Merge remote-tracking branch 'origin/master' into entity-sets-dev

Adding Piecewise coefficient classes [pw-coefs-dev]

.github/workflows/mfem-analysis.yml

@@ -82,9 +82,9 @@ jobs:
       uses: mfem/github-actions/build-mfem@v2.0
       with:
         os: ${{ runner.os }}
-        target: optim
+        target: opt
         codecov: NO
-        mpi: parallel
+        mpi: par
         build-system: make
         hypre-dir: ${{ env.HYPRE_TOP_DIR }}
         metis-dir: ${{ env.METIS_TOP_DIR }}

.gitignore

@@ -51,6 +51,8 @@ examples/ex1[04-9]
 examples/ex1[0-9]p
 examples/ex2[0-9]
 examples/ex2[0-9]p
+examples/ex30
+examples/ex30p
 
 examples/refined.mesh
 examples/displaced.mesh
@@ -223,6 +225,14 @@ miniapps/mtop/ParHeat*
 miniapps/mtop/seqheat
 miniapps/mtop/SeqHeat*
 
+miniapps/autodiff/paradiff
+miniapps/autodiff/seqadiff
+miniapps/autodiff/seqtest
+miniapps/autodiff/par_example
+miniapps/autodiff/seq_example
+miniapps/autodiff/seq_test
+miniapps/autodiff/Exampl*
+
 miniapps/navier/navier_mms
 miniapps/navier/navier_kovasznay
 miniapps/navier/navier_kovasznay_vs

.gitlab-ci.yml

@@ -29,12 +29,34 @@ stages:
 
 variables:
   CUSTOM_CI_BUILDS_DIR: "/usr/workspace/mfem/gitlab-runner"
+  USER_CI_TOP_DIR: "${CUSTOM_CI_BUILDS_DIR}/${GITLAB_USER_LOGIN}"
+  SHARED_REPOS_DIR: "${USER_CI_TOP_DIR}/repos"
+  AUTOTEST_ROOT: "${SHARED_REPOS_DIR}"
+  # MFEM_DATA_DIR is setup in '.gitlab/configs/setup-build-and-test.yml' and
+  # used in '.gitlab/configs/<machine>-config.yml':
+  MFEM_DATA_DIR: "${SHARED_REPOS_DIR}/mfem-data"
+
+# Defines the default choice for updating the saved baseline results. By default
+# the baseline can only be updated from the master branch. This variable offers
+# the option to manually ask for rebaselining from another branch if necessary.
+  REBASELINE: "NO"
+  AUTOTEST: "NO"
+  # AUTOTEST_COMMIT: used only when AUTOTEST is set to YES.
+  # * If AUTOTEST_COMMIT is NOT set to NO, reporting jobs will commit their
+  #   files to the MFEM/autotest repo.
+  # * If AUTOTEST_COMMIT is set to NO, reporting jobs will NOT commit their
+  #   files to the MFEM/autotest repo. Instead they will just show the contents
+  #   of the report files and remove them.
+  AUTOTEST_COMMIT: "YES"
 
 # Trigger subpipelines:
 quartz-build-and-test:
   stage: sub-pipelines
   variables:
-    _AUTOTEST: $AUTOTEST
+    # Explicitly pass down values that we want to be able to set when triggering
+    # pipelines manually or using scheduling
+    AUTOTEST: "${AUTOTEST}"
+    AUTOTEST_COMMIT: "${AUTOTEST_COMMIT}"
   trigger:
     include: .gitlab/quartz-build-and-test.yml
     strategy: depend
@@ -42,7 +64,11 @@ quartz-build-and-test:
 quartz-baseline:
   stage: sub-pipelines
   variables:
-    _AUTOTEST: $AUTOTEST
+    # Explicitly pass down values that we want to be able to set when triggering
+    # pipelines manually or using scheduling
+    REBASELINE: "${REBASELINE}"
+    AUTOTEST: "${AUTOTEST}"
+    AUTOTEST_COMMIT: "${AUTOTEST_COMMIT}"
   trigger:
     include: .gitlab/quartz-baseline.yml
     strategy: depend
@@ -50,7 +76,10 @@ quartz-baseline:
 lassen-build-and-test:
   stage: sub-pipelines
   variables:
-    _AUTOTEST: $AUTOTEST
+    # Explicitly pass down values that we want to be able to set when triggering
+    # pipelines manually or using scheduling
+    AUTOTEST: "${AUTOTEST}"
+    AUTOTEST_COMMIT: "${AUTOTEST_COMMIT}"
   trigger:
     include: .gitlab/lassen-build-and-test.yml
     strategy: depend
@@ -58,7 +87,10 @@ lassen-build-and-test:
 corona-build-and-test:
   stage: sub-pipelines
   variables:
-    _AUTOTEST: $AUTOTEST
+    # Explicitly pass down values that we want to be able to set when triggering
+    # pipelines manually or using scheduling
+    AUTOTEST: "${AUTOTEST}"
+    AUTOTEST_COMMIT: "${AUTOTEST_COMMIT}"
   trigger:
     include: .gitlab/corona-build-and-test.yml
     strategy: depend

.gitlab/configs/common.yml

@@ -18,19 +18,13 @@ variables:
 # the pipeline, preventing any form of concurrency with other pipelines. This
 # also means that the BUILD_ROOT directory will never be cleaned.
 # TODO: add a clean-up mechanism
-  BUILD_ROOT: ${CI_BUILDS_DIR}/MFEM_${MACHINE_NAME}/${CI_PROJECT_NAME}_${CI_COMMIT_REF_SLUG}_${CI_PIPELINE_ID}
+  BUILD_ROOT: ${USER_CI_TOP_DIR}/${CI_PROJECT_NAME}-${MACHINE_NAME}-pipeline-${CI_PIPELINE_ID}
 
 # On LLNL's quartz, there is only one allocation shared among jobs in order to
 # save time and resource. This allocation has to be uniquely named so that we
 # are sure to retrieve it.
   ALLOC_NAME: ${CI_PROJECT_NAME}_ci_${CI_PIPELINE_ID}
 
-# Defines the default choice for updating the saved baseline results. By default
-# the baseline can only be updated from the master branch. This variable offers
-# the option to manually ask for rebaselining from another branch if necessary.
-  _REBASELINE: "NO"
-  _AUTOTEST: "NO"
-
 # Git repositories used in the pipeline
   TPLS_REPO: ssh://git@mybitbucket.llnl.gov:7999/mfem/tpls.git
   TESTS_REPO: ssh://git@mybitbucket.llnl.gov:7999/mfem/tests.git
@@ -40,5 +34,3 @@ variables:
 # Directory used to place artifacts.
   ARTIFACTS_DIR: artifacts
   SLURM_OVERLAP: 1
-
-

.gitlab/configs/corona-config.yml

@@ -26,17 +26,20 @@ variables:
     - if: '$CI_COMMIT_BRANCH =~ /_cnone/ || $ON_CORONA != "ON"'
       when: never
     # Don’t run autotest update if...
-    - if: '$CI_JOB_NAME =~ /report/ && $_AUTOTEST != "YES"'
+    - if: '$CI_JOB_NAME =~ /report/ && $AUTOTEST != "YES"'
       when: never
     # Report success on success status
-    - if: '$CI_JOB_NAME =~ /report_job_success/ && $_AUTOTEST == "YES"'
+    - if: '$CI_JOB_NAME =~ /report_job_success/ && $AUTOTEST == "YES"'
       when: on_success
     # Report failure on failure status
-    - if: '$CI_JOB_NAME =~ /report_job_failure/ && $_AUTOTEST == "YES"'
+    - if: '$CI_JOB_NAME =~ /report_job_failure/ && $AUTOTEST == "YES"'
       when: on_failure
     # Always release resource
     - if: '$CI_JOB_NAME =~ /release_resource/'
       when: always
+    # Always cleanup
+    - if: '$CI_JOB_NAME =~ /cleanup/'
+      when: always
     # Default is to run if previous stage succeeded
     - when: on_success
 
@@ -46,9 +49,11 @@ variables:
   extends: [.on_corona]
   stage: build_and_test
   script:
+    # THREADS is used by 'tests/gitlab/build_and_test', run below
     - export THREADS=12
     - echo ${ALLOC_NAME}
     - export JOBID=$(squeue -h --name=${ALLOC_NAME} --format=%A)
     - echo ${JOBID}
-    - srun $( [[ -n "${JOBID}" ]] && echo "--jobid=${JOBID}" ) -t 15 -N 1 tests/gitlab/build_and_test --spec "${SPEC}" --build-root "${BUILD_ROOT}" --data
-
+    - echo ${MFEM_DATA_DIR}
+    - echo ${SPEC}
+    - srun $( [[ -n "${JOBID}" ]] && echo "--jobid=${JOBID}" ) -t 15 -N 1 tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data

.gitlab/configs/lassen-config.yml

@@ -21,14 +21,17 @@ variables:
     - if: '$CI_COMMIT_BRANCH =~ /_lnone/ || $ON_LASSEN == "OFF"' #run except if ...
       when: never
     # Don't run autotest update if...
-    - if: '$CI_JOB_NAME =~ /report/ && $_AUTOTEST != "YES"'
+    - if: '$CI_JOB_NAME =~ /report/ && $AUTOTEST != "YES"'
       when: never
     # Report success on success status
-    - if: '$CI_JOB_NAME =~ /report_job_success/ && $_AUTOTEST == "YES"'
+    - if: '$CI_JOB_NAME =~ /report_job_success/ && $AUTOTEST == "YES"'
       when: on_success
     # Report failure on failure status
-    - if: '$CI_JOB_NAME =~ /report_job_failure/ && $_AUTOTEST == "YES"'
+    - if: '$CI_JOB_NAME =~ /report_job_failure/ && $AUTOTEST == "YES"'
       when: on_failure
+    # Always cleanup
+    - if: '$CI_JOB_NAME =~ /cleanup/'
+      when: always
     - when: on_success
 
 # Lassen uses a different job scheduler (spectrum lsf) that does not allow
@@ -39,5 +42,8 @@ variables:
   extends: [.on_lassen]
   stage: build_and_test
   script:
-    - lalloc 1 -W 30 -q pdebug tests/gitlab/build_and_test --spec "${SPEC}" --build-root "${BUILD_ROOT}" --data
+    - echo ${MFEM_DATA_DIR}
+    - echo ${SPEC}
+    # Next script uses 'THREADS': leaving it empty --> it uses 'make all -j'
+    - lalloc 1 -W 30 -q pdebug tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data
   needs: [setup]

.gitlab/configs/quartz-config.yml

@@ -22,17 +22,20 @@ variables:
     - if: '$CI_COMMIT_BRANCH =~ /_qnone/ || $ON_QUARTZ == "OFF"'
       when: never
     # Don't run autotest update if...
-    - if: '$CI_JOB_NAME =~ /report/ && $_AUTOTEST != "YES"'
+    - if: '$CI_JOB_NAME =~ /report/ && $AUTOTEST != "YES"'
       when: never
     # Report success on success status
-    - if: '$CI_JOB_NAME =~ /report_job_success/ && $_AUTOTEST == "YES"'
+    - if: '$CI_JOB_NAME =~ /report_job_success/ && $AUTOTEST == "YES"'
       when: on_success
     # Report failure on failure status
-    - if: '$CI_JOB_NAME =~ /report_job_failure/ && $_AUTOTEST == "YES"'
+    - if: '$CI_JOB_NAME =~ /report_job_failure/ && $AUTOTEST == "YES"'
       when: on_failure
     # Always release resource
     - if: '$CI_JOB_NAME =~ /release_resource/'
       when: always
+    # Always cleanup
+    - if: '$CI_JOB_NAME =~ /cleanup/'
+      when: always
     # Default is to run if previous stage succeeded
     - when: on_success
 
@@ -42,9 +45,11 @@ variables:
   extends: [.on_quartz]
   stage: build_and_test
   script:
+    # THREADS is used by 'tests/gitlab/build_and_test', run below
     - export THREADS=12
     - echo ${ALLOC_NAME}
     - export JOBID=$(squeue -h --name=${ALLOC_NAME} --format=%A)
     - echo ${JOBID}
-    - srun $( [[ -n "${JOBID}" ]] && echo "--jobid=${JOBID}" ) -t 30 -N 1 tests/gitlab/build_and_test --spec "${SPEC}" --build-root "${BUILD_ROOT}" --data
-
+    - echo ${MFEM_DATA_DIR}
+    - echo ${SPEC}
+    - srun $( [[ -n "${JOBID}" ]] && echo "--jobid=${JOBID}" ) -t 30 -N 1 tests/gitlab/build_and_test --spec "${SPEC}" --data-dir "${MFEM_DATA_DIR}" --data

.gitlab/configs/report-build-and-test.yml

@@ -0,0 +1,81 @@
+# Copyright (c) 2010-2021, 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.
+
+# Jobs report
+.report_job_success:
+  script:
+    - echo ${MACHINE_NAME}
+    - echo ${AUTOTEST}
+    - echo ${AUTOTEST_COMMIT}
+    - echo "AUTOTEST_ROOT ${AUTOTEST_ROOT}"
+    - cd ${AUTOTEST_ROOT}
+    - |
+      (
+        date
+        echo "Waiting to aquire lock on '$PWD/autotest.lock' ..."
+        # try to get an excusive lock on fd 9 (autotest.lock) repeating the try
+        # every 5 seconds; simply using no timeout, i.e. 'flock 9', causes the
+        # command to hang indefinitely sometimes, so we use the timeout & retry
+        # as a workaround; we may want to add a counter for the number of
+        # retries to interrupt a potential infinite loop
+        while ! flock -w 5 9; do
+          true
+        done
+        echo "Aquired lock on '$PWD/autotest.lock'"
+        date
+        # Report SUCCESS while holding the file lock on 'autotest.lock'.
+        # The next script uses the following environment variables:
+        # - MACHINE_NAME, AUTOTEST_ROOT, AUTOTEST_COMMIT
+        # - CI_COMMIT_REF_SLUG, CI_PROJECT_DIR, CI_PIPELINE_URL
+        # It also calls the script '.gitlab/scripts/safe_create_rundir'.
+        ${CI_PROJECT_DIR}/.gitlab/scripts/report_build_and_test_success
+        err=$?
+        # sleep for a period to allow NFS to propagate the above changes;
+        # clearly, there is no guarantee that other NFS clients will see the
+        # changes even after the timeout
+        sleep 10
+        exit $err
+      ) 9> autotest.lock
+
+.report_job_failure:
+  script:
+    - echo ${MACHINE_NAME}
+    - echo ${AUTOTEST}
+    - echo ${AUTOTEST_COMMIT}
+    - echo "AUTOTEST_ROOT ${AUTOTEST_ROOT}"
+    - cd ${AUTOTEST_ROOT}
+    - |
+      (
+        date
+        echo "Waiting to aquire lock on '$PWD/autotest.lock' ..."
+        # try to get an excusive lock on fd 9 (autotest.lock) repeating the try
+        # every 5 seconds; simply using no timeout, i.e. 'flock 9', causes the
+        # command to hang indefinitely sometimes, so we use the timeout & retry
+        # as a workaround; we may want to add a counter for the number of
+        # retries to interrupt a potential infinite loop
+        while ! flock -w 5 9; do
+          true
+        done
+        echo "Aquired lock on '$PWD/autotest.lock'"
+        date
+        # Report FAILURE while holding the file lock on 'autotest.lock'.
+        # The next script uses the following environment variables:
+        # - MACHINE_NAME, AUTOTEST_ROOT, AUTOTEST_COMMIT
+        # - CI_COMMIT_REF_SLUG, CI_PROJECT_DIR, CI_PIPELINE_URL
+        # It also calls the script '.gitlab/scripts/safe_create_rundir'.
+        ${CI_PROJECT_DIR}/.gitlab/scripts/report_build_and_test_failure
+        err=$?
+        # sleep for a period to allow NFS to propagate the above changes;
+        # clearly, there is no guarantee that other NFS clients will see the
+        # changes even after the timeout
+        sleep 10
+        exit $err
+      ) 9> autotest.lock

.gitlab/configs/setup-baseline.yml

@@ -9,13 +9,6 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.
 
-# TPLS_DIR is used in .gitlab/scripts/baseline to provide the tpls location
-# when call the runtest script in MFEM test repo.
-# Note: the value must be consistent with what setup_baseline does.
-variables:
-  TPLS_DIR: ${BUILD_ROOT}/tpls
-  AUTOTEST_ROOT: ${CI_BUILDS_DIR}/MFEM_${MACHINE_NAME}_baseline
-
 # The setup_baseline job doesn't rely on MFEM git repo. It prepares a
 # pipeline-wide working directory downloading/updating external repos.
 # TODO:
@@ -30,13 +23,50 @@ setup_baseline:
   variables:
     GIT_STRATEGY: none
   script:
+    #
+    # Setup ${BUILD_ROOT}/tpls and ${BUILD_ROOT}/tests:
+    #
+    - echo "MACHINE_NAME = ${MACHINE_NAME}"
+    - echo "REBASELINE = ${REBASELINE}"
+    - echo "AUTOTEST = ${AUTOTEST}"
+    - echo "AUTOTEST_COMMIT = ${AUTOTEST_COMMIT}"
     - echo "BUILD_ROOT ${BUILD_ROOT}"
     - mkdir -p ${BUILD_ROOT} && cd ${BUILD_ROOT}
     - if [ ! -d "tpls" ]; then git clone ${TPLS_REPO}; fi
     - if [ ! -d "tests" ]; then git clone ${TESTS_REPO}; fi
     - cd tpls && git pull && cd ..
     - cd tests && git pull origin && cd ..
+    #
+    # Setup ${AUTOTEST_ROOT}/autotest:
+    #
     - echo "AUTOTEST_ROOT ${AUTOTEST_ROOT}"
     - mkdir -p ${AUTOTEST_ROOT} && cd ${AUTOTEST_ROOT}
-    - if [ ! -d "autotest" ]; then git clone ${AUTOTEST_REPO}; fi
-    - cd autotest && git pull && cd ..
+    - command -v flock || echo "Required command 'flock' not found"
+    - |
+      (
+        date
+        echo "Waiting to aquire lock on '$PWD/autotest.lock' ..."
+        # try to get an excusive lock on fd 9 (autotest.lock) repeating the try
+        # every 5 seconds; simply using no timeout, i.e. 'flock 9', causes the
+        # command to hang indefinitely sometimes, so we use the timeout & retry
+        # as a workaround; we may want to add a counter for the number of
+        # retries to interrupt a potential infinite loop
+        while ! flock -w 5 9; do
+          true
+        done
+        echo "Aquired lock on '$PWD/autotest.lock'"
+        date
+        # clone/update the autotest repo while holding the file lock on
+        # 'autotest.lock'
+        err=0
+        if [[ ! -d "autotest" ]]; then
+          git clone ${AUTOTEST_REPO}
+        else
+          cd autotest && git pull && cd ..
+        fi || err=1
+        # sleep for a period to allow NFS to propagate the above changes;
+        # clearly, there is no guarantee that other NFS clients will see the
+        # changes even after the timeout
+        sleep 10
+        exit $err
+      ) 9> autotest.lock

.gitlab/configs/setup-build-and-test.yml

@@ -9,13 +9,10 @@
 # terms of the BSD-3 license. We welcome feedback and contributions, see file
 # CONTRIBUTING.md for details.
 
-variables:
-  AUTOTEST_ROOT: ${CI_BUILDS_DIR}/MFEM_${MACHINE_NAME}_build_and_test
-
-# setup clones the mfem/data repo in ${BUILD_ROOT}. The build_and_test script
-# then symlinks the repo to the parent directory of the MFEM source directory.
-# Unit tests that depend on the mfem/data repo will then detect that this
-# directory is present and be enabled.
+# Setup clones the mfem/data repo in ${SHARED_REPOS_DIR}. The build_and_test
+# script then symlinks the repo to the parent directory of the MFEM source
+# directory. Unit tests that depend on the mfem/data repo will then detect that
+# this directory is present and be enabled.
 setup:
   tags:
     - shell
@@ -24,11 +21,74 @@ setup:
   variables:
     GIT_STRATEGY: none
   script:
-    - echo "BUILD_ROOT ${BUILD_ROOT}"
-    - mkdir -p ${BUILD_ROOT} && cd ${BUILD_ROOT}
-    - if [ ! -d data ]; then git clone ${MFEM_DATA_REPO}; fi
+    #
+    # Setup MFEM_DATA_DIR=${SHARED_REPOS_DIR}/mfem-data, see '.gitlab-ci.yml'
+    # and '.gitlab/configs/<machine>-config.yml'
+    #
+    - echo "MACHINE_NAME = ${MACHINE_NAME}"
+    - echo "AUTOTEST = ${AUTOTEST}"
+    - echo "AUTOTEST_COMMIT = ${AUTOTEST_COMMIT}"
+    - echo "SHARED_REPOS_DIR ${SHARED_REPOS_DIR}"
+    - mkdir -p ${SHARED_REPOS_DIR} && cd ${SHARED_REPOS_DIR}
+    - command -v flock || echo "Required command 'flock' not found"
+    - |
+      (
+        date
+        echo "Waiting to aquire lock on '$PWD/mfem-data.lock' ..."
+        # try to get an excusive lock on fd 9 (mfem-data.lock) repeating the try
+        # every 5 seconds; simply using no timeout, i.e. 'flock 9', causes the
+        # command to hang indefinitely sometimes, so we use the timeout & retry
+        # as a workaround; we may want to add a counter for the number of
+        # retries to interrupt a potential infinite loop
+        while ! flock -w 5 9; do
+          true
+        done
+        echo "Aquired lock on '$PWD/mfem-data.lock'"
+        date
+        # clone/update the mfem/data repo while holding the file lock on
+        # 'mfem-data.lock'
+        err=0
+        if [[ ! -d "mfem-data" ]]; then
+          git clone ${MFEM_DATA_REPO} "mfem-data"
+        else
+          cd "mfem-data" && git pull && cd ..
+        fi || err=1
+        # sleep for a period to allow NFS to propagate the above changes;
+        # clearly, there is no guarantee that other NFS clients will see the
+        # changes even after the timeout
+        sleep 10
+        exit $err
+      ) 9> mfem-data.lock
+    #
+    # Setup ${AUTOTEST_ROOT}/autotest:
+    #
     - echo "AUTOTEST_ROOT ${AUTOTEST_ROOT}"
     - mkdir -p ${AUTOTEST_ROOT} && cd ${AUTOTEST_ROOT}
-    - if [ ! -d "autotest" ]; then git clone ${AUTOTEST_REPO}; fi
-    - cd autotest && git pull && cd ..
-
+    - |
+      (
+        date
+        echo "Waiting to aquire lock on '$PWD/autotest.lock' ..."
+        # try to get an excusive lock on fd 9 (autotest.lock) repeating the try
+        # every 5 seconds; simply using no timeout, i.e. 'flock 9', causes the
+        # command to hang indefinitely sometimes, so we use the timeout & retry
+        # as a workaround; we may want to add a counter for the number of
+        # retries to interrupt a potential infinite loop
+        while ! flock -w 5 9; do
+          true
+        done
+        echo "Aquired lock on '$PWD/autotest.lock'"
+        date
+        # clone/update the autotest repo while holding the file lock on
+        # 'autotest.lock'
+        err=0
+        if [[ ! -d "autotest" ]]; then
+          git clone ${AUTOTEST_REPO}
+        else
+          cd autotest && git pull && cd ..
+        fi || err=1
+        # sleep for a period to allow NFS to propagate the above changes;
+        # clearly, there is no guarantee that other NFS clients will see the
+        # changes even after the timeout
+        sleep 10
+        exit $err
+      ) 9> autotest.lock

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

@@ -22,6 +22,7 @@ allocate_resource:
   extends: .on_corona
   stage: allocate_resource
   script:
+    - echo ${ALLOC_NAME}
     - salloc --exclusive --nodes=1 --partition=mi60 --time=30 --no-shell --job-name=${ALLOC_NAME}
   timeout: 6h
   needs: [setup]
@@ -40,24 +41,27 @@ release_resource:
   extends: .on_corona
   stage: release_resource_and_report
   script:
+    - echo ${ALLOC_NAME}
     - export JOBID=$(squeue -h --name=${ALLOC_NAME} --format=%A)
+    - echo ${JOBID}
     - ([[ -n "${JOBID}" ]] && scancel ${JOBID})
   needs: [rocm_gcc_8.3.1]
 
 # Jobs report
 report_job_success:
-  extends: .on_corona
   stage: release_resource_and_report
-  script:
-    - .gitlab/scripts/report_build_and_test_success
+  extends:
+    - .on_corona
+    - .report_job_success
 
 report_job_failure:
-  extends: .on_corona
   stage: release_resource_and_report
-  script:
-    - .gitlab/scripts/report_build_and_test_failure
+  extends:
+    - .on_corona
+    - .report_job_failure
 
 include:
   - local: .gitlab/configs/common.yml
   - local: .gitlab/configs/corona-config.yml
   - local: .gitlab/configs/setup-build-and-test.yml
+  - local: .gitlab/configs/report-build-and-test.yml

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

@@ -21,18 +21,19 @@ opt_mpi_cuda_xl_16_1_1_8:
 
 # Jobs report
 report_job_success:
-  extends: .on_lassen
   stage: report
-  script:
-    - .gitlab/scripts/report_build_and_test_success
+  extends:
+    - .on_lassen
+    - .report_job_success
 
 report_job_failure:
-  extends: .on_lassen
   stage: report
-  script:
-    - .gitlab/scripts/report_build_and_test_failure
+  extends:
+    - .on_lassen
+    - .report_job_failure
 
 include:
   - local: .gitlab/configs/common.yml
   - local: .gitlab/configs/lassen-config.yml
   - local: .gitlab/configs/setup-build-and-test.yml
+  - local: .gitlab/configs/report-build-and-test.yml

.gitlab/quartz-baseline.yml

@@ -16,12 +16,26 @@ stages:
   - setup
   - baseline_check
   - baseline_report
+  - cleanup
   - baseline_publish
 
 baselinecheck_mfem_intel_quartz:
   extends: [.on_quartz]
   stage: baseline_check
+  variables:
+    # TPLS_DIR is used in .gitlab/scripts/baseline to provide the tpls location
+    # when call the runtest script in MFEM test repo.
+    # Note: the value must be consistent with the setup performed in
+    #       .gitlab/configs/setup-baseline.yml.
+    TPLS_DIR: ${BUILD_ROOT}/tpls
   script:
+    - echo ${BUILD_ROOT}
+    - echo ${TPLS_DIR}
+    # Used by the tests in MFEM/tests:
+    - export MFEM_TEST_NP=32
+    # The next script uses the following environment variables:
+    # * BASELINE_TEST, SYS_TYPE, CI_PROJECT_DIR, ARTIFACTS_DIR,
+    # * BUILD_ROOT, TPLS_DIR, MACHINE_NAME
     - .gitlab/scripts/baseline
   artifacts:
     when: always
@@ -29,33 +43,88 @@ baselinecheck_mfem_intel_quartz:
       - ${ARTIFACTS_DIR}
   allow_failure: true
 
+cleanup:
+  extends: .on_quartz
+  stage: cleanup
+  variables:
+    GIT_STRATEGY: none
+  script:
+    - echo "BUILD_ROOT=${BUILD_ROOT}"
+    - rm -rf "${BUILD_ROOT}" || true
+
 report_baseline:
   extends: [.on_quartz]
   stage: baseline_report
   script:
-    - cd ${AUTOTEST_ROOT}/autotest && git pull
-    - mkdir -p ${MACHINE_NAME}
-    - rundir="${MACHINE_NAME}/$(date +%Y-%m-%d)-gitlab-${BASELINE_TEST}-${CI_COMMIT_REF_SLUG}"
-    - rundir=$(${CI_PROJECT_DIR}/.gitlab/scripts/safe_create_rundir ${rundir})
-    - cp ${CI_PROJECT_DIR}/${ARTIFACTS_DIR}/* ${rundir}
-    # We create an autotest-email.html file, because that's how we signal that there was a diff (temporary).
+    - echo ${MACHINE_NAME}
+    - echo ${AUTOTEST}
+    - echo ${AUTOTEST_COMMIT}
+    - echo "AUTOTEST_ROOT ${AUTOTEST_ROOT}"
+    - cd ${AUTOTEST_ROOT}
     - |
-      if [[ -f ${rundir}/*.err ]]
-      then
-        echo "See the pipeline here -> $CI_PIPELINE_URL" >> ${rundir}/*.err
-        cp ${rundir}/*.err ${rundir}/autotest-email.html
-      fi
-    - git add ${rundir}
-    - git commit -am "GitLab CI log for ${BASELINE_TEST} on ${MACHINE_NAME} with intel ($(date +%Y-%m-%d))"
-    - git push origin master
+      (
+        date
+        echo "Waiting to aquire lock on '$PWD/autotest.lock' ..."
+        # try to get an excusive lock on fd 9 (autotest.lock) repeating the try
+        # every 5 seconds; simply using no timeout, i.e. 'flock 9', causes the
+        # command to hang indefinitely sometimes, so we use the timeout & retry
+        # as a workaround; we may want to add a counter for the number of
+        # retries to interrupt a potential infinite loop
+        while ! flock -w 5 9; do
+          true
+        done
+        echo "Aquired lock on '$PWD/autotest.lock'"
+        date
+        # ----------------------
+        cd ${AUTOTEST_ROOT}/autotest || \
+          { echo "Invalid 'autotest' dir: ${AUTOTEST_ROOT}/autotest"; exit 1; }
+        mkdir -p ${MACHINE_NAME}
+        rundir="${MACHINE_NAME}/$(date +%Y-%m-%d)-gitlab-${BASELINE_TEST}-${CI_COMMIT_REF_SLUG}"
+        rundir=$(${CI_PROJECT_DIR}/.gitlab/scripts/safe_create_rundir ${rundir})
+        cp ${CI_PROJECT_DIR}/${ARTIFACTS_DIR}/* ${rundir}
+        # We create an autotest-email.html file, because that's how we signal that there was a diff (temporary).
+        if [[ -f ${rundir}/${BASELINE_TEST}.err ]]; then
+          cp ${rundir}/${BASELINE_TEST}.err ${rundir}/autotest-email.html
+        fi
+        printf "%s\n" "" "Pipeline URL:" "$CI_PIPELINE_URL" \
+          >> ${rundir}/pipeline.txt
+        msg="GitLab CI log for ${BASELINE_TEST} on ${MACHINE_NAME} ($(date +%Y-%m-%d))"
+        if [[ "$AUTOTEST_COMMIT" != "NO" ]]; then
+          git pull && \
+          git add ${rundir} && \
+          git commit -m "${msg}" && \
+          git push origin master
+        else
+          for file in ${rundir}/*; do
+            echo "------------------------------"
+            echo "Content of '$file'"
+            echo "******************************"
+            cat $file
+            echo "******************************"
+          done
+          rm -rf ${rundir} || true
+        fi
+        err=$?
+        # ----------------------
+        # sleep for a period to allow NFS to propagate the above changes;
+        # clearly, there is no guarantee that other NFS clients will see the
+        # changes even after the timeout
+        sleep 10
+        exit $err
+      ) 9> autotest.lock
 
 baselinepublish_mfem_quartz:
   extends: [.on_quartz]
   stage: baseline_publish
   rules:
-    - if: '$CI_COMMIT_BRANCH == "master" || $REBASELINE == "YES"'
+    # - if: '$CI_COMMIT_BRANCH == "master" || $REBASELINE == "YES"'
+    - if: '$REBASELINE == "YES"'
       when: manual
   script:
+    - echo ${BUILD_ROOT}
+    - echo ${PWD}
+    - echo ${ARTIFACTS_DIR}
+    - ls -lA ${ARTIFACTS_DIR}
     - .gitlab/scripts/rebaseline
 
 include:

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

@@ -22,6 +22,7 @@ allocate_resource:
   extends: .on_quartz
   stage: allocate_resource
   script:
+    - echo ${ALLOC_NAME}
     - salloc --exclusive --nodes=1 --partition=pdebug --time=30 --no-shell --job-name=${ALLOC_NAME}
   timeout: 6h
 
@@ -73,23 +74,26 @@ release_resource:
   extends: .on_quartz
   stage: release_resource_and_report
   script:
+    - echo ${ALLOC_NAME}
     - export JOBID=$(squeue -h --name=${ALLOC_NAME} --format=%A)
+    - echo ${JOBID}
     - ([[ -n "${JOBID}" ]] && scancel ${JOBID})
 
 # Jobs report
 report_job_success:
-  extends: .on_quartz
   stage: release_resource_and_report
-  script:
-    - .gitlab/scripts/report_build_and_test_success
+  extends:
+    - .on_quartz
+    - .report_job_success
 
 report_job_failure:
-  extends: .on_quartz
   stage: release_resource_and_report
-  script:
-    - .gitlab/scripts/report_build_and_test_failure
+  extends:
+    - .on_quartz
+    - .report_job_failure
 
 include:
   - local: .gitlab/configs/common.yml
   - local: .gitlab/configs/quartz-config.yml
   - local: .gitlab/configs/setup-build-and-test.yml
+  - local: .gitlab/configs/report-build-and-test.yml

.gitlab/scripts/baseline

@@ -20,7 +20,8 @@ base_out=${base}.out
 artifacts_path=${CI_PROJECT_DIR}/${ARTIFACTS_DIR}
 
 # prepare
-cd ${BUILD_ROOT}
+cd ${BUILD_ROOT} || \
+  { echo "Invalid BUILD_ROOT=$BUILD_ROOT"; exit 1; }
 ln -snf ${CI_PROJECT_DIR} mfem
 cd tests
 [[ -d _${BASELINE_TEST} ]] && rm -rf _${BASELINE_TEST}
@@ -33,6 +34,9 @@ elif [[ ${MACHINE_NAME} == "corona" ]]; then
   srun --nodes=1 -t 60 -p mi60 ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 elif [[ ${MACHINE_NAME} == "lassen" ]]; then
   lalloc 1 -q pdebug ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+else
+  echo "Unknown machine: MACHINE_NAME=$MACHINE_NAME"
+  exit 1
 fi
 
 # post
@@ -60,6 +64,10 @@ then
   cp ${base_out} ${artifacts_path}/${base_out}
 fi
 
+if [[ -f ${BASELINE_TEST}.out ]]; then
+  cp ${BASELINE_TEST}.out ${artifacts_path}
+fi
+
 # base_diff won't even exist if there is no difference.
 if [[ -f ${base_diff} ]]
 then

.gitlab/scripts/report_build_and_test_failure

@@ -13,20 +13,33 @@
 
 echo "Runs if there was at least one failure on ${MACHINE_NAME}"
 
-cd ${AUTOTEST_ROOT}/autotest && git pull
+cd ${AUTOTEST_ROOT}/autotest || \
+  { echo "Invalid 'autotest' dir: ${AUTOTEST_ROOT}/autotest"; exit 1; }
 mkdir -p ${MACHINE_NAME}
 
 rundir="${MACHINE_NAME}/$(date +%Y-%m-%d)-gitlab-ci-${CI_COMMIT_REF_SLUG}"
 rundir=$(${CI_PROJECT_DIR}/.gitlab/scripts/safe_create_rundir $rundir)
 
-echo "There was an error while running CI on ${MACHINE_NAME}" > ${rundir}/gitlab.err
-echo "See the pipeline here -> $CI_PIPELINE_URL" >> ${rundir}/gitlab.err
+printf "%s\n" "Some 'build-and-test' jobs on ${MACHINE_NAME} FAILED." \
+  "Pipeline URL:" "$CI_PIPELINE_URL" > ${rundir}/gitlab.err
 
 msg="GitLab CI log for build-and-test on ${MACHINE_NAME} ($(date +%Y-%m-%d))"
 
+# Create 'autotest-email.html' to indicate failure:
 cp ${rundir}/gitlab.err ${rundir}/autotest-email.html
 
-git pull
-git add ${rundir}
-git commit -am "${msg}"
-git push origin master
+if [[ "$AUTOTEST_COMMIT" != "NO" ]]; then
+  git pull && \
+  git add ${rundir} && \
+  git commit -m "${msg}" && \
+  git push origin master
+else
+  for file in ${rundir}/*; do
+    echo "------------------------------"
+    echo "Content of '$file'"
+    echo "******************************"
+    cat $file
+    echo "******************************"
+  done
+  rm -rf ${rundir} || true
+fi

.gitlab/scripts/report_build_and_test_success

@@ -13,18 +13,30 @@
 
 echo "Can only run if all the ${MACHINE_NAME} jobs passed"
 
-cd ${AUTOTEST_ROOT}/autotest && git pull
+cd ${AUTOTEST_ROOT}/autotest || \
+  { echo "Invalid 'autotest' dir: ${AUTOTEST_ROOT}/autotest"; exit 1; }
 mkdir -p ${MACHINE_NAME}
 
 rundir="${MACHINE_NAME}/$(date +%Y-%m-%d)-gitlab-ci-${CI_COMMIT_REF_SLUG}"
 rundir=$(${CI_PROJECT_DIR}/.gitlab/scripts/safe_create_rundir $rundir)
 
-echo "The ${MACHINE_NAME} jobs were successful" > ${rundir}/gitlab.out
-echo "See the pipeline here -> $CI_PIPELINE_URL" >> ${rundir}/gitlab.err
+printf "%s\n" "The 'build-and-test' jobs on ${MACHINE_NAME} were SUCCESSFUL." \
+  "Pipeline URL:" "$CI_PIPELINE_URL" > ${rundir}/gitlab.out
 
 msg="GitLab CI log for build-and-test on ${MACHINE_NAME} ($(date +%Y-%m-%d))"
 
-git pull
-git add ${rundir}
-git commit -am "${msg}"
-git push origin master
+if [[ "$AUTOTEST_COMMIT" != "NO" ]]; then
+  git pull && \
+  git add ${rundir} && \
+  git commit -m "${msg}" && \
+  git push origin master
+else
+  for file in ${rundir}/*; do
+    echo "------------------------------"
+    echo "Content of '$file'"
+    echo "******************************"
+    cat $file
+    echo "******************************"
+  done
+  rm -rf ${rundir} || true
+fi

CHANGELOG

@@ -10,6 +10,15 @@
 
 Version 4.3.1 (development)
 ===========================
+- Added support for automatic differentiation. Users can select between 
+  native implementation and external library implementation at the 
+  configuration phase. A parallel and two serial examples are implemented
+  in the autodiff miniapp directory.
+
+- Added support for mesh preprocessing to resolve fine scale problem data
+  before simulation. This feature uses adaptive mesh refinement to control the
+  associated data oscillation error. See the new Example 30/30p.
+
 - Switched from Artistic Style (astyle) version 2.05.1 to version 3.1 for code
   formatting. See the "make style" target.
 
@@ -60,6 +69,16 @@ Version 4.3.1 (development)
 - Remove the 'u' flag in the ar command, to update all files in the archive,
   avoiding file name collisions from different subdirectories.
 
+- Added initial TMOP-based capabilities for surface fitting and tangential
+  relaxation in the mesh-optimizer and pmesh-optimizer miniapps.
+ 
+- Added ParMesh Adjaceny Set (adjset) creation support to the Conduit Mesh
+  Blueprint MFEM wrapper functions in ConduitDataCollection.
+
+- `HypreParVector` and `Vector` now support move semantics, and the copy
+  constructor for `HypreParVector` now copies the local vector data.
+
+
 Version 4.3, released on July 29, 2021
 ======================================
 

CMakeLists.txt

@@ -252,6 +252,11 @@ if (MFEM_USE_OPENMP OR MFEM_USE_LEGACY_OPENMP)
   endif()
   find_package(OpenMP REQUIRED)
   set(OPENMP_LIBRARIES ${OpenMP_CXX_LIBRARIES})
+  if(APPLE)
+    # On macOS, the compiler needs additional help to find the <omp.h> header.
+    # See issue #2642 for more information.
+    include_directories(${OpenMP_CXX_INCLUDE_DIRS})
+  endif(APPLE)
 endif()
 
 # SuiteSparse (before SUNDIALS which may depend on KLU)
@@ -367,6 +372,12 @@ if (MFEM_USE_HIOP)
   # find_package updates HIOP_FOUND, HIOP_INCLUDE_DIRS, HIOP_LIBRARIES
 endif()
 
+# CoDiPack package
+if (MFEM_USE_CODIPACK)
+  find_package(CODIPACK REQUIRED)
+  # find_package updates CODIPACK_FOUND, CODIPACK_INCLUDE_DIRS, CODIPACK_LIBRARIES
+endif()
+
 # OCCA
 if (MFEM_USE_OCCA)
    find_package(OCCA REQUIRED)
@@ -446,7 +457,7 @@ endif()
 set(MFEM_TPLS OPENMP HYPRE BLAS LAPACK SuperLUDist METIS SuiteSparse SUNDIALS PETSC
     SLEPC MESQUITE MUMPS STRUMPACK AXOM FMS CONDUIT Ginkgo GNUTLS GSLIB NETCDF
     MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE ADIOS2
-    CUSPARSE MKL_CPARDISO AMGX CALIPER BENCHMARK PARELAG MPI_CXX)
+    CUSPARSE MKL_CPARDISO AMGX CALIPER CODIPACK BENCHMARK PARELAG MPI_CXX)
 
 # Add all *_FOUND libraries in the variable TPL_LIBRARIES.
 set(TPL_LIBRARIES "")
@@ -576,7 +587,11 @@ add_custom_target(${MFEM_EXEC_PREREQUISITES_TARGET_NAME})
 # Create a target for all examples and, optionally, enable it.
 set(MFEM_ALL_EXAMPLES_TARGET_NAME examples)
 add_mfem_target(${MFEM_ALL_EXAMPLES_TARGET_NAME} ${MFEM_ENABLE_EXAMPLES})
-add_subdirectory(examples EXCLUDE_FROM_ALL)
+if (MFEM_ENABLE_EXAMPLES)
+  add_subdirectory(examples) #install examples if enabled
+else()
+  add_subdirectory(examples EXCLUDE_FROM_ALL)
+endif()
 
 # Create a target for all miniapps and, optionally, enable it.
 set(MFEM_ALL_MINIAPPS_TARGET_NAME miniapps)

CONTRIBUTING.md

@@ -42,6 +42,7 @@ back to them before issuing pull requests:
   - [New Feature Development](#new-feature-development)
   - [Developer Guidelines](#developer-guidelines)
   - [Pull Requests](#pull-requests)
+  - [MFEM PR Rules](#mfem-pr-rules)
   - [Pull Request Checklist](#pull-request-checklist)
   - [Master/Next Workflow](#masternext-workflow)
   - [Releases](#releases)
@@ -67,8 +68,9 @@ Origin](#developers-certificate-of-origin-11) at the end of this file.*
   with regards to documentation and code styling.
 - Please do not commit large/binary files to the central repository (use a fork
   instead).
-- Pull requests  should be issued toward `mfem:master`. Make sure
-  to check the items off the [Pull Request Checklist](#pull-request-checklist).
+- Pull requests should be issued toward `mfem:master`. Make sure
+  to check the items off the [Pull Request Checklist](#pull-request-checklist) and
+  follow the [MFEM PR Rules](#mfem-pr-rules).
 - When your contribution is fully working and ready to be reviewed, add
   the `ready-for-review` label.
 - PRs are treated similarly to journal submission with an "editor" assigning two
@@ -121,6 +123,7 @@ The MFEM source code has the following structure:
   ├── mesh
   ├── miniapps
   │   ├── adjoint
+  │   ├── autodiff
   │   ├── common
   │   ├── electromagnetics
   │   ├── gslib
@@ -326,8 +329,12 @@ Before you can start, you need a GitHub account, here are a few suggestions:
     change the code by default.
 
