NanoVDB: remove multi-GPU diagnostics now that the race is fixed

Drop the per-device stripe/voxel-count fprintf in DistributedPointsToGrid and the "--repeat until-fail:20" mgpu loop in the CodeBuild buildspec. Both were temporary scaffolding for diagnosing the shared-tile race, which is now fixed and covered by dedicated regression tests. Signed-off-by: Jonathan Swartz <jonathan@jswartz.info>
NanoVDB: drop FewerVoxelsThanDevices multi-GPU test
2026-08-07 15:45:12 +12:00 · 2026-08-07 15:40:23 +12:00 · 2026-08-07 14:23:44 +12:00 · 2026-08-07 14:20:03 +12:00 · 2026-08-07 12:14:22 +12:00 · 2026-08-07 12:14:22 +12:00
517 changed files with 108433 additions and 18039 deletions
@@ -30,9 +30,6 @@ on:
      - 'fvdb/**'
      - 'pendingchanges/**'
      - '**.md'
-  schedule:
-    # run this workflow every day 7am UTC
-    - cron:  '0 7 * * *'
  workflow_dispatch:
    inputs:
      type:
@@ -67,17 +64,22 @@ jobs:
      matrix:
        # Only test unified builds (core+ax). weekly CI should test standalone
        config:
-          - { image: '2023-clang15', cxx: 'clang++', build: 'Release', cmake: '' }
-          - { image: '2023-clang15', cxx: 'g++',     build: 'Release', cmake: '' }
-          - { image: '2023-clang15', cxx: 'clang++', build: 'Debug',   cmake: '' }
+          - { image: '2026',           cxx: 'clang++', build: 'Release', cmake: '' }
+          - { image: '2026',           cxx: 'g++',     build: 'Release', cmake: '' }
+          - { image: '2026',           cxx: 'clang++', build: 'Debug',   cmake: '' }
+          - { image: '2025-clang19.1', cxx: 'clang++', build: 'Release', cmake: '' }
+          - { image: '2024-clang16.2', cxx: 'clang++', build: 'Release', cmake: '' }
+          # Needed to test LLVM 15.X
+          - { image: '2023-clang15',   cxx: 'clang++', build: 'Release', cmake: '-DDISABLE_DEPENDENCY_VERSION_CHECKS=ON' }
      fail-fast: false
    steps:
      - uses: actions/checkout@v3
      - name: nanobind
-        #if: contains(matrix.config.image, '2023') == false
-        run: ./ci/install_nanobind.sh 2.0.0
-      - name: install_gtest
-        run: ./ci/install_gtest.sh 1.15.2
+        run: ./ci/install_nanobind.sh 2.5.0
+        # 2026 images are broken with gtest/glfw installs
+      - name: deps
+        if: contains(matrix.config.image, '2026')
+        run: ./ci/install_gtest.sh latest
      - name: timestamp
        id: timestamp
        run: echo "timestamp=$(date -u +'%Y-%m-%dT%H:%M:%SZ')" >> $GITHUB_OUTPUT
@@ -94,12 +96,11 @@ jobs:
        run: >
          ./ci/build.sh -v
          --build-type=${{ matrix.config.build }}
-          --components='core,python,bin,axcore,axbin,axtest'
+          --components='core,python,axcore,axbin,axtest'
          --cargs=\"
          ${{ matrix.config.cmake }}
          -DOPENVDB_AX_TEST_CMD_DOWNLOADS=ON
          -DUSE_EXPLICIT_INSTANTIATION=OFF
-          -DOPENVDB_CXX_STRICT=ON
          \"
      - name: test
        run: cd build && ctest -V
@@ -122,9 +123,13 @@ jobs:
    strategy:
      matrix:
        config:
-          # Last macos runner before M1 (macos-14)
-          - { runner: 'macos-13', cxx: 'clang++', build: 'Release', llvm: '15' }
-          - { runner: 'macos-13', cxx: 'clang++', build: 'Release', llvm: '13' }
+          - { runner: 'macos-15', cxx: 'clang++', build: 'Release', llvm: '21' }
+          - { runner: 'macos-14', cxx: 'clang++', build: 'Release', llvm: '21' }
+          - { runner: 'macos-14', cxx: 'clang++', build: 'Release', llvm: '20' }
+          - { runner: 'macos-14', cxx: 'clang++', build: 'Release', llvm: '19' }
+          - { runner: 'macos-14', cxx: 'clang++', build: 'Release', llvm: '18' }
+          - { runner: 'macos-14', cxx: 'clang++', build: 'Release', llvm: '17' }
+          - { runner: 'macos-14', cxx: 'clang++', build: 'Release', llvm: '16' }
      fail-fast: false
    steps:
      - uses: actions/checkout@v3
@@ -136,7 +141,7 @@ jobs:
        run: >
          ./ci/build.sh -v
          --build-type=${{ matrix.config.build }}
-          --components="core,python,bin,axcore,axbin,axtest"
+          --components="core,python,axcore,axbin,axtest"
          --cargs=\"
          -DOPENVDB_AX_TEST_CMD_DOWNLOADS=ON
          -DUSE_EXPLICIT_INSTANTIATION=OFF
@@ -154,7 +159,7 @@ jobs:
      github.event.inputs.type == 'grammar'
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'ubuntu-22.04-8c-32g-300h') || 'ubuntu-latest' }}
    container:
-      image: aswf/ci-openvdb:2023-clang15
+      image: aswf/ci-openvdb:2025-clang19.1
    steps:
    - uses: actions/checkout@v3
    - name: build
@@ -19,6 +19,8 @@ on:
      - 'pendingchanges/**'
      - '**.md'
  pull_request:
+    # Add 'ready_for_review' to run CI when a draft PR is marked ready.
+    types: [opened, synchronize, reopened, ready_for_review]
    branches:
      - '**'
    paths-ignore:
@@ -32,12 +34,6 @@ on:
      - 'fvdb/**'
      - 'pendingchanges/**'
      - '**.md'
-  schedule:
-    # run this workflow every day at 7am UTC except Monday
-    - cron:  '0 7 * * 0,2-6'
-    # run this workflow Monday 7am UTC
-    # warning: This pattern is checked in various places below
-    - cron:  '0 7 * * 1'
  workflow_dispatch:
    inputs:
      type:
@@ -59,15 +55,13 @@ jobs:
    # VFX platform jobs. These are run on the appropriate CentOS images from
    # the provided ASWF docker containers
    if: |
-      github.event_name != 'workflow_dispatch' ||
-      github.event.inputs.type == 'all' ||
-      github.event.inputs.type == 'linux'
+      (github.event_name == 'push' ||
+       (github.event_name == 'pull_request' && github.event.pull_request.draft == false) ||
+       (github.event_name == 'workflow_dispatch' &&
+        (github.event.inputs.type == 'all' || github.event.inputs.type == 'linux')))
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'ubuntu-22.04-8c-32g-300h') || 'ubuntu-latest' }}
    name: >
-      linux-vfx:${{ matrix.config.image }}-
-      abi:${{ matrix.config.abi }}-
-      cxx:${{ matrix.config.cxx }}-
-      type:${{ matrix.config.build }}
+      linux:${{ matrix.config.image }}-abi:${{ matrix.config.abi }}-cxx:${{ matrix.config.cxx }}-type:${{ matrix.config.build }}
    container:
      image: aswf/ci-openvdb:${{ matrix.config.image }}
    env:
@@ -76,21 +70,22 @@ jobs:
    strategy:
      matrix:
        config:
-          - { cxx: clang++, image: '2024', abi: '12', build: 'Release', cmake: '' }
-          - { cxx: g++,     image: '2024', abi: '12', build: 'Release', cmake: '' }
-          - { cxx: clang++, image: '2024', abi: '12', build: 'Debug',   cmake: '' }
-          - { cxx: clang++, image: '2023', abi: '11', build: 'Release', cmake: '-DDISABLE_DEPENDENCY_VERSION_CHECKS=ON' }
-          - { cxx: g++,     image: '2023', abi: '11', build: 'Release', cmake: '-DDISABLE_DEPENDENCY_VERSION_CHECKS=ON' }
+          - { cxx: clang++, image: '2026',           abi: '13', build: 'Release', cmake: '-DOPENVDB_USE_FUTURE_ABI_13=ON -DOPENVDB_ENABLE_ASSERTS=ON' }
+          - { cxx: g++,     image: '2026',           abi: '13', build: 'Release', cmake: '-DOPENVDB_USE_FUTURE_ABI_13=ON' }
+          - { cxx: clang++, image: '2026',           abi: '13', build: 'Debug',   cmake: '-DOPENVDB_USE_FUTURE_ABI_13=ON' }
+          - { cxx: clang++, image: '2025-clang19.1', abi: '12', build: 'Release', cmake: '' }
+          - { cxx: clang++, image: '2024-clang17.2', abi: '11', build: 'Release', cmake: '-DDISABLE_DEPENDENCY_VERSION_CHECKS=ON' }
      fail-fast: false
    steps:
    - uses: actions/checkout@v3
    - name: nanobind
-      run: ./ci/install_nanobind.sh 2.0.0
-    - name: glfw
-      if: contains(matrix.config.image, '2023') == true
-      run: ./ci/install_glfw.sh 3.3.10
-    - name: install_gtest
-      run: ./ci/install_gtest.sh 1.15.2
+      run: ./ci/install_nanobind.sh 2.5.0
+      # 2026 images are broken with gtest/glfw installs
+    - name: deps
+      if: contains(matrix.config.image, '2026')
+      run: |
+        ./ci/install_gtest.sh latest
+        ./ci/install_glfw.sh latest
    - name: timestamp
      id: timestamp
      run: echo "timestamp=$(date -u +'%Y-%m-%dT%H:%M:%SZ')" >> $GITHUB_OUTPUT
@@ -114,9 +109,8 @@ jobs:
        -DOPENVDB_ABI_VERSION_NUMBER=${{ matrix.config.abi }}
        \"
    - name: test
-      # Always run tests on weekly builds but skip Debug on commits as they take a while.
-      # https://github.community/t/distinct-job-for-each-schedule/17811/2
-      if: contains(github.event.schedule, '0 7 * * 1') || matrix.config.build == 'Release'
+      # Skip Debug on commits as they take a while.
+      if: github.event_name == 'workflow_dispatch' || matrix.config.build == 'Release'
      run: |
        cd build && ctest -V
        cd - && ./ci/test_install.sh
@@ -128,9 +122,10 @@ jobs:
  windows:
    # Windows CI. Tests a dynamic build with MD.
    if: |
-      github.event_name != 'workflow_dispatch' ||
-      github.event.inputs.type == 'all' ||
-      github.event.inputs.type == 'win'
+      (github.event_name == 'push' ||
+       (github.event_name == 'pull_request' && github.event.pull_request.draft == false) ||
+       (github.event_name == 'workflow_dispatch' &&
+        (github.event.inputs.type == 'all' || github.event.inputs.type == 'win')))
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'windows-2022-8c-32g-300h') || 'windows-latest' }}
    name: windows
    env:
@@ -148,6 +143,9 @@ jobs:
    - name: install
      shell: powershell
      run: .\ci\install_windows.ps1
+    - name: install_numpy
+      shell: powershell
+      run: .\ci\install_windows_numpy.ps1
    - name: build
      run: >
        ./ci/build.sh -v
@@ -169,12 +167,19 @@ jobs:

  macos:
    if: |
-      github.event_name != 'workflow_dispatch' ||
-      github.event.inputs.type == 'all' ||
-      github.event.inputs.type == 'mac'
-    runs-on: macos-13 # Last macos runner befor M1 (macos-14)
+      (github.event_name == 'push' ||
+       (github.event_name == 'pull_request' && github.event.pull_request.draft == false) ||
+       (github.event_name == 'workflow_dispatch' &&
+        (github.event.inputs.type == 'all' || github.event.inputs.type == 'mac')))
+    runs-on: ${{ matrix.config.image }}
+    name: ${{ matrix.config.image }}
    env:
      CXX: clang++
+    strategy:
+      matrix:
+        config:
+          - { image: macos-14 }
+          - { image: macos-15 }
    steps:
    - uses: actions/checkout@v3
    - name: install
@@ -191,3 +196,4 @@ jobs:
        \'
    - name: test
      run: cd build && ctest -V
+
@@ -11,7 +11,6 @@ on:
      - 'CHANGES'
      - 'openvdb_maya/**'
      - 'pendingchanges/**'
-      - '**.md'
  pull_request:
    branches:
      - '**'
@@ -19,7 +18,6 @@ on:
      - 'CHANGES'
      - 'openvdb_maya/**'
      - 'pendingchanges/**'
-      - '**.md'
  workflow_dispatch:
    inputs:
      deploy:
@@ -33,27 +31,41 @@ concurrency:
  cancel-in-progress: true

 jobs:
-  doxygen:
+  validate_docs:
+    # Cheap static checks that run before the heavier doxygen build:
+    #  * every #include <...> referenced from a fenced code block in doc/ resolves
+    #  * every github.com/.../blob/<branch>/<path> URL points to a file on disk
+    # Catches the class of docs-rot from issue #2161 (e.g. nanovdb::util ->
+    # nanovdb::tools rename) before the doxygen job runs.
    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v4
+    - name: set_up_python
+      uses: actions/setup-python@v5
+      with:
+        python-version: '3.x'
+    - name: run_validator
+      run: python3 ci/validate_docs_references.py
+
+  doxygen:
+    needs: validate_docs
+    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'ubuntu-22.04-8c-32g-300h') || 'ubuntu-latest' }}
    env:
      CXX: g++
    container:
-      # @note We can't update this as epydoc doesn't support python3. We'll
-      # need to re-write the python docs to use sphinx
-      image: aswf/ci-openvdb:2024
+      image: aswf/ci-openvdb:2026
    steps:
    - uses: actions/checkout@v4
    - name: install_doxygen
-      run: ./ci/install_doxygen.sh 1_8_11
+      run: ./ci/install_doxygen.sh 1_14_0
    - name: nanobind
-      #if: contains(matrix.config.image, '2023') == false
-      run: ./ci/install_nanobind.sh 2.0.0
-    # - name: install_epydoc
-    #   run: pip install epydoc
-    - name: install_latex
-      run: |
-        yum -y install texlive-latex-bin texlive-dvips texlive-collection-fontsrecommended texlive-collection-latexrecommended
+      run: ./ci/install_nanobind.sh 2.5.0
    - name: build
+      # Doxygen's native WARN_AS_ERROR=FAIL_ON_WARNINGS gates the docs build:
+      # any warning fails the job. FAIL_ON_WARNINGS still generates the docs and
+      # writes the full warning log first, so the artifact step below captures
+      # it for debugging. Enabled via OPENVDB_DOXYGEN_WARN_AS_ERROR (see
+      # doc/CMakeLists.txt).
      run: >
        ./ci/build.sh -v
        --build-type=Release
@@ -64,10 +76,18 @@ jobs:
        -DOPENVDB_PYTHON_WRAP_ALL_GRID_TYPES=ON
        -DUSE_EXPLICIT_INSTANTIATION=OFF
        -DDISABLE_DEPENDENCY_VERSION_CHECKS=ON
+        -DOPENVDB_DOXYGEN_WARN_AS_ERROR=ON
        \"
-    # - name: epydoc
-    #   run: |
-    #     epydoc --html -o /usr/local/share/doc/OpenVDB/html/python pyopenvdb
+    - name: upload_doxygen_warning_log
+      # Always upload so contributors can debug failed PR builds without
+      # having to reproduce the doxygen environment locally.
+      if: always()
+      uses: actions/upload-artifact@v4
+      with:
+        name: doxygen-warnings
+        path: build/doc/doxygen.warnings.log
+        if-no-files-found: ignore
+        retention-days: 14
    - name: pre_deploy
      # Overwrite global SSH configuration
      #  https://github.com/peaceiris/actions-gh-pages/issues/719
@@ -106,7 +126,7 @@ jobs:
      github.event.inputs.deploy == 'coverage'
    runs-on: ubuntu-latest
    container:
-      image: aswf/ci-openvdb:2023
+      image: aswf/ci-openvdb:2025-clang19.1
    env:
      CXX: g++
    steps:
@@ -156,4 +176,3 @@ jobs:
          - Deployed from: AcademySoftwareFoundation/openvdb ${{ github.sha }}

          Signed-off-by: ${{ github.actor }} <${{ github.actor }}@users.noreply.github.com>"
-
@@ -28,9 +28,6 @@ on:
      - 'fvdb/**'
      - 'pendingchanges/**'
      - '**.md'
-  schedule:
-    # run this workflow every day 7am UTC
-    - cron:  '0 7 * * *'
  workflow_dispatch:

 # Allow subsequent pushes to the same PR or REF to cancel any previous jobs.
@@ -66,7 +63,7 @@ jobs:
      ${{ needs.checksecret.outputs.HOUDINI_SECRETS == 'true' ||
          github.repository_owner == 'AcademySoftwareFoundation' }}
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'ubuntu-22.04-8c-32g-300h') || 'ubuntu-latest' }}
-    name: hou:${{ matrix.config.hou_hash }}-vfx:${{ matrix.config.image }}-cxx:${{ matrix.config.cxx }}
+    name: hou:${{ matrix.config.hou_hash }}-vfx:${{ matrix.config.image }}-cxx:${{ matrix.config.cxx }}-build:${{ matrix.config.build }}
    container:
      image: aswf/ci-base:${{ matrix.config.image }}
    env:
@@ -75,26 +72,18 @@ jobs:
    strategy:
      matrix:
        config:
-          - { cxx: clang++, image: '2024', hou_hash: '20_5', build: 'Release', components: 'core,hou,bin,view,render,python,test' }
-          - { cxx: clang++, image: '2023', hou_hash: '20_5', build: 'Debug',   components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
-          - { cxx: g++,     image: '2023', hou_hash: '20_5', build: 'Release', components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
-          - { cxx: clang++, image: '2023', hou_hash: '20_0-newabi', build: 'Release', components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
-          - { cxx: g++,     image: '2023', hou_hash: '20_0-newabi', build: 'Release', components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
+          - { cxx: clang++, image: '2026', hou_hash: '21_0', build: 'Release', components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
+          - { cxx: clang++, image: '2026', hou_hash: '21_0', build: 'Debug',   components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
+          - { cxx: g++,     image: '2026', hou_hash: '21_0', build: 'Release', components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
+          - { cxx: clang++, image: '2024', hou_hash: '20_5', build: 'Release', components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
+          - { cxx: g++,     image: '2024', hou_hash: '20_5', build: 'Release', components: 'core,hou,bin,view,render,python,test,axcore,axbin,axtest' }
      fail-fast: false
    steps:
-    - name: remove zstd
-      run: yum -y remove zstd
    - uses: actions/checkout@v3
    - name: nanobind
-      #if: contains(matrix.config.image, '2023') == false
-      run: ./ci/install_nanobind.sh 2.0.0
-    - name: glfw
-      if: contains(matrix.config.image, '2023') == true
-      run: ./ci/install_glfw.sh 3.3.10
-    - name: cppunit
-      run: ./ci/install_cppunit.sh 1.15.1
+      run: ./ci/install_nanobind.sh 2.5.0
    - name: install_gtest
-      run: ./ci/install_gtest.sh 1.15.2
+      run: ./ci/install_gtest.sh latest
    - name: timestamp
      id: timestamp
      run: echo "timestamp=$(date -u +'%Y-%m-%dT%H:%M:%SZ')" >> $GITHUB_OUTPUT
@@ -125,7 +114,7 @@ jobs:
    - name: build
      run: |
        ./ci/build.sh -v \
-          --build-type=Release \
+          --build-type="${{ matrix.config.build }}" \
          --components="${{ matrix.config.components }}" \
          --cargs=\" \
            -DHOUDINI_ROOT=$HOME/houdini_install/hou \
@@ -135,6 +124,7 @@ jobs:
            -DDISABLE_DEPENDENCY_VERSION_CHECKS=ON \
            \"
    - name: test
+      if: matrix.config.build == 'Release'
      run: cd build && ctest -V
    # Keep ccache light by stripping out any caches not accessed in the last day
    - name: ccache_clean
@@ -165,12 +155,12 @@ jobs:
        cp hou/hou.tar.gz $HOME/houdini_install/hou.tar.gz
        cd $HOME/houdini_install && tar -xzf hou.tar.gz && cd -
    - name: install_deps
-      run: ./ci/install_macos.sh 15
+      run: ./ci/install_macos.sh 20
    - name: build
      run: |
        ./ci/build.sh -v \
          --build-type=Release \
-          --components="core,hou,bin,view,render,python,test,axcore,axbin" \
+          --components="core,hou,bin,view,render,test,axcore,axbin" \
          --cargs=\" \
            -DHOUDINI_ROOT=$HOME/houdini_install/hou \
            -DOPENVDB_BUILD_HOUDINI_ABITESTS=OFF \
@@ -178,8 +168,7 @@ jobs:
            -DDISABLE_CMAKE_SEARCH_PATHS=ON \
            -DDISABLE_DEPENDENCY_VERSION_CHECKS=ON \
            -DUSE_EXPLICIT_INSTANTIATION=OFF \
-            -DTbb_INCLUDE_DIR=$HOME/houdini_install/hou/Frameworks/Houdini.framework/Versions/Current/Resources/toolkit/include/tbb \
-            -DLLVM_DIR=/opt/homebrew/opt/llvm@15/lib/cmake/llvm \
+            -DLLVM_DIR=/opt/homebrew/opt/llvm@20/lib/cmake/llvm \
            -DCMAKE_INSTALL_PREFIX=${{ github.workspace }}/install \
            \"
    - name: test
@@ -17,7 +17,9 @@ on:
      - 'fvdb/**'
      - 'pendingchanges/**'
      - '**.md'
-  pull_request:
+  pull_request_target:
+    # Add 'ready_for_review' to run CI when a draft PR is marked ready.
+    types: [opened, synchronize, reopened, ready_for_review]
    branches:
      - '**'
    paths-ignore:
@@ -30,9 +32,6 @@ on:
      - 'fvdb/**'
      - 'pendingchanges/**'
      - '**.md'
-  schedule:
-    # run this workflow every day 7am UTC
-    - cron:  '0 7 * * *'
  workflow_dispatch:
    inputs:
      type:
@@ -47,56 +46,141 @@ concurrency:

