(SUNDIALS_DIR)/lib\
+SUNDIALS_DIR    = @MFEM_DIR@/../sundials-5.0.0/instdir
+SUNDIALS_OPT    = -I$(SUNDIALS_DIR)/include
+SUNDIALS_LIBDIR = $(wildcard $(SUNDIALS_DIR)/lib*)
+SUNDIALS_LIB    = $(XLINKER)-rpath,$(SUNDIALS_LIBDIR) -L$(SUNDIALS_LIBDIR)\
 -lsundials_arkode -lsundials_cvodes -lsundials_nvecserial -lsundials_kinsol
+
 ifeq ($(MFEM_USE_MPI),YES)
   SUNDIALS_LIB += -lsundials_nvecparallel -lsundials_nvecmpiplusx
 endif
@@ -313,7 +305,7 @@ SCALAPACK_LIB = -L$(SCALAPACK_DIR)/lib -lscalapack $(LAPACK_LIB)
 MPI_FORTRAN_LIB = -lmpifort
 # OpenMPI:
 # MPI_FORTRAN_LIB = -lmpi_mpifh
-# Additional Fortran library:
+# Additional Fortan library:
 # MPI_FORTRAN_LIB += -lgfortran

 # MUMPS library configuration
@@ -528,22 +520,6 @@ PARELAG_DIR = @MFEM_DIR@/../parelag
 PARELAG_OPT = -I$(PARELAG_DIR)/src -I$(PARELAG_DIR)/build/src
 PARELAG_LIB = -L$(PARELAG_DIR)/build/src -lParELAG

-# Enzyme configuration
-
-# If you want to enable automatic differentiation at compile time, use the
-# options below, adapted to your configuration. To be more flexible, we
-# recommend using the Enzyme plugin during link time optimization. One option is
-# to add your options to the global compiler/linker flags like
-#
-# BASE_FLAGS += -flto
-# CXX_XLINKER += -fuse-ld=lld -Wl,--lto-legacy-pass-manager\
-#                -Wl,-mllvm=-load=$(ENZYME_DIR)/LLDEnzyme-$(ENZYME_VERSION).so -Wl,
-#
-ENZYME_DIR ?= @MFEM_DIR@/../enzyme
-ENZYME_VERSION ?= 14
-ENZYME_OPT = -fno-experimental-new-pass-manager -Xclang -load -Xclang $(ENZYME_DIR)/ClangEnzyme-$(ENZYME_VERSION).so
-ENZYME_LIB = ""
-
 # If YES, enable some informational messages
 VERBOSE = NO

@@ -1,27 +1,30 @@
-FROM ghcr.io/mfem/mfem-ubuntu-base:latest as builder
+FROM ghcr.io/rse-ops/cuda-ubuntu-20.04:cuda-11.0.3

-# docker build -t ghcr.io/mfem/mfem-ubuntu .
-
-COPY ./config/docker/spack.yaml /opt/mfem-env/spack.yaml
-RUN apt-get install -y python3 && \
-    cd /opt/mfem-env && \
-    . /opt/spack/share/spack/setup-env.sh && \
-    spack env activate . && \
-    spack env view regenerate
-
-FROM ubuntu:22.04
-
-COPY --from=builder /opt/view /opt/view
-COPY --from=builder /opt/mfem-view /opt/mfem-view
+# docker build -t ghcr.io/mfem/mfem-ubuntu-base .

 RUN apt-get update && \
    apt-get install -y unzip gfortran && \
+    spack compiler find && \
    apt-get install -y libcurl4-openssl-dev libssl-dev

-ENV PATH=$PATH:/opt/mfem-view/bin
-ENV LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/opt/mfem-view/lib:/opt/mfem-view/lib64
-ENV DEBIAN_FRONTEND=noninteractive
+# /code is the working directory for code
+WORKDIR /code
+COPY . /code
+
+# This is for a spack environment/view to install from there
+WORKDIR /opt/mfem-env
+RUN . /opt/spack/share/spack/setup-env.sh && \
+    spack env create -d . && \
+    echo "  concretization: together" >> spack.yaml && \
+    spack env activate . && \
+    spack develop --path /code mfem@master+examples+miniapps && \
+    spack add mfem@master+examples+miniapps && \
+    spack install
+
+# ensure mfem always on various paths
+RUN cd /opt/mfem-env && \
+    spack env activate --sh -d . >> /etc/profile.d/z10_spack_environment.sh

 # The user will see the view on shell into the container
-WORKDIR /opt/mfem-view
-ENTRYPOINT ["/bin/bash"]
+WORKDIR /opt/mfem-env/.spack-env/view/
+ENTRYPOINT ["/bin/bash", "--rcfile", "/etc/profile", "-l", "-c"]
@@ -1,47 +0,0 @@
-FROM ghcr.io/rse-ops/cuda-ubuntu-20.04:cuda-11.0.3
-
-# docker build -f Dockerfile.base -t ghcr.io/mfem/mfem-ubuntu-base .
-
-RUN apt-get update && \
-    apt-get install -y unzip gfortran && \
-    spack compiler find && \
-    apt-get install -y libcurl4-openssl-dev libssl-dev
-
-# /code is the working directory for code
-WORKDIR /code
-COPY . /code
-
-# This is for a spack environment/view to install from there
-RUN mkdir -p /opt/mfem-env \
-&&  (echo "spack:" \
-&&   echo "  view:" \
-&&   echo "    mfem:" \
-&&   echo "      root: /opt/mfem-view" \
-&&   echo "      link_type: copy" \
-&&   echo "  packages:" \
-&&   echo "    all:" \
-&&   echo "      target:" \
-&&   echo "      - x86_64_v3" \
-&&   echo "  config:" \
-&&   echo "    concretizer: clingo" \
-&&   echo "    compiler:" \
-&&   echo "      target:" \
-&&   echo "      - x86_64_v3" \
-&&   echo "    install_missing_compilers: true" \
-&&   echo "  concretization: together") > /opt/mfem-env/spack.yaml
-
-RUN cd /opt/mfem-env && \
-    . /opt/spack/share/spack/setup-env.sh && \
-    spack env activate . && \
-    spack develop --path /code mfem@master+examples+miniapps && \
-    spack add mfem@master+examples+miniapps # && \
-#    spack install
-
-# ensure mfem always on various paths
-#RUN cd /opt/mfem-env && \
-#    spack env activate --sh -d . >> /etc/profile.d/z10_spack_environment.sh
-
-# Present the software install when we shell in
-# The view is at /opt/mfem-env/.spack-env/view
-#WORKDIR /opt/software
-#ENTRYPOINT ["/bin/bash", "--rcfile", "/etc/profile", "-l", "-c"]
@@ -1,8 +1,7 @@
 # mfem Docker

-We provide a [Dockerfile.base](Dockerfile.base) to build an ubuntu base image,
-and a [Dockerfile](Dockerfile) to build a smaller one with a multi-stage build.
-You can use this image for a demo of using mfem! 🎉️
+We provide a [Dockerfile](Dockerfile) to build an ubuntu base image. You can use
+this image for a demo of using mfem! 🎉️

 Updated containers are built and deployed on merges to the main branch and releases.
 If you want to request a build on demand, you can [manually run the workflow](https://docs.github.com/en/actions/managing-workflow-runs/manually-running-a-workflow) thanks to the workflow dispatch event.
@@ -15,33 +14,18 @@ is the [GitHub packages](https://github.com/features/packages) registry that sup
 Docker images and other OCI artifacts. From the root of the repository:

 ```bash
-$ docker build -f config/docker/Dockerfile -t ghcr.io/mfem/mfem-ubuntu .
-$ docker build -f config/docker/Dockerfile.base -t ghcr.io/mfem/mfem-ubuntu-base .
+$ docker build -f config/docker/Dockerfile -t ghcr.io/mfem/mfem-ubuntu-base .
 ```

-### Shell Ubuntu
-
-To shell into the container:
+or this directory:

 ```bash
-$ docker run -it ghcr.io/mfem/mfem-ubuntu
+$ docker build -f Dockerfile -t ghcr.io/mfem/mfem-ubuntu-base ../../
 ```

-This smaller image has a view where everything is installed.
+### Shell

-```bash
-$ ls
-bin  etc  include  lib  libexec  sbin  share  var
-```
-
- - Examples are in share/mfem/examples
- - Examples are in share/mfem/miniapps
-
-You can read more about interaction with these examples and miniapps below.
-
-### Shell Ubuntu Base
-
-To shell into the container:
+To shell into a container (here is an example with ubuntu):

 ```bash
 $ docker run -it ghcr.io/mfem/mfem-ubuntu-base bash
@@ -144,4 +128,3 @@ $ docker run -it ghcr.io/mfem/mfem-ubuntu-base -v $PWD:/src bash
 In the above, we can pretend your project is in the present working directory (PWD) and we are
 binding to source. You can then use the mfem in the container for development, and if you
 want to distribute your library or app in a container, you can use the mfem container as the base.
-
@@ -1,11 +0,0 @@
-spack:
-  specs: [mfem@master+examples+miniapps]
-  view: 
-    mfem:
-      root: /opt/mfem-view 
-      link_type: copy
-  concretization: together
-  develop:
-    mfem:
-      path: /code
-      spec: mfem@master+examples+miniapps
@@ -554,14 +554,15 @@ function go()
   local cmd_line="${1##+( )}"
   cmd_line="${cmd_line%%+( )}"
   shopt -u extglob
+   eval local cmd=(${cmd_line})
   local res=""
   echo $sep
   echo "<${group}>" "${cmd_line}"
   echo $sep
   if [ "${timing}" == "yes" ]; then
-      timed_run eval "${cmd_line}"
+      timed_run "${cmd[@]}"
   else
-      eval "${cmd_line}"
+      "${cmd[@]}"
   fi
   if [ "$?" -eq 0 ]; then
      res="${green}  OK  ${none}"
@@ -1,8 +0,0 @@
--- a/CMakeLists.txt	Wed Dec 21 18:24:22 2016
-+++ b/CMakeLists.txt	Wed Dec 21 18:24:26 2016
-@@ -20,4 +20,4 @@
- # Recursively look for CMakeLists.txt in subdirs.
- add_subdirectory("include")
- add_subdirectory("libmetis")
-add_subdirectory("programs")
-+# add_subdirectory("programs")
@@ -1,15 +0,0 @@
--- a/CMakeLists.txt	Sat Mar 30 17:24:45 2013
-+++ b/CMakeLists.txt	Wed Dec 21 18:23:43 2016
-@@ -4,11 +4,7 @@
- set(GKLIB_PATH "GKlib" CACHE PATH "path to GKlib")
- set(SHARED FALSE CACHE BOOL "build a shared library")
- 
-if(MSVC)
-  set(METIS_INSTALL FALSE)
-else()
-  set(METIS_INSTALL TRUE)
-endif()
-+set(METIS_INSTALL TRUE)
- 
- # Configure libmetis library.
- if(SHARED)
@@ -1,34 +0,0 @@
-diff --git a/include/metis.h b/include/metis.h
-index dc5406a..7732437 100644
--- a/include/metis.h
-+++ b/include/metis.h
-@@ -72,10 +72,14 @@ typedef __int64 int64_t;
- #define PRId64       "I64d"
- #define SCNd32       "ld"
- #define SCNd64       "I64d"
-+#ifdef _WIN32
-+#include <stdint.h>
-+#else
- #define INT32_MIN    ((int32_t)_I32_MIN)
- #define INT32_MAX    _I32_MAX
- #define INT64_MIN    ((int64_t)_I64_MIN)
- #define INT64_MAX    _I64_MAX
-+#endif
- #else
- #include <inttypes.h>
- #endif
-diff --git a/GKlib/gk_arch.h b/GKlib/gk_arch.h
-index 78b1431..7258763 100644
--- a/GKlib/gk_arch.h
-+++ b/GKlib/gk_arch.h
-@@ -32,8 +32,8 @@
- 
- 
- #ifdef __MSC__ 
-  #include "ms_stdint.h"
-  #include "ms_inttypes.h"
-+  #include <stdint.h>
-+  #include <inttypes.h>
-   #include "ms_stat.h"
- #else
- #ifndef SUNOS
@@ -1,11 +0,0 @@
--- a/GKlib/gk_arch.h	Wed Dec 21 18:34:18 2016
-+++ b/GKlib/gk_arch.h	Wed Dec 21 18:30:49 2016
-@@ -58,7 +58,7 @@
- #define PTRDIFF_MAX  INT64_MAX
- #endif
- 
-#ifdef __MSC__
-+#if defined(__MSC__) && (_MSC_VER < 1900)
- /* MSC does not have rint() function */
- #define rint(x) ((int)((x)+0.5))  
- 
@@ -1,14 +0,0 @@
-diff --git a/CMakeLists.txt b/CMakeLists.txt
-index e94f050..b9613a7 100644
--- a/CMakeLists.txt
-+++ b/CMakeLists.txt
-@@ -1,7 +1,8 @@
- cmake_minimum_required(VERSION 2.8)
- project(METIS)
- 
-set(GKLIB_PATH "GKlib" CACHE PATH "path to GKlib")
-+set(GKLIB_PATH "${CMAKE_SOURCE_DIR}/GKlib" CACHE PATH "path to GKlib")
-+
- set(SHARED FALSE CACHE BOOL "build a shared library")
- 
- set(METIS_INSTALL TRUE)
@@ -1,11 +0,0 @@
--- a/libmetis/metislib.h	Sat Mar 30 17:24:45 2013
-+++ b/libmetis/metislib.h	Wed Dec 21 18:30:59 2016
-@@ -31,7 +31,7 @@
- #include <proto.h>
- 
- 
-#if defined(COMPILER_MSC)
-+#if defined(COMPILER_MSC) && (_MSC_VER < 1900)
- #if defined(rint)
-   #undef rint
- #endif
@@ -1,10 +0,0 @@
--- a/libmetis/CMakeLists.txt	Sat Mar 30 17:24:45 2013
-+++ b/libmetis/CMakeLists.txt	Wed Dec 21 17:41:37 2016
-@@ -11,6 +11,6 @@
- if(METIS_INSTALL)
-   install(TARGETS metis
-     LIBRARY DESTINATION lib
-    RUNTIME DESTINATION lib
-+    RUNTIME DESTINATION bin
-     ARCHIVE DESTINATION lib)
- endif()
@@ -1,44 +0,0 @@
-diff --git a/CMakeLists.txt b/CMakeLists.txt
-index b9613a7..e43ffee 100644
--- a/CMakeLists.txt
-+++ b/CMakeLists.txt
-@@ -22,3 +22,23 @@ include_directories(include)
- add_subdirectory("include")
- add_subdirectory("libmetis")
- # add_subdirectory("programs")
-+
-+if(METIS_INSTALL)
-+  set(PRJ_NAME metis)
-+  set(PRJ_VER 5.1.0)
-+  install(EXPORT metisTargets 
-+    FILE ${PRJ_NAME}Targets.cmake
-+    DESTINATION lib/cmake/${PRJ_NAME})
-+  include(CMakePackageConfigHelpers)
-+  write_basic_package_version_file(
-+    ${CMAKE_CURRENT_BINARY_DIR}/${PRJ_NAME}ConfigVersion.cmake
-+    VERSION ${PRJ_VER}
-+    COMPATIBILITY SameMajorVersion)
-+  file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/${PRJ_NAME}Config.cmake
-+    "include(\${CMAKE_CURRENT_LIST_DIR}/${PRJ_NAME}Targets.cmake)")
-+  install(FILES 
-+    ${CMAKE_CURRENT_BINARY_DIR}/${PRJ_NAME}ConfigVersion.cmake
-+    ${CMAKE_CURRENT_BINARY_DIR}/${PRJ_NAME}Config.cmake
-+    DESTINATION lib/cmake/${PRJ_NAME})
-+endif()
-+
-diff --git a/libmetis/CMakeLists.txt b/libmetis/CMakeLists.txt
-index 7a5fc74..5a68cf0 100644
--- a/libmetis/CMakeLists.txt
-+++ b/libmetis/CMakeLists.txt
-@@ -9,8 +9,9 @@ if(UNIX)
- endif()
- 
- if(METIS_INSTALL)
-  install(TARGETS metis
-+  install(TARGETS metis EXPORT metisTargets
-     LIBRARY DESTINATION lib
-     RUNTIME DESTINATION bin
-    ARCHIVE DESTINATION lib)
-+    ARCHIVE DESTINATION lib
-+    INCLUDES DESTINATION include)
- endif()
@@ -1,41 +0,0 @@
-vcpkg_check_linkage(ONLY_STATIC_LIBRARY)
-set(OPTIONS -DSHARED=OFF)
-
-set(METIS_VERSION 5.1.0)
-
-vcpkg_download_distfile(ARCHIVE
-    URLS "https://github.com/mfem/tpls/raw/gh-pages/metis-${METIS_VERSION}.tar.gz"
-    FILENAME "metis-${METIS_VERSION}.tar.gz"
-    SHA512 deea47749d13bd06fbeaf98a53c6c0b61603ddc17a43dae81d72c8015576f6495fd83c11b0ef68d024879ed5415c14ebdbd87ce49c181bdac680573bea8bdb25
-)
-
-vcpkg_extract_source_archive_ex(
-    OUT_SOURCE_PATH SOURCE_PATH
-    ARCHIVE ${ARCHIVE}
-    REF ${METIS_VERSION}
-    PATCHES
-        enable-install.patch
-        disable-programs.patch
-        fix-runtime-install-destination.patch
-        fix-metis-vs14-math.patch
-        fix-gklib-vs14-math.patch
-        fix-linux-build-error.patch
-        install-metisConfig.patch
-        fix-INT_MIN_define.patch
-)
-
-vcpkg_configure_cmake(
-    SOURCE_PATH ${SOURCE_PATH}
-    PREFER_NINJA
-    OPTIONS ${OPTIONS}
-)
-
-vcpkg_install_cmake()
-vcpkg_copy_pdbs()
-vcpkg_fixup_cmake_targets(CONFIG_PATH lib/cmake/metis)
-
-file(REMOVE_RECURSE ${CURRENT_PACKAGES_DIR}/debug/include)
-
-# Handle copyright
-file(COPY ${SOURCE_PATH}/LICENSE.txt DESTINATION ${CURRENT_PACKAGES_DIR}/share/metis)
-file(INSTALL ${SOURCE_PATH}/LICENSE.txt DESTINATION ${CURRENT_PACKAGES_DIR}/share/${PORT} RENAME copyright)
@@ -1,7 +0,0 @@
-{
-  "name": "metis-mfem",
-  "version-string": "5.1.0",
-  "port-version": 0,
-  "description": "Serial Graph Partitioning and Fill-reducing Matrix Ordering",
-  "homepage": "http://glaros.dtc.umn.edu/gkhome/metis/metis/overview"
-}
@@ -1,7 +1,7 @@
 MFEM mesh v1.0

 #
-# MFEM Geometry Types (see mesh/geom.hpp):
+# MFEM Geomety Types (see mesh/geom.hpp):
 #
 # POINT       = 0
 # SEGMENT     = 1
@@ -38,7 +38,7 @@ PROJECT_NAME           = "MFEM"
 # could be handy for archiving the generated documentation or if some version
 # control system is used.

-PROJECT_NUMBER         = v4.5.0
+PROJECT_NUMBER         = v4.4.1

 # Using the PROJECT_BRIEF tag one can provide an optional one line description
 # for a project that appears at the top of each page and should give viewer a
@@ -2349,7 +2349,7 @@ PLANTUML_INCLUDE_PATH  =
 # Minimum value: 0, maximum value: 10000, default value: 50.
 # This tag requires that the tag HAVE_DOT is set to YES.

-DOT_GRAPH_MAX_NODES    = 100
+DOT_GRAPH_MAX_NODES    = 50

 # The MAX_DOT_GRAPH_DEPTH tag can be used to set the maximum depth of the graphs
 # generated by dot. A depth value of 3 means that only nodes reachable from the
@@ -30,7 +30,6 @@
 //
 // Device sample runs:
 //               ex1 -pa -d cuda
-//             * ex1 -fa -d cuda
 //               ex1 -pa -d raja-cuda
 //             * ex1 -pa -d raja-hip
 //               ex1 -pa -d occa-cuda
@@ -38,13 +37,9 @@
 //               ex1 -pa -d occa-omp
 //               ex1 -pa -d ceed-cpu
 //               ex1 -pa -d ceed-cpu -o 4 -a
-//               ex1 -pa -d ceed-cpu -m ../data/square-mixed.mesh
-//               ex1 -pa -d ceed-cpu -m ../data/fichera-mixed.mesh
 //             * ex1 -pa -d ceed-cuda
 //             * ex1 -pa -d ceed-hip
 //               ex1 -pa -d ceed-cuda:/gpu/cuda/shared
-//               ex1 -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/square-mixed.mesh
-//               ex1 -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/fichera-mixed.mesh
 //               ex1 -m ../data/beam-hex.mesh -pa -d cuda
 //               ex1 -m ../data/beam-tet.mesh -pa -d ceed-cpu
 //               ex1 -m ../data/beam-tet.mesh -pa -d ceed-cuda:/gpu/cuda/ref
@@ -78,7 +73,6 @@ int main(int argc, char *argv[])
   int order = 1;
   bool static_cond = false;
   bool pa = false;
-   bool fa = false;
   const char *device_config = "cpu";
   bool visualization = true;
   bool algebraic_ceed = false;
@@ -93,8 +87,6 @@ int main(int argc, char *argv[])
                  "--no-static-condensation", "Enable static condensation.");
   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
                  "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&fa, "-fa", "--full-assembly", "-no-fa",
-                  "--no-full-assembly", "Enable Full Assembly.");
   args.AddOption(&device_config, "-d", "--device",
                  "Device configuration string, see Device::Configure().");
 #ifdef MFEM_USE_CEED
@@ -192,7 +184,6 @@ int main(int argc, char *argv[])
   //    domain integrator.
   BilinearForm a(&fespace);
   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   if (fa) { a.SetAssemblyLevel(AssemblyLevel::FULL); }
   a.AddDomainIntegrator(new DiffusionIntegrator(one));

   // 10. Assemble the bilinear form and the corresponding linear system,
@@ -30,18 +30,13 @@
 //
 // Device sample runs:
 //               mpirun -np 4 ex1p -pa -d cuda
-//             * mpirun -np 4 ex1p -fa -d cuda
 //               mpirun -np 4 ex1p -pa -d occa-cuda
 //               mpirun -np 4 ex1p -pa -d raja-omp
 //               mpirun -np 4 ex1p -pa -d ceed-cpu
 //               mpirun -np 4 ex1p -pa -d ceed-cpu -o 4 -a
-//               mpirun -np 4 ex1p -pa -d ceed-cpu -m ../data/square-mixed.mesh
-//               mpirun -np 4 ex1p -pa -d ceed-cpu -m ../data/fichera-mixed.mesh
 //             * mpirun -np 4 ex1p -pa -d ceed-cuda
 //             * mpirun -np 4 ex1p -pa -d ceed-hip
 //               mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared
-//               mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/square-mixed.mesh
-//               mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/fichera-mixed.mesh
 //               mpirun -np 4 ex1p -m ../data/beam-tet.mesh -pa -d ceed-cpu
 //
 // Description:  This example code demonstrates the use of MFEM to define a
@@ -79,7 +74,6 @@ int main(int argc, char *argv[])
   int order = 1;
   bool static_cond = false;
   bool pa = false;
-   bool fa = false;
   const char *device_config = "cpu";
   bool visualization = true;
   bool algebraic_ceed = false;
@@ -94,8 +88,6 @@ int main(int argc, char *argv[])
                  "--no-static-condensation", "Enable static condensation.");
   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
                  "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&fa, "-fa", "--full-assembly", "-no-fa",
-                  "--no-full-assembly", "Enable Full Assembly.");
   args.AddOption(&device_config, "-d", "--device",
                  "Device configuration string, see Device::Configure().");
 #ifdef MFEM_USE_CEED
@@ -219,7 +211,6 @@ int main(int argc, char *argv[])
   //     Diffusion domain integrator.
   ParBilinearForm a(&fespace);
   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   if (fa) { a.SetAssemblyLevel(AssemblyLevel::FULL); }
   a.AddDomainIntegrator(new DiffusionIntegrator(one));

   // 12. Assemble the parallel bilinear form and the corresponding linear
@@ -182,7 +182,7 @@ int main(int argc, char *argv[])
   }
   for (int level = 0; level < order_refinements; ++level)
   {
-      collections.Append(new H1_FECollection((int)std::pow(2, level+1), dim));
+      collections.Append(new H1_FECollection(std::pow(2, level+1), dim));
      fespaces.AddOrderRefinedLevel(collections.Last());
   }

@@ -219,7 +219,7 @@ int main(int argc, char *argv[])
   }
   for (int level = 0; level < order_refinements; ++level)
   {
-      collections.Append(new H1_FECollection((int)std::pow(2, level+1), dim));
+      collections.Append(new H1_FECollection(std::pow(2, level+1), dim));
      fespaces->AddOrderRefinedLevel(collections.Last());
   }

@@ -195,7 +195,7 @@ int main(int argc, char *argv[])
   Array<int> ess_tdof_list(0);
   if (h1 && pmesh.bdr_attributes.Size())
   {
-      // For a continuous basis the linear system must be modified to enforce an
+      // For a continuous basis the linear system must be modifed to enforce an
      // essential (Dirichlet) boundary condition. In the DG case this is not
      // necessary as the boundary condition will only be enforced weakly.
      fespace.GetEssentialTrueDofs(dbc_bdr, ess_tdof_list);
@@ -197,6 +197,7 @@ int main(int argc, char *argv[])
      SparseMatrix &M(mVarf->SpMat());
      SparseMatrix &B(bVarf->SpMat());
      B *= -1.;
+      B.EnsureMultTranspose();
      Bt = new TransposeOperator(&B);

      darcyOp.SetBlock(0,0, &M);
@@ -187,6 +187,7 @@ int main(int argc, char *argv[])
 {
   // 1. Initialize MPI and HYPRE.
   Mpi::Init(argc, argv);
+   int num_procs = Mpi::WorldSize();
   int myid = Mpi::WorldRank();
   Hypre::Init();

@@ -248,10 +248,7 @@ int main(int argc, char *argv[])
   //     constraints for non-conforming AMR, static condensation, etc.
   if (myid == 0) { cout << "matrix ... " << flush; }
   if (static_cond) { a->EnableStaticCondensation(); }
-   // Here we want to try out block-size aware AMG solver in PETSc.
-   // For that to work properly, we need a fully-compliant block-size
-   // structure and we do not skip zeros when assembling.
-   a->Assemble(use_petsc ? 0 : 1);
+   a->Assemble();

   Vector B, X;
   if (!use_petsc)
@@ -297,14 +294,13 @@ int main(int argc, char *argv[])
         cout << "done." << endl;
         cout << "Size of linear system: " << A.M() << endl;
      }
-      // Tell PETSc the matrix has a block structure
-      A.SetBlockSize(dim);
-
-      // The preconditioner for the PCG solver can be specified in the
-      // PETSc config file
      PetscPCGSolver *pcg = new PetscPCGSolver(A);
+
+      // The preconditioner for the PCG solver defined below is specified in the
+      // PETSc config file, rc_ex2p, since a Krylov solver in PETSc can also
+      // customize its preconditioner.
      PetscPreconditioner *prec = NULL;
-      if (use_nonoverlapping) // Specialized BDDC construction
+      if (use_nonoverlapping)
      {
         // Compute dofs belonging to the natural boundary
         Array<int> nat_tdof_list, nat_bdr(pmesh->bdr_attributes.Max());
@@ -450,7 +450,7 @@ int main(int argc, char *argv[])
      for (int ti = 0; !done; )
      {
         // We cannot match exactly the time history of the Run method
-         // since we are explicitly telling PETSc to use a time step
+         // since we are explictly telling PETSc to use a time step
         double dt_real = min(dt, t_final - t);
         ode_solver->Step(*U, t, dt_real);
         ti++;
@@ -78,7 +78,6 @@ EX1_ARGS_CUDA         := -m ../../data/star.mesh --usepetsc --partial-assembly -
 EX1_ARGS_CUDAAMG      := -m ../../data/star.mesh --usepetsc --device cuda --petscopts rc_ex1p_cudaamg
 EX2_ARGS              := -m ../../data/beam-quad.mesh --usepetsc --petscopts rc_ex2p
 EX2_ARGS_BDDC         := -m ../../data/beam-tri.mesh --usepetsc --nonoverlapping --petscopts rc_ex2p_bddc
-EX2_ARGS_ASM          := -m ../../data/beam-quad.mesh --usepetsc --petscopts rc_ex2p_asm
 EX3_ARGS              := -m ../../data/klein-bottle.mesh -o 2 -f 0.1 --usepetsc --petscopts rc_ex3p_bddc --nonoverlapping
 EX4_ARGS              := -m ../../data/klein-bottle.mesh -o 2 --usepetsc --petscopts rc_ex4p_bddc --nonoverlapping
 EX4_HYB_ARGS          := -m ../../data/klein-bottle.mesh -o 2 --usepetsc --petscopts rc_ex4p_bddc --nonoverlapping --hybridization
@@ -110,7 +109,6 @@ endif
 ex2p-test-par: ex2p
 	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX2_ARGS))
 	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX2_ARGS_BDDC))
-	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX2_ARGS_ASM))
 ex3p-test-par: ex3p
 	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX3_ARGS))
 ex4p-test-par: ex4p
@@ -1,7 +1,8 @@
 -ksp_converged_reason

 # GAMG is still not used at its best,
-# since we are not exploiting the RBMs
+# since we are not exploiting the
+# block size (Ordering::byVDIM) and the RBMs

 -ksp_view
 -pc_type gamg
@@ -1,10 +0,0 @@
-# Additive Schwarz with Overlap
-# This is not a good solver for elasticity
-# These options are here only to describe
-# the setup of the solver
-ksp_converged_reason
-ksp_view
-ksp_max_it 10
-pc_type asm
-pc_asm_overlap 1
-sub_pc_type icc
@@ -210,9 +210,6 @@ void visualize(ostream &os, Mesh *mesh, GridFunction *deformed_nodes,

 int main(int argc, char *argv[])
 {
-   // 0. Initialize SUNDIALS.
-   Sundials::Init();
-
   // 1. Parse command-line options.
   const char *mesh_file = "../../data/beam-quad.mesh";
   int ref_levels = 2;
@@ -215,11 +215,10 @@ void visualize(ostream &os, ParMesh *mesh, ParGridFunction *deformed_nodes,

 int main(int argc, char *argv[])
 {
-   // 1. Initialize MPI, HYPRE, and SUNDIALS.
+   // 1. Initialize MPI and HYPRE.
   Mpi::Init(argc, argv);
   int myid = Mpi::WorldRank();
   Hypre::Init();
-   Sundials::Init();

   // 2. Parse command-line options.
   const char *mesh_file = "../../data/beam-quad.mesh";
@@ -109,9 +109,6 @@ double InitialTemperature(const Vector &x);

 int main(int argc, char *argv[])
 {
-   // 0. Initialize SUNDIALS.
-   Sundials::Init();
-
   // 1. Parse command-line options.
   const char *mesh_file = "../../data/star.mesh";
   int ref_levels = 2;
@@ -293,10 +290,7 @@ int main(int argc, char *argv[])
         arkode->Init(oper);
         arkode->SetSStolerances(reltol, abstol);
         arkode->SetMaxStep(dt);
-         if (ode_solver_type == 11)
-         {
-            arkode->SetERKTableNum(ARKODE_FEHLBERG_13_7_8);
-         }
+         if (ode_solver_type == 11) { arkode->SetERKTableNum(FEHLBERG_13_7_8); }
         ode_solver = arkode; break;
      case 12:
         arkode = new ARKStepSolver(ARKStepSolver::IMPLICIT);
@@ -101,12 +101,11 @@ double InitialTemperature(const Vector &x);

 int main(int argc, char *argv[])
 {
-   // 1. Initialize MPI, HYPRE, and SUNDIALS.
+   // 1. Initialize MPI and HYPRE.
   Mpi::Init(argc, argv);
   int num_procs = Mpi::WorldSize();
   int myid = Mpi::WorldRank();
   Hypre::Init();
-   Sundials::Init();

   // 2. Parse command-line options.
   const char *mesh_file = "../../data/star.mesh";
@@ -328,10 +327,7 @@ int main(int argc, char *argv[])
         arkode->Init(oper);
         arkode->SetSStolerances(reltol, abstol);
         arkode->SetMaxStep(dt);
-         if (ode_solver_type == 11)
-         {
-            arkode->SetERKTableNum(ARKODE_FEHLBERG_13_7_8);
-         }
+         if (ode_solver_type == 11) { arkode->SetERKTableNum(FEHLBERG_13_7_8); }
         ode_solver = arkode; break;
      case 12:
         arkode = new ARKStepSolver(MPI_COMM_WORLD, ARKStepSolver::IMPLICIT);
@@ -140,9 +140,6 @@ public:

 int main(int argc, char *argv[])
 {
-   // 0. Initialize SUNDIALS.
-   Sundials::Init();
-
   // 1. Parse command-line options.
   problem = 0;
   const char *mesh_file = "../../data/periodic-hexagon.mesh";
@@ -411,7 +408,7 @@ int main(int argc, char *argv[])
         arkode->Init(adv);
         arkode->SetSStolerances(reltol, abstol);
         arkode->SetMaxStep(dt);
-         arkode->SetERKTableNum(ARKODE_FEHLBERG_13_7_8);
+         arkode->SetERKTableNum(FEHLBERG_13_7_8);
         ode_solver = arkode; break;
   }

@@ -152,12 +152,11 @@ public:

 int main(int argc, char *argv[])
 {
-   // 1. Initialize MPI, HYPRE, and SUNDIALS.
+   // 1. Initialize MPI and HYPRE.
