Fix merge with global API

Merge branch 'dfem-multiple-outputs' into dfem-kernels
Remove vscode files
2026-05-01 10:28:29 -07:00 · 2026-05-01 10:03:48 -07:00 · 2026-05-01 09:49:26 -07:00 · 2026-05-01 09:47:16 -07:00 · 2026-05-01 09:30:41 -07:00 · 2026-04-29 13:09:32 -07:00
250 changed files with 12672 additions and 25420 deletions
@@ -94,16 +94,6 @@ inputs:
    description: If true, do not set any CXXFLAGS or LDFLAGS.
    default: false

-  # Unfortunately, "uses:" fields cannot have references to variables like
-  # ${{env.MFEM_ACTIONS_VERSION}}, so the branch/tag name has to be hard coded.
-  # Therefore, in the future, when updating the version of the
-  # mfem/github-actions to use, we'll have to replace:
-  # - all definitions of MFEM_ACTIONS_VERSION and
-  # - all "uses:" fields that refer to mfem/github-actions.
-  MFEM_ACTIONS_VERSION:
-    description: Version (branch or tag) of the mfem/github-actions to use.
-    default: v2.7
-
 runs:
  using: 'composite'
  steps:
@@ -128,7 +118,6 @@ runs:
        echo UBSAN_LDFLAGS=${{inputs.UBSAN_LDFLAGS}} >> $GITHUB_ENV
        echo MSAN_CXXFLAGS=${{inputs.MSAN_CXXFLAGS}} >> $GITHUB_ENV
        echo MSAN_LDFLAGS=${{inputs.MSAN_LDFLAGS}} >> $GITHUB_ENV
-        echo MFEM_ACTIONS_VERSION=${{inputs.MFEM_ACTIONS_VERSION}} >> $GITHUB_ENV
      shell: bash

    - name: Env (dir)
@@ -53,7 +53,7 @@ runs:
      run: echo CXXFLAGS=${{env.CXXFLAGS}} ${{env.UBSAN_CXXFLAGS}} >> $GITHUB_ENV
      shell: bash

-    - uses: mfem/github-actions/build-mfem@v2.7
+    - uses: mfem/github-actions/build-mfem@v2.5
      if: ${{steps.debug.outputs.cache-hit != 'true'}}
      env:
        CXXFLAGS: ${{env.CXXFLAGS}}
@@ -82,7 +82,7 @@ runs:
      run: find . -type f -name '*.o' -delete
      shell: bash

-    - uses: actions/upload-artifact@v7
+    - uses: actions/upload-artifact@v4
      with:
        name: build-${{inputs.par}}-${{inputs.sanitizer}}
        path: mfem/build
@@ -12,11 +12,6 @@
 name: 'Install MPI'
 description: 'Installs MPI and set up its environment variables'

-inputs:
-  NO_FLAGS:
-    description: If true, do not set any CXXFLAGS or LDFLAGS.
-    default: false
-
 runs:
  using: 'composite'
  steps:
@@ -32,7 +27,6 @@ runs:
      shell: bash

    - name: Env (bis)
-      if: ${{ inputs.NO_FLAGS != 'true' }}
      run: |
        echo CXXFLAGS=${{env.CXXFLAGS}} ${{env.MPI_INC}} >> $GITHUB_ENV
        echo LDFLAGS=${{env.LDFLAGS}} ${{env.MPI_LIB}} >> $GITHUB_ENV
@@ -49,7 +49,7 @@ runs:
        par: ${{inputs.par}}
        sanitizer: ${{inputs.sanitizer}}

-    - uses: actions/download-artifact@v8
+    - uses: actions/download-artifact@v4
      with:
        name: build-${{inputs.par}}-${{inputs.sanitizer}}
        path: mfem/build
@@ -37,14 +37,14 @@ runs:
      with:
        path: ${{env.HYPRE_DIR}}
        fail-on-cache-miss: true
-        key: ${{runner.os}}-ompi-build-${{env.HYPRE_DIR}}-int32-fp64-${{env.MFEM_ACTIONS_VERSION}}
+        key: ${{runner.os}}-ompi-build-${{env.HYPRE_DIR}}-int32-fp64-v2.5

    - uses: actions/cache/restore@v5  # Cache for Metis
      if: ${{inputs.par == 'true'}}
      with:
        path: ${{env.METIS_DIR}}
        fail-on-cache-miss: true
-        key: ${{runner.os}}-build-${{env.METIS_DIR}}-${{env.MFEM_ACTIONS_VERSION}}
+        key: ${{runner.os}}-build-${{env.METIS_DIR}}-v2.5

    - name: Hypre/Metis links
      if: ${{inputs.par == 'true'}}
@@ -1,42 +0,0 @@
-# MFEM Pull Request Review Agent Guide
-
-## Purpose and scope
-Review MFEM PRs for correctness, maintainability, performance, portability, test coverage, and MFEM consistency. Use the diff and PR context; reference source files, tests, and CI results when available. Follow `CONTRIBUTING.md`, especially Developer Guidelines, PR rules, checklist, and testing.
-
-## Critical review pillars
- Correctness and numerical behavior
- API and user-facing impact
- Performance implications
- Maintainability and portability
-
-## Review workflow
-1. Read the PR description, linked issues, and intended behavior.
-2. Inspect the diff before commenting.
-3. Identify affected MFEM components, examples, tests, build or docs changes, and downstream APIs.
-4. Analyze the code against the critical review pillars.
-5. Compare the change against nearby code and MFEM patterns; flag unmotivated deviations.
-6. Check whether tests and documentation were updated appropriately.
-7. Review CI results and suggest actions.
-8. Produce a structured review with prioritized findings.
-9. Always limit conclusions to available evidence.
-
-## MFEM-specific review checklist
- Component-aware scope: identify the touched subsystem (FEM, solvers, preconditioners, linear algebra, mesh, examples, miniapps, build, or docs) and assess its impact against the review pillars.
- Numerical and algorithmic behavior: assess issues in convergence, stability, tolerances, precision, iteration limits, and failure handling. If clear opportunities exist to improve the algorithmic approach, call them out with expected impact.
- API and user-facing impact: assess backward compatibility, user-visible behavior and default changes, migration impact, deprecations, and whether documentation clearly explains user-facing API changes.
- Data structure and memory semantics: assess ownership, lifetime, aliasing, container behavior, and device-host synchronization.
- Parallel and serial behavior: assess whether the change preserves equivalent semantics in serial and parallel modes where applicable; if logic is currently mode-specific, check whether extension to the other mode is straightforward (clear abstractions, no hard-wired assumptions), document constraints, and call out expected behavior differences explicitly.
- Backend and portability impact: assess likely cross-backend risks in CPU, CUDA, HIP, OCCA, RAJA, partial assembly, fallback paths, compiler compatibility, and platform assumptions.
- Build, dependency, and configuration impact: assess CMake or make changes, optional dependency behavior, and feature-flag interactions.
- Tests and docs alignment: check available regression or unit coverage evidence for changed behavior, and ensure docs are updated for new flags, APIs, options, or behavior changes.
- MFEM developer-guideline fit: keep code lean, simple, general, logically separated, and portable; suggest C++17 improvements when they clearly improve safety, clarity, or maintainability.
- New source files, examples, or miniapps: if a PR adds source/header files, verify they are properly wired into the relevant `makefile` and `CMakeLists.txt`, referenced in docs where applicable (including `doc/CodeDocumentation.dox`), and added to top-level `.gitignore` only when generated artifacts require it.
- Changelog: verify `CHANGELOG` is updated if the PR introduces significant new features or user-facing changes.
- MFEM conventions: use `real_t`; use `mfem::out`/`mfem::err` instead of `std::cout`/`std::cerr` in library code; flag large/binary files; if AI assistance is apparent but undisclosed, suggest following `CONTRIBUTING.md`.
- Edge cases: if the PR touches complex or error-prone areas, suggest additional tests for edge cases, failure modes, and parallel behavior.
-
-## Commenting guidelines
- Keep comments concise, actionable, and grounded in the diff.
- Focus on correctness, behavior changes, and user impact over style nits.
- Be professional, concise, collaborative, technically precise, and avoid unsupported assumptions.
-
@@ -13,7 +13,7 @@ Note that some of these scripts use the shared MFEM GitHub Actions from the exte

  <https://github.com/mfem/github-actions>

-For a particular action, e.g. `mfem/github-actions/build-mfem@v2.5`, the `v2.5` suffix denotes the branch (or tag) in the above from which the action is taken.
+For a particular action, e.g. `mfem/github-actions/build-mfem@v2.5`, the `v2.5` suffix denotes the branch in the above from which the action is taken.

 The current CI workflows are:

@@ -29,12 +29,16 @@ Runs a number of static repository-level sanity checks.

 - `branch-history` guards against accidental commits of large files using the `--history` option of the `config/githooks/pre-push` script.

+## `mfem-analysis.yml` (`build-analysis`)
+
+Checks if the code builds and satisfies minimal requirements.
+
+- `gitignore` builds hypre, METIS, and MFEM using `mfem/github-actions/build-hypre`, `mfem/github-actions/build-metis`, and `mfem/github-actions/build-mfem` and checks for correct `.gitignore` settings by running the `tests/scripts/gitignore` script.
+
 ## `builds-and-tests.yml`

 Runs a matrix of builds and tests runs with different compilers, OS, mfem/hypre settings, etc. Also processes and upload Codecov reports.

-One matrix job runs `tests/scripts/gitignore` after `make test-noclean` to check generated artifacts against `.gitignore`.
-
 Uses the following GitHub Actions from <https://github.com/mfem/github-actions>:

 - `mfem/github-actions/build-hypre`
@@ -40,7 +40,6 @@ env:
  METIS_ARCHIVE_MAC: metis-4.0.3-mac.tgz
  METIS_TOP_DIR: metis-4.0.3
  MFEM_TOP_DIR: mfem
-  MFEM_ACTIONS_VERSION: v2.7

 # Note for future improvements:
 #
@@ -111,7 +110,6 @@ jobs:
            build-system: make
            hypre-target: int64
            precision: fp64
-            gitignore-check: YES
          - os: ubuntu-latest
            target: opt
            codecov: NO
@@ -172,6 +170,20 @@ jobs:
          env
        shell: bash

+      # For info on Xcode see:
+      #  - https://github.com/actions/runner-images/issues/12541
+      #  - https://github.com/actions/runner-images/blob/releases/macos-15-arm64/20250811/images/macos/macos-15-arm64-Readme.md#xcode
+      - name: Xcode version setup (MacOS)
+        if: matrix.os == 'macos-latest'
+        run: |
+          XCODE_PATH="/Applications/Xcode_16.4.app"
+          echo "> sudo xcode-select -s ${XCODE_PATH}"
+          sudo xcode-select -s ${XCODE_PATH}
+          echo "> g++ -v"
+          g++ -v
+          echo "> clang++ -v"
+          clang++ -v
+
      # Only get MPI if defined for the job.
      # TODO: It would be nice to have only one step, e.g. with a dedicated
      # action, but I (@adrienbernede) don't see how at the moment.
@@ -216,11 +228,11 @@ jobs:
        uses: actions/cache@v5
        with:
          path: ${{ env.HYPRE_TOP_DIR }}
-          key: ${{ runner.os }}-ompi-build-${{ env.HYPRE_TOP_DIR }}-${{ matrix.hypre-target }}-${{ matrix.precision }}-${{ env.MFEM_ACTIONS_VERSION }}
+          key: ${{ runner.os }}-ompi-build-${{ env.HYPRE_TOP_DIR }}-${{ matrix.hypre-target }}-${{ matrix.precision }}-v2.5

      - name: get hypre
        if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os != 'windows-latest'
-        uses: mfem/github-actions/build-hypre@v2.7
+        uses: mfem/github-actions/build-hypre@v2.5
        with:
          archive: ${{ env.HYPRE_ARCHIVE }}
          dir: ${{ env.HYPRE_TOP_DIR }}
@@ -230,7 +242,7 @@ jobs:

      - name: get hypre (Windows)
        if: matrix.mpi == 'par' && steps.hypre-cache.outputs.cache-hit != 'true' && matrix.os == 'windows-latest'
-        uses: mfem/github-actions/build-hypre@v2.7
+        uses: mfem/github-actions/build-hypre@v2.5
        with:
          archive: ${{ env.HYPRE_ARCHIVE }}
          dir: ${{ env.HYPRE_TOP_DIR }}
@@ -246,11 +258,11 @@ jobs:
        uses: actions/cache@v5
        with:
          path: ${{ env.METIS_TOP_DIR }}
-          key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-${{ env.MFEM_ACTIONS_VERSION }}
+          key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.5

      - name: install metis
        if: matrix.mpi == 'par' && matrix.os != 'windows-latest' && steps.metis-cache.outputs.cache-hit != 'true'
-        uses: mfem/github-actions/build-metis@v2.7
+        uses: mfem/github-actions/build-metis@v2.5
        with:
          archive: ${{ matrix.os != 'macos-latest' && env.METIS_ARCHIVE || env.METIS_ARCHIVE_MAC }}
          dir: ${{ env.METIS_TOP_DIR }}
@@ -292,7 +304,7 @@ jobs:

      # MFEM build and test
      - name: build
-        uses: mfem/github-actions/build-mfem@v2.7
+        uses: mfem/github-actions/build-mfem@v2.5
        env:
          VCPKG_DEFAULT_BINARY_CACHE: ${{ github.workspace }}/vcpkg_cache
        with:
@@ -318,13 +330,7 @@ jobs:
      - name: tests
        if: matrix.build-system == 'make' && (matrix.target == 'opt' || matrix.os == 'ubuntu-latest')
        run: |
-          cd ${{ env.MFEM_TOP_DIR }}
-          if [[ "${{ matrix.gitignore-check }}" == "YES" ]]; then
-            make test-noclean
-          else
-            make test
-          fi
-        shell: bash
+          cd ${{ env.MFEM_TOP_DIR }} && make test

      - name: cmake checks
        if: matrix.build-system == 'cmake' && matrix.target == 'dbg'
@@ -369,16 +375,8 @@ jobs:
      # Code coverage (process and upload reports)
      - name: codecov
        if: matrix.codecov == 'YES'
-        uses: mfem/github-actions/upload-coverage@v2.7
+        uses: mfem/github-actions/upload-coverage@v2.5
        with:
-          name: ${{ matrix.os }}-${{ matrix.build-system }}-${{ matrix.target }}-${{ matrix.mpi }}-${{ matrix.hypre-target }}-${{ matrix.precision }}
+          name: ${{ matrix.os }}-${{ matrix.build-system }}-${{ matrix.target }}-${{ matrix.mpi }}-${{ matrix.hypre-target }}
          project_dir: ${{ env.MFEM_TOP_DIR }}
          directories: "fem general linalg mesh"
-        env:
-          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
-
-      - name: gitignore
-        if: matrix.gitignore-check == 'YES'
-        run: |
-          cd ${{ env.MFEM_TOP_DIR }}/tests/scripts
-          ./runtest gitignore
@@ -14,19 +14,9 @@ name: "Static Analysis"
 on:
  push:
    branches: ["master", "next"]
-    paths-ignore: &docs-only-paths
-      - "**/*.md"
-      - "doc/**"
-      - ".binder/**"
-      - "CITATION.cff"
-      - "LICENSE"
-      - "NOTICE"
-      - "CHANGELOG"
-      - "INSTALL"
  pull_request:
    # The branches below must be a subset of the branches above
    branches: ["master"]
-    paths-ignore: *docs-only-paths

 concurrency:
  group: ${{ github.workflow }}-${{ github.ref }}
@@ -0,0 +1,100 @@
+# Copyright (c) 2010-2025, 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.
+
+name: "Build Analysis"
+
+permissions:
+  actions: write
+
+on:
+  push:
+    branches:
+      - master
+      - next
+  pull_request:
+  workflow_dispatch:
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.ref }}
+  cancel-in-progress: true
+
+env:
+  HYPRE_ARCHIVE: v2.19.0.tar.gz
+  HYPRE_TOP_DIR: hypre-2.19.0
+  METIS_ARCHIVE: metis-4.0.3.tar.gz
+  METIS_TOP_DIR: metis-4.0.3
+  COVERAGE_ENV: mfem-coverage
+
+jobs:
+  gitignore:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: checkout MFEM
+        uses: actions/checkout@v6
+        with:
+          path: mfem
+
+      - name: Get MPI (Linux)
+        run: |
+          sudo apt-get install openmpi-bin libopenmpi-dev
+          export OMPI_MCA_rmaps_base_oversubscribe=1
+
+      - name: Cache Hypre Install
+        id: hypre-cache
+        uses: actions/cache@v5
+        with:
+          path: ${{ env.HYPRE_TOP_DIR }}
+          key: ${{ runner.os }}-ompi-build-${{ env.HYPRE_TOP_DIR }}-v2.5
+
+      - name: Get Hypre
+        if: steps.hypre-cache.outputs.cache-hit != 'true'
+        uses: mfem/github-actions/build-hypre@v2.5
+        with:
+          archive: ${{ env.HYPRE_ARCHIVE }}
+          dir: ${{ env.HYPRE_TOP_DIR }}
+          target: int32
+
+      - name: Cache Metis Install
+        id: metis-cache
+        uses: actions/cache@v5
+        with:
+          path: ${{ env.METIS_TOP_DIR }}
+          key: ${{ runner.os }}-build-${{ env.METIS_TOP_DIR }}-v2.5
+
+      - name: Install Metis
+        if: steps.metis-cache.outputs.cache-hit != 'true'
+        uses: mfem/github-actions/build-metis@v2.5
+        with:
+          archive: ${{ env.METIS_ARCHIVE }}
+          dir: ${{ env.METIS_TOP_DIR }}
+
+      # MFEM build and test
+      - name: build-mfem
+        uses: mfem/github-actions/build-mfem@v2.5
+        with:
+          os: ${{ runner.os }}
+          target: opt
+          codecov: NO
+          mpi: par
+          build-system: make
+          hypre-dir: ${{ env.HYPRE_TOP_DIR }}
+          metis-dir: ${{ env.METIS_TOP_DIR }}
+          mfem-dir: mfem
+
+      - name: test (no clean)
+        run: |
+          cd mfem && make test-noclean
+
+      - name: gitignore
+        run: |
+          cd mfem/tests/scripts
+          ./runtest gitignore
@@ -19,22 +19,18 @@ jobs:
    steps:
      - uses: actions/checkout@v6
      - uses: ./.github/actions/sanitize/config
-        with:
-          NO_FLAGS: true
      - name: Cache
        id: cache
        uses: actions/cache@v5
        with:
          path: ${{env.HYPRE_DIR}}
-          key: ${{runner.os}}-ompi-build-${{env.HYPRE_DIR}}-int32-fp64-${{ env.MFEM_ACTIONS_VERSION }}
+          key: ${{runner.os}}-ompi-build-${{env.HYPRE_DIR}}-int32-fp64-v2.5
      - name: Setup
        if: steps.cache.outputs.cache-hit != 'true'
        uses: ./.github/actions/sanitize/mpi
-        with:
-          NO_FLAGS: true
      - name: Build
        if: steps.cache.outputs.cache-hit != 'true'
-        uses: mfem/github-actions/build-hypre@v2.7
+        uses: mfem/github-actions/build-hypre@v2.5
        with:
          archive: ${{env.HYPRE_TGZ}}
          dir: ${{env.HYPRE_DIR}}
@@ -19,22 +19,18 @@ jobs:
    steps:
      - uses: actions/checkout@v6
      - uses: ./.github/actions/sanitize/config
-        with:
-          NO_FLAGS: true
      - name: Cache
        id: cache
        uses: actions/cache@v5
        with:
          path: ${{env.METIS_DIR}}
-          key: ${{runner.os}}-build-${{env.METIS_DIR}}-${{env.MFEM_ACTIONS_VERSION}}
+          key: ${{runner.os}}-build-${{env.METIS_DIR}}-v2.5
      - name: Setup
        if: steps.cache.outputs.cache-hit != 'true'
        uses: ./.github/actions/sanitize/mpi
-        with:
-          NO_FLAGS: true
      - name: Build
        if: steps.cache.outputs.cache-hit != 'true'
-        uses: mfem/github-actions/build-metis@v2.7
+        uses: mfem/github-actions/build-metis@v2.5
        with:
          archive: ${{env.METIS_TGZ}}
          dir: ${{env.METIS_DIR}}
@@ -146,7 +146,7 @@ jobs:
        if: ${{steps.restore.outputs.cache-hit != 'true'}}
        working-directory: mfem/build/tests/unit
        run: find . -type f -name '*.o' -delete
-      - uses: actions/upload-artifact@v7
+      - uses: actions/upload-artifact@v4
        with:
          name: tests-${{inputs.par}}-${{inputs.sanitizer}}
          path: mfem/build/tests/unit/${{env.unit_tests}}
@@ -172,7 +172,7 @@ jobs:
          par: ${{inputs.par}}
          sanitizer: ${{inputs.sanitizer}}
          cache-path: mfem/build/tests/unit/${{env.unit_tests}}
-      - uses: actions/download-artifact@v8
+      - uses: actions/download-artifact@v4
        if: ${{steps.restore.outputs.cache-hit != 'true'}}
        with:
          name: tests-${{inputs.par}}-${{inputs.sanitizer}}
@@ -17,17 +17,7 @@ permissions:
 on:
  push:
    branches: ["master", "next"]
-    paths-ignore: &docs-only-paths
-      - "**/*.md"
-      - "doc/**"
-      - ".binder/**"
-      - "CITATION.cff"
-      - "LICENSE"
-      - "NOTICE"
-      - "CHANGELOG"
-      - "INSTALL"
  pull_request:
-    paths-ignore: *docs-only-paths
  workflow_dispatch:

 concurrency:
@@ -451,8 +451,8 @@ miniapps/plasma/pic/*.csv
 tests/unit/output_meshes
 tests/unit/unit_tests
 tests/unit/punit_tests
-tests/unit/gpu_unit_tests
-tests/unit/pgpu_unit_tests
+tests/unit/cunit_tests
+tests/unit/pcunit_tests
 tests/unit/sedov_tests_*
 tests/unit/psedov_tests_*
 tests/unit/tmop_pa_tests_*
@@ -85,8 +85,3 @@ opt_par_gcc_10_pumi:
  extends: .mfem_job_on_dane
  variables:
    SPEC: "%gcc@10.3.1 +pumi"
-
-opt_par_gcc_10_gslib:
-  extends: .mfem_job_on_dane
-  variables:
-    SPEC: "%gcc@10.3.1 +gslib"
@@ -63,8 +63,3 @@ opt_mpi_cuda_hypre_cuda_gcc:
  extends: .mfem_job_on_matrix
  variables:
    SPEC: "%gcc@10.3.1 +mpi +cuda cuda_arch=90 ^hypre+cuda"
-
-opt_mpi_cuda_gcc_gslib:
-  extends: .mfem_job_on_matrix
-  variables:
-    SPEC: "%gcc@10.3.1 +mpi +cuda +gslib cuda_arch=90 ^hypre+cuda"
@@ -32,9 +32,9 @@ mkdir _${BASELINE_TEST} && cd _${BASELINE_TEST}

 # run
 if [[ "${MACHINE_NAME}" == "dane" ]]; then
-  srun --nodes=1 -t 60 --exclusive --reservation=ci ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+  salloc --nodes=1 -t 60 --exclusive --reservation=ci ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 elif [[ ${MACHINE_NAME} == "corona" ]]; then
-  srun --nodes=1 -t 60 -p pbatch ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
+  salloc --nodes=1 -t 60 -p pbatch ../runtest ../../mfem "${BASELINE_TEST} ${TPLS_DIR}"
 else
  echo "Unknown machine: MACHINE_NAME=$MACHINE_NAME"
  exit 1
@@ -8,89 +8,43 @@
                               https://mfem.org


-Version 4.9.1 (development)
-===========================
-
- Added policy for AI-assisted contribution to CONTRIBUTING.md.
+Version 4.10 (development)
+==========================

 Discretization improvements
 ---------------------------
- Improved FindPointsGSLIB surface mesh capability with support for simplices
-  and an option to specify axis-aligned bounding box padding for near-surface
-  point queries.
+- Replaced legacy simplex quadrature rules with symmetric positive-weight
+  rules for triangles (orders 0-25) and tetrahedra (orders 0-20). These
+  rules guarantee all-positive weights and interior quadrature points,
+  improving numerical stability. Higher orders fall back to Grundmann-Moller.
+  Triangle rules: Witherden & Vincent, Comput. Math. Appl. 69(10):1232-1241,
+  2015.
+  Tet rules (d=1-13): Witherden & Vincent (ibid).
+  Tet rules (d=14-20): Chuluunbaatar et al., Comput. Math. Appl. 124:89-97,
+  2022.

- Added GPU-enabled partial assembly for simplicial Bernstein H1 basis based on
-  ragged tensor algorithms (see DOI: 10.1137/11082539X) for mass and diffusion
-  integrators.

- Replaced legacy simplex quadrature rules with symmetric positive weight rules
-  for triangles (orders 0-25) and tetrahedra (orders 0-20). These rules
-  guarantee all-positive weights and interior quadrature points, improving
-  numerical stability. Higher orders fall back to Grundmann-Moller.
-  * Triangle rules: Witherden and Vincent, DOI: 10.1016/j.camwa.2015.03.017
-  * Tet rules (d=1-13): Witherden and  Vincent (same as above)
-  * Tet rules (d=14-20): Chuluunbaatar et al., DOI: 10.1016/j.camwa.2022.08.016
+Version 4.9.1 (development)
+===========================

- Added support for general 1D Gauss-Jacobi quadrature rules and Stroud conical
-  quadrature rules on triangles and tetrahedra.
-
- Improved the GridFunction projection routines. Projections work for Scalar,
+Discretization improvements
+---------------------------
+- Improved the gridfunction projection routines. Projections work for Scalar,
  Vector and VectorFE, also NURBS versions. Optionally different types of
-  projections can be selected, default behavior has not changed.
+  projections can be selected, default behaviour has not changed.

- Added GridFunction projection methods for trace spaces, i.e., project
-  coefficients on the mesh skeleton.
-
- Added methods to estimate function extremum using piecewise linear bounds plus
+- Added methods to estimate function extremum using piecewise linear bounds +
  recursive subdivision.

- Extend FindPointsGSLIB to support surface meshes.
-
 Meshing improvements
 --------------------
- Added option to guarantee mesh validity during TMOP-based r-adaptivity, using
-  bounds on the determinant of the mesh transformation Jacobian.
-
- Added PA support for TMOP's adaptive limiting functionality. Multiple
-  GridFunctions and Coefficients can be combined to form a composite term.
-
 - Improved support for 1D NURBS meshes with variable order, including using
  the patches construct for 1D NURBS meshes.

- Added the option to include material interfaces (faces separating elements
-  with different element attributes) as additional boundary elements, for
-  parallel visualization, e.g. with GLVis. This is supported by both the Print
-  and PrintAsOne methods of ParMesh. See ParMesh::SetPrintInterfaces().
-
-Linear and nonlinear solvers
----------------------------
- Added support for trace spaces in PRefinementTransferOperator. This is used in
-  PRefinement multigrid methods for problems posed on trace spaces (see e.g. the
-  DPG miniapps).
-
-GPU computing
-------------
- Added device assembly support for 3D H(curl) VectorFEDomainLFIntegrator.
-
- Added NVIDIA cuDSS library interface. Implementation examples have been
-  added to ex1 and ex1p. See https://developer.nvidia.com/cudss for more
-  details. Supported versions >= 0.6.0.
-
- Allow specifying GPU kernel launch bounds for native and RAJA GPU backends.
-
 New and updated examples and miniapps
 -------------------------------------
- The Lorentz miniapp (in miniapps/electromagnetics) has been updated to
-  leverage the ParticleSet capability.
-
- Added (Complex)PRefinementMultigrid solver option in the DPG miniapps.
-
-Miscellaneous
-------------
- Fixed signed DOF handling in ParGridFunction reading (read constructor) and
-  saving via SaveAsOne(). Simplified the process of applying the DOF signs by
-  using the new method ApplyDofSigns() in class ParFiniteElementSpace: the
-  method will return immediately if no sign flips are needed.
+- Electromagnetics/lorentz miniapp has been updated to leverage the ParticleSet
+  capability.


 Version 4.9, released on Dec 11, 2025
@@ -433,15 +433,6 @@ if (MFEM_USE_STRUMPACK)
  endif()
 endif()

-# cuDSS can only be enabled in CUDA
-if (MFEM_USE_CUDSS)
-  if (MFEM_USE_CUDA)
-    find_package(CUDSS REQUIRED)
-  else()
-    message(FATAL_ERROR " *** cuDSS requires that CUDA be enabled.")
-  endif()
-endif()
-
 # GnuTLS
 if (MFEM_USE_GNUTLS)
  find_package(_GnuTLS REQUIRED)
@@ -601,6 +592,13 @@ if (MFEM_USE_ENZYME)
    set(ENZYME_INCLUDE_DIRS ${ENZYME_DIR}/include)
 endif()

+if (MFEM_USE_PROTEUS)
+    enable_language(C)
+    find_package(proteus REQUIRED PATHS "${PROTEUS_DIR}")
+    message(STATUS "${PROTEUS_DIR}/include")
+    include_directories("${PROTEUS_DIR}/include")
+endif()
+
 # MFEM_TIMER_TYPE
 if (NOT DEFINED MFEM_TIMER_TYPE)
  if (APPLE)
@@ -640,7 +638,7 @@ find_package(Threads REQUIRED)
 set(MFEM_TPLS OPENMP HYPRE LAPACK BLAS SuperLUDist STRUMPACK METIS SuiteSparse
    SUNDIALS PETSC SLEPC MUMPS AXOM FMS CONDUIT Ginkgo GNUTLS GSLIB HDF5
    NETCDF MPFR PUMI HIOP POSIXCLOCKS MFEMBacktrace ZLIB OCCA CEED RAJA UMPIRE
-    ADIOS2 MKL_CPARDISO MKL_PARDISO AMGX MAGMA CUSPARSE CUBLAS CUDSS CALIPER CODIPACK
+    ADIOS2 MKL_CPARDISO MKL_PARDISO AMGX MAGMA CUSPARSE CUBLAS CALIPER CODIPACK
    BENCHMARK PARELAG TRIBOL MPI_CXX HIP HIPBLAS HIPSPARSE MOONOLITH BLITZ
    ALGOIM ENZYME CUDA::cudart)

@@ -737,6 +735,16 @@ mfem_add_library(mfem ${SOURCES} ${HEADERS} ${MASTER_HEADERS})
 target_compile_features(mfem PUBLIC cxx_std_${CMAKE_CXX_STANDARD})
 # message(STATUS "TPL_LIBRARIES = ${TPL_LIBRARIES}")
 target_link_libraries(mfem PUBLIC ${TPL_LIBRARIES} ${TPL_TARGETS})
+
+if (MFEM_USE_PROTEUS)
+    add_library(ClangProteusFlags INTERFACE IMPORTED)
+    set_target_properties(ClangProteusFlags PROPERTIES
+        INTERFACE_COMPILE_OPTIONS "-fpass-plugin=$<TARGET_FILE:ProteusPass>"
+    )
+    target_link_libraries(mfem PUBLIC ClangProteusFlags)
+    target_link_libraries(mfem PUBLIC proteus)
+endif()
+
 if (TPL_TARGETS)
  add_dependencies(mfem ${TPL_TARGETS})
 endif()
@@ -3,12 +3,12 @@
 </p>

 <p align="center">
-<a href="https://github.com/mfem/mfem/blob/master/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-BSD-blue.svg"></a>
-<a href="https://github.com/mfem/mfem/releases/latest"><img alt="GitHub release" src="https://img.shields.io/github/v/release/mfem/mfem"></a>
-<a href="https://github.com/mfem/mfem/actions/workflows/repo-check.yml?query=branch%3Amaster"><img alt="Repo check" src="https://github.com/mfem/mfem/actions/workflows/repo-check.yml/badge.svg?branch=master"></a>
-<a href="https://github.com/mfem/mfem/actions/workflows/builds-and-tests.yml?query=branch%3Amaster"><img alt="Builds and Tests" src="https://github.com/mfem/mfem/actions/workflows/builds-and-tests.yml/badge.svg?branch=master"></a>
+<a href="https://github.com/mfem/mfem/blob/master/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-BSD-brightgreen.svg"></a>
+<a href="https://github.com/mfem/mfem/actions?query=workflow%3Arepo-check+branch%3Amaster"><img alt="Repo check" src="https://github.com/mfem/mfem/actions/workflows/repo-check.yml/badge.svg?branch=master"></a>
+<a href="https://github.com/mfem/mfem/actions?query=workflow%3Abuild-analysis+branch%3Amaster"><img alt="Build Analysis" src="https://github.com/mfem/mfem/actions/workflows/mfem-analysis.yml/badge.svg?branch=master"></a>
+<a href="https://github.com/mfem/mfem/actions?query=workflow%3Abuilds-and-tests+branch%3Amaster"><img alt="Builds and Tests" src="https://github.com/mfem/mfem/actions/workflows/builds-and-tests.yml/badge.svg?branch=master"></a>
 <a href="https://ci.appveyor.com/project/mfem/mfem"><img alt="Build Status" src="https://ci.appveyor.com/api/projects/status/19non9sqm6msi2wy?svg=true"></a>
-<a href="https://docs.mfem.org/html/index.html"><img alt="Documentation" src="https://img.shields.io/badge/code-documented-brightgreen.svg"></a>
+<a href="https://docs.mfem.org/html/index.html"><img alt="Doxygen" src="https://img.shields.io/badge/code-documented-brightgreen.svg"></a>
 </p>


@@ -24,14 +24,6 @@ must be made under this license.
 Note also that MFEM has a [Code of Conduct](CODE_OF_CONDUCT.md). By participating
 in the MFEM community, you agree to abide by its rules.

-## AI Policy
- Use of AI code generation in MFEM is allowed but must be disclosed, e.g. by
-  selecting the `AI-assisted` label on the PR.
- By submitting a PR, the author acknowledges that they have reviewed and
-  understand the changes they are proposing.
- PR authors are still responsible for correctness, licensing, and attribution
-  of all changes.
-
 If you plan on contributing to MFEM, consider reviewing the
 [issue tracker](https://github.com/mfem/mfem/issues) first to check if a thread
 already exists for your desired feature or the bug you ran into. Use a pull
@@ -84,7 +76,7 @@ Origin](#developers-certificate-of-origin-11) at the end of this file.*
  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
+- PRs are treated similarly to journal submission with an "editor" assigning two
  reviewers to evaluate the changes.
 - The reviewers have 3 weeks to evaluate the PR and work with the author to
  fix issues and implement improvements.
@@ -125,7 +117,7 @@ The MFEM source code has the following structure:
  │   ├── petsc
  │   ├── pumi
  │   ├── sundials
-  │   └── superlu
+  |   └── superlu
  ├── fem
  │   ├── ceed
  │   ├── dfem
@@ -137,6 +129,10 @@ The MFEM source code has the following structure:
  │   ├── moonolith
  │   ├── qinterp
  │   └── tmop
+  │   |   ├── assemble
+  │   |   ├── metrics
+  │   |   ├── mult
+  │   |   └── tools
  ├── general
  ├── linalg
  │   ├── batched
@@ -149,10 +145,11 @@ The MFEM source code has the following structure:
  │   ├── common
  │   ├── contact
  │   ├── dfem
-  │   ├── diag-smoothers
  │   ├── dpg
  │   ├── electromagnetics
  │   ├── fluids
+  │   │   ├── navier
+  │   │   └── schrodinger-flow
  │   ├── gslib
  │   ├── hdiv-linear-solver
  │   ├── hooke
@@ -162,7 +159,6 @@ The MFEM source code has the following structure:
  │   ├── nurbs
  │   ├── parelag
  │   ├── performance
-  │   ├── plasma
  │   ├── shifted
  │   ├── solvers
  │   ├── spde
@@ -193,15 +189,15 @@ respectively.

 - The main finite element classes are:
  + [`FiniteElement`](https://docs.mfem.org/html/classmfem_1_1FiniteElement.html)
-  + [`FiniteElementCollection`](https://docs.mfem.org/html/classmfem_1_1FiniteElementCollection.html)
+  + [`FiniteElementCollection`](https://docs.mfem.org/html/classmfem_1_1FiniteElement.html)
  + [`FiniteElementSpace`](https://docs.mfem.org/html/classmfem_1_1FiniteElementSpace.html)
  + [`GridFunction`](https://docs.mfem.org/html/classmfem_1_1GridFunction.html)
  + [`BilinearFormIntegrator`](https://docs.mfem.org/html/classmfem_1_1BilinearFormIntegrator.html) and [`LinearFormIntegrator`](https://docs.mfem.org/html/classmfem_1_1LinearFormIntegrator.html)
-  + [`LinearForm`](https://docs.mfem.org/html/classmfem_1_1LinearForm.html), [`BilinearForm`](https://docs.mfem.org/html/classmfem_1_1BilinearForm.html) and [`MixedBilinearForm`](https://docs.mfem.org/html/classmfem_1_1MixedBilinearForm.html)
+  + [`LinearForm`](https://docs.mfem.org/html/classmfem_1_1LinearFormIntegrator.html), [`BilinearForm`](https://docs.mfem.org/html/classmfem_1_1BilinearForm.html) and [`MixedBilinearForm`](https://docs.mfem.org/html/classmfem_1_1MixedBilinearForm.html)

 - The main linear algebra classes and sources are
  + [`Operator`](https://docs.mfem.org/html/classmfem_1_1Operator.html) and [`BilinearForm`](https://docs.mfem.org/html/classmfem_1_1BilinearForm.html)
-  + [`Vector`](https://docs.mfem.org/html/classmfem_1_1Vector.html) and [`LinearForm`](https://docs.mfem.org/html/classmfem_1_1LinearForm.html)
+  + [`Vector`](https://docs.mfem.org/html/classmfem_1_1BilinearForm.html) and [`LinearForm`](https://docs.mfem.org/html/classmfem_1_1LinearForm.html)
  + [`DenseMatrix`](https://docs.mfem.org/html/classmfem_1_1DenseMatrix.html) and [`SparseMatrix`](https://docs.mfem.org/html/classmfem_1_1SparseMatrix.html)
  + Sparse [smoothers](https://docs.mfem.org/html/sparsesmoothers_8hpp.html) and linear [solvers](https://docs.mfem.org/html/solvers_8hpp.html)

@@ -213,8 +209,8 @@ shared geometric entities between different tasks. The parallel source files
 have a `p` prefix, e.g. `pmesh.cpp` vs. the serial `mesh.cpp`.

 - The main parallel classes are
-  + [`ParMesh`](https://docs.mfem.org/html/classmfem_1_1ParMesh.html)
-  + [`ParNCMesh`](https://docs.mfem.org/html/classmfem_1_1ParNCMesh.html)
+  + [`ParMesh`](https://docs.mfem.org/html/solvers_8hpp.html)
+  + [`ParNCMesh`](https://docs.mfem.org/html/classmfem_1_1ParMesh.html)
  + [`ParFiniteElementSpace`](https://docs.mfem.org/html/classmfem_1_1ParFiniteElementSpace.html)
  + [`ParGridFunction`](https://docs.mfem.org/html/classmfem_1_1ParGridFunction.html)
  + [`ParBilinearForm`](https://docs.mfem.org/html/classmfem_1_1ParBilinearForm.html) and [`ParLinearForm`](https://docs.mfem.org/html/classmfem_1_1ParLinearForm.html)
@@ -224,14 +220,14 @@ have a `p` prefix, e.g. `pmesh.cpp` vs. the serial `mesh.cpp`.
 #### GPU and general device support

 GPU and multi-core CPU support is based on device kernels supporting different
-backends (CUDA, HIP, OCCA, RAJA, OpenMP, etc.) and an internal lightweight
+backends (CUDA, OCCA, RAJA, OpenMP, etc.) and an internal lightweight
 device/host memory manager.

 - The main device-relevant classes and sources are:
  + [`Device`](https://docs.mfem.org/html/device_8hpp.html)
  + [`MemoryManager`](https://docs.mfem.org/html/mem_manager_8hpp.html)
  + the [`mfem::forall`](https://docs.mfem.org/html/forall_8hpp.html) function
-  + the [`cuda.hpp`](https://docs.mfem.org/html/cuda_8hpp.html), [`hip.hpp`](https://docs.mfem.org/html/hip_8hpp.html) and [`occa.hpp`](https://docs.mfem.org/html/occa_8hpp.html) files
+  + the [`cuda.hpp`](https://docs.mfem.org/html/cuda_8hpp.html) and [`occa.hpp`](https://docs.mfem.org/html/occa_8hpp.html) files

 #### Utilities, building and documentation
 - The `general/` directory contains C++ classes that serve as utilities for
@@ -245,8 +241,8 @@ device/host memory manager.
 - `examples` and `miniapps` respectively gather simple and more fully-featured
  demonstrations of the usage on MFEM. They both rely on `data/` for the
  collection of meshes.
- The `tests/` directory contains a unit test suite, additional tests, and
-  benchmarks.
+- The `tests/` directory contains a unit test suite and will later contain more
+  tests that run example codes.

 See also the [code overview](https://mfem.org/code-overview/) section on the MFEM
 website.
@@ -280,8 +276,8 @@ Before you can start, you need a GitHub account, here are a few suggestions:
  the top of https://github.com/mfem.
 - Consider making your membership public by going to https://github.com/orgs/mfem/people
  and clicking on the organization visibility drop box next to your name.
- Project discussions and announcements will be posted at https://github.com/orgs/mfem/discussions,
-  tagging the `@mfem/everyone` team when appropriate.
+- Project discussions and announcements will be posted at
+  https://github.com/orgs/mfem/teams/everyone.

 #### Structure
 - The MFEM source code is in the [mfem](https://github.com/mfem/mfem)
@@ -341,12 +337,11 @@ Before you can start, you need a GitHub account, here are a few suggestions:
  - Well-designed simple code is frequently more general and powerful.
  - Lean code base is easier to understand by new collaborators.
  - New features should be added only if they are necessary or generally useful.
-  - Introduction of language constructs not currently used in MFEM should be
+  - Introduction of language constructions not currently used in MFEM should be
    justified and generally avoided (to maintain portability to various systems
    and compilers, including early access hardware).
-  - We prefer basic C++. Use C++17 features judiciously, prioritizing readability,
-    consistency with existing MFEM code, and portability to different systems,
-    compilers and device backends.
+  - We prefer basic C++ and the C++03 standard, to keep the code readable by
+    a large audience and to make sure it compiles anywhere.

 - *Keep the code general and reasonably efficient*
  - The main goal is fast prototyping for research and application development.
@@ -389,7 +384,7 @@ Before you can start, you need a GitHub account, here are a few suggestions:
 - When your branch is ready for other developers to review / comment on
  the code, create a pull request towards `mfem:master`.

- Pull requests typically have titles like:
+- Pull request typically have titles like:

     `Description [new-feature-dev]`

@@ -410,12 +405,12 @@ Before you can start, you need a GitHub account, here are a few suggestions:
 - Add a description, appropriate labels and assign yourself to the PR. The MFEM
  team will add reviewers as appropriate.

- List outstanding TODO items in the description.
+- List outstanding TODO items in the description, see PR #222 for an example.

 - When your contribution is fully working and ready to be reviewed, add
-  or request the `ready-for-review` label.
+  the `ready-for-review` label.

- PRs are treated similarly to journal submission, with an "editor" assigning
+- PRs are treated similarly to journal submission with an "editor" assigning
  two reviewers to evaluate the changes. The reviewers have 3 weeks to evaluate
  the PR and work with the author to implement improvements and fix issues.

@@ -441,7 +436,7 @@ Before you can start, you need a GitHub account, here are a few suggestions:
  checks in GitHub Actions enforce MFEM-specific rules which are explained in
  the error messages and the `tests/scripts` directory.

- Also note that the tests `branch-history` and `repo-check` found in GitHub
+- Also note that the tests `branch-history` and `repos-checks` found in GitHub
  Actions can be triggered automatically before each push using git hooks. See
  the [git hooks README](config/githooks/README.md) for a detailed explanation.

@@ -498,15 +493,15 @@ Everyone on the MFEM team can be asked to serve as a reviewer on a PR in their a

 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.

-4. To seek help from the editors in case of difficulties.
+3. To seek help from the editors in case of difficulties.

-5. To complete the review in a timely manner: 3 weeks from assignment.
+4. To complete the review in a timely manner: 3 weeks from assignment.

-6. 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.
+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.

-7. 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*.
+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*.

-8. The review of bugfixes should be expedited proportional to their importance. The review window can be much less than three weeks in such cases.
+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

@@ -532,30 +527,30 @@ Before a PR can be merged, it should satisfy the following:
 - [ ] Code builds.
 - [ ] Code passes `make style`.
 - [ ] Update `CHANGELOG`:
-   - [ ] Is this a new feature users need to be aware of? New or updated example or miniapp?
-   - [ ] Does it make sense to create a new section in the `CHANGELOG` to group with other related features?
+    - [ ] Is this a new feature users need to be aware of? New or updated example or miniapp?
+    - [ ] Does it make sense to create a new section in the `CHANGELOG` to group with other related features?
 - [ ] Update `INSTALL`:
-   - [ ] Has a new optional library been added? If so, what range of versions of this library are required? (*Make sure the external library is compatible with our BSD license, e.g. it is not licensed under GPL!*)
-   - [ ] Have the version ranges for any required or optional libraries changed?
-   - [ ] Does `make` or `cmake` have a new target?
-   - [ ] Did the requirements or the installation process change? *(rare)*
+    - [ ] Had a new optional library been added? If so, what range of versions of this library are required? (*Make sure the external library is compatible with our BSD license, e.g. it is not licensed under GPL!*)
+    - [ ] Have the version ranges for any required or optional libraries changed?
+    - [ ] Does `make` or `cmake` have a new target?
+    - [ ] Did the requirements or the installation process change? *(rare)*
 - [ ] Update continuous integration server configurations if necessary (e.g. with new version requirements for each of MFEM's dependencies)
-   - [ ] `.github`
-   - [ ] `.appveyor.yml`
+    - [ ] `.github`
+    - [ ] `.appveyor.yml`
 - [ ] Update `.gitignore`:
-   - [ ] Check if `make distclean; git status` shows any files that were generated from the source by the project (not an IDE) but we don't want to track in the repository.
-   - [ ] Add new patterns (just for the new files above) and re-run the above test.
+    - [ ] Check if `make distclean; git status` shows any files that were generated from the source by the project (not an IDE) but we don't want to track in the repository.
+    - [ ] Add new patterns (just for the new files above) and re-run the above test.
 - [ ] New examples:
-   - [ ] All sample runs at the top of the example source file work.
-   - [ ] Update `examples/makefile`:
+    - [ ] All sample runs at the top of the example source file work.
+    - [ ] Update `examples/makefile`:
      - [ ] Add the example code to the appropriate `SEQ_EXAMPLES` and `PAR_EXAMPLES` variables.
      - [ ] Add any files generated by it to the `clean` target.
      - [ ] Add the example binary and any files generated by it to the top-level `.gitignore` file.
-   - [ ] Update `examples/CMakeLists.txt`:
+    - [ ] Update `examples/CMakeLists.txt`:
      - [ ] Add the example code to the `ALL_EXE_SRCS` variable.
      - [ ] Make sure `THIS_TEST_OPTIONS` is set correctly for the new example.
   - [ ] List the new example in `doc/CodeDocumentation.dox`.
-   - [ ] If new examples directory (e.g. `examples/pumi`), list it in `doc/CodeDocumentation.conf.in`
+   - [ ] If new examples directory (e.g.`examples/pumi`), list it in `doc/CodeDocumentation.conf.in`
   - [ ] Companion pull request for documentation in [mfem/web](https://github.com/mfem/web) repo:
      - [ ] Update or add example-specific documentation, see e.g. the `src/examples.md`.
      - [ ] Add the description, labels and screenshots in `src/examples.md` and `src/img`.
@@ -572,13 +567,13 @@ Before a PR can be merged, it should satisfy the following:
      - [ ] Add/update the `CMakeLists.txt` file in the new miniapp directory.
      - [ ] Consider adding a new test for the new miniapp.
   - [ ] List the new miniapp in `doc/CodeDocumentation.dox`
-   - [ ] If new miniapps directory (e.g. `miniapps/nurbs`), add it to `MINIAPP_SUBDIRS` in the `makefile`.
-   - [ ] If new miniapps directory (e.g. `miniapps/nurbs`), list it in `doc/CodeDocumentation.conf.in`
+   - [ ] If new miniapps directory (e.g.`miniapps/nurbs`), add it to `MINIAPP_SUBDIRS` in the `makefile`.
+   - [ ] If new miniapps directory (e.g.`miniapps/nurbs`), list it in `doc/CodeDocumentation.conf.in`
   - [ ] Companion pull request for documentation in [mfem/web](https://github.com/mfem/web) repo:
-      - [ ] Update or add miniapp-specific documentation, see e.g. the `src/meshing.md` and `src/electromagnetics.md` files.
-      - [ ] Add the description, labels and screenshots in `src/examples.md` and `src/img`.
-      - [ ] 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`.
+     - [ ] Update or add miniapp-specific documentation, see e.g. the `src/meshing.md` and `src/electromagnetics.md` files.
+     - [ ] Add the description, labels and screenshots in `src/examples.md` and `src/img`.
+     - [ ] 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 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.
@@ -680,7 +675,7 @@ MFEM uses a `master`/`next`-branch workflow as described below:
 - [ ] Update URL shortlinks:
    - [ ] Create a shortlink at [http://bit.ly/](http://bit.ly/) for the release tarball, e.g. https://mfem.github.io/releases/mfem-3.1.tgz.
    - [ ] (LLNL only) Add and commit the new shortlink in the `links` and `links-mfem` files of the internal `mfem/downloads` repo.
-    - [ ] Add the new shortlinks to the MFEM package in `spack`.
+    - [ ] Add the new shortlinks to the MFEM packages in `spack`, `homebrew/science`, `VisIt`, etc.
 - [ ] Update website in `mfem/web` repo:
    - 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`.
@@ -732,24 +727,22 @@ commit or push, see the [README](config/githooks/README.md) in the `config/githo
 directory.


-### GitHub Actions smoke tests
-
+### 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
 [https://github.com/mfem/mfem/actions](https://github.com/mfem/mfem/actions).

-GitHub Actions testing should be kept lightweight, as there is a time
-constraint on jobs. The current workflows cover Linux, macOS, and Windows
-configurations.
+Testing using GitHub Actions should be kept lightweight, as there is a time
+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.

-### Additional Windows smoke test

-We also 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.yml` file
-and the build logs at
+### 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
+build logs at
 [https://ci.appveyor.com/project/mfem/mfem](https://ci.appveyor.com/project/mfem/mfem).

 CMake is used to generate the MSVC Project files and drive the build.  A release
@@ -38,13 +38,14 @@ the option MFEM_USE_METIS.
 MFEM also includes support for devices such as GPUs, and programming models such
 as CUDA, HIP, OCCA, OpenMP and RAJA.

- Starting with version 4.9, MFEM requires a C++17 compiler.
+- Starting with version 4.0, MFEM requires a C++11 compiler. We recommend using
+  a newer compiler, e.g. GCC version 4.9 or higher.

 - CUDA support requires an NVIDIA GPU and an installation of the CUDA Toolkit
  https://developer.nvidia.com/cuda-toolkit

 - HIP support requires an AMD GPU and an installation of the ROCm software stack
-  https://rocm.docs.amd.com
+  https://rocmdocs.amd.com

 - OCCA support requires the OCCA library
  https://libocca.org
@@ -82,9 +83,9 @@ Serial build:
 Parallel build:
   (download hypre and METIS 4 from above URLs)
   (build METIS 4 in ../metis-4.0 relative to mfem/)
-   (for METIS 5, see https://mfem.org/building/#parallel-build-using-metis-5)
   (build hypre in ../hypre relative to mfem/)
   make parallel -j 4
+(For METIS 5, see https://mfem.org/building/#parallel-build-using-metis-5)

 CUDA build:
   make cuda -j 4
@@ -114,14 +115,14 @@ Serial build:
 Parallel build:
   (download hypre and METIS 4 from above URLs)
   (build METIS 4 in ../metis-4.0 relative to mfem/)
-   (for METIS 5, see https://mfem.org/building/#parallel-build-using-metis-5)
   (build hypre in ../hypre relative to mfem/)
   mkdir <mfem-build-dir> ; cd <mfem-build-dir>
   cmake <mfem-source-dir> -DMFEM_USE_MPI=YES
   make -j 4
+(For METIS 5, see https://mfem.org/building/#parallel-build-using-metis-5)

 Parallel build with fetching of hypre and METIS:
-   mkdir <mfem-build-dir> ; cd <mfem-build-dir>
+   mkdir <mfem-buil-dir> ; cd <mfem-build-dir>
   cmake <mfem-source-dir> -DMFEM_USE_MPI=YES -DMFEM_FETCH_TPLS=YES
   make -j 4

@@ -133,8 +134,7 @@ CUDA build:

 HIP build:
   mkdir <mfem-build-dir> ; cd <mfem-build-dir>
-   cmake <mfem-source-dir> -DMFEM_USE_HIP=YES -DHIP_ARCH=gfx942 \
-         -DCMAKE_CXX_COMPILER=amdclang++ -DCMAKE_HIP_COMPILER=amdclang++
+   cmake <mfem-source-dir> -DMFEM_USE_HIP=YES -DHIP_ARCH=gfx942 -DCMAKE_CXX_COMPILER=amdclang++ -DCMAKE_HIP_COMPILER=amdclang++
   make -j 4

 Example codes (serial/parallel, depending on the build):
@@ -269,7 +269,6 @@ Compilers:
   CXX      - C++ compiler, serial build
   MPICXX   - MPI C++ compiler, parallel build
   CUDA_CXX - The CUDA compiler, 'nvcc' or 'clang++'
-   HIP_CXX  - The HIP compiler, e.g. 'hipcc'

 Compiler options:
   OPTIM_FLAGS - Options for optimized build
@@ -396,11 +395,6 @@ MFEM_USE_STRUMPACK = YES/NO
   classes. When enabled, this option uses the STRUMPACK_* library options, see
   below.

-MFEM_USE_CUDSS = YES/NO
-   Enable MFEM functionality based on the cuDSS library. When using cuDSS, CUDA
-   support must be also enabled in MFEM, i.e. MFEM_USE_CUDA=YES must be set.
-   When enabled, this option uses the CUDSS_* library options, see below.
-
 MFEM_USE_GINKGO = YES/NO
   Enable MFEM functionality based on the Ginkgo library, which provides
   iterative linear solvers and preconditioners with OpenMP, CUDA backends, see
@@ -560,13 +554,13 @@ MFEM_USE_RAJA = YES/NO
 MFEM_USE_OCCA = YES/NO
   Enables support for the OCCA library in MFEM. OCCA is an open-source library
   which aims to make it easy to program different types of devices (e.g. CPU,
-   GPU, FPGA) by providing a unified API for interacting with JIT-compiled
+   GPU, FPGA) by providing an unified API for interacting with JIT-compiled
   backends. In order to use the OCCA CUDA backend, CUDA support must be enabled
   in MFEM as well, i.e. MFEM_USE_CUDA=YES must be set.

 MFEM_USE_GSLIB = YES/NO
   Enables MFEM functionality based on the GSLIB library, and specifically its
-   FindPoints component, which provides robust algorithms to evaluate finite
+   FindPoints component, which provides a robust algorithms to evaluate finite
   element functions in a collection of points in physical space. When enabled,
   the user can use the GSLIB-FindPoints methods as shown in miniapps/gslib.

@@ -725,18 +719,9 @@ The specific libraries and their options are:
  Options: STRUMPACK_OPT, STRUMPACK_LIB.
  Versions: STRUMPACK >= 3.0.0.

- CUDSS (optional), used when MFEM_USE_CUDSS = YES. Note that CUDSS requires 
-  CUDA 12.x toolkit and the cuDSS libraries. The supported communication backend
-  is OpenMPI 4.x (default), and OpenMPI 4.x or a later version must be pre-built.
-  The source files in the cuDSS tarball provide guidance for developing custom 
-  MPI implementations.
-  URL: https://developer.nvidia.com/cudss
-       https://docs.nvidia.com/cuda/cudss/advanced_features.html#communication-layer-library-in-cudss
-  Options: CUDSS_OPT, CUDSS_LIB.
-  Versions: cuDSS >= 0.6.0.
-
- Ginkgo (optional), used when MFEM_USE_GINKGO = YES. Ginkgo may have additional
-  requirements and module-specific dependencies; see the webpage below.
+- Ginkgo (optional), used when MFEM_USE_GINKGO = YES. Note that Ginkgo needs a
+  C++ compiler that supports the C++-17 standard. For additional requirements
+  and dependencies of specific modules, see the Ginkgo webpage below.
  URL: https://ginkgo-project.github.io
  Options: GINKGO_OPT, GINKGO_LIB, GINKGO_DIR, GINKGO_BUILD_TYPE (Release or
           Debug).
@@ -808,7 +793,7 @@ The specific libraries and their options are:
  Options: CONDUIT_OPT, CONDUIT_LIB.
  Versions: Conduit >= 0.3.1.

- ADIOS2 (optional), used when MFEM_USE_ADIOS2 = YES.
+- ADIOS2 (optional) used when MFEM_USE_ADIOS2 = YES.
  URL: https://adios2.readthedocs.io/
  Versions: ADIOS >= 2.5.0.

@@ -884,7 +869,7 @@ The specific libraries and their options are:
  Options: RAJA_DIR, RAJA_OPT, RAJA_LIB.
  Versions: RAJA >= 2022.10.3.

- Moonolith (optional), used when MFEM_USE_MOONOLITH = YES.
+- Moonolith (optional), use when MFEM_USE_MOONOLITH = YES.
  URL: https://bitbucket.org/zulianp/par_moonolith
  Options: MOONOLITH_DIR
  Versions: MOONOLITH >= 1.1.0.
@@ -972,7 +957,7 @@ CMAKE_BUILD_TYPE which can be set to standard values like "Debug", and "Release"
 To use a specific generator use the "-G <generator>" option of cmake:

   cmake <mfem-source-dir> -G "Xcode"
-   cmake <mfem-source-dir> -G "Visual Studio 17 2022"
+   cmake <mfem-source-dir> -G "Visual Studio 12 2013"
   cmake <mfem-source-dir> -G "MinGW Makefiles"

 With CMake it is possible to build MFEM as a shared library using the standard
@@ -1217,7 +1202,7 @@ larger problems, there are two options:
 Specific options for HIP
 ========================
 MFEM expects the `ROCM_PATH` environment variable to be set to the path of the
-ROCm install, as well as having `$ROCM_PATH/bin` in `PATH`.
+ROCM install, as well as having `$ROCM_PATH/bin` in `PATH`.