 jobs:
  linux-nanovdb:
+    # push OR non-draft PR OR matching manual run
    if: |
-      github.event_name != 'workflow_dispatch' ||
-      github.event.inputs.type == 'all' ||
-      github.event.inputs.type == 'linux'
+      (github.event_name == 'push' ||
+       (github.event_name == 'pull_request_target' && github.event.pull_request.draft == false) ||
+       (github.event_name == 'workflow_dispatch' &&
+        (github.event.inputs.type == 'all' || github.event.inputs.type == 'linux')))
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'ubuntu-22.04-8c-32g-300h') || 'ubuntu-latest' }}
    name: >
      linux-nanovdb:cxx:${{ matrix.config.cxx }}-${{ matrix.config.build }}
+    # Build in the same image used by the AWS CodeBuild GPU runner so the
+    # uploaded build tree is binary/runtime compatible when executed there.
    container:
-      image: aswf/ci-openvdb:${{ matrix.config.image }}
+      image: aswf/ci-openvdb:2024-clang17.2
    env:
      CXX: ${{ matrix.config.cxx }}
    strategy:
      matrix:
        config:
-          - { cxx: g++,     image: '2024', build: 'Release' }
-          - { cxx: g++,     image: '2024', build: 'Debug' }
-          - { cxx: clang++, image: '2024', build: 'Release' }
-          - { cxx: clang++, image: '2024', build: 'Debug' }
+          - { cxx: g++,     label: gcc,   build: 'Release' }
+          - { cxx: g++,     label: gcc,   build: 'Debug' }
+          - { cxx: clang++, label: clang, build: 'Release' }
+          - { cxx: clang++, label: clang, build: 'Debug' }
      fail-fast: false
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v7
+        with:
+          ref: ${{ github.event_name == 'pull_request_target' && github.event.pull_request.head.sha || github.sha }}
      - name: setup_cuda_12
        run: |
-          yum -y install yum-utils
          echo "/usr/local/cuda-12/bin" >> $GITHUB_PATH
          echo "LD_LIBRARY_PATH=/usr/local/cuda-12/lib64:$LD_LIBRARY_PATH" >> $GITHUB_ENV
+      - name: nanobind
+        run: ./ci/install_nanobind.sh 2.5.0
      - name: build
      # NOTE: CMAKE_POSITION_INDEPENDENT_CODE set to fix default behaviour change in clang 14
      #   https://github.com/AcademySoftwareFoundation/aswf-docker/issues/228
+      # NOTE: CUDA architecture 86 targets the AWS CodeBuild A10G GPUs used by
+      #   the linux-nanovdb-gpu test stage below.
        run: >
          ./ci/build.sh -v
          --build-type=${{ matrix.config.build }}
-          --components=core,nano,nanotest,nanoexam,nanobench,nanotool
+          --components=core,nano,nanotest,nanoexam,nanobench,nanotool,nanopython,nanopytest
          --cargs=\'
          -DUSE_EXPLICIT_INSTANTIATION=OFF
+          -DOPENVDB_CORE_STATIC=OFF
          -DNANOVDB_USE_CUDA=ON
-          -DCMAKE_CUDA_ARCHITECTURES="80"
+          -DCMAKE_CUDA_ARCHITECTURES="86"
          -DNANOVDB_USE_OPENVDB=ON
          -DCMAKE_INSTALL_PREFIX=`pwd`
-          -DUSE_BLOSC=OFF
+          -DUSE_BLOSC=ON
          -DCMAKE_POSITION_INDEPENDENT_CODE=ON
          \'
+      # Run the non-GPU tests here. GPU-only (cuda/mgpu) tests are executed in
+      # the linux-nanovdb-gpu job. This step also generates the shared test
+      # data files (e.g. data/*.nvdb) that the GPU tests consume, so it must
+      # run before the build tree is uploaded.
      - name: test
-        run: cd build && sudo ctest -V -E ".*cuda.*"
+        run: cd build && ctest -V -E ".*cuda.*|.*mgpu.*"
+      - name: package_gpu_build
+        # Record the absolute workspace path so the GPU runner can restore the
+        # build tree to the same location, keeping RPATHs and CTest-generated
+        # absolute paths valid.
+        run: |
+          META="$GITHUB_WORKSPACE/nanovdb-gpu-meta.env"
+          echo "BUILD_WORKSPACE=$GITHUB_WORKSPACE" > "$META"
+          echo "GITHUB_SHA=$GITHUB_SHA" >> "$META"
+          echo "CONFIG=${{ matrix.config.label }}-${{ matrix.config.build }}" >> "$META"
+          cat "$META"
+      - name: upload_gpu_build
+        uses: actions/upload-artifact@v7
+        with:
+          name: nanovdb-gpu-build-${{ matrix.config.label }}-${{ matrix.config.build }}
+          path: |
+            build
+            nanovdb/nanovdb/python/test/TestNanoVDB.py
+            nanovdb-gpu-meta.env
+          retention-days: 1
+          if-no-files-found: error
+
+  linux-nanovdb-gpu:
+    # Execute the CUDA / multi-GPU tests on AWS CodeBuild (4x A10G) using the
+    # prebuilt artifacts from linux-nanovdb. Repository secrets are not shared
+    # with fork PRs, so restrict this job to same-repo events.
+    if: |
+      github.repository == 'AcademySoftwareFoundation/openvdb' &&
+      (github.event_name == 'push' ||
+       (github.event_name == 'pull_request_target' &&
+        github.event.pull_request.draft == false &&
+        github.event.pull_request.head.repo.full_name == github.repository) ||
+       (github.event_name == 'workflow_dispatch' &&
+        (github.event.inputs.type == 'all' || github.event.inputs.type == 'linux')))
+    needs: linux-nanovdb
+    runs-on: ubuntu-latest
+    name: >
+      linux-nanovdb-gpu:${{ matrix.config.label }}-${{ matrix.config.build }}
+    # GITHUB_TOKEN needs actions:read to fetch the workflow-run artifacts.
+    permissions:
+      contents: read
+      actions: read
+    strategy:
+      matrix:
+        config:
+          - { label: gcc,   build: 'Release' }
+          - { label: gcc,   build: 'Debug' }
+          - { label: clang, build: 'Release' }
+          - { label: clang, build: 'Debug' }
+      fail-fast: false
+    env:
+      # Consumed by buildspec.yml on the CodeBuild side.
+      ARTIFACT_NAME: nanovdb-gpu-build-${{ matrix.config.label }}-${{ matrix.config.build }}
+      GH_TOKEN: ${{ github.token }}
+    steps:
+      - name: Configure AWS Credentials
+        uses: aws-actions/configure-aws-credentials@v6
+        with:
+          aws-access-key-id: ${{ secrets.CODEBUILD_ID }}
+          aws-secret-access-key: ${{ secrets.CODEBUILD_SECRET }}
+          aws-region: us-west-2
+      - name: Run CodeBuild GPU tests
+        uses: aws-actions/aws-codebuild-run-build@v1
+        with:
+          project-name: OpenVDB
+          buildspec-override: ci/buildspec_nanovdb.yml
+          image-override: aswf/ci-openvdb:2024-clang17.2
+          # GITHUB_* variables (GITHUB_RUN_ID, GITHUB_REPOSITORY, ...) are
+          # forwarded automatically. The job-scoped Actions runtime values let
+          # CodeBuild upload report.tgz directly to this workflow run.
+          env-vars-for-codebuild: ARTIFACT_NAME,GH_TOKEN,ACTIONS_RUNTIME_TOKEN,ACTIONS_RESULTS_URL

  windows-nanovdb:
    if: |
-      github.event_name != 'workflow_dispatch' ||
-      github.event.inputs.type == 'all' ||
-      github.event.inputs.type == 'win'
+      (github.event_name == 'push' ||
+       (github.event_name == 'pull_request_target' && github.event.pull_request.draft == false) ||
+       (github.event_name == 'workflow_dispatch' &&
+        (github.event.inputs.type == 'all' || github.event.inputs.type == 'win')))
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'windows-2022-8c-32g-300h') || 'windows-latest' }}
    env:
      VCPKG_DEFAULT_TRIPLET: 'x64-windows'
@@ -105,7 +189,9 @@ jobs:
    strategy:
      fail-fast: false
    steps:
-    - uses: actions/checkout@v3
+    - uses: actions/checkout@v7
+      with:
+        ref: ${{ github.event_name == 'pull_request_target' && github.event.pull_request.head.sha || github.sha }}
    - name: path
      run: |
        # note: system path must be modified in a previous step to it's use
@@ -118,41 +204,46 @@ jobs:
      shell: powershell
      run: .\ci\install_windows.ps1
    - name: build
+    # nvcc doesn't set _WIN32 when run in bash so we need to set it manually
      shell: bash
      run: >
        ./ci/build.sh -v
        --config=Release
-        --components=core,nano,nanotest,nanoexam,nanobench,nanotool
+        --components=core,nano,nanotest,nanoexam,nanobench,nanotool,nanopython,nanopytest
        --cargs=\'
        -A x64 -G \"Visual Studio 17 2022\" -DOPENVDB_CORE_STATIC=OFF
        -DMSVC_COMPRESS_PDB=ON
        -DUSE_EXPLICIT_INSTANTIATION=OFF
        -DNANOVDB_USE_CUDA=ON
        -DCMAKE_CUDA_ARCHITECTURES="80"
+        -DCMAKE_CUDA_FLAGS="-D_WIN32"
        -DNANOVDB_USE_OPENVDB=ON
        -DVCPKG_TARGET_TRIPLET=${VCPKG_DEFAULT_TRIPLET}
        -DCMAKE_TOOLCHAIN_FILE=\"${VCPKG_INSTALLATION_ROOT}\\scripts\\buildsystems\\vcpkg.cmake\"
        \'
    - name: test
      shell: bash
-      run: cd build && ctest -V -E ".*cuda.*"
+      run: cd build && ctest -V -C Release -E ".*cuda.*|.*mgpu.*"

  macos-nanovdb:
    if: |
-      github.event_name != 'workflow_dispatch' ||
-      github.event.inputs.type == 'all' ||
-      github.event.inputs.type == 'mac'
+      (github.event_name == 'push' ||
+       (github.event_name == 'pull_request_target' && github.event.pull_request.draft == false) ||
+       (github.event_name == 'workflow_dispatch' &&
+        (github.event.inputs.type == 'all' || github.event.inputs.type == 'mac')))
    runs-on: ${{ matrix.config.runner }}
    env:
      CXX: ${{ matrix.config.cxx }}
    strategy:
      matrix:
        config:
-          - { runner: 'macos-13', cxx: 'clang++', build: 'Release' }
-          - { runner: 'macos-13', cxx: 'clang++', build: 'Debug'   }
+          - { runner: 'macos-14', cxx: 'clang++', build: 'Release' }
+          - { runner: 'macos-14', cxx: 'clang++', build: 'Debug'   }
      fail-fast: false
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v7
+        with:
+          ref: ${{ github.event_name == 'pull_request_target' && github.event.pull_request.head.sha || github.sha }}
      - name: install_deps
        run: |
          ./ci/install_macos.sh
@@ -161,28 +252,31 @@ jobs:
        run: >
          ./ci/build.sh -v
          --build-type=${{ matrix.config.build }}
-          --components=core,nano,nanotest,nanoexam,nanobench,nanotool
-          --cargs=\'-DUSE_EXPLICIT_INSTANTIATION=OFF -DNANOVDB_USE_CUDA=OFF -DNANOVDB_USE_OPENVDB=ON\'
+          --components=core,nano,nanotest,nanoexam,nanobench,nanotool,nanopython,nanopytest
+          --cargs=\'-DUSE_EXPLICIT_INSTANTIATION=OFF -DNANOVDB_USE_CUDA=OFF -DNANOVDB_USE_OPENVDB=ON -DCMAKE_INSTALL_PREFIX=${{ github.workspace }}/install \'
      - name: test
-        run: cd build && ctest -V -E ".*cuda.*"
+        run: cd build && ctest -V -E ".*cuda.*|.*mgpu.*"

  nanovdb-lite:
    if: |
-      github.event_name != 'workflow_dispatch' ||
-      github.event.inputs.type == 'all' ||
-      github.event.inputs.type == 'linux'
+      (github.event_name == 'push' ||
+       (github.event_name == 'pull_request_target' && github.event.pull_request.draft == false) ||
+       (github.event_name == 'workflow_dispatch' &&
+        (github.event.inputs.type == 'all' || github.event.inputs.type == 'linux')))
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'ubuntu-22.04-8c-32g-300h') || 'ubuntu-latest' }}
    container:
-      image: aswf/ci-openvdb:2024
+      image: aswf/ci-openvdb:2024-clang17.2
    steps:
-      - uses: actions/checkout@v3
+      - uses: actions/checkout@v7
+        with:
+          ref: ${{ github.event_name == 'pull_request_target' && github.event.pull_request.head.sha || github.sha }}
      - name: install_gtest
-        run: ./ci/install_gtest.sh 1.15.2
+        run: ./ci/install_gtest.sh latest
      - name: build_and_test
        run: |
          cd nanovdb/nanovdb
-          sudo mkdir .build
+          mkdir .build
          cd .build
-          sudo cmake -DUSE_EXPLICIT_INSTANTIATION=OFF -DNANOVDB_BUILD_UNITTESTS=ON -DNANOVDB_USE_OPENVDB=OFF -DNANOVDB_USE_CUDA=OFF ../
-          sudo make -j8 install
-          sudo ctest -V
+          cmake -DUSE_EXPLICIT_INSTANTIATION=OFF -DNANOVDB_BUILD_UNITTESTS=ON -DNANOVDB_USE_OPENVDB=OFF -DNANOVDB_USE_CUDA=OFF -DCMAKE_INSTALL_PREFIX="$(pwd)/install" ../
+          make -j8 install
+          ctest -V
@@ -70,23 +70,13 @@ jobs:
    strategy:
      matrix:
        config:
-          - { houdini_version: '20.0', platform: 'linux_x86_64_gcc11.2', hou_hash: '20_0-newabi' }
+          - { houdini_version: '21.0', platform: 'linux_x86_64_gcc11.2', hou_hash: '21_0' }
          - { houdini_version: '20.5', platform: 'linux_x86_64_gcc11.2', hou_hash: '20_5' }
      fail-fast: false
    container:
      image: aswf/ci-base:2024
    steps:
    - uses: actions/checkout@v3
-    # We bumped from the 2021 CI image to 2023 here to fix some OpenSSL issues
-    # with the Houdini download script. In so doing we broke some of the caching
-    # between this job and the jobs in houdini.yml which _don't_ use the 2023
-    # image yet. The issue is that the cache action will use zstd if it's
-    # available to zip the cache and this causes it to be inserted with a unique
-    # hash which images without zstd (i.e. the 2021/2022 images don't have
-    # access to). For now, uninstall zstd here instead of installing it
-    # everywhere and ask the LF to add zstd to the older base images.
-    - name: remove zstd
-      run: yum -y remove zstd
    - name: timestamp
      id: timestamp
      run: echo "timestamp=$(date -u +'%Y-%m-%dT%H:%M:%SZ')" >> $GITHUB_OUTPUT
@@ -113,7 +103,7 @@ jobs:
    # Note that macos-14 (current macos-latest) switches to M1. We could instead test
    # the arm build here instead of the x86 one.
    runs-on: macos-latest
-    name: macos-houdini-20
+    name: macos-houdini
    env:
      HOUDINI_CLIENT_ID: ${{ secrets.HOUDINI_CLIENT_ID }}
      HOUDINI_SECRET_KEY: ${{ secrets.HOUDINI_SECRET_KEY }}
@@ -123,7 +113,7 @@ jobs:
      id: timestamp
      run: echo "timestamp=$(date -u +'%Y-%m-%dT%H:%M:%SZ')" >> $GITHUB_OUTPUT
    - name: download_houdini
-      run: ./ci/download_houdini.sh 20.0 macosx_arm64_clang14.0_13 --prod
+      run: ./ci/download_houdini.sh 21.0 macosx_arm64_clang15.0_14 --prod
    - name: install_houdini
      run: |
        mkdir $HOME/houdini_install
@@ -149,28 +139,34 @@ jobs:
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'ubuntu-22.04-8c-32g-300h') || 'ubuntu-latest' }}
    name: linux-extra:${{ matrix.config.name }}
    container:
-      image: aswf/ci-openvdb:2024
+      image: aswf/ci-openvdb:${{ matrix.config.image }}
    env:
      CXX: clang++
    strategy:
      matrix:
        config:
-          - { name: 'all',   build: 'Release', components: 'core,python,bin,view,render,test',               cmake: '-DUSE_BLOSC=ON  -DUSE_ZLIB=ON  -DUSE_EXR=ON  -DUSE_PNG=ON'  }
-          - { name: 'lite',  build: 'Release', components: 'core,python,bin,view,render,test',               cmake: '-DUSE_BLOSC=OFF -DUSE_ZLIB=OFF -DUSE_EXR=OFF -DUSE_PNG=OFF -DOPENVDB_USE_DELAYED_LOADING=OFF' }
-          - { name: 'half',  build: 'Release', components: 'core,python,bin,view,render,test',               cmake: '-DUSE_BLOSC=OFF -DUSE_IMATH_HALF=ON' }
-          - { name: 'sse',   build: 'Release', components: 'core,python,bin,view,render,test',               cmake: '-DOPENVDB_SIMD=SSE42' }
-          - { name: 'avx',   build: 'Release', components: 'core,python,bin,view,render,test',               cmake: '-DOPENVDB_SIMD=AVX' }
-          - { name: 'numpy', build: 'Release', components: 'core,python,bin,view,render,test',               cmake: '-DUSE_NUMPY=ON -DOPENVDB_PYTHON_WRAP_ALL_GRID_TYPES=ON' }
-          - { name: 'asan',  build: 'asan',    components: 'core,test',                                      cmake: '-DNANOVDB_USE_OPENVDB=ON -DOPENVDB_AX_STATIC=OFF -DOPENVDB_CORE_STATIC=OFF -DUSE_BLOSC=OFF' } # We never called blosc_destroy(), so disable blosc to silence these errors
-          - { name: 'ubsan', build: 'ubsan',   components: 'core,test',                                      cmake: '-DCMAKE_CXX_FLAGS="-Wno-deprecated-declarations" ' }
-          - { name: 'c++20', build: 'Release', components: 'core,test',                                      cmake: '-DCMAKE_CXX_STANDARD=20' }
-          - { name: 'conf',  build: 'Release', components: 'core,python,bin,view,render,test',               cmake: '-DCMAKE_FIND_PACKAGE_PREFER_CONFIG=ON' }
+          - { name: 'all',    image: '2026', build: 'Release', components: 'core,python,bin,view,render,test', cmake: '-DUSE_BLOSC=ON  -DUSE_ZLIB=ON  -DUSE_EXR=ON  -DUSE_PNG=ON'  }
+          - { name: 'lite',   image: '2026', build: 'Release', components: 'core,python,bin,view,render,test', cmake: '-DUSE_BLOSC=OFF -DUSE_ZLIB=OFF -DUSE_EXR=OFF -DUSE_PNG=OFF -DOPENVDB_USE_DELAYED_LOADING=OFF' }
+          - { name: 'half',   image: '2026', build: 'Release', components: 'core,python,bin,view,render,test', cmake: '-DUSE_BLOSC=OFF -DUSE_IMATH_HALF=ON' }
+          - { name: 'simd',   image: '2026', build: 'Release', components: 'core,python,bin,view,render,test', cmake: '-DOPENVDB_X86_INSTRSET=7 -DUSE_VCL=ON' } # 7=AVX
+          - { name: 'pygrid', image: '2026', build: 'Release', components: 'core,python,bin,view,render,test', cmake: '-DOPENVDB_PYTHON_WRAP_ALL_GRID_TYPES=ON' }
+          - { name: 'asan',   image: '2026', build: 'asan',    components: 'core,test',                        cmake: '-DDISABLE_DEPENDENCY_VERSION_CHECKS=ON -DNANOVDB_USE_OPENVDB=ON -DOPENVDB_AX_STATIC=OFF -DOPENVDB_CORE_STATIC=OFF -DUSE_BLOSC=OFF' } # We never called blosc_destroy(), so disable blosc to silence these errors
+          - { name: 'ubsan',  image: '2026', build: 'ubsan',   components: 'core,test',                        cmake: '-DDISABLE_DEPENDENCY_VERSION_CHECKS=ON -DCMAKE_CXX_FLAGS="-Wno-deprecated-declarations" ' }
+          - { name: 'c++20',  image: '2026', build: 'Release', components: 'core,test',                        cmake: '-DCMAKE_CXX_STANDARD=20' }
+          - { name: 'conf',   image: '2026', build: 'Release', components: 'core,python,bin,view,render,test', cmake: '-DCMAKE_FIND_PACKAGE_PREFER_CONFIG=ON' }
      fail-fast: false
    steps:
    - uses: actions/checkout@v3
    - name: nanobind
-      #if: contains(container.image, '2023') == false
-      run: ./ci/install_nanobind.sh 2.0.0
+      run: ./ci/install_nanobind.sh 2.5.0
+      # 2026 images are broken with gtest/glfw/exr installs
+    - name: deps
+      if: contains(matrix.config.image, '2026')
+      run: |
+        ./ci/install_gtest.sh latest
+        ./ci/install_glfw.sh latest
+        ./ci/install_imath.sh v3.2.2
+        ./ci/install_exr.sh v3.2.2
    - name: build
      run: >
        ./ci/build.sh -v
@@ -192,21 +188,22 @@ jobs:
    strategy:
      matrix:
        config:
-          - { runson: ubuntu-latest, cxx: g++,     cmake: '' }
+          # llvm-17 is currently the default which has symbol issues with the python module
+          - { runson: ubuntu-latest, cxx: g++, cmake: '-DOPENVDB_PYTHON_USE_AX=OFF' }
          # Disable the clang job for now. See https://github.com/actions/runner-images/issues/8659
          # - { runson: ubuntu-latest, cxx: clang++, cmake: '' }
          # @todo gcc on macos
-          - { runson: macos-latest,  cxx: '',      cmake: '-DCMAKE_CXX_COMPILER=/opt/homebrew/opt/llvm@15/bin/clang++ -DLLVM_DIR=/opt/homebrew/opt/llvm@15/lib/cmake/llvm' }
+          - { runson: macos-latest, cxx: '', cmake: '-DLLVM_DIR=/opt/homebrew/opt/llvm@20/lib/cmake/llvm' }
      fail-fast: false
    steps:
      - uses: actions/checkout@v3
      - name: install_deps
        run: |
          if [ "$RUNNER_OS" == "Linux" ]; then
-            sudo apt-get -q install -y libboost-dev libboost-iostreams-dev libtbb-dev libblosc-dev llvm-15-dev libgtest-dev libgmock-dev libcppunit-dev
-            ./ci/install_nanobind.sh 2.0.0
+            sudo apt-get -q install -y libboost-dev libboost-iostreams-dev libtbb-dev libblosc-dev libgtest-dev libgmock-dev python3-numpy
+            ./ci/install_nanobind.sh 2.5.0
          elif [ "$RUNNER_OS" == "macOS" ]; then
-            ./ci/install_macos.sh 15
+            ./ci/install_macos.sh 20
            ./ci/install_tbb_macos.sh
          else
            echo "$RUNNER_OS not supported"; exit 1
@@ -216,7 +213,12 @@ jobs:
          ./ci/build.sh -v
          --build-type=Release
          --components=\"core,axcore,python,bin,render,test,axbin\"
-          --cargs=\"-DCMAKE_CXX_STANDARD=20 -DOPENVDB_USE_DELAYED_LOADING=OFF -DCMAKE_INSTALL_PREFIX=${{ github.workspace }}/install ${{ matrix.config.cmake }}\"
+          --cargs=\'
+          -DCMAKE_CXX_STANDARD=20
+          -DOPENVDB_USE_DELAYED_LOADING=OFF
+          -DCMAKE_INSTALL_PREFIX=${{ github.workspace }}/install
+          ${{ matrix.config.cmake }}
+          \'
      - name: test
        run: cd build && ctest -V