   Mpi::Init(argc, argv);
   int num_procs = Mpi::WorldSize();
   int myid = Mpi::WorldRank();
   Hypre::Init();
-   Sundials::Init();

   // 2. Parse command-line options.
   problem = 0;
@@ -488,10 +487,7 @@ int main(int argc, char *argv[])
         arkode->Init(adv);
         arkode->SetSStolerances(reltol, abstol);
         arkode->SetMaxStep(dt);
-         if (ode_solver_type == 9)
-         {
-            arkode->SetERKTableNum(ARKODE_FEHLBERG_13_7_8);
-         }
+         if (ode_solver_type == 9) { arkode->SetERKTableNum(FEHLBERG_13_7_8); }
         ode_solver = arkode; break;
   }

@@ -35,7 +35,7 @@ add_mfem_examples(SUPERLU_EXAMPLES_SRCS ${PFX} "" test_superlu)
 if (MFEM_ENABLE_TESTING)
  # Command line options for the tests.
  # Example 1: Test SuperLU on the simple Poisson problem
-  set(EX1_COMMON_OPTS -m ../../data/star.mesh)
+  set(EX1_COMMON_OPTS -m ../../data/star.mesh -p 2)
  set(EX1P_TEST_OPTS ${EX1_COMMON_OPTS})

  # Add the tests: one test per source file.
@@ -39,7 +39,6 @@ set(SRCS
  complex_fem.cpp
  convergence.cpp
  datacollection.cpp
-  dgmassinv.cpp
  doftrans.cpp
  eltrans.cpp
  estimators.cpp
@@ -73,7 +72,6 @@ set(SRCS
  linearform.cpp
  linearform_ext.cpp
  lininteg.cpp
-  lininteg_boundary.cpp
  lininteg_domain.cpp
  lininteg_domain_grad.cpp
  lor/lor.cpp
@@ -90,7 +88,6 @@ set(SRCS
  fespacehierarchy.cpp
  nonlininteg_vectorconvection.cpp
  nonlininteg_vectorconvection_mf.cpp
-  qfunction.cpp
  qinterp/det.cpp
  qinterp/eval_by_nodes.cpp
  qinterp/eval_by_vdim.cpp
@@ -98,7 +95,6 @@ set(SRCS
  qinterp/grad_by_vdim.cpp
  qinterp/grad_phys_by_nodes.cpp
  qinterp/grad_phys_by_vdim.cpp
-  qspace.cpp
  quadinterpolator.cpp
  quadinterpolator_face.cpp
  restriction.cpp
@@ -140,13 +136,10 @@ set(HDRS
  bilinearform.hpp
  bilinearform_ext.hpp
  bilininteg.hpp
-  bilininteg_mass_pa.hpp
  coefficient.hpp
  complex_fem.hpp
  convergence.hpp
  datacollection.hpp
-  dgmassinv.hpp
-  dgmassinv_kernels.hpp
  doftrans.hpp
  eltrans.hpp
  estimators.hpp
@@ -196,11 +189,9 @@ set(HDRS
  nonlinearform.hpp
  nonlinearform_ext.hpp
  nonlininteg.hpp
-  qfunction.hpp
  qinterp/dispatch.hpp
  qinterp/eval.hpp
  qinterp/grad.hpp
-  qspace.hpp
  quadinterpolator.hpp
  quadinterpolator_face.hpp
  restriction.hpp
@@ -100,6 +100,7 @@ BilinearForm::BilinearForm (FiniteElementSpace * f, BilinearForm * bf, int ps)

   // Copy the pointers to the integrators
   domain_integs = bf->domain_integs;
+   domain_integs_marker = bf->domain_integs_marker;

   boundary_integs = bf->boundary_integs;
   boundary_integs_marker = bf->boundary_integs_marker;
@@ -112,6 +113,138 @@ BilinearForm::BilinearForm (FiniteElementSpace * f, BilinearForm * bf, int ps)
   AllocMat();
 }

+BilinearForm::BilinearForm(BilinearForm &&other)
+   : Matrix(other.fes->GetVSize()), mat(other.mat), mat_e(other.mat_e),
+     fes(other.fes), assembly(other.assembly), batch(other.batch),
+     ext(other.ext), sequence(other.sequence), extern_bfs(other.extern_bfs),
+     element_matrices(other.element_matrices), static_cond(other.static_cond),
+     hybridization(other.hybridization), diag_policy(other.diag_policy),
+     precompute_sparsity(other.precompute_sparsity)
+{
+   // We swap stored integrators and markers with the moved nonlinear form
+   mfem::Swap(domain_integs, other.domain_integs);
+   mfem::Swap(domain_integs_marker, other.domain_integs_marker);
+   mfem::Swap(boundary_integs, other.boundary_integs);
+   mfem::Swap(boundary_integs_marker, other.boundary_integs_marker);
+   mfem::Swap(interior_face_integs, other.interior_face_integs);
+   mfem::Swap(boundary_face_integs, other.boundary_face_integs);
+   mfem::Swap(boundary_face_integs_marker, other.boundary_face_integs_marker);
+
+   /// Leave the moved nonlinear form in a state as if it was just constructed
+   /// with fes
+   other.sequence = fes->GetSequence();
+   other.mat = nullptr;
+   other.mat_e = nullptr;
+   other.extern_bfs = 0;
+   other.element_matrices = nullptr;
+   other.static_cond = nullptr;
+   other.hybridization = nullptr;
+   other.precompute_sparsity = 0;
+   other.diag_policy = DIAG_KEEP;
+
+   other.assembly = AssemblyLevel::LEGACY;
+   other.batch = 1;
+   other.ext = nullptr;
+}
+
+BilinearForm& BilinearForm::operator=(BilinearForm &&other)
+{
+   if (this != &other)
+   {
+      /// Cleanup current bilinear form first
+      delete mat_e;
+      delete mat;
+      delete element_matrices;
+      delete static_cond;
+      delete hybridization;
+      for (int i = 0; i < domain_integs.Size(); i++) { delete domain_integs[i]; }
+      for (int i = 0; i < boundary_integs.Size(); i++) { delete boundary_integs[i]; }
+      for (int i = 0; i < interior_face_integs.Size(); i++) { delete interior_face_integs[i]; }
+      for (int i = 0; i < boundary_face_integs.Size(); i++) { delete boundary_face_integs[i]; }
+      delete ext;
+
+      /// Null out all our integs and set size of their arrays to zero
+      for (int k = 0; k < domain_integs.Size(); k++)
+      {
+         domain_integs[k] = nullptr;
+      }
+      domain_integs.SetSize(0);
+      for (int k = 0; k < boundary_integs.Size(); k++)
+      {
+         boundary_integs[k] = nullptr;
+      }
+      boundary_integs.SetSize(0);
+      for (int k = 0; k < interior_face_integs.Size(); k++)
+      {
+         interior_face_integs[k] = nullptr;
+      }
+      interior_face_integs.SetSize(0);
+      for (int k = 0; k < boundary_face_integs.Size(); ++k)
+      {
+         boundary_face_integs[k] = nullptr;
+      }
+      boundary_face_integs.SetSize(0);
+
+      /// Null out all our markers and set size of their arrays to zero
+      for (int k = 0; k < domain_integs_marker.Size(); ++k)
+      {
+         domain_integs_marker[k] = nullptr;
+      }
+      domain_integs_marker.SetSize(0);
+      for (int k = 0; k < boundary_integs_marker.Size(); ++k)
+      {
+         boundary_integs_marker[k] = nullptr;
+      }
+      boundary_integs_marker.SetSize(0);
+      for (int k = 0; k < boundary_face_integs_marker.Size(); ++k)
+      {
+         boundary_face_integs_marker[k] = nullptr;
+      }
+      boundary_face_integs_marker.SetSize(0);
+
+      /// Now steal data from other bilinear form leaving it in a state as if
+      /// it was just constructed with fes
+      Matrix::operator=(std::move(other));
+
+      mat = other.mat;
+      other.mat = nullptr;
+      mat_e = other.mat_e;
+      other.mat_e = nullptr;
+      fes = other.fes;
+      assembly = other.assembly;
+      other.assembly = AssemblyLevel::LEGACY;
+      batch = other.batch;
+      other.batch = 1;
+      ext = other.ext;
+      other.ext = nullptr;
+      sequence = other.sequence;
+      other.sequence = fes->GetSequence();
+      extern_bfs = other.extern_bfs;
+      other.extern_bfs = 0;
+
+      // Swap our empty integ and marker arrays with the moved bilinear form
+      mfem::Swap(domain_integs, other.domain_integs);
+      mfem::Swap(domain_integs_marker, other.domain_integs_marker);
+      mfem::Swap(boundary_integs, other.boundary_integs);
+      mfem::Swap(boundary_integs_marker, other.boundary_integs_marker);
+      mfem::Swap(interior_face_integs, other.interior_face_integs);
+      mfem::Swap(boundary_face_integs, other.boundary_face_integs);
+      mfem::Swap(boundary_face_integs_marker, other.boundary_face_integs_marker);
+
+      element_matrices = other.element_matrices;
+      other.element_matrices = nullptr;
+      static_cond = other.static_cond;
+      other.static_cond = nullptr;
+      hybridization = other.hybridization;
+      other.hybridization = nullptr;
+      diag_policy = other.diag_policy;
+      other.diag_policy = DIAG_KEEP;
+      precompute_sparsity = other.precompute_sparsity;
+      other.precompute_sparsity = 0;
+   }
+   return *this;
+}
+
 void BilinearForm::SetAssemblyLevel(AssemblyLevel assembly_level)
 {
   if (ext)
@@ -136,7 +269,7 @@ void BilinearForm::SetAssemblyLevel(AssemblyLevel assembly_level)
         ext = new MFBilinearFormExtension(this);
         break;
      default:
-         MFEM_ABORT("BilinearForm: unknown assembly level");
+         mfem_error("Unknown assembly level");
   }
 }

@@ -992,7 +1125,6 @@ void BilinearForm::EliminateVDofs(const Array<int> &vdofs_,
      mat_e = new SparseMatrix(height);
   }

-   vdofs_.HostRead();
   for (int i = 0; i < vdofs_.Size(); i++)
   {
      int vdof = vdofs_[i];
@@ -26,8 +26,7 @@ namespace mfem
 {

 /** @brief Enumeration defining the assembly level for bilinear and nonlinear
-    form classes derived from Operator. For more details, see
-    https://mfem.org/howto/assembly_levels */
+    form classes derived from Operator. */
 enum class AssemblyLevel
 {
   /// In the case of a BilinearForm LEGACY corresponds to a fully assembled
@@ -91,23 +90,23 @@ protected:
   int extern_bfs;

   /// Set of Domain Integrators to be applied.
-   Array<BilinearFormIntegrator*> domain_integs;
+   Array<BilinearFormIntegrator*> domain_integs; // owned
   /// Element attribute marker (should be of length mesh->attributes.Max() or
   /// 0 if mesh->attributes is empty)
   /// Includes all by default.
   /// 0 - ignore attribute
   /// 1 - include attribute
-   Array<Array<int>*>             domain_integs_marker;
+   Array<Array<int>*>             domain_integs_marker; // not owned

   /// Set of Boundary Integrators to be applied.
-   Array<BilinearFormIntegrator*> boundary_integs;
+   Array<BilinearFormIntegrator*> boundary_integs; // owned
   Array<Array<int>*> boundary_integs_marker; ///< Entries are not owned.

   /// Set of interior face Integrators to be applied.
-   Array<BilinearFormIntegrator*> interior_face_integs;
+   Array<BilinearFormIntegrator*> interior_face_integs; // owned

   /// Set of boundary face Integrators to be applied.
-   Array<BilinearFormIntegrator*> boundary_face_integs;
+   Array<BilinearFormIntegrator*> boundary_face_integs; // owned
   Array<Array<int>*> boundary_face_integs_marker; ///< Entries are not owned.

   DenseMatrix elemmat;
@@ -142,13 +141,6 @@ protected:
      ext = NULL;
   }

-private:
-   /// Copy construction is not supported; body is undefined.
-   BilinearForm(const BilinearForm &);
-
-   /// Copy assignment is not supported; body is undefined.
-   BilinearForm &operator=(const BilinearForm &);
-
 public:
   /// Creates bilinear form associated with FE space @a *f.
   /** The pointer @a f is not owned by the newly constructed object. */
@@ -166,6 +158,20 @@ public:
       #precompute_sparsity, see UsePrecomputedSparsity() for details. */
   BilinearForm(FiniteElementSpace *f, BilinearForm *bf, int ps = 0);

+   /// Explicitly prohibit copy construction/assignment of BilinearForms
+   BilinearForm(const BilinearForm &) = delete;
+   BilinearForm &operator=(const BilinearForm &) = delete;
+
+   /// Move constructor for BilinearForm.
+   /** This constructor "steals" the owned data members from the @a other
+       BilinearForm. */
+   BilinearForm(BilinearForm &&other);
+
+   /// Move assignment operator for BilinearForm
+   /** This assignment first frees all owned data, then "steals" the owned data
+       members from the @a other BilinearForm. */
+   BilinearForm& operator=(BilinearForm &&other);
+
   /// Get the size of the BilinearForm as a square matrix.
   int Size() const { return height; }

@@ -178,7 +184,7 @@ public:
       - AssemblyLevel::ELEMENT
       - AssemblyLevel::NONE

-       If used, this method must be called before assembly. */
+       This method must be called before assembly. */
   void SetAssemblyLevel(AssemblyLevel assembly_level);

   /// Returns the assembly level
@@ -334,7 +340,7 @@ public:


   /**  @brief Nullifies the internal matrix \f$ M \f$ and returns a pointer
-        to it.  Used for transferring ownership. */
+        to it.  Used for transfering ownership. */
   SparseMatrix *LoseMat() { SparseMatrix *tmp = mat; mat = NULL; return tmp; }

   /** @brief Returns a const reference to the sparse matrix of eliminated b.c.:
@@ -439,14 +445,6 @@ public:
   virtual const Operator *GetOutputRestriction() const
   { return GetRestriction(); }

-   /// @brief Compute serial RAP operator and store it in @a A as a SparseMatrix.
-   void SerialRAP(OperatorHandle &A)
-   {
-      MFEM_ASSERT(mat, "SerialRAP requires the SparseMatrix to be assembled.");
-      ConformingAssemble();
-      A.Reset(mat, false);
-   }
-
   /** @brief Form the linear system A X = B, corresponding to this bilinear
       form and the linear form @a b(.). */
   /** This method applies any necessary transformations to the linear system
@@ -775,7 +773,7 @@ public:
   SparseMatrix &SpMat() { return *mat; }

   /**  @brief Nullifies the internal matrix \f$ M \f$ and returns a pointer
-        to it.  Used for transferring ownership. */
+        to it.  Used for transfering ownership. */
   SparseMatrix *LoseMat() { SparseMatrix *tmp = mat; mat = NULL; return tmp; }

   /// Adds a domain integrator. Assumes ownership of @a bfi.
@@ -18,8 +18,6 @@
 #include "pgridfunc.hpp"
 #include "ceed/interface/util.hpp"

-#include "../general/nvtx.hpp"
-
 namespace mfem
 {

@@ -162,7 +160,7 @@ void MFBilinearFormExtension::Mult(const Vector &x, Vector &y) const
         {
            intFaceIntegrators[i]->AddMultMF(int_face_X, int_face_Y);
         }
-         int_face_restrict_lex->AddMultTransposeInPlace(int_face_Y, y);
+         int_face_restrict_lex->AddMultTranspose(int_face_Y, y);
      }
   }

@@ -178,7 +176,7 @@ void MFBilinearFormExtension::Mult(const Vector &x, Vector &y) const
         {
            bdrFaceIntegrators[i]->AddMultMF(bdr_face_X, bdr_face_Y);
         }
-         bdr_face_restrict_lex->AddMultTransposeInPlace(bdr_face_Y, y);
+         bdr_face_restrict_lex->AddMultTranspose(bdr_face_Y, y);
      }
   }
 }
@@ -219,7 +217,7 @@ void MFBilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
         {
            intFaceIntegrators[i]->AddMultTransposeMF(int_face_X, int_face_Y);
         }
-         int_face_restrict_lex->AddMultTransposeInPlace(int_face_Y, y);
+         int_face_restrict_lex->AddMultTranspose(int_face_Y, y);
      }
   }

@@ -235,7 +233,7 @@ void MFBilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
         {
            bdrFaceIntegrators[i]->AddMultTransposeMF(bdr_face_X, bdr_face_Y);
         }
-         bdr_face_restrict_lex->AddMultTransposeInPlace(bdr_face_Y, y);
+         bdr_face_restrict_lex->AddMultTranspose(bdr_face_Y, y);
      }
   }
 }
@@ -253,7 +251,6 @@ PABilinearFormExtension::PABilinearFormExtension(BilinearForm *form)

 void PABilinearFormExtension::SetupRestrictionOperators(const L2FaceValues m)
 {
-   if ( Device::Allows(Backend::CEED_MASK) ) { return; }
   ElementDofOrdering ordering = UsesTensorBasis(*a->FESpace())?
                                 ElementDofOrdering::LEXICOGRAPHIC:
                                 ElementDofOrdering::NATIVE;
@@ -291,10 +288,6 @@ void PABilinearFormExtension::SetupRestrictionOperators(const L2FaceValues m)

 void PABilinearFormExtension::Assemble()
 {
-#undef MFEM_NVTX_COLOR
-#define MFEM_NVTX_COLOR NavyBlue
-   NVTX("HO Assemble");
-
   SetupRestrictionOperators(L2FaceValues::DoubleValued);

   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
@@ -389,10 +382,6 @@ void PABilinearFormExtension::FormLinearSystem(const Array<int> &ess_tdof_list,

 void PABilinearFormExtension::Mult(const Vector &x, Vector &y) const
 {
-#undef MFEM_NVTX_COLOR
-#define MFEM_NVTX_COLOR MediumSpringGreen
-   NVTX("HO Apply");
-
   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();

   const int iSz = integrators.Size();
@@ -428,7 +417,7 @@ void PABilinearFormExtension::Mult(const Vector &x, Vector &y) const
         {
            intFaceIntegrators[i]->AddMultPA(int_face_X, int_face_Y);
         }
-         int_face_restrict_lex->AddMultTransposeInPlace(int_face_Y, y);
+         int_face_restrict_lex->AddMultTranspose(int_face_Y, y);
      }
   }

@@ -444,7 +433,7 @@ void PABilinearFormExtension::Mult(const Vector &x, Vector &y) const
         {
            bdrFaceIntegrators[i]->AddMultPA(bdr_face_X, bdr_face_Y);
         }
-         bdr_face_restrict_lex->AddMultTransposeInPlace(bdr_face_Y, y);
+         bdr_face_restrict_lex->AddMultTranspose(bdr_face_Y, y);
      }
   }
 }
@@ -485,7 +474,7 @@ void PABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
         {
            intFaceIntegrators[i]->AddMultTransposePA(int_face_X, int_face_Y);
         }
-         int_face_restrict_lex->AddMultTransposeInPlace(int_face_Y, y);
+         int_face_restrict_lex->AddMultTranspose(int_face_Y, y);
      }
   }

@@ -501,7 +490,7 @@ void PABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
         {
            bdrFaceIntegrators[i]->AddMultTransposePA(bdr_face_X, bdr_face_Y);
         }
-         bdr_face_restrict_lex->AddMultTransposeInPlace(bdr_face_Y, y);
+         bdr_face_restrict_lex->AddMultTranspose(bdr_face_Y, y);
      }
   }
 }
@@ -678,7 +667,7 @@ void EABilinearFormExtension::Mult(const Vector &x, Vector &y) const
            Y(j, 0, f) += res;
         });
         // Apply the Interior Face Restriction transposed
-         int_face_restrict_lex->AddMultTransposeInPlace(int_face_Y, y);
+         int_face_restrict_lex->AddMultTranspose(int_face_Y, y);
      }
   }

@@ -709,7 +698,7 @@ void EABilinearFormExtension::Mult(const Vector &x, Vector &y) const
            Y(j, f) += res;
         });
         // Apply the Boundary Face Restriction transposed
-         bdr_face_restrict_lex->AddMultTransposeInPlace(bdr_face_Y, y);
+         bdr_face_restrict_lex->AddMultTranspose(bdr_face_Y, y);
      }
   }
 }
@@ -806,7 +795,7 @@ void EABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
            Y(j, 0, f) += res;
         });
         // Apply the Interior Face Restriction transposed
-         int_face_restrict_lex->AddMultTransposeInPlace(int_face_Y, y);
+         int_face_restrict_lex->AddMultTranspose(int_face_Y, y);
      }
   }

@@ -837,7 +826,7 @@ void EABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
            Y(j, f) += res;
         });
         // Apply the Boundary Face Restriction transposed
-         bdr_face_restrict_lex->AddMultTransposeInPlace(bdr_face_Y, y);
+         bdr_face_restrict_lex->AddMultTranspose(bdr_face_Y, y);
      }
   }
 }
@@ -967,60 +956,6 @@ void FABilinearFormExtension::Assemble()
   }
 }

-
-void FABilinearFormExtension::RAP(OperatorHandle &A)
-{
-#ifdef MFEM_USE_MPI
-   if ( auto pa = dynamic_cast<ParBilinearForm*>(a) )
-   {
-      pa->ParallelRAP(*pa->mat, A);
-   }
-   else
-#endif
-   {
-      a->SerialRAP(A);
-   }
-}
-
-void FABilinearFormExtension::EliminateBC(const Array<int> &ess_dofs,
-                                          OperatorHandle &A)
-{
-   MFEM_VERIFY(a->diag_policy == DiagonalPolicy::DIAG_ONE,
-               "Only DiagonalPolicy::DIAG_ONE supported with"
-               " FABilinearFormExtension.");
-#ifdef MFEM_USE_MPI
-   if ( dynamic_cast<ParBilinearForm*>(a) )
-   {
-      A.As<HypreParMatrix>()->EliminateBC(ess_dofs,
-                                          DiagonalPolicy::DIAG_ONE);
-   }
-   else
-#endif
-   {
-      A.As<SparseMatrix>()->EliminateBC(ess_dofs,
-                                        DiagonalPolicy::DIAG_ONE);
-   }
-}
-
-void FABilinearFormExtension::FormSystemMatrix(const Array<int> &ess_dofs,
-                                               OperatorHandle &A)
-{
-   RAP(A);
-   EliminateBC(ess_dofs, A);
-}
-
-void FABilinearFormExtension::FormLinearSystem(const Array<int> &ess_tdof_list,
-                                               Vector &x, Vector &b,
-                                               OperatorHandle &A,
-                                               Vector &X, Vector &B,
-                                               int copy_interior)
-{
-   Operator *A_out;
-   Operator::FormLinearSystem(ess_tdof_list, x, b, A_out, X, B, copy_interior);
-   delete A_out;
-   FormSystemMatrix(ess_tdof_list, A);
-}
-
 void FABilinearFormExtension::DGMult(const Vector &x, Vector &y) const
 {
 #ifdef MFEM_USE_MPI
@@ -125,15 +125,6 @@ public:
   FABilinearFormExtension(BilinearForm *form);

   void Assemble();
-   void RAP(OperatorHandle &A);
-   /** @note Always does `DIAG_ONE` policy to be consistent with
-       `Operator::FormConstrainedSystemOperator`. */
-   void EliminateBC(const Array<int> &ess_dofs, OperatorHandle &A);
-   void FormSystemMatrix(const Array<int> &ess_tdof_list, OperatorHandle &A);
-   void FormLinearSystem(const Array<int> &ess_tdof_list,
-                         Vector &x, Vector &b,
-                         OperatorHandle &A, Vector &X, Vector &B,
-                         int copy_interior = 0);
   void Mult(const Vector &x, Vector &y) const;
   void MultTranspose(const Vector &x, Vector &y) const;

@@ -2003,83 +2003,6 @@ void CurlCurlIntegrator::AssembleElementMatrix
   }
 }

-void CurlCurlIntegrator::AssembleElementMatrix2(const FiniteElement &trial_fe,
-                                                const FiniteElement &test_fe,
-                                                ElementTransformation &Trans,
-                                                DenseMatrix &elmat)
-{
-   int tr_nd = trial_fe.GetDof();
-   int te_nd = test_fe.GetDof();
-   dim = trial_fe.GetDim();
-   int dimc = trial_fe.GetCurlDim();
-   double w;
-
-#ifdef MFEM_THREAD_SAFE
-   Vector D;
-   DenseMatrix curlshape(tr_nd,dimc), curlshape_dFt(tr_nd,dimc), M;
-   DenseMatrix te_curlshape(te_nd,dimc), te_curlshape_dFt(te_nd,dimc);
-#else
-   curlshape.SetSize(tr_nd,dimc);
-   curlshape_dFt.SetSize(tr_nd,dimc);
-   te_curlshape.SetSize(te_nd,dimc);
-   te_curlshape_dFt.SetSize(te_nd,dimc);
-#endif
-   elmat.SetSize(te_nd, tr_nd);
-
-   if (MQ) { M.SetSize(dimc); }
-   if (DQ) { D.SetSize(dimc); }
-
-   const IntegrationRule *ir = IntRule;
-   if (ir == NULL)
-   {
-      int order;
-      if (trial_fe.Space() == FunctionSpace::Pk)
-      {
-         order = test_fe.GetOrder() + trial_fe.GetOrder() - 2;
-      }
-      else
-      {
-         order = test_fe.GetOrder() + trial_fe.GetOrder() + trial_fe.GetDim() - 1;
-      }
-      ir = &IntRules.Get(trial_fe.GetGeomType(), order);
-   }
-
-   elmat = 0.0;
-   for (int i = 0; i < ir->GetNPoints(); i++)
-   {
-      const IntegrationPoint &ip = ir->IntPoint(i);
-
-      Trans.SetIntPoint(&ip);
-
-      w = ip.weight * Trans.Weight();
-      trial_fe.CalcPhysCurlShape(Trans, curlshape_dFt);
-      test_fe.CalcPhysCurlShape(Trans, te_curlshape_dFt);
-
-      if (MQ)
-      {
-         MQ->Eval(M, Trans, ip);
-         M *= w;
-         Mult(te_curlshape_dFt, M, te_curlshape);
-         AddMultABt(te_curlshape, curlshape_dFt, elmat);
-      }
-      else if (DQ)
-      {
-         DQ->Eval(D, Trans, ip);
-         D *= w;
-         AddMultADBt(te_curlshape_dFt,D,curlshape_dFt,elmat);
-      }
-      else
-      {
-         if (Q)
-         {
-            w *= Q->Eval(Trans, ip);
-         }
-         curlshape_dFt *= w;
-         AddMultABt(te_curlshape_dFt, curlshape_dFt, elmat);
-      }
-   }
-}
-
 void CurlCurlIntegrator
 ::ComputeElementFlux(const FiniteElement &el, ElementTransformation &Trans,
                     Vector &u, const FiniteElement &fluxelem, Vector &flux,
@@ -2317,84 +2240,6 @@ double VectorCurlCurlIntegrator::GetElementEnergy(
   return 0.5 * energy;
 }

-void MixedCurlIntegrator::AssembleElementMatrix2(
-   const FiniteElement &trial_fe, const FiniteElement &test_fe,
-   ElementTransformation &Trans, DenseMatrix &elmat)
-{
-   int dim = trial_fe.GetDim();
-   int trial_dof = trial_fe.GetDof();
-   int test_dof = test_fe.GetDof();
-   int dimc = (dim == 3) ? 3 : 1;
-
-   MFEM_VERIFY(trial_fe.GetMapType() == mfem::FiniteElement::H_CURL ||
-               (dim == 2 && trial_fe.GetMapType() == mfem::FiniteElement::VALUE),
-               "Trial finite element must be either 2D/3D H(Curl) or 2D H1");
-   MFEM_VERIFY(test_fe.GetMapType() == mfem::FiniteElement::VALUE ||
-               test_fe.GetMapType() == mfem::FiniteElement::INTEGRAL,
-               "Test finite element must be in H1/L2");
-
-   bool spaceH1 = (trial_fe.GetMapType() == mfem::FiniteElement::VALUE);
-
-   if (spaceH1)
-   {
-      dshape.SetSize(trial_dof,dim);
-      curlshape.SetSize(dim*trial_dof,1);
-      dimc = dim;
-   }
-   else
-   {
-      curlshape.SetSize(trial_dof,dimc);
-      elmat_comp.SetSize(test_dof, trial_dof);
-   }
-   elmat.SetSize(dimc * test_dof, trial_dof);
-   shape.SetSize(test_dof);
-   elmat = 0.0;
-
-   double c;
-   Vector d_col;
-   const IntegrationRule *ir = IntRule;
-
-   if (ir == NULL)
-   {
-      int order = trial_fe.GetOrder() + test_fe.GetOrder() + Trans.OrderJ();
-      ir = &IntRules.Get(trial_fe.GetGeomType(), order);
-   }
-
-   for (int i = 0; i < ir->GetNPoints(); i++)
-   {
-      const IntegrationPoint &ip = ir->IntPoint(i);
-      Trans.SetIntPoint(&ip);
-      if (spaceH1)
-      {
-         trial_fe.CalcPhysDShape(Trans, dshape);
-         dshape.GradToCurl(curlshape);
-      }
-      else
-      {
-         trial_fe.CalcPhysCurlShape(Trans, curlshape);
-      }
-      test_fe.CalcPhysShape(Trans, shape);
-      c = ip.weight*Trans.Weight();
-      if (Q)
-      {
-         c *= Q->Eval(Trans, ip);
-      }
-      shape *= c;
-
-      for (int d = 0; d < dimc; ++d)
-      {
-         double * curldata = &(curlshape.GetData())[d*trial_dof];
-         for (int jj = 0; jj < trial_dof; ++jj)
-         {
-            for (int ii = 0; ii < test_dof; ++ii)
-            {
-               elmat(d * test_dof + ii, jj) += shape(ii) * curldata[jj];
-            }
-         }
-      }
-   }
-}
-

 void VectorFEMassIntegrator::AssembleElementMatrix(
   const FiniteElement &el,
@@ -2741,54 +2586,6 @@ void DivDivIntegrator::AssembleElementMatrix(
   }
 }

-void DivDivIntegrator::AssembleElementMatrix2(
-   const FiniteElement &trial_fe,
-   const FiniteElement &test_fe,
-   ElementTransformation &Trans,
-   DenseMatrix &elmat)
-{
-   int tr_nd = trial_fe.GetDof();
-   int te_nd = test_fe.GetDof();
-   double c;
-
-#ifdef MFEM_THREAD_SAFE
-   Vector divshape(tr_nd);
-   Vector te_divshape(te_nd);
-#else
-   divshape.SetSize(tr_nd);
-   te_divshape.SetSize(te_nd);
-#endif
-   elmat.SetSize(te_nd,tr_nd);
-
-   const IntegrationRule *ir = IntRule;
-   if (ir == NULL)
-   {
-      int order = 2 * max(test_fe.GetOrder(),
-                          trial_fe.GetOrder()) - 2; // <--- OK for RTk
-      ir = &IntRules.Get(test_fe.GetGeomType(), order);
-   }
-
-   elmat = 0.0;
-
-   for (int i = 0; i < ir -> GetNPoints(); i++)
-   {
-      const IntegrationPoint &ip = ir->IntPoint(i);
-
-      trial_fe.CalcDivShape(ip,divshape);
-      test_fe.CalcDivShape(ip,te_divshape);
-
-      Trans.SetIntPoint (&ip);
-      c = ip.weight / Trans.Weight();
-
-      if (Q)
-      {
-         c *= Q -> Eval (Trans, ip);
-      }
-
-      te_divshape *= c;
-      AddMultVWt(te_divshape, divshape, elmat);
-   }
-}

 void VectorDiffusionIntegrator::AssembleElementMatrix(
   const FiniteElement &el,
@@ -3983,7 +3780,7 @@ void NormalTraceJumpIntegrator::AssembleFaceMatrix(
      for (i = 0; i < ndof1; i++)
         for (j = 0; j < face_ndof; j++)
         {
-            elmat(i, j) += shape1_n(i) * face_shape(j);
+            elmat(i, j) -= shape1_n(i) * face_shape(j);
         }
      if (ndof2)
      {
@@ -3991,7 +3788,7 @@ void NormalTraceJumpIntegrator::AssembleFaceMatrix(
         for (i = 0; i < ndof2; i++)
            for (j = 0; j < face_ndof; j++)
            {
-               elmat(ndof1+i, j) -= shape2_n(i) * face_shape(j);
+               elmat(ndof1+i, j) += shape2_n(i) * face_shape(j);
            }
      }
   }
@@ -215,10 +215,10 @@ public:
                             function by any coefficients describing the
                             integrator.
       @param[in] ir  If passed (the default value is NULL), the implementation
-                      of the method will ignore the integration rule provided
-                      by the @a fluxelem parameter and, instead, compute the
-                      discrete flux at the points specified by the integration
-                      rule @a ir.
+                        of the method will ignore the integration rule provided
+                        by the @a fluxelem parameter and, instead, compute the
+                        discrete flux at the points specified by the integration
+                        rule @a ir.
    */
   virtual void ComputeElementFlux(const FiniteElement &el,
                                   ElementTransformation &Trans,
@@ -2174,7 +2174,6 @@ public:
 /** Class for local mass matrix assembling a(u,v) := (Q u, v) */
 class MassIntegrator: public BilinearFormIntegrator
 {
-   friend class DGMassInverse;
 protected:
 #ifndef MFEM_THREAD_SAFE
   Vector shape, te_shape;
@@ -2525,7 +2524,6 @@ private:
 #ifndef MFEM_THREAD_SAFE
   Vector D;
   DenseMatrix curlshape, curlshape_dFt, M;
-   DenseMatrix te_curlshape, te_curlshape_dFt;
   DenseMatrix vshape, projcurl;
 #endif

@@ -2559,11 +2557,6 @@ public:
                                      ElementTransformation &Trans,
                                      DenseMatrix &elmat);

-   virtual void AssembleElementMatrix2(const FiniteElement &trial_fe,
-                                       const FiniteElement &test_fe,
-                                       ElementTransformation &Trans,
-                                       DenseMatrix &elmat);
-
   virtual void ComputeElementFlux(const FiniteElement &el,
                                   ElementTransformation &Trans,
                                   Vector &u, const FiniteElement &fluxelem,
@@ -2609,35 +2602,6 @@ public:
                                   const Vector &elfun);
 };

-/** Class for integrating the bilinear form a(u,v) := (Q curl u, v) where Q is
-    an optional scalar coefficient, and v is a vector with components v_i in
-    the L2 or H1 space. This integrator handles 3 cases:
-    (a) u ∈ H(curl) in 3D, v is a 3D vector with components v_i in L^2 or H^1
-    (b) u ∈ H(curl) in 2D, v is a scalar field in L^2 or H^1
-    (c) u is a scalar field in H^1, i.e, curl u := [0 1;-1 0]grad u and v is a
-        2D vector field with components v_i in L^2 or H^1 space.
-    Note: Case (b) can also be handled by MixedScalarCurlIntegrator  */
-class MixedCurlIntegrator : public BilinearFormIntegrator
-{
-protected:
-   Coefficient *Q;
-
-private:
-   Vector shape;
-   DenseMatrix dshape;
-   DenseMatrix curlshape;
-   DenseMatrix elmat_comp;
-public:
-   MixedCurlIntegrator() : Q{NULL} { }
-   MixedCurlIntegrator(Coefficient *q_) :  Q{q_} { }
-   MixedCurlIntegrator(Coefficient &q) :  Q{&q} { }
-
-   virtual void AssembleElementMatrix2(const FiniteElement &trial_fe,
-                                       const FiniteElement &test_fe,
-                                       ElementTransformation &Trans,
-                                       DenseMatrix &elmat);
-};
-
 /** Integrator for (Q u, v), where Q is an optional coefficient (of type scalar,
    vector (diagonal matrix), or matrix), trial function u is in H(Curl) or
    H(Div), and test function v is in H(Curl), H(Div), or v=(v1,...,vn), where
@@ -2761,7 +2725,7 @@ protected:

 private:
 #ifndef MFEM_THREAD_SAFE
-   Vector divshape, te_divshape;
+   Vector divshape;
 #endif

   // PA extension
@@ -2773,18 +2737,11 @@ private:

 public:
   DivDivIntegrator() { Q = NULL; }
-   DivDivIntegrator(Coefficient &q, const IntegrationRule *ir = NULL) :
-      BilinearFormIntegrator(ir), Q(&q) { }
+   DivDivIntegrator(Coefficient &q) : Q(&q) { }

   virtual void AssembleElementMatrix(const FiniteElement &el,
                                      ElementTransformation &Trans,
                                      DenseMatrix &elmat);
-
-   virtual void AssembleElementMatrix2(const FiniteElement &trial_fe,
-                                       const FiniteElement &test_fe,
-                                       ElementTransformation &Trans,
-                                       DenseMatrix &elmat);
-
   const Coefficient *GetCoefficient() const { return Q; }
 };

@@ -30,16 +30,7 @@ void ConvectionIntegrator::AssembleMF(const FiniteElementSpace &fes)
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedMFConvectionIntegrator(*this, fes, Q, alpha);
-      }
-      else
-      {
-         ceedOp = new ceed::MFConvectionIntegrator(fes, *ir, Q, alpha);
-      }
+      ceedOp = new ceed::MFConvectionIntegrator(fes, *ir, Q, alpha);
      return;
   }
   MFEM_ABORT("Error: ConvectionIntegrator::AssembleMF only implemented with"
@@ -12,7 +12,6 @@
 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
-#include "qfunction.hpp"
 #include "ceed/integrators/convection/convection.hpp"
 #include "quadinterpolator.hpp"

@@ -1387,16 +1386,7 @@ void ConvectionIntegrator::AssemblePA(const FiniteElementSpace &fes)
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedPAConvectionIntegrator(*this, fes, Q, alpha);
-      }
-      else
-      {
-         ceedOp = new ceed::PAConvectionIntegrator(fes, *ir, Q, alpha);
-      }
+      ceedOp = new ceed::PAConvectionIntegrator(fes, *ir, Q, alpha);
      return;
   }
   const int dims = el.GetDim();
@@ -1409,10 +1399,66 @@ void ConvectionIntegrator::AssemblePA(const FiniteElementSpace &fes)
   dofs1D = maps->ndof;
   quad1D = maps->nqpt;
   pa_data.SetSize(symmDims * nq * ne, mt);
+   Vector vel;
+   if (VectorConstantCoefficient *cQ =
+          dynamic_cast<VectorConstantCoefficient*>(Q))
+   {
+      vel = cQ->GetVec();
+   }
+   else if (VectorGridFunctionCoefficient *vgfQ =
+               dynamic_cast<VectorGridFunctionCoefficient*>(Q))
+   {
+      vel.SetSize(dim * nq * ne, mt);

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector vel(*Q, qs, CoefficientStorage::COMPRESSED);
+      const GridFunction *gf = vgfQ->GetGridFunction();
+      const FiniteElementSpace &gf_fes = *gf->FESpace();
+      const QuadratureInterpolator *qi(gf_fes.GetQuadratureInterpolator(*ir));
+      const bool use_tensor_products = UsesTensorBasis(gf_fes);
+      const ElementDofOrdering ordering = use_tensor_products ?