 Specific options for RAJA+HIP+MPI
 =================================
@@ -28,7 +28,6 @@ license files. These software products and their licenses are as follows:
 * AmgXWrapper (linalg/amgxsolver.{hpp,cpp}) -- MIT license
 * Catch++ (tests/unit/catch.hpp) -- Boost 1.0 license
 * Gecko (general/gecko.{cpp,hpp}) -- BSD 3-clause license
-* gslib (fem/gslib.{cpp,hpp}, mesh/bb_grid_map.{cpp,hpp}) -- BSD 3-clause license
 * Picojson (fem/picojson.h) -- Custom 2-clause license
 * TinyXML2 (general/tinyxml2.{cpp,h}) -- zlib license
 * Zstr (general/zstr.hpp) -- MIT license
@@ -35,7 +35,6 @@ set(MFEM_USE_SUITESPARSE @MFEM_USE_SUITESPARSE@)
 set(MFEM_USE_SUPERLU @MFEM_USE_SUPERLU@)
 set(MFEM_USE_MUMPS @MFEM_USE_MUMPS@)
 set(MFEM_USE_STRUMPACK @MFEM_USE_STRUMPACK@)
-set(MFEM_USE_CUDSS @MFEM_USE_CUDSS@)
 set(MFEM_USE_GINKGO @MFEM_USE_GINKGO@)
 set(MFEM_USE_AMGX @MFEM_USE_AMGX@)
 set(MFEM_USE_MAGMA @MFEM_USE_MAGMA@)
@@ -110,10 +109,6 @@ if (MFEM_USE_RAJA)
   find_dependency(RAJA)
 endif()

-if (MFEM_USE_CUDSS)
-  find_dependency(cudss)
-endif (MFEM_USE_CUDSS)
-
 if (MFEM_USE_UMPIRE)
   find_dependency(umpire)
 endif()
@@ -108,15 +108,6 @@
 // Enable MFEM functionality based on the STRUMPACK library.
 #cmakedefine MFEM_USE_STRUMPACK

-// Enable MFEM functionality based on the cuDSS library.
-#cmakedefine MFEM_USE_CUDSS
-
-// CUDSS communication layer library path
-#cmakedefine MFEM_CUDSS_COMM_LIB "@MFEM_CUDSS_COMM_LIB@"
-
-// CUDSS threading layer library path
-#cmakedefine MFEM_CUDSS_THREADING_LIB "@MFEM_CUDSS_THREADING_LIB@"
-
 // Enable functionality based on the Ginkgo library.
 #cmakedefine MFEM_USE_GINKGO

@@ -1,68 +0,0 @@
-if (NOT cudss_DIR AND CUDSS_DIR)
-  set(cudss_DIR ${CUDSS_DIR}/lib/cmake/cudss)
-endif()
-message(STATUS "Looking for CUDSS ...")
-message(STATUS "   in CUDSS_DIR = ${CUDSS_DIR}")
-message(STATUS "      cudss_DIR = ${cudss_DIR}")
-find_package(cudss)
-set(CUDSS_FOUND ${cudss_FOUND})
-set(CUDSS_LIBRARIES "cudss")
-if (CUDSS_FOUND)
-  message(STATUS
-    "Found CUDSS target: ${CUDSS_LIBRARIES} (version: ${cudss_VERSION})")
-else()
-  set(msg STATUS)
-  if (CUDSS_FIND_REQUIRED)
-    set(msg FATAL_ERROR)
-  endif()
-  message(${msg}
-    "CUDSS not found. Please set CUDSS_DIR to the install prefix.")
-endif()
-
-if(CUDSS_FOUND AND TARGET cudss)
-  get_target_property(CUDSS_LIBRARY_LOCATION cudss IMPORTED_LOCATION)
-  if(NOT CUDSS_LIBRARY_LOCATION)
-    get_target_property(CUDSS_LIBRARY_LOCATION cudss IMPORTED_LOCATION_RELEASE)
-  endif()
-  if(CUDSS_LIBRARY_LOCATION)
-    get_filename_component(CUDSS_LIBRARY_DIR "${CUDSS_LIBRARY_LOCATION}" DIRECTORY)
-  else()
-    message(WARNING "Could not determine the location of the cuDSS library.")
-  endif()
-else()
-    message(WARNING "cuDSS target not available; cannot determine library directory.")
-endif()
-
-# Set the full name of the cuDSS threading library if OpenMP is enabled.
-# The threading layer library (libcudss_mtlayer_gomp.so) is located under the
-# cuDSS library directory by default.
-if (MFEM_USE_OPENMP)
-  find_file(
-    CUDSS_THREADING_LIB
-    NAMES libcudss_mtlayer_gomp.so
-    PATHS ${CUDSS_LIBRARY_DIR}
-    NO_DEFAULT_PATH
-  )
-  if (NOT DEFINED MFEM_CUDSS_THREADING_LIB AND CUDSS_THREADING_LIB)
-    set(MFEM_CUDSS_THREADING_LIB "${CUDSS_THREADING_LIB}")
-  endif()
-  message(STATUS "CUDSS threading layer library: ${MFEM_CUDSS_THREADING_LIB}")
-endif()
-
-# Set the full name of the cuDSS communication library if MFEM use OpenMPI.
-# The communication layer library (libcudss_commlayer_mpi.so) is located under the
-# cuDSS library directory by default.
-# The communication layer library is used pre-built communication layers for OpenMPI 
-# by default.
-if (MFEM_USE_MPI)
-  find_file(
-    CUDSS_COMM_LIB
-    NAMES libcudss_commlayer_openmpi.so
-    PATHS ${CUDSS_LIBRARY_DIR}
-    NO_DEFAULT_PATH
-  )
-  if (NOT DEFINED MFEM_CUDSS_COMM_LIB AND CUDSS_COMM_LIB)
-    set(MFEM_CUDSS_COMM_LIB "${CUDSS_COMM_LIB}")
-  endif()
-  message(STATUS "CUDSS communication layer library: ${MFEM_CUDSS_COMM_LIB}")
-endif()
@@ -18,17 +18,19 @@

 if (MFEM_FETCH_GSLIB OR MFEM_FETCH_TPLS)
  enable_language(C)
-  set(GSLIB_FETCH_VERSION 1.0.9)
-  add_library(GSLIB STATIC IMPORTED)
-  # set options (technically flags because GSLIB does not use cmake)
  string(TOUPPER "${CMAKE_BUILD_TYPE}" BUILD_TYPE)
-  set(GSLIB_FLAGS "${CMAKE_C_FLAGS} ${CMAKE_C_FLAGS_${BUILD_TYPE}}")
-  if (BUILD_SHARED_LIBS)
-    set(GSLIB_FLAGS "${GSLIB_FLAGS} -fPIC")
+  set(GSLIB_FETCH_VERSION 1.0.9)
+  set(GSLIB_C_FLAGS ${CMAKE_C_FLAGS_${BUILD_TYPE}})
+  if (CMAKE_C_FLAGS)
+    set(GSLIB_C_FLAGS "${CMAKE_C_FLAGS} ${CMAKE_C_FLAGS_${BUILD_TYPE}}")
  endif()
+  if (BUILD_SHARED_LIBS)
+    set(GSLIB_C_FLAGS "${GSLIB_C_FLAGS} -fPIC")
+  endif()
+  add_library(GSLIB STATIC IMPORTED)
  # define external project and create future include directory so it is present
  # to pass CMake checks at end of MFEM configuration step
-  message(STATUS "Will fetch GSLIB ${GSLIB_FETCH_VERSION} to be built with ${GSLIB_FLAGS}")
+  message(STATUS "Will fetch GSLIB ${GSLIB_FETCH_VERSION} to be built with ${GSLIB_C_FLAGS}")
  set(PREFIX ${CMAKE_BINARY_DIR}/fetch/gslib)
  include(ExternalProject)
  ExternalProject_Add(gslib
@@ -38,7 +40,7 @@ if (MFEM_FETCH_GSLIB OR MFEM_FETCH_TPLS)
    UPDATE_DISCONNECTED TRUE
    PREFIX ${PREFIX}
    CONFIGURE_COMMAND ""
-    BUILD_COMMAND cd ${PREFIX}/src/gslib && $(MAKE) clean && $(MAKE) DESTDIR=${PREFIX} MPI=$<BOOL:${MFEM_USE_MPI}> "CFLAGS=${GSLIB_FLAGS}"
+    BUILD_COMMAND cd ${PREFIX}/src/gslib && $(MAKE) clean && $(MAKE) DESTDIR=${PREFIX} MPI=$<BOOL:${MFEM_USE_MPI}> "CFLAGS= ${GSLIB_C_FLAGS}"
    INSTALL_COMMAND "")
  file(MAKE_DIRECTORY ${PREFIX}/include)
  # set imported library target properties
@@ -44,9 +44,6 @@ if (MFEM_FETCH_HYPRE OR MFEM_FETCH_TPLS)
  # set options and associated dependencies
  set(HYPRE_CMAKE_OPTIONS "")
  list(APPEND HYPRE_CMAKE_OPTIONS -DCMAKE_BUILD_TYPE:STRING=${CMAKE_BUILD_TYPE})
-  if (BUILD_SHARED_LIBS)
-    list(APPEND HYPRE_CMAKE_OPTIONS -DCMAKE_POSITION_INDEPENDENT_CODE:BOOL=ON)
-  endif()
  # collect all HYPRE_ENABLE variables and pass them to hypre, assuming they are BOOL.
  get_cmake_property(all_vars VARIABLES)
  foreach(var ${all_vars})
@@ -98,6 +95,7 @@ if (MFEM_FETCH_HYPRE OR MFEM_FETCH_TPLS)
    UPDATE_DISCONNECTED TRUE
    SOURCE_SUBDIR src
    PREFIX ${HYPRE_INSTALL}
+    BUILD_COMMAND ${CMAKE_COMMAND} --build . -- -j${CMAKE_BUILD_PARALLEL_LEVEL}
    CMAKE_CACHE_ARGS -DCMAKE_INSTALL_PREFIX:PATH=${HYPRE_INSTALL} -DCMAKE_INSTALL_LIBDIR:PATH=lib ${HYPRE_CMAKE_OPTIONS})
  file(MAKE_DIRECTORY ${HYPRE_INSTALL}/include)
  # set imported library target properties
@@ -19,18 +19,10 @@
 #   - METIS_VERSION_5 (cache variable)

 if (MFEM_FETCH_METIS OR MFEM_FETCH_TPLS)
-  enable_language(C)
  set(METIS_FETCH_VERSION 4.0.3)
  add_library(METIS STATIC IMPORTED)
-  # set options (technically flags because METIS does not use cmake)
-  set(METIS_FLAGS "-Wno-implicit-int -Wno-incompatible-pointer-types")
-  string(TOUPPER "${CMAKE_BUILD_TYPE}" BUILD_TYPE)
-  set(METIS_FLAGS "${METIS_FLAGS} ${CMAKE_C_FLAGS} ${CMAKE_C_FLAGS_${BUILD_TYPE}}")
-  if (BUILD_SHARED_LIBS)
-    set(METIS_FLAGS "${METIS_FLAGS} -fPIC")
-  endif()
  # define external project
-  message(STATUS "Will fetch METIS ${METIS_FETCH_VERSION} to be built with ${METIS_FLAGS}")
+  message(STATUS "Will fetch METIS ${METIS_FETCH_VERSION} to be built with default options")
  set(PREFIX ${CMAKE_BINARY_DIR}/fetch/metis)
  include(ExternalProject)
  ExternalProject_Add(metis
@@ -40,7 +32,7 @@ if (MFEM_FETCH_METIS OR MFEM_FETCH_TPLS)
    UPDATE_DISCONNECTED TRUE
    PREFIX ${PREFIX}
    CONFIGURE_COMMAND tar -xzf ../metis/metis-${METIS_FETCH_VERSION}-mac.tgz --strip=1
-    BUILD_COMMAND $(MAKE) clean && $(MAKE) "OPTFLAGS=${METIS_FLAGS}"
+    BUILD_COMMAND $(MAKE) COPTIONS=-Wno-incompatible-pointer-types
    INSTALL_COMMAND mkdir -p ${PREFIX}/lib && cp libmetis.a ${PREFIX}/lib/)
  # set imported library target properties
  add_dependencies(METIS metis)
@@ -22,15 +22,15 @@ include(MfemCmakeUtilities)
 mfem_find_package(SuiteSparse SuiteSparse SuiteSparse_DIR "" "" "" ""
  "Paths to headers required by SuiteSparse."
  "Libraries required by SuiteSparse."
-  ADD_COMPONENT "UMFPACK" "include;include/suitesparse;suitesparse" umfpack.h "lib" umfpack
-  ADD_COMPONENT "KLU" "include;include/suitesparse;suitesparse" klu.h "lib" klu
-  ADD_COMPONENT "AMD" "include;include/suitesparse;suitesparse" amd.h "lib" amd
-  ADD_COMPONENT "BTF" "include;include/suitesparse;suitesparse" btf.h "lib" btf
-  ADD_COMPONENT "CHOLMOD" "include;include/suitesparse;suitesparse" cholmod.h "lib" cholmod
-  ADD_COMPONENT "COLAMD" "include;include/suitesparse;suitesparse" colamd.h "lib" colamd
-  ADD_COMPONENT "CAMD" "include;include/suitesparse;suitesparse" camd.h "lib" camd
-  ADD_COMPONENT "CCOLAMD" "include;include/suitesparse;suitesparse" ccolamd.h "lib" ccolamd
-  ADD_COMPONENT "config" "include;include/suitesparse;suitesparse" SuiteSparse_config.h "lib"
+  ADD_COMPONENT "UMFPACK" "include;suitesparse" umfpack.h "lib" umfpack
+  ADD_COMPONENT "KLU" "include;suitesparse" klu.h "lib" klu
+  ADD_COMPONENT "AMD" "include;suitesparse" amd.h "lib" amd
+  ADD_COMPONENT "BTF" "include;suitesparse" btf.h "lib" btf
+  ADD_COMPONENT "CHOLMOD" "include;suitesparse" cholmod.h "lib" cholmod
+  ADD_COMPONENT "COLAMD" "include;suitesparse" colamd.h "lib" colamd
+  ADD_COMPONENT "CAMD" "include;suitesparse" camd.h "lib" camd
+  ADD_COMPONENT "CCOLAMD" "include;suitesparse" ccolamd.h "lib" ccolamd
+  ADD_COMPONENT "config" "include;suitesparse" SuiteSparse_config.h "lib"
    suitesparseconfig)

 if (SuiteSparse_FOUND AND METIS_VERSION_5)
@@ -157,10 +157,22 @@ constexpr real_t operator""_r(unsigned long long v)
 #endif
 #endif // MFEM_USE_MPI not defined

-#ifndef MFEM_USE_CUDA
-#ifdef MFEM_USE_CUDSS
-#error Building with cuDSS (MFEM_USE_CUDSS=YES) requires CUDA (MFEM_USE_CUDA=YES)
+#ifdef NVTX_DBG_HPP
+#include NVTX_DBG_HPP
+#else
+#define db1(...)
+#define dbg(...)
+#define dbl(...)
+#define dba(...)
+#define dbc(...)
+#define NVTX_MARK_FUNCTION
+#define NVTX_MARK_BEGIN(...)
+#define NVTX_INI(...)
+#define NVTX_END(...)
+#define NVTX_MARK_INI(...)
+#define NVTX_MARK_END(...)
+#define NVTX_MARK(...)
+#define NVTX(...)
 #endif
-#endif // MFEM_USE_CUDSS not defined

 #endif // MFEM_CONFIG_HPP
@@ -108,15 +108,6 @@
 // Enable MFEM functionality based on the STRUMPACK library.
 // #define MFEM_USE_STRUMPACK

-// Enable MFEM functionality based on the cuDSS library.
-// #define MFEM_USE_CUDSS
-
-// CUDSS communication layer library path
-// #define MFEM_CUDSS_COMM_LIB "@MFEM_CUDSS_COMM_LIB@"
-
-// CUDSS threading layer library path
-// #define MFEM_CUDSS_THREADING_LIB "@MFEM_CUDSS_THREADING_LIB@"
-
 // Enable MFEM features based on the Ginkgo library.
 // #define MFEM_USE_GINKGO

@@ -36,9 +36,6 @@ MFEM_USE_SUPERLU       = @MFEM_USE_SUPERLU@
 MFEM_USE_SUPERLU5      = @MFEM_USE_SUPERLU5@
 MFEM_USE_MUMPS         = @MFEM_USE_MUMPS@
 MFEM_USE_STRUMPACK     = @MFEM_USE_STRUMPACK@
-MFEM_USE_CUDSS         = @MFEM_USE_CUDSS@
-MFEM_CUDSS_COMM_LIB    = @MFEM_CUDSS_COMM_LIB@
-MFEM_CUDSS_THREADING_LIB = @MFEM_CUDSS_THREADING_LIB@
 MFEM_USE_GINKGO        = @MFEM_USE_GINKGO@
 MFEM_USE_AMGX          = @MFEM_USE_AMGX@
 MFEM_USE_MAGMA         = @MFEM_USE_MAGMA@
@@ -38,7 +38,6 @@ option(MFEM_USE_SUPERLU "Enable SuperLU_DIST usage" OFF)
 option(MFEM_USE_SUPERLU5 "Use the old SuperLU_DIST 5.1 version" OFF)
 option(MFEM_USE_MUMPS "Enable MUMPS usage" OFF)
 option(MFEM_USE_STRUMPACK "Enable STRUMPACK usage" OFF)
-option(MFEM_USE_CUDSS "Enable cuDSS usage" OFF)
 option(MFEM_USE_GINKGO "Enable Ginkgo usage" OFF)
 option(MFEM_USE_AMGX "Enable AmgX usage" OFF)
 option(MFEM_USE_MAGMA "Enable MAGMA usage" OFF)
@@ -153,7 +153,6 @@ MFEM_USE_SUPERLU       = NO
 MFEM_USE_SUPERLU5      = NO
 MFEM_USE_MUMPS         = NO
 MFEM_USE_STRUMPACK     = NO
-MFEM_USE_CUDSS         = NO
 MFEM_USE_GINKGO        = NO
 MFEM_USE_AMGX          = NO
 MFEM_USE_MAGMA         = NO
@@ -369,19 +368,6 @@ STRUMPACK_OPT = -I$(STRUMPACK_DIR)/include $(SCOTCH_OPT)
 STRUMPACK_LIB = -L$(STRUMPACK_DIR)/lib -lstrumpack $(MPI_FORTRAN_LIB)\
 $(SCOTCH_LIB) $(SCALAPACK_LIB)

-# CUDSS library configuration
-CUDSS_DIR         = @MFEM_DIR@/../cudss
-CUDSS_INCLUDE_DIR = $(CUDSS_DIR)/include
-CUDSS_LIBRARY_DIR = $(CUDSS_DIR)/lib
-CUDSS_OPT         = -I$(CUDSS_INCLUDE_DIR)
-CUDSS_LIB         = \
- $(XLINKER)-rpath,$(CUDSS_LIBRARY_DIR) -L$(CUDSS_LIBRARY_DIR) -lcudss
-# The cuDSS communication and threading libraries. 
-MFEM_CUDSS_COMM_LIB = $(abspath $(wildcard $(or $(CUDSS_COMM_LIB),\
-   $(subst @MFEM_DIR@,$(MFEM_DIR), $(CUDSS_LIBRARY_DIR)/libcudss_commlayer_openmpi.so))))
-MFEM_CUDSS_THREADING_LIB = $(abspath $(wildcard $(or $(CUDSS_THREADING_LIB),\
-   $(subst @MFEM_DIR@,$(MFEM_DIR),$(CUDSS_LIBRARY_DIR)/libcudss_mtlayer_gomp.so))))
-
 # Ginkgo library configuration
 GINKGO_DIR = @MFEM_DIR@/../ginkgo/install
 GINKGO_SEARCH_DIR = $(subst @MFEM_DIR@,$(MFEM_DIR),$(GINKGO_DIR))
@@ -635,7 +621,7 @@ PARELAG_LIB = -L$(PARELAG_DIR)/build/src -lParELAG
 AXOM_DIR = @MFEM_DIR@/../axom
 TRIBOL_DIR = @MFEM_DIR@/../tribol
 TRIBOL_OPT = -I$(TRIBOL_DIR)/include -I$(AXOM_DIR)/include
-TRIBOL_LIB = -L$(TRIBOL_DIR)/lib -ltribol -ltribol_shared -lredecomp -L$(AXOM_DIR)/lib -laxom_mint\
+TRIBOL_LIB = -L$(TRIBOL_DIR)/lib -ltribol -lredecomp -L$(AXOM_DIR)/lib -laxom_mint\
   -laxom_slam -laxom_slic -laxom_core

 # Enzyme configuration
@@ -47,6 +47,7 @@ list(APPEND ALL_EXE_SRCS
  ex39.cpp
  ex40.cpp
  ex41.cpp
+  jitplayground.cpp
  )

 if (MFEM_USE_MPI)
@@ -215,7 +216,7 @@ if (MFEM_ENABLE_TESTING)
    add_test(NAME ex1p_ceed_np=${MFEM_MPI_NP}
      COMMAND ${MPIEXEC} ${MPIEXEC_NUMPROC_FLAG} ${MFEM_MPI_NP}
      ${MPIEXEC_PREFLAGS}
-      $<TARGET_FILE:ex1p> "-no-vis" "-d" "ceed-cpu" "-pa" "-a"
+      $<TARGET_FILE:ex1p> "-no-vis" "-d ceed-cpu" "-pa" "-a"
      ${MPIEXEC_POSTFLAGS})
  endif()
 endif()
@@ -64,7 +64,7 @@ PARALLEL_NAME := Parallel AMGX example
 $(MFEM_LIB_FILE):
 	$(error The MFEM library is not build)

-clean: clean-build clean-exec
+clean: clean-build

 clean-build:
 	rm -f *.o *~ $(SEQ_EXAMPLES) $(PAR_EXAMPLES)
@@ -64,12 +64,12 @@ ex1p-test-par: ex1p
 $(MFEM_LIB_FILE):
 	$(error The MFEM library is not built)

-clean: clean-build clean-exec
+clean: clean-build clean-exec $(SUBDIRS_CLEAN)

 clean-build:
 	rm -f *.o *~ $(SEQ_EXAMPLES) $(PAR_EXAMPLES)
 	rm -rf *.dSYM *.TVD.*breakpoints

 clean-exec:
-	@rm -f refined.mesh mesh.*
-	@rm -f sol.*
+	@rm -f refined.mesh displaced.mesh mesh.* ex5.mesh
+	@rm -f sphere_refined.* sol.* sol_u.* sol_p.* sol_r.* sol_i.*
@@ -50,10 +50,6 @@
 //               ex1 -m ../data/beam-tet.mesh -pa -d ceed-cpu
 //               ex1 -m ../data/beam-tet.mesh -pa -d ceed-cuda:/gpu/cuda/ref
 //
-// Device simplices sample runs:
-//               ex1 -pa -d gpu -m ../data/inline-tet.mesh
-//               ex1 -pa -d gpu -m ../data/inline-tri.mesh
-//
 // Description:  This example code demonstrates the use of MFEM to define a
 //               simple finite element discretization of the Poisson problem
 //               -Delta u = 1 with homogeneous Dirichlet boundary conditions.
@@ -142,25 +138,25 @@ int main(int argc, char *argv[])
   }

   // 5. 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.
-   //    - If the mesh is simplicial and partial assembly is requested,
-   //      we use the positive basis, which supports device execution.
+   //    Lagrange finite elements of the specified order. If order < 1, we
+   //    instead use an isoparametric/isogeometric space.
   FiniteElementCollection *fec;
-   auto basis_type = (pa && mesh.IsSimplexMesh()) ?
-                     BasisType::Positive : BasisType::GaussLobatto;
+   bool delete_fec;
   if (order > 0)
   {
-      fec = new H1_FECollection(order, dim, basis_type);
+      fec = new H1_FECollection(order, dim);
+      delete_fec = true;
   }
   else if (mesh.GetNodes())
   {
      fec = mesh.GetNodes()->OwnFEC();
+      delete_fec = false;
      cout << "Using isoparametric FEs: " << fec->Name() << endl;
   }
   else
   {
-      fec = new H1_FECollection(order = 1, dim, basis_type);
+      fec = new H1_FECollection(order = 1, dim);
+      delete_fec = true;
   }
   FiniteElementSpace fespace(&mesh, fec);
   cout << "Number of finite element unknowns: "
@@ -228,29 +224,17 @@ int main(int argc, char *argv[])
   // 11. Solve the linear system A X = B.
   if (!pa)
   {
-#ifdef MFEM_USE_CUDSS
-      if (Device::Allows(Backend::CUDA_MASK))
-      {
-         // Use cuDSS to solve the system.
-         CuDSSSolver cudss_solver;
-         cudss_solver.SetOperator(*A);
-         cudss_solver.Mult(B, X);
-      }
-      else
-#endif
-      {
 #ifndef MFEM_USE_SUITESPARSE
-         // Use a simple symmetric Gauss-Seidel preconditioner with PCG.
-         GSSmoother M((SparseMatrix&)(*A));
-         PCG(*A, M, B, X, 1, 200, 1e-12, 0.0);
+      // Use a simple symmetric Gauss-Seidel preconditioner with PCG.
+      GSSmoother M((SparseMatrix&)(*A));
+      PCG(*A, M, B, X, 1, 200, 1e-12, 0.0);
 #else
-         // If MFEM was compiled with SuiteSparse, use UMFPACK to solve the system.
-         UMFPackSolver umf_solver;
-         umf_solver.Control[UMFPACK_ORDERING] = UMFPACK_ORDERING_METIS;
-         umf_solver.SetOperator(*A);
-         umf_solver.Mult(B, X);
+      // 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
-      }
   }
   else
   {
@@ -289,14 +273,17 @@ int main(int argc, char *argv[])
   if (visualization)
   {
      char vishost[] = "localhost";
-      int visport = 19916;
+      int  visport   = 19916;
      socketstream sol_sock(vishost, visport);
      sol_sock.precision(8);
      sol_sock << "solution\n" << mesh << x << flush;
   }

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

   return 0;
 }
@@ -42,11 +42,7 @@
 //               mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared
 //               mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/square-mixed.mesh
 //               mpirun -np 4 ex1p -pa -d ceed-cuda:/gpu/cuda/shared -m ../data/fichera-mixed.mesh
-//               mpirun -np 4 ex1p -pa -d ceed-cpu -m ../data/beam-tet.mesh
-//
-// Device simplices sample runs:
-//               mpirun -np 4 ex1p -pa -d gpu -m ../data/inline-tet.mesh
-//               mpirun -np 4 ex1p -pa -d gpu -m ../data/inline-tri.mesh
+//               mpirun -np 4 ex1p -m ../data/beam-tet.mesh -pa -d ceed-cpu
 //
 // Description:  This example code demonstrates the use of MFEM to define a
 //               simple finite element discretization of the Poisson problem
@@ -87,9 +83,6 @@ int main(int argc, char *argv[])
   const char *device_config = "cpu";
   bool visualization = true;
   bool algebraic_ceed = false;
-#ifdef MFEM_USE_CUDSS
-   bool cudss_solver = false;
-#endif

   OptionsParser args(argc, argv);
   args.AddOption(&mesh_file, "-m", "--mesh",
@@ -109,10 +102,6 @@ int main(int argc, char *argv[])
   args.AddOption(&algebraic_ceed, "-a", "--algebraic",
                  "-no-a", "--no-algebraic",
                  "Use algebraic Ceed solver");
-#endif
-#ifdef MFEM_USE_CUDSS
-   args.AddOption(&cudss_solver, "-cudss", "--cudss-solver", "-no-cudss",
-                  "--no-cudss-solver", "Use the cuDSS Solver.");
 #endif
   args.AddOption(&visualization, "-vis", "--visualization", "-no-vis",
                  "--no-visualization",
@@ -169,20 +158,19 @@ int main(int argc, char *argv[])
   }

   // 7. Define a parallel finite element space on the parallel mesh. Here we
-   //    use continuous Lagrange finite elements of the specified order.
-   //    - If order < 1, we instead use an isoparametric/isogeometric space.
-   //    - If the mesh is simplicial and partial assembly is requested,
-   //      we use the positive basis, which supports device execution.
+   //    use continuous Lagrange finite elements of the specified order. If
+   //    order < 1, we instead use an isoparametric/isogeometric space.
   FiniteElementCollection *fec;
-   auto basis_type = (pa && pmesh.IsSimplexMesh()) ?
-                     BasisType::Positive : BasisType::GaussLobatto;
+   bool delete_fec;
   if (order > 0)
   {
-      fec = new H1_FECollection(order, dim, basis_type);
+      fec = new H1_FECollection(order, dim);
+      delete_fec = true;
   }
   else if (pmesh.GetNodes())
   {
      fec = pmesh.GetNodes()->OwnFEC();
+      delete_fec = false;
      if (myid == 0)
      {
         cout << "Using isoparametric FEs: " << fec->Name() << endl;
@@ -190,7 +178,8 @@ int main(int argc, char *argv[])
   }
   else
   {
-      fec = new H1_FECollection(order = 1, dim, basis_type);
+      fec = new H1_FECollection(order = 1, dim);
+      delete_fec = true;
   }
   ParFiniteElementSpace fespace(&pmesh, fec);
   HYPRE_BigInt size = fespace.GlobalTrueVSize();
@@ -259,51 +248,33 @@ int main(int argc, char *argv[])
   // 13. Solve the linear system A X = B.
   //     * With full assembly, use the BoomerAMG preconditioner from hypre.
   //     * With partial assembly, use Jacobi smoothing, for now.
-#ifdef MFEM_USE_CUDSS
-   if (!pa && (Device::Allows(Backend::CUDA_MASK) && cudss_solver))
+   Solver *prec = NULL;
+   if (pa)
   {
-      // Solve using a direct solver with cuDSS
-      CuDSSSolver cudss_solver(MPI_COMM_WORLD);
-      cudss_solver.SetMatrixSymType(
-         CuDSSSolver::SYMMETRIC_POSITIVE_DEFINITE);
-      cudss_solver.SetMatrixViewType(CuDSSSolver::UPPER);
-      cudss_solver.SetOperator(*A);
-      cudss_solver.Mult(B, X);
-   }
-   else
-#endif
-   {
-      Solver *prec = NULL;
-      if (pa)
+      if (UsesTensorBasis(fespace))
      {
-         if (UsesTensorBasis(fespace))
+         if (algebraic_ceed)
         {
-            if (algebraic_ceed)
-            {
-               prec = new ceed::AlgebraicSolver(a, ess_tdof_list);
-            }
-            else
-            {
-               prec = new OperatorJacobiSmoother(a, ess_tdof_list);
-            }
+            prec = new ceed::AlgebraicSolver(a, ess_tdof_list);
+         }
+         else
+         {
+            prec = new OperatorJacobiSmoother(a, ess_tdof_list);
         }
      }
-      else
-      {
-         prec = new HypreBoomerAMG;
-      }
-      CGSolver cg(MPI_COMM_WORLD);
-      cg.SetRelTol(1e-12);
-      cg.SetMaxIter(2000);
-      cg.SetPrintLevel(1);
-      if (prec)
-      {
-         cg.SetPreconditioner(*prec);
-      }
-      cg.SetOperator(*A);
-      cg.Mult(B, X);
-      delete prec;
   }
+   else
+   {
+      prec = new HypreBoomerAMG;
+   }
+   CGSolver cg(MPI_COMM_WORLD);
+   cg.SetRelTol(1e-12);
+   cg.SetMaxIter(2000);
+   cg.SetPrintLevel(1);
+   if (prec) { cg.SetPreconditioner(*prec); }
+   cg.SetOperator(*A);
+   cg.Mult(B, X);
+   delete prec;

   // 14. Recover the parallel grid function corresponding to X. This is the
   //     local finite element solution on each processor.
@@ -337,7 +308,10 @@ int main(int argc, char *argv[])
   }

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

   return 0;
 }
@@ -95,15 +95,6 @@ int main(int argc, char *argv[])
      args.PrintOptions(cout);
   }

-   if (amg_elast && !static_cond && reorder_space)
-   {
-      if (myid == 0)
-         cerr << "\nThe AMG elasticity solver requires ordering byVDIM! "
-              << "Ignoring the specified option -nodes/--by-nodes.\n"
-              << endl;
-      reorder_space = false;
-   }
-
   // 3. Enable hardware devices such as GPUs, and programming models such as
   //    CUDA, OCCA, RAJA and OpenMP based on command line options.
   Device device(device_config);
@@ -76,4 +76,4 @@ clean-build:
 	rm -rf *.dSYM *.TVD.*breakpoints

 clean-exec:
-	@rm -f refined.mesh sol.gf mesh.* sol.*
+	@rm -f refined.mesh sol.gf
@@ -0,0 +1,536 @@
+#include <mfem.hpp>
+
+#include "../fem/dfem/util.hpp"
+
+#include <proteus/CppJitModule.h>
+
+#include "jitplayground.hpp"
+
+#include <algorithm>
+#include <array>
+#include <cctype>
+#include <cmath>
+#include <fstream>
+#include <initializer_list>
+#include <iostream>
+#include <memory>
+#include <sstream>
+#include <string>
+#include <string_view>
+#include <type_traits>
+#include <unordered_map>
+#include <unordered_set>
+#include <utility>
+#include <vector>
+
+namespace util
+{
+constexpr std::string_view Dirname(std::string_view path)
+{
+   const size_t last_sep = path.find_last_of("/\\");
+   if (last_sep == std::string_view::npos) { return {}; }
+   return path.substr(0, last_sep);
+}
+
+constexpr std::string_view thisFileDir = Dirname(__FILE__);
+}
+
+template <typename T>
+static std::string TypeNameString()
+{
+   return std::string(mfem::future::get_type_name<T>());
+}
+
+template <typename Tuple, size_t... Is>
+static auto ParamTypeStringsImpl(std::index_sequence<Is...>)
+{
+   return std::array<std::string, sizeof...(Is)>
+   {
+      TypeNameString<std::remove_reference_t<decltype(mfem::future::get<Is>(std::declval<Tuple&>()))>>()...
+   };
+}
+
+template <typename Tuple>
+static auto ParamTypeStrings()
+{
+   return ParamTypeStringsImpl<Tuple>(
+             std::make_index_sequence<mfem::future::tuple_size<Tuple>::value> {});
+}
+
+static std::string_view Trim(std::string_view s)
+{
+   size_t begin = 0;
+   while (begin < s.size() && std::isspace(static_cast<unsigned char>(s[begin])))
+   {
+      ++begin;
+   }
+   size_t end = s.size();
+   while (end > begin &&
+          std::isspace(static_cast<unsigned char>(s[end - 1])))
+   {
+      --end;
+   }
+   return s.substr(begin, end - begin);
+}
+
+static bool IsValidIdentifier(std::string_view s)
+{
+   if (s.empty()) { return false; }
+   const unsigned char c0 = static_cast<unsigned char>(s[0]);
+   if (!(std::isalpha(c0) || c0 == '_')) { return false; }
+   for (size_t i = 1; i < s.size(); ++i)
+   {
+      const unsigned char c = static_cast<unsigned char>(s[i]);
+      if (!(std::isalnum(c) || c == '_')) { return false; }
+   }
+   return true;
+}
+
+static bool ParseJitDirective(std::string_view line,
+                              std::string &type,
+                              std::string &var,
+                              std::string &kind)
+{
+   const size_t jit_pos = line.find("$JIT");
+   if (jit_pos == std::string_view::npos) { return false; }
+
+   const size_t open = line.find('[', jit_pos);
+   const size_t close = line.find(']', jit_pos);
+   MFEM_VERIFY(open != std::string_view::npos &&
+               close != std::string_view::npos &&
+               close > open,
+               "malformed $JIT directive (expected brackets): " << line);
+
+   const std::string_view payload = line.substr(open + 1, close - open - 1);
+   const size_t comma1 = payload.find(',');
+   const size_t comma2 = (comma1 == std::string_view::npos)
+                         ? std::string_view::npos
+                         : payload.find(',', comma1 + 1);
+   MFEM_VERIFY(comma1 != std::string_view::npos &&
+               comma2 != std::string_view::npos,
+               "malformed $JIT directive (expected 3 comma-separated fields): "
+               << line);
+
+   const std::string_view f0 = Trim(payload.substr(0, comma1));
+   const std::string_view f1 = Trim(payload.substr(comma1 + 1,
+                                                   comma2 - comma1 - 1));
+   const std::string_view f2 = Trim(payload.substr(comma2 + 1));
+   MFEM_VERIFY(!f0.empty() && !f1.empty() && !f2.empty(),
+               "malformed $JIT directive (empty field): " << line);
+
+   type.assign(f0);
+   var.assign(f1);
+   kind.assign(f2);
+   return true;
+}
+
+static std::string ReadFileOrEmpty(const std::string &fn)
+{
+   std::ifstream file(fn);
+   if (!file.is_open())
+   {
+      std::cerr << "could not open file " << fn << "\n";
+      return {};
+   }
+   std::stringstream buffer;
+   buffer << file.rdbuf();
+   return buffer.str();
+}
+
+static std::vector<std::string> ExtractJitVarNames(const std::string
+                                                   &kernel_code)
+{
+   std::stringstream ss(kernel_code);
+   std::string line;
+   std::vector<std::string> var_names;
+   std::unordered_set<std::string> seen_vars;
+
+   while (std::getline(ss, line))
+   {
+      std::string type, var, kind;
+      if (ParseJitDirective(line, type, var, kind))
+      {
+         MFEM_VERIFY(IsValidIdentifier(var),
+                     "$JIT variable must be a valid identifier: " << var);
+         MFEM_VERIFY(seen_vars.insert(var).second,
+                     "duplicate $JIT variable name: " << var);
+         var_names.push_back(var);
+      }
+   }
+   return var_names;
+}
+
+static std::string RewriteKernelForJit(std::string kernel_code,
+                                       const std::vector<std::string> &jit_values)
+{
+   std::stringstream ss(kernel_code);
+   std::string line;
+
+   std::string out;
+   out.reserve(kernel_code.size() + 128);
+
+   bool have_pending = false;
+   size_t pending_index = 0;
+   std::string pending_type;
+   std::string pending_var;
+   std::unordered_set<std::string> seen_vars;
+
+   while (std::getline(ss, line))
+   {
+      line.push_back('\n');
+
+      if (have_pending)
+      {
+         MFEM_VERIFY(pending_index < jit_values.size(),
+                     "not enough JIT values provided");
+         const size_t indent_end = line.find_first_not_of(" \t");
+         const std::string indent =
+            (indent_end == std::string::npos) ? std::string() :
+            line.substr(0, indent_end);
+         out += indent + "const " + pending_type + " " + pending_var + " = " +
+                jit_values[pending_index] + ";\n";
+         have_pending = false;
+         ++pending_index;
+         continue;
+      }
+
+      std::string type, var, kind;
+      if (ParseJitDirective(line, type, var, kind))
+      {
+         MFEM_VERIFY(IsValidIdentifier(var),
+                     "$JIT variable must be a valid identifier: " << var);
+         MFEM_VERIFY(kind == "generic",
+                     "unsupported $JIT kind: " << kind);
+         MFEM_VERIFY(seen_vars.insert(var).second,
+                     "duplicate $JIT variable name: " << var);
+
+         pending_type = std::move(type);
+         pending_var = std::move(var);
+         have_pending = true;
+         continue; // drop directive line
+      }
+
+      out += line;
+   }
+
+   MFEM_VERIFY(!have_pending,
+               "$JIT directive must annotate a following line");
+   MFEM_VERIFY(jit_values.size() == pending_index,
+               "JIT value count must match number of $JIT directives");
+   return out;
+}
+
+static std::string GeneratedOutputPath(std::string_view original_path)
+{
+   const size_t last_sep = original_path.find_last_of("/\\");
+   const size_t dot = original_path.find_last_of('.');
+   const bool dot_in_filename =
+      (dot != std::string_view::npos) &&
+      (last_sep == std::string_view::npos || dot > last_sep);
+
+   const std::string_view base =
+      dot_in_filename ? original_path.substr(0, dot) : original_path;
+   return std::string(base) + "_generated.hpp";
+}
+
+static void WriteFileOrWarn(const std::string &path,
+                            const std::string &contents)
+{
+   std::ofstream out(path);
+   if (!out.is_open())
+   {
+      std::cerr << "could not write generated file " << path << "\n";
+      return;
+   }
+   out << contents;
+}
+
+class JitQFunction
+{
+public:
+   template <typename ImplT, size_t N>
+   JitQFunction(ImplT, const std::string &fn,
+                const std::array<bool, N> &activity_map)
+   {
+      using qf_signature = typename
+                           mfem::future::get_function_signature<
+                           decltype(&ImplT::operator())>::type;
+      using qf_param_ts = typename qf_signature::parameter_ts;
+      constexpr size_t nparams = mfem::future::tuple_size<qf_param_ts>::value;
+      static_assert(N == nparams, "activity_map size must match qfunc arity");
+
+      this->fn = fn;
+      this->nparams = nparams;
+      this->activity_map.reserve(N);
+      for (size_t i = 0; i < N; ++i)
+      {
+         this->activity_map.push_back(activity_map[i]);
+      }
+      {
+         const auto param_types_arr = ParamTypeStrings<qf_param_ts>();
+         this->param_types.assign(param_types_arr.begin(), param_types_arr.end());
+      }
+      this->return_type = TypeNameString<typename qf_signature::return_t>();
+      this->return_is_void = std::is_same_v<typename qf_signature::return_t, void>;
+      this->impl_type_name = TypeNameString<ImplT>();
+      this->jit_var_names = ExtractJitVarNames(ReadFileOrEmpty(fn));
+   }
+
+   template <typename ReturnT, typename... Args>
+   ReturnT run(std::string_view name,
+               std::initializer_list<std::pair<std::string_view, std::string_view>> jit_values,
+               Args&&... args)
+   {
+      auto ordered_values = MatchJitValues(jit_values);
+      auto &mod = GetOrCreateModule(ordered_values);
+      auto &instance = mod.instantiate(std::string(name), std::string());
+      return instance.template run<ReturnT>(std::forward<Args>(args)...);
+   }
+
+   template <typename ReturnT, typename... Args>
+   ReturnT run_primal(
+      std::initializer_list<std::pair<std::string_view, std::string_view>> jit_values,
+      Args&&... args)
+   {
+      return run<ReturnT>(qfunc_name, jit_values,
+                          std::forward<Args>(args)...);
+   }
+
+   template <typename ReturnT, typename... Args>
+   ReturnT run_derivative(
+      std::initializer_list<std::pair<std::string_view, std::string_view>> jit_values,
+      Args&&... args)
+   {
+      return run<ReturnT>(qfunc_name + "_fwddiff", jit_values,
+                          std::forward<Args>(args)...);
+   }
+
+private:
+   std::vector<std::string_view> MatchJitValues(
+      std::initializer_list<std::pair<std::string_view, std::string_view>>
+      named_values) const
+   {
+      std::unordered_map<std::string_view, std::string_view> value_map;
+      for (const auto &[name, value] : named_values)
+      {
+         value_map[name] = value;
+      }
+
+      std::vector<std::string_view> ordered_values;
+      ordered_values.reserve(jit_var_names.size());
+      for (const auto &var_name : jit_var_names)
+      {
+         auto it = value_map.find(var_name);
+         MFEM_VERIFY(it != value_map.end(),
+                     "missing JIT value for variable: " << var_name);
+         ordered_values.push_back(it->second);
+      }
+
+      MFEM_VERIFY(ordered_values.size() == named_values.size(),
+                  "provided " << named_values.size() << " JIT values but expected "
+                  << jit_var_names.size());
+      return ordered_values;
+   }
+
+
+   std::string BuildModuleCode(const std::vector<std::string> &jit_values) const
+   {
+      std::string module_code =
+         RewriteKernelForJit(ReadFileOrEmpty(fn), jit_values);
+      module_code += "\n\n";
+      module_code += "// --- generated ---\n";
+      module_code +=
+         "template <typename return_type, typename... Args>\n"
+         "return_type __enzyme_fwddiff(Args...);\n"
+         "\n"
+         "extern int enzyme_const;\n"
+         "extern int enzyme_dup;\n"
+         "\n";
+
+      // Generate a primal wrapper with the requested symbol name, so the kernel
+      // header can just define the qfunc as a functor.
+      //
+      // Note: Proteus instantiates entrypoints via `qfunc_wrapper<>(...)` even
+      // when there are no user template args, so keep the wrapper itself a
+      // template (with a default parameter) while still doing literal `$JIT`
+      // replacements in the kernel code.
+      module_code += "template <typename = void>\n";
+      module_code += return_type + " " +
+                     std::string(qfunc_name) + "(";
+      bool first = true;
+      for (size_t i = 0; i < nparams; ++i)
+      {
+         if (!first) { module_code += ", "; }
+         first = false;
+         module_code += param_types[i] + " Arg" + std::to_string(i);
+      }
+      module_code += ")\n";
+      module_code += "{\n";
+      module_code += "   " + impl_type_name + " qf;\n";
+      if (return_is_void)
+      {
+         module_code += "   ";
+      }
+      else
+      {
+         module_code += "   return ";
+      }
+      module_code += "qf(";
+      for (size_t i = 0; i < nparams; ++i)
+      {
+         if (i) { module_code += ", "; }
+         module_code += "Arg" + std::to_string(i);
+      }
+      module_code += ");\n";
+      module_code += "}\n\n";
+
+      module_code += "template <typename = void>\n";
+      module_code += return_type + " " +
+                     std::string(qfunc_name) + "_fwddiff(";
+
+      first = true;
+      for (size_t i = 0; i < nparams; ++i)
+      {
+         if (!first) { module_code += ", "; }
+         first = false;
+         module_code += param_types[i] + " Arg" + std::to_string(i);
+         if (activity_map[i])
+         {
+            module_code += ", " + param_types[i] + " dArg" + std::to_string(i);
+         }
+      }
+      module_code += ")\n";
+      module_code += "{\n";
+      if (return_is_void)
+      {
+         module_code += "   __enzyme_fwddiff<void>(\n";
+      }
+      else
+      {
+         module_code += "   return __enzyme_fwddiff<" +
+                        return_type + ">(\n";
+      }
+      module_code += "      (void*)" + std::string(qfunc_name) + "<>";
+      module_code += ",\n";
+      for (size_t i = 0; i < nparams; ++i)
+      {
+         if (activity_map[i])
+         {
+            module_code += "      enzyme_dup, Arg" + std::to_string(i) +
+                           ", dArg" + std::to_string(i);
+         }
+         else
+         {
+            module_code += "      enzyme_const, Arg" + std::to_string(i);
+         }
+         module_code += (i + 1 == nparams) ? ");\n" : ",\n";
+      }
+      module_code += "}\n";
+
+      WriteFileOrWarn(GeneratedOutputPath(fn), module_code);
+      return module_code;
+   }
+
+   proteus::CppJitModule &GetOrCreateModule(
+      const std::vector<std::string_view> &jit_values)
+   {
+      std::string key;
+      for (const auto &val : jit_values)
+      {
+         if (!key.empty()) { key += ","; }
+         key += val;
+      }
+
+      auto it = modules.find(key);
+      if (it != modules.end())
+      {
+         return *it->second;
+      }
+
+      std::vector<std::string> values(jit_values.begin(), jit_values.end());
+      std::string code = BuildModuleCode(values);
+      auto mod = std::make_unique<proteus::CppJitModule>("host", code,
+                                                         DefaultExtraArgs());
+      auto [inserted, ok] = modules.emplace(key, std::move(mod));
+      MFEM_VERIFY(ok, "failed to cache JIT module");
+      return *inserted->second;
+   }
+
+   static std::vector<std::string> DefaultExtraArgs()
+   {
+      return {"-fplugin=/Users/andrej1/local/enzyme/lib/ClangEnzyme-20.dylib"};
+   }
+
+   std::string qfunc_name = "qfunc_wrapper";
+   std::string fn;
+   size_t nparams = 0;
+   std::vector<bool> activity_map;
+   std::vector<std::string> param_types;
+   std::string return_type;
+   bool return_is_void = false;
+   std::string impl_type_name;
+   std::vector<std::string> jit_var_names;
+   std::unordered_map<std::string, std::unique_ptr<proteus::CppJitModule>> modules;
+};
+
+int main()
+{
+   const size_t N = 4;
+   const size_t M = 5;
+   const double A = 123.4;
+
+   std::vector<double> X(N);
+   std::vector<double> Y(N);
+   for (size_t i = 0; i < N; ++i)
+   {
+      X[i] = static_cast<double>(i + 1);
+      Y[i] = static_cast<double>(N - i);
+   }
+
+   // // >>> user interface calls
+   // const std::string kernel_path = std::string(util::thisFileDir) +
+   //                                 "/jitplayground.hpp";
+   // JitQFunction qf(daxpy_op{}, kernel_path, std::array{false, true, false});
+   // // <<< user interface calls
+
+   // // this will happen internally in dFEM
+
+   daxpy_op op;
+   printf("\n\nfunction call\n");
+   op(&A, X.data(), Y.data(), &N);
+
+   // reset X for the derivative test
+   for (size_t i = 0; i < N; ++i)
+   {
+      X[i] = static_cast<double>(i + 1);
+      Y[i] = static_cast<double>(N - i);
+   }
+
+   std::vector<double> dX(N, 1.0);
+   printf("\n\nforward diff call\n");
+   daxpy_op_fwddiff(&A, X.data(), dX.data(), Y.data(), &N);
+
+   std::vector<double> dX_manual(N, A);
+
+   printf("\n\nderivative checks\n");
+   std::cout << "dX: ";
+   for (size_t i = 0; i < N; ++i)
+   {
+      std::cout << dX[i] << (i + 1 == N ? '\n' : ' ');
+   }
+
+   std::cout << "dX_manual: ";
+   for (size_t i = 0; i < N; ++i)
+   {
+      std::cout << dX_manual[i] << (i + 1 == N ? '\n' : ' ');
+   }
+
+   double max_abs_err = 0.0;
+   for (size_t i = 0; i < N; ++i)
+   {
+      max_abs_err = std::max(max_abs_err, std::abs(dX[i] - dX_manual[i]));
+   }
+   std::cout << "max |dX - dX_manual| = " << max_abs_err << "\n";
+
+   return 0;
+}
@@ -0,0 +1,58 @@
+#pragma once
+
+#include <cstddef>
+#include <vector>
+#include <type_traits>
+
+#include "proteus/JitInterface.h"
+
+struct daxpy_op
+{
+   void operator()(
+      const double *a,
+      double *x,
+      const double *y,
+      const size_t *N) const
+   {
+      const size_t n = *N;
+      auto lam = [=, n = proteus::jit_variable(n)]
+                 () __attribute__((annotate("jit")))
+      {
+         printf("N = %zu\n", n);
+         for (size_t i = 0; i < n; ++i)
+         {
+            printf("x[%zu] = %f, y[%zu] = %f\n", i, x[i], i, y[i]);
+            x[i] = *a * x[i] + y[i];
+            printf("updated x[%zu] = %f\n", i, x[i]);
+         }
+      };
+
+      proteus::register_lambda(lam);
+
+      lam();
+   }
+};
+
+template <typename return_type, typename... Args>
+return_type __enzyme_fwddiff(Args...);
+
+extern int enzyme_const;
+extern int enzyme_dup;
+
+void daxpy_op_wrapper(const double * Arg0, double * Arg1,
+                      const double * Arg2, const size_t *Arg3)
+{
+   daxpy_op qf;
+   qf(Arg0, Arg1, Arg2, Arg3);
+}
+
+void daxpy_op_fwddiff(const double * Arg0, double * Arg1,
+                      double * dArg1, const double * Arg2, const size_t *Arg3)
+{
+   __enzyme_fwddiff<void>(
+      (void*)daxpy_op_wrapper,
+      enzyme_const, Arg0,
+      enzyme_dup, Arg1, dArg1,
+      enzyme_const, Arg2,
+      enzyme_const, Arg3);
+}
@@ -71,7 +71,6 @@ endif

 SUBDIRS_ALL = $(addsuffix /all,$(SUBDIRS))
 SUBDIRS_TEST = $(addsuffix /test,$(SUBDIRS))
-SUBDIRS_TEST_NOCLEAN = $(addsuffix /test-noclean,$(SUBDIRS))
 SUBDIRS_CLEAN = $(addsuffix /clean,$(SUBDIRS))
 SUBDIRS_TPRINT = $(addsuffix /test-print,$(SUBDIRS))

@@ -88,9 +87,8 @@ SUBDIRS_TPRINT = $(addsuffix /test-print,$(SUBDIRS))

 all: $(EXAMPLES) $(SUBDIRS_ALL)

-.PHONY: $(SUBDIRS_ALL) $(SUBDIRS_TEST) $(SUBDIRS_TEST_NOCLEAN) \
-   $(SUBDIRS_CLEAN) $(SUBDIRS_TPRINT)
-$(SUBDIRS_ALL) $(SUBDIRS_TEST) $(SUBDIRS_TEST_NOCLEAN) $(SUBDIRS_CLEAN):
+.PHONY: $(SUBDIRS_ALL) $(SUBDIRS_TEST) $(SUBDIRS_CLEAN) $(SUBDIRS_TPRINT)
+$(SUBDIRS_ALL) $(SUBDIRS_TEST) $(SUBDIRS_CLEAN):
 	$(MAKE) -C $(@D) $(@F)
 $(SUBDIRS_TPRINT):
 	@$(MAKE) -C $(@D) $(@F)
@@ -109,7 +107,6 @@ endif
 MFEM_TESTS = EXAMPLES
 include $(MFEM_TEST_MK)
 test: $(SUBDIRS_TEST)
-test-noclean: $(SUBDIRS_TEST_NOCLEAN)
 test-print: $(SUBDIRS_TPRINT)

 # Testing: Parallel vs. serial runs
@@ -121,6 +121,11 @@ set(SRCS
  qinterp/eval_hdiv.cpp
  qinterp/grad_by_nodes.cpp
  qinterp/grad_by_vdim.cpp
+  qinterp/grad_transpose.cpp
+  qinterp/grad_transpose_by_nodes.cpp
+  qinterp/grad_transpose_by_vdim.cpp
+  qinterp/eval_transpose.cpp
+  qinterp/eval_transpose_by_vdim.cpp
  qspace.cpp
  quadinterpolator.cpp
  quadinterpolator_face.cpp
@@ -171,12 +176,8 @@ set(SRCS
  tmop_tools.cpp
  tmop_amr.cpp
  gslib.cpp
-  gslib/findptsedge_local_2.cpp
-  gslib/findptsedge_local_3.cpp
-  gslib/findptssurf_local_3.cpp
  gslib/findpts_local_2.cpp
  gslib/findpts_local_3.cpp
-  gslib/interpolate_local_1.cpp
  gslib/interpolate_local_2.cpp
  gslib/interpolate_local_3.cpp
  transfer.cpp
@@ -195,14 +196,12 @@ set(HDRS
  integ/bilininteg_dgtrace_kernels.hpp
  integ/bilininteg_vecdiffusion_kernels.hpp
  integ/bilininteg_convection_kernels.hpp
-  integ/bilininteg_diffusion_pa_simplices.hpp
  integ/bilininteg_diffusion_kernels.hpp
  integ/bilininteg_elasticity_kernels.hpp
  integ/bilininteg_hcurl_kernels.hpp
  integ/bilininteg_hdiv_kernels.hpp
  integ/bilininteg_hcurlhdiv_kernels.hpp
  integ/bilininteg_mass_kernels.hpp
-  integ/bilininteg_mass_pa_simplices.hpp
  integ/bilininteg_vecdiffusion_pa.hpp
  integ/bilininteg_vecmass_pa.hpp
  coefficient.hpp
@@ -284,8 +283,10 @@ set(HDRS
  qfunction.hpp
  qinterp/det.hpp
  qinterp/eval.hpp
+  qinterp/eval_transpose.hpp
  qinterp/eval_hdiv.hpp
  qinterp/grad.hpp
+  qinterp/grad_transpose.hpp
  qspace.hpp
  quadinterpolator.hpp
  quadinterpolator_face.hpp
@@ -311,7 +312,6 @@ set(HDRS
  tmop_tools.hpp
  tmop_amr.hpp
  gslib.hpp
-  gslib/gslib_kernel_helpers.hpp
  transfer.hpp
  hyperbolic.hpp
  integrator.hpp
@@ -320,36 +320,36 @@ set(HDRS
  )

 if (MFEM_USE_SIDRE)
-  list(APPEND SRCS sidredatacollection.cpp)
-  list(APPEND HDRS sidredatacollection.hpp)
+    list(APPEND SRCS sidredatacollection.cpp)
+    list(APPEND HDRS sidredatacollection.hpp)
 endif()

 if (MFEM_USE_CONDUIT)
-  list(APPEND SRCS conduitdatacollection.cpp)
-  list(APPEND HDRS conduitdatacollection.hpp)
+    list(APPEND SRCS conduitdatacollection.cpp)
+    list(APPEND HDRS conduitdatacollection.hpp)
 endif()

 if (MFEM_USE_ADIOS2)
-  list(APPEND SRCS adios2datacollection.cpp)
-  list(APPEND HDRS adios2datacollection.hpp)
+    list(APPEND SRCS adios2datacollection.cpp)
+    list(APPEND HDRS adios2datacollection.hpp)
 endif()

 if (MFEM_USE_FMS)
-  list(APPEND SRCS fmsdatacollection.cpp fmsconvert.cpp)
-  list(APPEND HDRS fmsdatacollection.hpp fmsconvert.hpp)
+    list(APPEND SRCS fmsdatacollection.cpp fmsconvert.cpp)
+    list(APPEND HDRS fmsdatacollection.hpp fmsconvert.hpp)
 endif()

 if (MFEM_USE_MPI)
-  list(APPEND SRCS
+    list(APPEND SRCS
    pbilinearform.cpp
    pfespace.cpp
    pgridfunc.cpp
    plinearform.cpp
    pnonlinearform.cpp
    prestriction.cpp)
-  # If this list (HDRS -> HEADERS) is used for install, we probably want the
-  # headers added all the time.
-  list(APPEND HDRS
+    # If this list (HDRS -> HEADERS) is used for install, we probably want the
+    # headers added all the time.
+    list(APPEND HDRS
    pbilinearform.hpp
    pfespace.hpp
    pgridfunc.hpp
@@ -1345,8 +1345,7 @@ real_t DiffusionIntegrator::ComputeFluxEnergy
 }

 const IntegrationRule &DiffusionIntegrator::GetRule(
-   const FiniteElement &trial_fe, const FiniteElement &test_fe,
-   const bool stroud)
+   const FiniteElement &trial_fe, const FiniteElement &test_fe)
 {
   int order;
   if (trial_fe.Space() == FunctionSpace::Pk)
@@ -1363,15 +1362,7 @@ const IntegrationRule &DiffusionIntegrator::GetRule(
   {
      return RefinedIntRules.Get(trial_fe.GetGeomType(), order);
   }
-
-   if (stroud)
-   {
-      return StroudIntRules.Get(trial_fe.GetGeomType(), order);
-   }
-   else
-   {
-      return IntRules.Get(trial_fe.GetGeomType(), order);
-   }
+   return IntRules.Get(trial_fe.GetGeomType(), order);
 }

 MassIntegrator::MassIntegrator(const IntegrationRule *ir)
@@ -1458,8 +1449,7 @@ void MassIntegrator::AssembleElementMatrix2(

 const IntegrationRule &MassIntegrator::GetRule(const FiniteElement &trial_fe,
                                               const FiniteElement &test_fe,
-                                               const ElementTransformation &Trans,
-                                               const bool stroud)
+                                               const ElementTransformation &Trans)
 {
   // int order = trial_fe.GetOrder() + test_fe.GetOrder();
   const int order = trial_fe.GetOrder() + test_fe.GetOrder() + Trans.OrderW();
@@ -1468,15 +1458,7 @@ const IntegrationRule &MassIntegrator::GetRule(const FiniteElement &trial_fe,
   {
      return RefinedIntRules.Get(trial_fe.GetGeomType(), order);
   }
-
-   if (stroud)
-   {
-      return StroudIntRules.Get(trial_fe.GetGeomType(), order);
-   }
-   else
-   {
-      return IntRules.Get(trial_fe.GetGeomType(), order);
-   }
+   return IntRules.Get(trial_fe.GetGeomType(), order);
 }


@@ -2178,29 +2178,22 @@ class DiffusionIntegrator: public BilinearFormIntegrator
 {
 public:

-   using ApplyKernelType = void(*)(const int, const bool, const Array<real_t>&,
-                                   const Array<real_t>&, const Array<real_t>&,
-                                   const Array<real_t>&,
-                                   const Vector&, const Vector&,
-                                   Vector&, const int, const int);
+   using DiffusionApplyKernelType = void(*)(const int, const bool,
+                                            const Array<real_t>&,
+                                            const Array<real_t>&, const Array<real_t>&,
+                                            const Array<real_t>&,
+                                            const Vector&, const Vector&,
+                                            Vector&, const int, const int);

-   using ApplySimplexKernelType = void(*)(const int, const bool, const Array<int>&,
-                                          const Array<int>&,
-                                          const Array<int>&, const Array<int>&, const Array<int>&,
-                                          const Array<real_t>&, const Array<real_t>&,
-                                          const Array<real_t>&, const Array<real_t>&,
-                                          const Array<real_t>&, const Array<real_t>&,
-                                          const Vector&, const Vector&,
-                                          Vector&, const int, const int);
+   using DiffusionDiagonalKernelType = void(*)(const int, const bool,
+                                               const Array<real_t>&,
+                                               const Array<real_t>&, const Vector&, Vector&,
+                                               const int, const int);