 - Code specifics
-  - All significant new classes, methods and functions have Doxygen-style
-    documentation in source comments.
+  - All new public, protected, and private classes, methods, data members, and
+    functions have Doxygen-style documentation in source comments.
+  - In addition to arguments and functionality, documentation should include the
+    current limitations of the code, any background information that is
+    implicitly assumed in the implementation, and the ownership and lifetime
+    of data.
   - Consistent code styling is enforced with `make style` in the top-level
     directory. This requires [Artistic Style](http://astyle.sourceforge.net) (we
     specifically use version 3.1). See also the file `config/mfem.astylerc`.
@@ -335,6 +342,9 @@ Before you can start, you need a GitHub account, here are a few suggestions:
     internal library code. (You can use `std` in examples and miniapps.)
   - When manually resolving conflicts during a merge, make sure to mention the
     conflicted files in the commit message.
+  - All significant new features and changes should be documented in CHANGELOG.
+  - New examples and miniapps should have documentation on the MFEM webpage.
+
 
 ### Pull Requests
 
@@ -400,6 +410,83 @@ Before you can start, you need a GitHub account, here are a few suggestions:
 - If triggered, track the status of the LLNL GitLab tests. If failing, ask
   one of the _LLNL developers_ for details.
 
+
+### MFEM PR Rules
+
+The Pull Request (PR) approval process in MFEM is similar to the approval of papers in a peer-reviewed journal. In particular:
+
+1. There is an MFEM board of "editors" that evaluates new PRs and assigns "reviewers" for each PR.
+
+2. The assigned reviewers are responsible to carefully review and test the proposed PR.
+
+3. A PR can be (manually) merged in the *next* branch only if 2 of the assigned reviewers have approved it and it has passed internal testing. This merge can be performed by any of the assigned reviewers or by any of the editors.
+
+4. A PR can be merged in the *master* branch only if it has been tested successfully for a week in *next* and an editor has (optionally) taken a final look. This merge can be performed only by one of the editors.
+
+#### Responsibilities of Editors
+
+The current list of MFEM editors is:
+
+- @v-dobrev (Veselin Dobrev)
+- @tzanio (Tzanio Kolev)
+- @pazner (Will Pazner)
+- @mlstowell (Mark Stowell)
+
+**The responsibilities of the editors are:**
+
+1. To assign appropriate milestone and labels for new PRs, e.g. *bugfix*, *minor*, *api-change*, *high-impact*, etc.
+
+2. To assign at least 2 reviewers for new PRs. An editor can also be a reviewer. The editor, reviewers, and author should be listed as "Assignees" on the GitHub PR page. After assignment, the `in-review` label should be added.
+
+3. To complete the initial PR evaluation and assignments in a timely manner: 1 week from submission.
+
+4. To assist reviewers when they need help with their reviews (but also to stay out of the way when they don't).
+
+5. To remind the reviewers about timely completion of their review.
+
+6. To take a final look and complete the PR merge in *master*. The final look step is optional and shouldn't take more than 3 days.
+
+7. The assignment of bugfixes should be expedited proportional to their importance, e.g. in some cases the editor can assign much shorter review window.
+
+#### Responsibilities of Reviewers
+
+Everyone on the MFEM team can be asked to serve as a reviewer on a PR in their area of expertise.
+
+**The responsibilities of the reviewers are:**
+
+1. To let the editors know if the proposed assignment is not a good match for them.
+
+2. To communicate with the PR author, provide feedback and work with them to resolve issues.
+
+3. To ensure the quality of the PR by making sure that the code adheres to the [Developer Guidelines](#developer-guidelines), e.g. all methods, data members, and functions have documentation, including data ownership and lifetime, new examples/miniapps have a corresponding PR in mfem/web, major features have `CHANGELOG` entries, etc.
+
+3. To seek help from the editors in case of difficulties.
+
+4. To complete the review in a timely manner: 3 weeks from assignment.
+
+5. To test the PR thoroughly before merging in *next*. The PR author is also encouraged to perform testing and inform the reviewers about the results.
+
+6. To monitor the PR impact on the testing in the *next* branch and alert the editors that the PR is ready for merging in *master*.
+
+7. The review of bugfixes should be expedited proportional to their importance. The review window can be much less than three weeks in such cases.
+
+#### Responsibilities of Authors
+
+Authors should clearly indicate when a PR is ready for review (before that the PR should be marked as `Draft` or `[WIP]`).
+
+**The responsibilities of the authors are:**
+
+1. To follow the instructions and PR checklist in the `CONTRIBUTING.md` document in the MFEM repository.
+
+2. To respond to reviewer feedback in a timely manner.
+
+3. Authors are encouraged to perform testing and inform the reviewers about the results.
+
+4. Authors can use the "Reviewers" section of the GitHub PR page to suggest reviewers, but the "Assignees" section will show who the editor has assigned to do the reviews.
+
+5. To indicate when the PR is ready for review by adding the `ready-for-review` label.
+
+
 ### Pull Request Checklist
 
 Before a PR can be merged, it should satisfy the following:
@@ -453,7 +540,9 @@ Before a PR can be merged, it should satisfy the following:
      - [ ] The miniapps go at the end of the page, and are usually listed only under a specific "Application (PDE)" category.
      - [ ] Add a short description of the miniapp in the "Extensive Examples" section of `features.md`.
 - [ ] New capability:
-   - [ ] All significant new classes, methods and functions have Doxygen-style documentation in source comments.
+   - [ ] All new public, protected, and private classes, methods, data members, and functions have full Doxygen-style documentation in source comments. Documentation should include descriptions of member data, function arguments and return values, template parameters, and prerequisites for calling new functions.
+   - [ ] Pointer arguments and return values must specify whether ownership is being transferred or lent with the call.
+   - [ ] Any new functions should include descriptions of their intended use e.g. for internal use only, user-facing, etc., along with references to example code whenever possible/appropriate.
    - [ ] Consider adding new sample runs in existing examples to highlight the new capability.
    - [ ] Consider saving cool simulation pictures with the new capability in the Confluence gallery (LLNL only) or submitting them, via pull request, to the gallery section of the `mfem/web` repo.
    - [ ] If this is a major new feature, consider mentioning it in the short summary inside `README` *(rare)*.
@@ -464,6 +553,7 @@ Before a PR can be merged, it should satisfy the following:
 - [ ] (LLNL only) After merging:
    - [ ] Update internal tests to include the new features.
 
+
 ### Master/Next Workflow
 
 MFEM uses a `master`/`next`-branch workflow as described below:
@@ -555,8 +645,10 @@ MFEM uses a `master`/`next`-branch workflow as described below:
     - Update version and shortlinks in `src/index.md` and `src/download.md`.
     -  Use [cloc-1.62.pl](http://cloc.sourceforge.net/) and `ls -lh` to estimate the SLOC and the tarball size in `src/download.md`.
 
+
 ## LLNL Workflow
 
+
 ### Mirroring on Bitbucket
 
 - The GitHub `master` and `next` branches are mirrored to the LLNL institutional
@@ -576,6 +668,7 @@ MFEM uses a `master`/`next`-branch workflow as described below:
   - `mfem:gh-next` -- Bleeding-edge development version, may be broken, use at
      your own risk.
 
+
 ### Mirroring on GitLab
 
 - MFEM repository is also mirrored on the LLNL GitLab instance, in a
@@ -598,6 +691,7 @@ In addition, developers can set local git hooks to run some quick checks on
 commit or push, see the [README](config/githooks/README.md) in the `config/githooks`
 directory.
 
+
 ### Linux and Mac smoke tests
 We use GitHub Actions to drive the default tests on the `master` and `next`
 branches. See the `.github/workflows` files and the logs at
@@ -609,6 +703,7 @@ constraint on jobs. Two virtual machines are configured - Mac (OS X) and Linux.
 - Tests on the `master` branch are triggered whenever a PR is issued on this branch.
 - Tests on the `next` branch are currently scheduled to run each night.
 
+
 ### Windows smoke test
 We use Appveyor to test building with the MS Visual C++ compiler in a Windows
 environment, as well as to test the CMake build. See the `.appveyor` file and the
@@ -618,6 +713,7 @@ build logs at
 CMake is used to generate the MSVC Project files and drive the build.  A release
 and debug build is performed with a simple run of `ex1` to verify the executable.
 
+
 ### Tests at LLNL
 
 - We mirror the `master` and `next` branches internally (to `gh-master` and

INSTALL

@@ -467,6 +467,14 @@ 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_CODIPACK = YES/NO
+  Enable automatic differentiation using the CoDiPack library.
+  www.scicomp.uni-kl.de/codi/
+
+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
@@ -703,6 +711,11 @@ The specific libraries and their options are:
   Options: HIOP_OPT, HIOP_LIB.
   Versions: HIOP >= 0.4.6.
 
+- CoDiPack (optiobal), used with MFEM_USE_CODIPACK = YES
+  URL: https://www.scicomp.uni-kl.de/codi/
+  Options: CODIPACK_OPT
+  Versions: 1.9.3
+
 - GSLIB (optional), used when MFEM_USE_GSLIB = YES. The gslib library must be
   built prior to the MFEM build, as follows: download gslib-1.0.7, untar it at
   the same level as MFEM and create a symbolic link: "ln -s gslib-1.0.7 gslib".
@@ -908,6 +921,8 @@ MFEM_USE_MPFR
 MFEM_USE_ZLIB
 MFEM_USE_PUMI
 MFEM_USE_HIOP
+MFEM_USE_CODIPACK
+MFEM_USE_ADFORWARD
 MFEM_USE_CUDA
 MFEM_USE_OCCA
 MFEM_USE_CEED
@@ -967,6 +982,7 @@ The CMake build system adds auto-detection for the following packages/libraries:
  - POSIXCLOCKS
  - PUMI
  - HIOP
+ - CoDiPack
  - OCCA
  - RAJA
  - UMPIRE

config/cmake/MFEMConfig.cmake.in

@@ -54,6 +54,8 @@ 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_CODIPACK @MFEM_USE_CODIPACK@)
+set(MFEM_USE_ADFORWARD @MFEM_USE_ADFORWARD@)
 set(MFEM_USE_CALIPER @MFEM_USE_CALIPER@)
 set(MFEM_USE_BENCHMARK @MFEM_USE_BENCHMARK@)
 set(MFEM_USE_PARELAG @MFEM_USE_PARELAG@)

config/cmake/config.hpp.in

@@ -175,6 +175,12 @@
 // Enable interface to the MKL CPardiso library.
 #cmakedefine MFEM_USE_MKL_CPARDISO
 
+// Use forward mode for automatic differentiation
+#cmakedefine MFEM_USE_ADFORWARD
+
+// Enable the use of the CoDiPack library for AD
+#cmakedefine MFEM_USE_CODIPACK
+
 // Enable MFEM functionality based on the Google Benchmark library.
 #cmakedefine MFEM_USE_BENCHMARK
 

config/cmake/modules/FindCODIPACK.cmake

@@ -0,0 +1,24 @@
+# Copyright (c) 2010-2021, 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.
+
+# Automatic differentiation using the CoDiPack library.
+#  www.scicomp.uni-kl.de/codi/
+# Sets the following variables:
+#   - CODIPACK_FOUND
+#   - CODIPACK_INCLUDE_DIRS
+#   - CODIPACK_LIBRARIES
+
+include(MfemCmakeUtilities)
+mfem_find_package(CODIPACK CODIPACK CODIPACK_DIR
+  "include" "codi.h"
+  "lib" ""
+  "Paths to headers required by CODIPACK."
+  "Libraries required by CODIPACK.")

config/cmake/modules/MfemCmakeUtilities.cmake

@@ -100,6 +100,8 @@ macro(add_mfem_examples EXE_SRCS)
 
     string(REPLACE ".cpp" "" EXE_NAME "${EXE_PREFIX}${SRC_FILENAME}")
     mfem_add_executable(${EXE_NAME} ${SRC_FILE})
+    install(TARGETS ${EXE_NAME}
+            RUNTIME DESTINATION examples)
     add_dependencies(${MFEM_ALL_EXAMPLES_TARGET_NAME} ${EXE_NAME})
     if (EXE_NEEDED_BY)
       add_dependencies(${EXE_NEEDED_BY} ${EXE_NAME})

config/config.hpp.in

@@ -180,6 +180,12 @@
 // Enable interface to the MKL CPardiso library.
 // #define MFEM_USE_MKL_CPARDISO
 
+// Use forward mode for automatic differentiation
+// #define MFEM_USE_ADFORWARD
+
+// Enable the use of the CoDiPack library for AD
+// #define MFEM_USE_CODIPACK
+
 // Enable functionality based on the Google Benchmark library.
 // #define MFEM_USE_BENCHMARK
 

config/config.mk.in

@@ -58,6 +58,8 @@ MFEM_USE_UMPIRE        = @MFEM_USE_UMPIRE@
 MFEM_USE_SIMD          = @MFEM_USE_SIMD@
 MFEM_USE_ADIOS2        = @MFEM_USE_ADIOS2@
 MFEM_USE_MKL_CPARDISO  = @MFEM_USE_MKL_CPARDISO@
+MFEM_USE_ADFORWARD     = @MFEM_USE_ADFORWARD@
+MFEM_USE_CODIPACK      = @MFEM_USE_CODIPACK@
 MFEM_USE_BENCHMARK     = @MFEM_USE_BENCHMARK@
 MFEM_USE_PARELAG       = @MFEM_USE_PARELAG@
 

config/defaults.cmake

@@ -58,6 +58,8 @@ option(MFEM_USE_SIMD "Enable use of SIMD intrinsics" OFF)
 option(MFEM_USE_ADIOS2 "Enable ADIOS2" OFF)
 option(MFEM_USE_CALIPER "Enable Caliper support" OFF)
 option(MFEM_USE_MKL_CPARDISO "Enable MKL CPardiso" OFF)
+option(MFEM_USE_ADFORWARD "Enable forward mode for AD" OFF)
+option(MFEM_USE_CODIPACK "Enable automatic differentiation (AD) using CoDiPack" OFF)
 option(MFEM_USE_BENCHMARK "Enable Google Benchmark" OFF)
 option(MFEM_USE_PARELAG "Enable ParELAG" OFF)
 
@@ -243,6 +245,9 @@ set(BLAS_LIBRARIES "" CACHE STRING "The BLAS library.")
 set(LAPACK_INCLUDE_DIRS "" CACHE STRING "Path to LAPACK headers.")
 set(LAPACK_LIBRARIES "" CACHE STRING "The LAPACK library.")
 
+set(CODIPACK_INCLUDE_DIRS "${MFEM_DIR}/../CoDiPack/inlude" CACHE STRING "Path to CoDiPack headers.")
+set(CODIPACK_LIBRARIES "")
+
 # Some useful variables:
 set(CMAKE_SKIP_PREPROCESSED_SOURCE_RULES ON) # Skip *.i rules
 set(CMAKE_SKIP_ASSEMBLY_SOURCE_RULES  ON)    # Skip *.s rules

config/defaults.mk

@@ -59,6 +59,9 @@ HIP_FLAGS = --amdgpu-target=$(HIP_ARCH)
 HIP_XCOMPILER =
 HIP_XLINKER   = -Wl,
 
+# Flags for generating dependencies.
+DEP_FLAGS = -MM -MT
+
 ifneq ($(NOTMAC),)
    AR      = ar
    ARFLAGS = crv
@@ -86,6 +89,9 @@ else
    BUILD_RPATH = $(XLINKER)-undefined,dynamic_lookup
    INSTALL_SOFLAGS = $(subst $1 ,,$(call MAKE_SOFLAGS,$(MFEM_LIB_DIR)))
    INSTALL_RPATH = $(XLINKER)-undefined,dynamic_lookup
+   # Silence unused command line argument warnings when generating dependencies
+   # with mpicxx and clang
+   DEP_FLAGS := -Wno-unused-command-line-argument $(DEP_FLAGS)
 endif
 
 # Set CXXFLAGS to overwrite the default selection of DEBUG_FLAGS/OPTIM_FLAGS
@@ -151,6 +157,8 @@ MFEM_USE_UMPIRE        = NO
 MFEM_USE_SIMD          = NO
 MFEM_USE_ADIOS2        = NO
 MFEM_USE_MKL_CPARDISO  = NO
+MFEM_USE_ADFORWARD     = NO
+MFEM_USE_CODIPACK      = NO
 MFEM_USE_BENCHMARK     = NO
 MFEM_USE_PARELAG       = NO
 
@@ -408,6 +416,11 @@ HIOP_DIR = @MFEM_DIR@/../hiop/install
 HIOP_OPT = -I$(HIOP_DIR)/include
 HIOP_LIB = -L$(HIOP_DIR)/lib -lhiop $(LAPACK_LIB)
 
+# CoDiPack
+CODIPACK_DIR = @MFEM_DIR@/../CoDiPack
+CODIPACK_OPT = -I$(CODIPACK_DIR)
+CODIPACK_LIB =
+
 # GSLIB library
 GSLIB_DIR = @MFEM_DIR@/../gslib/build
 GSLIB_OPT = -I$(GSLIB_DIR)/include

doc/CodeDocumentation.conf.in

@@ -781,6 +781,7 @@ INPUT                  = @MFEM_SOURCE_DIR@/doc/CodeDocumentation.dox \
                          @MFEM_SOURCE_DIR@/miniapps/gslib \
                          @MFEM_SOURCE_DIR@/miniapps/meshing \
                          @MFEM_SOURCE_DIR@/miniapps/mtop \
+                         @MFEM_SOURCE_DIR@/miniapps/autodiff \
                          @MFEM_SOURCE_DIR@/miniapps/navier \
                          @MFEM_SOURCE_DIR@/miniapps/nurbs \
                          @MFEM_SOURCE_DIR@/miniapps/performance \

doc/CodeDocumentation.dox

@@ -194,6 +194,8 @@ namespace mfem {
  * - <a class="el" href="parheat_8cpp_source.html">Optimization gradients</a>: Gradients of PDE-constrained function
  * - <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
+ * - <a class="el" href="par__example_8cpp_source.html">Parallel pLaplacian example using AD</a>: Parallel pLaplacian example
+ * - <a class="el" href="seq__example_8cpp_source.html">Serial pLaplacian example using AD</a>: Serial pLaplacian example
  *
  * See also the <a class="el" href="https://mfem.org/examples/">examples documentation</a> online.
  */

examples/CMakeLists.txt

@@ -37,6 +37,7 @@ list(APPEND ALL_EXE_SRCS
   ex27.cpp
   ex28.cpp
   ex29.cpp
+  ex30.cpp
   )
 
 if (MFEM_USE_MPI)
@@ -70,6 +71,7 @@ if (MFEM_USE_MPI)
     ex27p.cpp
     ex28p.cpp
     ex29p.cpp
+    ex30p.cpp
     )
 endif()
 
@@ -80,78 +82,80 @@ include_directories(BEFORE ${PROJECT_BINARY_DIR})
 add_mfem_examples(ALL_EXE_SRCS)
 
 # Add a test for each example
-foreach(SRC_FILE ${ALL_EXE_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
+if (MFEM_ENABLE_TESTING)
+  foreach(SRC_FILE ${ALL_EXE_SRCS})
+    get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
+    string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
 
-  set(THIS_TEST_OPTIONS "-no-vis")
-  if (${TEST_NAME} MATCHES "ex0p?")
-    set(THIS_TEST_OPTIONS)
-  endif()
-  if (${TEST_NAME} MATCHES "ex10p*")
-    list(APPEND THIS_TEST_OPTIONS "-tf" "5")
-  elseif(${TEST_NAME} MATCHES "ex15p*")
-    list(APPEND THIS_TEST_OPTIONS "-e" "1")
-  elseif(${TEST_NAME} MATCHES "ex27p*")
-    list(APPEND THIS_TEST_OPTIONS "-dg")
-  endif()
+    set(THIS_TEST_OPTIONS "-no-vis")
+    if (${TEST_NAME} MATCHES "ex0p?")
+      set(THIS_TEST_OPTIONS)
+    endif()
+    if (${TEST_NAME} MATCHES "ex10p*")
+      list(APPEND THIS_TEST_OPTIONS "-tf" "5")
+    elseif(${TEST_NAME} MATCHES "ex15p*")
+      list(APPEND THIS_TEST_OPTIONS "-e" "1")
+    elseif(${TEST_NAME} MATCHES "ex27p*")
+      list(APPEND THIS_TEST_OPTIONS "-dg")
+    endif()
 
-  if (NOT (${TEST_NAME} MATCHES ".*p$"))
-    add_test(NAME ${TEST_NAME}_ser
-      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-  else()
-    add_test(NAME ${TEST_NAME}_np=${MFEM_MPI_NP}
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
-  endif()
-endforeach()
-
-# Add CUDA/HIP tests.
-set(DEVICE_EXAMPLES
-    # serial examples with device support:
-    ex1 ex3 ex4 ex5 ex6 ex9 ex22 ex24 ex25 ex26
-    # parallel examples with device support:
-    ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex22p ex24p ex25p ex26p)
-set(MFEM_TEST_DEVICE)
-if (MFEM_USE_CUDA)
-  set(MFEM_TEST_DEVICE "cuda")
-elseif (MFEM_USE_HIP)
-  set(MFEM_TEST_DEVICE "hip")
-endif()
-if (MFEM_TEST_DEVICE)
-  foreach(TEST_NAME ${DEVICE_EXAMPLES})
-    set(THIS_TEST_OPTIONS "-no-vis" "-d" "${MFEM_TEST_DEVICE}")
     if (NOT (${TEST_NAME} MATCHES ".*p$"))
-      add_test(NAME ${TEST_NAME}_${MFEM_TEST_DEVICE}_ser
+      add_test(NAME ${TEST_NAME}_ser
         COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-    elseif (MFEM_USE_MPI)
-      add_test(NAME ${TEST_NAME}_${MFEM_TEST_DEVICE}_np=${MFEM_MPI_NP}
+    else()
+      add_test(NAME ${TEST_NAME}_np=${MFEM_MPI_NP}
         COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
         ${MPIEXEC_PREFLAGS}
         $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
         ${MPIEXEC_POSTFLAGS})
     endif()
   endforeach()
-endif()
 
-# If STRUMPACK is enabled, add a test run that uses it.
-if (MFEM_USE_STRUMPACK)
-  add_test(NAME ex11p_strumpack_np=${MFEM_MPI_NP}
-    COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-    ${MPIEXEC_PREFLAGS}
-    $<TARGET_FILE:ex11p> "-no-vis" "--strumpack"
-    ${MPIEXEC_POSTFLAGS})
-endif()
+  # Add CUDA/HIP tests.
+  set(DEVICE_EXAMPLES
+      # serial examples with device support:
+      ex1 ex3 ex4 ex5 ex6 ex9 ex22 ex24 ex25 ex26
+      # parallel examples with device support:
+      ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex22p ex24p ex25p ex26p)
+  set(MFEM_TEST_DEVICE)
+  if (MFEM_USE_CUDA)
+    set(MFEM_TEST_DEVICE "cuda")
+  elseif (MFEM_USE_HIP)
+    set(MFEM_TEST_DEVICE "hip")
+  endif()
+  if (MFEM_TEST_DEVICE)
+    foreach(TEST_NAME ${DEVICE_EXAMPLES})
+      set(THIS_TEST_OPTIONS "-no-vis" "-d" "${MFEM_TEST_DEVICE}")
+      if (NOT (${TEST_NAME} MATCHES ".*p$"))
+        add_test(NAME ${TEST_NAME}_${MFEM_TEST_DEVICE}_ser
+          COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
+      elseif (MFEM_USE_MPI)
+        add_test(NAME ${TEST_NAME}_${MFEM_TEST_DEVICE}_np=${MFEM_MPI_NP}
+          COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
+          ${MPIEXEC_PREFLAGS}
+          $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
+          ${MPIEXEC_POSTFLAGS})
+      endif()
+    endforeach()
+  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"
-    ${MPIEXEC_POSTFLAGS})
+  # If STRUMPACK is enabled, add a test run that uses it.
+  if (MFEM_USE_STRUMPACK)
+    add_test(NAME ex11p_strumpack_np=${MFEM_MPI_NP}
+      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()
 endif()
 
 # Include the examples/amgx directory if AmgX is enabled

examples/amgx/CMakeLists.txt

@@ -50,30 +50,32 @@ add_mfem_examples(AMGX_EXAMPLES_SRCS ${PFX} copy_amgx_json_files test_amgx)
 # which builds the examples and runs:
 #   ctest -R amgx
 
-# Command line options for the tests.
-# Example 1/1p:
-set(EX1_TEST_OPTS)
-set(EX1P_TEST_OPTS)
+if (MFEM_ENABLE_TESTING)
+  # Command line options for the tests.
+  # Example 1/1p:
+  set(EX1_TEST_OPTS)
+  set(EX1P_TEST_OPTS)
 
-# Add the tests: one test per source file.
-foreach(SRC_FILE ${AMGX_EXAMPLES_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
-  string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
-  set(TEST_NAME ${PFX}${TEST_NAME})
+  # Add the tests: one test per source file.
+  foreach(SRC_FILE ${AMGX_EXAMPLES_SRCS})
+    get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
+    string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
+    string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
+    set(TEST_NAME ${PFX}${TEST_NAME})
 
-  set(THIS_TEST_OPTIONS "-no-vis")
-  list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
-  # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
+    set(THIS_TEST_OPTIONS "-no-vis")
+    list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
+    # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
 
-  if (NOT (${TEST_NAME} MATCHES ".*p$"))
-    add_test(NAME ${TEST_NAME}_ser
-      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-  else()
-    add_test(NAME ${TEST_NAME}_np=${MFEM_MPI_NP}
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
-  endif()
-endforeach()
+    if (NOT (${TEST_NAME} MATCHES ".*p$"))
+      add_test(NAME ${TEST_NAME}_ser
+        COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
+    else()
+      add_test(NAME ${TEST_NAME}_np=${MFEM_MPI_NP}
+        COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
+        ${MPIEXEC_PREFLAGS}
+        $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
+        ${MPIEXEC_POSTFLAGS})
+    endif()
+  endforeach()
+endif()

examples/caliper/CMakeLists.txt

@@ -30,22 +30,22 @@ set(PREFIX caliper_)
 add_mfem_examples(CALIPER_EXE_SRCS ${PREFIX})
 
 # Add a test for each example
-foreach(SRC_FILE ${CALIPER_EXE_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
-
-  set(THIS_TEST_OPTIONS "-no-vis")
-
-  if (NOT (${TEST_NAME} MATCHES ".*p$"))
-    add_test(NAME ${TEST_NAME}_ser
-      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-  else()
-    add_test(NAME ${TEST_NAME}_np=${MFEM_MPI_NP}
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
-  endif()
-endforeach()
+if (MFEM_ENABLE_TESTING)
+  foreach(SRC_FILE ${CALIPER_EXE_SRCS})
+    get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
+    string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
 
+    set(THIS_TEST_OPTIONS "-no-vis")
 
+    if (NOT (${TEST_NAME} MATCHES ".*p$"))
+      add_test(NAME ${TEST_NAME}_ser
+        COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
+    else()
+      add_test(NAME ${TEST_NAME}_np=${MFEM_MPI_NP}
+        COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
+        ${MPIEXEC_PREFLAGS}
+        $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
+        ${MPIEXEC_POSTFLAGS})
+    endif()
+  endforeach()
+endif()

examples/ex1_es.cpp

@@ -0,0 +1,266 @@
+//                                MFEM Example 1
+//
+// Compile with: make ex1
+//
+// Sample runs:  ex1 -m ../data/square-disc.mesh
+//               ex1 -m ../data/star.mesh
+//               ex1 -m ../data/escher.mesh
+//               ex1 -m ../data/fichera.mesh
+//               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
+//               ex1 -m ../data/disc-nurbs.mesh -o -1
+//               ex1 -m ../data/pipe-nurbs.mesh -o -1
+//               ex1 -m ../data/star-surf.mesh
+//               ex1 -m ../data/square-disc-surf.mesh
+//               ex1 -m ../data/inline-segment.mesh
+//               ex1 -m ../data/amr-quad.mesh
+//               ex1 -m ../data/amr-hex.mesh
+//               ex1 -m ../data/fichera-amr.mesh
+//               ex1 -m ../data/mobius-strip.mesh
+//               ex1 -m ../data/mobius-strip.mesh -o -1 -sc
+//
+// 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>
+
+using namespace std;
+using namespace mfem;
+
+int main(int argc, char *argv[])
+{
+   // 1. Parse command-line options.
+   const char *mesh_file = "./star-set.mesh";
+   int order = 1;
+   int rs = -1;
+   int ra = 0;
+   int bt = EntitySets::INVALID;
+   const char *bs = "Origin";
+   bool static_cond = false;
+   bool visualization = 1;
+
+   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(&rs, "-rs", "--refine-serial",
+                  "Number of serial refinement levels");
+   args.AddOption(&ra, "-ra", "--refine-adaptive",
+                  "Number of adaptive refinement levels");
+   args.AddOption(&bt, "-bt", "--bc-entity-type",
+                  "");
+   args.AddOption(&bs, "-bs", "--bc-entity-set-name",
+                  "");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(cout);
+      return 1;
+   }
+   args.PrintOptions(cout);
+
+   // 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();
+
+   // 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 50,000
+   //    elements.
+   {
+      int ref_levels = ( rs >= 0 ) ? rs :
+                       (int)floor(log(50000./mesh->GetNE())/log(2.)/dim);
+      for (int l = 0; l < ref_levels; l++)
+      {
+         mesh->UniformRefinement();
+      }
+   }
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+
+   if ( ra > 0 )
+   {
+      cout << "calling EnsureNCMesh" << endl;
+      mesh->EnsureNCMesh();
+      cout << "back from EnsureNCMesh" << endl;
+   }
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+   cout << "Calling RandomRefinement " << ra << " times." << endl;
+   for (int l = 0; l < ra; l++)
+   {
+      mesh->RandomRefinement(0.2);
+   }
+   cout << "Done with refinement" << endl;
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   if ( mesh->ncmesh )
+   {
+      mesh->ncmesh->PrintStats(cout);
+
+      ofstream ofsV("vp.out");
+      ofstream ofsE("ce.out");
+      mesh->ncmesh->PrintVertexParents(ofsV);
+      mesh->ncmesh->PrintCoarseElements(ofsE);
+   }
+
+   // 4. Define a finite element space on the 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 (mesh->GetNodes())
+   {
+      fec = mesh->GetNodes()->OwnFEC();
+      cout << "Using isoparametric FEs: " << fec->Name() << endl;
+   }
+   else
+   {
+      fec = new H1_FECollection(order = 1, dim);
+   }
+   FiniteElementSpace *fespace = new FiniteElementSpace(mesh, fec);
+   cout << "Number of finite element unknowns: "
+        << fespace->GetTrueVSize() << endl;
+
+   // 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.
+   Array<int> ess_tdof_list;
+   if ( bt == EntitySets::INVALID )
+   {
+      if (mesh->bdr_attributes.Size())
+      {
+         Array<int> ess_bdr(mesh->bdr_attributes.Max());
+         ess_bdr = 1;
+         fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      }
+   }
+   else
+   {
+      fespace->GetEssentialTrueDofs((EntitySets::EntityType)bt, bs,
+                                    ess_tdof_list);
+   }
+
+   cout << "Number of Dirichlet dofs: " << ess_tdof_list.Size() << endl;
+
+   // 6. 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 = new LinearForm(fespace);
+   ConstantCoefficient one(1.0);
+   b->AddDomainIntegrator(new DomainLFIntegrator(one));
+   b->Assemble();
+
+   // 7. 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);
+   x = 0.0;
+
+   // 8. 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 = new BilinearForm(fespace);
+   a->AddDomainIntegrator(new DiffusionIntegrator(one));
+
+   // 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, etc.
+   if (static_cond) { a->EnableStaticCondensation(); }
+   a->Assemble();
+
+   SparseMatrix A;
+   Vector B, X;
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);
+
+   cout << "Size of linear system: " << A.Height() << endl;
+
+#ifndef MFEM_USE_SUITESPARSE
+   // 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, 200, 1e-12, 0.0);
+#else
+   // 10. 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);
+#endif
+
+   // 11. Recover the solution as a finite element grid function.
+   a->RecoverFEMSolution(X, *b, x);
+
+   // 12. 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);
+   ofstream sol_ofs("sol.gf");
+   sol_ofs.precision(8);
+   x.Save(sol_ofs);
+
+   // 13. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream sol_sock(vishost, visport);
+      sol_sock.precision(8);
+      sol_sock << "solution\n" << *mesh << x << flush;
+   }
+
+   // 14. Free the used memory.
+   delete a;
+   delete b;
+   delete fespace;
+   if (order > 0) { delete fec; }
+   delete mesh;
+
+   return 0;
+}

examples/ex1p_es.cpp

@@ -0,0 +1,388 @@
+//                       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/escher.mesh
+//               mpirun -np 4 ex1p -m ../data/fichera.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/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/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
+//
+//               The following are examples of using EntitySets to define
+//               homogeneous Dirichlet boundary condition.  These examples
+//               require a modified mesh file and a specialized version of
+//               example 1 called "ex1p_es".
+//               mpirun -np 4 ex1p_es -m ./fichera-set.mesh -bt 0 -bs Origin
+//               mpirun -np 4 ex1p_es -m ./fichera-set.mesh -bt 1 -bs Axes
+//               mpirun -np 4 ex1p_es -m ./fichera-set.mesh
+//                                    -bt 1 -bs "Negative Axes"
+//               mpirun -np 4 ex1p_es -m ./fichera-set.mesh
+//                                    -bt 2 -bs "Interior Corner"
+//               mpirun -np 4 ex1p_es -m ./fichera-set.mesh
+//                                    -bt 2 -bs "Exterior Corner"
+//               mpirun -np 4 ex1p_es -m ./fichera-set.mesh
+//                                    -bt 3 -bs "Interior Corner"
+//               mpirun -np 4 ex1p_es -m ./fichera-set.mesh
+//                                    -bt 3 -bs "Exterior Corner"
+//               mpirun -np 4 ex1p_es -m ./fichera-set.mesh -bt 3 -bs "Steps"
+//
+// 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>
+
+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 = "./star-set.mesh";
+   int order = 1;
+   int rs = -1;
+   int rp = 2;
+   int ra = 0;
+   int bt = EntitySets::INVALID;
+   const char *bs = "Origin";
+   bool static_cond = false;
+   bool visualization = 1;
+
+   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(&rs, "-rs", "--refine-serial",
+                  "Number of serial refinement levels");
+   args.AddOption(&rp, "-rp", "--refine-parallel",
+                  "Number of parallel refinement levels");
+   args.AddOption(&ra, "-ra", "--refine-adaptive",
+                  "Number of adaptive refinement levels");
+   args.AddOption(&bt, "-bt", "--bc-entity-type",
+                  "");
+   args.AddOption(&bs, "-bs", "--bc-entity-set-name",
+                  "");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0)
+      {
+         args.PrintUsage(cout);
+      }
+      MPI_Finalize();
+      return 1;
+   }
+   if (myid == 0)
+   {
+      args.PrintOptions(cout);
+   }
+
+   // 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. We choose
+   //    'ref_levels' to be the largest number that gives a final mesh with no
+   //    more than 10,000 elements.
+   {
+      int ref_levels = ( rs >= 0 ) ? rs :
+                       (int)floor(log(10000./mesh->GetNE())/log(2.)/dim);
+      for (int l = 0; l < ref_levels; l++)
+      {
+         if ( myid == 0 ) { cout << "Uniform refinement in serial..."; }
+         mesh->UniformRefinement();
+      }
+      MPI_Barrier(MPI_COMM_WORLD);
+      if ( myid == 0 && rs > 0 ) { cout << "Done" << endl; }
+   }
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+   /*
+     At this point we have a serial mesh containing an EntitySets
+     object which stores the current node/edge/face/element indices
+     for each entity in each set.  This data is duplicated on each MPI
+     rank.
+    */
+   if ( ra > 0 )
+   {
+      cout << "calling EnsureNCMesh" << endl;
+      mesh->EnsureNCMesh();
+      cout << "back from EnsureNCMesh" << endl;
+   }
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+   /*
+     We now have an NCEntitySets object which stores the node indices
+     describing each enity in each node/edge/face set and the element
+     indices for the elements in each element set.  This data is
+     duplicated on each MPI rank.
+    */
+
+   // 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.
+   cout << "creating ParMesh from serial mesh" << endl;
+   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
+   cout << "done creating ParMesh from serial mesh" << endl;
+   delete mesh;
+   if ( pmesh->pent_sets )
+   {
+      cout << "pmesh->pent_sets is non NULL" << endl;
+      pmesh->pent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "pmesh->pent_sets is NULL" << endl;
+   }
+   /*
+     We now have a ParEntitySets object which marshals the data stored
+     in EntitySets objects.  The data has now been pruned so that each
+     rank only contains indices of local entities.
+
+     The NCEntitySets object remains unchanged...
+
+     If we have an NC mesh a different path is taken and the
+     EntitySets are ignored.
+
+     1) ParNCMesh is created from NCMesh
+       a) Creates a ParNCEntitySets object from ncmesh (every rank contains
+          information to find every entity)
+     2) ParNCMesh is pruned which involves renumbering elements and vertices
+     3) ParMesh is initialized from ParNCMesh
+     4) ParNCMesh::OnMeshUpdated is called
+     5) Mesh::GenerateNCFaceInfo is called
+    */
+   {
+      int par_ref_levels = rp;
+      for (int l = 0; l < par_ref_levels; l++)
+      {
+         if ( myid == 0 ) { cout << "Uniform refinement in parallel..."; }
+         pmesh->UniformRefinement();
+      }
+      MPI_Barrier(MPI_COMM_WORLD);
+      if ( myid == 0 && rs > 0 ) { cout << "Done" << endl; }
+   }
+   /*
+     RandomRefinement will end up calling
+     ParMesh::NonconformingRefinement which will create a new ParMesh
+     object using the ParNCMesh object and then call
+     ParMesh::OnMeshUpdated on this new mesh.
+    */
+
+   for (int l = 0; l < ra; l++)
+   {
+      pmesh->RandomRefinement(0.2);
+   }
+   if ( ra > 0 )
+   {
+      if ( pmesh->pent_sets )
+      {
+         cout << "pmesh->pent_sets is non NULL post random refinement" << endl;
+         pmesh->pent_sets->PrintSetInfo(cout);
+      }
+      else
+      {
+         cout << "pmesh->pent_sets is NULL post random refinement" << endl;
+      }
+   }
+
+   // 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();
+      if (myid == 0)
+      {
+         cout << "Using isoparametric FEs: " << fec->Name() << endl;
+      }
+   }
+   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 finite element unknowns: " << size << endl;
+   }
+
+   // 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.
+   Array<int> ess_tdof_list;
+   if ( bt == EntitySets::INVALID )
+   {
+      if (pmesh->bdr_attributes.Size())
+      {
+         Array<int> ess_bdr(pmesh->bdr_attributes.Max());
+         ess_bdr = 1;
+         fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      }
+   }
+   else
+   {
+      fespace->GetEssentialTrueDofs((EntitySets::EntityType)bt, bs,
+                                    ess_tdof_list);
+   }
+   for (int i=0; i<num_procs; i++)
+   {
+      if (myid == i)
+      {
+         cout << "Number of Dirichlet dofs on proc " << i << ": "
+              << ess_tdof_list.Size() << endl;
+      }
+   }
+
+   // 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
+   //    (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();
+
+   // 9. 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;
+
+   // 10. 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);
+   a->AddDomainIntegrator(new DiffusionIntegrator(one));
+
+   // 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, etc.
+   if (static_cond) { a->EnableStaticCondensation(); }
+   a->Assemble();
+
+   HypreParMatrix A;
+   Vector B, X;
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);
+
+   if (myid == 0)
+   {
+      cout << "Size of linear system: " << A.GetGlobalNumRows() << endl;
+   }
+
+   // 12. Define and apply a parallel PCG solver for AX=B with the BoomerAMG
+   //     preconditioner from hypre.
+   HypreSolver *amg = new HypreBoomerAMG(A);
+   HyprePCG *pcg = new HyprePCG(A);
+   pcg->SetTol(1e-12);
+   pcg->SetMaxIter(200);
+   pcg->SetPrintLevel(2);
+   pcg->SetPreconditioner(*amg);
+   pcg->Mult(B, X);
+
+   // 13. Recover the parallel grid function corresponding to X. This is the
+   //     local finite element solution on each processor.
+   a->RecoverFEMSolution(X, *b, x);
+
+   // 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, 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);
+
+      ofstream sol_ofs(sol_name.str().c_str());
+      sol_ofs.precision(8);
+      x.Save(sol_ofs);
+   }
+
+   // 15. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream sol_sock(vishost, visport);
+      sol_sock << "parallel " << num_procs << " " << myid << "\n";
+      sol_sock.precision(8);
+      sol_sock << "solution\n" << *pmesh << x << flush;
+   }
+
+   // 16. Free the used memory.
+   delete pcg;
+   delete amg;
+   delete a;
+   delete b;
+   delete fespace;
+   if (order > 0) { delete fec; }
+   delete pmesh;
+
+   MPI_Finalize();
+
+   return 0;
+}