@@ -253,6 +255,10 @@ jobs:
    - name: install
      shell: powershell
      run: .\ci\install_windows.ps1
+    - name: install_numpy
+      if: matrix.config.vc == 'x64-windows'
+      shell: powershell
+      run: .\ci\install_windows_numpy.ps1
    - name: build
      run: >
        ./ci/build.sh -v
@@ -294,15 +300,14 @@ jobs:
    strategy:
      matrix:
        config:
-          # Unified
-          - { image: '2023-clang15', cxx: 'clang++', build: 'Release', components: 'core,bin,axcore,axbin,axtest', cmake: '' }
-          - { image: '2023-clang15', cxx: 'g++',     build: 'Release', components: 'core,bin,axcore,axbin,axtest', cmake: '' }
+          - { image: '2025-clang18.1', cxx: 'clang++', build: 'Release', components: 'core,axcore,axbin,axtest', cmake: '' }
+          # @note  LLVM 17 builds are incompatible with the vdb python plugin on linux
+          - { image: '2024-clang17.2', cxx: 'clang++', build: 'Release', components: 'core,axcore,axbin,axtest', cmake: '-DOPENVDB_PYTHON_USE_AX=OFF' }
      fail-fast: false
    steps:
      - uses: actions/checkout@v3
      - name: nanobind
-        #f: contains(matrix.config.image, '2023') == false
-        run: ./ci/install_nanobind.sh 2.0.0
+        run: ./ci/install_nanobind.sh 2.5.0
      - name: build
        run: >
          ./ci/build.sh -v
@@ -341,14 +346,14 @@ jobs:
      github.event_name != 'workflow_dispatch' ||
      github.event.inputs.type == 'all' ||
      github.event.inputs.type == 'ax'
-    runs-on: macos-13
+    runs-on: ${{ matrix.config.runner }}
    name: macos-cxx:${{ matrix.config.cxx }}-llvm:${{ matrix.config.llvm }}-${{ matrix.config.build }}
    env:
      CXX: ${{ matrix.config.cxx }}
    strategy:
      matrix:
        config:
-          - { cxx: 'clang++', build: 'Release', llvm: '15' }
+          - { runner: 'macos-latest', cxx: 'clang++', build: 'Release', llvm: '20' }
      fail-fast: false
    steps:
      - uses: actions/checkout@v3
@@ -378,7 +383,7 @@ jobs:
      github.event.inputs.type == 'all' ||
      github.event.inputs.type == 'ax'
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'windows-2022-8c-32g-300h') || 'windows-latest' }}
-    name: windows-vc:${{ matrix.config.vc }}-type:${{ matrix.config.build }}
+    name: ${{ matrix.config.vc }}-${{ matrix.config.crt }}-${{ matrix.config.build }}-llvm${{ matrix.config.llvm }}
    env:
      VCPKG_DEFAULT_TRIPLET: ${{ matrix.config.vc }}
      # Export this with '' avoid bash treating \ as escape
@@ -392,17 +397,22 @@ jobs:
          # @todo  We don't currently run the axtests with shared builds of ax
          # due to symbol issues using LLVM as a static lib (which is the only
          # option on Windows).
-          - { vc: 'x64-windows',        crt: 'MD',  components: 'core,bin,axcore,axbin,python', build: 'Release', cmake: '-DOPENVDB_CORE_STATIC=OFF -DOPENVDB_AX_STATIC=OFF' }
-          - { vc: 'x64-windows-static', crt: 'MT',  components: 'core,bin,axcore,axbin,axtest', build: 'Release', cmake: '-DOPENVDB_CORE_SHARED=OFF -DOPENVDB_AX_SHARED=OFF -DUSE_STATIC_DEPENDENCIES=ON -DBLOSC_USE_EXTERNAL_SOURCES=ON' }
-          - { vc: 'x64-windows-static', crt: 'MTd', components: 'core,bin,axcore,axbin,axtest', build: 'Debug',   cmake: '-DOPENVDB_CORE_SHARED=OFF -DOPENVDB_AX_SHARED=OFF -DUSE_STATIC_DEPENDENCIES=ON -DBLOSC_USE_EXTERNAL_SOURCES=ON' }
+          # @note  currently only test static builds with MT as building LLVM as
+          # a shared lib on windows is a bit dicey
+          - { vc: 'x64-windows-static', llvm: '18.1.0', crt: 'MultiThreaded',      components: 'core,axcore,axbin,axtest', build: 'Release', cmake: '-DOPENVDB_CORE_SHARED=OFF -DOPENVDB_AX_SHARED=OFF -DUSE_STATIC_DEPENDENCIES=ON -DBLOSC_USE_EXTERNAL_SOURCES=ON' }
+          - { vc: 'x64-windows-static', llvm: '18.1.0', crt: 'MultiThreadedDebug', components: 'core,axcore,axbin,axtest', build: 'Debug',   cmake: '-DOPENVDB_CORE_SHARED=OFF -DOPENVDB_AX_SHARED=OFF -DUSE_STATIC_DEPENDENCIES=ON -DBLOSC_USE_EXTERNAL_SOURCES=ON' }
+          - { vc: 'x64-windows',        llvm: '18.1.0', crt: 'MultiThreadedDLL',   components: 'core,axcore,axbin',        build: 'Release', cmake: '-DOPENVDB_CORE_STATIC=OFF -DOPENVDB_AX_STATIC=OFF' }
+          - { vc: 'x64-windows-static', llvm: '15.0.0', crt: 'MultiThreaded',      components: 'core,axcore,axbin,axtest', build: 'Release', cmake: '-DOPENVDB_CORE_SHARED=OFF -DOPENVDB_AX_SHARED=OFF -DUSE_STATIC_DEPENDENCIES=ON -DBLOSC_USE_EXTERNAL_SOURCES=ON' }
      fail-fast: false
    steps:
    - uses: actions/checkout@v3
    - name: llvm
-      run: ./ci/install_llvm_windows.sh ${{ matrix.config.crt }}
+      run: ./ci/install_llvm_windows.sh ${{ matrix.config.llvm }} ${{ matrix.config.crt }}
    - name: install
      shell: powershell
-      run: .\ci\install_windows.ps1
+      run: |
+        vcpkg update
+        vcpkg install tbb gtest
    - name: build
      run: >
        ./ci/build.sh -v
@@ -410,10 +420,13 @@ jobs:
        --components="${{ matrix.config.components }}"
        --cargs=\'
        -A x64 -G \"Visual Studio 17 2022\"
+        -DOPENVDB_USE_DELAYED_LOADING=OFF
+        -DUSE_BLOSC=OFF \
+        -DUSE_ZLIB=OFF \
        -DVCPKG_TARGET_TRIPLET=${VCPKG_DEFAULT_TRIPLET}
        -DCMAKE_TOOLCHAIN_FILE=\"${VCPKG_INSTALLATION_ROOT}\\scripts\\buildsystems\\vcpkg.cmake\"
        -DMSVC_COMPRESS_PDB=ON
-        -DOPENVDB_AX_TEST_CMD_DOWNLOADS=ON
+        -DOPENVDB_AX_TEST_CMD_DOWNLOADS=OFF
        -DUSE_EXPLICIT_INSTANTIATION=OFF
        -DLLVM_DIR=\"${HOME}\\llvm_install\\lib\\cmake\\llvm\"
        -DCMAKE_INSTALL_PREFIX=\"${VDB_INSTALL_PREFIX}\"
@@ -485,7 +498,7 @@ jobs:
        \'
    - name: test
      shell: bash
-      run: cd build && ctest -V -E ".*cuda.*"
+      run: cd build && ctest -V -E ".*cuda.*|.*mgpu.*"

  #############################################################################
  ################################## Blosc ####################################
@@ -499,10 +512,10 @@ jobs:
    runs-on: ${{ (github.repository_owner == 'AcademySoftwareFoundation' && 'ubuntu-22.04-8c-32g-300h') || 'ubuntu-latest' }}
    name: linux-blosc:${{ matrix.blosc }}
    container:
-      image: aswf/ci-base:2023
+      image: aswf/ci-openvdb:2025-clang19.1
    strategy:
      matrix:
-        blosc: ['1.18.0','1.19.0','1.20.0','1.21.0']
+        blosc: ['1.21.0', 'latest']
      fail-fast: false
    steps:
    - uses: actions/checkout@v3
@@ -4,10 +4,12 @@ name: Whitespace
 on:
  push:
    paths-ignore:
+      - 'ext/**'
      - 'pendingchanges/**'
      - 'tsc/meetings/**'
  pull_request:
    paths-ignore:
+      - 'ext/**'
      - 'pendingchanges/**'
      - 'tsc/meetings/**'

@@ -25,7 +27,7 @@ jobs:
    - name: test
      run: |
          set +e
-          git grep -n -I -E '^.+[ ]+$' -- ':!tsc/meetings/*' ':!pendingchanges/*' ':!*.wlt'
+          git grep -n -I -E '^.+[ ]+$' -- ':!tsc/meetings/*' ':!pendingchanges/*' ':!ext/*' ':!*.wlt'
          test $? -eq 1

  # Search for any tabs excluding meeting notes, image files and a few others
@@ -36,5 +38,5 @@ jobs:
    - name: test
      run: |
          set +e
-          git grep -n "	" -- ':!*/whitespace.yml' ':!tsc/meetings/*' ':!*.svg' ':!*.cmd' ':!*.png' ':!pendingchanges/*' ':!*.wlt' ':!*.jpg' ':!*.gif' ':!*.mp4' ':!*.pt' ':!*.pth' ':!*.nvdb' ':!*.npz'
+          git grep -n "	" -- ':!*/whitespace.yml' ':!tsc/meetings/*' ':!*.svg' ':!*.cmd' ':!*.png' ':!pendingchanges/*' ':!ext/*' ':!*.wlt' ':!*.jpg' ':!*.gif' ':!*.mp4' ':!*.pt' ':!*.pth' ':!*.nvdb' ':!*.npz'
          test $? -eq 1
@@ -1,18 +1,306 @@
 OpenVDB Version History
 =======================

