updting io benchmark

Merge branch 'PCFinalProject' of github.com:mfem/mfem into PCFinalProject
adding io benchmark
2020-05-05 15:44:44 -04:00 · 2020-05-04 11:14:22 -07:00 · 2020-05-04 11:13:42 -07:00 · 2020-05-04 12:42:18 -04:00 · 2020-05-04 12:42:13 -04:00 · 2020-05-04 09:40:57 -07:00
300 changed files with 7646 additions and 39266 deletions
@@ -15,10 +15,8 @@ install:
 - msmpisdk.msi /passive
 - set PATH=C:\Program Files\Microsoft MPI\Bin;%PATH%

-# 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'
+# Install METIS
+- ps: Start-FileDownload 'http://glaros.dtc.umn.edu/gkhome/fetch/sw/metis/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
 - ps: ( get-content "GKlib\gk_arch.h") | % { If ($_.ReadCount -ge 52) {$_ -replace "#ifdef __MSC__","#ifdef DISABLE_THIS_ANCIENT_MSC_CHECK"} Else {$_} } | set-content "GKlib\gk_arch.h"
@@ -28,17 +26,17 @@ install:
 - cd ..

 # Install hypre
- ps: Start-FileDownload 'https://github.com/hypre-space/hypre/archive/v2.19.0.tar.gz'
- 7z x v2.19.0.tar.gz -so | 7z x -si -ttar > nul
- cd hypre-2.19.0/src
- cmake -H. -Bbuild -DMPI_C_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include" -DMPI_C_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include"
+- ps: Start-FileDownload 'https://github.com/hypre-space/hypre/archive/V2-10-0b.tar.gz'
+- 7z x V2-10-0b.tar.gz -so | 7z x -si -ttar > nul
+- cd hypre-2-10-0b
+- cmake -H. -Bbuild -DHYPRE_USING_FEI=OFF -DMPI_C_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include" -DMPI_C_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include"
 - cmake --build build
 - cmake --build build --target install
- cd ../..
+- cd ..

 # MFEM
 before_build:
- cmake -H. -DCMAKE_INSTALL_PREFIX=install -Bbuild_parallel -DMFEM_USE_MPI=TRUE -DMFEM_USE_METIS_5=TRUE -DMPI_CXX_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include" -DHYPRE_DIR=%cd%\hypre-2.19.0\src\hypre -DMETIS_LIBRARIES=%cd%\metis-5.1.0\build\libmetis\Debug\metis.lib -DMETIS_INCLUDE_DIRS=%cd%\metis-5.1.0\include
+- cmake -H. -DCMAKE_INSTALL_PREFIX=install -Bbuild_parallel -DMFEM_USE_MPI=TRUE -DMFEM_USE_METIS_5=TRUE -DMPI_CXX_LIBRARIES="C:\Program Files (x86)\Microsoft SDKs\MPI\Lib\x86\msmpi.lib" -DMPI_CXX_INCLUDE_PATH="C:\Program Files (x86)\Microsoft SDKs\MPI\Include" -DHYPRE_LIBRARIES=%cd%\hypre-2-10-0b\hypre\lib\HYPRE.lib -DHYPRE_INCLUDE_DIRS=%cd%\hypre-2-10-0b\hypre\include -DHYPRE_VERSION=21000 -DMETIS_LIBRARIES=%cd%\metis-5.1.0\build\libmetis\Debug\metis.lib -DMETIS_INCLUDE_DIRS=%cd%\metis-5.1.0\include
 - cmake -H. -DCMAKE_INSTALL_PREFIX=install -Bbuild_serial -DMFEM_USE_MPI=FALSE

 build_script:
@@ -29,8 +29,6 @@ config/sample-runs-build.log
 doc/CodeDocumentation.conf
 doc/CodeDocumentation.html
 doc/CodeDocumentation
-doc/undoc.log
-doc/warnings.log

 # Temporary files created by the tests.
 *.stderr
@@ -122,7 +120,7 @@ examples/sundials/ex16-final.*
 examples/sundials/Example16*

 examples/petsc/ex[1-69]p
-examples/petsc/ex1[0-1]p
+examples/petsc/ex10p

 examples/petsc/mesh.*
 examples/petsc/sol.*
@@ -137,7 +135,6 @@ examples/petsc/Example9*
 examples/petsc/deformed.*
 examples/petsc/velocity.*
 examples/petsc/elastic_energy.*
-examples/petsc/mode_*

 examples/pumi/ex1
 examples/pumi/ex[126]p
@@ -147,11 +144,6 @@ examples/hiop/ex9-mesh.*
 examples/hiop/ex9-init.*
 examples/hiop/ex9-final.*

-
-examples/ex71
-examples/ex71p
-examples/Example71*
-
 examples/pumi/refined.mesh
 examples/pumi/sol.gf
 examples/pumi/mesh.*
@@ -175,7 +167,6 @@ miniapps/meshing/twist
 miniapps/meshing/mesh-explorer
 miniapps/meshing/shaper
 miniapps/meshing/extruder
-miniapps/meshing/trimmer
 miniapps/meshing/mesh-optimizer
 miniapps/meshing/pmesh-optimizer
 miniapps/meshing/minimal-surface
@@ -189,7 +180,6 @@ miniapps/meshing/mesh-explorer.mesh
 miniapps/meshing/partitioning.txt
 miniapps/meshing/shaper.mesh
 miniapps/meshing/extruder.mesh
-miniapps/meshing/trimmer.mesh
 miniapps/meshing/optimized*
 miniapps/meshing/perturbed*

@@ -249,9 +239,6 @@ miniapps/navier/navier_3dfoc
 miniapps/navier/tgv_out*.txt
 miniapps/navier/*_output

-miniapps/adjoint/cvsRoberts_ASAi_dns
-miniapps/adjoint/adjoint_advection_diffusion
-
 # Unit test binary and outputs
 tests/unit/output_meshes
 tests/unit/unit_tests
@@ -11,20 +11,13 @@

 language: cpp

-os: linux
-dist: bionic
+sudo: false

 stages:
  - checks
  - tests
  - optional

-env:
-  global:
-    - HYPRE_ARCHIVE=v2.19.0.tar.gz
-      HYPRE_URL=https://github.com/hypre-space/hypre/archive/$HYPRE_ARCHIVE
-      HYPRE_TOP_DIR=hypre-2.19.0
-
 jobs:
  include:

@@ -37,7 +30,6 @@ jobs:

    - stage: checks
      os: linux
-      dist: xenial
      name: "code-style"
      addons:
        apt:
@@ -56,6 +48,9 @@ jobs:
          packages:
            - doxygen
            - graphviz
+            - mpich
+            - libmpich-dev
+      env: MPI=YES
      script:
        - cd ${TRAVIS_BUILD_DIR}
        - cd tests/scripts
@@ -70,24 +65,13 @@ jobs:
            - mpich
            - libmpich-dev
      env: MPI=YES
-      before_script:
+      script:
        - cd ${TRAVIS_BUILD_DIR}
-        - mpicxx -v
        - make config MFEM_USE_MPI=YES MFEM_MPI_NP=2
        - make all -j3
        - make test-noclean
-      script:
        - cd tests/scripts
        - ./runtest gitignore
-      cache:
-        ccache: true
-        directories:
-          - $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
-          - $TRAVIS_BUILD_DIR/../metis-4.0
-      before_cache:
-        - cd $TRAVIS_BUILD_DIR/../metis-4.0;
-          mv libmetis.a Lib ..; rm -rf * ; mv ../libmetis.a ../Lib .;
-          rm -f Lib/*.{c,o}

    # ========================
    #  Optional Checks/Tests
@@ -96,7 +80,6 @@ jobs:

    - stage: optional
      name: "branch-history"
-      if: branch != next
      # need full git history for the binary/big files check
      git:
        depth: false
@@ -125,8 +108,6 @@ jobs:
           MPI=NO
           CODECOV=NO
           MFEM_TEST_TARGET=check
-      cache:
-        ccache: true

    - os: linux
      compiler: gcc
@@ -135,8 +116,6 @@ jobs:
           MPI=NO
           CODECOV=NO
           MFEM_TEST_TARGET=test
-      cache:
-        ccache: true

    - os: linux
      compiler: gcc
@@ -160,9 +139,9 @@ jobs:
           MFEM_TEST_TARGET=check
           NPROCS=2
      cache:
-        ccache: true
        directories:
-          - $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/lib
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/include
          - $TRAVIS_BUILD_DIR/../metis-4.0
      before_cache:
        - cd $TRAVIS_BUILD_DIR/../metis-4.0;
@@ -191,9 +170,9 @@ jobs:
           MFEM_TEST_TARGET=test
           NPROCS=2
      cache:
-        ccache: true
        directories:
-          - $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/lib
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/include
          - $TRAVIS_BUILD_DIR/../metis-4.0
      before_cache:
        - cd $TRAVIS_BUILD_DIR/../metis-4.0;
@@ -216,16 +195,16 @@ jobs:
        - cd ${TRAVIS_BUILD_DIR}/build
        - cmake ..
             -DMFEM_USE_MPI=ON
-             -DHYPRE_DIR=${TRAVIS_BUILD_DIR}/../$HYPRE_TOP_DIR/src/hypre
+             -DHYPRE_DIR=${TRAVIS_BUILD_DIR}/../hypre-2.10.0b/src/hypre
             -DMFEM_MPI_NP=$NPROCS
        - make -j3 mfem examples
        - cd ${TRAVIS_BUILD_DIR}/build/tests/unit
        - make -j3
        - ctest --output-on-failure
      cache:
-        ccache: true
        directories:
-          - $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/lib
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/include
          - $TRAVIS_BUILD_DIR/../metis-4.0
      before_cache:
        - cd $TRAVIS_BUILD_DIR/../metis-4.0;
@@ -241,43 +220,27 @@ jobs:
    # - parallel

    - os: osx
-      osx_image: xcode11.2
+      # osx_image: xcode7.3
      compiler: clang
      name: "Mac: Serial + Debug"
-      addons:
-        homebrew:
-          packages:
-            - ccache
      env: DEBUG=YES
           MPI=NO
           CODECOV=NO
           MFEM_TEST_TARGET=check
-      cache:
-        ccache: true

    - os: osx
-      osx_image: xcode11.2
+      # osx_image: xcode7.3
      compiler: clang
      name: "Mac: Serial"
-      addons:
-        homebrew:
-          packages:
-            - ccache
      env: DEBUG=NO
           MPI=NO
           CODECOV=NO
           MFEM_TEST_TARGET=test
-      cache:
-        ccache: true

    - os: osx
-      osx_image: xcode11.2
+      # osx_image: xcode7.3
      compiler: clang
      name: "Mac: Parallel + Debug"
-      addons:
-        homebrew:
-          packages:
-            - ccache
      env: DEBUG=YES
           MPI=YES
           CODECOV=NO
@@ -285,9 +248,9 @@ jobs:
           NPROCS=4
           TMPDIR=/tmp
      cache:
-        ccache: true
        directories:
-          - $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/lib
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/include
          - $TRAVIS_BUILD_DIR/../metis-4.0
          - $HOME/local-cached
      before_cache:
@@ -296,13 +259,9 @@ jobs:
          rm -f Lib/*.{c,o}

    - os: osx
-      osx_image: xcode11.2
+      # osx_image: xcode7.3
      compiler: clang
      name: "Mac: Parallel"
-      addons:
-        homebrew:
-          packages:
-            - ccache
      env: DEBUG=NO
           MPI=YES
           CODECOV=YES
@@ -310,9 +269,9 @@ jobs:
           NPROCS=4
           TMPDIR=/tmp
      cache:
-        ccache: true
        directories:
-          - $TRAVIS_BUILD_DIR/../$HYPRE_TOP_DIR/src/hypre
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/lib
+          - $TRAVIS_BUILD_DIR/../hypre-2.10.0b/src/hypre/include
          - $TRAVIS_BUILD_DIR/../metis-4.0
          - $HOME/local-cached
      before_cache:
@@ -326,19 +285,14 @@ before_install:
   #      brew install open-mpi;
   #   fi

-   # Disable ccache while building dependencies that are cached:
-   - echo "before \$PATH = $PATH";
-     export PATH=${PATH//\/usr\/lib\/ccache:/};
-     echo "after  \$PATH = $PATH"
-
-   # On Mac OS X, build and cache OpenMPI 2.1.6:
+   # On Mac OS X, build and cache OpenMPI 2.1.1:
   - if [ $TRAVIS_OS_NAME == "osx" ] && [ $MPI == "YES" ]; then
        if [ ! -e $HOME/local-cached/bin/mpicc ]; then
           mkdir -p $HOME/builds && cd $HOME/builds &&
-           wget https://download.open-mpi.org/release/open-mpi/v2.1/openmpi-2.1.6.tar.bz2 &&
-           tar jxf openmpi-2.1.6.tar.bz2 &&
+           wget https://www.open-mpi.org/software/ompi/v2.1/downloads/openmpi-2.1.1.tar.bz2 &&
+           tar jxf openmpi-2.1.1.tar.bz2 &&
           mkdir openmpi-build && cd openmpi-build &&
-           ../openmpi-2.1.6/configure --prefix=$HOME/local-cached &&
+           ../openmpi-2.1.1/configure --prefix=$HOME/local-cached &&
           make -j3 all && make install;
        fi;
        PATH=$HOME/local-cached/bin:$PATH;
@@ -383,28 +337,26 @@ install:

   # hypre
   - if [ $MPI == "YES" ]; then
-        if [ ! -e $HYPRE_TOP_DIR/src/hypre/lib/libHYPRE.a ]; then
-           wget $HYPRE_URL;
-           rm -rf $HYPRE_TOP_DIR;
-           tar xvzf $HYPRE_ARCHIVE;
-           cd $HYPRE_TOP_DIR/src;
-           ./configure --disable-fortran CC=mpicc CXX=mpic++;
+        if [ ! -e hypre-2.10.0b/src/hypre/lib/libHYPRE.a ]; then
+           wget https://computation.llnl.gov/project/linear_solvers/download/hypre-2.10.0b.tar.gz --no-check-certificate;
+           rm -rf hypre-2.10.0b;
+           tar xvzf hypre-2.10.0b.tar.gz;
+           cd hypre-2.10.0b/src;
+           ./configure --disable-fortran --without-fei CC=mpicc CXX=mpic++;
           make -j3;
           cd ../..;
        else
-            echo "Reusing cached $HYPRE_TOP_DIR/";
+            echo "Reusing cached hypre-2.10.0b/";
        fi;
-        ln -s $HYPRE_TOP_DIR hypre;
+        ln -s hypre-2.10.0b hypre;
     else
        echo "Serial build, not using hypre";
     fi

-   # METIS, use a mirror because the original source server is not always up.
-   # Original url:
-   #    http://glaros.dtc.umn.edu/gkhome/fetch/sw/metis/OLD/metis-4.0.3.tar.gz
+   # METIS
   - if [ $MPI == "YES" ]; then
        if [ ! -e metis-4.0/libmetis.a ]; then
-           wget https://mfem.github.io/tpls/metis-4.0.3.tar.gz;
+           wget http://glaros.dtc.umn.edu/gkhome/fetch/sw/metis/OLD/metis-4.0.3.tar.gz;
           tar xvzf metis-4.0.3.tar.gz;
           make -j3 -C metis-4.0.3/Lib CC="$CC" OPTFLAGS="-O2";
           rm -rf metis-4.0;
@@ -414,26 +366,12 @@ install:
        fi;
     fi

-   # Re-enable ccache on linux; enable ccache on mac os:
-   - if [ $TRAVIS_OS_NAME == "linux" ]; then
-        export PATH="/usr/lib/ccache:$PATH";
-     else
-        if [ $TRAVIS_OS_NAME == "osx" ]; then
-           export PATH="/usr/local/opt/ccache/libexec:$PATH";
-        fi;
-     fi
-
-   - printf "which \$CC  = "; which $CC;
-     printf "which \$CXX = "; which $CXX
-
 script:
   # Compiler
   - if [ $MPI == "YES" ]; then
        export MYCXX=mpic++;
-        export MAKE_CXX_FLAG=MPICXX=$MYCXX;
     else
        export MYCXX="$CXX";
-        export MAKE_CXX_FLAG=CXX=$MYCXX;
     fi

   # Print the compiler version
@@ -446,12 +384,12 @@ script:
     if [ "$CODECOV" == "YES" ]; then
        CPPFLAGS="--coverage -g";
     fi;
-     if [ "$TRAVIS_OS_NAME" != "linux" ] || [ "$DEBUG" == "YES" ]; then
-        CPPFLAGS+=" -pedantic -Wall -Werror";
+     if [ "$CXX" == "clang++" ]; then
+        export MFEM_PERF_SW=clang;
     fi

   # Configure the library
-   - make config MFEM_USE_MPI=$MPI MFEM_DEBUG=$DEBUG $MAKE_CXX_FLAG
+   - make config MFEM_USE_MPI=$MPI MFEM_DEBUG=$DEBUG MFEM_CXX="$MYCXX"
        MFEM_MPI_NP=$NPROCS CPPFLAGS="$CPPFLAGS"
   # Show the configuration
   - make info
@@ -23,43 +23,10 @@ Meshing improvements
  Hessian for r-adaptivity using discrete fields, and allows use of skewness
  and orientation based metrics.

- Added support for r-adaptivity with more than one discrete field. This allows
-  the user to specify different discrete functions for controlling the
-  size, aspect-ratio, orientation, and skew of elements in the mesh.
-
- Added TMOP capability for approximate tangential mesh relaxation.
-
- Added support for reading periodic meshes in Gmsh format (version 2.2). See
-  for example the periodic-annulus-sector and periodic-torus-sector files in
-  the data directory.
-
- Added complete action of the TMOP Integrator to account for the spatial
-  derivatives of discrete and analytic targets.
-
-Performance improvements
------------------------
- Added support for explicit vectorization in the high-performance templated
-  code, which can now take advantage of specific intrinsics classes on the
-  following architectures:
-    - x86 (SSE/AVX/AVX2/AVX512),
-    - Power8 & Power9 (VSX),
-    - BG/Q (QPX).
-  These are disabled by default, and can be enabled with MFEM_USE_SIMD=YES.
-  See the new file linalg/simd.hpp and the new directory linalg/simd.
-
 Improved GPU capabilities
 -------------------------
 - Added support for Chebyshev accelerated polynomial smoother on GPU.

- Optimized AMD/HIP kernel support.
-
- Added a Full Assembly mode compatible with Device kernel execution. This
-  assembly level builds on top of the current Element Assembly kernels to
-  compute a global sparse matrix. All integrators supported by element assembly
-  are also supported by full assembly. See the '-fa' option in Example 9.
-
- Added support for BlockOperator on GPU. See the updated Example 5.
-
 Discretization improvements
 ---------------------------
 - Added support for matrix-free interpolation and restriction operators between
@@ -68,26 +35,6 @@ Discretization improvements

 - Added support for simplices in GSLIB-FindPoints.

- Added support for H1 and L2 element matrix assembly in the mass, convection,
-  diffusion, transpose, and the face DG trace integrators. This is compatible
-  with GPU device execution and is illustrated in Example 9/9p, see the option
-  '-ea'. When enabled, this level of assembly stores independent dense matrices
-  for the elements, and independent dense matrices for the faces in the DG case.
-
- Added new partial assembly kernels for H(div) bilinear forms, as well as
-  VectorFEDivergenceIntegrator.
-
- Improved the documentation of the GridFunction GetValue and GetVectorValue
-  methods. Expanded the GetValue and GetVectorValue methods which accept an
-  ElementTransformation argument to support evaluation on boundary elements
-  and, in the continuous field case, arbitrary mesh edges and faces.
-
- Added new coefficient and vector coefficient classes for QuadratureFunctions.
-  Additionally, new LinearForm integrators were also added which make use of
-  these new QuadratureFunction coefficient classes.
-
- Added support face integrals on the boundaries of NURBS meshes.
-
 Linear and nonlinear solvers
 ----------------------------
 - Added power method to iteratively estimate the largest eigenvalue and the
@@ -96,23 +43,6 @@ Linear and nonlinear solvers
 - Added initial support for h- and p-multigrid solvers and preconditioners for
  matrix-based and matrix-free discretizations with basic GPU capability.

- Added a new IterativeSolverMonitor class that allows to monitor the residual
-  and solution during the solving process of an IterativeSolver after every
-  iteration.
-
- Added support for the CVODES package in SUNDIALS which provides ODE
-  solvers with sensitivity analysis capabilities. See the CVODESSolver
-  class and the new adjoint miniapps below.
-
- Block arrays of parallel matrices can now be merged into a single parallel
-  matrix with the function HypreParMatrixFromBlocks. This could be useful for
-  solving block systems with parallel direct solvers such as STRUMPACK.
-
- In SLISolver, changed the residual inner product from (Br,r) to (Br,Br) so the
-  solver can work with non-SPD preconditioner B.
-
- Added support for the SLEPc eigensolver package.
-
 New and updated examples and miniapps
 -------------------------------------
 - Added a new example, Example 25/25p, to demonstrate the use of a Perfectly
@@ -122,25 +52,11 @@ New and updated examples and miniapps
 - Added a new Example 26/26p to demonstrate the construction of a matrix-free
  geometric and p-multigrid preconditioner for the Laplace problem.

- Added a new example, Example 27/27p, to demonstrate the enforcement of various
-  boundary conditions with the Laplace operator. The example shows the procedure
-  for applying Dirichlet, Neumann (both homogeneous and inhomogeneous), Robin,
-  and periodic boundary conditions with either H1 or DG discretizations.
-
- Added a new miniapp, Navier, that solves the time-dependent Navier-Stokes
-  equations of incompressible fluid dynamics. See the miniapps/navier directory
-  for more details.
-
- Added a new miniapps/adjoint directory with two miniapps demonstrating how to
-  solve adjoint problems in MFEM using the CVODES package in SUNDIALS. Both of
-  these miniapps require the MFEM_USE_SUNDIALS configuration option.
-  * The cvsRoberts_ASAi_dns miniapp solves a backward adjoint problem for a
-    system of ODEs, evaluating both forward and adjoint quadratures in serial.
-  * The adjoint_advection_diffusion miniapp solves a backward adjoint problem
-    for an advection diffusion PDE, evaluating adjoint quadratures in parallel.
-
- Ported Example 11p to SLEPc, to demonstrate solving the Laplace eigenvalue
-  equation with the shift-and-invert spectral transformation method.
+- Added a new example, Example 27/27p, to demonstrate the enforcement of
+  various boundary conditions with the Laplace operator.  The example shows the
+  procedures for applying Dirichlet, Neumann (both homogeneous and
+  inhomogeneous), Robin, and periodic boundary conditions with either H1 or DG
+  discretizations.

 - Added a simple meshing miniapp, Twist, which demonstrates MFEM's strategy of
  stitching together opposite surfaces of a mesh to create a topologically
@@ -149,22 +65,6 @@ New and updated examples and miniapps
 - Added a new meshing miniapp, Minimal Surface, which solves Plateau's problem:
  the Dirichlet problem for the minimal surface equation.

- Added partial assembly support to Example 4/4p and Example 5/5p, with diagonal
-  preconditioning.
-
- Added full assembly support in Example 9/9p.
-
- Added a new test problem in Example 24/24p, demonstrating a mixed bilinear
-  form for H1, H(curl), H(div) and L_2, with partial assembly support.
-
- Added weak Dirichlet boundary conditions (Nitsche) to the NURBS miniapp.
-
- Added a simple mesh editing miniapp, Trimmer, which trims away portions of a
-  mesh based on element attributes. Any newly exposed boundary elements are
-  assigned attribute numbers related to the trimmed element attributes.
-
- Added device support in Example 5/5p.
-
 Improved testing
 ----------------
 - Added a GitLab pipeline that automates PR testing on supercomputing systems
@@ -174,17 +74,15 @@ Improved testing

 Miscellaneous
 -------------
+- In SLISolver, changed the residual inner product from (Br,r) to (Br,Br) so the
+  solver can work with non-SPD preconditioner B.
+
 - Added support for ADIOS2 for parallel I/O with ParaView visualization. The
  classes adios2stream and ADIOS2DataCollection are introduced in mfem as the
  interfaces to generate ADIOS2 Binary Pack (BP4) directory datasets for the
  entire spatial and temporal data. In addition, ADIOS2 allows for setting a
  user-defined number of data substreams/subfiles. See examples 5, 9, 12, 16.

- The integration order used in the ComputeLpError and ComputeElementLpError
-  methods of class GridFunction has been increased.
-
- Various other simplifications, extensions, and bugfixes in the code.
-

 Version 4.1, released on March 10, 2020
 =======================================
@@ -149,13 +149,9 @@ if (MFEM_USE_MPI)
      message(FATAL_ERROR "PETSc version >= 3.8.0 is required")
    endif()
    set(PETSC_INCLUDE_DIRS ${PETSC_INCLUDES})
-    if (MFEM_USE_SLEPC)
-      find_package(SLEPc REQUIRED config)
-      message(STATUS "Found SLEPc version ${SLEPC_VERSION}")
-    endif()
  endif()
 else()
-  set(PKGS_NEED_MPI SUPERLU PETSC SLEPC STRUMPACK PUMI)
+  set(PKGS_NEED_MPI SUPERLU PETSC STRUMPACK PUMI)
  foreach(PKG IN LISTS PKGS_NEED_MPI)
    if (MFEM_USE_${PKG})
      message(STATUS "Disabling package ${PKG} - requires MPI")
@@ -211,10 +207,10 @@ endif()
 # SUNDIALS
 if (MFEM_USE_SUNDIALS)
  if (NOT MFEM_USE_MPI)
-    find_package(SUNDIALS REQUIRED NVector_Serial CVODES ARKODE KINSOL)
+    find_package(SUNDIALS REQUIRED NVector_Serial CVODE ARKODE KINSOL)
  else()
    find_package(SUNDIALS REQUIRED
-      NVector_Serial NVector_Parallel NVector_ParHyp CVODES ARKODE KINSOL)
+      NVector_Serial NVector_Parallel NVector_ParHyp CVODE ARKODE KINSOL)
  endif()
 endif()

@@ -296,18 +292,6 @@ if (MFEM_USE_HIOP)
  # find_package updates HIOP_FOUND, HIOP_INCLUDE_DIRS, HIOP_LIBRARIES
 endif()

-# ADEPT package
-if (MFEM_USE_ADEPT)
-        find_package(ADEPT REQUIRED)
-        # find_package updates ADEPT_FOUND, ADEPT_INCLUDE_DIRS, ADEPT_LIBRARIES
-endif()
-
-# FADBAD++ package
-if (MFEM_USE_FADBADPP)
-	find_package(FADBADPP REQUIRED)
-        # find_package updates FADBADPP_FOUND, FADBADPP_INCLUDE_DIRS, FADBADPP_LIBRARIES
-endif()
-
 # CUDA
 if (MFEM_USE_CUDA)
   set(CMAKE_CUDA_STANDARD 11)
@@ -368,9 +352,8 @@ endif()
 #    integers, the METIS header (with 32-bit indices, as used by mfem) needs to
 #    be before SuiteSparse.
 set(MFEM_TPLS MPI_CXX OPENMP BLAS LAPACK METIS HYPRE SuiteSparse SUNDIALS PETSC
-    SLEPC MESQUITE SuperLUDist STRUMPACK AXOM CONDUIT Ginkgo GNUTLS GSLIB NETCDF
-    ADEPT FADBADPP MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA 
-    UMPIRE ADIOS2)
+    MESQUITE SuperLUDist STRUMPACK AXOM CONDUIT Ginkgo GNUTLS GSLIB NETCDF
+    MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE ADIOS2)
 # Add all *_FOUND libraries in the variable TPL_LIBRARIES.
 set(TPL_LIBRARIES "")
 set(TPL_INCLUDE_DIRS "")
@@ -109,7 +109,6 @@ The MFEM source code has the following structure:
  ├── linalg
  ├── mesh
  ├── miniapps
-  │   ├── adjoint
  │   ├── common
  │   ├── electromagnetics
  │   ├── gslib
@@ -383,10 +383,6 @@ MFEM_USE_PETSC = YES/NO
   and other features based on the PETSc package. When enabled, this option uses
   the PETSC_* library options, see below.

-MFEM_USE_SLEPC = YES/NO
-   Enable MFEM eigensolvers based on the SLEPc package. When enabled, this
-   option uses the SLEPC_* library options, see below.
-
 MFEM_USE_MPFR = YES/NO
   MPFR is a library for multiple-precision floating-point computations. This
   option enables the use of MPFR in MFEM, e.g. for precise computation of 1D
@@ -400,12 +396,6 @@ MFEM_USE_SIDRE = YES/NO
   blueprint specification. When enabled, this option requires installation of
   HDF5 (see also MFEM_USE_NETCDF), Conduit and LLNL's axom project.

-MFEM_USE_SIMD = YES/NO
-   Enables the high performance templated classes to use architecture dependent
-   SIMD intrinsics instead of the generic implementation of class AutoSIMD in
-   linalg/simd/auto.hpp. This option should be combined with suitable
-   compiler options, such as -march=native, to enable optimal vectorization.
-
 MFEM_USE_CONDUIT = YES/NO
   Enables support for converting MFEM Mesh and Grid Function objects to and
   from Conduit Mesh Blueprint Descriptions (https://github.com/LLNL/conduit/)
@@ -436,8 +426,6 @@ MFEM_USE_PUMI = YES/NO
   data management system that is capable of handling general non-manifold
   models and effectively supports automated adaptive analysis. PUMI enables
   support for parallel unstructured mesh modifications in MFEM.
-   The develop branch of PUMI repository (https://github.com/SCOREC/core)
-   should be used for most updated features.

 MFEM_USE_UMPIRE = YES/NO
   Enables support for Umpire, a resource management library that allows the
@@ -448,19 +436,6 @@ MFEM_USE_HIOP = YES/NO
  Enable the usage of HiOp (https://github.com/LLNL/hiop) in MFEM. HiOp is an
  HPC solver for nonlinear optimization problems.

-MFEM_USE_ADEPT = YES/NO
-  Enable automatic differentiation using the ADEPT library. 
-  (http://www.met.reading.ac.uk/clouds/adept) 
-  Please, compile the library with flag --disable-openmp.
-
-MFEM_USE_FADBADPP = YES/NO
-  Enable automatic differentiation using the FADBAD++ library.
-  www.fadbad.com/fadbad.html 
-
-MFEM_USE_ADFORWARD = YES/NO
-  Enable forward mode for AD packages. This option is valid
-  only if the AD package supports two modes (backward/forward).
-
 MFEM_USE_CUDA = YES/NO
   Enables support for CUDA devices in MFEM. CUDA is a parallel computing
   platform and programming model for general computing on graphical processing
@@ -614,12 +589,6 @@ The specific libraries and their options are:
  Options: PETSC_OPT, PETSC_LIB.
  Versions: PETSc >= 3.8.0.

- SLEPc (optional), used when MFEM_USE_SLEPC = YES. SLEPc depends on PETSc and
-  uses some of the PETSc options when compiled.
-  URL: https://slepc.upv.es/
-  Options: SLEPC_OPT, SLEPC_LIB.
-  Versions: SLEPc >= 3.8.0.
-
 - Sidre (optional), part of LLNL's axom project, used when MFEM_USE_SIDRE = YES.
  Starting with MFEM v4.1, Axom version 0.3.1 or later is required.
  URL: https://github.com/LLNL/axom
@@ -640,24 +609,13 @@ The specific libraries and their options are:

 - PUMI (optional), used when MFEM_USE_PUMI = YES.
  URL: https://scorec.rpi.edu/pumi
-       https://github.com/SCOREC/core
  Options: PUMI_OPT, PUMI_LIB.
-  Versions: PUMI >= 2.2.3.
+  Versions: PUMI >= 2.2.0.

 - HiOp (optional), used when MFEM_USE_HIOP = YES.
  URL: https://github.com/LLNL/hiop
  Options: HIOP_OPT, HIOP_LIB.

- ADEPT (optional), used with MFEM_USE_ADEPT = YES
-  URL: www.met.reading.ac.uk/clouds/adept/
-  Options: ADEPT_OPT, ADEPT_LIB
-  Versions: 1.1 and 2.0.5
-
- FADBAD++ (optiobal), used with MFEM_USE_FADBADPP = YES
-  URL: www.fadbad.com/fadbad.html
-  Options: FADBADPP_OPT
-  Versions: 2.1
-
 - GSLIB (optional), used when MFEM_USE_GSLIB = YES. The gslib library must be
  built prior to the MFEM build, as follows: download gslib-1.0.5, untar it at
  the same level as MFEM and create a symbolic link: "ln -s gslib-1.0.5 gslib".
@@ -682,11 +640,12 @@ The specific libraries and their options are:
  Options: OCCA_DIR, OCCA_OPT, OCCA_LIB.
  Versions: OCCA >= 1.0.9.

- libCEED (optional), used when MFEM_USE_CEED = YES.
+- libCEED (optional), used when MFEM_USE_CEED = YES. Requires libCEED v0.6
+  or later version, specifically, git-hash 3d05795 or later.
  URL: https://github.com/CEED/libCEED
       https://ceed.exascaleproject.org/libceed
  Options: CEED_DIR, CEED_OPT, CEED_LIB.
-  Versions: libCEED >= 0.6, git-hash a970f63.
+  Versions: libCEED >= 0.6.

 - RAJA (optional), used when MFEM_USE_RAJA = YES.
  Beginning with MFEM v4.1, only RAJA v0.10.0+ is supported.
@@ -836,9 +795,6 @@ MFEM_USE_MPFR
 MFEM_USE_ZLIB
 MFEM_USE_PUMI
 MFEM_USE_HIOP
-MFEM_USE_ADEPT
-MFEM_USE_FADBADPP
-MFEM_USE_ADFORWARD
 MFEM_USE_CUDA
 MFEM_USE_OCCA
 MFEM_USE_CEED
@@ -893,8 +849,6 @@ The CMake build system adds auto-detection for the following packages/libraries:
 - POSIXCLOCKS
 - PUMI
 - HIOP
- - ADEPT
- - FADBAD++
 - OCCA
 - RAJA
 - UMPIRE
@@ -244,10 +244,6 @@ IF (DEFINED TPL_ENABLE_PETSC)
  SET(MFEM_USE_PETSC ${TPL_ENABLE_PETSC} CACHE BOOL "Enable PETSc support." FORCE)
 ENDIF()

-IF (DEFINED TPL_ENABLE_SLEPC)
-  SET(MFEM_USE_SLEPC ${TPL_ENABLE_SLEPC} CACHE BOOL "Enable SLEPc support." FORCE)
-ENDIF()
-
 IF (DEFINED TPL_ENABLE_MPFR)
  SET(MFEM_USE_MPFR ${TPL_ENABLE_MPFR} CACHE BOOL "Enable MPFR usage." FORCE)
 ENDIF()
@@ -38,7 +38,6 @@ set(MFEM_USE_GNUTLS @MFEM_USE_GNUTLS@)
 set(MFEM_USE_GSLIB @MFEM_USE_GSLIB@)
 set(MFEM_USE_NETCDF @MFEM_USE_NETCDF@)
 set(MFEM_USE_PETSC @MFEM_USE_PETSC@)
-set(MFEM_USE_SLEPC @MFEM_USE_SLEPC@)
 set(MFEM_USE_MPFR @MFEM_USE_MPFR@)
 set(MFEM_USE_SIDRE @MFEM_USE_SIDRE@)
 set(MFEM_USE_CONDUIT @MFEM_USE_CONDUIT@)
@@ -48,11 +47,7 @@ set(MFEM_USE_OCCA @MFEM_USE_OCCA@)
 set(MFEM_USE_RAJA @MFEM_USE_RAJA@)
 set(MFEM_USE_CEED @MFEM_USE_CEED@)
 set(MFEM_USE_UMPIRE @MFEM_USE_UMPIRE@)
-set(MFEM_USE_SIMD @MFEM_USE_SIMD@)
 set(MFEM_USE_ADIOS2 @MFEM_USE_ADIOS2@)
-set(MFEM_USE_ADEPT @MFEM_USE_ADEPT@)
-set(MFEM_USE_FADBADPP @MFEM_USE_FADBADPP@)
-set(MFEM_USE_ADFORWARD @MFEM_USE_ADFORWARD@)

 set(MFEM_CXX_COMPILER "@CMAKE_CXX_COMPILER@")
 set(MFEM_CXX_FLAGS "@CMAKE_CXX_FLAGS@")
@@ -104,15 +104,9 @@
 // Enable MFEM functionality based on the PETSc library
 #cmakedefine MFEM_USE_PETSC

-// Enable MFEM functionality based on the SLEPc library
-#cmakedefine MFEM_USE_SLEPC
-
 // Enable MFEM functionality based on the Sidre library
 #cmakedefine MFEM_USE_SIDRE

-// Enable the use of SIMD in the high performance templated classes
-#cmakedefine MFEM_USE_SIMD
-
 // Enable MFEM functionality based on Conduit
 #cmakedefine MFEM_USE_CONDUIT

@@ -156,13 +150,4 @@
 // library.
 #cmakedefine MFEM_USE_SIMMETRIX

-// use ADEPT library for AD
-#cmakedefine MFEM_USE_ADEPT
-
-// use FADBAD++ library for AD
-#cmakedefine MFEM_USE_FADBADPP
-
-// use forward mode for automatic differentiation
-#cmakedefine MFEM_USE_ADFORWARD
-
 #endif // MFEM_CONFIG_HEADER
@@ -1,23 +0,0 @@
-# Copyright (c) 2010, Lawrence Livermore National Security, LLC. Produced at the
-# Lawrence Livermore National Laboratory. LLNL-CODE-443211. All Rights reserved.
-# See file COPYRIGHT for details.
-#
-# This file is part of the MFEM library. For more information and source code
-# availability see http://mfem.org.
-#
-# MFEM is free software; you can redistribute it and/or modify it under the
-# terms of the GNU Lesser General Public License (as published by the Free
-# Software Foundation) version 2.1 dated February 1999.
-
-# Sets the following variables:
-#   - ADEPT_FOUND
-#   - ADEPT_INCLUDE_DIRS
-#   - ADEPT_LIBRARIES
-
-include(MfemCmakeUtilities)
-mfem_find_package(ADEPT ADEPT ADEPT_DIR
-  "include" "adept.hpp"
-  "lib" "libadept.so"
-  "Paths to headers required by ADEPT."
-  "Libraries required by ADEPT.")
-
@@ -1,23 +0,0 @@
-# Copyright (c) 2010, Lawrence Livermore National Security, LLC. Produced at the
-# Lawrence Livermore National Laboratory. LLNL-CODE-443211. All Rights reserved.
-# See file COPYRIGHT for details.
-#
-# This file is part of the MFEM library. For more information and source code
-# availability see http://mfem.org.
-#
-# MFEM is free software; you can redistribute it and/or modify it under the
-# terms of the GNU Lesser General Public License (as published by the Free
-# Software Foundation) version 2.1 dated February 1999.
-
-# Sets the following variables:
-#   - FADBADPP_FOUND
-#   - FADBADPP_INCLUDE_DIRS
-#   - FADBADPP_LIBRARIES
-
-include(MfemCmakeUtilities)
-mfem_find_package(FADBADPP FADBADPP FADBADPP_DIR
-  "include" "fadiff.h"
-  "lib" ""
-  "Paths to headers required by FADBADPP."
-  "Libraries required by FADBADPP.")
-
@@ -1,44 +0,0 @@
-# Copyright (c) 2010-2020, 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.
-
-# Sets the following variables:
-#   - SLEPC_FOUND
-#   - SLEPC_INCLUDE_DIRS
-#   - SLEPC_LIBRARIES
-
-set(SLEPc_REQUIRED_PACKAGES "PETSC" CACHE STRING
-  "Additional packages required by SLEPc")
-
-include(MfemCmakeUtilities)
-mfem_find_package(SLEPc SLEPC SLEPC_DIR
-  "include" "slepceps.h"
-  "${PETSC_ARCH}/lib" "slepc" # add NAMES_PER_DIR?
-  "Paths to headers required by SLEPc."
-  "Libraries required by SLEPc."
-  ADD_COMPONENT "config" "${PETSC_ARCH}/include" "slepcconf.h" "" ""
-  CHECK_BUILD SLEPC_VERSION_OK TRUE
-"
-#include \"petsc.h\"
-#include \"slepceps.h\"
-int main()
-{
-  PetscErrorCode ierr;
-  int argc = 0;
-  char** argv = NULL;
-  ierr = SlepcInitialize(&argc, &argv, PETSC_NULL, PETSC_NULL);
-  EPS eps;
-  ierr = EPSCreate(PETSC_COMM_SELF, &eps); CHKERRQ(ierr);
-  ierr = EPSDestroy(&eps); CHKERRQ(ierr);
-  ierr = SlepcFinalize(); CHKERRQ(ierr);
-  return 0;
-}
-"
-  )
@@ -25,6 +25,5 @@ mfem_find_package(SUNDIALS SUNDIALS SUNDIALS_DIR
  ADD_COMPONENT NVector_ParHyp
    "include" nvector/nvector_parhyp.h "lib" sundials_nvecparhyp
  ADD_COMPONENT CVODE "include" cvode/cvode.h "lib" sundials_cvode
-  ADD_COMPONENT CVODES "include" cvodes/cvodes.h "lib" sundials_cvodes
  ADD_COMPONENT ARKODE "include" arkode/arkode.h "lib" sundials_arkode
  ADD_COMPONENT KINSOL "include" kinsol/kinsol.h "lib" sundials_kinsol)
@@ -731,10 +731,9 @@ function(mfem_export_mk_files)
      MFEM_USE_LAPACK MFEM_THREAD_SAFE MFEM_USE_OPENMP MFEM_USE_LEGACY_OPENMP
      MFEM_USE_MEMALLOC MFEM_USE_SUNDIALS MFEM_USE_MESQUITE MFEM_USE_SUITESPARSE
      MFEM_USE_SUPERLU MFEM_USE_STRUMPACK MFEM_USE_GNUTLS
-      MFEM_USE_GSLIB MFEM_USE_NETCDF MFEM_USE_PETSC MFEM_USE_SLEPC MFEM_USE_MPFR MFEM_USE_SIDRE
+      MFEM_USE_GSLIB MFEM_USE_NETCDF MFEM_USE_PETSC MFEM_USE_MPFR MFEM_USE_SIDRE
      MFEM_USE_CONDUIT MFEM_USE_PUMI MFEM_USE_CUDA MFEM_USE_OCCA MFEM_USE_RAJA
-      MFEM_USE_UMPIRE MFEM_USE_SIMD MFEM_USE_ADEPT MFEM_USE_FADBADPP 
-      MFEM_USE_ADFORWARD MFEM_USE_ADIOS2)
+      MFEM_USE_UMPIRE)
  foreach(var ${CONFIG_MK_BOOL_VARS})
    if (${var})
      set(${var} YES)
@@ -744,7 +743,6 @@ function(mfem_export_mk_files)
  endforeach()
  # TODO: Add support for MFEM_USE_CUDA=YES
  set(MFEM_CXX ${CMAKE_CXX_COMPILER})
-  set(MFEM_HOST_CXX ${MFEM_CXX})
  set(MFEM_CPPFLAGS "")
  string(STRIP "${CMAKE_CXX_FLAGS_${BUILD_TYPE}} ${CMAKE_CXX_FLAGS}"
         MFEM_CXXFLAGS)
@@ -48,9 +48,6 @@
 #ifdef MFEM_USE_PETSC
 #error Building with PETSc (MFEM_USE_PETSC=YES) requires MPI (MFEM_USE_MPI=YES)
 #endif
-#ifdef MFEM_USE_SLEPC
-#error Building with SLEPc (MFEM_USE_SLEPC=YES) requires MPI (MFEM_USE_MPI=YES)
-#endif
 #ifdef MFEM_USE_PUMI
 #error Building with PUMI (MFEM_USE_PUMI=YES) requires MPI (MFEM_USE_MPI=YES)
 #endif
@@ -106,9 +106,6 @@
 // Enable Sidre support
 // #define MFEM_USE_SIDRE

-// Enable the use of SIMD in the high performance templated classes
-// #define MFEM_USE_SIMD
-
 // Enable Conduit support
 // #define MFEM_USE_CONDUIT

@@ -118,9 +115,6 @@
 // Enable functionality based on the PETSc library
 // #define MFEM_USE_PETSC

-// Enable functionality based on the SLEPc library
-// #define MFEM_USE_SLEPC
-
 // Enable functionality based on the MPFR library.
 // #define MFEM_USE_MPFR

@@ -163,14 +157,4 @@
 // library.
 // #define MFEM_USE_SIMMETRIX

-// use ADEPT library for AD
-// #define MFEM_USE_ADEPT
-
-// use FADBAD++ library for AD
-// #define MFEM_USE_FADBADPP
-
-// use forward mode for automatic differentiation
-// #define MFEM_USE_ADFORWARD
-
-
 #endif // MFEM_CONFIG_HEADER
@@ -37,15 +37,11 @@ MFEM_USE_GINKGO        = @MFEM_USE_GINKGO@
 MFEM_USE_GNUTLS        = @MFEM_USE_GNUTLS@
 MFEM_USE_NETCDF        = @MFEM_USE_NETCDF@
 MFEM_USE_PETSC         = @MFEM_USE_PETSC@
-MFEM_USE_SLEPC         = @MFEM_USE_SLEPC@
 MFEM_USE_MPFR          = @MFEM_USE_MPFR@
 MFEM_USE_SIDRE         = @MFEM_USE_SIDRE@
 MFEM_USE_CONDUIT       = @MFEM_USE_CONDUIT@
 MFEM_USE_PUMI          = @MFEM_USE_PUMI@
 MFEM_USE_HIOP          = @MFEM_USE_HIOP@
-MFEM_USE_ADEPT         = @MFEM_USE_ADEPT@
-MFEM_USE_FADBADPP      = @MFEM_USE_FADBADPP@
-MFEM_USE_ADFORWARD     = @MFEM_USE_ADFORWARD@
 MFEM_USE_GSLIB         = @MFEM_USE_GSLIB@
 MFEM_USE_CUDA          = @MFEM_USE_CUDA@
 MFEM_USE_HIP           = @MFEM_USE_HIP@
@@ -53,12 +49,10 @@ MFEM_USE_RAJA          = @MFEM_USE_RAJA@
 MFEM_USE_OCCA          = @MFEM_USE_OCCA@
 MFEM_USE_CEED          = @MFEM_USE_CEED@
 MFEM_USE_UMPIRE        = @MFEM_USE_UMPIRE@
-MFEM_USE_SIMD          = @MFEM_USE_SIMD@
 MFEM_USE_ADIOS2        = @MFEM_USE_ADIOS2@

 # Compiler, compile options, and link options
 MFEM_CXX       = @MFEM_CXX@
-MFEM_HOST_CXX  = @MFEM_HOST_CXX@
 MFEM_CPPFLAGS  = @MFEM_CPPFLAGS@
 MFEM_CXXFLAGS  = @MFEM_CXXFLAGS@
 MFEM_TPLFLAGS  = @MFEM_TPLFLAGS@
@@ -39,7 +39,6 @@ option(MFEM_USE_GNUTLS "Enable GNUTLS usage" OFF)
 option(MFEM_USE_GSLIB "Enable GSLIB usage" OFF)
 option(MFEM_USE_NETCDF "Enable NETCDF usage" OFF)
 option(MFEM_USE_PETSC "Enable PETSc support." OFF)
-option(MFEM_USE_SLEPC "Enable SLEPc support." OFF)
 option(MFEM_USE_MPFR "Enable MPFR usage." OFF)
 option(MFEM_USE_SIDRE "Enable Axom/Sidre usage" OFF)
 option(MFEM_USE_CONDUIT "Enable Conduit usage" OFF)
@@ -50,11 +49,7 @@ option(MFEM_USE_OCCA "Enable OCCA" OFF)
 option(MFEM_USE_RAJA "Enable RAJA" OFF)
 option(MFEM_USE_CEED "Enable CEED" OFF)
 option(MFEM_USE_UMPIRE "Enable Umpire" OFF)
-option(MFEM_USE_SIMD "Enable use of SIMD intrinsics" OFF)
 option(MFEM_USE_ADIOS2 "Enable ADIOS2" OFF)
-option(MFEM_USE_ADEPT "Enable AD using ADEPT" OFF)
-option(MFEM_USE_FADBADPP "Enable AD using FADBAD++" OFF)
-option(MFEM_USE_ADFORWARD "Enable forward mode for AD" OFF)

 set(MFEM_MPI_NP 4 CACHE STRING "Number of processes used for MPI tests")

@@ -91,8 +86,6 @@ set(METIS_DIR "${MFEM_DIR}/../metis-4.0" CACHE PATH "Path to the METIS library."

 set(LIBUNWIND_DIR "" CACHE PATH "Path to Libunwind.")

-# For sundials_nvecparhyp and nvecparallel remember to build with MPI_ENABLED=ON
-# and modify cmake variables for hypre for sundials
 set(SUNDIALS_DIR "${MFEM_DIR}/../sundials-5.0.0/instdir" CACHE PATH
    "Path to the SUNDIALS library.")
 # The following may be necessary, if SUNDIALS was built with KLU:
@@ -161,10 +154,6 @@ set(PETSC_DIR "${MFEM_DIR}/../petsc" CACHE PATH
    "Path to the PETSc main directory.")
 set(PETSC_ARCH "arch-linux2-c-debug" CACHE STRING "PETSc build architecture.")

-set(SLEPC_DIR "${MFEM_DIR}/../slepc" CACHE PATH
-    "Path to the SLEPc main directory.")
-set(SLEPC_ARCH "arch-linux2-c-debug" CACHE STRING "SLEPC build architecture.")
-
 set(MPFR_DIR "" CACHE PATH "Path to the MPFR library.")

 set(CONDUIT_DIR "${MFEM_DIR}/../conduit" CACHE PATH
@@ -193,13 +182,6 @@ set(BLAS_LIBRARIES "" CACHE STRING "The BLAS library.")
 set(LAPACK_INCLUDE_DIRS "" CACHE STRING "Path to LAPACK headers.")
 set(LAPACK_LIBRARIES "" CACHE STRING "The LAPACK library.")

-set(ADEPT_INCLUDE_DIRS "${MFEM_DIR}/../adept-1.1/include" CACHE STRING "Path to ADEPT headers.")
-set(ADEPT_LIBRARIES "-L${MFEM_DIR}/../adept-1.1/lib -ladept" CACHE STRING "The ADEPT library.")
-
-set(FADBADPP_INCLUDE_DIRS "${MFEM_DIR}/../FADBAD++" CACHE STRING "Path to FADBAD++ headers.")
-set(FADBADPP_LIBRARIES "")
-
-
 # Some useful variables:
 set(CMAKE_SKIP_PREPROCESSED_SOURCE_RULES ON) # Skip *.i rules
 set(CMAKE_SKIP_ASSEMBLY_SOURCE_RULES  ON)    # Skip *.s rules
@@ -125,7 +125,6 @@ MFEM_USE_GINKGO        = NO
 MFEM_USE_GNUTLS        = NO
 MFEM_USE_NETCDF        = NO
 MFEM_USE_PETSC         = NO
-MFEM_USE_SLEPC         = NO
 MFEM_USE_MPFR          = NO
 MFEM_USE_SIDRE         = NO
 MFEM_USE_CONDUIT       = NO
@@ -138,11 +137,7 @@ MFEM_USE_RAJA          = NO
 MFEM_USE_OCCA          = NO
 MFEM_USE_CEED          = NO
 MFEM_USE_UMPIRE        = NO
-MFEM_USE_SIMD          = NO
 MFEM_USE_ADIOS2        = NO
-MFEM_USE_ADEPT         = NO
-MFEM_USE_FADBADPP      = NO
-MFEM_USE_ADFORWARD     = NO

 # Compile and link options for zlib.
 ZLIB_DIR =
@@ -192,12 +187,10 @@ OPENMP_LIB =
 POSIX_CLOCKS_LIB = -lrt

 # SUNDIALS library configuration
-# For sundials_nvecparhyp and nvecparallel remember to build with MPI_ENABLED=ON
-# and modify cmake variables for hypre for sundials
 SUNDIALS_DIR = @MFEM_DIR@/../sundials-5.0.0/instdir
 SUNDIALS_OPT = -I$(SUNDIALS_DIR)/include
 SUNDIALS_LIB = -Wl,-rpath,$(SUNDIALS_DIR)/lib64 -L$(SUNDIALS_DIR)/lib64\
- -lsundials_arkode -lsundials_cvodes -lsundials_nvecserial -lsundials_kinsol
+ -lsundials_arkode -lsundials_cvode -lsundials_nvecserial -lsundials_kinsol

 ifeq ($(MFEM_USE_MPI),YES)
   SUNDIALS_LIB += -lsundials_nvecparhyp -lsundials_nvecparallel
@@ -282,20 +275,6 @@ ifeq ($(PETSC_FOUND),YES)
      -L$(abspath $(PETSC_DIR))/lib -lpetsc $(PETSC_LIB)
 endif

-SLEPC_DIR := $(MFEM_DIR)/../slepc
-SLEPC_VARS := $(SLEPC_DIR)/lib/slepc/conf/slepc_variables
-SLEPC_FOUND := $(if $(wildcard $(SLEPC_VARS)),YES,)
-SLEPC_INC_VAR = SLEPC_INCLUDE
-SLEPC_LIB_VAR = SLEPC_EXTERNAL_LIB
-ifeq ($(SLEPC_FOUND),YES)
-   SLEPC_OPT := $(shell sed -n "s/$(SLEPC_INC_VAR) *= *//p" $(SLEPC_VARS))
-   # Some additional external libraries might be defined in this file
-   -include ${SLEPC_DIR}/${PETSC_ARCH}/lib/slepc/conf/slepcvariables
-   SLEPC_LIB := $(shell sed -n "s/$(SLEPC_LIB_VAR) *= *//p" $(SLEPC_VARS))
-   SLEPC_LIB := -Wl,-rpath,$(abspath $(SLEPC_DIR))/$(PETSC_ARCH)/lib\
-      -L$(abspath $(SLEPC_DIR))/$(PETSC_ARCH)/lib -lslepc $(SLEPC_LIB)
-endif
-
 # MPFR library configuration
 MPFR_OPT =
 MPFR_LIB = -lmpfr
@@ -339,16 +318,6 @@ HIOP_DIR = @MFEM_DIR@/../hiop/install
 HIOP_OPT = -I$(HIOP_DIR)/include
 HIOP_LIB = -L$(HIOP_DIR)/lib -lhiop $(LAPACK_LIB)

-# ADEPT
-ADEPT_DIR = @MFEM_DIR@/../adept-1.1
-ADEPT_OPT = -I$(ADEPT_DIR)/include
-ADEPT_LIB = -L$(ADEPT_DIR)/lib -ladept 
-
-# FADBAD++
-FADBADPP_DIR = @MFEM_DIR@/../FADBAD++
-FADBADPP_OPT = -I$(FADBADPP_DIR)
-FADBADPP_LIB = -L.
-
 # GSLIB library
 GSLIB_DIR = @MFEM_DIR@/../gslib/build
 GSLIB_OPT = -I$(GSLIB_DIR)/include
@@ -29,20 +29,8 @@
 #define MFEM_ALWAYS_INLINE
 #endif

-// --- MFEM_VECTORIZE_LOOP (disabled)
-#if (__cplusplus >= 201103L) && !defined(MFEM_DEBUG) && defined(__GNUC__)
-//#define MFEM_VECTORIZE_LOOP _Pragma("GCC ivdep")
-#define MFEM_VECTORIZE_LOOP
-#else
-#define MFEM_VECTORIZE_LOOP
-#endif
-
-// MFEM_TEMPLATE_BLOCK_SIZE is the block size used by the template matrix-matrix
-// multiply, Mult_AB, defined in tmatrix.hpp. This parameter will generally
-// require tuning to determine good value. It is probably highly influenced by
-// the SIMD width when Mult_AB is used with a SIMD type like AutoSIMD.
 #define MFEM_TEMPLATE_BLOCK_SIZE 4
-
+#define MFEM_SIMD_SIZE 32
 #define MFEM_TEMPLATE_ENABLE_SERIALIZE

 // #define MFEM_TEMPLATE_ELTRANS_HAS_NODE_DOFS
@@ -50,6 +38,11 @@
 // #define MFEM_TEMPLATE_FIELD_EVAL_DATA_HAS_DOFS
 #define MFEM_TEMPLATE_INTRULE_COEFF_PRECOMP

+// derived macros
+#define MFEM_ROUNDUP(val,base) ((((val)+(base)-1)/(base))*(base))
+#define MFEM_ALIGN_SIZE(size,type) \
+   MFEM_ROUNDUP(size,(MFEM_SIMD_SIZE)/sizeof(type))
+
 #ifdef MFEM_COUNT_FLOPS
 namespace mfem
 {
@@ -1,37 +0,0 @@
-SetFactory("OpenCASCADE");
-
-R1 = 1.0;
-R2 = 2.0;
-
-Point(1) = {0.0, 0, 0, 1.0};
-Point(2) = {R1, 0, 0, 1.0};
-Point(3) = {R2, 0, 0, 1.0};
-Point(4) = {R1*Cos(Pi/3), R1*Sin(Pi/3), 0, 1.0};
-Point(5) = {R2*Cos(Pi/3), R2*Sin(Pi/3), 0, 1.0};
-Line(1) = {2, 3};
-Line(2) = {4, 5};
-Circle(3) = {2, 1, 4};
-Circle(4) = {3, 1, 5};
-Curve Loop(5) = {1, 4, -2, -3};
-Plane Surface(1) = {5};
-
-Transfinite Curve{1} = 7;
-Transfinite Curve{2} = 7;
-Transfinite Curve{3} = 4;
-Transfinite Curve{4} = 10;
-
-// Set a rotation periodicity constraint:
-Periodic Line{1} = {2} Rotate{{0,0,1}, {0,0,0}, -Pi/3};
-
-// Tag surfaces and volumes with positive integers
-Physical Curve(1) = {3};
-Physical Curve(2) = {4};
-Physical Curve(3) = {1};
-Physical Curve(4) = {2};
-Physical Surface(1) = {1};
-
-// Generate 2D mesh
-Mesh 2;
-Mesh.MshFileVersion = 2.2;
-
-Save "periodic-annulus-sector.msh";
@@ -1,185 +0,0 @@
-$MeshFormat
-2.2 0 8
-$EndMeshFormat
-$Nodes
-55
-1 1 0 0
-2 2 0 0
-3 0.5000000000000001 0.8660254037844386 0
-4 1 1.732050807568877 0
-5 1.166666666666667 0 0
-6 1.333333333333333 0 0
-7 1.5 0 0
-8 1.666666666666667 0 0
-9 1.833333333333333 0 0
-10 0.5833333333333335 1.010362971081845 0
-11 0.6666666666666667 1.154700538379251 0
-12 0.7500000000000002 1.299038105676658 0
-13 0.8333333333333335 1.443375672974064 0
-14 0.9166666666666669 1.587713240271471 0
-15 0.9396926207859085 0.3420201433256683 0
-16 0.7660444431189786 0.6427876096865386 0
-17 1.986476715483886 0.2321858282504602 0
-18 1.946089741159648 0.4612317414848793 0
-19 1.879385241571817 0.6840402866513365 0
-20 1.787265280646825 0.8975983604009234 0
-21 1.670975622825874 1.09901795614161 0
-22 1.532088886237958 1.285575219373077 0
-23 1.372483275737469 1.454747283146095 0
-24 1.194317183405575 1.604246385510085 0
-25 1.425989114816062 0.1915326920916892 0
-26 0.8788667344146573 1.13917645290495 0
-27 1.630372059110754 0.7154531062316609 0
-28 1.436395769298814 1.053728612482506 0
-29 1.081023776188756 0.6241293681829633 0
-30 1.168737372335971 1.428012728596308 0
-31 1.821063986059922 0.298149890497067 0
-32 1.234707097211386 0.3469796339295647 0
-33 1.377747393186519 0.6200150626754309 0
-34 1.457047681210906 0.3890895843559762 0
-35 0.917846726184522 0.8957978954532204 0
-36 1.218335619030348 0.9017812086952638 0
-37 1.066623110765233 1.061857005744772 0
-38 1.587029716281926 0.1355955181472859 0
-39 1.744445799211916 0.1441515753740107 0
-40 1.25 0.1443375672974065 0
-41 1.453660070628011 0.8435769396609902 0
-42 1.741367044061892 0.499612708014486 0
-43 1.30550638526547 1.257610469847477 0
-44 1.118213276932792 0.1666674689105279 0
-45 0.9109440214958271 1.306610291787315 0
-46 0.9970618258753989 1.438658589955562 0
-47 0.7499999999999998 1.010362971081845 0
-48 0.7034449005273667 0.8850673702175776 0
-49 1.605449512513618 0.9269067082200894 0
-50 1.561654019115059 0.5298592532912715 0
-51 1.229782222487711 1.096820457143683 0
-52 1.617066998712459 0.3090202662210922 0
-53 1.079645953234324 1.246963713711438 0
-54 1.877063966817811 0.1348974588243076 0
-55 1.055356609656722 1.558136350380461 0
-$EndNodes
-$Elements
-108
-1 1 2 3 1 1 5
-2 1 2 3 1 5 6
-3 1 2 3 1 6 7
-4 1 2 3 1 7 8
-5 1 2 3 1 8 9
-6 1 2 3 1 9 2
-7 1 2 4 2 3 10
-8 1 2 4 2 10 11
-9 1 2 4 2 11 12
-10 1 2 4 2 12 13
-11 1 2 4 2 13 14
-12 1 2 4 2 14 4
-13 1 2 1 3 1 15
-14 1 2 1 3 15 16
-15 1 2 1 3 16 3
-16 1 2 2 4 2 17
-17 1 2 2 4 17 18
-18 1 2 2 4 18 19
-19 1 2 2 4 19 20
-20 1 2 2 4 20 21
-21 1 2 2 4 21 22
-22 1 2 2 4 22 23
-23 1 2 2 4 23 24
-24 1 2 2 4 24 4
-25 2 2 1 1 32 40 25
-26 2 2 1 1 25 34 32
-27 2 2 1 1 33 41 36
-28 2 2 1 1 38 52 25
-29 2 2 1 1 33 36 29
-30 2 2 1 1 26 47 35
-31 2 2 1 1 35 37 26
-32 2 2 1 1 25 52 34
-33 2 2 1 1 32 44 40
-34 2 2 1 1 15 32 29
-35 2 2 1 1 15 29 16
-36 2 2 1 1 36 41 28
-37 2 2 1 1 32 33 29
-38 2 2 1 1 50 52 42
-39 2 2 1 1 32 34 33
-40 2 2 1 1 42 52 31
-41 2 2 1 1 43 53 51
-42 2 2 1 1 27 41 33
-43 2 2 1 1 26 53 45
-44 2 2 1 1 18 31 17
-45 2 2 1 1 29 35 16
-46 2 2 1 1 29 36 35
-47 2 2 1 1 24 30 23
-48 2 2 1 1 30 53 43
-49 2 2 1 1 17 54 2
-50 2 2 1 1 4 55 24
-51 2 2 1 1 28 51 36
-52 2 2 1 1 47 48 35
-53 2 2 1 1 36 37 35
-54 2 2 1 1 37 53 26
-55 2 2 1 1 22 28 21
-56 2 2 1 1 20 27 19
-57 2 2 1 1 33 50 27
-58 2 2 1 1 15 44 32
-59 2 2 1 1 18 42 31
-60 2 2 1 1 30 43 23
-61 2 2 1 1 35 48 16
-62 2 2 1 1 31 54 17
-63 2 2 1 1 9 39 8
-64 2 2 1 1 8 38 7
-65 2 2 1 1 7 25 6
-66 2 2 1 1 22 43 28
-67 2 2 1 1 23 43 22
-68 2 2 1 1 39 54 31
-69 2 2 1 1 19 42 18
-70 2 2 1 1 24 55 30
-71 2 2 1 1 27 42 19
-72 2 2 1 1 13 46 14
-73 2 2 1 1 51 53 37
-74 2 2 1 1 39 52 38
-75 2 2 1 1 6 40 5
-76 2 2 1 1 34 52 50
-77 2 2 1 1 12 45 13
-78 2 2 1 1 30 55 46
-79 2 2 1 1 10 47 11
-80 2 2 1 1 8 39 38
-81 2 2 1 1 28 49 21
-82 2 2 1 1 7 38 25
-83 2 2 1 1 41 49 28
-84 2 2 1 1 20 49 27
-85 2 2 1 1 11 26 12
-86 2 2 1 1 27 49 41
-87 2 2 1 1 31 52 39
-88 2 2 1 1 25 40 6
-89 2 2 1 1 2 54 9
-90 2 2 1 1 14 55 4
-91 2 2 1 1 45 53 46
-92 2 2 1 1 45 46 13
-93 2 2 1 1 5 44 1
-94 2 2 1 1 21 49 20
-95 2 2 1 1 46 53 30
-96 2 2 1 1 3 48 10
-97 2 2 1 1 34 50 33
-98 2 2 1 1 36 51 37
-99 2 2 1 1 26 45 12
-100 2 2 1 1 11 47 26
-101 2 2 1 1 27 50 42
-102 2 2 1 1 40 44 5
-103 2 2 1 1 43 51 28
-104 2 2 1 1 10 48 47
-105 2 2 1 1 9 54 39
-106 2 2 1 1 46 55 14
-107 2 2 1 1 1 44 15
-108 2 2 1 1 16 48 3
-$EndElements
-$Periodic
-1
-1 1 2
-Affine 0.5000000000000001 0.8660254037844386 0 0 -0.8660254037844386 0.5000000000000001 0 0 0 0 1 0 0 0 0 1
-7
-9 14
-6 11
-8 13
-5 10
-7 12
-2 4
-1 3
-$EndPeriodic
@@ -1,25 +0,0 @@
-SetFactory("OpenCASCADE");
-
-R = 1.5;
-r = 0.5;
-
-Torus(1) = {0,0,0, R, r, Pi/3};
-
-pts() = PointsOf{ Volume{1}; };
-
-Characteristic Length{ pts() } = 0.25;
-
-// Set a rotation periodicity constraint:
-Periodic Surface{3} = {2} Rotate{{0,0,1}, {0,0,0}, Pi/3};
-
-// Tag surfaces and volumes with positive integers
-Physical Surface(1) = {1};
-Physical Surface(2) = {2};
-Physical Surface(3) = {3};
-Physical Volume(1) = {1};
-
-// Generate 3D mesh
-Mesh 3;
-
-Mesh.MshFileVersion = 2.2;
-Save "periodic-torus-sector.msh";
@@ -1,155 +0,0 @@
-MFEM NURBS mesh v1.0
-
-dimension
-2
-
-elements
-5
-1 3 0 3 7 4
-1 3 3 2 6 7
-1 3 2 1 5 6
-1 3 1 0 4 5
-1 3 2 8 9 1
-
-boundary
-10
-1 1 0 3
-2 1 3 2
-2 1 1 0
-2 1 2 8
-2 1 9 1
-3 1 7 4
-3 1 6 7
-3 1 5 6
-3 1 4 5
-4 1 8 9
-
-edges
-15
-0 0 4
-0 3 7
-0 1 5
-0 2 6
-1 0 3
-1 4 7
-2 3 2
-2 7 6
-2 1 0
-2 5 4
-1 2 1
-1 6 5
-1 8 9
-3 2 8
-3 1 9
-
-vertices
-10
-
-patches
-
-knotvectors
-2
-2 3 0 0 0 1 1 1
-2 4 0 0 0 0.5 1 1 1
-
-dimension
-2
-
-controlpoints_cartesian
-5 5 1
-5 3.92523e-16 1
-5 -5 1
-2.47593 2.47593 1
-4.95187 6.06429e-16 0.707107
-2.47593 -2.47593 1
-0.424264 0.424264 1
-0.848528 1.03915e-16 0.707107
-0.424264 -0.424264 1
-0.353553 0.353553 1
-0.707107 8.65956e-17 0.707107
-0.353553 -0.353553 1
-
-knotvectors
-2
-2 3 0 0 0 1 1 1
-2 4 0 0 0 0.5 1 1 1
-
-dimension
-2
-
-controlpoints_cartesian
-5 -5 1
-1.17757e-15 -5 1
-5 -5 1
-2.47593 -2.47593 1
-9.09644e-16 -4.95187 0.707107
-2.47593 -2.47593 1
-0.424264 -0.424264 1
-1.55872e-16 -0.848528 0.707107
-0.424264 -0.424264 1
-0.353553 -0.353553 1
-1.29893e-16 -0.707107 0.707107
-0.353553 -0.353553 1
-
-knotvectors
-2
-2 3 0 0 0 1 1 1
-2 4 0 0 0 0.5 1 1 1
-
-dimension
-2
-
-controlpoints_cartesian
-5 -5 1
-5 -1.17757e-15 1
-5 5 1
-2.47593 -2.47593 1
-4.95187 -1.21286e-15 0.707107
-2.47593 2.47593 1
-0.424264 -0.424264 1
-0.848528 -2.07829e-16 0.707107
-0.424264 0.424264 1
-0.353553 -0.353553 1
-0.707107 -1.73191e-16 0.707107
-0.353553 0.353553 1
-
-knotvectors
-2
-2 3 0 0 0 1 1 1
-2 4 0 0 0 0.5 1 1 1
-
-dimension
-2
-
-controlpoints_cartesian
-5 5 1
-3.92523e-16 5 1
-5 5 1
-2.47593 2.47593 1
-3.03215e-16 4.95187 0.707107
-2.47593 2.47593 1
-0.424264 0.424264 1
-5.19574e-17 0.848528 0.707107
-0.424264 0.424264 1
-0.353553 0.353553 1
-4.32978e-17 0.707107 0.707107
-0.353553 0.353553 1
-
-knotvectors
-2
-2 3 0 0 0 1 1 1
-2 3 0 0 0 1 1 1
-
-dimension
-2
-
-controlpoints_cartesian
-5 -5 1
-10 -5 1
-15 -5 1
-5 0 1
-10 0 1
-15 0 1
-5 5 1
-10 5 1
-15 5 1
@@ -16,21 +16,36 @@ if (DOXYGEN_FOUND)
  configure_file(${CMAKE_CURRENT_SOURCE_DIR}/CodeDocumentation.conf.in
    ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.conf @ONLY)

+  if (UNIX)
+    # Only create symlinks if UNIX operating system
+    add_custom_target(doc
+      COMMAND ${DOXYGEN_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.conf
+      COMMAND ${CMAKE_COMMAND} -E remove -f ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.html
+      COMMAND ${CMAKE_COMMAND} -E create_symlink
+      ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation/html/index.html
+      ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.html
+      BYPRODUCTS ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation/html/index.html
+      WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
+      COMMENT "Generating API documentation with Doxygen to CodeDocumentation.html"
+      VERBATIM)

-  add_custom_target(doc
-    COMMAND ${DOXYGEN_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.conf
-    COMMAND echo "<meta http-equiv=\"REFRESH\" content=\"0;URL=CodeDocumentation/html/index.html\">" > ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.html
-    BYPRODUCTS ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation/html/index.html
-    WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
-    COMMENT "Generating API documentation with Doxygen to CodeDocumentation.html"
-    VERBATIM)
-
-  add_custom_target(clean-doc
-    COMMAND ${CMAKE_COMMAND} -E remove -f ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.html
-    COMMAND ${CMAKE_COMMAND} -E remove -f ${CMAKE_CURRENT_BINARY_DIR}/warnings.log
-    COMMAND ${CMAKE_COMMAND} -E remove_directory ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation
-    COMMENT "Removing API documentation"
-    VERBATIM)
+    add_custom_target(clean-doc
+      COMMAND ${CMAKE_COMMAND} -E remove -f ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.html
+      COMMAND ${CMAKE_COMMAND} -E remove_directory ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation
+      COMMENT "Removing API documentation"
+      VERBATIM)

+  else (UNIX)
+    add_custom_target(doc
+      COMMAND ${DOXYGEN_EXECUTABLE} ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation.conf
+      BYPRODUCTS ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation/html/index.html
+      WORKING_DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}
+      COMMENT "Generating API documentation with Doxygen to CodeDocumentation/html/index.html"
+      VERBATIM)

+    add_custom_target(clean-doc
+      COMMAND ${CMAKE_COMMAND} -E remove_directory ${CMAKE_CURRENT_BINARY_DIR}/CodeDocumentation
+      COMMENT "Removing API documentation"
+      VERBATIM)
+  endif (UNIX)
 endif (DOXYGEN_FOUND)
@@ -51,7 +51,7 @@ PROJECT_BRIEF          = "Finite element discretization library"
 # pixels and the maximum width should not exceed 200 pixels. Doxygen will copy
 # the logo to the output directory.