examples/ex30.cpp

@@ -0,0 +1,195 @@
+//                            MFEM Example 30
+//
+// Compile with: make ex30
+//
+// Sample runs:  ex30 -m ../data/square-disc.mesh -o 1
+//               ex30 -m ../data/square-disc.mesh -o 2
+//               ex30 -m ../data/square-disc.mesh -o 2 -me 1e3
+//               ex30 -m ../data/square-disc-nurbs.mesh -o 2
+//               ex30 -m ../data/star.mesh -o 2 -eo 4
+//               ex30 -m ../data/fichera.mesh -o 2 -me 1e4
+//               ex30 -m ../data/disc-nurbs.mesh -o 2
+//               ex30 -m ../data/ball-nurbs.mesh -o 2 -eo 3 -e 1e-2 -me 1e4
+//               ex30 -m ../data/star-surf.mesh -o 2
+//               ex30 -m ../data/square-disc-surf.mesh -o 2
+//               ex30 -m ../data/amr-quad.mesh -l 2
+//
+// Description:  This is an example of adaptive mesh refinement preprocessing
+//               which lowers the data oscillation [1] to a user-defined
+//               relative threshold. There is no PDE being solved.
+//
+//               MFEM's capability to work with both conforming and
+//               nonconforming meshes is demonstrated in example 6. In some
+//               problems, the material data or loading data is not sufficiently
+//               resolved on the initial mesh. This missing fine scale data
+//               reduces the accuracy of the solution as well as the accuracy
+//               of some local error estimators. By preprocessing the mesh
+//               before solving the PDE, many issues can be avoided.
+//
+//               [1] Morin, P., Nochetto, R. H., & Siebert, K. G. (2000).
+//                   Data oscillation and convergence of adaptive FEM. SIAM
+//                   Journal on Numerical Analysis, 38(2), 466-488.
+//
+//               [2] Mitchell, W. F. (2013). A collection of 2D elliptic
+//                   problems for testing adaptive grid refinement algorithms.
+//                   Applied mathematics and computation, 220, 350-364.
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+// Piecewise-affine function which is sometimes mesh-conforming
+double affine_function(const Vector &p)
+{
+   double x = p(0), y = p(1);
+   if (x < 0.0)
+   {
+      return 1.0 + x + y;
+   }
+   else
+   {
+      return 1.0;
+   }
+}
+
+// Piecewise-constant function which is never mesh-conforming
+double jump_function(const Vector &p)
+{
+   if (p.Normlp(2.0) > 0.4 && p.Normlp(2.0) < 0.6) { return 1.0; }
+   return 5.0;
+}
+
+// Singular function derived from the Laplacian of the "steep wavefront"
+// problem in [2].
+double singular_function(const Vector &p)
+{
+   double x = p(0), y = p(1);
+   double alpha = 1000.0;
+   double xc = 0.75, yc = 0.5;
+   double r0 = 0.7;
+   double r = sqrt(pow(x - xc,2.0) + pow(y - yc,2.0));
+   double num = - ( alpha - pow(alpha,3) * (pow(r,2) - pow(r0,2)) );
+   double denom = pow(r * ( pow(alpha,2) * pow(r0,2) + pow(alpha,2) * pow(r,2) \
+                            - 2 * pow(alpha,2) * r0 * r + 1.0 ),2);
+   denom = max(denom,1e-8);
+   return num / denom;
+}
+
+int main(int argc, char *argv[])
+{
+   // 1. Parse command-line options.
+   const char *mesh_file = "../data/star.mesh";
+   int order = 1;
+   int nc_limit = 1;
+   int max_elems = 1e5;
+   double double_max_elems = double(max_elems);
+   bool visualization = true;
+   double osc_threshold = 1e-3;
+   int enriched_order = 5;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&nc_limit, "-l", "--nc-limit",
+                  "Maximum level of hanging nodes.");
+   args.AddOption(&double_max_elems, "-me", "--max-elems",
+                  "Stop after reaching this many elements.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.AddOption(&osc_threshold, "-e", "--error",
+                  "relative data oscillation threshold.");
+   args.AddOption(&enriched_order, "-eo", "--enriched_order",
+                  "Enriched quadrature order.");
+
+   args.Parse();
+   if (!args.Good())
+   {
+      args.PrintUsage(cout);
+      return 1;
+   }
+   args.PrintOptions(cout);
+
+   max_elems = int(double_max_elems);
+   Mesh mesh(mesh_file, 1, 1);
+
+   // 2. Since a NURBS mesh can currently only be refined uniformly, we need to
+   //    convert it to a piecewise-polynomial curved mesh. First we refine the
+   //    NURBS mesh a bit more and then project the curvature to quadratic Nodes.
+   if (mesh.NURBSext)
+   {
+      for (int i = 0; i < 2; i++)
+      {
+         mesh.UniformRefinement();
+      }
+      mesh.SetCurvature(2);
+   }
+
+   // 3. Define functions and refiner.
+   FunctionCoefficient affine_coeff(affine_function);
+   FunctionCoefficient jump_coeff(jump_function);
+   FunctionCoefficient singular_coeff(singular_function);
+   CoefficientRefiner coeffrefiner(affine_coeff, order);
+
+   // 4. Connect to GLVis.
+   char vishost[] = "localhost";
+   int  visport   = 19916;
+   socketstream sol_sock;
+   if (visualization)
+   {
+      sol_sock.open(vishost, visport);
+   }
+
+   // 5. Define custom integration rule (optional).
+   const IntegrationRule *irs[Geometry::NumGeom];
+   int order_quad = 2*order + enriched_order;
+   for (int i = 0; i < Geometry::NumGeom; ++i)
+   {
+      irs[i] = &(IntRules.Get(i, order_quad));
+   }
+
+   // 6. Apply custom refiner settings.
+   coeffrefiner.SetIntRule(irs);
+   coeffrefiner.SetMaxElements(max_elems);
+   coeffrefiner.SetThreshold(osc_threshold);
+   coeffrefiner.SetNCLimit(nc_limit);
+   coeffrefiner.PrintWarnings();
+
+   // 7. Preprocess mesh to control osc (piecewise-affine function).
+   //    This is mostly just a verification check. The oscillation should
+   //    be zero if the function is mesh-conforming and order > 0.
+   coeffrefiner.PreprocessMesh(mesh);
+
+   mfem::out  << "\n";
+   mfem::out << "Function 0 (affine) \n";
+   mfem::out << "Number of Elements " << mesh.GetNE() << "\n";
+   mfem::out << "Osc error " << coeffrefiner.GetOsc() << "\n";
+
+   // 8. Preprocess mesh to control osc (jump function).
+   coeffrefiner.ResetCoefficient(jump_coeff);
+   coeffrefiner.PreprocessMesh(mesh);
+
+   mfem::out  << "\n";
+   mfem::out << "Function 1 (discontinuous) \n";
+   mfem::out << "Number of Elements " << mesh.GetNE() << "\n";
+   mfem::out << "Osc error " << coeffrefiner.GetOsc() << "\n";
+
+   // 9. Preprocess mesh to control osc (singular function).
+   coeffrefiner.ResetCoefficient(singular_coeff);
+   coeffrefiner.PreprocessMesh(mesh);
+
+   mfem::out  << "\n";
+   mfem::out << "Function 2 (singular) \n";
+   mfem::out << "Number of Elements " << mesh.GetNE() << "\n";
+   mfem::out << "Osc error " << coeffrefiner.GetOsc() << "\n";
+
+   sol_sock.precision(8);
+   sol_sock << "mesh\n" << mesh << flush;
+
+   return 0;
+}

examples/ex30p.cpp

@@ -0,0 +1,241 @@
+//                        MFEM Example 30 - Parallel Version
+//
+// Compile with: make ex30p
+//
+// Sample runs:  mpirun -np 4 ex30p -m ../data/square-disc.mesh -o 1
+//               mpirun -np 4 ex30p -m ../data/square-disc.mesh -o 2
+//               mpirun -np 4 ex30p -m ../data/square-disc.mesh -o 2 -me 1e3
+//               mpirun -np 4 ex30p -m ../data/square-disc-nurbs.mesh -o 2
+//               mpirun -np 4 ex30p -m ../data/star.mesh -o 2 -eo 4
+//               mpirun -np 4 oscp -m ../data/fichera.mesh -o 2 -me 1e4
+//               mpirun -np 4 ex30p -m ../data/disc-nurbs.mesh -o 2
+//               mpirun -np 4 ex30p -m ../data/ball-nurbs.mesh -o 2 -eo 3 -e 1e-2
+//               mpirun -np 4 ex30p -m ../data/star-surf.mesh -o 2
+//               mpirun -np 4 ex30p -m ../data/square-disc-surf.mesh -o 2
+//               mpirun -np 4 ex30p -m ../data/amr-quad.mesh -l 2
+//
+// Description:  This is an example of adaptive mesh refinement preprocessing
+//               which lowers the data oscillation [1] to a user-defined
+//               relative threshold. There is no PDE being solved.
+//
+//               MFEM's capability to work with both conforming and
+//               nonconforming meshes is demonstrated in example 6. In some
+//               problems, the material data or loading data is not sufficiently
+//               resolved on the initial mesh. This missing fine scale data
+//               reduces the accuracy of the solution as well as the accuracy
+//               of some local error estimators. By preprocessing the mesh
+//               before solving the PDE, many issues can be avoided.
+//
+//               [1] Morin, P., Nochetto, R. H., & Siebert, K. G. (2000).
+//                   Data oscillation and convergence of adaptive FEM. SIAM
+//                   Journal on Numerical Analysis, 38(2), 466-488.
+//
+//               [2] Mitchell, W. F. (2013). A collection of 2D elliptic
+//                   problems for testing adaptive grid refinement algorithms.
+//                   Applied mathematics and computation, 220, 350-364.
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+// Piecewise-affine function which is sometimes mesh-conforming
+double affine_function(const Vector &p)
+{
+   double x = p(0), y = p(1);
+   if (x < 0.0)
+   {
+      return 1.0 + x + y;
+   }
+   else
+   {
+      return 1.0;
+   }
+}
+
+// Piecewise-constant function which is never mesh-conforming
+double jump_function(const Vector &p)
+{
+   if (p.Normlp(2.0) > 0.4 && p.Normlp(2.0) < 0.6) { return 1.0; }
+   return 5.0;
+}
+
+// Singular function derived from the Laplacian of the "steep wavefront"
+// problem in [2].
+double singular_function(const Vector &p)
+{
+   double x = p(0), y = p(1);
+   double alpha = 1000.0;
+   double xc = 0.75, yc = 0.5;
+   double r0 = 0.7;
+   double r = sqrt(pow(x - xc,2.0) + pow(y - yc,2.0));
+   double num = - ( alpha - pow(alpha,3) * (pow(r,2) - pow(r0,2)) );
+   double denom = pow(r * ( pow(alpha,2) * pow(r0,2) + pow(alpha,2) * pow(r,2) \
+                            - 2 * pow(alpha,2) * r0 * r + 1.0 ),2);
+   denom = max(denom,1e-8);
+   return num / denom;
+}
+
+int main(int argc, char *argv[])
+{
+   // 0. 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);
+
+   // 1. Parse command-line options.
+   const char *mesh_file = "../data/star.mesh";
+   int order = 1;
+   int nc_limit = 1;
+   int max_elems = 1e5;
+   double double_max_elems = double(max_elems);
+   bool visualization = true;
+   bool nc_simplices = true;
+   double osc_threshold = 1e-3;
+   int enriched_order = 5;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&nc_limit, "-l", "--nc-limit",
+                  "Maximum level of hanging nodes.");
+   args.AddOption(&double_max_elems, "-me", "--max-elems",
+                  "Stop after reaching this many elements.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.AddOption(&osc_threshold, "-e", "--error",
+                  "relative data oscillation threshold.");
+   args.AddOption(&enriched_order, "-eo", "--enriched_order",
+                  "Enriched quadrature order.");
+   args.AddOption(&nc_simplices, "-ns", "--nonconforming-simplices",
+                  "-cs", "--conforming-simplices",
+                  "For simplicial meshes, enable/disable nonconforming"
+                  " refinement");
+
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0)
+      {
+         args.PrintUsage(cout);
+      }
+      MPI_Finalize();
+      return 1;
+   }
+   if (myid == 0)
+   {
+      args.PrintOptions(cout);
+   }
+
+   max_elems = int(double_max_elems);
+   Mesh mesh(mesh_file, 1, 1);
+
+   // 2. Since a NURBS mesh can currently only be refined uniformly, we need to
+   //    convert it to a piecewise-polynomial curved mesh. First we refine the
+   //    NURBS mesh a bit more and then project the curvature to quadratic Nodes.
+   if (mesh.NURBSext)
+   {
+      for (int i = 0; i < 2; i++)
+      {
+         mesh.UniformRefinement();
+      }
+      mesh.SetCurvature(2);
+   }
+
+   // 3. Make sure the mesh is in the non-conforming mode to enable local
+   //    refinement of quadrilaterals/hexahedra. Simplices can be refined
+   //    either in conforming or in non-conforming mode. The conforming
+   //    mode however does not support dynamic partitioning.
+   mesh.EnsureNCMesh(nc_simplices);
+
+   // 4. Define a parallel mesh by partitioning the serial mesh.
+   //    Once the parallel mesh is defined, the serial mesh can be deleted.
+   ParMesh pmesh(MPI_COMM_WORLD, mesh);
+   mesh.Clear();
+
+   // 5. Define functions and refiner.
+   FunctionCoefficient affine_coeff(affine_function);
+   FunctionCoefficient jump_coeff(jump_function);
+   FunctionCoefficient singular_coeff(singular_function);
+   CoefficientRefiner coeffrefiner(affine_coeff,order);
+
+   // 6. Connect to GLVis.
+   char vishost[] = "localhost";
+   int  visport   = 19916;
+   socketstream sol_sock;
+   if (visualization)
+   {
+      sol_sock.open(vishost, visport);
+   }
+
+   // 7. Define custom integration rule (optional).
+   const IntegrationRule *irs[Geometry::NumGeom];
+   int order_quad = 2*order + enriched_order;
+   for (int i=0; i < Geometry::NumGeom; ++i)
+   {
+      irs[i] = &(IntRules.Get(i, order_quad));
+   }
+
+   // 8. Apply custom refiner settings.
+   coeffrefiner.SetIntRule(irs);
+   coeffrefiner.SetMaxElements(max_elems);
+   coeffrefiner.SetThreshold(osc_threshold);
+   coeffrefiner.SetNCLimit(nc_limit);
+   coeffrefiner.PrintWarnings();
+
+   // 9. Preprocess mesh to control osc (piecewise-affine function).
+   //    This is mostly just a verification check. The oscillation should
+   //    be zero if the function is mesh-conforming and order > 0.
+   coeffrefiner.PreprocessMesh(pmesh);
+
+   int globalNE = pmesh.GetGlobalNE();
+   double osc = coeffrefiner.GetOsc();
+   if (myid == 0)
+   {
+      mfem::out  << "\n";
+      mfem::out << "Function 0 (affine) \n";
+      mfem::out << "Number of Elements " << globalNE << "\n";
+      mfem::out << "Osc error " << osc << "\n";
+   }
+
+   // 10. Preprocess mesh to control osc (jump function).
+   coeffrefiner.ResetCoefficient(jump_coeff);
+   coeffrefiner.PreprocessMesh(pmesh);
+
+   globalNE = pmesh.GetGlobalNE();
+   osc = coeffrefiner.GetOsc();
+   if (myid == 0)
+   {
+      mfem::out  << "\n";
+      mfem::out << "Function 1 (discontinuous) \n";
+      mfem::out << "Number of Elements " << globalNE << "\n";
+      mfem::out << "Osc error " << osc << "\n";
+   }
+
+   // 11. Preprocess mesh to control osc (singular function).
+   coeffrefiner.ResetCoefficient(singular_coeff);
+   coeffrefiner.PreprocessMesh(pmesh);
+
+   globalNE = pmesh.GetGlobalNE();
+   osc = coeffrefiner.GetOsc();
+   if (myid == 0)
+   {
+      mfem::out  << "\n";
+      mfem::out << "Function 2 (singular) \n";
+      mfem::out << "Number of Elements " << globalNE << "\n";
+      mfem::out << "Osc error " << osc << "\n";
+   }
+
+   sol_sock.precision(8);
+   sol_sock << "parallel " << num_procs << " " << myid << "\n";
+   sol_sock << "mesh\n" << pmesh << flush;
+
+   MPI_Finalize();
+   return 0;
+}

examples/ex3p_es.cpp

@@ -0,0 +1,411 @@
+//                       MFEM Example 3 - Parallel Version
+//
+// Compile with: make ex3p
+//
+// Sample runs:  mpirun -np 4 ex3p -m ../data/star.mesh
+//               mpirun -np 4 ex3p -m ../data/square-disc.mesh -o 2
+//               mpirun -np 4 ex3p -m ../data/beam-tet.mesh
+//               mpirun -np 4 ex3p -m ../data/beam-hex.mesh
+//               mpirun -np 4 ex3p -m ../data/escher.mesh
+//               mpirun -np 4 ex3p -m ../data/fichera.mesh
+//               mpirun -np 4 ex3p -m ../data/fichera-q2.vtk
+//               mpirun -np 4 ex3p -m ../data/fichera-q3.mesh
+//               mpirun -np 4 ex3p -m ../data/square-disc-nurbs.mesh
+//               mpirun -np 4 ex3p -m ../data/beam-hex-nurbs.mesh
+//               mpirun -np 4 ex3p -m ../data/amr-quad.mesh -o 2
+//               mpirun -np 4 ex3p -m ../data/amr-hex.mesh
+//               mpirun -np 4 ex3p -m ../data/star-surf.mesh -o 2
+//               mpirun -np 4 ex3p -m ../data/mobius-strip.mesh -o 2 -f 0.1
+//               mpirun -np 4 ex3p -m ../data/klein-bottle.mesh -o 2 -f 0.1
+//
+// Description:  This example code solves a simple electromagnetic diffusion
+//               problem corresponding to the second order definite Maxwell
+//               equation curl curl E + E = f with boundary condition
+//               E x n = <given tangential field>. Here, we use a given exact
+//               solution E and compute the corresponding r.h.s. f.
+//               We discretize with Nedelec finite elements in 2D or 3D.
+//
+//               The example demonstrates the use of H(curl) finite element
+//               spaces with the curl-curl and the (vector finite element) mass
+//               bilinear form, as well as the computation of discretization
+//               error when the exact solution is known. Static condensation is
+//               also illustrated.
+//
+//               We recommend viewing examples 1-2 before viewing this example.
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+// Exact solution, E, and r.h.s., f. See below for implementation.
+//void E_exact(const Vector &, Vector &);
+//void f_exact(const Vector &, Vector &);
+//double freq = 1.0, kappa;
+void f_const(const Vector &, Vector &);
+
+int dim;
+
+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/beam-tet.mesh";
+   int order = 1;
+   int rs = -1;
+   int rp = 2;
+   int ra = 0;
+   int bt = EntitySets::INVALID;
+   const char *bs = "Origin";
+   bool static_cond = false;
+   bool visualization = 1;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   /*
+   args.AddOption(&freq, "-f", "--frequency", "Set the frequency for the exact"
+                  " solution.");
+   */
+   args.AddOption(&rs, "-rs", "--refine-serial",
+                  "Number of serial refinement levels");
+   args.AddOption(&rp, "-rp", "--refine-parallel",
+                  "Number of parallel refinement levels");
+   args.AddOption(&ra, "-ra", "--refine-adaptive",
+                  "Number of adaptive refinement levels");
+   args.AddOption(&bt, "-bt", "--bc-entity-type",
+                  "");
+   args.AddOption(&bs, "-bs", "--bc-entity-set-name",
+                  "");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0)
+      {
+         args.PrintUsage(cout);
+      }
+      MPI_Finalize();
+      return 1;
+   }
+   if (myid == 0)
+   {
+      args.PrintOptions(cout);
+   }
+   // kappa = freq * M_PI;
+
+   // 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);
+   dim = mesh->Dimension();
+   int sdim = mesh->SpaceDimension();
+
+   // 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.
+   {
+      int ref_levels = ( rs >= 0 ) ? rs :
+                       (int)floor(log(1000./mesh->GetNE())/log(2.)/dim);
+      for (int l = 0; l < ref_levels; l++)
+      {
+         if ( myid == 0 ) { cout << "Uniform refinement in serial..."; }
+         mesh->UniformRefinement();
+      }
+      MPI_Barrier(MPI_COMM_WORLD);
+      if ( myid == 0 && rs > 0 ) { cout << "Done" << endl; }
+   }
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+   /*
+     At this point we have a serial mesh containing an EntitySets
+     object which stores the current node/edge/face/element indices
+     for each entity in each set.  This data is duplicated on each MPI
+     rank.
+    */
+   if ( ra > 0 )
+   {
+      cout << "calling EnsureNCMesh" << endl;
+      mesh->EnsureNCMesh();
+      cout << "back from EnsureNCMesh" << endl;
+   }
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+   /*
+     We now have an NCEntitySets object which stores the node indices
+     describing each enity in each node/edge/face set and the element
+     indices for the elements in each element set.  This data is
+     duplicated on each MPI rank.
+    */
+
+   // 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
+   //    spaces on them.
+   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
+   delete mesh;
+   if ( pmesh->pent_sets )
+   {
+      cout << "pmesh->pent_sets is non NULL" << endl;
+      pmesh->pent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "pmesh->pent_sets is NULL" << endl;
+   }
+   {
+      int par_ref_levels = rp;
+      for (int l = 0; l < par_ref_levels; l++)
+      {
+         if ( myid == 0 ) { cout << "Uniform refinement in parallel..."; }
+         pmesh->UniformRefinement();
+      }
+      MPI_Barrier(MPI_COMM_WORLD);
+      if ( myid == 0 && rs > 0 ) { cout << "Done" << endl; }
+   }
+   pmesh->ReorientTetMesh();
+   pmesh->ent_sets->PrintSetInfo(cout);
+
+   for (int l = 0; l < ra; l++)
+   {
+      pmesh->RandomRefinement(0.2);
+   }
+   if ( ra > 0 )
+   {
+      if ( pmesh->pent_sets )
+      {
+         cout << "pmesh->pent_sets is non NULL post random refinement" << endl;
+         pmesh->pent_sets->PrintSetInfo(cout);
+      }
+      else
+      {
+         cout << "pmesh->pent_sets is NULL post random refinement" << endl;
+      }
+   }
+
+   // 6. 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);
+   ParFiniteElementSpace *fespace = new ParFiniteElementSpace(pmesh, fec);
+   HYPRE_Int size = fespace->GlobalTrueVSize();
+   if (myid == 0)
+   {
+      cout << "Number of finite element unknowns: " << size << endl;
+   }
+
+   // 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.
+   Array<int> ess_tdof_list;
+   if ( bt == EntitySets::INVALID )
+   {
+      if (pmesh->bdr_attributes.Size())
+      {
+         Array<int> ess_bdr(pmesh->bdr_attributes.Max());
+         ess_bdr = 1;
+         fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      }
+   }
+   else
+   {
+      fespace->GetEssentialTrueDofs((EntitySets::EntityType)bt, bs,
+                                    ess_tdof_list);
+   }
+   if (myid == 0)
+   {
+      cout << "Number of Dirichlet dofs: " << ess_tdof_list.Size() << endl;
+   }
+
+
+   // 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.
+   VectorFunctionCoefficient f(sdim, f_const);
+   ParLinearForm *b = new ParLinearForm(fespace);
+   b->AddDomainIntegrator(new VectorFEDomainLFIntegrator(f));
+   b->Assemble();
+
+   // 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 edges will be used
+   //    when eliminating the non-homogeneous boundary condition to modify the
+   //    r.h.s. vector b.
+   ParGridFunction x(fespace);
+   // VectorFunctionCoefficient E(sdim, E_exact);
+   // x.ProjectCoefficient(E);
+   x = 0.0;
+
+   // 10. Set up the parallel bilinear form corresponding to the EM diffusion
+   //     operator curl muinv curl + sigma I, by adding the curl-curl and the
+   //     mass domain integrators.
+   Coefficient *muinv = new ConstantCoefficient(1.0);
+   Coefficient *sigma = new ConstantCoefficient(1.0);
+   ParBilinearForm *a = new ParBilinearForm(fespace);
+   a->AddDomainIntegrator(new CurlCurlIntegrator(*muinv));
+   a->AddDomainIntegrator(new VectorFEMassIntegrator(*sigma));
+
+   // 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, etc.
+   if (static_cond) { a->EnableStaticCondensation(); }
+   a->Assemble();
+
+   HypreParMatrix A;
+   Vector B, X;
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);
+
+   if (myid == 0)
+   {
+      cout << "Size of linear system: " << A.GetGlobalNumRows() << endl;
+   }
+
+   // 12. Define and apply a parallel PCG solver for AX=B with the AMS
+   //     preconditioner from hypre.
+   ParFiniteElementSpace *prec_fespace =
+      (a->StaticCondensationIsEnabled() ? a->SCParFESpace() : fespace);
+   HypreSolver *ams = new HypreAMS(A, prec_fespace);
+   HyprePCG *pcg = new HyprePCG(A);
+   pcg->SetTol(1e-12);
+   pcg->SetMaxIter(500);
+   pcg->SetPrintLevel(2);
+   pcg->SetPreconditioner(*ams);
+   pcg->Mult(B, X);
+
+   // 13. Recover the parallel grid function corresponding to X. This is the
+   //     local finite element solution on each processor.
+   a->RecoverFEMSolution(X, *b, x);
+
+   /*
+   // 14. Compute and print the L^2 norm of the error.
+   {
+      double err = x.ComputeL2Error(E);
+      if (myid == 0)
+      {
+         cout << "\n|| E_h - E ||_{L^2} = " << err << '\n' << 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);
+
+      ofstream sol_ofs(sol_name.str().c_str());
+      sol_ofs.precision(8);
+      x.Save(sol_ofs);
+   }
+
+   // 16. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream sol_sock(vishost, visport);
+      sol_sock << "parallel " << num_procs << " " << myid << "\n";
+      sol_sock.precision(8);
+      sol_sock << "solution\n" << *pmesh << x << flush;
+   }
+
+   // 17. Free the used memory.
+   delete pcg;
+   delete ams;
+   delete a;
+   delete sigma;
+   delete muinv;
+   delete b;
+   delete fespace;
+   delete fec;
+   delete pmesh;
+
+   MPI_Finalize();
+
+   return 0;
+}
+
+/*
+void E_exact(const Vector &x, Vector &E)
+{
+   if (dim == 3)
+   {
+      E(0) = sin(kappa * x(1));
+      E(1) = sin(kappa * x(2));
+      E(2) = sin(kappa * x(0));
+   }
+   else
+   {
+      E(0) = sin(kappa * x(1));
+      E(1) = sin(kappa * x(0));
+      if (x.Size() == 3) { E(2) = 0.0; }
+   }
+}
+
+void f_exact(const Vector &x, Vector &f)
+{
+   if (dim == 3)
+   {
+      f(0) = (1. + kappa * kappa) * sin(kappa * x(1));
+      f(1) = (1. + kappa * kappa) * sin(kappa * x(2));
+      f(2) = (1. + kappa * kappa) * sin(kappa * x(0));
+   }
+   else
+   {
+      f(0) = (1. + kappa * kappa) * sin(kappa * x(1));
+      f(1) = (1. + kappa * kappa) * sin(kappa * x(0));
+      if (x.Size() == 3) { f(2) = 0.0; }
+   }
+}
+*/
+void f_const(const Vector &x, Vector &f)
+{
+   if (dim == 3)
+   {
+      f(0) = 1.0;
+      f(1) = 1.0;
+      f(2) = 1.0;
+   }
+   else
+   {
+      f(0) = 1.0;
+      f(1) = 1.0;
+      if (x.Size() == 3) { f(2) = 0.0; }
+   }
+}

examples/ex4p_es.cpp

@@ -0,0 +1,438 @@
+//                       MFEM Example 4 - Parallel Version
+//
+// Compile with: make ex4p
+//
+// Sample runs:  mpirun -np 4 ex4p -m ../data/square-disc.mesh
+//               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/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
+//               mpirun -np 4 ex4p -m ../data/fichera-q3.mesh -o 2 -sc
+//               mpirun -np 4 ex4p -m ../data/square-disc-nurbs.mesh -o 3
+//               mpirun -np 4 ex4p -m ../data/beam-hex-nurbs.mesh -o 3
+//               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 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
+//
+// 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
+//               = <given normal field>. Here, we use a given exact solution F
+//               and compute the corresponding r.h.s. f.  We discretize with
+//               Raviart-Thomas finite elements.
+//
+//               The example demonstrates the use of H(div) finite element
+//               spaces with the grad-div and H(div) vector finite element mass
+//               bilinear form, as well as the computation of discretization
+//               error when the exact solution is known. Bilinear form
+//               hybridization and static condensation are also illustrated.
+//
+//               We recommend viewing examples 1-3 before viewing this example.
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+// Exact solution, F, and r.h.s., f. See below for implementation.
+//void F_exact(const Vector &, Vector &);
+//void f_exact(const Vector &, Vector &);
+//double freq = 1.0, kappa;
+void f_const(const Vector &, Vector &);
+
+int dim;
+
+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 order = 1;
+   int rs = -1;
+   int rp = 2;
+   int ra = 0;
+   int bt = EntitySets::INVALID;
+   const char *bs = "Origin";
+   bool set_bc = true;
+   bool static_cond = false;
+   bool hybridization = false;
+   bool visualization = 1;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&set_bc, "-bc", "--impose-bc", "-no-bc", "--dont-impose-bc",
+                  "Impose or not essential boundary conditions.");
+   args.AddOption(&rs, "-rs", "--refine-serial",
+                  "Number of serial refinement levels");
+   args.AddOption(&rp, "-rp", "--refine-parallel",
+                  "Number of parallel refinement levels");
+   args.AddOption(&ra, "-ra", "--refine-adaptive",
+                  "Number of adaptive refinement levels");
+   args.AddOption(&bt, "-bt", "--bc-entity-type",
+                  "");
+   args.AddOption(&bs, "-bs", "--bc-entity-set-name",
+                  "");
+   // args.AddOption(&freq, "-f", "--frequency", "Set the frequency for the exact"
+   //                 " solution.");
+   args.AddOption(&static_cond, "-sc", "--static-condensation", "-no-sc",
+                  "--no-static-condensation", "Enable static condensation.");
+   args.AddOption(&hybridization, "-hb", "--hybridization", "-no-hb",
+                  "--no-hybridization", "Enable hybridization.");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0)
+      {
+         args.PrintUsage(cout);
+      }
+      MPI_Finalize();
+      return 1;
+   }
+   if (myid == 0)
+   {
+      args.PrintOptions(cout);
+   }
+   // kappa = freq * M_PI;
+
+   // 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);
+   dim = mesh->Dimension();
+   int sdim = mesh->SpaceDimension();
+
+   // 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.
+   {
+      int ref_levels = ( rs >= 0 ) ? rs :
+                       (int)floor(log(1000./mesh->GetNE())/log(2.)/dim);
+      for (int l = 0; l < ref_levels; l++)
+      {
+         if ( myid == 0 ) { cout << "Uniform refinement in serial..."; }
+         mesh->UniformRefinement();
+      }
+      MPI_Barrier(MPI_COMM_WORLD);
+      if ( myid == 0 && rs > 0 ) { cout << "Done" << endl; }
+   }
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+   /*
+     At this point we have a serial mesh containing an EntitySets
+     object which stores the current node/edge/face/element indices
+     for each entity in each set.  This data is duplicated on each MPI
+     rank.
+    */
+   if ( ra > 0 )
+   {
+      cout << "calling EnsureNCMesh" << endl;
+      mesh->EnsureNCMesh();
+      cout << "back from EnsureNCMesh" << endl;
+   }
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+   /*
+     We now have an NCEntitySets object which stores the node indices
+     describing each enity in each node/edge/face set and the element
+     indices for the elements in each element set.  This data is
+     duplicated on each MPI rank.
+    */
+
+   // 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
+   //    spaces on them (this is needed in the ADS solver below).
+   ParMesh *pmesh = new ParMesh(MPI_COMM_WORLD, *mesh);
+   delete mesh;
+   if ( pmesh->pent_sets )
+   {
+      cout << "pmesh->pent_sets is non NULL" << endl;
+      pmesh->pent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "pmesh->pent_sets is NULL" << endl;
+   }
+   {
+      int par_ref_levels = rp;
+      for (int l = 0; l < par_ref_levels; l++)
+      {
+         if ( myid == 0 ) { cout << "Uniform refinement in parallel..."; }
+         pmesh->UniformRefinement();
+      }
+   }
+   pmesh->ReorientTetMesh();
+
+   for (int l = 0; l < ra; l++)
+   {
+      pmesh->RandomRefinement(0.2);
+   }
+   if ( ra > 0 )
+   {
+      if ( pmesh->pent_sets )
+      {
+         cout << "pmesh->pent_sets is non NULL post random refinement" << endl;
+         pmesh->pent_sets->PrintSetInfo(cout);
+      }
+      else
+      {
+         cout << "pmesh->pent_sets is NULL post random refinement" << endl;
+      }
+   }
+
+   // 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);
+   HYPRE_Int size = fespace->GlobalTrueVSize();
+   if (myid == 0)
+   {
+      cout << "Number of finite element unknowns: " << size << endl;
+   }
+
+   // 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.
+   Array<int> ess_tdof_list;
+   if ( bt == EntitySets::INVALID )
+   {
+      if (pmesh->bdr_attributes.Size())
+      {
+         Array<int> ess_bdr(pmesh->bdr_attributes.Max());
+         ess_bdr = set_bc ? 1 : 0;
+         fespace->GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      }
+   }
+   else
+   {
+      fespace->GetEssentialTrueDofs((EntitySets::EntityType)bt, bs,
+                                    ess_tdof_list);
+   }
+   if (myid == 0)
+   {
+      cout << "Number of Dirichlet dofs: " << ess_tdof_list.Size() << endl;
+   }
+
+   // 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.
+   VectorFunctionCoefficient f(sdim, f_const);
+   ParLinearForm *b = new ParLinearForm(fespace);
+   b->AddDomainIntegrator(new VectorFEDomainLFIntegrator(f));
+   b->Assemble();
+
+   // 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
+   //    r.h.s. vector b.
+   ParGridFunction x(fespace);
+   // VectorFunctionCoefficient F(sdim, F_exact);
+   // x.ProjectCoefficient(F);
+   x = 0.0;
+
+   // 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);
+   a->AddDomainIntegrator(new DivDivIntegrator(*alpha));
+   a->AddDomainIntegrator(new VectorFEMassIntegrator(*beta));
+
+   // 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,
+   //     hybridization, etc.
+   FiniteElementCollection *hfec = NULL;
+   ParFiniteElementSpace *hfes = NULL;
+   if (static_cond)
+   {
+      a->EnableStaticCondensation();
+   }
+   else if (hybridization)
+   {
+      hfec = new DG_Interface_FECollection(order-1, dim);
+      hfes = new ParFiniteElementSpace(pmesh, hfec);
+      a->EnableHybridization(hfes, new NormalTraceJumpIntegrator(),
+                             ess_tdof_list);
+   }
+   a->Assemble();
+
+   HypreParMatrix A;
+   Vector B, X;
+   a->FormLinearSystem(ess_tdof_list, x, *b, A, X, B);
+
+   HYPRE_Int glob_size = A.GetGlobalNumRows();
+   if (myid == 0)
+   {
+      cout << "Size of linear system: " << glob_size << endl;
+   }
+
+   // 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.
+   HypreSolver *prec = NULL;
+   CGSolver *pcg = new CGSolver(A.GetComm());
+   pcg->SetOperator(A);
+   pcg->SetRelTol(1e-12);
+   pcg->SetMaxIter(500);
+   pcg->SetPrintLevel(1);
+   if (hybridization) { prec = new HypreBoomerAMG(A); }
+   else
+   {
+      ParFiniteElementSpace *prec_fespace =
+         (a->StaticCondensationIsEnabled() ? a->SCParFESpace() : fespace);
+      if (dim == 2)   { prec = new HypreAMS(A, prec_fespace); }
+      else            { prec = new HypreADS(A, prec_fespace); }
+   }
+   pcg->SetPreconditioner(*prec);
+   pcg->Mult(B, X);
+
+   // 13. Recover the parallel grid function corresponding to X. This is the
+   //     local finite element solution on each processor.
+   a->RecoverFEMSolution(X, *b, x);
+   /*
+   // 14. Compute and print the L^2 norm of the error.
+   {
+      double err = x.ComputeL2Error(F);
+      if (myid == 0)
+      {
+         cout << "\n|| F_h - F ||_{L^2} = " << err << '\n' << 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);
+
+      ofstream sol_ofs(sol_name.str().c_str());
+      sol_ofs.precision(8);
+      x.Save(sol_ofs);
+   }
+
+   // 16. Send the solution by socket to a GLVis server.
+   if (visualization)
+   {
+      char vishost[] = "localhost";
+      int  visport   = 19916;
+      socketstream sol_sock(vishost, visport);
+      sol_sock << "parallel " << num_procs << " " << myid << "\n";
+      sol_sock.precision(8);
+      sol_sock << "solution\n" << *pmesh << x << flush;
+   }
+
+   // 17. Free the used memory.
+   delete pcg;
+   delete prec;
+   delete hfes;
+   delete hfec;
+   delete a;
+   delete alpha;
+   delete beta;
+   delete b;
+   delete fespace;
+   delete fec;
+   delete pmesh;
+
+   MPI_Finalize();
+
+   return 0;
+}
+
+/*
+// The exact solution (for non-surface meshes)
+void F_exact(const Vector &p, Vector &F)
+{
+   int dim = p.Size();
+
+   double x = p(0);
+   double y = p(1);
+   // double z = (dim == 3) ? p(2) : 0.0;
+
+   F(0) = cos(kappa*x)*sin(kappa*y);
+   F(1) = cos(kappa*y)*sin(kappa*x);
+   if (dim == 3)
+   {
+      F(2) = 0.0;
+   }
+}
+
+// The right hand side
+void f_exact(const Vector &p, Vector &f)
+{
+   int dim = p.Size();
+
+   double x = p(0);
+   double y = p(1);
+   // double z = (dim == 3) ? p(2) : 0.0;
+
+   double temp = 1 + 2*kappa*kappa;
+
+   f(0) = temp*cos(kappa*x)*sin(kappa*y);
+   f(1) = temp*cos(kappa*y)*sin(kappa*x);
+   if (dim == 3)
+   {
+      f(2) = 0;
+   }
+}
+*/
+void f_const(const Vector &x, Vector &f)
+{
+   if (dim == 3)
+   {
+      f(0) = 1.0;
+      f(1) = 1.0;
+      f(2) = 1.0;
+   }
+   else
+   {
+      f(0) = 1.0;
+      f(1) = 1.0;
+      if (x.Size() == 3) { f(2) = 0.0; }
+   }
+}