-   using DiagonalKernelType = void(*)(const int, const bool, const Array<real_t>&,
-                                      const Array<real_t>&, const Vector&, Vector&,
-                                      const int, const int);
-
-   MFEM_REGISTER_KERNELS(ApplyPAKernels, ApplyKernelType, (int, int, int));
-   MFEM_REGISTER_KERNELS(ApplySimplexPAKernels, ApplySimplexKernelType, (int, int,
-                                                                         int));
-   MFEM_REGISTER_KERNELS(DiagonalPAKernels, DiagonalKernelType, (int, int, int));
+   MFEM_REGISTER_KERNELS(DiffusionApplyPAKernel, DiffusionApplyKernelType,
+                         (int, int, int));
+   MFEM_REGISTER_KERNELS(DiffusionDiagonalPAKernel, DiffusionDiagonalKernelType,
+                         (int, int, int));
   struct Kernels { Kernels(); };

 protected:
@@ -2220,6 +2213,7 @@ private:
   const FiniteElementSpace *fespace;
   const DofToQuad *maps;         ///< Not owned
   const GeometricFactors *geom;  ///< Not owned
+public:
   int dim, ne, dofs1D, quad1D;
   Vector pa_data;
   bool symmetric = true; ///< False if using a nonsymmetric matrix coefficient
@@ -2352,8 +2346,7 @@ public:
   void AddMultPatchPA(const int patch, const Vector &x, Vector &y) const;

   static const IntegrationRule &GetRule(const FiniteElement &trial_fe,
-                                         const FiniteElement &test_fe,
-                                         const bool stroud = false);
+                                         const FiniteElement &test_fe);

   bool SupportsCeed() const override { return DeviceCanUseCeed(); }

@@ -2362,15 +2355,8 @@ public:
   template <int DIM, int D1D, int Q1D>
   static void AddSpecialization()
   {
-      ApplyPAKernels::Specialization<DIM,D1D,Q1D>::Add();
-      DiagonalPAKernels::Specialization<DIM,D1D,Q1D>::Add();
-      AddSimplexSpecialization<DIM,D1D,Q1D>();
-   }
-
-   template <int DIM, int D1D, int Q1D>
-   static void AddSimplexSpecialization()
-   {
-      ApplySimplexPAKernels::Specialization<DIM,D1D,Q1D>::Add();
+      DiffusionApplyPAKernel::Specialization<DIM,D1D,Q1D>::Add();
+      DiffusionDiagonalPAKernel::Specialization<DIM,D1D,Q1D>::Add();
   }
 protected:
   const IntegrationRule* GetDefaultIntegrationRule(
@@ -2407,22 +2393,11 @@ public:
                                   const Array<real_t>&, const Vector&,
                                   const Vector&, Vector&, const int, const int);

-   using ApplySimplexKernelType = void(*)(const int, const Array<int>&,
-                                          const Array<int>&,
-                                          const Array<int>&, const Array<int>&, const Array<int>&,
-                                          const Array<real_t>&, const Array<real_t>&,
-                                          const Array<real_t>&, const Array<real_t>&,
-                                          const Array<real_t>&, const Array<real_t>&,
-                                          const Vector&, const Vector&, Vector&,
-                                          const int, const int);
-
   using DiagonalKernelType =  void(*)(const int, const Array<real_t>&,
                                       const Vector&, Vector&, const int,
                                       const int);

   MFEM_REGISTER_KERNELS(ApplyPAKernels, ApplyKernelType, (int, int, int));
-   MFEM_REGISTER_KERNELS(ApplySimplexPAKernels, ApplySimplexKernelType, (int, int,
-                                                                         int));
   MFEM_REGISTER_KERNELS(DiagonalPAKernels, DiagonalKernelType, (int, int, int));
   struct Kernels { Kernels(); };

@@ -2471,8 +2446,7 @@ public:

   static const IntegrationRule &GetRule(const FiniteElement &trial_fe,
                                         const FiniteElement &test_fe,
-                                         const ElementTransformation &Trans,
-                                         const bool stroud = false);
+                                         const ElementTransformation &Trans);

   bool SupportsCeed() const override { return DeviceCanUseCeed(); }

@@ -2483,13 +2457,6 @@ public:
   {
      ApplyPAKernels::Specialization<DIM,D1D,Q1D>::Add();
      DiagonalPAKernels::Specialization<DIM,D1D,Q1D>::Add();
-      AddSimplexSpecialization<DIM,D1D,Q1D>();
-   }
-
-   template <int DIM, int D1D, int Q1D>
-   static void AddSimplexSpecialization()
-   {
-      ApplySimplexPAKernels::Specialization<DIM,D1D,Q1D>::Add();
   }

 protected:
@@ -54,8 +54,6 @@ void Coefficient::Project(QuadratureFunction &qf)
   QuadratureSpaceBase &qspace = *qf.GetSpace();
   const int ne = qspace.GetNE();
   Vector values;
-   // GetValues makes a reference, but we need it to be valid on Host
-   qf.HostWrite();
   for (int iel = 0; iel < ne; ++iel)
   {
      qf.GetValues(iel, values);
@@ -329,8 +327,6 @@ void VectorCoefficient::Project(QuadratureFunction &qf)
   const int ne = qspace.GetNE();
   DenseMatrix values;
   Vector col;
-   // GetValues makes a reference, but we need it to be valid on Host
-   qf.HostWrite();
   for (int iel = 0; iel < ne; ++iel)
   {
      qf.GetValues(iel, values);
@@ -699,8 +695,6 @@ void MatrixCoefficient::Project(QuadratureFunction &qf, bool transpose)
   QuadratureSpaceBase &qspace = *qf.GetSpace();
   const int ne = qspace.GetNE();
   DenseMatrix values, matrix;
-   // GetValues makes a reference, but we need it to be valid on Host
-   qf.HostWrite();
   for (int iel = 0; iel < ne; ++iel)
   {
      qf.GetValues(iel, values);
@@ -237,81 +237,6 @@ ComplexGridFunction::ProjectBdrCoefficientTangent(VectorCoefficient
   gfi->SyncAliasMemory(*this);
 }

-real_t
-ComplexGridFunction::ComputeLpError(const real_t p,
-                                    Coefficient &exsolr,
-                                    Coefficient &exsoli,
-                                    Coefficient *weight,
-                                    const IntegrationRule *irs[],
-                                    const Array<int> *elems) const
-{
-   real_t error = 0.0;
-   const FiniteElement *fe;
-   ElementTransformation *T;
-   Vector valsr;
-   Vector valsi;
-
-   const GridFunction& gf_r = real();
-   const GridFunction& gf_i = imag();
-
-   for (int i = 0; i < fes->GetNE(); i++)
-   {
-      if (elems != NULL && (*elems)[i] == 0) { continue; }
-      fe = fes->GetFE(i);
-      const IntegrationRule *ir;
-      if (irs)
-      {
-         ir = irs[fe->GetGeomType()];
-      }
-      else
-      {
-         int intorder = 2*fe->GetOrder() + 3;
-         ir = &(IntRules.Get(fe->GetGeomType(), intorder));
-      }
-      real_t elem_error = 0.0;
-      gf_r.GetValues(i, *ir, valsr);
-      gf_i.GetValues(i, *ir, valsi);
-      T = fes->GetElementTransformation(i);
-      for (int j = 0; j < ir->GetNPoints(); j++)
-      {
-         const IntegrationPoint &ip = ir->IntPoint(j);
-         T->SetIntPoint(&ip);
-         real_t diffr = valsr(j) - exsolr.Eval(*T, ip);
-         real_t diffi = valsi(j) - exsoli.Eval(*T, ip);
-         real_t diff = hypot(diffr, diffi);
-         if (p < infinity())
-         {
-            diff = pow(diff, p);
-            if (weight)
-            {
-               diff *= weight->Eval(*T, ip);
-            }
-            elem_error += ip.weight * T->Weight() * diff;
-         }
-         else
-         {
-            if (weight)
-            {
-               diff *= weight->Eval(*T, ip);
-            }
-            error = std::max(error, diff);
-         }
-      }
-      if (p < infinity())
-      {
-         // negative quadrature weights may cause the error to be negative
-         error += fabs(elem_error);
-      }
-   }
-
-   if (p < infinity())
-   {
-      error = pow(error, 1./p);
-   }
-
-   return error;
-}
-
 void ComplexGridFunction::Save(std::ostream &os) const
 {
   os << "ComplexGridFunction\n";
@@ -905,9 +830,15 @@ ParComplexGridFunction::ParComplexGridFunction(ParMesh *m, std::istream &input)
      int vsize = pfes->GetVSize();
      Vector::Load(input, 2*vsize);

-      real_t *h_data = HostReadWrite();
-      pfes->ApplyDofSigns(h_data);
-      pfes->ApplyDofSigns(h_data + vsize);
+      real_t *data_  = const_cast<real_t*>(HostRead());
+      for (int i = 0; i < vsize; i++)
+      {
+         if (pfes->GetDofSign(i) < 0)
+         {
+            data_[i] = -data_[i];
+            data_[i+vsize] = -data_[i+vsize];
+         }
+      }


      // if the mesh is a legacy (v1.1) NC mesh, it has old vertex ordering
@@ -1120,14 +1051,15 @@ void ParComplexGridFunction::Save(std::ostream &os) const
   os << '\n';

   int vsize = pfes->GetVSize();
-   // We use const_cast + HostRead (instead of HostReadWrite) because we only
-   // need to change the host data temporarily and this way we do not invalidate
-   // the data if it is on device. If we use HostReadWrite here, later calls to
-   // Read or ReadWrite will need to copy the data from host to device. With the
-   // approach used here, the host-to-device copy is avoided.
-   real_t *h_data = const_cast<real_t*>(HostRead());
-   pfes->ApplyDofSigns(h_data);
-   pfes->ApplyDofSigns(h_data + vsize);
+   real_t *data_  = const_cast<real_t*>(HostRead());
+   for (int i = 0; i < vsize; i++)
+   {
+      if (pfes->GetDofSign(i) < 0)
+      {
+         data_[i] = -data_[i];
+         data_[i+vsize] = -data_[i+vsize];
+      }
+   }

   if (pfes->GetOrdering() == Ordering::byNODES)
   {
@@ -1138,8 +1070,14 @@ void ParComplexGridFunction::Save(std::ostream &os) const
      Vector::Print(os, pfes->GetVDim());
   }

-   pfes->ApplyDofSigns(h_data);
-   pfes->ApplyDofSigns(h_data + vsize);
+   for (int i = 0; i < vsize; i++)
+   {
+      if (pfes->GetDofSign(i) < 0)
+      {
+         data_[i] = -data_[i];
+         data_[i+vsize] = -data_[i+vsize];
+      }
+   }

   os.flush();
 }
@@ -166,75 +166,6 @@ public:
      return sqrt(err_r * err_r + err_i * err_i);
   }

-   /// @brief Returns Max|u_ex - u_h| error for complex-valued H1 or L2 elements
-   ///
-   /// Compute the $L_\infty$ error across the entire domain.
-   ///
-   /// @param[in] exsolr  Coefficient object reproducing the real part of the
-   ///                    anticipated values of the scalar field, Re(u_ex).
-   /// @param[in] exsoli  Coefficient object reproducing the imaginary part of
-   ///                    the anticipated values of the scalar field, Im(u_ex).
-   /// @param[in] irs        Optional pointer to an array of custom integration
-   ///                       rules e.g. higher order than the default rules. If
-   ///                       present the array will be indexed by
-   ///                       Geometry::Type.
-   ///
-   /// @note Uses ComputeLpError internally. See the ComputeLpError
-   ///       documentation for generalizations of this error computation.
-   ///
-   /// @note If an array of integration rules is provided through @a irs, be
-   ///       sure to include valid rules for each element type that may occur
-   ///       in the list of elements.
-   ///
-   virtual real_t ComputeMaxError(Coefficient &exsolr,
-                                  Coefficient &exsoli,
-                                  const IntegrationRule *irs[] = NULL) const
-   {
-      return ComputeLpError(infinity(), exsolr, exsoli, NULL, irs);
-   }
-
-   /// @brief Returns ||u_ex - u_h||_Lp for complex-valued H1 or L2 elements
-   ///
-   /// Computes:
-   ///    $$(\sum_{elems} \int_{elem} w \, |u_{ex} - u_h|^p)^{1/p}$$
-   /// Where:
-   ///    $$|u_{ex} - u_h| = \sqrt{Re(u_{ex} - u_h)^2 + Im(u_{ex} - u_h)^2}$$
-   ///
-   /// @param[in] p       Real value indicating the exponent of the $L^p$ norm.
-   ///                    To avoid domain errors p should have a positive value,
-   ///                    either finite or infinite.
-   /// @param[in] exsolr  Coefficient object reproducing the real part of the
-   ///                    anticipated values of the scalar field, Re(u_ex).
-   /// @param[in] exsoli  Coefficient object reproducing the imaginary part of
-   ///                    the anticipated values of the scalar field, Im(u_ex).
-   /// @param[in] weight  Optional pointer to a Coefficient object reproducing
-   ///                    a weighting function, w.
-   /// @param[in] irs     Optional pointer to an array of custom integration
-   ///                    rules e.g. higher order than the default rules. If
-   ///                    present the array will be indexed by Geometry::Type.
-   /// @param[in] elems   Optional pointer to a marker array, with a length
-   ///                    equal to the number of local elements, indicating
-   ///                    which elements to integrate over. Only those elements
-   ///                    corresponding to non-zero entries in @a elems will
-   ///                    contribute to the computed L2 error.
-   ///
-   /// @note If an array of integration rules is provided through @a irs, be
-   ///       sure to include valid rules for each element type that may occur
-   ///       in the list of elements.
-   ///
-   /// @note Quadratures with negative weights (as in some simplex integration
-   ///       rules in MFEM) can produce negative integrals even with
-   ///       non-negative integrands. To avoid returning negative errors this
-   ///       function uses the absolute values of the element-wise integrals.
-   ///       This may lead to results which are not entirely consistent with
-   ///       such integration rules.
-   virtual real_t ComputeLpError(const real_t p,
-                                 Coefficient &exsolr,
-                                 Coefficient &exsoli,
-                                 Coefficient *weight = NULL,
-                                 const IntegrationRule *irs[] = NULL,
-                                 const Array<int> *elems = NULL) const;
-
   /// Save the ComplexGridFunction to an output stream.
   virtual void Save(std::ostream &out) const;

@@ -114,10 +114,6 @@ void ConduitDataCollection::Save()
                                         n_mesh["fields"][name]);
   }

-   // TODO: in parallel, we need to call ParFiniteElementSpace::ApplyDofSigns
-   // for all ParGridFunction objects before and after saving, see
-   // ParGridFunction::Save.
-
   // save mesh data
   SaveMeshAndFields(myid,
                     n_mesh,
@@ -1181,14 +1181,12 @@ void ParaViewDataCollection::SaveGFieldVTU(std::ostream &os, int ref_,
   DenseMatrix vval, pmat;
   std::vector<char> buf;
   int vec_dim = it->second->VectorDim();
-   int map_type = it->second->FESpace()->GetTypicalFE()->GetMapType();
   os << "<DataArray type=\"" << GetDataTypeString()
      << "\" Name=\"" << it->first
      << "\" NumberOfComponents=\"" << vec_dim << "\" "
      << VTKComponentLabels(vec_dim) << " "
      << "format=\"" << GetDataFormatString() << "\" >" << '\n';
-   if (vec_dim == 1 && (map_type == FiniteElement::VALUE ||
-                        map_type == FiniteElement::INTEGRAL))
+   if (vec_dim == 1)
   {
      for (int i = 0; i < mesh->GetNE(); i++)
      {
@@ -25,35 +25,21 @@
 namespace mfem
 {

-/// Lightweight adaptor over an std::map from type K to type to V
-template<typename K, typename V,
-         typename = typename std::enable_if<std::is_default_constructible<V>::value>::type>
-class GenericFieldMap
+/// Lightweight adaptor over an std::map from strings to pointer to T
+template<typename T>
+class NamedFieldsMap
 {
-private:
-   static constexpr bool ValueIsPointer = std::is_pointer<V>::value;
-
 public:
-   typedef std::map<K, V> MapType;
+   typedef std::map<std::string, T*> MapType;
   typedef typename MapType::iterator iterator;
   typedef typename MapType::const_iterator const_iterator;

-   /// Register field @a field with name @a key
-   /// Only enabled if the template parameter V is not a pointer
-   template<typename = std::enable_if<!ValueIsPointer, bool>>
-   void Register(const K& key, V field)
+   /// Register field @a field with name @a fname
+   /** Replace existing field associated with @a fname (and optionally
+       delete associated pointer if @a own_data is true) */
+   void Register(const std::string& fname, T* field, bool own_data)
   {
-      field_map[key] = field;
-   }
-
-   /// Register field @a field with name @a key
-   /** Replace existing field associated with @a key (and optionally
-       delete associated pointer if @a own_data is true).
-       Only enabled if the template parameter V is a pointer*/
-   template<typename = std::enable_if<ValueIsPointer, bool>>
-   void Register(const K& key, V field, bool own_data)
-   {
-      V& ref = field_map[key];
+      T*& ref = field_map[fname];
      if (own_data)
      {
         delete ref; // if newly allocated -> ref is null -> OK
@@ -61,40 +47,23 @@ public:
      ref = field;
   }

-   /// Unregister association between field @a field and name @a key
-   /// Only enabled if the template parameter V is not a pointer
-   template<typename = std::enable_if<!ValueIsPointer, bool>>
-   void Deregister(const K& key)
+   /// Unregister association between field @a field and name @a fname
+   /** Optionally delete associated pointer if @a own_data is true */
+   void Deregister(const std::string& fname, bool own_data)
   {
-      iterator it = field_map.find(key);
-      if ( it != field_map.end() )
-      {
-         field_map.erase(it);
-      }
-   }
-
-   /// Unregister association between field @a field and name @a key
-   /** Optionally delete associated pointer if @a own_data is true.
-       Only enabled if the template parameter V is a pointer */
-   template<typename = std::enable_if<ValueIsPointer, bool>>
-   void Deregister(const K& key, bool own_data)
-   {
-      iterator it = field_map.find(key);
+      iterator it = field_map.find(fname);
      if ( it != field_map.end() )
      {
         if (own_data)
         {
            delete it->second;
-            it->second = nullptr;
         }
         field_map.erase(it);
      }
   }

   /// Clear all associations between names and fields
-   /** Delete associated pointers when @a own_data is true.
-       Only enabled if the template parameter V is a pointer */
-   template<typename = std::enable_if<ValueIsPointer, bool>>
+   /** Delete associated pointers when @a own_data is true */
   void DeleteData(bool own_data)
   {
      for (iterator it = field_map.begin(); it != field_map.end(); ++it)
@@ -107,37 +76,22 @@ public:
      }
   }

-   /// Predicate to check if a field is associated with name @a key
-   bool Has(const K& key) const
+   /// Predicate to check if a field is associated with name @a fname
+   bool Has(const std::string& fname) const
   {
-      return field_map.find(key) != field_map.end();
+      return field_map.find(fname) != field_map.end();
   }

-   /// Get a pointer to the field associated with name @a key
-   /** @return Field associated with @a key or NULL,
-       if value is pointer and key not found */
-   V Get(const K& key) const
+   /// Get a pointer to the field associated with name @a fname
+   /** @return Pointer to field associated with @a fname or NULL */
+   T* Get(const std::string& fname) const
   {
-      const_iterator it = field_map.find(key);
-      if (it != field_map.end())
-      {
-         return it->second;
-      }
-      else
-      {
-         if constexpr (ValueIsPointer)
-         {
-            return nullptr;
-         }
-         else
-         {
-            return V(); // Return default-constructed value for non-pointer types
-         }
-      }
+      const_iterator it = field_map.find(fname);
+      return it != field_map.end() ? it->second : NULL;
   }

   /// Returns a const reference to the underlying map
-   const MapType &GetMap() const { return field_map; }
+   const MapType& GetMap() const { return field_map; }

   /// Returns the number of registered fields
   int NumFields() const { return field_map.size(); }
@@ -152,24 +106,21 @@ public:
   /// Returns an end const iterator to the registered fields
   const_iterator end() const { return field_map.end(); }

-   /// Returns an iterator to the field @a key
-   iterator find(const K& key)
-   { return field_map.find(key); }
+   /// Returns an iterator to the field @a fname
+   iterator find(const std::string& fname)
+   { return field_map.find(fname); }

-   /// Returns a const iterator to the field @a key
-   const_iterator find(const K& key) const
-   { return field_map.find(key); }
+   /// Returns a const iterator to the field @a fname
+   const_iterator find(const std::string& fname) const
+   { return field_map.find(fname); }

-   /// Clears the map of registered fields
+   /// Clears the map of registered fields without reclaiming memory
   void clear() { field_map.clear(); }

 protected:
   MapType field_map;
 };

-/// Lightweight adaptor over an std::map from strings to pointer to T
-template<typename T>
-using NamedFieldsMap = GenericFieldMap<std::string, T*>;

 /** A class for collecting finite element data that is part of the same
    simulation. Currently, this class groups together grid functions (fields),
@@ -0,0 +1,587 @@
+// Copyright (c) 2010-2025, 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.
+#pragma once
+
+#include <cassert>
+#include <cstddef>
+
+// #include "fem/kernels.hpp"
+#include "fem/kernels3d.hpp"
+namespace ker = mfem::kernels::internal;
+namespace low = mfem::kernels::internal::low;
+#include "fem/kernel_dispatch.hpp"
+
+// #include "linalg/kernels.hpp"
+
+#include "util.hpp"
+
+#undef NVTX_COLOR
+#define NVTX_COLOR ::nvtx::kOrchid
+
+namespace mfem::future
+{
+
+/** @brief Zero-copy view of a contiguous block as a `tensor<T, n1>` */
+template<typename T, int n1>
+MFEM_HOST_DEVICE
+const tensor<T, n1>& as_tensor(const T* ptr)
+{
+   // std::launder makes this defined behavior under strict aliasing rules
+   return *std::launder(reinterpret_cast<const tensor<T, n1>*>(ptr));
+}
+
+// convenience overload if you prefer a mutable view
+template<typename T, int n1>
+MFEM_HOST_DEVICE
+tensor<T, n1>& as_tensor(T* ptr)
+{
+   return *std::launder(reinterpret_cast<tensor<T, n1>*>(ptr));
+}
+
+/** @brief Zero-copy view of a contiguous block as a `tensor<T, n1, n2>` */
+template<typename T, int n1, int n2>
+MFEM_HOST_DEVICE
+const tensor<T, n1, n2>& as_tensor(const T* ptr)
+{
+   // std::launder makes this defined behavior under strict aliasing rules
+   return *std::launder(reinterpret_cast<const tensor<T, n1, n2>*>(ptr));
+}
+
+// convenience overload if you prefer a mutable view
+template<typename T, int n1, int n2>
+MFEM_HOST_DEVICE
+tensor<T, n1, n2>& as_tensor(T* ptr)
+{
+   return *std::launder(reinterpret_cast<tensor<T, n1, n2>*>(ptr));
+}
+
+/** @brief Zero-copy view of a contiguous block as a `tensor<T, n1, n2, n3>` */
+template<typename T, int n1, int n2, int n3>
+MFEM_HOST_DEVICE
+const tensor<T, n1, n2, n3>& as_tensor(const T* ptr)
+{
+   // std::launder makes this defined behavior under strict aliasing rules
+   return *std::launder(reinterpret_cast<const tensor<T, n1, n2, n3>*>(ptr));
+}
+
+// convenience overload if you prefer a mutable view
+template<typename T, int n1, int n2, int n3>
+MFEM_HOST_DEVICE
+tensor<T, n1, n2, n3>& as_tensor(T* ptr)
+{
+   return *std::launder(reinterpret_cast<tensor<T, n1, n2, n3>*>(ptr));
+}
+
+/** @brief Zero-copy view of a contiguous block as a `tensor<T, n1, n2, n3, n4>` */
+template<typename T, int n1, int n2, int n3, int n4>
+MFEM_HOST_DEVICE
+const tensor<T, n1, n2, n3, n4>& as_tensor(const T* ptr)
+{
+   // std::launder makes this defined behavior under strict aliasing rules
+   return *std::launder(reinterpret_cast<const tensor<T, n1, n2, n3, n4>*>(ptr));
+}
+
+// convenience overload if you prefer a mutable view
+template<typename T, int n1, int n2, int n3, int n4>
+MFEM_HOST_DEVICE
+tensor<T, n1, n2, n3, n4>& as_tensor(T* ptr)
+{
+   return *std::launder(reinterpret_cast<tensor<T, n1, n2, n3, n4>*>(ptr));
+}
+
+
+template <std::size_t N>
+MFEM_HOST_DEVICE inline
+std::array<real_t*, N>
+load_field_e_ptr(const std::array<DeviceTensor<2>, N> &fields_e,
+                 const int e)
+{
+   std::array<real_t*, N> f;
+   for_constexpr<N>([&](auto i) { f[i] = &fields_e[i](0, e); });
+   return f;
+}
+
+namespace qf
+{
+
+template <int T_Q1D,
+          size_t num_args,
+          typename reg_t,
+          typename qfunc_t,
+          typename args_ts>
+MFEM_HOST_DEVICE inline
+void apply_kernel(reg_t &res /*output*/,
+                  reg_t &reg,
+                  const real_t *rd,
+                  const int qx, const int qy, const int qz,
+                  const qfunc_t &qfunc, args_ts &args)
+{
+   if constexpr (num_args == 2) // PAApply
+   {
+      // ∇u
+      tensor<real_t, 3> &arg_0 = get<0>(args);
+      arg_0[0] = reg[qz][qy][qx][0];
+      arg_0[1] = reg[qz][qy][qx][1];
+      arg_0[2] = reg[qz][qy][qx][2];
+
+      // D (PA data)
+      tensor<real_t, 3, 3> &arg_1 = get<1>(args);
+
+      if constexpr (T_Q1D > 0)
+      {
+         const auto *D = (const real_t (*)[T_Q1D][T_Q1D][3][3]) rd;
+         for (int k = 0; k < 3; k++)
+         {
+            for (int j = 0; j < 3; j++)
+            {
+               arg_1[k][j] = D[qx][qy][qz][k][j];
+            }
+         }
+      }
+      else
+      {
+         static_assert(false);
+         // const auto D = Reshape(r2, 3, 3, Q1D, Q1D, Q1D);
+         // for (int j = 0; j < 3; j++)
+         // {
+         //    for (int k = 0; k < 3; k++)
+         //    {
+         //       arg_1[k][j] = D(j, k, qz, qy, qx);
+         //    }
+         // }
+      }
+   }
+   else
+   {
+      // MFApply comes here
+      assert(false);
+      // MFEM_ABORT("Only two arguments (∇u and D) are supported in apply_kernel for now");
+   }
+
+   const auto r = get<0>(apply(qfunc, args));
+
+   if constexpr (decltype(r)::ndim == 1)
+   {
+      // process_qf_result_from_reg(r0, qx, qy, qz, r);
+      as_tensor<real_t, 3>(&res[qz][qy][qx][0]) = r;
+   }
+   else
+   {
+      static_assert(false);
+   }
+}
+
+} // namespace qf
+
+#define MFEM_D2Q_MAX_SIZE 4
+static MFEM_CONSTANT real_t Bi[MFEM_D2Q_MAX_SIZE][8*8], Bo[8*8];
+static MFEM_CONSTANT real_t Gi[MFEM_D2Q_MAX_SIZE][8*8], Go[8*8];
+
+template<size_t num_fields,
+         size_t num_inputs,
+         size_t num_outputs,
+         typename restriction_cb_t,
+         typename qfunc_t,
+         typename input_t,
+         typename output_fop_t>
+class NewActionCallback
+{
+   restriction_cb_t &restriction_cb;
+   qfunc_t &qfunc;
+   input_t &inputs;
+   const std::array<size_t, num_inputs> &input_to_field;
+   const std::array<DofToQuadMap, num_inputs> &input_dtq_maps;
+   const std::array<DofToQuadMap, num_outputs> &output_dtq_maps;
+   const int num_entities;
+   const int test_vdim;
+   const int num_test_dof;
+   const int dimension;
+   const ThreadBlocks &thread_blocks;
+   SharedMemoryInfo<num_fields, num_inputs, num_outputs> &shmem_info;
+   const Array<int> &attributes;
+   const output_fop_t &output_fop;
+   const Array<int> *elem_attributes;
+   // refs
+   std::vector<Vector> &fields_e;
+   Vector &residual_e;
+   std::function<void(Vector &, Vector &)> &output_restriction_transpose;
+   // args
+   std::vector<Vector> &solutions_l;
+   const std::vector<Vector> &parameters_l;
+   Vector &residual_l;
+
+public:
+   NewActionCallback() = delete;
+
+   NewActionCallback(const bool use_kernels_specialization,
+                     restriction_cb_t &restriction_cb,
+                     qfunc_t &qfunc,
+                     input_t &inputs,
+                     const std::array<size_t, num_inputs> &input_to_field,
+                     const std::array<DofToQuadMap, num_inputs> &input_dtq_maps,
+                     const std::array<DofToQuadMap, num_outputs> &output_dtq_maps,
+                     const int num_entities,
+                     const int test_vdim,
+                     const int num_test_dof,
+                     const int dimension,
+                     const ThreadBlocks &thread_blocks,
+                     SharedMemoryInfo<num_fields, num_inputs, num_outputs> &shmem_info,
+                     const Array<int> &attributes,
+                     const output_fop_t &output_fop,
+                     const Array<int> *elem_attributes,
+                     // refs
+                     std::vector<Vector> &fields_e,
+                     Vector &residual_e,
+                     std::function<void(Vector &, Vector &)> &output_restriction_transpose,
+                     // args
+                     std::vector<Vector> &solutions_l,
+                     const std::vector<Vector> &parameters_l,
+                     Vector &residual_l):
+      restriction_cb(restriction_cb),
+      qfunc(qfunc),
+      inputs(inputs),
+      input_to_field(input_to_field),
+      input_dtq_maps(input_dtq_maps),
+      output_dtq_maps(output_dtq_maps),
+      num_entities(num_entities),
+      test_vdim(test_vdim),
+      num_test_dof(num_test_dof),
+      dimension(dimension),
+      thread_blocks(thread_blocks),
+      shmem_info(shmem_info),
+      attributes(attributes),
+      output_fop(output_fop),
+      elem_attributes(elem_attributes),
+      fields_e(fields_e),
+      residual_e(residual_e),
+      output_restriction_transpose(output_restriction_transpose),
+      solutions_l(solutions_l),
+      parameters_l(parameters_l),
+      residual_l(residual_l)
+   {
+      if (!use_kernels_specialization) { return; }
+      NewActionCallbackKernels::template Specialization<3>::Add(); // 1
+      NewActionCallbackKernels::template Specialization<4>::Add(); // 2
+      NewActionCallbackKernels::template Specialization<5>::Add(); // 3
+      NewActionCallbackKernels::template Specialization<6>::Add(); // 4
+      NewActionCallbackKernels::template Specialization<7>::Add(); // 5
+      NewActionCallbackKernels::template Specialization<8>::Add(); // 6
+   }
+
+   template<int T_Q1D = 0>
+   static void action_callback_new(const int d1d,
+                                   restriction_cb_t &restriction_cb,
+                                   qfunc_t &qfunc,
+                                   [[maybe_unused]] input_t &inputs,
+                                   [[maybe_unused]] const std::array<size_t, num_inputs> &input_to_field,
+                                   const std::array<DofToQuadMap, num_inputs> &input_dtq_maps,
+                                   const std::array<DofToQuadMap, num_outputs> &output_dtq_maps,
+                                   [[maybe_unused]] const int dimension,
+                                   const int num_entities,
+                                   [[maybe_unused]] const int test_vdim,
+                                   [[maybe_unused]] const int num_test_dof,
+                                   const ThreadBlocks &thread_blocks,
+                                   [[maybe_unused]] SharedMemoryInfo<num_fields, num_inputs, num_outputs>
+                                   &shmem_info,
+                                   [[maybe_unused]] const Array<int> &attributes,
+                                   [[maybe_unused]] const output_fop_t &output_fop,
+                                   [[maybe_unused]] const Array<int> *elem_attributes,
+                                   // refs
+                                   std::vector<Vector> &fields_e,
+                                   Vector &residual_e,
+                                   std::function<void(Vector &, Vector &)> &output_restriction_transpose,
+                                   // args
+                                   std::vector<Vector> &solutions_l,
+                                   const std::vector<Vector> &parameters_l,
+                                   Vector &residual_l,
+                                   // fallback arguments
+                                   const int q1d)
+   {
+      NVTX_MARK_FUNCTION;
+      assert(dimension == 3);
+      static_assert(MFEM_D2Q_MAX_SIZE >= num_inputs, "MFEM_D2Q_MAX_SIZE error");
+
+      constexpr int DIM = 3;
+
+      [[maybe_unused]] static bool ini = (for_constexpr<num_inputs>([&](auto i)
+      {
+         const auto dtq = input_dtq_maps[i];
+         {
+            const auto [q, _, p] = dtq.B.GetShape();
+            const auto B = (const real_t*)input_dtq_maps[i].B;
+            dbg("Loading Bi[{}]: q={} p={}", i.value, q, p);
+            if (B) { Gpu(MemcpyToSymbol)(Bi[i], B, (p*q)*sizeof(real_t)); }
+         }
+         {
+            const auto [q, _, p] = dtq.G.GetShape();
+            const auto G = (const real_t*)input_dtq_maps[i].G;
+            if (G) { Gpu(MemcpyToSymbol)(Gi[i], G, (p*q)*sizeof(real_t)); }
+         }
+         if constexpr (i == 0) // output B
+         {
+            const auto dtq_o = output_dtq_maps[0];
+            const auto [q, _, p] = dtq_o.B.GetShape();
+            const auto B = (const real_t*)dtq_o.B;
+            if (B) { Gpu(MemcpyToSymbol)(Bo, B, (p*q)*sizeof(real_t)); }
+         }
+         if constexpr (i == 0) // output G
+         {
+            const auto dtq_o = output_dtq_maps[0];
+            const auto [q, _, p] = dtq_o.G.GetShape();
+            const auto G = (const real_t*)dtq_o.G;
+            if (G) { Gpu(MemcpyToSymbol)(Go, G, (p*q)*sizeof(real_t)); }
+            dbg("Loaded B and G to constant memory");
+         }
+      }), true);
+
+      // types
+      using qf_signature =
+         typename create_function_signature<decltype(&qfunc_t::operator())>::type;
+      using qf_param_ts = typename qf_signature::parameter_ts;
+
+      restriction_cb(solutions_l, parameters_l, fields_e);
+
+      NVTX_INI("res=0");
+      residual_e = 0.0;
+      NVTX_END("res=0");
+
+      // auto wrapped_fields_e =
+      // wrap_fields(fields_e, shmem_info.field_sizes, num_entities);
+
+      const bool has_attr = attributes.Size() > 0;
+      const auto d_attr = attributes.Read();
+      const auto d_elem_attr = elem_attributes->Read();
+
+      // const int vdim = input.vdim;
+      // const auto fields_e_ptr = load_field_e_ptr(wrapped_fields_e, e);
+      // const real_t *field_e_r = fields_e_ptr[input_to_field[i]];
+      // const auto fields_e_ptr = load_field_e_ptr(wrapped_fields_e, e);
+      const int NE = num_entities;
+      constexpr int VDIM = 1;
+
+      const auto XE = Reshape(fields_e[0].Read(), d1d, d1d, d1d, VDIM, NE);
+      const real_t *dx_ptr = fields_e[1].Read();
+
+      auto YE = Reshape(residual_e.ReadWrite(), d1d, d1d, d1d, VDIM, NE);
+
+      const auto B = (const real_t*)input_dtq_maps[0/*i*/].B;
+      const auto G = (const real_t*)input_dtq_maps[0/*i*/].G;
+
+      NVTX_INI("forall");
+      dfem::forall<T_Q1D*T_Q1D*T_Q1D>([=] MFEM_HOST_DEVICE (int e, void *)
+      {
+         if (has_attr && !d_attr[d_elem_attr[e] - 1]) { return; }
+
+         constexpr int MQ1 = T_Q1D > 0 ? T_Q1D : 8;
+
+         MFEM_SHARED real_t sm0[MQ1][MQ1][MQ1][3];
+         MFEM_SHARED real_t sm1[MQ1][MQ1][MQ1][3];
+         // real_t (&sm0_ptr)[MQ1][MQ1][MQ1][3] = sm0;
+         // real_t (&sm1_ptr)[MQ1][MQ1][MQ1][3] = sm1;
+
+         low::regs3d_t<DIM, MQ1> reg;
+         const real_t *rd = dx_ptr;
+
+         // const auto fields_e_ptr = load_field_e_ptr(wrapped_fields_e, e);
+
+         MFEM_SHARED real_t sB[MQ1][MQ1], sG[MQ1][MQ1];
+         // real_t (&sB_ptr)[MD1][MQ1] = sB;
+         // real_t (&sG_ptr)[MD1][MQ1] = sG;
+
+         // Interpolate
+         // for_constexpr<num_inputs>(
+         //    [ D1D, Q1D, MQ1, e,
+         //           &input_dtq_maps,
+         //           &sm0_ptr, &sm1_ptr,
+         //           &sB = sB_ptr, &sG = sG_ptr,
+         //           &inputs,
+         //           //  &fields_e_ptr,
+         //           &reg, &rd,
+         //           &input_to_field ] (auto i)
+         {
+            // const auto input = get<0/*i*/>(inputs);
+            // using field_operator_t = std::decay_t<decltype(input)>;
+
+            // if constexpr (is_gradient_fop<field_operator_t>::value) // Grad
+            {
+               // const int vdim = input.vdim;
+               // const real_t *field_e_r = fields_e_ptr[input_to_field[i]];
+               // const auto XE = Reshape(field_e_r, D1D, D1D, D1D, vdim);
+               // const auto sB = reinterpret_cast<const real_t (*)[MQ1]>(Bi[i]);
+               // const auto sG = reinterpret_cast<const real_t (*)[MQ1]>(Gi[i]);
+               low::LoadMatrix(d1d, q1d, B, sB);
+               low::LoadMatrix(d1d, q1d, G, sG);
+               // for (int c = 0; c < vdim; c++)
+               // constexpr int c = 0;
+               {
+                  low::LoadDofs3d(e, d1d, XE, sm0);
+                  low::Grad3d(d1d, q1d, sB, sG, sm0, sm1, reg);
+               }
+            }
+            // else if constexpr (is_identity_fop<field_operator_t>::value)   // Identity
+            {
+               // db1("Identity");
+               // rd = fields_e_ptr[input_to_field[i]];
+               // rd = dx_ptr;
+            }
+            // else if constexpr (is_weight_fop<field_operator_t>::value)   // Weight
+            // {
+            //    dbg("Weight");
+            //    rw = fields_e_ptr[input_to_field[i]]; // 🔥
+            // }
+            // else
+            {
+               // MFApply comes here
+               // assert(false);
+               // MFEM_ABORT("Only Grad and Identity field operators are supported");
+            }
+         }//); // for_constexpr<num_inputs>
+
+         MFEM_FOREACH_THREAD_DIRECT(qz,z,q1d)
+         {
+            MFEM_FOREACH_THREAD_DIRECT(qy,y,q1d)
+            {
+               MFEM_FOREACH_THREAD_DIRECT(qx,x,q1d)
+               {
+#if 0
+                  auto qf_args = decay_tuple<qf_param_ts> {};
+                  qf::apply_kernel<T_Q1D, num_inputs>
+                  (reg, reg, rd, qx, qy, qz, qfunc, qf_args);
+#elif 0
+                  real_t v[3], u[3] = { reg[qz][qy][qx][0],
+                                        reg[qz][qy][qx][1],
+                                        reg[qz][qy][qx][2]
+                                      };
+                  const auto *D = (real_t (*)[T_Q1D][T_Q1D][3][3]) rd;
+                  kernels::Mult(3, 3, &D[qx][qy][qz][0][0], u, v);
+                  reg[qz][qy][qx][0] = v[0];
+                  reg[qz][qy][qx][1] = v[1];
+                  reg[qz][qy][qx][2] = v[2];
+#elif 0
+                  const auto *D = (real_t (*)[T_Q1D][T_Q1D][3][3]) rd;
+                  const auto args = decay_tuple<qf_param_ts>
+                  {
+                     {{ reg[qz][qy][qx][0], reg[qz][qy][qx][1], reg[qz][qy][qx][2] }},
+                     {{
+                           {{ D[qx][qy][qz][0][0], D[qx][qy][qz][0][1], D[qx][qy][qz][0][2] }},
+                           {{ D[qx][qy][qz][1][0], D[qx][qy][qz][1][1], D[qx][qy][qz][1][2] }},
+                           {{ D[qx][qy][qz][2][0], D[qx][qy][qz][2][1], D[qx][qy][qz][2][2] }}
+                        }
+                     }
+                  };
+                  const auto r = get<0>(apply(qfunc, args));
+                  reg[qz][qy][qx][0] = r[0];
+                  reg[qz][qy][qx][1] = r[1];
+                  reg[qz][qy][qx][2] = r[2];
+#elif 0
+                  auto u = as_tensor<real_t, 3>(&reg[qz][qy][qx][0]);
+                  const auto *d = (real_t (*)[T_Q1D][T_Q1D][3][3]) rd;
+                  auto D = as_tensor<real_t, 3, 3>(&d[qx][qy][qz][0][0]);
+                  auto r = D * u;
+                  reg[qz][qy][qx][0] = r[0];
+                  reg[qz][qy][qx][1] = r[1];
+                  reg[qz][qy][qx][2] = r[2];
+#else
+                  auto args = decay_tuple<qf_param_ts> {};
+                  get<0>(args) = as_tensor<real_t, 3>(&reg[qz][qy][qx][0]);
+                  if constexpr (T_Q1D > 0)
+                  {
+                     get<1>(args) = as_tensor<real_t, 3, 3>(rd + 9*(qx*T_Q1D*T_Q1D + qy*T_Q1D + qz));
+                  }
+                  else
+                  {
+                     get<1>(args) = as_tensor<real_t, 3, 3>(rd + 9*(qx*q1d*q1d + qy*q1d + qz));
+                  }
+                  auto r = get<0>(apply(qfunc, args));
+                  if constexpr (decltype(r)::ndim == 1)
+                  {
+                     as_tensor<real_t, 3>(&reg[qz][qy][qx][0]) = r;
+                  }
+                  else { static_assert(false); }
+#endif
+               }
+            }
+         }
+         MFEM_SYNC_THREAD;
+         // Integrate
+         // if constexpr (is_gradient_fop<std::decay_t<output_fop_t>>::value) // Gradient
+         {
+            // const auto sB = reinterpret_cast<const real_t (*)[MQ1]>(Bo);
+            // const auto sG = reinterpret_cast<const real_t (*)[MQ1]>(Go);
+            low::GradTranspose3d(d1d, q1d, sB, sG, reg, sm1, sm0);
+            low::WriteDofs3d(d1d, 0, e, reg, YE);
+         }
+      },
+      num_entities, thread_blocks, 0, nullptr);
+      NVTX_END("forall");
+
+      NVTX_INI("out^T");
+      output_restriction_transpose(residual_e, residual_l);
+      NVTX_END("out^T");
+   }
+
+   using NewActionKernelType = decltype(&NewActionCallback::action_callback_new<>);
+   MFEM_REGISTER_KERNELS(NewActionCallbackKernels, NewActionKernelType, (int));
+
+   void Apply(const int d1d, const int q1d)
+   {
+      db1();
+      NewActionCallbackKernels::Run(q1d,
+                                    // args
+                                    d1d,
+                                    restriction_cb,
+                                    qfunc,
+                                    inputs,
+                                    input_to_field,
+                                    input_dtq_maps,
+                                    output_dtq_maps,
+                                    dimension,
+                                    num_entities,
+                                    test_vdim,
+                                    num_test_dof,
+                                    thread_blocks,
+                                    shmem_info,
+                                    attributes,
+                                    output_fop,
+                                    elem_attributes,
+                                    fields_e,
+                                    residual_e,
+                                    output_restriction_transpose,
+                                    solutions_l,
+                                    parameters_l,
+                                    residual_l,
+                                    // fallback arguments
+                                    q1d);
+   }
+};
+
+template<size_t num_fields, size_t num_inputs, size_t num_outputs,
+         typename restriction_cb_t, typename qfunc_t, typename input_t, typename output_fop_t>
+template<int T_Q1D>
+typename NewActionCallback<num_fields, num_inputs, num_outputs, restriction_cb_t, qfunc_t, input_t, output_fop_t>::NewActionKernelType
+NewActionCallback<num_fields, num_inputs, num_outputs, restriction_cb_t, qfunc_t, input_t, output_fop_t>::NewActionCallbackKernels::Kernel()
+{
+   return action_callback_new<T_Q1D>;
+}
+
+template<size_t num_fields, size_t num_inputs, size_t num_outputs,
+         typename restriction_cb_t, typename qfunc_t, typename input_t, typename output_fop_t>
+typename NewActionCallback<num_fields, num_inputs, num_outputs, restriction_cb_t, qfunc_t, input_t, output_fop_t>::NewActionKernelType
+NewActionCallback<num_fields, num_inputs, num_outputs, restriction_cb_t, qfunc_t, input_t, output_fop_t>::NewActionCallbackKernels::Fallback
+(int q1d)
+{
+   dbg("\x1b[33mFallback q1d:{}", q1d);
+   // MFEM_ABORT("No kernel for q1d=" << q1d);
+   // return nullptr;
+   return action_callback_new<>;
+}
+
+} // namespace mfem::future
@@ -0,0 +1,111 @@
+#pragma once
+
+#include "../util.hpp"
+#include "../../integrator_ctx.hpp"
+
+#include <utility>
+
+namespace mfem::future
+{
+
+namespace GlobalQFImpl
+{
+
+template<
+   typename qfunc_t,
+   typename inputs_t,
+   typename outputs_t,
+   size_t ninputs = tuple_size<inputs_t>::value,
+   size_t noutputs = tuple_size<outputs_t>::value>
+struct Action
+{
+   Action(
+      IntegratorContext ctx,
+      qfunc_t qfunc,
+      inputs_t inputs,
+      outputs_t outputs) :
+      ctx(ctx),
+      qfunc(std::move(qfunc)),
+      inputs(inputs),
+      outputs(outputs)
+   {
+      create_fop_to_fd(inputs, ctx.infds, input_to_infd);
+      create_fop_to_fd(outputs, ctx.outfds, output_to_outfd);
+
+      check_consistency(inputs, input_to_infd, ctx.infds);
+      check_consistency(outputs, output_to_outfd, ctx.outfds);
+
+      create_fieldbases(inputs, input_to_infd, ctx.infds, ctx.ir, input_bases);
+      create_fieldbases(outputs, output_to_outfd, ctx.outfds, ctx.ir, output_bases);
+
+      create_qlayouts(inputs, ctx.in_qlayouts, input_qlayouts);
+      create_qlayouts(outputs, ctx.out_qlayouts, output_qlayouts);
+
+      const int nqp = ctx.ir.GetNPoints();
+      gnqp = nqp * ctx.nentities;
+
+      xq_offsets.SetSize(ninputs + 1);
+      xq_offsets[0] = 0;
+      constexpr_for<0, ninputs>([&](auto i)
+      {
+         const auto input = get<i>(inputs);
+         xq_offsets[i + 1] = nqp * input.size_on_qp * ctx.nentities;
+      });
+      xq_offsets.PartialSum();
+      xq.Update(xq_offsets);
+
+      yq_offsets.SetSize(noutputs + 1);
+      yq_offsets[0] = 0;
+      constexpr_for<0, noutputs>([&](auto i)
+      {
+         const auto output = get<i>(outputs);
+         yq_offsets[i + 1] = nqp * output.size_on_qp * ctx.nentities;
+      });
+      yq_offsets.PartialSum();
+      yq.Update(yq_offsets);
+   }
+
+   void operator()(
+      const std::vector<Vector *> &xe,
+      std::vector<Vector *> &ye) const
+   {
+      if (ctx.attr.Size() == 0) { return; }
+
+      // E -> Q
+      interpolate(input_to_infd, input_bases, xe, xq);
+
+      // Q -> Q
+      static_assert(
+         detail::supports_tensor_array_qfunc<qfunc_t, inputs_t, outputs_t>::value,
+         "qfunc signature not supported by default backend Action");
+
+      detail::call_qfunc(
+         qfunc, xq, yq, gnqp, input_qlayouts, output_qlayouts,
+         std::make_index_sequence<ninputs> {},
+         std::make_index_sequence<noutputs> {});
+
+      // Q -> E
+      integrate(output_to_outfd, output_bases, yq, ye);
+   }
+
+   IntegratorContext ctx;
+   qfunc_t qfunc;
+   inputs_t inputs;
+   outputs_t outputs;
+
+   std::array<size_t, ninputs> input_to_infd;
+   std::array<size_t, noutputs> output_to_outfd;
+
+   std::array<FieldBasis, ninputs> input_bases;
+   std::array<FieldBasis, noutputs> output_bases;
+
+   std::array<std::vector<int>, ninputs>  input_qlayouts;
+   std::array<std::vector<int>, noutputs> output_qlayouts;
+
+   int gnqp = 0;
+   Array<int> xq_offsets, yq_offsets;
+   mutable BlockVector xq, yq;
+};
+
+}
+}
@@ -0,0 +1,131 @@
+#pragma once
+
+#include "../fem/quadinterpolator.hpp"
+#include "../../integrator_ctx.hpp"
+#include "../util.hpp"
+#include <utility>
+
+namespace mfem::future
+{
+
+namespace GlobalQFImpl
+{
+
+template<
+   int derivative_id,
+   typename qfunc_t,
+   typename inputs_t,
+   typename outputs_t,
+   size_t ninputs = tuple_size<inputs_t>::value,
+   size_t noutputs = tuple_size<outputs_t>::value>
+struct DerivativeActionEnzyme
+{
+   DerivativeActionEnzyme(
+      IntegratorContext ctx,
+      qfunc_t &qfunc,
+      inputs_t inputs,
+      outputs_t outputs) :
+      ctx(ctx),
+      qfunc(qfunc),
+      inputs(inputs),
+      outputs(outputs)
+   {
+      create_fop_to_fd(inputs, ctx.infds, input_to_infd);
+      create_fop_to_fd(outputs, ctx.outfds, output_to_outfd);
+
+      check_consistency(inputs, input_to_infd, ctx.infds);
+      check_consistency(outputs, output_to_outfd, ctx.outfds);
+
+      create_fieldbases(inputs, input_to_infd, ctx.infds, ctx.ir, input_bases);
+      create_fieldbases(outputs, output_to_outfd, ctx.outfds, ctx.ir, output_bases);
+
+      create_qlayouts(inputs, ctx.in_qlayouts, input_qlayouts);
+      create_qlayouts(outputs, ctx.out_qlayouts, output_qlayouts);
+
+      const int nqp = ctx.ir.GetNPoints();
+      gnqp = nqp * ctx.nentities;
+
+      xq_offsets.SetSize(ninputs + 1);
+      xq_offsets[0] = 0;
+      constexpr_for<0, ninputs>([&](auto i)
+      {
+         const auto input = get<i>(inputs);
+         xq_offsets[i + 1] = nqp * input.size_on_qp * ctx.nentities;
+      });
+      xq_offsets.PartialSum();
+      xq.Update(xq_offsets);
+
+      yq_offsets.SetSize(noutputs + 1);
+      yq_offsets[0] = 0;
+      constexpr_for<0, noutputs>([&](auto i)
+      {
+         const auto output = get<i>(outputs);
+         yq_offsets[i + 1] = nqp * output.size_on_qp * ctx.nentities;
+      });
+      yq_offsets.PartialSum();
+      yq.Update(yq_offsets);
+
+      // For each dependent input in the dependency map we create a shadow
+      // memory variable at the quadrature point level.
+      const auto activity_map = detail::make_activity_map<derivative_id>(inputs);
+      shadow_xq_offsets.SetSize(ninputs + 1);
+      shadow_xq_offsets = 0;
+      constexpr_for<0, ninputs>([&](auto i)
+      {
+         if (activity_map[i])
+         {
+            shadow_xq_offsets[i + 1] =
+               xq_offsets[i + 1] - xq_offsets[i];;
+         }
+      });
+      shadow_xq_offsets.PartialSum();
+      shadow_xq.Update(shadow_xq_offsets);
+   }
+
+   void operator()(
+      const std::vector<Vector *> &xe,
+      const Vector *de,
+      std::vector<Vector *> &ye) const
+   {
+      if (ctx.attr.Size() == 0) { return; }
+      // E -> Q
+      interpolate(input_to_infd, input_bases, xe, xq);
+
+      const auto activity_map = detail::make_activity_map<derivative_id>(inputs);
+      interpolate(input_to_infd, input_bases, xe, shadow_xq, activity_map);
+
+      // Q -> Q
+      static_assert(
+         detail::supports_tensor_array_qfunc<qfunc_t, inputs_t, outputs_t>::value,
+         "qfunc signature not supported by default backend Action");
+
+      detail::enzyme_fwddiff<derivative_id, qfunc_t, inputs_t, outputs_t>(
+         qfunc, xq, shadow_xq, yq, gnqp, input_qlayouts, output_qlayouts,
+         std::make_index_sequence<ninputs> {},
+         std::make_index_sequence<noutputs> {});
+
+      // Q -> E
+      integrate(output_to_outfd, output_bases, yq, ye);
+   }
+
+   IntegratorContext ctx;
+   qfunc_t &qfunc;
+   inputs_t inputs;
+   outputs_t outputs;
+
+   std::array<size_t, ninputs> input_to_infd;
+   std::array<size_t, noutputs> output_to_outfd;
+
+   std::array<FieldBasis, ninputs> input_bases;
+   std::array<FieldBasis, noutputs> output_bases;
+
+   std::array<std::vector<int>, ninputs>  input_qlayouts;
+   std::array<std::vector<int>, noutputs> output_qlayouts;
+
+   int gnqp = 0;
+   Array<int> xq_offsets, shadow_xq_offsets, yq_offsets;
+   mutable BlockVector xq, shadow_xq, yq;
+};
+
+}
+}
@@ -0,0 +1,42 @@
+#pragma once
+
+#include "action.hpp"
+#include "derivative_action_enzyme.hpp"
+
+namespace mfem::future
+{
+
+struct GlobalQFBackend
+{
+   template<
+      typename qfunc_t,
+      typename inputs_t,
+      typename outputs_t>
+   auto static MakeAction(
+      const IntegratorContext &ctx,
+      qfunc_t qfunc,
+      inputs_t inputs,
+      outputs_t outputs)
+   {
+      return GlobalQFImpl::Action(ctx, qfunc, inputs, outputs);
+   }
+
+   template<
+      int derivative_id,
+      typename qfunc_t,
+      typename inputs_t,
+      typename outputs_t>
+   auto static MakeDerivativeAction(
+      const IntegratorContext &ctx,
+      qfunc_t qfunc,
+      inputs_t inputs,
+      outputs_t outputs)
+   {
+      return GlobalQFImpl::DerivativeActionEnzyme<
+             derivative_id, qfunc_t, inputs_t, outputs_t>(
+                ctx, qfunc, inputs, outputs);
+   }
+
+};
+
+}
@@ -0,0 +1,166 @@
+#pragma once
+
+#include "../util.hpp"
+#include "../../integrator_ctx.hpp"
+
+#include <utility>
+
+namespace mfem::future
+{
+
+namespace LocalQFImpl
+{
+
+template<
+   typename qfunc_t,
+   typename inputs_t,
+   typename outputs_t,
+   size_t ninputs = tuple_size<inputs_t>::value,
+   size_t noutputs = tuple_size<outputs_t>::value>
+struct Action
+{
+   Action(
+      IntegratorContext ctx,
+      qfunc_t qfunc,
+      inputs_t inputs,
+      outputs_t outputs) :
+      ctx(ctx),
+      qfunc(std::move(qfunc)),
+      inputs(inputs),
+      outputs(outputs)
+   {
+      create_fop_to_fd(inputs, ctx.infds, input_to_infd);
+      create_fop_to_fd(outputs, ctx.outfds, output_to_outfd);
+
+      check_consistency(inputs, input_to_infd, ctx.infds);
+      check_consistency(outputs, output_to_outfd, ctx.outfds);
+
+      const int nqp = ctx.ir.GetNPoints();
+
+      // Initialize DofToQuad maps for inputs
+      for_constexpr<ninputs>([&](auto i)
+      {
+         const auto &fd = ctx.infds[input_to_infd[i]];
+         std::visit([&](auto* space_ptr)
+         {
+            using T = std::decay_t<decltype(*space_ptr)>;
+            if constexpr (std::is_same_v<T, FiniteElementSpace> ||
+                          std::is_same_v<T, ParFiniteElementSpace>)
+            {
+               const auto *fe = space_ptr->GetTypicalFE();
+               input_dtq_maps[i] = &fe->GetDofToQuad(ctx.ir, DofToQuad::TENSOR);
+            }
+         }, fd.data);
+      });
+
+      // Initialize DofToQuad maps for outputs
+      for_constexpr<noutputs>([&](auto i)
+      {
+         const auto &fd = ctx.outfds[output_to_outfd[i]];
+         std::visit([&](auto* space_ptr)
+         {
+            using T = std::decay_t<decltype(*space_ptr)>;
+            if constexpr (std::is_same_v<T, FiniteElementSpace> ||
+                          std::is_same_v<T, ParFiniteElementSpace>)
+            {
+               const auto *fe = space_ptr->GetTypicalFE();
+               output_dtq_maps[i] = &fe->GetDofToQuad(ctx.ir, DofToQuad::TENSOR);
+            }
+         }, fd.data);
+      });
+   }
+
+   void operator()(
+      const std::vector<Vector *> &xe,
+      std::vector<Vector *> &ye) const
+   {
+      if (ctx.attr.Size() == 0) { return; }
+
+      // input_dtq_maps
+
+      // const auto B = (const real_t*)input_dtq_maps[0/*i*/].B;
+      // const auto G = (const real_t*)input_dtq_maps[0/*i*/].G;
+
+      //    dfem::forall<T_Q1D*T_Q1D*T_Q1D>([=] MFEM_HOST_DEVICE (int e, void *)
+      //    {
+      //       if (has_attr && !d_attr[d_elem_attr[e] - 1]) { return; }
+
+      //       constexpr int MQ1 = T_Q1D > 0 ? T_Q1D : 8;
+
+      //       MFEM_SHARED real_t sm0[MQ1][MQ1][MQ1][3];
+      //       MFEM_SHARED real_t sm1[MQ1][MQ1][MQ1][3];
+
+      //       low::regs3d_t<DIM, MQ1> reg;
+      //       const real_t *rd = dx_ptr;
+
+      //       MFEM_SHARED real_t sB[MQ1][MQ1], sG[MQ1][MQ1];
+      //       {
+      //          low::LoadMatrix(d1d, q1d, B, sB);
+      //          low::LoadMatrix(d1d, q1d, G, sG);
+      //          {
+      //             low::LoadDofs3d(e, d1d, XE, sm0);
+      //             low::Grad3d(d1d, q1d, sB, sG, sm0, sm1, reg);
+      //          }
+      //       }
+      //       // else if constexpr (is_identity_fop<field_operator_t>::value)   // Identity
+      //       {
+      //          // db1("Identity");
+      //          // rd = fields_e_ptr[input_to_field[i]];
+      //          // rd = dx_ptr;
+      //       }
+      //    }
+
+      //    MFEM_FOREACH_THREAD_DIRECT(qz,z,q1d)
+      //    {
+      //       MFEM_FOREACH_THREAD_DIRECT(qy,y,q1d)
+      //       {
+      //          MFEM_FOREACH_THREAD_DIRECT(qx,x,q1d)
+      //          {
+
+      //             auto args = decay_tuple<qf_param_ts> {};
+      //             get<0>(args) = as_tensor<real_t, 3>(&reg[qz][qy][qx][0]);
+      //             if constexpr (T_Q1D > 0)
+      //             {
+      //                get<1>(args) = as_tensor<real_t, 3, 3>(rd + 9*(qx*T_Q1D*T_Q1D + qy*T_Q1D + qz));
+      //             }
+      //             else
+      //             {
+      //                get<1>(args) = as_tensor<real_t, 3, 3>(rd + 9*(qx*q1d*q1d + qy*q1d + qz));
+      //             }
+      //             auto r = get<0>(apply(qfunc, args));
+      //             if constexpr (decltype(r)::ndim == 1)
+      //             {
+      //                as_tensor<real_t, 3>(&reg[qz][qy][qx][0]) = r;
+      //             }
+      //             else { static_assert(false); }
+      //          }
+      //       }
+      //    }
+      //    MFEM_SYNC_THREAD;
+      //    // Integrate
+      //    // if constexpr (is_gradient_fop<std::decay_t<output_fop_t>>::value) // Gradient
+      //    {
+      //       // const auto sB = reinterpret_cast<const real_t (*)[MQ1]>(Bo);
+      //       // const auto sG = reinterpret_cast<const real_t (*)[MQ1]>(Go);
+      //       low::GradTranspose3d(d1d, q1d, sB, sG, reg, sm1, sm0);
+      //       low::WriteDofs3d(d1d, 0, e, reg, YE);
+      //    }
+      // },
+      // num_entities, thread_blocks, 0, nullptr);
+   }
+
+
+   IntegratorContext ctx;
+   qfunc_t qfunc;
+   inputs_t inputs;
+   outputs_t outputs;
+
+   std::array<size_t, ninputs> input_to_infd;
+   std::array<size_t, noutputs> output_to_outfd;
+
+   std::array<const DofToQuad*, ninputs> input_dtq_maps;
+   std::array<const DofToQuad*, noutputs> output_dtq_maps;
+};
+
+}
+}
@@ -0,0 +1,39 @@
+#pragma once
+
+#include "../../integrator_ctx.hpp"
+#include "action.hpp"
+
+namespace mfem::future
+{
+
+struct LocalQFBackend
+{
+   template<
+      typename qfunc_t,
+      typename inputs_t,
+      typename outputs_t>
+   auto static MakeAction(
+      const IntegratorContext &ctx,
+      qfunc_t qfunc,
+      inputs_t inputs,
+      outputs_t outputs)
+   {
+      return LocalQFImpl::Action(ctx, qfunc, inputs, outputs);
+   }
+
+   template<
+      int derivative_id,
+      typename qfunc_t,
+      typename inputs_t,
+      typename outputs_t>
+   auto static MakeDerivativeAction(
+      const IntegratorContext &ctx,
+      qfunc_t qfunc,
+      inputs_t inputs,
+      outputs_t outputs)
+   {
+      MFEM_ABORT("LocalQFBackend does not support derivative actions.");
+   }
+};
+
+}
@@ -0,0 +1,659 @@
+#pragma once
+
+#include "../fem/quadinterpolator.hpp"
+#include "../util.hpp"
+#include "general/enzyme.hpp"
+
+namespace mfem::future
+{
+
+template <size_t N, size_t... Is>
+constexpr std::array<bool, N> all_true_impl(std::index_sequence<Is...>)
+{
+   return {{((void)Is, true)...}};
+}
+
+template <size_t N>
+constexpr std::array<bool, N> all_true()
+{
+   return all_true_impl<N>(std::make_index_sequence<N> {});
+}
+
+struct FieldBasis
+{
+   // E-vector -> Q-vector
+   std::function<void(const Vector &, Vector &)> forward;
+
+   // Q-vector -> E-vector
+   std::function<void(const Vector &, Vector &)> transpose;
+};
+
+inline FieldBasis FromQI(const QuadratureInterpolator *qi,
+                         QuadratureInterpolator::EvalFlags mode)
+{
+   return
+   {
+      [qi, mode](const Vector &xe, Vector &xq)
+      {
+         qi->SetOutputLayout(QVectorLayout::byVDIM);
+         if (mode == QuadratureInterpolator::VALUES)
+         {
+            qi->Values(xe, xq);
+         }
+         else
+         {
+            qi->Derivatives(xe, xq);
+         }
+      },
+      [qi, mode](const Vector &yq, Vector &ye)
+      {
+         Vector empty;
+         qi->SetOutputLayout(QVectorLayout::byVDIM);
+         if (mode == QuadratureInterpolator::VALUES)
+         {
+            qi->AddMultTranspose(QuadratureInterpolator::VALUES, yq, empty, ye);
+         }
+         else
+         {
+            qi->AddMultTranspose(QuadratureInterpolator::DERIVATIVES, empty, yq, ye);
+         }
+      }
+   };
+}
+
+// QuadratureFunction identity copy
+inline FieldBasis FromQF()
+{
+   return
+   {
+      [](const Vector &xe, Vector &xq) { xq = xe; },
+      [](const Vector &yq, Vector &ye) { ye = yq; }
+   };
+}
+
+// User-defined parameter space B
+inline FieldBasis FromPS(const Operator *B, const Operator *Bt)
+{
+   return
+   {
+      [B](const Vector &xe, Vector &xq) { B->Mult(xe, xq); },
+      [Bt](const Vector &yq, Vector &ye) { Bt->Mult(yq, ye); }
+   };
+}
+
+inline FieldBasis FieldBasisFromWeight(const IntegrationRule &ir)
+{
+   return
+   {
+      [&ir](const Vector &, Vector &xq)
+      {
+         const int nqp = ir.GetNPoints();
+         MFEM_ASSERT(xq.Size() % nqp == 0, "weight block has unexpected size");
+
+         const int ne = xq.Size() / nqp;
+         const real_t *wref = ir.GetWeights().Read();
+
+         for (int e = 0; e < ne; e++)
+         {
+            std::memcpy(xq.GetData() + e*nqp, wref, nqp*sizeof(real_t));
+         }
+      },
+      [](const Vector &, Vector &) {}
+   };
+}
+
+inline const FieldBasis GetFieldBasis(const FieldDescriptor &f,
+                                      const IntegrationRule &ir,
+                                      QuadratureInterpolator::EvalFlags mode)
+{
+   return std::visit([&ir, &mode](auto && arg) -> FieldBasis
+   {
+      using T = std::decay_t<decltype(arg)>;
+
+      if constexpr (std::is_same_v<T, const FiniteElementSpace *>)
+      {
+         return FromQI(arg->GetQuadratureInterpolator(ir), mode);
+      }
+      else if constexpr (std::is_same_v<T, const ParFiniteElementSpace *>)
+      {
+         return FromQI(arg->GetQuadratureInterpolator(ir), mode);
+      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         return FromQF();
+      }
+      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
+      {
+         return FromPS(arg->GetB(), arg->GetBt());
+      }
+      else if constexpr (std::is_same_v<T, const IntegrationRule *>)
+      {
+         return FieldBasis{};
+      }
+      else
+      {
+         static_assert(dfem::always_false<T>, "internal error");
+      }
+   }, f.data);
+}
+
+template <typename fops_t, size_t nfops>
+void create_fieldbases(
+   fops_t &fops,
+   const std::array<size_t, nfops> &fop_to_fd,
+   const std::vector<FieldDescriptor> &fds,
+   const IntegrationRule &ir,
+   std::array<FieldBasis, nfops> &bases)
+{
+   constexpr_for<0, nfops>([&](auto i)
+   {
+      const auto fop = get<i>(fops);
+      using fop_t = std::decay_t<decltype(fop)>;
+
+      const auto fd = fds[fop_to_fd[i]];
+
+      constexpr QuadratureInterpolator::EvalFlags dummy_mode =
+         QuadratureInterpolator::VALUES;
+      if constexpr (is_identity_fop<fop_t>::value)
+      {
+         bases[i] = GetFieldBasis(fd, ir, dummy_mode);
+      }
+      else if constexpr (is_weight_fop<fop_t>::value)
+      {
+         bases[i] = FieldBasisFromWeight(ir);
+      }
+      else if constexpr (is_value_fop<fop_t>::value)
+      {
+         bases[i] = GetFieldBasis(fd, ir, QuadratureInterpolator::VALUES);
+      }
+      else if constexpr (is_gradient_fop<fop_t>::value)
+      {
+         bases[i] = GetFieldBasis(fd, ir, QuadratureInterpolator::DERIVATIVES);
+      }
+   });
+}
+
+template <typename fops_t, size_t nfops>
+void check_consistency(
+   fops_t &fops,
+   const std::array<size_t, nfops> &fop_to_fd,
+   const std::vector<FieldDescriptor> &fields)
+{
+   constexpr_for<0, nfops>([&](auto i)
+   {
+      const auto input = get<i>(fops);
+      using input_t = std::decay_t<decltype(input)>;
+
+      const auto fd = fields[fop_to_fd[i]];
+
+      if constexpr (is_identity_fop<input_t>::value)
+      {
+         MFEM_ASSERT(std::holds_alternative<const QuadratureFunction *>(fd.data),
+                     "Identity FieldOperator requested on non "
+                     "QuadratureFunction");
+      }
+      else if constexpr (is_weight_fop<input_t>::value)
+      {
+      }
+      else if constexpr (is_value_fop<input_t>::value)
+      {
+         MFEM_ASSERT(std::holds_alternative<const FiniteElementSpace *>(fd.data) ||
+                     std::holds_alternative<const ParFiniteElementSpace *>(fd.data) ||
+                     std::holds_alternative<const ParameterSpace *>(fd.data),
+                     "Value FieldOperator requested on non "
+                     "QuadratureFunction");
+      }
+      else if constexpr (is_gradient_fop<input_t>::value)
+      {
+         MFEM_ASSERT(std::holds_alternative<const FiniteElementSpace *>(fd.data) ||
+                     std::holds_alternative<const ParFiniteElementSpace *>(fd.data),
+                     "Value FieldOperator requested on non "
+                     "QuadratureFunction");
+      }
+   });
+}
+
+template <size_t ninputs>
+void interpolate(
+   const std::array<size_t, ninputs> &input_to_infd,
+   const std::array<FieldBasis, ninputs> &input_bases,
+   const std::vector<Vector *> &xe,
+   BlockVector &xq,
+   const std::array<bool, ninputs> &conditional = all_true<ninputs>())
+{
+   constexpr_for<0, ninputs>([&](auto i)
+   {
+      if (!conditional.empty() && !conditional[i]) { return; }
+
+      input_bases[i].forward(*xe[input_to_infd[i]], xq.GetBlock(i));
+   });
+}
+
+template <size_t noutputs>
+void integrate(
+   const std::array<size_t, noutputs> &output_to_outfd,
+   const std::array<FieldBasis, noutputs> &output_bases,
+   const BlockVector &yq,
+   std::vector<Vector *> &ye)
+{
+   for (auto v : ye) { *v = 0.0; }
+
+   constexpr_for<0, noutputs>([&](auto i)
+   {
+      output_bases[i].transpose(yq.GetBlock(i), *ye[output_to_outfd[i]]);
+   });
+}
+
+
+namespace detail
+{
+
+template <typename T>
+struct is_tensor_array : std::false_type {};
+
+template <typename scalar_t, int... Dims>
+struct is_tensor_array<tensor_array<scalar_t, Dims...>> : std::true_type {};
+
+template <typename T>
+struct is_tensor_array_mut : std::false_type {};
+
+template <typename scalar_t, int... Dims>
+struct is_tensor_array_mut<tensor_array<scalar_t, Dims...>> :
+                                                            std::bool_constant<!std::is_const_v<scalar_t>> {};
+
+
+template <typename ndarray_t>
+inline void set_layout_default(ndarray_t &a)
+{
+   if constexpr (ndarray_t::tensor_rank() == 0) { return; }
+
+   constexpr std::size_t nd = ndarray_t::rank();
+   constexpr std::size_t td = ndarray_t::tensor_rank();
+   std::array<std::size_t, nd + td> perm{};
+
+   for (std::size_t i = 0; i < td; i++) { perm[i] = nd + i; }
+   for (std::size_t i = 0; i < nd; i++) { perm[td + i] = i; }
+
+   a.set_layout(perm);
+}
+
+template <typename ndarray_t>
+inline void set_layout(ndarray_t& a, const std::vector<int>& layout)
+{
+   if constexpr (ndarray_t::tensor_rank() == 0) { return; }
+
+   constexpr std::size_t nd = ndarray_t::rank();
+   constexpr std::size_t td = ndarray_t::tensor_rank();
+   constexpr std::size_t N  = nd + td;
+
+   // missing means default
+   if (layout.empty()) { set_layout_default(a); return; }
+
+   MFEM_VERIFY(layout.size() == N,
+               "layout size mismatch: expected " << N << " got " << layout.size());
+
+   // TODO: make a version of set_layout that takes `std::vector<int>`
+   std::array<std::size_t, N> perm{};
+   for (std::size_t i = 0; i < N; i++)
+   {
+      MFEM_VERIFY(layout[i] >= 0, "layout index must be >=0");
+      perm[i] = static_cast<std::size_t>(layout[i]);
+   }
+
+   a.set_layout(perm);
+}
+
+/// Primary template: intentionally undefined — gives a clear error for unsupported types.
+template <typename T>
+struct tensor_array_traits;
+
+/// Matches tensor<scalar_t, sizes...>
+template <typename scalar_t, int... sizes>
+struct tensor_array_traits<tensor<scalar_t, sizes...>>
+{
+   using scalar_type = scalar_t;
+   template <std::size_t ndims>
+   using array_type = tensor_ndarray<scalar_t, ndims, sizes...>;
+};
+
+/// Matches tensor_ndarray<scalar_t, ndims, tensor_sizes...>
+template <typename scalar_t, int ndims, int... tensor_sizes>
+struct tensor_array_traits<tensor_ndarray<scalar_t, ndims, tensor_sizes...>>
+{
+   using scalar_type = scalar_t;
+   template <std::size_t N>
+   using array_type = tensor_ndarray<scalar_t, N, tensor_sizes...>;
+};
+
+/// Entry point: explicit tensor type T as template argument.
+template <typename T, typename ptr_scalar_t, typename... dyn_sizes_t>
+decltype(auto) make_tensor_array(ptr_scalar_t *ptr,
+                                 const std::vector<int>* layout,
+                                 dyn_sizes_t... dynamic_sizes)
+{
+   using traits = tensor_array_traits<T>;
+   using array_t = typename traits::template array_type<sizeof...(dynamic_sizes)>;
+   auto a = array_t(ptr, {std::size_t(dynamic_sizes)...});
+   if (layout) { set_layout(a, *layout); }
+   else        { set_layout_default(a); }
+   return a;
+}
+
+template <typename qfunc_t, typename inputs_t, typename outputs_t>
+struct supports_tensor_array_qfunc
+{
+   using qf_signature = typename get_function_signature<qfunc_t>::type;
+   using qf_param_ts = typename qf_signature::parameter_ts;
+
+   static constexpr int ninputs = tuple_size<inputs_t>::value;
+   static constexpr int noutputs = tuple_size<outputs_t>::value;
+   static constexpr int nparams = tuple_size<qf_param_ts>::value;
+
+   template <std::size_t... Is>
+   static constexpr bool InputsOk(std::index_sequence<Is...>)
+   {
+      return (is_tensor_array<std::remove_cv_t<std::remove_reference_t<
+              typename tuple_element<Is, qf_param_ts>::type>>>::value && ...);
+   }
+
+   template <std::size_t... Is>
+   static constexpr bool OutputsOk(std::index_sequence<Is...>)
+   {
+      return (is_tensor_array_mut<std::remove_cv_t<std::remove_reference_t<
+              typename tuple_element<ninputs + Is, qf_param_ts>::type>>>::value && ...);
+   }
+
+   static constexpr bool value =
+      (nparams == ninputs + noutputs) &&
+      InputsOk(std::make_index_sequence<ninputs> {}) &&
+      OutputsOk(std::make_index_sequence<noutputs> {});
+};
+
+template <typename qfunc_t, std::size_t... Is, std::size_t... Os>
+inline void call_qfunc(
+   const qfunc_t &qfunc,
+   const BlockVector &xq,
+   BlockVector &yq,
+   int gnqp,
+   const std::array<std::vector<int>, sizeof...(Is)>& in_layouts,
+   const std::array<std::vector<int>, sizeof...(Os)>& out_layouts,
+   std::index_sequence<Is...>,
+   std::index_sequence<Os...>)
+{
+   constexpr std::size_t ninputs = sizeof...(Is);
+
+   using qf_signature = typename get_function_signature<qfunc_t>::type;
+   using qf_param_ts = typename qf_signature::parameter_ts;
+
+   auto inputs = std::make_tuple(
+                    make_tensor_array<std::remove_cv_t<std::remove_reference_t<
+                    typename tuple_element<Is, qf_param_ts>::type>>>(
+                       xq.GetBlock(Is).Read(), &in_layouts[Is], gnqp)...);
+
+   auto outputs = std::make_tuple(
+                     make_tensor_array<std::remove_cv_t<std::remove_reference_t<
+                     typename tuple_element<ninputs + Os, qf_param_ts>::type>>>(
+                        yq.GetBlock(Os).ReadWrite(), &out_layouts[Os], gnqp)...);
+
+   std::apply([&](auto&&... args)
+   {
+      qfunc(args...);
+   }, std::tuple_cat(inputs, outputs));
+}
+
+template <typename func_t, typename... arg_ts>
+MFEM_HOST_DEVICE inline
+auto qfunction_wrapper(const func_t &f, arg_ts...args)
+{
+   return f(args...);
+}
+
+template <std::size_t derivative_id, std::size_t I, typename Tuple, std::size_t... Is>
+constexpr std::array<bool, sizeof...(Is)>
+make_activity_array(std::index_sequence<Is...>)
+{
+   return { (std::decay_t<typename tuple_element<Is, Tuple>::type>::GetFieldId() == derivative_id)... };
+}
+
+template <std::size_t derivative_id, typename inputs_t, std::size_t... Is>
+constexpr auto make_activity_map_impl(std::index_sequence<Is...>)
+{
+   constexpr std::size_t N = sizeof...(Is);
+
+   if constexpr (N == 0)
+      return std::array<bool, 0> {};
+
+   return make_activity_array<derivative_id, 0, inputs_t>
+          (std::make_index_sequence<N> {});
+}
+
+template <std::size_t derivative_id, typename inputs_t>
+constexpr auto make_activity_map(inputs_t)
+{
+   return make_activity_map_impl<derivative_id, inputs_t>(
+             std::make_index_sequence<tuple_size<inputs_t>::value> {});
+}
+
+namespace enzyme_detail
+{
+
+template <auto wrapper_fn, typename qf_return_t, typename... AccArgs>
+__attribute__((always_inline)) inline void
+do_enzyme_call(AccArgs... acc)
+{
+   __enzyme_fwddiff<qf_return_t>(wrapper_fn, acc...);
+}
+
+template <auto wrapper_fn, typename qf_return_t,
+          size_t CurO, size_t NO,
+          typename primals_t, typename derivs_t,
+          typename... AccArgs>
+__attribute__((always_inline)) inline void
+process_outputs(primals_t &primals, derivs_t &derivs, AccArgs... acc)
+{
+   if constexpr (CurO == NO)
+   {
+      do_enzyme_call<wrapper_fn, qf_return_t>(acc...);
+   }
+   else
+   {
+      process_outputs<wrapper_fn, qf_return_t, CurO + 1, NO>(
+         primals, derivs,
+         acc...,
+         enzyme_dupnoneed,
+         &std::get<CurO>(primals),
+         &std::get<CurO>(derivs));
+   }
+}
+
+template <auto wrapper_fn, typename qf_return_t,
+          size_t CurI, size_t NI, bool... ActivityMap,
+          typename inputs_t, typename shadows_t,
+          typename primals_t, typename derivs_t,
+          typename... AccArgs>
+__attribute__((always_inline)) inline void
+process_inputs(inputs_t &inputs, shadows_t &shadows,
+               primals_t &primals, derivs_t &derivs,
+               AccArgs... acc)
+{
+   if constexpr (CurI == NI)
+   {
+      constexpr size_t NO = std::tuple_size_v<primals_t>;
+      process_outputs<wrapper_fn, qf_return_t, 0, NO>(
+         primals, derivs, acc...);
+   }
+   else
+   {
+      constexpr bool active =
+         std::array<bool, sizeof...(ActivityMap)> {ActivityMap...} [CurI];
+
+      if constexpr (active)
+      {
+         std::cout << "Input[" << CurI << "]: ACTIVE (enzyme_dup)\n"
+                   << "  primal ptr type:  "
+                   << get_type_name<decltype(&std::get<CurI>(inputs))>() << "\n"
+                   << "  shadow ptr type:  "
+                   << get_type_name<decltype(&std::get<CurI>(shadows))>() << "\n";
+      }
+      else
+      {
+         std::cout << "Input[" << CurI << "]: INACTIVE (enzyme_const)\n"
+                   << "  primal ptr type:  "
+                   << get_type_name<decltype(&std::get<CurI>(inputs))>() << "\n";
+      }
+
+      if constexpr (active)
+      {
+         process_inputs<wrapper_fn, qf_return_t, CurI + 1, NI, ActivityMap...>(
+            inputs, shadows, primals, derivs,
+            acc...,
+            enzyme_dup,
+            &std::get<CurI>(inputs),
+            &std::get<CurI>(shadows));
+      }
+      else
+      {
+         process_inputs<wrapper_fn, qf_return_t, CurI + 1, NI, ActivityMap...>(
+            inputs, shadows, primals, derivs,
+            acc...,
+            enzyme_const,
+            &std::get<CurI>(inputs));
+      }
+   }
+}
+
+} // namespace enzyme_detail
+
+template <size_t derivative_id, typename qfunc_t, typename inputs_t, typename outputs_t,
+          std::size_t... Is, std::size_t... Os>
+inline void enzyme_fwddiff(
+   qfunc_t &qfunc,
+   const BlockVector &xq,
+   const BlockVector &shadow_xq,
+   BlockVector &yq,
+   const int &gnqp,
+   const std::array<std::vector<int>, sizeof...(Is)>& in_layouts,
+   const std::array<std::vector<int>, sizeof...(Os)>& out_layouts,
+   std::index_sequence<Is...>,
+   std::index_sequence<Os...>)
+{
+#ifdef MFEM_USE_ENZYME
+   constexpr std::size_t ninputs  = sizeof...(Is);
+   constexpr std::size_t noutputs = sizeof...(Os);
+
+   using qf_signature = typename get_function_signature<qfunc_t>::type;
+   using qf_param_ts  = typename qf_signature::parameter_ts;
+   using qf_return_t  = typename qf_signature::return_t;
+
+   constexpr auto activity_map = make_activity_map<derivative_id>(inputs_t{});
+   static_assert(activity_map.size() == ninputs, "activity map size mismatch");
+
+   std::cout << "activity_map: ";
+   for (const auto &v : activity_map)
+   {
+      std::cout << v << " ";
+   }
+   std::cout << "\n";
+
+   auto inputs = std::make_tuple(
+                    make_tensor_array<std::remove_cv_t<std::remove_reference_t<
+                    typename tuple_element<Is, qf_param_ts>::type>>>(
+                       xq.GetBlock(Is).Read(), &in_layouts[Is], gnqp)...);
+
+   auto shadows = std::make_tuple(
+                     make_tensor_array<std::remove_cv_t<std::remove_reference_t<
+                     typename tuple_element<Is, qf_param_ts>::type>>>(
+                        shadow_xq.GetBlock(Is).Read(), &in_layouts[Is], gnqp)...);
+
+   std::array<Vector, noutputs> primal_storage;
+   ((primal_storage[Os].SetSize(yq.GetBlock(Os).Size())), ...);
+
+   auto primals_out = std::make_tuple(
+                         make_tensor_array<std::remove_cv_t<std::remove_reference_t<
+                         typename tuple_element<ninputs + Os, qf_param_ts>::type>>>(
+                            primal_storage[Os].ReadWrite(), &out_layouts[Os], gnqp)...);
+
+   auto derivs_out = std::make_tuple(
+                        make_tensor_array<std::remove_cv_t<std::remove_reference_t<
+                        typename tuple_element<ninputs + Os, qf_param_ts>::type>>>(
+                           yq.GetBlock(Os).ReadWrite(), &out_layouts[Os], gnqp)...);
+
+   using wrapper_fn_t = qf_return_t (*)(
+                           const qfunc_t &,
+                           std::remove_reference_t<decltype(std::get<Is>(inputs))>...,
+                           std::remove_reference_t<decltype(std::get<Os>(primals_out))>...);
+
+   constexpr wrapper_fn_t wrapper_fn =
+      qfunction_wrapper<qfunc_t,
+      std::remove_reference_t<decltype(std::get<Is>(inputs))>...,
+      std::remove_reference_t<decltype(std::get<Os>(primals_out))>...>;
+
+   // wrapper_fn travels as a non-type template parameter throughout without
+   // being stored.
+   enzyme_detail::process_inputs<
+   wrapper_fn,
+   qf_return_t,
+   0,
+   ninputs,
+   activity_map[Is]...
+   >(inputs, shadows,
+     primals_out, derivs_out,
+     enzyme_const, &qfunc // seed: qfunc is always inactive
+    );
+
+#else
+   MFEM_ABORT("enzyme_fwddiff requires MFEM_USE_ENZYME");
+#endif
+}
+
+} // namespace detail
+
+// Create quadrature function fop to fields map
+template <typename fops_t, size_t N = tuple_size<fops_t>::value, size_t M>
+void create_fop_to_fd(const fops_t &fops,
+                      const std::vector<FieldDescriptor> &fields,
+                      std::array<size_t, M> &fop_to_fd)
+{
+   static_assert(N == M, "sizes must match");
+   constexpr_for<0, N>([&](auto i)
+   {
+      const auto fop = get<i>(fops);
+      fop_to_fd[i] = std::numeric_limits<size_t>::max();
+      for (size_t j = 0; j < fields.size(); j++)
+      {
+         // TODO: output.GetFieldId() should probably store/return size_t
+         if (static_cast<int>(fields[j].id) == fop.GetFieldId())
+         {
+            fop_to_fd[i] = j;
+         }
+      }
+      // Handle Weight type. There is no FieldDescriptor for the weight.
+      // TODO: Create weight descriptor for the weight for internal use?
+      // TODO: this is a hack...
+      if (is_weight_fop<std::remove_cv_t<decltype(fop)>>::value)
+      {
+         fop_to_fd[i] = 0;
+      }
+      else if (fop_to_fd[i] == std::numeric_limits<size_t>::max())
+      {
+         MFEM_ABORT("not found");
+      }
+   });
+}
+
+template <typename fops_t, size_t nfops>
+void create_qlayouts(const fops_t &fops,
+                     const std::unordered_map<std::type_index, std::vector<int>> &a,
+                     std::array<std::vector<int>, nfops> &b)
+{
+   constexpr_for<0, nfops>([&](auto i)
+   {
+      using fop_t =
+         std::remove_cv_t<std::remove_reference_t<decltype(get<i>(fops))>>;
+      auto it = a.find(std::type_index(typeid(fop_t)));
+      if (it != a.end()) { b[i] = it->second; }
+      else               { b[i].clear(); }
+   });
+}
+
+}
@@ -11,44 +11,119 @@