-PROJECT_LOGO           = web/logo-small.png
+PROJECT_LOGO           =

 # The OUTPUT_DIRECTORY tag is used to specify the (relative or absolute) path
 # into which the generated documentation will be written. If a relative path is
@@ -746,7 +746,7 @@ WARN_FORMAT            = "$file:$line: $text"
 # messages should be written. If left blank the output is written to standard
 # error (stderr).

-WARN_LOGFILE           = warnings.log
+WARN_LOGFILE           =

 #---------------------------------------------------------------------------
 # Configuration options related to the input files
@@ -770,7 +770,6 @@ INPUT                  = @MFEM_SOURCE_DIR@/doc/CodeDocumentation.dox \
                         @MFEM_SOURCE_DIR@/examples/pumi \
                         @MFEM_SOURCE_DIR@/examples/hiop \
                         @MFEM_SOURCE_DIR@/examples/sundials \
-                         @MFEM_SOURCE_DIR@/miniapps/adjoint \			 
                         @MFEM_SOURCE_DIR@/miniapps/common \
                         @MFEM_SOURCE_DIR@/miniapps/electromagnetics \
                         @MFEM_SOURCE_DIR@/miniapps/gslib \
@@ -1471,7 +1470,7 @@ MATHJAX_FORMAT         = HTML-CSS
 # The default value is: http://cdn.mathjax.org/mathjax/latest.
 # This tag requires that the tag USE_MATHJAX is set to YES.

-MATHJAX_RELPATH        = http://cdn.mathjax.org/mathjax/latest
+MATHJAX_RELPATH        = https://cdn.llnl.gov/mathjax/2.7.2

 # The MATHJAX_EXTENSIONS tag can be used to specify one or more MathJax
 # extension names that should be enabled during MathJax rendering. For example
@@ -88,8 +88,8 @@ namespace mfem {
 * - <a class="el" href="ex24p_8cpp_source.html">Example 24p</a>: parallel mixed finite element spaces and interpolators
 * - <a class="el" href="ex25_8cpp_source.html">Example 25</a>: simulation of electromagnetic wave propagation using a Perfectly Matched Layer (PML)
 * - <a class="el" href="ex25p_8cpp_source.html">Example 25p</a>: parallel simulation of electromagnetic wave propagation using a Perfectly Matched Layer (PML)
- * - <a class="el" href="ex26_8cpp_source.html">Example 26</a>: multigrid preconditioner for the Laplace problem using nodal H1 FEM
- * - <a class="el" href="ex26p_8cpp_source.html">Example 26p</a>: parallel multigrid preconditioner for the Laplace problem using nodal H1 FEM
+ * - <a class="el" href="ex26_8cpp_source.html">Example 26</a>: multigrid preconditioner for the Laplace problem using nodal H1 FEM 
+ * - <a class="el" href="ex26p_8cpp_source.html">Example 26p</a>: parallel multigrid preconditioner for the Laplace problem using nodal H1 FEM 
 *
 * <H4>SUNDIALS Examples</H4>
 * - Variants of Examples
@@ -101,9 +101,6 @@ namespace mfem {
 *   and
 *   <a class="el" href="sundials_2ex16p_8cpp_source.html">16p</a>
 *   demonstrating the use of MFEM's \link sundials.hpp SUNDIALS classes\endlink
- * - CVODES adjoint examples:
- *   <a class="el" href="cvsRoberts__ASAi__dns_8cpp_source.html">serial ODE system</a>,
- *   <a class="el" href="adjoint__advection__diffusion_8cpp_source.html">parallel advection-diffusion</a>
 *
 * <H4>PETSc Examples</H4>
 * - Variants of Examples
@@ -143,9 +140,7 @@ namespace mfem {
 * - <a class="el" href="tesla_8cpp_source.html">Tesla</a>: simple magnetostatics simulation code
 * - <a class="el" href="maxwell_8cpp_source.html">Maxwell</a>: simple transient full-wave electromagnetics simulation code
 * - <a class="el" href="joule_8cpp_source.html">Joule</a>: transient magnetics and Joule heating miniapp
- * - <a class="el" href="classmfem_1_1navier_1_1NavierSolver.html">Navier</a>: solve the transient incompressible Navier-Stokes equations
-
-* - <a class="el" href="mobius-strip_8cpp_source.html">Mobius Strip</a>: generate various Mobius strip-like meshes
+ * - <a class="el" href="mobius-strip_8cpp_source.html">Mobius Strip</a>: generate various Mobius strip-like meshes
 * - <a class="el" href="klein-bottle_8cpp_source.html">Klein Bottle</a>: generate three types of Klein bottle surfaces
 * - <a class="el" href="toroid_8cpp_source.html">Toroid</a>: generate simple toroidal meshes
 * - <a class="el" href="twist_8cpp_source.html">Twist</a>: generate simple periodic meshes
@@ -154,7 +149,6 @@ namespace mfem {
 * - <a class="el" href="extruder_8cpp_source.html">Extruder</a>: extrude a low-dimensional mesh into a higher dimension
 * - <a class="el" href="mesh-explorer_8cpp_source.html">Mesh Explorer</a>: visualize and manipulate meshes
 * - <a class="el" href="mesh-optimizer_8cpp_source.html">Mesh Optimizer</a>: optimize high-order meshes, <a class="el" href="mesh-optimizer_8cpp_source.html">serial</a> and <a class="el" href="pmesh-optimizer_8cpp_source.html">parallel</a> versions
- * - <a class="el" href="trimmer_8cpp_source.html">Trimmer</a>: trim elements from existing meshes
 * - <a class="el" href="display-basis_8cpp_source.html">Display Basis</a>: visualize finite element basis functions
 * - <a class="el" href="get-values_8cpp_source.html">Get Values</a>: extract field values via DataCollection classes
 * - <a class="el" href="load-dc_8cpp_source.html">Load DC</a>: visualize fields saved via DataCollection classes
@@ -162,6 +156,7 @@ namespace mfem {
 * - <a class="el" href="lor-transfer_8cpp_source.html">LOR Transfer</a>: map functions between high-order and low-order refined spaces
 * - <a class="el" href="findpts_8cpp_source.html">Find Points</a>: evaluate grid function in physical space, <a class="el" href="findpts_8cpp_source.html">serial</a> and <a class="el" href="pfindpts_8cpp_source.html">parallel</a> versions
 * - <a class="el" href="field-diff_8cpp_source.html">Field Diff</a>: compare grid functions on different meshes
+ * - <a class="el" href="classmfem_1_1navier_1_1NavierSolver.html">Navier</a>: solve the transient incompressible Navier-Stokes equations
 * - <a class="el" href="miniapps_2performance_2ex1_8cpp_source.html">HPC Example 1</a>: high-performance nodal H1 FEM for the Laplace problem
 * - <a class="el" href="miniapps_2performance_2ex1p_8cpp_source.html">HPC Example 1p</a>: high-performance parallel nodal H1 FEM for the Laplace problem
 *
@@ -9,25 +9,18 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.

-SHELL = /bin/bash
 MFEM_DIR ?= ..
 DOXYGEN_CONF = CodeDocumentation.conf

-
 # doxygen uses: graphviz, latex
 html: $(DOXYGEN_CONF)
-	@# Generate the html documentation
-	@doxygen $(DOXYGEN_CONF)
-	@echo "<meta http-equiv=\"REFRESH\" content=\"0;URL=CodeDocumentation/html/index.html\">" > CodeDocumentation.html
-	@cat warnings.log
-	@# Generate the log of undocumented methods
-	@( cat $(DOXYGEN_CONF) ; echo "GENERATE_HTML=NO" ; echo "EXTRACT_ALL=NO" ; echo "WARN_LOGFILE=undoc.log" ; echo "QUIET=YES" ) | doxygen - &> /dev/null
+	doxygen $(DOXYGEN_CONF)
+	rm -f CodeDocumentation.html
+	ln -s CodeDocumentation/html/index.html CodeDocumentation.html

 clean:
 	rm -rf $(DOXYGEN_CONF) CodeDocumentation CodeDocumentation.html *~
-	rm -rf  undoc.log warnings.log

 $(DOXYGEN_CONF): $(MFEM_DIR)/doc/$(DOXYGEN_CONF).in
-	@sed -e 's%@MFEM_SOURCE_DIR@%$(MFEM_DIR)%g' $(<) \
+	sed -e 's%@MFEM_SOURCE_DIR@%$(MFEM_DIR)%g' $(<) \
 	  > $(DOXYGEN_CONF)
-
@@ -34,7 +34,6 @@ list(APPEND ALL_EXE_SRCS
  ex25.cpp
  ex26.cpp
  ex27.cpp
-  ex71.cpp
  )

 if (MFEM_USE_MPI)
@@ -65,7 +64,8 @@ if (MFEM_USE_MPI)
    ex25p.cpp
    ex26p.cpp
    ex27p.cpp
-    ex71p.cpp
+    pa_oper.cpp
+    io_benchmark.cpp
    )
 endif()

@@ -93,7 +93,7 @@ foreach(SRC_FILE ${ALL_EXE_SRCS})
    add_test(NAME ${TEST_NAME}_ser
      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
  else()
-    add_test(NAME ${TEST_NAME}_np=${MFEM_MPI_NP}
+    add_test(NAME ${TEST_NAME}_np=4
      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
      ${MPIEXEC_PREFLAGS}
      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
@@ -103,22 +103,13 @@ endforeach()

 # If STRUMPACK is enabled, add a test run that uses it.
 if (MFEM_USE_STRUMPACK)
-  add_test(NAME ex11p_strumpack_np=${MFEM_MPI_NP}
+  add_test(NAME ex11p_strumpack_np=4
    COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
    ${MPIEXEC_PREFLAGS}
    $<TARGET_FILE:ex11p> "-no-vis" "--strumpack"
    ${MPIEXEC_POSTFLAGS})
 endif()

-# If SuperLU_DIST is enabled, add a test run that uses it.
-if (MFEM_USE_SUPERLU)
-  add_test(NAME ex11p_superlu_np=${MFEM_MPI_NP}
-    COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-    ${MPIEXEC_PREFLAGS}
-    $<TARGET_FILE:ex11p> "-no-vis" "--superlu"
-    ${MPIEXEC_POSTFLAGS})
-endif()
-
 # Include the examples/sundials directory if SUNDIALS is enabled.
 if (MFEM_USE_SUNDIALS)
  add_subdirectory(sundials)
@@ -9,8 +9,6 @@
 //               ex1 -m ../data/fichera.mesh
 //               ex1 -m ../data/fichera-mixed.mesh
 //               ex1 -m ../data/toroid-wedge.mesh
-//               ex1 -m ../data/periodic-annulus-sector.msh
-//               ex1 -m ../data/periodic-torus-sector.msh
 //               ex1 -m ../data/square-disc-p2.vtk -o 2
 //               ex1 -m ../data/square-disc-p3.mesh -o 3
 //               ex1 -m ../data/square-disc-nurbs.mesh -o -1
@@ -34,8 +32,7 @@
 //               ex1 -pa -d raja-omp
 //               ex1 -pa -d occa-omp
 //               ex1 -pa -d ceed-cpu
-//             * ex1 -pa -d ceed-cuda
-//               ex1 -pa -d ceed-cuda:/gpu/cuda/shared
+//               ex1 -pa -d ceed-cuda
 //               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
@@ -103,8 +100,8 @@ int main(int argc, char *argv[])
   // 3. Read the mesh from the given mesh file. We can handle triangular,
   //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
   //    the same code.
-   Mesh mesh(mesh_file, 1, 1);
-   int dim = mesh.Dimension();
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();

   // 4. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement. We choose 'ref_levels' to be the
@@ -112,10 +109,10 @@ int main(int argc, char *argv[])
   //    elements.
   {
      int ref_levels =
-         (int)floor(log(50000./mesh.GetNE())/log(2.)/dim);
+         (int)floor(log(50000./mesh->GetNE())/log(2.)/dim);
      for (int l = 0; l < ref_levels; l++)
      {
-         mesh.UniformRefinement();
+         mesh->UniformRefinement();
      }
   }

@@ -123,70 +120,66 @@ int main(int argc, char *argv[])
   //    Lagrange finite elements of the specified order. If order < 1, we
   //    instead use an isoparametric/isogeometric space.
   FiniteElementCollection *fec;
-   bool delete_fec;
   if (order > 0)
   {
      fec = new H1_FECollection(order, dim);
-      delete_fec = true;
   }
-   else if (mesh.GetNodes())
+   else if (mesh->GetNodes())
   {
-      fec = mesh.GetNodes()->OwnFEC();
-      delete_fec = false;
+      fec = mesh->GetNodes()->OwnFEC();
      cout << "Using isoparametric FEs: " << fec->Name() << endl;
   }
   else
   {
      fec = new H1_FECollection(order = 1, dim);
-      delete_fec = true;
   }
-   FiniteElementSpace fespace(&mesh, fec);
+   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, fec);
   cout << "Number of finite element unknowns: "
-        << fespace.GetTrueVSize() << endl;
+        << fespace->GetTrueVSize() << endl;

   // 6. Determine the list of true (i.e. conforming) essential boundary dofs.
   //    In this example, the boundary conditions are defined by marking all
   //    the boundary attributes from the mesh as essential (Dirichlet) and
   //    converting them to a list of true dofs.
   Array<int> ess_tdof_list;
-   if (mesh.bdr_attributes.Size())
+   if (mesh->bdr_attributes.Size())
   {
-      Array<int> ess_bdr(mesh.bdr_attributes.Max());
+      Array<int> ess_bdr(mesh->bdr_attributes.Max());
      ess_bdr = 1;
-      fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
   }

   // 7. Set up the linear form b(.) which corresponds to the right-hand side of
   //    the FEM linear system, which in this case is (1,phi_i) where phi_i are
   //    the basis functions in the finite element fespace.
-   LinearForm b(&fespace);
+   LinearForm *b = new LinearForm(fespace);
   ConstantCoefficient one(1.0);
-   b.AddDomainIntegrator(new DomainLFIntegrator(one));
-   b.Assemble();
+   b->AddDomainIntegrator(new DomainLFIntegrator(one));
+   b->Assemble();

   // 8. Define the solution vector x as a finite element grid function
   //    corresponding to fespace. Initialize x with initial guess of zero,
   //    which satisfies the boundary conditions.
-   GridFunction x(&fespace);
+   GridFunction x(fespace);
   x = 0.0;

   // 9. Set up the bilinear form a(.,.) on the finite element space
   //    corresponding to the Laplacian operator -Delta, by adding the Diffusion
   //    domain integrator.
-   BilinearForm a(&fespace);
-   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a.AddDomainIntegrator(new DiffusionIntegrator(one));
+   BilinearForm *a = new BilinearForm(fespace);
+   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
+   a->AddDomainIntegrator(new DiffusionIntegrator(one));

   // 10. Assemble the bilinear form and the corresponding linear system,
   //     applying any necessary transformations such as: eliminating boundary
   //     conditions, applying conforming constraints for non-conforming AMR,
   //     static condensation, etc.
-   if (static_cond) { a.EnableStaticCondensation(); }
-   a.Assemble();
+   if (static_cond) { a->EnableStaticCondensation(); }
+   a->Assemble();

   OperatorPtr A;
   Vector B, X;
-   a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);

   cout << "Size of linear system: " << A->Height() << endl;

@@ -207,9 +200,9 @@ int main(int argc, char *argv[])
   }
   else // Jacobi preconditioning in partial assembly mode
   {
-      if (UsesTensorBasis(fespace))
+      if (UsesTensorBasis(*fespace))
      {
-         OperatorJacobiSmoother M(a, ess_tdof_list);
+         OperatorJacobiSmoother M(*a, ess_tdof_list);
         PCG(*A, M, B, X, 1, 400, 1e-12, 0.0);
      }
      else
@@ -219,13 +212,13 @@ int main(int argc, char *argv[])
   }

   // 12. Recover the solution as a finite element grid function.
-   a.RecoverFEMSolution(X, b, x);
+   a->RecoverFEMSolution(X, *b, x);

   // 13. Save the refined mesh and the solution. This output can be viewed later
   //     using GLVis: "glvis -m refined.mesh -g sol.gf".
   ofstream mesh_ofs("refined.mesh");
   mesh_ofs.precision(8);
-   mesh.Print(mesh_ofs);
+   mesh->Print(mesh_ofs);
   ofstream sol_ofs("sol.gf");
   sol_ofs.precision(8);
   x.Save(sol_ofs);
@@ -237,14 +230,15 @@ int main(int argc, char *argv[])
      int  visport   = 19916;
      socketstream sol_sock(vishost, visport);
      sol_sock.precision(8);
-      sol_sock << "solution\n" << mesh << x << flush;
+      sol_sock << "solution\n" << *mesh << x << flush;
   }

   // 15. Free the used memory.
-   if (delete_fec)
-   {
-      delete fec;
-   }
+   delete a;
+   delete b;
+   delete fespace;
+   if (order > 0) { delete fec; }
+   delete mesh;

   return 0;
 }
@@ -8,8 +8,6 @@
 //               mpirun -np 4 ex11p -m ../data/escher.mesh
 //               mpirun -np 4 ex11p -m ../data/fichera.mesh
 //               mpirun -np 4 ex11p -m ../data/fichera-mixed.mesh
-//               mpirun -np 4 ex11p -m ../data/periodic-annulus-sector.msh
-//               mpirun -np 4 ex11p -m ../data/periodic-torus-sector.msh -rs 1
 //               mpirun -np 4 ex11p -m ../data/toroid-wedge.mesh -o 2
 //               mpirun -np 4 ex11p -m ../data/square-disc-p2.vtk -o 2
 //               mpirun -np 4 ex11p -m ../data/square-disc-p3.mesh -o 3
@@ -35,38 +35,6 @@
 using namespace std;
 using namespace mfem;

-class CustomSolverMonitor : public IterativeSolverMonitor
-{
-public:
-   CustomSolverMonitor(const ParMesh *m,
-                       ParGridFunction *f) :
-      pmesh(m),
-      pgf(f) {}
-
-   void MonitorSolution(int i, double norm, const Vector &x, bool final)
-   {
-      char vishost[] = "localhost";
-      int  visport   = 19916;
-      int  num_procs, myid;
-
-      MPI_Comm_size(pmesh->GetComm(),&num_procs);
-      MPI_Comm_rank(pmesh->GetComm(),&myid);
-
-      pgf->SetFromTrueDofs(x);
-
-      socketstream sol_sock(vishost, visport);
-      sol_sock << "parallel " << num_procs << " " << myid << "\n";
-      sol_sock.precision(8);
-      sol_sock << "solution\n" << *pmesh << *pgf
-               << "window_title 'Iteration no " << i << "'"
-               << "keys rRjlc\n" << flush;
-   }
-
-private:
-   const ParMesh *pmesh;
-   ParGridFunction *pgf;
-};
-
 int main(int argc, char *argv[])
 {
   // 1. Initialize MPI.
@@ -220,7 +188,6 @@ int main(int argc, char *argv[])
   }
   else
   {
-      CustomSolverMonitor monitor(pmesh, &x);
      GMRESSolver gmres(MPI_COMM_WORLD);
      gmres.SetAbsTol(0.0);
      gmres.SetRelTol(1e-12);
@@ -229,7 +196,6 @@ int main(int argc, char *argv[])
      gmres.SetPrintLevel(1);
      gmres.SetOperator(*A);
      gmres.SetPreconditioner(*amg);
-      gmres.SetMonitor(monitor);
      gmres.Mult(*B, *X);
   }
   delete amg;
@@ -88,6 +88,8 @@ private:
   Vector funval2;
   Vector nor;
   Vector fluxN;
+   IntegrationPoint eip1;
+   IntegrationPoint eip2;

 public:
   FaceIntegrator(RiemannSolver &rsolver_, const int dim);
@@ -416,24 +418,27 @@ void FaceIntegrator::AssembleFaceVector(const FiniteElement &el1,
   {
      intorder++;
   }
-   const IntegrationRule *ir = &IntRules.Get(Tr.GetGeometryType(), intorder);
+   const IntegrationRule *ir = &IntRules.Get(Tr.FaceGeom, intorder);

   for (int i = 0; i < ir->GetNPoints(); i++)
   {
      const IntegrationPoint &ip = ir->IntPoint(i);

-      Tr.SetAllIntPoints(&ip); // set face and element int. points
+      Tr.Loc1.Transform(ip, eip1);
+      Tr.Loc2.Transform(ip, eip2);

      // Calculate basis functions on both elements at the face
-      el1.CalcShape(Tr.GetElement1IntPoint(), shape1);
-      el2.CalcShape(Tr.GetElement2IntPoint(), shape2);
+      el1.CalcShape(eip1, shape1);
+      el2.CalcShape(eip2, shape2);

      // Interpolate elfun at the point
      elfun1_mat.MultTranspose(shape1, funval1);
      elfun2_mat.MultTranspose(shape2, funval2);

+      Tr.Face->SetIntPoint(&ip);
+
      // Get the normal vector and the flux on the face
-      CalcOrtho(Tr.Jacobian(), nor);
+      CalcOrtho(Tr.Face->Jacobian(), nor);
      const double mcs = rsolver.Eval(funval1, funval2, nor, fluxN);

      // Update max char speed
@@ -38,42 +38,6 @@
 using namespace std;
 using namespace mfem;

-class GeneralResidualMonitor : public IterativeSolverMonitor
-{
-public:
-   GeneralResidualMonitor(const std::string& prefix_, int print_lvl)
-      : prefix(prefix_)
-   {
-      print_level = print_lvl;
-   }
-
-   virtual void MonitorResidual(int it, double norm, const Vector &r, bool final);
-
-private:
-   const std::string prefix;
-   int print_level;
-   mutable double norm0;
-};
-
-void GeneralResidualMonitor::MonitorResidual(int it, double norm,
-                                             const Vector &r, bool final)
-{
-   if (print_level == 1 || (print_level == 3 && (final || it == 0)))
-   {
-      mfem::out << prefix << " iteration " << setw(2) << it
-                << " : ||r|| = " << norm;
-      if (it > 0)
-      {
-         mfem::out << ",  ||r||/||r_0|| = " << norm/norm0;
-      }
-      else
-      {
-         norm0 = norm;
-      }
-      mfem::out << '\n';
-   }
-}
-
 // Custom block preconditioner for the Jacobian of the incompressible nonlinear
 // elasticity operator. It has the form
 //
@@ -139,11 +103,9 @@ protected:

   // Newton solver for the hyperelastic operator
   NewtonSolver newton_solver;
-   GeneralResidualMonitor newton_monitor;

   // Solver for the Jacobian solve in the Newton method
   Solver *j_solver;
-   GeneralResidualMonitor j_monitor;

   // Preconditioner for the Jacobian
   Solver *j_prec;
@@ -448,8 +410,7 @@ RubberOperator::RubberOperator(Array<FiniteElementSpace *> &fes,
                               int iter,
                               Coefficient &c_mu)
   : Operator(fes[0]->GetVSize() + fes[1]->GetVSize()),
-     newton_solver(), newton_monitor("Newton", 1),
-     j_monitor("  GMRES", 3), mu(c_mu), block_offsets(offsets)
+     newton_solver(), mu(c_mu), block_offsets(offsets)
 {
   Array<Vector *> rhs(2);
   rhs = NULL; // Set all entries in the array
@@ -485,8 +446,7 @@ RubberOperator::RubberOperator(Array<FiniteElementSpace *> &fes,
   j_gmres->SetRelTol(1e-12);
   j_gmres->SetAbsTol(1e-12);
   j_gmres->SetMaxIter(300);
-   j_gmres->SetPrintLevel(-1);
-   j_gmres->SetMonitor(j_monitor);
+   j_gmres->SetPrintLevel(0);
   j_gmres->SetPreconditioner(*j_prec);
   j_solver = j_gmres;

@@ -494,8 +454,7 @@ RubberOperator::RubberOperator(Array<FiniteElementSpace *> &fes,
   newton_solver.iterative_mode = true;
   newton_solver.SetSolver(*j_solver);
   newton_solver.SetOperator(*this);
-   newton_solver.SetPrintLevel(-1);
-   newton_solver.SetMonitor(newton_monitor);
+   newton_solver.SetPrintLevel(1);
   newton_solver.SetRelTol(rel_tol);
   newton_solver.SetAbsTol(abs_tol);
   newton_solver.SetMaxIter(iter);
@@ -38,56 +38,6 @@
 using namespace std;
 using namespace mfem;

-class GeneralResidualMonitor : public IterativeSolverMonitor
-{
-public:
-   GeneralResidualMonitor(MPI_Comm comm, const std::string& prefix_,
-                          int print_lvl)
-      : prefix(prefix_)
-   {
-#ifndef MFEM_USE_MPI
-      print_level = print_lvl;
-#else
-      int rank;
-      MPI_Comm_rank(comm, &rank);
-      if (rank == 0)
-      {
-         print_level = print_lvl;
-      }
-      else
-      {
-         print_level = -1;
-      }
-#endif
-   }
-
-   virtual void MonitorResidual(int it, double norm, const Vector &r, bool final);
-
-private:
-   const std::string prefix;
-   int print_level;
-   mutable double norm0;
-};
-
-void GeneralResidualMonitor::MonitorResidual(int it, double norm,
-                                             const Vector &r, bool final)
-{
-   if (print_level == 1 || (print_level == 3 && (final || it == 0)))
-   {
-      mfem::out << prefix << " iteration " << setw(2) << it
-                << " : ||r|| = " << norm;
-      if (it > 0)
-      {
-         mfem::out << ",  ||r||/||r_0|| = " << norm/norm0;
-      }
-      else
-      {
-         norm0 = norm;
-      }
-      mfem::out << '\n';
-   }
-}
-
 // Custom block preconditioner for the Jacobian of the incompressible nonlinear
 // elasticity operator. It has the form
 //
@@ -153,11 +103,9 @@ protected:

   // Newton solver for the hyperelastic operator
   NewtonSolver newton_solver;
-   GeneralResidualMonitor newton_monitor;

   // Solver for the Jacobian solve in the Newton method
   Solver *j_solver;
-   GeneralResidualMonitor j_monitor;

   // Preconditioner for the Jacobian
   Solver *j_prec;
@@ -511,10 +459,7 @@ RubberOperator::RubberOperator(Array<ParFiniteElementSpace *> &fes,
                               int iter,
                               Coefficient &c_mu)
   : Operator(fes[0]->TrueVSize() + fes[1]->TrueVSize()),
-     newton_solver(fes[0]->GetComm()),
-     newton_monitor(fes[0]->GetComm(), "Newton", 1),
-     j_monitor(fes[0]->GetComm(), "  GMRES", 3),
-     mu(c_mu), block_trueOffsets(trueOffsets)
+     newton_solver(fes[0]->GetComm()), mu(c_mu), block_trueOffsets(trueOffsets)
 {
   Array<Vector *> rhs(2);
   rhs = NULL; // Set all entries in the array
@@ -554,8 +499,7 @@ RubberOperator::RubberOperator(Array<ParFiniteElementSpace *> &fes,
   j_gmres->SetRelTol(1e-12);
   j_gmres->SetAbsTol(1e-12);
   j_gmres->SetMaxIter(300);
-   j_gmres->SetPrintLevel(-1);
-   j_gmres->SetMonitor(j_monitor);
+   j_gmres->SetPrintLevel(0);
   j_gmres->SetPreconditioner(*j_prec);
   j_solver = j_gmres;

@@ -563,8 +507,7 @@ RubberOperator::RubberOperator(Array<ParFiniteElementSpace *> &fes,
   newton_solver.iterative_mode = true;
   newton_solver.SetSolver(*j_solver);
   newton_solver.SetOperator(*this);
-   newton_solver.SetPrintLevel(-1);
-   newton_solver.SetMonitor(newton_monitor);
+   newton_solver.SetPrintLevel(1);
   newton_solver.SetRelTol(rel_tol);
   newton_solver.SetAbsTol(abs_tol);
   newton_solver.SetMaxIter(iter);
@@ -9,8 +9,6 @@
 //               mpirun -np 4 ex1p -m ../data/fichera.mesh
 //               mpirun -np 4 ex1p -m ../data/fichera-mixed.mesh
 //               mpirun -np 4 ex1p -m ../data/toroid-wedge.mesh
-//               mpirun -np 4 ex1p -m ../data/periodic-annulus-sector.msh
-//               mpirun -np 4 ex1p -m ../data/periodic-torus-sector.msh
 //               mpirun -np 4 ex1p -m ../data/square-disc-p2.vtk -o 2
 //               mpirun -np 4 ex1p -m ../data/square-disc-p3.mesh -o 3
 //               mpirun -np 4 ex1p -m ../data/square-disc-nurbs.mesh -o -1
@@ -32,8 +30,7 @@
 //               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-cuda
-//               mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared
+//               mpirun -np 4 ex1p -pa -d ceed-cuda
 //               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
@@ -73,6 +70,7 @@ int main(int argc, char *argv[])
   bool pa = false;
   const char *device_config = "cpu";
   bool visualization = true;
+   int nfiles = 1;

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -89,6 +87,7 @@ int main(int argc, char *argv[])
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
+   args.AddOption(&nfiles, "-nf", "--num-files", "Number of files to write.");
   args.Parse();
   if (!args.Good())
   {
@@ -112,8 +111,8 @@ int main(int argc, char *argv[])
   // 4. Read the (serial) mesh from the given mesh file on all processors.  We
   //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
   //    and volume meshes with the same code.
-   Mesh mesh(mesh_file, 1, 1);
-   int dim = mesh.Dimension();
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();

   // 5. Refine the serial mesh on all processors to increase the resolution. In
   //    this example we do 'ref_levels' of uniform refinement. We choose
@@ -121,23 +120,23 @@ int main(int argc, char *argv[])
   //    more than 10,000 elements.
   {
      int ref_levels =
-         (int)floor(log(10000./mesh.GetNE())/log(2.)/dim);
+         (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
      for (int l = 0; l < ref_levels; l++)
      {
-         mesh.UniformRefinement();
+         mesh->UniformRefinement();
      }
   }

   // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
   //    this mesh further in parallel to increase the resolution. Once the
   //    parallel mesh is defined, the serial mesh can be deleted.
-   ParMesh pmesh(MPI_COMM_WORLD, mesh);
-   mesh.Clear();
+   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
+   delete mesh;
   {
      int par_ref_levels = 2;
      for (int l = 0; l < par_ref_levels; l++)
      {
-         pmesh.UniformRefinement();
+         pmesh->UniformRefinement();
      }
   }

@@ -145,16 +144,13 @@ int main(int argc, char *argv[])
   //    use continuous Lagrange finite elements of the specified order. If
   //    order < 1, we instead use an isoparametric/isogeometric space.
   FiniteElementCollection *fec;
-   bool delete_fec;
   if (order > 0)
   {
      fec = new H1_FECollection(order, dim);
-      delete_fec = true;
   }
-   else if (pmesh.GetNodes())
+   else if (pmesh->GetNodes())
   {
-      fec = pmesh.GetNodes()->OwnFEC();
-      delete_fec = false;
+      fec = pmesh->GetNodes()->OwnFEC();
      if (myid == 0)
      {
         cout << "Using isoparametric FEs: " << fec->Name() << endl;
@@ -163,10 +159,9 @@ int main(int argc, char *argv[])
   else
   {
      fec = new H1_FECollection(order = 1, dim);
-      delete_fec = true;
   }
-   ParFiniteElementSpace fespace(&pmesh, fec);
-   HYPRE_Int size = fespace.GlobalTrueVSize();
+   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec, 1, 0);
+   HYPRE_Int size = fespace->GlobalTrueVSize();
   if (myid == 0)
   {
      cout << "Number of finite element unknowns: " << size << endl;
@@ -177,44 +172,44 @@ int main(int argc, char *argv[])
   //    by marking all the boundary attributes from the mesh as essential
   //    (Dirichlet) and converting them to a list of true dofs.
   Array<int> ess_tdof_list;
-   if (pmesh.bdr_attributes.Size())
+   if (pmesh->bdr_attributes.Size())
   {
-      Array<int> ess_bdr(pmesh.bdr_attributes.Max());
+      Array<int> ess_bdr(pmesh->bdr_attributes.Max());
      ess_bdr = 1;
-      fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
   }

   // 9. Set up the parallel linear form b(.) which corresponds to the
   //    right-hand side of the FEM linear system, which in this case is
   //    (1,phi_i) where phi_i are the basis functions in fespace.
-   ParLinearForm b(&fespace);
+   ParLinearForm *b = new ParLinearForm(fespace);
   ConstantCoefficient one(1.0);
-   b.AddDomainIntegrator(new DomainLFIntegrator(one));
-   b.Assemble();
+   b->AddDomainIntegrator(new DomainLFIntegrator(one));
+   b->Assemble();

   // 10. Define the solution vector x as a parallel finite element grid function
   //     corresponding to fespace. Initialize x with initial guess of zero,
   //     which satisfies the boundary conditions.
-   ParGridFunction x(&fespace);
+   ParGridFunction x(fespace);
   x = 0.0;

   // 11. Set up the parallel bilinear form a(.,.) on the finite element space
   //     corresponding to the Laplacian operator -Delta, by adding the Diffusion
   //     domain integrator.
-   ParBilinearForm a(&fespace);
-   if (pa) { a.SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-   a.AddDomainIntegrator(new DiffusionIntegrator(one));
+   ParBilinearForm *a = new ParBilinearForm(fespace);
+   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
+   a->AddDomainIntegrator(new DiffusionIntegrator(one));

   // 12. Assemble the parallel bilinear form and the corresponding linear
   //     system, applying any necessary transformations such as: parallel
   //     assembly, eliminating boundary conditions, applying conforming
   //     constraints for non-conforming AMR, static condensation, etc.
-   if (static_cond) { a.EnableStaticCondensation(); }
-   a.Assemble();
+   if (static_cond) { a->EnableStaticCondensation(); }
+   a->Assemble();

   OperatorPtr A;
   Vector B, X;
-   a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);

   // 13. Solve the linear system A X = B.
   //     * With full assembly, use the BoomerAMG preconditioner from hypre.
@@ -222,9 +217,9 @@ int main(int argc, char *argv[])
   Solver *prec = NULL;
   if (pa)
   {
-      if (UsesTensorBasis(fespace))
+      if (UsesTensorBasis(*fespace))
      {
-         prec = new OperatorJacobiSmoother(a, ess_tdof_list);
+         prec = new OperatorJacobiSmoother(*a, ess_tdof_list);
      }
   }
   else
@@ -242,22 +237,54 @@ int main(int argc, char *argv[])

   // 14. Recover the parallel grid function corresponding to X. This is the
   //     local finite element solution on each processor.
-   a.RecoverFEMSolution(X, b, x);
+   a->RecoverFEMSolution(X, *b, x);
+
+   std::string filename("nranks_");
+   filename += to_string(num_procs);
+   filename += ".gf";
+   {
+      double t1;
+      t1 = MPI_Wtime();
+      x.Save(filename.c_str(), nfiles);
+      double t2 = MPI_Wtime();
+
+      if (myid == 0)
+      {
+         err << "elapsed write time: " << t2 - t1 << endl;
+      }
+   }
+   {
+      double t1;
+      t1 = MPI_Wtime();
+      ParGridFunction new_x(fespace, filename.c_str());
+      double t2 = MPI_Wtime();
+      if (myid == 0)
+      {
+         err << "elapsed read time: " << t2 - t1 << endl;
+      }
+      // new_x -= x;
+      // out << "GF difference: " << new_x.Norml1() << endl;
+   }

   // 15. Save the refined mesh and the solution in parallel. This output can
   //     be viewed later using GLVis: "glvis -np <np> -m mesh -g sol".
   {
      ostringstream mesh_name, sol_name;
-      mesh_name << "mesh." << setfill('0') << setw(6) << myid;
-      sol_name << "sol." << setfill('0') << setw(6) << myid;
-
-      ofstream mesh_ofs(mesh_name.str().c_str());
-      mesh_ofs.precision(8);
-      pmesh.Print(mesh_ofs);
+      //mesh_name << "mesh." << setfill('0') << setw(6) << myid;
+      sol_name << "sol." << num_procs << setfill('0') << setw(6) << myid;

+      //ofstream mesh_ofs(mesh_name.str().c_str());
+      //mesh_ofs.precision(8);
+      //pmesh->Print(mesh_ofs);
+      double t1 = MPI_Wtime();
      ofstream sol_ofs(sol_name.str().c_str());
      sol_ofs.precision(8);
      x.Save(sol_ofs);
+      double t2 = MPI_Wtime();
+      if (myid == 0)
+      {
+         err << t2 - t1 << endl;
+      }
   }

   // 16. Send the solution by socket to a GLVis server.
@@ -268,14 +295,16 @@ int main(int argc, char *argv[])
      socketstream sol_sock(vishost, visport);
      sol_sock << "parallel " << num_procs << " " << myid << "\n";
      sol_sock.precision(8);
-      sol_sock << "solution\n" << pmesh << x << flush;
+      sol_sock << "solution\n" << *pmesh << x << flush;
   }

   // 17. Free the used memory.
-   if (delete_fec)
-   {
-      delete fec;
-   }
+   delete a;
+   delete b;
+   delete fespace;
+   if (order > 0) { delete fec; }
+   delete pmesh;
+
   MPI_Finalize();

   return 0;
@@ -13,11 +13,6 @@
 //               ex22 -m ../data/inline-hex.mesh -o 2 -p 2
 //               ex22 -m ../data/star.mesh -r 1 -o 2 -sigma 10.0
 //
-// With partial assembly:
-//               ex22 -m ../data/inline-quad.mesh -o 3 -p 1 -pa
-//               ex22 -m ../data/inline-hex.mesh -o 2 -p 2 -pa
-//               ex22 -m ../data/star.mesh -r 1 -o 2 -sigma 10.0 -pa
-//
 // Description:  This example code demonstrates the use of MFEM to define and
 //               solve simple complex-valued linear systems. It implements three
 //               variants of a damped harmonic oscillator:
@@ -81,7 +76,6 @@ int main(int argc, char *argv[])
   bool visualization = 1;
   bool herm_conv = true;
   bool exact_sol = true;
-   bool pa = false;

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -112,8 +106,6 @@ int main(int argc, char *argv[])
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
-   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
-                  "--no-partial-assembly", "Enable Partial Assembly.");
   args.Parse();
   if (!args.Good())
   {
@@ -290,7 +282,6 @@ int main(int argc, char *argv[])
   ConstantCoefficient negMassCoef(omega_ * omega_ * epsilon_);

   SesquilinearForm *a = new SesquilinearForm(fespace, conv);
-   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   switch (prob)
   {
      case 0:
@@ -327,8 +318,6 @@ int main(int argc, char *argv[])
   //         -Grad(a Div) - omega^2 b + omega c
   //
   BilinearForm *pcOp = new BilinearForm(fespace);
-   if (pa) { pcOp->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
-
   switch (prob)
   {
      case 0:
@@ -359,8 +348,19 @@ int main(int argc, char *argv[])
   Vector B, U;

   a->FormLinearSystem(ess_tdof_list, u, b, A, U, B);
+   u = 0.0;
+   U = 0.0;

-   cout << "Size of linear system: " << A->Width() << endl << endl;
+   OperatorHandle PCOp;
+   pcOp->FormSystemMatrix(ess_tdof_list, PCOp);
+
+   {
+      ComplexSparseMatrix * Asp =
+         dynamic_cast<ComplexSparseMatrix*>(A.Ptr());
+
+      cout << "Size of linear system: "
+           << 2 * Asp->real().Width() << endl << endl;
+   }

   // 10. Define and apply a GMRES solver for AU=B with a block diagonal
   //     preconditioner based on the appropriate sparse smoother.
@@ -368,8 +368,8 @@ int main(int argc, char *argv[])
      Array<int> blockOffsets;
      blockOffsets.SetSize(3);
      blockOffsets[0] = 0;
-      blockOffsets[1] = A->Height() / 2;
-      blockOffsets[2] = A->Height() / 2;
+      blockOffsets[1] = PCOp.Ptr()->Height();
+      blockOffsets[2] = PCOp.Ptr()->Height();
      blockOffsets.PartialSum();

      BlockDiagonalPreconditioner BDP(blockOffsets);
@@ -377,31 +377,22 @@ int main(int argc, char *argv[])
      Operator * pc_r = NULL;
      Operator * pc_i = NULL;

-      if (pa)
+      double s = 1.0;
+      switch (prob)
      {
-         pc_r = new OperatorJacobiSmoother(*pcOp, ess_tdof_list);
+         case 0:
+            pc_r = new DSmoother(*PCOp.As<SparseMatrix>());
+            break;
+         case 1:
+            pc_r = new GSSmoother(*PCOp.As<SparseMatrix>());
+            s = -1.0;
+            break;
+         case 2:
+            pc_r = new DSmoother(*PCOp.As<SparseMatrix>());
+            break;
+
+         default: break; // This should be unreachable
      }
-      else
-      {
-         OperatorHandle PCOp;
-         pcOp->SetDiagonalPolicy(mfem::Operator::DIAG_ONE);
-         pcOp->FormSystemMatrix(ess_tdof_list, PCOp);
-         switch (prob)
-         {
-            case 0:
-               pc_r = new DSmoother(*PCOp.As<SparseMatrix>());
-               break;
-            case 1:
-               pc_r = new GSSmoother(*PCOp.As<SparseMatrix>());
-               break;
-            case 2:
-               pc_r = new DSmoother(*PCOp.As<SparseMatrix>());
-               break;
-            default:
-               break; // This should be unreachable
-         }
-      }
-      double s = (prob != 1) ? 1.0 : -1.0;
      pc_i = new ScaledOperator(pc_r,
                                (conv == ComplexOperator::HERMITIAN) ?
                                s:-s);
@@ -13,11 +13,6 @@
 //               mpirun -np 4 ex22p -m ../data/inline-hex.mesh -o 2 -p 2
 //               mpirun -np 4 ex22p -m ../data/star.mesh -o 2 -sigma 10.0
 //
-// With partial assembly:
-//               mpirun -np 4 ex22p -m ../data/inline-quad.mesh -o 1 -p 1 -pa
-//               mpirun -np 4 ex22p -m ../data/inline-hex.mesh -o 1 -p 2 -pa
-//               mpirun -np 4 ex22p -m ../data/star.mesh -o 2 -sigma 10.0 -pa
-//
 // Description:  This example code demonstrates the use of MFEM to define and
 //               solve simple complex-valued linear systems. It implements three
 //               variants of a damped harmonic oscillator:
@@ -89,7 +84,6 @@ int main(int argc, char *argv[])
   bool visualization = 1;
   bool herm_conv = true;
   bool exact_sol = true;
-   bool pa = false;

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -122,8 +116,6 @@ int main(int argc, char *argv[])
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
-   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
-                  "--no-partial-assembly", "Enable Partial Assembly.");
   args.Parse();
   if (!args.Good())
   {
@@ -323,7 +315,6 @@ int main(int argc, char *argv[])
   ConstantCoefficient negMassCoef(omega_ * omega_ * epsilon_);

   ParSesquilinearForm *a = new ParSesquilinearForm(fespace, conv);
-   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   switch (prob)
   {
      case 0:
@@ -360,7 +351,6 @@ int main(int argc, char *argv[])
   //         -Grad(a Div) - omega^2 b + omega c
   //
   ParBilinearForm *pcOp = new ParBilinearForm(fespace);
-   if (pa) { pcOp->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   switch (prob)
   {
      case 0:
@@ -392,11 +382,19 @@ int main(int argc, char *argv[])
   Vector B, U;

   a->FormLinearSystem(ess_tdof_list, u, b, A, U, B);
+   u = 0.0;
+   U = 0.0;
+
+   OperatorHandle PCOp;
+   pcOp->FormSystemMatrix(ess_tdof_list, PCOp);

   if (myid == 0)
   {
+      ComplexHypreParMatrix * Ahyp =
+         dynamic_cast<ComplexHypreParMatrix*>(A.Ptr());
+
      cout << "Size of linear system: "
-           << 2 * fespace->GlobalTrueVSize() << endl << endl;
+           << 2 * Ahyp->real().GetGlobalNumRows() << endl << endl;
   }

   // 12. Define and apply a parallel FGMRES solver for AU=B with a block
@@ -406,8 +404,8 @@ int main(int argc, char *argv[])
      Array<int> blockTrueOffsets;
      blockTrueOffsets.SetSize(3);
      blockTrueOffsets[0] = 0;
-      blockTrueOffsets[1] = A->Height() / 2;
-      blockTrueOffsets[2] = A->Height() / 2;
+      blockTrueOffsets[1] = PCOp.Ptr()->Height();
+      blockTrueOffsets[2] = PCOp.Ptr()->Height();
      blockTrueOffsets.PartialSum();

      BlockDiagonalPreconditioner BDP(blockTrueOffsets);
@@ -415,34 +413,25 @@ int main(int argc, char *argv[])
      Operator * pc_r = NULL;
      Operator * pc_i = NULL;

-      if (pa)
+      switch (prob)
      {
-         pc_r = new OperatorJacobiSmoother(*pcOp, ess_tdof_list);
-      }
-      else
-      {
-         OperatorHandle PCOp;
-         pcOp->FormSystemMatrix(ess_tdof_list, PCOp);
-         switch (prob)
-         {
-            case 0:
-               pc_r = new HypreBoomerAMG(*PCOp.As<HypreParMatrix>());
-               break;
-            case 1:
+         case 0:
+            pc_r = new HypreBoomerAMG(*PCOp.As<HypreParMatrix>());
+            break;
+         case 1:
+            pc_r = new HypreAMS(*PCOp.As<HypreParMatrix>(), fespace);
+            break;
+         case 2:
+            if (dim == 2 )
+            {
               pc_r = new HypreAMS(*PCOp.As<HypreParMatrix>(), fespace);
-               break;
-            case 2:
-               if (dim == 2 )
-               {
-                  pc_r = new HypreAMS(*PCOp.As<HypreParMatrix>(), fespace);
-               }
-               else
-               {
-                  pc_r = new HypreADS(*PCOp.As<HypreParMatrix>(), fespace);
-               }
-               break;
-            default: break; // This should be unreachable
-         }
+            }
+            else
+            {
+               pc_r = new HypreADS(*PCOp.As<HypreParMatrix>(), fespace);
+            }
+            break;
+         default: break; // This should be unreachable
      }
      pc_i = new ScaledOperator(pc_r,
                                (conv == ComplexOperator::HERMITIAN) ?
@@ -6,8 +6,6 @@
 //               ex24 -m ../data/square-disc.mesh -o 2
 //               ex24 -m ../data/beam-tet.mesh
 //               ex24 -m ../data/beam-hex.mesh -o 2 -pa
-//               ex24 -m ../data/beam-hex.mesh -o 2 -pa -p 1
-//               ex24 -m ../data/beam-hex.mesh -o 2 -pa -p 2
 //               ex24 -m ../data/escher.mesh
 //               ex24 -m ../data/escher.mesh -o 2
 //               ex24 -m ../data/fichera.mesh
@@ -25,16 +23,11 @@
 //               ex24 -m ../data/beam-hex.mesh -pa -d cuda
 //
 // Description:  This example code illustrates usage of mixed finite element
-//               spaces, with three variants:
+//               spaces. Using two different approaches, we project a gradient
+//               of a function in H^1 to H(curl). Other spaces and example
+//               computations are to be added in the future.
 //
-//               1) (grad p, u) for p in H^1 tested against u in H(curl)
-//               2) (curl v, u) for v in H(curl) tested against u in H(div), 3D
-//               3) (div v, q) for v in H(div) tested against q in L_2
-//
-//               Using different approaches, we project the gradient, curl, or
-//               divergence to the appropriate space.
-//
-//               We recommend viewing examples 1, 3, and 5 before viewing this
+//               We recommend viewing examples 1 and 3 before viewing this
 //               example.

 #include "mfem.hpp"
@@ -46,19 +39,14 @@ using namespace mfem;

 double p_exact(const Vector &x);
 void gradp_exact(const Vector &, Vector &);
-double div_gradp_exact(const Vector &x);
-void v_exact(const Vector &x, Vector &v);
-void curlv_exact(const Vector &x, Vector &cv);

 int dim;
-double freq = 1.0, kappa;

 int main(int argc, char *argv[])
 {
   // 1. Parse command-line options.
   const char *mesh_file = "../data/beam-hex.mesh";
   int order = 1;
-   int prob = 0;
   bool static_cond = false;
   bool pa = false;
   const char *device_config = "cpu";
@@ -69,8 +57,6 @@ int main(int argc, char *argv[])
                  "Mesh file to use.");
   args.AddOption(&order, "-o", "--order",
                  "Finite element order (polynomial degree).");
-   args.AddOption(&prob, "-p", "--problem-type",
-                  "Choose between 0: H(Curl) or 1: H(Div)");
   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
                  "--no-static-condensation", "Enable static condensation.");
   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
@@ -88,7 +74,6 @@ int main(int argc, char *argv[])
      return 1;
   }
   args.PrintOptions(cout);
-   kappa = freq * M_PI;

   // 2. Enable hardware devices such as GPUs, and programming models such as
   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
@@ -115,129 +100,72 @@ int main(int argc, char *argv[])
   }
   mesh->ReorientTetMesh();

-   // 5. Define a finite element space on the mesh. Here we use Nedelec or
-   //    Raviart-Thomas finite elements of the specified order.
-   FiniteElementCollection *trial_fec = NULL;
-   FiniteElementCollection *test_fec = NULL;
+   // 5. Define a parallel finite element space on the parallel mesh. Here we
+   //    use the Nedelec finite elements of the specified order.
+   FiniteElementCollection *fec = new ND_FECollection(order, dim);
+   FiniteElementCollection *H1fec = new H1_FECollection(order, dim);
+   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, fec);
+   FiniteElementSpace *H1fespace = new FiniteElementSpace(mesh, H1fec);

-   if (prob == 0)
-   {
-      trial_fec = new H1_FECollection(order, dim);
-      test_fec = new ND_FECollection(order, dim);
-   }
-   else if (prob == 1)
-   {
-      trial_fec = new ND_FECollection(order, dim);
-      test_fec = new RT_FECollection(order-1, dim);
-   }
-   else
-   {
-      trial_fec = new RT_FECollection(order-1, dim);
-      test_fec = new L2_FECollection(order-1, dim);
-   }
+   int size = fespace->GetTrueVSize();
+   int H1size = H1fespace->GetTrueVSize();
+   cout << "Number of Nedelec finite element unknowns: " << size << endl;
+   cout << "Number of H1 finite element unknowns: " << H1size << endl;

-   FiniteElementSpace trial_fes(mesh, trial_fec);
-   FiniteElementSpace test_fes(mesh, test_fec);
-
-   int trial_size = trial_fes.GetTrueVSize();
-   int test_size = test_fes.GetTrueVSize();
-
-   if (prob == 0)
-   {
-      cout << "Number of Nedelec finite element unknowns: " << test_size << endl;
-      cout << "Number of H1 finite element unknowns: " << trial_size << endl;
-   }
-   else if (prob == 1)
-   {
-      cout << "Number of Nedelec finite element unknowns: " << trial_size << endl;
-      cout << "Number of Raviart-Thomas finite element unknowns: " << test_size <<
-           endl;
-   }
-   else
-   {
-      cout << "Number of Raviart-Thomas finite element unknowns: "
-           << trial_size << endl;
-      cout << "Number of L2 finite element unknowns: " << test_size << endl;
-   }
-
-   // 6. Define the solution vector as a finite element grid function
-   //    corresponding to the trial fespace.
-   GridFunction gftest(&test_fes);
-   GridFunction gftrial(&trial_fes);
-   GridFunction x(&test_fes);
+   // 6. Define the solution vector x as a parallel finite element grid function
+   //    corresponding to fespace. Initialize x by projecting the exact
+   //    solution. Note that only values from the boundary edges will be used
+   //    when eliminating the non-homogeneous boundary condition to modify the
+   //    r.h.s. vector b.
+   GridFunction x(fespace);
   FunctionCoefficient p_coef(p_exact);
+   GridFunction p(H1fespace);
+   p.ProjectCoefficient(p_coef);
+   p.SetTrueVector();
+   p.SetFromTrueVector();
+
   VectorFunctionCoefficient gradp_coef(sdim, gradp_exact);
-   VectorFunctionCoefficient v_coef(sdim, v_exact);
-   VectorFunctionCoefficient curlv_coef(sdim, curlv_exact);
-   FunctionCoefficient divgradp_coef(div_gradp_exact);

-   if (prob == 0)
-   {
-      gftrial.ProjectCoefficient(p_coef);
-   }
-   else if (prob == 1)
-   {
-      gftrial.ProjectCoefficient(v_coef);
-   }
-   else
-   {
-      gftrial.ProjectCoefficient(gradp_coef);
-   }
-
-   gftrial.SetTrueVector();
-   gftrial.SetFromTrueVector();
-
-   // 7. Set up the bilinear forms for L2 projection.
-   ConstantCoefficient one(1.0);
-   BilinearForm a(&test_fes);
-   MixedBilinearForm a_mixed(&trial_fes, &test_fes);
+   // 7. Set up the bilinear forms.
+   Coefficient *muinv = new ConstantCoefficient(1.0);
+   Coefficient *sigma = new ConstantCoefficient(1.0);
+   BilinearForm *a = new BilinearForm(fespace);
+   MixedBilinearForm *a_NDH1 = new MixedBilinearForm(H1fespace, fespace);
   if (pa)
   {
-      a.SetAssemblyLevel(AssemblyLevel::PARTIAL);
-      a_mixed.SetAssemblyLevel(AssemblyLevel::PARTIAL);
+      a->SetAssemblyLevel(AssemblyLevel::PARTIAL);
+      a_NDH1->SetAssemblyLevel(AssemblyLevel::PARTIAL);
   }

-   if (prob == 0)
-   {
-      a.AddDomainIntegrator(new VectorFEMassIntegrator(one));
-      a_mixed.AddDomainIntegrator(new MixedVectorGradientIntegrator(one));
-   }
-   else if (prob == 1)
-   {
-      a.AddDomainIntegrator(new VectorFEMassIntegrator(one));
-      a_mixed.AddDomainIntegrator(new MixedVectorCurlIntegrator(one));
-   }
-   else
-   {
-      a.AddDomainIntegrator(new MassIntegrator(one));
-      a_mixed.AddDomainIntegrator(new VectorFEDivergenceIntegrator(one));
-   }
+   // First approach: L2 projection
+   a->AddDomainIntegrator(new VectorFEMassIntegrator(*sigma));
+   a_NDH1->AddDomainIntegrator(new MixedVectorGradientIntegrator(*muinv));

-   // 8. Assemble the bilinear form and the corresponding linear system,
-   //    applying any necessary transformations such as: eliminating boundary
-   //    conditions, applying conforming constraints for non-conforming AMR,
-   //    static condensation, etc.
-   if (static_cond) { a.EnableStaticCondensation(); }
+   // 8. Assemble the parallel bilinear form and the corresponding linear
+   //    system, applying any necessary transformations such as: parallel
+   //    assembly, eliminating boundary conditions, applying conforming
+   //    constraints for non-conforming AMR, static condensation, etc.
+   if (static_cond) { a->EnableStaticCondensation(); }

-   a.Assemble();
-   if (!pa) { a.Finalize(); }
+   a->Assemble();
+   if (!pa) { a->Finalize(); }

-   a_mixed.Assemble();
-   if (!pa) { a_mixed.Finalize(); }
+   a_NDH1->Assemble();
+   if (!pa) { a_NDH1->Finalize(); }

   if (pa)
   {
-      a_mixed.Mult(gftrial, x);
+      a_NDH1->Mult(p, x);
   }
   else
   {
-      SparseMatrix& mixed = a_mixed.SpMat();
-      mixed.Mult(gftrial, x);
+      SparseMatrix& NDH1 = a_NDH1->SpMat();
+      NDH1.Mult(p, x);
   }

   // 9. Define and apply a PCG solver for Ax = b with Jacobi preconditioner.
   {
-      GridFunction rhs(&test_fes);
+      GridFunction rhs(fespace);
      rhs = x;
      x = 0.0;

@@ -248,15 +176,15 @@ int main(int argc, char *argv[])
      if (pa)
      {
         Array<int> ess_tdof_list; // empty
-         OperatorJacobiSmoother Jacobi(a, ess_tdof_list);
+         OperatorJacobiSmoother Jacobi(*a, ess_tdof_list);

-         cg.SetOperator(a);
+         cg.SetOperator(*a);
         cg.SetPreconditioner(Jacobi);
         cg.Mult(rhs, x);
      }
      else
      {
-         SparseMatrix& Amat = a.SpMat();
+         SparseMatrix& Amat = a->SpMat();
         DSmoother Jacobi(Amat);

         cg.SetOperator(Amat);
@@ -265,89 +193,33 @@ int main(int argc, char *argv[])
      }
   }

-   // 10. Compute the same field by applying a DiscreteInterpolator.
-   GridFunction discreteInterpolant(&test_fes);
-   DiscreteLinearOperator dlo(&trial_fes, &test_fes);
-   if (prob == 0)
-   {
-      dlo.AddDomainInterpolator(new GradientInterpolator());
-   }
-   else if (prob == 1)
-   {
-      dlo.AddDomainInterpolator(new CurlInterpolator());
-   }
-   else
-   {
-      dlo.AddDomainInterpolator(new DivergenceInterpolator());
-   }
+   // 10. Second approach: compute the same solution by applying
+   //     GradientInterpolator in H(curl).
+   DiscreteLinearOperator grad(H1fespace, fespace);
+   grad.AddDomainInterpolator(new GradientInterpolator());
+   grad.Assemble();

-   dlo.Assemble();
-   dlo.Mult(gftrial, discreteInterpolant);
+   GridFunction gradp(fespace);
+   grad.Mult(p, gradp);

-   // 11. Compute the projection of the exact field.
-   GridFunction exact_proj(&test_fes);
-   if (prob == 0)
-   {
-      exact_proj.ProjectCoefficient(gradp_coef);
-   }
-   else if (prob == 1)
-   {
-      exact_proj.ProjectCoefficient(curlv_coef);
-   }
-   else
-   {
-      exact_proj.ProjectCoefficient(divgradp_coef);
-   }
+   // 11. Compute the projection of the exact grad p.
+   GridFunction exact_gradp(fespace);
+   exact_gradp.ProjectCoefficient(gradp_coef);
+   exact_gradp.SetTrueVector();
+   exact_gradp.SetFromTrueVector();

-   exact_proj.SetTrueVector();
-   exact_proj.SetFromTrueVector();
-
-   // 12. Compute and print the L_2 norm of the error.
-   if (prob == 0)
+   // 12. Compute and print the L^2 norm of the error.
   {
      double errSol = x.ComputeL2Error(gradp_coef);
-      double errInterp = discreteInterpolant.ComputeL2Error(gradp_coef);
-      double errProj = exact_proj.ComputeL2Error(gradp_coef);
+      double errInterp = gradp.ComputeL2Error(gradp_coef);
+      double errProj = exact_gradp.ComputeL2Error(gradp_coef);

      cout << "\n Solution of (E_h,v) = (grad p_h,v) for E_h and v in H(curl): "
-           "|| E_h - grad p ||_{L_2} = " << errSol << '\n' << endl;
+           "|| E_h - grad p ||_{L^2} = " << errSol << '\n' << endl;
      cout << " Gradient interpolant E_h = grad p_h in H(curl): || E_h - grad p"
-           " ||_{L_2} = " << errInterp << '\n' << endl;
+           "||_{L^2} = " << errInterp << '\n' << endl;
      cout << " Projection E_h of exact grad p in H(curl): || E_h - grad p "
-           "||_{L_2} = " << errProj << '\n' << endl;
-   }
-   else if (prob == 1)
-   {
-      double errSol = x.ComputeL2Error(curlv_coef);
-      double errInterp = discreteInterpolant.ComputeL2Error(curlv_coef);
-      double errProj = exact_proj.ComputeL2Error(curlv_coef);
-
-      cout << "\n Solution of (E_h,w) = (curl v_h,w) for E_h and w in H(div): "
-           "|| E_h - curl v ||_{L_2} = " << errSol << '\n' << endl;
-      cout << " Curl interpolant E_h = curl v_h in H(div): || E_h - curl v "
-           "||_{L_2} = " << errInterp << '\n' << endl;
-      cout << " Projection E_h of exact curl v in H(div): || E_h - curl v "
-           "||_{L_2} = " << errProj << '\n' << endl;
-   }
-   else
-   {
-      int order_quad = max(2, 2*order+1);
-      const IntegrationRule *irs[Geometry::NumGeom];
-      for (int i=0; i < Geometry::NumGeom; ++i)
-      {
-         irs[i] = &(IntRules.Get(i, order_quad));
-      }
-
-      double errSol = x.ComputeL2Error(divgradp_coef, irs);
-      double errInterp = discreteInterpolant.ComputeL2Error(divgradp_coef, irs);
-      double errProj = exact_proj.ComputeL2Error(divgradp_coef, irs);
-
-      cout << "\n Solution of (f_h,q) = (div v_h,q) for f_h and q in L_2: "
-           "|| f_h - div v ||_{L_2} = " << errSol << '\n' << endl;
-      cout << " Divergence interpolant f_h = div v_h in L_2: || f_h - div v "
-           "||_{L_2} = " << errInterp << '\n' << endl;
-      cout << " Projection f_h of exact div v in L_2: || f_h - div v "
-           "||_{L_2} = " << errProj << '\n' << endl;
+           "||_{L^2} = " << errProj << '\n' << endl;
   }

   // 13. Save the refined mesh and the solution. This output can be viewed
@@ -370,8 +242,14 @@ int main(int argc, char *argv[])
   }

   // 15. Free the used memory.
-   delete trial_fec;
-   delete test_fec;
+   delete a;
+   delete a_NDH1;
+   delete sigma;
+   delete muinv;
+   delete fespace;
+   delete H1fespace;
+   delete fec;
+   delete H1fec;
   delete mesh;

   return 0;
@@ -406,47 +284,3 @@ void gradp_exact(const Vector &x, Vector &f)
      if (x.Size() == 3) { f(2) = 0.0; }
   }
 }
-
-double div_gradp_exact(const Vector &x)
-{
-   if (dim == 3)
-   {
-      return -3.0 * sin(x(0)) * sin(x(1)) * sin(x(2));
-   }
-   else if (dim == 2)
-   {
-      return -2.0 * sin(x(0)) * sin(x(1));
-   }
-
-   return 0.0;
-}
-
-void v_exact(const Vector &x, Vector &v)
-{
-   if (dim == 3)
-   {
-      v(0) = sin(kappa * x(1));
-      v(1) = sin(kappa * x(2));
-      v(2) = sin(kappa * x(0));
-   }
-   else
-   {
-      v(0) = sin(kappa * x(1));
-      v(1) = sin(kappa * x(0));
-      if (x.Size() == 3) { v(2) = 0.0; }
-   }
-}
-
-void curlv_exact(const Vector &x, Vector &cv)
-{
-   if (dim == 3)
-   {
-      cv(0) = -kappa * cos(kappa * x(2));
-      cv(1) = -kappa * cos(kappa * x(0));
-      cv(2) = -kappa * cos(kappa * x(1));
-   }
-   else
-   {
-      cv = 0.0;
-   }
-}
@@ -6,8 +6,6 @@
 //               mpirun -np 4 ex24p -m ../data/square-disc.mesh -o 2
 //               mpirun -np 4 ex24p -m ../data/beam-tet.mesh
 //               mpirun -np 4 ex24p -m ../data/beam-hex.mesh -o 2 -pa
-//               mpirun -np 4 ex24p -m ../data/beam-hex.mesh -o 2 -pa -p 1
-//               mpirun -np 4 ex24p -m ../data/beam-hex.mesh -o 2 -pa -p 2
 //               mpirun -np 4 ex24p -m ../data/escher.mesh
 //               mpirun -np 4 ex24p -m ../data/escher.mesh -o 2
 //               mpirun -np 4 ex24p -m ../data/fichera.mesh
@@ -25,16 +23,11 @@
 //               mpirun -np 4 ex24p -m ../data/beam-hex.mesh -pa -d cuda
 //
 // Description:  This example code illustrates usage of mixed finite element
-//               spaces, with three variants:
+//               spaces. Using two different approaches, we project a gradient
+//               of a function in H^1 to H(curl). Other spaces and example
+//               computations are to be added in the future.
 //
-//               1) (grad p, u) for p in H^1 tested against u in H(curl)
-//               2) (curl v, u) for v in H(curl) tested against u in H(div), 3D
-//               3) (div v, q) for v in H(div) tested against q in L_2
-//
-//               Using different approaches, we project the gradient, curl, or
-//               divergence to the appropriate space.
-//
-//               We recommend viewing examples 1, 3, and 5 before viewing this
+//               We recommend viewing examples 1 and 3 before viewing this
 //               example.