examples/ex6p_es.cpp

@@ -0,0 +1,325 @@
+//                       MFEM Example 6 - Parallel Version
+//
+// Compile with: make ex6p
+//
+// Sample runs:  mpirun -np 4 ex6p -m ../data/square-disc.mesh -o 1
+//               mpirun -np 4 ex6p -m ../data/square-disc.mesh -o 2
+//               mpirun -np 4 ex6p -m ../data/square-disc-nurbs.mesh -o 2
+//               mpirun -np 4 ex6p -m ../data/star.mesh -o 3
+//               mpirun -np 4 ex6p -m ../data/escher.mesh -o 2
+//               mpirun -np 4 ex6p -m ../data/fichera.mesh -o 2
+//               mpirun -np 4 ex6p -m ../data/disc-nurbs.mesh -o 2
+//               mpirun -np 4 ex6p -m ../data/ball-nurbs.mesh
+//               mpirun -np 4 ex6p -m ../data/pipe-nurbs.mesh
+//               mpirun -np 4 ex6p -m ../data/star-surf.mesh -o 2
+//               mpirun -np 4 ex6p -m ../data/square-disc-surf.mesh -o 2
+//               mpirun -np 4 ex6p -m ../data/amr-quad.mesh
+//
+// Description:  This is a version of Example 1 with a simple adaptive mesh
+//               refinement loop. The problem being solved is again the Laplace
+//               equation -Delta u = 1 with homogeneous Dirichlet boundary
+//               conditions. The problem is solved on a sequence of meshes which
+//               are locally refined in a conforming (triangles, tetrahedrons)
+//               or non-conforming (quadrilaterals, hexahedra) manner according
+//               to a simple ZZ error estimator.
+//
+//               The example demonstrates MFEM's capability to work with both
+//               conforming and nonconforming refinements, in 2D and 3D, on
+//               linear, curved and surface meshes. Interpolation of functions
+//               from coarse to fine meshes, as well as persistent GLVis
+//               visualization are also illustrated.
+//
+//               We recommend viewing Example 1 before viewing this example.
+
+#include "mfem.hpp"
+#include <fstream>
+#include <iostream>
+
+using namespace std;
+using namespace mfem;
+
+static int max_dofs = 100000;
+
+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 = "./star-set.mesh";
+   int order = 1;
+   int bt = EntitySets::INVALID;
+   const char *bs = "";
+   bool visualization = true;
+
+   OptionsParser args(argc, argv);
+   args.AddOption(&mesh_file, "-m", "--mesh",
+                  "Mesh file to use.");
+   args.AddOption(&order, "-o", "--order",
+                  "Finite element order (polynomial degree).");
+   args.AddOption(&max_dofs, "-md", "--max-dofs",
+                  "Maximum number of degrees of freedom.");
+   args.AddOption(&bt, "-bt", "--bc-entity-type",
+                  "");
+   args.AddOption(&bs, "-bs", "--bc-entity-set-name",
+                  "");
+   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
+                  "--no-visualization",
+                  "Enable or disable GLVis visualization.");
+   args.Parse();
+   if (!args.Good())
+   {
+      if (myid == 0)
+      {
+         args.PrintUsage(cout);
+      }
+      MPI_Finalize();
+      return 1;
+   }
+   if (myid == 0)
+   {
+      args.PrintOptions(cout);
+   }
+
+   // 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();
+   int sdim = mesh->SpaceDimension();
+
+   // 4. Refine the serial mesh on all processors to increase the resolution.
+   //    Also project a NURBS mesh to a piecewise-quadratic curved mesh. Make
+   //    sure that the mesh is non-conforming.
+   if (mesh->NURBSext)
+   {
+      mesh->UniformRefinement();
+      mesh->SetCurvature(2);
+   }
+   mesh->EnsureNCMesh();
+   if ( mesh->ent_sets )
+   {
+      cout << "mesh->ent_sets is non NULL" << endl;
+      mesh->ent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "mesh->ent_sets is NULL" << endl;
+   }
+
+   // 5. Define a parallel mesh by partitioning the serial mesh.
+   //    Once the parallel mesh is defined, the serial mesh can be deleted.
+   ParMesh pmesh(MPI_COMM_WORLD, *mesh);
+   delete mesh;
+   if ( pmesh.pent_sets )
+   {
+      cout << "pmesh->pent_sets is non NULL" << endl;
+      pmesh.pent_sets->PrintSetInfo(cout);
+   }
+   else
+   {
+      cout << "pmesh->pent_sets is NULL" << endl;
+   }
+
+   // 6. Define a finite element space on the mesh. The polynomial order is
+   //    one (linear) by default, but this can be changed on the command line.
+   H1_FECollection fec(order, dim);
+   ParFiniteElementSpace fespace(&pmesh, &fec);
+
+   Array<int> ess_tdof_list;
+   if ( bt == EntitySets::INVALID )
+   {
+      if (pmesh.bdr_attributes.Size())
+      {
+         Array<int> ess_bdr(pmesh.bdr_attributes.Max());
+         ess_bdr = 1;
+         fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+      }
+   }
+   else
+   {
+      fespace.GetEssentialTrueDofs((EntitySets::EntityType)bt, bs,
+                                   ess_tdof_list);
+   }
+
+   // 7. As in Example 1p, we set up bilinear and linear forms corresponding to
+   //    the Laplace problem -\Delta u = 1. We don't assemble the discrete
+   //    problem yet, this will be done in the main loop.
+   ParBilinearForm a(&fespace);
+   ParLinearForm b(&fespace);
+
+   ConstantCoefficient one(1.0);
+
+   BilinearFormIntegrator *integ = new DiffusionIntegrator(one);
+   a.AddDomainIntegrator(integ);
+   b.AddDomainIntegrator(new DomainLFIntegrator(one));
+
+   // 8. The solution vector x and the associated finite element grid function
+   //    will be maintained over the AMR iterations. We initialize it to zero.
+   ParGridFunction x(&fespace);
+   x = 0;
+
+   // 9. Connect to GLVis.
+   char vishost[] = "localhost";
+   int  visport   = 19916;
+
+   socketstream sout;
+   if (visualization)
+   {
+      sout.open(vishost, visport);
+      if (!sout)
+      {
+         if (myid == 0)
+         {
+            cout << "Unable to connect to GLVis server at "
+                 << vishost << ':' << visport << endl;
+            cout << "GLVis visualization disabled.\n";
+         }
+         visualization = false;
+      }
+
+      sout.precision(8);
+   }
+
+   // 10. Set up an error estimator. Here we use the Zienkiewicz-Zhu estimator
+   //     with L2 projection in the smoothing step to better handle hanging
+   //     nodes and parallel partitioning. We need to supply a space for the
+   //     discontinuous flux (L2) and a space for the smoothed flux (H(div) is
+   //     used here).
+   L2_FECollection flux_fec(order, dim);
+   ParFiniteElementSpace flux_fes(&pmesh, &flux_fec, sdim);
+   RT_FECollection smooth_flux_fec(order-1, dim);
+   ParFiniteElementSpace smooth_flux_fes(&pmesh, &smooth_flux_fec);
+   // Another possible option for the smoothed flux space:
+   // H1_FECollection smooth_flux_fec(order, dim);
+   // ParFiniteElementSpace smooth_flux_fes(&pmesh, &smooth_flux_fec, dim);
+   L2ZienkiewiczZhuEstimator estimator(*integ, x, flux_fes, smooth_flux_fes);
+
+   // 11. A refiner selects and refines elements based on a refinement strategy.
+   //     The strategy here is to refine elements with errors larger than a
+   //     fraction of the maximum element error. Other strategies are possible.
+   //     The refiner will call the given error estimator.
+   ThresholdRefiner refiner(estimator);
+   refiner.SetTotalErrorFraction(0.7);
+
+   // 12. The main AMR loop. In each iteration we solve the problem on the
+   //     current mesh, visualize the solution, and refine the mesh.
+   // const int max_dofs = 100000;
+   for (int it = 0; ; it++)
+   {
+      HYPRE_Int global_dofs = fespace.GlobalTrueVSize();
+      if (myid == 0)
+      {
+         cout << "\nAMR iteration " << it << endl;
+         cout << "Number of unknowns: " << global_dofs << endl;
+      }
+
+      // 13. Assemble the stiffness matrix and the right-hand side. Note that
+      //     MFEM doesn't care at this point that the mesh is nonconforming
+      //     and parallel. The FE space is considered 'cut' along hanging
+      //     edges/faces, and also across processor boundaries.
+      a.Assemble();
+      b.Assemble();
+
+      // 14. Create the parallel linear system: eliminate boundary conditions,
+      //     constrain hanging nodes and nodes across processor boundaries.
+      //     The system will be solved for true (unconstrained/unique) DOFs only.
+      // Array<int> ess_tdof_list;
+      if ( bt == EntitySets::INVALID )
+      {
+         if (pmesh.bdr_attributes.Size())
+         {
+            Array<int> ess_bdr(pmesh.bdr_attributes.Max());
+            ess_bdr = 1;
+            fespace.GetEssentialTrueDofs(ess_bdr, ess_tdof_list);
+         }
+      }
+      else
+      {
+         fespace.GetEssentialTrueDofs((EntitySets::EntityType)bt, bs,
+                                      ess_tdof_list);
+      }
+
+      HypreParMatrix A;
+      Vector B, X;
+      const int copy_interior = 1;
+      a.FormLinearSystem(ess_tdof_list, x, b, A, X, B, copy_interior);
+
+      // 15. Define and apply a parallel PCG solver for AX=B with the BoomerAMG
+      //     preconditioner from hypre.
+      HypreBoomerAMG amg;
+      amg.SetPrintLevel(0);
+      CGSolver pcg(A.GetComm());
+      pcg.SetPreconditioner(amg);
+      pcg.SetOperator(A);
+      pcg.SetRelTol(1e-6);
+      pcg.SetMaxIter(200);
+      pcg.SetPrintLevel(3); // print the first and the last iterations only
+      pcg.Mult(B, X);
+
+      // 16. Extract the parallel grid function corresponding to the finite element
+      //     approximation X. This is the local solution on each processor.
+      a.RecoverFEMSolution(X, b, x);
+
+      // 17. Send the solution by socket to a GLVis server.
+      if (visualization)
+      {
+         sout << "parallel " << num_procs << " " << myid << "\n";
+         sout << "solution\n" << pmesh << x << flush;
+      }
+
+      if (global_dofs > max_dofs)
+      {
+         if (myid == 0)
+         {
+            cout << "Reached the maximum number of dofs. Stop." << endl;
+         }
+         break;
+      }
+
+      // 18. Call the refiner to modify the mesh. The refiner calls the error
+      //     estimator to obtain element errors, then it selects elements to be
+      //     refined and finally it modifies the mesh. The Stop() method can be
+      //     used to determine if a stopping criterion was met.
+      refiner.Apply(pmesh);
+      if (refiner.Stop())
+      {
+         if (myid == 0)
+         {
+            cout << "Stopping criterion satisfied. Stop." << endl;
+         }
+         break;
+      }
+
+      // 19. Update the finite element space (recalculate the number of DOFs,
+      //     etc.) and create a grid function update matrix. Apply the matrix
+      //     to any GridFunctions over the space. In this case, the update
+      //     matrix is an interpolation matrix so the updated GridFunction will
+      //     still represent the same function as before refinement.
+      fespace.Update();
+      x.Update();
+
+      // 20. Load balance the mesh, and update the space and solution. Currently
+      //     available only for nonconforming meshes.
+      if (pmesh.Nonconforming())
+      {
+         pmesh.Rebalance();
+
+         // Update the space and the GridFunction. This time the update matrix
+         // redistributes the GridFunction among the processors.
+         fespace.Update();
+         x.Update();
+      }
+
+      // 21. Inform also the bilinear and linear forms that the space has
+      //     changed.
+      a.Update();
+      b.Update();
+   }
+
+   MPI_Finalize();
+   return 0;
+}

examples/fichera-mixed-set.mesh

@@ -0,0 +1,162 @@
+MFEM mesh v1.0
+
+#
+# MFEM Geometry Types (see mesh/geom.hpp):
+#
+# POINT       = 0
+# SEGMENT     = 1
+# TRIANGLE    = 2
+# SQUARE      = 3
+# TETRAHEDRON = 4
+# CUBE        = 5
+# PRISM       = 6
+#
+
+dimension
+3
+
+elements
+14
+1 4 13 15 21 25
+1 4 12 13 15 21
+1 4 13 21 22 25
+1 4 15 24 21 25
+1 4 13 15 25 16
+1 5  0  1  4  3  9 10 13 12
+1 5  8  9 12 11 17 18 21 20
+1 5  2  3  6  5 11 12 15 14
+1 6  3  4  6 12 13 15
+1 6  4  7  6 13 16 15
+1 6 12 13 21  9 10 18
+1 6 13 22 21 10 19 18
+1 6 11 14 20 12 15 21
+1 6 15 21 24 14 20 23
+
+boundary
+30
+ 1 3  5  6  3  2
+ 2 2  3  6  4
+ 2 2  4  6  7
+ 3 3  3  4  1  0
+ 4 3 11 12  9  8
+ 5 3  2  3 12 11
+ 6 3  0  1 10  9
+ 7 2  9 10 18
+ 7 2 10 19 18
+ 8 3  8  9 18 17
+ 9 3  1  4 13 10
+10 3  4  7 16 13
+11 2 13 16 25
+11 2 13 25 22
+12 3 10 13 22 19
+13 3  7  6 15 16
+14 3  6  5 14 15
+15 3 15 14 23 24
+16 2 16 15 25
+16 2 15 24 25
+17 3  5  2 11 14
+18 3  3  0  9 12
+19 3 11  8 17 20
+20 2 11 20 14
+20 2 14 20 23
+21 3 17 18 21 20
+22 3 18 19 22 21
+23 2 21 22 25
+23 2 21 25 24
+24 3 20 21 24 23
+
+vertices
+26
+3
+0 -1 -1
+1 -1 -1
+-1 0 -1
+0 0 -1
+1 0 -1
+-1 1 -1
+0 1 -1
+1 1 -1
+-1 -1 0
+0 -1 0
+1 -1 0
+-1 0 0
+0 0 0
+1 0 0
+-1 1 0
+0 1 0
+1 1 0
+-1 -1 1
+0 -1 1
+1 -1 1
+-1 0 1
+0 0 1
+1 0 1
+-1 1 1
+0 1 1
+1 1 1
+
+MFEM sets v1.0
+
+vertex_sets
+1
+
+Origin
+1
+12
+
+edge_sets
+2
+
+Axes
+3
+12 13
+12 15
+12 21
+
+Negative Axes
+3
+12 9
+12 11
+12 3
+
+face_sets
+2
+
+Interior Corner
+3
+3 11 12  9 8
+3  2  3 12 11
+3  3  0  9 12
+
+Exterior Corner
+15
+2 13 16 25
+2 13 25 22
+2 16 15 25
+2 15 24 25
+2 21 22 25
+2 21 25 24
+3 10 13 22 19
+3 4 7 16 13
+3 1 4 13 10
+3 7 6 15 16
+3 6 5 14 15
+3 15 14 23 24
+3 20 21 24 23
+3 18 19 22 21
+3 17 18 21 20
+
+element_sets
+3
+
+Interior Corner
+3
+5 6 7
+
+Exterior Corner
+5
+0 1 2 3 4
+
+Steps
+3
+6 8 9

examples/fichera-set.mesh

@@ -0,0 +1,145 @@
+MFEM mesh v1.0
+
+#
+# MFEM Geometry Types (see mesh/geom.hpp):
+#
+# POINT       = 0
+# SEGMENT     = 1
+# TRIANGLE    = 2
+# SQUARE      = 3
+# TETRAHEDRON = 4
+# CUBE        = 5
+#
+
+dimension
+3
+
+elements
+7
+1 5 0 1 4 3 9 10 13 12
+1 5 3 4 7 6 12 13 16 15
+1 5 2 3 6 5 11 12 15 14
+1 5 8 9 12 11 17 18 21 20
+1 5 9 10 13 12 18 19 22 21
+1 5 12 13 16 15 21 22 25 24
+1 5 11 12 15 14 20 21 24 23
+
+boundary
+24
+1 3 5 6 3 2
+2 3 6 7 4 3
+3 3 3 4 1 0
+4 3 11 12 9 8
+5 3 2 3 12 11
+6 3 0 1 10 9
+7 3 9 10 19 18
+8 3 8 9 18 17
+9 3 1 4 13 10
+10 3 4 7 16 13
+11 3 13 16 25 22
+12 3 10 13 22 19
+13 3 7 6 15 16
+14 3 6 5 14 15
+15 3 15 14 23 24
+16 3 16 15 24 25
+17 3 5 2 11 14
+18 3 3 0 9 12
+19 3 11 8 17 20
+20 3 14 11 20 23
+21 3 17 18 21 20
+22 3 18 19 22 21
+23 3 21 22 25 24
+24 3 20 21 24 23
+
+vertices
+26
+3
+0 -1 -1
+1 -1 -1
+-1 0 -1
+0 0 -1
+1 0 -1
+-1 1 -1
+0 1 -1
+1 1 -1
+-1 -1 0
+0 -1 0
+1 -1 0
+-1 0 0
+0 0 0
+1 0 0
+-1 1 0
+0 1 0
+1 1 0
+-1 -1 1
+0 -1 1
+1 -1 1
+-1 0 1
+0 0 1
+1 0 1
+-1 1 1
+0 1 1
+1 1 1
+
+MFEM sets v1.0
+
+vertex_sets
+1
+
+Origin
+1
+12
+
+edge_sets
+2
+
+Axes
+3
+12 13
+12 15
+12 21
+
+Negative Axes
+3
+12 9
+12 11
+12 3
+
+face_sets
+2
+
+Interior Corner
+3
+3 11 12  9 8
+3  2  3 12 11
+3  3  0  9 12
+
+Exterior Corner
+12
+3 13 16 25 22
+3 16 15 24 25
+3 21 22 25 24
+3 10 13 22 19
+3 4 7 16 13
+3 1 4 13 10
+3 7 6 15 16
+3 6 5 14 15
+3 15 14 23 24
+3 20 21 24 23
+3 18 19 22 21
+3 17 18 21 20
+
+element_sets
+3
+
+Interior Corner
+3
+0 2 3
+
+Exterior Corner
+1
+5
+
+Steps
+2
+1 3

examples/ginkgo/CMakeLists.txt

@@ -31,29 +31,31 @@ add_mfem_examples(GINKGO_EXAMPLES_SRCS ${PFX} "" test_ginkgo)
 # which builds the examples and runs:
 #   ctest -R ginkgo
 
-# Command line options for the tests.
-set(EX1_COMMON_OPTS ex1 -m ../data/star.mesh --use_gko_solver)
-set(EX1_TEST_OPTS ${EX9_COMMON_OPTS})
+if (MFEM_ENABLE_TESTING)
+  # Command line options for the tests.
+  set(EX1_COMMON_OPTS ex1 -m ../data/star.mesh --use_gko_solver)
+  set(EX1_TEST_OPTS ${EX9_COMMON_OPTS})
 
-# Add the tests: one test per source file.
-foreach(SRC_FILE ${GINKGO_EXAMPLES_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
-  string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
-  set(TEST_NAME ${PFX}${TEST_NAME})
+  # Add the tests: one test per source file.
+  foreach(SRC_FILE ${GINKGO_EXAMPLES_SRCS})
+    get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
+    string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
+    string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
+    set(TEST_NAME ${PFX}${TEST_NAME})
 
-  set(THIS_TEST_OPTIONS "-no-vis")
-  list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
-  # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
+    set(THIS_TEST_OPTIONS "-no-vis")
+    list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
+    # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
 
-  if (NOT (${TEST_NAME} MATCHES ".*p$"))
-    add_test(NAME ${TEST_NAME}_ser
-      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-  else()
-    add_test(NAME ${TEST_NAME}_np=4
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
-  endif()
-endforeach()
+    if (NOT (${TEST_NAME} MATCHES ".*p$"))
+      add_test(NAME ${TEST_NAME}_ser
+        COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
+    else()
+      add_test(NAME ${TEST_NAME}_np=4
+        COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
+        ${MPIEXEC_PREFLAGS}
+        $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
+        ${MPIEXEC_POSTFLAGS})
+    endif()
+  endforeach()
+endif()

examples/hiop/CMakeLists.txt

@@ -33,31 +33,33 @@ add_mfem_examples(HIOP_EXAMPLES_SRCS ${PFX} "" test_hiop)
 # which builds the examples and runs:
 #   ctest -R hiop
 
-# Command line options for the tests.
-# Example 9:
-set(EX9_COMMON_OPTS -m ../../data/periodic-segment.mesh -p 0 -dt 0.005)
-set(EX9_TEST_OPTS ${EX9_COMMON_OPTS}  -r  2 )
-set(EX9P_TEST_OPTS ${EX9_COMMON_OPTS})
+if (MFEM_ENABLE_TESTING)
+  # Command line options for the tests.
+  # Example 9:
+  set(EX9_COMMON_OPTS -m ../../data/periodic-segment.mesh -p 0 -dt 0.005)
+  set(EX9_TEST_OPTS ${EX9_COMMON_OPTS}  -r  2 )
+  set(EX9P_TEST_OPTS ${EX9_COMMON_OPTS})
 
-# Add the tests: one test per source file.
-foreach(SRC_FILE ${HIOP_EXAMPLES_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
-  string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
-  set(TEST_NAME ${PFX}${TEST_NAME})
+  # Add the tests: one test per source file.
+  foreach(SRC_FILE ${HIOP_EXAMPLES_SRCS})
+    get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
+    string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
+    string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
+    set(TEST_NAME ${PFX}${TEST_NAME})
 
-  set(THIS_TEST_OPTIONS "-no-vis")
-  list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
-  # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
+    set(THIS_TEST_OPTIONS "-no-vis")
+    list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
+    # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
 
-  if (NOT (${TEST_NAME} MATCHES ".*p$"))
-    add_test(NAME ${TEST_NAME}_ser
-      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-  else()
-    add_test(NAME ${TEST_NAME}_np=4
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} 4
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
-  endif()
-endforeach()
+    if (NOT (${TEST_NAME} MATCHES ".*p$"))
+      add_test(NAME ${TEST_NAME}_ser
+        COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
+    else()
+      add_test(NAME ${TEST_NAME}_np=4
+        COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} 4
+        ${MPIEXEC_PREFLAGS}
+        $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
+        ${MPIEXEC_POSTFLAGS})
+    endif()
+  endforeach()
+endif()

examples/makefile

@@ -22,10 +22,10 @@ MFEM_LIB_FILE = mfem_is_not_built
 -include $(CONFIG_MK)
 
 SEQ_EXAMPLES = ex0 ex1 ex2 ex3 ex4 ex5 ex6 ex7 ex8 ex9 ex10 ex14 ex15 ex16 \
-  ex17 ex18 ex19 ex20 ex21 ex22 ex23 ex24 ex25 ex26 ex27 ex28 ex29
+  ex17 ex18 ex19 ex20 ex21 ex22 ex23 ex24 ex25 ex26 ex27 ex28 ex29 ex30
 PAR_EXAMPLES = ex0p 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 ex28p ex29p
+  ex25p ex26p ex27p ex28p ex29p ex30p
 SEQ_DEVICE_EXAMPLES = ex1 ex3 ex4 ex5 ex6 ex9 ex22 ex24 ex25 ex26
 PAR_DEVICE_EXAMPLES = ex1p ex2p ex3p ex4p ex5p ex6p ex7p ex9p ex13p ex22p \
   ex24p ex25p ex26p

examples/petsc/CMakeLists.txt

@@ -94,30 +94,32 @@ if (MFEM_USE_SLEPC)
 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})
-  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)
-  string(REGEX REPLACE "^([^_]+).*" "\\1" TEST_NAME ${TEST_NAME_FULL})
-  set(TEST_NAME_FULL ${PFX}${TEST_NAME_FULL})
-  set(TEST_NAME ${PFX}${TEST_NAME})
-  set(TEST_OPTIONS "-no-vis" ${${TEST_OPTIONS_VAR}})
-  # message(STATUS "${TEST_NAME_FULL} --> ${TEST_NAME} ${TEST_OPTIONS}")
-
-  # All PETSC tests are parallel.
-  if (MFEM_USE_MPI)
-    add_test(NAME ${TEST_NAME_FULL}_np=4
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
+if (MFEM_ENABLE_TESTING)
+  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()
-endforeach()
+
+  foreach(TEST_OPTIONS_VAR ${TEST_OPTIONS_VARS})
+    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)
+    string(REGEX REPLACE "^([^_]+).*" "\\1" TEST_NAME ${TEST_NAME_FULL})
+    set(TEST_NAME_FULL ${PFX}${TEST_NAME_FULL})
+    set(TEST_NAME ${PFX}${TEST_NAME})
+    set(TEST_OPTIONS "-no-vis" ${${TEST_OPTIONS_VAR}})
+    # message(STATUS "${TEST_NAME_FULL} --> ${TEST_NAME} ${TEST_OPTIONS}")
+
+    # All PETSC tests are parallel.
+    if (MFEM_USE_MPI)
+      add_test(NAME ${TEST_NAME_FULL}_np=4
+        COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
+        ${MPIEXEC_PREFLAGS}
+        $<TARGET_FILE:${TEST_NAME}> ${TEST_OPTIONS}
+        ${MPIEXEC_POSTFLAGS})
+    endif()
+  endforeach()
+endif()

examples/pumi/CMakeLists.txt

@@ -37,37 +37,39 @@ add_mfem_examples(PUMI_EXAMPLES_SRCS ${PFX} "" test_pumi)
 # which builds the examples and runs:
 #   ctest -R pumi
 
-# Command line options for the tests.
-# TODO...
+if (MFEM_ENABLE_TESTING)
+  # Command line options for the tests.
+  # TODO...
 
-# Set the number of processors for the parallel examples. The value of
-# MFEM_MPI_NP is ignored.
-set(EX1_TEST_NP 1)
-set(EX1P_TEST_NP 8)
-set(EX2_TEST_NP 1)
-set(EX6P_TEST_NP 8)
+  # Set the number of processors for the parallel examples. The value of
+  # MFEM_MPI_NP is ignored.
+  set(EX1_TEST_NP 1)
+  set(EX1P_TEST_NP 8)
+  set(EX2_TEST_NP 1)
+  set(EX6P_TEST_NP 8)
 
-# Add the tests: one test per source file.
-foreach(SRC_FILE ${PUMI_EXAMPLES_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
-  string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
-  set(TEST_NAME ${PFX}${TEST_NAME})
+  # Add the tests: one test per source file.
+  foreach(SRC_FILE ${PUMI_EXAMPLES_SRCS})
+    get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
+    string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
+    string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
+    set(TEST_NAME ${PFX}${TEST_NAME})
 
-  set(THIS_TEST_OPTIONS "-no-vis")
-  list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
-  # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
+    set(THIS_TEST_OPTIONS "-no-vis")
+    list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
+    # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
 
-  # All PUMI examples require MPI
-  if (FALSE)
-    add_test(NAME ${TEST_NAME}_ser
-      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-  else()
-    set(TEST_NP ${${UP_TEST_NAME}_TEST_NP})
-    add_test(NAME ${TEST_NAME}_np=${TEST_NP}
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${TEST_NP}
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
-  endif()
-endforeach()
+    # All PUMI examples require MPI
+    if (FALSE)
+      add_test(NAME ${TEST_NAME}_ser
+        COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
+    else()
+      set(TEST_NP ${${UP_TEST_NAME}_TEST_NP})
+      add_test(NAME ${TEST_NAME}_np=${TEST_NP}
+        COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${TEST_NP}
+        ${MPIEXEC_PREFLAGS}
+        $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
+        ${MPIEXEC_POSTFLAGS})
+    endif()
+  endforeach()
+endif()

examples/star-mixed-set.mesh

@@ -0,0 +1,158 @@
+MFEM mesh v1.0
+
+#
+# MFEM Geometry Types (see mesh/geom.hpp):
+#
+# POINT       = 0
+# SEGMENT     = 1
+# TRIANGLE    = 2
+# SQUARE      = 3
+# TETRAHEDRON = 4
+# CUBE        = 5
+# PRISM       = 6
+#
+
+dimension
+2
+
+elements
+30
+1 3 0 11 26 14
+1 3 0 14 27 17
+1 3 0 17 28 20
+1 3 0 20 29 23
+1 3 0 23 30 11
+1 2 11 1 26
+1 2 1 12 26
+1 3 26 12 3 13
+1 2 26 13 2
+1 2 14 26 2
+1 2 14 2 27
+1 2 2 15 27
+1 3 27 15 5 16
+1 2 27 16 4
+1 2 17 27 4
+1 2 17 4 28
+1 2 4 18 28
+1 3 28 18 7 19
+1 2 28 19 6
+1 2 20 28 6
+1 2 20 6 29
+1 2 6 21 29
+1 3 29 21 9 22
+1 2 29 22 8
+1 2 23 29 8
+1 2 23 8 30
+1 2 8 24 30
+1 3 30 24 10 25
+1 2 30 25 1
+1 2 11 30 1
+
+boundary
+20
+1 1 13 2
+1 1 12 3
+1 1 16 4
+1 1 15 5
+1 1 19 6
+1 1 18 7
+1 1 22 8
+1 1 21 9
+1 1 25 1
+1 1 24 10
+1 1 3 13
+1 1 1 12
+1 1 5 16
+1 1 2 15
+1 1 7 19
+1 1 4 18
+1 1 9 22
+1 1 6 21
+1 1 10 25
+1 1 8 24
+
+vertices
+31
+2
+0 0
+1 0
+0.309017 0.951057
+1.30902 0.951057
+-0.809017 0.587785
+-0.5 1.53884
+-0.809017 -0.587785
+-1.61803 0
+0.309017 -0.951057
+-0.5 -1.53884
+1.30902 -0.951057
+0.5 0
+1.15451 0.475529
+0.809019 0.951057
+0.154508 0.475529
+-0.0954915 1.24495
+-0.654508 1.06331
+-0.404508 0.293893
+-1.21352 0.293893
+-1.21352 -0.293892
+-0.404508 -0.293893
+-0.654508 -1.06331
+-0.0954915 -1.24495
+0.154508 -0.475529
+0.809019 -0.951057
+1.15451 -0.475529
+0.654509 0.475529
+-0.25 0.769421
+-0.809016 0
+-0.25 -0.76942
+0.654509 -0.475529
+
+MFEM sets v1.0
+
+vertex_sets
+3
+
+Origin
+1
+0
+
+Tent
+5
+1 2 4 6 8
+
+Gazebo
+5
+3 5 7 9 10
+
+edge_sets
+2
+
+Columbine
+5
+1 11
+2 14
+4 17
+6 20
+8 23
+
+Lily
+5
+0 11
+0 14
+0 17
+0 20
+0 23
+
+element_sets
+3
+
+Flying Squirrel
+3
+7 17 27
+
+Sea Lion
+4
+12 17 22 27
+
+Pinwheel
+5
+8 13 18 23 28

examples/star-set.mesh

@@ -0,0 +1,143 @@
+MFEM mesh v1.0
+
+#
+# MFEM Geometry Types (see mesh/geom.hpp):
+#
+# POINT       = 0
+# SEGMENT     = 1
+# TRIANGLE    = 2
+# SQUARE      = 3
+# TETRAHEDRON = 4
+# CUBE        = 5
+#
+
+dimension
+2
+
+elements
+20
+1 3 0 11 26 14
+1 3 0 14 27 17
+1 3 0 17 28 20
+1 3 0 20 29 23
+1 3 0 23 30 11
+1 3 11 1 12 26
+1 3 26 12 3 13
+1 3 14 26 13 2
+1 3 14 2 15 27
+1 3 27 15 5 16
+1 3 17 27 16 4
+1 3 17 4 18 28
+1 3 28 18 7 19
+1 3 20 28 19 6
+1 3 20 6 21 29
+1 3 29 21 9 22
+1 3 23 29 22 8
+1 3 23 8 24 30
+1 3 30 24 10 25
+1 3 11 30 25 1
+
+boundary
+20
+1 1 13 2
+1 1 12 3
+1 1 16 4
+1 1 15 5
+1 1 19 6
+1 1 18 7
+1 1 22 8
+1 1 21 9
+1 1 25 1
+1 1 24 10
+1 1 3 13
+1 1 1 12
+1 1 5 16
+1 1 2 15
+1 1 7 19
+1 1 4 18
+1 1 9 22
+1 1 6 21
+1 1 10 25
+1 1 8 24
+
+vertices
+31
+2
+0 0
+1 0
+0.309017 0.951057
+1.30902 0.951057
+-0.809017 0.587785
+-0.5 1.53884
+-0.809017 -0.587785
+-1.61803 0
+0.309017 -0.951057
+-0.5 -1.53884
+1.30902 -0.951057
+0.5 0
+1.15451 0.475529
+0.809019 0.951057
+0.154508 0.475529
+-0.0954915 1.24495
+-0.654508 1.06331
+-0.404508 0.293893
+-1.21352 0.293893
+-1.21352 -0.293892
+-0.404508 -0.293893
+-0.654508 -1.06331
+-0.0954915 -1.24495
+0.154508 -0.475529
+0.809019 -0.951057
+1.15451 -0.475529
+0.654509 0.475529
+-0.25 0.769421
+-0.809016 0
+-0.25 -0.76942
+0.654509 -0.475529
+
+MFEM sets v1.0
+
+vertex_sets
+3
+
+Origin
+1
+0
+
+Tent
+5
+1 2 4 6 8
+
+Gazebo
+5
+3 5 7 9 10
+
+edge_sets
+2
+
+Columbine
+5
+1 11
+2 14
+4 17
+6 20
+8 23
+
+Lily
+5
+0 11
+0 14
+0 17
+0 20
+0 23
+
+element_sets
+2
+
+Flying Squirrel
+3
+6 12 18
+
+Sea Lion
+4
+9 12 15 18

examples/sundials/CMakeLists.txt

@@ -41,36 +41,38 @@ add_mfem_examples(SUNDIALS_EXAMPLES_SRCS ${PFX} "" test_sundials)
 # which builds the examples and runs:
 #   ctest -R sundials
 
-# Command line options for the tests.
-# Example 9: test CVODE with CV_ADAMS (non-stiff implicit) time stepping
-set(EX9_COMMON_OPTS -m ../../data/periodic-hexagon.mesh -p 0 -s 7)
-set(EX9_TEST_OPTS ${EX9_COMMON_OPTS}  -r  2 -dt 0.0018 -vs 25)
-set(EX9P_TEST_OPTS ${EX9_COMMON_OPTS} -rp 1 -dt 0.0009 -vs 50)
-# Example 10: test CVODE with CV_BDF (stiff implicit) time stepping
-set(EX10_COMMON_OPTS -m ../../data/beam-quad.mesh -o 2 -s 5 -dt 0.15 -tf 6 -vs 10)
-set(EX10_TEST_OPTS ${EX10_COMMON_OPTS}  -r  2)
-set(EX10P_TEST_OPTS ${EX10_COMMON_OPTS} -rp 1)
-# Example 16: use the default options
+if (MFEM_ENABLE_TESTING)
+  # Command line options for the tests.
+  # Example 9: test CVODE with CV_ADAMS (non-stiff implicit) time stepping
+  set(EX9_COMMON_OPTS -m ../../data/periodic-hexagon.mesh -p 0 -s 7)
+  set(EX9_TEST_OPTS ${EX9_COMMON_OPTS}  -r  2 -dt 0.0018 -vs 25)
+  set(EX9P_TEST_OPTS ${EX9_COMMON_OPTS} -rp 1 -dt 0.0009 -vs 50)
+  # Example 10: test CVODE with CV_BDF (stiff implicit) time stepping
+  set(EX10_COMMON_OPTS -m ../../data/beam-quad.mesh -o 2 -s 5 -dt 0.15 -tf 6 -vs 10)
+  set(EX10_TEST_OPTS ${EX10_COMMON_OPTS}  -r  2)
+  set(EX10P_TEST_OPTS ${EX10_COMMON_OPTS} -rp 1)
+  # Example 16: use the default options
 
-# Add the tests: one test per source file.
-foreach(SRC_FILE ${SUNDIALS_EXAMPLES_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
-  string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
-  set(TEST_NAME ${PFX}${TEST_NAME})
+  # Add the tests: one test per source file.
+  foreach(SRC_FILE ${SUNDIALS_EXAMPLES_SRCS})
+    get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
+    string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
+    string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
+    set(TEST_NAME ${PFX}${TEST_NAME})
 
-  set(THIS_TEST_OPTIONS "-no-vis")
-  list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
-  # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
+    set(THIS_TEST_OPTIONS "-no-vis")
+    list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
+    # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
 
-  if (NOT (${TEST_NAME} MATCHES ".*p$"))
-    add_test(NAME ${TEST_NAME}_ser
-      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-  else()
-    add_test(NAME ${TEST_NAME}_np=4
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
-  endif()
-endforeach()
+    if (NOT (${TEST_NAME} MATCHES ".*p$"))
+      add_test(NAME ${TEST_NAME}_ser
+        COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
+    else()
+      add_test(NAME ${TEST_NAME}_np=4
+        COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
+        ${MPIEXEC_PREFLAGS}
+        $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
+        ${MPIEXEC_POSTFLAGS})
+    endif()
+  endforeach()
+endif()

examples/superlu/CMakeLists.txt

@@ -32,31 +32,32 @@ add_mfem_examples(SUPERLU_EXAMPLES_SRCS ${PFX} "" test_superlu)
 # The SuperLU tests can be run separately using the target "test_superlu"
 # which builds the examples and runs:
 #   ctest -R superlu
+if (MFEM_ENABLE_TESTING)
+  # Command line options for the tests.
+  # Example 1: Test SuperLU on the simple Poisson problem
+  set(EX1_COMMON_OPTS -m ../../data/star.mesh -p 2)
+  set(EX1P_TEST_OPTS ${EX1_COMMON_OPTS})
 
-# Command line options for the tests.
-# Example 1: Test SuperLU on the simple Poisson problem
-set(EX1_COMMON_OPTS -m ../../data/star.mesh -p 2)
-set(EX1P_TEST_OPTS ${EX1_COMMON_OPTS})
+  # Add the tests: one test per source file.
+  foreach(SRC_FILE ${SUPERLU_EXAMPLES_SRCS})
+    get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
+    string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
+    string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
+    set(TEST_NAME ${PFX}${TEST_NAME})
 
-# Add the tests: one test per source file.
-foreach(SRC_FILE ${SUPERLU_EXAMPLES_SRCS})
-  get_filename_component(SRC_FILENAME ${SRC_FILE} NAME)
-  string(REPLACE ".cpp" "" TEST_NAME ${SRC_FILENAME})
-  string(TOUPPER ${TEST_NAME} UP_TEST_NAME)
-  set(TEST_NAME ${PFX}${TEST_NAME})
+    set(THIS_TEST_OPTIONS "-no-vis")
+    list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
+    # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
 
-  set(THIS_TEST_OPTIONS "-no-vis")
-  list(APPEND THIS_TEST_OPTIONS ${${UP_TEST_NAME}_TEST_OPTS})
-  # message(STATUS "Test ${TEST_NAME} options: ${THIS_TEST_OPTIONS}")
-
-  if (NOT (${TEST_NAME} MATCHES ".*p$"))
-    add_test(NAME ${TEST_NAME}_ser
-      COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
-  else()
-    add_test(NAME ${TEST_NAME}_np=4
-      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
-      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
-      ${MPIEXEC_POSTFLAGS})
-  endif()
-endforeach()
+    if (NOT (${TEST_NAME} MATCHES ".*p$"))
+      add_test(NAME ${TEST_NAME}_ser
+        COMMAND ${TEST_NAME} ${THIS_TEST_OPTIONS})
+    else()
+      add_test(NAME ${TEST_NAME}_np=4
+        COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
+        ${MPIEXEC_PREFLAGS}
+        $<TARGET_FILE:${TEST_NAME}> ${THIS_TEST_OPTIONS}
+        ${MPIEXEC_POSTFLAGS})
+    endif()
+  endforeach()
+endif()

fem/bilininteg.cpp

@@ -985,6 +985,8 @@ void DiffusionIntegrator::ComputeElementFlux
                   "Unexpected height for MatrixCoefficient");
    }
 
+   MFEM_VERIFY(!SMQ, "SymmetricMatrixCoefficient not supported here");
+
 #ifdef MFEM_THREAD_SAFE
    DenseMatrix dshape(nd,dim), invdfdx(dim, spaceDim);
    DenseMatrix M(MQ ? spaceDim : 0);
@@ -997,7 +999,7 @@ void DiffusionIntegrator::ComputeElementFlux
 #endif
    vec.SetSize(dim);
    vecdxt.SetSize(spaceDim);
-   pointflux.SetSize(MQ ? spaceDim : 0);
+   pointflux.SetSize(MQ || VQ ? spaceDim : 0);
 
    const IntegrationRule &ir = fluxelem.GetNodes();
    fnd = ir.GetNPoints();
@@ -1013,36 +1015,45 @@ void DiffusionIntegrator::ComputeElementFlux
       CalcInverse(Trans.Jacobian(), invdfdx);
       invdfdx.MultTranspose(vec, vecdxt);
 
-      if (!MQ && !VQ)
+      if (with_coef)
       {
-         if (Q && with_coef)
+         if (!MQ && !VQ)
          {
-            vecdxt *= Q->Eval(Trans,ip);
+            if (Q)
+            {
+               vecdxt *= Q->Eval(Trans,ip);
+            }
+            for (j = 0; j < spaceDim; j++)
+            {
+               flux(fnd*j+i) = vecdxt(j);
+            }
          }
-         for (j = 0; j < spaceDim; j++)
+         else
          {
-            flux(fnd*j+i) = vecdxt(j);
+            if (MQ)
+            {
+               MQ->Eval(M, Trans, ip);
+               M.Mult(vecdxt, pointflux);
+            }
+            else
+            {
+               VQ->Eval(D, Trans, ip);
+               for (int j=0; j<spaceDim; ++j)
+               {
+                  pointflux[j] = D[j] * vecdxt[j];
+               }
+            }
+            for (j = 0; j < spaceDim; j++)
+            {
+               flux(fnd*j+i) = pointflux(j);
+            }
          }
       }
       else
       {
-         if (MQ)
-         {
-            MQ->Eval(M, Trans, ip);
-            M.Mult(vecdxt, pointflux);
-         }
-         else
-         {
-            VQ->Eval(D, Trans, ip);
-            for (int j=0; j<spaceDim; ++j)
-            {
-               pointflux[j] = D[j] * vecdxt[j];
-            }
-
-         }
          for (j = 0; j < spaceDim; j++)
          {
-            flux(fnd*j+i) = pointflux(j);
+            flux(fnd*j+i) = vecdxt(j);
          }
       }
    }
@@ -1058,8 +1069,13 @@ double DiffusionIntegrator::ComputeFluxEnergy
 
 #ifdef MFEM_THREAD_SAFE
    DenseMatrix M;
+   Vector D(VQ ? VQ->GetVDim() : 0);
+#else
+   D.SetSize(VQ ? VQ->GetVDim() : 0);
 #endif
 
+   MFEM_VERIFY(!SMQ, "SymmetricMatrixCoefficient not supported here");
+
    shape.SetSize(nd);
    pointflux.SetSize(spaceDim);
    if (d_energy) { vec.SetSize(spaceDim); }
@@ -1088,17 +1104,23 @@ double DiffusionIntegrator::ComputeFluxEnergy
       Trans.SetIntPoint(&ip);
       double w = Trans.Weight() * ip.weight;
 