+                                          ElementDofOrdering::LEXICOGRAPHIC :
+                                          ElementDofOrdering::NATIVE;
+      const Operator *R = gf_fes.GetElementRestriction(ordering);

+      Vector xe(R->Height(), mt);
+      xe.UseDevice(true);
+
+      R->Mult(*gf, xe);
+      qi->SetOutputLayout(QVectorLayout::byVDIM);
+      qi->DisableTensorProducts(!use_tensor_products);
+      qi->Values(xe,vel);
+   }
+   else if (VectorQuadratureFunctionCoefficient* vqfQ =
+               dynamic_cast<VectorQuadratureFunctionCoefficient*>(Q))
+   {
+      const QuadratureFunction &qFun = vqfQ->GetQuadFunction();
+      MFEM_VERIFY(qFun.Size() == dim * nq * ne,
+                  "Incompatible QuadratureFunction dimension \n");
+
+      MFEM_VERIFY(ir == &qFun.GetSpace()->GetElementIntRule(0),
+                  "IntegrationRule used within integrator and in"
+                  " QuadratureFunction appear to be different");
+
+      qFun.Read();
+      vel.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
+   }
+   else
+   {
+      vel.SetSize(dim * nq * ne);
+      auto C = Reshape(vel.HostWrite(), dim, nq, ne);
+      DenseMatrix MQ_ir;
+      for (int e = 0; e < ne; ++e)
+      {
+         ElementTransformation& T = *fes.GetElementTransformation(e);
+         Q->Eval(MQ_ir, T, *ir);
+         for (int q = 0; q < nq; ++q)
+         {
+            for (int i = 0; i < dim; ++i)
+            {
+               C(i,q,e) = MQ_ir(i,q);
+            }
+         }
+      }
+   }
   PAConvectionSetup(dim, nq, ne, ir->GetWeights(), geom->J,
                     vel, alpha, pa_data);
 }
@@ -12,7 +12,6 @@
 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
-#include "qfunction.hpp"
 #include "restriction.hpp"

 using namespace std;
@@ -162,24 +161,88 @@ void DGTraceIntegrator::SetupPA(const FiniteElementSpace &fes, FaceType type)
   dofs1D = maps->ndof;
   quad1D = maps->nqpt;
   pa_data.SetSize(symmDims * nq * nf, Device::GetMemoryType());
-
-   FaceQuadratureSpace qs(*mesh, *ir, type);
-   CoefficientVector vel(*u, qs, CoefficientStorage::COMPRESSED);
-
-   CoefficientVector r(qs, CoefficientStorage::COMPRESSED);
-   if (rho == nullptr)
+   Vector vel;
+   if (VectorConstantCoefficient *c_u = dynamic_cast<VectorConstantCoefficient*>
+                                        (u))
   {
-      r.SetConstant(1.0);
+      vel = c_u->GetVec();
   }
-   else if (ConstantCoefficient *const_rho = dynamic_cast<ConstantCoefficient*>
-                                             (rho))
+   else if (VectorQuadratureFunctionCoefficient* qf_u =
+               dynamic_cast<VectorQuadratureFunctionCoefficient*>(u))
   {
-      r.SetConstant(const_rho->constant);
+      // Assumed to be in lexicographical ordering
+      const QuadratureFunction &qFun = qf_u->GetQuadFunction();
+      MFEM_VERIFY(qFun.Size() == dim * nq * nf,
+                  "Incompatible QuadratureFunction dimension \n");
+
+      MFEM_VERIFY(ir == &qFun.GetSpace()->GetElementIntRule(0),
+                  "IntegrationRule used within integrator and in"
+                  " QuadratureFunction appear to be different");
+      qFun.Read();
+      vel.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
+   }
+   else
+   {
+      vel.SetSize(dim * nq * nf);
+      auto C = Reshape(vel.HostWrite(), dim, nq, nf);
+      Vector Vq(dim);
+      int f_ind = 0;
+      for (int f = 0; f < mesh->GetNumFacesWithGhost(); ++f)
+      {
+         Mesh::FaceInformation face = mesh->GetFaceInformation(f);
+         if (face.IsNonconformingCoarse())
+         {
+            // We skip nonconforming coarse faces as they are treated
+            // by the corresponding nonconforming fine faces.
+            continue;
+         }
+         else if ( face.IsOfFaceType(type) )
+         {
+            const int mask = FaceElementTransformations::HAVE_ELEM1 |
+                             FaceElementTransformations::HAVE_LOC1;
+            FaceElementTransformations &T =
+               *fes.GetMesh()->GetFaceElementTransformations(f, mask);
+            for (int q = 0; q < nq; ++q)
+            {
+               // Convert to lexicographic ordering
+               int iq = ToLexOrdering(dim, face.element[0].local_face_id,
+                                      quad1D, q);
+               T.SetAllIntPoints(&ir->IntPoint(q));
+               const IntegrationPoint &eip1 = T.GetElement1IntPoint();
+               u->Eval(Vq, *T.Elem1, eip1);
+               for (int i = 0; i < dim; ++i)
+               {
+                  C(i,iq,f_ind) = Vq(i);
+               }
+            }
+            f_ind++;
+         }
+      }
+      MFEM_VERIFY(f_ind==nf, "Incorrect number of faces.");
+   }
+   Vector r;
+   if (rho==nullptr)
+   {
+      r.SetSize(1);
+      r(0) = 1.0;
+   }
+   else if (ConstantCoefficient *c_rho = dynamic_cast<ConstantCoefficient*>(rho))
+   {
+      r.SetSize(1);
+      r(0) = c_rho->constant;
   }
   else if (QuadratureFunctionCoefficient* qf_rho =
               dynamic_cast<QuadratureFunctionCoefficient*>(rho))
   {
-      r.MakeRef(qf_rho->GetQuadFunction());
+      const QuadratureFunction &qFun = qf_rho->GetQuadFunction();
+      MFEM_VERIFY(qFun.Size() == nq * nf,
+                  "Incompatible QuadratureFunction dimension \n");
+
+      MFEM_VERIFY(ir == &qFun.GetSpace()->GetElementIntRule(0),
+                  "IntegrationRule used within integrator and in"
+                  " QuadratureFunction appear to be different");
+      qFun.Read();
+      r.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
   }
   else
   {
@@ -191,42 +254,45 @@ void DGTraceIntegrator::SetupPA(const FiniteElementSpace &fes, FaceType type)
      for (int f = 0; f < mesh->GetNumFacesWithGhost(); ++f)
      {
         Mesh::FaceInformation face = mesh->GetFaceInformation(f);
-         if (face.IsNonconformingCoarse() || !face.IsOfFaceType(type))
+         if (face.IsNonconformingCoarse())
         {
            // We skip nonconforming coarse faces as they are treated
            // by the corresponding nonconforming fine faces.
            continue;
         }
-         FaceElementTransformations &T =
-            *fes.GetMesh()->GetFaceElementTransformations(f);
-         for (int q = 0; q < nq; ++q)
+         else if ( face.IsOfFaceType(type) )
         {
-            // Convert to lexicographic ordering
-            int iq = ToLexOrdering(dim, face.element[0].local_face_id,
-                                   quad1D, q);
-
-            T.SetAllIntPoints(&ir->IntPoint(q));
-            const IntegrationPoint &eip1 = T.GetElement1IntPoint();
-            const IntegrationPoint &eip2 = T.GetElement2IntPoint();
-            double rq;
-
-            if (face.IsBoundary())
+            FaceElementTransformations &T =
+               *fes.GetMesh()->GetFaceElementTransformations(f);
+            for (int q = 0; q < nq; ++q)
            {
-               rq = rho->Eval(*T.Elem1, eip1);
-            }
-            else
-            {
-               double udotn = 0.0;
-               for (int d=0; d<dim; ++d)
+               // Convert to lexicographic ordering
+               int iq = ToLexOrdering(dim, face.element[0].local_face_id,
+                                      quad1D, q);
+
+               T.SetAllIntPoints(&ir->IntPoint(q));
+               const IntegrationPoint &eip1 = T.GetElement1IntPoint();
+               const IntegrationPoint &eip2 = T.GetElement2IntPoint();
+               double rq;
+
+               if ( face.IsBoundary() )
               {
-                  udotn += C_vel(d,iq,f_ind)*n(iq,d,f_ind);
+                  rq = rho->Eval(*T.Elem1, eip1);
               }
-               if (udotn >= 0.0) { rq = rho->Eval(*T.Elem2, eip2); }
-               else { rq = rho->Eval(*T.Elem1, eip1); }
+               else
+               {
+                  double udotn = 0.0;
+                  for (int d=0; d<dim; ++d)
+                  {
+                     udotn += C_vel(d,iq,f_ind)*n(iq,d,f_ind);
+                  }
+                  if (udotn >= 0.0) { rq = rho->Eval(*T.Elem2, eip2); }
+                  else { rq = rho->Eval(*T.Elem1, eip1); }
+               }
+               C(iq,f_ind) = rq;
            }
-            C(iq,f_ind) = rq;
+            f_ind++;
         }
-         f_ind++;
      }
      MFEM_VERIFY(f_ind==nf, "Incorrect number of faces.");
   }
@@ -33,16 +33,7 @@ void DiffusionIntegrator::AssembleMF(const FiniteElementSpace &fes)
      MFEM_VERIFY(!VQ && !MQ,
                  "Only scalar coefficient supported for DiffusionIntegrator"
                  " with libCEED");
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedMFDiffusionIntegrator(*this, fes, Q);
-      }
-      else
-      {
-         ceedOp = new ceed::MFDiffusionIntegrator(fes, *ir, Q);
-      }
+      ceedOp = new ceed::MFDiffusionIntegrator(fes, *ir, Q);
      return;
   }
   MFEM_ABORT("Error: DiffusionIntegrator::AssembleMF only implemented with"
@@ -12,7 +12,6 @@
 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
-#include "qfunction.hpp"
 #include "ceed/integrators/diffusion/diffusion.hpp"

 using namespace std;
@@ -272,21 +271,18 @@ void PADiffusionSetup3D(const int Q1D,
                  D(qx,qy,qz,1,e) = D12; // 1,2
                  D(qx,qy,qz,2,e) = w_detJ * (A11*R13 + A12*R23 + A13*R33); // 1,3

+                  const double D21 = w_detJ * (A21*R11 + A22*R21 + A23*R31);
                  const double D22 = w_detJ * (A21*R12 + A22*R22 + A23*R32);
                  const double D23 = w_detJ * (A21*R13 + A22*R23 + A23*R33);

                  const double D33 = w_detJ * (A31*R13 + A32*R23 + A33*R33);

+                  D(qx,qy,qz,3,e) = symmetric ? D22 : D21; // 2,2 or 2,1
                  D(qx,qy,qz,4,e) = symmetric ? D23 : D22; // 2,3 or 2,2
                  D(qx,qy,qz,5,e) = symmetric ? D33 : D23; // 3,3 or 2,3

-                  if (symmetric)
+                  if (!symmetric)
                  {
-                     D(qx,qy,qz,3,e) = D22; // 2,2
-                  }
-                  else
-                  {
-                     D(qx,qy,qz,3,e) = w_detJ * (A21*R11 + A22*R21 + A23*R31); // 2,1
                     D(qx,qy,qz,6,e) = w_detJ * (A31*R11 + A32*R21 + A33*R31); // 3,1
                     D(qx,qy,qz,7,e) = w_detJ * (A31*R12 + A32*R22 + A33*R32); // 3,2
                     D(qx,qy,qz,8,e) = D33; // 3,3
@@ -369,16 +365,7 @@ void DiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)
      MFEM_VERIFY(!VQ && !MQ,
                  "Only scalar coefficient supported for DiffusionIntegrator"
                  " with libCEED");
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedPADiffusionIntegrator(*this, fes, Q);
-      }
-      else
-      {
-         ceedOp = new ceed::PADiffusionIntegrator(fes, *ir, Q);
-      }
+      ceedOp = new ceed::PADiffusionIntegrator(fes, *ir, Q);
      return;
   }
   const int dims = el.GetDim();
@@ -391,21 +378,120 @@ void DiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)
   maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
   dofs1D = maps->ndof;
   quad1D = maps->nqpt;
+   int coeffDim = 1;
+   Vector coeff;
+   const int MQfullDim = MQ ? MQ->GetHeight() * MQ->GetWidth() : 0;
+   if (auto *SMQ = dynamic_cast<SymmetricMatrixCoefficient *>(MQ))
+   {
+      MFEM_VERIFY(SMQ->GetSize() == dim, "");
+      coeffDim = symmDims;
+      coeff.SetSize(symmDims * nq * ne);

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(qs, CoefficientStorage::COMPRESSED);
+      DenseSymmetricMatrix sym_mat;
+      sym_mat.SetSize(dim);

-   if (MQ) { coeff.ProjectTranspose(*MQ); }
-   else if (VQ) { coeff.Project(*VQ); }
-   else if (Q) { coeff.Project(*Q); }
-   else { coeff.SetConstant(1.0); }
+      auto C = Reshape(coeff.HostWrite(), symmDims, nq, ne);

-   const int coeff_dim = coeff.GetVDim();
-   symmetric = (coeff_dim != dims*dims);
-   const int pa_size = symmetric ? symmDims : dims*dims;
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            SMQ->Eval(sym_mat, *tr, ir->IntPoint(p));
+            int cnt = 0;
+            for (int i=0; i<dim; ++i)
+               for (int j=i; j<dim; ++j, ++cnt)
+               {
+                  C(cnt, p, e) = sym_mat(i,j);
+               }
+         }
+      }
+   }
+   else if (MQ)
+   {
+      symmetric = false;
+      MFEM_VERIFY(MQ->GetHeight() == dim && MQ->GetWidth() == dim, "");

-   pa_data.SetSize(pa_size * nq * ne, mt);
-   PADiffusionSetup(dim, sdim, dofs1D, quad1D, coeff_dim, ne, ir->GetWeights(),
+      coeffDim = MQfullDim;
+
+      coeff.SetSize(MQfullDim * nq * ne);
+
+      DenseMatrix mat;
+      mat.SetSize(dim);
+
+      auto C = Reshape(coeff.HostWrite(), MQfullDim, nq, ne);
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            MQ->Eval(mat, *tr, ir->IntPoint(p));
+            for (int i=0; i<dim; ++i)
+               for (int j=0; j<dim; ++j)
+               {
+                  C(j+(i*dim), p, e) = mat(i,j);
+               }
+         }
+      }
+   }
+   else if (VQ)
+   {
+      MFEM_VERIFY(VQ->GetVDim() == dim, "");
+      coeffDim = VQ->GetVDim();
+      coeff.SetSize(coeffDim * nq * ne);
+      auto C = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
+      Vector DM(coeffDim);
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            VQ->Eval(DM, *tr, ir->IntPoint(p));
+            for (int i=0; i<coeffDim; ++i)
+            {
+               C(i, p, e) = DM[i];
+            }
+         }
+      }
+   }
+   else if (Q == nullptr)
+   {
+      coeff.SetSize(1);
+      coeff(0) = 1.0;
+   }
+   else if (ConstantCoefficient* cQ = dynamic_cast<ConstantCoefficient*>(Q))
+   {
+      coeff.SetSize(1);
+      coeff(0) = cQ->constant;
+   }
+   else if (QuadratureFunctionCoefficient* qfQ =
+               dynamic_cast<QuadratureFunctionCoefficient*>(Q))
+   {
+      const QuadratureFunction &qFun = qfQ->GetQuadFunction();
+      MFEM_VERIFY(qFun.Size() == ne*nq,
+                  "Incompatible QuadratureFunction dimension \n");
+
+      MFEM_VERIFY(ir == &qFun.GetSpace()->GetElementIntRule(0),
+                  "IntegrationRule used within integrator and in"
+                  " QuadratureFunction appear to be different");
+      qFun.Read();
+      coeff.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
+   }
+   else
+   {
+      coeff.SetSize(nq * ne);
+      auto C = Reshape(coeff.HostWrite(), nq, ne);
+      for (int e = 0; e < ne; ++e)
+      {
+         ElementTransformation& T = *fes.GetElementTransformation(e);
+         for (int q = 0; q < nq; ++q)
+         {
+            C(q,e) = Q->Eval(T, ir->IntPoint(q));
+         }
+      }
+   }
+   pa_data.SetSize((symmetric ? symmDims : MQfullDim) * nq * ne, mt);
+   PADiffusionSetup(dim, sdim, dofs1D, quad1D, coeffDim, ne, ir->GetWeights(),
                    geom->J, coeff, pa_data);
 }

@@ -1686,7 +1772,7 @@ static void PADiffusionApply(const int dim,
         case 0x77: return SmemPADiffusionApply2D<7,7,4>(NE,symm,B,G,D,X,Y);
         case 0x88: return SmemPADiffusionApply2D<8,8,2>(NE,symm,B,G,D,X,Y);
         case 0x99: return SmemPADiffusionApply2D<9,9,2>(NE,symm,B,G,D,X,Y);
-            // default:   return PADiffusionApply2D(NE,symm,B,G,Bt,Gt,D,X,Y,D1D,Q1D);
+         default:   return PADiffusionApply2D(NE,symm,B,G,Bt,Gt,D,X,Y,D1D,Q1D);
      }
   }

@@ -1704,14 +1790,7 @@ static void PADiffusionApply(const int dim,
         case 0x67: return SmemPADiffusionApply3D<6,7>(NE,symm,B,G,D,X,Y);
         case 0x78: return SmemPADiffusionApply3D<7,8>(NE,symm,B,G,D,X,Y);
         case 0x89: return SmemPADiffusionApply3D<8,9>(NE,symm,B,G,D,X,Y);
-         case 0x33: return SmemPADiffusionApply3D<3,3>(NE,symm,B,G,D,X,Y);
-         case 0x44: return SmemPADiffusionApply3D<4,4>(NE,symm,B,G,D,X,Y);
-         case 0x55: return SmemPADiffusionApply3D<5,5>(NE,symm,B,G,D,X,Y);
-         case 0x66: return SmemPADiffusionApply3D<6,6>(NE,symm,B,G,D,X,Y);
-         case 0x77: return SmemPADiffusionApply3D<7,7>(NE,symm,B,G,D,X,Y);
-         case 0x88: return SmemPADiffusionApply3D<8,8>(NE,symm,B,G,D,X,Y);
-         case 0x99: return SmemPADiffusionApply3D<9,9>(NE,symm,B,G,D,X,Y);
-            // default:   return PADiffusionApply3D(NE,symm,B,G,Bt,Gt,D,X,Y,D1D,Q1D);
+         default:   return PADiffusionApply3D(NE,symm,B,G,Bt,Gt,D,X,Y,D1D,Q1D);
      }
   }
   MFEM_ABORT("Unknown kernel: 0x"<<std::hex << id << std::dec);
@@ -12,7 +12,6 @@
 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
-#include "qfunction.hpp"

 using namespace std;

@@ -210,8 +209,44 @@ void GradientIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
               "PA requires test and trial space to have same number of quadrature points!");
   pa_data.SetSize(nq * dimsToStore * ne, Device::GetMemoryType());

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(Q, qs, CoefficientStorage::COMPRESSED);
+   Vector coeff;
+
+   if (Q == nullptr)
+   {
+      coeff.SetSize(1);
+      coeff(0) = 1.0;
+   }
+   else if (ConstantCoefficient* cQ = dynamic_cast<ConstantCoefficient*>(Q))
+   {
+      coeff.SetSize(1);
+      coeff(0) = cQ->constant;
+   }
+   else if (QuadratureFunctionCoefficient* qfQ =
+               dynamic_cast<QuadratureFunctionCoefficient*>(Q))
+   {
+      const QuadratureFunction &qFun = qfQ->GetQuadFunction();
+      MFEM_VERIFY(qFun.Size() == ne*nq,
+                  "Incompatible QuadratureFunction dimension \n");
+
+      MFEM_VERIFY(ir == &qFun.GetSpace()->GetElementIntRule(0),
+                  "IntegrationRule used within integrator and in"
+                  " QuadratureFunction appear to be different");
+      qFun.Read();
+      coeff.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
+   }
+   else
+   {
+      coeff.SetSize(nq * ne);
+      auto C = Reshape(coeff.HostWrite(), nq, ne);
+      for (int e = 0; e < ne; ++e)
+      {
+         ElementTransformation& T = *trial_fes.GetElementTransformation(e);
+         for (int q = 0; q < nq; ++q)
+         {
+            C(q,e) = Q->Eval(T, ir->IntPoint(q));
+         }
+      }
+   }

   PAGradientSetup(dim, trial_dofs1D, test_dofs1D, quad1D,
                   ne, ir->GetWeights(), geom->J, coeff, pa_data);
@@ -830,3 +865,4 @@ void GradientIntegrator::AddMultTransposePA(const Vector &x, Vector &y) const
 }

 } // namespace mfem
+
@@ -12,7 +12,6 @@
 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
-#include "qspace.hpp"

 using namespace std;

@@ -968,6 +967,8 @@ void CurlCurlIntegrator::AssemblePA(const FiniteElementSpace &fes)
   dim = mesh->Dimension();
   MFEM_VERIFY(dim == 2 || dim == 3, "");

+   const int dimc = (dim == 3) ? 3 : 1;
+
   ne = fes.GetNE();
   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::JACOBIANS);
   mapsC = &el->GetDofToQuad(*ir, DofToQuad::TENSOR);
@@ -977,19 +978,88 @@ void CurlCurlIntegrator::AssemblePA(const FiniteElementSpace &fes)

   MFEM_VERIFY(dofs1D == mapsO->ndof + 1 && quad1D == mapsO->nqpt, "");

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(qs, CoefficientStorage::SYMMETRIC);
-   if (Q) { coeff.Project(*Q); }
-   else if (MQ) { coeff.ProjectTranspose(*MQ); }
-   else if (DQ) { coeff.Project(*DQ); }
-   else { coeff.SetConstant(1.0); }
+   auto SMQ = dynamic_cast<SymmetricMatrixCoefficient *>(MQ);

-   const int coeff_dim = coeff.GetVDim();
-   symmetric = (coeff_dim != dim*dim);
-   const int sym_dims = (dims * (dims + 1)) / 2; // 1x1: 1, 2x2: 3, 3x3: 6
-   const int ndata = (dim == 2) ? 1 : (symmetric ? sym_dims : dim*dim);
+   const int MQsymmDim = SMQ ? (SMQ->GetSize() * (SMQ->GetSize() + 1)) / 2 : 0;
+   const int MQfullDim = MQ ? (MQ->GetHeight() * MQ->GetWidth()) : 0;
+   const int MQdim = SMQ ? MQsymmDim : MQfullDim;
+   const int coeffDim = MQ ? MQdim : (DQ ? DQ->GetVDim() : 1);
+
+   symmetric = (SMQ || MQ == NULL);
+
+   const int symmDims = (dims * (dims + 1)) / 2; // 1x1: 1, 2x2: 3, 3x3: 6
+   const int ndata = (dim == 2) ? 1 : (symmetric ? symmDims : MQfullDim);
   pa_data.SetSize(ndata * nq * ne, Device::GetMemoryType());

+   Vector coeff(coeffDim * ne * nq);
+   coeff = 1.0;
+   auto coeffh = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
+   if (Q || DQ || MQ)
+   {
+      Vector DM(DQ ? coeffDim : 0);
+      DenseMatrix GM;
+      DenseSymmetricMatrix SM;
+
+      if (DQ)
+      {
+         MFEM_VERIFY(coeffDim == dimc, "");
+      }
+      if (SMQ)
+      {
+         SM.SetSize(dimc);
+         MFEM_VERIFY(SMQ->GetSize() == dimc, "");
+      }
+      else if (MQ)
+      {
+         GM.SetSize(dimc);
+         MFEM_VERIFY(coeffDim == MQdim, "");
+         MFEM_VERIFY(MQ->GetHeight() == dimc && MQ->GetWidth() == dimc, "");
+      }
+
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            if (SMQ)
+            {
+               SMQ->Eval(SM, *tr, ir->IntPoint(p));
+
+               int cnt = 0;
+               for (int i=0; i<dimc; ++i)
+                  for (int j=i; j<dimc; ++j, ++cnt)
+                  {
+                     coeffh(cnt, p, e) = SM(i,j);
+                  }
+
+            }
+            else if (MQ)
+            {
+               MQ->Eval(GM, *tr, ir->IntPoint(p));
+
+               for (int i=0; i<dimc; ++i)
+                  for (int j=0; j<dimc; ++j)
+                  {
+                     coeffh(j+(i*dimc), p, e) = GM(i,j);
+                  }
+
+            }
+            else if (DQ)
+            {
+               DQ->Eval(DM, *tr, ir->IntPoint(p));
+               for (int i=0; i<coeffDim; ++i)
+               {
+                  coeffh(i, p, e) = DM[i];
+               }
+            }
+            else
+            {
+               coeffh(0, p, e) = Q->Eval(*tr, ir->IntPoint(p));
+            }
+         }
+      }
+   }
+
   if (el->GetDerivType() != mfem::FiniteElement::CURL)
   {
      MFEM_ABORT("Unknown kernel.");
@@ -997,7 +1067,7 @@ void CurlCurlIntegrator::AssemblePA(const FiniteElementSpace &fes)

   if (dim == 3)
   {
-      PACurlCurlSetup3D(quad1D, coeff_dim, ne, ir->GetWeights(), geom->J, coeff,
+      PACurlCurlSetup3D(quad1D, coeffDim, ne, ir->GetWeights(), geom->J, coeff,
                        pa_data);
   }
   else
@@ -2710,7 +2780,7 @@ void CurlCurlIntegrator::AssembleDiagonalPA(Vector& diag)
   }
 }

-// Apply to x corresponding to DOFs in H^1 (trial), whose gradients are
+// Apply to x corresponding to DOF's in H^1 (trial), whose gradients are
 // integrated against H(curl) test functions corresponding to y.
 void PAHcurlH1Apply3D(const int D1D,
                      const int Q1D,
@@ -2900,7 +2970,7 @@ void PAHcurlH1Apply3D(const int D1D,
   }); // end of element loop
 }

-// Apply to x corresponding to DOFs in H(curl), integrated
+// Apply to x corresponding to DOF's in H(curl), integrated
 // against gradients of H^1 functions corresponding to y.
 void PAHcurlH1ApplyTranspose3D(const int D1D,
                               const int Q1D,
@@ -3099,7 +3169,7 @@ void PAHcurlH1ApplyTranspose3D(const int D1D,
   }); // end of element loop
 }

-// Apply to x corresponding to DOFs in H^1 (trial), whose gradients are
+// Apply to x corresponding to DOF's in H^1 (trial), whose gradients are
 // integrated against H(curl) test functions corresponding to y.
 void PAHcurlH1Apply2D(const int D1D,
                      const int Q1D,
@@ -3223,7 +3293,7 @@ void PAHcurlH1Apply2D(const int D1D,
   }); // end of element loop
 }

-// Apply to x corresponding to DOFs in H(curl), integrated
+// Apply to x corresponding to DOF's in H(curl), integrated
 // against gradients of H^1 functions corresponding to y.
 void PAHcurlH1ApplyTranspose2D(const int D1D,
                               const int Q1D,
@@ -3419,8 +3489,20 @@ void MixedScalarCurlIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,

   pa_data.SetSize(nq * ne, Device::GetMemoryType());

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(Q, qs, CoefficientStorage::FULL);
+   Vector coeff(ne * nq);
+   coeff = 1.0;
+   auto coeffh = Reshape(coeff.HostWrite(), nq, ne);
+   if (Q)
+   {
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            coeffh(p, e) = Q->Eval(*tr, ir->IntPoint(p));
+         }
+      }
+   }

   if (dim == 2)
   {
@@ -3511,11 +3593,38 @@ void MixedVectorCurlIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
   const int ndata = curlSpaces ? (coeffDim == 1 ? 1 : 9) : symmDims;
   pa_data.SetSize(ndata * nq * ne, Device::GetMemoryType());

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(qs, CoefficientStorage::FULL);
-   if (Q) { coeff.Project(*Q); }
-   else if (DQ) { coeff.Project(*DQ); }
-   else { coeff.SetConstant(1.0); }
+   Vector coeff(coeffDim * nq * ne);
+   coeff = 1.0;
+   auto coeffh = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
+   if (Q || DQ)
+   {
+      Vector V(coeffDim);
+      if (DQ)
+      {
+         MFEM_VERIFY(DQ->GetVDim() == coeffDim, "");
+      }
+
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+
+         for (int p=0; p<nq; ++p)
+         {
+            if (DQ)
+            {
+               DQ->Eval(V, *tr, ir->IntPoint(p));
+               for (int i=0; i<coeffDim; ++i)
+               {
+                  coeffh(i, p, e) = V[i];
+               }
+            }
+            else
+            {
+               coeffh(0, p, e) = Q->Eval(*tr, ir->IntPoint(p));
+            }
+         }
+      }
+   }

   if (testType == mfem::FiniteElement::CURL &&
       trialType == mfem::FiniteElement::CURL && dim == 3)
@@ -3543,7 +3652,7 @@ void MixedVectorCurlIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
   }
 }

-// Apply to x corresponding to DOFs in H(curl) (trial), whose curl is
+// Apply to x corresponding to DOF's in H(curl) (trial), whose curl is
 // integrated against H(curl) test functions corresponding to y.
 template<int MAX_D1D = HCURL_MAX_D1D, int MAX_Q1D = HCURL_MAX_Q1D>
 static void PAHcurlL2Apply3D(const int D1D,
@@ -3906,7 +4015,7 @@ static void PAHcurlL2Apply3D(const int D1D,
   }); // end of element loop
 }

-// Apply to x corresponding to DOFs in H(curl) (trial), whose curl is
+// Apply to x corresponding to DOF's in H(curl) (trial), whose curl is
 // integrated against H(curl) test functions corresponding to y.
 template<int MAX_D1D = HCURL_MAX_D1D, int MAX_Q1D = HCURL_MAX_Q1D>
 static void SmemPAHcurlL2Apply3D(const int D1D,
@@ -4216,7 +4325,7 @@ static void SmemPAHcurlL2Apply3D(const int D1D,
   ForallWrap<3>(true, NE, device_kernel, host_kernel, Q1D, Q1D, Q1D);
 }

-// Apply to x corresponding to DOFs in H(curl) (trial), whose curl is
+// Apply to x corresponding to DOF's in H(curl) (trial), whose curl is
 // integrated against H(div) test functions corresponding to y.
 template<int MAX_D1D = HCURL_MAX_D1D, int MAX_Q1D = HCURL_MAX_Q1D>
 static void PAHcurlHdivApply3D(const int D1D,
@@ -4572,7 +4681,7 @@ static void PAHcurlHdivApply3D(const int D1D,
   }); // end of element loop
 }

-// Apply to x corresponding to DOFs in H(div) (test), integrated against the
+// Apply to x corresponding to DOF's in H(div) (test), integrated against the
 // curl of H(curl) trial functions corresponding to y.
 template<int MAX_D1D = HCURL_MAX_D1D, int MAX_Q1D = HCURL_MAX_Q1D>
 static void PAHcurlHdivApply3DTranspose(const int D1D,
@@ -5037,11 +5146,38 @@ void MixedVectorWeakCurlIntegrator::AssemblePA(const FiniteElementSpace

   pa_data.SetSize(ndata * nq * ne, Device::GetMemoryType());

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(qs, CoefficientStorage::FULL);
-   if (Q) { coeff.Project(*Q); }
-   else if (DQ) { coeff.Project(*DQ); }
-   else { coeff.SetConstant(1.0); }
+   Vector coeff(coeffDim * nq * ne);
+   coeff = 1.0;
+   auto coeffh = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
+   if (Q || DQ)
+   {
+      Vector V(coeffDim);
+      if (DQ)
+      {
+         MFEM_VERIFY(DQ->GetVDim() == coeffDim, "");
+      }
+
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+
+         for (int p=0; p<nq; ++p)
+         {
+            if (DQ)
+            {
+               DQ->Eval(V, *tr, ir->IntPoint(p));
+               for (int i=0; i<coeffDim; ++i)
+               {
+                  coeffh(i, p, e) = V[i];
+               }
+            }
+            else
+            {
+               coeffh(0, p, e) = Q->Eval(*tr, ir->IntPoint(p));
+            }
+         }
+      }
+   }

   if (trialType == mfem::FiniteElement::CURL && dim == 3)
   {
@@ -5067,7 +5203,7 @@ void MixedVectorWeakCurlIntegrator::AssemblePA(const FiniteElementSpace
   }
 }

-// Apply to x corresponding to DOFs in H(curl) (trial), integrated against curl
+// Apply to x corresponding to DOF's in H(curl) (trial), integrated against curl
 // of H(curl) test functions corresponding to y.