 #include "doperator.hpp"

+#include <algorithm>
+
 #ifdef MFEM_USE_MPI

 using namespace mfem;
 using namespace mfem::future;

-void DifferentiableOperator::SetParameters(std::vector<Vector *> p) const
+DifferentiableOperator::DifferentiableOperator(
+   const std::vector<FieldDescriptor> &infds,
+   const std::vector<FieldDescriptor> &outfds,
+   const ParMesh &mesh) :
+   Operator(),
+   mesh(mesh),
+   infds(infds),
+   outfds(outfds)
 {
-   MFEM_ASSERT(parameters.size() == p.size(),
-               "number of parameters doesn't match descriptors");
-   for (size_t i = 0; i < parameters.size(); i++)
+   unionfds.clear();
+   unionfds.insert(unionfds.end(), infds.begin(), infds.end());
+   unionfds.insert(unionfds.end(), outfds.begin(), outfds.end());
+   std::sort(unionfds.begin(), unionfds.end());
+   auto last = std::unique(unionfds.begin(), unionfds.end());
+   unionfds.erase(last, unionfds.end());
+
+   infields_l.resize(infds.size());
+   for (size_t i = 0; i < infds.size(); i++)
   {
-      p[i]->Read();
-      parameters_l[i] = *p[i];
+      infields_l[i] = new Vector(GetVSize(infds[i]));
   }
+
+   infields_e.resize(infds.size());
 }

-DifferentiableOperator::DifferentiableOperator(
-   const std::vector<FieldDescriptor> &solutions,
-   const std::vector<FieldDescriptor> &parameters,
-   const ParMesh &mesh) :
-   mesh(mesh),
-   solutions(solutions),
-   parameters(parameters)
+void DifferentiableOperator::SetMultLevel(MultLevel level)
 {
-   fields.resize(solutions.size() + parameters.size());
-   fields_e.resize(fields.size());
-   solutions_l.resize(solutions.size());
-   parameters_l.resize(parameters.size());
+   mult_level = level;
+}

-   for (size_t i = 0; i < solutions.size(); i++)
+void DifferentiableOperator::Mult(const Vector &x, Vector &y) const
+{
+   MFEM_ASSERT(!action_callbacks.empty(),
+               "no integrators have been set");
+
+   MFEM_ASSERT(dynamic_cast<const BlockVector*>(&x),
+               "x needs to be a BlockVector");
+
+   MFEM_ASSERT(dynamic_cast<const BlockVector*>(&y),
+               "y needs to be a BlockVector");
+
+   const auto &bx = static_cast<const BlockVector &>(x);
+   auto &by = static_cast<BlockVector &>(y);
+
+   Mult(bx, by);
+}
+
+void DifferentiableOperator::DisableTensorProductStructure(bool disable)
+{
+   use_tensor_product_structure = !disable;
+}
+
+std::shared_ptr<DerivativeOperator> DifferentiableOperator::GetDerivative(
+   size_t derivative_id, const Vector &x)
+{
+   MFEM_ASSERT(derivative_action_callbacks.find(derivative_id) !=
+               derivative_action_callbacks.end(),
+               "no derivative action has been found for ID " << derivative_id);
+
+   const size_t dfidx = FindIdx(derivative_id, infds);
+
+   // Get transpose callbacks if available, otherwise pass empty vector
+   std::vector<derivative_action_t> transpose_callbacks;
+   auto it = daction_transpose_callbacks.find(derivative_id);
+   if (it != daction_transpose_callbacks.end())
   {
-      fields[i] = solutions[i];
+      transpose_callbacks = it->second;
   }

-   for (size_t i = 0; i < parameters.size(); i++)
+   return std::make_shared<DerivativeOperator>(
+             height,
+             GetTrueVSize(infds[dfidx]),
+             derivative_action_callbacks[derivative_id],
+             transpose_callbacks,
+             infds[dfidx],
+             x,
+             infds,
+             outfds);
+}
+
+std::shared_ptr<DerivativeOperator> DifferentiableOperator::GetDerivative(
+   size_t derivative_id, const MultiVector &x)
+{
+   MFEM_ASSERT(derivative_action_callbacks.find(derivative_id) !=
+               derivative_action_callbacks.end(),
+               "no derivative action has been found for ID " << derivative_id);
+
+   const size_t dfidx = FindIdx(derivative_id, infds);
+
+   // Get transpose callbacks if available, otherwise pass empty vector
+   std::vector<derivative_action_t> transpose_callbacks;
+   auto it = daction_transpose_callbacks.find(derivative_id);
+   if (it != daction_transpose_callbacks.end())
   {
-      fields[i + solutions.size()] = parameters[i];
+      transpose_callbacks = it->second;
   }
+
+   return std::make_shared<DerivativeOperator>(
+             height,
+             GetTrueVSize(infds[dfidx]),
+             derivative_action_callbacks[derivative_id],
+             transpose_callbacks,
+             infds[dfidx],
+             x,
+             infds,
+             outfds);
 }

 #endif // MFEM_USE_MPI
@@ -0,0 +1,63 @@
+// Copyright (c) 2010-2025, 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.
+#pragma once
+
+#include "../fespace.hpp"
+#include "parameterspace.hpp"
+
+namespace mfem::future
+{
+
+/// @brief FieldDescriptor struct
+///
+/// This struct is used to store information about a field.
+struct FieldDescriptor
+{
+   using data_variant_t =
+      std::variant<const FiniteElementSpace *,
+      const ParFiniteElementSpace *,
+      const QuadratureFunction *,
+      const ParameterSpace *>;
+
+   /// Field ID
+   std::size_t id;
+
+   /// Field variant
+   data_variant_t data;
+
+   /// Default constructor
+   FieldDescriptor() :
+      id(SIZE_MAX), data(data_variant_t{}) {}
+
+   /// Constructor
+   template <typename T>
+   FieldDescriptor(std::size_t field_id, const T* v) :
+      id(field_id), data(v) {}
+
+   bool operator==(const FieldDescriptor& other) const
+   {
+      return id == other.id;
+   }
+
+   bool operator<(const FieldDescriptor& other) const
+   {
+      return id < other.id;
+   }
+
+   friend void swap(FieldDescriptor& a, FieldDescriptor& b)
+   {
+      using std::swap;
+      swap(a.id, b.id);
+      swap(a.data, b.data);
+   }
+};
+
+}
@@ -0,0 +1,22 @@
+#pragma once
+
+#include "util.hpp"
+
+namespace mfem::future
+{
+
+struct IntegratorContext
+{
+   const ParMesh &mesh;
+   const Array<int> *elem_attr;
+   Array<int> attr;
+   int nentities;
+   const std::vector<FieldDescriptor> &infds;
+   const std::vector<FieldDescriptor> &outfds;
+   const std::vector<FieldDescriptor> &unionfds;
+   const IntegrationRule &ir;
+   std::unordered_map<std::type_index, std::vector<int>> &in_qlayouts;
+   std::unordered_map<std::type_index, std::vector<int>> &out_qlayouts;
+};
+
+}
@@ -9,8 +9,23 @@
 // terms of the BSD-3 license. We welcome feedback and contributions, see file
 // CONTRIBUTING.md for details.
 #pragma once
+// #define NVTX_COLOR nvtx::kPeru

 #include "util.hpp"
+#include "fem/kernels.hpp"
+
+///////////////////////////////////////////////////////////////////////////////
+template <class T>
+inline std::enable_if_t<!std::numeric_limits<T>::is_integer, bool>
+AlmostEq(T x, T y, T tolerance = 15.0 * std::numeric_limits<T>::epsilon())
+{
+   const T neg = std::abs(x - y);
+   constexpr T min = std::numeric_limits<T>::min();
+   constexpr T eps = std::numeric_limits<T>::epsilon();
+   const T min_abs = std::min(std::abs(x), std::abs(y));
+   if (std::abs(min_abs) == 0.0) { return neg < eps; }
+   return (neg / (1.0 + std::max(min, min_abs))) < tolerance;
+}

 namespace mfem::future
 {
@@ -30,6 +45,7 @@ void map_field_to_quadrature_data_tensor_product_3d(

   if constexpr (is_value_fop<std::decay_t<field_operator_t>>::value)
   {
+      dbg("Value");
      auto [q1d, unused, d1d] = B.GetShape();
      const int vdim = input.vdim;
      const auto field = Reshape(&field_e[0], d1d, d1d, d1d, vdim);
@@ -94,10 +110,11 @@ void map_field_to_quadrature_data_tensor_product_3d(
   else if constexpr (
      is_gradient_fop<std::decay_t<field_operator_t>>::value)
   {
-      const auto [q1d, unused, d1d] = B.GetShape();
+      // dbg("Gradient");
+      const auto [q1d, B_dim, d1d] = B.GetShape();
      const int vdim = input.vdim;
      const int dim = input.dim;
-      const auto field = Reshape(&field_e[0], d1d, d1d, d1d, vdim);
+      const auto field = Reshape(&std::as_const(field_e[0]), d1d, d1d, d1d, vdim);
      auto fqp = Reshape(&field_qp[0], vdim, dim, q1d, q1d, q1d);

      auto s0 = Reshape(&scratch_mem[0](0), d1d, d1d, q1d);
@@ -106,7 +123,30 @@ void map_field_to_quadrature_data_tensor_product_3d(
      auto s3 = Reshape(&scratch_mem[3](0), d1d, q1d, q1d);
      auto s4 = Reshape(&scratch_mem[4](0), d1d, q1d, q1d);

-      for (int vd = 0; vd < vdim; vd++)
+      // constexpr int MQ1 = T_Q1D > 0 ? T_Q1D : 8;
+      // static constexpr int DIM = 3;
+      // MFEM_VERIFY(q1d <= MQ1, "q1d > MQ1");
+      // MFEM_SHARED real_t smem[MQ1][MQ1];
+
+      // kernels::internal::d_regs3d_t<DIM, MQ1> r0, r1;
+      // real_t sB[MQ1][MQ1], sG[MQ1][MQ1];
+
+      /*
+      {
+         assert(B_dim == 1 && "1D B required!");
+         kernels::internal::LoadMatrix(d1d, q1d, B, sB);
+         kernels::internal::LoadMatrix(d1d, q1d, G, sG);
+         for (int qx = 0; qx < q1d; qx++)
+         {
+            for (int dx = 0; dx < d1d; dx++)
+            {
+               assert(AlmostEq(B(qx, 0, dx), sB[dx][qx]));
+               assert(AlmostEq(G(qx, 0, dx), sG[dx][qx]));
+            }
+         }
+      }*/
+
+      for (int c = 0; c < vdim; c++)
      {
         MFEM_FOREACH_THREAD(dz, z, d1d)
         {
@@ -117,7 +157,7 @@ void map_field_to_quadrature_data_tensor_product_3d(
                  real_t uv[2] = {0.0, 0.0};
                  for (int dx = 0; dx < d1d; dx++)
                  {
-                     const real_t f = field(dx, dy, dz, vd);
+                     const real_t f = field(dx, dy, dz, c);
                     uv[0] += f * B(qx, 0, dx);
                     uv[1] += f * G(qx, 0, dx);
                  }
@@ -163,19 +203,59 @@ void map_field_to_quadrature_data_tensor_product_3d(
                     uvw[1] += s3(dz, qy, qx) * B(qz, 0, dz);
                     uvw[2] += s4(dz, qy, qx) * G(qz, 0, dz);
                  }
-                  fqp(vd, 0, qx, qy, qz) = uvw[0];
-                  fqp(vd, 1, qx, qy, qz) = uvw[1];
-                  fqp(vd, 2, qx, qy, qz) = uvw[2];
+                  fqp(c, 0, qx, qy, qz) = uvw[0];
+                  fqp(c, 1, qx, qy, qz) = uvw[1];
+                  fqp(c, 2, qx, qy, qz) = uvw[2];
               }
            }
         }
         MFEM_SYNC_THREAD;
      }
+
+      /*
+      {
+         for (int c = 0; c < vdim; c++)
+         {
+            kernels::internal::LoadDofs3d(d1d, c, field, r0);
+            for (int d = 0; d < DIM; d++)
+            {
+               for (int dz = 0; dz < d1d; dz++)
+               {
+                  for (int dy = 0; dy < d1d; dy++)
+                  {
+                     for (int dx = 0; dx < d1d; dx++)
+                     {
+                        const real_t f = field(dx, dy, dz, c);
+                        assert(AlmostEq(f, r0[d][dz][dy][dx]));
+                     }
+                  }
+               }
+            }
+
+            kernels::internal::Grad3d(d1d, q1d, smem, sB, sG, r0, r1, c);
+            for (int qz = 0; qz < q1d; qz++)
+            {
+               for (int qy = 0; qy < q1d; qy++)
+               {
+                  for (int qx = 0; qx < q1d; qx++)
+                  {
+                     if (!AlmostEq(fqp(c, d, qx, qy, qz), r1[d][qz][qy][qx]))
+                     {
+                        dbg("\x1b[31m[{}:d] {} {}", c, fqp(c, d, qx, qy, qz), r1[d][qz][qy][qx]);
+                        dbg("❌❌❌"), std::exit(EXIT_FAILURE);
+                     }
+                  }
+               }
+            }
+         }
+         // dbg("✅✅✅✅✅✅✅✅✅✅✅✅✅✅✅");//, std::exit(EXIT_SUCCESS);
+      }*/
   }
   // TODO: Create separate function for clarity
   else if constexpr (
      std::is_same_v<std::decay_t<field_operator_t>, Weight>)
   {
+      // dbg("None");
      const int num_qp = integration_weights.GetShape()[0];
      // TODO: eeek
      const int q1d = (int)floor(std::pow(num_qp, 1.0/input.dim) + 0.5);
@@ -518,6 +598,9 @@ void map_fields_to_quadrature_data(
   const int &dimension,
   const bool &use_sum_factorization = false)
 {
+   // dbg();
+   assert(use_sum_factorization && "❌ use_sum_factorization required");
+
   // When the input_to_field map returns -1, this means the requested input
   // is the integration weight. Weights don't have a user defined field
   // attached to them and we create a dummy field which is not accessed
@@ -578,6 +661,7 @@ void map_field_to_quadrature_data_conditional(
   const int &dimension,
   const bool &use_sum_factorization = false)
 {
+   assert(false && "❌ condition not implemented");
   if (condition)
   {
      if (use_sum_factorization)
@@ -619,6 +703,7 @@ void map_fields_to_quadrature_data_conditional(
   const std::array<bool, num_inputs> &conditions,
   const bool &use_sum_factorization = false)
 {
+   assert(false && "❌ condition not implemented");
   for_constexpr<num_inputs>([&](auto i)
   {
      map_field_to_quadrature_data_conditional(
@@ -627,7 +712,7 @@ void map_fields_to_quadrature_data_conditional(
   });
 }

-template <size_t num_inputs, typename field_operator_ts>
+template <int T_Q1D, size_t num_inputs, typename field_operator_ts>
 MFEM_HOST_DEVICE
 void map_direction_to_quadrature_data_conditional(
   std::array<DeviceTensor<2>, num_inputs> &directions_qp,
@@ -660,7 +745,7 @@ void map_direction_to_quadrature_data_conditional(
            }
            else if (dimension == 3)
            {
-               map_field_to_quadrature_data_tensor_product_3d(
+               map_field_to_quadrature_data_tensor_product_3d<T_Q1D>(
                  directions_qp[i], dtqmaps[i], direction_e, get<i>(fops),
                  integration_weights, scratch_mem);
            }
@@ -43,7 +43,7 @@ public:
   /// Get spatial dimension
   ///
   /// returns always 1.
-   int Dimension() const
+   constexpr int Dimension() const
   {
      return 1;
   }
@@ -74,11 +74,14 @@ public:
      return elem_restr.get();
   }

+   virtual const Operator* GetB() const = 0;
+
+   virtual const Operator* GetBt() const = 0;
+
 protected:
   int vdim;
   DofToQuad dtq;
-   mutable std::unique_ptr<Operator> prolongation;
-   mutable std::unique_ptr<Operator> elem_restr;
+   mutable std::unique_ptr<Operator> prolongation, elem_restr, B, Bt;
 };

 /// @brief Uniform parameter space
@@ -122,6 +125,18 @@ public:
      return lsize;
   }

+   const Operator* GetB() const override
+   {
+      MFEM_ABORT("UniformParameterSpace does not support GetB");
+      return nullptr;
+   }
+
+   const Operator* GetBt() const override
+   {
+      MFEM_ABORT("UniformParameterSpace does not support GetBt");
+      return nullptr;
+   }
+
 private:
   /// T-vector size
   int tsize;
@@ -243,6 +243,8 @@ void process_qf_arg(
   }
 }

+// const tensor<real_t, DIM>       ∇u
+// const tensor<real_t, DIM, DIM>  D (PA_DATA)
 template <typename arg_type>
 MFEM_HOST_DEVICE inline
 void process_qf_arg(const DeviceTensor<2> &u, arg_type &arg, int qp)
@@ -0,0 +1,76 @@
+// Copyright (c) 2010-2025, 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.
+#pragma once
+
+#include "tuple.hpp"
+#include "../linalg/tensor.hpp"
+
+using namespace mfem::future;
+using mfem::future::tensor;
+
+// Helper to add dimension to tensor type
+template<typename T, int qp>
+struct AddQPDimension;
+
+// Specialization for tensor<real_t, dim>
+template<typename real_t, int dim, int qp>
+struct AddQPDimension<tensor<real_t, dim>, qp>
+{
+   using type = tensor<real_t, dim, qp>;
+};
+
+// Specialization for tensor<real_t, dim, dim>
+template<typename real_t, int dim, int qp>
+struct AddQPDimension<tensor<real_t, dim, dim>, qp>
+{
+   using type = tensor<real_t, dim, dim, qp>;
+};
+
+// Specialization for real_t (transforms to tensor<real_t, qp>)
+template<typename real_t, int qp>
+struct AddQPDimension
+{
+   using type = tensor<real_t, qp>;
+};
+
+// Helper to transform tuple
+template<typename Tuple, int qp>
+struct TransformTupleQP {};
+
+// Specialization for mfem::future::tuple
+template<int qp, typename... Types>
+struct TransformTupleQP<mfem::future::tuple<Types...>, qp>
+{
+   using type = mfem::future::tuple<typename AddQPDimension<Types, qp>::type...>;
+};
+
+template<int qp, typename... Types>
+struct TransformTupleQP<std::tuple<Types...>, qp>
+{
+   using type = std::tuple<typename AddQPDimension<Types, qp>::type...>;
+};
+
+// Function to transform tuple type with qp dimension
+template<int qp, typename qf_param_ts>
+struct add_qp_dimension
+{
+   using type = typename TransformTupleQP<qf_param_ts, qp>::type;
+};
+
+// Helper alias template for cleaner usage
+template<int qp, typename qf_param_ts>
+using add_qp_dimension_t = typename add_qp_dimension<qp, qf_param_ts>::type;
+
+// ...AddDomainIntegrator...
+// {
+//    constexpr int Q1D = 4;
+//    using qf_param_augmentd_ts = add_qp_dimension_t<Q1D, decay_tuple<qf_param_ts>>;
+// }
@@ -21,6 +21,7 @@
 #include <type_traits>
 #include <numeric>
 #include <iomanip>
+#include <typeindex>

 #include "../../general/communication.hpp"
 #include "../../general/forall.hpp"
@@ -28,13 +29,19 @@
 #include "../fe/fe_base.hpp"
 #include "../fespace.hpp"
 #include "../pfespace.hpp"
+#include "../qfunction.hpp"
 #include "../../mesh/mesh.hpp"
 #include "../../linalg/dtensor.hpp"
+#include "../quadinterpolator.hpp"

+#include "fielddescriptor.hpp"
 #include "fieldoperator.hpp"
 #include "parameterspace.hpp"
 #include "tuple.hpp"

+#undef NVTX_COLOR
+#define NVTX_COLOR ::nvtx::kLightBlue
+
 namespace mfem::future
 {

@@ -75,7 +82,7 @@ constexpr void for_constexpr(lambda&& f,
 }

 template <typename lambda>
-constexpr void for_constexpr(lambda&& f, std::integer_sequence<std::size_t>) {}
+constexpr void for_constexpr(lambda&&, std::integer_sequence<std::size_t>) {}

 template <int... n, typename lambda>
 constexpr void for_constexpr(lambda&& f)
@@ -84,7 +91,7 @@ constexpr void for_constexpr(lambda&& f)
 }

 template <typename lambda, typename arg_t>
-constexpr void for_constexpr_with_arg(lambda&& f, arg_t&& arg,
+constexpr void for_constexpr_with_arg(lambda&&, arg_t&&,
                                      std::integer_sequence<std::size_t>)
 {
   // Base case - do nothing for empty sequence
@@ -108,6 +115,16 @@ constexpr void for_constexpr_with_arg(lambda&& f, arg_t&& arg)
                          indices{});
 }

+template <auto start, auto end, auto inc = 1, typename F>
+constexpr void constexpr_for(F&& f)
+{
+   if constexpr (start < end)
+   {
+      f(std::integral_constant<decltype(start), start>());
+      constexpr_for<start + inc, end, inc>(f);
+   }
+}
+
 template <std::size_t I, typename Tuple, std::size_t... Is>
 std::array<bool, sizeof...(Is)>
 make_dependency_array(const Tuple& inputs, std::index_sequence<Is...>)
@@ -444,6 +461,21 @@ struct create_function_signature<output_t (*)(input_ts...)>
   using type = FunctionSignature<output_t(input_ts...)>;
 };

+template <typename...>
+using void_t = void;
+
+template <typename T, typename = void>
+struct get_function_signature
+{
+   using type = typename create_function_signature<T>::type;
+};
+
+template <typename T>
+struct get_function_signature<T, void_t<decltype(&T::operator())>>
+{
+   using type = typename create_function_signature<decltype(&T::operator())>::type;
+};
+
 template <typename T>
 constexpr int GetFieldId()
 {
@@ -538,38 +570,12 @@ auto get_marked_entries(
 /// @param t the tuple to filter fields from.
 /// @returns a tuple containing only the fields with field IDs not equal to -1.
 template <typename... Ts>
-constexpr auto filter_fields(const std::tuple<Ts...>& t)
+constexpr auto filter_fields(const std::tuple<Ts...>&)
 {
   return std::tuple_cat(
             std::conditional_t<Ts::GetFieldId() != -1, std::tuple<Ts>, std::tuple<>> {}...);
 }

-/// @brief FieldDescriptor struct
-///
-/// This struct is used to store information about a field.
-struct FieldDescriptor
-{
-   using data_variant_t =
-      std::variant<const FiniteElementSpace *,
-      const ParFiniteElementSpace *,
-      const ParameterSpace *>;
-
-   /// Field ID
-   std::size_t id;
-
-   /// Field variant
-   data_variant_t data;
-
-   /// Default constructor
-   FieldDescriptor() :
-      id(SIZE_MAX), data(data_variant_t{}) {}
-
-   /// Constructor
-   template <typename T>
-   FieldDescriptor(std::size_t field_id, const T* v) :
-      id(field_id), data(v) {}
-};
-
 namespace dfem
 {
 template <class... T> constexpr bool always_false = false;
@@ -599,7 +605,7 @@ struct ThreadBlocks

 #if defined(MFEM_USE_CUDA_OR_HIP)
 template <typename func_t>
-__global__ void forall_kernel_shmem(func_t f, int n)
+__global__ void forall_kernel_extern_shmem(func_t f, int n)
 {
   int i = blockIdx.x;
   extern __shared__ real_t shmem[];
@@ -608,23 +614,48 @@ __global__ void forall_kernel_shmem(func_t f, int n)
      f(i, shmem);
   }
 }
+template <typename func_t>
+__global__ void forall_kernel_static_smem(func_t f, int n)
+{
+   int i = blockIdx.x;
+   if (i >= n) { return; }
+   f(i, nullptr);
+}
+template <int MAX_THREADS_PER_BLOCK, typename func_t>
+__global__
+MFEM_LAUNCH_BOUNDS(MAX_THREADS_PER_BLOCK)
+static void forall_kernel_static_smem_launch_bounds(func_t f, int n)
+{
+   for (int k = blockIdx.x; k < n; k += gridDim.x) { f(k, nullptr); }
+}
 #endif

-template <typename func_t>
+template </*typename kernel_tag,*/ typename func_t>
 void forall(func_t f,
            const int &N,
-            const ThreadBlocks &blocks,
-            int num_shmem = 0,
+            [[maybe_unused]] const ThreadBlocks &blocks,
+            [[maybe_unused]] int num_shmem = 0,
            real_t *shmem = nullptr)
 {
+   db1();
   if (Device::Allows(Backend::CUDA_MASK) ||
       Device::Allows(Backend::HIP_MASK))
   {
 #if defined(MFEM_USE_CUDA_OR_HIP)
      // int gridsize = (N + Z - 1) / Z;
      int num_bytes = num_shmem * sizeof(decltype(shmem));
+      db1("num_bytes:{}", num_bytes);
+      db1("block: {}x{}x{}", blocks.x, blocks.y, blocks.z);
      dim3 block_size(blocks.x, blocks.y, blocks.z);
-      forall_kernel_shmem<<<N, block_size, num_bytes>>>(f, N);
+      // ForallKernel<kernel_tag>::run<<<N, block_size, num_bytes>>>(f, N);
+      if (num_bytes > 0)
+      {
+         forall_kernel_extern_shmem<<<N, block_size, num_bytes>>>(f, N);
+      }
+      else
+      {
+         forall_kernel_static_smem<<<N, block_size>>>(f, N);
+      }
 #if defined(MFEM_USE_CUDA)
      MFEM_GPU_CHECK(cudaGetLastError());
 #elif defined(MFEM_USE_HIP)
@@ -635,6 +666,7 @@ void forall(func_t f,
   }
   else if (Device::Allows(Backend::CPU_MASK))
   {
+      db1("CPU_MASK");
      MFEM_ASSERT(!((bool)num_shmem != (bool)shmem),
                  "Backend::CPU needs a pre-allocated shared memory block");
      for (int i = 0; i < N; i++)
@@ -648,6 +680,69 @@ void forall(func_t f,
   }
 }

+namespace dfem
+{
+
+template <int MAX_THREADS_PER_BLOCK = 0, typename func_t>
+void forall(func_t f,
+            const int &N,
+            [[maybe_unused]] const ThreadBlocks &blocks,
+            [[maybe_unused]] int num_shmem = 0,
+            real_t *shmem = nullptr)
+{
+   db1();
+   if (Device::Allows(Backend::CUDA_MASK) ||
+       Device::Allows(Backend::HIP_MASK))
+   {
+#if defined(MFEM_USE_CUDA_OR_HIP)
+      int num_bytes = num_shmem * sizeof(decltype(shmem));
+      db1("num_bytes:{}", num_bytes);
+      db1("block: {}x{}x{}", blocks.x, blocks.y, blocks.z);
+      db1("MAX_THREADS_PER_BLOCK:{}", MAX_THREADS_PER_BLOCK);
+      dim3 block_size(blocks.x, blocks.y, blocks.z);
+      if constexpr (MAX_THREADS_PER_BLOCK > 0)
+      {
+         assert(num_bytes == 0);
+         forall_kernel_static_smem_launch_bounds
+         <MAX_THREADS_PER_BLOCK><<<N, block_size>>> (f, N);
+      }
+      else
+      {
+         static_assert(MAX_THREADS_PER_BLOCK == 0);
+         if (num_bytes == 0)
+         {
+            forall_kernel_static_smem<<<N, block_size>>>(f, N);
+         }
+         else
+         {
+            forall_kernel_extern_shmem<<<N, block_size, num_bytes>>>(f, N);
+         }
+      }
+#if defined(MFEM_USE_CUDA)
+      MFEM_GPU_CHECK(cudaGetLastError());
+#elif defined(MFEM_USE_HIP)
+      MFEM_GPU_CHECK(hipGetLastError());
+#endif
+      // MFEM_DEVICE_SYNC; // ⚠️
+#endif
+   }
+   else if (Device::Allows(Backend::CPU_MASK))
+   {
+      db1("CPU_MASK");
+      MFEM_ASSERT(!((bool)num_shmem != (bool)shmem),
+                  "Backend::CPU needs a pre-allocated shared memory block");
+      for (int i = 0; i < N; i++)
+      {
+         f(i, shmem);
+      }
+   }
+   else
+   {
+      MFEM_ABORT("no compute backend available");
+   }
+}
+}
+
 /// @todo To be removed.
 class FDJacobian : public Operator
 {
@@ -671,20 +766,6 @@ public:
                    MPI_COMM_WORLD);
   }

-   Operator& GetGradient(const Vector &x0) const override
-   {
-      x = x0;
-      f.UseDevice(x.UseDevice());
-      xpev.UseDevice(x.UseDevice());
-
-      op.Mult(x, f);
-      const real_t xnorm_local = x.Norml2();
-      MPI_Allreduce(&xnorm_local, &xnorm, 1, MPITypeMap<real_t>::mpi_type, MPI_SUM,
-                    MPI_COMM_WORLD);
-
-      return const_cast<FDJacobian&>(*this);
-   }
-
   void Mult(const Vector &v, Vector &y) const override
   {
      // See [1] for choice of eps.
@@ -739,11 +820,11 @@ public:

 private:
   const Operator &op;
-   mutable Vector x, f;
+   Vector x, f;
   mutable Vector xpev;
   real_t lambda = 1.0e-6;
   real_t fixed_eps;
-   mutable real_t xnorm;
+   real_t xnorm;
 };

 /// @brief Find the index of a field descriptor in a vector of field descriptors.
@@ -786,6 +867,10 @@ int GetVSize(const FieldDescriptor &f)
      {
         return arg->GetVSize();
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         return arg->Size();
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         return arg->GetVSize();
@@ -824,6 +909,10 @@ void GetElementVDofs(const FieldDescriptor &f, int el, Array<int> &vdofs)
      {
         arg->GetElementVDofs(el, vdofs);
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         MFEM_ABORT("internal error");
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         MFEM_ABORT("internal error");
@@ -858,6 +947,10 @@ int GetTrueVSize(const FieldDescriptor &f)
      {
         return arg->GetTrueVSize();
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         return arg->Size();
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         return arg->GetTrueVSize();
@@ -888,6 +981,10 @@ int GetVDim(const FieldDescriptor &f)
      {
         return arg->GetVDim();
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         return arg->GetVDim();
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         return arg->GetVDim();
@@ -923,6 +1020,10 @@ int GetDimension(const FieldDescriptor &f)
            return arg->GetMesh()->Dimension() - 1;
         }
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         return arg->GetSpace()->GetMesh()->Dimension();
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         return arg->Dimension();
@@ -935,6 +1036,36 @@ int GetDimension(const FieldDescriptor &f)
   }, f.data);
 }

+inline
+std::variant<const QuadratureInterpolator *, const Operator *>get_qinterp(
+   const FieldDescriptor &f,
+   const IntegrationRule &ir)
+{
+   return std::visit([&ir](auto && arg) -> const QuadratureInterpolator*
+   {
+      using T = std::decay_t<decltype(arg)>;
+      if constexpr (std::is_same_v<T, const FiniteElementSpace *> ||
+                    std::is_same_v<T, const ParFiniteElementSpace *>)
+      {
+         return arg->GetQuadratureInterpolator(ir);
+      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         // QuadratureFunction doesn't need a QuadratureInterpolator
+         return nullptr;
+      }
+      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
+      {
+         return nullptr;
+      }
+      else
+      {
+         static_assert(dfem::always_false<T>, "internal error");
+      }
+
+      return nullptr; // Unreachable, but avoids compiler warning
+   }, f.data);
+}