-Version 12.0.2 - In development
+Version 13.0.1 - In Development
+
+Version 13.0.0 - November 3, 2025
+
+      This version introduces ABI changes relative to older major releases,
+      so to preserve ABI compatibility it might be necessary to define the
+      macro OPENVDB_ABI_VERSION_NUMBER=N, where, for example, N is 11 for
+      Houdini 20.5 and 12 for Houdini 21.0.
+
+      Support for reading .vdb files written prior to VDB 1.0 has been removed.
+      To read older .vdb files, please use VDB 12.x or older and re-write them,
+      so that they use the newer file format and can be read in VDB 13+.
+
+  Highlights:
+  - OpenVDB now includes in-memory support for Half Grids and the file format
+    has been extended to add I/O serialization for half grids. More work is
+    planned on adapting algorithms when using Half Grids for better performance
+    and improved precision.
+  - NanoVDB is no longer limited to static applications, such as rendering, but
+    can increasingly be applied to problems that involve dynamic topology, such
+    as level set tracking, grid building, morphology operations and merging of
+    grids. This is due to the fact that we keep adding more and more tools to
+    NanoVDB that facilitates tree modification. However, unlike OpenVDB, these
+    operations generate new NanoVDB grids (as oppose to in-palace modifications
+    in OpenVDB), yet are typically much faster since they are performed on the
+    GPU. Many of these exciting new features are the foundation of fVDB.
+
+  OpenVDB:
+    New features:
+    - Added native registration of HalfGrid type.
+    - Added customization of the output transformation attribute for
+      points::pca(). Can be output as a quaternion or as a combined affine
+      transform with the scale/stretch ommitted.
+    - Added support for input quaternions attributes and affine
+      transformations (that include scale) in points::rasterizeSdf() for
+      ellipsoidal rasterizations.
+
+    API Changes:
+    - Added LeafBuffer::storage() for direct const access to the raw data
+      buffer when using the bool LeafNode.
+    - Removed Grid::evalMinMax(), use tools::minMax for superior threaded
+      performance.
+    - Removed points::rasterize() that took a custom filter and interrupter.
+    - Removed LeafNode<bool, ...>() TopologyCopy constructor in favor of using
+      explicit mask/buffer constructors.
+    - Removed ValueAccessor NodeT0, NodeT1, NodeT2 aliases, use NodeTypeAtLevel
+      instead.
+    - Add InternalNode::setValueMaskUnsafe() to write a value mask directly.
+    - LegacyFrustum and RootNodeMask have been removed. These classes were only
+      used for reading old .vdb files, which is no longer supported.
+    - io::writeCompressedValues() now takes a const ValueT* source buffer for
+      improved const-correctness.
+
+    Bug Fixes:
+    - Fix a bug in the LevelSetRayIntersector class where two of the methods
+      were incorrectly adjusting the time even when an intersection was not
+      found.
+    - Fixed an assertion in math::Quat where a vector length check was using
+      too small a tolerance.
+    - Fixed a bug in the unit tests where detecting if assertions were enabled
+      was using NDEBUG rather than OpenVDB assertions.
+
+    Build:
+    - Removed the unnecessary USE_NUMPY build option for the OpenVDB Python
+      module. All functions which take numpy arrays are now always generated
+      (numpy is not required at compile time for this support).
+    - Bumped minimum supported CMake version to 3.24, and new future minimum
+      to 4.0.0.
+    - Bumped minimum supported OpenEXR and Imath versions to 3.2, and new
+      future minimum to 3.3.
+    - Bumped minimum supported Boost version to 1.82, and new future minimum
+      to 1.85.
+    - Bumped minimum supported Python version to 3.11
+    - Bumped minimum supported Log4cplus version to 2.0
+    - Bumped minimum supported Houdini version to 20.5
+    - Bumped minimum supported GLFW version to 3.3
+    - Bumped future minimum for TBB to 2021 (oneapi).
+    - Bumped future minimum for nanobind to 2.5.0.
+
+  OpenVDB AX:
+    New:
+    - Added support for LLVM 21.
+
+  NanoVDB:
+    Highlights:
+    - NanoVDB is no longer limited to static applications, such as rendering,
+      but can increasingly be applied to problems that involve dynamic
+      topology, such as level set tracking, grid building, morphology
+      operations and merging of grids. This is due to the fact that we keep
+      adding more and more tools to NanoVDB that facilitates tree modification.
+      However, unlike OpenVDB, these operations generate new NanoVDB grids (as
+      oppose to in-palace modifications in OpenVDB), yet are typically much
+      faster since they are performed on the GPU. Many of these exciting new
+      features are the foundation of fVDB.
+
+    New features:
+    - Added nanovdb/tools/cuda/DilateGrid.cuh, which performs topology dilation
+      on the GPU.
+    - Added nanovdb/tools/cuda/MergeGrids.cuh, which merges the topology of two
+      grids on the GPU.
+    - Added nanovdb/tools/cuda/CoarsenGrid.cuh, which performs topological
+      coarsening of grids on the GPU.
+    - Added nanovdb/tools/cuda/RefineGrid.cuh, which performs topological
+      refinement of grids on the GPU.
+    - Added nanovdb/tools/cuda/PruneGrid.cuh, which prunes the topology of a
+      grid using a sidecar mask on the GPU.
+    - Added nanovdb/util/cuda/Injection.cuh, which facilitates sidecar data
+      copying - on the GPU - in the overlap of two grids with different
+      topology.
+    - Added Mat classes and other math types to NanoVDB.
+    - Major rewrite of nanovdb::NanoToOpenVDB, which now allows
+      nanovdb::IndexGrids and blind (or sidecar) data to be converted into an
+      openvdb grid.
+    - Added support for LevelSet, FogVolume, and Staggered grid semantics.
+      Added toGridClass and toSemantic to map Grid Blind data semantic to Grid
+      class and vice versa.
+
+    Improvements:
+    - Improvements to mGPU grid building and introduction of a separate test
+      executable for mGPU tests
+    - Added stand-alone function nanovdb::writeUncompressedGrid that allows
+      nanovdb::IndexGrids and blind (or sidecar) data to be written directly to
+      a stream.
+    - nanovdb_print now (optionally) prints the blind data count per grid.
+    - Extended nanovdb::GridBlindDataSemantic with nanovdb::GridClass types to
+      support conversion between IndexGrid and regular grids.
+    - Improvements and bugfixes to PNanoVDB.
+
+    API Changes:
+    - Removed nanovdb::ValueIndexMask and nanovdb::ValueOnIndexMask since they
+      are no longer used by fVDB (or elsewhere).
+    - NanoVDB version is now 32.9.0, so the ABI of the grid is preserved but
+      not the API. However, the API changes are very minor!
+
+    Bug Fixes:
+    - Several fixes for compiler warnings.
+    - Fixed error in nanovdb/io/IO.h related to missing throws.
+
+  Houdini:
+  - Remove support for openvdb_houdini::Interrupter, use
+    openvdb_houdini::HoudiniInterrupter instead.
+  - Remove input-stealing via duplicateSourceStealable(), SideFX have since
+    introduced verbification and compiled SOPs to provide zero-copy node
+    evaluation.
+
+
+Version 12.1.1 - September 30, 2025
+
+  OpenVDB AX:
+    Fixes:
+    - Fixed an issue which was causing host specific optimization passes to be
+      omitted when using LLVM 16 and higher.
+
+  Houdini:
+    - Fixed a compilation issue with the OpenVDB Activate SOP with Houdini 21
+    [Reported by @callen128]
+
+
+Version 12.1.0 - August 8, 2025
+
+  OpenVDB:
+    New features:
+    - New level set constructor tools::createLevelSetCapsule creates a grid
+      containing a narrow-band level set representation of a capsule (tube
+      with constant radius and sphere caps).
+    - New level set constructor tools::createLevelSetTaperedCapsule creates a
+      grid containing a narrow-band level set representation of a tapered
+      capsule (tube with sphere caps and different radii at both ends, or
+      equivalently the convex hull of two spheres with possibly different
+      centers and radii).
+    - New level set constructor tools::createLevelSetTubeComplex creates a grid
+      containing a narrow-band level set representation of a tube complex (a
+      collection of capsules or tapered capsules defined by endpoint coordinates
+      and segment indices).
+    - New level set constructor tools::createLevelSetDilatedMesh creates a grid
+      containing a narrow-band level set representation of a dilated surface
+      mesh (dilated by a radius in all directions).
+    - New class tools::ConvexVoxelizer, used by the above methods, which can be
+      inherited to implement a convex level set constructor.
+    - Introduced anisotropic surfacing tools in points/PointRasterizeSDF.h and
+      principal component analysis (PCA) methods in points/PrincipalComponentAnalysis.h.
+      The latter tool analyses point neighborhoods and computes affine
+      transformations representing elliptical distributions. The prior tool
+      takes rotational and stretch point attributes (optionally computed from
+      the prior PCA methods) and generates anisotropic surfaces.
+
+    Improvements:
+    - Significantly improved the performance of all LeafNode ValueIterators,
+      up to 5x on some platforms and up to 10x when delay loading is enabled.
+      Construction of a ValueIterator from a leaf node now requests the leaf
+      buffers ahead of iteration to avoid potentially expensive API calls.
+    - Improved the performance of PointRasterizeSDF.h tools with larger radii
+      with better bound computation approximations.
+    - Significant performance improvement to delay loaded files - the warning
+      logic which checks the underlying mapped file for the last modified time
+      has been removed, resulting in performance gains upwards of 10x when
+      loading in node data buffers. Note that the check that was removed is
+      entirely redundant on Windows and would only print a warning on other
+      system.
+      [Contributed by jjtParadox]
+    - Added OPENVDB_HAS_INVOKABLE_MEMBER_FUNCTION and OPENVDB_HAS_MEMBER_FUNCTION
+      macros to help determine the presence and compatibility of interface
+      member functions at compile time.
+    - Added OPENVDB_ASSUME macros to mimic builtin assume and C++23 assume
+      attributes.
+    - Improved read and write of PLY files with vdb_tool so they support Vec3d
+      and 64 bit attributes. Changed vdb_tool version number from 10.6.1 to
+      10.7.0.
+
+    API Changes:
+    - Mat3 and Mat4 functions have been marked [[nodiscard]] where it may be
+      confusing if they are operating in place or returning a result, in
+      particular tranpose, invert, adjoint, and timesDiagonal.
+    - Re-worked the API in PointRasterizeSDF.h with the addition of anisotropic
+      surfacing kernels. The new API exposes more concise free functions which
+      take structs of parameters. The old API remains supported but is
+      deprecated.
+    - Deprecated providing a filter or interrupter as arguments to
+      points::rasterize. These can instead be implemented on user transfer
+      schemes.
+    - Transfer schemes provided to points::rasterize can now optionally
+      implement a rasterizePoints callback that takes a range of points per
+      voxel.
+
+    Fixes:
+    - Fixed a bug in points::rasterizeSpheres and points::rasterizeSmoothSpheres
+      which could cause sections of the generated surface to incorrectly be
+      marked as negative interior values.
+    - Fixed a bug in points::rasterizeSmoothSpheres which could result in
+      incomplete surface reconstruction with significantly larger search
+      distances.
+    - Fixed some precision issues in various Matrix methods.
+    - Fixed bugs in vdb_tool that prevented build with newer namespaces in
+      nanovdb.
+
+  OpenVDB AX:
+    Improvements:
+    - Added support for LLVM versions greater than 15. Upgraded CI to test with
+      LLVM versions 16, 17, 18, 19 and 20.
+    - As of this release, LLVM 15 is the minimum required version for OpenVDB
+      AX. Note that there is an issue with LLVM 17 symbols which makes it
+      incompatible with the OpenVDB Python plugin. AS such, LLVM version 18 and
+      greater are the recommended LLVM versions to use, with LLVM 18 being the
+      new future minimum version.
+    - Significant performance improvement to the compilation of AX kernels when
+      using LLVM 18 and greater.
+    - Significant backend refactoring to AX IR generation to support newer
+      versions of LLVM. In particular, parts of the codegen::Function and
+      codegen::FunctionBuilder API require new usage from LLVM 16 due to
+      the LLVM opaque pointer changes. If you're not using the OpenVDB AX
+      Function codegen API then no client changes should be necessary.
+    - Migrated OpenVDB AX unit tests from cppunit to gtest
+      [Contributed by Tim Straubinger]
+
+    Fixes:
+    - Fixed a case of undefined behaviour when explicitly casting variables to
+      the same type as their storage type e.g:
+        double a; double b = double(a);
+    - Fixed an AX compiler error which would produce invalid IR when using 64bit
+      integer variables for matrix array offsetting e.g:
+        int64 a = 0; mat4f b = 1; float c = b[a,a];
+
+  NanoVDB:
+    - Added nanovdb::openToIndexVDB, which can convert an OpenVDB grid into a
+      NanoVDB IndexGrid.
+    - Added new options "--index" and "--onIndex" to the command-line tool
+      nanovdb_convert so it can produce IndexGrids.
+    - Fixed memory leak in DeviceBuffer.
+    - Added constructor for DeviceBuffer from an existing HostBuffer.
+    - Added VoxelBlockManager, an acceleration structure for sequential access
+      and stencil ops over active voxels.
+    - Added helper functions for kernel launches with shared memory.
+    - Added support for CUDA 13.0.
+    - Fix warnings and compilation errors for GCC and Clang.
+    - Fixed inconsistent usage of util::cuda::mallocAsync/freeAsync that
+      affected vGPUs.
+    - Added support for custom memory resources in PointsToGrid.
+    - Fixed race condition in split/mergeGridHandles.
+    - Fixed offset bug in nanovdb::tools::cuda::IndexToGrid related to non-POD
+      build types.
+      [Contributed by Andreas Henne]
+
+  Houdini:
+    - Define OP_InputIdx and OP_OutputIdx for Houdini 21.0 compatibility
+    - Add Houdini 21 style camera interface to support LOP cameras
+    - Include env flag for avoiding stream writing (was backported to 19.5)


 Version 12.0.1 - April 3, 2025

  OpenVDB:
    New features:
-    - Added read and write support for OFF (Object File Format) files to vdb_tool
+    - Added read and write support for OFF (Object File Format) files to
+      vdb_tool

    Build:
-    - Fixed a build issue with vdb_render on Windows with newer versions of glew not including glu.h by default
-    - Fixed issues with newer versions of clang refusing to accept older ::template syntax
+    - Fixed a build issue with vdb_render on Windows with newer versions of
+      glew not including glu.h by default
+    - Fixed issues with newer versions of clang refusing to accept older
+      ::template syntax

  NanoVDB:
    New features:
@@ -20,10 +308,12 @@ Version 12.0.1 - April 3, 2025
    - Added a UnifiedBuffer class that wraps CUDA unified memory
    - Added example for multiGPU sparse convolution
    - Added CUDA utility functions for device queries
-    - Added functions to independently stop and compute the elapsed time for timer classes
+    - Added functions to independently stop and compute the elapsed time for
+      timer classes

    Improvements:
-    - Replaced CUB's CachingDeviceAllocator with the default asynchronous stream ordered allocator in PointsToGrid for improved performance
+    - Replaced CUB's CachingDeviceAllocator with the default asynchronous
+      stream ordered allocator in PointsToGrid for improved performance

    Bug fix:
    - Map is now properly transposed when converting NanoVDB to OpenVDB.
@@ -42,7 +42,7 @@

 # note: cmake_minimum_required must be called before project commands to
 #  ensure policy scope is set up correctly
-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)

 # CMP0091 allows for MSVC ABI targetting via CMAKE_MSVC_RUNTIME_LIBRARY
 # from CMake 3.15 and above. Must come before project().
@@ -52,9 +52,9 @@ endif()

 ###### Version

-set(OpenVDB_MAJOR_VERSION 12)
+set(OpenVDB_MAJOR_VERSION 13)
 set(OpenVDB_MINOR_VERSION 0)
-set(OpenVDB_PATCH_VERSION 2)
+set(OpenVDB_PATCH_VERSION 1)
 set(OpenVDB_VERSION "${OpenVDB_MAJOR_VERSION}.${OpenVDB_MINOR_VERSION}.${OpenVDB_PATCH_VERSION}")

 project(OpenVDB LANGUAGES CXX VERSION ${OpenVDB_VERSION})
@@ -80,6 +80,10 @@ option(OPENVDB_BUILD_NANOVDB "Build the NanoVDB library. Turns ON if USE_NANOVDB
 # Global options
 option(OPENVDB_ENABLE_RPATH "Build with RPATH information" ON)
 option(OPENVDB_ENABLE_ASSERTS "Build with asserts in OpenVDB code enabled" OFF)
+option(OPENVDB_ENABLE_VEC_RELATIONAL_OPERATIONS [=[
+Enabled deprecated relational < and > operators on Vector types. This option
+will be switched to OFF and the feature will be subsequently removed in future
+releases.]=] ON)
 option(OPENVDB_USE_DELAYED_LOADING "Build the core OpenVDB library with delayed-loading." ON)
 option(OPENVDB_CXX_STRICT "Enable or disable pre-defined compiler warnings" OFF)
 cmake_dependent_option(OPENVDB_INSTALL_CMAKE_MODULES
@@ -131,6 +135,30 @@ required.]=] ${USE_EXR})
 option(USE_PNG "Use PNG while building openvdb components." OFF)
 option(USE_AX "Use OpenVDB AX while building openvdb components." ${OPENVDB_BUILD_AX})
 option(USE_NANOVDB "Use NanoVDB while building openvdb components." ${OPENVDB_BUILD_NANOVDB})
+option(USE_VCL [=[
+Use the internally supplied copy of Agner Fog's vectorclass (VCL) SIMD wrapper
+library to instrument OpenVDB tools. This also requires VCL to be installed
+along with OpenVDB headers. Only applicable which compiling for x86 targets.]=] OFF)
+set(OPENVDB_X86_INSTRSET 0 CACHE STRING [=[
+The x86 instruction set to use. When unset, if USE_VCL is ON, this is
+automatically configured based on provided build flags. When unset and USE_VCL
+is OFF, defaults to 0. It is expected for you to provide this flag. Available
+values are:
+    0:  Unknown or unspecified
+    1:  SSE             -msse
+    2:  SSE2            -msse2 # Minimum on x86_64
+    3:  SSE3            -msse3
+    4:  SSSE3           -mssse3
+    5:  SSE4.1          -msse4.1
+    6:  SSE4.2          -msse4.2 # Minimum recommended where supported
+    7:  AVX             -mavx
+    8:  AVX2            -mavx2
+    9:  AVX512F         -mavx512f
+    10: AVX512BW/DQ/VL  -mavx512bw -mavx512dq -mavx512vl
+Note that (while not strictly enforced by vendors) each option implies all
+preceeding flags. If you need more granular control you can provide your
+desired flags via CXX_FLAGS. Only applicable which compiling for x86 targets.
+]=])

 cmake_dependent_option(OPENVDB_DISABLE_BOOST_IMPLICIT_LINKING
  "Disable the implicit linking of Boost libraries on Windows" ON "WIN32" OFF)
@@ -172,19 +200,6 @@ elseif(NOT ${CONCURRENT_MALLOC} IN_LIST _CONCURRENT_MALLOC_OPTIONS)
  set(CONCURRENT_MALLOC Auto CACHE STRING FORCE)
 endif()

-set(_OPENVDB_SIMD_OPTIONS None SSE42 AVX)
-if(NOT OPENVDB_SIMD)
-  set(OPENVDB_SIMD None CACHE STRING
-    "Choose whether to enable SIMD compiler flags or not, options are: None SSE42 AVX.
-    Although not required, it is strongly recommended to enable SIMD. AVX implies SSE42.
-    None is the default." FORCE
-  )
-elseif(NOT ${OPENVDB_SIMD} IN_LIST _OPENVDB_SIMD_OPTIONS)
-  message(WARNING "Unrecognized or unsupported value for OPENVDB_SIMD, "
-    "using None instead.")
-  set(OPENVDB_SIMD None CACHE STRING FORCE)
-endif()
-
 if(USE_BLOSC AND NOT USE_ZLIB)
  message(WARNING "ZLib can only be disabled if Blosc is also disabled. Enabling ZLib.")
 endif()
@@ -268,6 +283,13 @@ mark_as_advanced(

 include(cmake/config/OpenVDBVersions.cmake)

+if(OPENVDB_FUTURE_DEPRECATION AND FUTURE_MINIMUM_CMAKE_VERSION)
+  if(CMAKE_VERSION VERSION_LESS ${FUTURE_MINIMUM_CMAKE_VERSION})
+    message(DEPRECATION "Support for CMake versions < ${FUTURE_MINIMUM_CMAKE_VERSION} "
+      "is deprecated and will be removed.")
+  endif()
+endif()
+
 foreach(DEPRECATED_ABI ${OPENVDB_DEPRECATED_ABI_LIST})
  option(OPENVDB_USE_DEPRECATED_ABI_${DEPRECATED_ABI} "Bypass minimum ABI requirement checks" OFF)
  mark_as_advanced(OPENVDB_USE_DEPRECATED_ABI_${DEPRECATED_ABI})
@@ -378,6 +400,28 @@ endif()

 #########################################################################

+# Still support -DOPENVDB_SIMD=SSE42 and -DOPENVDB_SIMD=AVX, these alias to
+# their respective OPENVDB_X86_INSTRSET value but, as a build option, are
+# deprecated. Not that the defines in version.h need to persist in case
+# downstream software was using them before including VDB headers
+set(_OPENVDB_SIMD_OPTIONS None SSE42 AVX)
+if(NOT OPENVDB_SIMD)
+  set(OPENVDB_SIMD None CACHE STRING "Deprecated, see OPENVDB_X86_INSTRSET." FORCE)
+else()
+  if(NOT ${OPENVDB_SIMD} IN_LIST _OPENVDB_SIMD_OPTIONS)
+    message(WARNING "Unrecognized or unsupported value for OPENVDB_SIMD, using None instead.")
+    set(OPENVDB_SIMD None CACHE STRING "Deprecated, see OPENVDB_X86_INSTRSET." FORCE)
+  endif()
+  message(DEPRECATION "The OPENVDB_SIMD option is deprecated and will be removed. Use OPENVDB_X86_INSTRSET.")
+  if(OPENVDB_SIMD STREQUAL "AVX")
+    set(OPENVDB_X86_INSTRSET 7)
+  elseif(OPENVDB_SIMD STREQUAL "SSE42")
+    set(OPENVDB_X86_INSTRSET 6)
+  endif()
+endif()
+
+#########################################################################
+
 # Configure general CMake and CXX settings

 include(cmake/config/OpenVDBCXX.cmake)
@@ -397,22 +441,6 @@ if(CONCURRENT_MALLOC STREQUAL "Auto")
  endif()
 endif()

-# Configure SIMD. AVX implies SSE 4.2.
-
-if(OPENVDB_SIMD STREQUAL "AVX")
-  if (CMAKE_SYSTEM_PROCESSOR MATCHES "(x86)|(X86)|(amd64)|(AMD64)")
-    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-mavx>")
-    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-msse4.2>")
-  endif()
-  add_compile_definitions("$<$<COMPILE_LANGUAGE:CXX>:OPENVDB_USE_AVX>")
-  add_compile_definitions("$<$<COMPILE_LANGUAGE:CXX>:OPENVDB_USE_SSE42>")
-elseif(OPENVDB_SIMD STREQUAL "SSE42")
-  if (CMAKE_SYSTEM_PROCESSOR MATCHES "(x86)|(X86)|(amd64)|(AMD64)")
-    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-msse4.2>")
-  endif()
-  add_compile_definitions("$<$<COMPILE_LANGUAGE:CXX>:OPENVDB_USE_SSE42>")
-endif()
-
 #########################################################################

 # Configure our cmake modules to only search for static libraries
@@ -478,10 +506,7 @@ endif()
 # Locate Python and nanobind if necessary
 if(OPENVDB_BUILD_PYTHON_MODULE OR (OPENVDB_BUILD_NANOVDB AND NANOVDB_BUILD_PYTHON_MODULE))

-  # Call find_package(Python ...)
  find_package(Python ${MINIMUM_PYTHON_VERSION} REQUIRED COMPONENTS Development Interpreter)
-  find_package(nanobind ${MINIMUM_NANOBIND_VERSION} REQUIRED)
-
  if(OPENVDB_FUTURE_DEPRECATION AND FUTURE_MINIMUM_PYTHON_VERSION)
    if(Python_VERSION VERSION_LESS ${FUTURE_MINIMUM_PYTHON_VERSION})
      message(DEPRECATION "Support for Python versions < ${FUTURE_MINIMUM_GLFW_VERSION} "
@@ -489,6 +514,20 @@ if(OPENVDB_BUILD_PYTHON_MODULE OR (OPENVDB_BUILD_NANOVDB AND NANOVDB_BUILD_PYTHO
    endif()
  endif()

+  find_package(nanobind REQUIRED CONFIG)
+  include(FindPackageHandleStandardArgs)
+  find_package_handle_standard_args(nanobind
+    FOUND_VAR nanobind_FOUND
+    REQUIRED_VARS NB_DIR
+    VERSION_VAR nanobind_VERSION)
+
+  if(OPENVDB_FUTURE_DEPRECATION AND FUTURE_MINIMUM_NANOBIND_VERSION)
+    if(nanobind_VERSION VERSION_LESS ${FUTURE_MINIMUM_NANOBIND_VERSION})
+      message(DEPRECATION "Support for nanobind versions < ${FUTURE_MINIMUM_NANOBIND_VERSION} "
+        "is deprecated and will be removed.")
+    endif()
+  endif()
+
  if(NOT DEFINED VDB_PYTHON_INSTALL_DIRECTORY)
    if(DEFINED Python_SITELIB)
      get_filename_component(Python_PACKAGES_DIR ${Python_SITELIB} NAME)
@@ -57,6 +57,10 @@ code and has the right to release it under the
 [Apache License, version 2.0](LICENSE.md)
 license. See the [TAC documentation on contribution sign off](https://github.com/AcademySoftwareFoundation/tac/blob/master/process/contributing.md#contribution-sign-off) for more information on this requirement.