 template<int MAX_D1D = HCURL_MAX_D1D, int MAX_Q1D = HCURL_MAX_Q1D>
 static void PAHcurlL2Apply3DTranspose(const int D1D,
@@ -12,7 +12,6 @@
 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
-#include "qspace.hpp"

 using namespace std;

@@ -25,20 +24,18 @@ namespace mfem

 // PA H(div) Mass Assemble 2D kernel
 void PAHdivSetup2D(const int Q1D,
-                   const int coeffDim,
                   const int NE,
                   const Array<double> &w,
                   const Vector &j,
                   Vector &coeff_,
                   Vector &op)
 {
-   const bool symmetric = (coeffDim != 4);
   const int NQ = Q1D*Q1D;
   auto W = w.Read();

   auto J = Reshape(j.Read(), NQ, 2, 2, NE);
-   auto C = Reshape(coeff_.Read(), coeffDim, NQ, NE);
-   auto y = Reshape(op.Write(), NQ, symmetric ? 3 : 4, NE);
+   auto coeff = Reshape(coeff_.Read(), NQ, NE);
+   auto y = Reshape(op.Write(), NQ, 3, NE);

   MFEM_FORALL(e, NE,
   {
@@ -48,60 +45,28 @@ void PAHdivSetup2D(const int Q1D,
         const double J21 = J(q,1,0,e);
         const double J12 = J(q,0,1,e);
         const double J22 = J(q,1,1,e);
-         const double c_detJ = W[q] / ((J11*J22)-(J21*J12));
-
-         // (1/detJ) J^T C J
-         if (coeffDim == 3 || coeffDim == 4) // Matrix coefficient
-         {
-            const double C11 = C(0,q,e);
-            const double C12 = C(1,q,e);
-            const double C21 = symmetric ? C12 : C(2,q,e);
-            const double C22 = symmetric ? C(2,q,e) : C(3,q,e);
-            const double R11 = C11*J11 + C12*J21;
-            const double R21 = C21*J11 + C22*J21;
-            const double R12 = C11*J12 + C12*J22;
-            const double R22 = C21*J12 + C22*J22;
-
-            y(q,0,e) = c_detJ * (J11*R11 + J21*R21); // 1,1
-            y(q,1,e) = c_detJ * (J11*R12 + J21*R22); // 1,2
-
-            if (symmetric)
-            {
-               y(q,2,e) = c_detJ * (J12*R12 + J22*R22); // 2,2
-            }
-            else
-            {
-               y(q,2,e) = c_detJ * (J12*R11 + J22*R21); // 2,1
-               y(q,3,e) = c_detJ * (J12*R12 + J22*R22); // 2,2
-            }
-         }
-         else // Vector or scalar coefficient
-         {
-            const double C1 = C(0,q,e);
-            const double C2 = (coeffDim == 2 ? C(1,q,e) : C1);
-            y(q,0,e) = c_detJ * (J11*C1*J11 + J21*C2*J21); // 1,1
-            y(q,1,e) = c_detJ * (J11*C1*J12 + J21*C2*J22); // 1,2
-            y(q,2,e) = c_detJ * (J12*C1*J12 + J22*C2*J22); // 2,2
-         }
+         const double c_detJ = W[q] * coeff(q, e) / ((J11*J22)-(J21*J12));
+         // (c/detJ) J^T J
+         y(q,0,e) = c_detJ * (J11*J11 + J21*J21); // 1,1
+         y(q,1,e) = c_detJ * (J11*J12 + J21*J22); // 1,2
+         y(q,2,e) = c_detJ * (J12*J12 + J22*J22); // 2,2
      }
   });
 }

 // PA H(div) Mass Assemble 3D kernel
 void PAHdivSetup3D(const int Q1D,
-                   const int coeffDim,
                   const int NE,
                   const Array<double> &w,
                   const Vector &j,
                   Vector &coeff_,
                   Vector &op)
 {
-   const bool symmetric = (coeffDim != 9);
   const int NQ = Q1D*Q1D*Q1D;
   auto W = w.Read();
   auto J = Reshape(j.Read(), NQ, 3, 3, NE);
-   auto C = Reshape(coeff_.Read(), coeffDim, NQ, NE);
-   auto y = Reshape(op.Write(), NQ, symmetric ? 6 : 9, NE);
+   auto coeff = Reshape(coeff_.Read(), NQ, NE);
+   auto y = Reshape(op.Write(), NQ, 6, NE);

   MFEM_FORALL(e, NE,
   {
@@ -119,58 +84,14 @@ void PAHdivSetup3D(const int Q1D,
         const double detJ = J11 * (J22 * J33 - J32 * J23) -
         /* */               J21 * (J12 * J33 - J32 * J13) +
         /* */               J31 * (J12 * J23 - J22 * J13);
-         const double c_detJ = W[q] / detJ;
-
-         // (1/detJ) J^T C J
-         if (coeffDim == 6 || coeffDim == 9) // Matrix coefficient version
-         {
-            double M[3][3];
-            M[0][0] = C(0, q, e);
-            M[0][1] = C(1, q, e);
-            M[0][2] = C(2, q, e);
-            M[1][0] = (!symmetric) ? C(3, q, e) : M[0][1];
-            M[1][1] = (!symmetric) ? C(4, q, e) : C(3, q, e);
-            M[1][2] = (!symmetric) ? C(5, q, e) : C(4, q, e);
-            M[2][0] = (!symmetric) ? C(6, q, e) : M[0][2];
-            M[2][1] = (!symmetric) ? C(7, q, e) : M[1][2];
-            M[2][2] = (!symmetric) ? C(8, q, e) : C(5, q, e);
-
-            int idx = 0;
-            for (int i=0; i<3; ++i)
-               for (int j = (symmetric ? i : 0); j<3; ++j)
-               {
-                  y(q,idx,e) = 0.0;
-                  for (int k=0; k<3; ++k)
-                  {
-                     double MJ_kj = 0.0;
-                     for (int l=0; l<3; ++l)
-                     {
-                        MJ_kj += M[k][l] * J(q,l,j,e);
-                     }
-
-                     y(q,idx,e) += J(q,k,i,e) * MJ_kj;
-                  }
-
-                  y(q,idx,e) *= c_detJ;
-                  idx++;
-               }
-         }
-         else  // Vector or scalar coefficient version
-         {
-            int idx = 0;
-            for (int i=0; i<3; ++i)
-               for (int j=i; j<3; ++j)
-               {
-                  y(q,idx,e) = 0.0;
-                  for (int k=0; k<3; ++k)
-                  {
-                     y(q,idx,e) += J(q,k,i,e) * C(coeffDim == 3 ? k : 0, q, e) * J(q,k,j,e);
-                  }
-
-                  y(q,idx,e) *= c_detJ;
-                  idx++;
-               }
-         }
+         const double c_detJ = W[q] * coeff(q, e) / detJ;
+         // (c/detJ) J^T J
+         y(q,0,e) = c_detJ * (J11*J11 + J21*J21 + J31*J31); // 1,1
+         y(q,1,e) = c_detJ * (J12*J11 + J22*J21 + J32*J31); // 2,1
+         y(q,2,e) = c_detJ * (J13*J11 + J23*J21 + J33*J31); // 3,1
+         y(q,3,e) = c_detJ * (J12*J12 + J22*J22 + J32*J32); // 2,2
+         y(q,4,e) = c_detJ * (J13*J12 + J23*J22 + J33*J32); // 3,2
+         y(q,5,e) = c_detJ * (J13*J13 + J23*J23 + J33*J33); // 3,3
      }
   });
 }
@@ -178,7 +99,6 @@ void PAHdivSetup3D(const int Q1D,
 void PAHdivMassApply2D(const int D1D,
                       const int Q1D,
                       const int NE,
-                       const bool symmetric,
                       const Array<double> &Bo_,
                       const Array<double> &Bc_,
                       const Array<double> &Bot_,
@@ -195,7 +115,7 @@ void PAHdivMassApply2D(const int D1D,
   auto Bc = Reshape(Bc_.Read(), Q1D, D1D);
   auto Bot = Reshape(Bot_.Read(), D1D-1, Q1D);
   auto Bct = Reshape(Bct_.Read(), D1D, Q1D);
-   auto op = Reshape(op_.Read(), Q1D, Q1D, symmetric ? 3 : 4, NE);
+   auto op = Reshape(op_.Read(), Q1D, Q1D, 3, NE);
   auto x = Reshape(x_.Read(), 2*(D1D-1)*D1D, NE);
   auto y = Reshape(y_.ReadWrite(), 2*(D1D-1)*D1D, NE);

@@ -258,12 +178,11 @@ void PAHdivMassApply2D(const int D1D,
         {
            const double O11 = op(qx,qy,0,e);
            const double O12 = op(qx,qy,1,e);
-            const double O21 = symmetric ? O12 : op(qx,qy,2,e);
-            const double O22 = symmetric ? op(qx,qy,2,e) : op(qx,qy,3,e);
+            const double O22 = op(qx,qy,2,e);
            const double massX = mass[qy][qx][0];
            const double massY = mass[qy][qx][1];
            mass[qy][qx][0] = (O11*massX)+(O12*massY);
-            mass[qy][qx][1] = (O21*massX)+(O22*massY);
+            mass[qy][qx][1] = (O12*massX)+(O22*massY);
         }
      }

@@ -306,179 +225,9 @@ void PAHdivMassApply2D(const int D1D,
   }); // end of element loop
 }

-template<int T_D1D = 0, int T_Q1D = 0>
-void SmemPAHdivMassApply2D(const int NE,
-                           const bool symmetric,
-                           const Array<double> &Bo_,
-                           const Array<double> &Bc_,
-                           const Array<double> &Bot_,
-                           const Array<double> &Bct_,
-                           const Vector &op_,
-                           const Vector &x_,
-                           Vector &y_,
-                           const int d1d = 0,
-                           const int q1d = 0)
-{
-   MFEM_CONTRACT_VAR(Bot_);
-   MFEM_CONTRACT_VAR(Bct_);
-
-   static constexpr int VDIM = 2;
-
-   const int D1D = T_D1D ? T_D1D : d1d;
-   const int Q1D = T_Q1D ? T_Q1D : q1d;
-
-   const auto bo = Reshape(Bo_.Read(), Q1D, D1D-1);
-   const auto bc = Reshape(Bc_.Read(), Q1D, D1D);
-   const auto D = Reshape(op_.Read(), Q1D, Q1D, symmetric ? 3 : 4, NE);
-   const auto x = Reshape(x_.Read(), D1D*(D1D-1), VDIM, NE);
-   auto y = y_.ReadWrite();
-
-   MFEM_FORALL_3D(e, NE, Q1D, Q1D, VDIM,
-   {
-      const int tidz = MFEM_THREAD_ID(z);
-
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-
-      constexpr int MQ1 = T_Q1D ? T_Q1D : HDIV_MAX_Q1D;
-      constexpr int MD1 = T_D1D ? T_D1D : HDIV_MAX_D1D;
-      constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
-
-      MFEM_SHARED double smo[MQ1*(MD1-1)];
-      DeviceMatrix Bo(smo, D1D-1, Q1D);
-
-      MFEM_SHARED double smc[MQ1*MD1];
-      DeviceMatrix Bc(smc, D1D, Q1D);
-
-      MFEM_SHARED double sm0[VDIM*MDQ*MDQ];
-      MFEM_SHARED double sm1[VDIM*MDQ*MDQ];
-      DeviceMatrix X(sm0, D1D*(D1D-1), VDIM);
-      DeviceCube QD(sm1, Q1D, D1D, VDIM);
-      DeviceCube QQ(sm0, Q1D, Q1D, VDIM);
-
-      // Load X, Bo and Bc into shared memory
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         MFEM_FOREACH_THREAD(dy,y,D1D)
-         {
-            MFEM_FOREACH_THREAD(qx,x,Q1D)
-            {
-               if (qx < D1D && dy < (D1D-1)) { X(qx + dy*D1D,vd) = x(qx+dy*D1D,vd,e); }
-               if (tidz == 0)
-               {
-                  if (dy < (D1D-1)) { Bo(dy,qx) = bo(qx,dy); }
-                  Bc(dy,qx) = bc(qx,dy);
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      // Apply B operator
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int nx = (vd == 0) ? D1D : D1D-1;
-         const int ny = (vd == 1) ? D1D : D1D-1;
-         DeviceCube Xxy(X, nx, ny, VDIM);
-         DeviceMatrix Bx = (vd == 0) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(dy,y,ny)
-         {
-            MFEM_FOREACH_THREAD(qx,x,Q1D)
-            {
-               double dq = 0.0;
-               for (int dx = 0; dx < nx; ++dx)
-               {
-                  dq += Xxy(dx,dy,vd) * Bx(dx,qx);
-               }
-               QD(qx,dy,vd) = dq;
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int ny = (vd == 1) ? D1D : D1D-1;
-         DeviceMatrix By = (vd == 1) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            MFEM_FOREACH_THREAD(qx,x,Q1D)
-            {
-               double qq = 0.0;
-               for (int dy = 0; dy < ny; ++dy)
-               {
-                  qq += QD(qx,dy,vd) * By(dy,qy);
-               }
-               QQ(qx,qy,vd) = qq;
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      // Apply D operator
-      if (tidz == 0)
-      {
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            MFEM_FOREACH_THREAD(qx,x,Q1D)
-            {
-               const double Qx = QQ(qx,qy,0);
-               const double Qy = QQ(qx,qy,1);
-
-               const double D11 = D(qx,qy,0,e);
-               const double D12 = D(qx,qy,1,e);
-               const double D21 = symmetric ? D12 : D(qx,qy,2,e);
-               const double D22 = symmetric ? D(qx,qy,2,e) : D(qx,qy,3,e);
-
-               QQ(qx,qy,0) = D11*Qx + D12*Qy;
-               QQ(qx,qy,1) = D21*Qx + D22*Qy;
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      // Apply Bt operator
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int nx = (vd == 0) ? D1D : D1D-1;
-         DeviceMatrix Btx = (vd == 0) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               double qd = 0.0;
-               for (int qx = 0; qx < Q1D; ++qx)
-               {
-                  qd += QQ(qx,qy,vd) * Btx(dx,qx);
-               }
-               QD(dx,qy,vd) = qd;
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int nx = (vd == 0) ? D1D : D1D-1;
-         const int ny = (vd == 1) ? D1D : D1D-1;
-         DeviceMatrix Bty = (vd == 1) ? Bc : Bo;
-         DeviceTensor<4> Yxy(y, nx, ny, VDIM, NE);
-         MFEM_FOREACH_THREAD(dy,y,ny)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               double dd = 0.0;
-               for (int qy = 0; qy < Q1D; ++qy)
-               {
-                  dd += QD(dx,qy,vd) * Bty(dy,qy);
-               }
-               Yxy(dx,dy,vd,e) += dd;
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-   });
-}
-
 void PAHdivMassAssembleDiagonal2D(const int D1D,
                                  const int Q1D,
                                  const int NE,
-                                  const bool symmetric,
                                  const Array<double> &Bo_,
                                  const Array<double> &Bc_,
                                  const Vector &op_,
@@ -489,7 +238,7 @@ void PAHdivMassAssembleDiagonal2D(const int D1D,

   auto Bo = Reshape(Bo_.Read(), Q1D, D1D-1);
   auto Bc = Reshape(Bc_.Read(), Q1D, D1D);
-   auto op = Reshape(op_.Read(), Q1D, Q1D, symmetric ? 3 : 4, NE);
+   auto op = Reshape(op_.Read(), Q1D, Q1D, 3, NE);
   auto diag = Reshape(diag_.ReadWrite(), 2*(D1D-1)*D1D, NE);

   MFEM_FORALL(e, NE,
@@ -510,7 +259,7 @@ void PAHdivMassAssembleDiagonal2D(const int D1D,
               for (int qy = 0; qy < Q1D; ++qy)
               {
                  const double wy = (c == 1) ? Bc(qy,dy) : Bo(qy,dy);
-                  mass[qx] += wy*wy*((c == 0) ? op(qx,qy,0,e) : op(qx,qy,symmetric ? 2 : 3,e));
+                  mass[qx] += wy*wy*((c == 0) ? op(qx,qy,0,e) : op(qx,qy,2,e));
               }
            }

@@ -534,7 +283,6 @@ void PAHdivMassAssembleDiagonal2D(const int D1D,
 void PAHdivMassAssembleDiagonal3D(const int D1D,
                                  const int Q1D,
                                  const int NE,
-                                  const bool symmetric,
                                  const Array<double> &Bo_,
                                  const Array<double> &Bc_,
                                  const Vector &op_,
@@ -546,7 +294,7 @@ void PAHdivMassAssembleDiagonal3D(const int D1D,

   auto Bo = Reshape(Bo_.Read(), Q1D, D1D-1);
   auto Bc = Reshape(Bc_.Read(), Q1D, D1D);
-   auto op = Reshape(op_.Read(), Q1D, Q1D, Q1D, symmetric ? 6 : 9, NE);
+   auto op = Reshape(op_.Read(), Q1D, Q1D, Q1D, 6, NE);
   auto diag = Reshape(diag_.ReadWrite(), 3*(D1D-1)*(D1D-1)*D1D, NE);

   MFEM_FORALL(e, NE,
@@ -559,8 +307,7 @@ void PAHdivMassAssembleDiagonal3D(const int D1D,
         const int D1Dy = (c == 1) ? D1D : D1D - 1;
         const int D1Dx = (c == 0) ? D1D : D1D - 1;

-         const int opc = (c == 0) ? 0 : ((c == 1) ? (symmetric ? 3 : 4) :
-                                         (symmetric ? 5 : 8));
+         const int opc = (c == 0) ? 0 : ((c == 1) ? 3 : 5);

         double mass[HDIV_MAX_Q1D];

@@ -603,7 +350,6 @@ void PAHdivMassAssembleDiagonal3D(const int D1D,
 void PAHdivMassApply3D(const int D1D,
                       const int Q1D,
                       const int NE,
-                       const bool symmetric,
                       const Array<double> &Bo_,
                       const Array<double> &Bc_,
                       const Array<double> &Bot_,
@@ -620,7 +366,7 @@ void PAHdivMassApply3D(const int D1D,
   auto Bc = Reshape(Bc_.Read(), Q1D, D1D);
   auto Bot = Reshape(Bot_.Read(), D1D-1, Q1D);
   auto Bct = Reshape(Bct_.Read(), D1D, Q1D);
-   auto op = Reshape(op_.Read(), Q1D, Q1D, Q1D, symmetric ? 6 : 9, NE);
+   auto op = Reshape(op_.Read(), Q1D, Q1D, Q1D, 6, NE);
   auto x = Reshape(x_.Read(), 3*(D1D-1)*(D1D-1)*D1D, NE);
   auto y = Reshape(y_.ReadWrite(), 3*(D1D-1)*(D1D-1)*D1D, NE);

@@ -715,19 +461,15 @@ void PAHdivMassApply3D(const int D1D,
               const double O11 = op(qx,qy,qz,0,e);
               const double O12 = op(qx,qy,qz,1,e);
               const double O13 = op(qx,qy,qz,2,e);
-               const double O21 = symmetric ? O12 : op(qx,qy,qz,3,e);
-               const double O22 = symmetric ? op(qx,qy,qz,3,e) : op(qx,qy,qz,4,e);
-               const double O23 = symmetric ? op(qx,qy,qz,4,e) : op(qx,qy,qz,5,e);
-               const double O31 = symmetric ? O13 : op(qx,qy,qz,6,e);
-               const double O32 = symmetric ? O23 : op(qx,qy,qz,7,e);
-               const double O33 = symmetric ? op(qx,qy,qz,5,e) : op(qx,qy,qz,8,e);
-
+               const double O22 = op(qx,qy,qz,3,e);
+               const double O23 = op(qx,qy,qz,4,e);
+               const double O33 = op(qx,qy,qz,5,e);
               const double massX = mass[qz][qy][qx][0];
               const double massY = mass[qz][qy][qx][1];
               const double massZ = mass[qz][qy][qx][2];
               mass[qz][qy][qx][0] = (O11*massX)+(O12*massY)+(O13*massZ);
-               mass[qz][qy][qx][1] = (O21*massX)+(O22*massY)+(O23*massZ);
-               mass[qz][qy][qx][2] = (O31*massX)+(O32*massY)+(O33*massZ);
+               mass[qz][qy][qx][1] = (O12*massX)+(O22*massY)+(O23*massZ);
+               mass[qz][qy][qx][2] = (O13*massX)+(O23*massY)+(O33*massZ);
            }
         }
      }
@@ -795,337 +537,6 @@ void PAHdivMassApply3D(const int D1D,
   }); // end of element loop
 }

-template<int T_D1D = 0, int T_Q1D = 0>
-void SmemPAHdivMassApply3D(const int NE,
-                           const bool symmetric,
-                           const Array<double> &Bo_,
-                           const Array<double> &Bc_,
-                           const Array<double> &Bot_,
-                           const Array<double> &Bct_,
-                           const Vector &op_,
-                           const Vector &x_,
-                           Vector &y_,
-                           const int d1d = 0,
-                           const int q1d = 0)
-{
-   MFEM_CONTRACT_VAR(Bot_);
-   MFEM_CONTRACT_VAR(Bct_);
-
-   static constexpr int VDIM = 3;
-
-   const int D1D = T_D1D ? T_D1D : d1d;
-   const int Q1D = T_Q1D ? T_Q1D : q1d;
-
-   const auto bo = Reshape(Bo_.Read(), Q1D, D1D-1);
-   const auto bc = Reshape(Bc_.Read(), Q1D, D1D);
-   const auto D = Reshape(op_.Read(), Q1D, Q1D, Q1D, symmetric ? 6 : 9, NE);
-   const auto x = Reshape(x_.Read(), D1D*(D1D-1)*(D1D-1), VDIM, NE);
-   auto y = y_.ReadWrite();
-
-   MFEM_FORALL_3D(e, NE, Q1D, Q1D, VDIM,
-   {
-      const int tidz = MFEM_THREAD_ID(z);
-
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-
-      constexpr int MQ1 = T_Q1D ? T_Q1D : HDIV_MAX_Q1D;
-      constexpr int MD1 = T_D1D ? T_D1D : HDIV_MAX_D1D;
-      constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
-
-      MFEM_SHARED double smo[MQ1*(MD1-1)];
-      DeviceMatrix Bo(smo, D1D-1, Q1D);
-
-      MFEM_SHARED double smc[MQ1*MD1];
-      DeviceMatrix Bc(smc, D1D, Q1D);
-
-      MFEM_SHARED double sm0[VDIM*MDQ*MDQ*MDQ];
-      MFEM_SHARED double sm1[VDIM*MDQ*MDQ*MDQ];
-      DeviceMatrix X(sm0, D1D*(D1D-1)*(D1D-1), VDIM);
-      DeviceTensor<4> QDD(sm1, Q1D, D1D, D1D, VDIM);
-      DeviceTensor<4> QQD(sm0, Q1D, Q1D, D1D, VDIM);
-      DeviceTensor<4> QQQ(sm1, Q1D, Q1D, Q1D, VDIM);
-      DeviceTensor<4> DQQ(sm0, D1D, Q1D, Q1D, VDIM);
-      DeviceTensor<4> DDQ(sm1, D1D, D1D, Q1D, VDIM);
-
-      // Load X into shared memory
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         MFEM_FOREACH_THREAD(dz,y,D1D-1)
-         {
-            MFEM_FOREACH_THREAD(dy,x,D1D-1)
-            {
-               MFEM_UNROLL(MD1)
-               for (int dx = 0; dx < D1D; ++dx)
-               {
-                  X(dx+(dy+dz*(D1D-1))*D1D,vd) = x(dx+(dy+dz*(D1D-1))*D1D,vd,e);
-               }
-            }
-         }
-      }
-      // Load Bo and Bc into shared memory
-      if (tidz == 0)
-      {
-         MFEM_FOREACH_THREAD(d,y,D1D-1)
-         {
-            MFEM_FOREACH_THREAD(q,x,Q1D)
-            {
-               Bo(d,q) = bo(q,d);
-            }
-         }
-         MFEM_FOREACH_THREAD(d,y,D1D)
-         {
-            MFEM_FOREACH_THREAD(q,x,Q1D)
-            {
-               Bc(d,q) = bc(q,d);
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      // Apply B operator
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int nx = (vd == 0) ? D1D : D1D-1;
-         const int ny = (vd == 1) ? D1D : D1D-1;
-         const int nz = (vd == 2) ? D1D : D1D-1;
-         DeviceTensor<4> Xxyz(X, nx, ny, nz, VDIM);
-         DeviceMatrix Bx = (vd == 0) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(dy,y,ny)
-         {
-            MFEM_FOREACH_THREAD(qx,x,Q1D)
-            {
-               double u[D1D];
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < nz; ++dz) { u[dz] = 0.0; }
-               MFEM_UNROLL(MD1)
-               for (int dx = 0; dx < nx; ++dx)
-               {
-                  MFEM_UNROLL(MD1)
-                  for (int dz = 0; dz < nz; ++dz)
-                  {
-                     u[dz] += Xxyz(dx,dy,dz,vd) * Bx(dx,qx);
-                  }
-               }
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < nz; ++dz) { QDD(qx,dy,dz,vd) = u[dz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int ny = (vd == 1) ? D1D : D1D-1;
-         const int nz = (vd == 2) ? D1D : D1D-1;
-         DeviceMatrix By = (vd == 1) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            MFEM_FOREACH_THREAD(qx,x,Q1D)
-            {
-               double u[D1D];
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < nz; ++dz) { u[dz] = 0.0; }
-               MFEM_UNROLL(MD1)
-               for (int dy = 0; dy < ny; ++dy)
-               {
-                  MFEM_UNROLL(MD1)
-                  for (int dz = 0; dz < nz; ++dz)
-                  {
-                     u[dz] += QDD(qx,dy,dz,vd) * By(dy,qy);
-                  }
-               }
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < nz; ++dz) { QQD(qx,qy,dz,vd) = u[dz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int nz = (vd == 2) ? D1D : D1D-1;
-         DeviceMatrix Bz = (vd == 2) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            MFEM_FOREACH_THREAD(qx,x,Q1D)
-            {
-               double u[Q1D];
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz) { u[qz] = 0.0; }
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < nz; ++dz)
-               {
-                  MFEM_UNROLL(MQ1)
-                  for (int qz = 0; qz < Q1D; ++qz)
-                  {
-                     u[qz] += QQD(qx,qy,dz,vd) * Bz(dz,qz);
-                  }
-               }
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz) { QQQ(qx,qy,qz,vd) = u[qz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      // Apply D operator
-      if (tidz == 0)
-      {
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            MFEM_FOREACH_THREAD(qx,x,Q1D)
-            {
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz)
-               {
-                  const double Qx = QQQ(qx,qy,qz,0);
-                  const double Qy = QQQ(qx,qy,qz,1);
-                  const double Qz = QQQ(qx,qy,qz,2);
-
-                  const double D11 = D(qx,qy,qz,0,e);
-                  const double D12 = D(qx,qy,qz,1,e);
-                  const double D13 = D(qx,qy,qz,2,e);
-                  const double D21 = symmetric ? D12 : D(qx,qy,qz,3,e);
-                  const double D22 = symmetric ? D(qx,qy,qz,3,e) : D(qx,qy,qz,4,e);
-                  const double D23 = symmetric ? D(qx,qy,qz,4,e) : D(qx,qy,qz,5,e);
-                  const double D31 = symmetric ? D13 : D(qx,qy,qz,6,e);
-                  const double D32 = symmetric ? D23 : D(qx,qy,qz,7,e);
-                  const double D33 = symmetric ? D(qx,qy,qz,5,e) : D(qx,qy,qz,8,e);
-
-                  QQQ(qx,qy,qz,0) = D11*Qx + D12*Qy + D13*Qz;
-                  QQQ(qx,qy,qz,1) = D21*Qx + D22*Qy + D23*Qz;
-                  QQQ(qx,qy,qz,2) = D31*Qx + D32*Qy + D33*Qz;
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      // Apply Bt operator
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int nx = (vd == 0) ? D1D : D1D-1;
-         DeviceMatrix Btx = (vd == 0) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(qy,y,Q1D)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               double u[Q1D];
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz) { u[qz] = 0.0; }
-               MFEM_UNROLL(MQ1)
-               for (int qx = 0; qx < Q1D; ++qx)
-               {
-                  MFEM_UNROLL(MQ1)
-                  for (int qz = 0; qz < Q1D; ++qz)
-                  {
-                     u[qz] += QQQ(qx,qy,qz,vd) * Btx(dx,qx);
-                  }
-               }
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz) { DQQ(dx,qy,qz,vd) = u[qz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int nx = (vd == 0) ? D1D : D1D-1;
-         const int ny = (vd == 1) ? D1D : D1D-1;
-         DeviceMatrix Bty = (vd == 1) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(dy,y,ny)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               double u[Q1D];
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz) { u[qz] = 0.0; }
-               MFEM_UNROLL(MQ1)
-               for (int qy = 0; qy < Q1D; ++qy)
-               {
-                  MFEM_UNROLL(MQ1)
-                  for (int qz = 0; qz < Q1D; ++qz)
-                  {
-                     u[qz] += DQQ(dx,qy,qz,vd) * Bty(dy,qy);
-                  }
-               }
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz) { DDQ(dx,dy,qz,vd) = u[qz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,VDIM)
-      {
-         const int nx = (vd == 0) ? D1D : D1D-1;
-         const int ny = (vd == 1) ? D1D : D1D-1;
-         const int nz = (vd == 2) ? D1D : D1D-1;
-         DeviceTensor<5> Yxyz(y, nx, ny, nz, VDIM, NE);
-         DeviceMatrix Btz = (vd == 2) ? Bc : Bo;
-         MFEM_FOREACH_THREAD(dy,y,ny)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               double u[D1D];
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < nz; ++dz) { u[dz] = 0.0; }
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz)
-               {
-                  MFEM_UNROLL(MD1)
-                  for (int dz = 0; dz < nz; ++dz)
-                  {
-                     u[dz] += DDQ(dx,dy,qz,vd) * Btz(dz,qz);
-                  }
-               }
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < nz; ++dz) { Yxyz(dx,dy,dz,vd,e) += u[dz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-   });
-}
-
-void PAHdivMassApply(const int dim,
-                     const int D1D,
-                     const int Q1D,
-                     const int NE,
-                     const bool symmetric,
-                     const Array<double> &Bo,
-                     const Array<double> &Bc,
-                     const Array<double> &Bot,
-                     const Array<double> &Bct,
-                     const Vector &op,
-                     const Vector &x,
-                     Vector &y)
-{
-   const int id = (D1D << 4) | Q1D;
-
-   if (dim == 2)
-   {
-      switch (id)
-      {
-         case 0x22: return SmemPAHdivMassApply2D<2,2>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         case 0x33: return SmemPAHdivMassApply2D<3,3>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         case 0x44: return SmemPAHdivMassApply2D<4,4>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         case 0x55: return SmemPAHdivMassApply2D<5,5>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         default: // fallback
-            return PAHdivMassApply2D(D1D,Q1D,NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-      }
-   }
-   else if (dim == 3)
-   {
-      switch (id)
-      {
-         case 0x23: return SmemPAHdivMassApply3D<2,3>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         case 0x34: return SmemPAHdivMassApply3D<3,4>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         case 0x45: return SmemPAHdivMassApply3D<4,5>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         case 0x56: return SmemPAHdivMassApply3D<5,6>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         case 0x67: return SmemPAHdivMassApply3D<6,7>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         case 0x78: return SmemPAHdivMassApply3D<7,8>(NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-         default: // fallback
-            return PAHdivMassApply3D(D1D,Q1D,NE,symmetric,Bo,Bc,Bot,Bct,op,x,y);
-      }
-   }
-}
-
 // PA H(div) div-div assemble 2D kernel
 // NOTE: this is identical to PACurlCurlSetup3D
 static void PADivDivSetup2D(const int Q1D,
@@ -1215,7 +626,7 @@ static void PADivDivApply2D(const int D1D,
   {
      double div[MAX_Q1D][MAX_Q1D];

-      // div[qy][qx] will be computed as du_x/dx + du_y/dy
+      // div[qy][qx] will be computed as du_x/dx + duy_/dy

      for (int qy = 0; qy < Q1D; ++qy)
      {
@@ -1514,8 +925,19 @@ void DivDivIntegrator::AssemblePA(const FiniteElementSpace &fes)

   pa_data.SetSize(nq * ne, Device::GetMemoryType());

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(Q, qs, CoefficientStorage::FULL);
+   Vector coeff(ne * nq);
+   coeff = 1.0;
+   if (Q)
+   {
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            coeff[p + (e * nq)] = Q->Eval(*tr, ir->IntPoint(p));
+         }
+      }
+   }

   if (el->GetDerivType() == mfem::FiniteElement::DIV && dim == 3)
   {
@@ -1773,14 +1195,18 @@ VectorFEDivergenceIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,

   pa_data.SetSize(nq * ne, Device::GetMemoryType());

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(Q, qs, CoefficientStorage::FULL);
-
-   if (test_el->GetMapType() == FiniteElement::INTEGRAL)
+   Vector coeff(ne * nq);
+   coeff = 1.0;
+   if (Q)
   {
-      const GeometricFactors *geom =
-         mesh->GetGeometricFactors(*ir, GeometricFactors::DETERMINANTS);
-      coeff /= geom->detJ;
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            coeff[p + (e * nq)] = Q->Eval(*tr, ir->IntPoint(p));
+         }
+      }
   }

   if (trial_el->GetDerivType() == mfem::FiniteElement::DIV && dim == 3)
@@ -1797,7 +1223,7 @@ VectorFEDivergenceIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
   }
 }

-// Apply to x corresponding to DOFs in H(div) (trial), whose divergence is
+// Apply to x corresponding to DOF's in H(div) (trial), whose divergence is
 // integrated against L_2 test functions corresponding to y.
 static void PAHdivL2Apply3D(const int D1D,
                            const int Q1D,
@@ -1960,7 +1386,7 @@ static void PAHdivL2Apply3D(const int D1D,
   }); // end of element loop
 }

-// Apply to x corresponding to DOFs in H(div) (trial), whose divergence is
+// Apply to x corresponding to DOF's in H(div) (trial), whose divergence is
 // integrated against L_2 test functions corresponding to y.