-      if (!MQ)
+      if (MQ)
+      {
+         MQ->Eval(M, Trans, ip);
+         energy += w * M.InnerProduct(pointflux, pointflux);
+      }
+      else if (VQ)
+      {
+         VQ->Eval(D, Trans, ip);
+         D *= pointflux;
+         energy += w * (D * pointflux);
+      }
+      else
       {
          double e = (pointflux * pointflux);
          if (Q) { e *= Q->Eval(Trans, ip); }
          energy += w * e;
       }
-      else
-      {
-         MQ->Eval(M, Trans, ip);
-         energy += w * M.InnerProduct(pointflux, pointflux);
-      }
 
       if (d_energy)
       {
@@ -1108,7 +1130,7 @@ double DiffusionIntegrator::ComputeFluxEnergy
          {
             (*d_energy)[k] += w * vec[k] * vec[k];
          }
-         // TODO: Q, MQ
+         // TODO: Q, VQ, MQ
       }
    }
 

fem/bilininteg_diffusion_pa.cpp

@@ -903,9 +903,11 @@ static void PADiffusionAssembleDiagonal(const int dim,
    {
       switch ((D1D << 4 ) | Q1D)
       {
+         case 0x22: return SmemPADiffusionDiagonal3D<2,2>(NE,symm,B,G,D,Y);
          case 0x23: return SmemPADiffusionDiagonal3D<2,3>(NE,symm,B,G,D,Y);
          case 0x34: return SmemPADiffusionDiagonal3D<3,4>(NE,symm,B,G,D,Y);
          case 0x45: return SmemPADiffusionDiagonal3D<4,5>(NE,symm,B,G,D,Y);
+         case 0x46: return SmemPADiffusionDiagonal3D<4,6>(NE,symm,B,G,D,Y);
          case 0x56: return SmemPADiffusionDiagonal3D<5,6>(NE,symm,B,G,D,Y);
          case 0x67: return SmemPADiffusionDiagonal3D<6,7>(NE,symm,B,G,D,Y);
          case 0x78: return SmemPADiffusionDiagonal3D<7,8>(NE,symm,B,G,D,Y);
@@ -1554,7 +1556,7 @@ static void SmemPADiffusionApply3D(const int NE,
    auto d = Reshape(d_.Read(), Q1D, Q1D, Q1D, symmetric ? 6 : 9, 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 D1D = T_D1D ? T_D1D : d1d;
       const int Q1D = T_Q1D ? T_Q1D : q1d;
@@ -1583,118 +1585,102 @@ 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 (MFEM_THREAD_ID(z) == 0)
+      {
+         MFEM_FOREACH_THREAD(dy,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(qx,x,Q1D)
+            {
+               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_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);
+               double u = 0.0, v = 0.0;
+               MFEM_UNROLL(MD1)
+               for (int dx = 0; dx < D1D; ++dx)
+               {
+                  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);
+                  const double coords = X[dz][dy][dx];
+                  u += coords * B[i][j];
+                  v += coords * G[k][l] * s;
+               }
+               DDQ0[dz][dy][qx] = u;
+               DDQ1[dz][dy][qx] = v;
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(dz,z,D1D)
+      {
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
+            {
+               double u = 0.0, v = 0.0, w = 0.0;
+               MFEM_UNROLL(MD1)
+               for (int dy = 0; dy < D1D; ++dy)
+               {
+                  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);
+                  u += DDQ1[dz][dy][qx] * B[i][j];
+                  v += DDQ0[dz][dy][qx] * G[k][l] * s;
+                  w += DDQ0[dz][dy][qx] * B[i][j];
+               }
+               DQQ0[dz][qy][qx] = u;
+               DQQ1[dz][qy][qx] = v;
+               DQQ2[dz][qy][qx] = w;
+            }
+         }
+      }
+      MFEM_SYNC_THREAD;
+      MFEM_FOREACH_THREAD(qz,z,Q1D)
+      {
+         MFEM_FOREACH_THREAD(qy,y,Q1D)
+         {
+            MFEM_FOREACH_THREAD(qx,x,Q1D)
+            {
+               double u = 0.0, v = 0.0, w = 0.0;
                MFEM_UNROLL(MD1)
                for (int dz = 0; dz < D1D; ++dz)
-               {
-                  const double coords = X[dz][dy][dx];
-                  u[dz] += coords * B[i][j];
-                  v[dz] += coords * G[k][l] * s;
-               }
-            }
-            MFEM_UNROLL(MD1)
-            for (int dz = 0; dz < D1D; ++dz)
-            {
-               DDQ0[dz][dy][qx] = u[dz];
-               DDQ1[dz][dy][qx] = v[dz];
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(qy,y,Q1D)
-      {
-         MFEM_FOREACH_THREAD(qx,x,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)
-            {
-               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++)
-               {
-                  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];
-               }
-            }
-            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];
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(qy,y,Q1D)
-      {
-         MFEM_FOREACH_THREAD(qx,x,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_UNROLL(MQ1)
-               for (int qz = 0; qz < Q1D; qz++)
                {
                   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[i][j];
+                  v += DQQ1[dz][qy][qx] * B[i][j];
+                  w += DQQ2[dz][qy][qx] * G[k][l] * s;
                }
-            }
-            MFEM_UNROLL(MQ1)
-            for (int qz = 0; qz < Q1D; qz++)
-            {
                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);
@@ -1704,9 +1690,9 @@ static void SmemPADiffusionApply3D(const int NE,
                const double O31 = symmetric ? O13 : d(qx,qy,qz,6,e);
                const double O32 = symmetric ? O23 : d(qx,qy,qz,7,e);
                const double O33 = symmetric ? d(qx,qy,qz,5,e) : d(qx,qy,qz,8,e);
-               const double gX = u[qz];
-               const double gY = v[qz];
-               const double gZ = w[qz];
+               const double gX = u;
+               const double gY = v;
+               const double gZ = w;
                QQQ0[qz][qy][qx] = (O11*gX) + (O12*gY) + (O13*gZ);
                QQQ1[qz][qy][qx] = (O21*gX) + (O22*gY) + (O23*gZ);
                QQQ2[qz][qy][qx] = (O31*gX) + (O32*gY) + (O33*gZ);
@@ -1714,112 +1700,94 @@ static void SmemPADiffusionApply3D(const int NE,
          }
       }
       MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(d,y,D1D)
+      if (MFEM_THREAD_ID(z) == 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)
+            {
+               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_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);
+               double u = 0.0, v = 0.0, w = 0.0;
+               MFEM_UNROLL(MQ1)
+               for (int qx = 0; qx < Q1D; ++qx)
+               {
+                  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);
+                  u += QQQ0[qz][qy][qx] * Gt[l][k] * s;
+                  v += QQQ1[qz][qy][qx] * Bt[j][i];
+                  w += QQQ2[qz][qy][qx] * Bt[j][i];
+               }
+               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, v = 0.0, w = 0.0;
+               MFEM_UNROLL(Q1D)
+               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);
+                  u += QQD0[qz][qy][dx] * Bt[j][i];
+                  v += QQD1[qz][qy][dx] * Gt[l][k] * s;
+                  w += QQD2[qz][qy][dx] * Bt[j][i];
+               }
+               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, v = 0.0, w = 0.0;
                MFEM_UNROLL(MQ1)
                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];
-               }
-            }
-            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;
+                  u += QDD0[qz][dy][dx] * Bt[j][i];
+                  v += QDD1[qz][dy][dx] * Bt[j][i];
+                  w += 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);
             }
          }
       }
@@ -1877,6 +1845,7 @@ static void PADiffusionApply(const int dim,
    {
       switch (ID)
       {
+         case 0x22: return SmemPADiffusionApply3D<2,2>(NE,symm,B,G,D,X,Y);
          case 0x23: return SmemPADiffusionApply3D<2,3>(NE,symm,B,G,D,X,Y);
          case 0x34: return SmemPADiffusionApply3D<3,4>(NE,symm,B,G,D,X,Y);
          case 0x45: return SmemPADiffusionApply3D<4,5>(NE,symm,B,G,D,X,Y);

fem/bilininteg_mass_pa.cpp

@@ -1203,8 +1203,10 @@ static void PAMassApply(const int dim,
    {
       switch (id)
       {
+         case 0x22: return SmemPAMassApply3D<2,2>(NE,B,Bt,D,X,Y);
          case 0x23: return SmemPAMassApply3D<2,3>(NE,B,Bt,D,X,Y);
          case 0x24: return SmemPAMassApply3D<2,4>(NE,B,Bt,D,X,Y);
+         case 0x26: return SmemPAMassApply3D<2,6>(NE,B,Bt,D,X,Y);
          case 0x34: return SmemPAMassApply3D<3,4>(NE,B,Bt,D,X,Y);
          case 0x35: return SmemPAMassApply3D<3,5>(NE,B,Bt,D,X,Y);
          case 0x36: return SmemPAMassApply3D<3,6>(NE,B,Bt,D,X,Y);

fem/ceed/util.cpp

@@ -186,14 +186,18 @@ static void InitTensorBasis(const mfem::FiniteElementSpace &fes,
    const int ndofs = maps.ndof;
    const int nqpts = maps.nqpt;
    mfem::Vector qX(nqpts), qW(nqpts);
-   const mfem::IntegrationRule &ir1d =
-      IntRules.Get(Geometry::SEGMENT, ir.GetOrder());
+   // The x-coordinates of the first `nqpts` points of the integration rule are
+   // the points of the corresponding 1D rule. We also scale the weights
+   // accordingly.
+   double w_sum = 0.0;
    for (int i = 0; i < nqpts; i++)
    {
-      const mfem::IntegrationPoint &ip = ir1d.IntPoint(i);
+      const mfem::IntegrationPoint &ip = ir.IntPoint(i);
       qX(i) = ip.x;
       qW(i) = ip.weight;
+      w_sum += ip.weight;
    }
+   qW *= 1.0/w_sum;
    CeedBasisCreateTensorH1(ceed, mesh->Dimension(), fes.GetVDim(), ndofs,
                            nqpts, maps.Bt.GetData(),
                            maps.Gt.GetData(), qX.GetData(),

fem/coefficient.cpp

@@ -28,6 +28,52 @@ double PWConstCoefficient::Eval(ElementTransformation & T,
    return (constants(att-1));
 }
 
+void PWCoefficient::InitMap(const Array<int> & attr,
+                            const Array<Coefficient*> & coefs)
+{
+   MFEM_VERIFY(attr.Size() == coefs.Size(),
+               "PWCoefficient:  "
+               "Attribute and coefficient arrays have incompatible "
+               "dimensions.");
+
+   for (int i=0; i<attr.Size(); i++)
+   {
+      if (coefs[i] != NULL)
+      {
+         UpdateCoefficient(attr[i], *coefs[i]);
+      }
+   }
+}
+
+void PWCoefficient::SetTime(double t)
+{
+   Coefficient::SetTime(t);
+
+   std::map<int, Coefficient*>::iterator p = pieces.begin();
+   for (; p != pieces.end(); p++)
+   {
+      if (p->second != NULL)
+      {
+         p->second->SetTime(t);
+      }
+   }
+}
+
+double PWCoefficient::Eval(ElementTransformation &T,
+                           const IntegrationPoint &ip)
+{
+   const int att = T.Attribute;
+   std::map<int, Coefficient*>::const_iterator p = pieces.find(att);
+   if (p != pieces.end())
+   {
+      if ( p->second != NULL)
+      {
+         return p->second->Eval(T, ip);
+      }
+   }
+   return 0.0;
+}
+
 double FunctionCoefficient::Eval(ElementTransformation & T,
                                  const IntegrationPoint & ip)
 {
@@ -120,6 +166,63 @@ void VectorCoefficient::Eval(DenseMatrix &M, ElementTransformation &T,
    }
 }
 
+void PWVectorCoefficient::InitMap(const Array<int> & attr,
+                                  const Array<VectorCoefficient*> & coefs)
+{
+   MFEM_VERIFY(attr.Size() == coefs.Size(),
+               "PWVectorCoefficient:  "
+               "Attribute and coefficient arrays have incompatible "
+               "dimensions.");
+
+   for (int i=0; i<attr.Size(); i++)
+   {
+      if (coefs[i] != NULL)
+      {
+         UpdateCoefficient(attr[i], *coefs[i]);
+      }
+   }
+}
+
+void PWVectorCoefficient::UpdateCoefficient(int attr, VectorCoefficient & coef)
+{
+   MFEM_VERIFY(coef.GetVDim() == vdim,
+               "PWVectorCoefficient::UpdateCoefficient:  "
+               "VectorCoefficient has incompatible dimension.");
+   pieces[attr] = &coef;
+}
+
+void PWVectorCoefficient::SetTime(double t)
+{
+   VectorCoefficient::SetTime(t);
+
+   std::map<int, VectorCoefficient*>::iterator p = pieces.begin();
+   for (; p != pieces.end(); p++)
+   {
+      if (p->second != NULL)
+      {
+         p->second->SetTime(t);
+      }
+   }
+}
+
+void PWVectorCoefficient::Eval(Vector &V, ElementTransformation &T,
+                               const IntegrationPoint &ip)
+{
+   const int att = T.Attribute;
+   std::map<int, VectorCoefficient*>::const_iterator p = pieces.find(att);
+   if (p != pieces.end())
+   {
+      if ( p->second != NULL)
+      {
+         p->second->Eval(V, T, ip);
+         return;
+      }
+   }
+
+   V.SetSize(vdim);
+   V = 0.0;
+}
+
 void VectorFunctionCoefficient::Eval(Vector &V, ElementTransformation &T,
                                      const IntegrationPoint &ip)
 {
@@ -331,6 +434,72 @@ void VectorRestrictedCoefficient::Eval(
    }
 }
 
+void PWMatrixCoefficient::InitMap(const Array<int> & attr,
+                                  const Array<MatrixCoefficient*> & coefs)
+{
+   MFEM_VERIFY(attr.Size() == coefs.Size(),
+               "PWMatrixCoefficient:  "
+               "Attribute and coefficient arrays have incompatible "
+               "dimensions.");
+
+   for (int i=0; i<attr.Size(); i++)
+   {
+      if (coefs[i] != NULL)
+      {
+         UpdateCoefficient(attr[i], *coefs[i]);
+      }
+   }
+}
+
+void PWMatrixCoefficient::UpdateCoefficient(int attr, MatrixCoefficient & coef)
+{
+   MFEM_VERIFY(coef.GetHeight() == height,
+               "PWMatrixCoefficient::UpdateCoefficient:  "
+               "MatrixCoefficient has incompatible height.");
+   MFEM_VERIFY(coef.GetWidth() == width,
+               "PWMatrixCoefficient::UpdateCoefficient:  "
+               "MatrixCoefficient has incompatible width.");
+   if (symmetric)
+   {
+      MFEM_VERIFY(coef.IsSymmetric(),
+                  "PWMatrixCoefficient::UpdateCoefficient:  "
+                  "MatrixCoefficient has incompatible symmetry.");
+   }
+   pieces[attr] = &coef;
+}
+
+void PWMatrixCoefficient::SetTime(double t)
+{
+   MatrixCoefficient::SetTime(t);
+
+   std::map<int, MatrixCoefficient*>::iterator p = pieces.begin();
+   for (; p != pieces.end(); p++)
+   {
+      if (p->second != NULL)
+      {
+         p->second->SetTime(t);
+      }
+   }
+}
+
+void PWMatrixCoefficient::Eval(DenseMatrix &K, ElementTransformation &T,
+                               const IntegrationPoint &ip)
+{
+   const int att = T.Attribute;
+   std::map<int, MatrixCoefficient*>::const_iterator p = pieces.find(att);
+   if (p != pieces.end())
+   {
+      if ( p->second != NULL)
+      {
+         p->second->Eval(K, T, ip);
+         return;
+      }
+   }
+
+   K.SetSize(height, width);
+   K = 0.0;
+}
+
 void MatrixFunctionCoefficient::SetTime(double t)
 {
    if (Q) { Q->SetTime(t); }

fem/coefficient.hpp

@@ -125,6 +125,87 @@ public:
                        const IntegrationPoint &ip);
 };
 
+/** @brief A piecewise coefficient with the pieces keyed off the element
+    attribute numbers.
+
+    A value of zero will be returned for any missing attribute numbers.
+
+    This object will not assume ownership of any Coefficient objects
+    passed to it. Consequently, the caller must ensure that the
+    individual Coefficient objects are not deleted while this
+    PWCoefficient is still in use.
+
+    \note The keys may either be domain attribute numbers or boundary
+    attribute numbers. If the PWCoefficient is used with a domain
+    integrator the keys are assumed to be domain attribute
+    numbers. Similarly, if the PWCoefficient is used with a boundary
+    integrator the keys are assumed to be boundary attribute numbers.
+*/
+class PWCoefficient : public Coefficient
+{
+private:
+   /** Internal data structure to store pointers to the appropriate
+       coefficients for different regions of the mesh. The keys used
+       in the map are the mesh attribute numbers (either element
+       attribute or boundary element attribute depending upon
+       context). The values returned for any missing attributes will
+       be zero. The coefficient pointers may be NULL in which case a
+       value of zero is returned.
+
+       The Coefficient objects contained in this map are NOT owned by
+       this PWCoefficient object. This means that they will not be
+       deleted when this object is deleted also the caller must ensure
+       that the various Coefficient objects are not deleted while this
+       PWCoefficient is still needed.
+   */
+   std::map<int, Coefficient*> pieces;
+
+   /** Convenience function to check for compatible array lengths,
+       loop over the arrays, and add their attribute/Coefficient pairs
+       to the internal data structure.
+   */
+   void InitMap(const Array<int> & attr,
+                const Array<Coefficient*> & coefs);
+
+public:
+
+   /// Constructs a piecewise coefficient
+   explicit PWCoefficient() {}
+
+   /// Construct the coefficient using arrays describing the pieces
+   /** \param attr - an array of attribute numbers for each piece
+       \param coefs - the corresponding array of Coefficient pointers
+       Any missing attributes or NULL coefficient pointers will result in a
+       value of zero being returned for that attribute.
+
+       \note Ownership of the Coefficient objects will NOT be
+       transferred to this object.
+   */
+   PWCoefficient(const Array<int> & attr,
+                 const Array<Coefficient*> & coefs)
+   { InitMap(attr, coefs); }
+
+   /// Set the time for time dependent coefficients
+   virtual void SetTime(double t);
+
+   /// Replace a set of coefficients
+   void UpdateCoefficients(const Array<int> & attr,
+                           const Array<Coefficient*> & coefs)
+   { InitMap(attr, coefs); }
+
+   /// Replace a single Coefficient for a particular attribute
+   void UpdateCoefficient(int attr, Coefficient & coef)
+   { pieces[attr] = &coef; }
+
+   /// Remove a single Coefficient for a particular attribute
+   void ZeroCoefficient(int attr)
+   { pieces.erase(attr); }
+
+   /// Evaluate the coefficient.
+   virtual double Eval(ElementTransformation &T,
+                       const IntegrationPoint &ip);
+};
+
 /// A general function coefficient
 class FunctionCoefficient : public Coefficient
 {
@@ -413,6 +494,88 @@ public:
    const Vector& GetVec() { return vec; }
 };
 
+/** @brief A piecewise vector-valued coefficient with the pieces keyed off the
+    element attribute numbers.
+
+    A value of zero will be returned for any missing attribute numbers.
+
+    This object will not assume ownership of any VectorCoefficient
+    objects passed to it. Consequently, the caller must ensure that
+    the individual VectorCoefficient objects are not deleted while
+    this PWVectorCoefficient is still in use.
+
+    \note The keys may either be domain attribute numbers or boundary
+    attribute numbers. If the PWVectorCoefficient is used with a
+    domain integrator the keys are assumed to be domain attribute
+    numbers. Similarly, if the PWVectorCoefficient is used with a
+    boundary integrator the keys are assumed to be boundary attribute
+    numbers.
+*/
+class PWVectorCoefficient : public VectorCoefficient
+{
+private:
+   /** Internal data structure to store pointers to the appropriate
+       coefficients for different regions of the mesh. The keys used
+       in the map are the mesh attribute numbers (either element
+       attribute or boundary element attribute depending upon
+       context). The values returned for any missing attributes will
+       be zero. The coefficient pointers may be NULL in which case a
+       value of zero is returned.
+
+       The VectorCoefficient objects contained in this map are NOT
+       owned by this PWVectorCoefficient object. This means that they
+       will not be deleted when this object is deleted also the caller
+       must ensure that the various VectorCoefficient objects are not
+       deleted while this PWVectorCoefficient is still needed.
+   */
+   std::map<int, VectorCoefficient*> pieces;
+
+   /** Convenience function to check for compatible array lengths,
+       loop over the arrays, and add their attribute/VectorCoefficient
+       pairs to the internal data structure.
+   */
+   void InitMap(const Array<int> & attr,
+                const Array<VectorCoefficient*> & coefs);
+
+public:
+
+   /// Constructs a piecewise vector coefficient of dimension vd
+   explicit PWVectorCoefficient(int vd): VectorCoefficient(vd) {}
+
+   /// Construct the coefficient using arrays describing the pieces
+   /** \param vd - dimension of the vector-valued result
+       \param attr - an array of attribute numbers for each piece
+       \param coefs - the corresponding array of VectorCoefficient pointers
+       Any missing attributes or NULL coefficient pointers will result in a
+       zero vector being returned for that attribute.
+
+       \note Ownership of the VectorCoefficient objects will NOT be
+       transferred to this object.
+   */
+   PWVectorCoefficient(int vd, const Array<int> & attr,
+                       const Array<VectorCoefficient*> & coefs)
+      : VectorCoefficient(vd) { InitMap(attr, coefs); }
+
+   /// Set the time for time dependent coefficients
+   virtual void SetTime(double t);
+
+   /// Replace a set of coefficients
+   void UpdateCoefficients(const Array<int> & attr,
+                           const Array<VectorCoefficient*> & coefs)
+   { InitMap(attr, coefs); }
+
+   /// Replace a single Coefficient for a particular attribute
+   void UpdateCoefficient(int attr, VectorCoefficient & coef);
+
+   /// Remove a single VectorCoefficient for a particular attribute
+   void ZeroCoefficient(int attr)
+   { pieces.erase(attr); }
+
+   /// Evaluate the coefficient.
+   virtual void Eval(Vector &V, ElementTransformation &T,
+                     const IntegrationPoint &ip);
+};
+
 /// A general vector function coefficient
 class VectorFunctionCoefficient : public VectorCoefficient
 {
@@ -782,6 +945,112 @@ public:
 };
 
 
+/** @brief A piecewise matrix-valued coefficient with the pieces keyed off the
+    element attribute numbers.
+
+    A value of zero will be returned for any missing attribute numbers.
+
+    This object will not assume ownership of any MatrixCoefficient
+    objects passed to it. Consequently, the caller must ensure that
+    the individual MatrixCoefficient objects are not deleted while
+    this PWMatrixCoefficient is still in use.
+
+    \note The keys may either be domain attribute numbers or boundary
+    attribute numbers. If the PWMatrixCoefficient is used with a
+    domain integrator the keys are assumed to be domain attribute
+    numbers. Similarly, if the PWMatrixCoefficient is used with a
+    boundary integrator the keys are assumed to be boundary attribute
+    numbers.
+*/
+class PWMatrixCoefficient : public MatrixCoefficient
+{
+private:
+   /** Internal data structure to store pointers to the appropriate
+       coefficients for different regions of the mesh. The keys used
+       in the map are the mesh attribute numbers (either element
+       attribute or boundary element attribute depending upon
+       context). The values returned for any missing attributes will
+       be zero. The coefficient pointers may be NULL in which case a
+       value of zero is returned.
+
+       The MatrixCoefficient objects contained in this map are NOT
+       owned by this PWMatrixCoefficient object. This means that they
+       will not be deleted when this object is deleted also the caller
+       must ensure that the various MatrixCoefficient objects are not
+       deleted while this PWMatrixCoefficient is still needed.
+   */
+   std::map<int, MatrixCoefficient*> pieces;
+
+   /** Convenience function to check for compatible array lengths,
+       loop over the arrays, and add their attribute/MatrixCoefficient
+       pairs to the internal data structure.
+   */
+   void InitMap(const Array<int> & attr,
+                const Array<MatrixCoefficient*> & coefs);
+
+public:
+
+   /// Constructs a piecewise matrix coefficient of dimension dim by dim
+   explicit PWMatrixCoefficient(int dim, bool symm = false)
+      : MatrixCoefficient(dim, symm) {}
+
+   /// Constructs a piecewise matrix coefficient of dimension h by w
+   explicit PWMatrixCoefficient(int h, int w, bool symm = false)
+      : MatrixCoefficient(h, w, symm) {}
+
+   /// Construct the coefficient using arrays describing the pieces
+   /** \param dim - size of the square matrix-valued result
+       \param attr - an array of attribute numbers for each piece
+       \param coefs - the corresponding array of MatrixCoefficient pointers
+       \param symm - true if the result will be symmetric, false otherwise
+       Any missing attributes or NULL coefficient pointers will result in a
+       zero matrix being returned.
+
+       \note Ownership of the MatrixCoefficient objects will NOT be
+       transferred to this object.
+   */
+   PWMatrixCoefficient(int dim, const Array<int> & attr,
+                       const Array<MatrixCoefficient*> & coefs,
+                       bool symm=false)
+      : MatrixCoefficient(dim, symm) { InitMap(attr, coefs); }
+
+   /// Construct the coefficient using arrays describing the pieces
+   /** \param h - height of the matrix-valued result
+       \param w - width of the matrix-valued result
+       \param attr - an array of attribute numbers for each piece
+       \param coefs - the corresponding array of MatrixCoefficient pointers
+       \param symm - true if the result will be symmetric, false otherwise
+       Any missing attributes or NULL coefficient pointers will result in a
+       zero matrix being returned for that attribute.
+
+       \note Ownership of the MatrixCoefficient objects will NOT be
+       transferred to this object.
+   */
+   PWMatrixCoefficient(int h, int w, const Array<int> & attr,
+                       const Array<MatrixCoefficient*> & coefs,
+                       bool symm=false)
+      : MatrixCoefficient(h, w, symm) { InitMap(attr, coefs); }
+
+   /// Set the time for time dependent coefficients
+   virtual void SetTime(double t);
+
+   /// Replace a set of coefficients
+   void UpdateCoefficients(const Array<int> & attr,
+                           const Array<MatrixCoefficient*> & coefs)
+   { InitMap(attr, coefs); }
+
+   /// Replace a single coefficient for a particular attribute
+   void UpdateCoefficient(int attr, MatrixCoefficient & coef);
+
+   /// Remove a single MatrixCoefficient for a particular attribute
+   void ZeroCoefficient(int attr)
+   { pieces.erase(attr); }
+
+   /// Evaluate the coefficient.
+   virtual void Eval(DenseMatrix &K, ElementTransformation &T,
+                     const IntegrationPoint &ip);
+};
+
 /** @brief A matrix coefficient with an optional scalar coefficient multiplier
     \a q.  The matrix function can either be represented by a std function or
     a constant matrix provided when constructing this object.  */

fem/complex_fem.cpp

@@ -195,6 +195,15 @@ ComplexLinearForm::AddDomainIntegrator(LinearFormIntegrator *lfi_real,
    if ( lfi_imag ) { lfi->AddDomainIntegrator(lfi_imag); }
 }
 
+void
+ComplexLinearForm::AddDomainIntegrator(LinearFormIntegrator *lfi_real,
+                                       LinearFormIntegrator *lfi_imag,
+                                       Array<int> &elem_attr_marker)
+{
+   if ( lfi_real ) { lfr->AddDomainIntegrator(lfi_real, elem_attr_marker); }
+   if ( lfi_imag ) { lfi->AddDomainIntegrator(lfi_imag, elem_attr_marker); }
+}
+
 void
 ComplexLinearForm::AddBoundaryIntegrator(LinearFormIntegrator *lfi_real,
                                          LinearFormIntegrator *lfi_imag)
@@ -317,6 +326,14 @@ void SesquilinearForm::AddDomainIntegrator(BilinearFormIntegrator *bfi_real,
    if (bfi_imag) { blfi->AddDomainIntegrator(bfi_imag); }
 }
 
+void SesquilinearForm::AddDomainIntegrator(BilinearFormIntegrator *bfi_real,
+                                           BilinearFormIntegrator *bfi_imag,
+                                           Array<int> & elem_marker)
+{
+   if (bfi_real) { blfr->AddDomainIntegrator(bfi_real, elem_marker); }
+   if (bfi_imag) { blfi->AddDomainIntegrator(bfi_imag, elem_marker); }
+}
+
 void
 SesquilinearForm::AddBoundaryIntegrator(BilinearFormIntegrator *bfi_real,
                                         BilinearFormIntegrator *bfi_imag)
@@ -879,6 +896,15 @@ ParComplexLinearForm::AddDomainIntegrator(LinearFormIntegrator *lfi_real,
    if ( lfi_imag ) { plfi->AddDomainIntegrator(lfi_imag); }
 }
 
+void
+ParComplexLinearForm::AddDomainIntegrator(LinearFormIntegrator *lfi_real,
+                                          LinearFormIntegrator *lfi_imag,
+                                          Array<int> &elem_attr_marker)
+{
+   if ( lfi_real ) { plfr->AddDomainIntegrator(lfi_real, elem_attr_marker); }
+   if ( lfi_imag ) { plfi->AddDomainIntegrator(lfi_imag, elem_attr_marker); }
+}
+
 void
 ParComplexLinearForm::AddBoundaryIntegrator(LinearFormIntegrator *lfi_real,
                                             LinearFormIntegrator *lfi_imag)
@@ -1040,6 +1066,14 @@ void ParSesquilinearForm::AddDomainIntegrator(BilinearFormIntegrator *bfi_real,
    if (bfi_imag) { pblfi->AddDomainIntegrator(bfi_imag); }
 }
 
+void ParSesquilinearForm::AddDomainIntegrator(BilinearFormIntegrator *bfi_real,
+                                              BilinearFormIntegrator *bfi_imag,
+                                              Array<int> & elem_marker)
+{
+   if (bfi_real) { pblfr->AddDomainIntegrator(bfi_real, elem_marker); }
+   if (bfi_imag) { pblfi->AddDomainIntegrator(bfi_imag, elem_marker); }
+}
+
 void
 ParSesquilinearForm::AddBoundaryIntegrator(BilinearFormIntegrator *bfi_real,
                                            BilinearFormIntegrator *bfi_imag)

fem/complex_fem.hpp

@@ -128,6 +128,11 @@ public:
    void AddDomainIntegrator(LinearFormIntegrator *lfi_real,
                             LinearFormIntegrator *lfi_imag);
 
+   /// Adds new Domain Integrator, restricted to the given attributes.
+   void AddDomainIntegrator(LinearFormIntegrator *lfi_real,
+                            LinearFormIntegrator *lfi_imag,
+                            Array<int> &elem_attr_marker);
+
    /// Adds new Boundary Integrator.
    void AddBoundaryIntegrator(LinearFormIntegrator *lfi_real,
                               LinearFormIntegrator *lfi_imag);
@@ -260,6 +265,11 @@ public:
    void AddDomainIntegrator(BilinearFormIntegrator *bfi_real,
                             BilinearFormIntegrator *bfi_imag);
 
+   /// Adds new Domain Integrator, restricted to the given attributes.
+   void AddDomainIntegrator(BilinearFormIntegrator *bfi_real,
+                            BilinearFormIntegrator *bfi_imag,
+                            Array<int> &elem_marker);
+
    /// Adds new Boundary Integrator.
    void AddBoundaryIntegrator(BilinearFormIntegrator *bfi_real,
                               BilinearFormIntegrator *bfi_imag);
@@ -464,6 +474,11 @@ public:
    void AddDomainIntegrator(LinearFormIntegrator *lfi_real,
                             LinearFormIntegrator *lfi_imag);
 
+   /// Adds new Domain Integrator, restricted to specific attributes.
+   void AddDomainIntegrator(LinearFormIntegrator *lfi_real,
+                            LinearFormIntegrator *lfi_imag,
+                            Array<int> &elem_attr_marker);
+
    /// Adds new Boundary Integrator.
    void AddBoundaryIntegrator(LinearFormIntegrator *lfi_real,
                               LinearFormIntegrator *lfi_imag);
@@ -598,6 +613,11 @@ public:
    void AddDomainIntegrator(BilinearFormIntegrator *bfi_real,
                             BilinearFormIntegrator *bfi_imag);
 
+   /// Adds new Domain Integrator, restricted to specific attributes.
+   void AddDomainIntegrator(BilinearFormIntegrator *bfi_real,
+                            BilinearFormIntegrator *bfi_imag,
+                            Array<int> &elem_marker);
+
    /// Adds new Boundary Integrator.
    void AddBoundaryIntegrator(BilinearFormIntegrator *bfi_real,
                               BilinearFormIntegrator *bfi_imag);

fem/conduitdatacollection.cpp

@@ -645,7 +645,8 @@ ConduitDataCollection::MeshToBlueprintMesh(Mesh *mesh,
                                            Node &n_mesh,
                                            const std::string &coordset_name,
                                            const std::string &main_topology_name,
-                                           const std::string &boundary_topology_name)
+                                           const std::string &boundary_topology_name,
+                                           const std::string &main_adjset_name)
 {
    int dim = mesh->SpaceDimension();
 
@@ -815,6 +816,83 @@ ConduitDataCollection::MeshToBlueprintMesh(Mesh *mesh,
          bndry_att_vals[i] = mesh->GetBdrAttribute(i);
       }
    }
+
+   ////////////////////////////////////////////
+   // Setup adjsets
+   ////////////////////////////////////////////
+
+#ifdef MFEM_USE_MPI
+   ParMesh *pmesh = dynamic_cast<ParMesh*>(mesh);
+   if (pmesh)
+   {
+      ////////////////////////////////////////////
+      // Setup main adjset
+      ////////////////////////////////////////////
+
+      Node &n_adjset = n_mesh["adjsets"][main_adjset_name];
+
+      n_adjset["association"] = "vertex";
+      n_adjset["topology"] = main_topology_name;
+      n_adjset["groups"].set(DataType::object());
+
+      const GroupTopology &pmesh_gtopo = pmesh->gtopo;
+      const int local_rank = pmesh->GetMyRank();
+      const int num_groups = pmesh_gtopo.NGroups();
+      // NOTE: skip the first group since its the local-only group
+      for (int i = 1; i < num_groups; i++)
+      {
+         const int num_group_nbrs = pmesh_gtopo.GetGroupSize(i);
+         const int *group_nbrs = pmesh_gtopo.GetGroup(i);
+         const int num_group_verts = pmesh->GroupNVertices(i);
+
+         // NOTE: 'neighbor' values are local to this processor, but Blueprint
+         // expects global domain identifiers, so we collapse this layer of
+         // indirection
+         Array<int> group_ranks(num_group_nbrs);
+         std::string group_name = "group";
+         {
+            for (int j = 0; j < num_group_nbrs; j++)
+            {
+               group_ranks[j] = pmesh_gtopo.GetNeighborRank(group_nbrs[j]);
+            }
+            group_ranks.Sort();
+            for (int j = 0; j < num_group_nbrs; j++)
+            {
+               group_name += "_" + std::to_string(group_ranks[j]);
+            }
+
+            // NOTE: Blueprint only wants remote ranks in its neighbor list,
+            // so we remove the local rank after the canonicalized Blueprint
+            // group name is formed
+            group_ranks.DeleteFirst(local_rank);
+         }
+         Node &n_group = n_adjset["groups"][group_name];
+
+         n_group["neighbors"].set(group_ranks.GetData(), group_ranks.Size());
+         n_group["values"].set(DataType::c_int(num_group_verts));
+
+         int_array group_vals = n_group["values"].value();
+         for (int j = 0; j < num_group_verts; j++)
+         {
+            group_vals[j] = pmesh->GroupVertex(i, j);
+         }
+      }
+
+      // NOTE: We don't create an adjset for face neighbor data because
+      // these faces aren't listed in the 'boundary_topology_name' topology
+      // (this topology only covers the faces between 'main_topology_name'
+      // elements and void). To include a face neighbor data adjset, this
+      // function would need to export a topology with either (1) all faces
+      // in the mesh topology or (2) all boundary faces, including neighbors.
+
+      ////////////////////////////////////////////
+      // Setup distributed state
+      ////////////////////////////////////////////
+
+      Node &n_domid = n_mesh["state/domain_id"];
+      n_domid.set(local_rank);
+   }
+#endif
 }
 
 //---------------------------------------------------------------------------//

fem/conduitdatacollection.hpp

@@ -166,7 +166,8 @@ public:
                                    conduit::Node &out,
                                    const std::string &coordset_name = "coords",
                                    const std::string &main_topology_name = "main",
-                                   const std::string &boundary_topology_name = "boundary");
+                                   const std::string &boundary_topology_name = "boundary",
+                                   const std::string &main_adjset_name = "main_adjset");
 
    /// Describes a MFEM grid function using the mesh blueprint
    /** Sets up passed conduit::Node out to describe the given grid function

fem/doftrans.cpp

@@ -85,6 +85,11 @@ void TransformPrimal(const DofTransformation *ran_dof_trans,
    }
 }
 
+void DofTransformation::InvTransformDual(Vector &v) const
+{
+   InvTransformDual(v.GetData());
+}
+
 void TransformDual(const DofTransformation *ran_dof_trans,
                    const DofTransformation *dom_dof_trans,
                    DenseMatrix &elmat)
@@ -195,6 +200,35 @@ void VDofTransformation::TransformDual(double *v) const
    }
 }
 
+void VDofTransformation::InvTransformDual(double *v) const
+{
+   int size = doftrans_->Size();
+
+   if ((Ordering::Type)ordering_ == Ordering::byNODES)
+   {
+      for (int i=0; i<vdim_; i++)
+      {
+         doftrans_->InvTransformDual(&v[i*size]);
+      }
+   }
+   else
+   {
+      Vector vec(size);
+      for (int i=0; i<vdim_; i++)
+      {
+         for (int j=0; j<size; j++)
+         {
+            vec(j) = v[j*vdim_+i];
+         }
+         doftrans_->InvTransformDual(vec);
+         for (int j=0; j<size; j++)
+         {
+            v[j*vdim_+i] = vec(j);
+         }
+      }
+   }
+}
+
 const double ND_DofTransformation::T_data[24] =
 {
    1.0,  0.0,  0.0,  1.0,
@@ -291,6 +325,26 @@ ND_TriDofTransformation::TransformDual(double *v) const
    }
 }
 
+void
+ND_TriDofTransformation::InvTransformDual(double *v) const
+{
+   int nedofs = order; // number of DoFs per edge
+   int nfdofs = order*(order-1); // number of DoFs per face
+
+   double data[2];
+   Vector v2(data, 2);
+
+   // Transform face DoFs
+   for (int f=0; f<1; f++)
+   {
+      for (int i=0; i<nfdofs/2; i++)
+      {
+         v2 = &v[3*nedofs + f*nfdofs + 2*i];
+         T(Fo[f]).MultTranspose(v2, &v[3*nedofs + f*nfdofs + 2*i]);
+      }
+   }
+}
+
 ND_TetDofTransformation::ND_TetDofTransformation(int p)
    : ND_DofTransformation(p*(p + 2)*(p + 3)/2, p)
 {
@@ -355,4 +409,24 @@ ND_TetDofTransformation::TransformDual(double *v) const
    }
 }
 
+void
+ND_TetDofTransformation::InvTransformDual(double *v) const
+{
+   int nedofs = order; // number of DoFs per edge
+   int nfdofs = order*(order-1); // number of DoFs per face
+
+   double data[2];
+   Vector v2(data, 2);
+
+   // Transform face DoFs
+   for (int f=0; f<4; f++)
+   {
+      for (int i=0; i<nfdofs/2; i++)
+      {
+         v2 = &v[6*nedofs + f*nfdofs + 2*i];
+         T(Fo[f]).MultTranspose(v2, &v[6*nedofs + f*nfdofs + 2*i]);
+      }
+   }
+}
+
 } // namespace mfem

fem/doftrans.hpp

@@ -102,6 +102,10 @@ public:
    virtual void TransformDual(double *v) const = 0;
    virtual void TransformDual(Vector &v) const;
 
+   /** Inverse Transform dual DoFs */
+   virtual void InvTransformDual(double *v) const = 0;
+   virtual void InvTransformDual(Vector &v) const;
+
    /** Transform a matrix of dual DoFs entries as computed by a
        BilinearFormIntegrator before summing into a BilinearForm object. */
    virtual void TransformDual(DenseMatrix &V) const;
@@ -183,10 +187,12 @@ public:
    using DofTransformation::TransformPrimal;
    using DofTransformation::InvTransformPrimal;
    using DofTransformation::TransformDual;
+   using DofTransformation::InvTransformDual;
 
    void TransformPrimal(double *v) const;
    void InvTransformPrimal(double *v) const;
    void TransformDual(double *v) const;
+   void InvTransformDual(double *v) const;
 };
 
 /** Abstract base class for high-order Nedelec spaces on elements with
@@ -235,6 +241,8 @@ public:
    void InvTransformPrimal(double *v) const;
 
    void TransformDual(double *v) const;
+
+   void InvTransformDual(double *v) const;
 };
 
 /// DoF transformation implementation for the Nedelec basis on tetrahedra
@@ -246,12 +254,15 @@ public:
    using DofTransformation::TransformPrimal;
    using DofTransformation::InvTransformPrimal;
    using DofTransformation::TransformDual;
+   using DofTransformation::InvTransformDual;
 
    void TransformPrimal(double *v) const;
 
    void InvTransformPrimal(double *v) const;
 
    void TransformDual(double *v) const;
+
+   void InvTransformDual(double *v) const;
 };
 
 /// DoF transformation implementation for the Nedelec basis on wedge elements
@@ -264,12 +275,16 @@ public:
    using DofTransformation::TransformPrimal;
    using DofTransformation::InvTransformPrimal;
    using DofTransformation::TransformDual;
+   using DofTransformation::InvTransformDual;
 
    void TransformPrimal(double *v) const;
 
    void InvTransformPrimal(double *v) const;
 
    void TransformDual(double *v) const;
+
+   void InvTransformDual(double *v) const;
+
 };
 
 } // namespace mfem

fem/fespace.cpp

@@ -561,6 +561,155 @@ void FiniteElementSpace::GetEssentialVDofs(const Array<int> &bdr_attr_is_ess,
    }
 }
 