 /// @brief Get the prolongation operator for a field descriptor.
 ///
@@ -943,6 +1074,7 @@ int GetDimension(const FieldDescriptor &f)
 inline
 const Operator *get_prolongation(const FieldDescriptor &f)
 {
+   NVTX("get P");
   return std::visit([](auto&& arg) -> const Operator*
   {
      using T = std::decay_t<decltype(arg)>;
@@ -951,6 +1083,10 @@ const Operator *get_prolongation(const FieldDescriptor &f)
      {
         return arg->GetProlongationMatrix();
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         return nullptr;
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         return arg->GetProlongationMatrix();
@@ -973,6 +1109,7 @@ inline
 const Operator *get_element_restriction(const FieldDescriptor &f,
                                        ElementDofOrdering o)
 {
+   NVTX("get ER");
   return std::visit([&o](auto&& arg) -> const Operator*
   {
      using T = std::decay_t<decltype(arg)>;
@@ -981,6 +1118,10 @@ const Operator *get_element_restriction(const FieldDescriptor &f,
      {
         return arg->GetElementRestriction(o);
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         return nullptr;
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         return arg->GetElementRestriction(o);
@@ -1008,6 +1149,7 @@ const Operator *get_face_restriction(const FieldDescriptor &f,
                                     FaceType ft,
                                     L2FaceValues m)
 {
+   NVTX("get FR");
   return std::visit([&o, &ft, &m](auto&& arg) -> const Operator*
   {
      using T = std::decay_t<decltype(arg)>;
@@ -1016,6 +1158,11 @@ const Operator *get_face_restriction(const FieldDescriptor &f,
      {
         return arg->GetFaceRestriction(o, ft, m);
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         // QuadratureFunction does not support face restrictions
+         MFEM_ABORT("internal error");
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         // ParameterSpace does not support face restrictions
@@ -1041,6 +1188,7 @@ inline
 const Operator *get_restriction(const FieldDescriptor &f,
                                const ElementDofOrdering &o)
 {
+   NVTX("get R");
   if constexpr (std::is_same_v<entity_t, Entity::Element>)
   {
      return get_element_restriction(f, o);
@@ -1066,12 +1214,14 @@ inline std::tuple<std::function<void(const Vector&, Vector&)>, int>
 get_restriction_transpose(
   const FieldDescriptor &f,
   const ElementDofOrdering &o,
-   const fop_t &fop)
+   [[maybe_unused]] const fop_t &fop)
 {
+   NVTX("get R^T");
   if constexpr (is_sum_fop<fop_t>::value)
   {
      auto RT = [=](const Vector &v_e, Vector &v_l)
      {
+         NVTX("R^T sum");
         v_l += v_e;
      };
      return std::make_tuple(RT, 1);
@@ -1081,6 +1231,7 @@ get_restriction_transpose(
      const Operator *R = get_restriction<entity_t>(f, o);
      std::function<void(const Vector&, Vector&)> RT = [=](const Vector &x, Vector &y)
      {
+         NVTX("R^T+");
         R->AddMultTranspose(x, y);
      };
      return std::make_tuple(RT, R->Height());
@@ -1100,11 +1251,26 @@ get_restriction_transpose(
 inline
 void prolongation(const FieldDescriptor field, const Vector &x, Vector &field_l)
 {
+   NVTX("P");
   const auto P = get_prolongation(field);
+
+   NVTX_INI("SetSize");
   field_l.SetSize(P->Height());
+   NVTX_END("SetSize");
+
+   NVTX_INI("P->Mult");
   P->Mult(x, field_l);
 }

+inline
+void prolongation_transpose(
+   const FieldDescriptor &field, const Vector &field_l, Vector &x)
+{
+   const auto P = get_prolongation(field);
+   x.SetSize(P->Width());
+   P->MultTranspose(field_l, x);
+}
+
 /// @brief Apply the prolongation operator to a vector of fields.
 ///
 /// x is a long vector containing the data for all fields on tdofs and
@@ -1121,6 +1287,7 @@ void prolongation(const std::array<FieldDescriptor, N> fields,
                  const Vector &x,
                  std::array<Vector, M> &fields_l)
 {
+   NVTX("P");
   int data_offset = 0;
   for (int i = 0; i < N; i++)
   {
@@ -1128,9 +1295,14 @@ void prolongation(const std::array<FieldDescriptor, N> fields,
      const int width = P->Width();
      // const Vector x_i(x.GetData() + data_offset, width);
      const Vector x_i(const_cast<Vector&>(x), data_offset, width);
-      fields_l[i].SetSize(P->Height());

+      NVTX_INI("SetSize");
+      fields_l[i].SetSize(P->Height());
+      NVTX_END("SetSize");
+
+      NVTX_INI("P->Mult");
      P->Mult(x_i, fields_l[i]);
+      NVTX_END("P->Mult");
      data_offset += width;
   }
 }
@@ -1144,20 +1316,259 @@ void prolongation(const std::array<FieldDescriptor, N> fields,
 /// @param fields the array of field descriptors.
 /// @param x the input vector in tdofs.
 /// @param fields_l the array of output vectors in vdofs.
+// inline
+// void prolongation(const std::vector<FieldDescriptor> fields,
+//                   const Vector &x,
+//                   std::vector<Vector> &fields_l)
+// {
+//    int data_offset = 0;
+//    for (std::size_t i = 0; i < fields.size(); i++)
+//    {
+//       const auto P = get_prolongation(fields[i]);
+//       const int width = P->Width();
+//       const Vector x_i(const_cast<Vector&>(x), data_offset, width);
+//       fields_l[i].SetSize(P->Height());
+//       P->Mult(x_i, fields_l[i]);
+//       data_offset += width;
+//    }
+// }
+
 inline
-void prolongation(const std::vector<FieldDescriptor> fields,
-                  const Vector &x,
-                  std::vector<Vector> &fields_l)
+void prolongation(
+   const std::vector<FieldDescriptor> fields,
+   const BlockVector &x,
+   std::vector<Vector *> &x_l)
 {
-   int data_offset = 0;
-   for (std::size_t i = 0; i < fields.size(); i++)
+   MFEM_ASSERT(x.NumBlocks() == static_cast<int>(x_l.size()),
+               "error " << x.NumBlocks() << " vs " << x_l.size());
+   for (int i = 0; i < x.NumBlocks(); i++)
   {
      const auto P = get_prolongation(fields[i]);
-      const int width = P->Width();
-      const Vector x_i(const_cast<Vector&>(x), data_offset, width);
-      fields_l[i].SetSize(P->Height());
-      P->Mult(x_i, fields_l[i]);
-      data_offset += width;
+
+      // If nullptr, assume Identity.
+      if (P == nullptr)
+      {
+         *x_l[i] = x.GetBlock(i);
+      }
+      else
+      {
+         const auto P = get_prolongation(fields[i]);
+         MFEM_ASSERT(P->Width() == x.GetBlock(i).Size(),
+                     "prolongation not applicable to given input data size " <<
+                     P->Width() << " vs " << x.GetBlock(i).Size());
+         MFEM_ASSERT(P->Height() == x_l[i]->Size(),
+                     "prolongation not applicable to given output data size " <<
+                     P->Height() << " vs " << x_l[i]->Size());
+         P->Mult(x.GetBlock(i), *x_l[i]);
+      }
+   }
+}
+
+inline
+void prolongation(
+   const std::vector<FieldDescriptor> fields,
+   const MultiVector &x,
+   std::vector<Vector *> &x_l)
+{
+   MFEM_ASSERT(x.NumBlocks() == static_cast<int>(x_l.size()),
+               "error " << x.NumBlocks() << " vs " << x_l.size());
+   for (int i = 0; i < x.NumBlocks(); i++)
+   {
+      const auto P = get_prolongation(fields[i]);
+
+      // If nullptr, assume Identity.
+      if (P == nullptr)
+      {
+         *x_l[i] = x[i];
+      }
+      else
+      {
+         const auto P = get_prolongation(fields[i]);
+         MFEM_ASSERT(P->Width() == x[i].Size(),
+                     "prolongation not applicable to given input data size " <<
+                     P->Width() << " vs " << x[i].Size());
+         MFEM_ASSERT(P->Height() == x_l[i]->Size(),
+                     "prolongation not applicable to given output data size " <<
+                     P->Height() << " vs " << x_l[i]->Size());
+         P->Mult(x[i], *x_l[i]);
+      }
+   }
+}
+
+inline
+void prolongation_transpose(
+   const std::vector<FieldDescriptor> fields,
+   const std::vector<Vector *> &x_l,
+   BlockVector &x)
+{
+   MFEM_ASSERT(static_cast<int>(x_l.size()) == x.NumBlocks(),
+               "error " << x_l.size() << " vs " << x.NumBlocks());
+   for (size_t i = 0; i < x_l.size(); i++)
+   {
+      const auto P = get_prolongation(fields[i]);
+
+      // If nullptr, assume Identity.
+      if (P == nullptr)
+      {
+         x.GetBlock(i) = *x_l[i];
+      }
+      else
+      {
+         MFEM_ASSERT(P->Height() == x_l[i]->Size(),
+                     "prolongation not applicable to given input data size " <<
+                     P->Height() << " vs " << x_l[i]->Size());
+         MFEM_ASSERT(P->Width() == x.GetBlock(i).Size(),
+                     "prolongation not applicable to given output data size " <<
+                     P->Width() << " vs " << x.GetBlock(i).Size());
+         P->MultTranspose(*x_l[i], x.GetBlock(i));
+      }
+   }
+}
+
+inline
+void prolongation_transpose(
+   const std::vector<FieldDescriptor> fields,
+   const std::vector<Vector *> &x_l,
+   MultiVector &x)
+{
+   MFEM_ASSERT(static_cast<int>(x_l.size()) == x.NumBlocks(),
+               "error " << x_l.size() << " vs " << x.NumBlocks());
+   for (size_t i = 0; i < x_l.size(); i++)
+   {
+      const auto P = get_prolongation(fields[i]);
+
+      // If nullptr, assume Identity.
+      if (P == nullptr)
+      {
+         x[i] = *x_l[i];
+      }
+      else
+      {
+         MFEM_ASSERT(P->Height() == x_l[i]->Size(),
+                     "prolongation not applicable to given input data size " <<
+                     P->Height() << " vs " << x_l[i]->Size());
+         MFEM_ASSERT(P->Width() == x[i].Size(),
+                     "prolongation not applicable to given output data size " <<
+                     P->Width() << " vs " << x[i].Size());
+         P->MultTranspose(*x_l[i], x[i]);
+      }
+   }
+}
+
+template <typename entity_t>
+void restriction(
+   const std::vector<FieldDescriptor> fields,
+   const std::vector<Vector *> &x_l,
+   std::vector<Vector *> &x_e)
+{
+   MFEM_ASSERT(x_l.size() == x_e.size(),
+               "internal error " << x_l.size() << " vs " << x_e.size());
+   for (size_t i = 0; i < fields.size(); i++)
+   {
+      int s = 0;
+      const auto R = get_restriction<entity_t>(
+                        fields[i], ElementDofOrdering::LEXICOGRAPHIC);
+
+      // If nullptr, assume Identity.
+      if (R == nullptr)
+      {
+         s = x_l[i]->Size();
+      }
+      else
+      {
+         s = R->Height();
+      }
+
+      // TODO
+      if (x_e[i] == nullptr)
+      {
+         x_e[i] = new Vector(s);
+      }
+      x_e[i]->SetSize(s);
+
+      if (R == nullptr)
+      {
+         x_e[i] = x_l[i];
+      }
+      else
+      {
+         MFEM_ASSERT(R->Width() == x_l[i]->Size(),
+                     "restriction not applicable to given input data size " <<
+                     R->Width() << " vs " << x_l[i]->Size());
+         R->Mult(*x_l[i], *x_e[i]);
+      }
+   }
+}
+
+template <typename entity_t>
+void prepare_residual(
+   const std::vector<FieldDescriptor> &fields,
+   std::vector<Vector *> &r_e)
+{
+   for (size_t i = 0; i < fields.size(); i++)
+   {
+      int s = 0;
+      if (std::holds_alternative<const QuadratureFunction *>(fields[i].data))
+      {
+         const auto fd = std::get<const QuadratureFunction *>(fields[i].data);
+         s = fd->Size();
+      }
+      else
+      {
+         const auto R = get_restriction<entity_t>(
+                           fields[i], ElementDofOrdering::LEXICOGRAPHIC);
+         s = R->Height();
+      }
+
+      // TODO
+      if (r_e[i] == nullptr)
+      {
+         r_e[i] = new Vector(s);
+      }
+      else
+      {
+         r_e[i]->SetSize(s);
+      }
+   }
+}
+
+template <typename entity_t>
+void restriction_transpose(
+   const std::vector<FieldDescriptor> &fields,
+   const std::vector<Vector *> &x_e,
+   std::vector<Vector *> &x_l)
+{
+   for (size_t i = 0; i < fields.size(); i++)
+   {
+      int s = 0;
+      const auto R = get_restriction<entity_t>(
+                        fields[i], ElementDofOrdering::LEXICOGRAPHIC);
+      // TODO: if nullptr, assume Identity
+      if (R == nullptr)
+      {
+         s = x_e[i]->Size();
+      }
+      else
+      {
+         s = R->Width();
+      }
+
+      // TODO
+      if (x_l[i] == nullptr)
+      {
+         x_l[i] = new Vector(s);
+      }
+      x_l[i]->SetSize(s);
+
+      // TODO: if nullptr, assume Identity
+      if (R == nullptr)
+      {
+         x_l[i] = x_e[i];
+      }
+      else
+      {
+         R->MultTranspose(*x_e[i], *x_l[i]);
+      }
   }
 }

@@ -1166,6 +1577,7 @@ void get_lvectors(const std::vector<FieldDescriptor> fields,
                  const Vector &x,
                  std::vector<Vector> &fields_l)
 {
+   NVTX("get_lvectors");
   int data_offset = 0;
   for (std::size_t i = 0; i < fields.size(); i++)
   {
@@ -1192,13 +1604,15 @@ template <typename fop_t>
 inline
 std::function<void(const Vector&, Vector&)> get_prolongation_transpose(
   const FieldDescriptor &f,
-   const fop_t &fop,
+   [[maybe_unused]] const fop_t &fop,
   MPI_Comm mpi_comm)
 {
+   NVTX("get P^T");
   if constexpr (is_sum_fop<fop_t>::value)
   {
      auto PT = [=](const Vector &r_local, Vector &y)
      {
+         NVTX("P^T sum");
         MFEM_ASSERT(y.Size() == 1, "output size doesn't match kernel description");
         real_t local_sum = r_local.Sum();
         MPI_Allreduce(&local_sum, y.GetData(), 1, MPI_DOUBLE, MPI_SUM, mpi_comm);
@@ -1209,6 +1623,7 @@ std::function<void(const Vector&, Vector&)> get_prolongation_transpose(
   {
      auto PT = [=](const Vector &r_local, Vector &y)
      {
+         NVTX("P^T Identity");
         y = r_local;
      };
      return PT;
@@ -1216,6 +1631,7 @@ std::function<void(const Vector&, Vector&)> get_prolongation_transpose(
   const Operator *P = get_prolongation(f);
   auto PT = [=](const Vector &r_local, Vector &y)
   {
+      NVTX("P^T");
      P->MultTranspose(r_local, y);
   };
   return PT;
@@ -1234,12 +1650,19 @@ void restriction(const FieldDescriptor u,
                 Vector &field_e,
                 ElementDofOrdering ordering)
 {
+   NVTX("R");
   const auto R = get_restriction<entity_t>(u, ordering);
   MFEM_ASSERT(R->Width() == u_l.Size(),
               "restriction not applicable to given data size");
   const int height = R->Height();
+
+   NVTX_INI("SetSize");
   field_e.SetSize(height);
+   NVTX_END("SetSize");
+
+   NVTX_INI("R->Mult");
   R->Mult(u_l, field_e);
+   NVTX_END("R->Mult");
 }

 /// @brief Apply the restriction operator to a vector of fields.
@@ -1257,14 +1680,29 @@ void restriction(const std::vector<FieldDescriptor> u,
                 ElementDofOrdering ordering,
                 const int offset = 0)
 {
+   NVTX("R");
   for (std::size_t i = 0; i < u.size(); i++)
   {
      const auto R = get_restriction<entity_t>(u[i], ordering);
      MFEM_ASSERT(R->Width() == u_l[i].Size(),
                  "restriction not applicable to given data size");
      const int height = R->Height();
+
+      // NVTX_INI("SetSize");
      fields_e[i + offset].SetSize(height);
-      R->Mult(u_l[i], fields_e[i + offset]);
+      // NVTX_END("SetSize");
+
+      // NVTX_INI("R->Mult");
+      if (dynamic_cast<const IdentityOperator*>(R))
+      {
+         NVTX("Identity");
+         fields_e[i + offset].NewMemoryAndSize(u_l[i].GetMemory(), u_l[i].Size(), false);
+      }
+      else
+      {
+         R->Mult(u_l[i], fields_e[i + offset]);
+      }
+      // NVTX_END("R->Mult");
   }
 }

@@ -1276,14 +1714,21 @@ void element_restriction(const std::array<FieldDescriptor, N> u,
                         ElementDofOrdering ordering,
                         const int offset = 0)
 {
+   NVTX("ER");
   for (int i = 0; i < N; i++)
   {
      const auto R = get_element_restriction(u[i], ordering);
      MFEM_ASSERT(R->Width() == u_l[i].Size(),
                  "element restriction not applicable to given data size");
      const int height = R->Height();
+
+      NVTX_INI("SetSize");
      fields_e[i + offset].SetSize(height);
+      NVTX_END("SetSize");
+
+      NVTX_INI("R->Mult");
      R->Mult(u_l[i], fields_e[i + offset]);
+      NVTX_END("R->Mult");
   }
 }

@@ -1340,6 +1785,10 @@ const DofToQuad *GetDofToQuad(const FieldDescriptor &f,
            return &arg->GetTypicalTraceElement()->GetDofToQuad(ir, mode);
         }
      }
+      else if constexpr (std::is_same_v<T, const QuadratureFunction *>)
+      {
+         return nullptr;
+      }
      else if constexpr (std::is_same_v<T, const ParameterSpace *>)
      {
         return &arg->GetDofToQuad();
@@ -1471,7 +1920,7 @@ create_descriptors_to_fields_map(

   auto f = [&](auto &fop, auto &map)
   {
-      if constexpr (std::is_same_v<std::decay_t<decltype(fop)>, Weight>)
+      if constexpr (is_weight_fop<std::decay_t<decltype(fop)>>::value)
      {
         // TODO-bug: stealing dimension from the first field
         fop.dim = GetDimension<entity_t>(fields[0]);
@@ -1601,7 +2050,7 @@ get_shmem_info(
   const std::array<DofToQuadMap, num_outputs> &output_dtq_maps,
   const std::vector<FieldDescriptor> &fields,
   const int &num_entities,
-   const input_t &inputs,
+   [[maybe_unused]] const input_t &inputs,
   const int &num_qp,
   const std::vector<int> &input_size_on_qp,
   const int &residual_size_on_qp,
@@ -2356,5 +2805,25 @@ std::array<DofToQuadMap, num_fields> create_dtq_maps(
             std::make_index_sequence<num_fields> {});
 }

+struct QLayoutEntry
+{
+   std::type_index type;
+   std::vector<int> layout;
+
+   template <class Fop>
+   QLayoutEntry(Fop, std::initializer_list<int> idx) :
+      type(typeid(Fop)), layout(idx) {}
+};
+
+static void ExtractQLayouts(
+   const std::initializer_list<QLayoutEntry> entries,
+   std::unordered_map<std::type_index, std::vector<int>>& out)
+{
+   for (const auto& e : entries)
+   {
+      out[e.type] = e.layout;
+   }
+}
+
 } // namespace mfem::future
 #endif
@@ -57,7 +57,7 @@ void DGMassApply(const int e,
      }
      else if (DIM == 3)
      {
-         SmemPAMassApply3D_Element<TD1D,TQ1D,NBZ,ACCUM>(e, NE, B, pa_data, x, y);
+         SmemPAMassApply3D_Element<TD1D,TQ1D,ACCUM>(e, NE, B, pa_data, x, y);
      }
      else
      {
@@ -167,15 +167,7 @@ public:
      /** @brief Full multidimensional representation which does not use tensor
          product structure. The ordering of the degrees of freedom is the
          same as TENSOR, but the sizes of B and G are the same as FULL.*/
-      LEXICOGRAPHIC_FULL,
-
-      /** @brief Ragged tensor product representation using 1D matrices/tensors
-          with dimensions using 1D number of quadrature points and ragged tensor degrees of
-          freedom. */
-      /** Used only for partial assembly of the H1 positive basis. The
-          size of B is d1d x qnpt x dim. Since different Gauss-Jacobi quadrature rules
-          are employed in each dimension, we need to store dim arrays. */
-      RAGGED_TENSOR
+      LEXICOGRAPHIC_FULL
   };

   /// Describes the contents of the #B, #Bt, #G, and #Gt arrays, see #Mode.
@@ -236,39 +228,6 @@ public:
      const Array<DofToQuad*> &dof2quad_array,
      const IntegrationRule &ir,
      DofToQuad::Mode mode);
-
-   virtual ~DofToQuad() = default;
-};
-
-/** @brief Structure representing the matrices/tensors needed to evaluate (in
-    reference space) the values, gradients, divergences, or curls of a positive
-    FiniteElement on simplices at the quadrature points of Stroud conical quadrature. */
-class RaggedDofToQuad : public DofToQuad
-{
-public:
-   /** @brief Special basis function structures for positive (Bernstein) basis with
-      partial assembly. The storage layout of Ba1 is ndof x nqpt for scalar elements.
-      The storage layout of Ba2 is ndof x ndof x nqpt. In particular, we have
-            Ba2(iqpt, a1, a2) = B^{p-a1}_{a2}(x_{iqpt}). */
-   Array<real_t> Ba1, Ba2, Ba3;
-   Array<real_t> Ba1t, Ba2t, Ba3t;
-
-   /** @brief Special structures for gradients of positive basis with partial assembly.
-      The gradient arrays exploit properties of the Bernstein basis which allow grad(B^p_alpha)
-      to be expressed as the sum of products of B^{p-1}_alpha and the barycentric coordinates.
-      Thus, Ga1 and Ga2 simply contain the ragged tensor product components of B^{p-1}_alpha */
-   Array<real_t> Ga1, Ga2, Ga3;
-   Array<real_t> Ga1t, Ga2t, Ga3t;
-
-   /** @brief Mapping from the Bernstein multi-index (a_1, ..., a_d) to the lexicographic
-       dof index. */
-   Array<int> lex_map;
-
-   Array<int> forward_map2d_diff, forward_map3d_diff;
-   Array<int> inverse_map2d_diff, inverse_map3d_diff;
-
-   Array<int> forward_map2d_mass, forward_map3d_mass;
-   Array<int> inverse_map2d_mass, inverse_map3d_mass;
 };

 /// Describes the function space on each element
@@ -557,101 +557,6 @@ H1Pos_TriangleElement::H1Pos_TriangleElement(const int p)
      }
 }

-const DofToQuad &H1Pos_TriangleElement::GetRaggedTensorDofToQuad(
-   const FiniteElement &fe, const IntegrationRule &ir,
-   DofToQuad::Mode mode,
-   Array<DofToQuad*> &dof2quad_array)
-{
-   DofToQuad *d2q = nullptr;
-   MFEM_VERIFY(mode == DofToQuad::RAGGED_TENSOR, "invalid mode requested");
-
-#if defined(MFEM_THREAD_SAFE) && defined(MFEM_USE_OPENMP)
-   #pragma omp critical (DofToQuad)
-#endif
-   {
-      for (int i = 0; i < dof2quad_array.Size(); i++)
-      {
-         d2q = dof2quad_array[i];
-         if (d2q->IntRule != &ir || d2q->mode != mode) { d2q = nullptr; }
-      }
-      if (!d2q)
-      {
-         d2q = new RaggedDofToQuad;
-         const int ndof = fe.GetOrder() + 1; // verify
-         const int nqpt = (int)floor(pow(ir.GetNPoints(), 1.0/fe.GetDim()) + 0.5);
-         d2q->FE = &fe;
-         d2q->IntRule = &ir;
-         d2q->mode = mode;
-         d2q->ndof = ndof;
-         d2q->nqpt = nqpt;
-
-         RaggedDofToQuad *rd2q = static_cast<RaggedDofToQuad*>(d2q);
-         rd2q->Ba1.SetSize(nqpt*ndof);
-         // second component of ragged tensor basis, technically dof*(dof-1)/2 entries
-         rd2q->Ba2.SetSize((int)nqpt*ndof*ndof);
-         rd2q->Ba1t.SetSize(nqpt*ndof);
-         rd2q->Ba2t.SetSize((int)nqpt*ndof*ndof);
-         // stores first component of ragged tensor basis with order p-1, for gradients only
-         rd2q->Ga1.SetSize(nqpt*(ndof -1));
-         // stores second component of ragged tensor basis with order p-1
-         rd2q->Ga2.SetSize(nqpt*(ndof-1)*(ndof -1));
-         rd2q->Ga1t.SetSize(nqpt*(ndof -1));
-         rd2q->Ga2t.SetSize(nqpt*(ndof-1)*(ndof -1));
-         rd2q->lex_map.SetSize(ndof * ndof);
-         Vector shape_a1(ndof), shape_a2(ndof * ndof);
-         Vector shape_Ga1(ndof-1), shape_Ga2((ndof-1) * (ndof-1));
-         for (int i = 0; i < nqpt; i++)
-         {
-            // The first 'nqpt' points in the first dimension 'ir' have the same x-coordinates as those
-            // of the 1D rule (ie. (2,0) Gauss-Jacobi rule). The first 'nqpt' points in the second dimension
-            // 'ir' have the same y-coordinates as those of the 1D rule for second dimension (i.e. (1,0)
-            // Gauss-Jacobi rule). Additionally, the Bernstein PA algorithms expect evaluation of the
-            // component 1D bases at the Stroud nodes pulled back to the unit square, so perform the pullback
-            // on the fly.
-            const real_t x = ir.IntPoint(i).x;
-            const real_t y = ir.IntPoint(nqpt*i).y / (1.0 - ir.IntPoint(nqpt*i).x);
-
-            Poly_1D::CalcBernstein(ndof-1, x, shape_a1);
-            Poly_1D::CalcBernstein(ndof-2, x, shape_Ga1);
-            for (int j = 0; j < ndof; j++)
-            {
-               rd2q->Ba1t[i+nqpt*j] = rd2q->Ba1[j+ndof*i] = shape_a1(j);
-               if (j < ndof-1)
-               {
-                  rd2q->Ga1t[i+nqpt*j] = rd2q->Ga1[j+(ndof-1)*i] = shape_Ga1(j);
-                  Poly_1D::CalcBernstein(ndof-2-j, y, shape_Ga2);
-               }
-
-               Poly_1D::CalcBernstein(ndof-1-j, y, shape_a2);
-               for (int k = 0; k < ndof-j; k++)
-               {
-                  rd2q->Ba2t[i + nqpt*(j + ndof*k)] = rd2q->Ba2[k + ndof*(j + ndof*i)] = shape_a2(
-                                                                                            k);
-                  if (j < ndof-1 && k < ndof-j-1)
-                  {
-                     rd2q->Ga2t[i + nqpt*(j + (ndof-1)*k)] = rd2q->Ga2[k + (ndof-1)*(j +
-                                                                                     (ndof-1)*i)] = shape_Ga2(k);
-                  }
-               }
-            }
-         }
-
-         // stores the mapping from 2D Bernstein multi-index (i,j,p-i-j) to the
-         // lexicographic DOF ordering
-         for (int i = 0; i < ndof; i++)
-         {
-            for (int j = 0; j < ndof-i; j++)
-            {
-               int idx = ((2 * (ndof-1) + 3) - j) * j / 2 + i;
-               rd2q->lex_map[j + ndof*i] = idx;
-            }
-         }
-         dof2quad_array.Append(d2q);
-      }
-   }
-   return *d2q;
-}
-
 // static method
 void H1Pos_TriangleElement::CalcShape(
   const int p, const real_t l1, const real_t l2, real_t *shape)
@@ -844,213 +749,6 @@ H1Pos_TetrahedronElement::H1Pos_TetrahedronElement(const int p)
         }
 }

-const DofToQuad &H1Pos_TetrahedronElement::GetRaggedTensorDofToQuad(
-   const FiniteElement &fe, const IntegrationRule &ir,
-   DofToQuad::Mode mode,
-   Array<DofToQuad*> &dof2quad_array)
-{
-   DofToQuad *d2q = nullptr;
-   MFEM_VERIFY(mode == DofToQuad::RAGGED_TENSOR, "invalid mode requested");
-
-#if defined(MFEM_THREAD_SAFE) && defined(MFEM_USE_OPENMP)
-   #pragma omp critical (DofToQuad)
-#endif
-   {
-      for (int i = 0; i < dof2quad_array.Size(); i++)
-      {
-         d2q = dof2quad_array[i];
-         if (d2q->IntRule != &ir || d2q->mode != mode) { d2q = nullptr; }
-      }
-      if (!d2q)
-      {
-         d2q = new RaggedDofToQuad;
-         const int ndof = fe.GetOrder() + 1; // verify
-         const int nqpt = (int)floor(pow(ir.GetNPoints(), 1.0/fe.GetDim()) + 0.5);
-         const int basis_dim2d = ndof*(ndof+1) / 2;
-         const int basis_dim3d = ndof*(ndof+1)*(ndof+2) / 6;
-         const int basis_dim2d_diff = (ndof-1)*(ndof) / 2;
-         const int basis_dim3d_diff = (ndof-1)*(ndof)*(ndof+1) / 6;
-         d2q->FE = &fe;
-         d2q->IntRule = &ir;
-         d2q->mode = mode;
-         d2q->ndof = ndof;
-         d2q->nqpt = nqpt;
-
-         RaggedDofToQuad *rd2q = static_cast<RaggedDofToQuad*>(d2q);
-         rd2q->Ba1.SetSize(nqpt * ndof);
-         // second component of ragged tensor basis, technically dof*(dof-1)/2 entries
-         rd2q->Ba2.SetSize(nqpt * basis_dim2d);
-         // third component of ragged tensor basis, technically dof*(dof-1)/2 entries
-         rd2q->Ba3.SetSize(nqpt * basis_dim3d);
-         rd2q->Ba1t.SetSize(nqpt * ndof);
-         rd2q->Ba2t.SetSize(nqpt * basis_dim2d);
-         rd2q->Ba3t.SetSize(nqpt * basis_dim3d);
-         // stores first component of ragged tensor basis with order p-1, for gradients only
-         rd2q->Ga1.SetSize(nqpt * (ndof-1));
-         // stores second component of ragged tensor basis with order p-1
-         rd2q->Ga2.SetSize(nqpt * basis_dim2d_diff);
-         // stores third component of ragged tensor basis with order p-1
-         rd2q->Ga3.SetSize(nqpt * basis_dim3d_diff);
-         rd2q->Ga1t.SetSize(nqpt * (ndof-1));
-         rd2q->Ga2t.SetSize(nqpt * basis_dim2d_diff);
-         rd2q->Ga3t.SetSize(nqpt * basis_dim3d_diff);
-         rd2q->lex_map.SetSize(ndof * ndof * ndof);
-
-         rd2q->forward_map2d_diff.SetSize((ndof-1) * (ndof-1));
-         rd2q->forward_map3d_diff.SetSize((ndof-1) * (ndof-1) * (ndof-1));
-         rd2q->inverse_map2d_diff.SetSize(2 * basis_dim2d_diff);
-         rd2q->inverse_map3d_diff.SetSize(3 * basis_dim3d_diff);
-
-         rd2q->forward_map2d_mass.SetSize(ndof * ndof);
-         rd2q->forward_map3d_mass.SetSize(ndof * ndof * ndof);
-         rd2q->inverse_map2d_mass.SetSize(2 * basis_dim2d);
-         rd2q->inverse_map3d_mass.SetSize(2 * basis_dim3d);
-
-         // forward and inverse maps for multi-index to collpased 1d index for diffusion, can combine
-         // these four loops, but need four idx's and clause for shorter diff loops
-         int idx = 0;
-         for (int i = 0; i < ndof-1; i++)
-         {
-            for (int j = 0; j < ndof-i-1; j++)
-            {
-               rd2q->forward_map2d_diff[j + (ndof-1)*i] = idx;
-               rd2q->inverse_map2d_diff[2*idx] = i;
-               rd2q->inverse_map2d_diff[1 + 2*idx] = j;
-               idx++;
-            }
-         }
-
-         idx = 0;
-         for (int k = 0; k < ndof-1; k++)
-         {
-            for (int j = 0; j < ndof-k-1; j++)
-            {
-               for (int i = 0; i < ndof-k-j-1; i++)
-               {
-                  rd2q->forward_map3d_diff[k + (ndof-1)*(j + (ndof-1)*i)] = idx;
-                  rd2q->inverse_map3d_diff[3*idx] = i;
-                  rd2q->inverse_map3d_diff[1 + 3*idx] = j;
-                  rd2q->inverse_map3d_diff[2 + 3*idx] = k;
-                  idx++;
-               }
-            }
-         }
-
-         // forward and inverse maps for multi-index to collpased 1d index for mass
-         idx = 0;
-         for (int j = 0; j < ndof; j++)
-         {
-            for (int i = 0; i < ndof-j; i++)
-            {
-               rd2q->forward_map2d_mass[j + ndof*i] = idx;
-               rd2q->inverse_map2d_mass[2*idx] = i;
-               rd2q->inverse_map2d_mass[1 + 2*idx] = j;
-               idx++;
-            }
-         }
-
-         idx = 0;
-         for (int k = 0; k < ndof; k++)
-         {
-            for (int j = 0; j < ndof-k; j++)
-            {
-               for (int i = 0; i < ndof-k-j; i++)
-               {
-                  rd2q->forward_map3d_mass[k + ndof*(j + ndof*i)] = idx;
-                  rd2q->inverse_map3d_mass[2*idx] = i;
-                  rd2q->inverse_map3d_mass[1 + 2*idx] = j;
-                  // d2q->inverse_map3d_mass[2 + 3*idx] = k;
-                  idx++;
-               }
-            }
-         }
-
-         Vector shape_a1(ndof), shape_a2(ndof * ndof), shape_a3(ndof * ndof * ndof);
-         Vector shape_Ga1(ndof-1), shape_Ga2(ndof-1), shape_Ga3(ndof-1);
-         for (int i = 0; i < nqpt; i++)
-         {
-            // The first 'nqpt' points in the first dimension 'ir' have the same x-coordinates as those
-            // of the 1D rule (ie. (2,0) Gauss-Jacobi rule). The first 'nqpt' points in the second dimension
-            // 'ir' have the same y-coordinates as those of the 1D rule for second dimension (i.e. (1,0)
-            // Gauss-Jacobi rule). The first 'nqpt' points in the third dimension have the same z-coordinates
-            // as those of the 1D rule for the third dimension (i.e. Gauss-Legendre rule). Additionally,
-            // the Bernstein PA algorithms expect evaluation of the component 1D bases at the Stroud nodes
-            // pulled back to the unit cube, so perform the pullback on the fly.
-            const real_t x = ir.IntPoint(i).x;
-            const real_t y = ir.IntPoint(nqpt*i).y / (1.0 - ir.IntPoint(nqpt*i).x);
-            const real_t z = ir.IntPoint(nqpt*nqpt*i).z / (1.0 - ir.IntPoint(
-                                                              nqpt*nqpt*i).x - ir.IntPoint(nqpt*nqpt*i).y);
-            Poly_1D::CalcBernstein(ndof-1, x, shape_a1);
-            Poly_1D::CalcBernstein(ndof-2, x, shape_Ga1);
-            for (int j = 0; j < ndof; j++)
-            {
-               rd2q->Ba1t[i+nqpt*j] = rd2q->Ba1[j+ndof*i] = shape_a1(j);
-               if (j < ndof-1)
-               {
-                  rd2q->Ga1t[i+nqpt*j] = rd2q->Ga1[j+(ndof-1)*i] = shape_Ga1(j);
-                  Poly_1D::CalcBernstein(ndof-2-j, y, shape_Ga2);
-               }
-
-               Poly_1D::CalcBernstein(ndof-1-j, y, shape_a2);
-               for (int k = 0; k < ndof-j; k++)
-               {
-                  const int a_2d_mass = rd2q->forward_map2d_mass[k + ndof*j];
-                  rd2q->Ba2t[i + nqpt*a_2d_mass] = rd2q->Ba2[a_2d_mass + basis_dim2d*i] =
-                                                      shape_a2(
-                                                         k);
-                  if (j < ndof-1 && k < ndof-j-1)
-                  {
-                     const int a_2d_diff = rd2q->forward_map2d_diff[k + (ndof-1)*j];
-                     rd2q->Ga2t[i + nqpt*a_2d_diff] = rd2q->Ga2[a_2d_diff + basis_dim2d_diff*i] =
-                                                         shape_Ga2(k);
-                     Poly_1D::CalcBernstein(ndof-2-j-k, z, shape_Ga3);
-                  }
-
-                  Poly_1D::CalcBernstein(ndof-1-j-k, z, shape_a3);
-                  for (int m = 0; m < ndof-j-k; m++)
-                  {
-                     const int a_3d_mass = rd2q->forward_map3d_mass[m + ndof*(k + ndof*j)];
-                     rd2q->Ba3t[i + nqpt*a_3d_mass] = rd2q->Ba3[a_3d_mass + basis_dim3d*i] =
-                                                         shape_a3(
-                                                            m);
-                     if (j < ndof-1 && k < ndof-j-1 && m < ndof-j-k-1)
-                     {
-                        // // collapsed 1D access
-                        // d2q->Ga3[i + nqpt*(m + d2q->offset3d[k + (ndof-1)*j])] = shape_Ga3(m);
-                        // collapsed 1D access with forward mapping
-                        const int a_3d_diff = rd2q->forward_map3d_diff[m + (ndof-1)*(k + (ndof-1)*j)];
-                        rd2q->Ga3t[i + nqpt*a_3d_diff] = rd2q->Ga3[a_3d_diff + basis_dim3d_diff*i] =
-                                                            shape_Ga3(m);
-                     }
-                  }
-               }
-            }
-         }
-
-         // stores the mapping from 3D Bernstein multi-index (i,j,k,p-i-j-k) to the
-         // lexicographic DOF ordering
-         int p = ndof - 1;
-         for (int i = 0; i < ndof; i++)
-         {
-            for (int j = 0; j < ndof-i; j++)
-            {
-               for (int k = 0; k < ndof-i-j; k++)
-               {
-                  int dof = (p+1)*(p+2)*(p+3) / 6;
-                  int tet = (p-k)*(p-k+1)*(p-k+2) / 6;
-                  int tri = (p+1-k-j)*(p+2-k-j)/2;
-                  int multi_idx = dof - tet - tri + i;
-                  rd2q->lex_map[k + ndof*(j + ndof*i)] = multi_idx;
-               }
-            }
-         }
-
-         dof2quad_array.Append(d2q);
-      }
-   }
-   return *d2q;
-}
-
 // static method
 void H1Pos_TetrahedronElement::CalcShape(
   const int p, const real_t l1, const real_t l2, const real_t l3,
@@ -191,21 +191,6 @@ public:
   /// Construct the H1Pos_TriangleElement of order @a p
   H1Pos_TriangleElement(const int p);

-   const DofToQuad &GetDofToQuad(const IntegrationRule &ir,
-                                 DofToQuad::Mode mode) const override
-   {
-      return (mode == DofToQuad::RAGGED_TENSOR) ?
-             GetRaggedTensorDofToQuad(*this, ir, mode, dof2quad_array) :
-             FiniteElement::GetDofToQuad(ir, mode);
-   }
-
-   static const DofToQuad &GetRaggedTensorDofToQuad(
-      const FiniteElement &fe, const IntegrationRule &ir,
-      DofToQuad::Mode mode,
-      Array<DofToQuad*> &dof2quad_array);
-
-   const Array<int> &GetDofMap() const { return dof_map; }
-
   // The size of shape is (p+1)(p+2)/2 (dof).
   static void CalcShape(const int p, const real_t x, const real_t y,
                         real_t *shape);
@@ -235,21 +220,6 @@ public:
   /// Construct the H1Pos_TetrahedronElement of order @a p
   H1Pos_TetrahedronElement(const int p);

-   const DofToQuad &GetDofToQuad(const IntegrationRule &ir,
-                                 DofToQuad::Mode mode) const override
-   {
-      return (mode == DofToQuad::RAGGED_TENSOR) ?
-             GetRaggedTensorDofToQuad(*this, ir, mode, dof2quad_array) :
-             FiniteElement::GetDofToQuad(ir, mode);
-   }
-
-   static const DofToQuad &GetRaggedTensorDofToQuad(
-      const FiniteElement &fe, const IntegrationRule &ir,
-      DofToQuad::Mode mode,
-      Array<DofToQuad*> &dof2quad_array);
-
-   const Array<int> &GetDofMap() const { return dof_map; }
-
   // The size of shape is (p+1)(p+2)(p+3)/6 (dof).
   static void CalcShape(const int p, const real_t x, const real_t y,
                         const real_t z, real_t *shape);
@@ -250,14 +250,6 @@ public:
       its GetOrder() method. */
   virtual FiniteElementCollection *Clone(int p) const;

-   /** @brief Return the order parameter used to construct this collection.
-    *  This differs from GetOrder() depending on the collection type. */
-   virtual int GetConstructorOrder() const
-   {
-      MFEM_ABORT("Collection " << Name() << " does not support GetConstructorOrder");
-      return -1;
-   }
-
 protected:
   const int base_p; ///< Order as returned by GetOrder().

@@ -322,9 +314,6 @@ public:
   FiniteElementCollection *Clone(int p) const override
   { return new H1_FECollection(p, dim, b_type); }

-   int GetConstructorOrder() const override
-   { return base_p; }
-
   virtual ~H1_FECollection();
 };

@@ -354,10 +343,6 @@ class H1_Trace_FECollection : public H1_FECollection
 public:
   H1_Trace_FECollection(const int p, const int dim,
                         const int btype = BasisType::GaussLobatto);
-
-   FiniteElementCollection *Clone(int p) const override
-   { return new H1_Trace_FECollection(p, dim+1, b_type); }
-
 };

 /// Arbitrary order "L2-conforming" discontinuous finite elements.
@@ -411,9 +396,6 @@ public:
   FiniteElementCollection *Clone(int p) const override
   { return new L2_FECollection(p, dim, b_type, m_type); }

-   int GetConstructorOrder() const override
-   { return base_p; }
-
   virtual ~L2_FECollection();
 };

@@ -474,9 +456,6 @@ public:
   FiniteElementCollection *Clone(int p) const override
   { return new RT_FECollection(p, dim, cb_type, ob_type); }

-   int GetConstructorOrder() const override
-   { return base_p-1; }
-
   virtual ~RT_FECollection();
 };

@@ -557,9 +536,6 @@ public:
   FiniteElementCollection *Clone(int p) const override
   { return new ND_FECollection(p, dim, cb_type, ob_type); }

-   int GetConstructorOrder() const override
-   { return dim>1 ? base_p : base_p+1; }
-
   virtual ~ND_FECollection();
 };

@@ -572,9 +548,6 @@ public:
   ND_Trace_FECollection(const int p, const int dim,
                         const int cb_type = BasisType::GaussLobatto,
                         const int ob_type = BasisType::GaussLegendre);
-
-   FiniteElementCollection *Clone(int p) const override
-   { return new ND_Trace_FECollection(p, dim+1, cb_type, ob_type); }
 };

 /// Arbitrary order 3D H(curl)-conforming Nedelec finite elements in 1D.
@@ -52,7 +52,7 @@
 #include "bounds.hpp"
 #include "particleset.hpp"

-#include "dfem/doperator.hpp"
+// #include "dfem/doperator.hpp"

 #ifdef MFEM_USE_MPI
 #include "pfespace.hpp"
@@ -4631,8 +4631,9 @@ FiniteElementCollection *FiniteElementSpace::Load(Mesh *m, std::istream &input)

 ElementDofOrdering GetEVectorOrdering(const FiniteElementSpace& fes)
 {
-   return (UsesTensorBasis(fes) || fes.UsesRaggedTensorBasis()) ?
+   return UsesTensorBasis(fes)?
          ElementDofOrdering::LEXICOGRAPHIC:
          ElementDofOrdering::NATIVE;
 }
+
 } // namespace mfem
@@ -1514,18 +1514,6 @@ public:
      return dynamic_cast<const L2_FECollection*>(fec) != NULL;
   }

-   /// @brief Return true if the mesh contains only one topology, the elements are
-   /// all triangles or tetrahedrons, and the elements are ragged tensor elements
-   /// i.e. Bernstein/positive basis.
-   bool UsesRaggedTensorBasis() const
-   {
-      bool simplex = this->GetMesh()->IsSimplexMesh();
-      bool positive =
-         dynamic_cast<const mfem::H1Pos_TriangleElement *>(this->GetTypicalFE()) ||
-         dynamic_cast<const mfem::H1Pos_TetrahedronElement *>(this->GetTypicalFE());
-      return simplex && positive;
-   }
-
   /** In variable-order spaces on nonconforming (NC) meshes, this function
       controls whether strict conformity is enforced in cases where coarse
       edges/faces have higher polynomial order than their fine NC neighbors.
@@ -2256,104 +2256,6 @@ void GridFunction::AccumulateAndCountBdrTangentValues(
   }
 }

-void GridFunction::AccumulateAndCountTraceValues(
-   Coefficient *coeff[], VectorCoefficient *vcoeff,
-   Array<int> &values_counter)
-{
-   if (vcoeff)
-   {
-      MFEM_VERIFY(fes->GetVDim() == vcoeff->GetVDim(),
-                  "vcoeff vdim != fes VDim");
-      MFEM_VERIFY(fes->GetTypicalTraceElement()->GetMapType() ==
-                  FiniteElement::VALUE &&
-                  fes->GetTypicalTraceElement()->GetRangeType() ==
-                  FiniteElement::SCALAR,
-                  "Can only call ProjectTraceCoefficient on scalar value-type "
-                  "trace elements. "
-                  "Use ProjectTraceCoefficientNormal for RT and "
-                  "ProjectTraceCoefficientTangent for ND finite elements.");
-   }
-
-   Array<int> vdofs;
-   Vector vc;
-
-   values_counter.SetSize(Size());
-   values_counter = 0;
-
-   const int vdim = fes->GetVDim();
-   HostReadWrite();
-
-   for (int i = 0; i < fes->GetMesh()->GetNumFaces(); i++)
-   {
-
-      const FiniteElement *fe = fes->GetFaceElement(i);
-      const int fdof = fe->GetDof();
-      ElementTransformation *transf = fes->GetMesh()->GetFaceTransformation(i);
-      const IntegrationRule &ir = fe->GetNodes();
-      fes->GetFaceVDofs(i, vdofs);
-
-      for (int j = 0; j < fdof; j++)
-      {
-         const IntegrationPoint &ip = ir.IntPoint(j);
-         transf->SetIntPoint(&ip);
-         if (vcoeff) { vcoeff->Eval(vc, *transf, ip); }
-         for (int d = 0; d < vdim; d++)
-         {
-            if (!vcoeff && !coeff[d]) { continue; }
-
-            real_t val = vcoeff ? vc(d) : coeff[d]->Eval(*transf, ip);
-            int ind = vdofs[fdof*d+j];
-            if ( ind < 0 )
-            {
-               val = -val, ind = -1-ind;
-            }
-            if (++values_counter[ind] == 1)
-            {
-               (*this)(ind) = val;
-            }
-            else
-            {
-               (*this)(ind) += val;
-            }
-         }
-      }
-   }
-}
-
-void GridFunction::AccumulateAndCountTraceTangentValues(
-   VectorCoefficient &vcoeff, Array<int> &values_counter)
-{
-   MFEM_VERIFY(fes->GetVDim() == 1, "fespace VDim != 1");
-   MFEM_VERIFY(fes->GetTypicalTraceElement()
-               ->GetRangeType() == FiniteElement::VECTOR &&
-               fes->GetTypicalTraceElement()
-               ->GetMapType() == FiniteElement::H_CURL,
-               "Not an ND FE space!");
-   MFEM_VERIFY(fes->GetTypicalTraceElement()->GetPhysRangeDim(
-                  fes->GetMesh()->SpaceDimension()) == vcoeff.GetVDim(),
-               "vcoeff vdim != PhysRangeDim");
-
-   const FiniteElement *fe;
-   ElementTransformation *T;
-   Array<int> dofs;
-   Vector lvec;
-
-   values_counter.SetSize(Size());
-   values_counter = 0;
-
-   HostReadWrite();
-
-   for (int i = 0; i < fes->GetMesh()->GetNumFaces(); i++)
-   {
-      fe = fes->GetFaceElement(i);
-      T = fes->GetMesh()->GetFaceTransformation(i);
-      fes->GetFaceVDofs(i, dofs);
-      lvec.SetSize(fe->GetDof());
-      fe->Project(vcoeff, *T, lvec);
-      accumulate_dofs(dofs, lvec, *this, values_counter);
-   }
-}
-
 void GridFunction::ComputeMeans(AvgType type, Array<int> &zones_per_vdof)
 {
   switch (type)
@@ -2796,74 +2698,6 @@ void GridFunction::ProjectCoefficient(VectorCoefficient &vcoeff,
   }
 }

-void GridFunction::ProjectTraceCoefficient(Coefficient *coeff[])
-{
-   Array<int> values_counter;
-   AccumulateAndCountTraceValues(coeff, NULL, values_counter);
-   ComputeMeans(ARITHMETIC, values_counter);
-}
-
-void GridFunction::ProjectTraceCoefficient(Coefficient &coeff)
-{
-   MFEM_VERIFY(FESpace()->GetVDim() == 1, "ProjectTraceCoefficient(Coefficient&)"
-               "is only valid for scalar GridFunction");
-   Coefficient *coeff_p = &coeff;
-   ProjectTraceCoefficient(&coeff_p);
-}
-
-void GridFunction::ProjectTraceCoefficient(VectorCoefficient &vcoeff)
-{
-   MFEM_VERIFY(FESpace()->GetVDim() == vcoeff.GetVDim(),
-               "Incompatible vcoeff vdim and fes vdim");
-   Array<int> values_counter;
-   AccumulateAndCountTraceValues(NULL, &vcoeff, values_counter);
-   ComputeMeans(ARITHMETIC, values_counter);
-}
-
-void GridFunction::ProjectTraceCoefficientNormal(VectorCoefficient &vcoeff)
-{
-   MFEM_VERIFY(fes->GetVDim() == 1, "fespace VDim != 1");
-   MFEM_VERIFY(fes->GetTypicalTraceElement()->GetRangeType() ==
-               FiniteElement::SCALAR &&
-               fes->GetTypicalTraceElement()->GetMapType() ==
-               FiniteElement::INTEGRAL, "Not an RT FE space!");
-   MFEM_VERIFY(vcoeff.GetVDim() == fes->GetMesh()->SpaceDimension(),
-               "vcoeff vdim (" << vcoeff.GetVDim()
-               << ") != SpaceDimension ("
-               << fes->GetMesh()->SpaceDimension() << ")");
-
-   const FiniteElement *fe;
-   ElementTransformation *T;
-   Array<int> dofs;
-   int dim = vcoeff.GetVDim();
-   Vector vc(dim), nor(dim), lvec;
-
-   for (int i = 0; i < fes->GetMesh()->GetNumFaces(); i++)
-   {
-      fe = fes->GetFaceElement(i);
-      T = fes->GetMesh()->GetFaceTransformation(i);
-      const IntegrationRule &ir = fe->GetNodes();
-      lvec.SetSize(fe->GetDof());
-      for (int j = 0; j < ir.GetNPoints(); j++)
-      {
-         const IntegrationPoint &ip = ir.IntPoint(j);
-         T->SetIntPoint(&ip);
-         vcoeff.Eval(vc, *T, ip);
-         CalcOrtho(T->Jacobian(), nor);
-         lvec(j) = (vc * nor);
-      }
-      fes->GetFaceVDofs(i, dofs);
-      SetSubVector(dofs, lvec);
-   }
-}
-
-void GridFunction::ProjectTraceCoefficientTangent(VectorCoefficient &vcoeff)
-{
-   Array<int> values_counter;
-   AccumulateAndCountTraceTangentValues(vcoeff, values_counter);
-   ComputeMeans(ARITHMETIC, values_counter);
-}
-
 void GridFunction::ProjectCoefficientGlobalL2(VectorCoefficient &vcoeff,
                                              real_t rtol, int iter)
 {
@@ -5452,7 +5286,6 @@ PLBound GridFunction::GetBounds(Vector &lower, Vector &upper,
 {
   int max_order = fes->GetMaxElementOrder();
   PLBound plb(fes, ref_factor*(max_order+1));
-
   Vector lel, uel;
   GetElementBounds(plb, lel, uel, vdim);

@@ -578,13 +578,6 @@ protected:
                                           const Array<int> &bdr_attr,
                                           Array<int> &values_counter);

-   void AccumulateAndCountTraceValues(Coefficient *coeff[],
-                                      VectorCoefficient *vcoeff,
-                                      Array<int> &values_counter);
-
-   void AccumulateAndCountTraceTangentValues(VectorCoefficient &vcoeff,
-                                             Array<int> &values_counter);
-
   // Complete the computation of averages; called e.g. after
   // AccumulateAndCountZones().
   void ComputeMeans(AvgType type, Array<int> &zones_per_vdof);
@@ -670,23 +663,6 @@ public:
      ProjectBdrCoefficient(&coeff_p, attr);
   }

-   /// Project a Coefficient on a GridFunction defined on H1 trace space
-   void ProjectTraceCoefficient(Coefficient *coeff[]);
-   void ProjectTraceCoefficient(Coefficient &coeff);
-
-   /** @brief Project a VectorCoefficient @a vcoeff on a GridFunction
-       defined on a Vector H1 trace space. Note that this also works
-       for a scalar H1 trace space, where only the first component of
-       @a vcoeff is used. */
-   void ProjectTraceCoefficient(VectorCoefficient &vcoeff);
-   /** @brief Project a VectorCoefficient on a GridFunction
-       defined on an RT trace space */
-   void ProjectTraceCoefficientNormal(VectorCoefficient &vcoeff);
-   /** @brief Project a VectorCoefficient on a GridFunction
-       defined on an ND trace space */
-   void ProjectTraceCoefficientTangent(VectorCoefficient &vcoeff);
-
-
   /** @brief Project a VectorCoefficient on the GridFunction, modifying only
       DOFs on the boundary associated with the boundary attributes marked in
       the @a attr array. */
@@ -1791,8 +1767,8 @@ public:
                             const int ref_factor=1, const int vdim=-1) const;

   /// Computes the \ref PLBound for the gridfunction with number of control
-   /// points based on @a ref_factor, and returns the bounds for each element
-   /// ordered byNODES:
+   /// points based on \p ref_factor, and returns the bounds for each element
+   /// ordered byNodes:
   /// lower_{0,0}, lower_{1,0}, ..., lower_{ne-1,0},
   /// lower_{0,1}, ..., lower_{ne-1,vdim-1}. We also return the
   /// PLBound object used to compute the bounds.
@@ -1826,7 +1802,7 @@ public:
                         const int vdim = -1) const;

   /// Compute bounds on the grid function for all the elements. The bounds
-   /// are returned in @b lower and @b upper, ordered byNODES:
+   /// are returned in @b lower and @b upper, ordered byNodes:
   /// lower_{0,0}, lower_{1,0}, ..., lower_{ne-1,0},
   /// lower_{0,1}, ..., lower_{ne-1,vdim-1}
   void GetElementBounds(const PLBound &plb, Vector &lower, Vector &upper,
@@ -12,9 +12,6 @@
 #ifndef MFEM_GSLIB
 #define MFEM_GSLIB

-#include <map>
-#include <vector>
-
 #include "../config/config.hpp"
 #ifdef MFEM_USE_MPI
 #include "pgridfunc.hpp"
@@ -24,45 +21,6 @@

 #ifdef MFEM_USE_GSLIB

-/* gslib license and copyright statement for code adapted from gslib:
-
-Copyright (c) 2008-2024, UCHICAGO ARGONNE, LLC.
-
-The UChicago Argonne, LLC as Operator of Argonne National
-Laboratory holds copyright in the Software. The copyright holder
-reserves all rights except those expressly granted to licensees,
-and U.S. Government license rights.
-
-Redistribution and use in source and binary forms, with or without
-modification, are permitted provided that the following conditions
-are met:
-
-1. Redistributions of source code must retain the above copyright
-notice, this list of conditions and the disclaimer below.
-
-2. Redistributions in binary form must reproduce the above copyright
-notice, this list of conditions and the disclaimer (as noted below)
-in the documentation and/or other materials provided with the
-distribution.
-
-3. Neither the name of ANL nor the names of its contributors
-may be used to endorse or promote products derived from this software
-without specific prior written permission.
-
-THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
-"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
-LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
-FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL
-UCHICAGO ARGONNE, LLC, THE U.S. DEPARTMENT OF
-ENERGY OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
-SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
-TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
-DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
-THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
-(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
-OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
-*/
-
 namespace gslib
 {
 struct comm;
@@ -122,18 +80,13 @@ protected:
   // IntegrationRules for simplex->Quad/Hex and to project to p_max in-case of
   // p-refinement.
   Array<IntegrationRule *> ir_split;
-   /// Integration rules built at the field polynomial order (only for surface
-   /// meshes when mesh order is not the same as gridfunction order).
-   Array<IntegrationRule *> ir_split_sol;
-   /// Order at which #ir_split_sol was built; -1 means not built.
-   int ir_split_sol_order = -1;
   Array<FiniteElementSpace *> fes_rst_map; //FESpaces to map Quad/Hex->Simplex
   Array<GridFunction *> gf_rst_map; // GridFunctions to map Quad/Hex->Simplex
   FiniteElementCollection *fec_map_lin;
   void *fdataD;
   struct gslib::crystal *cr;             // gslib's internal data
   struct gslib::comm *gsl_comm;          // gslib's internal data
-   int dim, spacedim, points_cnt;         // mesh dimension and number of points
+   int dim, points_cnt;                   // mesh dimension and number of points
   Array<unsigned int> gsl_code, gsl_proc, gsl_elem, gsl_mfem_elem;
   Vector gsl_mesh, gsl_ref, gsl_dist, gsl_mfem_ref;
   Array<unsigned int> recv_proc, recv_index; // data for custom interpolation
@@ -142,8 +95,6 @@ protected:
   AvgType avgtype;             // average type used for L2 functions
   Array<int> split_element_map;
   Array<int> split_element_index;
-   // Geometry::Type (as int) of the original element for each split quad.
-   Array<int> split_element_geom;
   int        NE_split_total;   // total number of elements after mesh splitting
   int        mesh_points_cnt;  // number of mesh nodes
   // Tolerance to ignore points found beyond the mesh boundary.
@@ -151,235 +102,111 @@ protected:
   double     bdr_tol;
   // Use CPU functions for Mesh/GridFunction on device for gslib1.0.7
   bool       gpu_to_cpu_fallback = false;
-   // Check if a point is inside the oriented bounding box of an
-   // element before the Newton iteration.
-   // Note: only used in MFEM implementation (not in gslib) which currently
-   // supports GPU kernels for area meshes in 2D, volume meshes in 3D,
-   // and surface meshes in 1D/2D/3D.
-   bool       obb_check = true;

   // Device specific data used for FindPoints
-   struct DEV_STRUCT
+   struct
   {
      bool setup_device = false;
      bool find_device  = false;
-      int local_hash_size, dof1d, dof1d_sol, lh_nx, gh_nx;
+      int local_hash_size, dof1d, dof1d_sol, h_o_size, h_nx;
      double newt_tol; // Tolerance specified during setup for Newton solve
      struct gslib::crystal *cr;
      struct gslib::hash_data_3 *hash3;
      struct gslib::hash_data_2 *hash2;
      mutable Vector bb, wtend, gll1d, lagcoeff, gll1d_sol, lagcoeff_sol;
-      mutable Array<unsigned int> lh_offset, gh_offset;
-      mutable Vector lh_min, lh_fac, gh_min, gh_fac;
-      // Tolerance to mark points found on the surface as CODE_INTERNAL
-      // or CODE_BORDER. This is needed because we cannot only use reference
-      // space coordinates to determine if a point is located inside the
-      // element or not.
-      mutable double surf_dist_tol;
+      mutable Array<unsigned int> loc_hash_offset;
+      mutable Vector loc_hash_min, loc_hash_fac;
   } DEV;

-   // Helper function to setup and free gslib's crystal router.
-   void SetupCrystal(); // Called inside Setup and SetupSurf_base
-   void FreeCrystal();  // Called inside FreeData
-
-   /// Use GSLIB for communication and interpolation. Updates field_out on
-   /// host.
+   /// Use GSLIB for communication and interpolation
   virtual void InterpolateH1(const GridFunction &field_in, Vector &field_out,
                              const int field_out_ordering);
   /// Uses GSLIB Crystal Router for communication followed by MFEM's
-   /// interpolation functions. Updates field_out on host.
+   /// interpolation functions
   virtual void InterpolateGeneral(const GridFunction &field_in,
                                   Vector &field_out,
                                   const int field_out_ordering);

-   /** @brief Since GSLIB is designed to work with quads/hexes, we split every
-    *  triangle/tet/prism/pyramid element into quads/hexes. */
+   /// Since GSLIB is designed to work with quads/hexes, we split every
+   /// triangle/tet/prism/pyramid element into quads/hexes.
   virtual void SetupSplitMeshes();