 #include "mfem.hpp"
@@ -46,12 +39,8 @@ using namespace mfem;

 double p_exact(const Vector &x);
 void gradp_exact(const Vector &, Vector &);
-double div_gradp_exact(const Vector &x);
-void v_exact(const Vector &x, Vector &v);
-void curlv_exact(const Vector &x, Vector &cv);

 int dim;
-double freq = 1.0, kappa;

 int main(int argc, char *argv[])
 {
@@ -64,7 +53,6 @@ int main(int argc, char *argv[])
   // 2. Parse command-line options.
   const char *mesh_file = "../data/beam-hex.mesh";
   int order = 1;
-   int prob = 0;
   bool static_cond = false;
   bool pa = false;
   const char *device_config = "cpu";
@@ -75,8 +63,6 @@ int main(int argc, char *argv[])
                  "Mesh file to use.");
   args.AddOption(&order, "-o", "--order",
                  "Finite element order (polynomial degree).");
-   args.AddOption(&prob, "-p", "--problem-type",
-                  "Choose between 0: H(Curl) or 1: H(Div)");
   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
                  "--no-static-condensation", "Enable static condensation.");
   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
@@ -101,7 +87,6 @@ int main(int argc, char *argv[])
   {
      args.PrintOptions(cout);
   }
-   kappa = freq * M_PI;

   // 3. Enable hardware devices such as GPUs, and programming models such as
   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
@@ -144,137 +129,80 @@ int main(int argc, char *argv[])
   pmesh->ReorientTetMesh();

   // 7. Define a parallel finite element space on the parallel mesh. Here we
-   //    use Nedelec or Raviart-Thomas finite elements of the specified order.
-   FiniteElementCollection *trial_fec = NULL;
-   FiniteElementCollection *test_fec = NULL;
-
-   if (prob == 0)
-   {
-      trial_fec = new H1_FECollection(order, dim);
-      test_fec = new ND_FECollection(order, dim);
-   }
-   else if (prob == 1)
-   {
-      trial_fec = new ND_FECollection(order, dim);
-      test_fec = new RT_FECollection(order-1, dim);
-   }
-   else
-   {
-      trial_fec = new RT_FECollection(order-1, dim);
-      test_fec = new L2_FECollection(order-1, dim);
-   }
-
-   ParFiniteElementSpace trial_fes(pmesh, trial_fec);
-   ParFiniteElementSpace test_fes(pmesh, test_fec);
-
-   HYPRE_Int trial_size = trial_fes.GlobalTrueVSize();
-   HYPRE_Int test_size = test_fes.GlobalTrueVSize();
-
+   //    use the Nedelec finite elements of the specified order.
+   FiniteElementCollection *fec = new ND_FECollection(order, dim);
+   FiniteElementCollection *H1fec = new H1_FECollection(order, dim);
+   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
+   ParFiniteElementSpace *H1fespace = new ParFiniteElementSpace(pmesh, H1fec);
+   HYPRE_Int size = fespace->GlobalTrueVSize();
+   HYPRE_Int H1size = H1fespace->GlobalTrueVSize();
   if (myid == 0)
   {
-      if (prob == 0)
-      {
-         cout << "Number of Nedelec finite element unknowns: " << test_size << endl;
-         cout << "Number of H1 finite element unknowns: " << trial_size << endl;
-      }
-      else if (prob == 1)
-      {
-         cout << "Number of Nedelec finite element unknowns: " << trial_size << endl;
-         cout << "Number of Raviart-Thomas finite element unknowns: " << test_size <<
-              endl;
-      }
-      else
-      {
-         cout << "Number of Raviart-Thomas finite element unknowns: "
-              << trial_size << endl;
-         cout << "Number of L2 finite element unknowns: " << test_size << endl;
-      }
+      cout << "Number of Nedelec finite element unknowns: " << size << endl;
+      cout << "Number of H1 finite element unknowns: " << H1size << endl;
   }

-   // 8. Define the solution vector as a parallel finite element grid function
-   //    corresponding to the trial fespace.
-   ParGridFunction gftest(&test_fes);
-   ParGridFunction gftrial(&trial_fes);
-   ParGridFunction x(&test_fes);
+   // 8. Define the solution vector x as a parallel finite element grid function
+   //    corresponding to fespace. Initialize x by projecting the exact
+   //    solution. Note that only values from the boundary edges will be used
+   //    when eliminating the non-homogeneous boundary condition to modify the
+   //    r.h.s. vector b.
+   ParGridFunction x(fespace);
   FunctionCoefficient p_coef(p_exact);
+   ParGridFunction p(H1fespace);
+   p.ProjectCoefficient(p_coef);
+   p.SetTrueVector();
+   p.SetFromTrueVector();
+
   VectorFunctionCoefficient gradp_coef(sdim, gradp_exact);
-   VectorFunctionCoefficient v_coef(sdim, v_exact);
-   VectorFunctionCoefficient curlv_coef(sdim, curlv_exact);
-   FunctionCoefficient divgradp_coef(div_gradp_exact);

-   if (prob == 0)
-   {
-      gftrial.ProjectCoefficient(p_coef);
-   }
-   else if (prob == 1)
-   {
-      gftrial.ProjectCoefficient(v_coef);
-   }
-   else
-   {
-      gftrial.ProjectCoefficient(gradp_coef);
-   }
-
-   gftrial.SetTrueVector();
-   gftrial.SetFromTrueVector();
-
-   // 9. Set up the parallel bilinear forms for L2 projection.
-   ConstantCoefficient one(1.0);
-   ParBilinearForm a(&test_fes);
-   ParMixedBilinearForm a_mixed(&trial_fes, &test_fes);
+   // 9. Set up the parallel bilinear forms.
+   Coefficient *muinv = new ConstantCoefficient(1.0);
+   Coefficient *sigma = new ConstantCoefficient(1.0);
+   ParBilinearForm *a = new ParBilinearForm(fespace);
+   ParMixedBilinearForm *a_NDH1 = new ParMixedBilinearForm(H1fespace, fespace);
   if (pa)
   {
-      a.SetAssemblyLevel(AssemblyLevel::PARTIAL);
-      a_mixed.SetAssemblyLevel(AssemblyLevel::PARTIAL);
+      a->SetAssemblyLevel(AssemblyLevel::PARTIAL);
+      a_NDH1->SetAssemblyLevel(AssemblyLevel::PARTIAL);
   }

-   if (prob == 0)
-   {
-      a.AddDomainIntegrator(new VectorFEMassIntegrator(one));
-      a_mixed.AddDomainIntegrator(new MixedVectorGradientIntegrator(one));
-   }
-   else if (prob == 1)
-   {
-      a.AddDomainIntegrator(new VectorFEMassIntegrator(one));
-      a_mixed.AddDomainIntegrator(new MixedVectorCurlIntegrator(one));
-   }
-   else
-   {
-      a.AddDomainIntegrator(new MassIntegrator(one));
-      a_mixed.AddDomainIntegrator(new VectorFEDivergenceIntegrator(one));
-   }
+   // First approach: L2 projection
+   a->AddDomainIntegrator(new VectorFEMassIntegrator(*sigma));
+   a_NDH1->AddDomainIntegrator(new MixedVectorGradientIntegrator(*muinv));

   // 10. Assemble the parallel bilinear form and the corresponding linear
   //     system, applying any necessary transformations such as: parallel
   //     assembly, eliminating boundary conditions, applying conforming
   //     constraints for non-conforming AMR, static condensation, etc.
-   if (static_cond) { a.EnableStaticCondensation(); }
+   if (static_cond) { a->EnableStaticCondensation(); }

-   a.Assemble();
-   if (!pa) { a.Finalize(); }
+   a->Assemble();
+   if (!pa) { a->Finalize(); }

-   a_mixed.Assemble();
-   if (!pa) { a_mixed.Finalize(); }
+   a_NDH1->Assemble();
+   if (!pa) { a_NDH1->Finalize(); }

-   Vector B(test_fes.GetTrueVSize());
-   Vector X(test_fes.GetTrueVSize());
+   Vector B(fespace->GetTrueVSize());
+   Vector X(fespace->GetTrueVSize());

   if (pa)
   {
-      ParLinearForm b(&test_fes); // used as a vector
-      a_mixed.Mult(gftrial, b); // process-local multiplication
-      b.ParallelAssemble(B);
+      ParLinearForm *b = new ParLinearForm(fespace); // used as a vector
+      a_NDH1->Mult(p, *b); // process-local multiplication
+      b->ParallelAssemble(B);
+      delete b;
   }
   else
   {
-      HypreParMatrix *mixed = a_mixed.ParallelAssemble();
+      HypreParMatrix *NDH1 = a_NDH1->ParallelAssemble();

-      Vector P(trial_fes.GetTrueVSize());
-      gftrial.GetTrueDofs(P);
+      Vector P(H1fespace->GetTrueVSize());
+      p.GetTrueDofs(P);

-      mixed->Mult(P,B);
+      NDH1->Mult(P,B);

-      delete mixed;
+      delete NDH1;
   }

   // 11. Define and apply a parallel PCG solver for AX=B with Jacobi
@@ -284,9 +212,9 @@ int main(int argc, char *argv[])
      Array<int> ess_tdof_list; // empty

      OperatorPtr A;
-      a.FormSystemMatrix(ess_tdof_list, A);
+      a->FormSystemMatrix(ess_tdof_list, A);

-      OperatorJacobiSmoother Jacobi(a, ess_tdof_list);
+      OperatorJacobiSmoother Jacobi(*a, ess_tdof_list);

      CGSolver cg(MPI_COMM_WORLD);
      cg.SetRelTol(1e-12);
@@ -299,7 +227,7 @@ int main(int argc, char *argv[])
   }
   else
   {
-      HypreParMatrix *Amat = a.ParallelAssemble();
+      HypreParMatrix *Amat = a->ParallelAssemble();
      HypreDiagScale Jacobi(*Amat);
      HyprePCG pcg(*Amat);
      pcg.SetTol(1e-12);
@@ -314,97 +242,35 @@ int main(int argc, char *argv[])

   x.SetFromTrueDofs(X);

-   // 12. Compute the same field by applying a DiscreteInterpolator.
-   ParGridFunction discreteInterpolant(&test_fes);
-   ParDiscreteLinearOperator dlo(&trial_fes, &test_fes);
-   if (prob == 0)
-   {
-      dlo.AddDomainInterpolator(new GradientInterpolator());
-   }
-   else if (prob == 1)
-   {
-      dlo.AddDomainInterpolator(new CurlInterpolator());
-   }
-   else
-   {
-      dlo.AddDomainInterpolator(new DivergenceInterpolator());
-   }
+   // 12. Second approach: compute the same solution by applying
+   //     GradientInterpolator in H(curl).
+   ParDiscreteLinearOperator grad(H1fespace, fespace);
+   grad.AddDomainInterpolator(new GradientInterpolator());
+   grad.Assemble();

-   dlo.Assemble();
-   dlo.Mult(gftrial, discreteInterpolant);
+   ParGridFunction gradp(fespace);
+   grad.Mult(p, gradp);

-   // 13. Compute the projection of the exact field.
-   ParGridFunction exact_proj(&test_fes);
-   if (prob == 0)
-   {
-      exact_proj.ProjectCoefficient(gradp_coef);
-   }
-   else if (prob == 1)
-   {
-      exact_proj.ProjectCoefficient(curlv_coef);
-   }
-   else
-   {
-      exact_proj.ProjectCoefficient(divgradp_coef);
-   }
+   // 13. Compute the projection of the exact grad p.
+   ParGridFunction exact_gradp(fespace);
+   exact_gradp.ProjectCoefficient(gradp_coef);
+   exact_gradp.SetTrueVector();
+   exact_gradp.SetFromTrueVector();

-   exact_proj.SetTrueVector();
-   exact_proj.SetFromTrueVector();
-
-   // 14. Compute and print the L_2 norm of the error.
-   if (prob == 0)
+   // 14. Compute and print the L^2 norm of the error.
   {
      double errSol = x.ComputeL2Error(gradp_coef);
-      double errInterp = discreteInterpolant.ComputeL2Error(gradp_coef);
-      double errProj = exact_proj.ComputeL2Error(gradp_coef);
+      double errInterp = gradp.ComputeL2Error(gradp_coef);
+      double errProj = exact_gradp.ComputeL2Error(gradp_coef);

      if (myid == 0)
      {
-         cout << "\n Solution of (E_h,v) = (grad p_h,v) for E_h and v in H(curl)"
-              ": || E_h - grad p ||_{L_2} = " << errSol << '\n' << endl;
-         cout << " Gradient interpolant E_h = grad p_h in H(curl): || E_h - grad"
-              " p ||_{L_2} = " << errInterp << '\n' << endl;
+         cout << "\n Solution of (E_h,v) = (grad p_h,v) for E_h and v in "
+              "H(curl): || E_h - grad p ||_{L^2} = " << errSol << '\n' << endl;
+         cout << " Gradient interpolant E_h = grad p_h in H(curl): || E_h - "
+              "grad p ||_{L^2} = " << errInterp << '\n' << endl;
         cout << " Projection E_h of exact grad p in H(curl): || E_h - grad p "
-              "||_{L_2} = " << errProj << '\n' << endl;
-      }
-   }
-   else if (prob == 1)
-   {
-      double errSol = x.ComputeL2Error(curlv_coef);
-      double errInterp = discreteInterpolant.ComputeL2Error(curlv_coef);
-      double errProj = exact_proj.ComputeL2Error(curlv_coef);
-
-      if (myid == 0)
-      {
-         cout << "\n Solution of (E_h,w) = (curl v_h,w) for E_h and w in "
-              "H(div): || E_h - curl v ||_{L_2} = " << errSol << '\n' << endl;
-         cout << " Curl interpolant E_h = curl v_h in H(div): || E_h - curl v "
-              "||_{L_2} = " << errInterp << '\n' << endl;
-         cout << " Projection E_h of exact curl v in H(div): || E_h - curl v "
-              "||_{L_2} = " << errProj << '\n' << endl;
-      }
-   }
-   else
-   {
-      int order_quad = max(2, 2*order+1);
-      const IntegrationRule *irs[Geometry::NumGeom];
-      for (int i=0; i < Geometry::NumGeom; ++i)
-      {
-         irs[i] = &(IntRules.Get(i, order_quad));
-      }
-
-      double errSol = x.ComputeL2Error(divgradp_coef, irs);
-      double errInterp = discreteInterpolant.ComputeL2Error(divgradp_coef, irs);
-      double errProj = exact_proj.ComputeL2Error(divgradp_coef, irs);
-
-      if (myid == 0)
-      {
-         cout << "\n Solution of (f_h,q) = (div v_h,q) for f_h and q in L_2: "
-              "|| f_h - div v ||_{L_2} = " << errSol << '\n' << endl;
-         cout << " Divergence interpolant f_h = div v_h in L_2: || f_h - div v"
-              " ||_{L_2} = " << errInterp << '\n' << endl;
-         cout << " Projection f_h of exact div v in L_2: || f_h - div v "
-              "||_{L_2} = " << errProj << '\n' << endl;
+              "||_{L^2} = " << errProj << '\n' << endl;
      }
   }

@@ -436,8 +302,14 @@ int main(int argc, char *argv[])
   }

   // 17. Free the used memory.
-   delete trial_fec;
-   delete test_fec;
+   delete a;
+   delete a_NDH1;
+   delete sigma;
+   delete muinv;
+   delete fespace;
+   delete H1fespace;
+   delete fec;
+   delete H1fec;
   delete pmesh;

   MPI_Finalize();
@@ -474,47 +346,3 @@ void gradp_exact(const Vector &x, Vector &f)
      if (x.Size() == 3) { f(2) = 0.0; }
   }
 }
-
-double div_gradp_exact(const Vector &x)
-{
-   if (dim == 3)
-   {
-      return -3.0 * sin(x(0)) * sin(x(1)) * sin(x(2));
-   }
-   else if (dim == 2)
-   {
-      return -2.0 * sin(x(0)) * sin(x(1));
-   }
-
-   return 0.0;
-}
-
-void v_exact(const Vector &x, Vector &v)
-{
-   if (dim == 3)
-   {
-      v(0) = sin(kappa * x(1));
-      v(1) = sin(kappa * x(2));
-      v(2) = sin(kappa * x(0));
-   }
-   else
-   {
-      v(0) = sin(kappa * x(1));
-      v(1) = sin(kappa * x(0));
-      if (x.Size() == 3) { v(2) = 0.0; }
-   }
-}
-
-void curlv_exact(const Vector &x, Vector &cv)
-{
-   if (dim == 3)
-   {
-      cv(0) = -kappa * cos(kappa * x(2));
-      cv(1) = -kappa * cos(kappa * x(0));
-      cv(2) = -kappa * cos(kappa * x(1));
-   }
-   else
-   {
-      cv = 0.0;
-   }
-}
@@ -389,22 +389,27 @@ int main(int argc, char *argv[])
   //     applying any necessary transformations such as: assembly, eliminating
   //     boundary conditions, applying conforming constraints for
   //     non-conforming AMR, etc.
-   a.Assemble(0);
+   a.Assemble();

-   OperatorPtr A;
+   OperatorHandle Ah;
   Vector B, X;
-   a.FormLinearSystem(ess_tdof_list, x, b, A, X, B);
+   a.FormLinearSystem(ess_tdof_list, x, b, Ah, X, B);

-   // 13. Solve using a direct or an iterative solver
+   // 13. Transform to monolithic SparseMatrix
+   SparseMatrix *A = Ah.As<ComplexSparseMatrix>()->GetSystemMatrix();
+
+   cout << "Size of linear system: " << A->Height() << endl;
+
+   // 14. Solve using a direct or an iterative solver
 #ifdef MFEM_USE_SUITESPARSE
   {
-      ComplexUMFPackSolver csolver(*A.As<ComplexSparseMatrix>());
-      csolver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
-      csolver.SetPrintLevel(1);
-      csolver.Mult(B, X);
+      UMFPackSolver  solver(*A);
+      solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
+      solver.Mult(B, X);
   }
 #else
-   // 13a. Set up the Bilinear form a(.,.) for the preconditioner
+
+   // 14a. Set up the Bilinear form a(.,.) for the preconditioner
   //
   //    In Comp
   //    Domain:   1/mu (Curl E, Curl F) + omega^2 * epsilon (E,F)
@@ -432,10 +437,10 @@ int main(int argc, char *argv[])

      prec.Assemble();

-      OperatorPtr PCOpAh;
+      OperatorHandle PCOpAh;
      prec.FormSystemMatrix(ess_tdof_list, PCOpAh);

-      // 13b. Define and apply a GMRES solver for AU=B with a block diagonal
+      // 14b. Define and apply a GMRES solver for AU=B with a block diagonal
      //      preconditioner based on the Gauss-Seidel sparse smoother.
      Array<int> offsets(3);
      offsets[0] = 0;
@@ -462,15 +467,17 @@ int main(int argc, char *argv[])
   }
 #endif

-   // 14. Recover the solution as a finite element grid function and compute the
+   // 15. Recover the solution as a finite element grid function and compute the
   //     errors if the exact solution is known.
   a.RecoverFEMSolution(X, b, x);

   // If exact is known compute the error
   if (exact_known)
   {
+      ComplexGridFunction x_gf(fespace);
      VectorFunctionCoefficient E_ex_Re(dim, E_exact_Re);
      VectorFunctionCoefficient E_ex_Im(dim, E_exact_Im);
+      x_gf.ProjectCoefficient(E_ex_Re, E_ex_Im);
      int order_quad = max(2, 2 * order + 1);
      const IntegrationRule *irs[Geometry::NumGeom];
      for (int i = 0; i < Geometry::NumGeom; ++i)
@@ -499,7 +506,7 @@ int main(int argc, char *argv[])
           << sqrt(L2Error_Re*L2Error_Re + L2Error_Im*L2Error_Im) << "\n\n";
   }

-   // 15. Save the refined mesh and the solution. This output can be viewed
+   // 16. Save the refined mesh and the solution. This output can be viewed
   //     later using GLVis: "glvis -m mesh -g sol".
   {
      ofstream mesh_ofs("ex25.mesh");
@@ -514,7 +521,7 @@ int main(int argc, char *argv[])
      x.imag().Save(sol_i_ofs);
   }

-   // 16. Send the solution by socket to a GLVis server.
+   // 17. Send the solution by socket to a GLVis server.
   if (visualization)
   {
      // Define visualization keys for GLVis (see GLVis documentation)
@@ -565,7 +572,8 @@ int main(int argc, char *argv[])
      }
   }

-   // 17. Free the used memory.
+   // 18. Free the used memory.
+   delete A;
   delete pml;
   delete fespace;
   delete fec;
@@ -419,15 +419,21 @@ int main(int argc, char *argv[])
   //     constraints for non-conforming AMR, etc.
   a.Assemble();

-   OperatorPtr Ah;
+   OperatorHandle Ah;
   Vector B, X;
   a.FormLinearSystem(ess_tdof_list, x, b, Ah, X, B);

-   // 15. Solve using a direct or an iterative solver
+   // 15. Transform to monolithic HypreParMatrix
+   HypreParMatrix *A = Ah.As<ComplexHypreParMatrix>()->GetSystemMatrix();
+
+   if (myid == 0)
+   {
+      cout << "Size of linear system: " << A->GetGlobalNumRows() << endl;
+   }
+
+   // 16. Solve using a direct or an iterative solver
 #ifdef MFEM_USE_SUPERLU
   {
-      // Transform to monolithic HypreParMatrix
-      HypreParMatrix *A = Ah.As<ComplexHypreParMatrix>()->GetSystemMatrix();
      SuperLURowLocMatrix SA(*A);
      SuperLUSolver superlu(MPI_COMM_WORLD);
      superlu.SetPrintStatistics(false);
@@ -435,9 +441,9 @@ int main(int argc, char *argv[])
      superlu.SetColumnPermutation(superlu::PARMETIS);
      superlu.SetOperator(SA);
      superlu.Mult(B, X);
-      delete A;
   }
 #else
+
   // 16a. Set up the parallel Bilinear form a(.,.) for the preconditioner
   //
   //    In Comp
@@ -466,7 +472,7 @@ int main(int argc, char *argv[])

      prec.Assemble();

-      OperatorPtr PCOpAh;
+      OperatorHandle PCOpAh;
      prec.FormSystemMatrix(ess_tdof_list, PCOpAh);

      // 16b. Define and apply a parallel GMRES solver for AU=B with a block
@@ -490,7 +496,7 @@ int main(int argc, char *argv[])
      gmres.SetMaxIter(2000);
      gmres.SetRelTol(1e-5);
      gmres.SetAbsTol(0.0);
-      gmres.SetOperator(*Ah);
+      gmres.SetOperator(*A);
      gmres.SetPreconditioner(BlockAMS);
      gmres.Mult(B, X);
   }
@@ -503,8 +509,10 @@ int main(int argc, char *argv[])
   // If exact is known compute the error
   if (exact_known)
   {
+      ParComplexGridFunction x_gf(fespace);
      VectorFunctionCoefficient E_ex_Re(dim, E_exact_Re);
      VectorFunctionCoefficient E_ex_Im(dim, E_exact_Im);
+      x_gf.ProjectCoefficient(E_ex_Re, E_ex_Im);
      int order_quad = max(2, 2 * order + 1);
      const IntegrationRule *irs[Geometry::NumGeom];
      for (int i = 0; i < Geometry::NumGeom; ++i)
@@ -621,6 +629,7 @@ int main(int argc, char *argv[])
   }

   // 20. Free the used memory.
+   delete A;
   delete pml;
   delete fespace;
   delete fec;
@@ -274,6 +274,13 @@ int main(int argc, char *argv[])
      pmesh->SetNodalFESpace(fespace);
   }

+   {
+      x.Save("ex2p.gf", 1);
+      ParGridFunction new_x(fespace, "ex2p.gf");
+      new_x -= x;
+      out << "GF difference: " << new_x.Norml1() << endl;
+   }
+
   // 16. Save in parallel the displaced mesh and the inverted solution (which
   //     gives the backward displacements to the original grid). This output
   //     can be viewed later using GLVis: "glvis -np <np> -m mesh -g sol".
@@ -6,7 +6,6 @@
 //               ex4 -m ../data/star.mesh
 //               ex4 -m ../data/beam-tet.mesh
 //               ex4 -m ../data/beam-hex.mesh
-//               ex4 -m ../data/beam-hex.mesh -o 2 -pa
 //               ex4 -m ../data/escher.mesh
 //               ex4 -m ../data/fichera.mesh -o 2 -hb
 //               ex4 -m ../data/fichera-q2.vtk
@@ -21,12 +20,6 @@
 //               ex4 -m ../data/fichera-amr.mesh -o 2 -sc
 //               ex4 -m ../data/star-surf.mesh -o 1
 //
-// Device sample runs:
-//               ex4 -m ../data/star.mesh -pa -d cuda
-//               ex4 -m ../data/star.mesh -pa -d raja-cuda
-//               ex4 -m ../data/star.mesh -pa -d raja-omp
-//               ex4 -m ../data/beam-hex.mesh -pa -d cuda
-//
 // Description:  This example code solves a simple 2D/3D H(div) diffusion
 //               problem corresponding to the second order definite equation
 //               -grad(alpha div F) + beta F = f with boundary condition F dot n
@@ -62,8 +55,6 @@ int main(int argc, char *argv[])
   bool set_bc = true;
   bool static_cond = false;
   bool hybridization = false;
-   bool pa = false;
-   const char *device_config = "cpu";
   bool visualization = 1;

   OptionsParser args(argc, argv);
@@ -79,10 +70,6 @@ int main(int argc, char *argv[])
                  "--no-static-condensation", "Enable static condensation.");
   args.AddOption(&hybridization, "-hb", "--hybridization", "-no-hb",
                  "--no-hybridization", "Enable hybridization.");
-   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
-                  "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -95,19 +82,14 @@ int main(int argc, char *argv[])
   args.PrintOptions(cout);
   kappa = freq * M_PI;

-   // 2. Enable hardware devices such as GPUs, and programming models such as
-   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
-   Device device(device_config);
-   device.Print();
-
-   // 3. Read the mesh from the given mesh file. We can handle triangular,
+   // 2. Read the mesh from the given mesh file. We can handle triangular,
   //    quadrilateral, tetrahedral, hexahedral, surface and volume, as well as
   //    periodic meshes with the same code.
   Mesh *mesh = new Mesh(mesh_file, 1, 1);
   int dim = mesh->Dimension();
   int sdim = mesh->SpaceDimension();

-   // 4. Refine the mesh to increase the resolution. In this example we do
+   // 3. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement. We choose 'ref_levels' to be the
   //    largest number that gives a final mesh with no more than 25,000
   //    elements.
@@ -120,14 +102,14 @@ int main(int argc, char *argv[])
      }
   }

-   // 5. Define a finite element space on the mesh. Here we use the
+   // 4. Define a finite element space on the mesh. Here we use the
   //    Raviart-Thomas finite elements of the specified order.
   FiniteElementCollection *fec = new RT_FECollection(order-1, dim);
   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, fec);
   cout << "Number of finite element unknowns: "
        << fespace->GetTrueVSize() << endl;

-   // 6. Determine the list of true (i.e. conforming) essential boundary dofs.
+   // 5. Determine the list of true (i.e. conforming) essential boundary dofs.
   //    In this example, the boundary conditions are defined by marking all
   //    the boundary attributes from the mesh as essential (Dirichlet) and
   //    converting them to a list of true dofs.
@@ -139,7 +121,7 @@ int main(int argc, char *argv[])
      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
   }

-   // 7. Set up the linear form b(.) which corresponds to the right-hand side
+   // 6. Set up the linear form b(.) which corresponds to the right-hand side
   //    of the FEM linear system, which in this case is (f,phi_i) where f is
   //    given by the function f_exact and phi_i are the basis functions in the
   //    finite element fespace.
@@ -148,7 +130,7 @@ int main(int argc, char *argv[])
   b->AddDomainIntegrator(new VectorFEDomainLFIntegrator(f));
   b->Assemble();

-   // 8. Define the solution vector x as a finite element grid function
+   // 7. Define the solution vector x as a finite element grid function
   //    corresponding to fespace. Initialize x by projecting the exact
   //    solution. Note that only values from the boundary faces will be used
   //    when eliminating the non-homogeneous boundary condition to modify the
@@ -157,17 +139,16 @@ int main(int argc, char *argv[])
   VectorFunctionCoefficient F(sdim, F_exact);
   x.ProjectCoefficient(F);

-   // 9. Set up the bilinear form corresponding to the H(div) diffusion operator
+   // 8. Set up the bilinear form corresponding to the H(div) diffusion operator
   //    grad alpha div + beta I, by adding the div-div and the mass domain
   //    integrators.
   Coefficient *alpha = new ConstantCoefficient(1.0);
   Coefficient *beta  = new ConstantCoefficient(1.0);
   BilinearForm *a = new BilinearForm(fespace);
-   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   a->AddDomainIntegrator(new DivDivIntegrator(*alpha));
   a->AddDomainIntegrator(new VectorFEMassIntegrator(*beta));

-   // 10. Assemble the bilinear form and the corresponding linear system,
+   // 9. Assemble the bilinear form and the corresponding linear system,
   //    applying any necessary transformations such as: eliminating boundary
   //    conditions, applying conforming constraints for non-conforming AMR,
   //    static condensation, hybridization, etc.
@@ -186,47 +167,32 @@ int main(int argc, char *argv[])
   }
   a->Assemble();

-   OperatorPtr A;
+   SparseMatrix A;
   Vector B, X;
   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);

-   cout << "Size of linear system: " << A->Height() << endl;
+   cout << "Size of linear system: " << A.Height() << endl;

-   // 11. Solve the linear system A X = B.
-   if (!pa)
-   {
 #ifndef MFEM_USE_SUITESPARSE
-      // Use a simple symmetric Gauss-Seidel preconditioner with PCG.
-      GSSmoother M((SparseMatrix&)(*A));
-      PCG(*A, M, B, X, 1, 10000, 1e-20, 0.0);
+   // 10. Define a simple symmetric Gauss-Seidel preconditioner and use it to
+   //     solve the system A X = B with PCG.
+   GSSmoother M(A);
+   PCG(A, M, B, X, 1, 10000, 1e-20, 0.0);
 #else
-      // If MFEM was compiled with SuiteSparse, use UMFPACK to solve the system.
-      UMFPackSolver umf_solver;
-      umf_solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
-      umf_solver.SetOperator(*A);
-      umf_solver.Mult(B, X);
+   // 10. If compiled with SuiteSparse support, use UMFPACK to solve the system.
+   UMFPackSolver umf_solver;
+   umf_solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
+   umf_solver.SetOperator(A);
+   umf_solver.Mult(B, X);
 #endif
-   }
-   else // Jacobi preconditioning in partial assembly mode
-   {
-      if (UsesTensorBasis(*fespace))
-      {
-         OperatorJacobiSmoother M(*a, ess_tdof_list);
-         PCG(*A, M, B, X, 1, 10000, 1e-20, 0.0);
-      }
-      else
-      {
-         CG(*A, B, X, 1, 10000, 1e-20, 0.0);
-      }
-   }

-   // 12. Recover the solution as a finite element grid function.
+   // 11. Recover the solution as a finite element grid function.
   a->RecoverFEMSolution(X, *b, x);

-   // 13. Compute and print the L^2 norm of the error.
+   // 12. Compute and print the L^2 norm of the error.
   cout << "\n|| F_h - F ||_{L^2} = " << x.ComputeL2Error(F) << '\n' << endl;

-   // 14. Save the refined mesh and the solution. This output can be viewed
+   // 13. Save the refined mesh and the solution. This output can be viewed
   //     later using GLVis: "glvis -m refined.mesh -g sol.gf".
   {
      ofstream mesh_ofs("refined.mesh");
@@ -237,7 +203,7 @@ int main(int argc, char *argv[])
      x.Save(sol_ofs);
   }

-   // 15. Send the solution by socket to a GLVis server.
+   // 14. Send the solution by socket to a GLVis server.
   if (visualization)
   {
      char vishost[] = "localhost";
@@ -247,7 +213,7 @@ int main(int argc, char *argv[])
      sol_sock << "solution\n" << *mesh << x << flush;
   }

-   // 16. Free the used memory.
+   // 15. Free the used memory.
   delete hfes;
   delete hfec;
   delete a;
@@ -269,7 +235,7 @@ void F_exact(const Vector &p, Vector &F)

   double x = p(0);
   double y = p(1);
-   // double z = (dim == 3) ? p(2) : 0.0; // Uncomment if F is changed to depend on z
+   // double z = (dim == 3) ? p(2) : 0.0;

   F(0) = cos(kappa*x)*sin(kappa*y);
   F(1) = cos(kappa*y)*sin(kappa*x);
@@ -286,7 +252,7 @@ void f_exact(const Vector &p, Vector &f)

   double x = p(0);
   double y = p(1);
-   // double z = (dim == 3) ? p(2) : 0.0; // Uncomment if f is changed to depend on z
+   // double z = (dim == 3) ? p(2) : 0.0;

   double temp = 1 + 2*kappa*kappa;

@@ -6,7 +6,6 @@
 //               mpirun -np 4 ex4p -m ../data/star.mesh
 //               mpirun -np 4 ex4p -m ../data/beam-tet.mesh
 //               mpirun -np 4 ex4p -m ../data/beam-hex.mesh
-//               mpirun -np 4 ex4p -m ../data/beam-hex.mesh -o 2 -pa
 //               mpirun -np 4 ex4p -m ../data/escher.mesh -o 2 -sc
 //               mpirun -np 4 ex4p -m ../data/fichera.mesh -o 2 -hb
 //               mpirun -np 4 ex4p -m ../data/fichera-q2.vtk
@@ -16,17 +15,10 @@
 //               mpirun -np 4 ex4p -m ../data/periodic-square.mesh -no-bc
 //               mpirun -np 4 ex4p -m ../data/periodic-cube.mesh -no-bc
 //               mpirun -np 4 ex4p -m ../data/amr-quad.mesh
-//               mpirun -np 3 ex4p -m ../data/amr-quad.mesh -o 2 -hb
 //               mpirun -np 4 ex4p -m ../data/amr-hex.mesh -o 2 -sc
 //               mpirun -np 4 ex4p -m ../data/amr-hex.mesh -o 2 -hb
 //               mpirun -np 4 ex4p -m ../data/star-surf.mesh -o 3 -hb
 //
-// Device sample runs:
-//               mpirun -np 4 ex4p -m ../data/star.mesh -pa -d cuda
-//               mpirun -np 4 ex4p -m ../data/star.mesh -pa -d raja-cuda
-//               mpirun -np 4 ex4p -m ../data/star.mesh -pa -d raja-omp
-//               mpirun -np 4 ex4p -m ../data/beam-hex.mesh -pa -d cuda
-//
 // Description:  This example code solves a simple 2D/3D H(div) diffusion
 //               problem corresponding to the second order definite equation
 //               -grad(alpha div F) + beta F = f with boundary condition F dot n
@@ -68,8 +60,6 @@ int main(int argc, char *argv[])
   bool set_bc = true;
   bool static_cond = false;
   bool hybridization = false;
-   bool pa = false;
-   const char *device_config = "cpu";
   bool visualization = 1;

   OptionsParser args(argc, argv);
@@ -85,10 +75,6 @@ int main(int argc, char *argv[])
                  "--no-static-condensation", "Enable static condensation.");
   args.AddOption(&hybridization, "-hb", "--hybridization", "-no-hb",
                  "--no-hybridization", "Enable hybridization.");
-   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
-                  "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -108,19 +94,14 @@ int main(int argc, char *argv[])
   }
   kappa = freq * M_PI;

-   // 3. Enable hardware devices such as GPUs, and programming models such as
-   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
-   Device device(device_config);
-   if (myid == 0) { device.Print(); }
-
-   // 4. Read the (serial) mesh from the given mesh file on all processors.  We
+   // 3. Read the (serial) mesh from the given mesh file on all processors.  We
   //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
   //    and volume, as well as periodic meshes with the same code.
   Mesh *mesh = new Mesh(mesh_file, 1, 1);
   int dim = mesh->Dimension();
   int sdim = mesh->SpaceDimension();

-   // 5. Refine the serial mesh on all processors to increase the resolution. In
+   // 4. Refine the serial mesh on all processors to increase the resolution. In
   //    this example we do 'ref_levels' of uniform refinement. We choose
   //    'ref_levels' to be the largest number that gives a final mesh with no
   //    more than 1,000 elements.
@@ -133,7 +114,7 @@ int main(int argc, char *argv[])
      }
   }

-   // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
+   // 5. Define a parallel mesh by a partitioning of the serial mesh. Refine
   //    this mesh further in parallel to increase the resolution. Once the
   //    parallel mesh is defined, the serial mesh can be deleted. Tetrahedral
   //    meshes need to be reoriented before we can define high-order Nedelec
@@ -149,7 +130,7 @@ int main(int argc, char *argv[])
   }
   pmesh->ReorientTetMesh();

-   // 7. Define a parallel finite element space on the parallel mesh. Here we
+   // 6. Define a parallel finite element space on the parallel mesh. Here we
   //    use the Raviart-Thomas finite elements of the specified order.
   FiniteElementCollection *fec = new RT_FECollection(order-1, dim);
   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
@@ -159,7 +140,7 @@ int main(int argc, char *argv[])
      cout << "Number of finite element unknowns: " << size << endl;
   }

-   // 8. Determine the list of true (i.e. parallel conforming) essential
+   // 7. Determine the list of true (i.e. parallel conforming) essential
   //    boundary dofs. In this example, the boundary conditions are defined
   //    by marking all the boundary attributes from the mesh as essential
   //    (Dirichlet) and converting them to a list of true dofs.
@@ -171,7 +152,7 @@ int main(int argc, char *argv[])
      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
   }

-   // 9. Set up the parallel linear form b(.) which corresponds to the
+   // 8. Set up the parallel linear form b(.) which corresponds to the
   //    right-hand side of the FEM linear system, which in this case is
   //    (f,phi_i) where f is given by the function f_exact and phi_i are the
   //    basis functions in the finite element fespace.
@@ -180,7 +161,7 @@ int main(int argc, char *argv[])
   b->AddDomainIntegrator(new VectorFEDomainLFIntegrator(f));
   b->Assemble();

-   // 10. Define the solution vector x as a parallel finite element grid function
+   // 9. Define the solution vector x as a parallel finite element grid function
   //    corresponding to fespace. Initialize x by projecting the exact
   //    solution. Note that only values from the boundary faces will be used
   //    when eliminating the non-homogeneous boundary condition to modify the
@@ -189,17 +170,16 @@ int main(int argc, char *argv[])
   VectorFunctionCoefficient F(sdim, F_exact);
   x.ProjectCoefficient(F);

-   // 11. Set up the parallel bilinear form corresponding to the H(div)
+   // 10. Set up the parallel bilinear form corresponding to the H(div)
   //     diffusion operator grad alpha div + beta I, by adding the div-div and
   //     the mass domain integrators.
   Coefficient *alpha = new ConstantCoefficient(1.0);
   Coefficient *beta  = new ConstantCoefficient(1.0);
   ParBilinearForm *a = new ParBilinearForm(fespace);
-   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   a->AddDomainIntegrator(new DivDivIntegrator(*alpha));
   a->AddDomainIntegrator(new VectorFEMassIntegrator(*beta));

-   // 12. Assemble the parallel bilinear form and the corresponding linear
+   // 11. Assemble the parallel bilinear form and the corresponding linear
   //     system, applying any necessary transformations such as: parallel
   //     assembly, eliminating boundary conditions, applying conforming
   //     constraints for non-conforming AMR, static condensation,
@@ -219,43 +199,41 @@ int main(int argc, char *argv[])
   }
   a->Assemble();

-   OperatorPtr A;
+   HypreParMatrix A;
   Vector B, X;
   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);

-   if (myid == 0 && !pa)
+   HYPRE_Int glob_size = A.GetGlobalNumRows();
+   if (myid == 0)
   {
-      cout << "Size of linear system: "
-           << A.As<HypreParMatrix>()->GetGlobalNumRows() << endl;
+      cout << "Size of linear system: " << glob_size << endl;
   }

-   // 13. Define and apply a parallel PCG solver for A X = B with the 2D AMS or
+   // 12. Define and apply a parallel PCG solver for A X = B with the 2D AMS or
   //     the 3D ADS preconditioners from hypre. If using hybridization, the
-   //     system is preconditioned with hypre's BoomerAMG. In the partial
-   //     assembly case, use Jacobi preconditioning.
-   Solver *prec = NULL;
-   CGSolver *pcg = new CGSolver(MPI_COMM_WORLD);
-   pcg->SetOperator(*A);
+   //     system is preconditioned with hypre's BoomerAMG.
+   HypreSolver *prec = NULL;
+   CGSolver *pcg = new CGSolver(A.GetComm());
+   pcg->SetOperator(A);
   pcg->SetRelTol(1e-12);
-   pcg->SetMaxIter(2000);
+   pcg->SetMaxIter(500);
   pcg->SetPrintLevel(1);
-   if (hybridization) { prec = new HypreBoomerAMG(*A.As<HypreParMatrix>()); }
-   else if (pa) { prec = new OperatorJacobiSmoother(*a, ess_tdof_list); }
+   if (hybridization) { prec = new HypreBoomerAMG(A); }
   else
   {
      ParFiniteElementSpace *prec_fespace =
         (a->StaticCondensationIsEnabled() ? a->SCParFESpace() : fespace);
-      if (dim == 2)   { prec = new HypreAMS(*A.As<HypreParMatrix>(), prec_fespace); }
-      else            { prec = new HypreADS(*A.As<HypreParMatrix>(), prec_fespace); }
+      if (dim == 2)   { prec = new HypreAMS(A, prec_fespace); }
+      else            { prec = new HypreADS(A, prec_fespace); }
   }
   pcg->SetPreconditioner(*prec);
   pcg->Mult(B, X);

-   // 14. Recover the parallel grid function corresponding to X. This is the
+   // 13. Recover the parallel grid function corresponding to X. This is the
   //     local finite element solution on each processor.
   a->RecoverFEMSolution(X, *b, x);

-   // 15. Compute and print the L^2 norm of the error.
+   // 14. Compute and print the L^2 norm of the error.
   {
      double err = x.ComputeL2Error(F);
      if (myid == 0)
@@ -264,7 +242,7 @@ int main(int argc, char *argv[])
      }
   }

-   // 16. Save the refined mesh and the solution in parallel. This output can
+   // 15. Save the refined mesh and the solution in parallel. This output can
   //     be viewed later using GLVis: "glvis -np <np> -m mesh -g sol".
   {
      ostringstream mesh_name, sol_name;
@@ -280,7 +258,7 @@ int main(int argc, char *argv[])
      x.Save(sol_ofs);
   }

-   // 17. Send the solution by socket to a GLVis server.
+   // 16. Send the solution by socket to a GLVis server.
   if (visualization)
   {
      char vishost[] = "localhost";
@@ -291,7 +269,7 @@ int main(int argc, char *argv[])
      sol_sock << "solution\n" << *pmesh << x << flush;
   }

-   // 18. Free the used memory.
+   // 17. Free the used memory.
   delete pcg;
   delete prec;
   delete hfes;
@@ -317,7 +295,7 @@ void F_exact(const Vector &p, Vector &F)

   double x = p(0);
   double y = p(1);
-   // double z = (dim == 3) ? p(2) : 0.0; // Uncomment if F is changed to depend on z
+   // double z = (dim == 3) ? p(2) : 0.0;

   F(0) = cos(kappa*x)*sin(kappa*y);
   F(1) = cos(kappa*y)*sin(kappa*x);
@@ -334,7 +312,7 @@ void f_exact(const Vector &p, Vector &f)

   double x = p(0);
   double y = p(1);
-   // double z = (dim == 3) ? p(2) : 0.0; // Uncomment if f is changed to depend on z
+   // double z = (dim == 3) ? p(2) : 0.0;

   double temp = 1 + 2*kappa*kappa;

@@ -4,19 +4,11 @@
 //
 // Sample runs:  ex5 -m ../data/square-disc.mesh
 //               ex5 -m ../data/star.mesh
-//               ex5 -m ../data/star.mesh -pa
 //               ex5 -m ../data/beam-tet.mesh
 //               ex5 -m ../data/beam-hex.mesh
-//               ex5 -m ../data/beam-hex.mesh -pa
 //               ex5 -m ../data/escher.mesh
 //               ex5 -m ../data/fichera.mesh
 //
-// Device sample runs:
-//               ex5 -m ../data/star.mesh -pa -d cuda
-//               ex5 -m ../data/star.mesh -pa -d raja-cuda
-//               ex5 -m ../data/star.mesh -pa -d raja-omp
-//               ex5 -m ../data/beam-hex.mesh -pa -d cuda
-//
 // Description:  This example code solves a simple 2D/3D mixed Darcy problem
 //               corresponding to the saddle point system
 //                                 k*u + grad p = f
@@ -55,8 +47,6 @@ int main(int argc, char *argv[])
   // 1. Parse command-line options.
   const char *mesh_file = "../data/star.mesh";
   int order = 1;
-   bool pa = false;
-   const char *device_config = "cpu";
   bool visualization = 1;

   OptionsParser args(argc, argv);
@@ -64,10 +54,6 @@ int main(int argc, char *argv[])
                  "Mesh file to use.");
   args.AddOption(&order, "-o", "--order",
                  "Finite element order (polynomial degree).");
-   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
-                  "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -79,18 +65,13 @@ int main(int argc, char *argv[])
   }
   args.PrintOptions(cout);

-   // 2. Enable hardware devices such as GPUs, and programming models such as
-   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
-   Device device(device_config);
-   device.Print();
-
-   // 3. Read the mesh from the given mesh file. We can handle triangular,
+   // 2. Read the mesh from the given mesh file. We can handle triangular,
   //    quadrilateral, tetrahedral, hexahedral, surface and volume meshes with
   //    the same code.
   Mesh *mesh = new Mesh(mesh_file, 1, 1);
   int dim = mesh->Dimension();

-   // 4. Refine the mesh to increase the resolution. In this example we do
+   // 3. Refine the mesh to increase the resolution. In this example we do
   //    'ref_levels' of uniform refinement. We choose 'ref_levels' to be the
   //    largest number that gives a final mesh with no more than 10,000
   //    elements.
@@ -103,7 +84,7 @@ int main(int argc, char *argv[])
      }
   }

-   // 5. Define a finite element space on the mesh. Here we use the
+   // 4. Define a finite element space on the mesh. Here we use the
   //    Raviart-Thomas finite elements of the specified order.
   FiniteElementCollection *hdiv_coll(new RT_FECollection(order, dim));
   FiniteElementCollection *l2_coll(new L2_FECollection(order, dim));
@@ -111,7 +92,7 @@ int main(int argc, char *argv[])
   FiniteElementSpace *R_space = new FiniteElementSpace(mesh, hdiv_coll);
   FiniteElementSpace *W_space = new FiniteElementSpace(mesh, l2_coll);

-   // 6. Define the BlockStructure of the problem, i.e. define the array of
+   // 5. Define the BlockStructure of the problem, i.e. define the array of
   //    offsets for each variable. The last component of the Array is the sum
   //    of the dimensions of each block.
   Array<int> block_offsets(3); // number of variables + 1
@@ -126,7 +107,7 @@ int main(int argc, char *argv[])
   std::cout << "dim(R+W) = " << block_offsets.Last() << "\n";
   std::cout << "***********************************************************\n";

-   // 7. Define the coefficients, analytical solution, and rhs of the PDE.
+   // 6. Define the coefficients, analytical solution, and rhs of the PDE.
   ConstantCoefficient k(1.0);

   VectorFunctionCoefficient fcoeff(dim, fFun);
@@ -136,28 +117,25 @@ int main(int argc, char *argv[])
   VectorFunctionCoefficient ucoeff(dim, uFun_ex);
   FunctionCoefficient pcoeff(pFun_ex);

-   // 8. Allocate memory (x, rhs) for the analytical solution and the right hand
+   // 7. Allocate memory (x, rhs) for the analytical solution and the right hand
   //    side.  Define the GridFunction u,p for the finite element solution and
   //    linear forms fform and gform for the right hand side.  The data
   //    allocated by x and rhs are passed as a reference to the grid functions
   //    (u,p) and the linear forms (fform, gform).
-   MemoryType mt = device.GetMemoryType();
-   BlockVector x(block_offsets, mt), rhs(block_offsets, mt);
+   BlockVector x(block_offsets), rhs(block_offsets);

   LinearForm *fform(new LinearForm);
   fform->Update(R_space, rhs.GetBlock(0), 0);
   fform->AddDomainIntegrator(new VectorFEDomainLFIntegrator(fcoeff));
   fform->AddBoundaryIntegrator(new VectorFEBoundaryFluxLFIntegrator(fnatcoeff));
   fform->Assemble();
-   fform->SyncAliasMemory(rhs);

   LinearForm *gform(new LinearForm);
   gform->Update(W_space, rhs.GetBlock(1), 0);
   gform->AddDomainIntegrator(new DomainLFIntegrator(gcoeff));
   gform->Assemble();
-   gform->SyncAliasMemory(rhs);

-   // 9. Assemble the finite element matrices for the Darcy operator
+   // 8. Assemble the finite element matrices for the Darcy operator
   //
   //                            D = [ M  B^T ]
   //                                [ B   0  ]
@@ -168,103 +146,55 @@ int main(int argc, char *argv[])
   BilinearForm *mVarf(new BilinearForm(R_space));
   MixedBilinearForm *bVarf(new MixedBilinearForm(R_space, W_space));

-   if (pa) { mVarf->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   mVarf->AddDomainIntegrator(new VectorFEMassIntegrator(k));
   mVarf->Assemble();
-   if (!pa) { mVarf->Finalize(); }
+   mVarf->Finalize();
+   SparseMatrix &M(mVarf->SpMat());

-   if (pa) { bVarf->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   bVarf->AddDomainIntegrator(new VectorFEDivergenceIntegrator);
   bVarf->Assemble();
-   if (!pa) { bVarf->Finalize(); }
+   bVarf->Finalize();
+   SparseMatrix & B(bVarf->SpMat());
+   B *= -1.;
+   SparseMatrix *BT = Transpose(B);

-   BlockOperator darcyOp(block_offsets);
+   BlockMatrix darcyMatrix(block_offsets);
+   darcyMatrix.SetBlock(0,0, &M);
+   darcyMatrix.SetBlock(0,1, BT);
+   darcyMatrix.SetBlock(1,0, &B);

-   TransposeOperator *Bt = NULL;
-
-   if (pa)
-   {
-      Bt = new TransposeOperator(bVarf);
-
-      darcyOp.SetBlock(0,0, mVarf);
-      darcyOp.SetBlock(0,1, Bt, -1.0);
-      darcyOp.SetBlock(1,0, bVarf, -1.0);
-   }
-   else
-   {
-      SparseMatrix &M(mVarf->SpMat());
-      SparseMatrix &B(bVarf->SpMat());
-      B *= -1.;
-      Bt = new TransposeOperator(&B);
-
-      darcyOp.SetBlock(0,0, &M);
-      darcyOp.SetBlock(0,1, Bt);
-      darcyOp.SetBlock(1,0, &B);
-   }
-
-   // 10. Construct the operators for preconditioner
+   // 9. Construct the operators for preconditioner
   //
   //                 P = [ diag(M)         0         ]
   //                     [  0       B diag(M)^-1 B^T ]
   //
   //     Here we use Symmetric Gauss-Seidel to approximate the inverse of the
   //     pressure Schur Complement
-   SparseMatrix *MinvBt = NULL;
-   Vector Md(mVarf->Height());
+   SparseMatrix *MinvBt = Transpose(B);
+   Vector Md(M.Height());
+   M.GetDiag(Md);
+   for (int i = 0; i < Md.Size(); i++)
+   {
+      MinvBt->ScaleRow(i, 1./Md(i));
+   }
+   SparseMatrix *S = Mult(B, *MinvBt);

-   BlockDiagonalPreconditioner darcyPrec(block_offsets);
   Solver *invM, *invS;
-   SparseMatrix *S = NULL;
-
-   if (pa)
-   {
-      mVarf->AssembleDiagonal(Md);
-      auto Md_host = Md.HostRead();
-      Vector invMd(mVarf->Height());
-      for (int i=0; i<mVarf->Height(); ++i)
-      {
-         invMd(i) = 1.0 / Md_host[i];
-      }
-
-      Vector BMBt_diag(bVarf->Height());
-      bVarf->AssembleDiagonal_ADAt(invMd, BMBt_diag);
-
-      Array<int> ess_tdof_list;  // empty
-
-      invM = new OperatorJacobiSmoother(Md, ess_tdof_list);
-      invS = new OperatorJacobiSmoother(BMBt_diag, ess_tdof_list);
-   }
-   else
-   {
-      SparseMatrix &M(mVarf->SpMat());
-      M.GetDiag(Md);
-
-      SparseMatrix &B(bVarf->SpMat());
-      MinvBt = Transpose(B);
-
-      for (int i = 0; i < Md.Size(); i++)
-      {
-         MinvBt->ScaleRow(i, 1./Md(i));
-      }
-
-      S = Mult(B, *MinvBt);
-
-      invM = new DSmoother(M);
-
+   invM = new DSmoother(M);
 #ifndef MFEM_USE_SUITESPARSE
-      invS = new GSSmoother(*S);
+   invS = new GSSmoother(*S);
 #else
-      invS = new UMFPackSolver(*S);
+   invS = new UMFPackSolver(*S);
 #endif
-   }

   invM->iterative_mode = false;
   invS->iterative_mode = false;

+   BlockDiagonalPreconditioner darcyPrec(block_offsets);
   darcyPrec.SetDiagonalBlock(0, invM);
   darcyPrec.SetDiagonalBlock(1, invS);

-   // 11. Solve the linear system with MINRES.
+   // 10. Solve the linear system with MINRES.
   //     Check the norm of the unpreconditioned residual.
   int maxIter(1000);
   double rtol(1.e-6);
@@ -276,12 +206,11 @@ int main(int argc, char *argv[])
   solver.SetAbsTol(atol);
   solver.SetRelTol(rtol);
   solver.SetMaxIter(maxIter);
-   solver.SetOperator(darcyOp);
+   solver.SetOperator(darcyMatrix);
   solver.SetPreconditioner(darcyPrec);
   solver.SetPrintLevel(1);
   x = 0.0;
   solver.Mult(rhs, x);
-   if (device.IsEnabled()) { x.HostRead(); }
   chrono.Stop();

   if (solver.GetConverged())
@@ -292,7 +221,7 @@ int main(int argc, char *argv[])
                << " iterations. Residual norm is " << solver.GetFinalNorm() << ".\n";
   std::cout << "MINRES solver took " << chrono.RealTime() << "s. \n";

-   // 12. Create the grid functions u and p. Compute the L2 error norms.
+   // 11. Create the grid functions u and p. Compute the L2 error norms.
   GridFunction u, p;
   u.MakeRef(R_space, x.GetBlock(0), 0);
   p.MakeRef(W_space, x.GetBlock(1), 0);
@@ -312,7 +241,7 @@ int main(int argc, char *argv[])
   std::cout << "|| u_h - u_ex || / || u_ex || = " << err_u / norm_u << "\n";
   std::cout << "|| p_h - p_ex || / || p_ex || = " << err_p / norm_p << "\n";

-   // 13. Save the mesh and the solution. This output can be viewed later using
+   // 12. Save the mesh and the solution. This output can be viewed later using
   //     GLVis: "glvis -m ex5.mesh -g sol_u.gf" or "glvis -m ex5.mesh -g
   //     sol_p.gf".
   {
@@ -329,13 +258,13 @@ int main(int argc, char *argv[])
      p.Save(p_ofs);
   }

-   // 14. Save data in the VisIt format
+   // 13. Save data in the VisIt format
   VisItDataCollection visit_dc("Example5", mesh);
   visit_dc.RegisterField("velocity", &u);
   visit_dc.RegisterField("pressure", &p);
   visit_dc.Save();

-   // 15. Save data in the ParaView format
+   // 14. Save data in the ParaView format
   ParaViewDataCollection paraview_dc("Example5", mesh);
   paraview_dc.SetPrefixPath("ParaView");
   paraview_dc.SetLevelsOfDetail(order);
@@ -347,7 +276,7 @@ int main(int argc, char *argv[])
   paraview_dc.RegisterField("pressure",&p);
   paraview_dc.Save();

-   // 16. Send the solution by socket to a GLVis server.
+   // 15. Send the solution by socket to a GLVis server.
   if (visualization)
   {
      char vishost[] = "localhost";
@@ -360,14 +289,14 @@ int main(int argc, char *argv[])
      p_sock << "solution\n" << *mesh << p << "window_title 'Pressure'" << endl;
   }

-   // 17. Free the used memory.
+   // 16. Free the used memory.
   delete fform;
   delete gform;
   delete invM;
   delete invS;
   delete S;
-   delete Bt;
   delete MinvBt;
+   delete BT;
   delete mVarf;
   delete bVarf;
   delete W_space;
@@ -4,19 +4,11 @@
 //
 // Sample runs:  mpirun -np 4 ex5p -m ../data/square-disc.mesh
 //               mpirun -np 4 ex5p -m ../data/star.mesh
-//               mpirun -np 4 ex5p -m ../data/star.mesh -r 2 -pa
 //               mpirun -np 4 ex5p -m ../data/beam-tet.mesh
 //               mpirun -np 4 ex5p -m ../data/beam-hex.mesh
-//               mpirun -np 4 ex5p -m ../data/beam-hex.mesh -pa
 //               mpirun -np 4 ex5p -m ../data/escher.mesh
 //               mpirun -np 4 ex5p -m ../data/fichera.mesh
 //
-// Device sample runs:
-//               mpirun -np 4 ex5p -m ../data/star.mesh -r 2 -pa -d cuda
-//               mpirun -np 4 ex5p -m ../data/star.mesh -r 2 -pa -d raja-cuda
-//               mpirun -np 4 ex5p -m ../data/star.mesh -r 2 -pa -d raja-omp
-//               mpirun -np 4 ex5p -m ../data/beam-hex.mesh -pa -d cuda
-//
 // Description:  This example code solves a simple 2D/3D mixed Darcy problem
 //               corresponding to the saddle point system
 //                                 k*u + grad p = f
@@ -62,28 +54,19 @@ int main(int argc, char *argv[])

   // 2. Parse command-line options.
   const char *mesh_file = "../data/star.mesh";
-   int ref_levels = -1;
   int order = 1;
   bool par_format = false;
-   bool pa = false;
-   const char *device_config = "cpu";
   bool visualization = 1;
   bool adios2 = false;

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
                  "Mesh file to use.");
-   args.AddOption(&ref_levels, "-r", "--refine",
-                  "Number of times to refine the mesh uniformly.");
   args.AddOption(&order, "-o", "--order",
                  "Finite element order (polynomial degree).");
   args.AddOption(&par_format, "-pf", "--parallel-format", "-sf",
                  "--serial-format",
                  "Format to use when saving the results for VisIt.");
-   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
-                  "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&device_config, "-d", "--device",
-                  "Device configuration string, see Device::Configure().");
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
                  "Enable or disable GLVis visualization.");
@@ -105,34 +88,26 @@ int main(int argc, char *argv[])
      args.PrintOptions(cout);
   }

-   // 3. Enable hardware devices such as GPUs, and programming models such as
-   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
-   Device device(device_config);
-   if (myid == 0) { device.Print(); }
-
-   // 4. Read the (serial) mesh from the given mesh file on all processors.  We
+   // 3. Read the (serial) mesh from the given mesh file on all processors.  We
   //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
   //    and volume meshes with the same code.
   Mesh *mesh = new Mesh(mesh_file, 1, 1);
   int dim = mesh->Dimension();

-   // 5. Refine the serial mesh on all processors to increase the resolution. In
+   // 4. Refine the serial mesh on all processors to increase the resolution. In
   //    this example we do 'ref_levels' of uniform refinement. We choose
   //    'ref_levels' to be the largest number that gives a final mesh with no
-   //    more than 10,000 elements, unless the user specifies it as input.
+   //    more than 10,000 elements.
   {
-      if (ref_levels == -1)
-      {
-         ref_levels = (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
-      }
-
+      int ref_levels =
+         (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
      for (int l = 0; l < ref_levels; l++)
      {
         mesh->UniformRefinement();
      }
   }

-   // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
+   // 5. Define a parallel mesh by a partitioning of the serial mesh. Refine
   //    this mesh further in parallel to increase the resolution. Once the
   //    parallel mesh is defined, the serial mesh can be deleted.
   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
@@ -145,7 +120,7 @@ int main(int argc, char *argv[])
      }
   }

-   // 7. Define a parallel finite element space on the parallel mesh. Here we
+   // 6. Define a parallel finite element space on the parallel mesh. Here we
   //    use the Raviart-Thomas finite elements of the specified order.
   FiniteElementCollection *hdiv_coll(new RT_FECollection(order, dim));
   FiniteElementCollection *l2_coll(new L2_FECollection(order, dim));
@@ -165,7 +140,7 @@ int main(int argc, char *argv[])
      std::cout << "***********************************************************\n";
   }

-   // 8. Define the two BlockStructure of the problem.  block_offsets is used
+   // 7. Define the two BlockStructure of the problem.  block_offsets is used
   //    for Vector based on dof (like ParGridFunction or ParLinearForm),
   //    block_trueOffstes is used for Vector based on trueDof (HypreParVector
   //    for the rhs and solution of the linear system).  The offsets computed
@@ -182,7 +157,7 @@ int main(int argc, char *argv[])
   block_trueOffsets[2] = W_space->TrueVSize();
   block_trueOffsets.PartialSum();

-   // 9. Define the coefficients, analytical solution, and rhs of the PDE.
+   // 8. Define the coefficients, analytical solution, and rhs of the PDE.
   ConstantCoefficient k(1.0);

   VectorFunctionCoefficient fcoeff(dim, fFun);
@@ -192,30 +167,25 @@ int main(int argc, char *argv[])
   VectorFunctionCoefficient ucoeff(dim, uFun_ex);
   FunctionCoefficient pcoeff(pFun_ex);

-   // 10. Define the parallel grid function and parallel linear forms, solution
-   //     vector and rhs.
-   MemoryType mt = device.GetMemoryType();
-   BlockVector x(block_offsets, mt), rhs(block_offsets, mt);
-   BlockVector trueX(block_trueOffsets, mt), trueRhs(block_trueOffsets, mt);
+   // 9. Define the parallel grid function and parallel linear forms, solution
+   //    vector and rhs.
+   BlockVector x(block_offsets), rhs(block_offsets);
+   BlockVector trueX(block_trueOffsets), trueRhs(block_trueOffsets);

   ParLinearForm *fform(new ParLinearForm);
   fform->Update(R_space, rhs.GetBlock(0), 0);
   fform->AddDomainIntegrator(new VectorFEDomainLFIntegrator(fcoeff));
   fform->AddBoundaryIntegrator(new VectorFEBoundaryFluxLFIntegrator(fnatcoeff));
   fform->Assemble();
-   fform->SyncAliasMemory(rhs);
   fform->ParallelAssemble(trueRhs.GetBlock(0));
-   trueRhs.GetBlock(0).SyncAliasMemory(trueRhs);

   ParLinearForm *gform(new ParLinearForm);
   gform->Update(W_space, rhs.GetBlock(1), 0);
   gform->AddDomainIntegrator(new DomainLFIntegrator(gcoeff));
   gform->Assemble();
-   gform->SyncAliasMemory(rhs);
   gform->ParallelAssemble(trueRhs.GetBlock(1));
-   trueRhs.GetBlock(1).SyncAliasMemory(trueRhs);

-   // 11. Assemble the finite element matrices for the Darcy operator
+   // 10. Assemble the finite element matrices for the Darcy operator
   //
   //                            D = [ M  B^T ]
   //                                [ B   0  ]
@@ -226,93 +196,44 @@ int main(int argc, char *argv[])
   ParBilinearForm *mVarf(new ParBilinearForm(R_space));
   ParMixedBilinearForm *bVarf(new ParMixedBilinearForm(R_space, W_space));

-   HypreParMatrix *M = NULL;
-   HypreParMatrix *B = NULL;
+   HypreParMatrix *M, *B;

-   if (pa) { mVarf->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   mVarf->AddDomainIntegrator(new VectorFEMassIntegrator(k));
   mVarf->Assemble();
-   if (!pa) { mVarf->Finalize(); }
+   mVarf->Finalize();
+   M = mVarf->ParallelAssemble();

-   if (pa) { bVarf->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
   bVarf->AddDomainIntegrator(new VectorFEDivergenceIntegrator);
   bVarf->Assemble();
-   if (!pa) { bVarf->Finalize(); }
+   bVarf->Finalize();
+   B = bVarf->ParallelAssemble();
+   (*B) *= -1;
+
+   HypreParMatrix *BT = B->Transpose();

   BlockOperator *darcyOp = new BlockOperator(block_trueOffsets);
+   darcyOp->SetBlock(0,0, M);
+   darcyOp->SetBlock(0,1, BT);
+   darcyOp->SetBlock(1,0, B);

-   Array<int> empty_tdof_list;  // empty
-   OperatorPtr opM, opB;
-
-   TransposeOperator *Bt = NULL;
-
-   if (pa)
-   {
-      mVarf->FormSystemMatrix(empty_tdof_list, opM);
-      bVarf->FormRectangularSystemMatrix(empty_tdof_list, empty_tdof_list, opB);
-      Bt = new TransposeOperator(opB.Ptr());
-
-      darcyOp->SetBlock(0,0, opM.Ptr());
-      darcyOp->SetBlock(0,1, Bt, -1.0);
-      darcyOp->SetBlock(1,0, opB.Ptr(), -1.0);
-   }
-   else
-   {
-      M = mVarf->ParallelAssemble();
-      B = bVarf->ParallelAssemble();
-      (*B) *= -1;
-      Bt = new TransposeOperator(B);
-
-      darcyOp->SetBlock(0,0, M);
-      darcyOp->SetBlock(0,1, Bt);
-      darcyOp->SetBlock(1,0, B);
-   }
-
-   // 12. Construct the operators for preconditioner
+   // 11. Construct the operators for preconditioner
   //
   //                 P = [ diag(M)         0         ]
   //                     [  0       B diag(M)^-1 B^T ]
   //
   //     Here we use Symmetric Gauss-Seidel to approximate the inverse of the
   //     pressure Schur Complement.
-   HypreParMatrix *MinvBt = NULL;
-   HypreParVector *Md = NULL;
-   HypreParMatrix *S = NULL;
-   Vector Md_PA;
-   Solver *invM, *invS;
+   HypreParMatrix *MinvBt = B->Transpose();
+   HypreParVector *Md = new HypreParVector(MPI_COMM_WORLD, M->GetGlobalNumRows(),
+                                           M->GetRowStarts());
+   M->GetDiag(*Md);

-   if (pa)
-   {
-      Md_PA.SetSize(R_space->GetTrueVSize());
-      mVarf->AssembleDiagonal(Md_PA);
-      auto Md_host = Md_PA.HostRead();
-      Vector invMd(Md_PA.Size());
-      for (int i=0; i<Md_PA.Size(); ++i)
-      {
-         invMd(i) = 1.0 / Md_host[i];
-      }
+   MinvBt->InvScaleRows(*Md);
+   HypreParMatrix *S = ParMult(B, MinvBt);

-      Vector BMBt_diag(W_space->GetTrueVSize());
-      bVarf->AssembleDiagonal_ADAt(invMd, BMBt_diag);
-
-      Array<int> ess_tdof_list;  // empty
-
-      invM = new OperatorJacobiSmoother(Md_PA, ess_tdof_list);
-      invS = new OperatorJacobiSmoother(BMBt_diag, ess_tdof_list);
-   }
-   else
-   {
-      Md = new HypreParVector(MPI_COMM_WORLD, M->GetGlobalNumRows(),
-                              M->GetRowStarts());
-      M->GetDiag(*Md);
-
-      MinvBt = B->Transpose();
-      MinvBt->InvScaleRows(*Md);
-      S = ParMult(B, MinvBt);
-
-      invM = new HypreDiagScale(*M);
-      invS = new HypreBoomerAMG(*S);
-   }
+   HypreSolver *invM, *invS;
+   invM = new HypreDiagScale(*M);
+   invS = new HypreBoomerAMG(*S);

   invM->iterative_mode = false;
   invS->iterative_mode = false;
@@ -322,9 +243,9 @@ int main(int argc, char *argv[])
   darcyPr->SetDiagonalBlock(0, invM);
   darcyPr->SetDiagonalBlock(1, invS);

-   // 13. Solve the linear system with MINRES.
+   // 12. Solve the linear system with MINRES.
   //     Check the norm of the unpreconditioned residual.
-   int maxIter(pa ? 1000 : 500);
+   int maxIter(500);
   double rtol(1.e-6);
   double atol(1.e-10);

@@ -339,7 +260,6 @@ int main(int argc, char *argv[])
   solver.SetPrintLevel(verbose);
   trueX = 0.0;
   solver.Mult(trueRhs, trueX);
-   if (device.IsEnabled()) { trueX.HostRead(); }
   chrono.Stop();

   if (verbose)
@@ -353,7 +273,7 @@ int main(int argc, char *argv[])
      std::cout << "MINRES solver took " << chrono.RealTime() << "s. \n";
   }

-   // 14. Extract the parallel grid function corresponding to the finite element
+   // 13. Extract the parallel grid function corresponding to the finite element
   //     approximation X. This is the local solution on each processor. Compute
   //     L2 error norms.
   ParGridFunction *u(new ParGridFunction);
@@ -381,7 +301,7 @@ int main(int argc, char *argv[])
      std::cout << "|| p_h - p_ex || / || p_ex || = " << err_p / norm_p << "\n";
   }

-   // 15. Save the refined mesh and the solution in parallel. This output can be
+   // 14. Save the refined mesh and the solution in parallel. This output can be
   //     viewed later using GLVis: "glvis -np <np> -m mesh -g sol_*".
   {
      ostringstream mesh_name, u_name, p_name;
@@ -402,7 +322,7 @@ int main(int argc, char *argv[])
      p->Save(p_ofs);
   }

-   // 16. Save data in the VisIt format
+   // 15. Save data in the VisIt format
   VisItDataCollection visit_dc("Example5-Parallel", pmesh);
   visit_dc.RegisterField("velocity", u);
   visit_dc.RegisterField("pressure", p);
@@ -411,7 +331,7 @@ int main(int argc, char *argv[])
                      DataCollection::PARALLEL_FORMAT);
   visit_dc.Save();

-   // 17. Save data in the ParaView format
+   // 16. Save data in the ParaView format
   ParaViewDataCollection paraview_dc("Example5P", pmesh);
   paraview_dc.SetPrefixPath("ParaView");
   paraview_dc.SetLevelsOfDetail(order);
@@ -423,7 +343,7 @@ int main(int argc, char *argv[])
   paraview_dc.RegisterField("pressure",p);
   paraview_dc.Save();

-   // 18. Optionally output a BP (binary pack) file using ADIOS2. This can be
+   // 17. Optionally output a BP (binary pack) file using ADIOS2. This can be
   //     visualized with the ParaView VTX reader.
 #ifdef MFEM_USE_ADIOS2
   if (adios2)
@@ -443,7 +363,7 @@ int main(int argc, char *argv[])
   }
 #endif

-   // 19. Send the solution by socket to a GLVis server.
+   // 18. Send the solution by socket to a GLVis server.
   if (visualization)
   {
      char vishost[] = "localhost";
@@ -463,7 +383,7 @@ int main(int argc, char *argv[])
             << endl;
   }