+#### Stale PRs
+
+A PR that does not have a signed CLA or DCO sign-off will be given at least one week in which to resolve the issue and then will be closed. The TSC cannot review or merge PRs without these requirements having been met. Please re-open PRs that are now expected to pass once this criteria has been met.
+
 ## Maintainer

 The maintainer role is the equivalent of the "Committer" role in the charter.
@@ -95,8 +95,6 @@ vcpkg install zlib:x64-windows
 vcpkg install blosc:x64-windows
 vcpkg install tbb:x64-windows
 vcpkg install boost-iostreams:x64-windows
-vcpkg install boost-any:x64-windows
-vcpkg install boost-algorithm:x64-windows
 vcpkg install boost-interprocess:x64-windows
 ```

@@ -33,8 +33,13 @@ PARMS[--target]=install
 PARMS[--build-dir]=build

 if [[ $RUNNER_NAME == *"8c-32g-300h"* ]]; then
-    # ASWF github actions runners have 8 threads
-    PARMS[-j]=8
+    if [[ $CXX == "g++" ]]; then
+        # GCC hits memory limits on runners, reduce threads
+        PARMS[-j]=6
+    else
+        # ASWF github actions runners have 8 threads
+        PARMS[-j]=8
+    fi
 else
    # Github actions runners have 2 threads
    # https://help.github.com/en/actions/reference/virtual-environments-for-github-hosted-runners
@@ -51,7 +56,8 @@ declare -A COMPONENTS
 COMPONENTS['core']='OPENVDB_BUILD_CORE'
 COMPONENTS['python']='OPENVDB_BUILD_PYTHON_MODULE'
 COMPONENTS['test']='OPENVDB_BUILD_UNITTESTS'
-COMPONENTS['bin']='OPENVDB_BUILD_BINARIES'
+# @todo split out vdb tool from this...
+COMPONENTS['bin']='OPENVDB_BUILD_VDB_PRINT,OPENVDB_BUILD_VDB_LOD,OPENVDB_BUILD_VDB_TOOL,OPENVDB_BUILD_VDB_TOOL_UNITTESTS'
 COMPONENTS['view']='OPENVDB_BUILD_VDB_VIEW'
 COMPONENTS['render']='OPENVDB_BUILD_VDB_RENDER'
 COMPONENTS['hou']='OPENVDB_BUILD_HOUDINI_PLUGIN'
@@ -67,6 +73,8 @@ COMPONENTS['nanotest']='NANOVDB_BUILD_UNITTESTS'
 COMPONENTS['nanoexam']='NANOVDB_BUILD_EXAMPLES'
 COMPONENTS['nanobench']='NANOVDB_BUILD_BENCHMARK'
 COMPONENTS['nanotool']='NANOVDB_BUILD_TOOLS'
+COMPONENTS['nanopython']='NANOVDB_BUILD_PYTHON_MODULE'
+COMPONENTS['nanopytest']='NANOVDB_BUILD_PYTHON_UNITTESTS'

 ################################################

@@ -150,7 +158,11 @@ for comp in "${!COMPONENTS[@]}"; do
            setting="ON"; break
        fi
    done
-    CMAKE_EXTRA+=("-D${COMPONENTS[$comp]}=$setting")
+    OPENVDB_TOGGLES="${COMPONENTS[$comp]}"
+    OPENVDB_TOGGLES=(${OPENVDB_TOGGLES//,/ })
+    for toggle in "${!OPENVDB_TOGGLES[@]}"; do
+        CMAKE_EXTRA+=("-D${OPENVDB_TOGGLES[$toggle]}=$setting")
+    done
 done

 ################################################
@@ -182,17 +194,14 @@ cd ${BUILD_DIR}
 set -x

 # Note:
-# - print and lod binary options are always on and can be toggles with: OPENVDB_BUILD_BINARIES=ON/OFF
+# - OPENVDB_BUILD_BINARIES is always ON, but each binary will be toggled depending on the provided comps
 # - always enabled the python tests with OPENVDB_BUILD_PYTHON_UNITTESTS if the python module is in use,
 #   regardless of the 'test' component being enabled or not (see the OPENVDB_BUILD_PYTHON_UNITTESTS option).
 cmake \
-    -DOPENVDB_USE_DEPRECATED_ABI_10=ON \
    -DOPENVDB_USE_DEPRECATED_ABI_11=ON \
-    -DOPENVDB_BUILD_VDB_PRINT=ON \
-    -DOPENVDB_BUILD_VDB_LOD=ON \
-    -DOPENVDB_BUILD_VDB_TOOL=ON \
-    -DOPENVDB_BUILD_VDB_TOOL_UNITTESTS=ON \
+    -DOPENVDB_USE_DEPRECATED_ABI_12=ON \
    -DOPENVDB_TOOL_USE_NANO=OFF \
+    -DOPENVDB_BUILD_BINARIES=ON \
    -DOPENVDB_BUILD_PYTHON_UNITTESTS=ON \
    -DMSVC_MP_THREAD_COUNT=${PARMS[-j]} \
    "${CMAKE_EXTRA[@]}" \
@@ -0,0 +1,208 @@
+version: 0.2
+
+# Runs the NanoVDB CUDA / multi-GPU unit tests on AWS CodeBuild (4x A10G).
+# This spec does NOT build the project. Instead it downloads the prebuilt
+# build tree produced by the matching GitHub Actions run (see the
+# linux-nanovdb / linux-nanovdb-gpu jobs in .github/workflows/nanovdb.yml),
+# restores it at its original absolute path, and executes the GPU and Python tests.
+#
+# Required environment variables (provided by aws-codebuild-run-build):
+#   ARTIFACT_NAME     - name of the GitHub Actions artifact to fetch
+#   GH_TOKEN          - ephemeral GITHUB_TOKEN with actions:read
+#   GITHUB_REPOSITORY - owner/repo (auto-forwarded GITHUB_* variable)
+#   GITHUB_RUN_ID     - workflow run that produced the artifact (auto-forwarded)
+#   ACTIONS_RUNTIME_TOKEN - job-scoped token used to upload the report
+#   ACTIONS_RESULTS_URL   - GitHub Actions artifact service endpoint
+
+env:
+  variables:
+    CUDA_HOME: /usr/local/cuda-12
+    DIAGNOSTIC_GIST_ID: c082e01bf80cd88d045e1e61bd78d43c
+    DIAGNOSTIC_GIST_REVISION: dcd46d4e3a65c5d36e1cde96ade0c102464d8a45
+
+phases:
+  install:
+    commands:
+      # Rocky Linux 8 image: tools required to fetch and unpack artifacts and
+      # clone the multi-GPU diagnostic gist.
+      - dnf install -y jq unzip tree 'dnf-command(config-manager)'
+      - dnf config-manager --add-repo https://cli.github.com/packages/rpm/gh-cli.repo
+      - dnf install -y gh
+      - dnf module reset -y nodejs
+      - dnf module enable -y nodejs:20
+      - dnf install -y nodejs
+      - npm install --prefix /tmp/actions-artifact-uploader --omit=dev --no-audit --no-fund @actions/artifact@6.2.1
+
+  pre_build:
+    commands:
+      # Confirm the GPUs are visible (expect 4x A10G).
+      - nvidia-smi
+      - nvidia-smi topo -m
+      - |
+        echo "=== IOMMU diagnostics ==="
+
+        kernel_cmdline=$(cat /proc/cmdline)
+        echo "Kernel command line:"
+        echo "${kernel_cmdline}"
+
+        iommu_args=$(echo "${kernel_cmdline}" |
+          grep -Eo '(^|[[:space:]])(iommu|amd_iommu|intel_iommu)=[^[:space:]]+' |
+          xargs || true)
+        if [ -n "${iommu_args}" ]; then
+          echo "Explicit IOMMU kernel arguments: ${iommu_args}"
+        else
+          echo "No explicit IOMMU kernel arguments found"
+        fi
+
+        if dmesg_output=$(dmesg 2>&1); then
+          echo "IOMMU kernel messages:"
+          echo "${dmesg_output}" |
+            grep -Ei 'DMAR|IOMMU|AMD-Vi|amd_iommu|intel_iommu' ||
+            echo "No IOMMU kernel messages found"
+
+          echo "IOMMU passthrough messages:"
+          echo "${dmesg_output}" |
+            grep -Ei 'iommu.*(pt|passthrough)|default domain type.*passthrough' ||
+            echo "No IOMMU passthrough messages found"
+        else
+          echo "dmesg unavailable in this container: ${dmesg_output}"
+        fi
+
+        if compgen -G '/sys/kernel/iommu_groups/*/devices/*' >/dev/null; then
+          echo "Populated IOMMU groups are visible"
+          group_count=0
+          typed_group_count=0
+          for group in /sys/kernel/iommu_groups/*; do
+            [ -d "${group}" ] || continue
+            group_count=$((group_count + 1))
+            if [ -r "${group}/type" ]; then
+              typed_group_count=$((typed_group_count + 1))
+              echo "IOMMU group ${group##*/} type: $(cat "${group}/type")"
+            fi
+          done
+          echo "Visible IOMMU group count: ${group_count}"
+          if [ "${typed_group_count}" -eq 0 ]; then
+            echo "Per-group IOMMU domain types are not exposed by this kernel"
+          fi
+        else
+          echo "No populated IOMMU groups are visible"
+        fi
+
+        if compgen -G '/sys/class/iommu/*' >/dev/null; then
+          echo "IOMMU devices visible in /sys/class/iommu:"
+          for iommu in /sys/class/iommu/*; do
+            echo "  ${iommu##*/}"
+          done
+        else
+          echo "No IOMMU devices are visible in /sys/class/iommu"
+        fi
+      # Locate, download and restore the prebuilt build tree.
+      - 'echo "Fetching artifact ${ARTIFACT_NAME} from run ${GITHUB_RUN_ID}"'
+      - |
+        set -euo pipefail
+        api="https://api.github.com/repos/${GITHUB_REPOSITORY}"
+        artifacts_json=$(curl -fsSL \
+          -H "Authorization: Bearer ${GH_TOKEN}" \
+          -H "Accept: application/vnd.github+json" \
+          -H "X-GitHub-Api-Version: 2022-11-28" \
+          "${api}/actions/runs/${GITHUB_RUN_ID}/artifacts?per_page=100")
+        count=$(echo "${artifacts_json}" | jq --arg n "${ARTIFACT_NAME}" \
+          '[.artifacts[] | select(.name==$n)] | length')
+        if [ "${count}" != "1" ]; then
+          echo "ERROR: expected exactly 1 artifact named '${ARTIFACT_NAME}', found ${count}"
+          exit 1
+        fi
+        artifact_id=$(echo "${artifacts_json}" | jq -r --arg n "${ARTIFACT_NAME}" \
+          '.artifacts[] | select(.name==$n) | .id')
+        echo "Resolved artifact id ${artifact_id}"
+        # The API returns a short-lived (60s) redirect to blob storage. Resolve
+        # the redirect first, then download without the Authorization header so
+        # it is not forwarded to the storage host.
+        download_url=$(curl -sS -o /dev/null -w '%{redirect_url}' \
+          -H "Authorization: Bearer ${GH_TOKEN}" \
+          -H "Accept: application/vnd.github+json" \
+          "${api}/actions/artifacts/${artifact_id}/zip")
+        curl -fsSL -o /tmp/nanovdb-gpu-build.zip "${download_url}"
+      - rm -rf /tmp/nanovdb-gpu-artifact
+      - mkdir -p /tmp/nanovdb-gpu-artifact
+      - unzip -q /tmp/nanovdb-gpu-build.zip -d /tmp/nanovdb-gpu-artifact
+      - |
+        set -euo pipefail
+        set -a
+        . /tmp/nanovdb-gpu-artifact/nanovdb-gpu-meta.env
+        set +a
+        echo "Restoring build tree and Python tests to ${BUILD_WORKSPACE}"
+        mkdir -p "${BUILD_WORKSPACE}"
+        rm -rf "${BUILD_WORKSPACE}/build"
+        mv /tmp/nanovdb-gpu-artifact/build "${BUILD_WORKSPACE}/build"
+        python_test_dir="${BUILD_WORKSPACE}/nanovdb/nanovdb/python/test"
+        mkdir -p "${python_test_dir}"
+        mv /tmp/nanovdb-gpu-artifact/nanovdb/nanovdb/python/test/TestNanoVDB.py \
+          "${python_test_dir}/TestNanoVDB.py"
+
+  build:
+    commands:
+      - |
+        set -euo pipefail
+        set -a
+        . /tmp/nanovdb-gpu-artifact/nanovdb-gpu-meta.env
+        set +a
+        export PATH="${CUDA_HOME}/bin:${PATH}"
+        export LD_LIBRARY_PATH="${CUDA_HOME}/lib64:${BUILD_WORKSPACE}/build/openvdb/openvdb:${BUILD_WORKSPACE}/build/nanovdb/nanovdb:${LD_LIBRARY_PATH:-}"
+        cd "${BUILD_WORKSPACE}/build"
+        ctest -V -R ".*cuda.*|.*mgpu.*|pytest_nanovdb" --output-on-failure
+
+  post_build:
+    commands:
+      - |
+        set -euo pipefail
+        diagnostic_dir="${CODEBUILD_SRC_DIR}/${DIAGNOSTIC_GIST_ID}"
+        rm -rf "${diagnostic_dir}" "${CODEBUILD_SRC_DIR}/report.tgz"
+        cd "${CODEBUILD_SRC_DIR}"
+        gh gist clone "${DIAGNOSTIC_GIST_ID}"
+        # Pin the executable diagnostic code rather than trusting the mutable
+        # head of the public gist on every CI run.
+        git -C "${diagnostic_dir}" checkout --detach "${DIAGNOSTIC_GIST_REVISION}"
+
+        diagnostic_status=0
+        (
+          cd "${diagnostic_dir}"
+          bash ./run_on_runner.sh
+        ) || diagnostic_status=$?
+
+        if [ -d "${diagnostic_dir}/report" ]; then
+          tar czf "${CODEBUILD_SRC_DIR}/report.tgz" \
+            -C "${diagnostic_dir}" report/
+        else
+          echo "ERROR: the diagnostic did not produce report/"
+          exit 1
+        fi
+
+        # Upload directly into the current GitHub Actions job. These values are
+        # short-lived and scoped to this job's artifact service.
+        : "${ACTIONS_RUNTIME_TOKEN:?ACTIONS_RUNTIME_TOKEN was not forwarded}"
+        : "${ACTIONS_RESULTS_URL:?ACTIONS_RESULTS_URL was not forwarded}"
+        cat > /tmp/actions-artifact-uploader/upload.mjs <<'EOF'
+        import {DefaultArtifactClient} from '@actions/artifact'
+
+        const reportPath = `${process.env.CODEBUILD_SRC_DIR}/report.tgz`
+        const artifactName = process.env.ARTIFACT_NAME.replace(
+          /^nanovdb-gpu-build-/,
+          'nanovdb-gpu-report-'
+        )
+        const client = new DefaultArtifactClient()
+        const result = await client.uploadArtifact(
+          artifactName,
+          [reportPath],
+          process.env.CODEBUILD_SRC_DIR,
+          {
+            retentionDays: 7,
+            compressionLevel: 0
+          }
+        )
+        console.log(
+          `Uploaded ${artifactName}: id=${result.id}, size=${result.size}, digest=${result.digest}`
+        )
+        EOF
+        node /tmp/actions-artifact-uploader/upload.mjs
+        exit "${diagnostic_status}"
@@ -8,8 +8,14 @@ HOUDINI_MAJOR="$1"
 PLATFORM="$2"
 OTHER_ARGS="$3"

-pip install --user requests
-python ci/download_houdini.py $HOUDINI_MAJOR $PLATFORM $OTHER_ARGS
+if [[ $PLATFORM =~ "linux" ]]; then
+    pip3 install --user requests
+elif [[ $PLATFORM =~ "macos" ]]; then
+    # Tell the homebrew install of pip3 that we don't care about installing directly through pip3
+    pip3 install --user requests --break-system-packages
+fi
+
+python3 ci/download_houdini.py $HOUDINI_MAJOR $PLATFORM $OTHER_ARGS

 if [[ $PLATFORM =~ "linux" ]]; then
    # create dir hierarchy
@@ -31,20 +37,20 @@ if [[ $PLATFORM =~ "linux" ]]; then
    ls -al dsolib/

    # copy required libraries
-    cp -r toolkit/cmake ../hou/toolkit/.
-    cp -r toolkit/include ../hou/toolkit/.
-    cp -r dsolib/libHoudini* ../hou/dsolib/.
-    cp -r dsolib/libopenvdb_sesi* ../hou/dsolib/.
-    cp -r dsolib/libblosc* ../hou/dsolib/.
-    cp -r dsolib/libhboost* ../hou/dsolib/.
-    cp -r dsolib/libz* ../hou/dsolib/.
-    cp -r dsolib/libbz2* ../hou/dsolib/.
-    cp -r dsolib/libtbb* ../hou/dsolib/.
-    cp -r dsolib/libjemalloc* ../hou/dsolib/.
-    cp -r dsolib/liblzma* ../hou/dsolib/.
-    cp -r dsolib/libIex* ../hou/dsolib/.
-    cp -r dsolib/libImath* ../hou/dsolib/.
-    cp -r dsolib/libIlmThread* ../hou/dsolib/.
+    mv toolkit/cmake ../hou/toolkit/.
+    mv toolkit/include ../hou/toolkit/.
+    mv dsolib/libHoudini* ../hou/dsolib/.
+    mv dsolib/libopenvdb_sesi* ../hou/dsolib/.
+    mv dsolib/libblosc* ../hou/dsolib/.
+    mv dsolib/libhboost* ../hou/dsolib/.
+    mv dsolib/libz* ../hou/dsolib/.
+    mv dsolib/libbz2* ../hou/dsolib/.
+    mv dsolib/libtbb* ../hou/dsolib/.
+    mv dsolib/libjemalloc* ../hou/dsolib/.
+    mv dsolib/liblzma* ../hou/dsolib/.
+    mv dsolib/libIex* ../hou/dsolib/.
+    mv dsolib/libImath* ../hou/dsolib/.
+    mv dsolib/libIlmThread* ../hou/dsolib/.
    cd ..

 elif [[ $PLATFORM =~ "macos" ]]; then
@@ -15,6 +15,6 @@ fi

 mkdir build
 cd build
-cmake ../.
+cmake -DCMAKE_POLICY_VERSION_MINIMUM=3.5 ../.
 make -j8
 make install
@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+# Copyright Contributors to the OpenVDB Project
+# SPDX-License-Identifier: Apache-2.0
+
+set -ex
+EXR_VERSION="$1"
+
+git clone https://github.com/AcademySoftwareFoundation/openexr.git
+cd openexr
+
+if [ "$EXR_VERSION" != "latest" ]; then
+    git checkout tags/${EXR_VERSION} -b ${EXR_VERSION}
+fi
+
+mkdir build
+cd build
+
+cmake ..
+
+make -j$(nproc)
+sudo make install
@@ -3,16 +3,19 @@
 # SPDX-License-Identifier: Apache-2.0

 set -ex
-
 GTEST_VERSION="$1"

-
-git clone https://github.com/google/googletest.git -b v${GTEST_VERSION}
+git clone https://github.com/google/googletest.git
 cd googletest
+
+if [ "$GTEST_VERSION" != "latest" ]; then
+    git checkout tags/${GTEST_VERSION} -b ${GTEST_VERSION}
+fi
+
 mkdir build
 cd build
+
 cmake ..

 make -j$(nproc)
-
 sudo make install
@@ -0,0 +1,21 @@
+#!/usr/bin/env bash
+# Copyright Contributors to the OpenVDB Project
+# SPDX-License-Identifier: Apache-2.0
+
+set -ex
+IMATH_VERSION="$1"
+
+git clone https://github.com/AcademySoftwareFoundation/Imath.git
+cd Imath
+
+if [ "$IMATH_VERSION" != "latest" ]; then
+    git checkout tags/${IMATH_VERSION} -b ${IMATH_VERSION}
+fi
+
+mkdir build
+cd build
+
+cmake ..
+
+make -j$(nproc)
+sudo make install
@@ -4,26 +4,55 @@

 set -ex

-LLVM_CRT="$1"
+LLVM_VER="$1"
+LLVM_CRT="$2"

-git clone -b llvmorg-14.0.0 --depth 1 https://github.com/llvm/llvm-project.git llvm
+# Legacy define for older versions of LLVM