-   /** @brief Setup integration points that will be used to interpolate the
-    *  nodal location at points expected by GSLIB. */
+   /// Setup integration points that will be used to interpolate the nodal
+   /// location at points expected by GSLIB.
   virtual void SetupIntegrationRuleForSplitMesh(Mesh *mesh,
                                                 IntegrationRule *irule,
                                                 int order);

-   /** @brief Build integration rules at the given @a order for each split mesh
-    *  and store them in @a ir_out. Requires that \ref SetupSplitMeshes has
-    *  already been called. */
-   virtual void SetupIntegrationRules(const int order,
-                                      Array<IntegrationRule *> &ir_out);
-
-   /** @brief Helper function that calls \ref SetupSplitMeshes and
-    * \ref SetupIntegrationRules. */
+   /// Helper function that calls \ref SetupSplitMeshes and
+   /// \ref SetupIntegrationRuleForSplitMesh.
   virtual void SetupSplitMeshesAndIntegrationRules(const int order);

-   /** @brief Get GridFunction value at the points expected by GSLIB.
-    *  @param[in]  gf_in       Grid function to evaluate.
-    *  @param[out] node_vals   Output values.
-    *  @param[in]  ir_in       If non-null, use these rules instead of #ir_split.
-    *  @param[in]  by_element  If true, output has element-major layout
-    *                          [nel][vdim][ndofs]; otherwise component-major
-    *                          layout [vdim][total_pts]. */
-   virtual void GetNodalValues(const GridFunction *gf_in, Vector &node_vals,
-                               const Array<IntegrationRule *> *ir_in = nullptr,
-                               bool by_element = false) const;
+   /// Get GridFunction value at the points expected by GSLIB.
+   virtual void GetNodalValues(const GridFunction *gf_in, Vector &node_vals) const;

-   /** @brief Map {r,s,t} coordinates from [-1,1] to [0,1] for MFEM. For
-    *  simplices, find the original element number (that was split into
-    *  micro quads/hexes) during the setup phase. */
+   /// Map {r,s,t} coordinates from [-1,1] to [0,1] for MFEM. For simplices,
+   /// find the original element number (that was split into micro quads/hexes)
+   /// during the setup phase.
   virtual void MapRefPosAndElemIndices();

-   /// FindPoints locally on device for 3D.
+   // Device functions
+   // FindPoints locally on device for 3D.
   void FindPointsLocal3(const Vector &point_pos, int point_pos_ordering,
                         Array<unsigned int> &gsl_code_dev_l,
                         Array<unsigned int> &gsl_elem_dev_l, Vector &gsl_ref_l,
                         Vector &gsl_dist_l, int npt);

-   /// FindPoints locally on device for 2D.
+   // FindPoints locally on device for 2D.
   void FindPointsLocal2(const Vector &point_pos, int point_pos_ordering,
                         Array<unsigned int> &gsl_code_dev_l,
                         Array<unsigned int> &gsl_elem_dev_l, Vector &gsl_ref_l,
                         Vector &gsl_dist_l, int npt);

-   /// FindPoints locally on device for 3D surface elements.
-   void FindPointsSurfLocal3(const Vector &point_pos,
-                             int point_pos_ordering,
-                             Array<unsigned int> &gsl_code_dev_l,
-                             Array<unsigned int> &gsl_elem_dev_l,
-                             Vector &gsl_ref_l,
-                             Vector &gsl_dist_l,
-                             int npt);
-
-   /// FindPoints locally on device for 3D edge elements.
-   void FindPointsEdgeLocal3(const Vector &point_pos,
-                             int point_pos_ordering,
-                             Array<unsigned int> &gsl_code_dev_l,
-                             Array<unsigned int> &gsl_elem_dev_l,
-                             Vector &gsl_ref_l,
-                             Vector &gsl_dist_l,
-                             int npt);
-
-   /// FindPoints locally on device for 2D edge elements.
-   void FindPointsEdgeLocal2(const Vector &point_pos,
-                             int point_pos_ordering,
-                             Array<unsigned int> &gsl_code_dev_l,
-                             Array<unsigned int> &gsl_elem_dev_l,
-                             Vector &gsl_ref_l,
-                             Vector &gsl_dist_l,
-                             int npt);
-
-   /// Interpolate on device for 3D.
+   // Interpolate on device for 3D.
   void InterpolateLocal3(const Vector &field_in,
                          Array<int> &gsl_elem_dev_l,
                          Vector &gsl_ref_l,
                          Vector &field_out,
                          int npt, int ncomp,
-                          int dof1dsol);
-
-   /// Interpolate on device for 2D.
+                          int nel, int dof1dsol);
+   // Interpolate on device for 2D.
   void InterpolateLocal2(const Vector &field_in,
                          Array<int> &gsl_elem_dev_l,
                          Vector &gsl_ref_l,
                          Vector &field_out,
                          int npt, int ncomp,
-                          int dof1dsol);
+                          int nel, int dof1dsol);

-   /// Interpolate on device for 1D.
-   void InterpolateLocal1(const Vector &field_in,
-                          Array<int> &gsl_elem_dev_l,
-                          Vector &gsl_ref_l,
-                          Vector &field_out,
-                          int npt, int ncomp, int dof1dsol);
-
-   /// Prepare data for device execution for volume meshes.
+   // Prepare data for device functions.
   void SetupDevice();

-   /** @brief Searches positions given in physical space by @a point_pos.
+   /** Searches positions given in physical space by @a point_pos.
       These positions can be ordered byNodes: (XXX...,YYY...,ZZZ) or
       byVDim: (XYZ,XYZ,....XYZ) specified by @a point_pos_ordering. */
   void FindPointsOnDevice(const Vector &point_pos,
                           const int point_pos_ordering = Ordering::byNODES);

-   /** @brief Interpolation of field values at prescribed reference space
-    *         positions.
-    *  @param[in] field_in_evec E-vector of grid function to be interpolated.
-    *                           Assumed ordering is NDOFSxVDIMxNEL
-    *  @param[in] nel           Number of elements in the mesh.
-    *  @param[in] ncomp         Number of components in the field.
-    *  @param[in] dof1dsol      Number of degrees of freedom in each reference
-    *                           space direction.
-    *  @param[in] ordering      Ordering of the out field values: byNodes/byVDIM
-    *
-    *  @param[out] field_out    Interpolated values. For points that are not
-    *                           found the value is set to
-    *                           #default_interp_value. */
+   /** Interpolation of field values at prescribed reference space positions.
+       @param[in] field_in_evec E-vector of grid function to be interpolated.
+                                Assumed ordering is NDOFSxVDIMxNEL
+       @param[in] nel           Number of elements in the mesh.
+       @param[in] ncomp         Number of components in the field.
+       @param[in] dof1dsol      Number of degrees of freedom in each reference
+                                space direction.
+       @param[in] ordering      Ordering of the out field values: byNodes/byVDIM
+
+       @param[out] field_out  Interpolated values. For points that are not found
+                              the value is set to #default_interp_value. */
   void InterpolateOnDevice(const Vector &field_in_evec, Vector &field_out,
                            const int nel, const int ncomp,
                            const int dof1dsol, const int ordering);

-   /** @brief Interpolation of field values at prescribed reference space
-    *         positions for surface meshes. */
-   void InterpolateSurfBase(const Vector &field_in, Vector &field_out,
-                            const int nel, const int ncomp,
-                            const int dof1dsol, const int field_out_ordering);
-
-   /// Preprocess 2D surface mesh needed for FindPoints.
-   void findptsedge_setup_2(DEV_STRUCT &devs,
-                            const double *const elx[2],
-                            const unsigned n,
-                            const unsigned int nel,
-                            const unsigned m,
-                            const double bbox_rel_size_inc,
-                            const unsigned int local_hash_size,
-                            const unsigned int global_hash_size,
-                            const Vector *aabb_sz_inc);
-
-   /// Preprocess 3D surface mesh needed for FindPoints.
-   void findptssurf_setup_3(DEV_STRUCT &devs,
-                            const double *const elx[3],
-                            const unsigned n,
-                            const unsigned int nel,
-                            const unsigned m,
-                            const double bbox_rel_size_inc,
-                            const unsigned int local_hash_size,
-                            const unsigned int global_hash_size,
-                            const int rD,
-                            const Vector *aabb_sz_inc);
-
-   /** @brief Shared implementation for the public surface-setup methods.
-    *
-    *  @details Initializes the surface-search data structures, builds the
-    *  split-element representation expected by gslib, and constructs the
-    *  element bounding boxes used by the MFEM surface kernels.
-    *
-    *  If @a aabb_sz_inc is null, the setup stores the default oriented
-    *  bounding boxes and uses @a bbox_rel_size_inc as their relative size
-    *  increase factor.
-    *
-    *  If @a aabb_sz_inc is non-null, the setup stores axis-aligned bounding
-    *  boxes only, applies the requested absolute AABB expansion in each
-    *  physical direction, and adjusts the tolerance @a bdr_tol so points
-    *  found in the expanded region are classified as border points.
-    *
-    *  @param[in] m                  Input surface mesh.
-    *  @param[in] bbox_rel_size_inc  Relative size increase applied when
-    *                                expanding each element bounding box during
-    *                                setup.
-    *  @param[in] aabb_sz_inc        Optional total absolute AABB expansion
-    *                                applied to the stored axis-aligned
-    *                                bounding boxes after construction.
-    *  @param[in] newt_tol           Newton tolerance for the point-search
-    *                                kernels.
-    */
-   void SetupSurf_Base(Mesh &m,
-                       const double bbox_rel_size_inc,
-                       const Vector *aabb_sz_inc,
-                       const double newt_tol);
 public:
   /// Serial constructor
   FindPointsGSLIB();

   /// Serial constructor + setup with given Mesh (see \ref Setup)
-   FindPointsGSLIB(Mesh &mesh_in, const double bbox_rel_size_inc = 0.1,
+   FindPointsGSLIB(Mesh &mesh_in, const double bb_t = 0.1,
                   const double newt_tol = 1.0e-12,
                   const int npt_max = 256);

@@ -388,7 +215,7 @@ public:
   FindPointsGSLIB(MPI_Comm comm_);

   /// Constructor + setup with given ParMesh (see \ref Setup)
-   FindPointsGSLIB(ParMesh &mesh_in, const double bbox_rel_size_inc = 0.1,
+   FindPointsGSLIB(ParMesh &mesh_in, const double bb_t = 0.1,
                   const double newt_tol = 1.0e-12,
                   const int npt_max = 256);
 #endif
@@ -397,72 +224,25 @@ public:
   FindPointsGSLIB(const FindPointsGSLIB&) = delete;
   FindPointsGSLIB& operator=(const FindPointsGSLIB&) = delete;

-   /** @brief Preprocess the internal mesh in gslib.
-
-       @details Initializes the internal mesh in gslib, by sending the
-       positions of the Gauss-Lobatto nodes of the input Mesh object \p m.
+   /** Initializes the internal mesh in gslib, by sending the positions of the
+       Gauss-Lobatto nodes of the input Mesh object \p m.
       Note: not tested with periodic (L2).
       Note: the input mesh \p m must have Nodes set.

-       @param[in] m                  Input mesh.
-       @param[in] bbox_rel_size_inc  (Optional) Relative size increase applied
-                                     when expanding each element bounding box.
-       @param[in] newt_tol           (Optional) Newton tolerance for the gslib
-                                     search methods.
-       @param[in] npt_max            (Optional) Number of points for
-                                     simultaneous iteration. This alters
-                                     performance and memory footprint.
-   */
-   void Setup(Mesh &m, const double bbox_rel_size_inc = 0.1,
-              const double newt_tol = 1.0e-12,
+       @param[in] m         Input mesh.
+       @param[in] bb_t      (Optional) Relative size of bounding box around
+                            each element.
+       @param[in] newt_tol  (Optional) Newton tolerance for the gslib
+                            search methods.
+       @param[in] npt_max   (Optional) Number of points for simultaneous
+                            iteration. This alters performance and
+                            memory footprint.*/
+
+   void Setup(Mesh &m, const double bb_t = 0.1, const double newt_tol = 1.0e-12,
              const int npt_max = 256);
-
-   /// Preprocess the surface mesh to compute data for FindPoints.
-   void SetupSurf(Mesh &m,
-                  const double bbox_rel_size_inc = 0.1,
-                  const double newt_tol = 1.0e-12);
-
-   /** @brief Preprocess the surface mesh to compute data for FindPoints using
-    *  absolute AABB expansion.
-    *
-    *  @details This method computes only axis-aligned bounding boxes and
-    *  increases their total length by a user-specified amount in each
-    *  physical direction. The absolute AABB expansion is applied
-    *  symmetrically to the lower and upper bounds.
-    *
-    *  The size of @a aabb_sz_inc determines how the expansion values are
-    *  interpreted:
-    *  - `1`: one expansion value used in every direction for every element
-    *  - `NElements`: one expansion value per element, reused in x/y/z
-    *    directions
-    *  - `SpaceDim`: one expansion value per physical direction, reused for
-    *    every element
-    *  - `NElements*SpaceDim`: one expansion value per element and direction,
-    *    ordered as `(dx1,dy1,dz1, ... dxN,dyN,dzN)`
-    *
-    *  This method disables the oriented bounding-box precheck because the
-    *  stored boxes are modified only in their axis-aligned representation.
-    *
-    *  @param[in] m                  Input surface mesh.
-    *  @param[in] aabb_sz_inc        Total absolute AABB expansion applied in
-    *                                each physical direction to the stored
-    *                                axis-aligned bounding boxes.
-    *  @param[in] newt_tol           Newton tolerance for the point-search
-    *                                kernels.
-    *
-    *  @note We disable the oriented bounding box check with this setup.
-    *        @a bdr_tol is also adjusted so that all points in the AABBs can
-    *        be found.
-    */
-   void SetupSurfWithAABBExpansion(Mesh &m, const Vector &aabb_sz_inc,
-                                   const double newt_tol = 1.0e-12);
-
-
-   /** @brief Searches positions given in physical space by \p point_pos.
-
-       @details These positions can be ordered byNodes: (XXX...,YYY...,ZZZ) or
+   /** Searches positions given in physical space by \p point_pos.
+       These positions can be ordered byNodes: (XXX...,YYY...,ZZZ) or
       byVDim: (XYZ,XYZ,....XYZ) specified by \p point_pos_ordering.
-
       This function populates the following member variables:
       #gsl_code        Return codes for each point: inside element (0),
                        element boundary (1), not found (2).
@@ -481,77 +261,40 @@ public:
       #gsl_dist        Distance between the sought and the found point
                        in physical space. */
   void FindPoints(const Vector &point_pos,
-                   int point_pos_ordering = Ordering::byNODES);
-
+                   const int point_pos_ordering = Ordering::byNODES);
   /// Convenience function when point positions are in a ParticleVector
   void FindPoints(const ParticleVector &point_pos)
   {
      FindPoints(point_pos, point_pos.GetOrdering());
   }
-
-   /** @brief Searches positions given in physical space by \p point_pos on
-    *  surface mesh. */
-   void FindPointsSurf(const Vector &point_pos,
-                       int point_pos_ordering = Ordering::byNODES);
-
-   /// Convenience function when point positions are in a ParticleVector
-   void FindPointsSurf(const ParticleVector &point_pos)
-   {
-      FindPointsSurf(point_pos, point_pos.GetOrdering());
-   }
-
   /// Setup FindPoints and search positions
   void FindPoints(Mesh &m, const Vector &point_pos,
                   const int point_pos_ordering = Ordering::byNODES,
-                   const double bbox_rel_size_inc = 0.1,
-                   const double newt_tol = 1.0e-12,
+                   const double bb_t = 0.1, const double newt_tol = 1.0e-12,
                   const int npt_max = 256);

-   /** @brief Interpolation of field values at prescribed reference space
-    *  positions.
-
+   /** Interpolation of field values at prescribed reference space positions.
       @param[in] field_in    Function values that will be interpolated on the
                              reference positions. Note: it is assumed that
                              \p field_in is in H1 and in the same space as the
                              mesh that was given to Setup().
       @param[out] field_out  Interpolated values. For points that are not found
                              the value is set to #default_interp_value.
-                              The output ordering is determined from field_in.
-
-       @note: field_out is moved to device if field_in is on device. Otherwise,
-              field_out memory allocation is not changed.
-   */
+                              The output ordering is determined from field_in.*/
   virtual void Interpolate(const GridFunction &field_in, Vector &field_out);
-
   /// Interpolation of field values, with output ordering specification.
   virtual void Interpolate(const GridFunction &field_in, Vector &field_out,
                            const int field_out_ordering);
-
-   /** @brief Same as Interpolate but for surface meshes */
-   virtual void InterpolateSurf(const GridFunction &field_in,
-                                Vector &field_out);
-
-   /** @brief Same as Interpolate but for surface meshes with specified output
-       ordering */
-   virtual void InterpolateSurf(const GridFunction &field_in,
-                                Vector &field_out,
-                                const int field_out_ordering);
-
-   /** @brief Search positions and interpolate.
-    *
-    *  @details The ordering (byNODES or byVDIM) of the output values in
-    *  \p field_out corresponds to the ordering used in the input
-    *  GridFunction \p field_in.
-    */
+   /** Search positions and interpolate. The ordering (byNODES or byVDIM) of
+       the output values in \p field_out corresponds to the ordering used
+       in the input GridFunction \p field_in. */
   void Interpolate(const Vector &point_pos, const GridFunction &field_in,
                    Vector &field_out,
-                    int point_pos_ordering = Ordering::byNODES);
-
+                    const int point_pos_ordering = Ordering::byNODES);
   /// Search positions and interpolate with given point and output ordering.
   void Interpolate(const Vector &point_pos, const GridFunction &field_in,
                    Vector &field_out, const int point_pos_ordering,
                    const int field_out_ordering);
-
   /** Setup FindPoints, search positions and interpolate. The ordering (byNODES
       or byVDIM) of the output values in \p field_out corresponds to the
       ordering used in the input GridFunction \p field_in. */
@@ -559,41 +302,32 @@ public:
                    const GridFunction &field_in, Vector &field_out,
                    const int point_pos_ordering = Ordering::byNODES);

-   /** @brief Average type to be used for L2 functions in-case a point is
-    *  located at an element boundary where the function might be multi-valued.
-    */
+   /// Average type to be used for L2 functions in-case a point is located at
+   /// an element boundary where the function might be multi-valued.
   virtual void SetL2AvgType(AvgType avgtype_) { avgtype = avgtype_; }

-   /** @brief Set the default interpolation value for points that are not found in the mesh. */
+   /// Set the default interpolation value for points that are not found in the
+   /// mesh.
   virtual void SetDefaultInterpolationValue(double interp_value_)
   {
      default_interp_value = interp_value_;
   }

-   /** @brief Tolerance for detecting points outside the 'curvilinear' boundary.
-    *
-    *  @details When using FindPoints, gslib may return points as found on the
-    *  boundary even when they are slightly outside the domain. This tolerance
-    *  is used to filter such points based on the distance^2 value and mark them
-    *  as not found.
-    *
-    *  @note When the SetupSurfWithAABBExpansion method is used for surface
-    *  meshes, this tolerance is automatically computed based on the size of
-    *  expanded AABBs. Using this method will override that computed tolerance.
-    *  */
+   /// Set the tolerance for detecting points outside the 'curvilinear' boundary
+   /// that gslib may return as found on the boundary. Points found on boundary
+   /// with distance greater than @ bdr_tol are marked as not found.
   virtual void SetDistanceToleranceForPointsFoundOnBoundary(double bdr_tol_)
   {
      bdr_tol = bdr_tol_;
   }

-   /** @brief Enable/Disable use of CPU functions for GPU data if the gslib
-    *  version is older. */
+   /// Enable/Disable use of CPU functions for GPU data if the gslib version
+   /// is older.
   virtual void SetGPUtoCPUFallback(bool mode) { gpu_to_cpu_fallback = mode; }

-   /** @brief Cleans up memory allocated internally by gslib.
-
-       @details Note that in parallel, this must be called before MPI_Finalize,
-       as it calls MPI_Comm_free() for internal gslib communicators. FreeData is
+   /** Cleans up memory allocated internally by gslib.
+       Note that in parallel, this must be called before MPI_Finalize(), as it
+       calls MPI_Comm_free() for internal gslib communicators. FreeData is
       also called by the class destructor and there are no memory leaks if the
       destructor is called before MPI_Finalize(). If the destructor is called
       after MPI_Finalize(), there will be an error because gslib will try to
@@ -601,8 +335,8 @@ public:
   */
   virtual void FreeData();

-   /** @brief Return code for each point searched by FindPoints:
-    *  inside element (0), element boundary (1), or not found (2). */
+   /// Return code for each point searched by FindPoints: inside element (0), on
+   /// element boundary (1), or not found (2).
   virtual const Array<unsigned int> &GetCode() const { return gsl_code; }
   /// Return element number for each point found by FindPoints.
   virtual const Array<unsigned int> &GetElem() const { return gsl_mfem_elem; }
@@ -610,15 +344,15 @@ public:
   virtual const Array<unsigned int> &GetProc() const { return gsl_proc; }
   /// Return reference coordinates for each point found by FindPoints.
   virtual const Vector &GetReferencePosition() const { return gsl_mfem_ref;  }
-   /// Return distance between the sought and the found point in physical space.
+   /// Return distance between the sought and the found point in physical space,
+   /// for each point found by FindPoints.
   virtual const Vector &GetDist()              const { return gsl_dist; }

-   /** @brief Return element number for each point found by FindPoints
-    *  corresponding to GSLIB mesh. gsl_mfem_elem != gsl_elem for mesh with
-    *  simplices. */
+   /// Return element number for each point found by FindPoints corresponding to
+   /// GSLIB mesh. gsl_mfem_elem != gsl_elem for mesh with simplices.
   virtual const Array<unsigned int> &GetGSLIBElem() const { return gsl_elem; }
-   /** @brief Return reference coordinates in [-1,1] (internal range in GSLIB)
-    *  for each point found by FindPoints. */
+   /// Return reference coordinates in [-1,1] (internal range in GSLIB) for each
+   /// point found by FindPoints.
   virtual const Vector &GetGSLIBReferencePosition() const { return gsl_ref; }

   /// Get array of indices of not-found points.
@@ -661,7 +395,7 @@ public:

   /// Return the axis-aligned bounding boxes (AABB) computed during \ref Setup.
   /// The size of the returned vector is (nel x nverts x dim), where nel is the
-   /// number of elements (after splitting for simplicies), nverts is number of
+   /// number of elements (after splitting for simplcies), nverts is number of
   /// vertices (4 in 2D, 8 in 3D), and dim is the spatial dimension.
   void GetAxisAlignedBoundingBoxes(Vector &aabb) const;

@@ -675,18 +409,6 @@ public:
   /// \p obbV, a vector of size (nel x nverts x dim) .
   void GetOrientedBoundingBoxes(DenseTensor &obbA, Vector &obbC,
                                 Vector &obbV) const;
-
-   /** @brief Return the bounding boxes as a mesh on rank 0.
-    *
-    *  @param[in] type  Bounding-box type: 0 - AABB, 1 - OBB.
-    *
-    *  @return On rank 0, returns a newly allocated mesh containing the
-    *  bounding boxes. The caller owns the returned pointer and is responsible
-    *  for deleting it. On other ranks, returns nullptr.
-    */
-   Mesh *GetBoundingBoxMesh(int type);
-
-   virtual const Vector &GetGLLMesh()           const { return gsl_mesh; }
 };

 /** \brief OversetFindPointsGSLIB enables use of findpts for arbitrary number of
@@ -715,28 +437,25 @@ public:
       Note: not tested with periodic meshes (L2).
       Note: the input mesh \p m must have Nodes set.

-       @param[in] m                  Input mesh.
-       @param[in] meshid             A unique # for each overlapping mesh.
-                                     This id is used to make sure that points
-                                     being searched are not looked for in the
-                                     mesh that they belong to.
-       @param[in] gfmax              (Optional) GridFunction in H1 that is used
-                                     as a discriminator when one point is
-                                     located in multiple meshes. The mesh that
-                                     maximizes gfmax is chosen. For example,
-                                     using the distance field based on the
-                                     overlapping boundaries is helpful for
-                                     convergence during Schwarz iterations.
-       @param[in] bbox_rel_size_inc  (Optional) Relative size increase applied
-                                     when expanding each element bounding box.
-       @param[in] newt_tol           (Optional) Newton tolerance for the gslib
-                                     search methods.
-       @param[in] npt_max            (Optional) Number of points for
-                                     simultaneous iteration. This alters
-                                     performance and memory footprint.*/
-   void Setup(Mesh &m, const int meshid, GridFunction *gfmax = nullptr,
-              const double bbox_rel_size_inc = 0.1,
-              const double newt_tol = 1.0e-12,
+       @param[in] m         Input mesh.
+       @param[in] meshid    A unique # for each overlapping mesh. This id is
+                            used to make sure that points being searched are not
+                            looked for in the mesh that they belong to.
+       @param[in] gfmax     (Optional) GridFunction in H1 that is used as a
+                            discriminator when one point is located in multiple
+                            meshes. The mesh that maximizes gfmax is chosen.
+                            For example, using the distance field based on the
+                            overlapping boundaries is helpful for convergence
+                            during Schwarz iterations.
+       @param[in] bb_t      (Optional) Relative size of bounding box around
+                            each element.
+       @param[in] newt_tol  (Optional) Newton tolerance for the gslib
+                            search methods.
+       @param[in] npt_max   (Optional) Number of points for simultaneous
+                            iteration. This alters performance and
+                            memory footprint.*/
+   void Setup(Mesh &m, const int meshid, GridFunction *gfmax = NULL,
+              const double bb_t = 0.1, const double newt_tol = 1.0e-12,
              const int npt_max = 256);

   /** Searches positions given in physical space by \p point_pos. All output
@@ -792,7 +511,7 @@ class GSOPGSLIB
 protected:
   struct gslib::crystal *cr;               // gslib's internal data
   struct gslib::comm *gsl_comm;            // gslib's internal data
-   struct gslib::gs_data *gsl_data = nullptr;
+   struct gslib::gs_data *gsl_data = NULL;
   int num_ids;

 public:
@@ -817,116 +536,6 @@ public:
   void GS(Vector &senddata, GSOp op);
 };

-#if defined(MFEM_USE_MPI)
-/** \brief Class to map a point in physical space to candidate ranks.
- *
- *  This class builds a Cartesian-aligned tensor grid that covers the entire
- *  domain and precomputes which ranks have elements intersecting each
- *  grid cell. Given a point in physical space, the grid cell containing
- *  the point is determined, and the list of candidate ranks whose
- *  elements intersect that cell is returned. This yields a fast, conservative
- *  point-to-rank candidate query. This is used internally by FindPointsGSLIB
- *  to speed up point searches in parallel.
- *
- *  See Mittal et al., "General Field Evaluation in High-Order Meshes on GPUs".
- *  (2025). Computers & Fluids. for technical details.
- *
- */
-class GlobalBBoxTensorGridMap
-{
-private:
-   struct gslib::crystal *cr = nullptr;               // gslib's internal data
-   struct gslib::comm *gsl_comm = nullptr;            // gslib's internal data
-   int sdim, n_local_cells, num_procs;
-   Array<int> gmap_n;
-   Vector gmap_bnd_min, gmap_bnd_max;
-   Vector gmap_fac;
-   Array<int> ggrid_map;
-
-   void SetupCrystal(const MPI_Comm &comm);
-public:
-   /// Constructor for a given mesh and number of tensor grid divisions
-   GlobalBBoxTensorGridMap(ParMesh &pmesh, int nx);
-
-   /** @brief Constructor for given element bounds and spatial dimension.
-    *
-    *  @details This constructor must be called collectively on \a comm.
-    *  Supports spatial dimensions 1, 2, and 3, and accepts nel == 0 on a rank.
-    *
-    *  Assumes elmin, elmax Ordering::byNodes:
-    *  elmin -> [x_{0,min},x_{1,min},... ,y_{0,min},y_{1,min},..,z_{nel-1,min}]
-    *  elmax -> [x_{0,max},x_{1,max},... ,y_{0,max},y_{1,max},..,z_{nel-1,max}]
-    *  Note elmin, elmax can be obtained using GridFunction::GetElementBounds()
-    *
-    *  When by_max_size=false, n gives the number of tensor-grid divisions in
-    *  each direction. When by_max_size=true, n is a per-rank size hint used to
-    *  derive a uniform global resolution. The communicator-wide sum of n is
-    *  converted to nx = ceil(pow(sum(n), 1./sdim)) in each direction, so n is
-    *  not a hard cap on ggrid_map.Size().
-    */
-   GlobalBBoxTensorGridMap(const MPI_Comm &comm, Vector &elmin,
-                           Vector &elmax, int nel, int sdim, int n,
-                           bool by_max_size);
-
-   /** @brief Constructor for given element bounds, spatial dimension, and
-    *  tensor-grid divisions in each direction.
-    *
-    *  @details This constructor must be called collectively on \a comm.
-    *  Supports spatial dimensions 1, 2, and 3, and accepts nel == 0 on a rank.
-    *  Requires nx.Size() == sdim and positive entries in nx.
-    *
-    *  Assumes elmin, elmax Ordering::byNodes:
-    *  elmin -> [x_{0,min},x_{1,min},... ,y_{0,min},y_{1,min},..,z_{nel-1,min}]
-    *  elmax -> [x_{0,max},x_{1,max},... ,y_{0,max},y_{1,max},..,z_{nel-1,max}]
-    *  Note elmin, elmax can be obtained using GridFunction::GetElementBounds()
-    */
-   GlobalBBoxTensorGridMap(const MPI_Comm &comm, Vector &elmin,
-                           Vector &elmax, int nel, int sdim, Array<int> &nx);
-
-   ~GlobalBBoxTensorGridMap();
-
-   /** @brief Get list of procs corresponding to the list of points.
-    *
-    *  @details This method must be called collectively on the communicator
-    *  used to construct the map. The input points can be ordered byNodes:
-    *  (XXX...,YYY...,ZZZ) or byVDIM: (XYZ,XYZ,...), as specified by
-    *  \a ordering.
-    *
-    *  The output map contains one entry for each input point, keyed by the
-    *  point's local index in \a xyz. Points with no candidate ranks, including
-    *  points outside the global bounding box, have an empty list of candidate
-    *  ranks.
-    */
-   void MapPointsToProcs(Vector &xyz, int ordering,
-                         std::map<int, std::vector<int>> &pt_to_procs) const;
-
-   // Some getters
-   const Array<int> &GetGridMap() const { return ggrid_map; }
-   const Vector &GetGridFac() const { return gmap_fac; }
-   const Vector &GetGridMin() const { return gmap_bnd_min; }
-   const Vector &GetGridMax() const { return gmap_bnd_max; }
-   const Array<int> &GetGridN() const { return gmap_n; }
-
-private:
-   /// Setup the map given element bounds and number of tensor grid divisions.
-   void Setup(const MPI_Comm &comm, Vector &elmin, Vector &elmax,
-              int nel, Array<int> &nx);
-
-   /// Get global hash cell index for a given point.
-   int GetGlobalGridCellFromPoint(Vector &xyz) const;
-
-   /** @brief Get owning proc and local index on that proc for given global
-    *  grid cell index. */
-   void GlobalGridCellToProcAndLocalIndex(int i, int &proc, int &idx) const;
-
-   /// Map a point to proc and local index of the corresponding grid cell
-   void GetProcAndLocalIndexFromPoint(Vector &xyz, int &proc, int &idx) const;
-
-   /// Given local cell index, return list of procs saved in the map
-   Array<int> MapCellToProcs(int l_idx) const;
-};
-#endif // MFEM_USE_MPI
-
 } // namespace mfem

 #endif // MFEM_USE_GSLIB
@@ -11,7 +11,7 @@

 #include "../gslib.hpp"
 #include "../../general/forall.hpp"
-#include "gslib_kernel_helpers.hpp"
+#include "../../linalg/kernels.hpp"

 #ifdef MFEM_USE_GSLIB

@@ -27,6 +27,8 @@
 #pragma GCC diagnostic pop
 #endif

+#include <climits>
+
 namespace mfem
 {
 #if GSLIB_RELEASE_VERSION >= 10009
@@ -52,14 +54,127 @@ struct findptsElementGPT_t
   double x[DIM], jac[DIM * DIM], hes[4];
 };

-using dbl_range_t = gslib::dbl_range_t;
-using obbox_t = gslib::obbox_t<DIM>;
-using findptsLocalHashData_t = gslib::findptsLocalHashData_t<DIM>;
-using gslib::bbox_test;
-using gslib::hash_index;
-using gslib::l2norm2;
-using gslib::lag_eval_first_der;
-using gslib::lag_eval_second_der;
+struct dbl_range_t
+{
+   double min, max;
+};
+struct obbox_t
+{
+   double c0[DIM], A[DIM * DIM];
+   dbl_range_t x[DIM];
+};
+
+struct findptsLocalHashData_t
+{
+   int hash_n;
+   dbl_range_t bnd[DIM];
+   double fac[DIM];
+   unsigned int *offset;
+   int max;
+};
+
+// Eval the ith Lagrange interpolant and its first derivative at x.
+// Note: lCoeff stores pre-computed coefficients for fast evaluation.
+static MFEM_HOST_DEVICE inline void lag_eval_first_der(double *p0, double x,
+                                                       int i, const double *z,
+                                                       const double *lCoeff,
+                                                       int pN)
+{
+   double u0 = 1, u1 = 0;
+   for (int j = 0; j < pN; ++j)
+   {
+      if (i != j)
+      {
+         double d_j = 2 * (x - z[j]);
+         u1 = d_j * u1 + u0;
+         u0 = d_j * u0;
+      }
+   }
+   p0[i] = lCoeff[i] * u0;
+   p0[pN+i] = 2.0 * lCoeff[i] * u1;
+}
+
+// Eval the ith Lagrange interpolant and its first and second derivative at x.
+// Note: lCoeff stores pre-computed coefficients for fast evaluation.
+static MFEM_HOST_DEVICE inline void lag_eval_second_der(double *p0, double x,
+                                                        int i, const double *z,
+                                                        const double *lCoeff,
+                                                        int pN)
+{
+   double u0 = 1, u1 = 0, u2 = 0;
+   for (int j = 0; j < pN; ++j)
+   {
+      if (i != j)
+      {
+         double d_j = 2 * (x - z[j]);
+         u2 = d_j * u2 + u1;
+         u1 = d_j * u1 + u0;
+         u0 = d_j * u0;
+      }
+   }
+   p0[i] = lCoeff[i] * u0;
+   p0[pN+i] = 2.0 * lCoeff[i] * u1;
+   p0[2*pN+i] = 8.0 * lCoeff[i] * u2;
+}
+
+// Axis-aligned bounding box test.
+static MFEM_HOST_DEVICE inline double AABB_test(const obbox_t *const b,
+                                                const double x[2])
+{
+   double test = 1;
+   for (int d = 0; d < 2; ++d)
+   {
+      double b_d = (x[d] - b->x[d].min) * (b->x[d].max - x[d]);
+      test = test < 0 ? test : b_d;
+   }
+   return test;
+}
+
+// Axis-aligned bounding box test followed by oriented bounding-box test.
+static MFEM_HOST_DEVICE inline double bbox_test(const obbox_t *const b,
+                                                const double x[2])
+{
+   const double bxyz = AABB_test(b, x);
+   if (bxyz < 0)
+   {
+      return bxyz;
+   }
+   else
+   {
+      double dxyz[2];
+      for (int d = 0; d < 2; ++d)
+      {
+         dxyz[d] = x[d] - b->c0[d];
+      }
+      double test = 1;
+      for (int d = 0; d < 2; ++d)
+      {
+         double rst = 0;
+         for (int e = 0; e < 2; ++e)
+         {
+            rst += b->A[d * 2 + e] * dxyz[e];
+         }
+         double brst = (rst + 1) * (1 - rst);
+         test = test < 0 ? test : brst;
+      }
+      return test;
+   }
+}
+
+// Element index corresponding to hash mesh that the point is located in.
+static MFEM_HOST_DEVICE inline int hash_index(const findptsLocalHashData_t *p,
+                                              const double x[2])
+{
+   const int n = p->hash_n;
+   int sum = 0;
+   for (int d = 2 - 1; d >= 0; --d)
+   {
+      sum *= n;
+      int i = (int)floor((x[d] - p->bnd[d].min) * p->fac[d]);
+      sum += i < 0 ? 0 : (n - 1 < i ? n - 1 : i);
+   }
+   return sum;
+}

 /*Solve Ax=y. A is row-major */
 static MFEM_HOST_DEVICE inline void lin_solve_2(double x[2], const double A[4],
@@ -70,6 +185,12 @@ static MFEM_HOST_DEVICE inline void lin_solve_2(double x[2], const double A[4],
   x[1] = idet*(A[0]*y[1] - A[2]*y[0]);
 }

+/* L2 norm squared. */
+static MFEM_HOST_DEVICE inline double l2norm2(const double x[2])
+{
+   return x[0] * x[0] + x[1] * x[1];
+}
+
 /* the bit structure of flags is CSSRR
   the C bit --- 1<<4 --- is set when the point is converged
   RR is 0 = 00b if r is unconstrained,
@@ -231,7 +352,7 @@ static MFEM_HOST_DEVICE bool reject_prior_step_q(findptsElementPoint_t *res,
                                                 const findptsElementPoint_t *p,
                                                 const double tol)
 {
-   const double dist2 = l2norm2<2>(resid);
+   const double dist2 = l2norm2(resid);
   const double decr = p->dist2 - dist2;
   const double pred = p->dist2p;
   for (int d = 0; d < 2; ++d)
@@ -441,7 +562,7 @@ newton_area_fin:
      int f = flags >> (2 * dd) & 3u;
      res->r[dd] = f == 0 ? r0[dd] + dr[dd] : (f == 1 ? -1 : 1);
   }
-   res->flags = flags | ((p->flags & FLAG_MASK) << 5);
+   res->flags = flags | (p->flags << 5);
 }

 // Full Newton solve on the face. One of r/s/t is constrained.
@@ -514,8 +635,7 @@ newton_edge_fin:
   res->r[de] = nr;
   res->r[dn]=p->r[dn];
   res->dist2p = -v;
-   res->flags = flags | new_flags | ((p->flags & FLAG_MASK) << 5);
-#undef EVAL
+   res->flags = flags | new_flags | (p->flags << 5);
 }

 // Find closest mesh node to the sought point.
@@ -574,26 +694,27 @@ static MFEM_HOST_DEVICE double tensor_ig2_j(double *g_partials,
 }

 template<int T_D1D = 0>
-static void FindPointsLocal2DKernel(const int npt,
-                                    const double tol,
-                                    const double *x,
-                                    const int point_pos_ordering,
-                                    const double *xElemCoord,
-                                    const int nel,
-                                    const double *wtend,
-                                    const double *boxinfo,
-                                    const int hash_n,
-                                    const double *hashMin,
-                                    const double *hashFac,
-                                    unsigned int *hashOffset,
-                                    unsigned int *const code_base,
-                                    unsigned int *const el_base,
-                                    double *const r_base,
-                                    double *const dist2_base,
-                                    const double *gll1D,
-                                    const double *lagcoeff,
-                                    const int pN = 0)
+static void FindPointsLocal2D_Kernel(const int npt,
+                                     const double tol,
+                                     const double *x,
+                                     const int point_pos_ordering,
+                                     const double *xElemCoord,
+                                     const int nel,
+                                     const double *wtend,
+                                     const double *boxinfo,
+                                     const int hash_n,
+                                     const double *hashMin,
+                                     const double *hashFac,
+                                     unsigned int *hashOffset,
+                                     unsigned int *const code_base,
+                                     unsigned int *const el_base,
+                                     double *const r_base,
+                                     double *const dist2_base,
+                                     const double *gll1D,
+                                     const double *lagcoeff,
+                                     const int pN = 0)
 {
+#define MAX_CONST(a, b) (((a) > (b)) ? (a) : (b))
   const int MD1 = T_D1D ? T_D1D : DofQuadLimits::MAX_D1D;
   const int D1D = T_D1D ? T_D1D : pN;
   const int p_NE = D1D*D1D;
@@ -608,7 +729,7 @@ static void FindPointsLocal2DKernel(const int npt,
      // 3D1D for seed, 10D1D+6 for area, 3D1D+9 for edge
      constexpr int size1 = 10*MD1 + 6;
      constexpr int size2 = MD1*4;            // edge constraints
-      constexpr int size3 = MD1*MD1*DIM;  // local element coordinates
+      constexpr int size3 = MD1*MD1*MD1*DIM;  // local element coordinates

      MFEM_SHARED double r_workspace[size1];
      MFEM_SHARED findptsElementPoint_t el_pts[2];
@@ -1041,9 +1162,9 @@ void FindPointsGSLIB::FindPointsLocal2(const Vector &point_pos,
   auto pgslm = gsl_mesh.Read();
   auto pwt = DEV.wtend.Read();
   auto pbb = DEV.bb.Read();
-   auto plhm = DEV.lh_min.Read();
-   auto plhf = DEV.lh_fac.Read();
-   auto plho = DEV.lh_offset.ReadWrite();
+   auto plhm = DEV.loc_hash_min.Read();
+   auto plhf = DEV.loc_hash_fac.Read();
+   auto plho = DEV.loc_hash_offset.ReadWrite();
   auto pcode = code.Write();
   auto pelem = elem.Write();
   auto pref = ref.Write();
@@ -1054,49 +1175,32 @@ void FindPointsGSLIB::FindPointsLocal2(const Vector &point_pos,
   switch (DEV.dof1d)
   {
      case 2:
-         FindPointsLocal2DKernel<2>(npt, DEV.newt_tol, pp,
-                                    point_pos_ordering, pgslm,
-                                    NE_split_total, pwt, pbb,
-                                    DEV.lh_nx, plhm, plhf, plho,
-                                    pcode, pelem, pref, pdist,
-                                    pgll1d, plc);
-         break;
+         return FindPointsLocal2D_Kernel<2>(
+                   npt, DEV.newt_tol, pp, point_pos_ordering, pgslm, NE_split_total, pwt,
+                   pbb, DEV.h_nx, plhm, plhf, plho, pcode, pelem, pref, pdist,
+                   pgll1d, plc);
      case 3:
-         FindPointsLocal2DKernel<3>(npt, DEV.newt_tol, pp,
-                                    point_pos_ordering, pgslm,
-                                    NE_split_total, pwt, pbb,
-                                    DEV.lh_nx, plhm, plhf, plho,
-                                    pcode, pelem, pref, pdist,
-                                    pgll1d, plc);
-         break;
+         return FindPointsLocal2D_Kernel<3>(
+                   npt, DEV.newt_tol, pp, point_pos_ordering, pgslm, NE_split_total, pwt,
+                   pbb, DEV.h_nx, plhm, plhf, plho, pcode, pelem, pref, pdist,
+                   pgll1d, plc);
      case 4:
-         FindPointsLocal2DKernel<4>(npt, DEV.newt_tol, pp,
-                                    point_pos_ordering, pgslm,
-                                    NE_split_total, pwt, pbb,
-                                    DEV.lh_nx, plhm, plhf, plho,
-                                    pcode, pelem, pref, pdist,
-                                    pgll1d, plc);
-         break;
+         return FindPointsLocal2D_Kernel<4>(
+                   npt, DEV.newt_tol, pp, point_pos_ordering, pgslm, NE_split_total, pwt,
+                   pbb, DEV.h_nx, plhm, plhf, plho, pcode, pelem, pref, pdist,
+                   pgll1d, plc);
      case 5:
-         FindPointsLocal2DKernel<5>(npt, DEV.newt_tol, pp,
-                                    point_pos_ordering, pgslm,
-                                    NE_split_total, pwt, pbb,
-                                    DEV.lh_nx, plhm, plhf, plho,
-                                    pcode, pelem, pref, pdist,
-                                    pgll1d, plc);
-         break;
+         return FindPointsLocal2D_Kernel<5>(
+                   npt, DEV.newt_tol, pp, point_pos_ordering, pgslm, NE_split_total, pwt,
+                   pbb, DEV.h_nx, plhm, plhf, plho, pcode, pelem, pref, pdist,
+                   pgll1d, plc);
      default:
-         FindPointsLocal2DKernel(npt, DEV.newt_tol, pp,
-                                 point_pos_ordering, pgslm,
-                                 NE_split_total, pwt, pbb,
-                                 DEV.lh_nx, plhm, plhf, plho,
-                                 pcode, pelem, pref, pdist,
-                                 pgll1d, plc, DEV.dof1d);
-         break;
+         return FindPointsLocal2D_Kernel(npt, DEV.newt_tol, pp, point_pos_ordering,
+                                         pgslm, NE_split_total, pwt, pbb, DEV.h_nx,
+                                         plhm, plhf, plho, pcode, pelem,
+                                         pref, pdist, pgll1d, plc, DEV.dof1d);
   }
 }
-#undef DIM2
-#undef DIM
 #undef CODE_INTERNAL
 #undef CODE_BORDER
 #undef CODE_NOT_FOUND
@@ -11,7 +11,9 @@

 #include "../gslib.hpp"
 #include "../../general/forall.hpp"
-#include "gslib_kernel_helpers.hpp"
+#include "../../linalg/kernels.hpp"
+
+#include <climits>

 #ifdef MFEM_USE_GSLIB

@@ -57,15 +59,128 @@ struct findptsElemPt
   double x[DIM], jac[DIM * DIM], hes[18];
 };

-using dbl_range_t = gslib::dbl_range_t;
-using obbox_t = gslib::obbox_t<DIM>;
-using findptsLocalHashData_t = gslib::findptsLocalHashData_t<DIM>;
-using gslib::bbox_test;
-using gslib::hash_index;
-using gslib::l2norm2;
-using gslib::lag_eval_first_der;
-using gslib::lag_eval_second_der;
-using gslib::lin_solve_sym_2;
+struct dbl_range_t
+{
+   double min, max;
+};
+
+struct obbox_t
+{
+   double c0[DIM], A[DIM * DIM];
+   dbl_range_t x[DIM];
+};
+
+struct findptsLocalHashData_t
+{
+   int hash_n;
+   dbl_range_t bnd[DIM];
+   double fac[DIM];
+   unsigned int *offset;
+   // int max;
+};
+
+// Eval the ith Lagrange interpolant and its first derivative at x.
+// Note: lCoeff stores pre-computed coefficients for fast evaluation.
+static MFEM_HOST_DEVICE inline void lag_eval_first_der(double *p0, double x,
+                                                       int i, const double *z,
+                                                       const double *lCoeff,
+                                                       int pN)
+{
+   double u0 = 1, u1 = 0;
+   for (int j = 0; j < pN; ++j)
+   {
+      if (i != j)
+      {
+         double d_j = 2*(x-z[j]);
+         u1 = d_j*u1+u0;
+         u0 = d_j*u0;
+      }
+   }
+   p0[i] = lCoeff[i]*u0;
+   p0[pN+i] = 2.0*lCoeff[i]*u1;
+}
+
+// Eval the ith Lagrange interpolant and its first and second derivative at x.
+// Note: lCoeff stores pre-computed coefficients for fast evaluation.
+static MFEM_HOST_DEVICE inline void lag_eval_second_der(double *p0, double x,
+                                                        int i, const double *z,
+                                                        const double *lCoeff,
+                                                        int pN)
+{
+   double u0 = 1, u1 = 0, u2 = 0;
+   for (int j = 0; j < pN; ++j)
+   {
+      if (i != j)
+      {
+         double d_j = 2*(x-z[j]);
+         u2 = d_j*u2+u1;
+         u1 = d_j*u1+u0;
+         u0 = d_j*u0;
+      }
+   }
+   p0[i] = lCoeff[i]*u0;
+   p0[pN+i] = 2.0*lCoeff[i]*u1;
+   p0[2*pN+i] = 8.0*lCoeff[i]*u2;
+}
+
+// Axis-aligned bounding box test.
+static MFEM_HOST_DEVICE inline double AABB_test(const obbox_t *const b,
+                                                const double x[3])
+{
+   double b_d;
+   for (int d = 0; d < 3; ++d)
+   {
+      b_d = (x[d]-b->x[d].min)*(b->x[d].max-x[d]);
+      if (b_d < 0) { return b_d; }
+   }
+   return b_d;
+}
+
+// Axis-aligned bounding box test followed by oriented bounding-box test.
+static MFEM_HOST_DEVICE inline double bbox_test(const obbox_t *const b,
+                                                const double x[3])
+{
+   const double bxyz = AABB_test(b, x);
+   if (bxyz < 0)
+   {
+      return bxyz;
+   }
+   else
+   {
+      double dxyz[3];
+      for (int d = 0; d < 3; ++d)
+      {
+         dxyz[d] = x[d]-b->c0[d];
+      }
+      double test = 1;
+      for (int d = 0; d < 3; ++d)
+      {
+         double rst = 0;
+         for (int e = 0; e < 3; ++e)
+         {
+            rst += b->A[d*3+e]*dxyz[e];
+         }
+         double brst = (rst+1)*(1-rst);
+         test = test < 0 ? test : brst;
+      }
+      return test;
+   }
+}
+
+// Element index corresponding to hash mesh that the point is located in.
+static MFEM_HOST_DEVICE inline int hash_index(const findptsLocalHashData_t *p,
+                                              const double x[3])
+{
+   const int n = p->hash_n;
+   int sum = 0;
+   for (int d = 3-1; d >= 0; --d)
+   {
+      sum *= n;
+      int i = (int)floor((x[d]-p->bnd[d].min)*p->fac[d]);
+      sum += i < 0 ? 0 : (n-1 < i ? n-1 : i);
+   }
+   return sum;
+}

 // Solve Ax=y. A is row-major.
 static MFEM_HOST_DEVICE inline void lin_solve_3(double x[3], const double A[9],
@@ -84,6 +199,22 @@ static MFEM_HOST_DEVICE inline void lin_solve_3(double x[3], const double A[9],
   x[2] = idet*(inv6*y[0]+inv7*y[1]+inv8*y[2]);
 }

+// Solve Ax=y. A is a symmetric 2x2 matrix.
+static MFEM_HOST_DEVICE inline void lin_solve_sym_2(double x[2],
+                                                    const double A[3],
+                                                    const double y[2])
+{
+   const double idet = 1 / (A[0]*A[2]-A[1]*A[1]);
+   x[0] = idet*(A[2]*y[0]-A[1]*y[1]);
+   x[1] = idet*(A[0]*y[1]-A[1]*y[0]);
+}
+
+// L2 norm.
+static MFEM_HOST_DEVICE inline double l2norm2(const double x[3])
+{
+   return x[0]*x[0]+x[1]*x[1]+x[2]*x[2];
+}
+
 /* the bit structure of flags is CTTSSRR
   the C bit --- 1<<6 --- is set when the point is converged
   RR is 0 = 00b if r is unconstrained,
@@ -328,7 +459,7 @@ static MFEM_HOST_DEVICE bool reject_prior_step_q(findptsPt *res,
                                                 const findptsPt *p,
                                                 const double tol)
 {
-   const double dist2 = l2norm2<3>(resid);
+   const double dist2 = l2norm2(resid);
   const double decr = p->dist2-dist2;
   const double pred = p->dist2p;
   for (int d = 0; d < 3; ++d)
@@ -575,7 +706,7 @@ newton_vol_fin:
      int f = flags >> (2*dd) & 3u;
      res->r[dd] = f == 0 ? r0[dd]+dr[dd] : (f == 1 ? -1 : 1);
   }
-   res->flags = flags | ((p->flags & FLAG_MASK) << 7);
+   res->flags = flags | (p->flags << 7);
 }

 // Full Newton solve on the face. One of r/s/t is constrained.
@@ -758,7 +889,7 @@ newton_face_fin:
   res->r[dn] = p->r[dn];
   res->r[d1] = r[0];
   res->r[d2] = r[1];
-   res->flags = new_flags | ((p->flags & FLAG_MASK) << 7);
+   res->flags = new_flags | (p->flags << 7);
 }

 // Full Newton solve on the edge. Two of r/s/t are constrained.
@@ -842,8 +973,7 @@ newton_edge_fin:
   res->r[dn1] = p->r[dn1];
   res->r[dn2] = p->r[dn2];
   res->dist2p = -v;
-   res->flags = flags | new_flags | ((p->flags & FLAG_MASK) << 7);
-#undef EVAL
+   res->flags = flags | new_flags | (p->flags << 7);
 }

 // Find closest mesh node to the sought point.
@@ -1122,6 +1252,7 @@ static void FindPointsLocal3DKernel(const int npt,
                     case 0: // findpt_vol
                     {
                        double *wtr = r_workspace_ptr;
+
                        double *resid = wtr+6*D1D;
                        double *jac = resid+3;
                        double *resid_temp = jac+9;
@@ -1372,7 +1503,7 @@ static void FindPointsLocal3DKernel(const int npt,
                              // Hes_T is transposed version (i.e. in col major)
                              // n1*[2, 1, 1, 0, 0]
                              // j==1 => wt_j = wt+n1
-                              double *wt_j = wt+D1D*(2 - (row+1)/2);
+                              double *wt_j = wt+D1D*(2-(row+1) / 2);
                              const double *x = e_x[row+1][d];
                              hes_T[j] = 0.0;
                              for (int k = 0; k < D1D; ++k)
@@ -1391,6 +1522,7 @@ static void FindPointsLocal3DKernel(const int npt,
                              hes[j] += resid[d]*hes_T[j*3+d];
                           }
                        }
+
                        MFEM_SYNC_THREAD;