-   // 20. Free the used memory.
+   // 19. Free the used memory.
   delete fform;
   delete gform;
   delete u;
@@ -475,7 +395,7 @@ int main(int argc, char *argv[])
   delete S;
   delete Md;
   delete MinvBt;
-   delete Bt;
+   delete BT;
   delete B;
   delete M;
   delete mVarf;
@@ -20,7 +20,7 @@
 //               ex6 -pa -d occa-cuda
 //               ex6 -pa -d raja-omp
 //               ex6 -pa -d ceed-cpu
-//             * ex6 -pa -d ceed-cuda
+//               * ex6 -pa -d ceed-cuda
 //               ex6 -pa -d ceed-cuda:/gpu/cuda/shared
 //
 // Description:  This is a version of Example 1 with a simple adaptive mesh
@@ -20,7 +20,7 @@
 //               mpirun -np 4 ex6p -pa -d occa-cuda
 //               mpirun -np 4 ex6p -pa -d raja-omp
 //               mpirun -np 4 ex6p -pa -d ceed-cpu
-//             * mpirun -np 4 ex6p -pa -d ceed-cuda
+//               * mpirun -np 4 ex6p -pa -d ceed-cuda
 //               mpirun -np 4 ex6p -pa -d ceed-cuda:/gpu/cuda/shared
 //
 // Description:  This is a version of Example 1 with a simple adaptive mesh
@@ -1,348 +0,0 @@
-//                       MFEM Example 71 - Serial Version
-//
-// Compile with: make ex71
-//
-// Sample runs:
-//    ex71 -m ../data/beam-quad.mesh -pp 3.5
-//    ex71 -m ../data/beam-tri.mesh  -pp 4.6
-//    ex71 -m ../data/beam-hex.mesh
-//    ex71 -m ../data/beam-tet.mesh
-//    ex71 -m ../data/beam-wedge.mesh
-//
-// Description:  This examples solves a quasi-static nonlinear
-//               p-Laplacian problem with zero Dirichlet boundary
-//               conditions applied on all defined boundaries
-//
-//           The example demonstrates the use of nonlinear operators
-//           combined with automatic differentiation (AD). The definitions
-//           of the integrators are written in the ex71.hpp.
-//           Selecting integrator=0 will use the handcoded integrator.
-//           Selecting integrator=1 will utilize the AD integrator.
-//           The AD integrator can be modifief to use ADQFunctionTJ.
-//
-//           qint (the integrand) is a function which is evaluated
-//           at every integration point. For implementations utilizing
-//           ADQFunctionTJ, the user has to implement the function and the
-//           residual evaluation. The Jacobian of the residual is evaluated
-//           using AD
-//
-//           For implementations utilizing ADQFunctionTH, the user has
-//           to implement only the function evaluation (as
-//           a template) and the first derivative (the residual) and the
-//           second derivatives (the Hessian) are evaluated using AD.
-//
-//           We recommend viewing examples 1 and 19, before viewing this
-//           example.
-
-#include "ex71.hpp"
-
-#undef MFEM_USE_SUITESPARSE
-
-int main(int argc, char *argv[])
-{
-   // 1. Parse command-line options
-   const char *mesh_file = "../data/beam-tet.mesh";
-   int ser_ref_levels = 3;
-   int order = 1;
-   bool visualization = true;
-   double newton_rel_tol = 1e-4;
-   double newton_abs_tol = 1e-6;
-   int newton_iter = 500;
-   int print_level = 0;
-   double pp = 2.0;
-   int integrator=1; //use AD
-   mfem::StopWatch* timer=new mfem::StopWatch();
-
-   mfem::OptionsParser args(argc, argv);
-   args.AddOption(&mesh_file, "-m", "--mesh",
-                  "Mesh file to use.");
-   args.AddOption(&ser_ref_levels, "-rs", "--refine-serial",
-                  "Number of times to refine the mesh uniformly in serial.");
-   args.AddOption(&order, "-o", "--order",
-                  "Order (degree) of the finite elements.");
-   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
-                  "--no-visualization",
-                  "Enable or disable GLVis visualization.");
-   args.AddOption(&newton_rel_tol, "-rel", "--relative-tolerance",
-                  "Relative tolerance for the Newton solve.");
-   args.AddOption(&newton_abs_tol, "-abs", "--absolute-tolerance",
-                  "Absolute tolerance for the Newton solve.");
-   args.AddOption(&newton_iter, "-it", "--newton-iterations",
-                  "Maximum iterations for the Newton solve.");
-   args.AddOption(&pp, "-pp", "--power-parameter",
-                  "Power parameter (>=2.0) for the p-Laplacian.");
-   args.AddOption((&print_level),"-prt","--print-level",
-                  "Print level.");
-   args.AddOption(&integrator, "-int","--integrator",
-                  "Integrator 0: standard; 1: AD;");
-
-   args.Parse();
-   if (!args.Good())
-   {
-      args.PrintUsage(std::cout);
-      return 1;
-   }
-   args.PrintOptions(std::cout);
-
-
-   // 2. Read the (serial) mesh from the given mesh file.
-   mfem::Mesh *mesh = new mfem::Mesh(mesh_file, 1, 1);
-   int dim = mesh->Dimension();
-
-   // 3. Refine the mesh in serial to increase the resolution. In this example
-   //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
-   //    a command-line parameter.
-   for (int lev = 0; lev < ser_ref_levels; lev++)
-   {
-      mesh->UniformRefinement();
-   }
-
-   // 4. Define the power parameter for the p-Laplacian and all other
-   //    coefficients
-   mfem::ConstantCoefficient c_pp(pp);
-   mfem::ConstantCoefficient load(1.000000000);
-   mfem::ConstantCoefficient c_ee(0.000000001);
-
-   // 5. Define the finite element spaces for the solution
-   mfem::H1_FECollection fec(order,dim);
-   mfem::FiniteElementSpace fespace(mesh,&fec,1,mfem::Ordering::byVDIM);
-   int glob_size=fespace.GetTrueVSize();
-   std::cout << "Number of finite element unknowns: " << glob_size << std::endl;
-
-   // 6. Define the Dirichlet conditions
-   mfem::Array<int> ess_bdr(mesh->bdr_attributes.Max());
-   ess_bdr = 1;
-
-   // 7. Define the nonlinear form
-   mfem::NonlinearForm* nf=new mfem::NonlinearForm(&fespace);
-
-   // 8. Define the solution vector x
-   mfem::GridFunction x(&fespace);
-   x = 0.0;
-   mfem::Vector tv(fespace.GetTrueVSize());
-   mfem::Vector sv(fespace.GetTrueVSize());
-   tv=0.0;
-   sv=0.0;
-
-   // 9. Define ParaView DataCollection
-   mfem::ParaViewDataCollection *dacol=new
-   mfem::ParaViewDataCollection("Example71",
-                                mesh);
-   dacol->SetLevelsOfDetail(order);
-   dacol->RegisterField("sol",&x);
-
-   // 11. Set domain integrators - start with linear diffusion
-   {
-      // the default power coefficient is 2.0
-      mfem::ConstantCoefficient lpp(2.0);
-      if (integrator==0)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplace(lpp,c_ee,load));
-      }
-      else if (integrator==1)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplaceAD(lpp,c_ee,load));
-      }
-      nf->SetEssentialBC(ess_bdr);
-      // compute the energy
-      double energy=nf->GetEnergy(tv);
-      std::cout<<"[2] The total energy of the system is E="<<energy<<std::endl;
-      // time the assembly
-      timer->Clear();
-      timer->Start();
-      nf->GetGradient(sv);
-      timer->Stop();
-      std::cout<<"[2] The assembly time is: "<<timer->RealTime()<<std::endl;
-      mfem::Solver *prec;
-#ifdef MFEM_USE_SUITESPARSE
-      prec=new mfem::UMFPackSolver();
-#else
-      prec=new mfem::GSSmoother();
-#endif
-      mfem::CGSolver *j_pcg = new mfem::CGSolver();
-      j_pcg->SetRelTol(1e-7);
-      j_pcg->SetAbsTol(1e-15);
-      j_pcg->SetMaxIter(500);
-      j_pcg->SetPrintLevel(print_level);
-      j_pcg->SetPreconditioner(*prec);
-
-      mfem::NewtonSolver* ns;
-      ns=new mfem::NewtonSolver();
-      ns->iterative_mode = true;
-      ns->SetSolver(*j_pcg);
-      ns->SetOperator(*nf);
-      ns->SetPrintLevel(print_level);
-      ns->SetRelTol(1e-6);
-      ns->SetAbsTol(1e-12);
-      ns->SetMaxIter(10);
-      //solve the problem
-      timer->Clear();
-      timer->Start();
-      ns->Mult(tv, sv);
-      timer->Stop();
-      std::cout<<"Time for the NewtonSolver: "<<timer->RealTime()<<std::endl;
-
-      energy=nf->GetEnergy(sv);
-      std::cout<<"[pp=2] The total energy of the system is E="<<energy<<std::endl;
-
-      delete ns;
-      delete j_pcg;
-      delete prec;
-
-      x.SetFromTrueDofs(sv);
-      dacol->SetTime(2.0);
-      dacol->SetCycle(2);
-      dacol->Save();
-   }
-
-   // 12. Continue with powers higher than 2
-   for (int i=3; i<pp; i++)
-   {
-      delete nf;
-      nf=new mfem::NonlinearForm(&fespace);
-      mfem::ConstantCoefficient lpp((double)i);
-      if (integrator==0)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplace(lpp,c_ee,load));
-      }
-      else if (integrator==1)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplaceAD(lpp,c_ee,load));
-      }
-      nf->SetEssentialBC(ess_bdr);
-      // compute the energy
-      double energy=nf->GetEnergy(sv);
-      std::cout<<"[pp="<<i<<"] The total energy of the system is E="<<energy<<std::endl;
-      // time the assembly
-      timer->Clear();
-      timer->Start();
-      nf->GetGradient(sv);
-      timer->Stop();
-      std::cout<<"[pp="<<i<<"] The assembly time is: "<<timer->RealTime()<<std::endl;
-      mfem::Solver *prec;
-#ifdef MFEM_USE_SUITESPARSE
-      prec=new mfem::UMFPackSolver();
-#else
-      prec=new mfem::GSSmoother();
-#endif
-      mfem::CGSolver *j_pcg = new mfem::CGSolver();
-      j_pcg->SetRelTol(1e-7);
-      j_pcg->SetAbsTol(1e-15);
-      j_pcg->SetMaxIter(500);
-      j_pcg->SetPrintLevel(print_level);
-      j_pcg->SetPreconditioner(*prec);
-
-      mfem::NewtonSolver* ns;
-      ns=new mfem::NewtonSolver();
-      ns->iterative_mode = true;
-      ns->SetSolver(*j_pcg);
-      ns->SetOperator(*nf);
-      ns->SetPrintLevel(print_level);
-      ns->SetRelTol(1e-6);
-      ns->SetAbsTol(1e-12);
-      ns->SetMaxIter(10);
-      //solve the problem
-      timer->Clear();
-      timer->Start();
-      ns->Mult(tv, sv);
-      timer->Stop();
-      std::cout<<"Time for the NewtonSolver: "<<timer->RealTime()<<std::endl;
-
-      energy=nf->GetEnergy(sv);
-      std::cout<<"[pp="<<i<<"] The total energy of the system is E="<<energy<<std::endl;
-
-      delete ns;
-      delete j_pcg;
-      delete prec;
-
-      x.SetFromTrueDofs(sv);
-      dacol->SetTime(i);
-      dacol->SetCycle(i);
-      dacol->Save();
-   }
-
-   // 13. Continue with the final power
-   if ( std::abs(pp-2.0) > std::numeric_limits<double>::epsilon())
-   {
-      delete nf;
-      nf=new mfem::NonlinearForm(&fespace);
-      if (integrator==0)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplace(c_pp,c_ee,load));
-      }
-      else if (integrator==1)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplaceAD(c_pp,c_ee,load));
-      }
-      nf->SetEssentialBC(ess_bdr);
-      // compute the energy
-      double energy=nf->GetEnergy(sv);
-      std::cout<<"[pp="<<pp<<"] The total energy of the system is E="<<energy<<std::endl;
-      // time the assembly
-      timer->Clear();
-      timer->Start();
-      nf->GetGradient(sv);
-      timer->Stop();
-      std::cout<<"[pp="<<pp<<"] The assembly time is: "<<timer->RealTime()<<std::endl;
-      mfem::Solver *prec;
-#ifdef MFEM_USE_SUITESPARSE
-      prec=new mfem::UMFPackSolver();
-#else
-      prec=new mfem::GSSmoother();
-#endif
-      mfem::CGSolver *j_pcg = new mfem::CGSolver();
-      j_pcg->SetRelTol(1e-7);
-      j_pcg->SetAbsTol(1e-15);
-      j_pcg->SetMaxIter(500);
-      j_pcg->SetPrintLevel(print_level);
-      j_pcg->SetPreconditioner(*prec);
-
-      mfem::NewtonSolver* ns;
-      ns=new mfem::NewtonSolver();
-      ns->iterative_mode = true;
-      ns->SetSolver(*j_pcg);
-      ns->SetOperator(*nf);
-      ns->SetPrintLevel(print_level);
-      ns->SetRelTol(1e-6);
-      ns->SetAbsTol(1e-12);
-      ns->SetMaxIter(10);
-      //solve the problem
-      timer->Clear();
-      timer->Start();
-      ns->Mult(tv, sv);
-      timer->Stop();
-      std::cout<<"Time for the NewtonSolver: "<<timer->RealTime()<<std::endl;
-
-      energy=nf->GetEnergy(sv);
-      std::cout<<"[pp="<<pp<<"] The total energy of the system is E="<<energy<<std::endl;
-
-      delete ns;
-      delete j_pcg;
-      delete prec;
-
-      x.SetFromTrueDofs(sv);
-      dacol->SetTime(pp);
-      if (pp<2.0)
-      {
-         dacol->SetCycle(std::floor(pp));
-      }
-      else
-      {
-         dacol->SetCycle(std::ceil(pp));
-      }
-      dacol->Save();
-   }
-
-
-
-   // 19. Free the used memory
-   delete dacol;
-   delete nf;
-   delete mesh;
-   delete timer;
-
-
-   return 0;
-}
-
-
@@ -1,587 +0,0 @@
-// shared implementation ex71p/ex71 for the AD integrands and
-// the handconded integrators
-
-
-#ifndef EXAMPLE71_H
-#define EXAMPLE71_H
-
-#include "mfem.hpp"
-#include <memory>
-#include <iostream>
-#include <fstream>
-
-
-namespace mfem
-{
-
-
-template<typename DType, typename MVType>
-class MyQFunctorJ
-{
-public:
-   DType operator()(const mfem::Vector& vparam, MVType& uu)
-   {
-      double pp=vparam[0];
-      double ee=vparam[1];
-      double ff=vparam[2];
-
-      DType  u=uu[3];
-      DType  norm2=uu[0]*uu[0]+uu[1]*uu[1]+uu[2]*uu[2];
-
-      DType rez= pow(ee*ee+norm2,pp/2.0)/pp-ff*u;
-      return rez;
-   }
-
-   void operator()(const mfem::Vector& vparam, MVType& uu, MVType& rr)
-   {
-      double pp=vparam[0];
-      double ee=vparam[1];
-      double ff=vparam[2];
-
-      DType norm2=uu[0]*uu[0]+uu[1]*uu[1]+uu[2]*uu[2];
-      DType tvar=pow(ee*ee+norm2,(pp-2.0)/2.0);
-
-      rr[0]=tvar*uu[0];
-      rr[1]=tvar*uu[1];
-      rr[2]=tvar*uu[2];
-      rr[3]=-ff;
-
-   }
-};
-
-
-typedef ADQFunctionTJ<MyQFunctorJ,4> pLapIntegrandTJ;
-
-
-template<typename DType, typename MVType>
-class MyQFunctorH
-{
-public:
-   DType operator()(const mfem::Vector& vparam, MVType& uu)
-   {
-      double pp=vparam[0];
-      double ee=vparam[1];
-      double ff=vparam[2];
-
-      DType  u=uu[3];
-      DType  norm2=uu[0]*uu[0]+uu[1]*uu[1]+uu[2]*uu[2];
-
-      DType rez= pow(ee*ee+norm2,pp/2.0)/pp-ff*u;
-      return rez;
-   }
-};
-
-typedef ADQFunctionTH<MyQFunctorH> pLapIntegrandTH;
-
-//comment the line below in order to use
-//pLapIntegrandTJ for differentiation
-//the user interface for both TH and TJ versions
-//is exacly the same
-//#define USE_ADH
-
-class pLaplaceAD: public mfem::NonlinearFormIntegrator
-{
-protected:
-   mfem::Coefficient* pp;
-   mfem::Coefficient* coeff;
-   mfem::Coefficient* load;
-#ifdef USE_ADH
-   pLapIntegrandTH qint;
-#else
-   pLapIntegrandTJ qint;
-#endif
-public:
-   pLaplaceAD()
-   {
-      coeff=nullptr;
-      pp=nullptr;
-   }
-
-   pLaplaceAD(mfem::Coefficient& pp_):pp(&pp_), coeff(nullptr), load(nullptr)
-   {
-
-   }
-
-   pLaplaceAD(mfem::Coefficient &pp_,mfem::Coefficient& q,
-              mfem::Coefficient& ld_): pp(&pp_), coeff(&q), load(&ld_)
-   {
-
-   }
-
-   virtual ~pLaplaceAD()
-   {
-
-   }
-
-   virtual double GetElementEnergy(const mfem::FiniteElement &el,
-                                   mfem::ElementTransformation &trans, const mfem::Vector &elfun) override
-   {
-      double energy=0.0;
-      int ndof = el.GetDof();
-      int ndim = el.GetDim();
-      int spaceDim = trans.GetSpaceDim();
-      bool square = (ndim == spaceDim);
-      const mfem::IntegrationRule *ir = NULL;
-      int order = 2 * trans.OrderGrad(&el) - 1; // correct order?
-      ir = &mfem::IntRules.Get(el.GetGeomType(), order);
-
-      mfem::Vector shapef(ndof);
-      mfem::DenseMatrix dshape_iso(ndof,ndim);
-      mfem::DenseMatrix dshape_xyz(ndof,spaceDim);
-      mfem::Vector grad(spaceDim);
-
-      mfem::Vector vparam(3);//[power, epsilon, load]
-      mfem::Vector uu(4);//[diff_x,diff_y,diff_z,u]
-
-      uu=0.0;
-      vparam[0]=2.0;  //default power
-      vparam[1]=1e-8; //default epsilon
-      vparam[2]=1.0;  //default load
-
-      double w;
-      double detJ;
-
-      for (int i = 0; i < ir -> GetNPoints(); i++)
-      {
-         const mfem::IntegrationPoint &ip = ir->IntPoint(i);
-         trans.SetIntPoint(&ip);
-         w = trans.Weight();
-         detJ = (square ? w : w*w);
-         w = ip.weight *w;
-
-         el.CalcDShape(ip,dshape_iso);
-         el.CalcShape(ip,shapef);
-         // AdjugateJacobian = / adj(J),         if J is square
-         //                    \ adj(J^t.J).J^t, otherwise
-         mfem::Mult(dshape_iso, trans.AdjugateJacobian(), dshape_xyz);
-         // dshape_xyz should be devided by detJ for obtaining the real value
-         // calculate the gradient
-         dshape_xyz.MultTranspose(elfun,grad);
-
-         //set the power
-         if (pp!=nullptr)
-         {
-            vparam[0]=pp->Eval(trans,ip);
-         }
-
-         //set the coefficient ensuring possitiveness of the tangent matrix
-         if (coeff!=nullptr)
-         {
-            vparam[1]=coeff->Eval(trans,ip);
-         }
-         //add the contribution from the load
-         if (load!=nullptr)
-         {
-            vparam[2]=load->Eval(trans,ip);
-         }
-         //fill the values of vector uu
-         for (int jj=0; jj<spaceDim; jj++)
-         {
-            uu[jj]=grad[jj]/detJ;
-         }
-         uu[3]=shapef*elfun;
-
-         energy = energy + w * (qint.QFunction(vparam,uu));
-
-      }
-      return energy;
-   }
-
-   virtual void AssembleElementVector(const mfem::FiniteElement & el,
-                                      mfem::ElementTransformation & trans,
-                                      const mfem::Vector & elfun,
-                                      mfem::Vector & elvect) override
-   {
-      int ndof = el.GetDof();
-      int ndim = el.GetDim();
-      int spaceDim = trans.GetSpaceDim();
-      const mfem::IntegrationRule *ir = NULL;
-      int order = 2 * trans.OrderGrad(&el) - 1; // correct order?
-      ir = &mfem::IntRules.Get(el.GetGeomType(), order);
-
-      mfem::Vector shapef(ndof);
-      mfem::DenseMatrix dshape_iso(ndof,ndim);
-      mfem::DenseMatrix dshape_xyz(ndof,spaceDim);
-      mfem::Vector lvec(ndof);
-      elvect.SetSize(ndof);
-      elvect=0.0;
-
-      mfem::DenseMatrix B(ndof,4); //[diff_x,diff_y,diff_z, shape]
-      mfem::Vector vparam(3);//[power, epsilon, load]
-      mfem::Vector uu(4);//[diff_x,diff_y,diff_z,u]
-      mfem::Vector du(4);
-      B=0.0;
-      uu=0.0;
-      //initialize the parameters - keep the same order
-      //utilized in the pLapIntegrator definition
-      vparam[0]=2.0;  //default power
-      vparam[1]=1e-8; //default epsilon
-      vparam[2]=1.0;  //default load
-
-      double w;
-
-      for (int i = 0; i < ir -> GetNPoints(); i++)
-      {
-         const mfem::IntegrationPoint &ip = ir->IntPoint(i);
-         trans.SetIntPoint(&ip);
-         w = trans.Weight();
-         //detJ = (square ? w : w*w);
-         w = ip.weight * w;
-
-         el.CalcDShape(ip,dshape_iso);
-         el.CalcShape(ip,shapef);
-         mfem::Mult(dshape_iso, trans.InverseJacobian(), dshape_xyz);
-
-         //set the matrix B
-         for (int jj=0; jj<spaceDim; jj++)
-         {
-            B.SetCol(jj,dshape_xyz.GetColumn(jj));
-         }
-         B.SetCol(3,shapef);
-
-
-         //set the power
-         if (pp!=nullptr)
-         {
-            vparam[0]=pp->Eval(trans,ip);
-         }
-         //set the coefficient ensuring possitiveness of the tangent matrix
-         if (coeff!=nullptr)
-         {
-            vparam[1]=coeff->Eval(trans,ip);
-         }
-         //add the contribution from the load
-         if (load!=nullptr)
-         {
-            vparam[2]=load->Eval(trans,ip);
-         }
-
-         //calculate uu
-         B.MultTranspose(elfun,uu);
-         //calculate derivative of the energy with respect to uu
-         qint.QFunctionDU(vparam,uu,du);
-
-         B.Mult(du,lvec);
-         elvect.Add( w, lvec);
-      }// end integration loop
-   }
-
-   virtual void AssembleElementGrad(const mfem::FiniteElement & el,
-                                    mfem::ElementTransformation & trans,
-                                    const mfem::Vector & elfun, mfem::DenseMatrix & elmat) override
-   {
-      int ndof = el.GetDof();
-      int ndim = el.GetDim();
-      int spaceDim = trans.GetSpaceDim();
-      const mfem::IntegrationRule *ir = NULL;
-      int order = 2 * trans.OrderGrad(&el) - 1; // correct order?
-      ir = &mfem::IntRules.Get(el.GetGeomType(), order);
-
-      mfem::Vector shapef(ndof);
-      mfem::DenseMatrix dshape_iso(ndof,ndim);
-      mfem::DenseMatrix dshape_xyz(ndof,spaceDim);
-      elmat.SetSize(ndof,ndof);
-      elmat=0.0;
-
-      mfem::DenseMatrix B(ndof,4); //[diff_x,diff_y,diff_z, shape]
-      mfem::DenseMatrix A(ndof,4);
-      mfem::Vector vparam(3);//[power, epsilon, load]
-      mfem::Vector uu(4);//[diff_x,diff_y,diff_z,u]
-      mfem::DenseMatrix duu(4,4);
-      B=0.0;
-      uu=0.0;
-      //initialize the parameters - keep the same order
-      //utilized in the pLapIntegrator definition
-      vparam[0]=2.0;  //default power
-      vparam[1]=1e-8; //default epsilon
-      vparam[2]=1.0;  //default load
-
-      double w;
-
-      for (int i = 0; i < ir -> GetNPoints(); i++)
-      {
-         const mfem::IntegrationPoint &ip = ir->IntPoint(i);
-         trans.SetIntPoint(&ip);
-         w = trans.Weight();
-         w = ip.weight * w;
-
-         el.CalcDShape(ip,dshape_iso);
-         el.CalcShape(ip,shapef);
-         mfem::Mult(dshape_iso, trans.InverseJacobian(), dshape_xyz);
-
-         //set the matrix B
-         for (int jj=0; jj<spaceDim; jj++)
-         {
-            B.SetCol(jj,dshape_xyz.GetColumn(jj));
-         }
-         B.SetCol(3,shapef);
-
-
-         //set the power
-         if (pp!=nullptr)
-         {
-            vparam[0]=pp->Eval(trans,ip);
-         }
-         //set the coefficient ensuring possitiveness of the tangent matrix
-         if (coeff!=nullptr)
-         {
-            vparam[1]=coeff->Eval(trans,ip);
-         }
-         //add the contribution from the load
-         if (load!=nullptr)
-         {
-            vparam[2]=load->Eval(trans,ip);
-         }
-
-         //calculate uu
-         B.MultTranspose(elfun,uu);
-         //calculate derivative of the energy with respect to uu
-         qint.QFunctionDD(vparam,uu,duu);
-
-         mfem::Mult(B,duu,A);
-         mfem::AddMult_a_ABt(w,A,B,elmat);
-
-      }//end integration loop
-   }
-
-
-};
-
-
-class pLaplace: public mfem::NonlinearFormIntegrator
-{
-protected:
-   mfem::Coefficient* pp;
-   mfem::Coefficient* coeff;
-   mfem::Coefficient* load;
-public:
-   pLaplace()
-   {
-      coeff=nullptr;
-      pp=nullptr;
-   }
-
-   pLaplace(mfem::Coefficient& pp_):pp(&pp_), coeff(nullptr), load(nullptr)
-   {
-
-   }
-
-   pLaplace(mfem::Coefficient &pp_,mfem::Coefficient& q,
-            mfem::Coefficient& ld_): pp(&pp_), coeff(&q), load(&ld_)
-   {
-
-   }
-
-   virtual ~pLaplace()
-   {
-
-   }
-
-   virtual double GetElementEnergy(const mfem::FiniteElement &el,
-                                   mfem::ElementTransformation &trans, const mfem::Vector &elfun) override
-   {
-      double energy=0.0;
-      int ndof = el.GetDof();
-      int ndim = el.GetDim();
-      int spaceDim = trans.GetSpaceDim();
-      bool square = (ndim == spaceDim);
-      const mfem::IntegrationRule *ir = NULL;
-      int order = 2 * trans.OrderGrad(&el) - 1; // correct order?
-      ir = &mfem::IntRules.Get(el.GetGeomType(), order);
-
-      mfem::Vector shapef(ndof);
-      mfem::DenseMatrix dshape_iso(ndof,ndim);
-      mfem::DenseMatrix dshape_xyz(ndof,spaceDim);
-      mfem::Vector grad(spaceDim);
-
-      double w;
-      double detJ;
-      double nrgrad2;
-      double ppp=2.0;
-      double eee=0.0;
-
-      for (int i = 0; i < ir -> GetNPoints(); i++)
-      {
-         const mfem::IntegrationPoint &ip = ir->IntPoint(i);
-         trans.SetIntPoint(&ip);
-         w = trans.Weight();
-         detJ = (square ? w : w*w);
-         w = ip.weight *w;
-
-         el.CalcDShape(ip,dshape_iso);
-         el.CalcShape(ip,shapef);
-         // AdjugateJacobian = / adj(J),         if J is square
-         //                    \ adj(J^t.J).J^t, otherwise
-         mfem::Mult(dshape_iso, trans.AdjugateJacobian(), dshape_xyz);
-         // dshape_xyz should be devided by detJ for obtaining the real value
-         // calculate the gradient
-         dshape_xyz.MultTranspose(elfun,grad);
-         nrgrad2=grad*grad/(detJ*detJ);
-
-         //set the power
-         if (pp!=nullptr)
-         {
-            ppp=pp->Eval(trans,ip);
-         }
-
-         //set the coefficient ensuring possitiveness of the tangent matrix
-         if (coeff!=nullptr)
-         {
-            eee=coeff->Eval(trans,ip);
-         }
-
-         energy = energy + w * std::pow( nrgrad2 + eee * eee , ppp / 2.0 ) / ppp;
-
-         //add the contribution from the load
-         if (load!=nullptr)
-         {
-            energy = energy - w *  (shapef*elfun) * load->Eval(trans,ip);
-         }
-      }
-      return energy;
-   }
-
-   virtual void  AssembleElementVector(const mfem::FiniteElement & el,
-                                       mfem::ElementTransformation & trans,
-                                       const mfem::Vector & elfun,
-                                       mfem::Vector & elvect) override
-   {
-      int ndof = el.GetDof();
-      int ndim = el.GetDim();
-      int spaceDim = trans.GetSpaceDim();
-      bool square = (ndim == spaceDim);
-      const mfem::IntegrationRule *ir = NULL;
-      int order = 2 * trans.OrderGrad(&el) - 1; // correct order?
-      ir = &mfem::IntRules.Get(el.GetGeomType(), order);
-
-      mfem::Vector shapef(ndof);
-      mfem::DenseMatrix dshape_iso(ndof,ndim);
-      mfem::DenseMatrix dshape_xyz(ndof,spaceDim);
-      mfem::Vector grad(spaceDim);
-      mfem::Vector lvec(ndof);
-      elvect.SetSize(ndof);
-      elvect=0.0;
-
-      double w;
-      double detJ;
-      double nrgrad;
-      double aa;
-      double ppp=2.0;
-      double eee=0.0;
-
-      for (int i = 0; i < ir -> GetNPoints(); i++)
-      {
-         const mfem::IntegrationPoint &ip = ir->IntPoint(i);
-         trans.SetIntPoint(&ip);
-         w = trans.Weight();
-         detJ = (square ? w : w*w);
-         w = ip.weight * w;//w;
-
-         el.CalcDShape(ip,dshape_iso);
-         el.CalcShape(ip,shapef);
-         // AdjugateJacobian = / adj(J),         if J is square
-         //                    \ adj(J^t.J).J^t, otherwise
-         mfem::Mult(dshape_iso, trans.AdjugateJacobian(), dshape_xyz);
-         // dshape_xyz should be devided by detJ for obtaining the real value
-
-         //calculate the gradient
-         dshape_xyz.MultTranspose(elfun,grad);
-         nrgrad=grad.Norml2()/detJ;
-         //grad is not scaled so far, i.e., grad=grad/detJ
-
-         //set the power
-         if (pp!=nullptr)
-         {
-            ppp=pp->Eval(trans,ip);
-         }
-
-         //set the coefficient ensuring possitiveness of the tangent matrix
-         if (coeff!=nullptr)
-         {
-            eee=coeff->Eval(trans,ip);
-         }
-
-         aa = nrgrad * nrgrad + eee * eee;
-         aa=std::pow( aa , ( ppp - 2.0 ) / 2.0 );
-         dshape_xyz.Mult(grad,lvec);
-         elvect.Add( w * aa / ( detJ * detJ ), lvec);
-
-
-         //add loading
-         if (load!=nullptr)
-         {
-            elvect.Add(-w*load->Eval(trans,ip),shapef);
-         }
-      }// end integration loop
-   }
-
-   virtual void AssembleElementGrad(const mfem::FiniteElement & el,
-                                    mfem::ElementTransformation & trans,
-                                    const mfem::Vector & elfun, mfem::DenseMatrix & elmat) override
-   {
-      int ndof = el.GetDof();
-      int ndim = el.GetDim();
-      int spaceDim = trans.GetSpaceDim();
-      bool square = (ndim == spaceDim);
-      const mfem::IntegrationRule *ir = NULL;
-      int order = 2 * trans.OrderGrad(&el) - 1; // correct order?
-      ir = &mfem::IntRules.Get(el.GetGeomType(), order);
-
-      mfem::DenseMatrix dshape_iso(ndof,ndim);
-      mfem::DenseMatrix dshape_xyz(ndof,spaceDim);
-      mfem::Vector grad(spaceDim);
-      mfem::Vector lvec(ndof);
-      elmat.SetSize(ndof,ndof);
-      elmat=0.0;
-
-      double w;
-      double detJ;
-      double nrgrad;
-      double aa0;
-      double aa1;
-      double ppp=2.0;
-      double eee=0.0;
-
-      for (int i = 0; i < ir -> GetNPoints(); i++)
-      {
-         const mfem::IntegrationPoint &ip = ir->IntPoint(i);
-         trans.SetIntPoint(&ip);
-         w = trans.Weight();
-         detJ = (square ? w : w*w);
-         w = ip.weight * w;
-
-         el.CalcDShape(ip,dshape_iso);
-         // AdjugateJacobian = / adj(J),         if J is square
-         //                    \ adj(J^t.J).J^t, otherwise
-         mfem::Mult(dshape_iso, trans.AdjugateJacobian(), dshape_xyz);
-         // dshape_xyz should be devided by detJ for obtaining the real value
-         // grad is not scaled so far,i.e., grad=grad/detJ
-
-         //set the power
-         if (pp!=nullptr)
-         {
-            ppp=pp->Eval(trans,ip);
-         }
-         //set the coefficient ensuring possitiveness of the tangent matrix
-         if (coeff!=nullptr)
-         {
-            eee=coeff->Eval(trans,ip);
-         }
-
-         //calculate the gradient
-         dshape_xyz.MultTranspose(elfun,grad);
-         nrgrad = grad.Norml2() / detJ;
-         aa0 = nrgrad * nrgrad + eee * eee;
-         aa1 = std::pow( aa0 , ( ppp - 2.0 ) / 2.0 );
-         aa0 = ( ppp - 2.0 ) * std::pow(aa0, ( ppp - 4.0 ) / 2.0 );
-         dshape_xyz.Mult(grad,lvec);
-         w = w / ( detJ * detJ );
-         mfem::AddMult_a_VVt( w * aa0 / ( detJ * detJ ), lvec, elmat);
-         mfem::AddMult_a_AAt( w * aa1 , dshape_xyz, elmat);
-
-      }//end integration loop
-   }
-};
-
-}
-#endif
@@ -1,401 +0,0 @@
-//                       MFEM Example 71 - Parallel Version
-//
-// Compile with: make ex71p
-//
-// Sample runs:
-//   mpirun -np 2 ex71p -m ../data/beam-quad.mesh -pp 3.8
-//   mpirun -np 2 ex71p -m ../data/beam-tri.mesh  -pp 7.2
-//   mpirun -np 2 ex71p -m ../data/beam-hex.mesh
-//   mpirun -np 2 ex71p -m ../data/beam-tet.mesh
-//   mpirun -np 2 ex71p -m ../data/beam-wedge.mesh
-//
-// Description:  This examples solves a quasi-static nonlinear
-//               p-Laplacian problem with zero Dirichlet boundary
-//               conditions applied on all defined boundaries
-//
-//           The example demonstrates the use of nonlinear operators
-//           combined with automatic differentiation (AD). The definitions
-//           of the integrators are written in the ex71.hpp.
-//           Selecting integrator=0 will use the handcoded integrator.
-//           Selecting integrator=1 will utilize the AD integrator.
-//           The AD integrator can be modifief to use ADQFunctionTJ.
-//
-//           qint (the integrand) is a function which is evaluated
-//           at every integration point. For implementations utilizing
-//           ADQFunctionTJ, the user has to implement the function and the
-//           residual evaluation. The Jacobian of the residual is evaluated
-//           using AD
-//
-//           For implementations utilizing ADQFunctionTH, the user has
-//           to implement only the function evaluation (as
-//           a template) and the first derivative (the residual) and the
-//           second derivatives (the Hessian) are evaluated using AD.
-//
-//           We recommend viewing examples 1 and 19, before viewing this
-//           example.
-
-#include "ex71.hpp"
-
-int main(int argc, char *argv[])
-{
-   // 1. Initialize MPI
-   int num_procs, myrank;
-   MPI_Init(&argc, &argv);
-   MPI_Comm_size(MPI_COMM_WORLD, &num_procs);
-   MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
-
-   // 2. Parse command-line options
-   const char *mesh_file = "../data/beam-tet.mesh";
-   int ser_ref_levels = 0;
-   int par_ref_levels = 0;
-   int order = 2;
-   bool visualization = true;
-   double newton_rel_tol = 1e-4;
-   double newton_abs_tol = 1e-6;
-   int newton_iter = 500;
-   int print_level = 0;
-   double pp = 2.0;
-   int integrator=1; //use AD
-   mfem::StopWatch* timer=new mfem::StopWatch();
-
-   mfem::OptionsParser args(argc, argv);
-   args.AddOption(&mesh_file, "-m", "--mesh",
-                  "Mesh file to use.");
-   args.AddOption(&ser_ref_levels, "-rs", "--refine-serial",
-                  "Number of times to refine the mesh uniformly in serial.");
-   args.AddOption(&par_ref_levels, "-rp", "--refine-parallel",
-                  "Number of times to refine the mesh uniformly in parallel.");
-   args.AddOption(&order, "-o", "--order",
-                  "Order (degree) of the finite elements.");
-   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
-                  "--no-visualization",
-                  "Enable or disable GLVis visualization.");
-   args.AddOption(&newton_rel_tol, "-rel", "--relative-tolerance",
-                  "Relative tolerance for the Newton solve.");
-   args.AddOption(&newton_abs_tol, "-abs", "--absolute-tolerance",
-                  "Absolute tolerance for the Newton solve.");
-   args.AddOption(&newton_iter, "-it", "--newton-iterations",
-                  "Maximum iterations for the Newton solve.");
-   args.AddOption(&pp, "-pp", "--power-parameter",
-                  "Power parameter (>=2.0) for the p-Laplacian.");
-   args.AddOption((&print_level),"-prt","--print-level",
-                  "Print level.");
-   args.AddOption(&integrator, "-int","--integrator",
-                  "Integrator 0: standard; 1: AD");
-
-   args.Parse();
-   if (!args.Good())
-   {
-      if (myrank == 0)
-      {
-         args.PrintUsage(std::cout);
-      }
-      MPI_Finalize();
-      return 1;
-   }
-   if (myrank == 0)
-   {
-      args.PrintOptions(std::cout);
-   }
-
-   // 3. Read the (serial) mesh from the given mesh file on all processors.  We
-   //    can handle triangular, quadrilateral, tetrahedral and hexahedral meshes
-   //    with the same code.
-   mfem::Mesh *mesh = new mfem::Mesh(mesh_file, 1, 1);
-   int dim = mesh->Dimension();
-
-   // 4. Refine the mesh in serial to increase the resolution. In this example
-   //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
-   //    a command-line parameter.
-   for (int lev = 0; lev < ser_ref_levels; lev++)
-   {
-      mesh->UniformRefinement();
-   }
-
-   // 5. Define a parallel mesh by a partitioning of the serial mesh. Refine
-   //    this mesh further in parallel to increase the resolution. Once the
-   //    parallel mesh is defined, the serial mesh can be deleted.
-   mfem::ParMesh *pmesh = new mfem::ParMesh(MPI_COMM_WORLD, *mesh);
-   delete mesh;
-   for (int lev = 0; lev < par_ref_levels; lev++)
-   {
-      pmesh->UniformRefinement();
-   }
-
-   // 6. Define the power parameter for the p-Laplacian and all other
-   //    coefficients
-   mfem::ConstantCoefficient c_pp(pp);
-   mfem::ConstantCoefficient load(1.000000000);
-   mfem::ConstantCoefficient c_ee(0.000000001);
-
-   // 7. Define the finite element spaces for the solution
-   mfem::H1_FECollection fec(order,dim);
-   mfem::ParFiniteElementSpace fespace(pmesh,&fec,1,mfem::Ordering::byVDIM);
-   HYPRE_Int glob_size=fespace.GlobalTrueVSize();
-   if (myrank == 0)
-   {
-      std::cout << "Number of finite element unknowns: " << glob_size << std::endl;
-   }
-
-   // 8. Define the Dirichlet conditions
-   mfem::Array<int> ess_bdr(pmesh->bdr_attributes.Max());
-   ess_bdr = 1;
-
-   // 9. Define the nonlinear form
-   mfem::ParNonlinearForm* nf=new mfem::ParNonlinearForm(&fespace);
-
-   // 10. Define the solution vector x as a parallel finite element grid function
-   //     corresponding to fespace. Initialize x with initial guess of zero,
-   //     which satisfies the boundary conditions.
-   mfem::ParGridFunction x(&fespace);
-   x = 0.0;
-   mfem::HypreParVector* tv=x.GetTrueDofs();
-   mfem::HypreParVector* sv=x.GetTrueDofs();
-
-   // 11. Define ParaView DataCollection
-   mfem::ParaViewDataCollection *dacol=new
-   mfem::ParaViewDataCollection("Example71",
-                                pmesh);
-   dacol->SetLevelsOfDetail(order);
-   dacol->RegisterField("sol",&x);
-
-
-   // 11. Set domain integrators - start with linear diffusion
-   {
-      // the default power coefficient is 2.0
-      mfem::ConstantCoefficient lpp(2.0);
-      if (integrator==0)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplace(lpp,c_ee,load));
-      }
-      else if (integrator==1)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplaceAD(lpp,c_ee,load));
-      }
-      nf->SetEssentialBC(ess_bdr);
-      // compute the energy
-      double energy=nf->GetEnergy(*tv);
-      if (myrank==0)
-      {
-         std::cout<<"[2] The total energy of the system is E="<<energy<<std::endl;
-      }
-      // time the assembly
-      timer->Clear();
-      timer->Start();
-      nf->GetGradient(*sv);
-      timer->Stop();
-      if (myrank==0)
-      {
-         std::cout<<"[2] The assembly time is: "<<timer->RealTime()<<std::endl;
-      }
-      mfem::Solver *prec=new mfem::HypreBoomerAMG();
-      mfem::GMRESSolver *j_gmres = new mfem::GMRESSolver(MPI_COMM_WORLD);
-      j_gmres->SetRelTol(1e-7);
-      j_gmres->SetAbsTol(1e-15);
-      j_gmres->SetMaxIter(300);
-      j_gmres->SetPrintLevel(print_level);
-      j_gmres->SetPreconditioner(*prec);
-
-      mfem::NewtonSolver* ns;
-      ns=new mfem::NewtonSolver(MPI_COMM_WORLD);
-      ns->iterative_mode = true;
-      ns->SetSolver(*j_gmres);
-      ns->SetOperator(*nf);
-      ns->SetPrintLevel(print_level);
-      ns->SetRelTol(1e-6);
-      ns->SetAbsTol(1e-12);
-      ns->SetMaxIter(3);
-      //solve the problem
-      timer->Clear();
-      timer->Start();
-      ns->Mult(*tv, *sv);
-      timer->Stop();
-      if (myrank==0)
-      {
-         std::cout<<"Time for the NewtonSolver: "<<timer->RealTime()<<std::endl;
-      }
-
-      energy=nf->GetEnergy(*sv);
-      if (myrank==0)
-      {
-         std::cout<<"[pp=2] The total energy of the system is E="<<energy<<std::endl;
-      }
-
-      delete ns;
-      delete j_gmres;
-      delete prec;
-
-      x.SetFromTrueDofs(*sv);
-      dacol->SetTime(2.0);
-      dacol->SetCycle(2);
-      dacol->Save();
-   }
-
-   // 12. Continue with powers higher than 2
-   for (int i=3; i<pp; i++)
-   {
-      delete nf;
-      nf=new mfem::ParNonlinearForm(&fespace);
-      mfem::ConstantCoefficient lpp((double)i);
-      if (integrator==0)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplace(lpp,c_ee,load));
-      }
-      else if (integrator==1)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplaceAD(lpp,c_ee,load));
-      }
-      nf->SetEssentialBC(ess_bdr);
-      // compute the energy
-      double energy=nf->GetEnergy(*sv);
-      if (myrank==0)
-      {
-         std::cout<<"[pp="<<i<<"] The total energy of the system is E="<<energy<<std::endl;
-      }
-      // time the assembly
-      timer->Clear();
-      timer->Start();
-      nf->GetGradient(*sv);
-      timer->Stop();
-      if (myrank==0)
-      {
-         std::cout<<"[pp="<<i<<"] The assembly time is: "<<timer->RealTime()<<std::endl;
-      }
-      mfem::Solver *prec=new mfem::HypreBoomerAMG();
-      mfem::GMRESSolver *j_gmres = new mfem::GMRESSolver(MPI_COMM_WORLD);
-      j_gmres->SetRelTol(1e-7);
-      j_gmres->SetAbsTol(1e-15);
-      j_gmres->SetMaxIter(300);
-      j_gmres->SetPrintLevel(print_level);
-      j_gmres->SetPreconditioner(*prec);
-
-      mfem::NewtonSolver* ns;
-      ns=new mfem::NewtonSolver(MPI_COMM_WORLD);
-      ns->iterative_mode = true;
-      ns->SetSolver(*j_gmres);
-      ns->SetOperator(*nf);
-      ns->SetPrintLevel(print_level);
-      ns->SetRelTol(1e-6);
-      ns->SetAbsTol(1e-12);
-      ns->SetMaxIter(3);
-      //solve the problem
-      timer->Clear();
-      timer->Start();
-      ns->Mult(*tv, *sv);
-      timer->Stop();
-      if (myrank==0)
-      {
-         std::cout<<"Time for the NewtonSolver: "<<timer->RealTime()<<std::endl;
-      }
-
-      energy=nf->GetEnergy(*sv);
-      if (myrank==0)
-      {
-         std::cout<<"[pp="<<i<<"] The total energy of the system is E="<<energy<<std::endl;
-      }
-
-      delete ns;
-      delete j_gmres;
-      delete prec;
-
-      x.SetFromTrueDofs(*sv);
-      dacol->SetTime(i);
-      dacol->SetCycle(i);
-      dacol->Save();
-   }
-
-   // 13. Continue with the final power
-   if ( std::abs(pp-2.0) > std::numeric_limits<double>::epsilon())
-   {
-      delete nf;
-      nf=new mfem::ParNonlinearForm(&fespace);
-      if (integrator==0)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplace(c_pp,c_ee,load));
-      }
-      else if (integrator==1)
-      {
-         nf->AddDomainIntegrator(new mfem::pLaplaceAD(c_pp,c_ee,load));
-      }
-      nf->SetEssentialBC(ess_bdr);
-      // compute the energy
-      double energy=nf->GetEnergy(*sv);
-      if (myrank==0)
-      {
-         std::cout<<"[pp="<<pp<<"] The total energy of the system is E="<<energy<<std::endl;
-      }
-      // time the assembly
-      timer->Clear();
-      timer->Start();
-      nf->GetGradient(*sv);
-      timer->Stop();
-      if (myrank==0)
-      {
-         std::cout<<"[pp="<<pp<<"] The assembly time is: "<<timer->RealTime()<<std::endl;
-      }
-      mfem::Solver *prec=new mfem::HypreBoomerAMG();
-      mfem::GMRESSolver *j_gmres = new mfem::GMRESSolver(MPI_COMM_WORLD);
-      j_gmres->SetRelTol(1e-8);
-      j_gmres->SetAbsTol(1e-15);
-      j_gmres->SetMaxIter(300);
-      j_gmres->SetPrintLevel(print_level);
-      j_gmres->SetPreconditioner(*prec);
-
-      mfem::NewtonSolver* ns;
-      ns=new mfem::NewtonSolver(MPI_COMM_WORLD);
-      ns->iterative_mode = true;
-      ns->SetSolver(*j_gmres);
-      ns->SetOperator(*nf);
-      ns->SetPrintLevel(print_level);
-      ns->SetRelTol(1e-6);
-      ns->SetAbsTol(1e-12);
-      ns->SetMaxIter(3);
-      //solve the problem
-      timer->Clear();
-      timer->Start();
-      ns->Mult(*tv, *sv);
-      timer->Stop();
-      if (myrank==0)
-      {
-         std::cout<<"Time for the NewtonSolver: "<<timer->RealTime()<<std::endl;
-      }
-
-      energy=nf->GetEnergy(*sv);
-      if (myrank==0)
-      {
-         std::cout<<"[pp="<<pp<<"] The total energy of the system is E="<<energy<<std::endl;
-      }
-
-      delete ns;
-      delete j_gmres;
-      delete prec;
-
-      x.SetFromTrueDofs(*sv);
-      dacol->SetTime(pp);
-      if (pp<2.0)
-      {
-         dacol->SetCycle(std::floor(pp));
-      }
-      else
-      {
-         dacol->SetCycle(std::ceil(pp));
-      }
-      dacol->Save();
-   }
-
-
-
-   // 19. Free the used memory
-   delete dacol;
-   delete sv;
-   delete tv;
-   delete nf;
-   delete pmesh;
-   delete timer;
-
-   MPI_Finalize();
-
-   return 0;
-}
-
-
@@ -19,12 +19,8 @@
 //
 // Device sample runs:
 //    ex9 -pa
-//    ex9 -ea
-//    ex9 -fa
 //    ex9 -pa -m ../data/periodic-cube.mesh
 //    ex9 -pa -m ../data/periodic-cube.mesh -d cuda
-//    ex9 -ea -m ../data/periodic-cube.mesh -d cuda
-//    ex9 -fa -m ../data/periodic-cube.mesh -d cuda
 //
 // Description:  This example code solves the time-dependent advection equation
 //               du/dt + v.grad(u) = 0, where v is a given fluid velocity, and
@@ -146,8 +142,6 @@ int main(int argc, char *argv[])
   int ref_levels = 2;
   int order = 3;
   bool pa = false;
-   bool ea = false;
-   bool fa = false;
   const char *device_config = "cpu";
   int ode_solver_type = 4;
   double t_final = 10.0;
@@ -172,10 +166,6 @@ int main(int argc, char *argv[])
                  "Order (degree) of the finite elements.");
   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
                  "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&ea, "-ea", "--element-assembly", "-no-ea",
-                  "--no-element-assembly", "Enable Element 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().");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
@@ -279,16 +269,6 @@ int main(int argc, char *argv[])
      m.SetAssemblyLevel(AssemblyLevel::PARTIAL);
      k.SetAssemblyLevel(AssemblyLevel::PARTIAL);
   }
-   else if (ea)
-   {
-      m.SetAssemblyLevel(AssemblyLevel::ELEMENT);
-      k.SetAssemblyLevel(AssemblyLevel::ELEMENT);
-   }
-   else if (fa)
-   {
-      m.SetAssemblyLevel(AssemblyLevel::FULL);
-      k.SetAssemblyLevel(AssemblyLevel::FULL);
-   }
   m.AddDomainIntegrator(new MassIntegrator);
   k.AddDomainIntegrator(new ConvectionIntegrator(velocity, -1.0));
   k.AddInteriorFaceIntegrator(
@@ -448,19 +428,20 @@ int main(int argc, char *argv[])
 FE_Evolution::FE_Evolution(BilinearForm &_M, BilinearForm &_K, const Vector &_b)
   : TimeDependentOperator(_M.Height()), M(_M), K(_K), b(_b), z(_M.Height())
 {
+   bool pa = M.GetAssemblyLevel() == AssemblyLevel::PARTIAL;
   Array<int> ess_tdof_list;
-   if (M.GetAssemblyLevel() == AssemblyLevel::LEGACYFULL)
-   {
-      M_prec = new DSmoother(M.SpMat());
-      M_solver.SetOperator(M.SpMat());
-      dg_solver = new DG_Solver(M.SpMat(), K.SpMat(), *M.FESpace());
-   }
-   else
+   if (pa)
   {
      M_prec = new OperatorJacobiSmoother(M, ess_tdof_list);
      M_solver.SetOperator(M);
      dg_solver = NULL;
   }
+   else
+   {
+      M_prec = new DSmoother(M.SpMat());
+      dg_solver = new DG_Solver(M.SpMat(), K.SpMat(), *M.FESpace());
+      M_solver.SetOperator(M.SpMat());
+   }
   M_solver.SetPreconditioner(*M_prec);
   M_solver.iterative_mode = false;
   M_solver.SetRelTol(1e-9);
@@ -16,16 +16,11 @@
 //    mpirun -np 4 ex9p -m ../data/disc-nurbs.mesh -p 2 -rp 1 -dt 0.005 -tf 9
 //    mpirun -np 4 ex9p -m ../data/periodic-square.mesh -p 3 -rp 2 -dt 0.0025 -tf 9 -vs 20
 //    mpirun -np 4 ex9p -m ../data/periodic-cube.mesh -p 0 -o 2 -rp 1 -dt 0.01 -tf 8
-//    mpirun -np 3 ex9p -m ../data/amr-hex.mesh -p 1 -rs 1 -rp 0 -dt 0.005 -tf 0.5
 //
 // Device sample runs:
 //    mpirun -np 4 ex9p -pa
-//    mpirun -np 4 ex9p -ea
-//    mpirun -np 4 ex9p -fa
 //    mpirun -np 4 ex9p -pa -m ../data/periodic-cube.mesh
 //    mpirun -np 4 ex9p -pa -m ../data/periodic-cube.mesh -d cuda
-//    mpirun -np 4 ex9p -ea -m ../data/periodic-cube.mesh -d cuda
-//    mpirun -np 4 ex9p -fa -m ../data/periodic-cube.mesh -d cuda
 //
 // Description:  This example code solves the time-dependent advection equation
 //               du/dt + v.grad(u) = 0, where v is a given fluid velocity, and
@@ -166,8 +161,6 @@ int main(int argc, char *argv[])
   int par_ref_levels = 0;
   int order = 3;
   bool pa = false;
-   bool ea = false;
-   bool fa = false;
   const char *device_config = "cpu";
   int ode_solver_type = 4;
   double t_final = 10.0;
@@ -195,10 +188,6 @@ int main(int argc, char *argv[])
                  "Order (degree) of the finite elements.");
   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
                  "--no-partial-assembly", "Enable Partial Assembly.");
-   args.AddOption(&ea, "-ea", "--element-assembly", "-no-ea",
-                  "--no-element-assembly", "Enable Element 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().");
   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
@@ -330,17 +319,6 @@ int main(int argc, char *argv[])
      m->SetAssemblyLevel(AssemblyLevel::PARTIAL);
      k->SetAssemblyLevel(AssemblyLevel::PARTIAL);
   }
-   else if (ea)
-   {
-      m->SetAssemblyLevel(AssemblyLevel::ELEMENT);
-      k->SetAssemblyLevel(AssemblyLevel::ELEMENT);
-   }
-   else if (fa)
-   {
-      m->SetAssemblyLevel(AssemblyLevel::FULL);
-      k->SetAssemblyLevel(AssemblyLevel::FULL);
-   }
-
   m->AddDomainIntegrator(new MassIntegrator);
   k->AddDomainIntegrator(new ConvectionIntegrator(velocity, -1.0));
   k->AddInteriorFaceIntegrator(
@@ -577,21 +555,28 @@ FE_Evolution::FE_Evolution(ParBilinearForm &_M, ParBilinearForm &_K,
     M_solver(_M.ParFESpace()->GetComm()),
     z(_M.Height())
 {
-   if (_M.GetAssemblyLevel()==AssemblyLevel::LEGACYFULL)
-   {
-      M.Reset(_M.ParallelAssemble(), true);
-      K.Reset(_K.ParallelAssemble(), true);
-   }
-   else
+   bool pa = _M.GetAssemblyLevel()==AssemblyLevel::PARTIAL;
+
+   if (pa)
   {
      M.Reset(&_M, false);
      K.Reset(&_K, false);
   }
+   else
+   {
+      M.Reset(_M.ParallelAssemble(), true);
+      K.Reset(_K.ParallelAssemble(), true);
+   }

   M_solver.SetOperator(*M);

   Array<int> ess_tdof_list;
-   if (_M.GetAssemblyLevel()==AssemblyLevel::LEGACYFULL)
+   if (pa)
+   {
+      M_prec = new OperatorJacobiSmoother(_M, ess_tdof_list);
+      dg_solver = NULL;
+   }
+   else
   {
      HypreParMatrix &M_mat = *M.As<HypreParMatrix>();
      HypreParMatrix &K_mat = *K.As<HypreParMatrix>();
@@ -600,11 +585,6 @@ FE_Evolution::FE_Evolution(ParBilinearForm &_M, ParBilinearForm &_K,

      dg_solver = new DG_Solver(M_mat, K_mat, *_M.FESpace());
   }
-   else
-   {
-      M_prec = new OperatorJacobiSmoother(_M, ess_tdof_list);
-      dg_solver = NULL;
-   }

   M_solver.SetPreconditioner(*M_prec);
   M_solver.iterative_mode = false;
@@ -0,0 +1,292 @@
+//                       MFEM Example 1 - Parallel Version
+//
+// Compile with: make ex1p
+//
+// Sample runs:  mpirun -np 4 ex1p -m ../data/square-disc.mesh
+//               mpirun -np 4 ex1p -m ../data/star.mesh
+//               mpirun -np 4 ex1p -m ../data/star-mixed.mesh
+//               mpirun -np 4 ex1p -m ../data/escher.mesh
+//               mpirun -np 4 ex1p -m ../data/fichera.mesh
+//               mpirun -np 4 ex1p -m ../data/fichera-mixed.mesh
+//               mpirun -np 4 ex1p -m ../data/toroid-wedge.mesh
+//               mpirun -np 4 ex1p -m ../data/square-disc-p2.vtk -o 2
+//               mpirun -np 4 ex1p -m ../data/square-disc-p3.mesh -o 3
+//               mpirun -np 4 ex1p -m ../data/square-disc-nurbs.mesh -o -1
+//               mpirun -np 4 ex1p -m ../data/star-mixed-p2.mesh -o 2
+//               mpirun -np 4 ex1p -m ../data/disc-nurbs.mesh -o -1
+//               mpirun -np 4 ex1p -m ../data/pipe-nurbs.mesh -o -1
+//               mpirun -np 4 ex1p -m ../data/ball-nurbs.mesh -o 2
+//               mpirun -np 4 ex1p -m ../data/fichera-mixed-p2.mesh -o 2
+//               mpirun -np 4 ex1p -m ../data/star-surf.mesh
+//               mpirun -np 4 ex1p -m ../data/square-disc-surf.mesh
+//               mpirun -np 4 ex1p -m ../data/inline-segment.mesh
+//               mpirun -np 4 ex1p -m ../data/amr-quad.mesh
+//               mpirun -np 4 ex1p -m ../data/amr-hex.mesh
+//               mpirun -np 4 ex1p -m ../data/mobius-strip.mesh
+//               mpirun -np 4 ex1p -m ../data/mobius-strip.mesh -o -1 -sc
+//
+// Device sample runs:
+//               mpirun -np 4 ex1p -pa -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-cuda
+//               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
+//               simple finite element discretization of the Laplace problem
+//               -Delta u = 1 with homogeneous Dirichlet boundary conditions.
+//               Specifically, we discretize using a FE space of the specified
+//               order, or if order < 1 using an isoparametric/isogeometric
+//               space (i.e. quadratic for quadratic curvilinear mesh, NURBS for
+//               NURBS mesh, etc.)
+//
+//               The example highlights the use of mesh refinement, finite
+//               element grid functions, as well as linear and bilinear forms
+//               corresponding to the left-hand side and right-hand side of the
+//               discrete linear system. We also cover the explicit elimination
+//               of essential boundary conditions, static condensation, and the
+//               optional connection to the GLVis tool for visualization.
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+#include "mpi.h"
+
+using namespace std;
+using namespace mfem;
+
+int main(int argc, char *argv[])
+{
+   // 1. Initialize MPI.
+   int num_procs, myid;
+   MPI_Init(&argc, &argv);
+   MPI_Comm_size(MPI_COMM_WORLD, &num_procs);
+   MPI_Comm_rank(MPI_COMM_WORLD, &myid);
+
+   // 2. Parse command-line options.
+   // const char *mesh_file = "../data/star.mesh";
+   const char *mesh_file = "../data/square-disc.mesh";
+   int order = 1;
+   bool static_cond = false;
+   bool pa = false;
+   const char *device_config = "cpu";
+   bool visualization = false;
+   int nfiles = 1;
+   // const char *out_file = "0_0.gf";
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree) or -1 for"
+                  " isoparametric space.");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&pa, "-pa", "--partial-assembly", "-no-pa",
+                  "--no-partial-assembly", "Enable Partial Assembly.");
+   args.AddOption(&device_config, "-d", "--device",
+                  "Device configuration string, see Device::Configure().");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.AddOption(&nfiles, "-nf", "--num-files", "Number of files to write.");
+   // args.AddOption(&out_file, "-o", "--outfile",
+   //                "Name of file to write.");
+
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0)
+      {
+         args.PrintUsage(cout);
+      }
+      MPI_Finalize();
+      return 1;
+   }
+   if (myid == 0)
+   {
+      args.PrintOptions(cout);
+   }
+
+   // 3. Enable hardware devices such as GPUs, and programming models such as
+   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
+   Device device(device_config);
+   if (myid == 0) { device.Print(); }
+
+   // 4. Read the (serial) mesh from the given mesh file on all processors.  We
+   //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
+   //    and volume meshes with the same code.
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();
+
+   // 5. Refine the serial mesh on all processors to increase the resolution. In
+   //    this example we do 'ref_levels' of uniform refinement. We choose
+   //    'ref_levels' to be the largest number that gives a final mesh with no
+   //    more than 10,000 elements.
+   {
+      int ref_levels =
+         (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh->UniformRefinement();
+      }
+   }
+
+   // 6. Define a parallel mesh by a partitioning of the serial mesh. Refine
+   //    this mesh further in parallel to increase the resolution. Once the
+   //    parallel mesh is defined, the serial mesh can be deleted.
+   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
+   delete mesh;
+   {
+      int par_ref_levels = 2;
+      for (int l = 0; l < par_ref_levels; l++)
+      {
+         pmesh->UniformRefinement();
+      }
+   }
+
+   // 7. Define a parallel finite element space on the parallel mesh. Here we
+   //    use continuous Lagrange finite elements of the specified order. If
+   //    order < 1, we instead use an isoparametric/isogeometric space.
+   FiniteElementCollection *fec;
+   if (order > 0)
+   {
+      fec = new H1_FECollection(order, dim);
+   }
+   else if (pmesh->GetNodes())
+   {
+      fec = pmesh->GetNodes()->OwnFEC();
+      if (myid == 0)
+      {
+         cout << "Using isoparametric FEs: " << fec->Name() << endl;
+      }
+   }
+   else
+   {
+      fec = new H1_FECollection(order = 1, dim);
+   }
+   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec, 1, 0);
+   HYPRE_Int size = fespace->GlobalTrueVSize();
+   if (myid == 0)
+   {
+      cout << "Number of finite element unknowns: " << size << endl;
+   }
+
+   // 8. Determine the list of true (i.e. parallel conforming) essential
+   //    boundary dofs. In this example, the boundary conditions are defined
+   //    by marking all the boundary attributes from the mesh as essential
+   //    (Dirichlet) and converting them to a list of true dofs.
+   Array<int> ess_tdof_list;
+   if (pmesh->bdr_attributes.Size())
+   {
+      Array<int> ess_bdr(pmesh->bdr_attributes.Max());
+      ess_bdr = 1;
+      fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+   }
+
+   // 9. Set up the parallel linear form b(.) which corresponds to the
+   //    right-hand side of the FEM linear system, which in this case is
+   //    (1,phi_i) where phi_i are the basis functions in fespace.
+   ParLinearForm *b = new ParLinearForm(fespace);
+   ConstantCoefficient one(1.0);
+   b->AddDomainIntegrator(new DomainLFIntegrator(one));
+   b->Assemble();
+
+   // 10. Define the solution vector x as a parallel finite element grid function
+   //     corresponding to fespace. Initialize x with initial guess of zero,
+   //     which satisfies the boundary conditions.
+   ParGridFunction x(fespace);
+   x = 0.0;
+
+   // 11. Set up the parallel bilinear form a(.,.) on the finite element space
+   //     corresponding to the Laplacian operator -Delta, by adding the Diffusion
+   //     domain integrator.
+   ParBilinearForm *a = new ParBilinearForm(fespace);
+   if (pa) { a->SetAssemblyLevel(AssemblyLevel::PARTIAL); }
+   a->AddDomainIntegrator(new DiffusionIntegrator(one));
+
+   // 12. Assemble the parallel bilinear form and the corresponding linear
+   //     system, applying any necessary transformations such as: parallel
+   //     assembly, eliminating boundary conditions, applying conforming
+   //     constraints for non-conforming AMR, static condensation, etc.
+   if (static_cond) { a->EnableStaticCondensation(); }
+   a->Assemble();
+
+   OperatorPtr A;
+   Vector B, X;
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);
+
+   // 13. Solve the linear system A X = B.
+   //     * With full assembly, use the BoomerAMG preconditioner from hypre.
+   //     * With partial assembly, use Jacobi smoothing, for now.
+   Solver *prec = NULL;
+   if (pa)
+   {
+      if (UsesTensorBasis(*fespace))
+      {
+         prec = new OperatorJacobiSmoother(*a, ess_tdof_list);
+      }
+   }
+   else
+   {
+      prec = new HypreBoomerAMG;
+   }
+   CGSolver cg(MPI_COMM_WORLD);
+   cg.SetRelTol(1e-12);
+   cg.SetMaxIter(2000);
+   cg.SetPrintLevel(1);
+   if (prec) { cg.SetPreconditioner(*prec); }
+   cg.SetOperator(*A);
+   cg.Mult(B, X);
+   delete prec;
+
+   // 14. Recover the parallel grid function corresponding to X. This is the
+   //     local finite element solution on each processor.
+   a->RecoverFEMSolution(X, *b, x);
+
+   std::string filename = to_string(num_procs) + "_" + to_string(nfiles) + "_";
+   {
+      double t1;
+      t1 = MPI_Wtime();
+      x.Save(filename.c_str(), nfiles);
+      double t2 = MPI_Wtime();
+
+      double write_time = t2 - t1;
+      double average_write_time;
+      MPI_Reduce(&write_time, &average_write_time, 1,
+               MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+      if (myid == 0)
+      {
+         std::cout << "Average write time: " << average_write_time / num_procs << " for "
+                   << nfiles << " files and "  << num_procs << " ranks\n";
+      }
+   }
+   {
+      double t1;
+      t1 = MPI_Wtime();
+      ParGridFunction temp_gf(fespace, filename.c_str());
+      double t2 = MPI_Wtime();
+
+      double read_time = t2 - t1;
+      double average_read_time;
+      MPI_Reduce(&read_time, &average_read_time, 1,
+                 MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+      if (myid == 0)
+      {
+         std::cout << "Average read time: " << average_read_time / num_procs << " for "
+                   << nfiles << " files and "  << num_procs << " ranks\n";
+      }
+   }
+
+   // 17. Free the used memory.
+   delete a;
+   delete b;
+   delete fespace;
+   if (order > 0) { delete fec; }
+   delete pmesh;
+
+   MPI_Finalize();
+
+   return 0;
+}
@@ -22,10 +22,10 @@ MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)

 SEQ_EXAMPLES = ex1 ex2 ex3 ex4 ex5 ex6 ex7 ex8 ex9 ex10 ex14 ex15 ex16 ex17\
-  ex18 ex19 ex20 ex21 ex22 ex23 ex24 ex25 ex26 ex27 ex71
+  ex18 ex19 ex20 ex21 ex22 ex23 ex24 ex25 ex26 ex27
 PAR_EXAMPLES = ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex8p ex9p ex10p ex11p ex12p\
  ex13p ex14p ex15p ex16p ex17p ex18p ex19p ex20p ex21p ex22p ex24p ex25p\
-  ex26p ex27p ex71p
+  ex26p ex27p

 ifeq ($(MFEM_USE_MPI),NO)
   EXAMPLES = $(SEQ_EXAMPLES)
@@ -114,11 +114,6 @@ ex11p-test-strumpack: ex11p
 	@$(call mfem-test,$<, $(RUN_MPI), STRUMPACK example,--strumpack)
 test-par-YES: ex11p-test-strumpack
 endif
-ifeq ($(MFEM_USE_SUPERLU),YES)
-ex11p-test-superlu: ex11p
-	@$(call mfem-test,$<, $(RUN_MPI), SuperLU_DIST example,--superlu)
-test-par-YES: ex11p-test-superlu
-endif