 static void PAHdivL2Apply2D(const int D1D,
                            const int Q1D,
@@ -31,16 +31,7 @@ void MassIntegrator::AssembleMF(const FiniteElementSpace &fes)
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedMFMassIntegrator(*this, fes, Q);
-      }
-      else
-      {
-         ceedOp = new ceed::MFMassIntegrator(fes, *ir, Q);
-      }
+      ceedOp = new ceed::MFMassIntegrator(fes, *ir, Q);
      return;
   }
   MFEM_ABORT("Error: MassIntegrator::AssembleMF only implemented with"
@@ -12,9 +12,7 @@
 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
-#include "qfunction.hpp"
 #include "ceed/integrators/mass/mass.hpp"
-#include "bilininteg_mass_pa.hpp"

 using namespace std;

@@ -40,16 +38,7 @@ void MassIntegrator::AssemblePA(const FiniteElementSpace &fes)
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedPAMassIntegrator(*this, fes, Q);
-      }
-      else
-      {
-         ceedOp = new ceed::PAMassIntegrator(fes, *ir, Q);
-      }
+      ceedOp = new ceed::PAMassIntegrator(fes, *ir, Q);
      return;
   }
   int map_type = el.GetMapType();
@@ -62,10 +51,43 @@ void MassIntegrator::AssemblePA(const FiniteElementSpace &fes)
   dofs1D = maps->ndof;
   quad1D = maps->nqpt;
   pa_data.SetSize(ne*nq, mt);
+   Vector coeff;
+   if (Q == nullptr)
+   {
+      coeff.SetSize(1);
+      coeff(0) = 1.0;
+   }
+   else if (ConstantCoefficient* cQ = dynamic_cast<ConstantCoefficient*>(Q))
+   {
+      coeff.SetSize(1);
+      coeff(0) = cQ->constant;
+   }
+   else if (QuadratureFunctionCoefficient* qfQ =
+               dynamic_cast<QuadratureFunctionCoefficient*>(Q))
+   {
+      const QuadratureFunction &qFun = qfQ->GetQuadFunction();
+      MFEM_VERIFY(qFun.Size() == nq * ne,
+                  "Incompatible QuadratureFunction dimension \n");

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(Q, qs, CoefficientStorage::COMPRESSED);
-
+      MFEM_VERIFY(ir == &qFun.GetSpace()->GetElementIntRule(0),
+                  "IntegrationRule used within integrator and in"
+                  " QuadratureFunction appear to be different");
+      qFun.Read();
+      coeff.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
+   }
+   else
+   {
+      coeff.SetSize(nq * ne);
+      auto C = Reshape(coeff.HostWrite(), nq, ne);
+      for (int e = 0; e < ne; ++e)
+      {
+         ElementTransformation& T = *fes.GetElementTransformation(e);
+         for (int q = 0; q < nq; ++q)
+         {
+            C(q,e) = Q->Eval(T, ir->IntPoint(q));
+         }
+      }
+   }
   if (dim==1) { MFEM_ABORT("Not supported yet... stay tuned!"); }
   if (dim==2)
   {
@@ -559,18 +581,85 @@ static void PAMassApply2D(const int NE,
                          const int d1d = 0,
                          const int q1d = 0)
 {
-   MFEM_VERIFY(T_D1D ? T_D1D : d1d <= MAX_D1D, "");
-   MFEM_VERIFY(T_Q1D ? T_Q1D : q1d <= MAX_Q1D, "");
-
-   const auto B = b_.Read();
-   const auto Bt = bt_.Read();
-   const auto D = d_.Read();
-   const auto X = x_.Read();
-   auto Y = y_.ReadWrite();
-
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   MFEM_VERIFY(D1D <= MAX_D1D, "");
+   MFEM_VERIFY(Q1D <= MAX_Q1D, "");
+   auto B = Reshape(b_.Read(), Q1D, D1D);
+   auto Bt = Reshape(bt_.Read(), D1D, Q1D);
+   auto D = Reshape(d_.Read(), Q1D, Q1D, NE);
+   auto X = Reshape(x_.Read(), D1D, D1D, NE);
+   auto Y = Reshape(y_.ReadWrite(), D1D, D1D, NE);
   MFEM_FORALL(e, NE,
   {
-      internal::PAMassApply2D_Element(e, NE, B, Bt, D, X, Y, d1d, q1d);
+      const int D1D = T_D1D ? T_D1D : d1d; // nvcc workaround
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      // the following variables are evaluated at compile time
+      constexpr int max_D1D = T_D1D ? T_D1D : MAX_D1D;
+      constexpr int max_Q1D = T_Q1D ? T_Q1D : MAX_Q1D;
+      double sol_xy[max_Q1D][max_Q1D];
+      for (int qy = 0; qy < Q1D; ++qy)
+      {
+         for (int qx = 0; qx < Q1D; ++qx)
+         {
+            sol_xy[qy][qx] = 0.0;
+         }
+      }
+      for (int dy = 0; dy < D1D; ++dy)
+      {
+         double sol_x[max_Q1D];
+         for (int qy = 0; qy < Q1D; ++qy)
+         {
+            sol_x[qy] = 0.0;
+         }
+         for (int dx = 0; dx < D1D; ++dx)
+         {
+            const double s = X(dx,dy,e);
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               sol_x[qx] += B(qx,dx)* s;
+            }
+         }
+         for (int qy = 0; qy < Q1D; ++qy)
+         {
+            const double d2q = B(qy,dy);
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               sol_xy[qy][qx] += d2q * sol_x[qx];
+            }
+         }
+      }
+      for (int qy = 0; qy < Q1D; ++qy)
+      {
+         for (int qx = 0; qx < Q1D; ++qx)
+         {
+            sol_xy[qy][qx] *= D(qx,qy,e);
+         }
+      }
+      for (int qy = 0; qy < Q1D; ++qy)
+      {
+         double sol_x[max_D1D];
+         for (int dx = 0; dx < D1D; ++dx)
+         {
+            sol_x[dx] = 0.0;
+         }
+         for (int qx = 0; qx < Q1D; ++qx)
+         {
+            const double s = sol_xy[qy][qx];
+            for (int dx = 0; dx < D1D; ++dx)
+            {
+               sol_x[dx] += Bt(dx,qx) * s;
+            }
+         }
+         for (int dy = 0; dy < D1D; ++dy)
+         {
+            const double q2d = Bt(dy,qy);
+            for (int dx = 0; dx < D1D; ++dx)
+            {
+               Y(dx,dy,e) += q2d * sol_x[dx];
+            }
+         }
+      }
   });
 }

@@ -592,13 +681,108 @@ static void SmemPAMassApply2D(const int NE,
   constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
   MFEM_VERIFY(D1D <= MD1, "");
   MFEM_VERIFY(Q1D <= MQ1, "");
-   const auto b = b_.Read();
-   const auto D = d_.Read();
-   const auto x = x_.Read();
-   auto Y = y_.ReadWrite();
+   auto b = Reshape(b_.Read(), Q1D, D1D);
+   auto D = Reshape(d_.Read(), Q1D, Q1D, NE);
+   auto x = Reshape(x_.Read(), D1D, D1D, NE);
+   auto Y = Reshape(y_.ReadWrite(), D1D, D1D, NE);
   MFEM_FORALL_2D(e, NE, Q1D, Q1D, NBZ,
   {
-      internal::SmemPAMassApply2D_Element<T_D1D,T_Q1D,T_NBZ>(e, NE, b, D, x, Y, d1d, q1d);
+      const int tidz = MFEM_THREAD_ID(z);
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      constexpr int NBZ = T_NBZ ? T_NBZ : 1;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+      constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
+      MFEM_SHARED double BBt[MQ1*MD1];
+      double (*B)[MD1] = (double (*)[MD1]) BBt;
+      double (*Bt)[MQ1] = (double (*)[MQ1]) BBt;
+      MFEM_SHARED double sm0[NBZ][MDQ*MDQ];
+      MFEM_SHARED double sm1[NBZ][MDQ*MDQ];
+      double (*X)[MD1] = (double (*)[MD1]) (sm0 + tidz);
+      double (*DQ)[MQ1] = (double (*)[MQ1]) (sm1 + tidz);
+      double (*QQ)[MQ1] = (double (*)[MQ1]) (sm0 + tidz);
+      double (*QD)[MD1] = (double (*)[MD1]) (sm1 + tidz);
+      MFEM_FOREACH_THREAD(dy,y,D1D)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            X[dy][dx] = x(dx,dy,e);
+         }
+      }
+      if (tidz == 0)
+      {
+         MFEM_FOREACH_THREAD(dy,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(q,x,Q1D)
+            {
+               B[q][dy] = b(q,dy);
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(dy,y,D1D)
+      {
+         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         {
+            double dq = 0.0;
+            for (int dx = 0; dx < D1D; ++dx)
+            {
+               dq += X[dy][dx] * B[qx][dx];
+            }
+            DQ[dy][qx] = dq;
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      {
+         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         {
+            double qq = 0.0;
+            for (int dy = 0; dy < D1D; ++dy)
+            {
+               qq += DQ[dy][qx] * B[qy][dy];
+            }
+            QQ[qy][qx] = qq * D(qx, qy, e);
+         }
+      }
+      MFEM_SYNC_THREAD;
+      if (tidz == 0)
+      {
+         MFEM_FOREACH_THREAD(dy,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(q,x,Q1D)
+            {
+               Bt[dy][q] = b(q,dy);
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            double dq = 0.0;
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               dq += QQ[qy][qx] * Bt[dx][qx];
+            }
+            QD[qy][dx] = dq;
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(dy,y,D1D)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            double dd = 0.0;
+            for (int qy = 0; qy < Q1D; ++qy)
+            {
+               dd += (QD[qy][dx] * Bt[dy][qy]);
+            }
+            Y(dx, dy, e) += dd;
+         }
+      }
   });
 }

@@ -612,18 +796,134 @@ static void PAMassApply3D(const int NE,
                          const int d1d = 0,
                          const int q1d = 0)
 {
-   MFEM_VERIFY(T_D1D ? T_D1D : d1d <= MAX_D1D, "");
-   MFEM_VERIFY(T_Q1D ? T_Q1D : q1d <= MAX_Q1D, "");
-
-   const auto B = b_.Read();
-   const auto Bt = bt_.Read();
-   const auto D = d_.Read();
-   const auto X = x_.Read();
-   auto Y = y_.ReadWrite();
-
+   const int D1D = T_D1D ? T_D1D : d1d;
+   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   MFEM_VERIFY(D1D <= MAX_D1D, "");
+   MFEM_VERIFY(Q1D <= MAX_Q1D, "");
+   auto B = Reshape(b_.Read(), Q1D, D1D);
+   auto Bt = Reshape(bt_.Read(), D1D, Q1D);
+   auto D = Reshape(d_.Read(), Q1D, Q1D, Q1D, NE);
+   auto X = Reshape(x_.Read(), D1D, D1D, D1D, NE);
+   auto Y = Reshape(y_.ReadWrite(), D1D, D1D, D1D, NE);
   MFEM_FORALL(e, NE,
   {
-      internal::PAMassApply3D_Element(e, NE, B, Bt, D, X, Y, d1d, q1d);
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      constexpr int max_D1D = T_D1D ? T_D1D : MAX_D1D;
+      constexpr int max_Q1D = T_Q1D ? T_Q1D : MAX_Q1D;
+      double sol_xyz[max_Q1D][max_Q1D][max_Q1D];
+      for (int qz = 0; qz < Q1D; ++qz)
+      {
+         for (int qy = 0; qy < Q1D; ++qy)
+         {
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               sol_xyz[qz][qy][qx] = 0.0;
+            }
+         }
+      }
+      for (int dz = 0; dz < D1D; ++dz)
+      {
+         double sol_xy[max_Q1D][max_Q1D];
+         for (int qy = 0; qy < Q1D; ++qy)
+         {
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               sol_xy[qy][qx] = 0.0;
+            }
+         }
+         for (int dy = 0; dy < D1D; ++dy)
+         {
+            double sol_x[max_Q1D];
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               sol_x[qx] = 0;
+            }
+            for (int dx = 0; dx < D1D; ++dx)
+            {
+               const double s = X(dx,dy,dz,e);
+               for (int qx = 0; qx < Q1D; ++qx)
+               {
+                  sol_x[qx] += B(qx,dx) * s;
+               }
+            }
+            for (int qy = 0; qy < Q1D; ++qy)
+            {
+               const double wy = B(qy,dy);
+               for (int qx = 0; qx < Q1D; ++qx)
+               {
+                  sol_xy[qy][qx] += wy * sol_x[qx];
+               }
+            }
+         }
+         for (int qz = 0; qz < Q1D; ++qz)
+         {
+            const double wz = B(qz,dz);
+            for (int qy = 0; qy < Q1D; ++qy)
+            {
+               for (int qx = 0; qx < Q1D; ++qx)
+               {
+                  sol_xyz[qz][qy][qx] += wz * sol_xy[qy][qx];
+               }
+            }
+         }
+      }
+      for (int qz = 0; qz < Q1D; ++qz)
+      {
+         for (int qy = 0; qy < Q1D; ++qy)
+         {
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               sol_xyz[qz][qy][qx] *= D(qx,qy,qz,e);
+            }
+         }
+      }
+      for (int qz = 0; qz < Q1D; ++qz)
+      {
+         double sol_xy[max_D1D][max_D1D];
+         for (int dy = 0; dy < D1D; ++dy)
+         {
+            for (int dx = 0; dx < D1D; ++dx)
+            {
+               sol_xy[dy][dx] = 0;
+            }
+         }
+         for (int qy = 0; qy < Q1D; ++qy)
+         {
+            double sol_x[max_D1D];
+            for (int dx = 0; dx < D1D; ++dx)
+            {
+               sol_x[dx] = 0;
+            }
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               const double s = sol_xyz[qz][qy][qx];
+               for (int dx = 0; dx < D1D; ++dx)
+               {
+                  sol_x[dx] += Bt(dx,qx) * s;
+               }
+            }
+            for (int dy = 0; dy < D1D; ++dy)
+            {
+               const double wy = Bt(dy,qy);
+               for (int dx = 0; dx < D1D; ++dx)
+               {
+                  sol_xy[dy][dx] += wy * sol_x[dx];
+               }
+            }
+         }
+         for (int dz = 0; dz < D1D; ++dz)
+         {
+            const double wz = Bt(dz,qz);
+            for (int dy = 0; dy < D1D; ++dy)
+            {
+               for (int dx = 0; dx < D1D; ++dx)
+               {
+                  Y(dx,dy,dz,e) += wz * sol_xy[dy][dx];
+               }
+            }
+         }
+      }
   });
 }

@@ -644,13 +944,213 @@ static void SmemPAMassApply3D(const int NE,
   constexpr int M1D = T_D1D ? T_D1D : MAX_D1D;
   MFEM_VERIFY(D1D <= M1D, "");
   MFEM_VERIFY(Q1D <= M1Q, "");
-   auto b = b_.Read();
-   auto d = d_.Read();
-   auto x = x_.Read();
-   auto y = y_.ReadWrite();
+   auto b = Reshape(b_.Read(), Q1D, D1D);
+   auto d = Reshape(d_.Read(), Q1D, Q1D, Q1D, NE);
+   auto x = Reshape(x_.Read(), D1D, D1D, D1D, NE);
+   auto y = Reshape(y_.ReadWrite(), D1D, D1D, D1D, NE);
   MFEM_FORALL_3D(e, NE, Q1D, Q1D, 1,
   {
-      internal::SmemPAMassApply3D_Element<T_D1D,T_Q1D>(e, NE, b, d, x, y, d1d, q1d);
+      const int D1D = T_D1D ? T_D1D : d1d;
+      const int Q1D = T_Q1D ? T_Q1D : q1d;
+      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+      constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
+      MFEM_SHARED double sDQ[MQ1*MD1];
+      double (*B)[MD1] = (double (*)[MD1]) sDQ;
+      double (*Bt)[MQ1] = (double (*)[MQ1]) sDQ;
+      MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
+      MFEM_SHARED double sm1[MDQ*MDQ*MDQ];
+      double (*X)[MD1][MD1]   = (double (*)[MD1][MD1]) sm0;
+      double (*DDQ)[MD1][MQ1] = (double (*)[MD1][MQ1]) sm1;
+      double (*DQQ)[MQ1][MQ1] = (double (*)[MQ1][MQ1]) sm0;
+      double (*QQQ)[MQ1][MQ1] = (double (*)[MQ1][MQ1]) sm1;
+      double (*QQD)[MQ1][MD1] = (double (*)[MQ1][MD1]) sm0;
+      double (*QDD)[MD1][MD1] = (double (*)[MD1][MD1]) sm1;
+      MFEM_FOREACH_THREAD(dy,y,D1D)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            MFEM_UNROLL(MD1)
+            for (int dz = 0; dz < D1D; ++dz)
+            {
+               X[dz][dy][dx] = x(dx,dy,dz,e);
+            }
+         }
+         MFEM_FOREACH_THREAD(dx,x,Q1D)
+         {
+            B[dx][dy] = b(dx,dy);
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(dy,y,D1D)
+      {
+         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         {
+            double u[D1D];
+            MFEM_UNROLL(MD1)
+            for (int dz = 0; dz < D1D; dz++)
+            {
+               u[dz] = 0;
+            }
+            MFEM_UNROLL(MD1)
+            for (int dx = 0; dx < D1D; ++dx)
+            {
+               MFEM_UNROLL(MD1)
+               for (int dz = 0; dz < D1D; ++dz)
+               {
+                  u[dz] += X[dz][dy][dx] * B[qx][dx];
+               }
+            }
+            MFEM_UNROLL(MD1)
+            for (int dz = 0; dz < D1D; ++dz)
+            {
+               DDQ[dz][dy][qx] = u[dz];
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      {
+         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         {
+            double u[D1D];
+            MFEM_UNROLL(MD1)
+            for (int dz = 0; dz < D1D; dz++)
+            {
+               u[dz] = 0;
+            }
+            MFEM_UNROLL(MD1)
+            for (int dy = 0; dy < D1D; ++dy)
+            {
+               MFEM_UNROLL(MD1)
+               for (int dz = 0; dz < D1D; dz++)
+               {
+                  u[dz] += DDQ[dz][dy][qx] * B[qy][dy];
+               }
+            }
+            MFEM_UNROLL(MD1)
+            for (int dz = 0; dz < D1D; dz++)
+            {
+               DQQ[dz][qy][qx] = u[dz];
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      {
+         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         {
+            double u[Q1D];
+            MFEM_UNROLL(MQ1)
+            for (int qz = 0; qz < Q1D; qz++)
+            {
+               u[qz] = 0;
+            }
+            MFEM_UNROLL(MD1)
+            for (int dz = 0; dz < D1D; ++dz)
+            {
+               MFEM_UNROLL(MQ1)
+               for (int qz = 0; qz < Q1D; qz++)
+               {
+                  u[qz] += DQQ[dz][qy][qx] * B[qz][dz];
+               }
+            }
+            MFEM_UNROLL(MQ1)
+            for (int qz = 0; qz < Q1D; qz++)
+            {
+               QQQ[qz][qy][qx] = u[qz] * d(qx,qy,qz,e);
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(d,y,D1D)
+      {
+         MFEM_FOREACH_THREAD(q,x,Q1D)
+         {
+            Bt[d][q] = b(q,d);
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            double u[Q1D];
+            MFEM_UNROLL(MQ1)
+            for (int qz = 0; qz < Q1D; ++qz)
+            {
+               u[qz] = 0;
+            }
+            MFEM_UNROLL(MQ1)
+            for (int qx = 0; qx < Q1D; ++qx)
+            {
+               MFEM_UNROLL(MQ1)
+               for (int qz = 0; qz < Q1D; ++qz)
+               {
+                  u[qz] += QQQ[qz][qy][qx] * Bt[dx][qx];
+               }
+            }
+            MFEM_UNROLL(MQ1)
+            for (int qz = 0; qz < Q1D; ++qz)
+            {
+               QQD[qz][qy][dx] = u[qz];
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(dy,y,D1D)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            double u[Q1D];
+            MFEM_UNROLL(MQ1)
+            for (int qz = 0; qz < Q1D; ++qz)
+            {
+               u[qz] = 0;
+            }
+            MFEM_UNROLL(MQ1)
+            for (int qy = 0; qy < Q1D; ++qy)
+            {
+               MFEM_UNROLL(MQ1)
+               for (int qz = 0; qz < Q1D; ++qz)
+               {
+                  u[qz] += QQD[qz][qy][dx] * Bt[dy][qy];
+               }
+            }
+            MFEM_UNROLL(MQ1)
+            for (int qz = 0; qz < Q1D; ++qz)
+            {
+               QDD[qz][dy][dx] = u[qz];
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(dy,y,D1D)
+      {
+         MFEM_FOREACH_THREAD(dx,x,D1D)
+         {
+            double u[D1D];
+            MFEM_UNROLL(MD1)
+            for (int dz = 0; dz < D1D; ++dz)
+            {
+               u[dz] = 0;
+            }
+            MFEM_UNROLL(MQ1)
+            for (int qz = 0; qz < Q1D; ++qz)
+            {
+               MFEM_UNROLL(MD1)
+               for (int dz = 0; dz < D1D; ++dz)
+               {
+                  u[dz] += QDD[qz][dy][dx] * Bt[dz][qz];
+               }
+            }
+            MFEM_UNROLL(MD1)
+            for (int dz = 0; dz < D1D; ++dz)
+            {
+               y(dx,dy,dz,e) += u[dz];
+            }
+         }
+      }
   });
 }

@@ -1,632 +0,0 @@
-// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
-// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-// LICENSE and NOTICE for details. LLNL-CODE-806117.
-//
-// This file is part of the MFEM library. For more information and source code
-// availability visit https://mfem.org.
-//
-// MFEM is free software; you can redistribute it and/or modify it under the
-// terms of the BSD-3 license. We welcome feedback and contributions, see file
-// CONTRIBUTING.md for details.
-
-#ifndef MFEM_BILININTEG_MASS_PA_HPP
-#define MFEM_BILININTEG_MASS_PA_HPP
-
-#include "../config/config.hpp"
-#include "../general/forall.hpp"
-#include "../linalg/dtensor.hpp"
-
-namespace mfem
-{
-
-namespace internal
-{
-
-template <bool ACCUMULATE = true>
-MFEM_HOST_DEVICE inline
-void PAMassApply2D_Element(const int e,
-                           const int NE,
-                           const double *b_,
-                           const double *bt_,
-                           const double *d_,
-                           const double *x_,
-                           double *y_,
-                           const int d1d = 0,
-                           const int q1d = 0)
-{
-   const int D1D = d1d;
-   const int Q1D = q1d;
-   auto B = ConstDeviceMatrix(b_, Q1D, D1D);
-   auto Bt = ConstDeviceMatrix(bt_, D1D, Q1D);
-   auto D = ConstDeviceCube(d_, Q1D, Q1D, NE);
-   auto X = ConstDeviceCube(x_, D1D, D1D, NE);
-   auto Y = DeviceCube(y_, D1D, D1D, NE);
-
-   if (!ACCUMULATE)
-   {
-      for (int dy = 0; dy < D1D; ++dy)
-      {
-         for (int dx = 0; dx < D1D; ++dx)
-         {
-            Y(dx, dy, e) = 0.0;
-         }
-      }
-   }
-
-   constexpr int max_D1D = MAX_D1D;
-   constexpr int max_Q1D = MAX_Q1D;
-   double sol_xy[max_Q1D][max_Q1D];
-   for (int qy = 0; qy < Q1D; ++qy)
-   {
-      for (int qx = 0; qx < Q1D; ++qx)
-      {
-         sol_xy[qy][qx] = 0.0;
-      }
-   }
-   for (int dy = 0; dy < D1D; ++dy)
-   {
-      double sol_x[max_Q1D];
-      for (int qy = 0; qy < Q1D; ++qy)
-      {
-         sol_x[qy] = 0.0;
-      }
-      for (int dx = 0; dx < D1D; ++dx)
-      {
-         const double s = X(dx,dy,e);
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            sol_x[qx] += B(qx,dx)* s;
-         }
-      }
-      for (int qy = 0; qy < Q1D; ++qy)
-      {
-         const double d2q = B(qy,dy);
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            sol_xy[qy][qx] += d2q * sol_x[qx];
-         }
-      }
-   }
-   for (int qy = 0; qy < Q1D; ++qy)
-   {
-      for (int qx = 0; qx < Q1D; ++qx)
-      {
-         sol_xy[qy][qx] *= D(qx,qy,e);
-      }
-   }
-   for (int qy = 0; qy < Q1D; ++qy)
-   {
-      double sol_x[max_D1D];
-      for (int dx = 0; dx < D1D; ++dx)
-      {
-         sol_x[dx] = 0.0;
-      }
-      for (int qx = 0; qx < Q1D; ++qx)
-      {
-         const double s = sol_xy[qy][qx];
-         for (int dx = 0; dx < D1D; ++dx)
-         {
-            sol_x[dx] += Bt(dx,qx) * s;
-         }
-      }
-      for (int dy = 0; dy < D1D; ++dy)
-      {
-         const double q2d = Bt(dy,qy);
-         for (int dx = 0; dx < D1D; ++dx)
-         {
-            Y(dx,dy,e) += q2d * sol_x[dx];
-         }
-      }
-   }
-}
-
-template<int T_D1D, int T_Q1D, int T_NBZ, bool ACCUMULATE = true>
-MFEM_HOST_DEVICE inline
-void SmemPAMassApply2D_Element(const int e,
-                               const int NE,
-                               const double *b_,
-                               const double *d_,
-                               const double *x_,
-                               double *y_,
-                               int d1d = 0,
-                               int q1d = 0)
-{
-   const int D1D = T_D1D ? T_D1D : d1d;
-   const int Q1D = T_Q1D ? T_Q1D : q1d;
-   constexpr int NBZ = T_NBZ ? T_NBZ : 1;
-
-   constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-   constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-   constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
-
-   auto b = ConstDeviceMatrix(b_, Q1D, D1D);
-   auto D = ConstDeviceCube(d_, Q1D, Q1D, NE);
-   auto x = ConstDeviceCube(x_, D1D, D1D, NE);
-   auto Y = DeviceCube(y_, D1D, D1D, NE);
-
-   const int tidz = MFEM_THREAD_ID(z);
-
-   MFEM_SHARED double BBt[MQ1*MD1];
-   double (*B)[MD1] = (double (*)[MD1]) BBt;
-   double (*Bt)[MQ1] = (double (*)[MQ1]) BBt;
-   MFEM_SHARED double sm0[NBZ][MDQ*MDQ];
-   MFEM_SHARED double sm1[NBZ][MDQ*MDQ];
-   double (*X)[MD1] = (double (*)[MD1]) (sm0 + tidz);
-   double (*DQ)[MQ1] = (double (*)[MQ1]) (sm1 + tidz);
-   double (*QQ)[MQ1] = (double (*)[MQ1]) (sm0 + tidz);
-   double (*QD)[MD1] = (double (*)[MD1]) (sm1 + tidz);
-
-
-   MFEM_FOREACH_THREAD(dy,y,D1D)
-   {
-      MFEM_FOREACH_THREAD(dx,x,D1D)
-      {
-         X[dy][dx] = x(dx,dy,e);
-      }
-   }
-   if (tidz == 0)
-   {
-      MFEM_FOREACH_THREAD(dy,y,D1D)
-      {
-         MFEM_FOREACH_THREAD(q,x,Q1D)
-         {
-            B[q][dy] = b(q,dy);
-         }
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(dy,y,D1D)
-   {
-      MFEM_FOREACH_THREAD(qx,x,Q1D)
-      {
-         double dq = 0.0;
-         for (int dx = 0; dx < D1D; ++dx)
-         {
-            dq += X[dy][dx] * B[qx][dx];
-         }
-         DQ[dy][qx] = dq;
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(qy,y,Q1D)
-   {
-      MFEM_FOREACH_THREAD(qx,x,Q1D)
-      {
-         double qq = 0.0;
-         for (int dy = 0; dy < D1D; ++dy)
-         {
-            qq += DQ[dy][qx] * B[qy][dy];
-         }
-         QQ[qy][qx] = qq * D(qx, qy, e);
-      }
-   }
-   MFEM_SYNC_THREAD;
-   if (tidz == 0)
-   {
-      MFEM_FOREACH_THREAD(dy,y,D1D)
-      {
-         MFEM_FOREACH_THREAD(q,x,Q1D)
-         {
-            Bt[dy][q] = b(q,dy);
-         }
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(qy,y,Q1D)
-   {
-      MFEM_FOREACH_THREAD(dx,x,D1D)
-      {
-         double dq = 0.0;
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            dq += QQ[qy][qx] * Bt[dx][qx];
-         }
-         QD[qy][dx] = dq;
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(dy,y,D1D)
-   {
-      MFEM_FOREACH_THREAD(dx,x,D1D)
-      {
-         double dd = 0.0;
-         for (int qy = 0; qy < Q1D; ++qy)
-         {
-            dd += (QD[qy][dx] * Bt[dy][qy]);
-         }
-         if (ACCUMULATE)
-         {
-            Y(dx, dy, e) += dd;
-         }
-         else
-         {
-            Y(dx, dy, e) = dd;
-         }
-      }
-   }
-}
-
-template <bool ACCUMULATE = true>
-MFEM_HOST_DEVICE inline
-void PAMassApply3D_Element(const int e,
-                           const int NE,
-                           const double *b_,
-                           const double *bt_,
-                           const double *d_,
-                           const double *x_,
-                           double *y_,
-                           const int d1d,
-                           const int q1d)
-{
-   const int D1D = d1d;
-   const int Q1D = q1d;
-   auto B = ConstDeviceMatrix(b_, Q1D, D1D);
-   auto Bt = ConstDeviceMatrix(bt_, D1D, Q1D);
-   auto D = DeviceTensor<4,const double>(d_, Q1D, Q1D, Q1D, NE);
-   auto X = DeviceTensor<4,const double>(x_, D1D, D1D, D1D, NE);
-   auto Y = DeviceTensor<4,double>(y_, D1D, D1D, D1D, NE);
-
-   if (!ACCUMULATE)
-   {
-      for (int dz = 0; dz < D1D; ++dz)
-      {
-         for (int dy = 0; dy < D1D; ++dy)
-         {
-            for (int dx = 0; dx < D1D; ++dx)
-            {
-               Y(dx, dy, dz, e) = 0.0;
-            }
-         }
-      }
-   }
-
-   constexpr int max_D1D = MAX_D1D;
-   constexpr int max_Q1D = MAX_Q1D;
-   double sol_xyz[max_Q1D][max_Q1D][max_Q1D];
-   for (int qz = 0; qz < Q1D; ++qz)
-   {
-      for (int qy = 0; qy < Q1D; ++qy)
-      {
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            sol_xyz[qz][qy][qx] = 0.0;
-         }
-      }
-   }
-   for (int dz = 0; dz < D1D; ++dz)
-   {
-      double sol_xy[max_Q1D][max_Q1D];
-      for (int qy = 0; qy < Q1D; ++qy)
-      {
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            sol_xy[qy][qx] = 0.0;
-         }
-      }
-      for (int dy = 0; dy < D1D; ++dy)
-      {
-         double sol_x[max_Q1D];
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            sol_x[qx] = 0;
-         }
-         for (int dx = 0; dx < D1D; ++dx)
-         {
-            const double s = X(dx,dy,dz,e);
-            for (int qx = 0; qx < Q1D; ++qx)
-            {
-               sol_x[qx] += B(qx,dx) * s;
-            }
-         }
-         for (int qy = 0; qy < Q1D; ++qy)
-         {
-            const double wy = B(qy,dy);
-            for (int qx = 0; qx < Q1D; ++qx)
-            {
-               sol_xy[qy][qx] += wy * sol_x[qx];
-            }
-         }
-      }
-      for (int qz = 0; qz < Q1D; ++qz)
-      {
-         const double wz = B(qz,dz);
-         for (int qy = 0; qy < Q1D; ++qy)
-         {
-            for (int qx = 0; qx < Q1D; ++qx)
-            {
-               sol_xyz[qz][qy][qx] += wz * sol_xy[qy][qx];
-            }
-         }
-      }
-   }
-   for (int qz = 0; qz < Q1D; ++qz)
-   {
-      for (int qy = 0; qy < Q1D; ++qy)
-      {
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            sol_xyz[qz][qy][qx] *= D(qx,qy,qz,e);
-         }
-      }
-   }
-   for (int qz = 0; qz < Q1D; ++qz)
-   {
-      double sol_xy[max_D1D][max_D1D];
-      for (int dy = 0; dy < D1D; ++dy)
-      {
-         for (int dx = 0; dx < D1D; ++dx)
-         {
-            sol_xy[dy][dx] = 0;
-         }
-      }
-      for (int qy = 0; qy < Q1D; ++qy)
-      {
-         double sol_x[max_D1D];
-         for (int dx = 0; dx < D1D; ++dx)
-         {
-            sol_x[dx] = 0;
-         }
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            const double s = sol_xyz[qz][qy][qx];
-            for (int dx = 0; dx < D1D; ++dx)
-            {
-               sol_x[dx] += Bt(dx,qx) * s;
-            }
-         }
-         for (int dy = 0; dy < D1D; ++dy)
-         {
-            const double wy = Bt(dy,qy);
-            for (int dx = 0; dx < D1D; ++dx)
-            {
-               sol_xy[dy][dx] += wy * sol_x[dx];
-            }
-         }
-      }
-      for (int dz = 0; dz < D1D; ++dz)
-      {
-         const double wz = Bt(dz,qz);
-         for (int dy = 0; dy < D1D; ++dy)
-         {
-            for (int dx = 0; dx < D1D; ++dx)
-            {
-               Y(dx,dy,dz,e) += wz * sol_xy[dy][dx];
-            }
-         }
-      }
-   }
-}
-
-template<int T_D1D, int T_Q1D, bool ACCUMULATE = true>
-MFEM_HOST_DEVICE inline
-void SmemPAMassApply3D_Element(const int e,
-                               const int NE,
-                               const double *b_,
-                               const double *d_,
-                               const double *x_,
-                               double *y_,
-                               const int d1d = 0,
-                               const int q1d = 0)
-{
-   constexpr int D1D = T_D1D ? T_D1D : d1d;
-   constexpr int Q1D = T_Q1D ? T_Q1D : q1d;
-   constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-   constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-   constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
-
-   auto b = ConstDeviceMatrix(b_, Q1D, D1D);
-   auto d = DeviceTensor<4,const double>(d_, Q1D, Q1D, Q1D, NE);
-   auto x = DeviceTensor<4,const double>(x_, D1D, D1D, D1D, NE);
-   auto y = DeviceTensor<4,double>(y_, D1D, D1D, D1D, NE);
-
-   MFEM_SHARED double sDQ[MQ1*MD1];
-   double (*B)[MD1] = (double (*)[MD1]) sDQ;
-   double (*Bt)[MQ1] = (double (*)[MQ1]) sDQ;
-   MFEM_SHARED double sm0[MDQ*MDQ*MDQ];
-   MFEM_SHARED double sm1[MDQ*MDQ*MDQ];
-   double (*X)[MD1][MD1]   = (double (*)[MD1][MD1]) sm0;
-   double (*DDQ)[MD1][MQ1] = (double (*)[MD1][MQ1]) sm1;
-   double (*DQQ)[MQ1][MQ1] = (double (*)[MQ1][MQ1]) sm0;
-   double (*QQQ)[MQ1][MQ1] = (double (*)[MQ1][MQ1]) sm1;
-   double (*QQD)[MQ1][MD1] = (double (*)[MQ1][MD1]) sm0;
-   double (*QDD)[MD1][MD1] = (double (*)[MD1][MD1]) sm1;
-   MFEM_FOREACH_THREAD(dy,y,D1D)
-   {
-      MFEM_FOREACH_THREAD(dx,x,D1D)
-      {
-         MFEM_UNROLL(MD1)
-         for (int dz = 0; dz < D1D; ++dz)
-         {
-            X[dz][dy][dx] = x(dx,dy,dz,e);
-         }
-      }
-      MFEM_FOREACH_THREAD(dx,x,Q1D)
-      {
-         B[dx][dy] = b(dx,dy);
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(dy,y,D1D)
-   {
-      MFEM_FOREACH_THREAD(qx,x,Q1D)
-      {
-         double u[D1D];
-         MFEM_UNROLL(MD1)
-         for (int dz = 0; dz < D1D; dz++)
-         {
-            u[dz] = 0;
-         }
-         MFEM_UNROLL(MD1)
-         for (int dx = 0; dx < D1D; ++dx)
-         {
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; ++dz)
-            {
-               u[dz] += X[dz][dy][dx] * B[qx][dx];
-            }
-         }
-         MFEM_UNROLL(MD1)
-         for (int dz = 0; dz < D1D; ++dz)
-         {
-            DDQ[dz][dy][qx] = u[dz];
-         }
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(qy,y,Q1D)
-   {
-      MFEM_FOREACH_THREAD(qx,x,Q1D)
-      {
-         double u[D1D];
-         MFEM_UNROLL(MD1)
-         for (int dz = 0; dz < D1D; dz++)
-         {
-            u[dz] = 0;
-         }
-         MFEM_UNROLL(MD1)
-         for (int dy = 0; dy < D1D; ++dy)
-         {
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; dz++)
-            {
-               u[dz] += DDQ[dz][dy][qx] * B[qy][dy];
-            }
-         }
-         MFEM_UNROLL(MD1)
-         for (int dz = 0; dz < D1D; dz++)
-         {
-            DQQ[dz][qy][qx] = u[dz];
-         }
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(qy,y,Q1D)
-   {
-      MFEM_FOREACH_THREAD(qx,x,Q1D)
-      {
-         double u[Q1D];
-         MFEM_UNROLL(MQ1)
-         for (int qz = 0; qz < Q1D; qz++)
-         {
-            u[qz] = 0;
-         }
-         MFEM_UNROLL(MD1)
-         for (int dz = 0; dz < D1D; ++dz)
-         {
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; qz++)
-            {
-               u[qz] += DQQ[dz][qy][qx] * B[qz][dz];
-            }
-         }
-         MFEM_UNROLL(MQ1)
-         for (int qz = 0; qz < Q1D; qz++)
-         {
-            QQQ[qz][qy][qx] = u[qz] * d(qx,qy,qz,e);
-         }
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(di,y,D1D)
-   {
-      MFEM_FOREACH_THREAD(q,x,Q1D)
-      {
-         Bt[di][q] = b(q,di);
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(qy,y,Q1D)
-   {
-      MFEM_FOREACH_THREAD(dx,x,D1D)
-      {
-         double u[Q1D];
-         MFEM_UNROLL(MQ1)
-         for (int qz = 0; qz < Q1D; ++qz)
-         {
-            u[qz] = 0;
-         }
-         MFEM_UNROLL(MQ1)
-         for (int qx = 0; qx < Q1D; ++qx)
-         {
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; ++qz)
-            {
-               u[qz] += QQQ[qz][qy][qx] * Bt[dx][qx];
-            }
-         }
-         MFEM_UNROLL(MQ1)
-         for (int qz = 0; qz < Q1D; ++qz)
-         {
-            QQD[qz][qy][dx] = u[qz];
-         }
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(dy,y,D1D)
-   {
-      MFEM_FOREACH_THREAD(dx,x,D1D)
-      {
-         double u[Q1D];
-         MFEM_UNROLL(MQ1)
-         for (int qz = 0; qz < Q1D; ++qz)
-         {
-            u[qz] = 0;
-         }
-         MFEM_UNROLL(MQ1)
-         for (int qy = 0; qy < Q1D; ++qy)
-         {
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; ++qz)
-            {
-               u[qz] += QQD[qz][qy][dx] * Bt[dy][qy];
-            }