+void FiniteElementSpace::GetEssentialVDofs(EntitySets::EntityType type,
+                                           int set_index,
+                                           Array<int> &ess_vdofs,
+                                           int component) const
+{
+   Array<int> vdofs, dofs;
+
+   ess_vdofs.SetSize(GetVSize());
+   ess_vdofs = 0;
+
+   MFEM_VERIFY(mesh->ent_sets != NULL, "Mesh object contains no "
+               "entity set information");
+   if (!mesh->ent_sets->SetExists(type, set_index))
+   {
+      ostringstream oss; oss << "Entity set of type \""
+                             << EntitySets::GetTypeName(type)
+                             << "\" and index " << set_index
+                             << " was not found.";
+
+      MFEM_VERIFY(false, oss.str().c_str());
+   }
+
+   set<int>::iterator it;
+   for (it=(*mesh->ent_sets)(type, set_index).begin();
+        it!=(*mesh->ent_sets)(type, set_index).end(); it++)
+   {
+      int ent_index = *it;
+      cout << "collecting vdofs for entity " << ent_index << "->";
+      if (component < 0)
+      {
+         switch (type)
+         {
+            case EntitySets::VERTEX:
+               GetVertexVDofs(ent_index, vdofs);
+               break;
+            case EntitySets::EDGE:
+               GetEdgeVDofs(ent_index, vdofs);
+               break;
+            case EntitySets::FACE:
+               GetFaceVDofs(ent_index, vdofs);
+               break;
+            case EntitySets::ELEMENT:
+               GetElementVDofs(ent_index, vdofs);
+               break;
+            default:
+               mfem_error("GetEssentialVDofs: Invalid entity type");
+         }
+         vdofs.Print(cout);
+         mark_dofs(vdofs, ess_vdofs);
+      }
+      else
+      {
+         switch (type)
+         {
+            case EntitySets::VERTEX:
+               GetVertexDofs(ent_index, dofs);
+               break;
+            case EntitySets::EDGE:
+               GetEdgeDofs(ent_index, dofs);
+               break;
+            case EntitySets::FACE:
+               GetFaceDofs(ent_index, dofs);
+               break;
+            case EntitySets::ELEMENT:
+               GetElementDofs(ent_index, dofs);
+               break;
+            default:
+               mfem_error("GetEssentialDofs: Invalid entity type");
+         }
+         for (int d = 0; d < dofs.Size(); d++)
+         { dofs[d] = DofToVDof(dofs[d], component); }
+         mark_dofs(dofs, ess_vdofs);
+      }
+   }
+
+   if (mesh->ncmesh)
+   {
+      Array<int> es_verts, es_edges, es_faces;
+      mesh->ncmesh->GetEntitySetClosure(type, set_index,
+                                        es_verts, es_edges, es_faces);
+      cout << "returned from get closure" << endl;
+      for (int i = 0; i < es_verts.Size(); i++)
+      {
+         if (es_verts[i] < GetNV())
+         {
+            if (component < 0)
+            {
+               GetVertexVDofs(es_verts[i], vdofs);
+               mark_dofs(vdofs, ess_vdofs);
+            }
+            else
+            {
+               GetVertexDofs(es_verts[i], dofs);
+               for (int d = 0; d < dofs.Size(); d++)
+               { dofs[d] = DofToVDof(dofs[d], component); }
+               mark_dofs(dofs, ess_vdofs);
+            }
+         }
+      }
+      for (int i = 0; i < es_edges.Size(); i++)
+      {
+         if (es_edges[i] < GetMesh()->GetNEdges())
+         {
+            if (component < 0)
+            {
+               GetEdgeVDofs(es_edges[i], vdofs);
+               mark_dofs(vdofs, ess_vdofs);
+            }
+            else
+            {
+               GetEdgeDofs(es_edges[i], dofs);
+               for (int d = 0; d < dofs.Size(); d++)
+               { dofs[d] = DofToVDof(dofs[d], component); }
+               mark_dofs(dofs, ess_vdofs);
+            }
+         }
+      }
+      for (int i = 0; i < es_faces.Size(); i++)
+      {
+         if (es_faces[i] < GetMesh()->GetNFaces())
+         {
+            if (component < 0)
+            {
+               GetFaceVDofs(es_faces[i], vdofs);
+               mark_dofs(vdofs, ess_vdofs);
+            }
+            else
+            {
+               GetFaceDofs(es_faces[i], dofs);
+               for (int d = 0; d < dofs.Size(); d++)
+               { dofs[d] = DofToVDof(dofs[d], component); }
+               mark_dofs(dofs, ess_vdofs);
+            }
+         }
+      }
+   }
+}
+
+void FiniteElementSpace::GetEssentialVDofs(EntitySets::EntityType type,
+                                           const string & set_name,
+                                           Array<int> &ess_vdofs,
+                                           int component) const
+{
+   MFEM_VERIFY(mesh->ent_sets != NULL, "Mesh object contains no "
+               "entity set information");
+   GetEssentialVDofs(type, mesh->ent_sets->GetSetIndex(type, set_name),
+                     ess_vdofs, component);
+}
+
 void FiniteElementSpace::GetEssentialTrueDofs(const Array<int> &bdr_attr_is_ess,
                                               Array<int> &ess_tdof_list,
                                               int component)
@@ -579,6 +728,36 @@ void FiniteElementSpace::GetEssentialTrueDofs(const Array<int> &bdr_attr_is_ess,
    MarkerToList(ess_tdofs, ess_tdof_list);
 }
 
+void FiniteElementSpace::GetEssentialTrueDofs(EntitySets::EntityType type,
+                                              int set_index,
+                                              Array<int> &ess_tdof_list,
+                                              int component)
+{
+   Array<int> ess_vdofs, ess_tdofs;
+   GetEssentialVDofs(type, set_index, ess_vdofs, component);
+   const SparseMatrix *R = GetConformingRestriction();
+   if (!R)
+   {
+      ess_tdofs.MakeRef(ess_vdofs);
+   }
+   else
+   {
+      R->BooleanMult(ess_vdofs, ess_tdofs);
+   }
+   MarkerToList(ess_tdofs, ess_tdof_list);
+}
+
+void FiniteElementSpace::GetEssentialTrueDofs(EntitySets::EntityType type,
+                                              const string & set_name,
+                                              Array<int> &ess_tdof_list,
+                                              int component)
+{
+   MFEM_VERIFY(mesh->ent_sets != NULL, "Mesh object contains no "
+               "entity set information");
+   GetEssentialTrueDofs(type, mesh->ent_sets->GetSetIndex(type, set_name),
+                        ess_tdof_list, component);
+}
+
 void FiniteElementSpace::GetBoundaryTrueDofs(Array<int> &boundary_dofs,
                                              int component)
 {
@@ -1709,8 +1888,7 @@ void FiniteElementSpace::RefinementOperator
             fespace->DofsToVDofs(vd, c_vdofs, old_ndofs);
 
             x.GetSubVector(f_vdofs, subX);
-            old_DoFTrans[geom]->InvTransformPrimal(subX);
-
+            doftrans->InvTransformDual(subX);
             for (int p = 0; p < f_dofs.Size(); ++p)
             {
                if (processed[DecodeDof(f_dofs[p])])
@@ -1719,9 +1897,9 @@ void FiniteElementSpace::RefinementOperator
                }
             }
 
-            lP.MultTranspose(subX, subY);
-            doftrans->TransformPrimal(subY);
-            y.AddElementVector(c_vdofs, subY);
+            lP.MultTranspose(subX, subYt);
+            old_DoFTrans[geom]->TransformDual(subYt);
+            y.AddElementVector(c_vdofs, subYt);
          }
 
          if (vdoftrans)
@@ -1737,6 +1915,122 @@ void FiniteElementSpace::RefinementOperator
    }
 }
 
+namespace internal
+{
+
+// Used in GetCoarseToFineMap() below.
+struct RefType
+{
+   Geometry::Type geom;
+   int num_children;
+   const Pair<int,int> *children;
+
+   RefType(Geometry::Type g, int n, const Pair<int,int> *c)
+      : geom(g), num_children(n), children(c) { }
+
+   bool operator<(const RefType &other) const
+   {
+      if (geom < other.geom) { return true; }
+      if (geom > other.geom) { return false; }
+      if (num_children < other.num_children) { return true; }
+      if (num_children > other.num_children) { return false; }
+      for (int i = 0; i < num_children; i++)
+      {
+         if (children[i].one < other.children[i].one) { return true; }
+         if (children[i].one > other.children[i].one) { return false; }
+      }
+      return false; // everything is equal
+   }
+};
+
+void GetCoarseToFineMap(const CoarseFineTransformations &cft,
+                        const mfem::Mesh &fine_mesh,
+                        Table &coarse_to_fine,
+                        Array<int> &coarse_to_ref_type,
+                        Table &ref_type_to_matrix,
+                        Array<Geometry::Type> &ref_type_to_geom)
+{
+   const int fine_ne = cft.embeddings.Size();
+   int coarse_ne = -1;
+   for (int i = 0; i < fine_ne; i++)
+   {
+      coarse_ne = std::max(coarse_ne, cft.embeddings[i].parent);
+   }
+   coarse_ne++;
+
+   coarse_to_ref_type.SetSize(coarse_ne);
+   coarse_to_fine.SetDims(coarse_ne, fine_ne);
+
+   Array<int> cf_i(coarse_to_fine.GetI(), coarse_ne+1);
+   Array<Pair<int,int> > cf_j(fine_ne);
+   cf_i = 0;
+   for (int i = 0; i < fine_ne; i++)
+   {
+      cf_i[cft.embeddings[i].parent+1]++;
+   }
+   cf_i.PartialSum();
+   MFEM_ASSERT(cf_i.Last() == cf_j.Size(), "internal error");
+   for (int i = 0; i < fine_ne; i++)
+   {
+      const Embedding &e = cft.embeddings[i];
+      cf_j[cf_i[e.parent]].one = e.matrix; // used as sort key below
+      cf_j[cf_i[e.parent]].two = i;
+      cf_i[e.parent]++;
+   }
+   std::copy_backward(cf_i.begin(), cf_i.end()-1, cf_i.end());
+   cf_i[0] = 0;
+   for (int i = 0; i < coarse_ne; i++)
+   {
+      std::sort(&cf_j[cf_i[i]], cf_j.GetData() + cf_i[i+1]);
+   }
+   for (int i = 0; i < fine_ne; i++)
+   {
+      coarse_to_fine.GetJ()[i] = cf_j[i].two;
+   }
+
+   using std::map;
+   using std::pair;
+
+   map<RefType,int> ref_type_map;
+   for (int i = 0; i < coarse_ne; i++)
+   {
+      const int num_children = cf_i[i+1]-cf_i[i];
+      MFEM_ASSERT(num_children > 0, "");
+      const int fine_el = cf_j[cf_i[i]].two;
+      // Assuming the coarse and the fine elements have the same geometry:
+      const Geometry::Type geom = fine_mesh.GetElementBaseGeometry(fine_el);
+      const RefType ref_type(geom, num_children, &cf_j[cf_i[i]]);
+      pair<map<RefType,int>::iterator,bool> res =
+         ref_type_map.insert(
+            pair<const RefType,int>(ref_type, (int)ref_type_map.size()));
+      coarse_to_ref_type[i] = res.first->second;
+   }
+
+   ref_type_to_matrix.MakeI((int)ref_type_map.size());
+   ref_type_to_geom.SetSize((int)ref_type_map.size());
+   for (map<RefType,int>::iterator it = ref_type_map.begin();
+        it != ref_type_map.end(); ++it)
+   {
+      ref_type_to_matrix.AddColumnsInRow(it->second, it->first.num_children);
+      ref_type_to_geom[it->second] = it->first.geom;
+   }
+
+   ref_type_to_matrix.MakeJ();
+   for (map<RefType,int>::iterator it = ref_type_map.begin();
+        it != ref_type_map.end(); ++it)
+   {
+      const RefType &rt = it->first;
+      for (int j = 0; j < rt.num_children; j++)
+      {
+         ref_type_to_matrix.AddConnection(it->second, rt.children[j].one);
+      }
+   }
+   ref_type_to_matrix.ShiftUpI();
+}
+
+} // namespace internal
+
+
 /// TODO: Implement DofTransformation support
 FiniteElementSpace::DerefinementOperator::DerefinementOperator(
    const FiniteElementSpace *f_fes, const FiniteElementSpace *c_fes,
@@ -1778,8 +2072,9 @@ FiniteElementSpace::DerefinementOperator::DerefinementOperator(
    }
 
    Table ref_type_to_matrix;
-   rtrans.GetCoarseToFineMap(*f_mesh, coarse_to_fine, coarse_to_ref_type,
-                             ref_type_to_matrix, ref_type_to_geom);
+   internal::GetCoarseToFineMap(rtrans, *f_mesh, coarse_to_fine,
+                                coarse_to_ref_type, ref_type_to_matrix,
+                                ref_type_to_geom);
    MFEM_ASSERT(coarse_to_fine.Size() == c_fes->GetNE(), "");
 
    const int total_ref_types = ref_type_to_geom.Size();

fem/fespace.hpp

@@ -778,6 +778,19 @@ public:
                                   Array<int> &ess_vdofs,
                                   int component = -1) const;
 
+   /** Mark degrees of freedom associated with the entity set with the
+       specified entity type and set index. */
+   virtual void GetEssentialVDofs(EntitySets::EntityType type, int set_index,
+                                  Array<int> &ess_vdofs,
+                                  int component = -1) const;
+
+   /** Mark degrees of freedom associated with the entity set with the
+       specified entity type and set index. */
+   virtual void GetEssentialVDofs(EntitySets::EntityType type,
+                                  const std::string & set_name,
+                                  Array<int> &ess_vdofs,
+                                  int component = -1) const;
+
    /** @brief Get a list of essential true dofs, ess_tdof_list, corresponding to the
        boundary attributes marked in the array bdr_attr_is_ess.
        For spaces with 'vdim' > 1, the 'component' parameter can be used
@@ -786,6 +799,19 @@ public:
                                      Array<int> &ess_tdof_list,
                                      int component = -1);
 
+   /** Get a list of essential true dofs, ess_tdof_list, corresponding to the
+       entity set specified by the given entity type and set index. */
+   virtual void GetEssentialTrueDofs(EntitySets::EntityType type, int set_index,
+                                     Array<int> &ess_tdof_list,
+                                     int component = -1);
+
+   /** Get a list of essential true dofs, ess_tdof_list, corresponding to the
+       entity set specified by the given entity type and set name. */
+   virtual void GetEssentialTrueDofs(EntitySets::EntityType type,
+                                     const std::string & set_name,
+                                     Array<int> &ess_tdof_list,
+                                     int component = -1);
+
    /** @brief Get a list of all boundary true dofs, @a boundary_dofs. For spaces
        with 'vdim' > 1, the 'component' parameter can be used to restricts the
        marked tDOFs to the specified component. Equivalent to

fem/nonlinearform.cpp

@@ -96,6 +96,7 @@ double NonlinearForm::GetGridFunctionEnergy(const Vector &x) const
    Vector el_x;
    const FiniteElement *fe;
    ElementTransformation *T;
+   DofTransformation *doftrans;
    double energy = 0.0;
 
    if (dnfi.Size())
@@ -103,9 +104,10 @@ double NonlinearForm::GetGridFunctionEnergy(const Vector &x) const
       for (int i = 0; i < fes->GetNE(); i++)
       {
          fe = fes->GetFE(i);
-         fes->GetElementVDofs(i, vdofs);
+         doftrans = fes->GetElementVDofs(i, vdofs);
          T = fes->GetElementTransformation(i);
          x.GetSubVector(vdofs, el_x);
+         if (doftrans) {doftrans->InvTransformPrimal(el_x); }
          for (int k = 0; k < dnfi.Size(); k++)
          {
             energy += dnfi[k]->GetElementEnergy(*fe, *T, el_x);
@@ -166,6 +168,7 @@ void NonlinearForm::Mult(const Vector &x, Vector &y) const
    Vector el_x, el_y;
    const FiniteElement *fe;
    ElementTransformation *T;
+   DofTransformation *doftrans;
    Mesh *mesh = fes->GetMesh();
 
    py = 0.0;
@@ -175,12 +178,14 @@ void NonlinearForm::Mult(const Vector &x, Vector &y) const
       for (int i = 0; i < fes->GetNE(); i++)
       {
          fe = fes->GetFE(i);
-         fes->GetElementVDofs(i, vdofs);
+         doftrans = fes->GetElementVDofs(i, vdofs);
          T = fes->GetElementTransformation(i);
          px.GetSubVector(vdofs, el_x);
+         if (doftrans) {doftrans->InvTransformPrimal(el_x); }
          for (int k = 0; k < dnfi.Size(); k++)
          {
             dnfi[k]->AssembleElementVector(*fe, *T, el_x, el_y);
+            if (doftrans) {doftrans->TransformDual(el_y); }
             py.AddElementVector(vdofs, el_y);
          }
       }
@@ -302,6 +307,7 @@ Operator &NonlinearForm::GetGradient(const Vector &x) const
    DenseMatrix elmat;
    const FiniteElement *fe;
    ElementTransformation *T;
+   DofTransformation *doftrans;
    Mesh *mesh = fes->GetMesh();
    const Vector &px = Prolongate(x);
 
@@ -319,12 +325,14 @@ Operator &NonlinearForm::GetGradient(const Vector &x) const
       for (int i = 0; i < fes->GetNE(); i++)
       {
          fe = fes->GetFE(i);
-         fes->GetElementVDofs(i, vdofs);
+         doftrans = fes->GetElementVDofs(i, vdofs);
          T = fes->GetElementTransformation(i);
          px.GetSubVector(vdofs, el_x);
+         if (doftrans) {doftrans->InvTransformPrimal(el_x); }
          for (int k = 0; k < dnfi.Size(); k++)
          {
             dnfi[k]->AssembleElementGrad(*fe, *T, el_x, elmat);
+            if (doftrans) { doftrans->TransformDual(elmat); }
             Grad->AddSubMatrix(vdofs, vdofs, elmat, skip_zeros);
             // Grad->AddSubMatrix(vdofs, vdofs, elmat, 1);
          }
@@ -583,6 +591,7 @@ double BlockNonlinearForm::GetEnergyBlocked(const BlockVector &bx) const
    Array<const Vector *> el_x_const(fes.Size());
    Array<const FiniteElement *> fe(fes.Size());
    ElementTransformation *T;
+   DofTransformation *doftrans;
    double energy = 0.0;
 
    for (int i=0; i<fes.Size(); ++i)
@@ -598,8 +607,9 @@ double BlockNonlinearForm::GetEnergyBlocked(const BlockVector &bx) const
          for (int s=0; s<fes.Size(); ++s)
          {
             fe[s] = fes[s]->GetFE(i);
-            fes[s]->GetElementVDofs(i, *vdofs[s]);
+            doftrans = fes[s]->GetElementVDofs(i, *vdofs[s]);
             bx.GetBlock(s).GetSubVector(*vdofs[s], *el_x[s]);
+            if (doftrans) {doftrans->InvTransformPrimal(*el_x[s]); }
          }
 
          for (int k = 0; k < dnfi.Size(); ++k)
@@ -645,6 +655,7 @@ void BlockNonlinearForm::MultBlocked(const BlockVector &bx,
    Array<const FiniteElement *> fe(fes.Size());
    Array<const FiniteElement *> fe2(fes.Size());
    ElementTransformation *T;
+   Array<DofTransformation *> doftrans(fes.Size()); doftrans = nullptr;
 
    by.UseDevice(true);
    by = 0.0;
@@ -664,9 +675,10 @@ void BlockNonlinearForm::MultBlocked(const BlockVector &bx,
          T = fes[0]->GetElementTransformation(i);
          for (int s = 0; s < fes.Size(); ++s)
          {
-            fes[s]->GetElementVDofs(i, *(vdofs[s]));
+            doftrans[s] = fes[s]->GetElementVDofs(i, *(vdofs[s]));
             fe[s] = fes[s]->GetFE(i);
             bx.GetBlock(s).GetSubVector(*(vdofs[s]), *el_x[s]);
+            if (doftrans[s]) {doftrans[s]->InvTransformPrimal(*el_x[s]); }
          }
 
          for (int k = 0; k < dnfi.Size(); ++k)
@@ -677,6 +689,7 @@ void BlockNonlinearForm::MultBlocked(const BlockVector &bx,
             for (int s=0; s<fes.Size(); ++s)
             {
                if (el_y[s]->Size() == 0) { continue; }
+               if (doftrans[s]) {doftrans[s]->TransformDual(*el_y[s]); }
                by.GetBlock(s).AddElementVector(*(vdofs[s]), *el_y[s]);
             }
          }
@@ -844,6 +857,7 @@ void BlockNonlinearForm::ComputeGradientBlocked(const BlockVector &bx) const
    Array<const FiniteElement *>fe(fes.Size());
    Array<const FiniteElement *>fe2(fes.Size());
    ElementTransformation * T;
+   Array<DofTransformation *> doftrans(fes.Size()); doftrans = nullptr;
 
    for (int i=0; i<fes.Size(); ++i)
    {
@@ -880,8 +894,9 @@ void BlockNonlinearForm::ComputeGradientBlocked(const BlockVector &bx) const
          for (int s = 0; s < fes.Size(); ++s)
          {
             fe[s] = fes[s]->GetFE(i);
-            fes[s]->GetElementVDofs(i, *vdofs[s]);
+            doftrans[s] = fes[s]->GetElementVDofs(i, *vdofs[s]);
             bx.GetBlock(s).GetSubVector(*vdofs[s], *el_x[s]);
+            if (doftrans[s]) {doftrans[s]->InvTransformPrimal(*el_x[s]); }
          }
 
          for (int k = 0; k < dnfi.Size(); ++k)
@@ -893,6 +908,10 @@ void BlockNonlinearForm::ComputeGradientBlocked(const BlockVector &bx) const
                for (int l=0; l<fes.Size(); ++l)
                {
                   if (elmats(j,l)->Height() == 0) { continue; }
+                  if (doftrans[j] || doftrans[l])
+                  {
+                     TransformDual(doftrans[j], doftrans[l], *elmats(j,l));
+                  }
                   Grads(j,l)->AddSubMatrix(*vdofs[j], *vdofs[l],
                                            *elmats(j,l), skip_zeros);
                }

fem/pfespace.cpp

@@ -23,6 +23,8 @@
 #include <limits>
 #include <list>
 
+using namespace std;
+
 namespace mfem
 {
 
@@ -1018,6 +1020,30 @@ void ParFiniteElementSpace::GetEssentialVDofs(const Array<int> &bdr_attr_is_ess,
    }
 }
 
+void ParFiniteElementSpace::GetEssentialVDofs(EntitySets::EntityType type,
+                                              int set_index,
+                                              Array<int> &ess_dofs,
+                                              int component) const
+{
+   FiniteElementSpace::GetEssentialVDofs(type, set_index, ess_dofs, component);
+
+   if (Conforming())
+   {
+      // Make sure that processors without boundary elements mark
+      // their boundary dofs (if they have any).
+      Synchronize(ess_dofs);
+   }
+}
+
+void ParFiniteElementSpace::GetEssentialVDofs(EntitySets::EntityType type,
+                                              const string & set_name,
+                                              Array<int> &ess_vdofs,
+                                              int component) const
+{
+   GetEssentialVDofs(type, pmesh->ent_sets->GetSetIndex(type, set_name),
+                     ess_vdofs, component);
+}
+
 void ParFiniteElementSpace::GetEssentialTrueDofs(const Array<int>
                                                  &bdr_attr_is_ess,
                                                  Array<int> &ess_tdof_list,
@@ -1047,6 +1073,27 @@ void ParFiniteElementSpace::GetEssentialTrueDofs(const Array<int>
    MarkerToList(true_ess_dofs, ess_tdof_list);
 }
 
+void ParFiniteElementSpace::GetEssentialTrueDofs(EntitySets::EntityType type,
+                                                 int set_index,
+                                                 Array<int> &ess_tdof_list,
+                                                 int component)
+{
+   Array<int> ess_dofs, true_ess_dofs;
+
+   GetEssentialVDofs(type, set_index, ess_dofs, component);
+   GetRestrictionMatrix()->BooleanMult(ess_dofs, true_ess_dofs);
+   MarkerToList(true_ess_dofs, ess_tdof_list);
+}
+
+void ParFiniteElementSpace::GetEssentialTrueDofs(EntitySets::EntityType type,
+                                                 const string & set_name,
+                                                 Array<int> &ess_tdof_list,
+                                                 int component)
+{
+   GetEssentialTrueDofs(type, pmesh->ent_sets->GetSetIndex(type, set_name),
+                        ess_tdof_list, component);
+}
+
 int ParFiniteElementSpace::GetLocalTDofNumber(int ldof) const
 {
    if (Nonconforming())

fem/pfespace.hpp

@@ -355,12 +355,38 @@ public:
                                   Array<int> &ess_dofs,
                                   int component = -1) const;
 
+   /** Mark degrees of freedom associated with the entity set with the
+       specified entity type and set index. */
+   virtual void GetEssentialVDofs(EntitySets::EntityType type, int set_index,
+                                  Array<int> &ess_vdofs,
+                                  int component = -1) const;
+
+   /** Mark degrees of freedom associated with the entity set with the
+       specified entity type and set index. */
+   virtual void GetEssentialVDofs(EntitySets::EntityType type,
+                                  const std::string & set_name,
+                                  Array<int> &ess_vdofs,
+                                  int component = -1) const;
+
    /** Get a list of essential true dofs, ess_tdof_list, corresponding to the
        boundary attributes marked in the array bdr_attr_is_ess. */
    virtual void GetEssentialTrueDofs(const Array<int> &bdr_attr_is_ess,
                                      Array<int> &ess_tdof_list,
                                      int component = -1);
 
+   /** Get a list of essential true dofs, ess_tdof_list, corresponding to the
+       entity set specified by the given entity type and set index. */
+   virtual void GetEssentialTrueDofs(EntitySets::EntityType type, int set_index,
+                                     Array<int> &ess_tdof_list,
+                                     int component = -1);
+
+   /** Get a list of essential true dofs, ess_tdof_list, corresponding to the
+       entity set specified by the given entity type and set name. */
+   virtual void GetEssentialTrueDofs(EntitySets::EntityType type,
+                                     const std::string & set_name,
+                                     Array<int> &ess_tdof_list,
+                                     int component = -1);
+
    /** If the given ldof is owned by the current processor, return its local
        tdof number, otherwise return -1 */
    int GetLocalTDofNumber(int ldof) const;

fem/staticcond.cpp

@@ -40,8 +40,6 @@ StaticCondensation::StaticCondensation(FiniteElementSpace *fespace)
 #endif
    S = S_e = NULL;
    symm = false;
-   A_data.Reset();
-   A_ipiv.Reset();
 
    Array<int> vdofs;
    const int NE = fes->GetNE();

fem/tmop.cpp

@@ -2326,6 +2326,8 @@ TMOP_Integrator::~TMOP_Integrator()
 {
    delete lim_func;
    delete zeta;
+   delete sigma;
+   delete sigma_bar;
    for (int i = 0; i < ElemDer.Size(); i++)
    {
       delete ElemDer[i];
@@ -2393,6 +2395,87 @@ void TMOP_Integrator::EnableAdaptiveLimiting(const ParGridFunction &z0,
 }
 #endif
 
+void TMOP_Integrator::EnableSurfaceFitting(const GridFunction &s0,
+                                           const Array<bool> &smarker,
+                                           Coefficient &coeff,
+                                           AdaptivityEvaluator &ae)
+{
+   delete sigma;
+   sigma = new GridFunction(s0);
+   sigma_marker = &smarker;
+   coeff_sigma = &coeff;
+   sigma_eval = &ae;
+
+   // Compute the restricted sigma.
+   delete sigma_bar;
+   sigma_bar = new GridFunction(*sigma);
+   for (int i = 0; i < sigma_marker->Size(); i++)
+   {
+      if ((*sigma_marker)[i] == false) { (*sigma_bar)(i) = 0.0; }
+   }
+
+   sigma_eval->SetSerialMetaInfo(*s0.FESpace()->GetMesh(),
+                                 *s0.FESpace()->FEColl(), 1);
+   sigma_eval->SetInitialField
+   (*sigma->FESpace()->GetMesh()->GetNodes(), *sigma);
+}
+
+#ifdef MFEM_USE_MPI
+void TMOP_Integrator::EnableSurfaceFitting(const ParGridFunction &s0,
+                                           const Array<bool> &smarker,
+                                           Coefficient &coeff,
+                                           AdaptivityEvaluator &ae)
+{
+   delete sigma;
+   sigma = new GridFunction(s0);
+   sigma_marker = &smarker;
+   coeff_sigma = &coeff;
+   sigma_eval = &ae;
+
+   // Compute the restricted sigma.
+   delete sigma_bar;
+   sigma_bar = new GridFunction(*sigma);
+   for (int i = 0; i < sigma_marker->Size(); i++)
+   {
+      if ((*sigma_marker)[i] == false) { (*sigma_bar)(i) = 0.0; }
+   }
+
+   sigma_eval->SetParMetaInfo(*s0.ParFESpace()->GetParMesh(),
+                              *s0.ParFESpace()->FEColl(), 1);
+   sigma_eval->SetInitialField
+   (*sigma->FESpace()->GetMesh()->GetNodes(), *sigma);
+}
+#endif
+
+void TMOP_Integrator::GetSurfaceFittingErrors(double &err_avg, double &err_max)
+{
+   MFEM_VERIFY(sigma, "Surface fitting has not been enabled.");
+
+   int loc_cnt = 0;
+   double loc_max = 0.0, loc_sum = 0.0;
+   for (int i = 0; i < sigma_marker->Size(); i++)
+   {
+      if ((*sigma_marker)[i] == true)
+      {
+         loc_cnt++;
+         loc_max  = std::max(loc_max, std::abs((*sigma_bar)(i)));
+         loc_sum += std::abs((*sigma_bar)(i));
+      }
+   }
+   err_avg = loc_sum / loc_cnt;
+   err_max = loc_max;
+
+#ifdef MFEM_USE_MPI
+   if (targetC->Parallel() == false) { return; }
+   int glob_cnt;
+   MPI_Comm comm = targetC->GetComm();
+   MPI_Allreduce(&loc_max, &err_max, 1, MPI_DOUBLE, MPI_MAX, comm);
+   MPI_Allreduce(&loc_cnt, &glob_cnt, 1, MPI_INT, MPI_SUM, comm);
+   MPI_Allreduce(&loc_sum, &err_avg, 1, MPI_DOUBLE, MPI_SUM, comm);
+   err_avg = err_avg / glob_cnt;
+#endif
+}
+
 void TMOP_Integrator::UpdateAfterMeshTopologyChange()
 {
    if (zeta)
@@ -2419,16 +2502,19 @@ void TMOP_Integrator::ParUpdateAfterMeshTopologyChange()
 }
 #endif
 
-
 double TMOP_Integrator::GetElementEnergy(const FiniteElement &el,
                                          ElementTransformation &T,
                                          const Vector &elfun)
 {
    const int dof = el.GetDof(), dim = el.GetDim();
+   const int el_id = T.ElementNo;
    double energy;
 
-   // No adaptive limiting terms if this is a FD computation.
+   // No adaptive limiting / surface fitting terms if the function is called
+   // as part of a FD derivative computation (because we include the exact
+   // derivatives of these terms in FD computations).
    const bool adaptive_limiting = (zeta && fd_call_flag == false);
+   const bool surface_fit = (sigma && fd_call_flag == false);
 
    DSh.SetSize(dof, dim);
    Jrt.SetSize(dim);
@@ -2440,7 +2526,7 @@ double TMOP_Integrator::GetElementEnergy(const FiniteElement &el,
 
    energy = 0.0;
    DenseTensor Jtr(dim, dim, ir.GetNPoints());
-   targetC->ComputeElementTargets(T.ElementNo, el, ir, elfun, Jtr);
+   targetC->ComputeElementTargets(el_id, el, ir, elfun, Jtr);
 
    // Limited case.
    Vector shape, p, p0, d_vals;
@@ -2453,11 +2539,11 @@ double TMOP_Integrator::GetElementEnergy(const FiniteElement &el,
       pos0.SetSize(dof, dim);
       Vector pos0V(pos0.Data(), dof * dim);
       Array<int> pos_dofs;
-      nodes0->FESpace()->GetElementVDofs(T.ElementNo, pos_dofs);
+      nodes0->FESpace()->GetElementVDofs(el_id, pos_dofs);
       nodes0->GetSubVector(pos_dofs, pos0V);
       if (lim_dist)
       {
-         lim_dist->GetValues(T.ElementNo, ir, d_vals);
+         lim_dist->GetValues(el_id, ir, d_vals);
       }
       else
       {
@@ -2467,11 +2553,11 @@ double TMOP_Integrator::GetElementEnergy(const FiniteElement &el,
 
    // Define ref->physical transformation, when a Coefficient is specified.
    IsoparametricTransformation *Tpr = NULL;
-   if (coeff1 || coeff0 || adaptive_limiting)
+   if (coeff1 || coeff0 || adaptive_limiting || surface_fit)
    {
       Tpr = new IsoparametricTransformation;
       Tpr->SetFE(&el);
-      Tpr->ElementNo = T.ElementNo;
+      Tpr->ElementNo = el_id;
       Tpr->ElementType = ElementTransformation::ELEMENT;
       Tpr->Attribute = T.Attribute;
       Tpr->GetPointMat().Transpose(PMatI); // PointMat = PMatI^T
@@ -2487,13 +2573,17 @@ double TMOP_Integrator::GetElementEnergy(const FiniteElement &el,
    Vector zeta_q, zeta0_q;
    if (adaptive_limiting)
    {
-      zeta->GetValues(T.ElementNo, ir, zeta_q);
-      zeta_0->GetValues(T.ElementNo, ir, zeta0_q);
+      zeta->GetValues(el_id, ir, zeta_q);
+      zeta_0->GetValues(el_id, ir, zeta0_q);
    }
 
+   Vector sigma_bar_q;
+   if (surface_fit) { sigma_bar->GetValues(el_id, ir, sigma_bar_q); }
+
    for (int i = 0; i < ir.GetNPoints(); i++)
    {
       const IntegrationPoint &ip = ir.IntPoint(i);
+
       const DenseMatrix &Jtr_i = Jtr(i);
       metric->SetTargetJacobian(Jtr_i);
       CalcInverse(Jtr_i, Jrt);
@@ -2516,16 +2606,24 @@ double TMOP_Integrator::GetElementEnergy(const FiniteElement &el,
                 coeff0->Eval(*Tpr, ip);
       }
 
+      // Contribution from the adaptive limiting term.
       if (adaptive_limiting)
       {
          const double diff = zeta_q(i) - zeta0_q(i);
          val += coeff_zeta->Eval(*Tpr, ip) * lim_normal * diff * diff;
       }
 
+      // Contribution from the surface fitting term.
+      if (surface_fit)
+      {
+         val += coeff_sigma->Eval(*Tpr, ip) * sigma_normal *
+                sigma_bar_q(i) * sigma_bar_q(i);
+      }
+
       energy += weight * val;
    }
-   delete Tpr;
 
+   delete Tpr;
    return energy;
 }
 
@@ -2747,7 +2845,7 @@ void TMOP_Integrator::AssembleElementVectorExact(const FiniteElement &el,
 
    // Define ref->physical transformation, when a Coefficient is specified.
    IsoparametricTransformation *Tpr = NULL;
-   if (coeff1 || coeff0 || zeta || exact_action)
+   if (coeff1 || coeff0 || zeta || sigma || exact_action)
    {
       Tpr = new IsoparametricTransformation;
       Tpr->SetFE(&el);
@@ -2829,7 +2927,8 @@ void TMOP_Integrator::AssembleElementVectorExact(const FiniteElement &el,
       }
    }
 
-   if (zeta) { AssembleElemVecAdaptLim(el, weights, *Tpr, ir, PMatO); }
+   if (zeta) { AssembleElemVecAdaptLim(el, *Tpr, ir, weights, PMatO); }
+   if (sigma) { AssembleElemVecSurfFit(el, *Tpr, ir, weights, PMatO); }
 
    delete Tpr;
 }
@@ -2881,7 +2980,7 @@ void TMOP_Integrator::AssembleElementGradExact(const FiniteElement &el,
 
    // Define ref->physical transformation, when a Coefficient is specified.
    IsoparametricTransformation *Tpr = NULL;
-   if (coeff1 || coeff0 || zeta)
+   if (coeff1 || coeff0 || zeta || sigma)
    {
       Tpr = new IsoparametricTransformation;
       Tpr->SetFE(&el);
@@ -2935,21 +3034,20 @@ void TMOP_Integrator::AssembleElementGradExact(const FiniteElement &el,
       }
    }
 
-   if (zeta) { AssembleElemGradAdaptLim(el, weights, *Tpr, ir, elmat); }
+   if (zeta) { AssembleElemGradAdaptLim(el, *Tpr, ir, weights, elmat); }
+   if (sigma) { AssembleElemGradSurfFit(el, *Tpr, ir, weights, elmat); }
 
    delete Tpr;
 }
 
 void TMOP_Integrator::AssembleElemVecAdaptLim(const FiniteElement &el,
-                                              const Vector &weights,
                                               IsoparametricTransformation &Tpr,
                                               const IntegrationRule &ir,
+                                              const Vector &weights,
                                               DenseMatrix &mat)
 {
-   if (zeta == NULL) { return; }
-
-   const int dof = el.GetDof(), dim = el.GetDim();
-   Vector shape(dof), zeta_e, zeta_q, zeta0_q;
+   const int dof = el.GetDof(), dim = el.GetDim(), nqp = weights.Size();
+   Vector shape(dof), zeta_e, zeta_q, zeta0_q(nqp);
 
    Array<int> dofs;
    zeta->FESpace()->GetElementDofs(Tpr.ElementNo, dofs);
@@ -2967,7 +3065,6 @@ void TMOP_Integrator::AssembleElemVecAdaptLim(const FiniteElement &el,
 
    Vector zeta_grad_q(dim);
 
-   const int nqp = weights.Size();
    for (int q = 0; q < nqp; q++)
    {
       const IntegrationPoint &ip = ir.IntPoint(q);
@@ -2980,15 +3077,13 @@ void TMOP_Integrator::AssembleElemVecAdaptLim(const FiniteElement &el,
 }
 
 void TMOP_Integrator::AssembleElemGradAdaptLim(const FiniteElement &el,
-                                               const Vector &weights,
                                                IsoparametricTransformation &Tpr,
                                                const IntegrationRule &ir,
+                                               const Vector &weights,
                                                DenseMatrix &mat)
 {
-   if (zeta == NULL) { return; }
-
-   const int dof = el.GetDof(), dim = el.GetDim();
-   Vector shape(dof), zeta_e, zeta_q, zeta0_q;
+   const int dof = el.GetDof(), dim = el.GetDim(), nqp = weights.Size();
+   Vector shape(dof), zeta_e, zeta_q, zeta0_q(nqp);
 
    Array<int> dofs;
    zeta->FESpace()->GetElementDofs(Tpr.ElementNo, dofs);
@@ -3014,7 +3109,6 @@ void TMOP_Integrator::AssembleElemGradAdaptLim(const FiniteElement &el,
    Vector zeta_grad_q(dim);
    DenseMatrix zeta_grad_grad_q(dim, dim);
 
-   const int nqp = weights.Size();
    for (int q = 0; q < nqp; q++)
    {
       const IntegrationPoint &ip = ir.IntPoint(q);
@@ -3043,6 +3137,169 @@ void TMOP_Integrator::AssembleElemGradAdaptLim(const FiniteElement &el,
    }
 }
 