 # Testing: "test" target and mfem-test* variables are defined in config/test.mk

@@ -146,4 +141,3 @@ clean-exec:
 	@rm -f ex21*.mesh ex21*.sol ex21p_*.*
 	@rm -f ex23.mesh ex23-*.gf
 	@rm -f ex25.mesh ex25-*.gf ex25p-*.*
-	@rm -rf Example71
@@ -0,0 +1,906 @@
+//                                MFEM Example 9
+//
+// Compile with: make serial_nogpu
+//
+// Description:  This code solves the time-dependent advection-diffusion
+//               equation:
+//               \frac(\partial u}{\partial t}
+//                                 = \mathbf{a} \cdot \Nabla u - \nu \Nabla^2 u
+//               where a is a given advection velocity, \nu is the diffusion
+//               parameter, and u0(x) = u(0,x) is a given initial condition.
+//
+//               The demonstrates explicit time marching with H1 elements of
+//               arbitrary order. Periodic boundary conditions are used through
+//               periodic meshes. GLVis can be used for visualization of a
+//               time-evolving solution.
+
+#include <fstream>
+#include <iostream>
+#include <algorithm>
+
+#include "mfem.hpp"
+#include "mpi.h"
+
+using namespace std;
+using namespace mfem;
+
+/** A time-dependent operator for the right-hand side of the ODE. The weak
+    form of du/dt = -a.grad(u) + nu Delta(u) is M du/dt = K u + b, where M and
+    K are the mass and advection-diffusion matrices, and b describes the flow
+    on the boundary. This can be written as a general ODE,
+    du/dt = M^{-1} (K u + b), and this class is used to evaluate the right-hand
+    side. */
+class AdvectionDiffusionEvolution : public mfem::TimeDependentOperator
+{
+public:
+   /// \param[in] M - bilinear form for mass matrix
+   /// \param[in] K - bilinear form for stiffness matrix
+   /// \param[in] b - load vector
+   AdvectionDiffusionEvolution(mfem::BilinearForm &M, mfem::BilinearForm &K,
+                               const mfem::Vector &b);
+
+   /// Perform the action of the operator: y = k = f(x, t), where k solves
+   /// Compute k = M^-1(Kx + l)
+   void Mult(const mfem::Vector &x, mfem::Vector &y) const override;
+
+   /// Solve the implicit equation: k = f(x + dt k, t), for the unknown k at
+   /// the current time t. 
+   void ImplicitSolve(const double dt, const mfem::Vector &x,
+                      mfem::Vector &k) override;
+
+   virtual ~AdvectionDiffusionEvolution();
+
+private:
+   mfem::BilinearForm &M, &K;
+   const mfem::Vector &b;
+   /// solver for inverting mass matrix for explicit time-marching
+   std::unique_ptr<mfem::Solver> M_prec;
+   mfem::CGSolver M_solver;
+   /// solver for implicit time-marching
+   mfem::GSSmoother prec;
+   mfem::GMRESSolver linear_solver;
+   mfem::NewtonSolver newton;
+
+   mutable mfem::Vector z;
+
+   /// pointer-to-implementation idiom
+   /// Hides implementation details of this operator
+   class SystemOperator;
+   /// Operator that combines the linear spatial discretization with
+   /// the load vector into one operator used for implicit solves
+   std::unique_ptr<SystemOperator> combined_oper;
+
+   /// sets the state and dt for the combined operator
+   /// \param[in] dt - time increment
+   /// \param[in] x - the current state
+   void setOperParameters(double dt, const mfem::Vector *x);
+
+};
+
+class PAJacobianOperator : public mfem::Operator
+{
+public:
+   PAJacobianOperator(mfem::ParBilinearForm &_mass,
+                      mfem::ParBilinearForm &_stiff);
+
+   /// Compute r = J@k = M@k + dt*K@k
+   /// \param[in] k - dx/dt
+   /// \param[out] r - J@k = M@k + dt*K@k
+   void Mult(const mfem::Vector &k, mfem::Vector &r) const override;
+      
+   /// Set current dt values - needed to compute action of Jacobian.
+   void setParameters(double dt);
+
+private:
+   mfem::ParBilinearForm &mass;
+   mfem::ParBilinearForm &stiff;
+
+   double dt;
+};
+
+class ParSystemOperator : public mfem::Operator
+{
+public:
+   /// Nonlinear operator of the form that combines the mass, res, stiff,
+   /// and load elements for implicit/explicit ODE integration
+   /// \param[in] ess_bdr - array of boundaries attributes marked essential
+   /// \param[in] mass - bilinear form for mass matrix (not owned)
+   /// \param[in] res - nonlinear residual operator (not owned)
+   /// \param[in] stiff - bilinear form for stiffness matrix (not owned)
+   /// \param[in] load - load vector (not owned)
+   /// \param[in] a - used to move the spatial residual to the rhs
+   ParSystemOperator(mfem::ParBilinearForm &_mass,
+                     mfem::ParBilinearForm &_stiff);
+
+   /// Compute r = M@k + K@(x+dt*k)
+   /// (with `@` denoting matrix-vector multiplication)
+   /// \param[in] k - dx/dt 
+   /// \param[out] r - the residual
+   /// \note the signs on each operator must be accounted for elsewhere
+   void Mult(const mfem::Vector &k, mfem::Vector &r) const override;
+
+   /// Compute J = M + dt * K
+   /// \param[in] k - dx/dt 
+   mfem::Operator &GetGradient(const mfem::Vector &k) const override;
+
+   /// Set current dt and x values - needed to compute action and Jacobian.
+   void setParameters(double _dt, const mfem::Vector *_x);
+
+   ~ParSystemOperator();
+
+private:
+   mfem::ParBilinearForm &mass;
+   mfem::ParBilinearForm &stiff;
+   mutable mfem::HypreParMatrix *jacobian, *stiff_jacobian;
+
+   double dt;
+   const mfem::Vector *x;
+
+   mutable mfem::Vector work, work2;
+
+   std::unique_ptr<PAJacobianOperator> pa_jac;
+
+};
+
+/** A time-dependent operator for the right-hand side of the ODE. The weak
+    form of du/dt = -a.grad(u) + nu Delta(u) is M du/dt = K u + b, where M and
+    K are the mass and advection-diffusion matrices, and b describes the flow
+    on the boundary. This can be written as a general ODE,
+    du/dt = M^{-1} (K u + b), and this class is used to evaluate the right-hand
+    side. */
+class ParAdvectionDiffusionEvolution : public mfem::TimeDependentOperator
+{
+public:
+   /// \param[in] M - parallel bilinear form for mass matrix
+   /// \param[in] K - parallel bilinear form for stiffness matrix
+   ParAdvectionDiffusionEvolution(mfem::ParBilinearForm &M,
+                                  mfem::ParBilinearForm &K);
+
+   /// Perform the action of the operator: y = k = f(x, t), where k solves
+   /// Compute k = M^-1(Kx + l)
+   void Mult(const mfem::Vector &x, mfem::Vector &y) const override;
+
+   /// Solve the implicit equation: k = f(x + dt k, t), for the unknown k at
+   /// the current time t. 
+   void ImplicitSolve(const double dt, const mfem::Vector &x,
+                      mfem::Vector &k) override;
+
+   virtual ~ParAdvectionDiffusionEvolution();
+
+private:
+   mfem::OperatorHandle M_;
+   mfem::ParBilinearForm &M, &K;
+   /// solver for inverting mass matrix for explicit time-marching
+   std::unique_ptr<mfem::Solver> M_prec;
+   mfem::CGSolver M_solver;
+   /// solver for implicit time-marching
+   mfem::Solver *prec;
+   mfem::GMRESSolver linear_solver;
+   mfem::NewtonSolver newton;
+
+   mfem::Vector diag;
+   mutable mfem::Vector z, work, work2;
+
+   /// pointer-to-implementation idiom
+   /// Hides implementation details of this operator
+   /// Operator that combines the linear spatial discretization with
+   /// the load vector into one operator used for implicit solves
+   std::unique_ptr<ParSystemOperator> combined_oper;
+
+   /// sets the state and dt for the combined operator
+   /// \param[in] dt - time increment
+   /// \param[in] x - the current state
+   void setOperParameters(double dt, const mfem::Vector *x);
+
+};
+
+// Choice for the problem setup. The fluid velocity, initial condition and
+// inflow boundary condition are chosen based on this parameter.
+int problem;
+
+// Velocity coefficient
+void velocity_function(const Vector &X, Vector &v);
+
+// Initial condition
+double u0_function(const Vector &X);
+
+// Inflow boundary condition
+double inflow_function(const Vector &X, const double t);
+
+// Mesh bounding box
+Vector bb_min, bb_max;
+
+int main(int argc, char *argv[])
+{
+   // 1. Initialize MPI.
+   int num_procs, myid;
+   MPI_Init(&argc, &argv);
+   MPI_Comm_size(MPI_COMM_WORLD, &num_procs);
+   MPI_Comm_rank(MPI_COMM_WORLD, &myid);
+
+   // 2. Parse command-line options.
+   problem = 3;
+   const char *mesh_file = "../data/periodic-square.mesh";
+   int ser_ref_levels = 0;
+   int par_ref_levels = 0;
+   int order = 3;
+   const char *device_config = "cpu";
+   int ode_solver_type = 22;
+   double t_final = 3 * 2*M_PI;
+   double dt = 0.01;
+   bool glvis = false;
+   bool paraview = false;
+   int vis_steps = 5;
+
+   double nu_val = 0.001;
+
+   int precision = 8;
+   cout.precision(precision);
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&problem, "-p", "--problem",
+                  "Problem setup to use. See options in velocity_function().");
+   args.AddOption(&ser_ref_levels, "-rs", "--refine-serial",
+                  "Number of times to refine the mesh uniformly in serial.");
+   args.AddOption(&par_ref_levels, "-rp", "--refine-parallel",
+                  "Number of times to refine the mesh uniformly in parallel.");
+   args.AddOption(&order, "-o", "--order",
+                  "Order (degree) of the finite elements.");
+   args.AddOption(&device_config, "-d", "--device",
+                  "Device configuration string, see Device::Configure().");
+   args.AddOption(&ode_solver_type, "-s", "--ode-solver",
+                  "ODE solver: 1 - Forward Euler,\n\t"
+                  "            2 - RK2 SSP, 3 - RK3 SSP, 4 - RK4, 6 - RK6.");
+   args.AddOption(&t_final, "-tf", "--t-final",
+                  "Final time; start time is 0.");
+   args.AddOption(&dt, "-dt", "--time-step",
+                  "Time step.");
+   args.AddOption(&glvis, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.AddOption(&paraview, "-paraview", "--paraview-datafiles", "-no-paraview",
+                  "--no-paraview-datafiles",
+                  "Save data files for ParaView (paraview.org) visualization.");
+   args.AddOption(&vis_steps, "-vs", "--visualization-steps",
+                  "Visualize every n-th timestep.");
+   args.AddOption(&nu_val, "-nu", "--nu-value",
+                  "Value for \nu, the parameter that controls diffusion.");
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0)
+      {
+         args.PrintUsage(cout);
+      }
+      MPI_Finalize();
+      return 1;
+   }
+   if (myid == 0)
+   {
+      std::cout << "Num ranks: " << num_procs << "\n";
+      args.PrintOptions(cout);
+   }
+   Device device(device_config);
+   if (myid == 0) { device.Print(); }
+
+   // 3. Read the serial mesh from the given mesh file on all processors. We can
+   //    handle geometrically periodic meshes in this code.
+   Mesh *mesh = new Mesh(mesh_file, 1, 1);
+   int dim = mesh->Dimension();
+
+   // 5. Refine the mesh in serial to increase the resolution. In this example
+   //    we do 'ser_ref_levels' of uniform refinement, where 'ser_ref_levels' is
+   //    a command-line parameter. If the mesh is of NURBS type, we convert it
+   //    to a (piecewise-polynomial) high-order mesh.
+   for (int lev = 0; lev < ser_ref_levels; lev++)
+   {
+      mesh->UniformRefinement();
+   }
+   mesh->GetBoundingBox(bb_min, bb_max, max(order, 1));
+
+   // 6. Define the parallel mesh by a partitioning of the serial mesh. Refine
+   //    this mesh further in parallel to increase the resolution. Once the
+   //    parallel mesh is defined, the serial mesh can be deleted.
+   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
+   delete mesh;
+   for (int lev = 0; lev < par_ref_levels; lev++)
+   {
+      pmesh->UniformRefinement();
+   }
+
+   // 7. Define the finite element space of the given
+   //    polynomial order on the refined mesh.
+   H1_FECollection fec(order, dim, BasisType::GaussLobatto);
+   ParFiniteElementSpace *fes = new ParFiniteElementSpace(pmesh, &fec);
+
+   HYPRE_Int global_vSize = fes->GlobalTrueVSize();
+   if (myid == 0)
+   {
+      cout << "Number of unknowns: " << global_vSize << endl;
+   }
+
+   // 8. Set up and assemble the bilinear and linear forms corresponding to the
+   //    CG discretization.
+   /// negative to move the diffusion terms to the right side
+   ConstantCoefficient nu(-nu_val);
+   ConstantCoefficient one(1.0);
+   VectorFunctionCoefficient velocity(dim, velocity_function);
+   FunctionCoefficient u0(u0_function);
+
+   ParBilinearForm *m_pa = new ParBilinearForm(fes);
+   ParBilinearForm *k_pa = new ParBilinearForm(fes);
+   m_pa->SetAssemblyLevel(AssemblyLevel::PARTIAL);
+   k_pa->SetAssemblyLevel(AssemblyLevel::PARTIAL);
+
+   /// create mass matrix
+   m_pa->AddDomainIntegrator(new MassIntegrator(one));
+   /// add advection terms to stiffness matrix
+   k_pa->AddDomainIntegrator(new ConvectionIntegrator(velocity, -1.0));
+   /// add diffusion terms to stiffness matrix
+   k_pa->AddDomainIntegrator(new DiffusionIntegrator(nu));
+
+   m_pa->Assemble();
+   int skip_zeros = 0;
+   k_pa->Assemble(skip_zeros);
+   m_pa->Finalize();
+   k_pa->Finalize(skip_zeros);
+
+   ParBilinearForm *m = new ParBilinearForm(fes);
+   ParBilinearForm *k = new ParBilinearForm(fes);
+   /// create mass matrix
+   m->AddDomainIntegrator(new MassIntegrator);
+   /// add advection terms to stiffness matrix
+   k->AddDomainIntegrator(new ConvectionIntegrator(velocity, -1.0));
+   /// add diffusion terms to stiffness matrix
+   k->AddDomainIntegrator(new DiffusionIntegrator(nu));
+
+   m->Assemble();
+   k->Assemble(skip_zeros);
+   m->Finalize();
+   k->Finalize(skip_zeros);
+
+
+   ParGridFunction *u = new ParGridFunction(fes);
+   u->UseDevice(true);
+   u->ProjectCoefficient(u0);
+
+
+   HypreParVector *U = u->GetTrueDofs();
+
+   ParSystemOperator pso(*m, *k);
+   ParSystemOperator pso_pa(*m_pa, *k_pa);
+
+   pso.setParameters(dt, U);
+   pso_pa.setParameters(dt, U);
+
+   MPI_Barrier(MPI_COMM_WORLD);
+   mfem::Vector pso_r(U->Size());
+   double t1 = MPI_Wtime();
+   pso.Mult(*U, pso_r);
+   double t2 = MPI_Wtime();
+   double fa_mult_time = t2 - t1;
+   double average_fa_mult_time;
+   MPI_Reduce(&fa_mult_time, &average_fa_mult_time, 1,
+              MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+   if (myid == 0)
+      std::cout << "FA Mult time: " << average_fa_mult_time / num_procs << endl;
+
+   MPI_Barrier(MPI_COMM_WORLD);
+   mfem::Vector pso_pa_r(U->Size());
+   double t3 = MPI_Wtime();
+   pso_pa.Mult(*U, pso_pa_r);
+   double t4 = MPI_Wtime();
+   double pa_mult_time = t4 - t3;
+   double average_pa_mult_time;
+   MPI_Reduce(&pa_mult_time, &average_pa_mult_time, 1,
+              MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+   if (myid == 0)
+      std::cout << "FA Mult time: " << average_pa_mult_time / num_procs << endl;
+   
+   double local_mult_speedup = (t2-t1) / (t4-t3);
+   double global_mult_speedup;
+   MPI_Reduce(&local_mult_speedup, &global_mult_speedup, 1,
+              MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+
+   if (myid == 0)
+      std::cout << "PA mult speedup: " << global_mult_speedup / num_procs << endl;
+
+   mfem::Vector diff_r(pso_pa_r);
+   diff_r -= pso_r;
+   // std::cout << "r diff: " << diff_r.Norml2() << std::endl;
+
+   mfem::Operator &pso_jac = pso.GetGradient(*U);
+   mfem::Operator &pso_pa_jac = pso_pa.GetGradient(*U);
+
+   MPI_Barrier(MPI_COMM_WORLD);
+   mfem::Vector pso_jac_r(U->Size());
+   double t5 = MPI_Wtime();
+   pso_jac.Mult(*U, pso_jac_r);
+   double t6 = MPI_Wtime();
+   double fa_jac_mult_time = t6-t5;
+   double average_fa_jac_time;
+   MPI_Reduce(&fa_jac_mult_time, &average_fa_jac_time, 1,
+              MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+   if (myid == 0)
+      std::cout << "FA Jac Mult time: " << average_fa_jac_time / num_procs << endl;
+
+   MPI_Barrier(MPI_COMM_WORLD);
+   mfem::Vector pso_pa_jac_r(U->Size());
+   double t7 = MPI_Wtime();
+   pso_pa_jac.Mult(*U, pso_pa_jac_r);
+   double t8 = MPI_Wtime();
+   double pa_jac_mult_time = t8-t7;
+   double average_pa_jac_time;
+   MPI_Reduce(&pa_jac_mult_time, &average_pa_jac_time, 1,
+              MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+   if (myid == 0)
+      std::cout << "PA Jac Mult time: " << average_pa_jac_time / num_procs << endl;
+
+   double local_jac_speedup = (t6-t5) / (t8-t7);
+   double global_jac_speedup;
+   MPI_Reduce(&local_jac_speedup, &global_jac_speedup, 1,
+              MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
+   if (myid == 0)
+      std::cout << "PA Jac mult speedup: " << global_jac_speedup / num_procs << endl;
+
+   // 13. Free the used memory.
+   delete U;
+   delete u;
+   delete k;
+   delete m;
+   delete fes;
+   delete pmesh;
+
+   MPI_Finalize();
+   return 0;
+}
+
+// Velocity coefficient
+void velocity_function(const Vector &x, Vector &v)
+{
+   int dim = x.Size();
+
+   // map to the reference [-1,1] domain
+   Vector X(dim);
+   for (int i = 0; i < dim; i++)
+   {
+      double center = (bb_min[i] + bb_max[i]) * 0.5;
+      X(i) = 2 * (x(i) - center) / (bb_max[i] - bb_min[i]);
+   }
+
+   switch (problem)
+   {
+      case 3:
+      {
+         // Translations in 1D, 2D, and 3D
+         switch (dim)
+         {
+            case 1: v(0) = 1.0; break;
+            case 2: v(0) = sqrt(2./3.); v(1) = sqrt(1./3.); break;
+            case 3: v(0) = sqrt(3./6.); v(1) = sqrt(2./6.); v(2) = sqrt(1./6.);
+               break;
+         }
+         break;
+      }
+      case 1:
+      case 2:
+      {
+         // Clockwise rotation in 2D around the origin
+         const double w = M_PI/2;
+         switch (dim)
+         {
+            case 1: v(0) = 1.0; break;
+            case 2: v(0) = w*X(1); v(1) = -w*X(0); break;
+            case 3: v(0) = w*X(1); v(1) = -w*X(0); v(2) = 0.0; break;
+         }
+         break;
+      }
+      case 0:
+      {
+         // Clockwise twisting rotation in 2D around the origin
+         const double w = M_PI/2;
+         double d = max((X(0)+1.)*(1.-X(0)),0.) * max((X(1)+1.)*(1.-X(1)),0.);
+         d = d*d;
+         switch (dim)
+         {
+            case 1: v(0) = 1.0; break;
+            case 2: v(0) = d*w*X(1); v(1) = -d*w*X(0); break;
+            case 3: v(0) = d*w*X(1); v(1) = -d*w*X(0); v(2) = 0.0; break;
+         }
+         break;
+      }
+   }
+}
+
+// Initial condition
+double u0_function(const Vector &x)
+{
+   int dim = x.Size();
+
+   // map to the reference [-1,1] domain
+   Vector X(dim);
+   for (int i = 0; i < dim; i++)
+   {
+      double center = (bb_min[i] + bb_max[i]) * 0.5;
+      X(i) = 2 * (x(i) - center) / (bb_max[i] - bb_min[i]);
+   }
+
+   switch (problem)
+   {
+      case 0:
+      case 1:
+      {
+         switch (dim)
+         {
+            case 1:
+               return exp(-40.*pow(X(0)-0.5,2));
+            case 2:
+            case 3:
+            {
+               double rx = 0.45, ry = 0.25, cx = 0., cy = -0.2, w = 10.;
+               if (dim == 3)
+               {
+                  const double s = (1. + 0.25*cos(2*M_PI*X(2)));
+                  rx *= s;
+                  ry *= s;
+               }
+               return ( erfc(w*(X(0)-cx-rx))*erfc(-w*(X(0)-cx+rx)) *
+                        erfc(w*(X(1)-cy-ry))*erfc(-w*(X(1)-cy+ry)) )/16;
+            }
+         }
+      }
+      case 2:
+      {
+         double x_ = X(0), y_ = X(1), rho, phi;
+         rho = hypot(x_, y_);
+         phi = atan2(y_, x_);
+         return pow(sin(M_PI*rho),2)*sin(3*phi);
+      }
+      case 3:
+      {
+         const double f = M_PI;
+         return sin(f*X(0))*sin(f*X(1));
+      }
+   }
+   return 0.0;
+}
+
+// Inflow boundary condition (zero for the problems considered in this example)
+double inflow_function(const Vector &x, const double t)
+{
+   switch (problem)
+   {
+      case 0:
+      case 1:
+      case 2:
+      case 3: return 0.0;
+   }
+   return 0.0;
+}
+
+class AdvectionDiffusionEvolution::SystemOperator : public mfem::Operator
+{
+public:
+   /// Nonlinear operator of the form that combines the mass, res, stiff,
+   /// and load elements for implicit/explicit ODE integration
+   /// \param[in] mass - bilinear form for mass matrix (not owned)
+   /// \param[in] res - nonlinear residual operator (not owned)
+   /// \param[in] stiff - bilinear form for stiffness matrix (not owned)
+   /// \param[in] load - load vector (not owned)
+   /// \param[in] a - used to move the spatial residual to the rhs
+   SystemOperator(BilinearForm &_mass, BilinearForm &_stiff,
+                  const mfem::Vector &b)
+      : Operator(_mass.Height()), mass(_mass), stiff(_stiff),
+        load(b), Jacobian(NULL), dt(0.0), x(NULL), work(height)
+      { }
+
+   /// Compute r = M@k + K@(x+dt*k) + l
+   /// (with `@` denoting matrix-vector multiplication)
+   /// \param[in] k - dx/dt 
+   /// \param[out] r - the residual
+   /// \note the signs on each operator must be accounted for elsewhere
+   void Mult(const mfem::Vector &k, mfem::Vector &r) const override
+   {
+      /// work = x+dt*k = x+dt*dx/dt = x+dx
+      add(1.0, *x, dt, k, work);
+      r = 0.0;
+      stiff.AddMult(work, r);
+      r += load;
+      mass.AddMult(k, r, -1.0);
+   }
+
+   /// Compute J = M + dt * K
+   /// \param[in] k - dx/dt 
+   mfem::Operator &GetGradient(const mfem::Vector &k) const override
+   {
+      delete Jacobian;
+      Jacobian = Add(-1.0, mass.SpMat(), dt, stiff.SpMat());
+      return *Jacobian;
+   }
+
+   /// Set current dt and x values - needed to compute action and Jacobian.
+   void setParameters(double _dt, const mfem::Vector *_x)
+   {
+      dt = _dt;
+      x = _x;
+   };
+
+   ~SystemOperator() {delete Jacobian;};
+
+private:
+   BilinearForm &mass;
+   BilinearForm &stiff;
+   const mfem::Vector &load;
+   mutable mfem::SparseMatrix *Jacobian;
+
+   double dt;
+   const mfem::Vector *x;
+
+   mutable mfem::Vector work, work2;
+
+};
+
+AdvectionDiffusionEvolution::AdvectionDiffusionEvolution(
+      BilinearForm &_M, BilinearForm &_K, const Vector &_b)
+   : TimeDependentOperator(_M.Height()), M(_M), K(_K), b(_b),
+     z(_M.Height())
+{
+   bool pa = M.GetAssemblyLevel() == AssemblyLevel::PARTIAL;
+   Array<int> ess_tdof_list;
+   if (pa)
+   {
+      M_prec.reset(new OperatorJacobiSmoother(M, ess_tdof_list));
+      M_solver.SetOperator(M);
+   }
+   else
+   {
+      M_prec.reset(new DSmoother(M.SpMat()));
+      M_solver.SetOperator(M.SpMat());
+   }
+
+   combined_oper.reset(new SystemOperator(_M, _K, _b));
+
+   M_solver.SetPreconditioner(*M_prec);
+   M_solver.iterative_mode = false;
+   M_solver.SetRelTol(1e-9);
+   M_solver.SetAbsTol(0.0);
+   M_solver.SetMaxIter(100);
+   M_solver.SetPrintLevel(0);
+
+   linear_solver.iterative_mode = true;
+   linear_solver.SetRelTol(1e-12);
+   linear_solver.SetAbsTol(0.0);
+   linear_solver.SetMaxIter(100);
+   linear_solver.SetPrintLevel(0);
+   linear_solver.SetPreconditioner(prec);
+
+   newton.iterative_mode = false;
+   newton.SetRelTol(1e-9);
+   newton.SetAbsTol(0.0);
+   newton.SetMaxIter(100);
+   newton.SetPrintLevel(-1);
+   newton.SetSolver(linear_solver);
+   newton.SetOperator(*combined_oper);
+}
+
+void AdvectionDiffusionEvolution::Mult(const Vector &x, Vector &y) const
+{
+   // y = M^{-1} (K x + b)
+   K.Mult(x, z);
+   z += b;
+   M_solver.Mult(z, y);
+}
+
+void AdvectionDiffusionEvolution::ImplicitSolve(const double dt,
+                                                const Vector &x,
+                                                Vector &k)
+{
+   setOperParameters(dt, &x);
+   Vector zero; // empty vector is interpreted as zero r.h.s. by NewtonSolver
+   newton.Mult(zero, k);
+   MFEM_VERIFY(newton.GetConverged(), "Newton solver did not converge!");
+}
+
+void AdvectionDiffusionEvolution::setOperParameters(double dt,
+                                                    const mfem::Vector *x)
+{
+   combined_oper->setParameters(dt, x);
+}
+
+AdvectionDiffusionEvolution::~AdvectionDiffusionEvolution() {}
+
+
+PAJacobianOperator::PAJacobianOperator(ParBilinearForm &_mass, ParBilinearForm &_stiff)
+ : Operator(_mass.ParFESpace()->GetTrueVSize()), mass(_mass), stiff(_stiff),
+   dt(0.0) { }
+
+
+void PAJacobianOperator::Mult(const mfem::Vector &k, mfem::Vector &r) const
+{
+   r.UseDevice(true);
+   r = 0.0;
+   stiff.TrueAddMult(k, r, dt);
+   mass.TrueAddMult(k, r, -1.0);
+}
+   
+void PAJacobianOperator::setParameters(const double _dt)
+{
+   dt = _dt;
+};
+
+ParSystemOperator::ParSystemOperator(ParBilinearForm &_mass, ParBilinearForm &_stiff)
+   : Operator(_mass.ParFESpace()->GetTrueVSize()), mass(_mass), stiff(_stiff),
+   jacobian(NULL), stiff_jacobian(NULL), dt(0.0), x(NULL),
+   work(height)
+{
+   pa_jac.reset(new PAJacobianOperator(mass, stiff));
+}
+
+/// Compute r = M@k + K@(x+dt*k)
+/// (with `@` denoting matrix-vector multiplication)
+/// \param[in] k - dx/dt 
+/// \param[out] r - the residual
+/// \note the signs on each operator must be accounted for elsewhere
+void ParSystemOperator::Mult(const mfem::Vector &k, mfem::Vector &r) const
+{
+   r = 0.0;
+   work.UseDevice(true);
+   work = 0.0;
+   /// work = x+dt*k = x+dt*dx/dt = x+dx
+   if (x)
+   {
+      add(1.0, *x, dt, k, work);
+   }
+
+   stiff.TrueAddMult(work, r);
+   mass.TrueAddMult(k, r, -1.0);
+}
+
+/// Compute J = M + dt * K
+/// \param[in] k - dx/dt 
+mfem::Operator &ParSystemOperator::GetGradient(const mfem::Vector &k) const
+{
+   bool mass_pa = mass.GetAssemblyLevel() == AssemblyLevel::PARTIAL;
+   bool stiff_pa = stiff.GetAssemblyLevel() == AssemblyLevel::PARTIAL;
+   
+   if (mass_pa && stiff_pa)
+   {
+      return *pa_jac.get();
+   }
+   else
+   {
+      delete stiff_jacobian;
+      delete jacobian;
+      jacobian = mass.ParallelAssemble();
+      *jacobian *= -1.0; //alpha;
+      stiff_jacobian = stiff.ParallelAssemble();
+      jacobian->Add(dt, *stiff_jacobian);
+      return *jacobian;
+   }
+}
+
+/// Set current dt and x values - needed to compute action and Jacobian.
+void ParSystemOperator::setParameters(const double _dt, const mfem::Vector *_x)
+{
+   dt = _dt;
+   x = _x;
+   pa_jac->setParameters(_dt);
+};
+
+ParSystemOperator::~ParSystemOperator()
+{
+   delete jacobian;
+   delete stiff_jacobian;
+};
+
+ParAdvectionDiffusionEvolution::ParAdvectionDiffusionEvolution(
+      ParBilinearForm &_M, ParBilinearForm &_K)
+   : TimeDependentOperator(_M.ParFESpace()->GetTrueVSize()), M(_M), K(_K), z(_M.Height())
+{
+   bool mass_pa = M.GetAssemblyLevel() == AssemblyLevel::PARTIAL;
+   bool stiff_pa = K.GetAssemblyLevel() == AssemblyLevel::PARTIAL;
+
+   Array<int> ess_tdof_list;
+   M_solver = CGSolver(MPI_COMM_WORLD);
+   if (mass_pa)
+   {
+      M_prec.reset(new OperatorJacobiSmoother(M, ess_tdof_list));
+      M_solver.SetOperator(M);
+   }
+   else
+   {
+      M_.Reset(_M.ParallelAssemble(), true);
+
+      // M_prec.reset(new HypreSmoother());
+      // M_solver.SetOperator(M.As<HypreParMatrix>());
+      HypreParMatrix &M_mat = *M_.As<HypreParMatrix>();
+      // HypreParMatrix &K_mat = *K.As<HypreParMatrix>();
+      M_prec.reset(new HypreSmoother(M_mat, HypreSmoother::Jacobi));
+   }
+
+   combined_oper.reset(new ParSystemOperator(_M, _K));
+
+   M_solver.SetPreconditioner(*M_prec);
+   M_solver.iterative_mode = false;
+   M_solver.SetRelTol(1e-9);
+   M_solver.SetAbsTol(0.0);
+   M_solver.SetMaxIter(100);
+   M_solver.SetPrintLevel(0);
+
+   if (mass_pa && stiff_pa)
+   {
+      diag.UseDevice(true);
+      diag.SetSize(M.ParFESpace()->GetTrueVSize());
+      diag = 0.0;
+      work.UseDevice(true);
+      work2.UseDevice(true);
+      work.SetSize(M.ParFESpace()->GetTrueVSize());
+      work2.SetSize(M.ParFESpace()->GetTrueVSize());
+      work = 0.0;
+      work2 = 0.0;
+      M.AssembleDiagonal(work);
+
+      ParBilinearForm k(M.ParFESpace());
+      ConstantCoefficient nu(-0.01);
+      k.AddDomainIntegrator(new mfem::DiffusionIntegrator(nu));
+      k.SetAssemblyLevel(AssemblyLevel::PARTIAL);
+      k.Assemble(0);
+      k.Finalize(0);
+      k.AssembleDiagonal(work2);
+
+      double dt = 0.1;
+      add(-1.0, work, dt, work2, diag);
+
+      prec = new OperatorChebyshevSmoother(combined_oper.get(), diag,
+                                           ess_tdof_list, 5,
+                                           M.ParFESpace()->GetComm());
+   }
+   else
+   {
+      prec = new HypreSmoother();
+   }
+   
+   linear_solver = GMRESSolver(MPI_COMM_WORLD);
+   linear_solver.iterative_mode = true;
+   linear_solver.SetRelTol(1e-12);
+   linear_solver.SetAbsTol(0.0);
+   linear_solver.SetMaxIter(2000);
+   linear_solver.SetPrintLevel(0);
+   linear_solver.SetPreconditioner(*prec);
+   linear_solver.SetKDim(2000);
+
+   newton.iterative_mode = true;
+   newton.SetRelTol(1e-9);
+   newton.SetAbsTol(0.0);
+   newton.SetMaxIter(10);
+   newton.SetPrintLevel(-1);
+   newton.SetSolver(linear_solver);
+   newton.SetOperator(*combined_oper);
+}
+
+void ParAdvectionDiffusionEvolution::Mult(const Vector &x, Vector &y) const
+{
+   // y = M^{-1} (K x + b)
+   K.Mult(x, z);
+   M_solver.Mult(z, y);
+}
+
+void ParAdvectionDiffusionEvolution::ImplicitSolve(const double dt,
+                                                   const Vector &x,
+                                                   Vector &k)
+{
+   setOperParameters(dt, &x);
+   Vector zero; // empty vector is interpreted as zero r.h.s. by NewtonSolver
+   newton.Mult(zero, k);
+   MFEM_VERIFY(newton.GetConverged(), "Newton solver did not converge!");
+}
+
+void ParAdvectionDiffusionEvolution::setOperParameters(const double dt,
+                                                       const mfem::Vector *x)
+{
+   combined_oper->setParameters(dt, x);
+}
+
+ParAdvectionDiffusionEvolution::~ParAdvectionDiffusionEvolution() {delete prec;}
@@ -34,15 +34,6 @@ if (MFEM_USE_MPI)
    )
 endif()

-if (MFEM_USE_SLEPC)
-  list(APPEND PETSC_EXAMPLES_SRCS
-    ex11p.cpp
-    )
-  list(APPEND PETSC_RC_FILES
-    rc_ex11p_lobpcg rc_ex11p_gd
-    )
-endif()
-
 # Include the source directory where mfem.hpp and mfem-performance.hpp are.
 include_directories(BEFORE ${PROJECT_BINARY_DIR})

@@ -87,22 +78,12 @@ set(EX9_E_ARGS       -m ../../data/periodic-hexagon.mesh --usepetsc --petscopts
 set(EX9_ES_ARGS      -m ../../data/periodic-hexagon.mesh --usepetsc --petscopts rc_ex9p_expl --no-step)
 set(EX9_IS_ARGS      -m ../../data/periodic-hexagon.mesh --usepetsc --petscopts rc_ex9p_impl --implicit -tf 0.5)
 set(EX10_ARGS        -m ../../data/beam-quad.mesh --usepetsc --petscopts rc_ex10p -tf 30 -s 3 -rs 2 -dt 3)
-if (MFEM_USE_SLEPC)
-  set(EX11_ARGS_SINV   -m ../../data/star.mesh --useslepc)
-  set(EX11_ARGS_LOBPCG -m ../../data/star.mesh --useslepc --slepcopts rc_ex11p_lobpcg)
-  set(EX11_ARGS_GD -m ../../data/star.mesh --useslepc --slepcopts rc_ex11p_gd)
-endif()

 # Add the tests: one test per command-line-variable.
-set(TEST_OPTIONS_VARS
-    EX1_ARGS_W EX1_ARGS_P EX2_ARGS EX3_ARGS EX4_ARGS EX4_HYB_ARGS
-    EX5_BDDC_LB_ARGS EX5_BDDC_GB_ARGS EX5_FSPL_ARGS EX6_ARGS EX6_NONOVL_ARGS
-    EX9_E_ARGS EX9_ES_ARGS EX9_IS_ARGS EX10_ARGS)
-if (MFEM_USE_SLEPC)
-    list(APPEND TEST_OPTIONS_VARS EX11_ARGS_SINV EX11_ARGS_LOBPCG EX11_ARGS_GD)
-endif()
-
-foreach(TEST_OPTIONS_VAR ${TEST_OPTIONS_VARS})
+foreach(TEST_OPTIONS_VAR
+        EX1_ARGS_W EX1_ARGS_P EX2_ARGS EX3_ARGS EX4_ARGS EX4_HYB_ARGS
+        EX5_BDDC_LB_ARGS EX5_BDDC_GB_ARGS EX5_FSPL_ARGS EX6_ARGS EX6_NONOVL_ARGS
+        EX9_E_ARGS EX9_ES_ARGS EX9_IS_ARGS EX10_ARGS)
  string(REGEX REPLACE "^(.+)_ARGS" "\\1" TEST_NAME_UC ${TEST_OPTIONS_VAR})
  string(REGEX REPLACE "^([^_]+)" "\\1P" TEST_NAME_UC ${TEST_NAME_UC})
  string(TOLOWER ${TEST_NAME_UC} TEST_NAME_FULL)
@@ -1,440 +0,0 @@
-//                       MFEM Example 11 - Parallel Version
-//                              PETSc Modification
-//
-// Compile with: make ex11p
-//
-// Sample runs:  mpirun -np 4 ex11p -m ../../data/star.mesh
-//               mpirun -np 4 ex11p -m ../../data/star.mesh --slepcopts rc_ex11p_lobpcg
-//               mpirun -np 4 ex11p -m ../../data/star.mesh --slepcopts rc_ex11p_gd
-//
-// Description:  This example code demonstrates the use of MFEM to solve the
-//               eigenvalue problem -Delta u = lambda u with homogeneous
-//               Dirichlet boundary conditions.
-//
-//               We compute a number of the lowest eigenmodes by discretizing
-//               the Laplacian and Mass operators using a FE space of the
-//               specified order, or an isoparametric/isogeometric space if
-//               order < 1 (quadratic for quadratic curvilinear mesh, NURBS for
-//               NURBS mesh, etc.)
-//
-//               The example demonstrates the use of the SLEPc eigensolver as an
-//               alternative to the LOBPCG eigenvalue solver. The shift and
-//               invert spectral transformation is used to help the convergence
-//               to the smaller eigenvalues. Alternative solver parameters can
-//               be passed in a file with "-slepcopts".
-//
-//               Reusing a single GLVis visualization window for multiple
-//               eigenfunctions is also illustrated.
-//
-//               We recommend viewing Example 1 before viewing this example.
-
-#include "mfem.hpp"
-#include <fstream>
-#include <iostream>
-
-#ifndef MFEM_USE_SLEPC
-#error This examples requires that MFEM is build with MFEM_USE_SLEPC=YES
-#endif
-
-using namespace std;
-using namespace mfem;
-
-int main(int argc, char *argv[])
-{
-   // 1. Initialize MPI.
-   int num_procs, myid;
-   MPI_Init(&argc, &argv);
-   MPI_Comm_size(MPI_COMM_WORLD, &num_procs);
-   MPI_Comm_rank(MPI_COMM_WORLD, &myid);
-
-   // 2. Parse command-line options.
-   const char *mesh_file = "../../data/star.mesh";
-   int ser_ref_levels = 2;
-   int par_ref_levels = 1;
-   int order = 1;
-   int nev = 5;
-   int seed = 75;
-   bool slu_solver  = false;
-   bool sp_solver = false;
-   bool visualization = 1;
-   bool use_slepc = true;
-   const char *slepcrc_file = "";
-
-   OptionsParser args(argc, argv);
-   args.AddOption(&mesh_file, "-m", "--mesh",
-                  "Mesh file to use.");
-   args.AddOption(&ser_ref_levels, "-rs", "--refine-serial",
-                  "Number of times to refine the mesh uniformly in serial.");
-   args.AddOption(&par_ref_levels, "-rp", "--refine-parallel",
-                  "Number of times to refine the mesh uniformly in parallel.");
-   args.AddOption(&order, "-o", "--order",
-                  "Finite element order (polynomial degree) or -1 for"
-                  " isoparametric space.");
-   args.AddOption(&nev, "-n", "--num-eigs",
-                  "Number of desired eigenmodes.");
-   args.AddOption(&seed, "-s", "--seed",
-                  "Random seed used to initialize LOBPCG.");
-#ifdef MFEM_USE_SUPERLU
-   args.AddOption(&slu_solver, "-slu", "--superlu", "-no-slu",
-                  "--no-superlu", "Use the SuperLU Solver.");
-#endif
-#ifdef MFEM_USE_STRUMPACK
-   args.AddOption(&sp_solver, "-sp", "--strumpack", "-no-sp",
-                  "--no-strumpack", "Use the STRUMPACK Solver.");
-#endif
-   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
-                  "--no-visualization",
-                  "Enable or disable GLVis visualization.");
-   args.AddOption(&use_slepc, "-useslepc","--useslepc","-no-slepc",
-                  "--no-slepc","Use or not SLEPc to solve the eigenvalue problem");
-   args.AddOption(&slepcrc_file, "-slepcopts", "--slepcopts",
-                  "SlepcOptions file to use.");
-   args.Parse();
-   if (slu_solver && sp_solver)
-   {
-      if (myid == 0)
-         cout << "WARNING: Both SuperLU and STRUMPACK have been selected,"
-              << " please choose either one." << endl
-              << "         Defaulting to SuperLU." << endl;
-      sp_solver = false;
-   }
-   // The command line options are also passed to the STRUMPACK
-   // solver. So do not exit if some options are not recognized.
-   if (!sp_solver)
-   {
-      if (!args.Good())
-      {
-         if (myid == 0)
-         {
-            args.PrintUsage(cout);
-         }
-         MPI_Finalize();
-         return 1;
-      }
-   }
-   if (myid == 0)
-   {
-      args.PrintOptions(cout);
-   }
-
-   // 2b. We initialize SLEPc. This internally initializes PETSc as well.
-   MFEMInitializeSlepc(NULL,NULL,slepcrc_file,NULL);
-
-   // 3. Read the (serial) mesh from the given mesh file on all processors. We
-   //    can handle triangular, quadrilateral, tetrahedral, hexahedral, surface
-   //    and volume meshes with the same code.
-   Mesh *mesh = new Mesh(mesh_file, 1, 1);
-   int dim = mesh->Dimension();
-
-   // 4. Refine the serial mesh on all processors to increase the resolution. In
-   //    this example we do 'ref_levels' of uniform refinement (2 by default, or
-   //    specified on the command line with -rs).
-   for (int lev = 0; lev < ser_ref_levels; lev++)
-   {
-      mesh->UniformRefinement();
-   }
-
-   // 5. Define a parallel mesh by a partitioning of the serial mesh. Refine
-   //    this mesh further in parallel to increase the resolution (1 time by
-   //    default, or specified on the command line with -rp). Once the parallel
-   //    mesh is defined, the serial mesh can be deleted.
-   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
-   delete mesh;
-   for (int lev = 0; lev < par_ref_levels; lev++)
-   {
-      pmesh->UniformRefinement();
-   }
-
-   // 6. Define a parallel finite element space on the parallel mesh. Here we
-   //    use continuous Lagrange finite elements of the specified order. If
-   //    order < 1, we instead use an isoparametric/isogeometric space.
-   FiniteElementCollection *fec;
-   if (order > 0)
-   {
-      fec = new H1_FECollection(order, dim);
-   }
-   else if (pmesh->GetNodes())
-   {
-      fec = pmesh->GetNodes()->OwnFEC();
-   }
-   else
-   {
-      fec = new H1_FECollection(order = 1, dim);
-   }
-   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
-   HYPRE_Int size = fespace->GlobalTrueVSize();
-   if (myid == 0)
-   {
-      cout << "Number of unknowns: " << size << endl;
-   }
-
-   // 7. Set up the parallel bilinear forms a(.,.) and m(.,.) on the finite
-   //    element space. The first corresponds to the Laplacian operator -Delta,
-   //    while the second is a simple mass matrix needed on the right hand side
-   //    of the generalized eigenvalue problem below. The boundary conditions
-   //    are implemented by elimination with special values on the diagonal to
-   //    shift the Dirichlet eigenvalues out of the computational range. After
-   //    serial and parallel assembly we extract the corresponding parallel
-   //    matrices A and M.
-   ConstantCoefficient one(1.0);
-   Array<int> ess_bdr;
-   if (pmesh->bdr_attributes.Size())
-   {
-      ess_bdr.SetSize(pmesh->bdr_attributes.Max());
-      ess_bdr = 1;
-   }
-
-   ParBilinearForm *a = new ParBilinearForm(fespace);
-   a->AddDomainIntegrator(new DiffusionIntegrator(one));
-   if (pmesh->bdr_attributes.Size() == 0)
-   {
-      // Add a mass term if the mesh has no boundary, e.g. periodic mesh or
-      // closed surface.
-      a->AddDomainIntegrator(new MassIntegrator(one));
-   }
-   a->Assemble();
-   a->EliminateEssentialBCDiag(ess_bdr, 1.0);
-   a->Finalize();
-
-   ParBilinearForm *m = new ParBilinearForm(fespace);
-   m->AddDomainIntegrator(new MassIntegrator(one));
-   m->Assemble();
-   // shift the eigenvalue corresponding to eliminated dofs to a large value
-   m->EliminateEssentialBCDiag(ess_bdr, numeric_limits<double>::min());
-   m->Finalize();
-
-   PetscParMatrix *pA = NULL, *pM = NULL;
-   HypreParMatrix *A = NULL, *M = NULL;
-   Operator::Type tid =
-      !use_slepc ? Operator::Hypre_ParCSR : Operator::PETSC_MATAIJ;
-   OperatorHandle Ah(tid), Mh(tid);
-
-   a->ParallelAssemble(Ah);
-   if (!use_slepc) { Ah.Get(A); }
-   else { Ah.Get(pA); }
-   Ah.SetOperatorOwner(false);
-
-   m->ParallelAssemble(Mh);
-   if (!use_slepc) {Mh.Get(M); }
-   else {Mh.Get(pM); }
-   Mh.SetOperatorOwner(false);
-
-#if defined(MFEM_USE_SUPERLU) || defined(MFEM_USE_STRUMPACK)
-   Operator * Arow = NULL;
-#ifdef MFEM_USE_SUPERLU
-   if (slu_solver)
-   {
-      Arow = new SuperLURowLocMatrix(*A);
-   }
-#endif
-#ifdef MFEM_USE_STRUMPACK
-   if (sp_solver)
-   {
-      Arow = new STRUMPACKRowLocMatrix(*A);
-   }
-#endif
-#endif
-
-   delete a;
-   delete m;
-
-   // 8. Define and configure the LOBPCG eigensolver and the BoomerAMG
-   //    preconditioner for A to be used within the solver. Set the matrices
-   //    which define the generalized eigenproblem A x = lambda M x.
-   Solver * precond = NULL;
-   if (!use_slepc)
-   {
-      if (!slu_solver && !sp_solver)
-      {
-         HypreBoomerAMG * amg = new HypreBoomerAMG(*A);
-         amg->SetPrintLevel(0);
-         precond = amg;
-      }
-      else
-      {
-#ifdef MFEM_USE_SUPERLU
-         if (slu_solver)
-         {
-            SuperLUSolver * superlu = new SuperLUSolver(MPI_COMM_WORLD);
-            superlu->SetPrintStatistics(false);
-            superlu->SetSymmetricPattern(true);
-            superlu->SetColumnPermutation(superlu::PARMETIS);
-            superlu->SetOperator(*Arow);
-            precond = superlu;
-         }
-#endif
-#ifdef MFEM_USE_STRUMPACK
-         if (sp_solver)
-         {
-            STRUMPACKSolver * strumpack = new STRUMPACKSolver(argc, argv, MPI_COMM_WORLD);
-            strumpack->SetPrintFactorStatistics(true);
-            strumpack->SetPrintSolveStatistics(false);
-            strumpack->SetKrylovSolver(strumpack::KrylovSolver::DIRECT);
-            strumpack->SetReorderingStrategy(strumpack::ReorderingStrategy::METIS);
-            strumpack->DisableMatching();
-            strumpack->SetOperator(*Arow);
-            strumpack->SetFromCommandLine();
-            precond = strumpack;
-         }
-#endif
-      }
-   }
-
-   HypreLOBPCG * lobpcg = NULL;
-   SlepcEigenSolver * slepc = NULL;
-   if (!use_slepc)
-   {
-
-      lobpcg = new HypreLOBPCG(MPI_COMM_WORLD);
-      lobpcg->SetNumModes(nev);
-      lobpcg->SetRandomSeed(seed);
-      lobpcg->SetPreconditioner(*precond);
-      lobpcg->SetMaxIter(200);
-      lobpcg->SetTol(1e-8);
-      lobpcg->SetPrecondUsageMode(1);
-      lobpcg->SetPrintLevel(1);
-      lobpcg->SetMassMatrix(*M);
-      lobpcg->SetOperator(*A);
-   }
-   else
-   {
-      slepc = new SlepcEigenSolver(MPI_COMM_WORLD);
-      slepc->SetNumModes(nev);
-      slepc->SetWhichEigenpairs(SlepcEigenSolver::TARGET_REAL);
-      slepc->SetTarget(0.0);
-      slepc->SetSpectralTransformation(SlepcEigenSolver::SHIFT_INVERT);
-      slepc->SetOperators(*pA,*pM);
-   }
-
-   // 9. Compute the eigenmodes and extract the array of eigenvalues. Define a
-   //    parallel grid function to represent each of the eigenmodes returned by
-   //    the solver.
-   Array<double> eigenvalues;
-   if (!use_slepc)
-   {
-      lobpcg->Solve();
-      lobpcg->GetEigenvalues(eigenvalues);
-   }
-   else
-   {
-      slepc->Solve();
-      eigenvalues.SetSize(nev);
-      for (int i=0; i<nev; i++)
-      {
-         slepc->GetEigenvalue(i,eigenvalues[i]);
-      }
-   }
-   Vector temp(fespace->GetTrueVSize());
-   ParGridFunction x(fespace);
-
-   // 10. Save the refined mesh and the modes in parallel. This output can be
-   //     viewed later using GLVis: "glvis -np <np> -m mesh -g mode".
-   {
-      ostringstream mesh_name, mode_name;
-      mesh_name << "mesh." << setfill('0') << setw(6) << myid;
-
-      ofstream mesh_ofs(mesh_name.str().c_str());
-      mesh_ofs.precision(8);
-      pmesh->Print(mesh_ofs);
-
-      for (int i=0; i<nev; i++)
-      {
-         // convert eigenvector from HypreParVector to ParGridFunction
-         if (!use_slepc)
-         {
-            x = lobpcg->GetEigenvector(i);
-         }
-         else
-         {
-            slepc->GetEigenvector(i,temp);
-            x.Distribute(temp);
-
-         }
-
-         mode_name << "mode_" << setfill('0') << setw(2) << i << "."
-                   << setfill('0') << setw(6) << myid;
-
-         ofstream mode_ofs(mode_name.str().c_str());
-         mode_ofs.precision(8);
-         x.Save(mode_ofs);
-         mode_name.str("");
-      }
-   }
-
-   // 11. Send the solution by socket to a GLVis server.
-   if (visualization)
-   {
-      char vishost[] = "localhost";
-      int  visport   = 19916;
-      socketstream mode_sock(vishost, visport);
-      mode_sock.precision(8);
-
-      for (int i=0; i<nev; i++)
-      {
-         if ( myid == 0 )
-         {
-            cout << "Eigenmode " << i+1 << '/' << nev
-                 << ", Lambda = " << eigenvalues[i] << endl;
-         }
-
-         // convert eigenvector from HypreParVector to ParGridFunction
-         if (!use_slepc)
-         {
-            x = lobpcg->GetEigenvector(i);
-         }
-         else
-         {
-            slepc->GetEigenvector(i,temp);
-            x.Distribute(temp);
-         }
-
-         mode_sock << "parallel " << num_procs << " " << myid << "\n"
-                   << "solution\n" << *pmesh << x << flush
-                   << "window_title 'Eigenmode " << i+1 << '/' << nev
-                   << ", Lambda = " << eigenvalues[i] << "'" << endl;
-
-         char c;
-         if (myid == 0)
-         {
-            cout << "press (q)uit or (c)ontinue --> " << flush;
-            cin >> c;
-         }
-         MPI_Bcast(&c, 1, MPI_CHAR, 0, MPI_COMM_WORLD);
-
-         if (c != 'c')
-         {
-            break;
-         }
-      }
-      mode_sock.close();
-   }
-
-   // 12. Free the used memory.
-   if (!use_slepc)
-   {
-      delete lobpcg;
-   }
-   else
-   {
-      delete slepc;
-   }
-   delete precond;
-   delete M;
-   delete A;
-#if defined(MFEM_USE_SUPERLU) || defined(MFEM_USE_STRUMPACK)
-   delete Arow;
-#endif
-
-   delete fespace;
-   if (order > 0)
-   {
-      delete fec;
-   }
-   delete pmesh;
-
-   // We finalize SLEPc
-   MFEMFinalizeSlepc();
-   MPI_Finalize();
-
-   return 0;
-}
@@ -23,9 +23,6 @@ MFEM_LIB_FILE = mfem_is_not_built

 SEQ_EXAMPLES =
 PAR_EXAMPLES = ex1p ex2p ex3p ex4p ex5p ex6p ex9p ex10p
-ifeq ($(MFEM_USE_SLEPC),YES)
-   PAR_EXAMPLES += ex11p
-endif
 ifeq ($(MFEM_USE_MPI),NO)
   EXAMPLES = $(SEQ_EXAMPLES)
 else
@@ -90,9 +87,6 @@ EX10_ARGS        := -m ../../data/beam-quad.mesh --usepetsc --petscopts rc_ex10p
 EX10_MF_ARGS     := -m ../../data/beam-quad.mesh --usepetsc --petscopts rc_ex10p_mf -tf 6 -s 3 -rs 0 -dt 3
 EX10_MFOP_ARGS   := -m ../../data/beam-quad.mesh --usepetsc --petscopts rc_ex10p_mfop -tf 6 -s 3 -rs 0 -dt 3
 EX10_JFNK_ARGS   := -m ../../data/beam-quad.mesh --usepetsc --petscopts rc_ex10p_jfnk --jfnk -tf 6 -s 3 -rs 0 -dt 3
-EX11_ARGS_SINV   := -m ../../data/star.mesh --useslepc
-EX11_ARGS_LOBPCG := -m ../../data/star.mesh --useslepc --slepcopts rc_ex11p_lobpcg
-EX11_ARGS_GD     := -m ../../data/star.mesh --useslepc --slepcopts rc_ex11p_gd

 ex1p-test-par: ex1p
 	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX1_ARGS_W))
@@ -120,12 +114,6 @@ ex10p-test-par: ex10p
 	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX10_MF_ARGS))
 	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX10_MFOP_ARGS))
 	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX10_JFNK_ARGS))
-ifeq ($(MFEM_USE_SLEPC),YES)
-ex11p-test-par: ex11p
-	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX11_ARGS_SINV))
-	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX11_ARGS_LOBPCG))
-	@$(call mfem-test,$<, $(RUN_MPI), $(TESTNAME),$(EX11_ARGS_GD))
-endif

 # Testing: "test" target and mfem-test* variables are defined in config/test.mk

@@ -1,6 +0,0 @@
-# Options for the eigenvalue solver
-eps_view
-eps_converged_reason
-eps_type gd
-# Options for the spectral transform
-st_type precond
@@ -1,11 +0,0 @@
-# Options for the eigenvalue solver
-eps_monitor
-eps_converged_reason
-eps_view_values
-eps_type lobpcg
-eps_gen_hermitian
-eps_smallest_real
-eps_lobpcg_blocksize 5
-# Options for the spectral transform
-st_type precond
-st_pc_type gamg
@@ -32,13 +32,6 @@
 //               is used for the Finite Element order and "-go" is used for the
 //               geometry order. Note that they can be used independently, i.e.
 //               "-o 8 -go 3" solves for 8th order FE on a third order geometry.
-//
-// NOTE:         Model/Mesh files for this example are in the (large) data file
-//               repository of MFEM here https://github.com/mfem/data under the
-//               folder named "pumi", which consists of the following sub-folders:
-//               a) geom -->  model files
-//               b) parallel --> parallel pumi mesh files
-//               c) serial --> serial pumi mesh files

 #include "mfem.hpp"
 #include <fstream>
@@ -36,14 +36,6 @@
 //               option "-o" is used for the Finite Element order and "-go" for
 //               the geometry order. Note that they can be used independently:
 //               "-o 8 -go 3" solves for 8th order FE on third order geometry.
-//
-// NOTE:         Model/Mesh files for this example are in the (large) data file
-//               repository of MFEM here https://github.com/mfem/data under the
-//               folder named "pumi", which consists of the following sub-folders:
-//               a) geom -->  model files
-//               b) parallel --> parallel pumi mesh files
-//               c) serial --> serial pumi mesh files
-

 #include "mfem.hpp"
 #include <fstream>
@@ -43,14 +43,6 @@
 //               also illustrated.
 //
 //               We recommend viewing Example 1 before viewing this example.
-//
-// NOTE:         Model/Mesh files for this example are in the (large) data file
-//               repository of MFEM here https://github.com/mfem/data under the
-//               folder named "pumi", which consists of the following sub-folders:
-//               a) geom -->  model files
-//               b) parallel --> parallel pumi mesh files
-//               c) serial --> serial pumi mesh files
-

 #include "mfem.hpp"
 #include <fstream>
@@ -1,7 +1,7 @@
 //                       MFEM Example 6 - Parallel Version
 //                              PUMI Modification
 //
-// Compile with: make ex6p
+// Compile with: make ex1p
 //
 // Sample runs:  mpirun -np 8 ex6p
 //
@@ -18,13 +18,6 @@
 //               is added to modify the "adapt_ratio" which is the fraction of
 //               allowable error that scales the output size field of the error
 //               estimator.
-//
-// NOTE:         Model/Mesh files for this example are in the (large) data file
-//               repository of MFEM here https://github.com/mfem/data under the
-//               folder named "pumi", which consists of the following sub-folders:
-//               a) geom -->  model files
-//               b) parallel --> parallel pumi mesh files
-//               c) serial --> serial pumi mesh files

 #include "mfem.hpp"
 #include <fstream>
@@ -339,6 +332,7 @@ int main(int argc, char *argv[])

      apf::destroyField(Tmag_field);
      apf::destroyField(ipfield);
+      apf::destroyNumbering(pumi_mesh->findNumbering("LocalVertexNumbering"));

      // 18. Perform MesAdapt.
      ma::Input* erinput = ma::configure(pumi_mesh, sizefield);
@@ -13,20 +13,13 @@ set(SRCS
  bilinearform.cpp
  bilinearform_ext.cpp
  bilininteg.cpp
-  bilininteg_convection_pa.cpp
-  bilininteg_convection_ea.cpp
-  bilininteg_dgtrace_pa.cpp
-  bilininteg_dgtrace_ea.cpp
-  bilininteg_diffusion_pa.cpp
-  bilininteg_diffusion_ea.cpp
+  bilininteg_convection.cpp
+  bilininteg_dgtrace.cpp
+  bilininteg_diffusion.cpp
  bilininteg_divergence.cpp
  bilininteg_hcurl.cpp
-  bilininteg_hdiv.cpp
-  bilininteg_vectorfe.cpp
  bilininteg_gradient.cpp
-  bilininteg_mass_pa.cpp
-  bilininteg_mass_ea.cpp
-  bilininteg_transpose_ea.cpp
+  bilininteg_mass.cpp
  bilininteg_vecdiffusion.cpp
  bilininteg_vecmass.cpp
  coefficient.cpp
@@ -98,7 +91,6 @@ set(HDRS
  tmop.hpp
  tmop_tools.hpp
  gslib.hpp
-  adnonlininteg.hpp
  transfer.hpp
  )

@@ -15,8 +15,6 @@

 #include "adios2datacollection.hpp"

-#ifdef MFEM_USE_ADIOS2
-
 namespace mfem
 {

@@ -89,4 +87,4 @@ noexcept

 } //end namespace mfem

-#endif // MFEM_USE_ADIOS2
+
@@ -17,9 +17,6 @@
 #define MFEM_ADIOS2DATACOLLECTION

 #include "../config/config.hpp"
-
-#ifdef MFEM_USE_ADIOS2
-
 #include "../general/adios2stream.hpp"
 #include "datacollection.hpp"

@@ -88,6 +85,4 @@ private:

 }  // namespace mfem

-#endif // MFEM_USE_ADIOS2
-
 #endif /* MFEM_ADIOS2DATACOLLECTION */
@@ -1,402 +0,0 @@
-// Copyright (c) 2010, Lawrence Livermore National Security, LLC. Produced at
-// the Lawrence Livermore National Laboratory. LLNL-CODE-443211. All Rights
-// reserved. See file COPYRIGHT for details.
-//
-// This file is part of the MFEM library. For more information and source code
-// availability see http://mfem.org.
-//
-// MFEM is free software; you can redistribute it and/or modify it under the
-// terms of the GNU Lesser General Public License (as published by the Free
-// Software Foundation) version 2.1 dated February 1999.
-
-
-
-#ifndef MFEM_ADNONLININTEG
-#define MFEM_ADNONLININTEG
-
-#include "../config/config.hpp"
-#include "fe.hpp"
-#include "coefficient.hpp"
-#include "fespace.hpp"
-#include "nonlininteg.hpp"
-#include "../linalg/tadvector.hpp"
-#include "../linalg/taddensemat.hpp"
-#include "../linalg/fdual.hpp"
-
-#if defined MFEM_USE_ADEPT
-#include <adept.h>
-#elif defined  MFEM_USE_FADBADPP
-#include <fadiff.h>
-#include <badiff.h>
-#endif
-
-//define Forward AD mode
-//#define MFEM_USE_ADFORWARD
-
-namespace mfem
-{
-
-// m - dimension of the residual vector
-// the Jacobian will have dimensions [m,length(uu)]
-template<template <typename, typename> class CTD, int m>
-class ADQFunctionTJ
-{
-protected:
-#ifdef MFEM_USE_ADEPT
-   adept::Stack  m_stack;
-#endif
-
-public:
-#if defined MFEM_USE_ADEPT
-   typedef adept::adouble             ADFType;
-   typedef TADVector<ADFType>         ADFVector;
-   typedef TADDenseMatrix<ADFType>    ADFDenseMatrix;
-#elif defined MFEM_USE_FADBADPP
-#ifdef MFEM_USE_ADFORWARD
-   typedef fadbad::F<double>         ADFType;
-   typedef TADVector<ADFType>         ADFVector;
-   typedef TADDenseMatrix<ADFType>    ADFDenseMatrix;
-#else
-   typedef fadbad::B<double>         ADFType;
-   typedef TADVector<ADFType>         ADFVector;
-   typedef TADDenseMatrix<ADFType>    ADFDenseMatrix;
-#endif
-#else
-   typedef mfem::ad::FDual<double>    ADFType;
-   typedef TADVector<ADFType>         ADFVector;
-   typedef TADDenseMatrix<ADFType>    ADFDenseMatrix;
-#endif
-
-#ifdef MFEM_USE_ADEPT
-   ADQFunctionTJ():m_stack(false)  {}
-#else
-   ADQFunctionTJ() {}
-#endif
-
-   ~ADQFunctionTJ() {}
-
-   double QFunction(const mfem::Vector& vparam, mfem::Vector& uu)
-   {
-      CTD<double,mfem::Vector> func;
-      return func(vparam,uu);
-   }
-
-   void   QFunctionDU(const mfem::Vector& vparam, ADFVector& uu,
-                      ADFVector& rr)
-   {
-      CTD<ADFType,ADFVector> func;
-      func(vparam,uu,rr);
-   }
-
-   void  QFunctionAU(const Vector &vparam, mfem::Vector &uu,
-                     mfem::Vector &rr)
-   {
-      //the result is computed automaticaly by differentiating
-      //QFunction with respect to uu
-      CTD<ADFType,ADFVector> func;
-      int n=uu.Size();
-      rr.SetSize(n);
-
-#if defined MFEM_USE_ADEPT
-      //use ADEPT package
-      adept::Stack* p_stack=adept::active_stack();
-      p_stack->deactivate();
-
-      m_stack.activate();
-      {
-         ADFVector aduu(uu);
-         ADFType   rez;
-         m_stack.new_recording();
-         rez=func(vparam,aduu);
-         m_stack.independent(aduu.GetData(), n);//independent variables
-         m_stack.dependent(&rez, 1);//dependent variables
-         m_stack.jacobian(rr.GetData());
-      }
-      m_stack.deactivate();
-#elif  defined MFEM_USE_FADBADPP
-      //use FADBAD++
-#ifdef MFEM_USE_ADFORWARD
-      {
-         ADFVector aduu(uu);
-         ADFType   rez;
-         for (int ii=0; ii<n; ii++)
-         {
-            aduu[ii].diff(ii,n);
-         }
-         rez=func(vparam,aduu);
-         for (int ii=0; ii<n; ii++)
-         {
-            rr[ii]=rez.d(ii);
-         }
-      }
-#else
-      {
-         ADFVector aduu(uu);
-         ADFType   rez;
-         rez=func(vparam,aduu);
-         rez.diff(0,1);
-         for (int ii=0; ii<n; ii++)
-         {
-            rr[ii]=aduu[ii].d(0);
-         }
-      }
-#endif
-#else
-      //use native AD package
-      {
-         ADFVector aduu(uu); //all dual numbers are initialized to zero
-         ADFType rez;
-
-         for (int ii=0; ii<n; ii++)
-         {
-            aduu[ii].dual(1.0);
-            rez=func(vparam,aduu);
-            rr[ii]=rez.dual();
-            aduu[ii].dual(0.0);
-         }
-      }
-#endif
-
-   }
-
-   void QFunctionDU(const mfem::Vector& vparam, mfem::Vector& uu,
-                    mfem::Vector& rr)
-   {
-      CTD<double,mfem::Vector> func;
-      func(vparam,uu,rr);
-   }
-
-   void QFunctionDD(const mfem::Vector& vparam, mfem::Vector& uu,
-                    mfem::DenseMatrix& jac)
-   {
-#if defined MFEM_USE_ADEPT
-      //use ADEPT package
-      adept::Stack* p_stack=adept::active_stack();
-      p_stack->deactivate();
-
-      int n=uu.Size();
-      jac.SetSize(m,n);
-      jac=0.0;
-      m_stack.activate();
-      {
-         ADFVector aduu(uu);
-         ADFVector rr(m); //residual vector
-         m_stack.new_recording();
-         QFunctionDU(vparam,aduu,rr);
-         m_stack.independent(aduu.GetData(), n);//independent variables
-         m_stack.dependent(rr.GetData(), m);//dependent variables
-         m_stack.jacobian(jac.Data());
-      }
-      m_stack.deactivate();
-#elif defined MFEM_USE_FADBADPP
-      //use FADBAD++
-#ifdef MFEM_USE_ADFORWARD
-      int n=uu.Size();
-      jac.SetSize(m,n);
-      jac=0.0;
-      {
-         ADFVector aduu(uu);
-         ADFVector rr(m);
-
-         for (int ii=0; ii<n; ii++)
-         {
-            aduu[ii].diff(ii,n);
-         }
-         QFunctionDU(vparam,aduu,rr);
-         for (int ii=0; ii<n; ii++)
-         {
-            for (int jj=0; jj<m; jj++)
-            {
-               jac(jj,ii)=rr[jj].d(ii);
-            }
-         }
-      }
-#else
-      int n=uu.Size();
-      jac.SetSize(m,n);
-      jac=0.0;
-      {
-         ADFVector aduu(uu);
-         ADFVector rr(m);
-         QFunctionDU(vparam,aduu,rr);
-         for (int ii=0; ii<m; ii++)
-         {
-            rr[ii].diff(ii,m);
-         }
-         for (int ii=0; ii<n; ii++)
-         {
-            for (int jj=0; jj<m; jj++)
-            {
-               jac(jj,ii)=aduu[ii].d(jj);
-            }
-         }
-
-      }
-#endif
-#else
-      //use native AD package
-      int n=uu.Size();
-      jac.SetSize(m,n);
-      jac=0.0;
-      {
-         ADFVector aduu(uu); //all dual numbers are initialized to zero
-         ADFVector rr(m);
-
-         for (int ii=0; ii<n; ii++)
-         {
-            aduu[ii].dual(1.0);
-            QFunctionDU(vparam,aduu,rr);
-            for (int jj=0; jj<m; jj++)
-            {
-               jac(jj,ii)=rr[jj].dual();
-            }
-            aduu[ii].dual(0.0);
-         }
-      }
-#endif
-   }
-
-};
-
-
-
-//template class for differentiation; the function
-//for differentiation is supplied as a functor
-//the operator()(scalar,vector) defines the actual function
-template<template <typename, typename> class CTD>
-class ADQFunctionTH
-{
-public:
-#if defined MFEM_USE_FADBADPP
-   typedef fadbad::B<double>             ADFType;
-   typedef TADVector<ADFType>            ADFVector;
-   typedef TADDenseMatrix<ADFType>       ADFDenseMatrix;
-
-   typedef fadbad::B<fadbad::F<double>>    ADSType;
-   typedef TADVector<ADSType>              ADSVector;
-   typedef TADDenseMatrix<ADSType>         ADSDenseMatrix;
-#else
-   typedef mfem::ad::FDual<double>    ADFType;
-   typedef TADVector<ADFType>         ADFVector;
-   typedef TADDenseMatrix<ADFType>    ADFDenseMatrix;
-
-   typedef mfem::ad::FDual<ADFType>   ADSType;
-   typedef TADVector<ADSType>         ADSVector;
-   typedef TADDenseMatrix<ADSType>    ADSDenseMatrix;
-#endif
-
-   ADQFunctionTH() {}
-
-   ~ADQFunctionTH() {}
-
-   double QFunction(const mfem::Vector& vparam, mfem::Vector& uu)
-   {
-      CTD<double, mfem::Vector> tf;
-      return tf(vparam, uu);
-   }
-
-   ADFType QFunction(const mfem::Vector& vparam, ADFVector& uu)
-   {
-      CTD<ADFType,ADFVector> tf;
-      return tf(vparam, uu);
-   }
-
-   ADSType QFunction(const mfem::Vector &vparam, ADSVector& uu)
-   {
-      CTD<ADSType,ADSVector> tf;
-      return tf(vparam, uu);
-   }
-
-   void QFunctionDU(const mfem::Vector& vparam, mfem::Vector& uu,
-                    mfem::Vector& rr)
-   {
-#if defined MFEM_USE_FADBADPP
-      int n=uu.Size();
-      rr.SetSize(n);
-      ADFVector aduu(uu);
-      ADFType   rez;
-      rez=QFunction(vparam,aduu);
-      rez.diff(0,1);
-      for (int ii=0; ii<n; ii++)
-      {
-         rr[ii]=aduu[ii].d(0);
-      }
-#else
-      int n=uu.Size();
-      rr.SetSize(n);
-      ADFVector aduu(uu);
-      ADFType   rez;
-      for (int ii=0; ii<n; ii++)
-      {
-         aduu[ii].dual(1.0);
-         rez=QFunction(vparam,aduu);
-         rr[ii]=rez.dual();
-         aduu[ii].dual(0.0);
-      }
-#endif
-   }
-
-   void QFunctionDD(const mfem::Vector& vparam, const mfem::Vector& uu,
-                    mfem::DenseMatrix& jac)
-   {
-#if defined MFEM_USE_FADBADPP
-      int n=uu.Size();
-      jac.SetSize(n);
-      jac=0.0;
-      {
-         ADSVector aduu(n);
-         for (int ii = 0;  ii < n ; ii++)
-         {
-            aduu[ii]=uu[ii];
-            aduu[ii].x().diff(ii,n);
-         }
-         ADSType rez=QFunction(vparam,aduu);
-         rez.diff(0,1);
-         for (int ii = 0; ii < n ; ii++)
-         {
-            for (int jj=0; jj<ii; jj++)
-            {
-               jac(ii,jj)=aduu[ii].d(0).d(jj);
-               jac(jj,ii)=aduu[jj].d(0).d(ii);
-            }
-            jac(ii,ii)=aduu[ii].d(0).d(ii);
-         }
-      }
-#else
-      int n=uu.Size();
-      jac.SetSize(n);
-      jac=0.0;
-      {
-         ADSVector aduu(n);
-         for (int ii = 0;  ii < n ; ii++)
-         {
-            aduu[ii].real(ADFType(uu[ii],0.0));
-            aduu[ii].dual(ADFType(0.0,0.0));
-         }
-
-         for (int ii = 0; ii < n ; ii++)
-         {
-            aduu[ii].real(ADFType(uu[ii],1.0));
-            for (int jj=0; jj<(ii+1); jj++)
-            {
-               aduu[jj].dual(ADFType(1.0,0.0));
-               ADSType rez=QFunction(vparam,aduu);
-               jac(ii,jj)=rez.dual().dual();
-               jac(jj,ii)=rez.dual().dual();
-               aduu[jj].dual(ADFType(0.0,0.0));
-            }
-            aduu[ii].real(ADFType(uu[ii],0.0));
-         }
-      }
-#endif
-   }
-
-
-};// end template ADFunctionTH
-
-
-
-}
-
-#endif
-
@@ -76,7 +76,7 @@ BilinearForm::BilinearForm(FiniteElementSpace * f)
   precompute_sparsity = 0;
   diag_policy = DIAG_KEEP;

-   assembly = AssemblyLevel::LEGACYFULL;
+   assembly = AssemblyLevel::FULL;
   batch = 1;
   ext = NULL;
 }
@@ -94,7 +94,7 @@ BilinearForm::BilinearForm (FiniteElementSpace * f, BilinearForm * bf, int ps)
   precompute_sparsity = ps;
   diag_policy = DIAG_KEEP;

-   assembly = AssemblyLevel::LEGACYFULL;
+   assembly = AssemblyLevel::FULL;
   batch = 1;
   ext = NULL;

@@ -121,13 +121,13 @@ void BilinearForm::SetAssemblyLevel(AssemblyLevel assembly_level)
   assembly = assembly_level;
   switch (assembly)
   {
-      case AssemblyLevel::LEGACYFULL:
-         break;
      case AssemblyLevel::FULL:
-         ext = new FABilinearFormExtension(this);
+         // ext = new FABilinearFormExtension(this);
+         // Use the original BilinearForm implementation for now
         break;
      case AssemblyLevel::ELEMENT:
-         ext = new EABilinearFormExtension(this);
+         mfem_error("Element assembly not supported yet... stay tuned!");
+         // ext = new EABilinearFormExtension(this);
         break;
      case AssemblyLevel::PARTIAL:
         ext = new PABilinearFormExtension(this);
@@ -144,7 +144,7 @@ void BilinearForm::SetAssemblyLevel(AssemblyLevel assembly_level)
 void BilinearForm::EnableStaticCondensation()
 {
   delete static_cond;
-   if (assembly != AssemblyLevel::LEGACYFULL)
+   if (assembly != AssemblyLevel::FULL)
   {
      static_cond = NULL;
      MFEM_WARNING("Static condensation not supported for this assembly level");
@@ -169,7 +169,7 @@ void BilinearForm::EnableHybridization(FiniteElementSpace *constr_space,
                                       const Array<int> &ess_tdof_list)
 {
   delete hybridization;
-   if (assembly != AssemblyLevel::LEGACYFULL)
+   if (assembly != AssemblyLevel::FULL)
   {
      delete constr_integ;
      hybridization = NULL;
@@ -224,7 +224,7 @@ MatrixInverse * BilinearForm::Inverse() const

 void BilinearForm::Finalize (int skip_zeros)
 {
-   if (assembly == AssemblyLevel::LEGACYFULL)
+   if (assembly == AssemblyLevel::FULL)
   {
      if (!static_cond) { mat->Finalize(skip_zeros); }
      if (mat_e) { mat_e->Finalize(skip_zeros); }
@@ -640,7 +640,8 @@ void BilinearForm::AssembleDiagonal(Vector &diag) const
   }
   else
   {
-      mat->GetDiag(diag);
+      MFEM_ABORT("Not implemented. Maybe assemble your bilinear form into a "
+                 "matrix and use SparseMatrix::GetDiag?");
   }
 }