-         }
-         MFEM_UNROLL(MQ1)
-         for (int qz = 0; qz < Q1D; ++qz)
-         {
-            QDD[qz][dy][dx] = u[qz];
-         }
-      }
-   }
-   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(dy,y,D1D)
-   {
-      MFEM_FOREACH_THREAD(dx,x,D1D)
-      {
-         double u[D1D];
-         MFEM_UNROLL(MD1)
-         for (int dz = 0; dz < D1D; ++dz)
-         {
-            u[dz] = 0;
-         }
-         MFEM_UNROLL(MQ1)
-         for (int qz = 0; qz < Q1D; ++qz)
-         {
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; ++dz)
-            {
-               u[dz] += QDD[qz][dy][dx] * Bt[dz][qz];
-            }
-         }
-         MFEM_UNROLL(MD1)
-         for (int dz = 0; dz < D1D; ++dz)
-         {
-            if (ACCUMULATE)
-            {
-               y(dx,dy,dz,e) += u[dz];
-            }
-            else
-            {
-               y(dx,dy,dz,e) = u[dz];
-            }
-         }
-      }
-   }
-   MFEM_SYNC_THREAD;
-}
-
-} // namespace internal
-
-} // namespace mfem
-
-#endif
@@ -12,7 +12,6 @@
 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
 #include "gridfunc.hpp"
-#include "qfunction.hpp"
 #include "ceed/integrators/diffusion/diffusion.hpp"

 using namespace std;
@@ -150,16 +149,7 @@ void VectorDiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedPADiffusionIntegrator(*this, fes, Q);
-      }
-      else
-      {
-         ceedOp = new ceed::PADiffusionIntegrator(fes, *ir, Q);
-      }
+      ceedOp = new ceed::PADiffusionIntegrator(fes, *ir, Q);
      return;
   }
   const int dims = el.GetDim();
@@ -176,9 +166,43 @@ void VectorDiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)

   MFEM_VERIFY(!VQ && !MQ,
               "Only scalar coefficient supported for partial assembly for VectorDiffusionIntegrator");
+   Vector coeff;
+   if (Q == nullptr)
+   {
+      coeff.SetSize(1);
+      coeff(0) = 1.0;
+   }
+   else if (ConstantCoefficient* cQ = dynamic_cast<ConstantCoefficient*>(Q))
+   {
+      coeff.SetSize(1);
+      coeff(0) = cQ->constant;
+   }
+   else if (QuadratureFunctionCoefficient* qfQ =
+               dynamic_cast<QuadratureFunctionCoefficient*>(Q))
+   {
+      const QuadratureFunction &qFun = qfQ->GetQuadFunction();
+      MFEM_VERIFY(qFun.Size() == ne*nq,
+                  "Incompatible QuadratureFunction dimension \n");

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(Q, qs, CoefficientStorage::COMPRESSED);
+      MFEM_VERIFY(ir == &qFun.GetSpace()->GetElementIntRule(0),
+                  "IntegrationRule used within integrator and in"
+                  " QuadratureFunction appear to be different");
+      qFun.Read();
+      coeff.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
+   }
+   else
+   {
+      coeff.SetSize(nq * ne);
+      auto Co = Reshape(coeff.HostWrite(), nq, ne);
+      for (int e = 0; e < ne; ++e)
+      {
+         ElementTransformation& T = *fes.GetElementTransformation(e);
+         for (int q = 0; q < nq; ++q)
+         {
+            Co(q,e) = Q->Eval(T, ir->IntPoint(q));
+         }
+      }
+   }

   const Array<double> &w = ir->GetWeights();
   const Vector &j = geom->J;
@@ -30,19 +30,7 @@ void VectorDiffusionIntegrator::AssembleMF(const FiniteElementSpace &fes)
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
-      MFEM_VERIFY(!VQ && !MQ,
-                  "Only scalar coefficient supported for DiffusionIntegrator"
-                  " with libCEED");
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedMFDiffusionIntegrator(*this, fes, Q);
-      }
-      else
-      {
-         ceedOp = new ceed::MFDiffusionIntegrator(fes, *ir, Q);
-      }
+      ceedOp = new ceed::MFDiffusionIntegrator(fes, *ir, Q);
      return;
   }
   MFEM_ABORT("Error: VectorDiffusionIntegrator::AssembleMF only implemented"
@@ -34,16 +34,7 @@ void VectorMassIntegrator::AssemblePA(const FiniteElementSpace &fes)
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedPAMassIntegrator(*this, fes, Q);
-      }
-      else
-      {
-         ceedOp = new ceed::PAMassIntegrator(fes, *ir, Q);
-      }
+      ceedOp = new ceed::PAMassIntegrator(fes, *ir, Q);
      return;
   }
   dim = mesh->Dimension();
@@ -34,16 +34,7 @@ void VectorMassIntegrator::AssembleMF(const FiniteElementSpace &fes)
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
-      const bool mixed = mesh->GetNumGeometries(mesh->Dimension()) > 1 ||
-                         fes.IsVariableOrder();
-      if (mixed)
-      {
-         ceedOp = new ceed::MixedMFMassIntegrator(*this, fes, Q);
-      }
-      else
-      {
-         ceedOp = new ceed::MFMassIntegrator(fes, *ir, Q);
-      }
+      ceedOp = new ceed::MFMassIntegrator(fes, *ir, Q);
      return;
   }
   MFEM_ABORT("Error: VectorMassIntegrator::AssembleMF only implemented with"
@@ -11,8 +11,6 @@

 #include "../general/forall.hpp"
 #include "bilininteg.hpp"
-#include "qspace.hpp"
-#include "gridfunc.hpp"

 namespace mfem
 {
@@ -91,7 +89,6 @@ void SmemPAHcurlMassApply3D(const int D1D,
                            Vector &y);

 void PAHdivSetup2D(const int Q1D,
-                   const int coeffDim,
                   const int NE,
                   const Array<double> &w,
                   const Vector &j,
@@ -99,7 +96,6 @@ void PAHdivSetup2D(const int Q1D,
                   Vector &op);

 void PAHdivSetup3D(const int Q1D,
-                   const int coeffDim,
                   const int NE,
                   const Array<double> &w,
                   const Vector &j,
@@ -153,7 +149,6 @@ void PAHcurlH1ApplyTranspose3D(const int D1D,
 void PAHdivMassAssembleDiagonal2D(const int D1D,
                                  const int Q1D,
                                  const int NE,
-                                  const bool symmetric,
                                  const Array<double> &Bo_,
                                  const Array<double> &Bc_,
                                  const Vector &op_,
@@ -162,24 +157,32 @@ void PAHdivMassAssembleDiagonal2D(const int D1D,
 void PAHdivMassAssembleDiagonal3D(const int D1D,
                                  const int Q1D,
                                  const int NE,
-                                  const bool symmetric,
                                  const Array<double> &Bo_,
                                  const Array<double> &Bc_,
                                  const Vector &op_,
                                  Vector &diag_);

-void PAHdivMassApply(const int dim,
-                     const int D1D,
-                     const int Q1D,
-                     const int NE,
-                     const bool symmetric,
-                     const Array<double> &Bo,
-                     const Array<double> &Bc,
-                     const Array<double> &Bot,
-                     const Array<double> &Bct,
-                     const Vector &op,
-                     const Vector &x,
-                     Vector &y);
+void PAHdivMassApply2D(const int D1D,
+                       const int Q1D,
+                       const int NE,
+                       const Array<double> &Bo_,
+                       const Array<double> &Bc_,
+                       const Array<double> &Bot_,
+                       const Array<double> &Bct_,
+                       const Vector &op_,
+                       const Vector &x_,
+                       Vector &y_);
+
+void PAHdivMassApply3D(const int D1D,
+                       const int Q1D,
+                       const int NE,
+                       const Array<double> &Bo_,
+                       const Array<double> &Bc_,
+                       const Array<double> &Bot_,
+                       const Array<double> &Bct_,
+                       const Vector &op_,
+                       const Vector &x_,
+                       Vector &y_);

 void PAHcurlL2Setup(const int NQ,
                    const int coeffDim,
@@ -794,63 +797,129 @@ void VectorFEMassIntegrator::AssemblePA(const FiniteElementSpace &trial_fes,
   trial_fetype = trial_el->GetDerivType();
   test_fetype = test_el->GetDerivType();

+   auto SMQ = dynamic_cast<SymmetricMatrixCoefficient *>(MQ);
+
+   const int MQsymmDim = SMQ ? (SMQ->GetSize() * (SMQ->GetSize() + 1)) / 2 : 0;
+   const int MQfullDim = MQ ? (MQ->GetHeight() * MQ->GetWidth()) : 0;
+   const int MQdim = SMQ ? MQsymmDim : MQfullDim;
+   const int coeffDim = MQ ? MQdim : (DQ ? DQ->GetVDim() : 1);
+
+   symmetric = (SMQ || MQ == NULL);
+
   const bool trial_curl = (trial_fetype == mfem::FiniteElement::CURL);
   const bool trial_div = (trial_fetype == mfem::FiniteElement::DIV);
   const bool test_curl = (test_fetype == mfem::FiniteElement::CURL);
   const bool test_div = (test_fetype == mfem::FiniteElement::DIV);

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(qs, CoefficientStorage::SYMMETRIC);
-   if (Q) { coeff.Project(*Q); }
-   else if (MQ) { coeff.ProjectTranspose(*MQ); }
-   else if (DQ) { coeff.Project(*DQ); }
-   else { coeff.SetConstant(1.0); }
-
-   const int coeff_dim = coeff.GetVDim();
-   symmetric = (coeff_dim != dim*dim);
-
   if ((trial_curl && test_div) || (trial_div && test_curl))
-      pa_data.SetSize((coeff_dim == 1 ? 1 : dim*dim) * nq * ne,
+      pa_data.SetSize((coeffDim == 1 ? 1 : dim*dim) * nq * ne,
                      Device::GetMemoryType());
   else
-      pa_data.SetSize((symmetric ? symmDims : dims*dims) * nq * ne,
+      pa_data.SetSize((symmetric ? symmDims : MQfullDim) * nq * ne,
                      Device::GetMemoryType());

+   Vector coeff(coeffDim * ne * nq);
+   coeff = 1.0;
+   auto coeffh = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
+   if (Q || DQ || MQ)
+   {
+      Vector DM(DQ ? coeffDim : 0);
+      DenseMatrix M;
+      DenseSymmetricMatrix SM;
+
+      if (DQ)
+      {
+         MFEM_VERIFY(coeffDim == dim, "");
+      }
+      if (SMQ)
+      {
+         MFEM_VERIFY(SMQ->GetSize() == dim, "");
+         SM.SetSize(dim);
+      }
+      else if (MQ)
+      {
+         MFEM_VERIFY(coeffDim == MQdim, "");
+         MFEM_VERIFY(MQ->GetHeight() == dim && MQ->GetWidth() == dim, "");
+         M.SetSize(dim);
+      }
+
+
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            if (SMQ)
+            {
+               SMQ->Eval(SM, *tr, ir->IntPoint(p));
+               int cnt = 0;
+               for (int i=0; i<dim; ++i)
+                  for (int j=i; j<dim; ++j, ++cnt)
+                  {
+                     coeffh(cnt, p, e) = SM(i,j);
+                  }
+            }
+            else if (MQ)
+            {
+               MQ->Eval(M, *tr, ir->IntPoint(p));
+
+               for (int i=0; i<dim; ++i)
+                  for (int j=0; j<dim; ++j)
+                  {
+                     coeffh(j+(i*dim), p, e) = M(i,j);
+                  }
+            }
+            else if (DQ)
+            {
+               DQ->Eval(DM, *tr, ir->IntPoint(p));
+               for (int i=0; i<coeffDim; ++i)
+               {
+                  coeffh(i, p, e) = DM[i];
+               }
+            }
+            else
+            {
+               coeffh(0, p, e) = Q->Eval(*tr, ir->IntPoint(p));
+            }
+         }
+      }
+   }
+
   if (trial_curl && test_curl && dim == 3)
   {
-      PADiffusionSetup3D(quad1D, coeff_dim, ne, ir->GetWeights(), geom->J,
+      PADiffusionSetup3D(quad1D, coeffDim, ne, ir->GetWeights(), geom->J,
                         coeff, pa_data);
   }
   else if (trial_curl && test_curl && dim == 2)
   {
-      PADiffusionSetup2D<2>(quad1D, coeff_dim, ne, ir->GetWeights(), geom->J,
+      PADiffusionSetup2D<2>(quad1D, coeffDim, ne, ir->GetWeights(), geom->J,
                            coeff, pa_data);
   }
   else if (trial_div && test_div && dim == 3)
   {
-      PAHdivSetup3D(quad1D, coeff_dim, ne, ir->GetWeights(), geom->J,
+      PAHdivSetup3D(quad1D, ne, ir->GetWeights(), geom->J,
                    coeff, pa_data);
   }
   else if (trial_div && test_div && dim == 2)
   {
-      PAHdivSetup2D(quad1D, coeff_dim, ne, ir->GetWeights(), geom->J,
+      PAHdivSetup2D(quad1D, ne, ir->GetWeights(), geom->J,
                    coeff, pa_data);
   }
   else if (((trial_curl && test_div) || (trial_div && test_curl)) &&
            test_fel->GetOrder() == trial_fel->GetOrder())
   {
-      if (coeff_dim == 1)
+      if (coeffDim == 1)
      {
-         PAHcurlL2Setup(nq, coeff_dim, ne, ir->GetWeights(), coeff, pa_data);
+         PAHcurlL2Setup(nq, coeffDim, ne, ir->GetWeights(), coeff, pa_data);
      }
      else
      {
         const bool tr = (trial_div && test_curl);
         if (dim == 3)
-            PAHcurlHdivSetup3D(quad1D, coeff_dim, ne, tr, ir->GetWeights(),
+            PAHcurlHdivSetup3D(quad1D, coeffDim, ne, tr, ir->GetWeights(),
                               geom->J, coeff, pa_data);
         else
-            PAHcurlHdivSetup2D(quad1D, coeff_dim, ne, tr, ir->GetWeights(),
+            PAHcurlHdivSetup2D(quad1D, coeffDim, ne, tr, ir->GetWeights(),
                               geom->J, coeff, pa_data);
      }
   }
@@ -894,7 +963,7 @@ void VectorFEMassIntegrator::AssembleDiagonalPA(Vector& diag)
      else if (trial_fetype == mfem::FiniteElement::DIV &&
               test_fetype == trial_fetype)
      {
-         PAHdivMassAssembleDiagonal3D(dofs1D, quad1D, ne, symmetric,
+         PAHdivMassAssembleDiagonal3D(dofs1D, quad1D, ne,
                                      mapsO->B, mapsC->B, pa_data, diag);
      }
      else
@@ -902,7 +971,7 @@ void VectorFEMassIntegrator::AssembleDiagonalPA(Vector& diag)
         MFEM_ABORT("Unknown kernel.");
      }
   }
-   else // 2D
+   else
   {
      if (trial_fetype == mfem::FiniteElement::CURL && test_fetype == trial_fetype)
      {
@@ -912,7 +981,7 @@ void VectorFEMassIntegrator::AssembleDiagonalPA(Vector& diag)
      else if (trial_fetype == mfem::FiniteElement::DIV &&
               test_fetype == trial_fetype)
      {
-         PAHdivMassAssembleDiagonal2D(dofs1D, quad1D, ne, symmetric,
+         PAHdivMassAssembleDiagonal2D(dofs1D, quad1D, ne,
                                      mapsO->B, mapsC->B, pa_data, diag);
      }
      else
@@ -965,8 +1034,8 @@ void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
      }
      else if (trial_div && test_div)
      {
-         PAHdivMassApply(3, dofs1D, quad1D, ne, symmetric, mapsO->B, mapsC->B, mapsO->Bt,
-                         mapsC->Bt, pa_data, x, y);
+         PAHdivMassApply3D(dofs1D, quad1D, ne, mapsO->B, mapsC->B, mapsO->Bt,
+                           mapsC->Bt, pa_data, x, y);
      }
      else if (trial_curl && test_div)
      {
@@ -987,7 +1056,7 @@ void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
         MFEM_ABORT("Unknown kernel.");
      }
   }
-   else // 2D
+   else
   {
      if (trial_curl && test_curl)
      {
@@ -996,8 +1065,8 @@ void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
      }
      else if (trial_div && test_div)
      {
-         PAHdivMassApply(2, dofs1D, quad1D, ne, symmetric, mapsO->B, mapsC->B, mapsO->Bt,
-                         mapsC->Bt, pa_data, x, y);
+         PAHdivMassApply2D(dofs1D, quad1D, ne, mapsO->B, mapsC->B, mapsO->Bt,
+                           mapsC->Bt, pa_data, x, y);
      }
      else if ((trial_curl && test_div) || (trial_div && test_curl))
      {
@@ -1042,11 +1111,6 @@ void VectorFEMassIntegrator::AddMultTransposePA(const Vector &x,

   if (symmetricSpaces)
   {
-      if (MQ && dynamic_cast<SymmetricMatrixCoefficient*>(MQ) == NULL)
-      {
-         MFEM_ABORT("VectorFEMassIntegrator transpose not implemented for asymmetric MatrixCoefficient");
-      }
-
      this->AddMultPA(x, y);
   }
 }
@@ -1092,8 +1156,19 @@ void MixedVectorGradientIntegrator::AssemblePA(const FiniteElementSpace

   pa_data.SetSize(symmDims * nq * ne, Device::GetMemoryType());

-   QuadratureSpace qs(*mesh, *ir);
-   CoefficientVector coeff(Q, qs, CoefficientStorage::FULL);
+   Vector coeff(ne * nq);
+   coeff = 1.0;
+   if (Q)
+   {
+      for (int e=0; e<ne; ++e)
+      {
+         ElementTransformation *tr = mesh->GetElementTransformation(e);
+         for (int p=0; p<nq; ++p)
+         {
+            coeff[p + (e * nq)] = Q->Eval(*tr, ir->IntPoint(p));
+         }
+      }
+   }

   // Use the same setup functions as VectorFEMassIntegrator.
   if (test_el->GetDerivType() == mfem::FiniteElement::CURL && dim == 3)
@@ -62,20 +62,6 @@ PAConvectionIntegrator::PAConvectionIntegrator(
 #endif
 }

-MixedPAConvectionIntegrator::MixedPAConvectionIntegrator(
-   const ConvectionIntegrator &integ,
-   const mfem::FiniteElementSpace &fes,
-   mfem::VectorCoefficient *Q,
-   const double alpha)
-{
-#ifdef MFEM_USE_CEED
-   ConvectionOperatorInfo info(fes.GetMesh()->Dimension(), alpha);
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
 MFConvectionIntegrator::MFConvectionIntegrator(
   const mfem::FiniteElementSpace &fes,
   const mfem::IntegrationRule &irm,
@@ -91,20 +77,6 @@ MFConvectionIntegrator::MFConvectionIntegrator(
 #endif
 }

-MixedMFConvectionIntegrator::MixedMFConvectionIntegrator(
-   const ConvectionIntegrator &integ,
-   const mfem::FiniteElementSpace &fes,
-   mfem::VectorCoefficient *Q,
-   const double alpha)
-{
-#ifdef MFEM_USE_CEED
-   ConvectionOperatorInfo info(fes.GetMesh()->Dimension(), alpha);
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
 } // namespace ceed

 } // namespace mfem
@@ -13,7 +13,6 @@
 #define MFEM_LIBCEED_CONV_HPP

 #include "../../interface/integrator.hpp"
-#include "../../interface/mixed_integrator.hpp"
 #include "../../../fespace.hpp"

 namespace mfem
@@ -27,39 +26,21 @@ class PAConvectionIntegrator : public PAIntegrator
 {
 public:
   PAConvectionIntegrator(const mfem::FiniteElementSpace &fes,
-                          const mfem::IntegrationRule &ir,
+                          const mfem::IntegrationRule &irm,
                          mfem::VectorCoefficient *Q,
                          const double alpha);
 };

-class MixedPAConvectionIntegrator : public MixedIntegrator<PAIntegrator>
-{
-public:
-   MixedPAConvectionIntegrator(const ConvectionIntegrator &integ,
-                               const mfem::FiniteElementSpace &fes,
-                               mfem::VectorCoefficient *Q,
-                               const double alpha);
-};
-
 /// Represent a ConvectionIntegrator with AssemblyLevel::None using libCEED.
 class MFConvectionIntegrator : public MFIntegrator
 {
 public:
   MFConvectionIntegrator(const mfem::FiniteElementSpace &fes,
-                          const mfem::IntegrationRule &ir,
+                          const mfem::IntegrationRule &irm,
                          mfem::VectorCoefficient *Q,
                          const double alpha);
 };

-class MixedMFConvectionIntegrator : public MixedIntegrator<MFIntegrator>
-{
-public:
-   MixedMFConvectionIntegrator(const ConvectionIntegrator &integ,
-                               const mfem::FiniteElementSpace &fes,
-                               mfem::VectorCoefficient *Q,
-                               const double alpha);
-};
-
 }

 }
@@ -60,32 +60,6 @@ PADiffusionIntegrator::PADiffusionIntegrator(
 #endif
 }

-MixedPADiffusionIntegrator::MixedPADiffusionIntegrator(
-   const DiffusionIntegrator &integ,
-   const mfem::FiniteElementSpace &fes,
-   mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   DiffusionOperatorInfo info(fes.GetMesh()->Dimension());
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
-MixedPADiffusionIntegrator::MixedPADiffusionIntegrator(
-   const VectorDiffusionIntegrator &integ,
-   const mfem::FiniteElementSpace &fes,
-   mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   DiffusionOperatorInfo info(fes.GetMesh()->Dimension());
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
 MFDiffusionIntegrator::MFDiffusionIntegrator(
   const mfem::FiniteElementSpace &fes,
   const mfem::IntegrationRule &irm,
@@ -100,32 +74,6 @@ MFDiffusionIntegrator::MFDiffusionIntegrator(
 #endif
 }

-MixedMFDiffusionIntegrator::MixedMFDiffusionIntegrator(
-   const DiffusionIntegrator &integ,
-   const mfem::FiniteElementSpace &fes,
-   mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   DiffusionOperatorInfo info(fes.GetMesh()->Dimension());
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
-MixedMFDiffusionIntegrator::MixedMFDiffusionIntegrator(
-   const VectorDiffusionIntegrator &integ,
-   const mfem::FiniteElementSpace &fes,
-   mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   DiffusionOperatorInfo info(fes.GetMesh()->Dimension());
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
 } // namespace ceed

 } // namespace mfem
@@ -13,7 +13,6 @@
 #define MFEM_LIBCEED_DIFF_HPP

 #include "../../interface/integrator.hpp"
-#include "../../interface/mixed_integrator.hpp"
 #include "../../../fespace.hpp"

 namespace mfem
@@ -27,43 +26,19 @@ class PADiffusionIntegrator : public PAIntegrator
 {
 public:
   PADiffusionIntegrator(const mfem::FiniteElementSpace &fes,
-                         const mfem::IntegrationRule &ir,
+                         const mfem::IntegrationRule &irm,
                         mfem::Coefficient *Q);
 };

-class MixedPADiffusionIntegrator : public MixedIntegrator<PAIntegrator>
-{
-public:
-   MixedPADiffusionIntegrator(const DiffusionIntegrator &integ,
-                              const mfem::FiniteElementSpace &fes,
-                              mfem::Coefficient *Q);
-
-   MixedPADiffusionIntegrator(const VectorDiffusionIntegrator &integ,
-                              const mfem::FiniteElementSpace &fes,
-                              mfem::Coefficient *Q);
-};
-
 /// Represent a DiffusionIntegrator with AssemblyLevel::None using libCEED.
 class MFDiffusionIntegrator : public MFIntegrator
 {
 public:
   MFDiffusionIntegrator(const mfem::FiniteElementSpace &fes,
-                         const mfem::IntegrationRule &ir,
+                         const mfem::IntegrationRule &irm,
                         mfem::Coefficient *Q);
 };

-class MixedMFDiffusionIntegrator : public MixedIntegrator<MFIntegrator>
-{
-public:
-   MixedMFDiffusionIntegrator(const DiffusionIntegrator &integ,
-                              const mfem::FiniteElementSpace &fes,
-                              mfem::Coefficient *Q);
-
-   MixedMFDiffusionIntegrator(const VectorDiffusionIntegrator &integ,
-                              const mfem::FiniteElementSpace &fes,
-                              mfem::Coefficient *Q);
-};
-
 }

 }
@@ -59,30 +59,6 @@ PAMassIntegrator::PAMassIntegrator(const mfem::FiniteElementSpace &fes,
 #endif
 }

-MixedPAMassIntegrator::MixedPAMassIntegrator(const MassIntegrator &integ,
-                                             const mfem::FiniteElementSpace &fes,
-                                             mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   MassOperatorInfo info;
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
-MixedPAMassIntegrator::MixedPAMassIntegrator(const VectorMassIntegrator &integ,
-                                             const mfem::FiniteElementSpace &fes,
-                                             mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   MassOperatorInfo info;
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
 MFMassIntegrator::MFMassIntegrator(const mfem::FiniteElementSpace &fes,
                                   const mfem::IntegrationRule &irm,
                                   mfem::Coefficient *Q)
@@ -96,30 +72,6 @@ MFMassIntegrator::MFMassIntegrator(const mfem::FiniteElementSpace &fes,
 #endif
 }

-MixedMFMassIntegrator::MixedMFMassIntegrator(const MassIntegrator &integ,
-                                             const mfem::FiniteElementSpace &fes,
-                                             mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   MassOperatorInfo info;
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
-MixedMFMassIntegrator::MixedMFMassIntegrator(const VectorMassIntegrator &integ,
-                                             const mfem::FiniteElementSpace &fes,
-                                             mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   MassOperatorInfo info;
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
 } // namespace ceed

 } // namespace mfem
@@ -13,7 +13,6 @@
 #define MFEM_LIBCEED_MASS_HPP

 #include "../../interface/integrator.hpp"
-#include "../../interface/mixed_integrator.hpp"
 #include "../../../fespace.hpp"

 namespace mfem
@@ -27,43 +26,19 @@ class PAMassIntegrator : public PAIntegrator
 {
 public:
   PAMassIntegrator(const mfem::FiniteElementSpace &fes,
-                    const mfem::IntegrationRule &ir,
+                    const mfem::IntegrationRule &irm,
                    mfem::Coefficient *Q);
 };

-class MixedPAMassIntegrator : public MixedIntegrator<PAIntegrator>
-{
-public:
-   MixedPAMassIntegrator(const MassIntegrator &integ,
-                         const mfem::FiniteElementSpace &fes,
-                         mfem::Coefficient *Q);
-
-   MixedPAMassIntegrator(const VectorMassIntegrator &integ,
-                         const mfem::FiniteElementSpace &fes,
-                         mfem::Coefficient *Q);
-};
-
 /// Represent a MassIntegrator with AssemblyLevel::None using libCEED.
 class MFMassIntegrator : public MFIntegrator
 {
 public:
   MFMassIntegrator(const mfem::FiniteElementSpace &fes,
-                    const mfem::IntegrationRule &ir,
+                    const mfem::IntegrationRule &irm,
                    mfem::Coefficient *Q);
 };

-class MixedMFMassIntegrator : public MixedIntegrator<MFIntegrator>
-{
-public:
-   MixedMFMassIntegrator(const MassIntegrator &integ,
-                         const mfem::FiniteElementSpace &fes,
-                         mfem::Coefficient *Q);
-
-   MixedMFMassIntegrator(const VectorMassIntegrator &integ,
-                         const mfem::FiniteElementSpace &fes,
-                         mfem::Coefficient *Q);
-};
-
 }

 }
@@ -60,19 +60,6 @@ PAVectorConvectionNLFIntegrator::PAVectorConvectionNLFIntegrator(
 #endif
 }

-MixedPAVectorConvectionNLIntegrator::MixedPAVectorConvectionNLIntegrator(
-   const VectorConvectionNLFIntegrator &integ,
-   const mfem::FiniteElementSpace &fes,
-   mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   NLConvectionOperatorInfo info(fes.GetMesh()->Dimension());
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
 MFVectorConvectionNLFIntegrator::MFVectorConvectionNLFIntegrator(
   const mfem::FiniteElementSpace &fes,
   const mfem::IntegrationRule &irm,
@@ -87,19 +74,6 @@ MFVectorConvectionNLFIntegrator::MFVectorConvectionNLFIntegrator(
 #endif
 }

-MixedMFVectorConvectionNLIntegrator::MixedMFVectorConvectionNLIntegrator(
-   const VectorConvectionNLFIntegrator &integ,
-   const mfem::FiniteElementSpace &fes,
-   mfem::Coefficient *Q)
-{
-#ifdef MFEM_USE_CEED
-   NLConvectionOperatorInfo info(fes.GetMesh()->Dimension());
-   Assemble(integ, info, fes, Q);
-#else
-   MFEM_ABORT("MFEM must be built with MFEM_USE_CEED=YES to use libCEED.");
-#endif
-}
-
 } // namespace ceed

 } // namespace mfem
@@ -13,7 +13,6 @@
 #define MFEM_LIBCEED_NLCONV_HPP

 #include "../../interface/integrator.hpp"
-#include "../../interface/mixed_integrator.hpp"
 #include "../../../fespace.hpp"

 namespace mfem
@@ -32,15 +31,6 @@ public:
                                   mfem::Coefficient *coeff);
 };

-class MixedPAVectorConvectionNLIntegrator : public MixedIntegrator<PAIntegrator>
-{
-public:
-   MixedPAVectorConvectionNLIntegrator(
-      const VectorConvectionNLFIntegrator &integ,
-      const mfem::FiniteElementSpace &fes,
-      mfem::Coefficient *Q);
-};
-
 /** Represent a VectorConvectionNLFIntegrator with AssemblyLevel::None
    using libCEED. */
 class MFVectorConvectionNLFIntegrator : public MFIntegrator
@@ -51,15 +41,6 @@ public:
                                   mfem::Coefficient *coeff);
 };

-class MixedMFVectorConvectionNLIntegrator : public MixedIntegrator<MFIntegrator>
-{
-public:
-   MixedMFVectorConvectionNLIntegrator(
-      const VectorConvectionNLFIntegrator &integ,
-      const mfem::FiniteElementSpace &fes,
-      mfem::Coefficient *Q);
-};
-
 }

 }
@@ -327,13 +327,13 @@ CEED_QFUNCTION(f_apply_conv_mf_const)(void *ctx, CeedInt Q,
            const CeedScalar A33 = J11 * J22 - J12 * J21;
            const CeedScalar w = qw[i] * coeff;
            const CeedScalar qd00 = w * A11;
-            const CeedScalar qd10 = w * A21;
-            const CeedScalar qd20 = w * A31;
-            const CeedScalar qd01 = w * A12;
+            const CeedScalar qd01 = w * A21;
+            const CeedScalar qd02 = w * A31;
+            const CeedScalar qd10 = w * A12;
            const CeedScalar qd11 = w * A22;
-            const CeedScalar qd21 = w * A32;
-            const CeedScalar qd02 = w * A13;
-            const CeedScalar qd12 = w * A23;
+            const CeedScalar qd12 = w * A32;
+            const CeedScalar qd20 = w * A13;
+            const CeedScalar qd21 = w * A23;
            const CeedScalar qd22 = w * A33;
            const CeedScalar u0   = u[i + Q * 0];
            const CeedScalar u1   = u[i + Q * 1];
@@ -440,13 +440,13 @@ CEED_QFUNCTION(f_apply_conv_mf_quad)(void *ctx, CeedInt Q,
            const CeedScalar A33 = J11 * J22 - J12 * J21;
            const CeedScalar w = qw[i] * c[i];
            const CeedScalar qd00 = w * A11;
-            const CeedScalar qd10 = w * A21;
-            const CeedScalar qd20 = w * A31;
-            const CeedScalar qd01 = w * A12;
+            const CeedScalar qd01 = w * A21;
+            const CeedScalar qd02 = w * A31;
+            const CeedScalar qd10 = w * A12;
            const CeedScalar qd11 = w * A22;
-            const CeedScalar qd21 = w * A32;
-            const CeedScalar qd02 = w * A13;
-            const CeedScalar qd12 = w * A23;
+            const CeedScalar qd12 = w * A32;
+            const CeedScalar qd20 = w * A13;
+            const CeedScalar qd21 = w * A23;
            const CeedScalar qd22 = w * A33;
            const CeedScalar u0   = u[i + Q * 0];
            const CeedScalar u1   = u[i + Q * 1];
@@ -36,8 +36,6 @@ static CeedElemTopology GetCeedTopology(Geometry::Type geom)
         return CEED_TOPOLOGY_HEX;
      case Geometry::PRISM:
         return CEED_TOPOLOGY_PRISM;
-      case Geometry::PYRAMID:
-         return CEED_TOPOLOGY_PYRAMID;
      default:
         MFEM_ABORT("This type of element is not supported");
         return CEED_TOPOLOGY_PRISM; // Silence warning
@@ -45,11 +43,11 @@ static CeedElemTopology GetCeedTopology(Geometry::Type geom)
 }

 static void InitNonTensorBasis(const mfem::FiniteElementSpace &fes,
-                               const mfem::FiniteElement &fe,
                               const mfem::IntegrationRule &ir,
                               Ceed ceed, CeedBasis *basis)
 {
-   const mfem::DofToQuad &maps = fe.GetDofToQuad(ir, mfem::DofToQuad::FULL);
+   const mfem::DofToQuad &maps = fes.GetFE(0)->
+                                 GetDofToQuad(ir,mfem::DofToQuad::FULL);
   mfem::Mesh *mesh = fes.GetMesh();
   const int dim = mesh->Dimension();
   const int ndofs = maps.ndof;
@@ -64,18 +62,18 @@ static void InitNonTensorBasis(const mfem::FiniteElementSpace &fes,
      if (dim>2) { qX(2,i) = ip.z; }
      qW(i) = ip.weight;
   }
-   CeedBasisCreateH1(ceed, GetCeedTopology(fe.GetGeomType()),
+   CeedBasisCreateH1(ceed, GetCeedTopology(fes.GetFE(0)->GetGeomType()),
                     fes.GetVDim(), ndofs, nqpts,
                     maps.Bt.GetData(), maps.Gt.GetData(),
                     qX.GetData(), qW.GetData(), basis);
 }

 static void InitTensorBasis(const mfem::FiniteElementSpace &fes,
-                            const mfem::FiniteElement &fe,
                            const mfem::IntegrationRule &ir,
                            Ceed ceed, CeedBasis *basis)
 {
-   const mfem::DofToQuad &maps = fe.GetDofToQuad(ir, mfem::DofToQuad::TENSOR);
+   const mfem::DofToQuad &maps =
+      fes.GetFE(0)->GetDofToQuad(ir, mfem::DofToQuad::TENSOR);
   mfem::Mesh *mesh = fes.GetMesh();
   const int ndofs = maps.ndof;
   const int nqpts = maps.nqpt;
@@ -98,30 +96,28 @@ static void InitTensorBasis(const mfem::FiniteElementSpace &fes,
                           qW.GetData(), basis);
 }

-static void InitBasisImpl(const FiniteElementSpace &fes,
-                          const FiniteElement &fe,
-                          const IntegrationRule &ir,
-                          Ceed ceed, CeedBasis *basis)
+void InitBasis(const FiniteElementSpace &fes,
+               const IntegrationRule &irm,
+               Ceed ceed, CeedBasis *basis)
 {
   // Check for FES -> basis, restriction in hash tables
-   const int P = fe.GetDof();
-   const int Q = ir.GetNPoints();
+   const mfem::FiniteElement *fe = fes.GetFE(0);
+   const int P = fe->GetDof();
+   const int Q = irm.GetNPoints();
   const int ncomp = fes.GetVDim();
-   BasisKey basis_key(&fes, &ir, ncomp, P, Q);
+   BasisKey basis_key(&fes, &irm, ncomp, P, Q);
   auto basis_itr = mfem::internal::ceed_basis_map.find(basis_key);
-   const bool tensor = dynamic_cast<const mfem::TensorBasisElement *>
-                       (&fe) != nullptr;

-   // Init or retrieve key values
+   // Init or retreive key values
   if (basis_itr == mfem::internal::ceed_basis_map.end())
   {
-      if ( tensor )
+      if (UsesTensorBasis(fes))
      {
-         InitTensorBasis(fes, fe, ir, ceed, basis);
+         InitTensorBasis(fes, irm, ceed, basis);
      }
      else
      {
-         InitNonTensorBasis(fes, fe, ir, ceed, basis);
+         InitNonTensorBasis(fes, irm, ceed, basis);
      }
      mfem::internal::ceed_basis_map[basis_key] = *basis;
   }
@@ -131,24 +127,6 @@ static void InitBasisImpl(const FiniteElementSpace &fes,
   }
 }

-void InitBasis(const FiniteElementSpace &fes,
-               const IntegrationRule &ir,
-               Ceed ceed, CeedBasis *basis)
-{
-   const mfem::FiniteElement &fe = *fes.GetFE(0);
-   InitBasisImpl(fes, fe, ir, ceed, basis);
-}
-
-void InitBasisWithIndices(const FiniteElementSpace &fes,
-                          const IntegrationRule &ir,
-                          int nelem,
-                          const int* indices,
-                          Ceed ceed, CeedBasis *basis)
-{
-   const mfem::FiniteElement &fe = *fes.GetFE(indices[0]);
-   InitBasisImpl(fes, fe, ir, ceed, basis);
-}
-
 #endif

 } // namespace ceed
@@ -22,32 +22,17 @@ namespace ceed

 #ifdef MFEM_USE_CEED

-/** @brief Initialize a CeedBasis for non-mixed meshes.