+LLVM_LEGACY_CRT_DEFINE=""
+if [[ "${LLVM_CRT}" == "MultiThreaded" ]]; then
+    LLVM_LEGACY_CRT_DEFINE="MT"
+elif [[ "${LLVM_CRT}" == "MultiThreadedDebug" ]]; then
+    LLVM_LEGACY_CRT_DEFINE="MTd"
+elif [[ "${LLVM_CRT}" == "MultiThreadedDLL" ]]; then
+    LLVM_LEGACY_CRT_DEFINE="MD"
+elif [[ "${LLVM_CRT}" == "MultiThreadedDebugDLL" ]]; then
+    LLVM_LEGACY_CRT_DEFINE="MDd"
+fi
+
+git clone -b llvmorg-${LLVM_VER} --depth 1 https://github.com/llvm/llvm-project.git llvm
 cd llvm

 mkdir .build
 cd .build

-# @note  currently only test static builds with MT
 cmake -B . -S ../llvm -A x64 -G "Visual Studio 17 2022" -Thost=x64 \
    -DCMAKE_INSTALL_PREFIX="${HOME}/llvm_install" \
+    -DCMAKE_BUILD_TYPE=Release \
+    -DLLVM_ENABLE_ASSERTIONS=OFF \
+    -DLLVM_ENABLE_PROJECTS="" \
+    -DLLVM_ENABLE_RUNTIMES="" \
+    -DLLVM_ENABLE_UNWIND_TABLES=OFF \
    -DLLVM_TARGETS_TO_BUILD="host" \
-    -DLLVM_BUILD_TOOLS=OFF \
    -DLLVM_INCLUDE_BENCHMARKS=OFF \
    -DLLVM_INCLUDE_EXAMPLES=OFF \
    -DLLVM_INCLUDE_TESTS=OFF \
+    -DLLVM_INCLUDE_DOCS=OFF \
    -DLLVM_INCLUDE_TOOLS=OFF \
-    -DLLVM_USE_CRT_RELEASE=${LLVM_CRT}
+    -DLLVM_INCLUDE_UTILS=OFF \
+    -DLLVM_BUILD_BENCHMARKS=OFF \
+    -DLLVM_BUILD_EXAMPLES=OFF \
+    -DLLVM_BUILD_TESTS=OFF \
+    -DLLVM_BUILD_DOCS=OFF \
+    -DLLVM_BUILD_TOOLS=OFF \
+    -DLLVM_BUILD_UTILS=OFF \
+    -DLLVM_ENABLE_BINDINGS=OFF \
+    -DLLVM_ENABLE_OCAMLDOC=OFF \
+    -DLLVM_PARALLEL_COMPILE_JOBS=4 \
+    -DLLVM_PARALLEL_LINK_JOBS=4 \
+    -DCMAKE_MSVC_RUNTIME_LIBRARY=${LLVM_CRT} \
+    -DLLVM_USE_CRT_RELEASE=${LLVM_LEGACY_CRT_DEFINE}

-cmake --build . --config Release --target install
+cmake --build . --parallel 4 --config Release --target install

 cd ..

@@ -22,6 +22,7 @@ brew install jq # for trivial parsing of brew json
 brew install openexr
 brew install nanobind # also installs the dependent python version
 brew install robin-map # required for nanobind
+brew install numpy # only for python tests
 brew install zlib
 brew install jemalloc

@@ -45,9 +46,7 @@ LLVM_VERSION=$1
 if [ ! -z "$LLVM_VERSION" ]; then
    if [ "$LLVM_VERSION" == "latest" ]; then
        brew install llvm
-        brew install cppunit
    else
        brew install llvm@$LLVM_VERSION
-        brew install cppunit
    fi
 fi
@@ -0,0 +1,28 @@
+#!/usr/bin/env bash
+# Copyright Contributors to the OpenVDB Project
+# SPDX-License-Identifier: Apache-2.0
+
+set -ex
+
+# Download the TeX Live installer
+wget https://mirrors.mit.edu/CTAN/systems/texlive/tlnet/install-tl-unx.tar.gz
+
+# Extract
+tar xzf install-tl-unx.tar.gz
+rm install-tl-unx.tar.gz
+cd install-tl-*
+
+# Run the installer with no docs or source installation
+./install-tl --no-interaction --no-doc-install --no-src-install --scheme=minimal --repository https://mirrors.mit.edu/CTAN/systems/texlive/tlnet
+
+# Export the texlive PATH to GITHUB_PATH
+TEXLIVE_PATH=$(echo /usr/local/texlive/*/bin/x86_64-linux)
+export PATH=${TEXLIVE_PATH}:${PATH}
+echo "${TEXLIVE_PATH}" >> $GITHUB_PATH
+
+# Install the required packages
+tlmgr install latex-bin dvips collection-fontsrecommended collection-latexrecommended
+
+# Validate that installing dvips and latex has been successful
+dvips --version
+latex --version
@@ -0,0 +1,9 @@
+# Enable verbose and stop on error
+$ErrorActionPreference = "Stop"
+$VerbosePreference = "Continue"
+
+# Install numpy for tests
+& "$env:VCPKG_INSTALLATION_ROOT\installed\$env:VCPKG_DEFAULT_TRIPLET\tools\python3\python.exe" -m ensurepip --upgrade
+& "$env:VCPKG_INSTALLATION_ROOT\installed\$env:VCPKG_DEFAULT_TRIPLET\tools\python3\python.exe" -m pip install numpy
+
+Write-Host "pip numpy install completed successfully"
@@ -10,7 +10,7 @@ set -e
 # the expected VDB installation

 cmakelists="
-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 project(TestInstall LANGUAGES CXX)
 find_package(OpenVDB REQUIRED COMPONENTS openvdb)
 add_executable(test_vdb_print \"../openvdb_cmd/vdb_print/main.cc\")
@@ -0,0 +1,236 @@
+#!/usr/bin/env python3
+# Copyright Contributors to the OpenVDB Project
+# SPDX-License-Identifier: Apache-2.0
+"""Validate cross-references inside the OpenVDB documentation tree.
+
+Run from the repo root (no arguments) or pass paths to validate explicitly.
+The script scans Markdown (`.md`) and Doxygen narrative (`.txt`) files under
+``doc/`` and checks two classes of cross-reference for staleness:
+
+1. ``#include <path/to/header.h>`` lines that appear inside Markdown
+   ``cpp``/``cuda``/``c``/``c++`` fenced code blocks or inside Doxygen
+   ``@code``/``\\code`` blocks.  Each include is resolved to a real file in
+   ``openvdb/``, ``openvdb_ax/``, ``openvdb_houdini/`` or ``nanovdb/``.
+
+2. ``https://github.com/AcademySoftwareFoundation/openvdb/blob/<ref>/<path>``
+   URLs.  The ``<path>`` portion is checked against the local checkout.
+
+The script exits non-zero on any unresolved reference and prints a
+``file:line`` style error so the failure is actionable in CI logs.  It is
+intentionally dependency-free so it can run in a minimal Python container.
+
+This guards against the class of bug tracked in issue #2161, where renames
+under ``nanovdb/util/ -> nanovdb/tools/`` and ``nanovdb/util/IO.h ->
+nanovdb/io/IO.h`` silently left the docs referencing paths that no longer
+exist.
+"""
+
+from __future__ import annotations
+
+import re
+import sys
+from pathlib import Path
+from typing import Iterable, List, Sequence, Tuple
+
+REPO_ROOT = Path(__file__).resolve().parent.parent
+
+# Map a top-level include namespace to one or more on-disk directories that it
+# resolves against.  ``#include <nanovdb/io/IO.h>`` resolves to
+# ``nanovdb/nanovdb/io/IO.h`` because the repo uses a double-named layout;
+# ``#include <houdini_utils/ParmFactory.h>`` resolves to
+# ``openvdb_houdini/openvdb_houdini/ParmFactory.h`` because the Houdini plugin
+# CMake configuration synthesises a ``houdini_utils/`` subtree at build time
+# (see ``openvdb_houdini/openvdb_houdini/CMakeLists.txt``).
+INCLUDE_NAMESPACE_ROOTS = {
+    "openvdb": ("openvdb/openvdb", "openvdb"),
+    "openvdb_ax": ("openvdb_ax/openvdb_ax", "openvdb_ax"),
+    "openvdb_houdini": ("openvdb_houdini/openvdb_houdini", "openvdb_houdini"),
+    "nanovdb": ("nanovdb/nanovdb", "nanovdb"),
+    "houdini_utils": ("openvdb_houdini/openvdb_houdini",),
+}
+
+# Top-level include path components that come from outside this repo.  When a
+# header's leading component matches one of these, we skip resolution entirely.
+# Houdini SDK headers (``UT/``, ``GA/``, ``GEO_*``, etc.) live in the Houdini
+# install, not in our checkout.
+EXTERNAL_INCLUDE_PREFIXES = (
+    "boost/",
+    "tbb/",
+    "cuda",            # cuda_runtime.h, cuda.h, ...
+    "thrust/",
+    "Eigen/",
+    "OpenEXR/",
+    "Imath/",
+    "openexr/",
+    "GL/",
+    # Houdini SDK prefixes
+    "UT/",
+    "GA/",
+    "GU/",
+    "GEO/",
+    "GT/",
+    "GR/",
+    "OP/",
+    "PRM/",
+    "SOP/",
+    "SYS/",
+    "CH/",
+    "DEP/",
+    "PI/",
+    "RE/",
+    # Standard library headers (no '/' is the usual signal, but list a few
+    # explicitly to be safe).
+    "stdio.h",
+    "iostream",
+    "vector",
+    "string",
+    "memory",
+    "cstdint",
+)
+
+# Match a Markdown fenced code block whose language is one of cpp/cuda/c/c++.
+_MARKDOWN_FENCE_RE = re.compile(
+    r"^```(?:cpp|cuda|c\+\+|c)\s*$(?P<body>.*?)^```\s*$",
+    re.DOTALL | re.MULTILINE,
+)
+
+# Doxygen ``@code``/``\code`` ... ``@endcode``/``\endcode`` blocks.  The
+# language hint (``@code{.cpp}``) is optional.
+_DOXY_CODE_RE = re.compile(
+    r"[@\\]code(?:\{[^}]*\})?\s*$(?P<body>.*?)^[@\\]endcode\s*$",
+    re.DOTALL | re.MULTILINE,
+)
+
+_INCLUDE_RE = re.compile(r"#\s*include\s*<([^>\n]+)>")
+
+# Match URLs pointing into a blob in this repo.  ``<ref>`` may be a branch
+# name or a sha; ``<path>`` is the portion that must exist on disk today.
+_GITHUB_BLOB_RE = re.compile(
+    r"https?://github\.com/AcademySoftwareFoundation/openvdb/blob/"
+    r"(?P<ref>[^/\s]+)/(?P<path>[^\s)\"'#?]+)"
+)
+
+
+def is_external_include(header: str) -> bool:
+    """Return True for headers we cannot resolve from this repo."""
+    head, _, _ = header.partition("/")
+    if head in INCLUDE_NAMESPACE_ROOTS:
+        return False
+    return any(header.startswith(p) for p in EXTERNAL_INCLUDE_PREFIXES) or "/" not in header
+
+
+def resolve_include(header: str) -> Path | None:
+    """Resolve ``#include <header>`` to a path on disk, or None if not found."""
+    head, _, rest = header.partition("/")
+    roots = INCLUDE_NAMESPACE_ROOTS.get(head)
+    if roots is None or not rest:
+        return None
+    for root in roots:
+        candidate = REPO_ROOT / root / rest
+        if candidate.is_file():
+            return candidate
+    return None
+
+
+def iter_code_block_lines(text: str) -> Iterable[Tuple[int, str]]:
+    """Yield (line_number, line_text) for every line inside a code block.
+
+    Line numbers are 1-based and refer to the original ``text``.  Handles both
+    Markdown fences (```cpp ...```) and Doxygen ``@code``/``\\code`` blocks.
+    """
+    for match in _MARKDOWN_FENCE_RE.finditer(text):
+        body_start = match.start("body")
+        yield from _lines_with_offsets(text, match.group("body"), body_start)
+    for match in _DOXY_CODE_RE.finditer(text):
+        body_start = match.start("body")
+        yield from _lines_with_offsets(text, match.group("body"), body_start)
+
+
+def _lines_with_offsets(full_text: str, body: str, body_offset: int) -> Iterable[Tuple[int, str]]:
+    base_line = full_text.count("\n", 0, body_offset) + 1
+    for i, line in enumerate(body.splitlines()):
+        yield base_line + i, line
+
+
+def find_doc_files(extra_paths: Sequence[str]) -> List[Path]:
+    """Return the list of doc files to scan.
+
+    Without arguments, scan ``doc/**/*.md`` and ``doc/**/*.txt``.  With
+    arguments, scan only those files (resolved relative to the repo root).
+    """
+    if extra_paths:
+        return [
+            (Path(p) if Path(p).is_absolute() else REPO_ROOT / p).resolve()
+            for p in extra_paths
+        ]
+    doc_dir = REPO_ROOT / "doc"
+    return sorted(
+        list(doc_dir.rglob("*.md")) + list(doc_dir.rglob("*.txt"))
+    )
+
+
+def check_file(path: Path) -> List[str]:
+    """Return a list of human-readable error strings for the given doc file."""
+    errors: List[str] = []
+    try:
+        text = path.read_text(encoding="utf-8")
+    except (OSError, UnicodeDecodeError) as exc:
+        return [f"{path}: failed to read: {exc}"]
+
+    rel = path.relative_to(REPO_ROOT) if path.is_relative_to(REPO_ROOT) else path
+
+    for line_no, line in iter_code_block_lines(text):
+        for include_match in _INCLUDE_RE.finditer(line):
+            header = include_match.group(1).strip()
+            if is_external_include(header):
+                continue
+            if resolve_include(header) is None:
+                errors.append(
+                    f"{rel}:{line_no}: include <{header}> does not resolve "
+                    "to a file in this checkout"
+                )
+
+    for url_match in _GITHUB_BLOB_RE.finditer(text):
+        ref = url_match.group("ref")
+        url_path = url_match.group("path")
+        on_disk = REPO_ROOT / url_path
+        if not on_disk.exists():
+            # Compute the 1-based line number of the URL for the error report.
+            line_no = text.count("\n", 0, url_match.start()) + 1
+            errors.append(
+                f"{rel}:{line_no}: github URL "
+                f"AcademySoftwareFoundation/openvdb/blob/{ref}/{url_path} "
+                "does not exist in this checkout"
+            )
+
+    return errors
+
+
+def main(argv: Sequence[str]) -> int:
+    files = find_doc_files(argv[1:])
+    total_errors: List[str] = []
+    for path in files:
+        if not path.exists():
+            total_errors.append(f"{path}: file does not exist")
+            continue
+        total_errors.extend(check_file(path))
+
+    if total_errors:
+        print("Documentation reference validation FAILED.", file=sys.stderr)
+        print("", file=sys.stderr)
+        for err in total_errors:
+            print(err, file=sys.stderr)
+        print("", file=sys.stderr)
+        print(
+            f"{len(total_errors)} broken reference(s) across "
+            f"{len(files)} file(s).",
+            file=sys.stderr,
+        )
+        return 1
+
+    print(f"OK: {len(files)} doc file(s) validated, no broken references.")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv))
@@ -0,0 +1,24 @@
+# SPDX-License-Identifier: MIT
+#
+# SPDX-FileCopyrightText: Copyright (c) 2019-2023 Lars Melchior and contributors
+
+set(CPM_DOWNLOAD_VERSION 0.40.2)
+set(CPM_HASH_SUM "c8cdc32c03816538ce22781ed72964dc864b2a34a310d3b7104812a5ca2d835d")
+
+if(CPM_SOURCE_CACHE)
+  set(CPM_DOWNLOAD_LOCATION "${CPM_SOURCE_CACHE}/cpm/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+elseif(DEFINED ENV{CPM_SOURCE_CACHE})
+  set(CPM_DOWNLOAD_LOCATION "$ENV{CPM_SOURCE_CACHE}/cpm/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+else()
+  set(CPM_DOWNLOAD_LOCATION "${CMAKE_BINARY_DIR}/cmake/CPM_${CPM_DOWNLOAD_VERSION}.cmake")
+endif()
+
+# Expand relative path. This is important if the provided path contains a tilde (~)
+get_filename_component(CPM_DOWNLOAD_LOCATION ${CPM_DOWNLOAD_LOCATION} ABSOLUTE)
+
+file(DOWNLOAD
+     https://github.com/cpm-cmake/CPM.cmake/releases/download/v${CPM_DOWNLOAD_VERSION}/CPM.cmake
+     ${CPM_DOWNLOAD_LOCATION} EXPECTED_HASH SHA256=${CPM_HASH_SUM}
+)
+
+include(${CPM_DOWNLOAD_LOCATION})
@@ -88,7 +88,7 @@ may be provided to tell this module where to look.

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 include(GNUInstallDirs)

 mark_as_advanced(
@@ -419,9 +419,11 @@ if(NOT TARGET Blosc::blosc)
        set(${BLOSC_EXTERNAL_LIB}_LIBRARIES       ${${BLOSC_EXTERNAL_LIB}_LIBRARY_DEBUG})
      endif()

-      target_link_libraries(Blosc::blosc INTERFACE
-        $<$<CONFIG:Release>:${${BLOSC_EXTERNAL_LIB}_LIBRARY_RELEASE}>
-        $<$<CONFIG:Debug>:${${BLOSC_EXTERNAL_LIB}_LIBRARY_DEBUG}>)
+      if(${BLOSC_EXTERNAL_LIB}_LIBRARY_RELEASE OR ${BLOSC_EXTERNAL_LIB}_LIBRARY_DEBUG)
+        target_link_libraries(Blosc::blosc INTERFACE
+          $<$<CONFIG:Release>:${${BLOSC_EXTERNAL_LIB}_LIBRARY_RELEASE}>
+          $<$<CONFIG:Debug>:${${BLOSC_EXTERNAL_LIB}_LIBRARY_DEBUG}>)
+      endif()

    endforeach()
  endif()
@@ -1,320 +0,0 @@
-# Copyright Contributors to the OpenVDB Project
-# SPDX-License-Identifier: Apache-2.0
-#
-#[=======================================================================[.rst:
-
-FindCppUnit
-----------
-
-Find CppUnit include dirs and libraries
-
-Use this module by invoking find_package with the form::
-
-  find_package(CppUnit
-    [version] [EXACT]      # Minimum or EXACT version
-    [REQUIRED]             # Fail with error if CppUnit is not found
-    )
-
-
-IMPORTED Targets
-^^^^^^^^^^^^^^^^
-
-``CppUnit::cppunit``
-  This module defines IMPORTED target CppUnit::cppunit, if CppUnit has been
-  found.
-
-Result Variables
-^^^^^^^^^^^^^^^^
-
-This will define the following variables:
-
-``CppUnit_FOUND``
-  True if the system has the CppUnit library.
-``CppUnit_VERSION``
-  The version of the CppUnit library which was found.
-``CppUnit_INCLUDE_DIRS``
-  Include directories needed to use CppUnit.
-``CppUnit_RELEASE_LIBRARIES``
-  Libraries needed to link to the release version of CppUnit.
-``CppUnit_RELEASE_LIBRARY_DIRS``
-  CppUnit release library directories.
-``CppUnit_DEBUG_LIBRARIES``
-  Libraries needed to link to the debug version of CppUnit.
-``CppUnit_DEBUG_LIBRARY_DIRS``
-  CppUnit debug library directories.
-
-Deprecated - use [RELEASE|DEBUG] variants:
-
-``CppUnit_LIBRARIES``
-  Libraries needed to link to CppUnit.
-``CppUnit_LIBRARY_DIRS``
-  CppUnit library directories.
-
-Cache Variables
-^^^^^^^^^^^^^^^
-
-The following cache variables may also be set:
-
-``CppUnit_INCLUDE_DIR``