@@ -1083,7 +1084,7 @@ MixedBilinearForm::MixedBilinearForm (FiniteElementSpace *tr_fes,
   mat = NULL;
   mat_e = NULL;
   extern_bfs = 0;
-   assembly = AssemblyLevel::LEGACYFULL;
+   assembly = AssemblyLevel::FULL;
   ext = NULL;
 }

@@ -1108,7 +1109,7 @@ MixedBilinearForm::MixedBilinearForm (FiniteElementSpace *tr_fes,
   bbfi_marker = mbf->bbfi_marker;
   btfbfi_marker = mbf->btfbfi_marker;

-   assembly = AssemblyLevel::LEGACYFULL;
+   assembly = AssemblyLevel::FULL;
   ext = NULL;
 }

@@ -1121,8 +1122,6 @@ void MixedBilinearForm::SetAssemblyLevel(AssemblyLevel assembly_level)
   assembly = assembly_level;
   switch (assembly)
   {
-      case AssemblyLevel::LEGACYFULL:
-         break;
      case AssemblyLevel::FULL:
         // ext = new FAMixedBilinearFormExtension(this);
         // Use the original BilinearForm implementation for now
@@ -1193,7 +1192,7 @@ void MixedBilinearForm::AddMultTranspose(const Vector & x, Vector & y,

 MatrixInverse * MixedBilinearForm::Inverse() const
 {
-   if (assembly != AssemblyLevel::LEGACYFULL)
+   if (assembly != AssemblyLevel::FULL)
   {
      MFEM_WARNING("MixedBilinearForm::Inverse not possible with this assembly level!");
      return NULL;
@@ -1206,7 +1205,7 @@ MatrixInverse * MixedBilinearForm::Inverse() const

 void MixedBilinearForm::Finalize (int skip_zeros)
 {
-   if (assembly == AssemblyLevel::LEGACYFULL)
+   if (assembly == AssemblyLevel::FULL)
   {
      mat -> Finalize (skip_zeros);
   }
@@ -1433,57 +1432,9 @@ void MixedBilinearForm::Assemble (int skip_zeros)
   }
 }

-void MixedBilinearForm::AssembleDiagonal_ADAt(const Vector &D,
-                                              Vector &diag) const
-{
-   if (ext)
-   {
-      MFEM_ASSERT(diag.Size() == test_fes->GetTrueVSize(),
-                  "Vector for holding diagonal has wrong size!");
-      MFEM_ASSERT(D.Size() == trial_fes->GetTrueVSize(),
-                  "Vector for holding diagonal has wrong size!");
-      const Operator *P_trial = trial_fes->GetProlongationMatrix();
-      const Operator *P_test = test_fes->GetProlongationMatrix();
-      if (!IsIdentityProlongation(P_trial))
-      {
-         Vector local_D(P_trial->Height());
-         P_trial->Mult(D, local_D);
-
-         if (!IsIdentityProlongation(P_test))
-         {
-            Vector local_diag(P_test->Height());
-            ext->AssembleDiagonal_ADAt(local_D, local_diag);
-            P_test->MultTranspose(local_diag, diag);
-         }
-         else
-         {
-            ext->AssembleDiagonal_ADAt(local_D, diag);
-         }
-      }
-      else
-      {
-         if (!IsIdentityProlongation(P_test))
-         {
-            Vector local_diag(P_test->Height());
-            ext->AssembleDiagonal_ADAt(D, local_diag);
-            P_test->MultTranspose(local_diag, diag);
-         }
-         else
-         {
-            ext->AssembleDiagonal_ADAt(D, diag);
-         }
-      }
-   }
-   else
-   {
-      MFEM_ABORT("Not implemented. Maybe assemble your bilinear form into a "
-                 "matrix and use SparseMatrix functions?");
-   }
-}
-
 void MixedBilinearForm::ConformingAssemble()
 {
-   if (assembly != AssemblyLevel::LEGACYFULL)
+   if (assembly != AssemblyLevel::FULL)
   {
      MFEM_WARNING("Conforming assemble not supported for this assembly level!");
      return;
@@ -25,15 +25,12 @@
 namespace mfem
 {

-/** @brief Enumeration defining the assembly level for bilinear and nonlinear
-    form classes derived from Operator. */
+/// Enumeration defining the assembly level for bilinear and nonlinear form
+/// classes derived from Operator.
 enum class AssemblyLevel
 {
-   /// Legacy fully assembled form, i.e. a global sparse matrix in MFEM, Hypre
-   /// or PETSC format. This assembly is ALWAYS performed on the host.
-   LEGACYFULL = 0,
-   /// Fully assembled form, i.e. a global sparse matrix in MFEM format. This
-   /// assembly is compatible with device execution.
+   /// Fully assembled form, i.e. a global sparse matrix in MFEM, Hypre or PETSC
+   /// format.
   FULL,
   /// Form assembled at element level, which computes and stores dense element
   /// matrices.
@@ -47,19 +44,15 @@ enum class AssemblyLevel
 };


-/** @brief A "square matrix" operator for the associated FE space and
-    BLFIntegrators The sum of all the BLFIntegrators can be used form the matrix
-    M. This class also supports other assembly levels specified via the
-    SetAssemblyLevel() function. */
+/** Class for bilinear form - "Matrix" with associated FE space and
+    BLFIntegrators. */
 class BilinearForm : public Matrix
 {
 protected:
-   /// Sparse matrix \f$ M \f$ to be associated with the form. Owned.
+   /// Sparse matrix to be associated with the form. Owned.
   SparseMatrix *mat;

-   /** @brief Sparse Matrix \f$ M_e \f$ used to store the eliminations
-        from the b.c.  Owned.
-       \f$ M + M_e = M_{original} \f$ */
+   /// Matrix used to eliminate b.c. Owned.
   SparseMatrix *mat_e;

   /// FE space on which the form lives. Not owned.
@@ -69,12 +62,12 @@ protected:
   AssemblyLevel assembly;
   /// Element batch size used in the form action (1, 8, num_elems, etc.)
   int batch;
-   /** @brief Extension for supporting Full Assembly (FA), Element Assembly (EA),
+   /** Extension for supporting Full Assembly (FA), Element Assembly (EA),
       Partial Assembly (PA), or Matrix Free assembly (MF). */
   BilinearFormExtension *ext;

-   /** @brief Indicates the Mesh::sequence corresponding to the current state of
-       the BilinearForm. */
+   /// Indicates the Mesh::sequence corresponding to the current state of the
+   /// BilinearForm.
   long sequence;

   /** @brief Indicates the BilinearFormIntegrator%s stored in #dbfi, #bbfi,
@@ -122,7 +115,7 @@ protected:
      static_cond = NULL; hybridization = NULL;
      precompute_sparsity = 0;
      diag_policy = DIAG_KEEP;
-      assembly = AssemblyLevel::LEGACYFULL;
+      assembly = AssemblyLevel::FULL;
      batch = 1;
      ext = NULL;
   }
@@ -154,43 +147,35 @@ public:
   /// Get the size of the BilinearForm as a square matrix.
   int Size() const { return height; }

-   /// Set the desired assembly level.
-   /** Valid choices are:
-
-       - AssemblyLevel::FULL  (default)
-       - AssemblyLevel::PARTIAL
-       - AssemblyLevel::ELEMENT
-       - AssemblyLevel::NONE
-
-       This method must be called before assembly. */
+   /// Set the desired assembly level. The default is AssemblyLevel::FULL.
+   /** This method must be called before assembly. */
   void SetAssemblyLevel(AssemblyLevel assembly_level);

   /// Returns the assembly level
   AssemblyLevel GetAssemblyLevel() const { return assembly; }

-   /** @brief Enable the use of static condensation. For details see the
-       description for class StaticCondensation in fem/staticcond.hpp This method
-       should be called before assembly. If the number of unknowns after static
+   /** Enable the use of static condensation. For details see the description
+       for class StaticCondensation in fem/staticcond.hpp This method should be
+       called before assembly. If the number of unknowns after static
       condensation is not reduced, it is not enabled. */
   void EnableStaticCondensation();

-   /** @brief Check if static condensation was actually enabled by a previous
-       call to EnableStaticCondensation(). */
+   /** Check if static condensation was actually enabled by a previous call to
+       EnableStaticCondensation(). */
   bool StaticCondensationIsEnabled() const { return static_cond; }

   /// Return the trace FE space associated with static condensation.
   FiniteElementSpace *SCFESpace() const
   { return static_cond ? static_cond->GetTraceFESpace() : NULL; }

-   /// Enable hybridization.
-   /** For details see the description for class
+   /** Enable hybridization; for details see the description for class
       Hybridization in fem/hybridization.hpp. This method should be called
       before assembly. */
   void EnableHybridization(FiniteElementSpace *constr_space,
                            BilinearFormIntegrator *constr_integ,
                            const Array<int> &ess_tdof_list);

-   /** @brief For scalar FE spaces, precompute the sparsity pattern of the matrix
+   /** For scalar FE spaces, precompute the sparsity pattern of the matrix
       (assuming dense element matrices) based on the types of integrators
       present in the bilinear form. */
   void UsePrecomputedSparsity(int ps = 1) { precompute_sparsity = ps; }
@@ -209,16 +194,15 @@ public:
   /// Use the sparsity of @a A to allocate the internal SparseMatrix.
   void UseSparsity(SparseMatrix &A);

-   /// Pre-allocate the internal SparseMatrix before assembly.
-   /**  If the flag 'precompute sparsity'
-       is set, the matrix is allocated in CSR format (i.e.
+   /** Pre-allocate the internal SparseMatrix before assembly. If the flag
+       'precompute sparsity' is set, the matrix is allocated in CSR format (i.e.
       finalized) and the entries are initialized with zeros. */
   void AllocateMatrix() { if (mat == NULL) { AllocMat(); } }

-   /// Access all the integrators added with AddDomainIntegrator().
+   /// Access all integrators added with AddDomainIntegrator().
   Array<BilinearFormIntegrator*> *GetDBFI() { return &dbfi; }

-   /// Access all the integrators added with AddBoundaryIntegrator().
+   /// Access all integrators added with AddBoundaryIntegrator().
   Array<BilinearFormIntegrator*> *GetBBFI() { return &bbfi; }
   /** @brief Access all boundary markers added with AddBoundaryIntegrator().
       If no marker was specified when the integrator was added, the
@@ -235,85 +219,64 @@ public:
       corresponding pointer (to Array<int>) will be NULL. */
   Array<Array<int>*> *GetBFBFI_Marker() { return &bfbfi_marker; }

-   /// Returns a reference to: \f$ M_{ij} \f$
   const double &operator()(int i, int j) { return (*mat)(i,j); }

-   /// Returns a reference to: \f$ M_{ij} \f$
+   /// Returns reference to a_{ij}.
   virtual double &Elem(int i, int j);

-   /// Returns constant reference to: \f$ M_{ij} \f$
+   /// Returns constant reference to a_{ij}.
   virtual const double &Elem(int i, int j) const;

-   /// Matrix vector multiplication:  \f$ y = M x \f$
+   /// Matrix vector multiplication.
   virtual void Mult(const Vector &x, Vector &y) const;

-   /** @brief Matrix vector multiplication with the original uneliminated
-       matrix.  The original matrix is \f$ M + M_e \f$ so we have:
-       \f$ y = M x + M_e x \f$ */
   void FullMult(const Vector &x, Vector &y) const
   { mat->Mult(x, y); mat_e->AddMult(x, y); }

-   /// Add the matrix vector multiple to a vector:  \f$ y += a M x \f$
   virtual void AddMult(const Vector &x, Vector &y, const double a = 1.0) const
   { mat -> AddMult (x, y, a); }

-   /** @brief Add the original uneliminated matrix vector multiple to a vector.
-       The original matrix is \f$ M + Me \f$ so we have:
-       \f$ y += M x + M_e x \f$ */
   void FullAddMult(const Vector &x, Vector &y) const
   { mat->AddMult(x, y); mat_e->AddMult(x, y); }

-   /// Add the matrix transpose vector multiplication:  \f$ y += a M^T x \f$
   virtual void AddMultTranspose(const Vector & x, Vector & y,
                                 const double a = 1.0) const
   { mat->AddMultTranspose(x, y, a); }

-   /** @brief Add the original uneliminated matrix transpose vector
-       multiple to a vector. The original matrix is \f$ M + M_e \f$
-       so we have: \f$ y += M^T x + {M_e}^T x \f$ */
   void FullAddMultTranspose(const Vector & x, Vector & y) const
   { mat->AddMultTranspose(x, y); mat_e->AddMultTranspose(x, y); }

-   /// Matrix transpose vector multiplication:  \f$ y = M^T x \f$
   virtual void MultTranspose(const Vector & x, Vector & y) const
   { y = 0.0; AddMultTranspose (x, y); }

-   /// Compute \f$ y^T M x \f$
   double InnerProduct(const Vector &x, const Vector &y) const
   { return mat->InnerProduct (x, y); }

-   /// Returns a pointer to (approximation) of the matrix inverse:  \f$ M^{-1} \f$
+   /// Returns a pointer to (approximation) of the matrix inverse.
   virtual MatrixInverse *Inverse() const;

   /// Finalizes the matrix initialization.
   virtual void Finalize(int skip_zeros = 1);

-   /// Returns a const reference to the sparse matrix.
+   /// Returns a reference to the sparse matrix
   const SparseMatrix &SpMat() const
   {
      MFEM_VERIFY(mat, "mat is NULL and can't be dereferenced");
      return *mat;
   }
-
-   /// Returns a reference to the sparse matrix:  \f$ M \f$
   SparseMatrix &SpMat()
   {
      MFEM_VERIFY(mat, "mat is NULL and can't be dereferenced");
      return *mat;
   }
-
-   /**  @brief Nullifies the internal matrix \f$ M \f$ and returns a pointer
-        to it.  Used for transfering ownership. */
   SparseMatrix *LoseMat() { SparseMatrix *tmp = mat; mat = NULL; return tmp; }

-   /// Returns a const reference to the sparse matrix of eliminated b.c.: \f$ M_e \f$
+   /// Returns a reference to the sparse matrix of eliminated b.c.
   const SparseMatrix &SpMatElim() const
   {
      MFEM_VERIFY(mat_e, "mat_e is NULL and can't be dereferenced");
      return *mat_e;
   }
-
-   /// Returns a reference to the sparse matrix of eliminated b.c.: \f$ M_e \f$
   SparseMatrix &SpMatElim()
   {
      MFEM_VERIFY(mat_e, "mat_e is NULL and can't be dereferenced");
@@ -348,7 +311,6 @@ public:
   void AddBdrFaceIntegrator(BilinearFormIntegrator *bfi,
                             Array<int> &bdr_marker);

-   /// Sets all sparse values of \f$ M \f$ and \f$ M_e \f$ to 'a'.
   void operator=(const double a)
   {
      if (mat != NULL) { *mat = a; }
@@ -366,10 +328,10 @@ public:
       for an AMR mesh. */
   void AssembleDiagonal(Vector &diag) const;

-   /// Get the finite element space prolongation operator.
+   /// Get the finite element space prolongation matrix
   virtual const Operator *GetProlongation() const
   { return fes->GetConformingProlongation(); }
-   /// Get the finite element space restriction operator
+   /// Get the finite element space restriction matrix
   virtual const Operator *GetRestriction() const
   { return fes->GetConformingRestriction(); }
   /// Get the output finite element space prolongation matrix
@@ -529,12 +491,10 @@ public:
                                 double value);

   /// Eliminate the given @a vdofs. NOTE: here, @a vdofs is a list of DOFs.
-   /** In this case the eliminations are applied to the internal \f$ M \f$
-       and @a rhs without storing the elimination matrix \f$ M_e \f$. */
   void EliminateVDofs(const Array<int> &vdofs, const Vector &sol, Vector &rhs,
                       DiagonalPolicy dpolicy = DIAG_ONE);

-   /// Eliminate the given @a vdofs, storing the eliminated part internally in \f$ M_e \f$.
+   /// Eliminate the given @a vdofs, storing the eliminated part internally.
   /** This method works in conjunction with EliminateVDofsInRHS() and allows
       elimination of boundary conditions in multiple right-hand sides. In this
       method, @a vdofs is a list of DOFs. */
@@ -563,11 +523,9 @@ public:
   void EliminateVDofsInRHS(const Array<int> &vdofs, const Vector &x,
                            Vector &b);

-   /// Compute inner product for full uneliminated matrix \f$ y^T M x + y^T M_e x \f$
   double FullInnerProduct(const Vector &x, const Vector &y) const
   { return mat->InnerProduct(x, y) + mat_e->InnerProduct(x, y); }

-   /// Update the @a FiniteElementSpace and delete all data associated with the old one.
   virtual void Update(FiniteElementSpace *nfes = NULL);

   /// (DEPRECATED) Return the FE space associated with the BilinearForm.
@@ -579,13 +537,7 @@ public:
   /// Read-only access to the associated FiniteElementSpace.
   const FiniteElementSpace *FESpace() const { return fes; }

-   /// Sets diagonal policy used upon construction of the linear system.
-   /** Policies include:
-
-       - DIAG_ZERO (Set the diagonal values to zero)
-       - DIAG_ONE  (Set the diagonal values to one)
-       - DIAG_KEEP (Keep the diagonal values)
-   */
+   /// Sets diagonal policy used upon construction of the linear system
   void SetDiagonalPolicy(DiagonalPolicy policy);

   /// Indicate that integrators are not owned by the BilinearForm
@@ -598,16 +550,16 @@ public:

 /**
   Class for assembling of bilinear forms `a(u,v)` defined on different
-   trial and test spaces. The assembled matrix `M` is such that
+   trial and test spaces. The assembled matrix `A` is such that

-       a(u,v) = V^t M U
+       a(u,v) = V^t A U

   where `U` and `V` are the vectors representing the functions `u` and `v`,
   respectively.  The first argument, `u`, of `a(,)` is in the trial space
   and the second argument, `v`, is in the test space. Thus,

-       # of rows of M = dimension of the test space and
-       # of cols of M = dimension of the trial space.
+       # of rows of A = dimension of the test space and
+       # of cols of A = dimension of the trial space.

   Both trial and test spaces should be defined on the same mesh.
 */
@@ -676,15 +628,11 @@ public:
                     FiniteElementSpace *te_fes,
                     MixedBilinearForm *mbf);

-   /// Returns a reference to: \f$ M_{ij} \f$
   virtual double &Elem(int i, int j);

-   /// Returns a reference to: \f$ M_{ij} \f$
   virtual const double &Elem(int i, int j) const;

-   /// Matrix multiplication: \f$ y = M x \f$
   virtual void Mult(const Vector & x, Vector & y) const;
-
   virtual void AddMult(const Vector & x, Vector & y,
                        const double a = 1.0) const;

@@ -694,7 +642,6 @@ public:

   virtual MatrixInverse *Inverse() const;

-   /// Finalizes the matrix initialization.
   virtual void Finalize(int skip_zeros = 1);

   /** Extract the associated matrix as SparseMatrix blocks. The number of
@@ -702,14 +649,8 @@ public:
       test and trial spaces, respectively. */
   void GetBlocks(Array2D<SparseMatrix *> &blocks) const;

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

   /// Adds a domain integrator. Assumes ownership of @a bfi.
@@ -756,7 +697,6 @@ public:
       corresponding pointer (to Array<int>) will be NULL. */
   Array<Array<int>*> *GetBTFBFI_Marker() { return &btfbfi_marker; }

-   /// Sets all sparse values of \f$ M \f$ to @a a.
   void operator=(const double a) { *mat = a; }

   /// Set the desired assembly level. The default is AssemblyLevel::FULL.
@@ -765,10 +705,6 @@ public:

   void Assemble(int skip_zeros = 1);

-   /** @brief Assemble the diagonal of ADA^T into diag, where A is this mixed
-       bilinear form and D is a diagonal. */
-   void AssembleDiagonal_ADAt(const Vector &D, Vector &diag) const;
-
   /// Get the input finite element space prolongation matrix
   virtual const Operator *GetProlongation() const
   { return trial_fes->GetProlongationMatrix(); }
@@ -15,7 +15,6 @@
 #include "../general/forall.hpp"
 #include "bilinearform.hpp"
 #include "libceed/ceed.hpp"
-#include "pgridfunc.hpp"

 namespace mfem
 {
@@ -48,7 +47,7 @@ PABilinearFormExtension::PABilinearFormExtension(BilinearForm *form)
   bdr_face_restrict_lex = NULL;
 }

-void PABilinearFormExtension::SetupRestrictionOperators(const L2FaceValues m)
+void PABilinearFormExtension::SetupRestrictionOperators()
 {
   ElementDofOrdering ordering = UsesTensorBasis(*a->FESpace())?
                                 ElementDofOrdering::LEXICOGRAPHIC:
@@ -66,8 +65,7 @@ void PABilinearFormExtension::SetupRestrictionOperators(const L2FaceValues m)
   if (int_face_restrict_lex == NULL && a->GetFBFI()->Size() > 0)
   {
      int_face_restrict_lex = trialFes->GetFaceRestriction(
-                                 ElementDofOrdering::LEXICOGRAPHIC,
-                                 FaceType::Interior);
+                                 ElementDofOrdering::LEXICOGRAPHIC, FaceType::Interior);
      faceIntX.SetSize(int_face_restrict_lex->Height(), Device::GetMemoryType());
      faceIntY.SetSize(int_face_restrict_lex->Height(), Device::GetMemoryType());
      faceIntY.UseDevice(true); // ensure 'faceIntY = 0.0' is done on device
@@ -76,9 +74,7 @@ void PABilinearFormExtension::SetupRestrictionOperators(const L2FaceValues m)
   if (bdr_face_restrict_lex == NULL && a->GetBFBFI()->Size() > 0)
   {
      bdr_face_restrict_lex = trialFes->GetFaceRestriction(
-                                 ElementDofOrdering::LEXICOGRAPHIC,
-                                 FaceType::Boundary,
-                                 m);
+                                 ElementDofOrdering::LEXICOGRAPHIC, FaceType::Boundary);
      faceBdrX.SetSize(bdr_face_restrict_lex->Height(), Device::GetMemoryType());
      faceBdrY.SetSize(bdr_face_restrict_lex->Height(), Device::GetMemoryType());
      faceBdrY.UseDevice(true); // ensure 'faceBoundY = 0.0' is done on device
@@ -87,7 +83,7 @@ void PABilinearFormExtension::SetupRestrictionOperators(const L2FaceValues m)

 void PABilinearFormExtension::Assemble()
 {
-   SetupRestrictionOperators(L2FaceValues::DoubleValued);
+   SetupRestrictionOperators();

   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
   const int integratorCount = integrators.Size();
@@ -116,7 +112,7 @@ void PABilinearFormExtension::AssembleDiagonal(Vector &y) const
   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();

   const int iSz = integrators.Size();
-   if (elem_restrict && !DeviceCanUseCeed())
+   if (elem_restrict)
   {
      localY = 0.0;
      for (int i = 0; i < iSz; ++i)
@@ -291,462 +287,6 @@ void PABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
   }
 }

-// Data and methods for element-assembled bilinear forms
-EABilinearFormExtension::EABilinearFormExtension(BilinearForm *form)
-   : PABilinearFormExtension(form),
-     factorize_face_terms(form->FESpace()->IsDGSpace())
-{
-}
-
-void EABilinearFormExtension::Assemble()
-{
-   SetupRestrictionOperators(L2FaceValues::SingleValued);
-
-   ne = trialFes->GetMesh()->GetNE();
-   elemDofs = trialFes->GetFE(0)->GetDof();
-
-   ea_data.SetSize(ne*elemDofs*elemDofs, Device::GetMemoryType());
-   ea_data.UseDevice(true);
-   ea_data = 0.0;
-
-   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
-   const int integratorCount = integrators.Size();
-   for (int i = 0; i < integratorCount; ++i)
-   {
-      integrators[i]->AssembleEA(*a->FESpace(), ea_data);
-   }
-
-   faceDofs = trialFes ->
-              GetTraceElement(0, trialFes->GetMesh()->GetFaceBaseGeometry(0)) ->
-              GetDof();
-
-   Array<BilinearFormIntegrator*> &intFaceIntegrators = *a->GetFBFI();
-   const int intFaceIntegratorCount = intFaceIntegrators.Size();
-   if (intFaceIntegratorCount>0)
-   {
-      nf_int = trialFes->GetNFbyType(FaceType::Interior);
-      ea_data_int.SetSize(2*nf_int*faceDofs*faceDofs, Device::GetMemoryType());
-      ea_data_ext.SetSize(2*nf_int*faceDofs*faceDofs, Device::GetMemoryType());
-      ea_data_int = 0.0;
-      ea_data_ext = 0.0;
-   }
-   for (int i = 0; i < intFaceIntegratorCount; ++i)
-   {
-      intFaceIntegrators[i]->AssembleEAInteriorFaces(*a->FESpace(),
-                                                     ea_data_int,
-                                                     ea_data_ext);
-   }
-
-   Array<BilinearFormIntegrator*> &bdrFaceIntegrators = *a->GetBFBFI();
-   const int boundFaceIntegratorCount = bdrFaceIntegrators.Size();
-   if (boundFaceIntegratorCount>0)
-   {
-      nf_bdr = trialFes->GetNFbyType(FaceType::Boundary);
-      ea_data_bdr.SetSize(nf_bdr*faceDofs*faceDofs, Device::GetMemoryType());
-      ea_data_bdr = 0.0;
-   }
-   for (int i = 0; i < boundFaceIntegratorCount; ++i)
-   {
-      bdrFaceIntegrators[i]->AssembleEABoundaryFaces(*a->FESpace(),ea_data_bdr);
-   }
-
-   if (factorize_face_terms && int_face_restrict_lex)
-   {
-      auto restFint = dynamic_cast<const L2FaceRestriction&>(*int_face_restrict_lex);
-      restFint.AddFaceMatricesToElementMatrices(ea_data_int, ea_data);
-   }
-   if (factorize_face_terms && bdr_face_restrict_lex)
-   {
-      auto restFbdr = dynamic_cast<const L2FaceRestriction&>(*bdr_face_restrict_lex);
-      restFbdr.AddFaceMatricesToElementMatrices(ea_data_bdr, ea_data);
-   }
-}
-
-void EABilinearFormExtension::Mult(const Vector &x, Vector &y) const
-{
-   // Apply the Element Restriction
-   const bool useRestrict = !DeviceCanUseCeed() && elem_restrict;
-   if (!useRestrict)
-   {
-      y.UseDevice(true); // typically this is a large vector, so store on device
-      y = 0.0;
-   }
-   else
-   {
-      elem_restrict->Mult(x, localX);
-      localY = 0.0;
-   }
-   // Apply the Element Matrices
-   const int NDOFS = elemDofs;
-   auto X = Reshape(useRestrict?localX.Read():x.Read(), NDOFS, ne);
-   auto Y = Reshape(useRestrict?localY.ReadWrite():y.ReadWrite(), NDOFS, ne);
-   auto A = Reshape(ea_data.Read(), NDOFS, NDOFS, ne);
-   MFEM_FORALL(glob_j, ne*NDOFS,
-   {
-      const int e = glob_j/NDOFS;
-      const int j = glob_j%NDOFS;
-      double res = 0.0;
-      for (int i = 0; i < NDOFS; i++)
-      {
-         res += A(i, j, e)*X(i, e);
-      }
-      Y(j, e) += res;
-   });
-   // Apply the Element Restriction transposed
-   if (useRestrict)
-   {
-      elem_restrict->MultTranspose(localY, y);
-   }
-
-   // Treatment of interior faces
-   Array<BilinearFormIntegrator*> &intFaceIntegrators = *a->GetFBFI();
-   const int iFISz = intFaceIntegrators.Size();
-   if (int_face_restrict_lex && iFISz>0)
-   {
-      // Apply the Interior Face Restriction
-      int_face_restrict_lex->Mult(x, faceIntX);
-      if (faceIntX.Size()>0)
-      {
-         faceIntY = 0.0;
-         // Apply the interior face matrices
-         const int NDOFS = faceDofs;
-         auto X = Reshape(faceIntX.Read(), NDOFS, 2, nf_int);
-         auto Y = Reshape(faceIntY.ReadWrite(), NDOFS, 2, nf_int);
-         if (!factorize_face_terms)
-         {
-            auto A_int = Reshape(ea_data_int.Read(), NDOFS, NDOFS, 2, nf_int);
-            MFEM_FORALL(glob_j, nf_int*NDOFS,
-            {
-               const int f = glob_j/NDOFS;
-               const int j = glob_j%NDOFS;
-               double res = 0.0;
-               for (int i = 0; i < NDOFS; i++)
-               {
-                  res += A_int(i, j, 0, f)*X(i, 0, f);
-               }
-               Y(j, 0, f) += res;
-               res = 0.0;
-               for (int i = 0; i < NDOFS; i++)
-               {
-                  res += A_int(i, j, 1, f)*X(i, 1, f);
-               }
-               Y(j, 1, f) += res;
-            });
-         }
-         auto A_ext = Reshape(ea_data_ext.Read(), NDOFS, NDOFS, 2, nf_int);
-         MFEM_FORALL(glob_j, nf_int*NDOFS,
-         {
-            const int f = glob_j/NDOFS;
-            const int j = glob_j%NDOFS;
-            double res = 0.0;
-            for (int i = 0; i < NDOFS; i++)
-            {
-               res += A_ext(i, j, 0, f)*X(i, 0, f);
-            }
-            Y(j, 1, f) += res;
-            res = 0.0;
-            for (int i = 0; i < NDOFS; i++)
-            {
-               res += A_ext(i, j, 1, f)*X(i, 1, f);
-            }
-            Y(j, 0, f) += res;
-         });
-         // Apply the Interior Face Restriction transposed
-         int_face_restrict_lex->MultTranspose(faceIntY, y);
-      }
-   }
-
-   // Treatment of boundary faces
-   Array<BilinearFormIntegrator*> &bdrFaceIntegrators = *a->GetBFBFI();
-   const int bFISz = bdrFaceIntegrators.Size();
-   if (!factorize_face_terms && bdr_face_restrict_lex && bFISz>0)
-   {
-      // Apply the Boundary Face Restriction
-      bdr_face_restrict_lex->Mult(x, faceBdrX);
-      if (faceBdrX.Size()>0)
-      {
-         faceBdrY = 0.0;
-         // Apply the boundary face matrices
-         const int NDOFS = faceDofs;
-         auto X = Reshape(faceBdrX.Read(), NDOFS, nf_bdr);
-         auto Y = Reshape(faceBdrY.ReadWrite(), NDOFS, nf_bdr);
-         auto A = Reshape(ea_data_bdr.Read(), NDOFS, NDOFS, nf_bdr);
-         MFEM_FORALL(glob_j, nf_bdr*NDOFS,
-         {
-            const int f = glob_j/NDOFS;
-            const int j = glob_j%NDOFS;
-            double res = 0.0;
-            for (int i = 0; i < NDOFS; i++)
-            {
-               res += A(i, j, f)*X(i, f);
-            }
-            Y(j, f) += res;
-         });
-         // Apply the Boundary Face Restriction transposed
-         bdr_face_restrict_lex->MultTranspose(faceBdrY, y);
-      }
-   }
-}
-
-void EABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
-{
-   // Apply the Element Restriction
-   const bool useRestrict = DeviceCanUseCeed() || !elem_restrict;
-   if (!useRestrict)
-   {
-      y.UseDevice(true); // typically this is a large vector, so store on device
-      y = 0.0;
-   }
-   else
-   {
-      elem_restrict->Mult(x, localX);
-      localY = 0.0;
-   }
-   // Apply the Element Matrices transposed
-   const int NDOFS = elemDofs;
-   auto X = Reshape(useRestrict?localX.Read():x.Read(), NDOFS, ne);
-   auto Y = Reshape(useRestrict?localY.ReadWrite():y.ReadWrite(), NDOFS, ne);
-   auto A = Reshape(ea_data.Read(), NDOFS, NDOFS, ne);
-   MFEM_FORALL(glob_j, ne*NDOFS,
-   {
-      const int e = glob_j/NDOFS;
-      const int j = glob_j%NDOFS;
-      double res = 0.0;
-      for (int i = 0; i < NDOFS; i++)
-      {
-         res += A(j, i, e)*X(i, e);
-      }
-      Y(j, e) += res;
-   });
-   // Apply the Element Restriction transposed
-   if (useRestrict)
-   {
-      elem_restrict->MultTranspose(localY, y);
-   }
-
-   // Treatment of interior faces
-   Array<BilinearFormIntegrator*> &intFaceIntegrators = *a->GetFBFI();
-   const int iFISz = intFaceIntegrators.Size();
-   if (int_face_restrict_lex && iFISz>0)
-   {
-      // Apply the Interior Face Restriction
-      int_face_restrict_lex->Mult(x, faceIntX);
-      if (faceIntX.Size()>0)
-      {
-         faceIntY = 0.0;
-         // Apply the interior face matrices transposed
-         const int NDOFS = faceDofs;
-         auto X = Reshape(faceIntX.Read(), NDOFS, 2, nf_int);
-         auto Y = Reshape(faceIntY.ReadWrite(), NDOFS, 2, nf_int);
-         if (!factorize_face_terms)
-         {
-            auto A_int = Reshape(ea_data_int.Read(), NDOFS, NDOFS, 2, nf_int);
-            MFEM_FORALL(glob_j, nf_int*NDOFS,
-            {
-               const int f = glob_j/NDOFS;
-               const int j = glob_j%NDOFS;
-               double res = 0.0;
-               for (int i = 0; i < NDOFS; i++)
-               {
-                  res += A_int(j, i, 0, f)*X(i, 0, f);
-               }
-               Y(j, 0, f) += res;
-               res = 0.0;
-               for (int i = 0; i < NDOFS; i++)
-               {
-                  res += A_int(j, i, 1, f)*X(i, 1, f);
-               }
-               Y(j, 1, f) += res;
-            });
-         }
-         auto A_ext = Reshape(ea_data_ext.Read(), NDOFS, NDOFS, 2, nf_int);
-         MFEM_FORALL(glob_j, nf_int*NDOFS,
-         {
-            const int f = glob_j/NDOFS;
-            const int j = glob_j%NDOFS;
-            double res = 0.0;
-            for (int i = 0; i < NDOFS; i++)
-            {
-               res += A_ext(j, i, 0, f)*X(i, 0, f);
-            }
-            Y(j, 1, f) += res;
-            res = 0.0;
-            for (int i = 0; i < NDOFS; i++)
-            {
-               res += A_ext(j, i, 1, f)*X(i, 1, f);
-            }
-            Y(j, 0, f) += res;
-         });
-         // Apply the Interior Face Restriction transposed
-         int_face_restrict_lex->MultTranspose(faceIntY, y);
-      }
-   }
-
-   // Treatment of boundary faces
-   Array<BilinearFormIntegrator*> &bdrFaceIntegrators = *a->GetBFBFI();
-   const int bFISz = bdrFaceIntegrators.Size();
-   if (!factorize_face_terms && bdr_face_restrict_lex && bFISz>0)
-   {
-      // Apply the Boundary Face Restriction
-      bdr_face_restrict_lex->Mult(x, faceBdrX);
-      if (faceBdrX.Size()>0)
-      {
-         faceBdrY = 0.0;
-         // Apply the boundary face matrices transposed
-         const int NDOFS = faceDofs;
-         auto X = Reshape(faceBdrX.Read(), NDOFS, nf_bdr);
-         auto Y = Reshape(faceBdrY.ReadWrite(), NDOFS, nf_bdr);
-         auto A = Reshape(ea_data_bdr.Read(), NDOFS, NDOFS, nf_bdr);
-         MFEM_FORALL(glob_j, nf_bdr*NDOFS,
-         {
-            const int f = glob_j/NDOFS;
-            const int j = glob_j%NDOFS;
-            double res = 0.0;
-            for (int i = 0; i < NDOFS; i++)
-            {
-               res += A(j, i, f)*X(i, f);
-            }
-            Y(j, f) += res;
-         });
-         // Apply the Boundary Face Restriction transposed
-         bdr_face_restrict_lex->MultTranspose(faceBdrY, y);
-      }
-   }
-}
-
-// Data and methods for fully-assembled bilinear forms
-FABilinearFormExtension::FABilinearFormExtension(BilinearForm *form)
-   : EABilinearFormExtension(form),
-     mat(form->FESpace()->GetVSize(),form->FESpace()->GetVSize(),0),
-     face_mat(form->FESpace()->GetVSize(),0,0),
-     use_face_mat(false)
-{
-#ifdef MFEM_USE_MPI
-   if ( ParFiniteElementSpace* pfes =
-           dynamic_cast<ParFiniteElementSpace*>(form->FESpace()) )
-   {
-      if (pfes->IsDGSpace())
-      {
-         use_face_mat = true;
-         pfes->ExchangeFaceNbrData();
-         face_mat.SetWidth(pfes->GetFaceNbrVSize());
-      }
-   }
-#endif
-}
-
-void FABilinearFormExtension::Assemble()
-{
-   EABilinearFormExtension::Assemble();
-   FiniteElementSpace &fes = *a->FESpace();
-   if (fes.IsDGSpace())
-   {
-      const L2ElementRestriction *restE =
-         static_cast<const L2ElementRestriction*>(elem_restrict);
-      const L2FaceRestriction *restF =
-         static_cast<const L2FaceRestriction*>(int_face_restrict_lex);
-      // 1. Fill I
-      //  1.1 Increment with restE
-      restE->FillI(mat);
-      //  1.2 Increment with restF
-      if (restF) { restF->FillI(mat, face_mat); }
-      //  1.3 Sum the non-zeros in I
-      auto h_I = mat.HostReadWriteI();
-      int cpt = 0;
-      const int vd = fes.GetVDim();
-      const int ndofs = ne*elemDofs*vd;
-      for (int i = 0; i < ndofs; i++)
-      {
-         const int nnz = h_I[i];
-         h_I[i] = cpt;
-         cpt += nnz;
-      }
-      const int nnz = cpt;
-      h_I[ndofs] = nnz;
-      mat.GetMemoryJ().New(nnz, mat.GetMemoryJ().GetMemoryType());
-      mat.GetMemoryData().New(nnz, mat.GetMemoryData().GetMemoryType());
-      if (use_face_mat && restF)
-      {
-         auto h_I_face = face_mat.HostReadWriteI();
-         int cpt = 0;
-         for (int i = 0; i < ndofs; i++)
-         {
-            const int nnz = h_I_face[i];
-            h_I_face[i] = cpt;
-            cpt += nnz;
-         }
-         const int nnz_face = cpt;
-         h_I_face[ndofs] = nnz_face;
-         face_mat.GetMemoryJ().New(nnz_face,
-                                   face_mat.GetMemoryJ().GetMemoryType());
-         face_mat.GetMemoryData().New(nnz_face,
-                                      face_mat.GetMemoryData().GetMemoryType());
-      }
-      // 2. Fill J and Data
-      // 2.1 Fill J and Data with Elem ea_data
-      restE->FillJAndData(ea_data, mat);
-      // 2.2 Fill J and Data with Face ea_data_ext
-      if (restF) { restF->FillJAndData(ea_data_ext, mat, face_mat); }
-      // 2.3 Shift indirections in I back to original
-      auto I = mat.HostReadWriteI();
-      for (int i = ndofs; i > 0; i--)
-      {
-         I[i] = I[i-1];
-      }
-      I[0] = 0;
-      if (use_face_mat && restF)
-      {
-         auto I_face = face_mat.HostReadWriteI();
-         for (int i = ndofs; i > 0; i--)
-         {
-            I_face[i] = I_face[i-1];
-         }
-         I_face[0] = 0;
-      }
-   }
-   else // continuous Galerkin case
-   {
-      const ElementRestriction &rest =
-         static_cast<const ElementRestriction&>(*elem_restrict);
-      rest.FillSparseMatrix(ea_data, mat);
-   }
-}
-
-void FABilinearFormExtension::Mult(const Vector &x, Vector &y) const
-{
-   mat.Mult(x, y);
-#ifdef MFEM_USE_MPI
-   if (const ParFiniteElementSpace *pfes =
-          dynamic_cast<const ParFiniteElementSpace*>(testFes))
-   {
-      ParGridFunction x_gf;
-      x_gf.MakeRef(const_cast<ParFiniteElementSpace*>(pfes),
-                   const_cast<Vector&>(x),0);
-      x_gf.ExchangeFaceNbrData();
-      Vector &shared_x = x_gf.FaceNbrData();
-      if (shared_x.Size()) { face_mat.AddMult(shared_x, y); }
-   }
-#endif
-}
-
-void FABilinearFormExtension::MultTranspose(const Vector &x, Vector &y) const
-{
-   mat.MultTranspose(x, y);
-#ifdef MFEM_USE_MPI
-   if (const ParFiniteElementSpace *pfes =
-          dynamic_cast<const ParFiniteElementSpace*>(testFes))
-   {
-      ParGridFunction x_gf;
-      x_gf.MakeRef(const_cast<ParFiniteElementSpace*>(pfes),
-                   const_cast<Vector&>(x),0);
-      x_gf.ExchangeFaceNbrData();
-      Vector &shared_x = x_gf.FaceNbrData();
-      if (shared_x.Size()) { face_mat.AddMultTranspose(shared_x, y); }
-   }
-#endif
-}
-
-
 MixedBilinearFormExtension::MixedBilinearFormExtension(MixedBilinearForm *form)
   : Operator(form->Height(), form->Width()), a(form)
 {
@@ -947,68 +487,4 @@ void PAMixedBilinearFormExtension::AddMultTranspose(const Vector &x, Vector &y,
   }
 }

-void PAMixedBilinearFormExtension::AssembleDiagonal_ADAt(const Vector &D,
-                                                         Vector &diag) const
-{
-   Array<BilinearFormIntegrator*> &integrators = *a->GetDBFI();
-
-   const int iSz = integrators.Size();
-
-   if (elem_restrict_trial)
-   {
-      const ElementRestriction* H1elem_restrict_trial =
-         dynamic_cast<const ElementRestriction*>(elem_restrict_trial);
-      if (H1elem_restrict_trial)
-      {
-         H1elem_restrict_trial->MultUnsigned(D, localTrial);
-      }
-      else
-      {
-         elem_restrict_trial->Mult(D, localTrial);
-      }
-   }
-
-   if (elem_restrict_test)
-   {
-      localTest = 0.0;
-      for (int i = 0; i < iSz; ++i)
-      {
-         if (elem_restrict_trial)
-         {
-            integrators[i]->AssembleDiagonalPA_ADAt(localTrial, localTest);
-         }
-         else
-         {
-            integrators[i]->AssembleDiagonalPA_ADAt(D, localTest);
-         }
-      }
-      const ElementRestriction* H1elem_restrict_test =
-         dynamic_cast<const ElementRestriction*>(elem_restrict_test);
-      if (H1elem_restrict_test)
-      {
-         H1elem_restrict_test->MultTransposeUnsigned(localTest, diag);
-      }
-      else
-      {
-         elem_restrict_test->MultTranspose(localTest, diag);
-      }
-   }
-   else
-   {
-      diag.UseDevice(true); // typically this is a large vector, so store on device
-      diag = 0.0;
-      for (int i = 0; i < iSz; ++i)
-      {
-         if (elem_restrict_trial)
-         {
-            integrators[i]->AssembleDiagonalPA_ADAt(localTrial, diag);
-         }
-         else
-         {
-            integrators[i]->AssembleDiagonalPA_ADAt(D, diag);
-         }
-      }
-   }
-}
-
 } // namespace mfem
@@ -22,12 +22,9 @@ namespace mfem
 class BilinearForm;
 class MixedBilinearForm;

-/// Class extending the BilinearForm class to support different AssemblyLevels.
-/**  FA - Full Assembly
-     PA - Partial Assembly
-     EA - Element Assembly
-     MF - Matrix Free
-*/
+
+/** @brief Class extending the BilinearForm class to support the different
+    AssemblyLevel%s. */
 class BilinearFormExtension : public Operator
 {
 protected:
@@ -45,7 +42,6 @@ public:
   /// Get the finite element space restriction matrix
   virtual const Operator *GetRestriction() const;

-   /// Assemble at the level given for the BilinearFormExtension subclass
   virtual void Assemble() = 0;

   virtual void AssembleDiagonal(Vector &diag) const
@@ -62,6 +58,46 @@ public:
   virtual void Update() = 0;
 };

+/// Data and methods for fully-assembled bilinear forms
+class FABilinearFormExtension : public BilinearFormExtension
+{
+public:
+   FABilinearFormExtension(BilinearForm *form)
+      : BilinearFormExtension(form) { }
+
+   /// TODO
+   void Assemble() {}
+   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 {}
+   void Update() {}
+   ~FABilinearFormExtension() {}
+};
+
+/// Data and methods for element-assembled bilinear forms
+class EABilinearFormExtension : public BilinearFormExtension
+{
+public:
+   EABilinearFormExtension(BilinearForm *form)
+      : BilinearFormExtension(form) { }
+
+   /// TODO
+   void Assemble() {}
+   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 {}
+   void Update() {}
+   ~EABilinearFormExtension() {}
+};
+
 /// Data and methods for partially-assembled bilinear forms
 class PABilinearFormExtension : public BilinearFormExtension
 {
@@ -77,6 +113,7 @@ protected:
 public:
   PABilinearFormExtension(BilinearForm*);

+   void SetupRestrictionOperators();
   void Assemble();
   void AssembleDiagonal(Vector &diag) const;
   void FormSystemMatrix(const Array<int> &ess_tdof_list, OperatorHandle &A);
@@ -84,53 +121,14 @@ public:
                         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;
   void Update();
-
-protected:
-   void SetupRestrictionOperators(const L2FaceValues m);
 };

-/// Data and methods for element-assembled bilinear forms
-class EABilinearFormExtension : public PABilinearFormExtension
-{
-protected:
-   int ne;
-   int elemDofs;
-   // The element matrices are stored row major
-   Vector ea_data;
-   int nf_int, nf_bdr;
-   int faceDofs;
-   Vector ea_data_int, ea_data_ext, ea_data_bdr;
-   bool factorize_face_terms;

-public:
-   EABilinearFormExtension(BilinearForm *form);
-
-   void Assemble();
-   void Mult(const Vector &x, Vector &y) const;
-   void MultTranspose(const Vector &x, Vector &y) const;
-};
-
-/// Data and methods for fully-assembled bilinear forms
-class FABilinearFormExtension : public EABilinearFormExtension
-{
-private:
-   SparseMatrix mat;
-   /// face_mat handles parallelism for DG face terms.
-   SparseMatrix face_mat;
-   bool use_face_mat;
-
-public:
-   FABilinearFormExtension(BilinearForm *form);
-
-   void Assemble();
-   void Mult(const Vector &x, Vector &y) const;
-   void MultTranspose(const Vector &x, Vector &y) const;
-};
-
-/// Data and methods for matrix-free bilinear forms NOT YET IMPLEMENTED.
+/// Data and methods for matrix-free bilinear forms
 class MFBilinearFormExtension : public BilinearFormExtension
 {
 public:
@@ -150,12 +148,8 @@ public:
   ~MFBilinearFormExtension() {}
 };

-/// Class extending the MixedBilinearForm class to support different AssemblyLevels.
-/**  FA - Full Assembly
-     PA - Partial Assembly
-     EA - Element Assembly
-     MF - Matrix Free
-*/
+/** @brief Class extending the MixedBilinearForm class to support the different
+    AssemblyLevel%s. */
 class MixedBilinearFormExtension : public Operator
 {
 protected:
@@ -192,8 +186,6 @@ public:
   virtual void AddMultTranspose(const Vector &x, Vector &y,
                                 const double c=1.0) const = 0;

-   virtual void AssembleDiagonal_ADAt(const Vector &D, Vector &diag) const = 0;
-
   virtual void Update() = 0;
 };

@@ -244,9 +236,6 @@ public:
   void MultTranspose(const Vector &x, Vector &y) const;
   /// y += c*A^T*x
   void AddMultTranspose(const Vector &x, Vector &y, const double c=1.0) const;
-   /// Assemble the diagonal of ADA^T for a diagonal vector D.
-   void AssembleDiagonal_ADAt(const Vector &D, Vector &diag) const;
-
   /// Update internals for when a new MixedBilinearForm is given to this class
   void Update();
 };
@@ -47,37 +47,7 @@ void BilinearFormIntegrator::AssemblePABoundaryFaces(const FiniteElementSpace&)

 void BilinearFormIntegrator::AssembleDiagonalPA(Vector &)
 {
-   mfem_error ("BilinearFormIntegrator::AssembleDiagonalPA(...)\n"
-               "   is not implemented for this class.");
-}
-
-void BilinearFormIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                        Vector &emat)
-{
-   mfem_error ("BilinearFormIntegrator::AssembleEA(...)\n"
-               "   is not implemented for this class.");
-}
-
-void BilinearFormIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace
-                                                     &fes,
-                                                     Vector &ea_data_int,
-                                                     Vector &ea_data_ext)
-{
-   mfem_error ("BilinearFormIntegrator::AssembleEAInteriorFaces(...)\n"
-               "   is not implemented for this class.");
-}
-
-void BilinearFormIntegrator::AssembleEABoundaryFaces(const FiniteElementSpace
-                                                     &fes,
-                                                     Vector &ea_data_bdr)
-{
-   mfem_error ("BilinearFormIntegrator::AssembleEABoundaryFaces(...)\n"
-               "   is not implemented for this class.");
-}
-
-void BilinearFormIntegrator::AssembleDiagonalPA_ADAt(const Vector &, Vector &)
-{
-   MFEM_ABORT("BilinearFormIntegrator::AssembleDiagonalPA_ADAt(...)\n"
+   MFEM_ABORT("BilinearFormIntegrator::AssembleDiagonalPA(...)\n"
              "   is not implemented for this class.");
 }

@@ -919,25 +889,22 @@ void BoundaryMassIntegrator::AssembleFaceMatrix(
   {
      int order = 2 * el1.GetOrder();

-      ir = &IntRules.Get(Trans.GetGeometryType(), order);
+      ir = &IntRules.Get(Trans.FaceGeom, order);
   }

   elmat = 0.0;
   for (int i = 0; i < ir->GetNPoints(); i++)
   {
      const IntegrationPoint &ip = ir->IntPoint(i);
-
-      // Set the integration point in the face and the neighboring element
-      Trans.SetAllIntPoints(&ip);
-
-      // Access the neighboring element's integration point
-      const IntegrationPoint &eip = Trans.GetElement1IntPoint();
+      IntegrationPoint eip;
+      Trans.Loc1.Transform(ip, eip);
      el1.CalcShape(eip, shape);

-      w = Trans.Weight() * ip.weight;
+      Trans.Face->SetIntPoint(&ip);
+      w = Trans.Face->Weight() * ip.weight;
      if (Q)
      {
-         w *= Q -> Eval(Trans, ip);
+         w *= Q -> Eval(*Trans.Face, ip);
      }

      AddMult_a_VVt(w, shape, elmat);
@@ -2007,7 +1974,7 @@ void VectorFEMassIntegrator::AssembleElementMatrix2(
      D.SetSize(VQ ? VQ->GetVDim() : 0);
      K.SetSize(MQ ? MQ->GetVDim() : 0, MQ ? MQ->GetVDim() : 0);
 #endif
-      DenseMatrix tmp(test_vshape.Height(), K.Width());
+      DenseMatrix tmp(trial_vshape.Height(), K.Width());

      elmat.SetSize (test_dof, trial_dof);

@@ -2568,23 +2535,23 @@ void DGTraceIntegrator::AssembleFaceMatrix(const FiniteElement &el1,
      {
         order++;
      }
-      ir = &IntRules.Get(Trans.GetGeometryType(), order);
+      ir = &IntRules.Get(Trans.FaceGeom, order);
   }

   for (int p = 0; p < ir->GetNPoints(); p++)
   {
      const IntegrationPoint &ip = ir->IntPoint(p);
-
-      // Set the integration point in the face and the neighboring elements
-      Trans.SetAllIntPoints(&ip);
-
-      // Access the neighboring elements' integration points
-      // Note: eip2 will only contain valid data if Elem2 exists
-      const IntegrationPoint &eip1 = Trans.GetElement1IntPoint();
-      const IntegrationPoint &eip2 = Trans.GetElement2IntPoint();
-
+      IntegrationPoint eip1, eip2;
+      Trans.Loc1.Transform(ip, eip1);
+      if (ndof2)
+      {
+         Trans.Loc2.Transform(ip, eip2);
+      }
      el1.CalcShape(eip1, shape1);

+      Trans.Face->SetIntPoint(&ip);
+      Trans.Elem1->SetIntPoint(&eip1);
+
      u->Eval(vu, *Trans.Elem1, eip1);

      if (dim == 1)
@@ -2593,7 +2560,7 @@ void DGTraceIntegrator::AssembleFaceMatrix(const FiniteElement &el1,
      }
      else
      {
-         CalcOrtho(Trans.Jacobian(), nor);
+         CalcOrtho(Trans.Face->Jacobian(), nor);
      }

      un = vu * nor;
@@ -2608,6 +2575,7 @@ void DGTraceIntegrator::AssembleFaceMatrix(const FiniteElement &el1,
         double rho_p;
         if (un >= 0.0 && ndof2)
         {
+            Trans.Elem2->SetIntPoint(&eip2);
            rho_p = rho->Eval(*Trans.Elem2, eip2);
         }
         else
@@ -2723,7 +2691,7 @@ void DGDiffusionIntegrator::AssembleFaceMatrix(
      {
         order = 2*el1.GetOrder();
      }
-      ir = &IntRules.Get(Trans.GetGeometryType(), order);
+      ir = &IntRules.Get(Trans.FaceGeom, order);
   }

   // assemble: < {(Q \nabla u).n},[v] >      --> elmat
@@ -2731,26 +2699,22 @@ void DGDiffusionIntegrator::AssembleFaceMatrix(
   for (int p = 0; p < ir->GetNPoints(); p++)
   {
      const IntegrationPoint &ip = ir->IntPoint(p);
+      IntegrationPoint eip1, eip2;

-      // Set the integration point in the face and the neighboring elements
-      Trans.SetAllIntPoints(&ip);
-
-      // Access the neighboring elements' integration points
-      // Note: eip2 will only contain valid data if Elem2 exists
-      const IntegrationPoint &eip1 = Trans.GetElement1IntPoint();
-      const IntegrationPoint &eip2 = Trans.GetElement2IntPoint();
-
+      Trans.Loc1.Transform(ip, eip1);
+      Trans.Face->SetIntPoint(&ip);
      if (dim == 1)
      {
         nor(0) = 2*eip1.x - 1.0;
      }
      else
      {
-         CalcOrtho(Trans.Jacobian(), nor);
+         CalcOrtho(Trans.Face->Jacobian(), nor);
      }

      el1.CalcShape(eip1, shape1);
      el1.CalcDShape(eip1, dshape1);
+      Trans.Elem1->SetIntPoint(&eip1);
      w = ip.weight/Trans.Elem1->Weight();
      if (ndof2)
      {
@@ -2796,8 +2760,10 @@ void DGDiffusionIntegrator::AssembleFaceMatrix(

      if (ndof2)
      {
+         Trans.Loc2.Transform(ip, eip2);
         el2.CalcShape(eip2, shape2);
         el2.CalcDShape(eip2, dshape2);
+         Trans.Elem2->SetIntPoint(&eip2);
         w = ip.weight/2/Trans.Elem2->Weight();
         if (!MQ)
         {
@@ -3007,20 +2973,16 @@ void DGElasticityIntegrator::AssembleFaceMatrix(
   {
      // a simple choice for the integration order; is this OK?
      const int order = 2 * max(el1.GetOrder(), ndofs2 ? el2.GetOrder() : 0);
-      ir = &IntRules.Get(Trans.GetGeometryType(), order);
+      ir = &IntRules.Get(Trans.FaceGeom, order);
   }

   for (int pind = 0; pind < ir->GetNPoints(); ++pind)
   {
      const IntegrationPoint &ip = ir->IntPoint(pind);
-
-      // Set the integration point in the face and the neighboring elements
-      Trans.SetAllIntPoints(&ip);
-
-      // Access the neighboring elements' integration points
-      // Note: eip2 will only contain valid data if Elem2 exists
-      const IntegrationPoint &eip1 = Trans.GetElement1IntPoint();
-      const IntegrationPoint &eip2 = Trans.GetElement2IntPoint();
+      IntegrationPoint eip1, eip2; // integration point in the reference space
+      Trans.Loc1.Transform(ip, eip1);
+      Trans.Face->SetIntPoint(&ip);
+      Trans.Elem1->SetIntPoint(&eip1);

      el1.CalcShape(eip1, shape1);
      el1.CalcDShape(eip1, dshape1);
@@ -3034,12 +2996,14 @@ void DGElasticityIntegrator::AssembleFaceMatrix(
      }
      else
      {
-         CalcOrtho(Trans.Jacobian(), nor);
+         CalcOrtho(Trans.Face->Jacobian(), nor);
      }

      double w, wLM;
      if (ndofs2)
      {
+         Trans.Loc2.Transform(ip, eip2);
+         Trans.Elem2->SetIntPoint(&eip2);
         el2.CalcShape(eip2, shape2);
         el2.CalcDShape(eip2, dshape2);
         CalcAdjugate(Trans.Elem2->Jacobian(), adjJ);
@@ -3169,36 +3133,33 @@ void TraceJumpIntegrator::AssembleFaceMatrix(
      order += trial_face_fe.GetOrder();
      if (trial_face_fe.GetMapType() == FiniteElement::VALUE)
      {
-         order += Trans.OrderW();
+         order += Trans.Face->OrderW();
      }
-      ir = &IntRules.Get(Trans.GetGeometryType(), order);
+      ir = &IntRules.Get(Trans.FaceGeom, order);
   }

   for (int p = 0; p < ir->GetNPoints(); p++)
   {
      const IntegrationPoint &ip = ir->IntPoint(p);
-
-      // Set the integration point in the face and the neighboring elements
-      Trans.SetAllIntPoints(&ip);
-
-      // Access the neighboring elements' integration points
-      // Note: eip2 will only contain valid data if Elem2 exists
-      const IntegrationPoint &eip1 = Trans.GetElement1IntPoint();
-      const IntegrationPoint &eip2 = Trans.GetElement2IntPoint();
-
+      IntegrationPoint eip1, eip2;
      // Trace finite element shape function
+      Trans.Face->SetIntPoint(&ip);
      trial_face_fe.CalcShape(ip, face_shape);
      // Side 1 finite element shape function
+      Trans.Loc1.Transform(ip, eip1);
      test_fe1.CalcShape(eip1, shape1);
+      Trans.Elem1->SetIntPoint(&eip1);
      if (ndof2)
      {
         // Side 2 finite element shape function
+         Trans.Loc2.Transform(ip, eip2);
         test_fe2.CalcShape(eip2, shape2);
+         Trans.Elem2->SetIntPoint(&eip2);
      }
      w = ip.weight;
      if (trial_face_fe.GetMapType() == FiniteElement::VALUE)
      {
-         w *= Trans.Weight();
+         w *= Trans.Face->Weight();
      }
      face_shape *= w;
      for (i = 0; i < ndof1; i++)
@@ -3263,7 +3224,7 @@ void NormalTraceJumpIntegrator::AssembleFaceMatrix(
         order = test_fe1.GetOrder() - 1;
      }
      order += trial_face_fe.GetOrder();
-      ir = &IntRules.Get(Trans.GetGeometryType(), order);
+      ir = &IntRules.Get(Trans.FaceGeom, order);
   }

   for (int p = 0; p < ir->GetNPoints(); p++)
@@ -57,9 +57,6 @@ public:
   /// Assemble diagonal and add it to Vector @a diag.
   virtual void AssembleDiagonalPA(Vector &diag);

-   /// Assemble diagonal of ADA^T (A is this integrator) and add it to @a diag.
-   virtual void AssembleDiagonalPA_ADAt(const Vector &D, Vector &diag);
-
   /// Method for partially assembled action.
   /** Perform the action of integrator on the input @a x and add the result to
       the output @a y. Both @a x and @a y are E-vectors, i.e. they represent
@@ -78,22 +75,6 @@ public:
       called. */
   virtual void AddMultTransposePA(const Vector &x, Vector &y) const;

-   /// Method defining element assembly.
-   /** The result of the element assembly is added and stored in the @a emat
-       Vector. */
-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
-   /** Used with BilinearFormIntegrators that have different spaces. */
-   // virtual void AssembleEA(const FiniteElementSpace &trial_fes,
-   //                         const FiniteElementSpace &test_fes,
-   //                         Vector &emat);
-
-   virtual void AssembleEAInteriorFaces(const FiniteElementSpace &fes,
-                                        Vector &ea_data_int,
-                                        Vector &ea_data_ext);
-
-   virtual void AssembleEABoundaryFaces(const FiniteElementSpace &fes,
-                                        Vector &ea_data_bdr);
-
   /// Given a particular Finite Element computes the element matrix elmat.
   virtual void AssembleElementMatrix(const FiniteElement &el,
                                      ElementTransformation &Trans,
@@ -199,8 +180,6 @@ public:
   virtual ~BilinearFormIntegrator() { }
 };

-/** Wraps a given @a BilinearFormIntegrator and transposes the resulting element
-    matrices. See for example ex9, ex9p. */
 class TransposeIntegrator : public BilinearFormIntegrator
 {
 private:
@@ -255,15 +234,6 @@ public:
      bfi->AddMultTransposePA(x, y);
   }

-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
-
-   virtual void AssembleEAInteriorFaces(const FiniteElementSpace &fes,
-                                        Vector &ea_data_int,
-                                        Vector &ea_data_ext);
-
-   virtual void AssembleEABoundaryFaces(const FiniteElementSpace &fes,
-                                        Vector &ea_data_bdr);
-
   virtual ~TransposeIntegrator() { if (own_bfi) { delete bfi; } }
 };

@@ -1565,7 +1535,7 @@ public:
 };

 /** Class for integrating the bilinear form a(u,v) := (-V u, Grad v) in 2D or 3D
-    and where V is a vector coefficient, u is in H1 or L2 and v is in H1. */
+    and where V is a vector coefficient, u is in H1 and v is in H1. */
 class MixedScalarWeakDivergenceIntegrator : public MixedScalarVectorIntegrator
 {
 public:
@@ -1685,22 +1655,6 @@ protected:
   {
      trial_fe.CalcPhysCurlShape(Trans, shape);
   }
-
-   using BilinearFormIntegrator::AssemblePA;
-   virtual void AssemblePA(const FiniteElementSpace &trial_fes,
-                           const FiniteElementSpace &test_fes);
-
-   virtual void AddMultPA(const Vector&, Vector&) const;
-
-private:
-   // PA extension
-   Vector pa_data;
-   const DofToQuad *mapsO;         ///< Not owned. DOF-to-quad map, open.
-   const DofToQuad *mapsC;         ///< Not owned. DOF-to-quad map, closed.
-   const DofToQuad *mapsOtest;     ///< Not owned. DOF-to-quad map, open.
-   const DofToQuad *mapsCtest;     ///< Not owned. DOF-to-quad map, closed.
-   const GeometricFactors *geom;   ///< Not owned
-   int dim, ne, dofs1D, dofs1Dtest,quad1D, testType, trialType, coeffDim;
 };

 /** Class for integrating the bilinear form a(u,v) := (Q u, curl v) in 3D and
@@ -1740,20 +1694,6 @@ protected:
   {
      test_fe.CalcPhysCurlShape(Trans, shape);
   }
-
-   using BilinearFormIntegrator::AssemblePA;
-   virtual void AssemblePA(const FiniteElementSpace &trial_fes,
-                           const FiniteElementSpace &test_fes);
-
-   virtual void AddMultPA(const Vector&, Vector&) const;
-
-private:
-   // PA extension
-   Vector pa_data;
-   const DofToQuad *mapsO;         ///< Not owned. DOF-to-quad map, open.
-   const DofToQuad *mapsC;         ///< Not owned. DOF-to-quad map, closed.
-   const GeometricFactors *geom;   ///< Not owned
-   int dim, ne, dofs1D, quad1D, testType, trialType, coeffDim;
 };

 /** Class for integrating the bilinear form a(u,v) := - (Q u, grad v) in either
@@ -1945,8 +1885,6 @@ public:

   virtual void AssemblePA(const FiniteElementSpace &fes);

-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
-
   virtual void AssembleDiagonalPA(Vector &diag);

   virtual void AddMultPA(const Vector&, Vector&) const;
@@ -1954,7 +1892,7 @@ public:
   static const IntegrationRule &GetRule(const FiniteElement &trial_fe,
                                         const FiniteElement &test_fe);

-   void SetupPA(const FiniteElementSpace &fes);
+   void SetupPA(const FiniteElementSpace &fes, const bool force = false);
 };

 /** Class for local mass matrix assembling a(u,v) := (Q u, v) */
@@ -2020,8 +1958,6 @@ public:

   virtual void AssemblePA(const FiniteElementSpace &fes);

-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
-
   virtual void AssembleDiagonalPA(Vector &diag);

   virtual void AddMultPA(const Vector&, Vector&) const;
@@ -2030,10 +1966,9 @@ public:
                                         const FiniteElement &test_fe,
                                         ElementTransformation &Trans);

-   void SetupPA(const FiniteElementSpace &fes);
+   void SetupPA(const FiniteElementSpace &fes, const bool force = false);
 };

-/** Mass integrator (u, v) restricted to the boundary of a domain */
 class BoundaryMassIntegrator : public MassIntegrator
 {
 public:
@@ -2076,8 +2011,6 @@ public:

   virtual void AssemblePA(const FiniteElementSpace&);

-   virtual void AssembleEA(const FiniteElementSpace &fes, Vector &emat);
-
   virtual void AddMultPA(const Vector&, Vector&) const;

   static const IntegrationRule &GetRule(const FiniteElement &el,
@@ -2177,25 +2110,11 @@ class VectorFEDivergenceIntegrator : public BilinearFormIntegrator
 protected:
   Coefficient *Q;

-   using BilinearFormIntegrator::AssemblePA;
-   virtual void AssemblePA(const FiniteElementSpace &trial_fes,
-                           const FiniteElementSpace &test_fes);
-
-   virtual void AddMultPA(const Vector&, Vector&) const;
-   virtual void AddMultTransposePA(const Vector&, Vector&) const;
-
 private:
 #ifndef MFEM_THREAD_SAFE
   Vector divshape, shape;
 #endif

-   // PA extension
-   Vector pa_data;
-   const DofToQuad *mapsO;         ///< Not owned. DOF-to-quad map, open.
-   const DofToQuad *L2mapsO;       ///< Not owned. DOF-to-quad map, open.
-   const DofToQuad *mapsC;         ///< Not owned. DOF-to-quad map, closed.
-   int dim, ne, dofs1D, L2dofs1D, quad1D;
-
 public:
   VectorFEDivergenceIntegrator() { Q = NULL; }
   VectorFEDivergenceIntegrator(Coefficient &q) { Q = &q; }
@@ -2206,8 +2125,6 @@ public:
                                       const FiniteElement &test_fe,
                                       ElementTransformation &Trans,
                                       DenseMatrix &elmat);
-
-   virtual void AssembleDiagonalPA_ADAt(const Vector &D, Vector &diag);
 };


@@ -2391,7 +2308,7 @@ protected:
   const DofToQuad *mapsO;         ///< Not owned. DOF-to-quad map, open.
   const DofToQuad *mapsC;         ///< Not owned. DOF-to-quad map, closed.
   const GeometricFactors *geom;   ///< Not owned
-   int dim, ne, nq, dofs1D, quad1D, fetype;
+   int dim, ne, nq, dofs1D, quad1D;

 public:
   VectorFEMassIntegrator() { Init(NULL, NULL, NULL); }
@@ -2470,23 +2387,11 @@ class DivDivIntegrator: public BilinearFormIntegrator
 protected:
   Coefficient *Q;

-   using BilinearFormIntegrator::AssemblePA;
-   virtual void AssemblePA(const FiniteElementSpace &fes);
-   virtual void AddMultPA(const Vector &x, Vector &y) const;
-   virtual void AssembleDiagonalPA(Vector& diag);
-
 private:
 #ifndef MFEM_THREAD_SAFE
   Vector divshape;
 #endif

-   // PA extension
-   Vector pa_data;
-   const DofToQuad *mapsO;         ///< Not owned. DOF-to-quad map, open.
-   const DofToQuad *mapsC;         ///< Not owned. DOF-to-quad map, closed.
-   const GeometricFactors *geom;   ///< Not owned
-   int dim, ne, dofs1D, quad1D;
-
 public:
   DivDivIntegrator() { Q = NULL; }
   DivDivIntegrator(Coefficient &q) : Q(&q) { }
@@ -2639,13 +2544,6 @@ public:

   virtual void AddMultPA(const Vector&, Vector&) const;

-   virtual void AssembleEAInteriorFaces(const FiniteElementSpace& fes,
-                                        Vector &ea_data_int,
-                                        Vector &ea_data_ext);
-
-   virtual void AssembleEABoundaryFaces(const FiniteElementSpace& fes,
-                                        Vector &ea_data_bdr);
-
   static const IntegrationRule &GetRule(Geometry::Type geom, int order,
                                         FaceElementTransformations &T);