                        MFEM_FOREACH_THREAD(l,x,1)
@@ -1648,7 +1780,6 @@ static void FindPointsLocal3DKernel(const int npt,
         } //findpts_local
      } //elp
   });
-#undef MAXC
 }

 void FindPointsGSLIB::FindPointsLocal3(const Vector &point_pos,
@@ -1665,9 +1796,9 @@ void FindPointsGSLIB::FindPointsLocal3(const Vector &point_pos,
   auto pgslm = gsl_mesh.Read();
   auto pwt = DEV.wtend.Read();
   auto pbb = DEV.bb.Read();
-   auto plhm = DEV.lh_min.Read();
-   auto plhf = DEV.lh_fac.Read();
-   auto plho = DEV.lh_offset.ReadWrite();
+   auto plhm = DEV.loc_hash_min.Read();
+   auto plhf = DEV.loc_hash_fac.Read();
+   auto plho = DEV.loc_hash_offset.ReadWrite();
   auto pcode = code.Write();
   auto pelem = elem.Write();
   auto pref = ref.Write();
@@ -1678,36 +1809,33 @@ void FindPointsGSLIB::FindPointsLocal3(const Vector &point_pos,
   {
      case 2:
         FindPointsLocal3DKernel<2>(npt, DEV.newt_tol, pp, point_pos_ordering,
-                                    pgslm, NE_split_total, pwt, pbb,
-                                    DEV.lh_nx, plhm, plhf, plho,
-                                    pcode, pelem, pref, pdist, pgll1d, plc);
+                                    pgslm, NE_split_total, pwt, pbb, DEV.h_nx, plhm,
+                                    plhf, plho, pcode, pelem, pref, pdist, pgll1d,
+                                    plc);
         break;
      case 3:
         FindPointsLocal3DKernel<3>(npt, DEV.newt_tol, pp, point_pos_ordering,
-                                    pgslm, NE_split_total, pwt, pbb,
-                                    DEV.lh_nx, plhm, plhf, plho,
-                                    pcode, pelem, pref, pdist, pgll1d, plc);
+                                    pgslm, NE_split_total, pwt, pbb, DEV.h_nx, plhm,
+                                    plhf, plho, pcode, pelem, pref, pdist, pgll1d,
+                                    plc);
         break;
      case 4:
         FindPointsLocal3DKernel<4>(npt, DEV.newt_tol, pp, point_pos_ordering,
-                                    pgslm, NE_split_total, pwt, pbb,
-                                    DEV.lh_nx, plhm, plhf, plho,
-                                    pcode, pelem, pref, pdist, pgll1d, plc);
+                                    pgslm, NE_split_total, pwt, pbb, DEV.h_nx, plhm,
+                                    plhf, plho, pcode, pelem, pref, pdist, pgll1d,
+                                    plc);
         break;
      case 5:
         FindPointsLocal3DKernel<5>(npt, DEV.newt_tol, pp, point_pos_ordering,
-                                    pgslm, NE_split_total, pwt, pbb,
-                                    DEV.lh_nx, plhm, plhf, plho,
-                                    pcode, pelem, pref, pdist, pgll1d, plc);
+                                    pgslm, NE_split_total, pwt, pbb, DEV.h_nx, plhm,
+                                    plhf, plho, pcode, pelem, pref, pdist, pgll1d,
+                                    plc);
         break;
      default:
-         FindPointsLocal3DKernel(npt, DEV.newt_tol, pp,
-                                 point_pos_ordering, pgslm,
-                                 NE_split_total, pwt, pbb,
-                                 DEV.lh_nx, plhm, plhf, plho,
-                                 pcode, pelem, pref, pdist, pgll1d, plc,
+         FindPointsLocal3DKernel(npt, DEV.newt_tol, pp, point_pos_ordering, pgslm,
+                                 NE_split_total, pwt, pbb, DEV.h_nx, plhm, plhf,
+                                 plho, pcode, pelem, pref, pdist, pgll1d, plc,
                                 DEV.dof1d);
-         break;
   }
 }
 #undef pMax
@@ -1,656 +0,0 @@
-// Copyright (c) 2010-2025, Lawrence Livermore National Security, LLC. Produced
-// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-// LICENSE and NOTICE for details. LLNL-CODE-806117.
-//
-// This file is part of the MFEM library. For more information and source code
-// availability visit https://mfem.org.
-//
-// MFEM is free software; you can redistribute it and/or modify it under the
-// terms of the BSD-3 license. We welcome feedback and contributions, see file
-// CONTRIBUTING.md for details.
-
-#include "../gslib.hpp"
-#include "../../general/forall.hpp"
-#include "gslib_kernel_helpers.hpp"
-
-#ifdef MFEM_USE_GSLIB
-
-#ifdef MFEM_HAVE_GCC_PRAGMA_DIAGNOSTIC
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wunused-function"
-#endif
-#include "gslib.h"
-#ifndef GSLIB_RELEASE_VERSION //gslib v1.0.7
-#define GSLIB_RELEASE_VERSION 10007
-#endif
-#ifdef MFEM_HAVE_GCC_PRAGMA_DIAGNOSTIC
-#pragma GCC diagnostic pop
-#endif
-
-namespace mfem
-{
-#if GSLIB_RELEASE_VERSION >= 10009
-#define CODE_INTERNAL 0
-#define CODE_BORDER 1
-#define CODE_NOT_FOUND 2
-#define sDIM 2
-#define sDIM2 4
-#define rDIM 1
-
-struct findptsElementPoint_t
-{
-   double x[sDIM], r, oldr, dist2, dist2p, tr;
-   int flags;
-};
-
-struct findptsElementGEdge_t
-{
-   double *x[sDIM];
-};
-
-struct findptsElementGPT_t
-{
-   double x[sDIM], jac[sDIM*rDIM], hes[sDIM*rDIM];
-};
-
-using dbl_range_t = gslib::dbl_range_t;
-using obbox_t = gslib::obbox_t<sDIM>;
-using findptsLocalHashData_t = gslib::findptsLocalHashData_t<sDIM>;
-using gslib::AABB_test;
-using gslib::bbox_test;
-using gslib::hash_index;
-using gslib::l2norm2;
-using gslib::lag_eval_second_der;
-
-/* the bit structure of flags is CRR
-   the C bit --- 1<<2 --- is set when the point is converged
-   RR is 0 = 00b if r is unconstrained,
-         1 = 01b if r is constrained at -1, i.e., rmin
-         2 = 10b if r is constrained at +1, i.e., rmax
-*/
-
-#define CONVERGED_FLAG (1u<<2)
-#define FLAG_MASK 0x07u // = 111b
-
-/* returns 1 if r direction (the only free direction in 2D) is constrained.
-   returns 1 if either 1st or 2nd bit of flags is set.
-*/
-static MFEM_HOST_DEVICE inline int num_constrained(const int flags)
-{
-   return ((flags | flags>>1) & 1u);
-}
-
-/* pi=0, r=-1; pi=1, r=+1 */
-static MFEM_HOST_DEVICE inline int point_index(const int x)
-{
-   return ((x>>1) & 1u);
-}
-
-/* check reduction in objective against prediction, and adjust
-   trust region radius (p->tr) accordingly;
-   may reject the prior step, returning 1; otherwise returns 0
-   sets out_pt->dist2, out_pt->index, out_pt->x, out_pt->oldr in any event,
-   leaving out_pt->r, out_pt->dr, out_pt->flags to be set when returning 0 */
-static MFEM_HOST_DEVICE bool reject_prior_step_q(findptsElementPoint_t *out_pt,
-                                                 const double resid[2],
-                                                 const findptsElementPoint_t *p,
-                                                 const double tol)
-{
-   const double dist2 = l2norm2<2>(resid);
-   const double decr = p->dist2 - dist2;
-   const double pred = p->dist2p;
-   out_pt->x[0] = p->x[0];
-   out_pt->x[1] = p->x[1];
-   out_pt->oldr = p->r;
-   out_pt->dist2 = dist2;
-   if (decr >= 0.01*pred)
-   {
-      if (decr >= 0.9*pred)   // very good iteration
-      {
-         out_pt->tr = p->tr*2;
-      }
-      else                    // somewhat good iteration
-      {
-         out_pt->tr = p->tr;
-      }
-      return false;
-   }
-   else
-   {
-      /* reject step; note: the point will pass through this routine
-         again, and we set things up here so it gets classed as a
-         "very good iteration" --- this doubles the trust radius,
-         which is why we divide by 4 below */
-      double v0 = fabs(p->r - p->oldr);
-      out_pt->tr = v0/4.0;
-      out_pt->dist2 = p->dist2;
-      out_pt->r = p->oldr;
-      out_pt->flags = p->flags>>3;
-      out_pt->dist2p = -HUGE_VAL;
-      if (pred < dist2*tol)
-      {
-         out_pt->flags |= CONVERGED_FLAG;
-      }
-      return true;
-   }
-}
-
-static MFEM_HOST_DEVICE inline void newton_edge( findptsElementPoint_t *const
-                                                 out_pt,
-                                                 const double jac[2],
-                                                 const double rhess,
-                                                 const double resid[2],
-                                                 int flags,
-                                                 const findptsElementPoint_t *const p,
-                                                 const double tol )
-{
-   const double tr = p->tr;
-   const double A = jac[0] * jac[0] + jac[1] * jac[1] -
-                    rhess; // A = J^T J - resid_d H_d
-   const double y = jac[0]*resid[0] + jac[1]*resid[1];        // y = J^T resid
-
-   const double oldr = p->r;
-   double dr, newr, tdr, tnewr, v, tv;
-   int new_flags=0, tnew_flags=0;
-
-#define EVAL(dr) ( (dr*A - 2*y) * dr )
-   if (A>0)
-   {
-      dr = y/A;
-      if (fabs(dr)<tol)
-      {
-         dr=0.0;
-         newr = oldr;
-      }
-      else
-      {
-         newr = oldr+dr;
-      }
-
-      if (fabs(dr)<tr && fabs(newr)<1)
-      {
-         v = EVAL(dr);
-         goto newton_edge_fin;
-      }
-   }
-
-   if ((newr=oldr-tr) > -1)
-   {
-      dr = -tr;
-   }
-   else
-   {
-      newr = -1, dr = -1-oldr, new_flags = flags|1u;
-   }
-   v = EVAL(dr);
-
-   if ((tnewr=oldr+tr) < 1)
-   {
-      tdr = tr;
-   }
-   else
-   {
-      tnewr = 1, tdr = 1-oldr, tnew_flags = flags|2u;
-   }
-   tv = EVAL(tdr);
-
-   if (tv<v)
-   {
-      newr = tnewr, dr = tdr, v = tv, new_flags = tnew_flags;
-   }
-#undef EVAL
-
-newton_edge_fin:
-   // check convergence by testing if change in r is less than tol
-   if (fabs(dr)<tol)
-   {
-      new_flags |= CONVERGED_FLAG;
-   }
-   out_pt->r = newr;
-   out_pt->dist2p = -v;
-   out_pt->flags = flags | new_flags | ((p->flags & FLAG_MASK)<<3);
-}
-
-static MFEM_HOST_DEVICE void seed_j( const double *elx[sDIM],
-                                     const double x[sDIM],
-                                     const double *z,
-                                     double       *dist2,
-                                     double       *r,
-                                     const int    ir,
-                                     const int    pN )
-{
-   double dx[sDIM];
-   for (int d=0; d<sDIM; ++d)
-   {
-      dx[d] = x[d] - elx[d][ir];
-   }
-   dist2[ir] = HUGE_VAL;
-   const double dist2_rs = l2norm2(dx);
-   if (dist2[ir]>dist2_rs)
-   {
-      dist2[ir] = dist2_rs;
-      r[ir] = z[ir];
-   }
-}
-
-template<int T_D1D = 0>
-static void FindPointsEdgeLocal2DKernel( const int     npt,
-                                         const double  tol,
-                                         const double  dist2tol,
-                                         const double  *x,
-                                         const int     point_pos_ordering,
-                                         const double  *xElemCoord,
-                                         const int     nel,
-                                         const double  *wtend,
-                                         const double  *boxinfo,
-                                         const bool    obb_check,
-                                         const int     hash_n,
-                                         const double  *hashMin,
-                                         const double  *hashFac,
-                                         unsigned int  *hashOffset,
-                                         unsigned int *const code_base,
-                                         unsigned int *const el_base,
-                                         double *const r_base,
-                                         double *const dist2_base,
-                                         const double  *gll1D,
-                                         const double  *lagcoeff,
-                                         const int     pN = 0 )
-{
-   const int MD1   = T_D1D ? T_D1D : DofQuadLimits::MAX_D1D;
-   const int D1D   = T_D1D ? T_D1D : pN;
-   const int p_NEL = nel*D1D;
-   MFEM_VERIFY(MD1<=DofQuadLimits::MAX_D1D,
-               "Increase Max allowable polynomial order.");
-   MFEM_VERIFY(pN<=DofQuadLimits::MAX_D1D,
-               "Increase Max allowable polynomial order.");
-   MFEM_VERIFY(D1D!=0, "Polynomial order not specified.");
-   const int nThreads = D1D*sDIM;
-
-   mfem::forall_2D(npt, nThreads, 1, [=] MFEM_HOST_DEVICE (int i)
-   {
-      // 2D1D for seed, 3D1D + 7 for edge
-      constexpr int size1 = 3*MD1 + 7;
-      // edge coordinates = D1D*2
-      constexpr int size2 = 2*MD1;
-      // local element coordinates in shared memory
-      constexpr int size3 = MD1*sDIM;
-
-      MFEM_SHARED findptsElementPoint_t el_pts[2];
-      MFEM_SHARED double r_workspace[size1];
-
-      MFEM_SHARED double constraint_workspace[size2];
-
-      MFEM_SHARED double elem_coords[MD1 <= 6 ? size3 : 1];
-
-      double *r_workspace_ptr = r_workspace;
-      findptsElementPoint_t *fpt, *tmp;
-      fpt = el_pts + 0;
-      tmp = el_pts + 1;
-
-      // x and y coord index within point_pos for point i
-      int id_x = point_pos_ordering == 0 ? i     : i*sDIM;
-      int id_y = point_pos_ordering == 0 ? i+npt : i*sDIM+1;
-      double x_i[2] = {x[id_x], x[id_y]};
-
-      unsigned int *code_i = code_base  + i;
-      double *dist2_i      = dist2_base + i;
-
-      //---------------- map_points_to_els --------------------
-      findptsLocalHashData_t hash;
-      for (int d=0; d<sDIM; ++d)
-      {
-         hash.bnd[d].min = hashMin[d];
-         hash.fac[d] = hashFac[d];
-      }
-      hash.hash_n = hash_n;
-      hash.offset = hashOffset;
-
-      const int hi                  = hash_index(&hash, x_i);
-      const unsigned int *elp       = hash.offset + hash.offset[hi];
-      const unsigned int *const ele = hash.offset + hash.offset[hi+1];
-      *code_i  = CODE_NOT_FOUND;
-      *dist2_i = HUGE_VAL;
-
-      for (; elp!=ele; ++elp)
-      {
-         const unsigned int el = *elp;
-
-         const int n_box_ents = obb_check ? (3*sDIM + sDIM2) : (2*sDIM);
-         bool pass_bb = true;
-         obbox_t box;
-         if (obb_check)
-         {
-            for (int idx = 0; idx < sDIM; ++idx)
-            {
-               box.c0[idx] = boxinfo[n_box_ents*el + idx];
-               box.x[idx].min = boxinfo[n_box_ents*el + sDIM + idx];
-               box.x[idx].max = boxinfo[n_box_ents*el + 2*sDIM + idx];
-            }
-            for (int idx = 0; idx < sDIM2; ++idx)
-            {
-               box.A[idx] = boxinfo[n_box_ents*el + 3*sDIM + idx];
-            }
-            pass_bb = (bbox_test(&box, x_i) >= 0);
-         }
-         else
-         {
-            for (int d = 0; d < sDIM; ++d)
-            {
-               box.x[d].min = boxinfo[n_box_ents*el + d];
-               box.x[d].max = boxinfo[n_box_ents*el + sDIM + d];
-            }
-            pass_bb = (AABB_test(&box, x_i) >= 0);
-         }
-
-         if (pass_bb)
-         {
-            //------------ findpts_local ------------------
-            {
-               if (MD1 <= 6)
-               {
-                  MFEM_FOREACH_THREAD(j,x,D1D*sDIM)
-                  {
-                     const int qp = j % D1D;
-                     const int d = j / D1D;
-                     elem_coords[qp + d*D1D] =
-                        xElemCoord[qp + el*D1D + d*p_NEL];
-                  }
-                  MFEM_SYNC_THREAD;
-               }
-
-               const double *elx[sDIM];
-               for (int d=0; d<sDIM; d++)
-               {
-                  elx[d] = MD1<= 6 ? &elem_coords[d*D1D] :
-                           xElemCoord + d*p_NEL + el*D1D;
-               }
-               MFEM_SYNC_THREAD;
-               //// findpts_el ////
-               {
-                  MFEM_FOREACH_THREAD(j,x,1)
-                  {
-                     fpt->dist2 = HUGE_VAL;
-                     fpt->dist2p = 0;
-                     fpt->tr = 1;
-                  }
-                  MFEM_FOREACH_THREAD(j,x,sDIM)
-                  {
-                     fpt->x[j] = x_i[j];
-                  }
-                  MFEM_SYNC_THREAD;
-
-                  {
-                     double *dist2_temp = r_workspace_ptr;
-                     double *r_temp = dist2_temp + D1D;
-                     MFEM_FOREACH_THREAD(j,x,D1D)
-                     {
-                        seed_j(elx, x_i, gll1D, dist2_temp, r_temp, j, D1D);
-                     }
-                     MFEM_SYNC_THREAD;
-
-                     MFEM_FOREACH_THREAD(j,x,1)
-                     {
-                        for (int ir=0; ir<D1D; ++ir)
-                        {
-                           if (dist2_temp[ir]<fpt->dist2)
-                           {
-                              fpt->dist2 = dist2_temp[ir];
-                              fpt->r = r_temp[ir];
-                           }
-                        }
-                     }
-                     MFEM_SYNC_THREAD;
-                  } //seed done
-
-                  // Initialize tmp struct with fpt values before starting Newton iterations
-                  MFEM_FOREACH_THREAD(j,x,1)
-                  {
-                     tmp->dist2 = HUGE_VAL;
-                     tmp->dist2p = 0;
-                     tmp->tr = 1;
-                     tmp->flags = 0;
-                     tmp->r = fpt->r;
-                  }
-                  MFEM_FOREACH_THREAD(j,x,sDIM)
-                  {
-                     tmp->x[j] = fpt->x[j];
-                  }
-                  MFEM_SYNC_THREAD;
-
-
-                  for (int step=0; step<50; step++)
-                  {
-                     int nc = num_constrained(tmp->flags & FLAG_MASK);
-                     switch (nc)
-                     {
-                        case 0:
-                        {
-                           double *wt = r_workspace_ptr;
-                           double *resid = wt + 3*D1D;
-                           double *jac = resid + sDIM;
-                           double *hess = jac + sDIM*rDIM;
-
-                           findptsElementGEdge_t edge;
-                           for (int d=0; d<sDIM; ++d)
-                           {
-                              edge.x[d] = constraint_workspace + d*D1D;
-                           }
-                           MFEM_FOREACH_THREAD(j,x,D1D)
-                           {
-                              for (int d=0; d<sDIM; ++d)
-                              {
-                                 edge.x[d][j] = elx[d][j];
-                              }
-                           }
-                           MFEM_SYNC_THREAD;
-
-                           // compute basis function info upto 2nd derivative
-                           MFEM_FOREACH_THREAD(j,x,D1D)
-                           {
-                              lag_eval_second_der(wt, tmp->r, j, gll1D,
-                                                  lagcoeff, D1D);
-                           }
-                           MFEM_SYNC_THREAD;
-
-                           MFEM_FOREACH_THREAD(j,x,sDIM)
-                           {
-                              resid[j] = tmp->x[j];
-                              jac[j] = 0.0;
-                              hess[j] = 0.0;
-                              for (int k=0; k<D1D; ++k)
-                              {
-                                 resid[j] -= wt[      k]*edge.x[j][k];
-                                 jac[j]   += wt[D1D+k]*edge.x[j][k];
-                                 hess[j]  += wt[2*D1D+k]*edge.x[j][k];
-                              }
-                           }
-                           MFEM_SYNC_THREAD;
-
-                           MFEM_FOREACH_THREAD(j,x,1)
-                           {
-                              hess[2] = resid[0]*hess[0] + resid[1]*hess[1];
-                           }
-                           MFEM_SYNC_THREAD;
-
-                           MFEM_FOREACH_THREAD(j,x,1)
-                           {
-                              if (!reject_prior_step_q(fpt, resid, tmp, tol))
-                              {
-                                 newton_edge(fpt, jac, hess[2], resid,
-                                             tmp->flags & FLAG_MASK, tmp, tol);
-                              }
-                           }
-                           MFEM_SYNC_THREAD;
-                           break;
-                        }
-                        case 1:    // r is constrained to either -1 or 1
-                        {
-                           MFEM_FOREACH_THREAD(j,x,1)
-                           {
-                              const int pi = point_index(tmp->flags &
-                                                         FLAG_MASK);
-                              const double *wt   = wtend + pi*3*D1D;
-                              findptsElementGPT_t gpt;
-                              for (int d=0; d<sDIM; ++d)
-                              {
-                                 gpt.x[d] = elx[d][pi*(D1D-1)];
-                                 gpt.jac[d] = 0.0;
-                                 gpt.hes[d] = 0.0;
-                                 for (int k=0; k<D1D; ++k)
-                                 {
-                                    gpt.jac[d] += wt[D1D  +k]*elx[d][k];
-                                    gpt.hes[d] += wt[2*D1D+k]*elx[d][k];
-                                 }
-                              }
-
-                              const double *const pt_x = gpt.x;
-                              const double *const jac = gpt.jac;
-                              const double *const hes = gpt.hes;
-                              double resid[sDIM], steep, sr;
-                              resid[0] = fpt->x[0] - pt_x[0];
-                              resid[1] = fpt->x[1] - pt_x[1];
-                              steep = jac[0]*resid[0] + jac[1]*resid[1];
-                              sr = steep*tmp->r;
-                              if ( !reject_prior_step_q(fpt, resid, tmp, tol) )
-                              {
-                                 if (sr<0)
-                                 {
-                                    const double rhess = resid[0]*hes[0] +
-                                                         resid[1]*hes[1];
-                                    newton_edge(fpt, jac, rhess,
-                                                resid, 0, tmp, tol);
-                                 }
-                                 else // sr==0
-                                 {
-                                    fpt->r = tmp->r;
-                                    fpt->dist2p = 0;
-                                    fpt->flags = tmp->flags | CONVERGED_FLAG;
-                                 }
-                              }
-                           }
-                           MFEM_SYNC_THREAD;
-                           break;
-                        } // case 1
-                     } //switch
-                     if (fpt->flags & CONVERGED_FLAG)
-                     {
-                        break;
-                     }
-                     MFEM_SYNC_THREAD;
-                     MFEM_FOREACH_THREAD(j,x,1)
-                     {
-                        *tmp = *fpt;
-                     }
-                     MFEM_SYNC_THREAD;
-                  } //for int step<50
-               } //findpts_el
-
-               bool converged_internal =
-                  ((fpt->flags&FLAG_MASK) == CONVERGED_FLAG) &&
-                  (fpt->dist2<dist2tol);
-
-               if (*code_i == CODE_NOT_FOUND || converged_internal ||
-                   fpt->dist2 < *dist2_i)
-               {
-                  MFEM_FOREACH_THREAD(j,x,1)
-                  {
-                     *(el_base+i) = el;
-                     *code_i = converged_internal ? CODE_INTERNAL : CODE_BORDER;
-                     *dist2_i = fpt->dist2;
-                     *(r_base+i) = fpt->r;
-                  }
-                  MFEM_SYNC_THREAD;
-                  if (converged_internal)
-                  {
-                     break;
-                  }
-               }
-            } //findpts_local
-         } //obbox_test
-      } //elp
-   });
-}
-
-void FindPointsGSLIB::FindPointsEdgeLocal2( const Vector &point_pos,
-                                            int point_pos_ordering,
-                                            Array<unsigned int> &code,
-                                            Array<unsigned int> &elem,
-                                            Vector &ref,
-                                            Vector &dist,
-                                            int npt )
-{
-   if (npt==0)
-   {
-      return;
-   }
-   MFEM_VERIFY(dim==1 && spacedim==2,"Function for 2D edges only");
-   bool use_dev = point_pos.UseDevice();
-   auto pp = point_pos.Read(use_dev);
-   auto pgslm = gsl_mesh.Read(use_dev);
-   auto pwt = DEV.wtend.Read(use_dev);
-   auto pbb = DEV.bb.Read(use_dev);
-   auto plhm = DEV.lh_min.Read(use_dev);
-   auto plhf = DEV.lh_fac.Read(use_dev);
-   auto plho = DEV.lh_offset.ReadWrite(use_dev);
-   auto pcode = code.Write(use_dev);
-   auto pelem = elem.Write(use_dev);
-   auto pref = ref.Write(use_dev);
-   auto pdist = dist.Write(use_dev);
-   auto pgll1d = DEV.gll1d.ReadWrite(use_dev);
-   auto plc = DEV.lagcoeff.Read(use_dev);
-   double dist2tol = DEV.surf_dist_tol;
-   const bool obb_chk = obb_check;
-   switch (DEV.dof1d)
-   {
-      case 2:
-         FindPointsEdgeLocal2DKernel<2>(npt, DEV.newt_tol, dist2tol,
-                                        pp, point_pos_ordering, pgslm,
-                                        NE_split_total, pwt, pbb, obb_chk,
-                                        DEV.lh_nx, plhm, plhf, plho,
-                                        pcode, pelem, pref, pdist,
-                                        pgll1d, plc);
-         break;
-      case 3:
-         FindPointsEdgeLocal2DKernel<3>(npt, DEV.newt_tol, dist2tol,
-                                        pp, point_pos_ordering, pgslm,
-                                        NE_split_total, pwt, pbb, obb_chk,
-                                        DEV.lh_nx, plhm, plhf, plho,
-                                        pcode, pelem, pref, pdist,
-                                        pgll1d, plc);
-         break;
-      case 4:
-         FindPointsEdgeLocal2DKernel<4>(npt, DEV.newt_tol, dist2tol,
-                                        pp, point_pos_ordering, pgslm,
-                                        NE_split_total, pwt, pbb, obb_chk,
-                                        DEV.lh_nx, plhm, plhf, plho,
-                                        pcode, pelem, pref, pdist,
-                                        pgll1d, plc);
-         break;
-      default:
-         FindPointsEdgeLocal2DKernel(npt, DEV.newt_tol, dist2tol, pp,
-                                     point_pos_ordering, pgslm,
-                                     NE_split_total, pwt, pbb, obb_chk,
-                                     DEV.lh_nx, plhm, plhf, plho,
-                                     pcode, pelem, pref, pdist,
-                                     pgll1d, plc, DEV.dof1d);
-         break;
-   }
-}
-#undef sDIM
-#undef rDIM
-#undef sDIM2
-#undef CODE_INTERNAL
-#undef CODE_BORDER
-#undef CODE_NOT_FOUND
-#else
-void FindPointsGSLIB::FindPointsEdgeLocal2( const Vector &point_pos,
-                                            int point_pos_ordering,
-                                            Array<unsigned int> &code,
-                                            Array<unsigned int> &elem,
-                                            Vector &ref,
-                                            Vector &dist,
-                                            int npt ) {} ;
-#endif
-} // namespace mfem
-
-#endif //ifdef MFEM_USE_GSLIB
@@ -1,661 +0,0 @@
-// Copyright (c) 2010-2025, Lawrence Livermore National Security, LLC. Produced
-// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-// LICENSE and NOTICE for details. LLNL-CODE-806117.
-//
-// This file is part of the MFEM library. For more information and source code
-// availability visit https://mfem.org.
-//
-// MFEM is free software; you can redistribute it and/or modify it under the
-// terms of the BSD-3 license. We welcome feedback and contributions, see file
-// CONTRIBUTING.md for details.
-
-#include "../gslib.hpp"
-#include "../../general/forall.hpp"
-#include "gslib_kernel_helpers.hpp"
-
-#ifdef MFEM_USE_GSLIB
-
-
-#ifdef MFEM_HAVE_GCC_PRAGMA_DIAGNOSTIC
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wunused-function"
-#endif
-#include "gslib.h"
-#ifndef GSLIB_RELEASE_VERSION //gslib v1.0.7
-#define GSLIB_RELEASE_VERSION 10007
-#endif
-#ifdef MFEM_HAVE_GCC_PRAGMA_DIAGNOSTIC
-#pragma GCC diagnostic pop
-#endif
-
-namespace mfem
-{
-#if GSLIB_RELEASE_VERSION >= 10009
-#define CODE_INTERNAL 0
-#define CODE_BORDER 1
-#define CODE_NOT_FOUND 2
-#define sDIM 3
-#define rDIM 1
-#define sDIM2 (sDIM*sDIM)
-#define rDIM2 (rDIM*rDIM)
-
-struct findptsElementPoint_t
-{
-   double x[sDIM], r, oldr, dist2, dist2p, tr;
-   int flags;
-};
-
-struct findptsElementGEdge_t
-{
-   double *x[sDIM], *dxdn[sDIM], *d2xdn[sDIM];
-};
-
-struct findptsElementGPT_t
-{
-   double x[sDIM], jac[sDIM], hes[sDIM*(1+1)];
-};
-
-using dbl_range_t = gslib::dbl_range_t;
-using obbox_t = gslib::obbox_t<sDIM>;
-using findptsLocalHashData_t = gslib::findptsLocalHashData_t<sDIM>;
-using gslib::AABB_test;
-using gslib::bbox_test;
-using gslib::hash_index;
-using gslib::l2norm2;
-using gslib::lag_eval_second_der;
-
-/* the bit structure of flags is CRR
-   the C bit --- 1<<2 --- is set when the point is converged
-   RR is 0 = 00b if r is unconstrained,
-         1 = 01b if r is constrained at -1, i.e., rmin
-         2 = 10b if r is constrained at +1, i.e., rmax
-*/
-#define CONVERGED_FLAG (1u<<2)
-#define FLAG_MASK 0x07u
-
-/* returns the number of constrained reference coordinates, max 1
-*/
-static MFEM_HOST_DEVICE inline int num_constrained(const int flags)
-{
-   return ((flags | flags>>1) & 1u);
-}
-
-static MFEM_HOST_DEVICE inline int point_index(const int x)
-{
-   return ((x>>1)&1u);
-}
-
-/* check reduction in objective against prediction, and adjust
-   trust region radius (p->tr) accordingly;
-   may reject the prior step, returning 1; otherwise returns 0
-   sets out_pt->dist2, out_pt->index, out_pt->x, out_pt->oldr in any event,
-   leaving out_pt->r, out_pt->dr, out_pt->flags to be set when returning 0 */
-static MFEM_HOST_DEVICE bool reject_prior_step_q(findptsElementPoint_t *out_pt,
-                                                 const double resid[3],
-                                                 const findptsElementPoint_t *p,
-                                                 const double tol)
-{
-   const double dist2 = l2norm2<sDIM>(resid);
-   const double decr  = p->dist2 - dist2;
-   const double pred  = p->dist2p;
-   for (int d=0; d<sDIM; ++d)
-   {
-      out_pt->x[d] = p->x[d];
-   }
-   out_pt->oldr = p->r;
-   out_pt->dist2 = dist2;
-   if (decr>=0.01*pred)
-   {
-      if (decr>=0.9*pred)   // very good iteration
-      {
-         out_pt->tr = 2*p->tr;
-      }
-      else   // good iteration
-      {
-         out_pt->tr = p->tr;
-      }
-      return false;
-   }
-   else   // if the iteration in not good
-   {
-      /* reject step; note: the point will pass through this routine
-         again, and we set things up here so it gets classed as a
-         "very good iteration" --- this doubles the trust radius,
-         which is why we divide by 4 below */
-      double v0 = fabs(p->r - p->oldr);
-      out_pt->tr = v0/4.0;
-      out_pt->dist2 = p->dist2;
-      out_pt->r = p->oldr;
-      out_pt->flags = p->flags>>3;
-      out_pt->dist2p = -HUGE_VAL;
-      if (pred<dist2*tol)
-      {
-         out_pt->flags |= CONVERGED_FLAG;
-      }
-      return true;
-   }
-}
-
-static MFEM_HOST_DEVICE inline void newton_edge(findptsElementPoint_t *const
-                                                out_pt,
-                                                const double jac[sDIM*rDIM],
-                                                const double rhes,
-                                                const double resid[sDIM],
-                                                int flags,
-                                                const findptsElementPoint_t *const p,
-                                                const double tol)
-{
-   const double tr = p->tr;
-   /* A = J^T J - resid_d H_d */
-   const double A = jac[0]*jac[0]+ jac[1] * jac[1] + jac[2] * jac[2]
-                    - rhes;
-   /* y = J^T r */
-   const double y = jac[0]*resid[0] + jac[1]*resid[1] + jac[0+2]*resid[2];
-
-   const double oldr = p->r;
-   double dr, nr, tdr, tnr;
-   double v, tv;
-   int new_flags = 0, tnew_flags = 0;
-
-#define EVAL(dr) (dr*A - 2*y)*dr
-
-   /* if A is not SPD, quadratic model has no minimum */
-   if (A>0)
-   {
-      dr = y/A;
-
-      if (fabs(dr)<tol)
-      {
-         dr=0.0;
-         nr = oldr;
-      }
-      else
-      {
-         nr = oldr+dr;
-      }
-      if ( fabs(dr)<tr && fabs(nr)<1 )
-      {
-         v = EVAL(dr);
-         goto newton_edge_fin;
-      }
-   }
-
-   if ( (nr=oldr-tr)>-1 )
-   {
-      dr = -tr;
-   }
-   else
-   {
-      nr = -1, dr = -1-oldr, new_flags = flags | 1u;
-   }
-   v = EVAL(dr);
-
-   if ( (tnr = oldr+tr)<1 )
-   {
-      tdr = tr;
-   }
-   else
-   {
-      tnr = 1, tdr = 1-oldr, tnew_flags = flags | 2u;
-   }
-   tv = EVAL(tdr);
-
-   if (tv<v)
-   {
-      nr = tnr, dr = tdr, v = tv, new_flags = tnew_flags;
-   }
-
-newton_edge_fin:
-   /* check convergence */
-   if ( fabs(dr)<tol )
-   {
-      new_flags |= CONVERGED_FLAG;
-   }
-   out_pt->r = nr;
-   out_pt->dist2p = -v;
-   out_pt->flags = flags | new_flags | ((p->flags & FLAG_MASK)<<3);
-#undef EVAL
-}
-
-static MFEM_HOST_DEVICE void seed_j(const double *elx[sDIM],
-                                    const double x[sDIM],
-                                    const double *z,
-                                    double *dist2,
-                                    double *r,
-                                    const int ir,
-                                    const int pN)
-{
-   if (ir>=pN)
-   {
-      return;
-   }
-
-   double dx[sDIM];
-   for (int d=0; d<sDIM; ++d)
-   {
-      dx[d] = x[d] - elx[d][ir];
-   }
-   dist2[ir] = l2norm2(dx);
-   r[ir] = z[ir];
-}
-
-template<int T_D1D = 0>
-static void FindPointsEdgeLocal3DKernel(const int npt,
-                                        const double tol,
-                                        const double dist2tol,
-                                        const double *x,
-                                        const int point_pos_ordering,
-                                        const double *xElemCoord,
-                                        const int nel,
-                                        const double *wtend,
-                                        const double *boxinfo,
-                                        const bool obb_check,
-                                        const int hash_n,
-                                        const double *hashMin,
-                                        const double *hashFac,
-                                        unsigned int *hashOffset,
-                                        unsigned int *const code_base,
-                                        unsigned int *const el_base,
-                                        double *const r_base,
-                                        double *const dist2_base,
-                                        const double *gll1D,
-                                        const double *lagcoeff,
-                                        const int pN = 0)
-{
-   const int MD1   = T_D1D ? T_D1D : DofQuadLimits::MAX_D1D;
-   const int D1D   = T_D1D ? T_D1D : pN;
-   const int p_NEL = nel*D1D;
-   MFEM_VERIFY(MD1<=DofQuadLimits::MAX_D1D,
-               "Increase Max allowable polynomial order.");
-   MFEM_VERIFY(pN<=DofQuadLimits::MAX_D1D,
-               "Increase Max allowable polynomial order.");
-   MFEM_VERIFY(D1D!=0, "Polynomial order not specified.");
-   const int nThreads = D1D*sDIM;
-
-   mfem::forall_2D(npt, nThreads, 1, [=] MFEM_HOST_DEVICE (int i)
-   {
-      constexpr int size1 = 3*MD1 + 13;
-      constexpr int size2 = 3*MD1;
-      constexpr int size3 = MD1*sDIM;
-
-      MFEM_SHARED findptsElementPoint_t el_pts[2];
-      MFEM_SHARED double r_workspace[size1];
-
-      MFEM_SHARED double constraint_workspace[size2];
-
-      MFEM_SHARED double elem_coords[MD1 <= 6 ? size3 : 1];
-
-      double *r_workspace_ptr = r_workspace;
-      findptsElementPoint_t *fpt, *tmp;
-      fpt = el_pts + 0;
-      tmp = el_pts + 1;
-
-      int id_x = point_pos_ordering==0 ?       i :   i*sDIM;
-      int id_y = point_pos_ordering==0 ?   npt+i : 1+i*sDIM;
-      int id_z = point_pos_ordering==0 ? 2*npt+i : 2+i*sDIM;
-      double x_i[3] = {x[id_x], x[id_y], x[id_z]};
-
-      unsigned int *code_i = code_base  + i;
-      double *dist2_i      = dist2_base + i;
-
-      //// map_points_to_els ////
-      findptsLocalHashData_t hash;
-      for (int d=0; d<sDIM; ++d)
-      {
-         hash.bnd[d].min = hashMin[d];
-         hash.fac[d]     = hashFac[d];
-      }
-      hash.hash_n = hash_n;
-      hash.offset = hashOffset;
-
-      const unsigned int hi   = hash_index(&hash, x_i);
-      const unsigned int *elp       = hash.offset + hash.offset[hi];
-      const unsigned int *const ele = hash.offset + hash.offset[hi+1];
-      *code_i  = CODE_NOT_FOUND;
-      *dist2_i = HUGE_VAL;
-
-      for (; elp!=ele; ++elp)
-      {
-         const unsigned int el = *elp;
-
-         const int n_box_ents = obb_check ? (3*sDIM + sDIM2) : (2*sDIM);
-         bool pass_bb = true;
-         obbox_t box;
-         if (obb_check)
-         {
-            for (int idx = 0; idx < sDIM; ++idx)
-            {
-               box.c0[idx] = boxinfo[n_box_ents*el + idx];
-               box.x[idx].min = boxinfo[n_box_ents*el + sDIM + idx];
-               box.x[idx].max = boxinfo[n_box_ents*el + 2*sDIM + idx];
-            }
-            for (int idx = 0; idx < sDIM2; ++idx)
-            {
-               box.A[idx] = boxinfo[n_box_ents*el + 3*sDIM + idx];
-            }
-            pass_bb = (bbox_test(&box, x_i) >= 0);
-         }
-         else
-         {
-            for (int d = 0; d < sDIM; ++d)
-            {
-               box.x[d].min = boxinfo[n_box_ents*el + d];
-               box.x[d].max = boxinfo[n_box_ents*el + sDIM + d];
-            }
-            pass_bb = (AABB_test(&box, x_i) >= 0);
-         }
-
-         if (pass_bb)
-         {
-            //// findpts_local ////
-            {
-               if (MD1 <= 6)
-               {
-                  MFEM_FOREACH_THREAD(j,x,D1D*sDIM)
-                  {
-                     const int qp = j % D1D;
-                     const int d = j / D1D;
-                     elem_coords[qp + d*D1D] =
-                        xElemCoord[qp + el*D1D + d*p_NEL];
-                  }
-                  MFEM_SYNC_THREAD;
-               }
-
-               const double *elx[sDIM];
-               for (int d=0; d<sDIM; d++)
-               {
-                  elx[d] = MD1<= 6 ? &elem_coords[d*D1D] :
-                           xElemCoord + d*p_NEL + el*D1D;
-               }
-               MFEM_SYNC_THREAD;
-               //// findpts_el ////
-               {
-                  MFEM_FOREACH_THREAD(j,x,1)
-                  {
-                     fpt->dist2 = HUGE_VAL;
-                     fpt->dist2p = 0;
-                     fpt->tr = 1.0;
-                  }
-                  MFEM_FOREACH_THREAD(j,x,sDIM)
-                  {
-                     fpt->x[j] = x_i[j];
-                  }
-                  MFEM_SYNC_THREAD;
-
-                  //// seed ////
-                  {
-                     double *dist2_temp = r_workspace_ptr;
-                     double *r_temp = dist2_temp + D1D;
-                     MFEM_FOREACH_THREAD(j,x,nThreads)
-                     {
-                        seed_j(elx, x_i, gll1D, dist2_temp, r_temp, j, D1D);
-                     }
-                     MFEM_SYNC_THREAD;
-
-                     MFEM_FOREACH_THREAD(j,x,1)
-                     {
-                        fpt->dist2 = HUGE_VAL;
-                        for (int ir=0; ir<D1D; ++ir)
-                        {
-                           if (dist2_temp[ir] < fpt->dist2)
-                           {
-                              fpt->dist2 = dist2_temp[ir];
-                              fpt->r = r_temp[ir];
-                           }
-                        }
-                     }
-                     MFEM_SYNC_THREAD;
-                  } //seed done
-
-                  MFEM_FOREACH_THREAD(j,x,1)
-                  {
-                     tmp->dist2 = HUGE_VAL;
-                     tmp->dist2p = 0;
-                     tmp->tr = 1;
-                     tmp->flags = 0;
-                     tmp->r = fpt->r;
-                  }
-                  MFEM_FOREACH_THREAD(j,x,sDIM)
-                  {
-                     tmp->x[j] = fpt->x[j];
-                  }
-                  MFEM_SYNC_THREAD;
-
-                  for (int step=0; step<50; step++)
-                  {
-                     switch (num_constrained(tmp->flags & FLAG_MASK))
-                     {
-                        case 0:
-                        {
-                           double *wt = r_workspace_ptr;
-                           double *resid = wt + 3*D1D;
-                           double *jac = resid + sDIM;
-                           double *hess = jac + sDIM*rDIM;
-
-                           findptsElementGEdge_t edge;
-                           for (int d=0; d<sDIM; ++d)
-                           {
-                              edge.x[d] = constraint_workspace + d*D1D;
-                           }
-                           MFEM_FOREACH_THREAD(j,x,D1D)
-                           {
-                              for (int d=0; d<sDIM; ++d)
-                              {
-                                 edge.x[d][j] = elx[d][j];
-                              }
-                           }
-                           MFEM_SYNC_THREAD;
-
-                           MFEM_FOREACH_THREAD(j,x,D1D)
-                           {
-                              lag_eval_second_der(wt, tmp->r, j, gll1D,
-                                                  lagcoeff, D1D);
-                           }
-                           MFEM_SYNC_THREAD;
-
-                           MFEM_FOREACH_THREAD(j,x,sDIM)
-                           {
-                              resid[j] = tmp->x[j];
-                              jac[j] = 0.0;
-                              hess[j] = 0.0;
-                              for (int k=0; k<D1D; ++k)
-                              {
-                                 resid[j] -= wt[      k]*edge.x[j][k];
-                                 jac[j]   += wt[D1D+k]*edge.x[j][k];
-                                 hess[j]  += wt[2*D1D+k]*edge.x[j][k];
-                              }
-                           }
-                           MFEM_SYNC_THREAD;
-
-                           MFEM_FOREACH_THREAD(j,x,1)
-                           {
-                              hess[3] = resid[0]*hess[0] + resid[1]*hess[1] +
-                                        resid[2]*hess[2];
-                           }
-
-                           MFEM_FOREACH_THREAD(l,x,1)
-                           {
-                              if (!reject_prior_step_q(fpt,resid,tmp,tol))
-                              {
-                                 newton_edge(fpt,jac,hess[3],resid,
-                                             tmp->flags&FLAG_MASK,tmp,tol);
-                              }
-                           }
-                           MFEM_SYNC_THREAD;
-                           break;
-                        }
-                        case 1:
-                        {
-                           MFEM_FOREACH_THREAD(j,x,1)
-                           {
-                              const int pi = point_index(tmp->flags &
-                                                         FLAG_MASK);
-                              const double *wt = wtend + pi*3*D1D;
-                              findptsElementGPT_t gpt;
-                              for (int d=0; d<sDIM; ++d)
-                              {
-                                 gpt.x[d] = elx[d][pi*(D1D-1)];
-                                 gpt.jac[d] = 0.0;
-                                 gpt.hes[d] = 0.0;
-                                 for (int k=0; k<D1D; ++k)
-                                 {
-                                    gpt.jac[d] += wt[D1D  +k]*elx[d][k];
-                                    gpt.hes[d] += wt[2*D1D+k]*elx[d][k];
-                                 }
-                              }
-
-                              const double *const pt_x = gpt.x;
-                              const double *const jac = gpt.jac;
-                              const double *const hes = gpt.hes;
-                              double resid[sDIM], steep, sr;
-                              resid[0] = fpt->x[0] - pt_x[0];
-                              resid[1] = fpt->x[1] - pt_x[1];
-                              resid[2] = fpt->x[2] - pt_x[2];
-                              steep = jac[0]*resid[0] + jac[1]*resid[1] +
-                                      jac[2]*resid[2];
-                              sr = steep*tmp->r;
-                              if (!reject_prior_step_q(fpt, resid, tmp, tol))
-                              {
-                                 if (sr<0)
-                                 {
-                                    const double rhess = resid[0]*hes[0] +
-                                                         resid[1]*hes[1] +
-                                                         resid[2]*hes[2];
-                                    newton_edge(fpt, jac, rhess,
-                                                resid, 0, tmp, tol);
-                                 }
-                                 else   // sr==0
-                                 {
-                                    fpt->r = tmp->r;
-                                    fpt->dist2p = 0;
-                                    fpt->flags = tmp->flags | CONVERGED_FLAG;
-                                 }
-                              }
-                           }
-                           MFEM_SYNC_THREAD;
-                           break;
-                        } // case 1
-                     } //switch
-                     if (fpt->flags & CONVERGED_FLAG)
-                     {
-                        break;
-                     }
-                     MFEM_SYNC_THREAD;
-
-                     MFEM_FOREACH_THREAD(j,x,1)
-                     {
-                        *tmp = *fpt;
-                     }
-                     MFEM_SYNC_THREAD;
-                  } // for step<50
-               } // findpts_el
-
-               bool converged_internal =
-                  ((fpt->flags&FLAG_MASK) == CONVERGED_FLAG) &&
-                  (fpt->dist2<dist2tol);
-               if (*code_i==CODE_NOT_FOUND || converged_internal ||
-                   fpt->dist2<*dist2_i)
-               {
-                  MFEM_FOREACH_THREAD(j,x,1)
-                  {
-                     *(el_base+i) = el;
-                     *code_i  = converged_internal?CODE_INTERNAL:CODE_BORDER;
-                     *dist2_i = fpt->dist2;
-                     *(r_base+i) = fpt->r;
-                  }
-                  MFEM_SYNC_THREAD;
-                  if (converged_internal)
-                  {
-                     break;
-                  }
-               }
-            } // findpts_local
-         } // obbox_test
-      } // elp
-   });
-}
-
-void FindPointsGSLIB::FindPointsEdgeLocal3(const Vector &point_pos,
-                                           int point_pos_ordering,
-                                           Array<unsigned int> &code,
-                                           Array<unsigned int> &elem,
-                                           Vector &ref,
-                                           Vector &dist,
-                                           int npt)
-{
-   if (npt == 0)
-   {
-      return;
-   }
-   MFEM_VERIFY(spacedim==3 && dim == 1,"Function for 3D edges only");
-   bool use_dev = point_pos.UseDevice();
-   auto pp = point_pos.Read(use_dev);
-   auto pgslm = gsl_mesh.Read(use_dev);
-   auto pwt = DEV.wtend.Read(use_dev);
-   auto pbb = DEV.bb.Read(use_dev);
-   auto plhm = DEV.lh_min.Read(use_dev);
-   auto plhf = DEV.lh_fac.Read(use_dev);
-   auto plho = DEV.lh_offset.ReadWrite(use_dev);
-   auto pcode = code.Write(use_dev);
-   auto pelem = elem.Write(use_dev);
-   auto pref = ref.Write(use_dev);
-   auto pdist = dist.Write(use_dev);
-   auto pgll1d = DEV.gll1d.ReadWrite(use_dev);
-   auto plc = DEV.lagcoeff.Read(use_dev);
-   double dist2tol = DEV.surf_dist_tol;
-   const bool obb_chk = obb_check;
-   switch (DEV.dof1d)
-   {
-      case 2:
-         FindPointsEdgeLocal3DKernel<2>(npt, DEV.newt_tol, dist2tol,
-                                        pp, point_pos_ordering, pgslm,
-                                        NE_split_total, pwt, pbb, obb_chk,
-                                        DEV.lh_nx, plhm, plhf, plho,
-                                        pcode, pelem, pref, pdist,
-                                        pgll1d, plc);
-         break;
-      case 3:
-         FindPointsEdgeLocal3DKernel<3>(npt, DEV.newt_tol, dist2tol,
-                                        pp, point_pos_ordering, pgslm,
-                                        NE_split_total, pwt, pbb, obb_chk,
-                                        DEV.lh_nx, plhm, plhf, plho,
-                                        pcode, pelem, pref, pdist,
-                                        pgll1d, plc);
-         break;
-      case 4:
-         FindPointsEdgeLocal3DKernel<4>(npt, DEV.newt_tol, dist2tol,
-                                        pp, point_pos_ordering, pgslm,
-                                        NE_split_total, pwt, pbb, obb_chk,
-                                        DEV.lh_nx, plhm, plhf, plho,
-                                        pcode, pelem, pref, pdist,
-                                        pgll1d, plc);
-         break;
-      default:
-         FindPointsEdgeLocal3DKernel(npt, DEV.newt_tol, dist2tol, pp,
-                                     point_pos_ordering, pgslm,
-                                     NE_split_total, pwt, pbb, obb_chk,
-                                     DEV.lh_nx, plhm, plhf, plho,
-                                     pcode, pelem, pref, pdist,
-                                     pgll1d, plc, DEV.dof1d);
-         break;
-   }
-}
-#undef rDIM2
-#undef sDIM2
-#undef rDIM
-#undef sDIM
-#undef CODE_INTERNAL
-#undef CODE_BORDER
-#undef CODE_NOT_FOUND
-#else
-void FindPointsGSLIB::FindPointsEdgeLocal3( const Vector &point_pos,
-                                            int point_pos_ordering,
-                                            Array<unsigned int> &code,
-                                            Array<unsigned int> &elem,
-                                            Vector &ref,
-                                            Vector &dist,
-                                            int npt ) {} ;
-#endif
-} // namespace mfem
-
-#endif //ifdef MFEM_USE_GSLIB
@@ -1,190 +0,0 @@
-#ifndef MFEM_GSLIB_KERNEL_HELPERS_HPP
-#define MFEM_GSLIB_KERNEL_HELPERS_HPP
-
-#include "../../config/config.hpp"
-
-#include <cmath>
-
-namespace mfem
-{
-
-namespace gslib
-{
-
-struct dbl_range_t
-{
-   double min, max;
-};
-
-template <int SDIM>
-struct obbox_t
-{
-   double c0[SDIM], A[SDIM * SDIM];
-   dbl_range_t x[SDIM];
-};
-
-template <int SDIM>
-struct findptsLocalHashData_t
-{
-   int hash_n;
-   dbl_range_t bnd[SDIM];
-   double fac[SDIM];
-   unsigned int *offset;
-};
-
-// Eval the ith Lagrange interpolant at x.
-MFEM_HOST_DEVICE inline void lagrange_eval(double *p0, double x,
-                                           int i, int p_Nq,
-                                           double *z, double *lagrangeCoeff)
-{
-   double p_i = (1 << (p_Nq - 1));
-   for (int j = 0; j < p_Nq; ++j)
-   {
-      const double d_j = x - z[j];
-      p_i *= j == i ? 1 : d_j;
-   }
-   p0[i] = lagrangeCoeff[i] * p_i;
-}
-
-// Eval the ith Lagrange interpolant and its first derivative at x.
-MFEM_HOST_DEVICE inline void lag_eval_first_der(double *p0, double x,
-                                                int i, const double *z,
-                                                const double *lCoeff,
-                                                int pN)
-{
-   double u0 = 1, u1 = 0;
-   for (int j = 0; j < pN; ++j)
-   {
-      if (i != j)
-      {
-         const double d_j = 2 * (x - z[j]);
-         u1 = d_j * u1 + u0;
-         u0 = d_j * u0;
-      }
-   }
-   p0[i] = lCoeff[i] * u0;
-   p0[pN + i] = 2.0 * lCoeff[i] * u1;
-}
-
-// Eval the ith Lagrange interpolant and its first and second derivative at x.
-MFEM_HOST_DEVICE inline void lag_eval_second_der(double *p0, double x,
-                                                 int i, const double *z,
-                                                 const double *lCoeff,
-                                                 int pN)
-{
-   double u0 = 1, u1 = 0, u2 = 0;
-   for (int j = 0; j < pN; ++j)
-   {
-      if (i != j)
-      {
-         const double d_j = 2 * (x - z[j]);
-         u2 = d_j * u2 + u1;
-         u1 = d_j * u1 + u0;
-         u0 = d_j * u0;
-      }
-   }
-   p0[i] = lCoeff[i] * u0;
-   p0[pN + i] = 2.0 * lCoeff[i] * u1;
-   p0[2 * pN + i] = 8.0 * lCoeff[i] * u2;
-}
-
-// Solve Ax=y where A is a symmetric 2x2 matrix packed as {a00, a01, a11}.
-MFEM_HOST_DEVICE inline void lin_solve_sym_2(double x[2],
-                                             const double A[3],
-                                             const double y[2])
-{
-   const double idet = 1 / (A[0] * A[2] - A[1] * A[1]);
-   x[0] = idet * (A[2] * y[0] - A[1] * y[1]);
-   x[1] = idet * (A[0] * y[1] - A[1] * y[0]);
-}
-
-// Positive when the point is inside the axis-aligned bounding box.
-template <int SDIM>
-MFEM_HOST_DEVICE inline double AABB_test(const obbox_t<SDIM> *const b,
-                                         const double (&x)[SDIM])
-{
-   double test = 1.0;
-   for (int d = 0; d < SDIM; ++d)
-   {
-      const double b_d = (x[d] - b->x[d].min) * (b->x[d].max - x[d]);
-      test = test < 0.0 ? test : b_d;
-   }
-   return test;
-}
-
-// Positive when the point is inside the oriented bounding box.
-template <int SDIM>
-MFEM_HOST_DEVICE inline double bbox_test(const obbox_t<SDIM> *const b,
-                                         const double (&x)[SDIM])
-{
-   const double bxyz = AABB_test(b, x);
-   if (bxyz < 0.0)
-   {
-      return bxyz;
-   }
-
-   double dxyz[SDIM];
-   for (int d = 0; d < SDIM; ++d)
-   {
-      dxyz[d] = x[d] - b->c0[d];
-   }
-
-   double test = 1.0;
-   for (int d = 0; d < SDIM; ++d)
-   {
-      double rst = 0.0;
-      for (int e = 0; e < SDIM; ++e)
-      {
-         rst += b->A[d * SDIM + e] * dxyz[e];
-      }
-      const double brst = (rst + 1.0) * (1.0 - rst);
-      test = test < 0.0 ? test : brst;
-   }
-   return test;
-}
-
-// Hash index in the hash table for the point x.
-template <int SDIM>
-MFEM_HOST_DEVICE inline int hash_index(
-   const findptsLocalHashData_t<SDIM> *const p,
-   const double (&x)[SDIM])
-{
-   const int n = p->hash_n;
-   int sum = 0;
-   for (int d = SDIM - 1; d >= 0; --d)
-   {
-      sum *= n;
-      const int i = (int)floor((x[d] - p->bnd[d].min) * p->fac[d]);
-      sum += i < 0 ? 0 : (n - 1 < i ? n - 1 : i);
-   }
-   return sum;
-}
-
-// Squared Euclidean norm.
-template <int SDIM>
-MFEM_HOST_DEVICE inline double l2norm2(const double (&x)[SDIM])
-{
-   double sum = 0.0;
-   for (int d = 0; d < SDIM; ++d)
-   {
-      sum += x[d] * x[d];
-   }
-   return sum;
-}
-
-template <int SDIM>
-MFEM_HOST_DEVICE inline double l2norm2(const double *x)
-{
-   double sum = 0.0;
-   for (int d = 0; d < SDIM; ++d)
-   {
-      sum += x[d] * x[d];
-   }
-   return sum;
-}
-
-} // namespace gslib
-
-} // namespace mfem
-
-#endif
@@ -1,152 +0,0 @@
-// Copyright (c) 2010-2025, Lawrence Livermore National Security, LLC. Produced
-// at the Lawrence Livermore National Laboratory. All Rights reserved. See files
-// LICENSE and NOTICE for details. LLNL-CODE-806117.
-//
-// This file is part of the MFEM library. For more information and source code
-// availability visit https://mfem.org.
-//
-// MFEM is free software; you can redistribute it and/or modify it under the
-// terms of the BSD-3 license. We welcome feedback and contributions, see file
-// CONTRIBUTING.md for details.
-
-#include "../gslib.hpp"
-#include "../../general/forall.hpp"
-#include "gslib_kernel_helpers.hpp"
-
-#ifdef MFEM_USE_GSLIB
-
-#ifdef MFEM_HAVE_GCC_PRAGMA_DIAGNOSTIC
-#pragma GCC diagnostic push
-#pragma GCC diagnostic ignored "-Wunused-function"
-#endif
-#include "gslib.h"
-#ifndef GSLIB_RELEASE_VERSION //gslib v1.0.7
-#define GSLIB_RELEASE_VERSION 10007
-#endif
-#ifdef MFEM_HAVE_GCC_PRAGMA_DIAGNOSTIC
-#pragma GCC diagnostic pop
-#endif
-namespace mfem
-{
-#if GSLIB_RELEASE_VERSION >= 10009
-#define CODE_INTERNAL 0
-#define CODE_BORDER 1
-#define CODE_NOT_FOUND 2
-
-using gslib::lagrange_eval;
-
-template<int T_D1D = 0>
-static void InterpolateLocal1DKernel(const double *const gf_in,
-                                     int *const el,
-                                     double *const r,
-                                     double *const int_out,
-                                     const int npt,
-                                     const int nfields,
-                                     double *gll1D,
-                                     double *lagcoeff,
-                                     const int pN = 0)
-{
-   const int MD1 = T_D1D ? T_D1D : DofQuadLimits::MAX_D1D;
-   const int D1D = T_D1D ? T_D1D : pN;
-   const int p_Nq = D1D;
-   MFEM_VERIFY(MD1 <= DofQuadLimits::MAX_D1D,
-               "Increase Max allowable polynomial order.");
-   MFEM_VERIFY(pN<=DofQuadLimits::MAX_D1D,
-               "Increase Max allowable polynomial order.");
-   MFEM_VERIFY(D1D != 0, "Polynomial order not specified.");
-   // for each point of the npt points, create a thread block of size dof1Dsol
-   mfem::forall_2D(npt, D1D, 1, [=] MFEM_HOST_DEVICE (int i)
-   {
-      MFEM_SHARED double wtr[MD1];
-      MFEM_SHARED double sums[MD1];
-
-      // Evaluate basis functions at the reference space coordinates
-      MFEM_FOREACH_THREAD(j,x,D1D)
-      {
-         lagrange_eval(wtr, r[i], j, p_Nq, gll1D, lagcoeff);
-      }
-      MFEM_SYNC_THREAD;
-
-      for (int fld=0; fld<nfields; ++fld)
-      {
-         // If using GetNodalValues, ordering is NDOFS x NEL x VDIM and the
-         // offset would be `el[i] * p_Nq + fld * gf_offset`.
-         // R->Mult produces element vectors in NDOFS x VDIM x NEL layout.
-         const int elemOffset = el[i]*nfields*p_Nq + fld*p_Nq;
-         MFEM_FOREACH_THREAD(j,x,D1D)
-         {
-            sums[j] = wtr[j] * gf_in[elemOffset + j];
-         }
-         MFEM_SYNC_THREAD;
-
-         MFEM_FOREACH_THREAD(j,x,1)
-         {
-            double sumv = 0.0;
-            // sum the contributions of each lagrange polynomial
-            for (int jj=0; jj<D1D; ++jj)
-            {
-               sumv += sums[jj];
-            }
-            int_out[fld*npt + i] = sumv;
-         }
-         MFEM_SYNC_THREAD;
-      }
-   });
-}
-
-void FindPointsGSLIB::InterpolateLocal1( const Vector &field_in,
-                                         Array<int> &gsl_elem_dev_l,
-                                         Vector &gsl_ref_l,
-                                         Vector &field_out,
-                                         int npt,
-                                         int ncomp,
-                                         int dof1Dsol )
-{
-   MFEM_VERIFY(dim == 1, "Kernel for edges only.");
-   if (npt == 0) { return; }
-   bool use_dev = field_in.UseDevice();
-   auto pfin = field_in.Read(use_dev);
-   auto pgsl = gsl_elem_dev_l.ReadWrite(use_dev);
-   auto pgslr = gsl_ref_l.ReadWrite(use_dev);
-   auto pfout = field_out.Write(use_dev);
-   auto pgll = DEV.gll1d_sol.ReadWrite(use_dev);
-   auto plcf = DEV.lagcoeff_sol.ReadWrite(use_dev);
-   switch (dof1Dsol)
-   {
-      case 2:
-         InterpolateLocal1DKernel<2>(pfin, pgsl, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 3:
-         InterpolateLocal1DKernel<3>(pfin, pgsl, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 4:
-         InterpolateLocal1DKernel<4>(pfin, pgsl, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 5:
-         InterpolateLocal1DKernel<5>(pfin, pgsl, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      default:
-         InterpolateLocal1DKernel(pfin, pgsl, pgslr, pfout,
-                                  npt, ncomp, pgll, plcf, dof1Dsol);
-         break;
-   }
-}
-#undef CODE_INTERNAL
-#undef CODE_BORDER
-#undef CODE_NOT_FOUND
-#else
-void FindPointsGSLIB::InterpolateLocal1(const Vector &field_in,
-                                        Array<int> &gsl_elem_dev_l,
-                                        Vector &gsl_ref_l,
-                                        Vector &field_out,
-                                        int npt, int ncomp,
-                                        int dof1Dsol) {};
-#endif
-
-} // namespace mfem
-
-#endif //ifdef MFEM_USE_GSLIB
@@ -11,7 +11,6 @@

 #include "../gslib.hpp"
 #include "../../general/forall.hpp"
-#include "gslib_kernel_helpers.hpp"

 #ifdef MFEM_USE_GSLIB

@@ -33,7 +32,18 @@ namespace mfem
 #define CODE_BORDER 1
 #define CODE_NOT_FOUND 2

-using gslib::lagrange_eval;
+static MFEM_HOST_DEVICE void lagrange_eval(double *p0, double x,
+                                           int i, int p_Nq,
+                                           double *z, double *lagrangeCoeff)
+{
+   double p_i = (1 << (p_Nq - 1));
+   for (int j = 0; j < p_Nq; ++j)
+   {
+      double d_j = x - z[j];
+      p_i *= j == i ? 1 : d_j;
+   }
+   p0[i] = lagrangeCoeff[i] * p_i;
+}

 template<int T_D1D = 0>
 static void InterpolateLocal2DKernel(const double *const gf_in,
@@ -42,6 +52,8 @@ static void InterpolateLocal2DKernel(const double *const gf_in,
                                     double *const int_out,
                                     const int npt,
                                     const int ncomp,
+                                     const int nel,
+                                     const int gf_offset,
                                     double *gll1D,
                                     double *lagcoeff,
                                     const int pN = 0)
@@ -52,8 +64,6 @@ static void InterpolateLocal2DKernel(const double *const gf_in,
   const int p_Np = D1D*D1D;
   MFEM_VERIFY(MD1 <= DofQuadLimits::MAX_D1D,
               "Increase Max allowable polynomial order.");
-   MFEM_VERIFY(pN<=DofQuadLimits::MAX_D1D,
-               "Increase Max allowable polynomial order.");
   MFEM_VERIFY(D1D != 0, "Polynomial order not specified.");
   mfem::forall_2D(npt, D1D, D1D, [=] MFEM_HOST_DEVICE (int i)
   {
@@ -72,9 +82,9 @@ static void InterpolateLocal2DKernel(const double *const gf_in,

      for (int fld = 0; fld < Nfields; ++fld)
      {
-         // If using GetNodalValues, ordering is NDOFS x NEL x VDIM and the
-         // offset would be `el[i] * p_Np + fld * gf_offset`.
-         // R->Mult produces element vectors in NDOFS x VDIM x NEL layout.
+         // If using GetNodalValues, ordering is NDOFSxNELxVDIM
+         // const int elemOffset = el[i] * p_Np + fld * gf_offset;
+         //if using R->Mult for L -> E-Vec use below: NDOFSxVDIMxNEL
         const int elemOffset = el[i] * p_Np * Nfields + fld * p_Np;
         MFEM_FOREACH_THREAD(j,x,D1D)
         {
@@ -110,38 +120,33 @@ void FindPointsGSLIB::InterpolateLocal2(const Vector &field_in,
                                        Vector &gsl_ref_l,
                                        Vector &field_out,
                                        int npt, int ncomp,
-                                        int dof1Dsol)
+                                        int nel, int dof1Dsol)
 {
   if (npt == 0) { return; }
-   bool use_dev = field_in.UseDevice();
-   auto pfin = field_in.Read(use_dev);
-   auto pgsl = gsl_elem_dev_l.ReadWrite(use_dev);
-   auto pgslr = gsl_ref_l.ReadWrite(use_dev);
-   auto pfout = field_out.Write(use_dev);
-   auto pgll = DEV.gll1d_sol.ReadWrite(use_dev);
-   auto plcf = DEV.lagcoeff_sol.ReadWrite(use_dev);
+   const int gf_offset = field_in.Size()/ncomp;
+   auto pfin = field_in.Read();
+   auto pgsl = gsl_elem_dev_l.ReadWrite();
+   auto pgslr = gsl_ref_l.ReadWrite();
+   auto pfout = field_out.Write();
+   auto pgll = DEV.gll1d_sol.ReadWrite();
+   auto plcf = DEV.lagcoeff_sol.ReadWrite();
   switch (dof1Dsol)
   {
-      case 2:
-         InterpolateLocal2DKernel<2>(pfin, pgsl, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 3:
-         InterpolateLocal2DKernel<3>(pfin, pgsl, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 4:
-         InterpolateLocal2DKernel<4>(pfin, pgsl, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 5:
-         InterpolateLocal2DKernel<5>(pfin, pgsl, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      default:
-         InterpolateLocal2DKernel(pfin, pgsl, pgslr, pfout,
-                                  npt, ncomp, pgll, plcf, dof1Dsol);
-         break;
+      case 2: return InterpolateLocal2DKernel<2>(pfin, pgsl, pgslr, pfout,
+                                                    npt, ncomp, nel, gf_offset,
+                                                    pgll, plcf);
+      case 3: return InterpolateLocal2DKernel<3>(pfin, pgsl, pgslr, pfout,
+                                                    npt, ncomp, nel, gf_offset,
+                                                    pgll, plcf);
+      case 4: return InterpolateLocal2DKernel<4>(pfin, pgsl, pgslr, pfout,
+                                                    npt, ncomp, nel, gf_offset,
+                                                    pgll, plcf);
+      case 5: return InterpolateLocal2DKernel<5>(pfin, pgsl, pgslr, pfout,
+                                                    npt, ncomp, nel, gf_offset,
+                                                    pgll, plcf);
+      default: return InterpolateLocal2DKernel(pfin, pgsl, pgslr, pfout,
+                                                  npt, ncomp, nel, gf_offset,
+                                                  pgll, plcf, dof1Dsol);
   }
 }

@@ -155,7 +160,7 @@ void FindPointsGSLIB::InterpolateLocal2(const Vector &field_in,
                                        Vector &gsl_ref_l,
                                        Vector &field_out,
                                        int npt, int ncomp,
-                                        int dof1Dsol) {};
+                                        int nel, int dof1Dsol) {};
 #endif
 } // namespace mfem

@@ -11,7 +11,6 @@

 #include "../gslib.hpp"
 #include "../../general/forall.hpp"
-#include "gslib_kernel_helpers.hpp"

 #ifdef MFEM_USE_GSLIB

@@ -33,7 +32,18 @@ namespace mfem
 #define CODE_BORDER 1
 #define CODE_NOT_FOUND 2

-using gslib::lagrange_eval;
+static MFEM_HOST_DEVICE void lagrange_eval(double *p0, double x,
+                                           int i, int p_Nq,
+                                           double *z, double *lagrangeCoeff)
+{
+   double p_i = (1 << (p_Nq - 1));
+   for (int j = 0; j < p_Nq; ++j)
+   {
+      double d_j = x - z[j];
+      p_i *= j == i ? 1 : d_j;
+   }
+   p0[i] = lagrangeCoeff[i] * p_i;
+}

 template<int T_D1D = 0>
 static void InterpolateLocal3DKernel(const double *const gf_in,
@@ -42,6 +52,8 @@ static void InterpolateLocal3DKernel(const double *const gf_in,
                                     double *const int_out,
                                     const int npt,
                                     const int ncomp,
+                                     const int nel,
+                                     const int gf_offset,
                                     double *gll1D,
                                     double *lagcoeff,
                                     const int pN = 0)
@@ -72,9 +84,9 @@ static void InterpolateLocal3DKernel(const double *const gf_in,

      for (int fld = 0; fld < Nfields; ++fld)
      {
-         // If using GetNodalValues, ordering is NDOFS x NEL x VDIM and the
-         // offset would be `el[i] * p_Np + fld * gf_offset`.
-         // R->Mult produces element vectors in NDOFS x VDIM x NEL layout.
+         // If using GetNodalValues, ordering is NDOFSxNELxVDIM
+         // const int elemOffset = el[i] * p_Np + fld * gf_offset;
+         //if using R->Mult for L -> E-Vec use below.
         const int elemOffset = el[i] * p_Np * Nfields + fld * p_Np;
         MFEM_FOREACH_THREAD(j,x,D1D)
         {
@@ -113,43 +125,37 @@ void FindPointsGSLIB::InterpolateLocal3(const Vector &field_in,
                                        Vector &gsl_ref_l,
                                        Vector &field_out,
                                        int npt, int ncomp,
-                                        int dof1Dsol)
+                                        int nel, int dof1Dsol)
 {
   if (npt == 0) { return; }
-   bool use_dev = field_in.UseDevice();
-   auto pfin = field_in.Read(use_dev);
-   auto pgsle = gsl_elem_dev_l.ReadWrite(use_dev);
-   auto pgslr = gsl_ref_l.ReadWrite(use_dev);
-   auto pfout = field_out.Write(use_dev);
-   auto pgll = DEV.gll1d_sol.ReadWrite(use_dev);
-   auto plcf = DEV.lagcoeff_sol.ReadWrite(use_dev);
+   const int gf_offset = field_in.Size()/ncomp;
+   auto pfin = field_in.Read();
+   auto pgsle = gsl_elem_dev_l.ReadWrite();
+   auto pgslr = gsl_ref_l.ReadWrite();
+   auto pfout = field_out.Write();
+   auto pgll = DEV.gll1d_sol.ReadWrite();
+   auto plcf = DEV.lagcoeff_sol.ReadWrite();
   switch (dof1Dsol)
   {
-      case 2:
-         InterpolateLocal3DKernel<2>(pfin, pgsle, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 3:
-         InterpolateLocal3DKernel<3>(pfin, pgsle, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 4:
-         InterpolateLocal3DKernel<4>(pfin, pgsle, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      case 5:
-         InterpolateLocal3DKernel<5>(pfin, pgsle, pgslr, pfout,
-                                     npt, ncomp, pgll, plcf);
-         break;
-      default:
-         InterpolateLocal3DKernel(pfin, pgsle, pgslr, pfout,
-                                  npt, ncomp, pgll, plcf, dof1Dsol);
-         break;
+      case 2: return InterpolateLocal3DKernel<2>(pfin, pgsle, pgslr, pfout,
+                                                    npt, ncomp, nel, gf_offset,
+                                                    pgll, plcf);
+      case 3: return InterpolateLocal3DKernel<3>(pfin, pgsle, pgslr, pfout,
+                                                    npt, ncomp, nel, gf_offset,
+                                                    pgll, plcf);
+      case 4: return InterpolateLocal3DKernel<4>(pfin, pgsle, pgslr, pfout,
+                                                    npt, ncomp, nel, gf_offset,
+                                                    pgll, plcf);
+      case 5: return InterpolateLocal3DKernel<5>(pfin, pgsle, pgslr, pfout,
+                                                    npt, ncomp, nel, gf_offset,
+                                                    pgll, plcf);
+      default: return InterpolateLocal3DKernel(pfin, pgsle, pgslr, pfout,
+                                                  npt, ncomp, nel, gf_offset,
+                                                  pgll, plcf, dof1Dsol);
   }
 }


-#undef MAXC
 #undef CODE_INTERNAL
 #undef CODE_BORDER
 #undef CODE_NOT_FOUND
@@ -159,7 +165,7 @@ void FindPointsGSLIB::InterpolateLocal3(const Vector &field_in,
                                        Vector &gsl_ref_l,
                                        Vector &field_out,
                                        int npt, int ncomp,
-                                        int dof1Dsol) {};
+                                        int nel, int dof1Dsol) {};
 #endif
 } // namespace mfem

@@ -10,7 +10,6 @@
 // CONTRIBUTING.md for details.