+/** @brief Initialize a CeedBasis.

   @param[in] fes Input finite element space.
-   @param[in] ir Input integration rule.
+   @param[in] irm Input integration rule.
   @param[in] ceed Input Ceed object.
   @param[out] basis The address of the initialized CeedBasis object.
 */
 void InitBasis(const FiniteElementSpace &fes,
-               const IntegrationRule &ir,
+               const IntegrationRule &irm,
               Ceed ceed, CeedBasis *basis);

-/** @brief Initialize a CeedBasis for mixed meshes.
-
-    @param[in] fes The finite element space.
-    @param[in] ir is the integration rule for the operator.
-    @param[in] nelem The number of elements.
-    @param[in] indices The indices of the elements of same type in the
-                       `FiniteElementSpace`.
-    @param[in] ceed The Ceed object.
-    @param[out] basis The `CeedBasis` to initialize. */
-void InitBasisWithIndices(const FiniteElementSpace &fes,
-                          const IntegrationRule &ir,
-                          int nelem,
-                          const int* indices,
-                          Ceed ceed, CeedBasis *basis);
-
 #endif

 } // namespace ceed
@@ -14,13 +14,11 @@

 #ifdef MFEM_USE_CEED

-#include "../../../general/forall.hpp"
 #include "../../../config/config.hpp"
 #include "../../../linalg/vector.hpp"
 #include "../../../linalg/dtensor.hpp"
 #include "../../../mesh/mesh.hpp"
 #include "../../gridfunc.hpp"
-#include "../../qfunction.hpp"
 #include "util.hpp"
 #include "ceed.hpp"

@@ -79,14 +77,7 @@ struct QuadCoefficient : VariableCoefficient

 /** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
    mfem::Coefficient @a Q, an mfem::Mesh @a mesh, and an mfem::IntegrationRule
-    @a ir.
-
-    @param[in] Q is the coefficient from the `Integrator`.
-    @param[in] mesh is the mesh.
-    @param[in] ir is the integration rule.
-    @param[out] coeff_ptr is the structure to store the coefficient for the
-                          `CeedOperator`.
-    @param[out] ctx is the Context associated to the QFunction. */
+    @a ir. */
 template <typename Context>
 void InitCoefficient(mfem::Coefficient *Q, mfem::Mesh &mesh,
                     const mfem::IntegrationRule &ir,
@@ -122,7 +113,7 @@ void InitCoefficient(mfem::Coefficient *Q, mfem::Mesh &mesh,
      MFEM_VERIFY(qFun.Size() == nq * ne,
                  "Incompatible QuadratureFunction dimension \n");

-      MFEM_VERIFY(&ir == &qFun.GetSpace()->GetIntRule(0),
+      MFEM_VERIFY(&ir == &qFun.GetSpace()->GetElementIntRule(0),
                  "IntegrationRule used within integrator and in"
                  " QuadratureFunction appear to be different");
      qFun.Read();
@@ -152,15 +143,8 @@ void InitCoefficient(mfem::Coefficient *Q, mfem::Mesh &mesh,


 /** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
-    mfem::VectorCoefficient @a VQ, an mfem::Mesh @a mesh, and an
-    mfem::IntegrationRule @a ir.
-
-    @param[in] VQ is the vector coefficient from the `Integrator`.
-    @param[in] mesh is the mesh.
-    @param[in] ir is the integration rule.
-    @param[out] coeff_ptr is the structure to store the coefficient for the
-                          `CeedOperator`.
-    @param[out] ctx is the Context associated to the QFunction. */
+    mfem::VectorCoefficient @a Q, an mfem::Mesh @a mesh, and an
+    mfem::IntegrationRule @a ir. */
 template <typename Context>
 void InitCoefficient(mfem::VectorCoefficient *VQ, mfem::Mesh &mesh,
                     const mfem::IntegrationRule &ir,
@@ -196,7 +180,7 @@ void InitCoefficient(mfem::VectorCoefficient *VQ, mfem::Mesh &mesh,
      MFEM_VERIFY(qFun.Size() == dim * nq * ne,
                  "Incompatible QuadratureFunction dimension \n");

-      MFEM_VERIFY(&ir == &qFun.GetSpace()->GetIntRule(0),
+      MFEM_VERIFY(&ir == &qFun.GetSpace()->GetElementIntRule(0),
                  "IntegrationRule used within integrator and in"
                  " QuadratureFunction appear to be different");
      qFun.Read();
@@ -230,209 +214,6 @@ void InitCoefficient(mfem::VectorCoefficient *VQ, mfem::Mesh &mesh,
   }
 }

-/** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
-    mfem::Coefficient @a Q, an mfem::Mesh @a mesh, and an mfem::IntegrationRule
-    @a ir for the elements given by the indices @a indices.
-
-    @param[in] Q is the coefficient from the `Integrator`.
-    @param[in] mesh is the mesh.
-    @param[in] ir is the integration rule.
-    @param[in] nelem The number of elements.
-    @param[in] indices The indices of the elements of same type in the
-                       `FiniteElementSpace`.
-    @param[out] coeff_ptr is the structure to store the coefficient for the
-                          `CeedOperator`.
-    @param[out] ctx is the Context associated to the QFunction. */
-template <typename Context>
-void InitCoefficientWithIndices(mfem::Coefficient *Q, mfem::Mesh &mesh,
-                                const mfem::IntegrationRule &ir,
-                                int nelem,
-                                const int* indices,
-                                Coefficient*& coeff_ptr, Context &ctx)
-{
-   if ( Q == nullptr )
-   {
-      Coefficient *ceedCoeff = new Coefficient(1);
-      ctx.coeff = 1.0;
-      coeff_ptr = ceedCoeff;
-   }
-   else if (ConstantCoefficient *const_coeff =
-               dynamic_cast<ConstantCoefficient*>(Q))
-   {
-      Coefficient *ceedCoeff = new Coefficient(1);
-      ctx.coeff = const_coeff->constant;
-      coeff_ptr = ceedCoeff;
-   }
-   else if (GridFunctionCoefficient* gf_coeff =
-               dynamic_cast<GridFunctionCoefficient*>(Q))
-   {
-      GridCoefficient *ceedCoeff =
-         new GridCoefficient(*gf_coeff->GetGridFunction());
-      coeff_ptr = ceedCoeff;
-   }
-   else if (QuadratureFunctionCoefficient *cQ =
-               dynamic_cast<QuadratureFunctionCoefficient*>(Q))
-   {
-      QuadCoefficient *ceedCoeff = new QuadCoefficient(1);
-      const int ne = mesh.GetNE();
-      const int nq = ir.GetNPoints();
-      const mfem::QuadratureFunction &qFun = cQ->GetQuadFunction();
-      MFEM_VERIFY(qFun.Size() == nq * ne,
-                  "Incompatible QuadratureFunction dimension \n");
-
-      MFEM_VERIFY(&ir == &qFun.GetSpace()->GetIntRule(0),
-                  "IntegrationRule used within integrator and in"
-                  " QuadratureFunction appear to be different");
-      ceedCoeff->coeff.SetSize(nq * nelem);
-      Memory<int> m_indices((int*)indices, nelem, false);
-      auto in = Reshape(qFun.Read(), nq, ne);
-      auto d_indices = Read(m_indices, nelem);
-      auto out = Reshape(ceedCoeff->coeff.Write(), nq, nelem);
-      MFEM_FORALL(i, nelem * nq,
-      {
-         const int q = i%nq;
-         const int sub_e = i/nq;
-         const int e = d_indices[sub_e];
-         out(q, sub_e) = in(q, e);
-      });
-      m_indices.DeleteDevice();
-      InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
-      coeff_ptr = ceedCoeff;
-   }
-   else
-   {
-      QuadCoefficient *ceedCoeff = new QuadCoefficient(1);
-      const int nq = ir.GetNPoints();
-      ceedCoeff->coeff.SetSize(nq * nelem);
-      auto C = Reshape(ceedCoeff->coeff.HostWrite(), nq, nelem);
-      for (int i = 0; i < nelem; ++i)
-      {
-         const int e = indices[i];
-         mfem::ElementTransformation &T = *mesh.GetElementTransformation(e);
-         for (int q = 0; q < nq; ++q)
-         {
-            C(q, i) = Q->Eval(T, ir.IntPoint(q));
-         }
-      }
-      InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
-      coeff_ptr = ceedCoeff;
-   }
-}
-
-
-/** @brief Initializes an mfem::ceed::Coefficient @a coeff_ptr from an
-    mfem::VectorCoefficient @a Q, an mfem::Mesh @a mesh, and an
-    mfem::IntegrationRule @a ir for the elements given by the indices @a indices.
-
-    @param[in] VQ is the vector coefficient from the `Integrator`.
-    @param[in] mesh is the mesh.
-    @param[in] ir is the integration rule.
-    @param[in] nelem The number of elements.
-    @param[in] indices The indices of the elements of same type in the
-                       `FiniteElementSpace`.
-    @param[out] coeff_ptr is the structure to store the coefficient for the
-                          `CeedOperator`.
-    @param[out] ctx is the Context associated to the QFunction. */
-template <typename Context>
-void InitCoefficientWithIndices(mfem::VectorCoefficient *VQ, mfem::Mesh &mesh,
-                                const mfem::IntegrationRule &ir,
-                                int nelem,
-                                const int* indices,
-                                Coefficient *&coeff_ptr, Context &ctx)
-{
-   if (VectorConstantCoefficient *const_coeff =
-          dynamic_cast<VectorConstantCoefficient*>(VQ))
-   {
-      const int vdim = const_coeff->GetVDim();
-      const mfem::Vector &val = const_coeff->GetVec();
-      Coefficient *ceedCoeff = new Coefficient(vdim);
-      for (int i = 0; i < vdim; i++)
-      {
-         ctx.coeff[i] = val[i];
-      }
-      coeff_ptr = ceedCoeff;
-   }
-   else if (VectorGridFunctionCoefficient* vgf_coeff =
-               dynamic_cast<VectorGridFunctionCoefficient*>(VQ))
-   {
-      GridCoefficient *ceedCoeff =
-         new GridCoefficient(*vgf_coeff->GetGridFunction());
-      coeff_ptr = ceedCoeff;
-   }
-   else if (VectorQuadratureFunctionCoefficient *cQ =
-               dynamic_cast<VectorQuadratureFunctionCoefficient*>(VQ))
-   {
-      QuadCoefficient *ceedCoeff = new QuadCoefficient(cQ->GetVDim());
-      const int dim = mesh.Dimension();
-      const int ne = mesh.GetNE();
-      const int nq = ir.GetNPoints();
-      const mfem::QuadratureFunction &qFun = cQ->GetQuadFunction();
-      MFEM_VERIFY(qFun.Size() == dim * nq * ne,
-                  "Incompatible QuadratureFunction dimension \n");
-
-      MFEM_VERIFY(&ir == &qFun.GetSpace()->GetIntRule(0),
-                  "IntegrationRule used within integrator and in"
-                  " QuadratureFunction appear to be different");
-      ceedCoeff->coeff.SetSize(dim * nq * nelem);
-      Memory<int> m_indices((int*)indices, nelem, false);
-      auto in = Reshape(qFun.Read(), dim, nq, ne);
-      auto d_indices = Read(m_indices, nelem);
-      auto out = Reshape(ceedCoeff->coeff.Write(), dim, nq, nelem);
-      MFEM_FORALL(i, nelem * nq,
-      {
-         const int q = i%nq;
-         const int sub_e = i/nq;
-         const int e = d_indices[sub_e];
-         for (int d = 0; d < dim; d++)
-         {
-            out(d, q, sub_e) = in(d, q, e);
-         }
-      });
-      m_indices.DeleteDevice();
-      InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
-      coeff_ptr = ceedCoeff;
-   }
-   else
-   {
-      const int dim = mesh.Dimension();
-      QuadCoefficient *ceedCoeff = new QuadCoefficient(dim);
-      const int nq = ir.GetNPoints();
-      ceedCoeff->coeff.SetSize(dim * nq * nelem);
-      auto C = Reshape(ceedCoeff->coeff.HostWrite(), dim, nq, nelem);
-      mfem::DenseMatrix Q_ir;
-      for (int i = 0; i < nelem; ++i)
-      {
-         const int e = indices[i];
-         mfem::ElementTransformation &T = *mesh.GetElementTransformation(e);
-         VQ->Eval(Q_ir, T, ir);
-         for (int q = 0; q < nq; ++q)
-         {
-            for (int d = 0; d < dim; ++d)
-            {
-               C(d, q, i) = Q_ir(d, q);
-            }
-         }
-      }
-      InitVector(ceedCoeff->coeff, ceedCoeff->coeffVector);
-      coeff_ptr = ceedCoeff;
-   }
-}
-
-template <typename Coeff, typename Context>
-void InitCoefficient(Coeff *Q, mfem::Mesh &mesh,
-                     const mfem::IntegrationRule &ir, int nelem,
-                     const int* indices, Coefficient *&coeff_ptr, Context &ctx)
-{
-   if (indices)
-   {
-      InitCoefficientWithIndices(Q, mesh, ir, nelem, indices, coeff_ptr, ctx);
-   }
-   else
-   {
-      InitCoefficient(Q, mesh, ir, coeff_ptr, ctx);
-   }
-}
-
 } // namespace ceed

 } // namespace mfem
@@ -18,7 +18,6 @@
 #include "operator.hpp"
 #include "coefficient.hpp"
 #include "restriction.hpp"
-#include "util.hpp"
 #include "ceed.hpp"

 namespace mfem
@@ -87,7 +86,6 @@ protected:
   CeedQFunctionContext build_ctx;
   CeedOperator build_oper;

-public:
   PAIntegrator()
      : Operator(),
        trial_basis(nullptr), test_basis(nullptr), mesh_basis(nullptr),
@@ -97,51 +95,23 @@ public:
        qdata(nullptr), coeff(nullptr), build_ctx(nullptr), build_oper(nullptr)
   { }

-   /** @brief This method assembles the `PAIntegrator` with the given
-       `CeedOperatorInfo` @a info, an `mfem::FiniteElementSpace` @a fes, an
-       `mfem::IntegrationRule` @a ir, and `mfem::Coefficient` or
-       `mfem::VectorCoefficient` @a Q.
-       The `CeedOperatorInfo` type is expected to inherit from `OperatorInfo`,
-       and contain a `Context` type relevant to the qFunctions.
+public:
+   /** This method assembles the PAIntegrator.

-       @param[in] info is the structure describing the CeedOperator to assemble.
-       @param[in] fes is the finite element space.
-       @param[in] ir is the integration rule for the operator.
-       @param[in] Q is the coefficient from the `Integrator`. */
+       @param[in] info the `CeedOperatorInfo` describing the `CeedOperator`,
+                       the `CeedOperatorInfo` type is expected to inherit from
+                       `OperatorInfo` and contain a `Context` type relevant to
+                       the qFunctions.
+       @param[in] fes the `FiniteElementSpace` for the form,
+       @param[in] ir the `IntegrationRule` for the numerical integration,
+       @param[in] Q `Coefficient` or `VectorCoefficient`. */
   template <typename CeedOperatorInfo, typename CoeffType>
   void Assemble(CeedOperatorInfo &info,
                 const mfem::FiniteElementSpace &fes,
-                 const mfem::IntegrationRule &ir,
+                 const mfem::IntegrationRule &irm,
                 CoeffType *Q)
   {
-      Assemble(info, fes, ir, fes.GetNE(), nullptr, Q);
-   }
-
-   /** @brief This method assembles the `PAIntegrator` with the given
-       `CeedOperatorInfo` @a info, an `mfem::FiniteElementSpace` @a fes, an
-       `mfem::IntegrationRule` @a ir, and `mfem::Coefficient` or
-       `mfem::VectorCoefficient` @a Q for the elements given by the indices
-       @a indices.
-       The `CeedOperatorInfo` type is expected to inherit from `OperatorInfo`,
-       and contain a `Context` type relevant to the qFunctions.
-
-       @param[in] info is the structure describing the CeedOperator to assemble.
-       @param[in] fes is the finite element space.
-       @param[in] ir is the integration rule for the operator.
-       @param[in] nelem The number of elements.
-       @param[in] indices The indices of the elements of same type in the
-                          `FiniteElementSpace`. If `indices == nullptr`, assumes
-                          that the `FiniteElementSpace` is not mixed.
-       @param[in] Q is the coefficient from the `Integrator`. */
-   template <typename CeedOperatorInfo, typename CoeffType>
-   void Assemble(CeedOperatorInfo &info,
-                 const mfem::FiniteElementSpace &fes,
-                 const mfem::IntegrationRule &ir,
-                 int nelem,
-                 const int* indices,
-                 CoeffType *Q)
-   {
-      Assemble(info, fes, fes, ir, nelem, indices, Q);
+      Assemble(info, fes, fes, irm, Q);
   }

   /** This method assembles the PAIntegrator for mixed forms.
@@ -158,40 +128,12 @@ public:
   void Assemble(CeedOperatorInfo &info,
                 const mfem::FiniteElementSpace &trial_fes,
                 const mfem::FiniteElementSpace &test_fes,
-                 const mfem::IntegrationRule &ir,
-                 CoeffType *Q)
-   {
-      Assemble(info, trial_fes, test_fes, ir, trial_fes.GetNE(), nullptr, Q);
-   }
-
-   /** This method assembles the PAIntegrator for mixed forms on mixed meshes.
-
-       @param[in] info the `CeedOperatorInfo` describing the `CeedOperator`,
-                       the `CeedOperatorInfo` type is expected to inherit from
-                       `OperatorInfo` and contain a `Context` type relevant to
-                       the qFunctions.
-       @param[in] trial_fes the trial `FiniteElementSpace` for the form,
-       @param[in] test_fes the test `FiniteElementSpace` for the form,
-       @param[in] ir the `IntegrationRule` for the numerical integration,
-       @param[in] nelem The number of elements,
-       @param[in] indices The indices of the elements of same type in the
-                          `FiniteElementSpace`. If `indices == nullptr`, assumes
-                          that the `FiniteElementSpace` is not mixed,
-       @param[in] Q `Coefficient` or `VectorCoefficient`. */
-   template <typename CeedOperatorInfo, typename CoeffType>
-   void Assemble(CeedOperatorInfo &info,
-                 const mfem::FiniteElementSpace &trial_fes,
-                 const mfem::FiniteElementSpace &test_fes,
-                 const mfem::IntegrationRule &ir,
-                 int nelem,
-                 const int* indices,
+                 const mfem::IntegrationRule &irm,
                 CoeffType *Q)
   {
      Ceed ceed(internal::ceed);
      mfem::Mesh &mesh = *trial_fes.GetMesh();
-      MFEM_VERIFY(!(!indices && mesh.GetNumGeometries(mesh.Dimension()) > 1),
-                  "Use ceed::MixedIntegrator on mixed meshes.");
-      InitCoefficient(Q, mesh, ir, nelem, indices, coeff, info.ctx);
+      InitCoefficient(Q, mesh, irm, coeff, info.ctx);
      bool const_coeff = coeff->IsConstant();
      std::string build_func = const_coeff ? info.build_func_const
                               : info.build_func_quad;
@@ -203,6 +145,7 @@ public:
                     info.trial_op,
                     info.test_op
                    };
+      CeedInt nqpts, nelem = mesh.GetNE();
      CeedInt dim = mesh.SpaceDimension();
      CeedInt trial_vdim = trial_fes.GetVDim();
      CeedInt test_vdim = test_fes.GetVDim();
@@ -210,23 +153,23 @@ public:
      mesh.EnsureNodes();
      if ( &trial_fes == &test_fes )
      {
-         InitBasisAndRestriction(trial_fes, ir, nelem, indices,
-                                 ceed, &trial_basis, &trial_restr);
+         InitBasisAndRestriction(trial_fes, irm, ceed,
+                                 &trial_basis, &trial_restr);
         test_basis = trial_basis;
         test_restr = trial_restr;
      }
      else
      {
-         InitBasisAndRestriction(trial_fes, ir, nelem, indices,
-                                 ceed, &trial_basis, &trial_restr);
-         InitBasisAndRestriction(test_fes, ir, nelem, indices,
-                                 ceed, &test_basis, &test_restr);
+         InitBasisAndRestriction(trial_fes, irm, ceed,
+                                 &trial_basis, &trial_restr);
+         InitBasisAndRestriction(test_fes, irm, ceed,
+                                 &test_basis, &test_restr);
      }

      const mfem::FiniteElementSpace *mesh_fes = mesh.GetNodalFESpace();
      MFEM_VERIFY(mesh_fes, "the Mesh has no nodal FE space");
-      InitBasisAndRestriction(*mesh_fes, ir, nelem, indices,
-                              ceed, &mesh_basis, &mesh_restr);
+      InitBasisAndRestriction(*mesh_fes, irm, ceed, &mesh_basis,
+                              &mesh_restr);

      CeedInt trial_nqpts, test_nqpts;
      CeedBasisGetNumQuadraturePoints(trial_basis, &trial_nqpts);
@@ -234,7 +177,7 @@ public:
      MFEM_VERIFY(trial_nqpts == test_nqpts,
                  "Trial and test basis must have the same number of quadrature"
                  " points.");
-      CeedInt nqpts = trial_nqpts;
+      nqpts = trial_nqpts;

      const int qdatasize = op.qdatasize;
      InitStridedRestriction(*mesh_fes, nelem, nqpts, qdatasize,
@@ -278,10 +221,8 @@ public:
      CeedOperatorCreate(ceed, build_qfunc, NULL, NULL, &build_oper);
      if (GridCoefficient *gridCoeff = dynamic_cast<GridCoefficient*>(coeff))
      {
-         InitBasisAndRestriction(*gridCoeff->gf.FESpace(), ir,
-                                 nelem, indices, ceed,
-                                 &gridCoeff->basis,
-                                 &gridCoeff->restr);
+         InitBasisAndRestriction(*gridCoeff->gf.FESpace(), irm, ceed,
+                                 &gridCoeff->basis, &gridCoeff->restr);
         CeedOperatorSetField(build_oper, "coeff", gridCoeff->restr,
                              gridCoeff->basis, gridCoeff->coeffVector);
      }
@@ -290,8 +231,7 @@ public:
      {
         const int ncomp = quadCoeff->ncomp;
         CeedInt strides[3] = {ncomp, 1, ncomp*nqpts};
-         InitStridedRestriction(*mesh.GetNodalFESpace(),
-                                nelem, nqpts, ncomp, strides,
+         InitStridedRestriction(*mesh_fes, nelem, nqpts, ncomp, strides,
                                &quadCoeff->restr);
         CeedOperatorSetField(build_oper, "coeff", quadCoeff->restr,
                              CEED_BASIS_COLLOCATED, quadCoeff->coeffVector);
@@ -314,17 +254,22 @@ public:
      switch (op.trial_op)
      {
         case EvalMode::None:
-            CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim, CEED_EVAL_NONE);
+            CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim,
+                                  CEED_EVAL_NONE);
            break;
         case EvalMode::Interp:
-            CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim, CEED_EVAL_INTERP);
+            CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim,
+                                  CEED_EVAL_INTERP);
            break;
         case EvalMode::Grad:
-            CeedQFunctionAddInput(apply_qfunc, "gu", trial_vdim*dim, CEED_EVAL_GRAD);
+            CeedQFunctionAddInput(apply_qfunc, "gu", trial_vdim*dim,
+                                  CEED_EVAL_GRAD);
            break;
         case EvalMode::InterpAndGrad:
-            CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim, CEED_EVAL_INTERP);
-            CeedQFunctionAddInput(apply_qfunc, "gu", trial_vdim*dim, CEED_EVAL_GRAD);
+            CeedQFunctionAddInput(apply_qfunc, "u", trial_vdim,
+                                  CEED_EVAL_INTERP);
+            CeedQFunctionAddInput(apply_qfunc, "gu", trial_vdim*dim,
+                                  CEED_EVAL_GRAD);
            break;
      }
      // qdata
@@ -333,17 +278,22 @@ public:
      switch (op.test_op)
      {
         case EvalMode::None:
-            CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim, CEED_EVAL_NONE);
+            CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim,
+                                   CEED_EVAL_NONE);
            break;
         case EvalMode::Interp:
-            CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim, CEED_EVAL_INTERP);
+            CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim,
+                                   CEED_EVAL_INTERP);
            break;
         case EvalMode::Grad:
-            CeedQFunctionAddOutput(apply_qfunc, "gv", test_vdim*dim, CEED_EVAL_GRAD);
+            CeedQFunctionAddOutput(apply_qfunc, "gv", test_vdim*dim,
+                                   CEED_EVAL_GRAD);
            break;
         case EvalMode::InterpAndGrad:
-            CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim, CEED_EVAL_INTERP);
-            CeedQFunctionAddOutput(apply_qfunc, "gv", test_vdim*dim, CEED_EVAL_GRAD);
+            CeedQFunctionAddOutput(apply_qfunc, "v", test_vdim,
+                                   CEED_EVAL_INTERP);
+            CeedQFunctionAddOutput(apply_qfunc, "gv", test_vdim*dim,
+                                   CEED_EVAL_GRAD);
            break;
      }
      CeedQFunctionSetContext(apply_qfunc, build_ctx);
@@ -358,14 +308,18 @@ public:
                                 CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE);
            break;
         case EvalMode::Interp:
-            CeedOperatorSetField(oper, "u", trial_restr, trial_basis, CEED_VECTOR_ACTIVE);
+            CeedOperatorSetField(oper, "u", trial_restr, trial_basis,
+                                 CEED_VECTOR_ACTIVE);
            break;
         case EvalMode::Grad:
-            CeedOperatorSetField(oper, "gu", trial_restr, trial_basis, CEED_VECTOR_ACTIVE);
+            CeedOperatorSetField(oper, "gu", trial_restr, trial_basis,
+                                 CEED_VECTOR_ACTIVE);
            break;
         case EvalMode::InterpAndGrad:
-            CeedOperatorSetField(oper, "u", trial_restr, trial_basis, CEED_VECTOR_ACTIVE);
-            CeedOperatorSetField(oper, "gu", trial_restr, trial_basis, CEED_VECTOR_ACTIVE);
+            CeedOperatorSetField(oper, "u", trial_restr, trial_basis,
+                                 CEED_VECTOR_ACTIVE);
+            CeedOperatorSetField(oper, "gu", trial_restr, trial_basis,
+                                 CEED_VECTOR_ACTIVE);
            break;
      }
      // qdata
@@ -379,14 +333,18 @@ public:
                                 CEED_BASIS_COLLOCATED, CEED_VECTOR_ACTIVE);
            break;
         case EvalMode::Interp:
-            CeedOperatorSetField(oper, "v", test_restr, test_basis, CEED_VECTOR_ACTIVE);
+            CeedOperatorSetField(oper, "v", test_restr, test_basis,
+                                 CEED_VECTOR_ACTIVE);
            break;
         case EvalMode::Grad:
-            CeedOperatorSetField(oper, "gv", test_restr, test_basis, CEED_VECTOR_ACTIVE);
+            CeedOperatorSetField(oper, "gv", test_restr, test_basis,
+                                 CEED_VECTOR_ACTIVE);
            break;
         case EvalMode::InterpAndGrad:
-            CeedOperatorSetField(oper, "v", test_restr, test_basis, CEED_VECTOR_ACTIVE);
-            CeedOperatorSetField(oper, "gv", test_restr, test_basis, CEED_VECTOR_ACTIVE);
+            CeedOperatorSetField(oper, "v", test_restr, test_basis,
+                                 CEED_VECTOR_ACTIVE);
+            CeedOperatorSetField(oper, "gv", test_restr, test_basis,
+                                 CEED_VECTOR_ACTIVE);
            break;
      }

@@ -444,7 +402,6 @@ protected:
   Coefficient *coeff;
   CeedQFunctionContext build_ctx;

-public:
   MFIntegrator()
      : Operator(),
        trial_basis(nullptr), test_basis(nullptr), mesh_basis(nullptr),
@@ -453,51 +410,23 @@ public:
        apply_qfunc(nullptr), node_coords(nullptr),
        qdata(nullptr), coeff(nullptr), build_ctx(nullptr) { }

-   /** @brief This method assembles the `MFIntegrator` with the given
-       `CeedOperatorInfo` @a info, an `mfem::FiniteElementSpace` @a fes, an
-       `mfem::IntegrationRule` @a ir, and `mfem::Coefficient` or
-       `mfem::VectorCoefficient` @a Q.
-       The `CeedOperatorInfo` type is expected to inherit from `OperatorInfo`,
-       and contain a `Context` type relevant to the qFunctions.
+public:
+   /** This method assembles the MFIntegrator.

-       @param[in] info is the structure describing the CeedOperator to assemble.
-       @param[in] fes is the finite element space.
-       @param[in] ir is the integration rule for the operator.
-       @param[in] Q is the coefficient from the `Integrator`. */
+       @param[in] info the `CeedOperatorInfo` describing the `CeedOperator`,
+                       the `CeedOperatorInfo` type is expected to inherit from
+                       `OperatorInfo` and contain a `Context` type relevant to
+                       the qFunctions.
+       @param[in] fes the `FiniteElementSpace` for the form,
+       @param[in] ir the `IntegrationRule` for the numerical integration,
+       @param[in] Q `Coefficient` or `VectorCoefficient`. */
   template <typename CeedOperatorInfo, typename CoeffType>
   void Assemble(CeedOperatorInfo &info,
                 const mfem::FiniteElementSpace &fes,
-                 const mfem::IntegrationRule &ir,
+                 const mfem::IntegrationRule &irm,
                 CoeffType *Q)
   {
-      Assemble(info, fes, ir, fes.GetNE(), nullptr, Q);
-   }
-
-   /** @brief This method assembles the `MFIntegrator` with the given
-       `CeedOperatorInfo` @a info, an `mfem::FiniteElementSpace` @a fes, an
-       `mfem::IntegrationRule` @a ir, and `mfem::Coefficient` or
-       `mfem::VectorCoefficient` @a Q for the elements given by the indices
-       @a indices.
-       The `CeedOperatorInfo` type is expected to inherit from `OperatorInfo`,
-       and contain a `Context` type relevant to the qFunctions.
-
-       @param[in] info is the structure describing the CeedOperator to assemble.
-       @param[in] fes is the finite element space.
-       @param[in] ir is the integration rule for the operator.
-       @param[in] nelem The number of elements.
-       @param[in] indices The indices of the elements of same type in the
-                          `FiniteElementSpace`. If `indices == nullptr`, assumes
-                          that the `FiniteElementSpace` is not mixed.
-       @param[in] Q is the coefficient from the `Integrator`. */
-   template <typename CeedOperatorInfo, typename CoeffType>
-   void Assemble(CeedOperatorInfo &info,
-                 const mfem::FiniteElementSpace &fes,
-                 const mfem::IntegrationRule &ir,
-                 int nelem,
-                 const int* indices,
-                 CoeffType *Q)
-   {
-      Assemble(info, fes, fes, ir, nelem, indices, Q);
+      Assemble(info, fes, fes, irm, Q);
   }

   /** This method assembles the MFIntegrator for mixed forms.