@@ -788,20 +788,6 @@ void ConvectionIntegrator::AssemblePA(const FiniteElementSpace &fes)
   {
      vel = cQ->GetVec();
   }
-   else if (VectorQuadratureFunctionCoefficient* cQ =
-               dynamic_cast<VectorQuadratureFunctionCoefficient*>(Q))
-   {
-      const QuadratureFunction &qFun = cQ->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);
@@ -1,258 +0,0 @@
-// Copyright (c) 2010-2020, 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.
-
-#include "../general/forall.hpp"
-#include "bilininteg.hpp"
-#include "gridfunc.hpp"
-
-namespace mfem
-{
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EAConvectionAssemble1D(const int NE,
-                                   const Array<double> &b,
-                                   const Array<double> &g,
-                                   const Vector &padata,
-                                   Vector &eadata,
-                                   const int d1d = 0,
-                                   const int q1d = 0)
-{
-   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 G = Reshape(g.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, NE);
-   auto A = Reshape(eadata.ReadWrite(), D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, 1,
-   {
-      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;
-      double r_Gi[MQ1];
-      double r_Bj[MQ1];
-      for (int q = 0; q < Q1D; q++)
-      {
-         r_Gi[q] = G(q,MFEM_THREAD_ID(x));
-         r_Bj[q] = B(q,MFEM_THREAD_ID(y));
-      }
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(j1,y,D1D)
-         {
-            double val = 0.0;
-            for (int k1 = 0; k1 < Q1D; ++k1)
-            {
-               val += r_Bj[k1] * D(k1, e) * r_Gi[k1];
-            }
-            A(i1, j1, e) += val;
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EAConvectionAssemble2D(const int NE,
-                                   const Array<double> &b,
-                                   const Array<double> &g,
-                                   const Vector &padata,
-                                   Vector &eadata,
-                                   const int d1d = 0,
-                                   const int q1d = 0)
-{
-   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 G = Reshape(g.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, Q1D, 2, NE);
-   auto A = Reshape(eadata.ReadWrite(), D1D, D1D, D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, 1,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-      double r_B[MQ1][MD1];
-      double r_G[MQ1][MD1];
-      for (int d = 0; d < D1D; d++)
-      {
-         for (int q = 0; q < Q1D; q++)
-         {
-            r_B[q][d] = B(q,d);
-            r_G[q][d] = G(q,d);
-         }
-      }
-      MFEM_SHARED double s_D[MQ1][MQ1][2];
-      MFEM_FOREACH_THREAD(k1,x,Q1D)
-      {
-         MFEM_FOREACH_THREAD(k2,y,Q1D)
-         {
-            s_D[k1][k2][0] = D(k1,k2,0,e);
-            s_D[k1][k2][1] = D(k1,k2,1,e);
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(i2,y,D1D)
-         {
-            for (int j1 = 0; j1 < D1D; ++j1)
-            {
-               for (int j2 = 0; j2 < D1D; ++j2)
-               {
-                  double val = 0.0;
-                  for (int k1 = 0; k1 < Q1D; ++k1)
-                  {
-                     for (int k2 = 0; k2 < Q1D; ++k2)
-                     {
-                        val += (r_G[k1][i1] * r_B[k2][i2] * s_D[k1][k2][0]
-                                + r_B[k1][i1] * r_G[k2][i2] * s_D[k1][k2][1])
-                               * r_B[k1][j1]* r_B[k2][j2];
-                     }
-                  }
-                  A(i1, i2, j1, j2, e) += val;
-               }
-            }
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EAConvectionAssemble3D(const int NE,
-                                   const Array<double> &b,
-                                   const Array<double> &g,
-                                   const Vector &padata,
-                                   Vector &eadata,
-                                   const int d1d = 0,
-                                   const int q1d = 0)
-{
-   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 G = Reshape(g.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, Q1D, Q1D, 3, NE);
-   auto A = Reshape(eadata.ReadWrite(), D1D, D1D, D1D, D1D, D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, D1D,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-      double r_B[MQ1][MD1];
-      double r_G[MQ1][MD1];
-      for (int d = 0; d < D1D; d++)
-      {
-         for (int q = 0; q < Q1D; q++)
-         {
-            r_B[q][d] = B(q,d);
-            r_G[q][d] = G(q,d);
-         }
-      }
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(i2,y,D1D)
-         {
-            MFEM_FOREACH_THREAD(i3,z,D1D)
-            {
-               for (int j1 = 0; j1 < D1D; ++j1)
-               {
-                  for (int j2 = 0; j2 < D1D; ++j2)
-                  {
-                     for (int j3 = 0; j3 < D1D; ++j3)
-                     {
-                        double val = 0.0;
-                        for (int k1 = 0; k1 < Q1D; ++k1)
-                        {
-                           for (int k2 = 0; k2 < Q1D; ++k2)
-                           {
-                              for (int k3 = 0; k3 < Q1D; ++k3)
-                              {
-                                 double D0 = D(k1,k2,k3,0,e);
-                                 double D1 = D(k1,k2,k3,1,e);
-                                 double D2 = D(k1,k2,k3,2,e);
-                                 val += (r_G[k1][i1] * r_B[k2][i2] * r_B[k3][i3] * D0
-                                         + r_B[k1][i1] * r_G[k2][i2] * r_B[k3][i3] * D1
-                                         + r_B[k1][i1] * r_B[k2][i2] * r_G[k3][i3] * D2)
-                                        * r_B[k1][j1] * r_B[k2][j2] * r_B[k3][j3];
-                              }
-                           }
-                        }
-                        A(i1, i2, i3, j1, j2, j3, e) += val;
-                     }
-                  }
-               }
-            }
-         }
-      }
-   });
-}
-
-void ConvectionIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                      Vector &ea_data)
-{
-   AssemblePA(fes);
-   const int ne = fes.GetMesh()->GetNE();
-   const Array<double> &B = maps->B;
-   const Array<double> &G = maps->G;
-   if (dim == 1)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x22: return EAConvectionAssemble1D<2,2>(ne,B,G,pa_data,ea_data);
-         case 0x33: return EAConvectionAssemble1D<3,3>(ne,B,G,pa_data,ea_data);
-         case 0x44: return EAConvectionAssemble1D<4,4>(ne,B,G,pa_data,ea_data);
-         case 0x55: return EAConvectionAssemble1D<5,5>(ne,B,G,pa_data,ea_data);
-         case 0x66: return EAConvectionAssemble1D<6,6>(ne,B,G,pa_data,ea_data);
-         case 0x77: return EAConvectionAssemble1D<7,7>(ne,B,G,pa_data,ea_data);
-         case 0x88: return EAConvectionAssemble1D<8,8>(ne,B,G,pa_data,ea_data);
-         case 0x99: return EAConvectionAssemble1D<9,9>(ne,B,G,pa_data,ea_data);
-         default:   return EAConvectionAssemble1D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   else if (dim == 2)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x22: return EAConvectionAssemble2D<2,2>(ne,B,G,pa_data,ea_data);
-         case 0x33: return EAConvectionAssemble2D<3,3>(ne,B,G,pa_data,ea_data);
-         case 0x44: return EAConvectionAssemble2D<4,4>(ne,B,G,pa_data,ea_data);
-         case 0x55: return EAConvectionAssemble2D<5,5>(ne,B,G,pa_data,ea_data);
-         case 0x66: return EAConvectionAssemble2D<6,6>(ne,B,G,pa_data,ea_data);
-         case 0x77: return EAConvectionAssemble2D<7,7>(ne,B,G,pa_data,ea_data);
-         case 0x88: return EAConvectionAssemble2D<8,8>(ne,B,G,pa_data,ea_data);
-         case 0x99: return EAConvectionAssemble2D<9,9>(ne,B,G,pa_data,ea_data);
-         default:   return EAConvectionAssemble2D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   else if (dim == 3)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x23: return EAConvectionAssemble3D<2,3>(ne,B,G,pa_data,ea_data);
-         case 0x34: return EAConvectionAssemble3D<3,4>(ne,B,G,pa_data,ea_data);
-         case 0x45: return EAConvectionAssemble3D<4,5>(ne,B,G,pa_data,ea_data);
-         case 0x56: return EAConvectionAssemble3D<5,6>(ne,B,G,pa_data,ea_data);
-         case 0x67: return EAConvectionAssemble3D<6,7>(ne,B,G,pa_data,ea_data);
-         case 0x78: return EAConvectionAssemble3D<7,8>(ne,B,G,pa_data,ea_data);
-         case 0x89: return EAConvectionAssemble3D<8,9>(ne,B,G,pa_data,ea_data);
-         default:   return EAConvectionAssemble3D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   MFEM_ABORT("Unknown kernel.");
-}
-
-}
@@ -167,19 +167,6 @@ void DGTraceIntegrator::SetupPA(const FiniteElementSpace &fes, FaceType type)
      r.SetSize(1);
      r(0) = c_rho->constant;
   }
-   else if (QuadratureFunctionCoefficient* c_rho =
-               dynamic_cast<QuadratureFunctionCoefficient*>(rho))
-   {
-      const QuadratureFunction &qFun = c_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
   {
      r.SetSize(nq * nf);
@@ -213,20 +200,6 @@ void DGTraceIntegrator::SetupPA(const FiniteElementSpace &fes, FaceType type)
   {
      vel = c_u->GetVec();
   }
-   else if (VectorQuadratureFunctionCoefficient* c_u =
-               dynamic_cast<VectorQuadratureFunctionCoefficient*>(u))
-   {
-      // Assumed to be in lexicographical ordering
-      const QuadratureFunction &qFun = c_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);
@@ -1,414 +0,0 @@
-// Copyright (c) 2010-2020, 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.
-
-#include "../general/forall.hpp"
-#include "bilininteg.hpp"
-#include "gridfunc.hpp"
-
-namespace mfem
-{
-
-static void EADGTraceAssemble1DInt(const int NF,
-                                   const Array<double> &basis,
-                                   const Vector &padata,
-                                   Vector &eadata_int,
-                                   Vector &eadata_ext)
-{
-   auto D = Reshape(padata.Read(), 2, 2, NF);
-   auto A_int = Reshape(eadata_int.ReadWrite(), 2, NF);
-   auto A_ext = Reshape(eadata_ext.ReadWrite(), 2, NF);
-   MFEM_FORALL(f, NF,
-   {
-      double val_int0, val_int1, val_ext01, val_ext10;
-      val_int0  = D(0, 0, f);
-      val_ext10 = D(1, 0, f);
-      val_ext01 = D(0, 1, f);
-      val_int1  = D(1, 1, f);
-      A_int(0, f) += val_int0;
-      A_int(1, f) += val_int1;
-      A_ext(0, f) += val_ext01;
-      A_ext(1, f) += val_ext10;
-   });
-}
-
-static void EADGTraceAssemble1DBdr(const int NF,
-                                   const Array<double> &basis,
-                                   const Vector &padata,
-                                   Vector &eadata_bdr)
-{
-   auto D = Reshape(padata.Read(), 2, 2, NF);
-   auto A_bdr = Reshape(eadata_bdr.ReadWrite(), NF);
-   MFEM_FORALL(f, NF,
-   {
-      A_bdr(f) += D(0, 0, f);
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EADGTraceAssemble2DInt(const int NF,
-                                   const Array<double> &basis,
-                                   const Vector &padata,
-                                   Vector &eadata_int,
-                                   Vector &eadata_ext,
-                                   const int d1d = 0,
-                                   const int q1d = 0)
-{
-   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(basis.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, 2, 2, NF);
-   auto A_int = Reshape(eadata_int.ReadWrite(), D1D, D1D, 2, NF);
-   auto A_ext = Reshape(eadata_ext.ReadWrite(), D1D, D1D, 2, NF);
-   MFEM_FORALL_3D(f, NF, D1D, D1D, 1,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(j1,y,D1D)
-         {
-            double val_int0 = 0.0;
-            double val_int1 = 0.0;
-            double val_ext01 = 0.0;
-            double val_ext10 = 0.0;
-            for (int k1 = 0; k1 < Q1D; ++k1)
-            {
-               val_int0  += B(k1,i1) * B(k1,j1) * D(k1, 0, 0, f);
-               val_ext01 += B(k1,i1) * B(k1,j1) * D(k1, 0, 1, f);
-               val_ext10 += B(k1,i1) * B(k1,j1) * D(k1, 1, 0, f);
-               val_int1  += B(k1,i1) * B(k1,j1) * D(k1, 1, 1, f);
-            }
-            A_int(i1, j1, 0, f) += val_int0;
-            A_int(i1, j1, 1, f) += val_int1;
-            A_ext(i1, j1, 0, f) += val_ext01;
-            A_ext(i1, j1, 1, f) += val_ext10;
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EADGTraceAssemble2DBdr(const int NF,
-                                   const Array<double> &basis,
-                                   const Vector &padata,
-                                   Vector &eadata_bdr,
-                                   const int d1d = 0,
-                                   const int q1d = 0)
-{
-   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(basis.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, 2, 2, NF);
-   auto A_bdr = Reshape(eadata_bdr.ReadWrite(), D1D, D1D, NF);
-   MFEM_FORALL_3D(f, NF, D1D, D1D, 1,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(j1,y,D1D)
-         {
-            double val_bdr = 0.0;
-            for (int k1 = 0; k1 < Q1D; ++k1)
-            {
-               val_bdr  += B(k1,i1) * B(k1,j1) * D(k1, 0, 0, f);
-            }
-            A_bdr(i1, j1, f) += val_bdr;
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EADGTraceAssemble3DInt(const int NF,
-                                   const Array<double> &basis,
-                                   const Vector &padata,
-                                   Vector &eadata_int,
-                                   Vector &eadata_ext,
-                                   const int d1d = 0,
-                                   const int q1d = 0)
-{
-   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(basis.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, Q1D, 2, 2, NF);
-   auto A_int = Reshape(eadata_int.ReadWrite(), D1D, D1D, D1D, D1D, 2, NF);
-   auto A_ext = Reshape(eadata_ext.ReadWrite(), D1D, D1D, D1D, D1D, 2, NF);
-   MFEM_FORALL_3D(f, NF, D1D, D1D, 1,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-      double r_B[MQ1][MD1];
-      for (int d = 0; d < D1D; d++)
-      {
-         for (int q = 0; q < Q1D; q++)
-         {
-            r_B[q][d] = B(q,d);
-         }
-      }
-      MFEM_SHARED double s_D[MQ1][MQ1][2][2];
-      for (int i=0; i < 2; i++)
-      {
-         for (int j=0; j < 2; j++)
-         {
-            MFEM_FOREACH_THREAD(k1,x,Q1D)
-            {
-               MFEM_FOREACH_THREAD(k2,y,Q1D)
-               {
-                  s_D[k1][k2][i][j] = D(k1,k2,i,j,f);
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(i2,y,D1D)
-         {
-            for (int j1 = 0; j1 < D1D; ++j1)
-            {
-               for (int j2 = 0; j2 < D1D; ++j2)
-               {
-                  double val_int0 = 0.0;
-                  double val_int1 = 0.0;
-                  double val_ext01 = 0.0;
-                  double val_ext10 = 0.0;
-                  for (int k1 = 0; k1 < Q1D; ++k1)
-                  {
-                     for (int k2 = 0; k2 < Q1D; ++k2)
-                     {
-                        val_int0 += r_B[k1][i1] * r_B[k1][j1]
-                                    * r_B[k2][i2] * r_B[k2][j2]
-                                    * s_D[k1][k2][0][0];
-                        val_int1 += r_B[k1][i1] * r_B[k1][j1]
-                                    * r_B[k2][i2] * r_B[k2][j2]
-                                    * s_D[k1][k2][1][1];
-                        val_ext01+= r_B[k1][i1] * r_B[k1][j1]
-                                    * r_B[k2][i2] * r_B[k2][j2]
-                                    * s_D[k1][k2][0][1];
-                        val_ext10+= r_B[k1][i1] * r_B[k1][j1]
-                                    * r_B[k2][i2] * r_B[k2][j2]
-                                    * s_D[k1][k2][1][0];
-                     }
-                  }
-                  A_int(i1, i2, j1, j2, 0, f) += val_int0;
-                  A_int(i1, i2, j1, j2, 1, f) += val_int1;
-                  A_ext(i1, i2, j1, j2, 0, f) += val_ext01;
-                  A_ext(i1, i2, j1, j2, 1, f) += val_ext10;
-               }
-            }
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EADGTraceAssemble3DBdr(const int NF,
-                                   const Array<double> &basis,
-                                   const Vector &padata,
-                                   Vector &eadata_bdr,
-                                   const int d1d = 0,
-                                   const int q1d = 0)
-{
-   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(basis.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, Q1D, 2, 2, NF);
-   auto A_bdr = Reshape(eadata_bdr.ReadWrite(), D1D, D1D, D1D, D1D, NF);
-   MFEM_FORALL_3D(f, NF, D1D, D1D, 1,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-      double r_B[MQ1][MD1];
-      for (int d = 0; d < D1D; d++)
-      {
-         for (int q = 0; q < Q1D; q++)
-         {
-            r_B[q][d] = B(q,d);
-         }
-      }
-      MFEM_SHARED double s_D[MQ1][MQ1][2][2];
-      for (int i=0; i < 2; i++)
-      {
-         for (int j=0; j < 2; j++)
-         {
-            MFEM_FOREACH_THREAD(k1,x,Q1D)
-            {
-               MFEM_FOREACH_THREAD(k2,y,Q1D)
-               {
-                  s_D[k1][k2][i][j] = D(k1,k2,i,j,f);
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(i2,y,D1D)
-         {
-            for (int j1 = 0; j1 < D1D; ++j1)
-            {
-               for (int j2 = 0; j2 < D1D; ++j2)
-               {
-                  double val_bdr = 0.0;
-                  for (int k1 = 0; k1 < Q1D; ++k1)
-                  {
-                     for (int k2 = 0; k2 < Q1D; ++k2)
-                     {
-                        val_bdr += r_B[k1][i1] * r_B[k1][j1]
-                                   * r_B[k2][i2] * r_B[k2][j2]
-                                   * s_D[k1][k2][0][0];
-                     }
-                  }
-                  A_bdr(i1, i2, j1, j2, f) += val_bdr;
-               }
-            }
-         }
-      }
-   });
-}
-
-void DGTraceIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace& fes,
-                                                Vector &ea_data_int,
-                                                Vector &ea_data_ext)
-{
-   SetupPA(fes, FaceType::Interior);
-   nf = fes.GetNFbyType(FaceType::Interior);
-   if (nf==0) { return; }
-   const Array<double> &B = maps->B;
-   if (dim == 1)
-   {
-      return EADGTraceAssemble1DInt(nf,B,pa_data,ea_data_int,ea_data_ext);
-   }
-   else if (dim == 2)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x22:
-            return EADGTraceAssemble2DInt<2,2>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x33:
-            return EADGTraceAssemble2DInt<3,3>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x44:
-            return EADGTraceAssemble2DInt<4,4>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x55:
-            return EADGTraceAssemble2DInt<5,5>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x66:
-            return EADGTraceAssemble2DInt<6,6>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x77:
-            return EADGTraceAssemble2DInt<7,7>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x88:
-            return EADGTraceAssemble2DInt<8,8>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x99:
-            return EADGTraceAssemble2DInt<9,9>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         default:
-            return EADGTraceAssemble2DInt(nf,B,pa_data,ea_data_int,
-                                          ea_data_ext,dofs1D,quad1D);
-      }
-   }
-   else if (dim == 3)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x23:
-            return EADGTraceAssemble3DInt<2,3>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x34:
-            return EADGTraceAssemble3DInt<3,4>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x45:
-            return EADGTraceAssemble3DInt<4,5>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x56:
-            return EADGTraceAssemble3DInt<5,6>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x67:
-            return EADGTraceAssemble3DInt<6,7>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x78:
-            return EADGTraceAssemble3DInt<7,8>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         case 0x89:
-            return EADGTraceAssemble3DInt<8,9>(nf,B,pa_data,ea_data_int,
-                                               ea_data_ext);
-         default:
-            return EADGTraceAssemble3DInt(nf,B,pa_data,ea_data_int,
-                                          ea_data_ext,dofs1D,quad1D);
-      }
-   }
-   MFEM_ABORT("Unknown kernel.");
-}
-
-void DGTraceIntegrator::AssembleEABoundaryFaces(const FiniteElementSpace& fes,
-                                                Vector &ea_data_bdr)
-{
-   SetupPA(fes, FaceType::Boundary);
-   nf = fes.GetNFbyType(FaceType::Boundary);
-   if (nf==0) { return; }
-   const Array<double> &B = maps->B;
-   if (dim == 1)
-   {
-      return EADGTraceAssemble1DBdr(nf,B,pa_data,ea_data_bdr);
-   }
-   else if (dim == 2)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x22: return EADGTraceAssemble2DBdr<2,2>(nf,B,pa_data,ea_data_bdr);
-         case 0x33: return EADGTraceAssemble2DBdr<3,3>(nf,B,pa_data,ea_data_bdr);
-         case 0x44: return EADGTraceAssemble2DBdr<4,4>(nf,B,pa_data,ea_data_bdr);
-         case 0x55: return EADGTraceAssemble2DBdr<5,5>(nf,B,pa_data,ea_data_bdr);
-         case 0x66: return EADGTraceAssemble2DBdr<6,6>(nf,B,pa_data,ea_data_bdr);
-         case 0x77: return EADGTraceAssemble2DBdr<7,7>(nf,B,pa_data,ea_data_bdr);
-         case 0x88: return EADGTraceAssemble2DBdr<8,8>(nf,B,pa_data,ea_data_bdr);
-         case 0x99: return EADGTraceAssemble2DBdr<9,9>(nf,B,pa_data,ea_data_bdr);
-         default:
-            return EADGTraceAssemble2DBdr(nf,B,pa_data,ea_data_bdr,dofs1D,quad1D);
-      }
-   }
-   else if (dim == 3)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x23: return EADGTraceAssemble3DBdr<2,3>(nf,B,pa_data,ea_data_bdr);
-         case 0x34: return EADGTraceAssemble3DBdr<3,4>(nf,B,pa_data,ea_data_bdr);
-         case 0x45: return EADGTraceAssemble3DBdr<4,5>(nf,B,pa_data,ea_data_bdr);
-         case 0x56: return EADGTraceAssemble3DBdr<5,6>(nf,B,pa_data,ea_data_bdr);
-         case 0x67: return EADGTraceAssemble3DBdr<6,7>(nf,B,pa_data,ea_data_bdr);
-         case 0x78: return EADGTraceAssemble3DBdr<7,8>(nf,B,pa_data,ea_data_bdr);
-         case 0x89: return EADGTraceAssemble3DBdr<8,9>(nf,B,pa_data,ea_data_bdr);
-         default:
-            return EADGTraceAssemble3DBdr(nf,B,pa_data,ea_data_bdr,dofs1D,quad1D);
-      }
-   }
-   MFEM_ABORT("Unknown kernel.");
-}
-
-}
@@ -253,7 +253,8 @@ static void PADiffusionSetup(const int dim,
   }
 }

-void DiffusionIntegrator::SetupPA(const FiniteElementSpace &fes)
+void DiffusionIntegrator::SetupPA(const FiniteElementSpace &fes,
+                                  const bool force)
 {
   // Assuming the same element type
   fespace = &fes;
@@ -262,7 +263,7 @@ void DiffusionIntegrator::SetupPA(const FiniteElementSpace &fes)
   const FiniteElement &el = *fes.GetFE(0);
   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el);
 #ifdef MFEM_USE_CEED
-   if (DeviceCanUseCeed())
+   if (DeviceCanUseCeed() && !force)
   {
      if (ceedDataPtr) { delete ceedDataPtr; }
      CeedData* ptr = new CeedData();
@@ -270,6 +271,8 @@ void DiffusionIntegrator::SetupPA(const FiniteElementSpace &fes)
      InitCeedCoeff(Q, ptr);
      return CeedPADiffusionAssemble(fes, *ir, *ptr);
   }
+#else
+   MFEM_CONTRACT_VAR(force);
 #endif
   const int dims = el.GetDim();
   const int symmDims = (dims * (dims + 1)) / 2; // 1x1: 1, 2x2: 3, 3x3: 6
@@ -293,19 +296,6 @@ void DiffusionIntegrator::SetupPA(const FiniteElementSpace &fes)
      coeff.SetSize(1);
      coeff(0) = cQ->constant;
   }
-   else if (QuadratureFunctionCoefficient* cQ =
-               dynamic_cast<QuadratureFunctionCoefficient*>(Q))
-   {
-      const QuadratureFunction &qFun = cQ->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);
@@ -746,17 +736,9 @@ static void PADiffusionAssembleDiagonal(const int dim,

 void DiffusionIntegrator::AssembleDiagonalPA(Vector &diag)
 {
-#ifdef MFEM_USE_CEED
-   if (DeviceCanUseCeed())
-   {
-      CeedAssembleDiagonalPA(ceedDataPtr, diag);
-   }
-   else
-#endif
-   {
-      PADiffusionAssembleDiagonal(dim, dofs1D, quad1D, ne,
-                                  maps->B, maps->G, pa_data, diag);
-   }
+   if (pa_data.Size()==0) { SetupPA(*fespace, true); }
+   PADiffusionAssembleDiagonal(dim, dofs1D, quad1D, ne,
+                               maps->B, maps->G, pa_data, diag);
 }


@@ -1325,33 +1307,7 @@ static void PADiffusionApply3D(const int NE,
   });
 }

-// Half of B and G are stored in shared to get B, Bt, G and Gt.
-// Indices computation for SmemPADiffusionApply3D.
-static MFEM_HOST_DEVICE inline int qi(const int q, const int d, const int Q)
-{
-   return (q<=d) ? q : Q-1-q;
-}
-
-static MFEM_HOST_DEVICE inline int dj(const int q, const int d, const int D)
-{
-   return (q<=d) ? d : D-1-d;
-}
-
-static MFEM_HOST_DEVICE inline int qk(const int q, const int d, const int Q)
-{
-   return (q<=d) ? Q-1-q : q;
-}
-
-static MFEM_HOST_DEVICE inline int dl(const int q, const int d, const int D)
-{
-   return (q<=d) ? D-1-d : d;
-}
-
-static MFEM_HOST_DEVICE inline double sign(const int q, const int d)
-{
-   return (q<=d) ? -1.0 : 1.0;
-}
-
+// Shared memory PA Diffusion Apply 3D kernel
 template<int T_D1D = 0, int T_Q1D = 0>
 static void SmemPADiffusionApply3D(const int NE,
                                   const Array<double> &b_,
@@ -1364,27 +1320,28 @@ static void SmemPADiffusionApply3D(const int NE,
 {
   const int D1D = T_D1D ? T_D1D : d1d;
   const int Q1D = T_Q1D ? T_Q1D : q1d;
-   constexpr int M1Q = T_Q1D ? T_Q1D : MAX_Q1D;
-   constexpr int M1D = T_D1D ? T_D1D : MAX_D1D;
-   MFEM_VERIFY(D1D <= M1D, "");
-   MFEM_VERIFY(Q1D <= M1Q, "");
+   constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
+   constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
+   MFEM_VERIFY(D1D <= MD1, "");
+   MFEM_VERIFY(Q1D <= MQ1, "");
   auto b = Reshape(b_.Read(), Q1D, D1D);
   auto g = Reshape(g_.Read(), Q1D, D1D);
-   auto d = Reshape(d_.Read(), Q1D, Q1D, Q1D, 6, NE);
+   auto d = Reshape(d_.Read(), Q1D*Q1D*Q1D, 6, 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,
+   MFEM_FORALL_3D(e, NE, Q1D, Q1D, 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 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 sBG[MQ1*MD1];
-      double (*B)[MD1] = (double (*)[MD1]) sBG;
-      double (*G)[MD1] = (double (*)[MD1]) sBG;
-      double (*Bt)[MQ1] = (double (*)[MQ1]) sBG;
-      double (*Gt)[MQ1] = (double (*)[MQ1]) sBG;
+      constexpr int MDQ = MQ1 > MD1 ? MQ1 : MD1;
+      MFEM_SHARED double sBG[2][MQ1*MD1];
+      double (*B)[MD1] = (double (*)[MD1]) (sBG+0);
+      double (*G)[MD1] = (double (*)[MD1]) (sBG+1);
+      double (*Bt)[MQ1] = (double (*)[MQ1]) (sBG+0);
+      double (*Gt)[MQ1] = (double (*)[MQ1]) (sBG+1);
      MFEM_SHARED double sm0[3][MDQ*MDQ*MDQ];
      MFEM_SHARED double sm1[3][MDQ*MDQ*MDQ];
      double (*X)[MD1][MD1]    = (double (*)[MD1][MD1]) (sm0+2);
@@ -1402,127 +1359,108 @@ static void SmemPADiffusionApply3D(const int NE,
      double (*QDD0)[MD1][MD1] = (double (*)[MD1][MD1]) (sm0+0);
      double (*QDD1)[MD1][MD1] = (double (*)[MD1][MD1]) (sm0+1);
      double (*QDD2)[MD1][MD1] = (double (*)[MD1][MD1]) (sm0+2);
-      MFEM_FOREACH_THREAD(dy,y,D1D)
+      MFEM_FOREACH_THREAD(dz,z,D1D)
      {
-         MFEM_FOREACH_THREAD(dx,x,D1D)
+         MFEM_FOREACH_THREAD(dy,y,D1D)
         {
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; ++dz)
+            MFEM_FOREACH_THREAD(dx,x,D1D)
            {
               X[dz][dy][dx] = x(dx,dy,dz,e);
            }
         }
-         MFEM_FOREACH_THREAD(qx,x,Q1D)
+      }
+      if (tidz == 0)
+      {
+         MFEM_FOREACH_THREAD(d,y,D1D)
         {
-            const int i = qi(qx,dy,Q1D);
-            const int j = dj(qx,dy,D1D);
-            const int k = qk(qx,dy,Q1D);
-            const int l = dl(qx,dy,D1D);
-            B[i][j] = b(qx,dy);
-            G[k][l] = g(qx,dy) * sign(qx,dy);
+            MFEM_FOREACH_THREAD(q,x,Q1D)
+            {
+               B[q][d] = b(q,d);
+               G[q][d] = g(q,d);
+            }
         }
      }
      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(dy,y,D1D)
+      MFEM_FOREACH_THREAD(dz,z,D1D)
      {
-         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         MFEM_FOREACH_THREAD(dy,y,D1D)
         {
-            double u[D1D], v[D1D];
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; dz++) { u[dz] = v[dz] = 0.0; }
-            MFEM_UNROLL(MD1)
-            for (int dx = 0; dx < D1D; ++dx)
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
            {
-               const int i = qi(qx,dx,Q1D);
-               const int j = dj(qx,dx,D1D);
-               const int k = qk(qx,dx,Q1D);
-               const int l = dl(qx,dx,D1D);
-               const double s = sign(qx,dx);
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < D1D; ++dz)
+               double u = 0.0;
+               double v = 0.0;
+               for (int dx = 0; dx < D1D; ++dx)
               {
                  const double coords = X[dz][dy][dx];
-                  u[dz] += coords * B[i][j];
-                  v[dz] += coords * G[k][l] * s;
+                  u += coords * B[qx][dx];
+                  v += coords * G[qx][dx];
               }
-            }
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; ++dz)
-            {
-               DDQ0[dz][dy][qx] = u[dz];
-               DDQ1[dz][dy][qx] = v[dz];
+               DDQ0[dz][dy][qx] = u;
+               DDQ1[dz][dy][qx] = v;
            }
         }
      }
      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      MFEM_FOREACH_THREAD(dz,z,D1D)
      {
-         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
         {
-            double u[D1D], v[D1D], w[D1D];
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; dz++) { u[dz] = v[dz] = w[dz] = 0.0; }
-            MFEM_UNROLL(MD1)
-            for (int dy = 0; dy < D1D; ++dy)
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
            {
-               const int i = qi(qy,dy,Q1D);
-               const int j = dj(qy,dy,D1D);
-               const int k = qk(qy,dy,Q1D);
-               const int l = dl(qy,dy,D1D);
-               const double s = sign(qy,dy);
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < D1D; dz++)
+               double u = 0.0;
+               double v = 0.0;
+               double w = 0.0;
+               for (int dy = 0; dy < D1D; ++dy)
               {
-                  u[dz] += DDQ1[dz][dy][qx] * B[i][j];
-                  v[dz] += DDQ0[dz][dy][qx] * G[k][l] * s;
-                  w[dz] += DDQ0[dz][dy][qx] * B[i][j];
+                  u += DDQ1[dz][dy][qx] * B[qy][dy];
+                  v += DDQ0[dz][dy][qx] * G[qy][dy];
+                  w += DDQ0[dz][dy][qx] * B[qy][dy];
               }
-            }
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; dz++)
-            {
-               DQQ0[dz][qy][qx] = u[dz];
-               DQQ1[dz][qy][qx] = v[dz];
-               DQQ2[dz][qy][qx] = w[dz];
+               DQQ0[dz][qy][qx] = u;
+               DQQ1[dz][qy][qx] = v;
+               DQQ2[dz][qy][qx] = w;
            }
         }
      }
      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      MFEM_FOREACH_THREAD(qz,z,Q1D)
      {
-         MFEM_FOREACH_THREAD(qx,x,Q1D)
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
         {
-            double u[Q1D], v[Q1D], w[Q1D];
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; qz++) { u[qz] = v[qz] = w[qz] = 0.0; }
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; ++dz)
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
            {
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; qz++)
+               double u = 0.0;
+               double v = 0.0;
+               double w = 0.0;
+               for (int dz = 0; dz < D1D; ++dz)
               {
-                  const int i = qi(qz,dz,Q1D);
-                  const int j = dj(qz,dz,D1D);
-                  const int k = qk(qz,dz,Q1D);
-                  const int l = dl(qz,dz,D1D);
-                  const double s = sign(qz,dz);
-                  u[qz] += DQQ0[dz][qy][qx] * B[i][j];
-                  v[qz] += DQQ1[dz][qy][qx] * B[i][j];
-                  w[qz] += DQQ2[dz][qy][qx] * G[k][l] * s;
+                  u += DQQ0[dz][qy][qx] * B[qz][dz];
+                  v += DQQ1[dz][qy][qx] * B[qz][dz];
+                  w += DQQ2[dz][qy][qx] * G[qz][dz];
               }
+               QQQ0[qz][qy][qx] = u;
+               QQQ1[qz][qy][qx] = v;
+               QQQ2[qz][qy][qx] = w;
            }
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; qz++)
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(qz,z,Q1D)
+      {
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
            {
-               const double O11 = d(qx,qy,qz,0,e);
-               const double O12 = d(qx,qy,qz,1,e);
-               const double O13 = d(qx,qy,qz,2,e);
-               const double O22 = d(qx,qy,qz,3,e);
-               const double O23 = d(qx,qy,qz,4,e);
-               const double O33 = d(qx,qy,qz,5,e);
-               const double gX = u[qz];
-               const double gY = v[qz];
-               const double gZ = w[qz];
+               const int q = qx + ((qy*Q1D) + (qz*Q1D*Q1D));
+               const double O11 = d(q,0,e);
+               const double O12 = d(q,1,e);
+               const double O13 = d(q,2,e);
+               const double O22 = d(q,3,e);
+               const double O23 = d(q,4,e);
+               const double O33 = d(q,5,e);
+               const double gX = QQQ0[qz][qy][qx];
+               const double gY = QQQ1[qz][qy][qx];
+               const double gZ = QQQ2[qz][qy][qx];
               QQQ0[qz][qy][qx] = (O11*gX) + (O12*gY) + (O13*gZ);
               QQQ1[qz][qy][qx] = (O12*gX) + (O22*gY) + (O23*gZ);
               QQQ2[qz][qy][qx] = (O13*gX) + (O23*gY) + (O33*gZ);
@@ -1530,112 +1468,78 @@ static void SmemPADiffusionApply3D(const int NE,
         }
      }
      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(d,y,D1D)
+      if (tidz == 0)
      {
-         MFEM_FOREACH_THREAD(q,x,Q1D)
+         MFEM_FOREACH_THREAD(d,y,D1D)
         {
-            const int i = qi(q,d,Q1D);
-            const int j = dj(q,d,D1D);
-            const int k = qk(q,d,Q1D);
-            const int l = dl(q,d,D1D);
-            Bt[j][i] = b(q,d);
-            Gt[l][k] = g(q,d) * sign(q,d);
+            MFEM_FOREACH_THREAD(q,x,Q1D)
+            {
+               Bt[d][q] = b(q,d);
+               Gt[d][q] = g(q,d);
+            }
         }
      }
      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(qy,y,Q1D)
+      MFEM_FOREACH_THREAD(qz,z,Q1D)
      {
-         MFEM_FOREACH_THREAD(dx,x,D1D)
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
         {
-            double u[Q1D], v[Q1D], w[Q1D];
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; ++qz) { u[qz] = v[qz] = w[qz] = 0.0; }
-            MFEM_UNROLL(MQ1)
-            for (int qx = 0; qx < Q1D; ++qx)
+            MFEM_FOREACH_THREAD(dx,x,D1D)
            {
-               const int i = qi(qx,dx,Q1D);
-               const int j = dj(qx,dx,D1D);
-               const int k = qk(qx,dx,Q1D);
-               const int l = dl(qx,dx,D1D);
-               const double s = sign(qx,dx);
-               MFEM_UNROLL(MQ1)
+               double u = 0.0;
+               double v = 0.0;
+               double w = 0.0;
+               for (int qx = 0; qx < Q1D; ++qx)
+               {
+                  u += QQQ0[qz][qy][qx] * Gt[dx][qx];
+                  v += QQQ1[qz][qy][qx] * Bt[dx][qx];
+                  w += QQQ2[qz][qy][qx] * Bt[dx][qx];
+               }
+               QQD0[qz][qy][dx] = u;
+               QQD1[qz][qy][dx] = v;
+               QQD2[qz][qy][dx] = w;
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(qz,z,Q1D)
+      {
+         MFEM_FOREACH_THREAD(dy,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(dx,x,D1D)
+            {
+               double u = 0.0;
+               double v = 0.0;
+               double w = 0.0;
+               for (int qy = 0; qy < Q1D; ++qy)
+               {
+                  u += QQD0[qz][qy][dx] * Bt[dy][qy];
+                  v += QQD1[qz][qy][dx] * Gt[dy][qy];
+                  w += QQD2[qz][qy][dx] * Bt[dy][qy];
+               }
+               QDD0[qz][dy][dx] = u;
+               QDD1[qz][dy][dx] = v;
+               QDD2[qz][dy][dx] = w;
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(dz,z,D1D)
+      {
+         MFEM_FOREACH_THREAD(dy,y,D1D)
+         {
+            MFEM_FOREACH_THREAD(dx,x,D1D)
+            {
+               double u = 0.0;
+               double v = 0.0;
+               double w = 0.0;
               for (int qz = 0; qz < Q1D; ++qz)
               {
-                  u[qz] += QQQ0[qz][qy][qx] * Gt[l][k] * s;
-                  v[qz] += QQQ1[qz][qy][qx] * Bt[j][i];
-                  w[qz] += QQQ2[qz][qy][qx] * Bt[j][i];
+                  u += QDD0[qz][dy][dx] * Bt[dz][qz];
+                  v += QDD1[qz][dy][dx] * Bt[dz][qz];
+                  w += QDD2[qz][dy][dx] * Gt[dz][qz];
               }
-            }
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; ++qz)
-            {
-               QQD0[qz][qy][dx] = u[qz];
-               QQD1[qz][qy][dx] = v[qz];
-               QQD2[qz][qy][dx] = w[qz];
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(dy,y,D1D)
-      {
-         MFEM_FOREACH_THREAD(dx,x,D1D)
-         {
-            double u[Q1D], v[Q1D], w[Q1D];
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; ++qz) { u[qz] = v[qz] = w[qz] = 0.0; }
-            MFEM_UNROLL(MQ1)
-            for (int qy = 0; qy < Q1D; ++qy)
-            {
-               const int i = qi(qy,dy,Q1D);
-               const int j = dj(qy,dy,D1D);
-               const int k = qk(qy,dy,Q1D);
-               const int l = dl(qy,dy,D1D);
-               const double s = sign(qy,dy);
-               MFEM_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; ++qz)
-               {
-                  u[qz] += QQD0[qz][qy][dx] * Bt[j][i];
-                  v[qz] += QQD1[qz][qy][dx] * Gt[l][k] * s;
-                  w[qz] += QQD2[qz][qy][dx] * Bt[j][i];
-               }
-            }
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; ++qz)
-            {
-               QDD0[qz][dy][dx] = u[qz];
-               QDD1[qz][dy][dx] = v[qz];
-               QDD2[qz][dy][dx] = w[qz];
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(dy,y,D1D)
-      {
-         MFEM_FOREACH_THREAD(dx,x,D1D)
-         {
-            double u[D1D], v[D1D], w[D1D];
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; ++dz) { u[dz] = v[dz] = w[dz] = 0.0; }
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; ++qz)
-            {
-               MFEM_UNROLL(MD1)
-               for (int dz = 0; dz < D1D; ++dz)
-               {
-                  const int i = qi(qz,dz,Q1D);
-                  const int j = dj(qz,dz,D1D);
-                  const int k = qk(qz,dz,Q1D);
-                  const int l = dl(qz,dz,D1D);
-                  const double s = sign(qz,dz);
-                  u[dz] += QDD0[qz][dy][dx] * Bt[j][i];
-                  v[dz] += QDD1[qz][dy][dx] * Bt[j][i];
-                  w[dz] += QDD2[qz][dy][dx] * Gt[l][k] * s;
-               }
-            }
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; ++dz)
-            {
-               y(dx,dy,dz,e) += (u[dz] + v[dz] + w[dz]);
+               y(dx,dy,dz,e) += (u + v + w);
            }
         }
      }
@@ -1670,11 +1574,9 @@ static void PADiffusionApply(const int dim,
      MFEM_ABORT("OCCA PADiffusionApply unknown kernel!");
   }
 #endif // MFEM_USE_OCCA
-   const int ID = (D1D << 4 ) | Q1D;
-
   if (dim == 2)
   {
-      switch (ID)
+      switch ((D1D << 4 ) | Q1D)
      {
         case 0x22: return SmemPADiffusionApply2D<2,2,16>(NE,B,G,D,X,Y);
         case 0x33: return SmemPADiffusionApply2D<3,3,16>(NE,B,G,D,X,Y);
@@ -1687,10 +1589,9 @@ static void PADiffusionApply(const int dim,
         default:   return PADiffusionApply2D(NE,B,G,Bt,Gt,D,X,Y,D1D,Q1D);
      }
   }
-
-   if (dim == 3)
+   else if (dim == 3)
   {
-      switch (ID)
+      switch ((D1D << 4 ) | Q1D)
      {
         case 0x23: return SmemPADiffusionApply3D<2,3>(NE,B,G,D,X,Y);
         case 0x34: return SmemPADiffusionApply3D<3,4>(NE,B,G,D,X,Y);
@@ -1713,7 +1614,29 @@ void DiffusionIntegrator::AddMultPA(const Vector &x, Vector &y) const
 #ifdef MFEM_USE_CEED
   if (DeviceCanUseCeed())
   {
-      CeedAddMultPA(ceedDataPtr, x, y);
+      const CeedScalar *x_ptr;
+      CeedScalar *y_ptr;
+      CeedMemType mem;
+      CeedGetPreferredMemType(internal::ceed, &mem);
+      if ( Device::Allows(Backend::CUDA) && mem==CEED_MEM_DEVICE )
+      {
+         x_ptr = x.Read();
+         y_ptr = y.ReadWrite();
+      }
+      else
+      {
+         x_ptr = x.HostRead();
+         y_ptr = y.HostReadWrite();
+         mem = CEED_MEM_HOST;
+      }
+      CeedVectorSetArray(ceedDataPtr->u, mem, CEED_USE_POINTER,
+                         const_cast<CeedScalar*>(x_ptr));
+      CeedVectorSetArray(ceedDataPtr->v, mem, CEED_USE_POINTER, y_ptr);
+
+      CeedOperatorApplyAdd(ceedDataPtr->oper, ceedDataPtr->u, ceedDataPtr->v,
+                           CEED_REQUEST_IMMEDIATE);
+
+      CeedVectorSyncArray(ceedDataPtr->v, mem);
   }
   else
 #endif
@@ -1,275 +0,0 @@
-// Copyright (c) 2010-2020, 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.
-
-#include "../general/forall.hpp"
-#include "bilininteg.hpp"
-#include "gridfunc.hpp"
-
-namespace mfem
-{
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EADiffusionAssemble1D(const int NE,
-                                  const Array<double> &b,
-                                  const Array<double> &g,
-                                  const Vector &padata,
-                                  Vector &eadata,
-                                  const int d1d = 0,
-                                  const int q1d = 0)
-{
-   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 G = Reshape(g.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, NE);
-   auto A = Reshape(eadata.ReadWrite(), D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, 1,
-   {
-      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;
-      double r_Gi[MQ1];
-      double r_Gj[MQ1];
-      for (int q = 0; q < Q1D; q++)
-      {
-         r_Gi[q] = G(q,MFEM_THREAD_ID(x));
-         r_Gj[q] = G(q,MFEM_THREAD_ID(y));
-      }
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(j1,y,D1D)
-         {
-            double val = 0.0;
-            for (int k1 = 0; k1 < Q1D; ++k1)
-            {
-               val += r_Gj[k1] * D(k1, e) * r_Gi[k1];
-            }
-            A(i1, j1, e) += val;
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EADiffusionAssemble2D(const int NE,
-                                  const Array<double> &b,
-                                  const Array<double> &g,
-                                  const Vector &padata,
-                                  Vector &eadata,
-                                  const int d1d = 0,
-                                  const int q1d = 0)
-{
-   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 G = Reshape(g.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, Q1D, 3, NE);
-   auto A = Reshape(eadata.ReadWrite(), D1D, D1D, D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, 1,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-      double r_B[MQ1][MD1];
-      double r_G[MQ1][MD1];
-      for (int d = 0; d < D1D; d++)
-      {
-         for (int q = 0; q < Q1D; q++)
-         {
-            r_B[q][d] = B(q,d);
-            r_G[q][d] = G(q,d);
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(i2,y,D1D)
-         {
-            for (int j1 = 0; j1 < D1D; ++j1)
-            {
-               for (int j2 = 0; j2 < D1D; ++j2)
-               {
-                  double val = 0.0;
-                  for (int k1 = 0; k1 < Q1D; ++k1)
-                  {
-                     for (int k2 = 0; k2 < Q1D; ++k2)
-                     {
-                        double bgi = r_G[k1][i1] * r_B[k2][i2];
-                        double gbi = r_B[k1][i1] * r_G[k2][i2];
-                        double bgj = r_G[k1][j1] * r_B[k2][j2];
-                        double gbj = r_B[k1][j1] * r_G[k2][j2];
-                        double D00 = D(k1,k2,0,e);
-                        double D10 = D(k1,k2,1,e);
-                        double D01 = D10;
-                        double D11 = D(k1,k2,2,e);
-                        val += bgi * D00 * bgj
-                               + gbi * D01 * bgj
-                               + bgi * D10 * gbj
-                               + gbi * D11 * gbj;
-                     }
-                  }
-                  A(i1, i2, j1, j2, e) += val;
-               }
-            }
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EADiffusionAssemble3D(const int NE,
-                                  const Array<double> &g,
-                                  const Array<double> &b,
-                                  const Vector &padata,
-                                  Vector &eadata,
-                                  const int d1d = 0,
-                                  const int q1d = 0)
-{
-   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 G = Reshape(g.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, Q1D, Q1D, 6, NE);
-   auto A = Reshape(eadata.ReadWrite(), D1D, D1D, D1D, D1D, D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, D1D,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-      double r_B[MQ1][MD1];
-      double r_G[MQ1][MD1];
-      for (int d = 0; d < D1D; d++)
-      {
-         for (int q = 0; q < Q1D; q++)
-         {
-            r_B[q][d] = B(q,d);
-            r_G[q][d] = G(q,d);
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(i2,y,D1D)
-         {
-            MFEM_FOREACH_THREAD(i3,z,D1D)
-            {
-               for (int j1 = 0; j1 < D1D; ++j1)
-               {
-                  for (int j2 = 0; j2 < D1D; ++j2)
-                  {
-                     for (int j3 = 0; j3 < D1D; ++j3)
-                     {
-                        double val = 0.0;
-                        for (int k1 = 0; k1 < Q1D; ++k1)
-                        {
-                           for (int k2 = 0; k2 < Q1D; ++k2)
-                           {
-                              for (int k3 = 0; k3 < Q1D; ++k3)
-                              {
-                                 double bbgi = r_G[k1][i1] * r_B[k2][i2] * r_B[k3][i3];
-                                 double bgbi = r_B[k1][i1] * r_G[k2][i2] * r_B[k3][i3];
-                                 double gbbi = r_B[k1][i1] * r_B[k2][i2] * r_G[k3][i3];
-                                 double bbgj = r_G[k1][j1] * r_B[k2][j2] * r_B[k3][j3];
-                                 double bgbj = r_B[k1][j1] * r_G[k2][j2] * r_B[k3][j3];
-                                 double gbbj = r_B[k1][j1] * r_B[k2][j2] * r_G[k3][j3];
-                                 double D00 = D(k1,k2,k3,0,e);
-                                 double D10 = D(k1,k2,k3,1,e);
-                                 double D20 = D(k1,k2,k3,2,e);
-                                 double D01 = D10;
-                                 double D11 = D(k1,k2,k3,3,e);
-                                 double D21 = D(k1,k2,k3,4,e);
-                                 double D02 = D20;
-                                 double D12 = D21;
-                                 double D22 = D(k1,k2,k3,5,e);
-                                 val += bbgi * D00 * bbgj
-                                        + bgbi * D10 * bbgj
-                                        + gbbi * D20 * bbgj
-                                        + bbgi * D01 * bgbj
-                                        + bgbi * D11 * bgbj
-                                        + gbbi * D21 * bgbj
-                                        + bbgi * D02 * gbbj
-                                        + bgbi * D12 * gbbj
-                                        + gbbi * D22 * gbbj;
-                              }
-                           }
-                        }
-                        A(i1, i2, i3, j1, j2, j3, e) += val;
-                     }
-                  }
-               }
-            }
-         }
-      }
-   });
-}
-
-void DiffusionIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                     Vector &ea_data)
-{
-   AssemblePA(fes);
-   const int ne = fes.GetMesh()->GetNE();
-   const Array<double> &B = maps->B;
-   const Array<double> &G = maps->G;
-   if (dim == 1)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x22: return EADiffusionAssemble1D<2,2>(ne,B,G,pa_data,ea_data);
-         case 0x33: return EADiffusionAssemble1D<3,3>(ne,B,G,pa_data,ea_data);
-         case 0x44: return EADiffusionAssemble1D<4,4>(ne,B,G,pa_data,ea_data);
-         case 0x55: return EADiffusionAssemble1D<5,5>(ne,B,G,pa_data,ea_data);
-         case 0x66: return EADiffusionAssemble1D<6,6>(ne,B,G,pa_data,ea_data);
-         case 0x77: return EADiffusionAssemble1D<7,7>(ne,B,G,pa_data,ea_data);
-         case 0x88: return EADiffusionAssemble1D<8,8>(ne,B,G,pa_data,ea_data);
-         case 0x99: return EADiffusionAssemble1D<9,9>(ne,B,G,pa_data,ea_data);
-         default:   return EADiffusionAssemble1D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   else if (dim == 2)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x22: return EADiffusionAssemble2D<2,2>(ne,B,G,pa_data,ea_data);
-         case 0x33: return EADiffusionAssemble2D<3,3>(ne,B,G,pa_data,ea_data);
-         case 0x44: return EADiffusionAssemble2D<4,4>(ne,B,G,pa_data,ea_data);
-         case 0x55: return EADiffusionAssemble2D<5,5>(ne,B,G,pa_data,ea_data);
-         case 0x66: return EADiffusionAssemble2D<6,6>(ne,B,G,pa_data,ea_data);
-         case 0x77: return EADiffusionAssemble2D<7,7>(ne,B,G,pa_data,ea_data);
-         case 0x88: return EADiffusionAssemble2D<8,8>(ne,B,G,pa_data,ea_data);
-         case 0x99: return EADiffusionAssemble2D<9,9>(ne,B,G,pa_data,ea_data);
-         default:   return EADiffusionAssemble2D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   else if (dim == 3)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x23: return EADiffusionAssemble3D<2,3>(ne,B,G,pa_data,ea_data);
-         case 0x34: return EADiffusionAssemble3D<3,4>(ne,B,G,pa_data,ea_data);
-         case 0x45: return EADiffusionAssemble3D<4,5>(ne,B,G,pa_data,ea_data);
-         case 0x56: return EADiffusionAssemble3D<5,6>(ne,B,G,pa_data,ea_data);
-         case 0x67: return EADiffusionAssemble3D<6,7>(ne,B,G,pa_data,ea_data);
-         case 0x78: return EADiffusionAssemble3D<7,8>(ne,B,G,pa_data,ea_data);
-         case 0x89: return EADiffusionAssemble3D<8,9>(ne,B,G,pa_data,ea_data);
-         default:   return EADiffusionAssemble3D(ne,B,G,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   MFEM_ABORT("Unknown kernel.");
-}
-
-}
@@ -23,7 +23,7 @@ namespace mfem

 // PA Mass Assemble kernel

-void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
+void MassIntegrator::SetupPA(const FiniteElementSpace &fes, const bool force)
 {
   // Assuming the same element type
   fespace = &fes;
@@ -33,7 +33,7 @@ void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
   ElementTransformation *T = mesh->GetElementTransformation(0);
   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el, *T);
 #ifdef MFEM_USE_CEED
-   if (DeviceCanUseCeed())
+   if (DeviceCanUseCeed() && !force)
   {
      if (ceedDataPtr) { delete ceedDataPtr; }
      CeedData* ptr = new CeedData();
@@ -62,19 +62,6 @@ void MassIntegrator::SetupPA(const FiniteElementSpace &fes)
      coeff.SetSize(1);
      coeff(0) = cQ->constant;
   }
-   else if (QuadratureFunctionCoefficient* cQ =
-               dynamic_cast<QuadratureFunctionCoefficient*>(Q))
-   {
-      const QuadratureFunction &qFun = cQ->GetQuadFunction();
-      MFEM_VERIFY(qFun.Size() == 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();
-      coeff.MakeRef(const_cast<QuadratureFunction &>(qFun),0);
-   }
   else
   {
      coeff.SetSize(nq * ne);
@@ -453,16 +440,8 @@ static void PAMassAssembleDiagonal(const int dim, const int D1D,

 void MassIntegrator::AssembleDiagonalPA(Vector &diag)
 {
-#ifdef MFEM_USE_CEED
-   if (DeviceCanUseCeed())
-   {
-      CeedAssembleDiagonalPA(ceedDataPtr, diag);
-   }
-   else
-#endif
-   {
-      PAMassAssembleDiagonal(dim, dofs1D, quad1D, ne, maps->B, pa_data, diag);
-   }
+   if (pa_data.Size()==0) { SetupPA(*fespace, true); }
+   PAMassAssembleDiagonal(dim, dofs1D, quad1D, ne, maps->B, pa_data, diag);
 }


@@ -660,7 +639,6 @@ static void SmemPAMassApply2D(const int NE,
                              const int d1d = 0,
                              const int q1d = 0)
 {
-   MFEM_CONTRACT_VAR(bt_);
   const int D1D = T_D1D ? T_D1D : d1d;
   const int Q1D = T_Q1D ? T_Q1D : q1d;
   constexpr int NBZ = T_NBZ ? T_NBZ : 1;
@@ -924,7 +902,6 @@ static void SmemPAMassApply3D(const int NE,
                              const int d1d = 0,
                              const int q1d = 0)
 {
-   MFEM_CONTRACT_VAR(bt_);
   const int D1D = T_D1D ? T_D1D : d1d;
   const int Q1D = T_Q1D ? T_Q1D : q1d;
   constexpr int M1Q = T_Q1D ? T_Q1D : MAX_Q1D;
@@ -1215,7 +1192,29 @@ void MassIntegrator::AddMultPA(const Vector &x, Vector &y) const
 #ifdef MFEM_USE_CEED
   if (DeviceCanUseCeed())
   {
-      CeedAddMultPA(ceedDataPtr, x, y);
+      const CeedScalar *x_ptr;
+      CeedScalar *y_ptr;
+      CeedMemType mem;
+      CeedGetPreferredMemType(internal::ceed, &mem);
+      if ( Device::Allows(Backend::CUDA) && mem==CEED_MEM_DEVICE )
+      {
+         x_ptr = x.Read();
+         y_ptr = y.ReadWrite();
+      }
+      else
+      {
+         x_ptr = x.HostRead();
+         y_ptr = y.HostReadWrite();
+         mem = CEED_MEM_HOST;
+      }
+      CeedVectorSetArray(ceedDataPtr->u, mem, CEED_USE_POINTER,
+                         const_cast<CeedScalar*>(x_ptr));
+      CeedVectorSetArray(ceedDataPtr->v, mem, CEED_USE_POINTER, y_ptr);
+
+      CeedOperatorApplyAdd(ceedDataPtr->oper, ceedDataPtr->u, ceedDataPtr->v,
+                           CEED_REQUEST_IMMEDIATE);
+
+      CeedVectorSyncArray(ceedDataPtr->v, mem);
   }
   else
 #endif
@@ -1,255 +0,0 @@
-// Copyright (c) 2010-2020, 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.
-
-#include "../general/forall.hpp"
-#include "bilininteg.hpp"
-#include "gridfunc.hpp"
-
-namespace mfem
-{
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EAMassAssemble1D(const int NE,
-                             const Array<double> &basis,
-                             const Vector &padata,
-                             Vector &eadata,
-                             const int d1d = 0,
-                             const int q1d = 0)
-{
-   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(basis.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, NE);
-   auto M = Reshape(eadata.ReadWrite(), D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, 1,
-   {
-      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;
-      double r_Bi[MQ1];
-      double r_Bj[MQ1];
-      for (int q = 0; q < Q1D; q++)
-      {
-         r_Bi[q] = B(q,MFEM_THREAD_ID(x));
-         r_Bj[q] = B(q,MFEM_THREAD_ID(y));
-      }
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(j1,y,D1D)
-         {
-            double val = 0.0;
-            for (int k1 = 0; k1 < Q1D; ++k1)
-            {
-               val += r_Bi[k1] * r_Bj[k1] * D(k1, e);
-            }
-            M(i1, j1, e) += val;
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EAMassAssemble2D(const int NE,
-                             const Array<double> &basis,
-                             const Vector &padata,
-                             Vector &eadata,
-                             const int d1d = 0,
-                             const int q1d = 0)
-{
-   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(basis.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, Q1D, NE);
-   auto M = Reshape(eadata.ReadWrite(), D1D, D1D, D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, 1,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-      double r_B[MQ1][MD1];
-      for (int d = 0; d < D1D; d++)
-      {
-         for (int q = 0; q < Q1D; q++)
-         {
-            r_B[q][d] = B(q,d);
-         }
-      }
-      MFEM_SHARED double s_D[MQ1][MQ1];
-      MFEM_FOREACH_THREAD(k1,x,Q1D)
-      {
-         MFEM_FOREACH_THREAD(k2,y,Q1D)
-         {
-            s_D[k1][k2] = D(k1,k2,e);
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(i2,y,D1D)
-         {
-            for (int j1 = 0; j1 < D1D; ++j1)
-            {
-               for (int j2 = 0; j2 < D1D; ++j2)
-               {
-                  double val = 0.0;
-                  for (int k1 = 0; k1 < Q1D; ++k1)
-                  {
-                     for (int k2 = 0; k2 < Q1D; ++k2)
-                     {
-                        val += r_B[k1][i1] * r_B[k1][j1]
-                               * r_B[k2][i2] * r_B[k2][j2]
-                               * s_D[k1][k2];
-                     }
-                  }
-                  M(i1, i2, j1, j2, e) += val;
-               }
-            }
-         }
-      }
-   });
-}
-
-template<int T_D1D = 0, int T_Q1D = 0>
-static void EAMassAssemble3D(const int NE,
-                             const Array<double> &basis,
-                             const Vector &padata,
-                             Vector &eadata,
-                             const int d1d = 0,
-                             const int q1d = 0)
-{
-   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(basis.Read(), Q1D, D1D);
-   auto D = Reshape(padata.Read(), Q1D, Q1D, Q1D, NE);
-   auto M = Reshape(eadata.ReadWrite(), D1D, D1D, D1D, D1D, D1D, D1D, NE);
-   MFEM_FORALL_3D(e, NE, D1D, D1D, D1D,
-   {
-      const int D1D = T_D1D ? T_D1D : d1d;
-      const int Q1D = T_Q1D ? T_Q1D : q1d;
-      constexpr int MD1 = T_D1D ? T_D1D : MAX_D1D;
-      constexpr int MQ1 = T_Q1D ? T_Q1D : MAX_Q1D;
-      double r_B[MQ1][MD1];
-      for (int d = 0; d < D1D; d++)
-      {
-         for (int q = 0; q < Q1D; q++)
-         {
-            r_B[q][d] = B(q,d);
-         }
-      }
-      MFEM_SHARED double s_D[MQ1][MQ1][MQ1];
-      MFEM_FOREACH_THREAD(k1,x,Q1D)
-      {
-         MFEM_FOREACH_THREAD(k2,y,Q1D)
-         {
-            MFEM_FOREACH_THREAD(k3,z,Q1D)
-            {
-               s_D[k1][k2][k3] = D(k1,k2,k3,e);
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(i1,x,D1D)
-      {
-         MFEM_FOREACH_THREAD(i2,y,D1D)
-         {
-            MFEM_FOREACH_THREAD(i3,z,D1D)
-            {
-               for (int j1 = 0; j1 < D1D; ++j1)
-               {
-                  for (int j2 = 0; j2 < D1D; ++j2)
-                  {
-                     for (int j3 = 0; j3 < D1D; ++j3)
-                     {
-                        double val = 0.0;
-                        for (int k1 = 0; k1 < Q1D; ++k1)
-                        {
-                           for (int k2 = 0; k2 < Q1D; ++k2)
-                           {
-                              for (int k3 = 0; k3 < Q1D; ++k3)
-                              {
-                                 val += r_B[k1][i1] * r_B[k1][j1]
-                                        * r_B[k2][i2] * r_B[k2][j2]
-                                        * r_B[k3][i3] * r_B[k3][j3]
-                                        * s_D[k1][k2][k3];
-                              }
-                           }
-                        }
-                        M(i1, i2, i3, j1, j2, j3, e) += val;
-                     }
-                  }
-               }
-            }
-         }
-      }
-   });
-}
-
-void MassIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                Vector &ea_data)
-{
-   AssemblePA(fes);
-   const int ne = fes.GetMesh()->GetNE();
-   const Array<double> &B = maps->B;
-   if (dim == 1)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x22: return EAMassAssemble1D<2,2>(ne,B,pa_data,ea_data);
-         case 0x33: return EAMassAssemble1D<3,3>(ne,B,pa_data,ea_data);
-         case 0x44: return EAMassAssemble1D<4,4>(ne,B,pa_data,ea_data);
-         case 0x55: return EAMassAssemble1D<5,5>(ne,B,pa_data,ea_data);
-         case 0x66: return EAMassAssemble1D<6,6>(ne,B,pa_data,ea_data);
-         case 0x77: return EAMassAssemble1D<7,7>(ne,B,pa_data,ea_data);
-         case 0x88: return EAMassAssemble1D<8,8>(ne,B,pa_data,ea_data);
-         case 0x99: return EAMassAssemble1D<9,9>(ne,B,pa_data,ea_data);
-         default:   return EAMassAssemble1D(ne,B,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   else if (dim == 2)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x22: return EAMassAssemble2D<2,2>(ne,B,pa_data,ea_data);
-         case 0x33: return EAMassAssemble2D<3,3>(ne,B,pa_data,ea_data);
-         case 0x44: return EAMassAssemble2D<4,4>(ne,B,pa_data,ea_data);
-         case 0x55: return EAMassAssemble2D<5,5>(ne,B,pa_data,ea_data);
-         case 0x66: return EAMassAssemble2D<6,6>(ne,B,pa_data,ea_data);
-         case 0x77: return EAMassAssemble2D<7,7>(ne,B,pa_data,ea_data);
-         case 0x88: return EAMassAssemble2D<8,8>(ne,B,pa_data,ea_data);
-         case 0x99: return EAMassAssemble2D<9,9>(ne,B,pa_data,ea_data);
-         default:   return EAMassAssemble2D(ne,B,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   else if (dim == 3)
-   {
-      switch ((dofs1D << 4 ) | quad1D)
-      {
-         case 0x23: return EAMassAssemble3D<2,3>(ne,B,pa_data,ea_data);
-         case 0x34: return EAMassAssemble3D<3,4>(ne,B,pa_data,ea_data);
-         case 0x45: return EAMassAssemble3D<4,5>(ne,B,pa_data,ea_data);
-         case 0x56: return EAMassAssemble3D<5,6>(ne,B,pa_data,ea_data);
-         case 0x67: return EAMassAssemble3D<6,7>(ne,B,pa_data,ea_data);
-         case 0x78: return EAMassAssemble3D<7,8>(ne,B,pa_data,ea_data);
-         case 0x89: return EAMassAssemble3D<8,9>(ne,B,pa_data,ea_data);
-         default:   return EAMassAssemble3D(ne,B,pa_data,ea_data,dofs1D,quad1D);
-      }
-   }
-   MFEM_ABORT("Unknown kernel.");
-}
-
-}
@@ -1,103 +0,0 @@
-// Copyright (c) 2010-2020, 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.
-
-#include "../general/forall.hpp"
-#include "bilininteg.hpp"
-
-namespace mfem
-{
-
-void TransposeIntegrator::AssembleEA(const FiniteElementSpace &fes,
-                                     Vector &ea_data)
-{
-   Vector ea_data_tmp(ea_data.Size());
-   ea_data_tmp = 0.0;
-   bfi->AssembleEA(fes, ea_data_tmp);
-   const int ne = fes.GetNE();
-   if (ne == 0) { return; }
-   const int dofs = fes.GetFE(0)->GetDof();
-   auto A = Reshape(ea_data_tmp.Write(), dofs, dofs, ne);
-   auto AT = Reshape(ea_data.ReadWrite(), dofs, dofs, ne);
-   MFEM_FORALL(e, ne,
-   {
-      for (int i = 0; i < dofs; i++)
-      {
-         for (int j = 0; j < dofs; j++)
-         {
-            const double a = A(i, j, e);
-            AT(j, i, e) += a;
-         }
-      }
-   });
-}
-
-void TransposeIntegrator::AssembleEAInteriorFaces(const FiniteElementSpace& fes,
-                                                  Vector &ea_data_int,
-                                                  Vector &ea_data_ext)
-{
-   const int nf = fes.GetNFbyType(FaceType::Interior);
-   if (nf == 0) { return; }
-   Vector ea_data_int_tmp(ea_data_int.Size());
-   Vector ea_data_ext_tmp(ea_data_ext.Size());
-   ea_data_int_tmp = 0.0;
-   ea_data_ext_tmp = 0.0;
-   bfi->AssembleEAInteriorFaces(fes, ea_data_int_tmp, ea_data_ext_tmp);
-   const int faceDofs = fes.GetTraceElement(0,
-                                            fes.GetMesh()->GetFaceBaseGeometry(0))->GetDof();
-   auto A_int = Reshape(ea_data_int_tmp.Read(), faceDofs, faceDofs, 2, nf);
-   auto A_ext = Reshape(ea_data_ext_tmp.Read(), faceDofs, faceDofs, 2, nf);
-   auto AT_int = Reshape(ea_data_int.ReadWrite(), faceDofs, faceDofs, 2, nf);
-   auto AT_ext = Reshape(ea_data_ext.ReadWrite(), faceDofs, faceDofs, 2, nf);
-   MFEM_FORALL(f, nf,
-   {
-      for (int i = 0; i < faceDofs; i++)
-      {
-         for (int j = 0; j < faceDofs; j++)
-         {
-            const double a_int0 = A_int(i, j, 0, f);
-            const double a_int1 = A_int(i, j, 1, f);
-            const double a_ext0 = A_ext(i, j, 0, f);
-            const double a_ext1 = A_ext(i, j, 1, f);
-            AT_int(j, i, 0, f) += a_int0;
-            AT_int(j, i, 1, f) += a_int1;
-            AT_ext(j, i, 0, f) += a_ext1;
-            AT_ext(j, i, 1, f) += a_ext0;
-         }
-      }
-   });
-}
-
-void TransposeIntegrator::AssembleEABoundaryFaces(const FiniteElementSpace& fes,
-                                                  Vector &ea_data_bdr)
-{
-   const int nf = fes.GetNFbyType(FaceType::Boundary);
-   if (nf == 0) { return; }
-   Vector ea_data_bdr_tmp(ea_data_bdr.Size());
-   ea_data_bdr_tmp = 0.0;
-   bfi->AssembleEABoundaryFaces(fes, ea_data_bdr_tmp);
-   const int faceDofs = fes.GetTraceElement(0,
-                                            fes.GetMesh()->GetFaceBaseGeometry(0))->GetDof();
-   auto A_bdr = Reshape(ea_data_bdr_tmp.Read(), faceDofs, faceDofs, nf);
-   auto AT_bdr = Reshape(ea_data_bdr.ReadWrite(), faceDofs, faceDofs, nf);
-   MFEM_FORALL(f, nf,
-   {
-      for (int i = 0; i < faceDofs; i++)
-      {
-         for (int j = 0; j < faceDofs; j++)
-         {
-            const double a_bdr = A_bdr(i, j, f);
-            AT_bdr(j, i, f) += a_bdr;
-         }
-      }
-   });
-}
-
-}
@@ -1,402 +0,0 @@
-// Copyright (c) 2010-2020, 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.
-
-#include "../general/forall.hpp"
-#include "bilininteg.hpp"
-
-namespace mfem
-{
-
-void PAHcurlSetup2D(const int Q1D,
-                    const int coeffDim,
-                    const int NE,
-                    const Array<double> &w,
-                    const Vector &j,
-                    Vector &_coeff,
-                    Vector &op);
-
-void PAHcurlSetup3D(const int Q1D,
-                    const int coeffDim,
-                    const int NE,
-                    const Array<double> &w,
-                    const Vector &j,
-                    Vector &_coeff,
-                    Vector &op);
-
-void PAHcurlMassAssembleDiagonal2D(const int D1D,
-                                   const int Q1D,
-                                   const int NE,
-                                   const Array<double> &_Bo,
-                                   const Array<double> &_Bc,
-                                   const Vector &_op,
-                                   Vector &_diag);
-
-void PAHcurlMassAssembleDiagonal3D(const int D1D,
-                                   const int Q1D,
-                                   const int NE,
-                                   const Array<double> &_Bo,
-                                   const Array<double> &_Bc,
-                                   const Vector &_op,
-                                   Vector &_diag);
-
-void PAHcurlMassApply2D(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 PAHcurlMassApply3D(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 PAHdivSetup2D(const int Q1D,
-                   const int NE,
-                   const Array<double> &w,
-                   const Vector &j,
-                   Vector &_coeff,
-                   Vector &op);
-
-void PAHdivSetup3D(const int Q1D,
-                   const int NE,
-                   const Array<double> &w,
-                   const Vector &j,
-                   Vector &_coeff,
-                   Vector &op);
-
-void PAHcurlH1Apply2D(const int D1D,
-                      const int Q1D,
-                      const int NE,
-                      const Array<double> &_Bc,
-                      const Array<double> &_Gc,
-                      const Array<double> &_Bot,
-                      const Array<double> &_Bct,
-                      const Vector &_op,
-                      const Vector &_x,
-                      Vector &_y);
-
-void PAHcurlH1Apply3D(const int D1D,
-                      const int Q1D,
-                      const int NE,
-                      const Array<double> &_Bc,
-                      const Array<double> &_Gc,
-                      const Array<double> &_Bot,
-                      const Array<double> &_Bct,
-                      const Vector &_op,
-                      const Vector &_x,
-                      Vector &_y);
-
-void PAHdivMassAssembleDiagonal2D(const int D1D,
-                                  const int Q1D,
-                                  const int NE,
-                                  const Array<double> &_Bo,
-                                  const Array<double> &_Bc,
-                                  const Vector &_op,
-                                  Vector &_diag);
-
-void PAHdivMassAssembleDiagonal3D(const int D1D,
-                                  const int Q1D,
-                                  const int NE,
-                                  const Array<double> &_Bo,
-                                  const Array<double> &_Bc,
-                                  const Vector &_op,
-                                  Vector &_diag);
-
-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 VectorFEMassIntegrator::AssemblePA(const FiniteElementSpace &fes)
-{
-   // Assumes tensor-product elements
-   Mesh *mesh = fes.GetMesh();
-   const FiniteElement *fel = fes.GetFE(0);
-
-   const VectorTensorFiniteElement *el =
-      dynamic_cast<const VectorTensorFiniteElement*>(fel);
-   MFEM_VERIFY(el != NULL, "Only VectorTensorFiniteElement is supported!");
-
-   const IntegrationRule *ir
-      = IntRule ? IntRule : &MassIntegrator::GetRule(*el, *el,
-                                                     *mesh->GetElementTransformation(0));
-   const int dims = el->GetDim();
-   MFEM_VERIFY(dims == 2 || dims == 3, "");
-
-   const int symmDims = (dims * (dims + 1)) / 2; // 1x1: 1, 2x2: 3, 3x3: 6
-   const int nq = ir->GetNPoints();
-   dim = mesh->Dimension();
-   MFEM_VERIFY(dim == 2 || dim == 3, "");
-
-   ne = fes.GetNE();
-   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::JACOBIANS);
-   mapsC = &el->GetDofToQuad(*ir, DofToQuad::TENSOR);
-   mapsO = &el->GetDofToQuadOpen(*ir, DofToQuad::TENSOR);
-   dofs1D = mapsC->ndof;
-   quad1D = mapsC->nqpt;
-
-   MFEM_VERIFY(dofs1D == mapsO->ndof + 1 && quad1D == mapsO->nqpt, "");
-
-   pa_data.SetSize(symmDims * nq * ne, Device::GetMemoryType());
-
-   const int coeffDim = VQ ? VQ->GetVDim() : 1;
-
-   Vector coeff(coeffDim * ne * nq);
-   coeff = 1.0;
-   auto coeffh = Reshape(coeff.HostWrite(), coeffDim, nq, ne);
-   if (Q || VQ)
-   {
-      Vector D(VQ ? coeffDim : 0);
-      if (VQ)
-      {
-         MFEM_VERIFY(coeffDim == dim, "");
-      }
-
-      for (int e=0; e<ne; ++e)
-      {
-         ElementTransformation *tr = mesh->GetElementTransformation(e);
-         for (int p=0; p<nq; ++p)
-         {
-            if (VQ)
-            {
-               VQ->Eval(D, *tr, ir->IntPoint(p));
-               for (int i=0; i<coeffDim; ++i)
-               {
-                  coeffh(i, p, e) = D[i];
-               }
-            }
-            else
-            {
-               coeffh(0, p, e) = Q->Eval(*tr, ir->IntPoint(p));
-            }
-         }
-      }
-   }
-
-   fetype = el->GetDerivType();
-
-   if (el->GetDerivType() == mfem::FiniteElement::CURL && dim == 3)
-   {
-      PAHcurlSetup3D(quad1D, coeffDim, ne, ir->GetWeights(), geom->J,
-                     coeff, pa_data);
-   }
-   else if (el->GetDerivType() == mfem::FiniteElement::CURL && dim == 2)
-   {
-      PAHcurlSetup2D(quad1D, coeffDim, ne, ir->GetWeights(), geom->J,
-                     coeff, pa_data);
-   }
-   else if (el->GetDerivType() == mfem::FiniteElement::DIV && dim == 3)
-   {
-      PAHdivSetup3D(quad1D, ne, ir->GetWeights(), geom->J,
-                    coeff, pa_data);
-   }
-   else if (el->GetDerivType() == mfem::FiniteElement::DIV && dim == 2)
-   {
-      PAHdivSetup2D(quad1D, ne, ir->GetWeights(), geom->J,
-                    coeff, pa_data);
-   }
-   else
-   {
-      MFEM_ABORT("Unknown kernel.");
-   }
-}
-
-void VectorFEMassIntegrator::AssembleDiagonalPA(Vector& diag)
-{
-   if (dim == 3)
-   {
-      if (fetype == mfem::FiniteElement::CURL)
-      {
-         PAHcurlMassAssembleDiagonal3D(dofs1D, quad1D, ne,
-                                       mapsO->B, mapsC->B, pa_data, diag);
-      }
-      else if (fetype == mfem::FiniteElement::DIV)
-      {
-         PAHdivMassAssembleDiagonal3D(dofs1D, quad1D, ne,
-                                      mapsO->B, mapsC->B, pa_data, diag);
-      }
-      else
-      {
-         MFEM_ABORT("Unknown kernel.");
-      }
-   }
-   else
-   {
-      if (fetype == mfem::FiniteElement::CURL)
-      {
-         PAHcurlMassAssembleDiagonal2D(dofs1D, quad1D, ne,
-                                       mapsO->B, mapsC->B, pa_data, diag);
-      }
-      else if (fetype == mfem::FiniteElement::DIV)
-      {
-         PAHdivMassAssembleDiagonal2D(dofs1D, quad1D, ne,
-                                      mapsO->B, mapsC->B, pa_data, diag);
-      }
-      else
-      {
-         MFEM_ABORT("Unknown kernel.");
-      }
-   }
-}
-
-void VectorFEMassIntegrator::AddMultPA(const Vector &x, Vector &y) const
-{
-   if (dim == 3)
-   {
-      if (fetype == mfem::FiniteElement::CURL)
-      {
-         PAHcurlMassApply3D(dofs1D, quad1D, ne, mapsO->B, mapsC->B, mapsO->Bt,
-                            mapsC->Bt, pa_data, x, y);
-      }
-      else if (fetype == mfem::FiniteElement::DIV)
-      {
-         PAHdivMassApply3D(dofs1D, quad1D, ne, mapsO->B, mapsC->B, mapsO->Bt,
-                           mapsC->Bt, pa_data, x, y);
-      }
-      else
-      {
-         MFEM_ABORT("Unknown kernel.");
-      }
-   }
-   else
-   {
-      if (fetype == mfem::FiniteElement::CURL)
-      {
-         PAHcurlMassApply2D(dofs1D, quad1D, ne, mapsO->B, mapsC->B, mapsO->Bt,
-                            mapsC->Bt, pa_data, x, y);
-      }
-      else if (fetype == mfem::FiniteElement::DIV)
-      {
-         PAHdivMassApply2D(dofs1D, quad1D, ne, mapsO->B, mapsC->B, mapsO->Bt,
-                           mapsC->Bt, pa_data, x, y);
-      }
-      else
-      {
-         MFEM_ABORT("Unknown kernel.");
-      }
-   }
-}
-
-void MixedVectorGradientIntegrator::AssemblePA(const FiniteElementSpace
-                                               &trial_fes,
-                                               const FiniteElementSpace &test_fes)
-{
-   // Assumes tensor-product elements, with a vector test space and H^1 trial space.
-   Mesh *mesh = trial_fes.GetMesh();
-   const FiniteElement *trial_fel = trial_fes.GetFE(0);
-   const FiniteElement *test_fel = test_fes.GetFE(0);
-
-   const NodalTensorFiniteElement *trial_el =
-      dynamic_cast<const NodalTensorFiniteElement*>(trial_fel);
-   MFEM_VERIFY(trial_el != NULL, "Only NodalTensorFiniteElement is supported!");
-
-   const VectorTensorFiniteElement *test_el =
-      dynamic_cast<const VectorTensorFiniteElement*>(test_fel);
-   MFEM_VERIFY(test_el != NULL, "Only VectorTensorFiniteElement is supported!");
-
-   const IntegrationRule *ir
-      = IntRule ? IntRule : &MassIntegrator::GetRule(*trial_el, *trial_el,
-                                                     *mesh->GetElementTransformation(0));
-   const int dims = trial_el->GetDim();
-   MFEM_VERIFY(dims == 2 || dims == 3, "");
-
-   const int symmDims = (dims * (dims + 1)) / 2; // 1x1: 1, 2x2: 3, 3x3: 6
-   const int nq = ir->GetNPoints();
-   dim = mesh->Dimension();
-   MFEM_VERIFY(dim == 2 || dim == 3, "");
-
-   MFEM_VERIFY(trial_el->GetOrder() == test_el->GetOrder(), "");
-
-   ne = trial_fes.GetNE();
-   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::JACOBIANS);
-   mapsC = &test_el->GetDofToQuad(*ir, DofToQuad::TENSOR);
-   mapsO = &test_el->GetDofToQuadOpen(*ir, DofToQuad::TENSOR);
-   dofs1D = mapsC->ndof;
-   quad1D = mapsC->nqpt;
-
-   MFEM_VERIFY(dofs1D == mapsO->ndof + 1 && quad1D == mapsO->nqpt, "");
-
-   pa_data.SetSize(symmDims * nq * ne, Device::GetMemoryType());
-
-   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)
-   {
-      PAHcurlSetup3D(quad1D, 1, ne, ir->GetWeights(), geom->J,
-                     coeff, pa_data);
-   }
-   else if (test_el->GetDerivType() == mfem::FiniteElement::CURL && dim == 2)
-   {
-      PAHcurlSetup2D(quad1D, 1, ne, ir->GetWeights(), geom->J,
-                     coeff, pa_data);
-   }
-   else
-   {
-      MFEM_ABORT("Unknown kernel.");
-   }
-}
-
-void MixedVectorGradientIntegrator::AddMultPA(const Vector &x, Vector &y) const
-{
-   if (dim == 3)
-      PAHcurlH1Apply3D(dofs1D, quad1D, ne, mapsC->B, mapsC->G,
-                       mapsO->Bt, mapsC->Bt, pa_data, x, y);
-   else if (dim == 2)
-      PAHcurlH1Apply2D(dofs1D, quad1D, ne, mapsC->B, mapsC->G,
-                       mapsO->Bt, mapsC->Bt, pa_data, x, y);
-   else
-   {
-      MFEM_ABORT("Unsupported dimension!");
-   }
-}
-
-} // namespace mfem
@@ -49,7 +49,7 @@ double FunctionCoefficient::Eval(ElementTransformation & T,
 double GridFunctionCoefficient::Eval (ElementTransformation &T,
                                      const IntegrationPoint &ip)
 {
-   return GridF -> GetValue (T, ip, Component);
+   return GridF -> GetValue (T.ElementNo, ip, Component);
 }

 double TransformedCoefficient::Eval(ElementTransformation &T,
@@ -160,13 +160,13 @@ void VectorArrayCoefficient::Eval(Vector &V, ElementTransformation &T,
 }