-  The directory containing ``cppunit/Portability.h``.
-``CppUnit_LIBRARY``
-  The path to the CppUnit library.
-
-Hints
-^^^^^
-
-Instead of explicitly setting the cache variables, the following variables
-may be provided to tell this module where to look.
-
-``CppUnit_ROOT``
-  Preferred installation prefix.
-``CPPUNIT_INCLUDEDIR``
-  Preferred include directory e.g. <prefix>/include
-``CPPUNIT_LIBRARYDIR``
-  Preferred library directory e.g. <prefix>/lib
-``CPPUNIT_DEBUG_SUFFIX``
-  Suffix of the debug version of CppUnit libs. Defaults to "_d".
-``SYSTEM_LIBRARY_PATHS``
-  Global list of library paths intended to be searched by and find_xxx call
-``CPPUNIT_USE_STATIC_LIBS``
-  Only search for static cppunit libraries
-``DISABLE_CMAKE_SEARCH_PATHS``
-  Disable CMakes default search paths for find_xxx calls in this module
-
-#]=======================================================================]
-
-cmake_minimum_required(VERSION 3.20)
-include(GNUInstallDirs)
-
-
-mark_as_advanced(
-  CppUnit_INCLUDE_DIR
-  CppUnit_LIBRARY
-)
-
-set(_FIND_CPPUNIT_ADDITIONAL_OPTIONS "")
-if(DISABLE_CMAKE_SEARCH_PATHS)
-  set(_FIND_CPPUNIT_ADDITIONAL_OPTIONS NO_DEFAULT_PATH)
-endif()
-
-# Set _CPPUNIT_ROOT based on a user provided root var. Xxx_ROOT and ENV{Xxx_ROOT}
-# are prioritised over the legacy capitalized XXX_ROOT variables for matching
-# CMake 3.12 behaviour
-# @todo  deprecate -D and ENV CPPUNIT_ROOT from CMake 3.12
-if(CppUnit_ROOT)
-  set(_CPPUNIT_ROOT ${CppUnit_ROOT})
-elseif(DEFINED ENV{CppUnit_ROOT})
-  set(_CPPUNIT_ROOT $ENV{CppUnit_ROOT})
-elseif(CPPUNIT_ROOT)
-  set(_CPPUNIT_ROOT ${CPPUNIT_ROOT})
-elseif(DEFINED ENV{CPPUNIT_ROOT})
-  set(_CPPUNIT_ROOT $ENV{CPPUNIT_ROOT})
-endif()
-
-# Additionally try and use pkconfig to find cppunit
-if(USE_PKGCONFIG)
-  if(NOT DEFINED PKG_CONFIG_FOUND)
-    find_package(PkgConfig)
-  endif()
-  pkg_check_modules(PC_CppUnit QUIET cppunit)
-endif()
-
-# ------------------------------------------------------------------------
-#  Search for CppUnit include DIR
-# ------------------------------------------------------------------------
-
-set(_CPPUNIT_INCLUDE_SEARCH_DIRS "")
-list(APPEND _CPPUNIT_INCLUDE_SEARCH_DIRS
-  ${CPPUNIT_INCLUDEDIR}
-  ${_CPPUNIT_ROOT}
-  ${PC_CppUnit_INCLUDEDIR}
-  ${SYSTEM_LIBRARY_PATHS}
-)
-
-# Look for a standard cppunit header file.
-find_path(CppUnit_INCLUDE_DIR cppunit/Portability.h
-  ${_FIND_CPPUNIT_ADDITIONAL_OPTIONS}
-  PATHS ${_CPPUNIT_INCLUDE_SEARCH_DIRS}
-  PATH_SUFFIXES ${CMAKE_INSTALL_INCLUDEDIR} include
-)
-
-if(EXISTS "${CppUnit_INCLUDE_DIR}/cppunit/Portability.h")
-  file(STRINGS "${CppUnit_INCLUDE_DIR}/cppunit/Portability.h"
-    _cppunit_version_string REGEX "#define CPPUNIT_VERSION "
-  )
-  string(REGEX REPLACE "#define CPPUNIT_VERSION +\"(.+)\".*$" "\\1"
-    _cppunit_version_string "${_cppunit_version_string}"
-  )
-  string(STRIP "${_cppunit_version_string}" CppUnit_VERSION)
-  unset(_cppunit_version_string )
-endif()
-
-# ------------------------------------------------------------------------
-#  Search for CppUnit lib DIR
-# ------------------------------------------------------------------------
-
-if(NOT DEFINED CPPUNIT_DEBUG_SUFFIX)
-  set(CPPUNIT_DEBUG_SUFFIX "d")
-endif()
-
-set(_CPPUNIT_LIBRARYDIR_SEARCH_DIRS "")
-list(APPEND _CPPUNIT_LIBRARYDIR_SEARCH_DIRS
-  ${CPPUNIT_LIBRARYDIR}
-  ${_CPPUNIT_ROOT}
-  ${PC_CppUnit_LIBDIR}
-  ${SYSTEM_LIBRARY_PATHS}
-)
-
-# Library suffix handling
-
-set(_CPPUNIT_ORIG_CMAKE_FIND_LIBRARY_SUFFIXES ${CMAKE_FIND_LIBRARY_SUFFIXES})
-
-if(MSVC)
-  if(CPPUNIT_USE_STATIC_LIBS)
-    set(CMAKE_FIND_LIBRARY_SUFFIXES ".lib")
-  else()
-    list(APPEND CMAKE_FIND_LIBRARY_SUFFIXES "_dll.lib")
-  endif()
-else()
-  if(CPPUNIT_USE_STATIC_LIBS)
-    set(CMAKE_FIND_LIBRARY_SUFFIXES ".a")
-  endif()
-endif()
-
-list(APPEND CPPUNIT_BUILD_TYPES RELEASE DEBUG)
-
-foreach(BUILD_TYPE ${CPPUNIT_BUILD_TYPES})
-  set(_CPPUNIT_LIB_NAME cppunit)
-  if(BUILD_TYPE STREQUAL DEBUG)
-    set(_CPPUNIT_LIB_NAME "${_CPPUNIT_LIB_NAME}${CPPUNIT_DEBUG_SUFFIX}")
-  endif()
-
-  # Find the lib
-  find_library(CppUnit_LIBRARY_${BUILD_TYPE} ${_CPPUNIT_LIB_NAME}
-    ${_FIND_CPPUNIT_ADDITIONAL_OPTIONS}
-    PATHS ${_CPPUNIT_LIBRARYDIR_SEARCH_DIRS}
-    PATH_SUFFIXES ${CMAKE_INSTALL_LIBDIR} lib64 lib
-  )
-endforeach()
-
-# Reset library suffix
-
-set(CMAKE_FIND_LIBRARY_SUFFIXES ${_CPPUNIT_ORIG_CMAKE_FIND_LIBRARY_SUFFIXES})
-unset(_CPPUNIT_ORIG_CMAKE_FIND_LIBRARY_SUFFIXES)
-
-if(CppUnit_LIBRARY_DEBUG AND CppUnit_LIBRARY_RELEASE)
-  # if the generator is multi-config or if CMAKE_BUILD_TYPE is set for
-  # single-config generators, set optimized and debug libraries
-  get_property(_isMultiConfig GLOBAL PROPERTY GENERATOR_IS_MULTI_CONFIG)
-  if(_isMultiConfig OR CMAKE_BUILD_TYPE)
-    set(CppUnit_LIBRARY optimized ${CppUnit_LIBRARY_RELEASE} debug ${CppUnit_LIBRARY_DEBUG})
-  else()
-    # For single-config generators where CMAKE_BUILD_TYPE has no value,
-    # just use the release libraries
-    set(CppUnit_LIBRARY ${CppUnit_LIBRARY_RELEASE})
-  endif()
-  # FIXME: This probably should be set for both cases
-  set(CppUnit_LIBRARIES optimized ${CppUnit_LIBRARY_RELEASE} debug ${CppUnit_LIBRARY_DEBUG})
-endif()
-
-# if only the release version was found, set the debug variable also to the release version
-if(CppUnit_LIBRARY_RELEASE AND NOT CppUnit_LIBRARY_DEBUG)
-  set(CppUnit_LIBRARY_DEBUG ${CppUnit_LIBRARY_RELEASE})
-  set(CppUnit_LIBRARY       ${CppUnit_LIBRARY_RELEASE})
-  set(CppUnit_LIBRARIES     ${CppUnit_LIBRARY_RELEASE})
-endif()
-
-# if only the debug version was found, set the release variable also to the debug version
-if(CppUnit_LIBRARY_DEBUG AND NOT CppUnit_LIBRARY_RELEASE)
-  set(CppUnit_LIBRARY_RELEASE ${CppUnit_LIBRARY_DEBUG})
-  set(CppUnit_LIBRARY         ${CppUnit_LIBRARY_DEBUG})
-  set(CppUnit_LIBRARIES       ${CppUnit_LIBRARY_DEBUG})
-endif()
-
-# If the debug & release library ends up being the same, omit the keywords
-if("${CppUnit_LIBRARY_RELEASE}" STREQUAL "${CppUnit_LIBRARY_DEBUG}")
-  set(CppUnit_LIBRARY   ${CppUnit_LIBRARY_RELEASE} )
-  set(CppUnit_LIBRARIES ${CppUnit_LIBRARY_RELEASE} )
-endif()
-
-if(CppUnit_LIBRARY)
-  set(CppUnit_FOUND TRUE)
-else()
-  set(CppUnit_FOUND FALSE)
-endif()
-
-# ------------------------------------------------------------------------
-#  Cache and set CppUnit_FOUND
-# ------------------------------------------------------------------------
-
-include(FindPackageHandleStandardArgs)
-find_package_handle_standard_args(CppUnit
-  FOUND_VAR CppUnit_FOUND
-  REQUIRED_VARS
-    CppUnit_LIBRARY
-    CppUnit_INCLUDE_DIR
-  VERSION_VAR CppUnit_VERSION
-)
-
-if(NOT CppUnit_FOUND)
-  if(CppUnit_FIND_REQUIRED)
-    message(FATAL_ERROR "Unable to find CppUnit")
-  endif()
-  return()
-endif()
-
-# Partition release/debug lib vars
-
-set(CppUnit_RELEASE_LIBRARIES ${CppUnit_LIBRARY_RELEASE})
-get_filename_component(CppUnit_RELEASE_LIBRARY_DIRS ${CppUnit_RELEASE_LIBRARIES} DIRECTORY)
-set(CppUnit_DEBUG_LIBRARIES ${CppUnit_LIBRARY_DEBUG})
-get_filename_component(CppUnit_DEBUG_LIBRARY_DIRS ${CppUnit_DEBUG_LIBRARIES} DIRECTORY)
-set(CppUnit_LIBRARIES ${CppUnit_RELEASE_LIBRARIES})
-set(CppUnit_LIBRARY_DIRS ${CppUnit_RELEASE_LIBRARY_DIRS})
-set(CppUnit_INCLUDE_DIRS ${CppUnit_INCLUDE_DIR})
-
-# Configure lib type. If XXX_USE_STATIC_LIBS, we always assume a static
-# lib is in use. If win32, we can't mark the import .libs as shared, so
-# these are always marked as UNKNOWN. Otherwise, infer from extension.
-set(CPPUNIT_LIB_TYPE UNKNOWN)
-if(CPPUNIT_USE_STATIC_LIBS)
-  set(CPPUNIT_LIB_TYPE STATIC)
-elseif(UNIX)
-  get_filename_component(_CPPUNIT_EXT ${CppUnit_LIBRARY_RELEASE} EXT)
-  if(_CPPUNIT_EXT STREQUAL ".a")
-    set(CPPUNIT_LIB_TYPE STATIC)
-  elseif(_CPPUNIT_EXT STREQUAL ".so" OR
-         _CPPUNIT_EXT STREQUAL ".dylib")
-    set(CPPUNIT_LIB_TYPE SHARED)
-  endif()
-endif()
-
-if(NOT TARGET CppUnit::cppunit)
-  add_library(CppUnit::cppunit ${CPPUNIT_LIB_TYPE} IMPORTED)
-  set_target_properties(CppUnit::cppunit PROPERTIES
-    INTERFACE_COMPILE_OPTIONS "${PC_CppUnit_CFLAGS_OTHER}"
-    INTERFACE_INCLUDE_DIRECTORIES "${CppUnit_INCLUDE_DIRS}")
-
-  # Standard location
-  set_target_properties(CppUnit::cppunit PROPERTIES
-    IMPORTED_LINK_INTERFACE_LANGUAGES "CXX"
-    IMPORTED_LOCATION "${CppUnit_LIBRARY}")
-
-  # Release location
-  if(EXISTS "${CppUnit_LIBRARY_RELEASE}")
-    set_property(TARGET CppUnit::cppunit APPEND PROPERTY
-      IMPORTED_CONFIGURATIONS RELEASE)
-    set_target_properties(CppUnit::cppunit PROPERTIES
-      IMPORTED_LINK_INTERFACE_LANGUAGES_RELEASE "CXX"
-      IMPORTED_LOCATION_RELEASE "${CppUnit_LIBRARY_RELEASE}")
-  endif()
-
-  # Debug location
-  if(EXISTS "${CppUnit_LIBRARY_DEBUG}")
-    set_property(TARGET CppUnit::cppunit APPEND PROPERTY
-      IMPORTED_CONFIGURATIONS DEBUG)
-    set_target_properties(CppUnit::cppunit PROPERTIES
-      IMPORTED_LINK_INTERFACE_LANGUAGES_DEBUG "CXX"
-      IMPORTED_LOCATION_DEBUG "${CppUnit_LIBRARY_DEBUG}")
-  endif()
-endif()
@@ -64,7 +64,7 @@ may be provided to tell this module where to look.

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 include(GNUInstallDirs)


@@ -87,7 +87,7 @@ may be provided to tell this module where to look.

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 include(GNUInstallDirs)


@@ -0,0 +1,44 @@
+include (FindPackageHandleStandardArgs)
+
+if(NCCL_ROOT)
+  set(_NCCL_ROOT ${NCCL_ROOT})
+elseif(DEFINED ENV{NCCL_ROOT})
+  set(_NCCL_ROOT $ENV{NCCL_ROOT})
+endif()
+
+set(_NCCL_INCLUDE_SEARCH_DIRS "")
+list(APPEND _NCCL_INCLUDE_SEARCH_DIRS
+  ${_NCCL_ROOT}/include
+)
+find_path(NCCL_INCLUDE_DIR
+    NAMES nccl.h
+    PATHS ${_NCCL_INCLUDE_SEARCH_DIRS}
+)
+
+set(_NCCL_LIBRARY_SEARCH_DIRS "")
+list(APPEND _NCCL_LIBRARY_SEARCH_DIRS
+  ${_NCCL_ROOT}/bin
+)
+find_library(NCCL_LIBRARY
+    NAMES nccl
+    PATHS ${_NCCL_LIBRARY_SEARCH_DIRS}
+)
+
+find_package_handle_standard_args(NCCL REQUIRED_VARS NCCL_INCLUDE_DIR NCCL_LIBRARY)
+
+if (NCCL_FOUND)
+    mark_as_advanced(NCCL_INCLUDE_DIR)
+    mark_as_advanced(NCCL_LIBRARY)
+endif()
+
+if (NCCL_FOUND AND NOT TARGET NCCL::NCCL)
+    add_library(NCCL::NCCL SHARED IMPORTED)
+    if(MSVC)
+      string(REPLACE ".lib" ".dll" NCCL_DLL ${NCCL_LIBRARY})
+      set_property(TARGET NCCL::NCCL PROPERTY IMPORTED_LOCATION ${NCCL_DLL})
+      set_property(TARGET NCCL::NCCL PROPERTY IMPORTED_IMPLIB ${NCCL_LIBRARY})
+    else()
+      set_property(TARGET NCCL::NCCL PROPERTY IMPORTED_LOCATION ${NCCL_LIBRARY})
+    endif()
+    target_include_directories(NCCL::NCCL INTERFACE ${NCCL_INCLUDE_DIR})
+endif()
@@ -93,7 +93,7 @@ may be provided to tell this module where to look.

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 include(GNUInstallDirs)


@@ -109,7 +109,7 @@ may be provided to tell this module where to look.

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 include(GNUInstallDirs)


@@ -93,7 +93,7 @@ may be provided to tell this module where to look.

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 include(GNUInstallDirs)


@@ -47,7 +47,7 @@ The following variables may be provided to tell this module where to look.
 # Find the glfw3 installation and use glfw's CMake to initialize
 # the glfw lib

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)

 set(_FIND_GLFW3_ADDITIONAL_OPTIONS "")
 if(DISABLE_CMAKE_SEARCH_PATHS)
@@ -73,7 +73,7 @@ may be provided to tell this module where to look.
 # Find the Houdini installation and use Houdini's CMake to initialize
 # the Houdini lib

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)

 # Include utility functions for version information
 include(${CMAKE_CURRENT_LIST_DIR}/OpenVDBUtils.cmake)
@@ -51,7 +51,7 @@ variables may be provided to tell this module where to look.
 # Find the Maya installation and use Maya's CMake to initialize
 # the Maya lib

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)


 set(_FIND_MAYA_ADDITIONAL_OPTIONS "")
@@ -57,7 +57,7 @@ The following functions are provided:

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)


 function(OPENVDB_GET_VERSION_DEFINE HEADER KEY VALUE)
@@ -83,8 +83,8 @@ function(OPENVDB_GET_VERSION_DEFINE HEADER KEY VALUE)
 endfunction()


-########################################################################
-########################################################################
+###############################################################################
+###############################################################################


 function(OPENVDB_VERSION_FROM_HEADER OPENVDB_VERSION_FILE)
@@ -120,8 +120,8 @@ function(OPENVDB_VERSION_FROM_HEADER OPENVDB_VERSION_FILE)
 endfunction()


-########################################################################
-########################################################################
+###############################################################################
+###############################################################################


 function(OPENVDB_ABI_VERSION_FROM_PRINT OPENVDB_PRINT)
@@ -162,3 +162,115 @@ function(OPENVDB_ABI_VERSION_FROM_PRINT OPENVDB_PRINT)
    set(${_VDB_ABI} ${_OpenVDB_ABI} PARENT_SCOPE)
  endif()
 endfunction()
+
+###############################################################################
+## @brief  Compute the enumerated ISA value for x86/x86_64 architectures that
+##   VCL will use during compilation of OpenVDB. The value is stored in the
+##   VCL_INSTRSET variable. Possible values are:
+##     0:  Unknown or error
+##     2:  SSE2
+##     3:  SSE3
+##     4:  SSSE3
+##     5:  SSE4.1
+##     6:  SSE4.2
+##     7:  AVX
+##     8:  AVX2
+##     9:  AVX512F
+##     10: AVX512BW/DQ/VL
+function(OPENVDB_COMPUTE_X86_INSTRSET_FROM_VCL VCL_INSTRSET)
+  # Append project specific flags
+  set(OLD_CMAKE_REQUIRED_FLAGS ${CMAKE_REQUIRED_FLAGS})
+  if(OPENVDB_SIMD STREQUAL "AVX")
+    set(CMAKE_REQUIRED_FLAGS "-mavx;-msse4.2")
+  elseif(OPENVDB_SIMD STREQUAL "SSE42")
+    set(CMAKE_REQUIRED_FLAGS "-msse4.2")
+  endif()
+
+  # Figure out that the INSTRSET macro would expand to.
+  # @note would be much faster with try_run but this way doesn't involve
+  #   invoking bad instruction sets on unsupported ISAs on the build machine.
+  include(CheckCXXSourceCompiles)
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 10); return 0; } ]] COMPUTE_VCL_INSTRSET__AVX512VL_DQ_BW__)
+  if(COMPUTE_VCL_INSTRSET__AVX512VL_DQ_BW__)
+    set(${VCL_INSTRSET} 10 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 9); return 0; } ]] COMPUTE_VCL_INSTRSET__AVX512F__)
+  if(COMPUTE_VCL_INSTRSET__AVX512F__)
+    set(${VCL_INSTRSET} 9 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 8); return 0; } ]] COMPUTE_VCL_INSTRSET__AVX2__)
+  if(COMPUTE_VCL_INSTRSET__AVX2__)
+    set(${VCL_INSTRSET} 8 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 7); return 0; } ]] COMPUTE_VCL_INSTRSET__AVX__)
+  if(COMPUTE_VCL_INSTRSET__AVX__)
+    set(${VCL_INSTRSET} 7 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 6); return 0; } ]] COMPUTE_VCL_INSTRSET__SSE4_2__)
+  if(COMPUTE_VCL_INSTRSET__SSE4_2__)
+    set(${VCL_INSTRSET} 6 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 5); return 0; } ]] COMPUTE_VCL_INSTRSET__SSE4_1__)
+  if(COMPUTE_VCL_INSTRSET__SSE4_1__)
+    set(${VCL_INSTRSET} 5 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 4); return 0; } ]] COMPUTE_VCL_INSTRSET__SSSE3__)
+  if(COMPUTE_VCL_INSTRSET__SSSE3__)
+    set(${VCL_INSTRSET} 4 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 3); return 0; } ]] COMPUTE_VCL_INSTRSET__SSE3__)
+  if(COMPUTE_VCL_INSTRSET__SSE3__)
+    set(${VCL_INSTRSET} 3 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 2); return 0; } ]] COMPUTE_VCL_INSTRSET__SSE2__)
+  if(COMPUTE_VCL_INSTRSET__SSE2__)
+    set(${VCL_INSTRSET} 2 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  check_cxx_source_compiles([[
+    #include "${CMAKE_SOURCE_DIR}/ext/vcl/openvdb/ext/vcl/instrset.h"
+    int main() { static_assert(INSTRSET == 1); return 0; } ]] COMPUTE_VCL_INSTRSET__SSE__)