@@ -514,40 +443,12 @@ public:
   void Assemble(CeedOperatorInfo &info,
                 const mfem::FiniteElementSpace &trial_fes,
                 const mfem::FiniteElementSpace &test_fes,
-                 const mfem::IntegrationRule &ir,
-                 CoeffType *Q)
-   {
-      Assemble(info, trial_fes, test_fes, ir, trial_fes.GetNE(), nullptr, Q);
-   }
-
-   /** This method assembles the MFIntegrator for mixed forms.
-
-       @param[in] info the `CeedOperatorInfo` describing the `CeedOperator`,
-                       the `CeedOperatorInfo` type is expected to inherit from
-                       `OperatorInfo` and contain a `Context` type relevant to
-                       the qFunctions.
-       @param[in] trial_fes the trial `FiniteElementSpace` for the form,
-       @param[in] test_fes the test `FiniteElementSpace` for the form,
-       @param[in] ir the `IntegrationRule` for the numerical integration,
-       @param[in] nelem The number of elements,
-       @param[in] indices The indices of the elements of same type in the
-                          `FiniteElementSpace`. If `indices == nullptr`, assumes
-                          that the `FiniteElementSpace` is not mixed,
-       @param[in] Q `Coefficient` or `VectorCoefficient`. */
-   template <typename CeedOperatorInfo, typename CoeffType>
-   void Assemble(CeedOperatorInfo &info,
-                 const mfem::FiniteElementSpace &trial_fes,
-                 const mfem::FiniteElementSpace &test_fes,
-                 const mfem::IntegrationRule &ir,
-                 int nelem,
-                 const int* indices,
+                 const mfem::IntegrationRule &irm,
                 CoeffType *Q)
   {
      Ceed ceed(internal::ceed);
      Mesh &mesh = *trial_fes.GetMesh();
-      MFEM_VERIFY(!(!indices && mesh.GetNumGeometries(mesh.Dimension()) > 1),
-                  "Use ceed::MixedIntegrator on mixed meshes.");
-      InitCoefficient(Q, mesh, ir, nelem, indices, coeff, info.ctx);
+      InitCoefficient(Q, mesh, irm, coeff, info.ctx);
      bool const_coeff = coeff->IsConstant();
      std::string apply_func = const_coeff ? info.apply_func_mf_const
                               : info.apply_func_mf_quad;
@@ -558,7 +459,7 @@ public:
                     info.trial_op,
                     info.test_op
                    };
-
+      CeedInt nqpts, nelem = mesh.GetNE();
      CeedInt dim = mesh.SpaceDimension();
      CeedInt trial_vdim = trial_fes.GetVDim();
      CeedInt test_vdim = test_fes.GetVDim();
@@ -566,22 +467,22 @@ public:
      mesh.EnsureNodes();
      if ( &trial_fes == &test_fes )
      {
-         InitBasisAndRestriction(trial_fes, ir, nelem, indices, ceed,
+         InitBasisAndRestriction(trial_fes, irm, ceed,
                                 &trial_basis, &trial_restr);
         test_basis = trial_basis;
         test_restr = trial_restr;
      }
      else
      {
-         InitBasisAndRestriction(trial_fes, ir, nelem, indices, ceed,
+         InitBasisAndRestriction(trial_fes, irm, ceed,
                                 &trial_basis, &trial_restr);
-         InitBasisAndRestriction(test_fes, ir, nelem, indices, ceed,
+         InitBasisAndRestriction(test_fes, irm, ceed,
                                 &test_basis, &test_restr);
      }

      const mfem::FiniteElementSpace *mesh_fes = mesh.GetNodalFESpace();
      MFEM_VERIFY(mesh_fes, "the Mesh has no nodal FE space");
-      InitBasisAndRestriction(*mesh_fes, ir, nelem, indices, ceed, &mesh_basis,
+      InitBasisAndRestriction(*mesh_fes, irm, ceed, &mesh_basis,
                              &mesh_restr);

      CeedInt trial_nqpts, test_nqpts;
@@ -590,7 +491,7 @@ public:
      MFEM_VERIFY(trial_nqpts == test_nqpts,
                  "Trial and test basis must have the same number of quadrature"
                  " points.");
-      CeedInt nqpts = trial_nqpts;
+      nqpts = trial_nqpts;

      InitVector(*mesh.GetNodes(), node_coords);

@@ -671,8 +572,8 @@ public:
      // coefficient
      if (GridCoefficient *gridCoeff = dynamic_cast<GridCoefficient*>(coeff))
      {
-         InitBasisAndRestriction(*gridCoeff->gf.FESpace(), ir, nelem, indices,
-                                 ceed, &gridCoeff->basis, &gridCoeff->restr);
+         InitBasisAndRestriction(*gridCoeff->gf.FESpace(), irm, ceed,
+                                 &gridCoeff->basis, &gridCoeff->restr);
         CeedOperatorSetField(oper, "coeff", gridCoeff->restr,
                              gridCoeff->basis, gridCoeff->coeffVector);
      }
@@ -22,8 +22,6 @@
 #include "coefficient.hpp"
 // PA or MF Operator using libCEED.
 #include "integrator.hpp"
-// PA Operator supporting mixed finite element spaces.
-#include "mixed_integrator.hpp"
 // Utility functions
 #include "util.hpp"
 // Wrapper to include <ceed.h>
@@ -1,126 +0,0 @@
-// Copyright (c) 2010-2022, Lawrence Livermore National Security, LLC. Produced
-// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-// LICENSE and NOTICE for details. LLNL-CODE-806117.
-//
-// This file is part of the MFEM library. For more information and source code
-// availability visit https://mfem.org.
-//
-// MFEM is free software; you can redistribute it and/or modify it under the
-// terms of the BSD-3 license. We welcome feedback and contributions, see file
-// CONTRIBUTING.md for details.
-
-#ifndef MFEM_LIBCEED_MIXED_INTEGRATOR
-#define MFEM_LIBCEED_MIXED_INTEGRATOR
-
-#include "ceed.hpp"
-#include "integrator.hpp"
-#include <unordered_map>
-
-namespace mfem
-{
-
-namespace ceed
-{
-
-/** @brief This class wraps a `ceed::PAIntegrator` or `ceed::MFIntegrator` to
-    support mixed finite element spaces. */
-template <typename CeedInteg>
-class MixedIntegrator : public ceed::Operator
-{
-#ifdef MFEM_USE_CEED
-   using ElementKey = std::pair<int, int>; //< Element::Type, Order >
-   struct key_hash
-   {
-      std::size_t operator()(const ElementKey& k) const
-      {
-         return k.first + 2 * k.second;
-      }
-   };
-   using ElementsMap = std::unordered_map<const ElementKey, int*, key_hash>;
-   std::vector<CeedInteg*> sub_ops;
-
-public:
-   template <typename Integrator, typename CeedOperatorInfo, typename CoeffType>
-   void Assemble(const Integrator &integ,
-                 CeedOperatorInfo &info,
-                 const mfem::FiniteElementSpace &fes,
-                 CoeffType *Q)
-   {
-      ElementsMap count;
-      ElementsMap element_indices;
-      ElementsMap offsets;
-
-      // Count the number of elements of each type
-      for (int i = 0; i < fes.GetNE(); i++)
-      {
-         ElementKey key(fes.GetElementType(i), fes.GetElementOrder(i));
-         auto value = count.find(key);
-         if (value == count.end())
-         {
-            count[key] = new int(1);
-         }
-         else
-         {
-            (*value->second)++;
-         }
-      }
-
-      // Initialization of the arrays
-      for ( const auto& value : count )
-      {
-         element_indices[value.first] = new int[*value.second];
-         offsets[value.first] = new int(0);
-      }
-
-      // Populates the indices arrays for each element type
-      for (int i = 0; i < fes.GetNE(); i++)
-      {
-         ElementKey key(fes.GetElementType(i), fes.GetElementOrder(i));
-         int &offset = *(offsets[key]);
-         int* indices_array = element_indices[key];
-         indices_array[offset] = i;
-         offset++;
-      }
-
-      // Create composite CeedOperator
-      CeedCompositeOperatorCreate(internal::ceed, &oper);
-
-      // Create each sub-CeedOperator
-      sub_ops.reserve(element_indices.size());
-      for (const auto& value : element_indices)
-      {
-         const int* indices = value.second;
-         const int first_index = indices[0];
-         const mfem::FiniteElement &el = *fes.GetFE(first_index);
-         auto &T = *fes.GetMesh()->GetElementTransformation(first_index);
-         MFEM_ASSERT(!integ.GetIntegrationRule(),
-                     "Mixed mesh integrators should not have an"
-                     " IntegrationRule.");
-         const IntegrationRule &ir = GetRule(integ, el, el, T);
-         auto sub_op = new CeedInteg();
-         int nelem = *count[value.first];
-         sub_op->Assemble(info, fes, ir, nelem, indices, Q);
-         sub_ops.push_back(sub_op);
-         CeedCompositeOperatorAddSub(oper, sub_op->GetCeedOperator());
-      }
-
-      const int ndofs = fes.GetVDim() * fes.GetNDofs();
-      CeedVectorCreate(internal::ceed, ndofs, &u);
-      CeedVectorCreate(internal::ceed, ndofs, &v);
-   }
-
-   virtual ~MixedIntegrator()
-   {
-      for (auto sub_op : sub_ops)
-      {
-         delete sub_op;
-      }
-   }
-#endif
-};
-
-} // namespace ceed
-
-} // namespace mfem
-
-#endif // MFEM_LIBCEED_MIXED_INTEGRATOR
@@ -20,8 +20,8 @@ namespace ceed

 #ifdef MFEM_USE_CEED

-static void InitNativeRestr(const mfem::FiniteElementSpace &fes,
-                            Ceed ceed, CeedElemRestriction *restr)
+static void InitNonTensorRestriction(const mfem::FiniteElementSpace &fes,
+                                     Ceed ceed, CeedElemRestriction *restr)
 {
   const mfem::FiniteElement *fe = fes.GetFE(0);
   const int P = fe->GetDof();
@@ -31,173 +31,77 @@ static void InitNativeRestr(const mfem::FiniteElementSpace &fes,
   const mfem::TensorBasisElement * tfe =
      dynamic_cast<const mfem::TensorBasisElement *>(fe);
   const int stride = compstride == 1 ? fes.GetVDim() : 1;
-   const mfem::Array<int>& dof_map = tfe->GetDofMap();
-
-   for (int i = 0; i < fes.GetNE(); i++)
+   if (tfe) // Lexicographic ordering using dof_map
   {
-      const int el_offset = P * i;
-      for (int j = 0; j < P; j++)
+      const mfem::Array<int>& dof_map = tfe->GetDofMap();
+      for (int i = 0; i < fes.GetNE(); i++)
      {
-         tp_el_dof[j+el_offset] = stride*el_dof.GetJ()[dof_map[j]+el_offset];
+         const int el_offset = P * i;
+         for (int j = 0; j < P; j++)
+         {
+            tp_el_dof[j+el_offset] = stride*el_dof.GetJ()[dof_map[j]+el_offset];
+         }
+      }
+   }
+   else  // Native ordering
+   {
+      for (int e = 0; e < fes.GetNE(); e++)
+      {
+         for (int i = 0; i < P; i++)
+         {
+            tp_el_dof[i + e*P] = stride*el_dof.GetJ()[i + e*P];
+         }
      }
   }
-
   CeedElemRestrictionCreate(ceed, fes.GetNE(), P, fes.GetVDim(),
                             compstride, (fes.GetVDim())*(fes.GetNDofs()),
                             CEED_MEM_HOST, CEED_COPY_VALUES,
                             tp_el_dof.GetData(), restr);
 }

-static void InitLexicoRestr(const mfem::FiniteElementSpace &fes,
-                            Ceed ceed, CeedElemRestriction *restr)
+// TODO fuse Tensor and NonTensor Restriction
+void InitTensorRestriction(const mfem::FiniteElementSpace &fes,
+                           Ceed ceed, CeedElemRestriction *restr)
 {
   const mfem::FiniteElement *fe = fes.GetFE(0);
-   const int P = fe->GetDof();
+   const mfem::TensorBasisElement * tfe =
+      dynamic_cast<const mfem::TensorBasisElement *>(fe);
+   MFEM_VERIFY(tfe, "invalid FE");
+   const mfem::Array<int>& dof_map = tfe->GetDofMap();
+
   CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
   const mfem::Table &el_dof = fes.GetElementToDofTable();
   mfem::Array<int> tp_el_dof(el_dof.Size_of_connections());
+   const int dof = fe->GetDof();
   const int stride = compstride == 1 ? fes.GetVDim() : 1;
-
-   for (int e = 0; e < fes.GetNE(); e++)
+   if (dof_map.Size()>0)
   {
-      for (int i = 0; i < P; i++)
+      for (int i = 0; i < fes.GetNE(); i++)
      {
-         tp_el_dof[i + e*P] = stride*el_dof.GetJ()[i + e*P];
+         const int el_offset = dof * i;
+         for (int j = 0; j < dof; j++)
+         {
+            tp_el_dof[j+el_offset] = stride*el_dof.GetJ()[dof_map[j]+el_offset];
+         }
      }
   }
-
-   CeedElemRestrictionCreate(ceed, fes.GetNE(), P, fes.GetVDim(),
+   else // dof_map.Size == 0, means dof_map[j]==j;
+   {
+      for (int i = 0; i < fes.GetNE(); i++)
+      {
+         const int el_offset = dof * i;
+         for (int j = 0; j < dof; j++)
+         {
+            tp_el_dof[j+el_offset] = stride*el_dof.GetJ()[j+el_offset];
+         }
+      }
+   }
+   CeedElemRestrictionCreate(ceed, fes.GetNE(), dof, fes.GetVDim(),
                             compstride, (fes.GetVDim())*(fes.GetNDofs()),
                             CEED_MEM_HOST, CEED_COPY_VALUES,
                             tp_el_dof.GetData(), restr);
 }

-static void InitRestrictionImpl(const mfem::FiniteElementSpace &fes,
-                                Ceed ceed, CeedElemRestriction *restr)
-{
-   const mfem::FiniteElement *fe = fes.GetFE(0);
-   const mfem::TensorBasisElement * tfe =
-      dynamic_cast<const mfem::TensorBasisElement *>(fe);
-   if ( tfe && tfe->GetDofMap().Size()>0 ) // Native ordering using dof_map
-   {
-      InitNativeRestr(fes, ceed, restr);
-   }
-   else  // Lexicographic ordering
-   {
-      InitLexicoRestr(fes, ceed, restr);
-   }
-}
-
-static void InitNativeRestrWithIndices(
-   const mfem::FiniteElementSpace &fes,
-   int nelem,
-   const int* indices,
-   Ceed ceed, CeedElemRestriction *restr)
-{
-   const mfem::FiniteElement *fe = fes.GetFE(indices[0]);
-   const int P = fe->GetDof();
-   CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
-   mfem::Array<int> tp_el_dof(nelem*P);
-   const mfem::TensorBasisElement * tfe =
-      dynamic_cast<const mfem::TensorBasisElement *>(fe);
-   Array<int> dofs;
-   const int stride = compstride == 1 ? fes.GetVDim() : 1;
-   const mfem::Array<int>& dof_map = tfe->GetDofMap();
-
-   for (int i = 0; i < nelem; i++)
-   {
-      const int elem_index = indices[i];
-      fes.GetElementDofs(elem_index, dofs);
-      const int el_offset = P * i;
-      for (int j = 0; j < P; j++)
-      {
-         tp_el_dof[j + el_offset] = stride*dofs[dof_map[j]];
-      }
-   }
-
-   CeedElemRestrictionCreate(ceed, nelem, P, fes.GetVDim(),
-                             compstride, (fes.GetVDim())*(fes.GetNDofs()),
-                             CEED_MEM_HOST, CEED_COPY_VALUES,
-                             tp_el_dof.GetData(), restr);
-}
-
-static void InitLexicoRestrWithIndices(
-   const mfem::FiniteElementSpace &fes,
-   int nelem,
-   const int* indices,
-   Ceed ceed, CeedElemRestriction *restr)
-{
-   const mfem::FiniteElement *fe = fes.GetFE(indices[0]);
-   const int P = fe->GetDof();
-   CeedInt compstride = fes.GetOrdering()==Ordering::byVDIM ? 1 : fes.GetNDofs();
-   mfem::Array<int> tp_el_dof(nelem*P);
-   Array<int> dofs;
-   const int stride = compstride == 1 ? fes.GetVDim() : 1;
-
-   for (int i = 0; i < nelem; i++)
-   {
-      const int elem_index = indices[i];
-      fes.GetElementDofs(elem_index, dofs);
-      const int el_offset = P * i;
-      for (int j = 0; j < P; j++)
-      {
-         tp_el_dof[j + el_offset] = stride*dofs[j];
-      }
-   }
-
-   CeedElemRestrictionCreate(ceed, nelem, P, fes.GetVDim(),
-                             compstride, (fes.GetVDim())*(fes.GetNDofs()),
-                             CEED_MEM_HOST, CEED_COPY_VALUES,
-                             tp_el_dof.GetData(), restr);
-}
-
-static void InitRestrictionWithIndicesImpl(
-   const mfem::FiniteElementSpace &fes,
-   int nelem,
-   const int* indices,
-   Ceed ceed, CeedElemRestriction *restr)
-{
-   const mfem::FiniteElement *fe = fes.GetFE(indices[0]);
-   const mfem::TensorBasisElement * tfe =
-      dynamic_cast<const mfem::TensorBasisElement *>(fe);
-   if ( tfe && tfe->GetDofMap().Size()>0 ) // Native ordering using dof_map
-   {
-      InitNativeRestrWithIndices(fes, nelem, indices, ceed, restr);
-   }
-   else  // Lexicographic ordering
-   {
-      InitLexicoRestrWithIndices(fes, nelem, indices, ceed, restr);
-   }
-}
-
-static void InitCoeffRestrictionWithIndicesImpl(
-   const mfem::FiniteElementSpace &fes,
-   int nelem,
-   const int* indices,
-   int nquads,
-   int ncomp,
-   Ceed ceed,
-   CeedElemRestriction *restr)
-{
-   mfem::Array<int> tp_el_dof(nelem*nquads);
-   const int stride_quad = ncomp;
-   const int stride_elem = ncomp*nquads;
-   // TODO generalize to support different #quads
-   for (int i = 0; i < nelem; i++)
-   {
-      const int elem_index = indices[i];
-      const int el_offset = elem_index * stride_elem;
-      for (int j = 0; j < nquads; j++)
-      {
-         tp_el_dof[j + nquads * i] = j * stride_quad + el_offset;
-      }
-   }
-   CeedElemRestrictionCreate(ceed, nelem, nquads, ncomp, 1,
-                             ncomp*fes.GetNE()*nquads,
-                             CEED_MEM_HOST, CEED_COPY_VALUES,
-                             tp_el_dof.GetData(), restr);
-}
-
 void InitStridedRestriction(const mfem::FiniteElementSpace &fes,
                            CeedInt nelem, CeedInt nqpts, CeedInt qdatasize,
                            const CeedInt *strides,
@@ -232,60 +136,17 @@ void InitRestriction(const FiniteElementSpace &fes,
   RestrKey restr_key(&fes, nelem, P, ncomp, restr_type::Standard);
   auto restr_itr = mfem::internal::ceed_restr_map.find(restr_key);

-   // Init or retrieve key values
+   // Init or retreive key values
   if (restr_itr == mfem::internal::ceed_restr_map.end())
   {
-      InitRestrictionImpl(fes, ceed, restr);
-      mfem::internal::ceed_restr_map[restr_key] = *restr;
-   }
-   else
-   {
-      *restr = restr_itr->second;
-   }
-}
-
-void InitRestrictionWithIndices(const FiniteElementSpace &fes,
-                                int nelem,
-                                const int* indices,
-                                Ceed ceed,
-                                CeedElemRestriction *restr)
-{
-   // Check for FES -> basis, restriction in hash tables
-   const mfem::FiniteElement *fe = fes.GetFE(indices[0]);
-   const int P = fe->GetDof();
-   const int ncomp = fes.GetVDim();
-   RestrKey restr_key(&fes, nelem, P, ncomp, restr_type::Standard);
-   auto restr_itr = mfem::internal::ceed_restr_map.find(restr_key);
-
-   // Init or retrieve key values
-   if (restr_itr == mfem::internal::ceed_restr_map.end())
-   {
-      InitRestrictionWithIndicesImpl(fes, nelem, indices, ceed, restr);
-      mfem::internal::ceed_restr_map[restr_key] = *restr;
-   }
-   else
-   {
-      *restr = restr_itr->second;
-   }
-}
-
-void InitCoeffRestrictionWithIndices(const FiniteElementSpace &fes,
-                                     int nelem,
-                                     const int* indices,
-                                     int nquads,
-                                     int ncomp,
-                                     Ceed ceed,
-                                     CeedElemRestriction *restr)
-{
-   // Check for FES -> basis, restriction in hash tables
-   RestrKey restr_key(&fes, nelem, nquads, ncomp, restr_type::Coeff);
-   auto restr_itr = mfem::internal::ceed_restr_map.find(restr_key);
-
-   // Init or retrieve key values
-   if (restr_itr == mfem::internal::ceed_restr_map.end())
-   {
-      InitCoeffRestrictionWithIndicesImpl(fes, nelem, indices, nquads, ncomp,
-                                          ceed, restr);
+      if (UsesTensorBasis(fes))
+      {
+         InitTensorRestriction(fes, ceed, restr);
+      }
+      else
+      {
+         InitNonTensorRestriction(fes, ceed, restr);
+      }
      mfem::internal::ceed_restr_map[restr_key] = *restr;
   }
   else
@@ -21,63 +21,37 @@ namespace ceed
 {

 #ifdef MFEM_USE_CEED
-/** @brief Initialize a CeedElemRestriction for non-mixed meshes.
-
-    @param[in] fes Input finite element space.
-    @param[in] ceed Input Ceed object.
-    @param[out] restr The address of the initialized CeedElemRestriction object.
-*/
-void InitRestriction(const FiniteElementSpace &fes,
-                     Ceed ceed,
-                     CeedElemRestriction *restr);
-
-/** @brief Initialize a CeedElemRestriction for mixed meshes.
-
-    @param[in] fes The finite element space.
-    @param[in] ceed The Ceed object.
-    @param[in] nelem The number of elements.
-    @param[in] indices The indices of the elements of same type in the
-                       `FiniteElementSpace`.
-    @param[out] restr The `CeedElemRestriction` to initialize. */
-void InitRestrictionWithIndices(const FiniteElementSpace &fes,
-                                int nelem,
-                                const int* indices,
-                                Ceed ceed,
-                                CeedElemRestriction *restr);
-
-/** @brief Initialize a strided CeedElemRestriction
-
-    @param[in] nelem is the number of elements.
-    @param[in] nqpts is the total number of quadrature points.
-    @param[in] qdatasize is the number of data per quadrature point.
-    @param[in] strides Array for strides between [nodes, components, elements].
+/// @brief Initialize a strided CeedElemRestriction
+/** @a nelem is the number of elements,
+    @a nqpts is the total number of quadrature points
+    @a qdatasize is the number of data per quadrature point
+    @a strides Array for strides between [nodes, components, elements].
    Data for node i, component j, element k can be found in the L-vector at
    index i*strides[0] + j*strides[1] + k*strides[2]. CEED_STRIDES_BACKEND may
-    be used with vectors created by a Ceed backend.
-    @param[out] restr The `CeedElemRestriction` to initialize. */
+    be used with vectors created by a Ceed backend. */
 void InitStridedRestriction(const mfem::FiniteElementSpace &fes,
                            CeedInt nelem, CeedInt nqpts, CeedInt qdatasize,
                            const CeedInt *strides,
                            CeedElemRestriction *restr);

-/** @brief Initialize a CeedElemRestriction for a mfem::Coefficient on a mixed
-    mesh.
+/** @brief Initialize a CeedElemRestriction.
+ *
+ * @param[in] fes Input finite element space.
+ * @param[in] ceed Input Ceed object.
+   @param[out] restr The address of the initialized CeedElemRestriction object.
+*/
+void InitRestriction(const FiniteElementSpace &fes,
+                     Ceed ceed,
+                     CeedElemRestriction *restr);

-    @param[in] fes The finite element space.
-    @param[in] nelem is the number of elements.
-    @param[in] indices The indices of the elements of same type in the
-                       `FiniteElementSpace`.
-    @param[in] nquads is the total number of quadrature points
-    @param[in] ncomp is the number of data per quadrature point
-    @param[in] ceed The Ceed object.
-    @param[out] restr The `CeedElemRestriction` to initialize. */
-void InitCoeffRestrictionWithIndices(const FiniteElementSpace &fes,
-                                     int nelem,
-                                     const int* indices,
-                                     int nquads,
-                                     int ncomp,
-                                     Ceed ceed,
-                                     CeedElemRestriction *restr);
+/** @brief Initialize a CeedElemRestriction.
+ *
+ * @param[in] fes Input finite element space.
+ * @param[in] ceed Input Ceed object.
+   @param[out] restr The address of the initialized CeedElemRestriction object.
+*/
+void InitTensorRestriction(const FiniteElementSpace &fes,
+                           Ceed ceed, CeedElemRestriction *restr);

 #endif

@@ -99,34 +99,6 @@ void InitBasisAndRestriction(const FiniteElementSpace &fes,
   InitRestriction(fes, ceed, restr);
 }

-void InitBasisAndRestrictionWithIndices(const FiniteElementSpace &fes,
-                                        const IntegrationRule &irm,
-                                        int nelem,
-                                        const int* indices,
-                                        Ceed ceed, CeedBasis *basis,
-                                        CeedElemRestriction *restr)
-{
-   InitBasisWithIndices(fes, irm, nelem, indices, ceed, basis);
-   InitRestrictionWithIndices(fes, nelem, indices, ceed, restr);
-}
-
-void InitBasisAndRestriction(const FiniteElementSpace &fes,
-                             const IntegrationRule &irm,
-                             int nelem,
-                             const int* indices,
-                             Ceed ceed, CeedBasis *basis,
-                             CeedElemRestriction *restr)
-{
-   if (indices)
-   {
-      InitBasisAndRestrictionWithIndices(fes,irm,nelem,indices,ceed,basis,restr);
-   }
-   else
-   {
-      InitBasisAndRestriction(fes,irm,ceed,basis,restr);
-   }
-}
-
 // Assumes a tensor-product operator with one active field
 int CeedOperatorGetActiveField(CeedOperator oper, CeedOperatorField *field)
 {
@@ -186,66 +158,6 @@ int CeedOperatorGetActiveField(CeedOperator oper, CeedOperatorField *field)
   return 0;
 }

-template <>
-const IntegrationRule & GetRule<MassIntegrator>(
-   const MassIntegrator &integ,
-   const FiniteElement &trial_fe,
-   const FiniteElement &test_fe,
-   ElementTransformation &trans)
-{
-   return MassIntegrator::GetRule(trial_fe, test_fe, trans);
-}
-
-template <>
-const IntegrationRule & GetRule<VectorMassIntegrator>(
-   const VectorMassIntegrator &integ,
-   const FiniteElement &trial_fe,
-   const FiniteElement &test_fe,
-   ElementTransformation &trans)
-{
-   return MassIntegrator::GetRule(trial_fe, test_fe, trans);
-}
-
-template <>
-const IntegrationRule & GetRule<ConvectionIntegrator>(
-   const ConvectionIntegrator &integ,
-   const FiniteElement &trial_fe,
-   const FiniteElement &test_fe,
-   ElementTransformation &trans)
-{
-   return ConvectionIntegrator::GetRule(trial_fe, test_fe, trans);
-}
-
-template <>
-const IntegrationRule & GetRule<VectorConvectionNLFIntegrator>(
-   const VectorConvectionNLFIntegrator &integ,
-   const FiniteElement &trial_fe,
-   const FiniteElement &test_fe,
-   ElementTransformation &trans)
-{
-   return VectorConvectionNLFIntegrator::GetRule(trial_fe, trans);
-}
-
-template <>
-const IntegrationRule & GetRule<DiffusionIntegrator>(
-   const DiffusionIntegrator &integ,
-   const FiniteElement &trial_fe,
-   const FiniteElement &test_fe,
-   ElementTransformation &trans)
-{
-   return DiffusionIntegrator::GetRule(trial_fe, test_fe);
-}
-
-template <>
-const IntegrationRule & GetRule<VectorDiffusionIntegrator>(
-   const VectorDiffusionIntegrator &integ,
-   const FiniteElement &trial_fe,
-   const FiniteElement &test_fe,
-   ElementTransformation &trans)
-{
-   return DiffusionIntegrator::GetRule(trial_fe, test_fe);
-}
-
 std::string ceed_path;

 const std::string &GetCeedPath()
@@ -26,9 +26,7 @@
 namespace mfem
 {

-class FiniteElement;
 class FiniteElementSpace;
-class ElementTransformation;
 class IntegrationRule;
 class Vector;

@@ -57,51 +55,15 @@ void RemoveBasisAndRestriction(const mfem::FiniteElementSpace *fes);
 /// Initialize a CeedVector from an mfem::Vector
 void InitVector(const mfem::Vector &v, CeedVector &cv);

-/** @brief Initialize a CeedBasis and a CeedElemRestriction based on an
-    mfem::FiniteElementSpace @a fes, and an mfem::IntegrationRule @a ir.
-
-    @param[in] fes The finite element space.
-    @param[in] ir The integration rule.
-    @param[in] ceed The Ceed object.
-    @param[out] basis The `CeedBasis` to initialize.
-    @param[out] restr The `CeedElemRestriction` to initialize.
-
-    @warning Only for non-mixed finite element spaces. */
+/** Initialize a CeedBasis and a CeedElemRestriction based on an
+    mfem::FiniteElementSpace @a fes, and an mfem::IntegrationRule @a ir. */
 void InitBasisAndRestriction(const mfem::FiniteElementSpace &fes,
                             const mfem::IntegrationRule &ir,
                             Ceed ceed, CeedBasis *basis,
                             CeedElemRestriction *restr);

-/** @brief Initialize a CeedBasis and a CeedElemRestriction based on an
-    mfem::FiniteElementSpace @a fes, and an mfem::IntegrationRule @a ir,
-    and a list of @a nelem elements of indices @a indices.
-
-    @param[in] fes The finite element space.
-    @param[in] ir The integration rule.
-    @param[in] nelem The number of elements.
-    @param[in] indices The indices of the elements of same type in the
-                       `FiniteElementSpace`. If `indices == nullptr`, assumes
-                       that the `FiniteElementSpace` is not mixed.
-    @param[in] ceed The Ceed object.
-    @param[out] basis The `CeedBasis` to initialize.
-    @param[out] restr The `CeedElemRestriction` to initialize. */
-void InitBasisAndRestriction(const FiniteElementSpace &fes,
-                             const IntegrationRule &ir,
-                             int nelem,
-                             const int* indices,
-                             Ceed ceed, CeedBasis *basis,
-                             CeedElemRestriction *restr);
-
 int CeedOperatorGetActiveField(CeedOperator oper, CeedOperatorField *field);

-
-template <typename Integrator>
-const IntegrationRule & GetRule(
-   const Integrator &integ,
-   const FiniteElement &trial_fe,
-   const FiniteElement &test_fe,
-   ElementTransformation &Trans);
-
 /// Return the path to the libCEED q-function headers.
 const std::string &GetCeedPath();

@@ -125,7 +87,7 @@ struct BasisHash
 };
 using BasisMap = std::unordered_map<const BasisKey, CeedBasis, BasisHash>;

-enum restr_type {Standard, Strided, Coeff};
+enum restr_type {Standard, Strided};

 // Hash table for CeedElemRestriction
 using RestrKey =
@@ -155,8 +117,6 @@ namespace internal
 {

 #ifdef MFEM_USE_CEED
-/** @warning These maps have a tendency to create bugs when adding new "types"
-    of CeedBasis and CeedElemRestriction. */
 extern ceed::BasisMap ceed_basis_map;
 extern ceed::RestrMap ceed_restr_map;
 #endif
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Tucker Babcock	0c6f51852b	make style	2022-06-30 12:33:11 -06:00
Tucker Babcock	737883992d	update linear form move ctor/assign to correctly move recently added LinearFormExtension member ext	2022-06-30 12:30:45 -06:00
Tucker Babcock	d896f164fe	Merge branch 'master' into move-nlf-dev	2022-06-30 12:27:48 -06:00
Tucker Babcock	d60c544e0f	merge master into move-nlf-dev	2022-03-24 10:11:05 -06:00
Tucker Babcock	c5fbe6cb27	correctly handle moving the bilinear form integ markers	2022-01-19 10:06:55 -07:00
Tucker Babcock	981304f4f3	add element_matrices to bilinearform move constructor, I had forgotten it before	2022-01-17 15:14:17 -07:00
Tucker Babcock	196125ef2e	make style	2022-01-15 20:08:31 -07:00
Tucker Babcock	57d196a963	reorder bilinear form constructor	2022-01-15 20:07:54 -07:00
Tucker Babcock	e9b4e2534b	first draft of bilinear form move semantics are complete, needs to be tested though	2022-01-15 19:18:33 -07:00
Tucker Babcock	96686455f6	Merge branch 'master' into move-nlf-dev	2022-01-15 18:27:24 -07:00
Tucker Babcock	f363bf7468	working on adding move semantics to bilinear forms	2022-01-15 18:27:02 -07:00
Tucker Babcock	bd6ac34b97	set assembly level of other nonlinear form to LEGACY in move assignment operator	2021-11-19 13:49:49 -07:00
Tucker Babcock	050bc5aacf	update nonlinear form move assignment/ctors to use mfem::Swap for integrator and marker arrays, also add support for moving domain_integs_marker	2021-11-19 13:46:57 -07:00
Tucker Babcock	751e200d83	make style	2021-11-19 13:14:41 -07:00
Tucker Babcock	d9fda017d4	merge master into move-nlf-dev	2021-11-19 13:13:36 -07:00
Tucker Babcock	2516c54386	add support for nonlinear form domain integrators restricted to certain attributes	2021-11-19 13:09:30 -07:00
Tucker Babcock	54e26b603a	make style	2021-11-19 12:19:16 -07:00
Tucker Babcock	d189d70da4	renamed nonlinear form integrator arrays to match bilinear form and linear form with more expressive names. Added method to add domain integrator restricted to certain element types for nonlinear form to match method for bilinear form. Need to update Mult, GetGradient, GetEnergy to use this and check for attributes	2021-11-19 12:18:36 -07:00
Tucker Babcock	7fd3ed3a94	make style	2021-11-09 15:23:59 -07:00
Tucker Babcock	cce25fced8	added LinearFrom and ParLinearForm move ctors and assignment operators, but need to test	2021-11-09 15:20:21 -07:00
Tucker Babcock	e25fb02eee	add non-default move ctor/assign for ParNonlinearForm	2021-10-15 09:12:49 -06:00
Tucker Babcock	c4f67c7dff	explicitly defaulting ParNonlinearForm move ctor/assign	2021-10-14 16:40:21 -06:00
Tucker Babcock	e5e92be833	make style	2021-10-14 15:26:30 -06:00
Tucker Babcock	88370b7fdc	reverting all operator changes, returning *this in nlf move assignment operator	2021-10-14 15:20:11 -06:00
Tucker Babcock	fc363ca778	added move ctors and assign operators to NonlinearForm classes, and explicitly delete copy ctor and assignment for NonlinearForm	2021-10-14 14:46:05 -06:00