 VectorGridFunctionCoefficient::VectorGridFunctionCoefficient (
-   const GridFunction *gf)
+   GridFunction *gf)
   : VectorCoefficient ((gf) ? gf -> VectorDim() : 0)
 {
   GridFunc = gf;
 }

-void VectorGridFunctionCoefficient::SetGridFunction(const GridFunction *gf)
+void VectorGridFunctionCoefficient::SetGridFunction(GridFunction *gf)
 {
   GridFunc = gf; vdim = (gf) ? gf -> VectorDim() : 0;
 }
@@ -174,7 +174,24 @@ void VectorGridFunctionCoefficient::SetGridFunction(const GridFunction *gf)
 void VectorGridFunctionCoefficient::Eval(Vector &V, ElementTransformation &T,
                                         const IntegrationPoint &ip)
 {
-   GridFunc->GetVectorValue(T, ip, V);
+   Mesh *mesh = GridFunc->FESpace()->GetMesh();
+   if (mesh->Dimension() == T.GetDimension())
+   {
+      GridFunc->GetVectorValue(T.ElementNo, ip, V);
+   }
+   else // Assuming T is a boundary element transformation
+   {
+      int el_id, el_info;
+      mesh->GetBdrElementAdjacentElement(T.ElementNo, el_id, el_info);
+      IntegrationPointTransformation loc_T;
+      mesh->GetLocalFaceTransformation(mesh->GetBdrElementType(T.ElementNo),
+                                       mesh->GetElementType(el_id),
+                                       loc_T.Transf,
+                                       el_info);
+      IntegrationPoint eip;
+      loc_T.Transform(ip, eip);
+      GridFunc->GetVectorValue(el_id, eip, V);
+   }
 }

 void VectorGridFunctionCoefficient::Eval(
@@ -184,14 +201,14 @@ void VectorGridFunctionCoefficient::Eval(
 }

 GradientGridFunctionCoefficient::GradientGridFunctionCoefficient (
-   const GridFunction *gf)
+   GridFunction *gf)
   : VectorCoefficient((gf) ?
                       gf -> FESpace() -> GetMesh() -> SpaceDimension() : 0)
 {
   GridFunc = gf;
 }

-void GradientGridFunctionCoefficient::SetGridFunction(const GridFunction *gf)
+void GradientGridFunctionCoefficient::SetGridFunction(GridFunction *gf)
 {
   GridFunc = gf; vdim = (gf) ?
                         gf -> FESpace() -> GetMesh() -> SpaceDimension() : 0;
@@ -209,24 +226,18 @@ void GradientGridFunctionCoefficient::Eval(
   GridFunc->GetGradients(T, ir, M);
 }

-CurlGridFunctionCoefficient::CurlGridFunctionCoefficient(
-   const GridFunction *gf)
-   : VectorCoefficient(0)
+CurlGridFunctionCoefficient::CurlGridFunctionCoefficient (
+   GridFunction *gf)
+   : VectorCoefficient ((gf) ?
+                        gf -> FESpace() -> GetMesh() -> SpaceDimension() : 0)
 {
-   SetGridFunction(gf);
+   GridFunc = gf;
 }

-void CurlGridFunctionCoefficient::SetGridFunction(const GridFunction *gf)
+void CurlGridFunctionCoefficient::SetGridFunction(GridFunction *gf)
 {
-   if (gf)
-   {
-      int sdim = gf -> FESpace() -> GetMesh() -> SpaceDimension();
-      MFEM_VERIFY(sdim == 2 || sdim == 3,
-                  "CurlGridFunctionCoefficient "
-                  "only defind for spaces of dimension 2 or 3.");
-   }
-   GridFunc = gf;
-   vdim = (gf) ? (2 * gf -> FESpace() -> GetMesh() -> SpaceDimension() - 3) : 0;
+   GridFunc = gf; vdim = (gf) ?
+                         gf -> FESpace() -> GetMesh() -> SpaceDimension() : 0;
 }

 void CurlGridFunctionCoefficient::Eval(Vector &V, ElementTransformation &T,
@@ -236,7 +247,7 @@ void CurlGridFunctionCoefficient::Eval(Vector &V, ElementTransformation &T,
 }

 DivergenceGridFunctionCoefficient::DivergenceGridFunctionCoefficient (
-   const GridFunction *gf) : Coefficient()
+   GridFunction *gf) : Coefficient()
 {
   GridFunc = gf;
 }
@@ -422,43 +433,13 @@ double DeterminantCoefficient::Eval(ElementTransformation &T,
   return ma.Det();
 }

-VectorSumCoefficient::VectorSumCoefficient(int dim)
-   : VectorCoefficient(dim),
-     ACoef(NULL), BCoef(NULL),
-     A(dim), B(dim),
-     alphaCoef(NULL), betaCoef(NULL),
-     alpha(1.0), beta(1.0)
-{
-   A = 0.0; B = 0.0;
-}
-
-VectorSumCoefficient::VectorSumCoefficient(VectorCoefficient &_A,
-                                           VectorCoefficient &_B,
+VectorSumCoefficient::VectorSumCoefficient(VectorCoefficient &A,
+                                           VectorCoefficient &B,
                                           double _alpha, double _beta)
-   : VectorCoefficient(_A.GetVDim()),
-     ACoef(&_A), BCoef(&_B),
-     A(_A.GetVDim()), B(_A.GetVDim()),
-     alphaCoef(NULL), betaCoef(NULL),
-     alpha(_alpha), beta(_beta)
+   : VectorCoefficient(A.GetVDim()), a(&A), b(&B), alpha(_alpha), beta(_beta),
+     va(A.GetVDim())
 {
-   MFEM_ASSERT(_A.GetVDim() == _B.GetVDim(),
-               "VectorSumCoefficient:  "
-               "Arguments must have the same dimension.");
-}
-
-VectorSumCoefficient::VectorSumCoefficient(VectorCoefficient &_A,
-                                           VectorCoefficient &_B,
-                                           Coefficient &_alpha,
-                                           Coefficient &_beta)
-   : VectorCoefficient(_A.GetVDim()),
-     ACoef(&_A), BCoef(&_B),
-     A(_A.GetVDim()),
-     B(_A.GetVDim()),
-     alphaCoef(&_alpha),
-     betaCoef(&_beta),
-     alpha(0.0), beta(0.0)
-{
-   MFEM_ASSERT(_A.GetVDim() == _B.GetVDim(),
+   MFEM_ASSERT(A.GetVDim() == B.GetVDim(),
               "VectorSumCoefficient:  "
               "Arguments must have the same dimension.");
 }
@@ -466,47 +447,26 @@ VectorSumCoefficient::VectorSumCoefficient(VectorCoefficient &_A,
 void VectorSumCoefficient::Eval(Vector &V, ElementTransformation &T,
                                const IntegrationPoint &ip)
 {
-   V.SetSize(A.Size());
-   if (    ACoef) { ACoef->Eval(A, T, ip); }
-   if (    BCoef) { BCoef->Eval(B, T, ip); }
-   if (alphaCoef) { alpha = alphaCoef->Eval(T, ip); }
-   if ( betaCoef) { beta  = betaCoef->Eval(T, ip); }
-   add(alpha, A, beta, B, V);
+   b->Eval(V, T, ip);
+   if ( beta != 1.0 ) { V *= beta; }
+   a->Eval(va, T, ip);
+   V.Add(alpha, va);
 }

-ScalarVectorProductCoefficient::ScalarVectorProductCoefficient(
-   double A,
-   VectorCoefficient &B)
-   : VectorCoefficient(B.GetVDim()), aConst(A), a(NULL), b(&B)
-{}
-
 ScalarVectorProductCoefficient::ScalarVectorProductCoefficient(
   Coefficient &A,
   VectorCoefficient &B)
-   : VectorCoefficient(B.GetVDim()), aConst(0.0), a(&A), b(&B)
+   : VectorCoefficient(B.GetVDim()), a(&A), b(&B)
 {}

 void ScalarVectorProductCoefficient::Eval(Vector &V, ElementTransformation &T,
                                          const IntegrationPoint &ip)
 {
-   double sa = (a == NULL) ? aConst : a->Eval(T, ip);
+   double sa = a->Eval(T, ip);
   b->Eval(V, T, ip);
   V *= sa;
 }

-NormalizedVectorCoefficient::NormalizedVectorCoefficient(VectorCoefficient &A,
-                                                         double _tol)
-   : VectorCoefficient(A.GetVDim()), a(&A), tol(_tol)
-{}
-
-void NormalizedVectorCoefficient::Eval(Vector &V, ElementTransformation &T,
-                                       const IntegrationPoint &ip)
-{
-   a->Eval(V, T, ip);
-   double nv = V.Norml2();
-   V *= (nv > tol) ? (1.0/nv) : 0.0;
-}
-
 VectorCrossProductCoefficient::VectorCrossProductCoefficient(
   VectorCoefficient &A,
   VectorCoefficient &B)
@@ -528,18 +488,17 @@ void VectorCrossProductCoefficient::Eval(Vector &V, ElementTransformation &T,
   V[2] = va[0] * vb[1] - va[1] * vb[0];
 }

-MatrixVectorProductCoefficient::MatrixVectorProductCoefficient(
-   MatrixCoefficient &A, VectorCoefficient &B)
+MatVecCoefficient::MatVecCoefficient(MatrixCoefficient &A,
+                                     VectorCoefficient &B)
   : VectorCoefficient(A.GetHeight()), a(&A), b(&B),
     ma(A.GetHeight(), A.GetWidth()), vb(B.GetVDim())
 {
   MFEM_ASSERT(A.GetWidth() == B.GetVDim(),
-               "MatrixVectorProductCoefficient:  "
-               "Arguments have incompatible dimensions.");
+               "MatVecCoefficient:  Arguments have incompatible dimensions.");
 }

-void MatrixVectorProductCoefficient::Eval(Vector &V, ElementTransformation &T,
-                                          const IntegrationPoint &ip)
+void MatVecCoefficient::Eval(Vector &V, ElementTransformation &T,
+                             const IntegrationPoint &ip)
 {
   a->Eval(ma, T, ip);
   b->Eval(vb, T, ip);
@@ -575,23 +534,17 @@ void MatrixSumCoefficient::Eval(DenseMatrix &M, ElementTransformation &T,
   M.Add(alpha, ma);
 }

-ScalarMatrixProductCoefficient::ScalarMatrixProductCoefficient(
-   double A,
-   MatrixCoefficient &B)
-   : MatrixCoefficient(B.GetHeight(), B.GetWidth()), aConst(A), a(NULL), b(&B)
-{}
-
 ScalarMatrixProductCoefficient::ScalarMatrixProductCoefficient(
   Coefficient &A,
   MatrixCoefficient &B)
-   : MatrixCoefficient(B.GetHeight(), B.GetWidth()), aConst(0.0), a(&A), b(&B)
+   : MatrixCoefficient(B.GetHeight(), B.GetWidth()), a(&A), b(&B)
 {}

 void ScalarMatrixProductCoefficient::Eval(DenseMatrix &M,
                                          ElementTransformation &T,
                                          const IntegrationPoint &ip)
 {
-   double sa = (a == NULL) ? aConst : a->Eval(T, ip);
+   double sa = a->Eval(T, ip);
   b->Eval(M, T, ip);
   M *= sa;
 }
@@ -645,30 +598,6 @@ void OuterProductCoefficient::Eval(DenseMatrix &M, ElementTransformation &T,
   }
 }

-CrossCrossCoefficient::CrossCrossCoefficient(Coefficient &A,
-                                             VectorCoefficient &K)
-   : MatrixCoefficient(K.GetVDim(), K.GetVDim()), aConst(0.0), a(&A), k(&K),
-     vk(K.GetVDim())
-{}
-
-void CrossCrossCoefficient::Eval(DenseMatrix &M, ElementTransformation &T,
-                                 const IntegrationPoint &ip)
-{
-   k->Eval(vk, T, ip);
-   M.SetSize(vk.Size(), vk.Size());
-   M = 0.0;
-   double k2 = vk*vk;
-   for (int i=0; i<vk.Size(); i++)
-   {
-      M(i, i) = k2;
-      for (int j=0; j<vk.Size(); j++)
-      {
-         M(i, j) -= vk[i] * vk[j];
-      }
-   }
-   M *= ((a == NULL ) ? aConst : a->Eval(T, ip) );
-}
-
 double LpNormLoop(double p, Coefficient &coeff, Mesh &mesh,
                  const IntegrationRule *irs[])
 {
@@ -846,61 +775,4 @@ double ComputeGlobalLpNorm(double p, VectorCoefficient &coeff, ParMesh &pmesh,
 }
 #endif

-VectorQuadratureFunctionCoefficient::VectorQuadratureFunctionCoefficient(
-   QuadratureFunction &qf)
-   : VectorCoefficient(qf.GetVDim()), QuadF(qf), index(0) { }
-
-void VectorQuadratureFunctionCoefficient::SetComponent(int _index, int _length)
-{
-   MFEM_VERIFY(_index >= 0, "Index must be >= 0");
-   MFEM_VERIFY(_index < QuadF.GetVDim(),
-               "Index must be < QuadratureFunction length");
-   index = _index;
-
-   MFEM_VERIFY(_length > 0, "Length must be > 0");
-   MFEM_VERIFY(_length <= QuadF.GetVDim() - index,
-               "Length must be <= (QuadratureFunction length - index)");
-
-   vdim = _length;
-}
-
-void VectorQuadratureFunctionCoefficient::Eval(Vector &V,
-                                               ElementTransformation &T,
-                                               const IntegrationPoint &ip)
-{
-   QuadF.HostRead();
-
-   if (index == 0 && vdim == QuadF.GetVDim())
-   {
-      QuadF.GetElementValues(T.ElementNo, ip.index, V);
-   }
-   else
-   {
-      Vector temp;
-      QuadF.GetElementValues(T.ElementNo, ip.index, temp);
-      V.SetSize(vdim);
-      for (int i = 0; i < vdim; i++)
-      {
-         V(i) = temp(index + i);
-      }
-   }
-
-   return;
-}
-
-QuadratureFunctionCoefficient::QuadratureFunctionCoefficient(
-   QuadratureFunction &qf) : QuadF(qf)
-{
-   MFEM_VERIFY(qf.GetVDim() == 1, "QuadratureFunction's vdim must be 1");
-}
-
-double QuadratureFunctionCoefficient::Eval(ElementTransformation &T,
-                                           const IntegrationPoint &ip)
-{
-   QuadF.HostRead();
-   Vector temp(1);
-   QuadF.GetElementValues(T.ElementNo, ip.index, temp);
-   return temp[0];
-}
-
 }
@@ -342,10 +342,11 @@ SesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
                                   int ci)
 {
   FiniteElementSpace * fes = blfr->FESpace();
+
   int vsize  = fes->GetVSize();

   // Allocate temporary vectors
-   Vector b_0(vsize); b_0 = 0.0;
+   Vector b_0(vsize);  b_0 = 0.0;

   // Extract the real and imaginary parts of the input vectors
   MFEM_ASSERT(x.Size() == 2 * vsize, "Input GridFunction of incorrect size!");
@@ -359,7 +360,8 @@ SesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
   if (conv == ComplexOperator::BLOCK_SYMMETRIC) { b_i *= -1.0; }

   int tvsize = fes->GetTrueVSize();
-   OperatorHandle A_r, A_i;
+   SparseMatrix * A_r = nullptr;
+   SparseMatrix * A_i = nullptr;

   X.SetSize(2 * tvsize);
   B.SetSize(2 * tvsize);
@@ -372,39 +374,42 @@ SesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,

   if (RealInteg())
   {
+      A_r = new SparseMatrix;
      blfr->SetDiagonalPolicy(diag_policy);

      b_0 = b_r;
-      blfr->FormLinearSystem(ess_tdof_list, x_r, b_0, A_r, X_0, B_0, ci);
+      blfr->FormLinearSystem(ess_tdof_list, x_r, b_0, *A_r, X_0, B_0, ci);
      X_r = X_0; B_r = B_0;

      b_0 = b_i;
-      blfr->FormLinearSystem(ess_tdof_list, x_i, b_0, A_r, X_0, B_0, ci);
+      blfr->FormLinearSystem(ess_tdof_list, x_i, b_0, *A_r, X_0, B_0, ci);
      X_i = X_0; B_i = B_0;

      if (ImagInteg())
      {
+         A_i = new SparseMatrix;
         blfi->SetDiagonalPolicy(mfem::Matrix::DiagonalPolicy::DIAG_ZERO);

         b_0 = 0.0;
-         blfi->FormLinearSystem(ess_tdof_list, x_i, b_0, A_i, X_0, B_0, false);
+         blfi->FormLinearSystem(ess_tdof_list, x_i, b_0, *A_i, X_0, B_0, false);
         B_r -= B_0;

         b_0 = 0.0;
-         blfi->FormLinearSystem(ess_tdof_list, x_r, b_0, A_i, X_0, B_0, false);
+         blfi->FormLinearSystem(ess_tdof_list, x_r, b_0, *A_i, X_0, B_0, false);
         B_i += B_0;
      }
   }
   else if (ImagInteg())
   {
+      A_i = new SparseMatrix;
      blfi->SetDiagonalPolicy(diag_policy);

      b_0 = b_i;
-      blfi->FormLinearSystem(ess_tdof_list, x_r, b_0, A_i, X_0, B_0, ci);
+      blfi->FormLinearSystem(ess_tdof_list, x_r, b_0, *A_i, X_0, B_0, ci);
      X_r = X_0; B_i = B_0;

      b_0 = b_r; b_0 *= -1.0;
-      blfi->FormLinearSystem(ess_tdof_list, x_i, b_0, A_i, X_0, B_0, ci);
+      blfi->FormLinearSystem(ess_tdof_list, x_i, b_0, *A_i, X_0, B_0, ci);
      X_i = X_0; B_r = B_0; B_r *= -1.0;
   }
   else
@@ -412,55 +417,16 @@ SesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
      MFEM_ABORT("Real and Imaginary part of the Sesquilinear form are empty");
   }

-   if (RealInteg() && ImagInteg())
-   {
-      // Modify RHS and offdiagonal blocks (imaginary parts of the matrix) to
-      // conform with standard essential BC treatment
-      if (A_i.Is<ConstrainedOperator>())
-      {
-         int n = ess_tdof_list.Size();
-         for (int k = 0; k < n; k++)
-         {
-            int j = ess_tdof_list[k];
-            B_r(j) = X_r(j);
-            B_i(j) = X_i(j);
-         }
-         A_i.As<ConstrainedOperator>()->SetDiagonalPolicy
-         (mfem::Operator::DiagonalPolicy::DIAG_ZERO);
-      }
-   }
-
   if (conv == ComplexOperator::BLOCK_SYMMETRIC)
   {
      B_i *= -1.0;
      b_i *= -1.0;
   }
-
   // A = A_r + i A_i
   A.Clear();
-   if ( A_r.Type() == Operator::MFEM_SPARSEMAT ||
-        A_i.Type() == Operator::MFEM_SPARSEMAT )
-   {
-      ComplexSparseMatrix * A_sp =
-         new ComplexSparseMatrix(A_r.As<SparseMatrix>(),
-                                 A_i.As<SparseMatrix>(),
-                                 A_r.OwnsOperator(),
-                                 A_i.OwnsOperator(),
-                                 conv);
-      A.Reset<ComplexSparseMatrix>(A_sp, true);
-   }
-   else
-   {
-      ComplexOperator * A_op =
-         new ComplexOperator(A_r.Ptr(),
-                             A_i.Ptr(),
-                             A_r.OwnsOperator(),
-                             A_i.OwnsOperator(),
-                             conv);
-      A.Reset<ComplexOperator>(A_op, true);
-   }
-   A_r.SetOperatorOwner(false);
-   A_i.SetOperatorOwner(false);
+   ComplexSparseMatrix * A_sp;
+   A_sp = new ComplexSparseMatrix(A_r, A_i, true, true, conv);
+   A.Reset<ComplexSparseMatrix>(A_sp, true);
 }

 void
@@ -468,60 +434,31 @@ SesquilinearForm::FormSystemMatrix(const Array<int> &ess_tdof_list,
                                   OperatorHandle &A)

 {
-   OperatorHandle A_r, A_i;
+   SparseMatrix * A_r = nullptr;
+   SparseMatrix * A_i = nullptr;
+
   if (RealInteg())
   {
+      A_r = new SparseMatrix;
      blfr->SetDiagonalPolicy(diag_policy);
-      blfr->FormSystemMatrix(ess_tdof_list, A_r);
+      blfr->FormSystemMatrix(ess_tdof_list, *A_r);
   }
   if (ImagInteg())
   {
-      blfi->SetDiagonalPolicy(RealInteg() ?
-                              mfem::Matrix::DiagonalPolicy::DIAG_ZERO :
-                              diag_policy);
-      blfi->FormSystemMatrix(ess_tdof_list, A_i);
+      A_i = new SparseMatrix;
+      blfr->SetDiagonalPolicy(diag_policy);
+      blfi->FormSystemMatrix(ess_tdof_list, *A_i);
   }
   if (!RealInteg() && !ImagInteg())
   {
      MFEM_ABORT("Both Real and Imaginary part of the Sesquilinear form are empty");
   }

-   if (RealInteg() && ImagInteg())
-   {
-      // Modify offdiagonal blocks (imaginary parts of the matrix) to conform
-      // with standard essential BC treatment
-      if (A_i.Is<ConstrainedOperator>())
-      {
-         A_i.As<ConstrainedOperator>()->SetDiagonalPolicy
-         (mfem::Operator::DiagonalPolicy::DIAG_ZERO);
-      }
-   }
-
   // A = A_r + i A_i
   A.Clear();
-   if ( A_r.Type() == Operator::MFEM_SPARSEMAT ||
-        A_i.Type() == Operator::MFEM_SPARSEMAT )
-   {
-      ComplexSparseMatrix * A_sp =
-         new ComplexSparseMatrix(A_r.As<SparseMatrix>(),
-                                 A_i.As<SparseMatrix>(),
-                                 A_r.OwnsOperator(),
-                                 A_i.OwnsOperator(),
-                                 conv);
-      A.Reset<ComplexSparseMatrix>(A_sp, true);
-   }
-   else
-   {
-      ComplexOperator * A_op =
-         new ComplexOperator(A_r.Ptr(),
-                             A_i.Ptr(),
-                             A_r.OwnsOperator(),
-                             A_i.OwnsOperator(),
-                             conv);
-      A.Reset<ComplexOperator>(A_op, true);
-   }
-   A_r.SetOperatorOwner(false);
-   A_i.SetOperatorOwner(false);
+   ComplexSparseMatrix * A_sp =
+      new ComplexSparseMatrix(A_r, A_i, true, true, conv);
+   A.Reset<ComplexSparseMatrix>(A_sp, true);
 }

 void
@@ -709,7 +646,7 @@ ParComplexLinearForm::ParComplexLinearForm(ParFiniteElementSpace *pfes,
   int n = (HYPRE_AssumedPartitionCheck()) ? 2 : pfes->GetNRanks();
   tdof_offsets = new HYPRE_Int[n+1];

-   for (int i = 0; i <= n; i++)
+   for (int i=0; i<=n; i++)
   {
      tdof_offsets[i] = 2 * tdof_offsets_fes[i];
   }
@@ -717,8 +654,7 @@ ParComplexLinearForm::ParComplexLinearForm(ParFiniteElementSpace *pfes,


 ParComplexLinearForm::ParComplexLinearForm(ParFiniteElementSpace *pfes,
-                                           ParLinearForm *plf_r,
-                                           ParLinearForm *plf_i,
+                                           ParLinearForm *plf_r, ParLinearForm *plf_i,
                                           ComplexOperator::Convention
                                           convention)
   : Vector(2*(pfes->GetVSize())),
@@ -734,7 +670,7 @@ ParComplexLinearForm::ParComplexLinearForm(ParFiniteElementSpace *pfes,
   int n = (HYPRE_AssumedPartitionCheck()) ? 2 : pfes->GetNRanks();
   tdof_offsets = new HYPRE_Int[n+1];

-   for (int i = 0; i <= n; i++)
+   for (int i=0; i<=n; i++)
   {
      tdof_offsets[i] = 2 * tdof_offsets_fes[i];
   }
@@ -881,8 +817,7 @@ ParSesquilinearForm::ParSesquilinearForm(ParFiniteElementSpace *pf,
 {}

 ParSesquilinearForm::ParSesquilinearForm(ParFiniteElementSpace *pf,
-                                         ParBilinearForm *pbfr,
-                                         ParBilinearForm *pbfi,
+                                         ParBilinearForm *pbfr, ParBilinearForm *pbfi,
                                         ComplexOperator::Convention convention)
   : conv(convention),
     pblfr(new ParBilinearForm(pf,pbfr)),
@@ -978,10 +913,9 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
   int vsize = pfes->GetVSize();

   // Allocate temporary vectors
-   Vector b_0(vsize); b_0 = 0.0;
+   Vector b_0(vsize);  b_0 = 0.0;

   // Extract the real and imaginary parts of the input vectors
-   MFEM_ASSERT(x.Size() == 2 * vsize, "Input GridFunction of incorrect size!");
   Vector x_r(x.GetData(), vsize);
   Vector x_i(&(x.GetData())[vsize], vsize);

@@ -1040,34 +974,25 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
      MFEM_ABORT("Real and Imaginary part of the Sesquilinear form are empty");
   }

+   // Modify RHS and offdiagonal blocks (Imaginary parts of the matrix) to
+   // conform with standard essential BC treatment i.e. zero out rows and
+   // columns and place ones on the diagonal.
   if (RealInteg() && ImagInteg())
   {
-      int n = ess_tdof_list.Size();
-      // Modify RHS to conform with standard essential BC treatment
-      for (int k = 0; k < n; k++)
-      {
-         int j=ess_tdof_list[k];
-         B_r(j) = X_r(j);
-         B_i(j) = X_i(j);
-      }
-      // Modify offdiagonal blocks (imaginary parts of the matrix) to conform
-      // with standard essential BC treatment
      if ( A_i.Type() == Operator::Hypre_ParCSR )
      {
-         HypreParMatrix * Ah;
-         A_i.Get(Ah);
-         hypre_ParCSRMatrix *Aih = *Ah;
-         for (int k = 0; k < n; k++)
+         HypreParMatrix * Ah;  A_i.Get(Ah);
+         int n = ess_tdof_list.Size();
+         hypre_ParCSRMatrix * Aih =
+            (hypre_ParCSRMatrix *)const_cast<HypreParMatrix&>(*Ah);
+         for (int k=0; k<n; k++)
         {
-            int j = ess_tdof_list[k];
+            int j=ess_tdof_list[k];
            Aih->diag->data[Aih->diag->i[j]] = 0.0;
+            B_r(j) = X_r(j);
+            B_i(j) = X_i(j);
         }
      }
-      else
-      {
-         A_i.As<ConstrainedOperator>()->SetDiagonalPolicy
-         (mfem::Operator::DiagonalPolicy::DIAG_ZERO);
-      }
   }

   if (conv == ComplexOperator::BLOCK_SYMMETRIC)
@@ -1075,7 +1000,6 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
      B_i *= -1.0;
      b_i *= -1.0;
   }
-
   // A = A_r + i A_i
   A.Clear();
   if ( A_r.Type() == Operator::Hypre_ParCSR ||
@@ -1099,8 +1023,6 @@ ParSesquilinearForm::FormLinearSystem(const Array<int> &ess_tdof_list,
                             conv);
      A.Reset<ComplexOperator>(A_op, true);
   }
-   A_r.SetOperatorOwner(false);
-   A_i.SetOperatorOwner(false);
 }

 void
@@ -1121,27 +1043,25 @@ ParSesquilinearForm::FormSystemMatrix(const Array<int> &ess_tdof_list,
      MFEM_ABORT("Both Real and Imaginary part of the Sesquilinear form are empty");
   }

+   // Modify offdiagonal blocks (Imaginary parts of the matrix) to conform with
+   // standard essential BC treatment i.e. zero out rows and columns and place
+   // ones on the diagonal.
   if (RealInteg() && ImagInteg())
   {
-      // Modify offdiagonal blocks (imaginary parts of the matrix) to conform
-      // with standard essential BC treatment
      if ( A_i.Type() == Operator::Hypre_ParCSR )
      {
         int n = ess_tdof_list.Size();
-         HypreParMatrix * Ah;
-         A_i.Get(Ah);
-         hypre_ParCSRMatrix * Aih = *Ah;
-         for (int k = 0; k < n; k++)
+         int j;
+
+         HypreParMatrix * Ah;  A_i.Get(Ah);
+         hypre_ParCSRMatrix * Aih =
+            (hypre_ParCSRMatrix *)const_cast<HypreParMatrix&>(*Ah);
+         for (int k=0; k<n; k++)
         {
-            int j = ess_tdof_list[k];
+            j=ess_tdof_list[k];
            Aih->diag->data[Aih->diag->i[j]] = 0.0;
         }
      }
-      else
-      {
-         A_i.As<ConstrainedOperator>()->SetDiagonalPolicy
-         (mfem::Operator::DiagonalPolicy::DIAG_ZERO);
-      }
   }

   // A = A_r + i A_i
@@ -1167,8 +1087,6 @@ ParSesquilinearForm::FormSystemMatrix(const Array<int> &ess_tdof_list,
                             conv);
      A.Reset<ComplexOperator>(A_op, true);
   }
-   A_r.SetOperatorOwner(false);
-   A_i.SetOperatorOwner(false);
 }

 void
@@ -219,21 +219,6 @@ public:
   void SetConvention(const ComplexOperator::Convention &
                      convention) { conv = convention; }

-   /// Set the desired assembly level.
-   /** Valid choices are:
-
-       - AssemblyLevel::FULL  (default)
-       - AssemblyLevel::PARTIAL
-       - AssemblyLevel::ELEMENT
-       - AssemblyLevel::NONE
-
-       This method must be called before assembly. */
-   void SetAssemblyLevel(AssemblyLevel assembly_level)
-   {
-      blfr->SetAssemblyLevel(assembly_level);
-      blfi->SetAssemblyLevel(assembly_level);
-   }
-
   BilinearForm & real() { return *blfr; }
   BilinearForm & imag() { return *blfi; }
   const BilinearForm & real() const { return *blfr; }
@@ -493,7 +478,7 @@ public:
 /** Class for a parallel sesquilinear form

    A sesquilinear form is a generalization of a bilinear form to complex-valued
-    fields. Sesquilinear forms are linear in the second argument but the
+    fields. Sesquilinear forms are linear in the second argument but but the
    first argument involves a complex conjugate in the sense that:

                a(alpha u, beta v) = conj(alpha) beta a(u, v)
@@ -539,21 +524,6 @@ public:
   void SetConvention(const ComplexOperator::Convention &
                      convention) { conv = convention; }

-   /// Set the desired assembly level.
-   /** Valid choices are:
-
-       - AssemblyLevel::FULL  (default)
-       - AssemblyLevel::PARTIAL
-       - AssemblyLevel::ELEMENT
-       - AssemblyLevel::NONE
-
-       This method must be called before assembly. */
-   void SetAssemblyLevel(AssemblyLevel assembly_level)
-   {
-      pblfr->SetAssemblyLevel(assembly_level);
-      pblfi->SetAssemblyLevel(assembly_level);
-   }
-
   ParBilinearForm & real() { return *pblfr; }
   ParBilinearForm & imag() { return *pblfi; }
   const ParBilinearForm & real() const { return *pblfr; }
@@ -415,6 +415,9 @@ void VisItDataCollection::SetMesh(MPI_Comm comm, Mesh *new_mesh)
 void VisItDataCollection::RegisterField(const std::string& name,
                                        GridFunction *gf)
 {
+   DataCollection::RegisterField(name, gf);
+   field_info_map[name] = VisItFieldInfo("nodes", gf->VectorDim());
+
   int LOD = 1;
   if (gf->FESpace()->GetNURBSext())
   {
@@ -428,27 +431,6 @@ void VisItDataCollection::RegisterField(const std::string& name,
      }
   }

-   DataCollection::RegisterField(name, gf);
-   field_info_map[name] = VisItFieldInfo("nodes", gf->VectorDim(), LOD);
-   visit_levels_of_detail = std::max(visit_levels_of_detail, LOD);
-}
-
-void VisItDataCollection::RegisterQField(const std::string& name,
-                                         QuadratureFunction *qf)
-{
-   int LOD = -1;
-   Mesh *mesh = qf->GetSpace()->GetMesh();
-   for (int e=0; e<qf->GetSpace()->GetNE(); e++)
-   {
-      int locLOD = GlobGeometryRefiner.GetRefinementLevelFromElems(
-                      mesh->GetElementBaseGeometry(e),
-                      qf->GetElementIntRule(e).GetNPoints());
-
-      LOD = std::max(LOD,locLOD);
-   }
-
-   DataCollection::RegisterQField(name, qf);
-   field_info_map[name] = VisItFieldInfo("elements", 1, LOD);
   visit_levels_of_detail = std::max(visit_levels_of_detail, LOD);
 }

@@ -616,28 +598,14 @@ void VisItDataCollection::LoadFields()
      // TODO: 1) load parallel GridFunction on one processor
      if (serial)
      {
-         if ((it->second).association == "nodes")
-         {
-            field_map.Register(it->first, new GridFunction(mesh, file), own_data);
-         }
-         else if ((it->second).association == "elements")
-         {
-            q_field_map.Register(it->first, new QuadratureFunction(mesh, file), own_data);
-         }
+         field_map.Register(it->first, new GridFunction(mesh, file), own_data);
      }
      else
      {
 #ifdef MFEM_USE_MPI
-         if ((it->second).association == "nodes")
-         {
-            field_map.Register(
-               it->first,
-               new ParGridFunction(dynamic_cast<ParMesh*>(mesh), file), own_data);
-         }
-         else if ((it->second).association == "elements")
-         {
-            q_field_map.Register(it->first, new QuadratureFunction(mesh, file), own_data);
-         }
+         field_map.Register(
+            it->first,
+            new ParGridFunction(dynamic_cast<ParMesh*>(mesh), file), own_data);
 #else
         error = READ_ERROR;
         MFEM_WARNING("Reading parallel format in serial is not supported");
@@ -672,7 +640,7 @@ std::string VisItDataCollection::GetVisItRootString()
   {
      ftags["assoc"] = picojson::value((it->second).association);
      ftags["comps"] = picojson::value(to_string((it->second).num_components));
-      ftags["lod"] = picojson::value(to_string((it->second).lod));
+      ftags["lod"] = picojson::value(to_string(visit_levels_of_detail));
      field["path"] = picojson::value(path_str + it->first + file_ext_format);
      field["tags"] = picojson::value(ftags);
      fields[it->first] = picojson::value(field);
@@ -771,6 +739,12 @@ ParaViewDataCollection::ParaViewDataCollection(const std::string&
 #endif
 }

+void ParaViewDataCollection::RegisterField(const std::string& field_name,
+                                           mfem::GridFunction *gf)
+{
+   DataCollection::RegisterField(field_name,gf);
+}
+
 void ParaViewDataCollection::SetLevelsOfDetail(int levels_of_detail_)
 {
   levels_of_detail = levels_of_detail_;
@@ -841,7 +815,7 @@ void ParaViewDataCollection::Save()
   // the directory is created

   // create pvd file if needed
-   if (myid == 0 && !pvd_stream.is_open())
+   if (!pvd_stream.is_open())
   {
      std::string dpath=GenerateCollectionPath();
      std::string pvdname=dpath+"/"+GeneratePVDFileName();
@@ -391,10 +391,9 @@ class VisItFieldInfo
 public:
   std::string association;
   int num_components;
-   int lod;
-   VisItFieldInfo() { association = ""; num_components = 0; lod = 1;}
-   VisItFieldInfo(std::string _association, int _num_components, int _lod = 1)
-   { association = _association; num_components = _num_components; lod =_lod;}
+   VisItFieldInfo() { association = ""; num_components = 0; }
+   VisItFieldInfo(std::string _association, int _num_components)
+   { association = _association; num_components = _num_components; }
 };

 /// Data collection with VisIt I/O routines
@@ -446,12 +445,6 @@ public:
   /// Add a grid function to the collection and update the root file
   virtual void RegisterField(const std::string& field_name, GridFunction *gf);

-   /// Add a quadrature function to the collection and update the root file.
-   /** Visualization of quadrature function is not supported in VisIt(3.12).
-       A patch has been sent to VisIt developers in June 2020. */
-   virtual void RegisterQField(const std::string& q_field_name,
-                               QuadratureFunction *qf);
-
   /// Set VisIt parameter: default levels of detail for the MultiresControl
   void SetLevelsOfDetail(int levels_of_detail);

@@ -508,6 +501,10 @@ public:
   ParaViewDataCollection(const std::string& collection_name,
                          mfem::Mesh *mesh_ = NULL);

+   /// Add a grid function to the collection
+   virtual void RegisterField(const std::string& field_name,
+                              mfem::GridFunction *gf) override;
+
   /// Set refinement levels - every element is uniformly split based on
   /// levels_of_detail_
   void SetLevelsOfDetail(int levels_of_detail_);
@@ -19,7 +19,6 @@ namespace mfem
 ElementTransformation::ElementTransformation()
   : IntPoint(static_cast<IntegrationPoint *>(NULL)),
     EvalState(0),
-     geom(Geometry::INVALID),
     Attribute(-1),
     ElementNo(-1)
 { }
@@ -552,155 +551,4 @@ void IntegrationPointTransformation::Transform (const IntegrationRule &ir1,
   }
 }

-void FaceElementTransformations::SetIntPoint(const IntegrationPoint *face_ip)
-{
-   IsoparametricTransformation::SetIntPoint(face_ip);
-
-   if (mask & 4)
-   {
-      Loc1.Transform(*face_ip, eip1);
-      if (Elem1)
-      {
-         Elem1->SetIntPoint(&eip1);
-      }
-   }
-   if (mask & 8)
-   {
-      Loc2.Transform(*face_ip, eip2);
-      if (Elem2)
-      {
-         Elem2->SetIntPoint(&eip2);
-      }
-   }
-}
-
-ElementTransformation &
-FaceElementTransformations::GetElement1Transformation()
-{
-   MFEM_VERIFY(mask & HAVE_ELEM1 && Elem1 != NULL, "The ElementTransformation "
-               "for the element has not been configured for side 1.");
-   return *Elem1;
-}
-
-ElementTransformation &
-FaceElementTransformations::GetElement2Transformation()
-{
-   MFEM_VERIFY(mask & HAVE_ELEM2 && Elem2 != NULL, "The ElementTransformation "
-               "for the element has not been configured for side 2.");
-   return *Elem2;
-}
-
-IntegrationPointTransformation &
-FaceElementTransformations::GetIntPoint1Transformation()
-{
-   MFEM_VERIFY(mask & HAVE_LOC1, "The IntegrationPointTransformation "
-               "for the element has not been configured for side 1.");
-   return Loc1;
-}
-
-IntegrationPointTransformation &
-FaceElementTransformations::GetIntPoint2Transformation()
-{
-   MFEM_VERIFY(mask & HAVE_LOC2, "The IntegrationPointTransformation "
-               "for the element has not been configured for side 2.");
-   return Loc2;
-}
-
-void FaceElementTransformations::Transform(const IntegrationPoint &ip,
-                                           Vector &trans)
-{
-   MFEM_VERIFY(mask & HAVE_FACE, "The ElementTransformation "
-               "for the face has not been configured.");
-   IsoparametricTransformation::Transform(ip, trans);
-}
-
-void FaceElementTransformations::Transform(const IntegrationRule &ir,
-                                           DenseMatrix &tr)
-{
-   MFEM_VERIFY(mask & HAVE_FACE, "The ElementTransformation "
-               "for the face has not been configured.");
-   IsoparametricTransformation::Transform(ir, tr);
-}
-
-void FaceElementTransformations::Transform(const DenseMatrix &matrix,
-                                           DenseMatrix &result)
-{
-   MFEM_VERIFY(mask & HAVE_FACE, "The ElementTransformation "
-               "for the face has not been configured.");
-   IsoparametricTransformation::Transform(matrix, result);
-}
-
-double FaceElementTransformations::CheckConsistency(int print_level,
-                                                    std::ostream &out)
-{
-   // Check that the face vertices are mapped to the same physical location
-   // when using the following three transformations:
-   // - the face transformation, *this
-   // - Loc1 + Elem1
-   // - Loc2 + Elem2, if present.
-
-   const bool have_face = (mask & 16);
-   const bool have_el1 = (mask & 1) && (mask & 4);
-   const bool have_el2 = (mask & 2) && (mask & 8) && (Elem2No >= 0);
-   if (int(have_face) + int(have_el1) + int(have_el2) < 2)
-   {
-      // need at least two different transformations to perform a check
-      return 0.0;
-   }
-
-   const IntegrationRule &v_ir = *Geometries.GetVertices(GetGeometryType());
-
-   double max_dist = 0.0;
-   Vector dist(v_ir.GetNPoints());
-   DenseMatrix coords_base, coords_el;
-   IntegrationRule v_eir(v_ir.GetNPoints());
-   if (have_face)
-   {
-      Transform(v_ir, coords_base);
-      if (print_level > 0)
-      {
-         out << "\nface vertex coordinates (from face transform):\n"
-             << "----------------------------------------------\n";
-         coords_base.PrintT(out, coords_base.Height());
-      }
-   }
-   if (have_el1)
-   {
-      Loc1.Transform(v_ir, v_eir);
-      Elem1->Transform(v_eir, coords_el);
-      if (print_level > 0)
-      {
-         out << "\nface vertex coordinates (from element 1 transform):\n"
-             << "---------------------------------------------------\n";
-         coords_el.PrintT(out, coords_el.Height());
-      }
-      if (have_face)
-      {
-         coords_el -= coords_base;
-         coords_el.Norm2(dist);
-         max_dist = std::max(max_dist, dist.Normlinf());
-      }
-      else
-      {
-         coords_base = coords_el;
-      }
-   }
-   if (have_el2)
-   {
-      Loc2.Transform(v_ir, v_eir);
-      Elem2->Transform(v_eir, coords_el);
-      if (print_level > 0)
-      {
-         out << "\nface vertex coordinates (from element 2 transform):\n"
-             << "---------------------------------------------------\n";
-         coords_el.PrintT(out, coords_el.Height());
-      }
-      coords_el -= coords_base;
-      coords_el.Norm2(dist);
-      max_dist = std::max(max_dist, dist.Normlinf());
-   }
-
-   return max_dist;
-}
-
 }
@@ -38,12 +38,9 @@ protected:
   };
   Geometry::Type geom;

-   /** @brief Evaluate the Jacobian of the transformation at the IntPoint and
-       store it in dFdx. */
+   // Evaluate the Jacobian of the transformation at the IntPoint and store it
+   // in dFdx.
   virtual const DenseMatrix &EvalJacobian() = 0;
-
-   /** @brief Evaluate the Hessian of the transformation at the IntPoint and
-       store it in d2Fdx2. */
   virtual const DenseMatrix &EvalHessian() = 0;

   double EvalWeight();
@@ -51,53 +48,18 @@ protected:
   const DenseMatrix &EvalInverseJ();

 public:
-
-   /** This enumeration declares the values stored in
-       ElementTransformation::ElementType and indicates which group of objects
-       the index stored in ElementTransformation::ElementNo refers:
-
-       | ElementType | Range of ElementNo
-       +-------------+-------------------------
-       | ELEMENT     | [0, Mesh::GetNE()     )
-       | BDR_ELEMENT | [0, Mesh::GetNBE()    )
-       | EDGE        | [0, Mesh::GetNEdges() )
-       | FACE        | [0, Mesh::GetNFaces() )
-       | BDR_FACE    | [0, Mesh::GetNBE()    )
-   */
-   enum
-   {
-      ELEMENT     = 1,
-      BDR_ELEMENT = 2,
-      EDGE        = 3,
-      FACE        = 4,
-      BDR_FACE    = 5
-   };
-
-   int Attribute, ElementNo, ElementType;
+   int Attribute, ElementNo;

   ElementTransformation();

-   /** @brief Set the integration point @a ip that weights and Jacobians will
-       be evaluated at. */
   void SetIntPoint(const IntegrationPoint *ip)
   { IntPoint = ip; EvalState = 0; }
-
-   /** @brief Get a const reference to the currently set integration point.  This
-       will return NULL if no integration point is set. */
   const IntegrationPoint &GetIntPoint() { return *IntPoint; }

-   /** @brief Transform integration point from reference coordinates to
-       physical coordinates and store them in the vector. */
   virtual void Transform(const IntegrationPoint &, Vector &) = 0;
-
-   /** @brief Transform all the integration points from the integration rule
-       from reference coordinates to physical
-       coordinates and store them as column vectors in the matrix. */
   virtual void Transform(const IntegrationRule &, DenseMatrix &) = 0;

-   /** @brief Transform all the integration points from the column vectors
-       of @a matrix from reference coordinates to physical
-       coordinates and store them as column vectors in @a result. */
+   /// Transform columns of 'matrix', store result in 'result'.
   virtual void Transform(const DenseMatrix &matrix, DenseMatrix &result) = 0;

   /** @brief Return the Jacobian matrix of the transformation at the currently
@@ -108,44 +70,27 @@ public:
   const DenseMatrix &Jacobian()
   { return (EvalState & JACOBIAN_MASK) ? dFdx : EvalJacobian(); }

-
-   /** @brief Return the Hessian matrix of the transformation at the currently
-       set IntegrationPoint, using the method SetIntPoint(). */
   const DenseMatrix &Hessian()
   { return (EvalState & HESSIAN_MASK) ? d2Fdx2 : EvalHessian(); }

-   /** @brief Return the weight of the Jacobian matrix of the transformation
-       at the currently set IntegrationPoint.
-       The Weight evaluates to \f$ \sqrt{\lvert J^T J \rvert} \f$. */
   double Weight() { return (EvalState & WEIGHT_MASK) ? Wght : EvalWeight(); }

-   /** @brief Return the adjugate of the Jacobian matrix of the transformation
-        at the currently set IntegrationPoint. */
   const DenseMatrix &AdjugateJacobian()
   { return (EvalState & ADJUGATE_MASK) ? adjJ : EvalAdjugateJ(); }

-   /** @brief Return the inverse of the Jacobian matrix of the transformation
-        at the currently set IntegrationPoint. */
   const DenseMatrix &InverseJacobian()
   { return (EvalState & INVERSE_MASK) ? invJ : EvalInverseJ(); }

-   /// Return the order of the current element we are using for the transformation.
   virtual int Order() const = 0;
-
-   /// Return the order of the elements of the Jacobian of the transformation.
   virtual int OrderJ() const = 0;
-
-   /** @brief Return the order of the determinant of the Jacobian (weight)
-       of the transformation. */
   virtual int OrderW() const = 0;
-
-   /// Return the order of \f$ adj(J)^T \nabla fi \f$
+   /// Order of adj(J)^t.grad(fi)
   virtual int OrderGrad(const FiniteElement *fe) const = 0;

   /// Return the Geometry::Type of the reference element.
   Geometry::Type GetGeometryType() const { return geom; }

-   /// Return the topological dimension of the reference element.
+   /// Return the dimension of the reference element.
   int GetDimension() const { return Geometry::Dimension[geom]; }

   /// Get the dimension of the target (physical) space.
@@ -341,7 +286,7 @@ public:
   virtual int Transform(const Vector &pt, IntegrationPoint &ip);
 };

-/// A standard isoparametric element transformation
+
 class IsoparametricTransformation : public ElementTransformation
 {
 private:
@@ -351,29 +296,26 @@ private:
   const FiniteElement *FElem;
   DenseMatrix PointMat; // dim x dof

-   /** @brief Evaluate the Jacobian of the transformation at the IntPoint and
-       store it in dFdx. */
+   // Evaluate the Jacobian of the transformation at the IntPoint and store it
+   // in dFdx.
   virtual const DenseMatrix &EvalJacobian();
   // Evaluate the Hessian of the transformation at the IntPoint and store it
   // in d2Fdx2.
   virtual const DenseMatrix &EvalHessian();
 public:
-   /// Set the element that will be used to compute the transformations
   void SetFE(const FiniteElement *FE) { FElem = FE; geom = FE->GetGeomType(); }
-
-   /// Get the current element used to compute the transformations
   const FiniteElement* GetFE() const { return FElem; }

   /// @brief Set the underlying point matrix describing the transformation.
   /** The dimensions of the matrix are space-dim x dof. The transformation is
       defined as
-           \f$ x = F( \hat x ) = P \phi( \hat x ) \f$

-       where \f$ \hat x \f$  is the reference point, @a x is the corresponding
-       physical point, @a P is the point matrix, and \f$ \phi( \hat x ) \f$ is
-       the column-vector of all basis functions evaluated at \f$ \hat x \f$ .
-       The columns of @a P represent the control points in physical space
-       defining the transformation. */
+           x = F(xh) = P . phi(xh),
+
+       where xh (x hat) is the reference point, x is the corresponding physical
+       point, P is the point matrix, and phi(xh) is the column-vector of all
+       basis functions evaluated at xh. The columns of P represent the control
+       points in physical space defining the transformation. */
   void SetPointMat(const DenseMatrix &pm) { PointMat = pm; }

   /// Return the stored point matrix.
@@ -382,44 +324,19 @@ public:
   /// Write access to the stored point matrix. Use with caution.
   DenseMatrix &GetPointMat() { return PointMat; }

-   /// Set the FiniteElement Geometry for the reference elements being used.
   void SetIdentityTransformation(Geometry::Type GeomType);

-   /** @brief Transform integration point from reference coordinates to
-       physical coordinates and store them in the vector. */
   virtual void Transform(const IntegrationPoint &, Vector &);
-
-   /** @brief Transform all the integration points from the integration rule
-       from reference coordinates to physical
-      coordinates and store them as column vectors in the matrix. */
   virtual void Transform(const IntegrationRule &, DenseMatrix &);
-
-   /** @brief Transform all the integration points from the column vectors
-       of @a matrix from reference coordinates to physical
-       coordinates and store them as column vectors in @a result. */
   virtual void Transform(const DenseMatrix &matrix, DenseMatrix &result);

-   /// Return the order of the current element we are using for the transformation.
   virtual int Order() const { return FElem->GetOrder(); }
-
-   /// Return the order of the elements of the Jacobian of the transformation.
   virtual int OrderJ() const;
-
-   /** @brief Return the order of the determinant of the Jacobian (weight)
-       of the transformation. */
   virtual int OrderW() const;
-
-   /// Return the order of \f$ adj(J)^T \nabla fi \f$
   virtual int OrderGrad(const FiniteElement *fe) const;

   virtual int GetSpaceDim() const { return PointMat.Height(); }

-   /** @brief Transform a point @a pt from physical space to a point @a ip in
-       reference space. */
-   /** Attempt to find the IntegrationPoint that is transformed into the given
-       point in physical space. If the inversion fails a non-zero value is
-       returned. This method is not 100 percent reliable for non-linear
-       transformations. */
   virtual int TransformBack(const Vector & v, IntegrationPoint & ip)
   {
      InverseElementTransformation inv_tr(this);
@@ -439,157 +356,15 @@ public:
   void Transform (const IntegrationRule  &, IntegrationRule  &);
 };

-/** @brief A specialized ElementTransformation class representing a face and
-    its two neighboring elements.
-
-    This class can be used as a container for the element transformation data
-    needed for integrating discontinuous fields on element interfaces in a
-    Discontinuous Galerkin (DG) context.
-
-    The secondary purpose of this class is to enable the
-    GridFunction::GetValue function, and various related functions, to properly
-    evaluate fields with limited continuity on boundary elements.
-*/
-class FaceElementTransformations : public IsoparametricTransformation
+class FaceElementTransformations
 {
-private:
-
-   // Bitwise OR of ConfigMasks
-   int mask;
-
-   IntegrationPoint eip1, eip2;
-
-protected: // interface for Mesh to be able to configure this object.
-
-   friend class Mesh;
-#ifdef MFEM_USE_MPI
-   friend class ParMesh;
-#endif
-
-   /// Set the mask indicating which portions of the object have been setup
-   /** The argument @a m is a bitmask used in
-       Mesh::GetFaceElementTransformations to indicate which portions of the
-       FaceElementTransformations object have been configured.
-
-       mask &  1: Elem1 is configured
-       mask &  2: Elem2 is configured
-       mask &  4: Loc1 is configured
-       mask &  8: Loc2 is configured
-       mask & 16: The Face transformation itself is configured
-   */
-   void SetConfigurationMask(int m) { mask = m; }
-
 public:
-
-   enum ConfigMasks
-   {
-      HAVE_ELEM1 =  1, ///< Element on side 1 is configured
-      HAVE_ELEM2 =  2, ///< Element on side 2 is configured
-      HAVE_LOC1  =  4, ///< Point transformation for side 1 is configured
-      HAVE_LOC2  =  8, ///< Point transformation for side 2 is configured
-      HAVE_FACE  = 16  ///< Face transformation is configured
-   };
-
-   int Elem1No, Elem2No;
-   Geometry::Type &FaceGeom; ///< @deprecated Use GetGeometryType instead
-   ElementTransformation *Elem1, *Elem2;
-   ElementTransformation *Face; ///< @deprecated No longer necessary
+   int Elem1No, Elem2No, FaceGeom;
+   ElementTransformation *Elem1, *Elem2, *Face;
   IntegrationPointTransformation Loc1, Loc2;
-
-   FaceElementTransformations() : FaceGeom(geom), Face(this) {}
-
-   /** @brief Method to set the geometry type of the face.
-
-       @note This method is designed to be used when
-       [Par]Mesh::GetFaceTransformation will not be called i.e. when the face
-       transformation will not be needed but the neighboring element
-       transformations will be.  Using this method to override the GeometryType
-       should only be done with great care.
-   */
-   void SetGeometryType(Geometry::Type g) { geom = g; }
-
-   /** @brief Return the mask defining the configuration state.
-
-       The mask value indicates which portions of FaceElementTransformations
-       object have been configured.
-
-       mask &  1: Elem1 is configured
-       mask &  2: Elem2 is configured
-       mask &  4: Loc1 is configured
-       mask &  8: Loc2 is configured
-       mask & 16: The Face transformation itself is configured
-   */
-   int GetConfigurationMask() const { return mask; }
-
-   /** @brief Set the integration point in the Face and the two neighboring
-       elements, if present.
-
-       The point @a face_ip must be in the reference coordinate system of the
-       face.
-   */
-   void SetIntPoint(const IntegrationPoint *face_ip);
-
-   /** @brief Set the integration point in the Face and the two neighboring
-       elements, if present.
-
-       This is a more expressive member function name than SetIntPoint, which
-       in this special case, does the same thing. This function can be used for
-       greater code clarity.
-   */
-   inline void SetAllIntPoints(const IntegrationPoint *face_ip)
-   { FaceElementTransformations::SetIntPoint(face_ip); }
-
-   /** @brief Get a const reference to the integration point in neighboring
-       element 1 corresponding to the currently set integration point on the
-       face.
-
-       This IntegrationPoint object will only contain up-to-date data if
-       SetIntPoint or SetAllIntPoints has been called with the latest
-       integration point for the face and the appropriate point transformation
-       has been configured. */
-   const IntegrationPoint &GetElement1IntPoint() { return eip1; }
-
-   /** @brief Get a const reference to the integration point in neighboring
-       element 2 corresponding to the currently set integration point on the
-       face.
-
-       This IntegrationPoint object will only contain up-to-date data if
-       SetIntPoint or SetAllIntPoints has been called with the latest
-       integration point for the face and the appropriate point transformation
-       has been configured. */
-   const IntegrationPoint &GetElement2IntPoint() { return eip2; }
-
-   virtual void Transform(const IntegrationPoint &, Vector &);
-   virtual void Transform(const IntegrationRule &, DenseMatrix &);
-   virtual void Transform(const DenseMatrix &matrix, DenseMatrix &result);
-
-   ElementTransformation & GetElement1Transformation();
-   ElementTransformation & GetElement2Transformation();
-   IntegrationPointTransformation & GetIntPoint1Transformation();
-   IntegrationPointTransformation & GetIntPoint2Transformation();
-
-   /** @brief Check for self-consistency: compares the result of mapping the
-       reference face vertices to physical coordinates using the three
-       transformations: face, element 1, and element 2.
-
-       @param[in] print_level  If set to a positive number, print the physical
-                               coordinates of the face vertices computed through
-                               all available transformations: face, element 1,
-                               and/or element 2.
-       @param[in,out] out      The output stream to use for printing.
-
-       @returns A maximal distance between physical coordinates of face vertices
-                that should coincide. A successful check should return a small
-                number relative to the mesh extents. If less than 2 of the three
-                transformations are set, returns 0.
-
-       @warning This check will generally fail on periodic boundary faces.
-   */
-   double CheckConsistency(int print_level = 0,
-                           std::ostream &out = mfem::out);
 };

-/**                Elem1(Loc1(x)) = Face(x) = Elem2(Loc2(x))
+/*                 Elem1(Loc1(x)) = Face(x) = Elem2(Loc2(x))


                                Physical Space
@@ -50,21 +50,4 @@ void L2ZienkiewiczZhuEstimator::ComputeEstimates()

 #endif // MFEM_USE_MPI

-void LpErrorEstimator::ComputeEstimates()
-{
-   MFEM_VERIFY(coef != NULL || vcoef != NULL,
-               "LpErrorEstimator has no coefficient!  Call SetCoef first.");
-
-   error_estimates.SetSize(sol->FESpace()->GetMesh()->GetNE());
-   if (coef)
-   {
-      sol->ComputeElementLpErrors(local_norm_p, *coef, error_estimates);
-   }
-   else
-   {
-      sol->ComputeElementLpErrors(local_norm_p, *vcoef, error_estimates);
-   }
-   current_sequence = sol->FESpace()->GetMesh()->GetSequence();
-}
-
 } // namespace mfem
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
Tucker Babcock	4adfa0fc86	updting io benchmark	2020-05-05 15:44:44 -04:00
Tucker Babcock	857a24f6c4	Merge branch 'PCFinalProject' of github.com:mfem/mfem into PCFinalProject	2020-05-04 11:14:22 -07:00
Tucker Babcock	1ae22c7c69	adding io benchmark	2020-05-04 11:13:42 -07:00
Tucker Babcock	161ebff2a1	Merge branch 'PCFinalProject' of https://github.com/mfem/mfem into PCFinalProject	2020-05-04 12:42:18 -04:00
Tucker Babcock	3548f2cb83	adding num ranks printing	2020-05-04 12:42:13 -04:00
Tucker Babcock	41a7730048	adding barriers ahead of timings and averaging timing over all ranks	2020-05-04 09:40:57 -07:00
Tucker Babcock	b638fb8960	adding all of the operator testing to one file	2020-05-03 22:02:26 -07:00
Tucker Babcock	dc80f42710	Merge branch 'PCFinalProject' of https://github.com/mfem/mfem into PCFinalProject	2020-05-04 00:51:51 -04:00
Tucker Babcock	4414a3fc01	adding test to mfem examples	2020-05-04 00:50:16 -04:00
Tucker Babcock	2f683f80fa	Merge branch 'mpiio-gf-dev' into PCFinalProject	2020-04-30 14:09:33 -07:00
Tucker Babcock	6ea2f7bf55	Merge branch 'mpiio-gf-dev' of github.com:mfem/mfem into mpiio-gf-dev	2020-04-30 14:06:16 -07:00
Tucker Babcock	581cafa7a7	updating documentation	2020-04-30 14:06:10 -07:00
Tucker Babcock	c5bab73f9a	Merge branch 'mpiio-gf-dev' into PCFinalProject	2020-04-30 15:45:12 -04:00
Tucker Babcock	b02eb71967	adding number of files printing control to example 1	2020-04-30 15:39:19 -04:00
Tucker Babcock	d236571e4a	cleaned up code in pgridfunc and added printing to example two.	2020-04-28 21:04:11 -07:00
Tucker Babcock	8d444d7f92	ordering by nodes appears to work now as well	2020-04-28 16:28:07 -07:00
Tucker Babcock	71937096f8	ordering by vdim works with high order	2020-04-28 16:26:30 -07:00
Tucker Babcock	4e0978cf3b	can save and load files correctly for p = 1, errors otherwise.	2020-04-28 15:10:12 -07:00
Tucker Babcock	e52fdd205a	initial commit adding MPI-IO writing of GridFunction supporting writing to arbitrary number of files. Reading support to come	2020-04-27 22:41:28 -07:00