+void TMOP_Integrator::AssembleElemVecSurfFit(const FiniteElement &el_x,
+                                             IsoparametricTransformation &Tpr,
+                                             const IntegrationRule &ir_quad,
+                                             const Vector &weights,
+                                             DenseMatrix &mat)
+{
+   const int el_id = Tpr.ElementNo;
+   const FiniteElement &el_s = *sigma->FESpace()->GetFE(el_id);
+
+   const int dof_x = el_x.GetDof(), dim = el_x.GetDim(),
+             dof_s = el_s.GetDof(), nqp = ir_quad.GetNPoints();
+
+   Vector sigma_e, sigma_bar_e;
+   Vector sigma_bar_q;
+   Array<int> dofs;
+   sigma->FESpace()->GetElementDofs(el_id, dofs);
+   sigma->GetSubVector(dofs, sigma_e);
+   sigma_bar->GetSubVector(dofs, sigma_bar_e);
+   sigma_bar->GetValues(el_id, ir_quad, sigma_bar_q);
+
+   // Project the gradient of sigma in the same space.
+   // The FE coefficients of the gradient go in sigma_grad_e.
+   DenseMatrix sigma_grad_e(dof_s, dim);
+   DenseMatrix grad_phys; // This will be (dof x dim, dof).
+   el_s.ProjectGrad(el_s, Tpr, grad_phys);
+   Vector grad_ptr(sigma_grad_e.GetData(), dof_s * dim);
+   grad_phys.Mult(sigma_e, grad_ptr);
+
+   // Gradient of sigma_bar.
+   DenseMatrix sigma_bar_grad_e(dof_s, dim);
+   Vector ptr(sigma_bar_grad_e.GetData(), dof_s * dim);
+   grad_phys.Mult(sigma_bar_e, ptr);
+
+   Vector shape_x(dof_x), shape_s(dof_s), grad_q(dim);
+
+   for (int q = 0; q < nqp; q++)
+   {
+      const IntegrationPoint &ip = ir_quad.IntPoint(q);
+      Tpr.SetIntPoint(&ip);
+      el_s.CalcShape(ip, shape_s);
+
+      // Grad of sigma_bar at the current quad point.
+      sigma_bar_grad_e.MultTranspose(shape_s, grad_q);
+
+      for (int s = 0; s < dof_s; s++)
+      {
+         if ((*sigma_marker)[dofs[s]] == false) { continue; }
+
+         for (int d = 0; d < dim; d++)
+         {
+            // Grad of sigma must be taken at the active DOFs.
+            grad_q(d) += sigma_grad_e(s, d) * shape_s(s);
+         }
+      }
+
+      grad_q *= 2.0 * sigma_normal * coeff_sigma->Eval(Tpr, ip) *
+                weights(q) * sigma_bar_q(q);
+
+      el_x.CalcShape(ip, shape_x);
+      AddMultVWt(shape_x, grad_q, mat);
+   }
+}
+
+void TMOP_Integrator::AssembleElemGradSurfFit(const FiniteElement &el_x,
+                                              IsoparametricTransformation &Tpr,
+                                              const IntegrationRule &ir_quad,
+                                              const Vector &weights,
+                                              DenseMatrix &mat)
+{
+   const int el_id = Tpr.ElementNo, nqp = ir_quad.GetNPoints();
+   const FiniteElement &el_s = *sigma->FESpace()->GetFE(el_id);
+
+   const int dof_x = el_x.GetDof(), dim = el_x.GetDim(),
+             dof_s = el_s.GetDof();
+
+   Vector sigma_e, sigma_bar_e;
+   Vector sigma_bar_q;
+
+   Array<int> dofs;
+   sigma->FESpace()->GetElementDofs(el_id, dofs);
+   sigma->GetSubVector(dofs, sigma_e);
+   sigma_bar->GetSubVector(dofs, sigma_bar_e);
+   sigma_bar->GetValues(el_id, ir_quad, sigma_bar_q);
+
+   // Project the gradient of sigma in the same space.
+   // The FE coefficients of the gradient go in sigma_grad_e.
+   DenseMatrix sigma_grad_e(dof_s, dim);
+   DenseMatrix grad_phys; // This will be (dof x dim, dof).
+   el_s.ProjectGrad(el_s, Tpr, grad_phys);
+   Vector grad_ptr(sigma_grad_e.GetData(), dof_s * dim);
+   grad_phys.Mult(sigma_e, grad_ptr);
+
+   // Gradient of sigma_bar.
+   DenseMatrix sigma_bar_grad_e(dof_s, dim);
+   Vector ptr(sigma_bar_grad_e.GetData(), dof_s * dim);
+   grad_phys.Mult(sigma_bar_e, ptr);
+
+   // Project the gradient of each gradient of sigma in the same space.
+   // The FE coefficients of the second derivatives go in sigma_grad_grad_e.
+   DenseMatrix sigma_grad_grad_e(dof_s * dim, dim);
+   Mult(grad_phys, sigma_grad_e, sigma_grad_grad_e);
+
+   // Project the gradient of each gradient of sigma in the same space.
+   // The FE coefficients of the second derivatives go in sigma_grad_grad_e.
+   DenseMatrix sigma_bar_grad_grad_e(dof_s * dim, dim);
+   Mult(grad_phys, sigma_bar_grad_e, sigma_bar_grad_grad_e);
+   // Reshape to be more convenient later (no change in the data).
+   sigma_bar_grad_grad_e.SetSize(dof_s, dim * dim);
+
+   DenseMatrix sigma_bar_grad_grad_q(dim, dim);
+
+   Vector shape_x(dof_x), shape_s(dof_s), sigma_bar_grad_q(dim);
+   DenseMatrix dshape_s(dof_s, dim);
+
+   for (int q = 0; q < nqp; q++)
+   {
+      const IntegrationPoint &ip = ir_quad.IntPoint(q);
+      Tpr.SetIntPoint(&ip);
+      el_s.CalcShape(ip, shape_s);
+      el_x.CalcShape(ip, shape_x);
+      // We could reuse grad_phys, but this is more accurate.
+      el_s.CalcPhysDShape(Tpr, dshape_s);
+
+      // Grad of sigma_bar at the current quad point.
+      sigma_bar_grad_e.MultTranspose(shape_s, sigma_bar_grad_q);
+
+      // Grad-grad of sigma_bar at the current quad point.
+      Vector gg_ptr(sigma_bar_grad_grad_q.GetData(), dim * dim);
+      sigma_bar_grad_grad_e.MultTranspose(shape_s, gg_ptr);
+
+      // Loops over the local matrix.
+      const double w = 2.0 * sigma_normal *
+                       coeff_sigma->Eval(Tpr, ip) * weights(q);
+      for (int i = 0; i < dof_x * dim; i++)
+      {
+         const int idof = i % dof_x, idim = i / dof_x;
+         for (int j = 0; j <= i; j++)
+         {
+            const int jdof = j % dof_x, jdim = j / dof_x;
+
+            double Di = sigma_bar_grad_q(idim),
+                   Dj = sigma_bar_grad_q(jdim),
+                   DD = sigma_bar_grad_grad_q(idim, jdim);
+            for (int s = 0; s < dof_s; s++)
+            {
+               if ((*sigma_marker)[dofs[s]] == false) { continue; }
+
+               Di += sigma_grad_e(s, idim) * shape_s(s);
+               Dj += sigma_grad_e(s, jdim) * shape_s(s);
+               DD += sigma_grad_e(s, idim) * dshape_s(s, jdim) +
+                     sigma_grad_grad_e(dof_s * idim + s, jdim) * shape_s(s) +
+                     sigma_grad_e(s, jdim) * dshape_s(s, idim);
+            }
+            const double entry = w * (Di * Dj + sigma_bar_q(q) * DD) *
+                                 shape_x(idof) * shape_x(jdof);
+
+            mat(i, j) += entry;
+            if (i != j) { mat(j, i) += entry; }
+         }
+      }
+   }
+}
+
 double TMOP_Integrator::GetFDDerivative(const FiniteElement &el,
                                         ElementTransformation &T,
                                         Vector &elfun, const int dofidx,
@@ -3103,8 +3360,8 @@ void TMOP_Integrator::AssembleElementVectorFD(const FiniteElement &el,
    }
    fd_call_flag = false;
 
-   // Contributions from adaptive limiting (exact derivatives).
-   if (zeta)
+   // Contributions from adaptive limiting, surface fitting (exact derivatives).
+   if (zeta || sigma)
    {
       const IntegrationRule &ir = ActionIntegrationRule(el);
       const int nqp = ir.GetNPoints();
@@ -3125,7 +3382,8 @@ void TMOP_Integrator::AssembleElementVectorFD(const FiniteElement &el,
       }
 
       PMatO.UseExternalData(elvect.GetData(), dof, dim);
-      AssembleElemVecAdaptLim(el, weights, Tpr, ir, PMatO);
+      if (zeta) { AssembleElemVecAdaptLim(el, Tpr, ir, weights, PMatO); }
+      if (sigma) { AssembleElemVecSurfFit(el, Tpr, ir, weights, PMatO); }
    }
 }
 
@@ -3200,7 +3458,7 @@ void TMOP_Integrator::AssembleElementGradFD(const FiniteElement &el,
    fd_call_flag = false;
 
    // Contributions from adaptive limiting.
-   if (zeta)
+   if (zeta || sigma)
    {
       const IntegrationRule &ir = GradientIntegrationRule(el);
       const int nqp = ir.GetNPoints();
@@ -3220,35 +3478,41 @@ void TMOP_Integrator::AssembleElementGradFD(const FiniteElement &el,
          weights(q) = ir.IntPoint(q).weight * Jtr(q).Det();
       }
 
-      AssembleElemGradAdaptLim(el, weights, Tpr, ir, elmat);
+      if (zeta) { AssembleElemGradAdaptLim(el, Tpr, ir, weights, elmat); }
+      if (sigma) { AssembleElemGradSurfFit(el, Tpr, ir, weights, elmat); }
    }
 }
 
 void TMOP_Integrator::EnableNormalization(const GridFunction &x)
 {
-   ComputeNormalizationEnergies(x, metric_normal, lim_normal);
+   ComputeNormalizationEnergies(x, metric_normal, lim_normal, sigma_normal);
    metric_normal = 1.0 / metric_normal;
    lim_normal = 1.0 / lim_normal;
+   //if (sigma) { sigma_normal = 1.0 / sigma_normal; }
+   if (sigma) { sigma_normal = lim_normal; }
 }
 
 #ifdef MFEM_USE_MPI
 void TMOP_Integrator::ParEnableNormalization(const ParGridFunction &x)
 {
-   double loc[2];
-   ComputeNormalizationEnergies(x, loc[0], loc[1]);
-   double rdc[2];
-   MPI_Allreduce(loc, rdc, 2, MPI_DOUBLE, MPI_SUM, x.ParFESpace()->GetComm());
+   double loc[3];
+   ComputeNormalizationEnergies(x, loc[0], loc[1], loc[2]);
+   double rdc[3];
+   MPI_Allreduce(loc, rdc, 3, MPI_DOUBLE, MPI_SUM, x.ParFESpace()->GetComm());
    metric_normal = 1.0 / rdc[0];
    lim_normal    = 1.0 / rdc[1];
+   // if (sigma) { sigma_normal = 1.0 / rdc[2]; }
+   if (sigma) { sigma_normal = lim_normal; }
 }
 #endif
 
 void TMOP_Integrator::ComputeNormalizationEnergies(const GridFunction &x,
                                                    double &metric_energy,
-                                                   double &lim_energy)
+                                                   double &lim_energy,
+                                                   double &sigma_energy)
 {
    Array<int> vdofs;
-   Vector x_vals;
+   Vector x_vals, sigma_bar_q;
    const FiniteElementSpace* const fes = x.FESpace();
 
    const int dim = fes->GetMesh()->Dimension();
@@ -3258,6 +3522,7 @@ void TMOP_Integrator::ComputeNormalizationEnergies(const GridFunction &x,
 
    metric_energy = 0.0;
    lim_energy = 0.0;
+   sigma_energy = 0.0;
    for (int i = 0; i < fes->GetNE(); i++)
    {
       const FiniteElement *fe = fes->GetFE(i);
@@ -3273,6 +3538,8 @@ void TMOP_Integrator::ComputeNormalizationEnergies(const GridFunction &x,
 
       targetC->ComputeElementTargets(i, *fe, ir, x_vals, Jtr);
 
+      if (sigma) { sigma_bar->GetValues(i, ir, sigma_bar_q); }
+
       for (int q = 0; q < nqp; q++)
       {
          const IntegrationPoint &ip = ir.IntPoint(q);
@@ -3286,8 +3553,15 @@ void TMOP_Integrator::ComputeNormalizationEnergies(const GridFunction &x,
 
          metric_energy += weight * metric->EvalW(Jpt);
          lim_energy += weight;
+
+         // Normalization of the surface fitting term.
+         if (sigma)
+         {
+            sigma_energy += weight * sigma_bar_q(q) * sigma_bar_q(q);
+         }
       }
    }
+
    if (targetC->ContainsVolumeInfo() == false)
    {
       // Special case when the targets don't contain volumetric information.
@@ -3336,6 +3610,17 @@ void TMOP_Integrator::UpdateAfterMeshPositionChange(const Vector &new_x)
    }
    // Update zeta if adaptive limiting is enabled.
    if (zeta) { adapt_eval->ComputeAtNewPosition(new_x, *zeta); }
+
+   // Update sigma if surface fitting is enabled.
+   if (sigma)
+   {
+      sigma_eval->ComputeAtNewPosition(new_x, *sigma);
+      // Update the restricted sigma.
+      for (int i = 0; i < sigma_marker->Size(); i++)
+      {
+         (*sigma_bar)(i) = ((*sigma_marker)[i] == true) ? (*sigma)(i) : 0.0;
+      }
+   }
 }
 
 void TMOP_Integrator::ComputeFDh(const Vector &x, const FiniteElementSpace &fes)

fem/tmop.hpp

@@ -592,6 +592,27 @@ public:
    virtual int Id() const { return 321; }
 };
 
+/// 3D barrier Shape+Size (VS) metric (polyconvex).
+class TMOP_Metric_328 : public TMOP_Combo_QualityMetric
+{
+protected:
+   mutable InvariantsEvaluator2D<double> ie;
+   double gamma;
+   TMOP_QualityMetric *sh_metric, *sz_metric;
+
+public:
+   TMOP_Metric_328(double gamma_) : gamma(gamma_),
+      sh_metric(new TMOP_Metric_301),
+      sz_metric(new TMOP_Metric_316)
+   {
+      // (1-gamma) mu_301 + gamma mu_316
+      AddQualityMetric(sh_metric, 1.-gamma_);
+      AddQualityMetric(sz_metric, gamma_);
+   }
+
+   virtual ~TMOP_Metric_328() { delete sh_metric; delete sz_metric; }
+};
+
 /// 3D barrier Shape+Size (VS) metric (polyconvex).
 class TMOP_Metric_332 : public TMOP_Combo_QualityMetric
 {
@@ -619,6 +640,7 @@ public:
 class TMOP_Metric_333 : public TMOP_Combo_QualityMetric
 {
 protected:
+   mutable InvariantsEvaluator2D<double> ie;
    double gamma;
    TMOP_QualityMetric *sh_metric, *sz_metric;
 
@@ -632,12 +654,30 @@ public:
       AddQualityMetric(sz_metric, gamma_);
    }
 
-   virtual int Id() const { return 333; }
-   double GetGamma() const { return gamma; }
-
    virtual ~TMOP_Metric_333() { delete sh_metric; delete sz_metric; }
 };
 
+/// 3D barrier Shape+Size (VS) metric (polyconvex).
+class TMOP_Metric_334 : public TMOP_Combo_QualityMetric
+{
+protected:
+   mutable InvariantsEvaluator2D<double> ie;
+   double gamma;
+   TMOP_QualityMetric *sh_metric, *sz_metric;
+
+public:
+   TMOP_Metric_334(double gamma_) : gamma(gamma_),
+      sh_metric(new TMOP_Metric_303),
+      sz_metric(new TMOP_Metric_316)
+   {
+      // (1-gamma) mu_303 + gamma mu_316
+      AddQualityMetric(sh_metric, 1.-gamma_);
+      AddQualityMetric(sz_metric, gamma_);
+   }
+
+   virtual ~TMOP_Metric_334() { delete sh_metric; delete sz_metric; }
+};
+
 /// Shifted barrier form of 3D metric 16 (volume, ideal barrier metric), 3D
 class TMOP_Metric_352 : public TMOP_QualityMetric
 {
@@ -897,9 +937,6 @@ protected:
 
 #ifdef MFEM_USE_MPI
    MPI_Comm comm;
-   bool Parallel() const { return (comm != MPI_COMM_NULL); }
-#else
-   bool Parallel() const { return false; }
 #endif
 
    // should be called only if avg_volume == 0.0, i.e. avg_volume is not
@@ -936,6 +973,13 @@ public:
 #endif
    virtual ~TargetConstructor() { }
 
+#ifdef MFEM_USE_MPI
+   bool Parallel() const { return (comm != MPI_COMM_NULL); }
+   MPI_Comm GetComm() const { return comm; }
+#else
+   bool Parallel() const { return false; }
+#endif
+
    /** @brief Set the nodes to be used in the target-matrix construction.
 
        This method should be called every time the target nodes are updated
@@ -1296,6 +1340,13 @@ protected:
    Coefficient *coeff_zeta;          // Not owned.
    AdaptivityEvaluator *adapt_eval;  // Not owned.
 
+   // Surface fitting.
+   GridFunction *sigma, *sigma_bar;  // Owned. Updated by sigma_eval.
+   const Array<bool> *sigma_marker;  // Not owned.
+   Coefficient *coeff_sigma;         // Not owned.
+   AdaptivityEvaluator *sigma_eval;  // Not owned.
+   double sigma_normal;
+
    DiscreteAdaptTC *discr_tc;
 
    // Parameters for FD-based Gradient & Hessian calculation.
@@ -1364,7 +1415,8 @@ protected:
    } PA;
 
    void ComputeNormalizationEnergies(const GridFunction &x,
-                                     double &metric_energy, double &lim_energy);
+                                     double &metric_energy, double &lim_energy,
+                                     double &sigma_energy);
 
    void AssembleElementVectorExact(const FiniteElement &el,
                                    ElementTransformation &T,
@@ -1383,12 +1435,25 @@ protected:
                               ElementTransformation &T,
                               const Vector &elfun, DenseMatrix &elmat);
 
-   void AssembleElemVecAdaptLim(const FiniteElement &el, const Vector &weights,
+   void AssembleElemVecAdaptLim(const FiniteElement &el,
                                 IsoparametricTransformation &Tpr,
-                                const IntegrationRule &ir, DenseMatrix &m);
-   void AssembleElemGradAdaptLim(const FiniteElement &el, const Vector &weights,
+                                const IntegrationRule &ir,
+                                const Vector &weights, DenseMatrix &mat);
+   void AssembleElemGradAdaptLim(const FiniteElement &el,
                                  IsoparametricTransformation &Tpr,
-                                 const IntegrationRule &ir, DenseMatrix &m);
+                                 const IntegrationRule &ir,
+                                 const Vector &weights, DenseMatrix &m);
+
+   // First derivative of the surface fitting term.
+   void AssembleElemVecSurfFit(const FiniteElement &el_x,
+                               IsoparametricTransformation &Tpr,
+                               const IntegrationRule &ir_quad,
+                               const Vector &weights, DenseMatrix &mat);
+   // Second derivative of the surface fitting term.
+   void AssembleElemGradSurfFit(const FiniteElement &el_x,
+                                IsoparametricTransformation &Tpr,
+                                const IntegrationRule &ir_quad,
+                                const Vector &weights, DenseMatrix &mat);
 
    double GetFDDerivative(const FiniteElement &el,
                           ElementTransformation &T,
@@ -1470,6 +1535,8 @@ public:
         nodes0(NULL), coeff0(NULL),
         lim_dist(NULL), lim_func(NULL), lim_normal(1.0),
         zeta_0(NULL), zeta(NULL), coeff_zeta(NULL), adapt_eval(NULL),
+        sigma(NULL), sigma_bar(NULL), sigma_marker(NULL), coeff_sigma(NULL),
+        sigma_eval(NULL), sigma_normal(1.0),
         discr_tc(dynamic_cast<DiscreteAdaptTC *>(tc)),
         fdflag(false), dxscale(1.0e3), fd_call_flag(false), exact_action(false)
    { PA.enabled = false; }
@@ -1522,7 +1589,7 @@ public:
 
        Adds the term @f$ \int c (z(x) - z_0(x_0))^2 @f$, where z0(x0) is a given
        function on the starting mesh, and z(x) is its image on the new mesh.
-       Minimizing this, means that a node at x0 is allowed to move to a
+       Minimizing this term means that a node at x0 is allowed to move to a
        position x(x0) only if z(x) ~ z0(x0).
        Such term can be used for tangential mesh relaxation.
 
@@ -1537,6 +1604,32 @@ public:
                                AdaptivityEvaluator &ae);
 #endif
 
+   /** @brief Fitting of certain DOFs to the zero level set of a function.
+
+       Having a level set function s0(x0) on the starting mesh, and a set of
+       marked nodes (or DOFs), we move these nodes to the zero level set of s0.
+       If s(x) is the image of s0(x0) on the current mesh, this function adds to
+       the TMOP functional the term @f$ \int c \bar{s}(x))^2 @f$, where
+       @f$\bar{s}(x)@f$ is the restriction of s(x) on the aligned DOFs.
+       Minimizing this term means that a marked node at x0 is allowed to move to
+       a position x(x0) only if s(x) ~ 0.
+       Such term can be used for surface fitting and tangential relaxation.
+
+       @param[in] s0      The level set function on the initial mesh.
+       @param[in] smarker Indicates which DOFs will be aligned.
+       @param[in] coeff   Coefficient c for the above integral.
+       @param[in] ae      AdaptivityEvaluator to compute s(x) from s0(x0). */
+   void EnableSurfaceFitting(const GridFunction &s0,
+                             const Array<bool> &smarker, Coefficient &coeff,
+                             AdaptivityEvaluator &ae);
+#ifdef MFEM_USE_MPI
+   /// Parallel support for surface fitting.
+   void EnableSurfaceFitting(const ParGridFunction &s0,
+                             const Array<bool> &smarker, Coefficient &coeff,
+                             AdaptivityEvaluator &ae);
+#endif
+   void GetSurfaceFittingErrors(double &err_avg, double &err_max);
+
    /// Update the original/reference nodes used for limiting.
    void SetLimitingNodes(const GridFunction &n0) { nodes0 = &n0; }
 

fem/tmop_amr.cpp

@@ -394,12 +394,13 @@ bool TMOPDeRefinerEstimator::GetDerefineEnergyForIntegrator(
 
       const CoarseFineTransformations &dtrans =
          meshcopy.ncmesh->GetDerefinementTransforms();
-      Table coarse_to_fine;
-      dtrans.GetCoarseToFineMap(meshcopy, coarse_to_fine);
 
+      Table coarse_to_fine;
+      dtrans.MakeCoarseToFineTable(coarse_to_fine);
+
+      Array<int> tabrow;
       for (int pe = 0; pe < coarse_to_fine.Size(); pe++)
       {
-         Array<int> tabrow;
          coarse_to_fine.GetRow(pe, tabrow);
          int nchild = tabrow.Size();
          double parent_energy = coarse_energy(pe);
@@ -446,12 +447,13 @@ bool TMOPDeRefinerEstimator::GetDerefineEnergyForIntegrator(
 
       const CoarseFineTransformations &dtrans =
          meshcopy.pncmesh->GetDerefinementTransforms();
-      Table coarse_to_fine;
-      dtrans.GetCoarseToFineMap(meshcopy, coarse_to_fine);
 
+      Table coarse_to_fine;
+      dtrans.MakeCoarseToFineTable(coarse_to_fine);
+
+      Array<int> tabrow;
       for (int pe = 0; pe < meshcopy.GetNE(); pe++)
       {
-         Array<int> tabrow;
          coarse_to_fine.GetRow(pe, tabrow);
          int nchild = tabrow.Size();
          double parent_energy = coarse_energy(pe);

fem/transfer.cpp

@@ -913,7 +913,11 @@ TransferOperator::TransferOperator(const FiniteElementSpace& lFESpace_,
    else if (lFESpace_.GetMesh()->GetNE() > 0
             && hFESpace_.GetMesh()->GetNE() > 0
             && dynamic_cast<const TensorBasisElement*>(lFESpace_.GetFE(0))
-            && dynamic_cast<const TensorBasisElement*>(hFESpace_.GetFE(0)))
+            && dynamic_cast<const TensorBasisElement*>(hFESpace_.GetFE(0))
+            && (hFESpace_.FEColl()->GetContType() ==
+                mfem::FiniteElementCollection::CONTINUOUS ||
+                hFESpace_.FEColl()->GetContType() ==
+                mfem::FiniteElementCollection::DISCONTINUOUS))
    {
       opr = new TensorProductPRefinementTransferOperator(lFESpace_, hFESpace_);
    }
@@ -961,8 +965,8 @@ void PRefinementTransferOperator::Mult(const Vector& x, Vector& y) const
 
    for (int i = 0; i < mesh->GetNE(); i++)
    {
-      hFESpace.GetElementDofs(i, h_dofs);
-      lFESpace.GetElementDofs(i, l_dofs);
+      DofTransformation * doftrans_h = hFESpace.GetElementDofs(i, h_dofs);
+      DofTransformation * doftrans_l = lFESpace.GetElementDofs(i, l_dofs);
 
       const Geometry::Type geom = mesh->GetElementBaseGeometry(i);
       if (geom != cached_geom)
@@ -982,7 +986,15 @@ void PRefinementTransferOperator::Mult(const Vector& x, Vector& y) const
          h_dofs.Copy(h_vdofs);
          hFESpace.DofsToVDofs(vd, h_vdofs);
          x.GetSubVector(l_vdofs, subX);
+         if (doftrans_l)
+         {
+            doftrans_l->InvTransformPrimal(subX);
+         }
          loc_prol.Mult(subX, subY);
+         if (doftrans_h)
+         {
+            doftrans_h->TransformPrimal(subY);
+         }
          y.SetSubVector(h_vdofs, subY);
       }
    }
@@ -1010,8 +1022,8 @@ void PRefinementTransferOperator::MultTranspose(const Vector& x,
 
    for (int i = 0; i < mesh->GetNE(); i++)
    {
-      hFESpace.GetElementDofs(i, h_dofs);
-      lFESpace.GetElementDofs(i, l_dofs);
+      DofTransformation * doftrans_h = hFESpace.GetElementDofs(i, h_dofs);
+      DofTransformation * doftrans_l = lFESpace.GetElementDofs(i, l_dofs);
 
       const Geometry::Type geom = mesh->GetElementBaseGeometry(i);
       if (geom != cached_geom)
@@ -1033,6 +1045,10 @@ void PRefinementTransferOperator::MultTranspose(const Vector& x,
          hFESpace.DofsToVDofs(vd, h_vdofs);
 
          x.GetSubVector(h_vdofs, subX);
+         if (doftrans_h)
+         {
+            doftrans_h->InvTransformDual(subX);
+         }
          for (int p = 0; p < h_dofs.Size(); ++p)
          {
             if (processed[lFESpace.DecodeDof(h_dofs[p])])
@@ -1042,6 +1058,10 @@ void PRefinementTransferOperator::MultTranspose(const Vector& x,
          }
 
          loc_prol.Mult(subX, subY);
+         if (doftrans_l)
+         {
+            doftrans_l->TransformDual(subY);
+         }
          y.AddElementVector(l_vdofs, subY);
       }
 
@@ -1085,7 +1105,8 @@ TensorProductPRefinementTransferOperator(
    // must be sorted in lexicographical order
    for (int i = 0; i < ir.GetNPoints(); ++i)
    {
-      irLex.IntPoint(i) = ir.IntPoint(hdofmap[i]);
+      int j = hdofmap[i] >=0 ? hdofmap[i] : -1 - hdofmap[i];
+      irLex.IntPoint(i) = ir.IntPoint(j);
    }
 
    NE = lFESpace.GetNE();

general/array.hpp

@@ -70,6 +70,10 @@ public:
    explicit inline Array(int asize)
       : size(asize) { asize > 0 ? data.New(asize) : data.Reset(); }
 
+   /// Creates array of @a asize elements with a given MemoryType
+   inline Array(int asize, MemoryType mt)
+      : size(asize) { asize > 0 ? data.New(asize, mt) : data.Reset(mt); }
+
    /** @brief Creates array using an existing c-array of asize elements;
        allocsize is set to -asize to indicate that the data will not
        be deleted. */

general/forall.hpp

@@ -183,6 +183,42 @@ void RajaCuWrap3D(const int N, DBODY &&d_body,
    MFEM_GPU_CHECK(cudaGetLastError());
 }
 
+template <int Dim>
+struct RajaCuWrap;
+
+template <>
+struct RajaCuWrap<1>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      RajaCuWrap1D<BLCK>(N, d_body);
+   }
+};
+
+template <>
+struct RajaCuWrap<2>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      RajaCuWrap2D(N, d_body, X, Y, Z);
+   }
+};
+
+template <>
+struct RajaCuWrap<3>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      RajaCuWrap3D(N, d_body, X, Y, Z, G);
+   }
+};
+
 #endif
 
 #if defined(MFEM_USE_RAJA) && defined(RAJA_ENABLE_HIP)
@@ -248,6 +284,43 @@ void RajaHipWrap3D(const int N, DBODY &&d_body,
 
    MFEM_GPU_CHECK(hipGetLastError());
 }
+
+template <int Dim>
+struct RajaHipWrap;
+
+template <>
+struct RajaHipWrap<1>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      RajaHipWrap1D<BLCK>(N, d_body);
+   }
+};
+
+template <>
+struct RajaHipWrap<2>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      RajaHipWrap2D(N, d_body, X, Y, Z);
+   }
+};
+
+template <>
+struct RajaHipWrap<3>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      RajaHipWrap3D(N, d_body, X, Y, Z, G);
+   }
+};
+
 #endif
 
 /// RAJA OpenMP backend
@@ -333,6 +406,42 @@ void CuWrap3D(const int N, DBODY &&d_body,
    MFEM_GPU_CHECK(cudaGetLastError());
 }
 
+template <int Dim>
+struct CuWrap;
+
+template <>
+struct CuWrap<1>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      CuWrap1D<BLCK>(N, d_body);
+   }
+};
+
+template <>
+struct CuWrap<2>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      CuWrap2D(N, d_body, X, Y, Z);
+   }
+};
+
+template <>
+struct CuWrap<3>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      CuWrap3D(N, d_body, X, Y, Z, G);
+   }
+};
+
 #endif // MFEM_USE_CUDA
 
 
@@ -392,6 +501,42 @@ void HipWrap3D(const int N, DBODY &&d_body,
    MFEM_GPU_CHECK(hipGetLastError());
 }
 
+template <int Dim>
+struct HipWrap;
+
+template <>
+struct HipWrap<1>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      HipWrap1D<BLCK>(N, d_body);
+   }
+};
+
+template <>
+struct HipWrap<2>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      HipWrap2D(N, d_body, X, Y, Z);
+   }
+};
+
+template <>
+struct HipWrap<3>
+{
+   template <const int BLCK = MFEM_CUDA_BLOCKS, typename DBODY>
+   static void run(const int N, DBODY &&d_body,
+                   const int X, const int Y, const int Z, const int G)
+   {
+      HipWrap3D(N, d_body, X, Y, Z, G);
+   }
+};
+
 #endif // MFEM_USE_HIP
 
 
@@ -413,9 +558,7 @@ inline void ForallWrap(const bool use_dev, const int N,
    // If Backend::RAJA_CUDA is allowed, use it
    if (Device::Allows(Backend::RAJA_CUDA))
    {
-      if (DIM == 1) { return RajaCuWrap1D(N, d_body); }
-      if (DIM == 2) { return RajaCuWrap2D(N, d_body, X, Y, Z); }
-      if (DIM == 3) { return RajaCuWrap3D(N, d_body, X, Y, Z, G); }
+      return RajaCuWrap<DIM>::run(N, d_body, X, Y, Z, G);
    }
 #endif
 
@@ -423,9 +566,7 @@ inline void ForallWrap(const bool use_dev, const int N,
    // If Backend::RAJA_HIP is allowed, use it
    if (Device::Allows(Backend::RAJA_HIP))
    {
-      if (DIM == 1) { return RajaHipWrap1D(N, d_body); }
-      if (DIM == 2) { return RajaHipWrap2D(N, d_body, X, Y, Z); }
-      if (DIM == 3) { return RajaHipWrap3D(N, d_body, X, Y, Z, G); }
+      return RajaHipWrap<DIM>::run(N, d_body, X, Y, Z, G);
    }
 #endif
 
@@ -433,9 +574,7 @@ inline void ForallWrap(const bool use_dev, const int N,
    // If Backend::CUDA is allowed, use it
    if (Device::Allows(Backend::CUDA))
    {
-      if (DIM == 1) { return CuWrap1D(N, d_body); }
-      if (DIM == 2) { return CuWrap2D(N, d_body, X, Y, Z); }
-      if (DIM == 3) { return CuWrap3D(N, d_body, X, Y, Z, G); }
+      return CuWrap<DIM>::run(N, d_body, X, Y, Z, G);
    }
 #endif
 
@@ -443,9 +582,7 @@ inline void ForallWrap(const bool use_dev, const int N,
    // If Backend::HIP is allowed, use it
    if (Device::Allows(Backend::HIP))
    {
-      if (DIM == 1) { return HipWrap1D(N, d_body); }
-      if (DIM == 2) { return HipWrap2D(N, d_body, X, Y, Z); }
-      if (DIM == 3) { return HipWrap3D(N, d_body, X, Y, Z, G); }
+      return HipWrap<DIM>::run(N, d_body, X, Y, Z, G);
    }
 #endif
 

general/mem_manager.hpp

@@ -175,8 +175,9 @@ protected:
    // Copy{From,To}, {ReadWrite,Read,Write}.
 
 public:
-   /// Default constructor: no initialization.
-   Memory() { }
+   /** Default constructor, sets the host pointer to nullptr and the metadata to
+       meaningful default values. */
+   Memory() { Reset(); }
 
    /// Copy constructor: default.
    Memory(const Memory &orig) = default;
@@ -368,8 +369,7 @@ public:
        be updated as described above. */
    inline void SetDeviceMemoryType(MemoryType d_mt);
 
-   /** @brief Delete the owned pointers. The Memory is not reset by this method,
-       i.e. it will, generally, not be Empty() after this call. */
+   /** @brief Delete the owned pointers and reset the Memory object. */
    inline void Delete();
 
    /** @brief Delete the device pointer, if owned. If @a copy_to_host is true
@@ -986,6 +986,7 @@ inline void Memory<T>::Delete()
    {
       if (flags & OWNS_HOST) { delete [] h_ptr; }
    }
+   Reset(h_mt);
 }
 
 template <typename T>

general/sets.cpp

@@ -69,6 +69,7 @@ void IntegerSet::Recreate(const int n, const int *p)
 
    me.Sort();
 
+   // Remove duplicate entries
    for (j = 0, i = 1; i < n; i++)
       if (me[i] != me[j])
       {

general/sets.hpp

@@ -36,7 +36,7 @@ public:
    IntegerSet(const int n, const int *p) { Recreate(n, p); }
 
    /// Return the size of the set.
-   int Size() { return me.Size(); }
+   int Size() const { return me.Size(); }
 
    /// Return a reference to the sorted array of all the set entries.
    operator Array<int>& () { return me; }
@@ -50,6 +50,8 @@ public:
    /// Return 1 if the sets are equal and 0 otherwise.
    int operator==(IntegerSet &s);
 
+   inline const int & operator[](int i) const { return me[i]; }
+
    /** @brief Create an integer set from C-array 'p' of 'n' integers.
        Overwrites any existing set data. */
    void Recreate(const int n, const int *p);
@@ -64,7 +66,7 @@ private:
 public:
 
    /// Return the number of integer sets in the list.
-   int Size() { return TheList.Size(); }
+   int Size() const { return TheList.Size(); }
 
    /// Return the value of the first element of the ith set.
    int PickElementInSet(int i) { return TheList[i]->PickElement(); }
@@ -84,6 +86,9 @@ public:
    /// Write the list of sets into table 't'.
    void AsTable(Table &t);
 
+   inline const IntegerSet & operator[](int i) const { return *TheList[i]; }
+   inline IntegerSet & operator[](int i) { return *TheList[i]; }
+
    ~ListOfIntegerSets();
 };
 

general/stable3d.cpp

@@ -61,7 +61,7 @@ inline void Sort3 (int &r, int &c, int &f)
    }
 }
 
-int STable3D::Push (int r, int c, int f)
+int STable3D::Push (int r, int c, int f, int t)
 {
    STable3DNode *node;
 
@@ -86,6 +86,7 @@ int STable3D::Push (int r, int c, int f)
 #endif
    node->Column = c;
    node->Floor  = f;
+   node->Tier   = t;
    node->Number = NElem;
    node->Prev   = Rows[r];
    Rows[r] = node;
@@ -109,9 +110,9 @@ int STable3D::operator() (int r, int c, int f) const
          }
    }
 
-   MFEM_ABORT("(r,c,f) = (" << r << "," << c << "," << f << ")");
+   // MFEM_ABORT("(r,c,f) = (" << r << "," << c << "," << f << ")");
 
-   return 0;
+   return -1;
 }
 
 int STable3D::Index (int r, int c, int f) const
@@ -152,13 +153,13 @@ int STable3D::Push4 (int r, int c, int f, int t)
    switch (i)
    {
       case 0:
-         return Push (c,f,t);
+         return Push (c,f,t,r);
       case 1:
-         return Push (r,f,t);
+         return Push (r,f,t,c);
       case 2:
-         return Push (r,c,t);
+         return Push (r,c,t,f);
       case 3:
-         return Push (r,c,f);
+         return Push (r,c,f,t);
    }
 
    return -1;
@@ -218,6 +219,7 @@ void STable3D::Print(std::ostream & out) const
          out << row
              << ' ' << node_p->Column
              << ' ' << node_p->Floor
+             << ' ' << node_p->Tier
              << ' ' << node_p->Number
              << endl;
          node_p = node_p->Prev;

general/stable3d.hpp

@@ -15,6 +15,8 @@
 #include "mem_alloc.hpp"
 #include "../general/globals.hpp"
 
+#include <iostream>
+
 namespace mfem
 {
 
@@ -22,7 +24,7 @@ class STable3DNode
 {
 public:
    STable3DNode *Prev;
-   int Column, Floor, Number;
+   int Column, Floor, Tier, Number;
 };
 
 /** @brief Symmetric 3D Table stored as an array of rows each of which has a
@@ -47,7 +49,7 @@ public:
 
    /** @brief Check to see if this entry is in the table and add it to the table
        if it is not there. Returns the number assigned to the table entry. */
-   int Push (int r, int c, int f);
+   int Push (int r, int c, int f, int t = -1);
 
    /// Return the number assigned to the table entry. Abort if it's not there.
    int operator() (int r, int c, int f) const;
@@ -66,13 +68,30 @@ public:
        not there. */
    int operator() (int r, int c, int f, int t) const;
 
+   /// Return the number of rows added to the table.
+   int NumberOfRows() const { return Size; }
+
    /// Return the number of elements added to the table.
-   int NumberOfElements() { return NElem; }
+   int NumberOfElements() const { return NElem; }
 
    /// Print out all of the table elements.
    void Print(std::ostream &out = mfem::out) const;
 
    ~STable3D ();
+
+   class RowIterator
+   {
+   private:
+      STable3DNode *n;
+   public:
+      RowIterator (const STable3D &t, int r) { n = t.Rows[r]; }
+      int operator!() { return (n != NULL); }
+      void operator++() { n = n->Prev; }
+      int Column() { return (n->Column); }
+      int Floor()  { return (n->Floor); }
+      int Tier()   { return (n->Tier); }
+      int Index()  { return (n->Number); }
+   };
 };
 
 }

general/table.cpp

@@ -35,10 +35,6 @@ Table::Table(const Table &table)
       I.CopyFrom(table.I, size+1);
       J.CopyFrom(table.J, nnz);
    }
-   else
-   {
-      I.Reset(); J.Reset();
-   }
 }
 
 Table& Table::operator=(const Table &rhs)

general/table.hpp

@@ -53,7 +53,7 @@ protected:
 
 public:
    /// Creates an empty table
-   Table() { size = -1; I.Reset(); J.Reset(); }
+   Table() { size = -1; }
 
    /// Copy constructor
    Table(const Table &);
@@ -66,7 +66,7 @@ public:
 
    /** Create a table from a list of connections, see MakeFromList(). */
    Table(int nrows, Array<Connection> &list) : size(-1)
-   { I.Reset(); J.Reset(); MakeFromList(nrows, list); }
+   { MakeFromList(nrows, list); }
 
    /** Create a table with one entry per row with column indices given
        by 'partitioning'. */

linalg/densemat.cpp

@@ -70,10 +70,7 @@ namespace mfem
 
 using namespace std;
 
-DenseMatrix::DenseMatrix() : Matrix(0)
-{
-   data.Reset();
-}
+DenseMatrix::DenseMatrix() : Matrix(0) { }
 
 DenseMatrix::DenseMatrix(const DenseMatrix &m) : Matrix(m.height, m.width)
 {
@@ -84,10 +81,6 @@ DenseMatrix::DenseMatrix(const DenseMatrix &m) : Matrix(m.height, m.width)
       data.New(hw);
       std::memcpy(data, m.data, sizeof(double)*hw);
    }
-   else
-   {
-      data.Reset();
-   }
 }
 
 DenseMatrix::DenseMatrix(int s) : Matrix(s)
@@ -98,10 +91,6 @@ DenseMatrix::DenseMatrix(int s) : Matrix(s)
       data.New(s*s);
       *this = 0.0; // init with zeroes
    }
-   else
-   {
-      data.Reset();
-   }
 }
 
 DenseMatrix::DenseMatrix(int m, int n) : Matrix(m, n)
@@ -114,10 +103,6 @@ DenseMatrix::DenseMatrix(int m, int n) : Matrix(m, n)
       data.New(capacity);
       *this = 0.0; // init with zeroes
    }
-   else
-   {
-      data.Reset();
-   }
 }
 
 DenseMatrix::DenseMatrix(const DenseMatrix &mat, char ch)
@@ -137,10 +122,6 @@ DenseMatrix::DenseMatrix(const DenseMatrix &mat, char ch)
          }
       }
    }
-   else
-   {
-      data.Reset();
-   }
 }
 
 void DenseMatrix::SetSize(int h, int w)

linalg/densemat.hpp

@@ -753,7 +753,6 @@ public:
    DenseTensor()
    {
       nk = 0;
-      tdata.Reset();
    }
 
    DenseTensor(int i, int j, int k)
@@ -787,10 +786,6 @@ public:
          tdata.New(size, other.tdata.GetMemoryType());
          tdata.CopyFrom(other.tdata, size);
       }
-      else
-      {
-         tdata.Reset();
-      }
    }
 
    int SizeI() const { return Mk.Height(); }

linalg/hypre.cpp

@@ -127,18 +127,19 @@ HypreParVector::HypreParVector(MPI_Comm comm, HYPRE_BigInt glob_size,
    own_ParVector = 1;
 }
 
-HypreParVector::HypreParVector(const HypreParVector &y) : Vector()
+// Call the move constructor on the "compatible" temp vector
+HypreParVector::HypreParVector(const HypreParVector &y) : HypreParVector(
+      y.CreateCompatibleVector())
 {
-   x = hypre_ParVectorCreate(y.x -> comm, y.x -> global_size,
-                             y.x -> partitioning);
-   hypre_ParVectorInitialize(x);
-#if MFEM_HYPRE_VERSION <= 22200
-   hypre_ParVectorSetPartitioningOwner(x,0);
-#endif
-   hypre_ParVectorSetDataOwner(x,1);
-   hypre_SeqVectorSetDataOwner(hypre_ParVectorLocalVector(x),1);
-   _SetDataAndSize_();
-   own_ParVector = 1;
+   // Deep copy the local data
+   hypre_SeqVectorCopy(hypre_ParVectorLocalVector(y.x),
+                       hypre_ParVectorLocalVector(x));
+}
+
+HypreParVector::HypreParVector(HypreParVector &&y)
+{
+   own_ParVector = 0;
+   *this = std::move(y);
 }
 
 HypreParVector::HypreParVector(const HypreParMatrix &A,
@@ -178,6 +179,23 @@ HypreParVector::HypreParVector(ParFiniteElementSpace *pfes)
    own_ParVector = 1;
 }
 