 #include "bilininteg_diffusion_kernels.hpp"
-#include "bilininteg_diffusion_pa_simplices.hpp" // IWYU pragma: keep

 namespace mfem
 {
@@ -20,13 +19,6 @@ namespace mfem
 DiffusionIntegrator::Kernels::Kernels()
 {
   // 2D
-   // Q = P, only for simplex
-   DiffusionIntegrator::AddSimplexSpecialization<2,2,1>();
-   DiffusionIntegrator::AddSimplexSpecialization<2,3,2>();
-   DiffusionIntegrator::AddSimplexSpecialization<2,4,3>();
-   DiffusionIntegrator::AddSimplexSpecialization<2,5,4>();
-   DiffusionIntegrator::AddSimplexSpecialization<2,6,5>();
-   DiffusionIntegrator::AddSimplexSpecialization<2,7,6>();
   // Q = P+1
   DiffusionIntegrator::AddSpecialization<2,1,1>();
   DiffusionIntegrator::AddSpecialization<2,2,2>();
@@ -48,18 +40,7 @@ DiffusionIntegrator::Kernels::Kernels()
   DiffusionIntegrator::AddSpecialization<2,8,9>();
   DiffusionIntegrator::AddSpecialization<2,9,10>();
   // others
-   DiffusionIntegrator::AddSimplexSpecialization<2,2,5>();
-   DiffusionIntegrator::AddSimplexSpecialization<2,3,6>();
-
   // 3D
-   // Q = P, only for simplex
-   DiffusionIntegrator::AddSimplexSpecialization<3,2,1>();
-   DiffusionIntegrator::AddSimplexSpecialization<3,3,2>();
-   DiffusionIntegrator::AddSimplexSpecialization<3,4,3>();
-   DiffusionIntegrator::AddSimplexSpecialization<3,5,4>();
-   DiffusionIntegrator::AddSimplexSpecialization<3,6,5>();
-   DiffusionIntegrator::AddSimplexSpecialization<3,7,6>();
-   DiffusionIntegrator::AddSimplexSpecialization<3,8,7>();
   // Q = P+1
   DiffusionIntegrator::AddSpecialization<3,1,1>();
   DiffusionIntegrator::AddSpecialization<3,2,2>();
@@ -12,6 +12,7 @@
 #ifndef MFEM_BILININTEG_DIFFUSION_KERNELS_HPP
 #define MFEM_BILININTEG_DIFFUSION_KERNELS_HPP

+#include "../kernel_dispatch.hpp"
 #include "../../config/config.hpp"
 #include "../../general/array.hpp"
 #include "../../general/forall.hpp"
@@ -19,8 +20,6 @@
 #include "../../linalg/vector.hpp"
 #include "../bilininteg.hpp"

-#include "bilininteg_diffusion_pa_simplices.hpp"
-
 namespace mfem
 {

@@ -638,8 +637,8 @@ inline void SmemPADiffusionApply2D(const int NE,
                                   const bool symmetric,
                                   const Array<real_t> &b_,
                                   const Array<real_t> &g_,
-                                   const Array<real_t> &,
-                                   const Array<real_t> &,
+                                   const Array<real_t> &bt_,
+                                   const Array<real_t> &gt_,
                                   const Vector &d_,
                                   const Vector &x_,
                                   Vector &y_,
@@ -1065,6 +1064,8 @@ inline void SmemPADiffusionApply3D(const int NE,
         }
      }
      MFEM_SYNC_THREAD;
+
+      // Grad X
      MFEM_FOREACH_THREAD_DIRECT(dz,z,D1D)
      {
         MFEM_FOREACH_THREAD_DIRECT(dy,y,D1D)
@@ -1085,6 +1086,8 @@ inline void SmemPADiffusionApply3D(const int NE,
         }
      }
      MFEM_SYNC_THREAD;
+
+      // Grad Y
      MFEM_FOREACH_THREAD_DIRECT(dz,z,D1D)
      {
         MFEM_FOREACH_THREAD_DIRECT(qy,y,Q1D)
@@ -1106,6 +1109,8 @@ inline void SmemPADiffusionApply3D(const int NE,
         }
      }
      MFEM_SYNC_THREAD;
+
+      // Grad Z + Q-function
      MFEM_FOREACH_THREAD_DIRECT(qz,z,Q1D)
      {
         MFEM_FOREACH_THREAD_DIRECT(qy,y,Q1D)
@@ -1218,48 +1223,48 @@ inline void SmemPADiffusionApply3D(const int NE,

 namespace
 {
-using ApplyKernelType = DiffusionIntegrator::ApplyKernelType;
-using ApplySimplexKernelType = DiffusionIntegrator::ApplySimplexKernelType;
-using DiagonalKernelType = DiffusionIntegrator::DiagonalKernelType;
+using DiffusionApplyKernelType =
+   DiffusionIntegrator::DiffusionApplyKernelType;
+
+using DiffusionDiagonalKernelType =
+   DiffusionIntegrator::DiffusionDiagonalKernelType;
 }

-template<int DIM, int D1D, int Q1D>
-ApplyKernelType DiffusionIntegrator::ApplyPAKernels::Kernel()
+template<int DIM, int T_D1D, int T_Q1D>
+DiffusionApplyKernelType DiffusionIntegrator::DiffusionApplyPAKernel::Kernel()
 {
-   if constexpr (DIM == 2) { return internal::SmemPADiffusionApply2D<D1D, Q1D>; }
-   else if constexpr (DIM == 3) { return internal::SmemPADiffusionApply3D<D1D, Q1D>; }
-   else { MFEM_ABORT(""); }
-   return nullptr;
+   if constexpr (DIM == 2) { return internal::SmemPADiffusionApply2D<T_D1D,T_Q1D>; }
+   else if constexpr (DIM == 3) { return internal::SmemPADiffusionApply3D<T_D1D, T_Q1D>; }
+   MFEM_ABORT("");
 }

-inline
-ApplyKernelType DiffusionIntegrator::ApplyPAKernels::Fallback(int dim, int, int)
+inline DiffusionApplyKernelType
+DiffusionIntegrator::DiffusionApplyPAKernel::Fallback(int DIM, int, int)
 {
-   if (dim == 2) { return internal::PADiffusionApply2D; }
-   else if (dim == 3) { return internal::PADiffusionApply3D; }
+   if (DIM == 2) { return internal::PADiffusionApply2D; }
+   else if (DIM == 3) { return internal::PADiffusionApply3D; }
   else { MFEM_ABORT(""); }
 }

 template<int DIM, int D1D, int Q1D>
-DiagonalKernelType DiffusionIntegrator::DiagonalPAKernels::Kernel()
+DiffusionDiagonalKernelType
+DiffusionIntegrator::DiffusionDiagonalPAKernel::Kernel()
 {
-   if constexpr (DIM == 2) { return internal::SmemPADiffusionDiagonal2D<D1D, Q1D>; }
+   if constexpr (DIM == 2) { return internal::SmemPADiffusionDiagonal2D<D1D,Q1D>; }
   else if constexpr (DIM == 3) { return internal::SmemPADiffusionDiagonal3D<D1D, Q1D>; }
-   else { MFEM_ABORT(""); }
-   return nullptr;
+   MFEM_ABORT("");
 }

-inline DiagonalKernelType
-DiffusionIntegrator::DiagonalPAKernels::Fallback(int dim, int, int)
+inline DiffusionDiagonalKernelType
+DiffusionIntegrator::DiffusionDiagonalPAKernel::Fallback(int DIM, int, int)
 {
-   if (dim == 2) { return internal::PADiffusionDiagonal2D; }
-   else if (dim == 3) { return internal::PADiffusionDiagonal3D; }
+   if (DIM == 2) { return internal::PADiffusionDiagonal2D; }
+   else if (DIM == 3) { return internal::PADiffusionDiagonal3D; }
   else { MFEM_ABORT(""); }
-   return nullptr;
 }
-
 /// \endcond DO_NOT_DOCUMENT

 } // namespace mfem

+
 #endif
@@ -15,7 +15,6 @@
 #include "../../mesh/nurbs.hpp"
 #include "../ceed/integrators/diffusion/diffusion.hpp"
 #include "bilininteg_diffusion_kernels.hpp"
-#include "bilininteg_diffusion_pa_simplices.hpp"

 namespace mfem
 {
@@ -32,8 +31,8 @@ void DiffusionIntegrator::AssembleDiagonalPA(Vector &diag)
      const Array<real_t> &B = maps->B;
      const Array<real_t> &G = maps->G;
      const Vector &Dv = pa_data;
-      DiagonalPAKernels::Run(dim, dofs1D, quad1D, ne, symmetric, B, G, Dv,
-                             diag, dofs1D, quad1D);
+      DiffusionDiagonalPAKernel::Run(dim, dofs1D, quad1D, ne, symmetric, B, G, Dv,
+                                     diag, dofs1D, quad1D);
   }
 }

@@ -69,26 +68,8 @@ void DiffusionIntegrator::AddMultPA(const Vector &x, Vector &y) const
      }
 #endif // MFEM_USE_OCCA

-      if (fespace->UsesRaggedTensorBasis())
-      {
-         const auto *rmaps = static_cast<const RaggedDofToQuad*>(maps);
-         return ApplySimplexPAKernels::Run(dim, dofs1D, quad1D, ne, symmetric,
-                                           rmaps->lex_map,
-                                           rmaps->forward_map2d_diff,
-                                           rmaps->inverse_map2d_diff,
-                                           rmaps->forward_map3d_diff,
-                                           rmaps->inverse_map3d_diff,
-                                           rmaps->Ga1,
-                                           rmaps->Ga2,
-                                           rmaps->Ga3,
-                                           rmaps->Ga1t,
-                                           rmaps->Ga2t,
-                                           rmaps->Ga3t,
-                                           Dv, x, y, dofs1D, quad1D);
-      }
-
-      ApplyPAKernels::Run(dim, dofs1D, quad1D, ne, symmetric, B, G, Bt,
-                          Gt, Dv, x, y, dofs1D, quad1D);
+      DiffusionApplyPAKernel::Run(dim, dofs1D, quad1D, ne, symmetric, B, G, Bt,
+                                  Gt, Dv, x, y, dofs1D, quad1D);
   }
 }

@@ -113,8 +94,7 @@ void DiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)
   fespace = &fes;
   Mesh *mesh = fes.GetMesh();
   const FiniteElement &el = *fes.GetTypicalFE();
-   const bool stroud = fes.UsesRaggedTensorBasis();
-   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el, stroud);
+   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el);
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
@@ -139,22 +119,13 @@ void DiffusionIntegrator::AssemblePA(const FiniteElementSpace &fes)
   dim = mesh->Dimension();
   ne = fes.GetNE();
   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::JACOBIANS, mt);
-   if (stroud)
-   {
-      maps = &el.GetDofToQuad(*ir, DofToQuad::RAGGED_TENSOR);
-   }
-   else
-   {
-      maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
-   }
   const int sdim = mesh->SpaceDimension();
+   maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
   dofs1D = maps->ndof;
   quad1D = maps->nqpt;

   QuadratureSpace qs(*mesh, *ir);
   CoefficientVector coeff(qs, CoefficientStorage::COMPRESSED);
-   // QuadratureSpace expects ir defined in reference simplex for Bernstein
-   // elements with partial assembly

   if (MQ) { coeff.ProjectTranspose(*MQ); }
   else if (VQ) { coeff.Project(*VQ); }
@@ -203,9 +174,9 @@ void DiffusionIntegrator::AddAbsMultPA(const Vector &x, Vector &y) const
   abs_pa_data.Abs();
   auto abs_maps = maps->Abs();

-   ApplyPAKernels::Run(dim, dofs1D, quad1D, ne, symmetric,
-                       abs_maps.B, abs_maps.G, abs_maps.Bt, abs_maps.Gt,
-                       abs_pa_data, x, y, dofs1D, quad1D);
+   DiffusionApplyPAKernel::Run(dim, dofs1D, quad1D, ne, symmetric,
+                               abs_maps.B, abs_maps.G, abs_maps.Bt, abs_maps.Gt,
+                               abs_pa_data, x, y, dofs1D, quad1D);
 }

 void DiffusionIntegrator::AddAbsMultTransposePA(const Vector &x,
@@ -10,7 +10,6 @@
 // CONTRIBUTING.md for details.

 #include "bilininteg_mass_kernels.hpp"
-#include "bilininteg_mass_pa_simplices.hpp" // IWYU pragma: keep

 namespace mfem
 {
@@ -40,10 +39,8 @@ MassIntegrator::Kernels::Kernels()
   MassIntegrator::AddSpecialization<2,9,10>();
   // others
   MassIntegrator::AddSpecialization<2,2,4>();
-   MassIntegrator::AddSpecialization<2,2,5>();
   MassIntegrator::AddSpecialization<2,3,6>();
   MassIntegrator::AddSpecialization<2,4,6>();
-
   // 3D
   // Q=P+1
   MassIntegrator::AddSpecialization<3,1,1>();
@@ -19,8 +19,6 @@
 #include "../../linalg/vector.hpp"
 #include "../bilininteg.hpp"

-#include "bilininteg_mass_pa_simplices.hpp"
-
 namespace mfem
 {

@@ -183,12 +181,6 @@ constexpr int NBZ(int D1D)
 {
   return ipow(2, D(D1D) >= 0 ? D(D1D) : 0);
 }
-constexpr int NBZ3D(int MDQ)
-{
-   return MDQ > 0 ? std::min<int>(
-             (128 + MDQ * MDQ * MDQ - 1) / (MDQ * MDQ * MDQ), 64)
-          : 1;
-}
 }

 // Shared memory PA Mass Diagonal 2D kernel
@@ -812,23 +804,19 @@ void PAMassApply3D_Element(const int e,
   }
 }

-template <int T_D1D, int T_Q1D, int TBATCH, bool ACCUMULATE = true>
-MFEM_HOST_DEVICE inline void
-SmemPAMassApply3D_Element(const int e, const int NE, const real_t *b_,
-                          const real_t *d_, const real_t *x_, real_t *y_,
-                          int d1d = 0, int q1d = 0)
+template<int T_D1D, int T_Q1D, bool ACCUMULATE = true>
+MFEM_HOST_DEVICE inline
+void SmemPAMassApply3D_Element(const int e,
+                               const int NE,
+                               const real_t *b_,
+                               const real_t *d_,
+                               const real_t *x_,
+                               real_t *y_,
+                               const int d1d = 0,
+                               const int q1d = 0)
 {
-   static_assert(TBATCH > 0, "TBATCH must be positive");
-#if defined(__CUDA_ARCH__) || defined(__HIP_DEVICE_COMPILE__)
-   constexpr int tbatch = TBATCH;
-   const int tidz = MFEM_THREAD_ID(z);
-#else
-   // host always batch size 1
-   constexpr int tbatch = 1;
-   constexpr int tidz = 0;
-#endif
-   const int D1D = T_D1D ? T_D1D : d1d;
-   const int Q1D = T_Q1D ? T_Q1D : q1d;
+   constexpr int D1D = T_D1D ? T_D1D : d1d;
+   constexpr int Q1D = T_Q1D ? T_Q1D : q1d;
   constexpr int MQ1 = T_Q1D ? T_Q1D : DofQuadLimits::MAX_Q1D;
   constexpr int MD1 = T_D1D ? T_D1D : DofQuadLimits::MAX_D1D;
   constexpr int MDQ = (MQ1 > MD1) ? MQ1 : MD1;
@@ -841,37 +829,33 @@ SmemPAMassApply3D_Element(const int e, const int NE, const real_t *b_,
   MFEM_SHARED real_t sDQ[MQ1*MD1];
   real_t (*B)[MD1] = (real_t (*)[MD1]) sDQ;
   real_t (*Bt)[MQ1] = (real_t (*)[MQ1]) sDQ;
-   MFEM_SHARED real_t sm0[tbatch][MDQ*MDQ*MDQ];
-   MFEM_SHARED real_t sm1[tbatch][MDQ*MDQ*MDQ];
-   real_t (*X)[MD1][MD1]   = (real_t (*)[MD1][MD1]) (sm0+tidz);
-   real_t (*DDQ)[MD1][MQ1] = (real_t (*)[MD1][MQ1]) (sm1+tidz);
-   real_t (*DQQ)[MQ1][MQ1] = (real_t (*)[MQ1][MQ1]) (sm0+tidz);
-   real_t (*QQQ)[MQ1][MQ1] = (real_t (*)[MQ1][MQ1]) (sm1+tidz);
-   real_t (*QQD)[MQ1][MD1] = (real_t (*)[MQ1][MD1]) (sm0+tidz);
-   real_t (*QDD)[MD1][MD1] = (real_t (*)[MD1][MD1]) (sm1+tidz);
-   MFEM_FOREACH_THREAD(dy, y, D1D)
+   MFEM_SHARED real_t sm0[MDQ*MDQ*MDQ];
+   MFEM_SHARED real_t sm1[MDQ*MDQ*MDQ];
+   real_t (*X)[MD1][MD1]   = (real_t (*)[MD1][MD1]) sm0;
+   real_t (*DDQ)[MD1][MQ1] = (real_t (*)[MD1][MQ1]) sm1;
+   real_t (*DQQ)[MQ1][MQ1] = (real_t (*)[MQ1][MQ1]) sm0;
+   real_t (*QQQ)[MQ1][MQ1] = (real_t (*)[MQ1][MQ1]) sm1;
+   real_t (*QQD)[MQ1][MD1] = (real_t (*)[MQ1][MD1]) sm0;
+   real_t (*QDD)[MD1][MD1] = (real_t (*)[MD1][MD1]) sm1;
+   MFEM_FOREACH_THREAD(dy,y,D1D)
   {
-      MFEM_FOREACH_THREAD(dx, x, D1D)
+      MFEM_FOREACH_THREAD(dx,x,D1D)
      {
         MFEM_UNROLL(MD1)
         for (int dz = 0; dz < D1D; ++dz)
         {
-            X[dz][dy][dx] = x(dx, dy, dz, e);
+            X[dz][dy][dx] = x(dx,dy,dz,e);
         }
      }
-      MFEM_FOREACH_THREAD(dx, x, Q1D) { B[dx][dy] = b(dx, dy); }
-   }
-   if (tidz == 0)
-   {
-      MFEM_FOREACH_THREAD(dy, y, D1D)
+      MFEM_FOREACH_THREAD(dx,x,Q1D)
      {
-         MFEM_FOREACH_THREAD(dx, x, Q1D) { B[dx][dy] = b(dx, dy); }
+         B[dx][dy] = b(dx,dy);
      }
   }
   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(dy, y, D1D)
+   MFEM_FOREACH_THREAD(dy,y,D1D)
   {
-      MFEM_FOREACH_THREAD(qx, x, Q1D)
+      MFEM_FOREACH_THREAD(qx,x,Q1D)
      {
         real_t u[D1D];
         MFEM_UNROLL(MD1)
@@ -896,9 +880,9 @@ SmemPAMassApply3D_Element(const int e, const int NE, const real_t *b_,
      }
   }
   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(qy, y, Q1D)
+   MFEM_FOREACH_THREAD(qy,y,Q1D)
   {
-      MFEM_FOREACH_THREAD(qx, x, Q1D)
+      MFEM_FOREACH_THREAD(qx,x,Q1D)
      {
         real_t u[D1D];
         MFEM_UNROLL(MD1)
@@ -923,9 +907,9 @@ SmemPAMassApply3D_Element(const int e, const int NE, const real_t *b_,
      }
   }
   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(qy, y, Q1D)
+   MFEM_FOREACH_THREAD(qy,y,Q1D)
   {
-      MFEM_FOREACH_THREAD(qx, x, Q1D)
+      MFEM_FOREACH_THREAD(qx,x,Q1D)
      {
         real_t u[Q1D];
         MFEM_UNROLL(MQ1)
@@ -945,22 +929,22 @@ SmemPAMassApply3D_Element(const int e, const int NE, const real_t *b_,
         MFEM_UNROLL(MQ1)
         for (int qz = 0; qz < Q1D; qz++)
         {
-            QQQ[qz][qy][qx] = u[qz] * d(qx, qy, qz, e);
+            QQQ[qz][qy][qx] = u[qz] * d(qx,qy,qz,e);
         }
      }
   }
   MFEM_SYNC_THREAD;
-   if (tidz == 0)
+   MFEM_FOREACH_THREAD(di,y,D1D)
   {
-      MFEM_FOREACH_THREAD(di, y, D1D)
+      MFEM_FOREACH_THREAD(q,x,Q1D)
      {
-         MFEM_FOREACH_THREAD(q, x, Q1D) { Bt[di][q] = b(q, di); }
+         Bt[di][q] = b(q,di);
      }
   }
   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(qy, y, Q1D)
+   MFEM_FOREACH_THREAD(qy,y,Q1D)
   {
-      MFEM_FOREACH_THREAD(dx, x, D1D)
+      MFEM_FOREACH_THREAD(dx,x,D1D)
      {
         real_t u[Q1D];
         MFEM_UNROLL(MQ1)
@@ -985,9 +969,9 @@ SmemPAMassApply3D_Element(const int e, const int NE, const real_t *b_,
      }
   }
   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(dy, y, D1D)
+   MFEM_FOREACH_THREAD(dy,y,D1D)
   {
-      MFEM_FOREACH_THREAD(dx, x, D1D)
+      MFEM_FOREACH_THREAD(dx,x,D1D)
      {
         real_t u[Q1D];
         MFEM_UNROLL(MQ1)
@@ -1012,9 +996,9 @@ SmemPAMassApply3D_Element(const int e, const int NE, const real_t *b_,
      }
   }
   MFEM_SYNC_THREAD;
-   MFEM_FOREACH_THREAD(dy, y, D1D)
+   MFEM_FOREACH_THREAD(dy,y,D1D)
   {
-      MFEM_FOREACH_THREAD(dx, x, D1D)
+      MFEM_FOREACH_THREAD(dx,x,D1D)
      {
         real_t u[D1D];
         MFEM_UNROLL(MD1)
@@ -1036,11 +1020,11 @@ SmemPAMassApply3D_Element(const int e, const int NE, const real_t *b_,
         {
            if (ACCUMULATE)
            {
-               y(dx, dy, dz, e) += u[dz];
+               y(dx,dy,dz,e) += u[dz];
            }
            else
            {
-               y(dx, dy, dz, e) = u[dz];
+               y(dx,dy,dz,e) = u[dz];
            }
         }
      }
@@ -1131,8 +1115,8 @@ inline void PAMassApply3D(const int NE,
   });
 }

-// Shared memory PA Mass Apply 3D kernel
-template<int T_D1D = 0, int T_Q1D = 0, int TBATCH=1>
+// Shared memory PA Mass Apply 2D kernel
+template<int T_D1D = 0, int T_Q1D = 0>
 inline void SmemPAMassApply3D(const int NE,
                              const Array<real_t> &b_,
                              const Array<real_t> &bt_,
@@ -1142,9 +1126,6 @@ inline void SmemPAMassApply3D(const int NE,
                              const int d1d = 0,
                              const int q1d = 0)
 {
-   static_assert(T_D1D > 0, "T_D1D must be positive");
-   static_assert(T_Q1D > 0, "T_Q1D must be positive");
-   static_assert(TBATCH > 0, "TBATCH must be positive");
   MFEM_CONTRACT_VAR(bt_);
   const int D1D = T_D1D ? T_D1D : d1d;
   const int Q1D = T_Q1D ? T_Q1D : q1d;
@@ -1156,11 +1137,9 @@ inline void SmemPAMassApply3D(const int NE,
   const auto d = d_.Read();
   const auto x = x_.Read();
   auto y = y_.ReadWrite();
-   mfem::forall_2D_batch<T_Q1D * T_Q1D * TBATCH>(NE, Q1D, Q1D, TBATCH,
-                                                 [=] MFEM_HOST_DEVICE(int e)
+   mfem::forall_2D<T_Q1D*T_Q1D>(NE, Q1D, Q1D, [=] MFEM_HOST_DEVICE (int e)
   {
-      internal::SmemPAMassApply3D_Element<T_D1D, T_Q1D, TBATCH>(e, NE, b, d, x,
-                                                                y, d1d, q1d);
+      internal::SmemPAMassApply3D_Element<T_D1D,T_Q1D>(e, NE, b, d, x, y, d1d, q1d);
   });
 }

@@ -1410,57 +1389,42 @@ using ApplyKernelType = MassIntegrator::ApplyKernelType;
 using DiagonalKernelType = MassIntegrator::DiagonalKernelType;
 }

-template<int DIM, int D1D, int Q1D>
+template<int DIM, int T_D1D, int T_Q1D>
 ApplyKernelType MassIntegrator::ApplyPAKernels::Kernel()
 {
   if constexpr (DIM == 1) { return internal::PAMassApply1D; }
-   else if constexpr (DIM == 2) { return internal::SmemPAMassApply2D<D1D, Q1D>; }
-   else if constexpr (DIM == 3)
-   {
-      constexpr int MDQ = D1D >= Q1D ? D1D : Q1D;
-      // max 64 threads in z limit in cuda and hip
-      if constexpr (MDQ > 0)
-      {
-         return internal::SmemPAMassApply3D<D1D, Q1D,
-                internal::mass::NBZ3D(MDQ)>;
-      }
-   }
-   else { MFEM_ABORT(""); }
-   return nullptr;
+   else if constexpr (DIM == 2) { return internal::SmemPAMassApply2D<T_D1D,T_Q1D>; }
+   else if constexpr (DIM == 3) { return internal::SmemPAMassApply3D<T_D1D, T_Q1D>; }
+   MFEM_ABORT("");
 }

 inline ApplyKernelType MassIntegrator::ApplyPAKernels::Fallback(
-   int dim, int, int)
+   int DIM, int, int)
 {
-   if (dim == 1) { return internal::PAMassApply1D; }
-   else if (dim == 2) { return internal::PAMassApply2D; }
-   else if (dim == 3) { return internal::PAMassApply3D; }
+   if (DIM == 1) { return internal::PAMassApply1D; }
+   else if (DIM == 2) { return internal::PAMassApply2D; }
+   else if (DIM == 3) { return internal::PAMassApply3D; }
   else { MFEM_ABORT(""); }
-   return nullptr;
 }

-template<int DIM, int D1D, int Q1D>
+template<int DIM, int T_D1D, int T_Q1D>
 DiagonalKernelType MassIntegrator::DiagonalPAKernels::Kernel()
 {
   if constexpr (DIM == 1) { return internal::PAMassAssembleDiagonal1D; }
-   else if constexpr (DIM == 2) { return internal::SmemPAMassAssembleDiagonal2D<D1D, Q1D>; }
-   else if constexpr (DIM == 3) { return internal::SmemPAMassAssembleDiagonal3D<D1D, Q1D>; }
-   else { MFEM_ABORT(""); }
-   return nullptr;
+   else if constexpr (DIM == 2) { return internal::SmemPAMassAssembleDiagonal2D<T_D1D,T_Q1D>; }
+   else if constexpr (DIM == 3) { return internal::SmemPAMassAssembleDiagonal3D<T_D1D, T_Q1D>; }
+   MFEM_ABORT("");
 }

 inline DiagonalKernelType MassIntegrator::DiagonalPAKernels::Fallback(
-   int dim, int, int)
+   int DIM, int, int)
 {
-   if (dim == 1) { return internal::PAMassAssembleDiagonal1D; }
-   else if (dim == 2) { return internal::PAMassAssembleDiagonal2D; }
-   else if (dim == 3) { return internal::PAMassAssembleDiagonal3D; }
+   if (DIM == 1) { return internal::PAMassAssembleDiagonal1D; }
+   else if (DIM == 2) { return internal::PAMassAssembleDiagonal2D; }
+   else if (DIM == 3) { return internal::PAMassAssembleDiagonal3D; }
   else { MFEM_ABORT(""); }
-   return nullptr;
 }
-
 /// \endcond DO_NOT_DOCUMENT
-
 } // namespace mfem

 #endif
@@ -15,7 +15,6 @@
 #include "../qfunction.hpp"
 #include "../ceed/integrators/mass/mass.hpp"
 #include "bilininteg_mass_kernels.hpp"
-#include "bilininteg_mass_pa_simplices.hpp"

 namespace mfem
 {
@@ -30,11 +29,9 @@ void MassIntegrator::AssemblePA(const FiniteElementSpace &fes)
   // Assuming the same element type
   fespace = &fes;
   Mesh *mesh = fes.GetMesh();
-   dim = mesh->Dimension();
   const FiniteElement &el = *fes.GetTypicalFE();
   ElementTransformation *T0 = mesh->GetTypicalElementTransformation();
-   const bool stroud = fes.UsesRaggedTensorBasis();
-   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el, *T0, stroud);
+   const IntegrationRule *ir = IntRule ? IntRule : &GetRule(el, el, *T0);
   if (DeviceCanUseCeed())
   {
      delete ceedOp;
@@ -51,25 +48,17 @@ void MassIntegrator::AssemblePA(const FiniteElementSpace &fes)
      return;
   }
   int map_type = el.GetMapType();
+   dim = mesh->Dimension();
   ne = fes.GetMesh()->GetNE();
   nq = ir->GetNPoints();
   geom = mesh->GetGeometricFactors(*ir, GeometricFactors::DETERMINANTS, mt);
-   if (stroud)
-   {
-      maps = &el.GetDofToQuad(*ir, DofToQuad::RAGGED_TENSOR);
-   }
-   else
-   {
-      maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
-   }
+   maps = &el.GetDofToQuad(*ir, DofToQuad::TENSOR);
   dofs1D = maps->ndof;
   quad1D = maps->nqpt;
   pa_data.SetSize(ne*nq, mt);

   QuadratureSpace qs(*mesh, *ir);
   CoefficientVector coeff(Q, qs, CoefficientStorage::COMPRESSED);
-   // QuadratureSpace expects ir defined in reference simplex for Bernstein
-   // elements with partial assembly
   {
      const int NE = ne;
      const int NQ = nq;
@@ -158,10 +147,9 @@ void MassIntegrator::AddMultPA(const Vector &x, Vector &y) const
   {
      const int D1D = dofs1D;
      const int Q1D = quad1D;
-      const Vector &D = pa_data;
      const Array<real_t> &B = maps->B;
      const Array<real_t> &Bt = maps->Bt;
-
+      const Vector &D = pa_data;
 #ifdef MFEM_USE_OCCA
      if (DeviceCanUseOcca())
      {
@@ -176,31 +164,7 @@ void MassIntegrator::AddMultPA(const Vector &x, Vector &y) const
         MFEM_ABORT("OCCA PA Mass Apply unknown kernel!");
      }
 #endif // MFEM_USE_OCCA
-
-      if (fespace->UsesRaggedTensorBasis())
-      {
-         const auto *rmaps = static_cast<const RaggedDofToQuad*>(maps);
-
-         const Array<real_t> &Ba1 = rmaps->Ba1;
-         const Array<real_t> &Ba2 = rmaps->Ba2;
-         const Array<real_t> &Ba3 = rmaps->Ba3;
-         const Array<real_t> &Ba1t = rmaps->Ba1t;
-         const Array<real_t> &Ba2t = rmaps->Ba2t;
-         const Array<real_t> &Ba3t = rmaps->Ba3t;
-         const Array<int> &lex_map = rmaps->lex_map;
-         const Array<int> &forward_map2d = rmaps->forward_map2d_mass;
-         const Array<int> &inverse_map2d = rmaps->inverse_map2d_mass;
-         const Array<int> &forward_map3d = rmaps->forward_map3d_mass;
-         const Array<int> &inverse_map3d = rmaps->inverse_map3d_mass;
-         ApplySimplexPAKernels::Run(dim, D1D, Q1D, ne, lex_map, forward_map2d,
-                                    inverse_map2d,
-                                    forward_map3d, inverse_map3d, Ba1, Ba2, Ba3, Ba1t, Ba2t, Ba3t,
-                                    D, x, y, D1D, Q1D);
-      }
-      else
-      {
-         ApplyPAKernels::Run(dim, D1D, Q1D, ne, B, Bt, D, x, y, D1D, Q1D);
-      }
+      ApplyPAKernels::Run(dim, D1D, Q1D, ne, B, Bt, D, x, y, D1D, Q1D);
   }
 }

@@ -213,8 +177,6 @@ void MassIntegrator::AddAbsMultPA(const Vector &x, Vector &y) const
   }
   else
   {
-      MFEM_VERIFY(!fespace->UsesRaggedTensorBasis(),
-                  "AbsMultPA not implemented for ragged tensor basis");
      Vector abs_pa_data(pa_data);
      abs_pa_data.Abs();
      Array<real_t> absB(maps->B);
@@ -307,506 +307,6 @@ DomainLFIntegrator::AssembleKernels::Kernel()
   MFEM_ABORT("");
 }

-template <int T_D1D = 0, int T_Q1D = 0>
-static void HdivDLFAssemble2D(const int ne, const Array<int> &markers,
-                              const Vector &jac, const Array<real_t> &weights,
-                              const Array<real_t> &testBO,
-                              const Array<real_t> &testBC, const Vector &coeff,
-                              Vector &y, const int d, const int q)
-{
-   MFEM_VERIFY(T_D1D || d <= DeviceDofQuadLimits::Get().HDIV_MAX_D1D,
-               "Problem size too large.");
-   MFEM_VERIFY(T_Q1D || q <= DeviceDofQuadLimits::Get().HDIV_MAX_Q1D,
-               "Problem size too large.");
-   MFEM_VERIFY(y.Size() == 2 * (d - 1) * d * ne, "");
-
-   constexpr int vdim = 2;
-   const auto F = coeff.Read();
-   const auto M = markers.Read();
-   const auto BO = Reshape(testBO.Read(), q, d-1);
-   const auto BC = Reshape(testBC.Read(), q, d);
-   const auto J = Reshape(jac.Read(), q, q, vdim, vdim, ne);
-   const auto W = Reshape(weights.Read(), q, q);
-   const bool cst = coeff.Size() == vdim;
-   const auto C = cst ? Reshape(F,vdim,1,1,1) : Reshape(F,vdim,q,q,ne);
-   auto Y = y.ReadWrite();
-
-   mfem::forall_3D(ne, q, q, vdim, [=] MFEM_HOST_DEVICE (int e)
-   {
-      constexpr int vdim = 2;
-      if (M[e] == 0) { return; } // ignore
-
-      constexpr int Q = T_Q1D ? T_Q1D : DofQuadLimits::HDIV_MAX_Q1D;
-      constexpr int D = T_D1D ? T_D1D : DofQuadLimits::HDIV_MAX_D1D;
-
-      MFEM_SHARED real_t sBot[Q*D];
-      MFEM_SHARED real_t sBct[Q*D];
-      MFEM_SHARED real_t sQQ[vdim*Q*Q];
-      MFEM_SHARED real_t sQD[vdim*Q*D];
-
-      // Bo and Bc into shared memory
-      const DeviceMatrix Bot(sBot, d-1, q);
-      kernels::internal::LoadB<D,Q>(d-1, q, BO, sBot);
-      const DeviceMatrix Bct(sBct, d, q);
-      kernels::internal::LoadB<D,Q>(d, q, BC, sBct);
-
-      const DeviceCube QQ(sQQ, q, q, vdim);
-      const DeviceCube QD(sQD, q, d, vdim);
-
-      MFEM_FOREACH_THREAD(vd,z,vdim)
-      {
-         const real_t cst_val_0 = C(0,0,0,0);
-         const real_t cst_val_1 = C(1,0,0,0);
-         MFEM_FOREACH_THREAD(y,y,q)
-         {
-            MFEM_FOREACH_THREAD(x,x,q)
-            {
-               const real_t J0 = J(x,y,0,vd,e);
-               const real_t J1 = J(x,y,1,vd,e);
-               const real_t C0 = cst ? cst_val_0 : C(0,x,y,e);
-               const real_t C1 = cst ? cst_val_1 : C(1,x,y,e);
-               QQ(x,y,vd) = W(x,y)*(J0*C0 + J1*C1);
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,vdim)
-      {
-         const int nx = (vd == 0) ? d : d-1;
-         DeviceMatrix Btx = (vd == 0) ? Bct : Bot;
-         MFEM_FOREACH_THREAD(qy,y,q)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               real_t qd = 0.0;
-               for (int qx = 0; qx < q; ++qx)
-               {
-                  qd += QQ(qx,qy,vd) * Btx(dx,qx);
-               }
-               QD(dx,qy,vd) = qd;
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,vdim)
-      {
-         const int nx = (vd == 0) ? d : d-1;
-         const int ny = (vd == 1) ? d : d-1;
-         DeviceMatrix Bty = (vd == 1) ? Bct : Bot;
-         DeviceTensor<4> Yxy(Y, nx, ny, vdim, ne);
-         MFEM_FOREACH_THREAD(dy,y,ny)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               real_t dd = 0.0;
-               for (int qy = 0; qy < q; ++qy)
-               {
-                  dd += QD(dx,qy,vd) * Bty(dy,qy);
-               }
-               Yxy(dx,dy,vd,e) += dd;
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-   });
-}
-
-template <int T_D1D = 0, int T_Q1D = 0>
-static void HdivDLFAssemble3D(const int ne, const Array<int> &markers,
-                              const Vector &jac, const Array<real_t> &weights,
-                              const Array<real_t> &testBO,
-                              const Array<real_t> &testBC, const Vector &coeff,
-                              Vector &y, const int d, const int q)
-{
-   MFEM_VERIFY(T_D1D || d <= DeviceDofQuadLimits::Get().HDIV_MAX_D1D,
-               "Problem size too large.");
-   MFEM_VERIFY(T_Q1D || q <= DeviceDofQuadLimits::Get().HDIV_MAX_Q1D,
-               "Problem size too large.");
-   MFEM_VERIFY(y.Size() == 3 * (d - 1) * (d - 1) * d * ne, "y wrong length");
-
-   constexpr int vdim = 3;
-   const auto F = coeff.Read();
-   const auto M = markers.Read();
-   const auto BO = Reshape(testBO.Read(), q, d-1);
-   const auto BC = Reshape(testBC.Read(), q, d);
-   const auto J = Reshape(jac.Read(), q, q, q, vdim, vdim, ne);
-   const auto W = Reshape(weights.Read(), q, q, q);
-   const bool cst = coeff.Size() == vdim;
-   const auto C = cst ? Reshape(F,vdim,1,1,1,1) : Reshape(F,vdim,q,q,q,ne);
-   auto Y = y.ReadWrite();
-
-   mfem::forall_3D(ne, q, q, vdim, [=] MFEM_HOST_DEVICE (int e)
-   {
-      constexpr int vdim = 3;
-      if (M[e] == 0) { return; } // ignore
-
-      constexpr int Q = T_Q1D ? T_Q1D : DofQuadLimits::HDIV_MAX_Q1D;
-      constexpr int D = T_D1D ? T_D1D : DofQuadLimits::HDIV_MAX_D1D;
-
-      MFEM_SHARED real_t sBot[Q*D];
-      MFEM_SHARED real_t sBct[Q*D];
-
-      // Bo and Bc into shared memory
-      const DeviceMatrix Bot(sBot, d-1, q);
-      kernels::internal::LoadB<D,Q>(d-1, q, BO, sBot);
-      const DeviceMatrix Bct(sBct, d, q);
-      kernels::internal::LoadB<D,Q>(d, q, BC, sBct);
-
-      MFEM_SHARED real_t sm0[vdim*Q*Q*Q];
-      MFEM_SHARED real_t sm1[vdim*Q*Q*Q];
-      DeviceTensor<4> QQQ(sm1, q, q, q, vdim);
-      DeviceTensor<4> DQQ(sm0, d, q, q, vdim);
-      DeviceTensor<4> DDQ(sm1, d, d, q, vdim);
-
-      MFEM_FOREACH_THREAD(vd,z,vdim)
-      {
-         const real_t cst_val_0 = C(0,0,0,0,0);
-         const real_t cst_val_1 = C(1,0,0,0,0);
-         const real_t cst_val_2 = C(2,0,0,0,0);
-         MFEM_FOREACH_THREAD(y,y,q)
-         {
-            MFEM_FOREACH_THREAD(x,x,q)
-            {
-               for (int z = 0; z < q; ++z)
-               {
-                  const real_t J0 = J(x,y,z,0,vd,e);
-                  const real_t J1 = J(x,y,z,1,vd,e);
-                  const real_t J2 = J(x,y,z,2,vd,e);
-                  const real_t C0 = cst ? cst_val_0 : C(0,x,y,z,e);
-                  const real_t C1 = cst ? cst_val_1 : C(1,x,y,z,e);
-                  const real_t C2 = cst ? cst_val_2 : C(2,x,y,z,e);
-                  QQQ(x,y,z,vd) = W(x,y,z)*(J0*C0 + J1*C1 + J2*C2);
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      // Apply Bt operator
-      MFEM_FOREACH_THREAD(vd,z,vdim)
-      {
-         const int nx = (vd == 0) ? d : d-1;
-         DeviceMatrix Btx = (vd == 0) ? Bct : Bot;
-         MFEM_FOREACH_THREAD(qy,y,q)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               real_t u[Q];
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz) { u[qz] = 0.0; }
-               MFEM_UNROLL(Q)
-               for (int qx = 0; qx < q; ++qx)
-               {
-                  MFEM_UNROLL(Q)
-                  for (int qz = 0; qz < q; ++qz)
-                  {
-                     u[qz] += QQQ(qx,qy,qz,vd) * Btx(dx,qx);
-                  }
-               }
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz) { DQQ(dx,qy,qz,vd) = u[qz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,vdim)
-      {
-         const int nx = (vd == 0) ? d : d-1;
-         const int ny = (vd == 1) ? d : d-1;
-         DeviceMatrix Bty = (vd == 1) ? Bct : Bot;
-         MFEM_FOREACH_THREAD(dy,y,ny)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               real_t u[Q];
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz) { u[qz] = 0.0; }
-               MFEM_UNROLL(Q)
-               for (int qy = 0; qy < q; ++qy)
-               {
-                  MFEM_UNROLL(Q)
-                  for (int qz = 0; qz < q; ++qz)
-                  {
-                     u[qz] += DQQ(dx,qy,qz,vd) * Bty(dy,qy);
-                  }
-               }
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz) { DDQ(dx,dy,qz,vd) = u[qz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd,z,vdim)
-      {
-         const int nx = (vd == 0) ? d : d-1;
-         const int ny = (vd == 1) ? d : d-1;
-         const int nz = (vd == 2) ? d : d-1;
-         DeviceTensor<5> Yxyz(Y, nx, ny, nz, vdim, ne);
-         DeviceMatrix Btz = (vd == 2) ? Bct : Bot;
-         MFEM_FOREACH_THREAD(dy,y,ny)
-         {
-            MFEM_FOREACH_THREAD(dx,x,nx)
-            {
-               real_t u[D];
-               MFEM_UNROLL(D)
-               for (int dz = 0; dz < nz; ++dz) { u[dz] = 0.0; }
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz)
-               {
-                  MFEM_UNROLL(D)
-                  for (int dz = 0; dz < nz; ++dz)
-                  {
-                     u[dz] += DDQ(dx,dy,qz,vd) * Btz(dz,qz);
-                  }
-               }
-               MFEM_UNROLL(D)
-               for (int dz = 0; dz < nz; ++dz) { Yxyz(dx,dy,dz,vd,e) += u[dz]; }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-   });
-}
-
-/// @param ne number of elements
-/// @param markers array where entry markers[e] == 0 to skip assembly over
-/// element e element
-/// @param jac Spatial Jacobians evaluated at all quadrature points
-/// @param weights 1D quadrature weights
-/// @param testBO 1D open basis test functions
-/// @param testBC 1D closed basis test functions
-/// @param coeff coefficient values evaluated at quadrature points, possibly
-/// compressed.
-/// @param d number of 1D closed dofs
-/// @param q number of 1D quadrature points
-/// @tparam T_D1D maximum number of dofs along any direction, or 0
-/// @tparam T_Q1D maximum number of quadrature points along any direction, or 0
-template <int T_D1D = 0, int T_Q1D = 0>
-static void HcurlDLFAssemble3D(const int ne, const Array<int> &markers,
-                               const Vector &jac, const Array<real_t> &weights,
-                               const Array<real_t> &testBO,
-                               const Array<real_t> &testBC, const Vector &coeff,
-                               Vector &y, const int d, const int q)
-{
-   MFEM_VERIFY(T_D1D || d <= DeviceDofQuadLimits::Get().HCURL_MAX_D1D,
-               "Problem size too large.");
-   MFEM_VERIFY(T_Q1D || q <= DeviceDofQuadLimits::Get().HCURL_MAX_Q1D,
-               "Problem size too large.");
-   MFEM_VERIFY(y.Size() == 3 * (d - 1) * d * d * ne, "y wrong length");
-
-   constexpr int vdim = 3;
-   const auto F = coeff.Read();
-   const auto M = markers.Read();
-   const auto BO = Reshape(testBO.Read(), q, d-1);
-   const auto BC = Reshape(testBC.Read(), q, d);
-   const auto J = Reshape(jac.Read(), q, q, q, vdim, vdim, ne);
-   const auto W = Reshape(weights.Read(), q, q, q);
-   const bool cst = coeff.Size() == vdim;
-   const auto C = cst ? Reshape(F,vdim,1,1,1,1) : Reshape(F,vdim,q,q,q,ne);
-   auto Y = y.ReadWrite();
-
-   mfem::forall_3D(ne, q, q, vdim, [=] MFEM_HOST_DEVICE(int e)
-   {
-      if (M[e] == 0)
-      {
-         // ignore
-         return;
-      }
-
-      constexpr int vdim = 3;
-      constexpr int Q = T_Q1D ? T_Q1D : DofQuadLimits::HCURL_MAX_Q1D;
-      constexpr int D = T_D1D ? T_D1D : DofQuadLimits::HCURL_MAX_D1D;
-
-      MFEM_SHARED real_t sBot[Q * D];
-      MFEM_SHARED real_t sBct[Q * D];
-
-      // Bo and Bc into shared memory
-      const DeviceMatrix Bot(sBot, d - 1, q);
-      kernels::internal::LoadB<D, Q>(d - 1, q, BO, sBot);
-      const DeviceMatrix Bct(sBct, d, q);
-      kernels::internal::LoadB<D, Q>(d, q, BC, sBct);
-
-      MFEM_SHARED real_t sm0[vdim * Q * Q * Q];
-      MFEM_SHARED real_t sm1[vdim * Q * Q * Q];
-      DeviceTensor<4> QQQ(sm1, q, q, q, vdim);
-      DeviceTensor<4> DQQ(sm0, d, q, q, vdim);
-      DeviceTensor<4> DDQ(sm1, d, d, q, vdim);
-
-      const real_t cst_val_0 = C(0, 0, 0, 0, 0);
-      const real_t cst_val_1 = C(1, 0, 0, 0, 0);
-      const real_t cst_val_2 = C(2, 0, 0, 0, 0);
-
-      MFEM_FOREACH_THREAD(vd, z, vdim)
-      {
-         MFEM_FOREACH_THREAD(y, y, q)
-         {
-            MFEM_FOREACH_THREAD(x, x, q)
-            {
-               for (int z = 0; z < q; ++z)
-               {
-                  real_t curr[3];
-                  curr[0] = cst ? cst_val_0 : C(0, x, y, z, e);
-                  curr[1] = cst ? cst_val_1 : C(1, x, y, z, e);
-                  curr[2] = cst ? cst_val_2 : C(2, x, y, z, e);
-
-                  const real_t J11 = J(x, y, z, 0, 0, e);
-                  const real_t J21 = J(x, y, z, 1, 0, e);
-                  const real_t J31 = J(x, y, z, 2, 0, e);
-                  const real_t J12 = J(x, y, z, 0, 1, e);
-                  const real_t J22 = J(x, y, z, 1, 1, e);
-                  const real_t J32 = J(x, y, z, 2, 1, e);
-                  const real_t J13 = J(x, y, z, 0, 2, e);
-                  const real_t J23 = J(x, y, z, 1, 2, e);
-                  const real_t J33 = J(x, y, z, 2, 2, e);
-                  // adj(J)
-                  const real_t A11 = (J22 * J33) - (J23 * J32);
-                  const real_t A12 = (J32 * J13) - (J12 * J33);
-                  const real_t A13 = (J12 * J23) - (J22 * J13);
-                  const real_t A21 = (J31 * J23) - (J21 * J33);
-                  const real_t A22 = (J11 * J33) - (J13 * J31);
-                  const real_t A23 = (J21 * J13) - (J11 * J23);
-                  const real_t A31 = (J21 * J32) - (J31 * J22);
-                  const real_t A32 = (J31 * J12) - (J11 * J32);
-                  const real_t A33 = (J11 * J22) - (J12 * J21);
-                  const real_t A[9] = {A11, A12, A13, A21, A22,
-                                       A23, A31, A32, A33
-                                      };
-                  QQQ(x, y, z, vd) = W(x, y, z) * (A[vd * vdim] * curr[0] +
-                                                   A[vd * vdim + 1] * curr[1] +
-                                                   A[vd * vdim + 2] * curr[2]);
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      // Apply Bt operator
-      MFEM_FOREACH_THREAD(vd, z, vdim)
-      {
-         const int nx = (vd == 0) ? d - 1 : d;
-         DeviceMatrix Btx = (vd == 0) ? Bot : Bct;
-         MFEM_FOREACH_THREAD(qy, y, q)
-         {
-            MFEM_FOREACH_THREAD(dx, x, nx)
-            {
-               real_t u[Q];
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz)
-               {
-                  u[qz] = 0.0;
-               }
-               MFEM_UNROLL(Q)
-               for (int qx = 0; qx < q; ++qx)
-               {
-                  MFEM_UNROLL(Q)
-                  for (int qz = 0; qz < q; ++qz)
-                  {
-                     u[qz] += QQQ(qx, qy, qz, vd) * Btx(dx, qx);
-                  }
-               }
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz)
-               {
-                  DQQ(dx, qy, qz, vd) = u[qz];
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd, z, vdim)
-      {
-         const int nx = (vd == 0) ? d - 1 : d;
-         const int ny = (vd == 1) ? d - 1 : d;
-         DeviceMatrix Bty = (vd == 1) ? Bot : Bct;
-         MFEM_FOREACH_THREAD(dy, y, ny)
-         {
-            MFEM_FOREACH_THREAD(dx, x, nx)
-            {
-               real_t u[Q];
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz)
-               {
-                  u[qz] = 0.0;
-               }
-               MFEM_UNROLL(Q)
-               for (int qy = 0; qy < q; ++qy)
-               {
-                  MFEM_UNROLL(Q)
-                  for (int qz = 0; qz < q; ++qz)
-                  {
-                     u[qz] += DQQ(dx, qy, qz, vd) * Bty(dy, qy);
-                  }
-               }
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz)
-               {
-                  DDQ(dx, dy, qz, vd) = u[qz];
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-      MFEM_FOREACH_THREAD(vd, z, vdim)
-      {
-         const int nx = (vd == 0) ? d - 1 : d;
-         const int ny = (vd == 1) ? d - 1 : d;
-         const int nz = (vd == 2) ? d - 1 : d;
-         DeviceTensor<5> Yxyz(Y, nx, ny, nz, vdim, ne);
-         DeviceMatrix Btz = (vd == 2) ? Bot : Bct;
-         MFEM_FOREACH_THREAD(dy, y, ny)
-         {
-            MFEM_FOREACH_THREAD(dx, x, nx)
-            {
-               real_t u[D];
-               MFEM_UNROLL(D)
-               for (int dz = 0; dz < nz; ++dz)
-               {
-                  u[dz] = 0.0;
-               }
-               MFEM_UNROLL(Q)
-               for (int qz = 0; qz < q; ++qz)
-               {
-                  MFEM_UNROLL(D)
-                  for (int dz = 0; dz < nz; ++dz)
-                  {
-                     u[dz] += DDQ(dx, dy, qz, vd) * Btz(dz, qz);
-                  }
-               }
-               MFEM_UNROLL(D)
-               for (int dz = 0; dz < nz; ++dz)
-               {
-                  Yxyz(dx, dy, dz, vd, e) += u[dz];
-               }
-            }
-         }
-      }
-      MFEM_SYNC_THREAD;
-   });
-}
-
-template <FiniteElement::DerivType TestType, int DIM, int TEST_D1D, int Q1D>
-VectorFEDomainLFIntegrator::AssembleKernelType
-VectorFEDomainLFIntegrator::AssembleKernels::Kernel()
-{
-   if constexpr (TestType == FiniteElement::DIV)
-   {
-      if constexpr (DIM == 2)
-      {
-         return HdivDLFAssemble2D<TEST_D1D, Q1D>;
-      }
-      if constexpr (DIM == 3)
-      {
-         return HdivDLFAssemble3D<TEST_D1D, Q1D>;
-      }
-   }
-   if constexpr (TestType == FiniteElement::CURL)
-   {
-      if constexpr (DIM == 3)
-      {
-         return HcurlDLFAssemble3D<TEST_D1D, Q1D>;
-      }
-   }
-   MFEM_ABORT("");
-}
-
 /// \endcond DO_NOT_DOCUMENT

 } // namespace mfem
--- a/Show More
+++ b/Show More