+HypreParVector HypreParVector::CreateCompatibleVector() const
+{
+   HypreParVector result;
+   result.x = hypre_ParVectorCreate(x -> comm, x -> global_size,
+                                    x -> partitioning);
+   hypre_ParVectorInitialize(result.x);
+#if MFEM_HYPRE_VERSION <= 22200
+   hypre_ParVectorSetPartitioningOwner(result.x,0);
+#endif
+   hypre_ParVectorSetDataOwner(result.x,1);
+   hypre_SeqVectorSetDataOwner(hypre_ParVectorLocalVector(result.x),1);
+   result._SetDataAndSize_();
+   result.own_ParVector = 1;
+
+   return result;
+}
+
 void HypreParVector::WrapHypreParVector(hypre_ParVector *y, bool owner)
 {
    if (own_ParVector) { hypre_ParVectorDestroy(x); }
@@ -216,6 +234,18 @@ HypreParVector& HypreParVector::operator=(const HypreParVector &y)
    return *this;
 }
 
+HypreParVector& HypreParVector::operator=(HypreParVector &&y)
+{
+   // If the argument vector owns its data, then the calling vector will as well
+   WrapHypreParVector(static_cast<hypre_ParVector*>(y), y.own_ParVector);
+   // Either way the argument vector will no longer own its data
+   y.own_ParVector = 0;
+   y.x = nullptr;
+   y.data.Reset();
+   y.size = 0;
+   return *this;
+}
+
 void HypreParVector::SetData(double *data_)
 {
    hypre_VectorData(hypre_ParVectorLocalVector(x)) = data_;
@@ -303,6 +333,18 @@ void HypreParVector::Print(const char *fname) const
    hypre_ParVectorPrint(x,fname);
 }
 
+void HypreParVector::Read(MPI_Comm comm, const char *fname)
+{
+   if (own_ParVector)
+   {
+      hypre_ParVectorDestroy(x);
+   }
+   data.Delete();
+   x = hypre_ParVectorRead(comm, fname);
+   own_ParVector = true;
+   _SetDataAndSize_();
+}
+
 HypreParVector::~HypreParVector()
 {
    if (own_ParVector)
@@ -1561,9 +1603,16 @@ HypreParMatrix *HypreParMatrix::ExtractSubmatrix(const Array<int> &indices,
    }
 
    // Construct cpts_global array on hypre matrix structure
+#if (MFEM_HYPRE_VERSION > 22300) || (MFEM_HYPRE_VERSION == 22300 && HYPRE_DEVELOP_NUMBER >=8)
+   HYPRE_BigInt cpts_global[2];
+
+   hypre_BoomerAMGCoarseParms(MPI_COMM_WORLD, local_num_vars, 1, NULL,
+                              CF_marker, NULL, cpts_global);
+#else
    HYPRE_BigInt *cpts_global;
    hypre_BoomerAMGCoarseParms(MPI_COMM_WORLD, local_num_vars, 1, NULL,
                               CF_marker, NULL, &cpts_global);
+#endif
 
    // Extract submatrix into *submat
 #ifdef hypre_IntArrayData
@@ -1575,7 +1624,9 @@ HypreParMatrix *HypreParMatrix::ExtractSubmatrix(const Array<int> &indices,
                                         "FF", &submat, threshold);
 #endif
 
+#if (MFEM_HYPRE_VERSION <= 22300) && !(MFEM_HYPRE_VERSION == 22300 && HYPRE_DEVELOP_NUMBER >=8)
    mfem_hypre_TFree(cpts_global);
+#endif
 #ifdef hypre_IntArrayData
    hypre_IntArrayDestroy(CF_marker);
 #endif

linalg/hypre.hpp

@@ -141,8 +141,10 @@ public:
        allocated in the memory location HYPRE_MEMORY_DEVICE. */
    HypreParVector(MPI_Comm comm, HYPRE_BigInt glob_size, double *data_,
                   HYPRE_BigInt *col, bool is_device_ptr = false);
-   /// Creates vector compatible with y
+   /// Creates a deep copy of @a y
    HypreParVector(const HypreParVector &y);
+   /// Move constructor for HypreParVector. "Steals" data from its argument.
+   HypreParVector(HypreParVector&& other);
    /// Creates vector compatible with (i.e. in the domain of) A or A^T
    explicit HypreParVector(const HypreParMatrix &A, int transpose = 0);
    /// Creates vector wrapping y
@@ -150,6 +152,10 @@ public:
    /// Create a true dof parallel vector on a given ParFiniteElementSpace
    explicit HypreParVector(ParFiniteElementSpace *pfes);
 
+   /// \brief Constructs a  @p HypreParVector *compatible* with the calling vector
+   /// - meaning that it will be the same size and have the same partitioning.
+   HypreParVector CreateCompatibleVector() const;
+
    /// MPI communicator
    MPI_Comm GetComm() const { return x->comm; }
 
@@ -192,6 +198,8 @@ public:
    HypreParVector& operator= (double d);
    /// Define '=' for hypre vectors.
    HypreParVector& operator= (const HypreParVector &y);
+   /// Move assignment
+   HypreParVector& operator= (HypreParVector &&y);
 
    using Vector::Read;
 
@@ -252,6 +260,9 @@ public:
    /// Prints the locally owned rows in parallel
    void Print(const char *fname) const;
 
+   /// Reads a HypreParVector from files saved with HypreParVector::Print
+   void Read(MPI_Comm comm, const char *fname);
+
    /// Calls hypre's destroy function
    ~HypreParVector();
 

linalg/kernels.hpp

@@ -160,7 +160,7 @@ double Norml2(const int size, const T *data)
     data of the input and output vectors. */
 template<typename TA, typename TX, typename TY>
 MFEM_HOST_DEVICE inline
-void Mult(const int height, const int width, TA *data, const TX *x, TY *y)
+void Mult(const int height, const int width, const TA *data, const TX *x, TY *y)
 {
    if (width == 0)
    {
@@ -170,7 +170,7 @@ void Mult(const int height, const int width, TA *data, const TX *x, TY *y)
       }
       return;
    }
-   TA *d_col = data;
+   const TA *d_col = data;
    TX x_col = x[0];
    for (int row = 0; row < height; row++)
    {
@@ -188,6 +188,35 @@ void Mult(const int height, const int width, TA *data, const TX *x, TY *y)
    }
 }
 
+/** @brief Matrix transpose vector multiplication: y = At x, where the matrix A
+    is of size @a height x @a width with given @a data, while @a x and @a y
+    specify the data of the input and output vectors. */
+template<typename TA, typename TX, typename TY>
+MFEM_HOST_DEVICE inline
+void MultTranspose(const int height, const int width, const TA *data,
+                   const TX *x, TY *y)
+{
+   if (height == 0)
+   {
+      for (int row = 0; row < width; row++)
+      {
+         y[row] = 0.0;
+      }
+      return;
+   }
+   TY *y_off = y;
+   for (int i = 0; i < width; ++i)
+   {
+      TY val = 0.0;
+      for (int j = 0; j < height; ++j)
+      {
+         val += x[j] * data[i * height + j];
+      }
+      *y_off = val;
+      y_off++;
+   }
+}
+
 /// Symmetrize a square matrix with given @a size and @a data: A -> (A+A^T)/2.
 template<typename T>
 MFEM_HOST_DEVICE inline
@@ -353,6 +382,30 @@ void MultABt(const int Aheight, const int Awidth, const int Bheight,
    }
 }
 
+/** @brief Multiply the transpose of a matrix of size @a Aheight x @a Awidth
+    and data @a Adata with a matrix of size @a Aheight x @a Bwidth and data @a
+    Bdata: At * B. Return the result in a matrix with data @a AtBdata. */
+template<typename TA, typename TB, typename TC>
+MFEM_HOST_DEVICE inline
+void MultAtB(const int Aheight, const int Awidth, const int Bwidth,
+             const TA *Adata, const TB *Bdata, TC *AtBdata)
+{
+   TC *c = AtBdata;
+   for (int i = 0; i < Bwidth; ++i)
+   {
+      for (int j = 0; j < Awidth; ++j)
+      {
+         TC val = 0.0;
+         for (int k = 0; k < Aheight; ++k)
+         {
+            val += Adata[j * Aheight + k] * Bdata[i * Aheight + k];
+         }
+         *c = val;
+         c++;
+      }
+   }
+}
+
 /// Compute the spectrum of the matrix of size dim with given @a data, returning
 /// the eigenvalues in the array @a lambda and the eigenvectors in the array @a
 /// vec (listed consecutively).

linalg/operator.hpp

@@ -242,7 +242,7 @@ public:
    void FormDiscreteOperator(Operator* &A);
 
    /// Prints operator with input size n and output size m in Matlab format.
-   void PrintMatlab(std::ostream & out, int n = 0, int m = 0) const;
+   void PrintMatlab(std::ostream & out, int n, int m = 0) const;
 
    /// Prints operator in Matlab format.
    virtual void PrintMatlab(std::ostream & out) const;

linalg/petsc.cpp

@@ -558,8 +558,6 @@ PetscParVector::PetscParVector(MPI_Comm comm, const Operator &op,
    else /* Vector intended to be used with Place/ResetMemory calls */
    {
       size = loc;
-      pdata.Reset();
-      data.Reset();
    }
 }
 
@@ -581,8 +579,6 @@ PetscParVector::PetscParVector(const PetscParMatrix &A,
       PetscInt n;
       ierr = VecGetLocalSize(x,&n); PCHKERRQ(x,ierr);
       size = n;
-      pdata.Reset();
-      data.Reset();
    }
    else
    {

linalg/sparsemat.cpp

@@ -84,9 +84,9 @@ SparseMatrix::SparseMatrix(int nrows, int ncols)
      isSorted(false)
 {
    // We probably do not need to set the ownership flags here.
-   I.Reset(); I.SetHostPtrOwner(true);
-   J.Reset(); J.SetHostPtrOwner(true);
-   A.Reset(); A.SetHostPtrOwner(true);
+   I.SetHostPtrOwner(true);
+   J.SetHostPtrOwner(true);
+   A.SetHostPtrOwner(true);
 
    for (int i = 0; i < nrows; i++)
    {
@@ -229,9 +229,9 @@ SparseMatrix::SparseMatrix(const SparseMatrix &mat, bool copy_graph,
       }
 
       // We probably do not need to set the ownership flags here.
-      I.Reset(); I.SetHostPtrOwner(true);
-      J.Reset(); J.SetHostPtrOwner(true);
-      A.Reset(); A.SetHostPtrOwner(true);
+      I.SetHostPtrOwner(true);
+      J.SetHostPtrOwner(true);
+      A.SetHostPtrOwner(true);
    }
 
    current_row = -1;

linalg/symmat.cpp

@@ -17,10 +17,7 @@
 namespace mfem
 {
 
-DenseSymmetricMatrix::DenseSymmetricMatrix() : Matrix(0)
-{
-   data.Reset();
-}
+DenseSymmetricMatrix::DenseSymmetricMatrix() : Matrix(0) { }
 
 DenseSymmetricMatrix::DenseSymmetricMatrix(int s) : Matrix(s)
 {
@@ -30,10 +27,6 @@ DenseSymmetricMatrix::DenseSymmetricMatrix(int s) : Matrix(s)
       data.New((s*(s+1))/2);
       *this = 0.0; // init with zeroes
    }
-   else
-   {
-      data.Reset();
-   }
 }
 
 void DenseSymmetricMatrix::SetSize(int s)

linalg/vector.cpp

@@ -39,21 +39,21 @@ namespace mfem
 Vector::Vector(const Vector &v)
 {
    const int s = v.Size();
+   size = s;
    if (s > 0)
    {
       MFEM_ASSERT(!v.data.Empty(), "invalid source vector");
-      size = s;
       data.New(s, v.data.GetMemoryType());
       data.CopyFrom(v.data, s);
    }
-   else
-   {
-      size = 0;
-      data.Reset();
-   }
    UseDevice(v.UseDevice());
 }
 
+Vector::Vector(Vector &&v)
+{
+   *this = std::move(v);
+}
+
 void Vector::Load(std::istream **in, int np, int *dim)
 {
    int i, j, s;
@@ -146,6 +146,15 @@ Vector &Vector::operator=(const Vector &v)
    return *this;
 }
 
+Vector &Vector::operator=(Vector &&v)
+{
+   data = std::move(v.data);
+   size = v.size;
+   v.data.Reset();
+   v.size = 0;
+   return *this;
+}
+
 Vector &Vector::operator=(double value)
 {
    const bool use_dev = UseDevice();

linalg/vector.hpp

@@ -66,12 +66,16 @@ protected:
 
 public:
 
-   /// Default constructor for Vector. Sets size = 0 and data = NULL.
-   Vector() { data.Reset(); size = 0; }
+   /** Default constructor for Vector. Sets size = 0, and calls Memory::Reset on
+       data through Memory<double>'s default constructor. */
+   Vector(): size(0) { }
 
    /// Copy constructor. Allocates a new data array and copies the data.
    Vector(const Vector &);
 
+   /// Move constructor. "Steals" data from its argument.
+   Vector(Vector&& v);
+
    /// @brief Creates vector of size s.
    /// @warning Entries are not initialized to zero!
    explicit Vector(int s);
@@ -278,6 +282,9 @@ public:
        assignment operator. */
    Vector &operator=(const Vector &v);
 
+   /// Move assignment
+   Vector &operator=(Vector&& v);
+
    /// Redefine '=' for vector = constant.
    Vector &operator=(double value);
 
@@ -503,16 +510,12 @@ inline int CheckFinite(const double *v, const int n)
 
 inline Vector::Vector(int s)
 {
+   MFEM_ASSERT(s>=0,"Unexpected negative size.");
+   size = s;
    if (s > 0)
    {
-      size = s;
       data.New(s);
    }
-   else
-   {
-      size = 0;
-      data.Reset();
-   }
 }
 
 inline void Vector::SetSize(int s)

makefile

@@ -123,7 +123,7 @@ EXAMPLE_SUBDIRS = amgx caliper ginkgo hiop petsc pumi sundials superlu
 EXAMPLE_DIRS := examples $(addprefix examples/,$(EXAMPLE_SUBDIRS))
 EXAMPLE_TEST_DIRS := examples
 
-MINIAPP_SUBDIRS = common electromagnetics meshing navier performance tools toys nurbs gslib adjoint solvers shifted mtop parelag
+MINIAPP_SUBDIRS = common electromagnetics meshing navier performance tools toys nurbs gslib adjoint solvers shifted mtop parelag autodiff
 MINIAPP_DIRS := $(addprefix miniapps/,$(MINIAPP_SUBDIRS))
 MINIAPP_TEST_DIRS := $(filter-out %/common,$(MINIAPP_DIRS))
 MINIAPP_USE_COMMON := $(addprefix miniapps/,electromagnetics meshing tools toys shifted)
@@ -274,7 +274,7 @@ endif
 # List of MFEM dependencies, that require the *_LIB variable to be non-empty
 MFEM_REQ_LIB_DEPS = SUPERLU MUMPS METIS FMS CONDUIT SIDRE LAPACK SUNDIALS MESQUITE\
  SUITESPARSE STRUMPACK GINKGO GNUTLS NETCDF PETSC SLEPC MPFR PUMI HIOP GSLIB\
- OCCA CEED RAJA UMPIRE MKL_CPARDISO AMGX CALIPER PARELAG BENCHMARK
+ OCCA CEED RAJA UMPIRE MKL_CPARDISO AMGX CALIPER PARELAG BENCHMARK 
 
 PETSC_ERROR_MSG = $(if $(PETSC_FOUND),,. PETSC config not found: $(PETSC_VARS))
 SLEPC_ERROR_MSG = $(if $(SLEPC_FOUND),,. SLEPC config not found: $(SLEPC_VARS))
@@ -340,8 +340,8 @@ MFEM_DEFINES = MFEM_VERSION MFEM_VERSION_STRING MFEM_GIT_STRING MFEM_USE_MPI\
  MFEM_USE_PUMI MFEM_USE_HIOP MFEM_USE_GSLIB MFEM_USE_CUDA MFEM_USE_HIP\
  MFEM_USE_OCCA MFEM_USE_CEED MFEM_USE_RAJA MFEM_USE_UMPIRE MFEM_USE_SIMD\
  MFEM_USE_ADIOS2 MFEM_USE_MKL_CPARDISO MFEM_USE_AMGX MFEM_USE_MUMPS\
- MFEM_USE_CALIPER MFEM_USE_BENCHMARK MFEM_USE_PARELAG\
- MFEM_SOURCE_DIR MFEM_INSTALL_DIR
+ MFEM_USE_ADFORWARD MFEM_USE_CODIPACK MFEM_USE_CALIPER MFEM_USE_BENCHMARK\
+ MFEM_USE_PARELAG MFEM_SOURCE_DIR MFEM_INSTALL_DIR
 
 # List of makefile variables that will be written to config.mk:
 MFEM_CONFIG_VARS = MFEM_CXX MFEM_HOST_CXX MFEM_CPPFLAGS MFEM_CXXFLAGS\
@@ -500,7 +500,7 @@ hpc:
 deps:
 	rm -f $(BLD)deps.mk
 	for i in $(RELSRC_FILES:.cpp=); do \
-	   $(DEP_CXX) $(MFEM_BUILD_FLAGS) -MM -MT $(BLD)$${i}.o $(SRC)$${i}.cpp\
+	   $(DEP_CXX) $(MFEM_BUILD_FLAGS) $(DEP_FLAGS) $(BLD)$${i}.o $(SRC)$${i}.cpp\
 	      >> $(BLD)deps.mk; done
 
 check: lib
@@ -679,6 +679,8 @@ status info:
 	$(info MFEM_USE_SIMD          = $(MFEM_USE_SIMD))
 	$(info MFEM_USE_ADIOS2        = $(MFEM_USE_ADIOS2))
 	$(info MFEM_USE_MKL_CPARDISO  = $(MFEM_USE_MKL_CPARDISO))
+	$(info MFEM_USE_ADFORWARD     = $(MFEM_USE_ADFORWARD))
+	$(info MFEM_USE_CODIPACK      = $(MFEM_USE_CODIPACK))
 	$(info MFEM_USE_BENCHMARK     = $(MFEM_USE_BENCHMARK))
 	$(info MFEM_USE_PARELAG       = $(MFEM_USE_PARELAG))
 	$(info MFEM_CXX               = $(value MFEM_CXX))

mesh/entsets.hpp

@@ -0,0 +1,219 @@
+// 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_ENTITY_SETS
+#define MFEM_ENTITY_SETS
+
+#include "../config/config.hpp"
+#include "../general/table.hpp"
+#include "../general/stable3d.hpp"
+#include <limits>
+#include <map>
+#include <set>
+#include <string>
+#include <vector>
+
+namespace mfem
+{
+
+class Mesh;
+class NCMesh;
+class NCEntitySets;
+
+class EntitySets
+{
+   friend class Mesh;
+   friend class NCMesh;
+   friend class NCEntitySets;
+
+public:
+   enum EntityType {INVALID = -1, VERTEX = 0, EDGE = 1, FACE = 2, ELEMENT = 3};
+
+   static std::map<EntityType,std::string> EntityTypeNames;
+
+   EntitySets(Mesh & mesh);
+   EntitySets(const EntitySets & ent_sets);
+   EntitySets(Mesh & mesh, NCMesh &ncmesh);
+
+   virtual ~EntitySets();
+
+   static const std::string & GetTypeName(EntityType t);
+
+   bool SetExists(EntityType t, unsigned int s) const;
+   bool SetExists(EntityType t, const std::string & s) const;
+
+   void Load(std::istream &input);
+   void Print(std::ostream &output) const;
+   virtual void PrintSetInfo(std::ostream &output) const;
+
+   inline Mesh *GetMesh() const { return mesh_; }
+
+   unsigned int GetNumSets(EntityType t) const;
+
+   const std::string & GetSetName(EntityType t, unsigned int s) const;
+   unsigned int        GetNumEntities(EntityType t, unsigned int s) const;
+
+   int          GetSetIndex(EntityType t, const std::string & s) const;
+   unsigned int GetNumEntities(EntityType t, const std::string & s) const;
+
+   inline std::set<int> & operator()(EntityType t, unsigned int s)
+   { return sets_[t][s]; }
+   inline const std::set<int> & operator()(EntityType t, unsigned int s) const
+   { return sets_[t][s]; }
+
+   const Table * GetEdgeVertexTable() const { return edge_vertex_; }
+   const Table * GetFaceVertexTable() const { return face_vertex_; }
+   const Table * GetFaceEdgeTable()   const { return face_edge_; }
+
+   // void Prune(int nelems);
+
+protected:
+
+   void SetNumSets(EntityType t, unsigned int n)
+   { sets_[t].resize(n); set_names_[t].resize(n); }
+   void SetSetName(EntityType t, int s, const std::string & name)
+   { set_names_[t][s] = name; set_index_by_name_[t][name] = s; }
+
+   /// Make local copies of edge_vertex, face_vertex, and face_edge tables.
+   void CopyMeshTables();
+
+   /// Refine quadrilateral mesh.
+   virtual void QuadUniformRefinement();
+
+   /// Refine hexahedral mesh.
+   virtual void HexUniformRefinement();
+
+   /// Refine 2D mesh.
+   virtual void UniformRefinement2D();
+
+   /// Refine 3D mesh.
+   virtual void UniformRefinement3D();
+
+private:
+
+   static void skip_comment_lines(std::istream &is, const char comment_char)
+   {
+      while (1)
+      {
+         is >> std::ws;
+         if (is.peek() != comment_char) { break; }
+         is.ignore(std::numeric_limits<std::streamsize>::max(), '\n');
+      }
+   }
+   // Check for, and remove, a trailing '\r'.
+   static void filter_dos(std::string &line)
+   {
+      if (!line.empty() && *line.rbegin() == '\r')
+      { line.resize(line.size()-1); }
+   }
+
+   static std::map<EntityType,std::string> init_type_names();
+
+   void LoadEntitySets(std::istream &input, EntityType t,
+                       const std::string & header);
+
+   void PrintEntitySets(std::ostream &output, EntityType t,
+                        const std::string & header) const;
+
+   void PrintEdgeSets(std::ostream &output) const;
+
+   void PrintFaceSets(std::ostream &output) const;
+
+   void PrintEntitySetInfo(std::ostream & output, EntityType t,
+                           const std::string & ent_name) const;
+
+   void CopyEntitySets(const EntitySets & ent_sets, EntityType t);
+   void BuildEntitySets(NCMesh &ncmesh, EntityType t);
+
+protected:
+
+   Mesh     * mesh_;
+   Table    * edge_vertex_;
+   Table    * face_vertex_;
+   Table    * face_edge_;
+
+   int NumOfVertices_;
+   int NumOfEdges_;
+   int NumOfElements_;
+
+   /** The node/edge/face/element indices needed by the finite element
+       space to look up DoFs. */
+   std::vector<std::vector<std::set<int> > > sets_;
+
+   /// Names of each entity set
+   std::vector<std::vector<std::string> >    set_names_;
+
+   /// Indices of each entity set indexed by set name
+   std::vector<std::map<std::string, int> >  set_index_by_name_;
+};
+
+class NCEntitySets
+{
+   friend class EntitySets;
+
+public:
+   NCEntitySets(const EntitySets & ent_sets, NCMesh &ncmesh);
+   NCEntitySets(const NCEntitySets & ncent_sets);
+
+   bool SetExists(EntitySets::EntityType t, unsigned int s) const;
+   bool SetExists(EntitySets::EntityType t, const std::string & s) const;
+
+   unsigned int GetNumSets(EntitySets::EntityType t) const;
+
+   static int GetEntitySize(EntitySets::EntityType t);
+
+   const std::string & GetSetName(EntitySets::EntityType t, int s) const;
+   unsigned int        GetNumEntities(EntitySets::EntityType t, int s) const;
+   void                GetEntityIndex(EntitySets::EntityType t, int s,
+                                      int i, Array<int> & inds) const;
+
+   int          GetSetIndex(EntitySets::EntityType t,
+                            const std::string & s) const;
+   unsigned int GetNumEntities(EntitySets::EntityType t,
+                               const std::string & s) const;
+   void         GetEntityIndex(EntitySets::EntityType t,
+                               const std::string & s, int i,
+                               Array<int> & inds) const;
+
+   inline std::vector<int> & operator()(EntitySets::EntityType t, int s)
+   { return sets_[t][s]; }
+   inline const std::vector<int> & operator()(EntitySets::EntityType t,
+                                              int s) const
+   { return sets_[t][s]; }
+   inline int & operator()(EntitySets::EntityType t, int s, int i)
+   { return sets_[t][s][i]; }
+   inline int operator()(EntitySets::EntityType t, int s, int i) const
+   { return sets_[t][s][i]; }
+
+private:
+   void CopyNCEntitySets(const NCEntitySets & ncent_sets,
+                         EntitySets::EntityType t);
+
+protected:
+
+   NCMesh * ncmesh_;
+
+   /// The nodes defining the node/edge/face/element sets
+   std::vector<std::vector<std::vector<int> > > sets_;
+
+   /// Names of each entity set
+   std::vector<std::vector<std::string> >      set_names_;
+
+   /// Indices of each entity set indexed by set name
+   std::vector<std::map<std::string, int> >    set_index_by_name_;
+
+   /// Number of indices per entity
+   static const int entity_size_[4];
+};
+
+} // namespace mfem
+
+#endif // MFEM_ENTITY_SETS

mesh/mesh.cpp

@@ -75,7 +75,7 @@ void Mesh::GetElementCenter(int i, Vector &center)
 
 double Mesh::GetElementSize(ElementTransformation *T, int type)
 {
-   DenseMatrix J(spaceDim,Dim);
+   DenseMatrix J(spaceDim, Dim);
 
    Geometry::Type geom = T->GetGeometryType();
    T->SetIntPoint(&Geometries.GetCenter(geom));
@@ -102,7 +102,7 @@ double Mesh::GetElementSize(int i, int type)
 
 double Mesh::GetElementSize(int i, const Vector &dir)
 {
-   DenseMatrix J(spaceDim,Dim);
+   DenseMatrix J(spaceDim, Dim);
    Vector d_hat(Dim);
    GetElementJacobian(i, J);
    J.MultTranspose(dir, d_hat);
@@ -1177,13 +1177,15 @@ void Mesh::Init()
    own_nodes = 1;
    NURBSext = NULL;
    ncmesh = NULL;
+   ent_sets = NULL;
    last_operation = Mesh::NONE;
 }
 
 void Mesh::InitTables()
 {
    el_to_edge =
-      el_to_face = el_to_el = bel_to_edge = face_edge = edge_vertex = NULL;
+      el_to_face = el_to_el = bel_to_edge = face_edge =
+                                               face_vertex = edge_vertex = NULL;
 }
 
 void Mesh::SetEmpty()
@@ -1205,6 +1207,7 @@ void Mesh::DestroyTables()
    }
 
    delete face_edge;
+   delete face_vertex;
    delete edge_vertex;
 }
 
@@ -1212,6 +1215,8 @@ void Mesh::DestroyPointers()
 {
    if (own_nodes) { delete Nodes; }
 
+   delete ent_sets;
+
    delete ncmesh;
 
    delete NURBSext;
@@ -3346,6 +3351,12 @@ Mesh::Mesh(const Mesh &mesh, bool copy_nodes)
    // Copy the edge-to-vertex Table, edge_vertex
    edge_vertex = (mesh.edge_vertex) ? new Table(*mesh.edge_vertex) : NULL;
 
+   // Copy the face-to-vertex Table, edge_vertex
+   face_vertex = (mesh.face_vertex) ? new Table(*mesh.face_vertex) : NULL;
+
+   // Do not copy any of the coarse (c_*), fine (f_*) or fine/coarse (fc_*)
+   // data members.
+
    // Copy the attributes and bdr_attributes
    mesh.attributes.Copy(attributes);
    mesh.bdr_attributes.Copy(bdr_attributes);
@@ -3396,6 +3407,9 @@ Mesh::Mesh(const Mesh &mesh, bool copy_nodes)
       Nodes = mesh.Nodes;
       own_nodes = 0;
    }
+
+   // Copy entity sets if present in the input mesh
+   ent_sets = (mesh.ent_sets) ? new EntitySets(*mesh.ent_sets) : NULL;
 }
 
 Mesh::Mesh(Mesh &&mesh) : Mesh()
@@ -5768,6 +5782,38 @@ Table *Mesh::GetEdgeVertexTable() const
    return edge_vertex;
 }
 
+Table *Mesh::GetFaceVertexTable() const
+{
+   if (face_vertex)
+   {
+      return face_vertex;
+   }
+
+   STable3D * faces_tbl = GetFacesTable();
+
+   int nfaces = faces_tbl->NumberOfElements();
+   face_vertex = new Table(nfaces, 4);
+   for (int i = 0; i < NumOfVertices; i++)
+   {
+      for (STable3D::RowIterator it(*faces_tbl, i); !it; ++it)
+      {
+         int j = it.Index();
+         face_vertex->Push(j, i);
+         face_vertex->Push(j, it.Column());
+         face_vertex->Push(j, it.Floor());
+         if ( it.Tier() > 0 )
+         {
+            face_vertex->Push(j, it.Tier());
+         }
+      }
+   }
+   face_vertex->Finalize();
+
+   delete faces_tbl;
+
+   return face_vertex;
+}
+
 Table *Mesh::GetVertexToElementTable()
 {
    int i, j, nv, *v;
@@ -6402,7 +6448,7 @@ void Mesh::GenerateNCFaceInfo()
    }
 }
 
-STable3D *Mesh::GetFacesTable()
+STable3D *Mesh::GetFacesTable() const
 {
    STable3D *faces_tbl = new STable3D(NumOfVertices);
    for (int i = 0; i < NumOfElements; i++)
@@ -7657,6 +7703,11 @@ void Mesh::UniformRefinement2D_base(bool update_nodes)
       NumOfEdges = GetElementToEdgeTable(*el_to_edge, be_to_edge);
    }
 
+   if ( ent_sets )
+   {
+      ent_sets->CopyMeshTables();
+   }
+
    int quad_counter = 0;
    for (int i = 0; i < NumOfElements; i++)
    {
@@ -7792,6 +7843,11 @@ void Mesh::UniformRefinement2D_base(bool update_nodes)
 
    if (update_nodes) { UpdateNodes(); }
 
+   if ( ent_sets )
+   {
+      ent_sets->UniformRefinement2D();
+   }
+
 #ifdef MFEM_DEBUG
    if (!Nodes || update_nodes)
    {
@@ -7822,6 +7878,11 @@ void Mesh::UniformRefinement3D_base(Array<int> *f2qf_ptr, DSTable *v_to_v_p,
       GetElementToFaceTable();
    }
 
+   if ( ent_sets )
+   {
+      ent_sets->CopyMeshTables();
+   }
+
    Array<int> f2qf_loc;
    Array<int> &f2qf = f2qf_ptr ? *f2qf_ptr : f2qf_loc;
    f2qf.SetSize(0);
@@ -8148,7 +8209,6 @@ void Mesh::UniformRefinement3D_base(Array<int> *f2qf_ptr, DSTable *v_to_v_p,
                }
                AverageVertices(vv, 4, oface + f2qf[f[fi]]);
             }
-
             for (int ei = 0; ei < 9; ei++)
             {
                for (int k = 0; k < 2; k++)
@@ -8492,6 +8552,11 @@ void Mesh::UniformRefinement3D_base(Array<int> *f2qf_ptr, DSTable *v_to_v_p,
    sequence++;
 
    if (update_nodes) { UpdateNodes(); }
+
+   if (ent_sets)
+   {
+      ent_sets->UniformRefinement3D();
+   }
 }
 
 void Mesh::LocalRefinement(const Array<int> &marked_el, int type)
@@ -8527,8 +8592,8 @@ void Mesh::LocalRefinement(const Array<int> &marked_el, int type)
          elements[new_e] = new Segment(new_v, vert[1], attr);
          vert[1] = new_v;
 
-         CoarseFineTr.embeddings[i] = Embedding(i, 1);
-         CoarseFineTr.embeddings[new_e] = Embedding(i, 2);
+         CoarseFineTr.embeddings[i] = Embedding(i, Geometry::SEGMENT, 1);
+         CoarseFineTr.embeddings[new_e] = Embedding(i, Geometry::SEGMENT, 2);
       }
 
       static double seg_children[3*2] = { 0.0,1.0, 0.0,0.5, 0.5,1.0 };
@@ -8961,6 +9026,8 @@ void Mesh::Swap(Mesh& other, bool non_geometry)
 
    mfem::Swap(geom_factors, other.geom_factors);
 
+   mfem::Swap(ent_sets, other.ent_sets);
+
 #ifdef MFEM_USE_MEMALLOC
    TetMemory.Swap(other.TetMemory);
 #endif
@@ -9276,7 +9343,7 @@ void Mesh::Bisection(int i, const DSTable &v_to_v,
 
       int coarse = FindCoarseElement(i);
       CoarseFineTr.embeddings[i].parent = coarse;
-      CoarseFineTr.embeddings.Append(Embedding(coarse));
+      CoarseFineTr.embeddings.Append(Embedding(coarse, Geometry::TRIANGLE));
 
       // 3. edge1 and edge2 may have to be changed for the second triangle.
       if (v[1][0] < v_to_v.NumberOfRows() && v[1][1] < v_to_v.NumberOfRows())
@@ -9396,7 +9463,7 @@ void Mesh::Bisection(int i, HashTable<Hashed2> &v_to_v)
 
       int coarse = FindCoarseElement(i);
       CoarseFineTr.embeddings[i].parent = coarse;
-      CoarseFineTr.embeddings.Append(Embedding(coarse));
+      CoarseFineTr.embeddings.Append(Embedding(coarse, Geometry::TETRAHEDRON));
 
       // 3. Set the bisection flag
       switch (type)
@@ -9534,10 +9601,10 @@ void Mesh::UniformRefinement(int i, const DSTable &v_to_v,
 
       // set parent indices
       int coarse = FindCoarseElement(i);
-      CoarseFineTr.embeddings[i] = Embedding(coarse);
-      CoarseFineTr.embeddings.Append(Embedding(coarse));
-      CoarseFineTr.embeddings.Append(Embedding(coarse));
-      CoarseFineTr.embeddings.Append(Embedding(coarse));
+      CoarseFineTr.embeddings[i] = Embedding(coarse, Geometry::TRIANGLE);
+      CoarseFineTr.embeddings.Append(Embedding(coarse, Geometry::TRIANGLE));
+      CoarseFineTr.embeddings.Append(Embedding(coarse, Geometry::TRIANGLE));
+      CoarseFineTr.embeddings.Append(Embedding(coarse, Geometry::TRIANGLE));
 
       NumOfElements += 3;
    }
@@ -9555,7 +9622,7 @@ void Mesh::InitRefinementTransforms()
    for (int i = 0; i < NumOfElements; i++)
    {
       elements[i]->ResetTransform(0);
-      CoarseFineTr.embeddings[i] = Embedding(i);
+      CoarseFineTr.embeddings[i] = Embedding(i, GetElementGeometry(i));
    }
 }
 

mesh/mesh.hpp

@@ -20,6 +20,7 @@
 #include "vertex.hpp"
 #include "vtk.hpp"
 #include "ncmesh.hpp"
+#include "entsets.hpp"
 #include "../fem/eltrans.hpp"
 #include "../fem/coefficient.hpp"
 #include "../general/zstr.hpp"
@@ -54,9 +55,11 @@ class Mesh
 #ifdef MFEM_USE_MPI
    friend class ParMesh;
    friend class ParNCMesh;
+   friend class ParEntitySets;
 #endif
    friend class NCMesh;
    friend class NURBSExtension;
+   friend class EntitySets;
 
 #ifdef MFEM_USE_ADIOS2
    friend class adios2stream;
@@ -166,6 +169,7 @@ protected:
    Array<int> be_to_face;
    mutable Table *face_edge;
    mutable Table *edge_vertex;
+   mutable Table *face_vertex;
 
    IsoparametricTransformation Transformation, Transformation2;
    IsoparametricTransformation BdrTransformation;
@@ -216,6 +220,8 @@ public:
    Array<FaceGeometricFactors*>
    face_geom_factors; ///< Optional face geometric factors.
 
+   EntitySets *ent_sets;
+
    // Global parameter that can be used to control the removal of unused
    // vertices performed when reading a mesh in MFEM format. The default value
    // (true) is set in mesh_readers.cpp.
@@ -287,7 +293,7 @@ protected:
    void PrepareNodeReorder(DSTable **old_v_to_v, Table **old_elem_vert);
    void DoNodeReorder(DSTable *old_v_to_v, Table *old_elem_vert);
 
-   STable3D *GetFacesTable();
+   STable3D *GetFacesTable() const;
    STable3D *GetElementToFaceTable(int ret_ftbl = 0);
 
    /** Red refinement. Element with index i is refined. The default
@@ -1067,9 +1073,12 @@ public:
    /// Returns the face-to-edge Table (3D)
    Table *GetFaceEdgeTable() const;
 
-   /// Returns the edge-to-vertex Table (3D)
+   /// Returns the edge-to-vertex Table (2D or 3D)
    Table *GetEdgeVertexTable() const;
 
+   /// Returns the face-to-vertex Table (2d or 3D)
+   Table *GetFaceVertexTable() const;
+
    /// Return the indices and the orientations of all faces of element i.
    void GetElementFaces(int i, Array<int> &faces, Array<int> &ori) const;
 

mesh/mesh_operators.cpp

@@ -157,6 +157,142 @@ int ThresholdDerefiner::ApplyImpl(Mesh &mesh)
 }
 
 
+int CoefficientRefiner::ApplyImpl(Mesh &mesh)
+{
+   int max_it = 1;
+   return PreprocessMesh(mesh, max_it);
+}
+
+int CoefficientRefiner::PreprocessMesh(Mesh &mesh, int max_it)
+{
+   int rank = 0;
+   MFEM_VERIFY(max_it > 0, "max_it must be strictly positive")
+
+   int dim = mesh.Dimension();
+   L2_FECollection l2fec(order, dim);
+   FiniteElementSpace* l2fes = NULL;
+
+   bool par = false;
+   GridFunction *gf = NULL;
+
+#ifdef MFEM_USE_MPI
+   ParMesh* pmesh = dynamic_cast<ParMesh*>(&mesh);
+   if (pmesh && pmesh->Nonconforming())
+   {
+      par = true;
+      l2fes = new ParFiniteElementSpace(pmesh, &l2fec);
+      gf = new ParGridFunction(static_cast<ParFiniteElementSpace*>(l2fes));
+   }
+#endif
+   if (!par)
+   {
+      l2fes = new FiniteElementSpace(&mesh, &l2fec);
+      gf = new GridFunction(l2fes);
+   }
+
+   // If custom integration rule has not been set,
+   // then use the default integration rule
+   if (!irs)
+   {
+      int order_quad = 2*order + 3;
+      for (int i=0; i < Geometry::NumGeom; ++i)
+      {
+         ir_default[i] = &(IntRules.Get(i, order_quad));
+      }
+      irs = ir_default;
+   }
+
+   for (int i = 0; i < max_it; i++)
+   {
+      // Compute number of elements and L2-norm of f.
+      int NE = mesh.GetNE();
+      int globalNE = 0;
+      double norm_of_coeff = 0.0;
+      if (par)
+      {
+#ifdef MFEM_USE_MPI
+         globalNE = pmesh->GetGlobalNE();
+         norm_of_coeff = ComputeGlobalLpNorm(2.0,*coeff,*pmesh,irs);
+#endif
+      }
+      else
+      {
+         globalNE = NE;
+         norm_of_coeff = ComputeLpNorm(2.0,*coeff,mesh,irs);
+      }
+
+      // Compute average L2-norm of f
+      double av_norm_of_coeff = norm_of_coeff / sqrt(globalNE);
+
+      // Compute element-wise L2-norms of (I - Π) f
+      Vector element_norms_of_fine_scale(NE);
+      gf->SetSpace(l2fes);
+      gf->ProjectCoefficient(*coeff);
+      gf->ComputeElementL2Errors(*coeff,element_norms_of_fine_scale,irs);
+
+      // Define osc_K(f) := || h ⋅ (I - Π) f ||_K and select elements
+      // for refinement based on threshold. Also record relative osc(f).
+      global_osc = 0.0;
+      mesh_refinements.SetSize(0);
+      element_oscs.Destroy();
+      element_oscs.SetSize(NE);
+      element_oscs = 0.0;
+      for (int j = 0; j < NE; j++)
+      {
+         double h = mesh.GetElementSize(j);
+         double element_osc = h * element_norms_of_fine_scale(j);
+         if ( element_osc > threshold * av_norm_of_coeff )
+         {
+            mesh_refinements.Append(j);
+         }
+         element_oscs(j) = element_osc/(norm_of_coeff + 1e-10);
+         global_osc += element_osc*element_osc;
+      }
+#ifdef MFEM_USE_MPI
+      if (par)
+      {
+         MPI_Comm comm = pmesh->GetComm();
+         MPI_Allreduce(MPI_IN_PLACE, &global_osc, 1, MPI_DOUBLE, MPI_SUM, comm);
+         MPI_Comm_rank(comm, &rank);
+      }
+#endif
+      global_osc = sqrt(global_osc)/(norm_of_coeff + 1e-10);
+
+      // Exit if the global threshold or maximum number of elements is reached.
+      if (global_osc < threshold || globalNE > max_elements)
+      {
+         if (global_osc > threshold && globalNE > max_elements && rank == 0 &&
+             print_level)
+         {
+            MFEM_WARNING("Reached maximum number of elements "
+                         "before resolving data to tolerance.");
+         }
+         delete l2fes;
+         delete gf;
+         return STOP;
+      }
+
+      // Refine elements.
+      mesh.GeneralRefinement(mesh_refinements, nonconforming, nc_limit);
+      l2fes->Update(false);
+      gf->Update();
+
+   }
+   delete l2fes;
+   delete gf;
+   return CONTINUE + REFINED;
+
+}
+
+void CoefficientRefiner::Reset()
+{
+   element_oscs.Destroy();
+   global_osc = 0.0;
+   coeff = NULL;
+   irs = NULL;
+}
+
+
 int Rebalancer::ApplyImpl(Mesh &mesh)
 {
 #ifdef MFEM_USE_MPI