+  if(COMPUTE_VCL_INSTRSET__SSE__)
+    set(${VCL_INSTRSET} 1 PARENT_SCOPE)
+    set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+    return()
+  endif()
+  # Unknown or error
+  set(${VCL_INSTRSET} 0 PARENT_SCOPE)
+  set(CMAKE_REQUIRED_FLAGS ${OLD_CMAKE_REQUIRED_FLAGS})
+endfunction()
@@ -16,7 +16,7 @@ existing from a previous run of cmake.

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)

 set(MANIFEST "${CMAKE_CURRENT_BINARY_DIR}/install_manifest.txt")

@@ -7,7 +7,7 @@

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)

 ###############################################################################

@@ -127,6 +127,47 @@ endif()

 ###############################################################################

+# x86 ISA selection
+
+if(CMAKE_SYSTEM_PROCESSOR MATCHES "(x86)|(X86)|(amd64)|(AMD64)")
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 1)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-msse>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 2)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-msse2>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 3)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-msse3>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 4)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-mssse3>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 5)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-msse4.1>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 6)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-msse4.2>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 7)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-mavx>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 8)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-mavx2>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 9)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-mavx512f>")
+  endif()
+  if(OPENVDB_X86_INSTRSET GREATER_EQUAL 10)
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-mavx512bw>")
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-mavx512dq>")
+    add_compile_options("$<$<COMPILE_LANGUAGE:CXX>:-mavx512vl>")
+  endif()
+elseif(OPENVDB_X86_INSTRSET GREATER 0)
+  message(WARNING "OPENVDB_X86_INSTRSET has been set but has no effect when building for ${CMAKE_SYSTEM_PROCESSOR}.")
+endif()
+
+###############################################################################
+
 # Increase the number of sections that an object file can contain
 add_compile_options("$<$<COMPILE_LANG_AND_ID:CXX,MSVC>:/bigobj>")
 # Enable MSVC options that make it behave like other compilers
@@ -8,7 +8,7 @@

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)

 ###############################################################################

@@ -27,7 +27,9 @@ set(OPENVDB_DEPRECATED_ABI_LIST ${MINIMUM_OPENVDB_ABI_VERSION} ${_PREV_ABI})
 unset(_PREV_ABI)

 if(NOT DISABLE_DEPENDENCY_VERSION_CHECKS)
-  # @note  Currently tracking CY2022 of the VFX platform where available
+  # @note  Currently tracking CY2024 of the VFX platform where available
+  # (that is, our minimum requirements are for the current VFX platform)
+  # minus 2 years to be able to build with legacy ABI)

  # @Note  Compiler versions are not really a hard and fast rule, you just need
  # a compiler with complete support for our MINIMUM_CXX_STANDARD (currently 17).
@@ -39,32 +41,27 @@ if(NOT DISABLE_DEPENDENCY_VERSION_CHECKS)
  set(MINIMUM_ICC_VERSION 19)
  set(MINIMUM_MSVC_VERSION 19.30) # 1928 (Visual Studio 2019 Version 16.8 + 16.9)

-  set(MINIMUM_BOOST_VERSION 1.80)
-  # Nanobind does not store the version in the CMake config file in 2.0.0.
-  # This issue was fixed in 2.1.0 so next time we bump the minimum version,
-  # we can define this.
-  #set(MINIMUM_NANOBIND_VERSION 2.0.0)
-  set(MINIMUM_IMATH_VERSION 3.1)
-  set(MINIMUM_OPENEXR_VERSION 3.1)
+  set(MINIMUM_BOOST_VERSION 1.82)
+  set(MINIMUM_OPENEXR_VERSION 3.2)
+  set(MINIMUM_IMATH_VERSION 3.2)
  set(MINIMUM_ZLIB_VERSION 1.2.7)
  set(MINIMUM_TBB_VERSION 2020.3)
-  set(MINIMUM_LLVM_VERSION 13.0.0)
+  set(MINIMUM_LLVM_VERSION 15.0.0)
  set(MINIMUM_BLOSC_VERSION 1.17.0)
-  set(MINIMUM_GLFW_VERSION 3.1)
+  set(MINIMUM_GLFW_VERSION 3.3)

-  set(MINIMUM_PYTHON_VERSION 3.10)
-  set(MINIMUM_NUMPY_VERSION 1.23.0)
+  set(MINIMUM_PYTHON_VERSION 3.11)

  set(MINIMUM_GOOGLETEST_VERSION 1.10)
-  set(MINIMUM_LOG4CPLUS_VERSION 1.1.2)
-  set(MINIMUM_HOUDINI_VERSION 19.5)
+  set(MINIMUM_LOG4CPLUS_VERSION 2.0)
+  set(MINIMUM_HOUDINI_VERSION 20.5)

  # These always promote warnings rather than errors
  set(MINIMUM_MAYA_VERSION 2017)
  set(MINIMUM_DOXYGEN_VERSION 1.8.8)
 endif()

-# VFX 23 deprecations to transition to MINIMUM_* variables in OpenVDB 12.0.0
+# VFX 24 deprecations to transition to MINIMUM_* variables in OpenVDB 14.0.0
 # @note  At the time of writing, any variables that are commented out don't
 #   have target transitional versions.

@@ -73,16 +70,16 @@ endif()
 # set(FUTURE_MINIMUM_ICC_VERSION 19)

 # set(FUTURE_MINIMUM_CXX_STANDARD 20)
-set(FUTURE_MINIMUM_CMAKE_VERSION 3.24)
-set(FUTURE_MINIMUM_OPENEXR_VERSION 3.2)
-set(FUTURE_MINIMUM_BOOST_VERSION 1.82)
-set(FUTURE_MINIMUM_GLFW_VERSION 3.3)
-set(FUTURE_MINIMUM_LOG4CPLUS_VERSION 2.0)
-# set(FUTURE_MINIMUM_NANOBIND_VERSION 2.1.0)
+set(FUTURE_MINIMUM_CMAKE_VERSION 4.0)
+set(FUTURE_MINIMUM_OPENEXR_VERSION 3.3)
+set(FUTURE_MINIMUM_IMATH_VERSION 3.3)
+set(FUTURE_MINIMUM_TBB_VERSION 2021)
+set(FUTURE_MINIMUM_BOOST_VERSION 1.85)
+#set(FUTURE_MINIMUM_GLFW_VERSION 3.3)
+#set(FUTURE_MINIMUM_LOG4CPLUS_VERSION 2.0)
+set(FUTURE_MINIMUM_NANOBIND_VERSION 2.5.0)

 # set(FUTURE_MINIMUM_BLOSC_VERSION 1.17.0)
-# set(FUTURE_MINIMUM_TBB_VERSION 2020.3)
-set(FUTURE_MINIMUM_PYTHON_VERSION 3.11)
-set(FUTURE_MINIMUM_NUMPY_VERSION 1.26.0)
+# set(FUTURE_MINIMUM_PYTHON_VERSION 3.11)
 # set(FUTURE_MINIMUM_HOUDINI_VERSION 20.0)
-set(FUTURE_MINIMUM_LLVM_VERSION 15.0.0)
+set(FUTURE_MINIMUM_LLVM_VERSION 18.0.0)
@@ -7,7 +7,7 @@

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 project(OpenVDBDocs LANGUAGES NONE)

 include(GNUInstallDirs)
@@ -25,8 +25,16 @@ option(OPENVDB_DOXYGEN_NANOVDB "Build NanoVDB documentation" ON)
 option(OPENVDB_DOXYGEN_INTERNAL [=[
 Enable documentation of methods and classes which have been explicitly marked as internal
 ]=] OFF)
+# When enabled, Doxygen's WARN_AS_ERROR=FAIL_ON_WARNINGS makes any warning fail
+# the docs build (it still generates the docs and writes the full warning log
+# first, then exits non-zero). CI turns this ON -- see .github/workflows/docs.yml
+# -- so net-new doc warnings fail the PR that introduces them. Defaults OFF so a
+# local doc build isn't blocked by a stray warning from a different Doxygen
+# version; pass -DOPENVDB_DOXYGEN_WARN_AS_ERROR=ON to gate locally too.
+option(OPENVDB_DOXYGEN_WARN_AS_ERROR
+  "Treat Doxygen warnings as errors" OFF)

-find_package(Doxygen REQUIRED doxygen)
+find_package(Doxygen 1.0...<1.15.0 REQUIRED doxygen)
 if(MINIMUM_DOXYGEN_VERSION)
  if(DOXYGEN_VERSION VERSION_LESS MINIMUM_DOXYGEN_VERSION)
    message(WARNING "The doxygen-config doxyfile has been generated with version "
@@ -41,12 +49,11 @@ set(DOXY_FILES
  openvdb/openvdb/io
  openvdb/openvdb/math
  openvdb/openvdb/points
-  openvdb/openvdb/points/impl
-  openvdb/openvdb/python/pyopenvdb.h
+  openvdb/openvdb/thread
  openvdb/openvdb/tools
-  openvdb/openvdb/tools/impl
  openvdb/openvdb/tree
  openvdb/openvdb/util
+  ext/imath/openvdb/math
  doc/build.txt
  doc/changes.txt
  doc/codingstyle.txt
@@ -58,28 +65,93 @@ set(DOXY_FILES
  doc/points.txt
  doc/python.txt)

+if(OPENVDB_DOXYGEN_INTERNAL)
+  list(APPEND
+    openvdb/openvdb/points/impl
+    openvdb/openvdb/tools/impl)
+endif()
+
+if(NOT OPENVDB_DOXYGEN_INTERNAL)
+  # Hide implementation-detail symbols in the lowercase `internal` namespaces
+  # (the project convention for non-public helpers). Note: do NOT add a
+  # capital-I "*Internal*" glob here -- it also matches the public
+  # tree::InternalNode (and nanovdb::InternalNode) class, silently dropping it
+  # from the docs and breaking every \link to it (e.g. InternalNode::childCount
+  # in doc/changes.txt) while its RootNode/LeafNode siblings resolve fine.
+  set(DOXYGEN_EXCLUDE_SYMBOLS
+    "*internal*")
+endif()
+
 set(DOXYGEN_PROJECT_NAME "OpenVDB")
+set(DOXYGEN_PROJECT_LOGO "${CMAKE_CURRENT_SOURCE_DIR}/img/openvdb_logo.png")
 set(DOXYGEN_PROJECT_NUMBER "${OpenVDB_VERSION}")
 set(DOXYGEN_PROJECT_BRIEF "")
 set(DOXYGEN_FILE_PATTERNS *.h *h.in) # headers only
 set(DOXYGEN_EXTENSION_MAPPING .in=C) # parse CMake config headers as C
-set(DOXYGEN_IMAGE_PATH "doc/img")
+set(DOXYGEN_IMAGE_PATH "${CMAKE_CURRENT_SOURCE_DIR}/img")
 set(DOXYGEN_RECURSIVE NO)
-set(DOXYGEN_QUIET YES)
+set(DOXYGEN_NUM_PROC_THREADS 0) # detects from system
+
+# Enable \snippet resolution against the on-disk example sources. Paths in
+# \snippet directives are resolved relative to EXAMPLE_PATH, so referring to
+# `nanovdb/nanovdb/examples/ex_xxx/yyy.cc` from a doc page works as written.
+set(DOXYGEN_EXAMPLE_PATH "${CMAKE_SOURCE_DIR}")
+set(DOXYGEN_EXAMPLE_PATTERNS *.cc *.cpp *.cu *.cuh *.h *.txt)
+set(DOXYGEN_EXAMPLE_RECURSIVE YES)
+
+# Always capture Doxygen warnings to a logfile so CI can publish it as an
+# artifact and contributors can see exactly what's broken.
+set(DOXYGEN_WARN_LOGFILE "${CMAKE_CURRENT_BINARY_DIR}/doxygen.warnings.log")
+if(OPENVDB_DOXYGEN_WARN_AS_ERROR)
+  # FAIL_ON_WARNINGS keeps generating docs but exits non-zero at the end if
+  # any warning was emitted. Requires Doxygen >= 1.9.x.
+  set(DOXYGEN_WARN_AS_ERROR FAIL_ON_WARNINGS)
+  # Don't suppress messages so contributors can see warnings during the build.
+  set(DOXYGEN_QUIET NO)
+else()
+  set(DOXYGEN_QUIET YES)
+endif()

 set(DOXYGEN_GENERATE_HTML YES)
 set(DOXYGEN_GENERATE_MAN NO)
 set(DOXYGEN_GENERATE_LATEX NO)
 set(DOXYGEN_GENERATE_TODOLIST NO)
-
-set(DOXYGEN_HTML_COLORSTYLE_HUE 4)
-set(DOXYGEN_HTML_COLORSTYLE_SAT 222)
 set(DOXYGEN_HTML_TIMESTAMP NO) # for simpler website diffs

+# Render math formulas via MathJax instead of LaTeX -> PNG. This removes the
+# entire `latex`/`dvips`/`gs` toolchain dependency from both local doc builds
+# and CI (see .github/workflows/docs.yml), eliminates a class of cross-platform
+# install pain, and produces crisper, selectable math in the rendered HTML.
+set(DOXYGEN_USE_MATHJAX YES)
+set(DOXYGEN_MATHJAX_VERSION MathJax_3) # default JS bundle loaded from a CDN
+
 set(DOXYGEN_EXTRACT_ALL YES)
 set(DOXYGEN_EXTRACT_STATIC YES)
 set(DOXYGEN_EXTRACT_LOCAL_CLASSES NO)

+# Configuration for doxygen-awesome-css
+# {
+set(DOXYGEN_HTML_COLORSTYLE_HUE 209)
+set(DOXYGEN_HTML_COLORSTYLE_SAT 255)
+set(DOXYGEN_HTML_COLORSTYLE_GAMMA 113)
+set(DOXYGEN_GENERATE_TREEVIEW YES)
+set(DOXYGEN_DISABLE_INDEX NO)
+set(DOXYGEN_FULL_SIDEBAR NO)
+set(DOXYGEN_HTML_HEADER
+  "${CMAKE_CURRENT_SOURCE_DIR}/header.html")
+set(DOXYGEN_HTML_EXTRA_STYLESHEET
+  "${CMAKE_SOURCE_DIR}/ext/doxygen-awesome/doxygen-awesome.css"
+  "${CMAKE_SOURCE_DIR}/ext/doxygen-awesome/doxygen-awesome-sidebar-only.css"
+  "${CMAKE_SOURCE_DIR}/ext/doxygen-awesome/doxygen-awesome-sidebar-only-darkmode-toggle.css"
+  "${CMAKE_CURRENT_SOURCE_DIR}/openvdb.css")
+set(DOXYGEN_HTML_EXTRA_FILES
+  "${CMAKE_SOURCE_DIR}/ext/doxygen-awesome/doxygen-awesome-paragraph-link.js"
+  "${CMAKE_SOURCE_DIR}/ext/doxygen-awesome/doxygen-awesome-fragment-copy-button.js"
+  "${CMAKE_SOURCE_DIR}/ext/doxygen-awesome/doxygen-awesome-darkmode-toggle.js")
+set(DOXYGEN_HTML_COPY_CLIPBOARD NO) # required with Doxygen >= 1.10.0
+set(DOXYGEN_HTML_COLORSTYLE LIGHT) # required with Doxygen >= 1.9.5
+# }
+
 # Allow for root directory paths to be listed (e.g. openvdb/io/File.h)
 # Rather than just the file names (e.g. File.h). This makes it far
 # easier to navigate the docs
@@ -110,12 +182,13 @@ set(DOXYGEN_HIDE_SCOPE_NAMES YES)
 set(DOXYGEN_INLINE_INHERITED_MEMB YES)
 set(DOXYGEN_MACRO_EXPANSION YES)
 # @todo use dot with class diagrams for relevant classes
+set(DOXYGEN_HAVE_DOT NO)
+set(DOXYGEN_CLASS_GRAPH NO)
 set(DOXYGEN_CLASS_DIAGRAMS NO)

-
 set(DOXYGEN_ALIASES
-  [[ijk="(<i>i</i>,&nbsp;<i>j</i>,&nbsp;<i>k</i>)"]]
-  [[xyz="(<i>x</i>,&nbsp;<i>y</i>,&nbsp;<i>z</i>)"]]
+  [[ijk="(<i>i</i>,&nbsp\;<i>j</i>,&nbsp\;<i>k</i>)"]]
+  [[xyz="(<i>x</i>,&nbsp\;<i>y</i>,&nbsp\;<i>z</i>)"]]
  [[const="<tt>const</tt>"]]
  "\"vdbnamespace=openvdb::v${OpenVDB_MAJOR_VERSION}_${OpenVDB_MINOR_VERSION}\""
  [["hunamespace=houdini_utils"]]
@@ -143,12 +216,18 @@ set(DOXYGEN_PREDEFINED
  "OPENVDB_VERSION_NAME=v${OpenVDB_MAJOR_VERSION}_${OpenVDB_MINOR_VERSION}"
  "OPENVDB_ABI_VERSION_NUMBER=${OpenVDB_MAJOR_VERSION}"
  [[__declspec(x):= __attribute__(x):=]]
+  # Strip the symbol-visibility/export macro so Doxygen doesn't fold it into a
+  # declaration's name. Doxygen tolerates `class OPENVDB_API X` but mis-parses
+  # `enum OPENVDB_API X` (e.g. half::FromBitsTag in ext/imath/.../Half.h),
+  # reporting the enum as a documented-but-undeclared symbol.
+  "OPENVDB_API="
  # Replace the deprecated macros with doxygen deprecated comments. This ensures
  # that every bit of deprecated C++ is listed in the doxygen deprecated list,
  # even if it was not properly tagged
  [["OPENVDB_DEPRECATED=/** \deprecated */"]]
  [["OPENVDB_DEPRECATED_MESSAGE(x)=/** \deprecated x */"]]
-  "OPENVDB_USE_LOG4CPLUS=")
+  "OPENVDB_USE_LOG4CPLUS="
+  "OPENVDB_UBSAN_SUPPRESS=")

 set(DOXYGEN_ENABLED_SECTIONS "")

@@ -190,9 +269,9 @@ if(OPENVDB_DOXYGEN_HOUDINI)
  list(APPEND DOXY_FILES
    doc/houdini.txt
    openvdb_houdini/openvdb_houdini/AttributeTransferUtil.h
-    openvdb_houdini/openvdb_houdini/GEO_PrimVDB.h
+    openvdb_houdini/openvdb_houdini/reference/GEO_PrimVDB.h
+    openvdb_houdini/openvdb_houdini/reference/GU_PrimVDB.h
    openvdb_houdini/openvdb_houdini/GT_GEOPrimCollectVDB.h
-    openvdb_houdini/openvdb_houdini/GU_PrimVDB.h
    openvdb_houdini/openvdb_houdini/GU_VDBPointTools.h
    openvdb_houdini/openvdb_houdini/GeometryUtil.h
    openvdb_houdini/openvdb_houdini/ParmFactory.h
@@ -205,7 +284,9 @@ if(OPENVDB_DOXYGEN_HOUDINI)
 endif()

 if(OPENVDB_DOXYGEN_AX)
-  set(DOXYGEN_EXTRA_PACKAGES amsmath) # for math formatting
+  # AX docs use @f$ inline and @f[ block math. Formula font size also applies
+  # under MathJax (set above). The amsmath package the AX docs assume is
+  # bundled by default in MathJax v3, so no extra extensions are required.
  set(DOXYGEN_FORMULA_FONTSIZE 14)
  list(APPEND DOXYGEN_ENABLED_SECTIONS "OPENVDB_AX")

@@ -221,8 +302,7 @@ if(OPENVDB_DOXYGEN_AX)
    openvdb_ax/openvdb_ax/compiler
    openvdb_ax/openvdb_ax/grammar
    openvdb_ax/openvdb_ax/grammar/generated
-    openvdb_ax/openvdb_ax/math
-    openvdb_ax/openvdb_ax/test)
+    openvdb_ax/openvdb_ax/math)
 endif()

 if(OPENVDB_DOXYGEN_NANOVDB)
@@ -236,7 +316,6 @@ if(OPENVDB_DOXYGEN_NANOVDB)
    doc/nanovdb/HelloWorld.md
    nanovdb/nanovdb
    nanovdb/nanovdb/util
-    nanovdb/nanovdb/examples/benchmark
    nanovdb/nanovdb/examples/ex_bump_pool_buffer
    nanovdb/nanovdb/examples/ex_collide_level_set
    nanovdb/nanovdb/examples/ex_make_custom_nanovdb
@@ -255,8 +334,7 @@ if(OPENVDB_DOXYGEN_NANOVDB)
    nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor_cuda
    nanovdb/nanovdb/examples/ex_util
    nanovdb/nanovdb/examples/ex_vox_to_nanovdb
-    nanovdb/nanovdb/examples/ex_write_nanovdb_grids
-    nanovdb/nanovdb/unittest)
+    nanovdb/nanovdb/examples/ex_write_nanovdb_grids)
 endif()

 set(DOXYGEN_VERBATIM_VARS DOXYGEN_ALIASES DOXYGEN_PREDEFINED)
@@ -1,14 +1,13 @@
 /**
-<div style="width:800px;text-align:justify;">

-@page ax AX
+@page ax AX Language Documentation

-@section axlanguage AX Language Documentation
+@section axlanguage Intro
@par
 Welcome to the AX Language documentation. These docs provide detailed
 information on the AX language including syntax, data types, available
 functionality (including OpenVDB specific methods) execution structure and
-control flow. See the @subpage axcplusplus documentation for the C++ developer
+control flow. See the @ref axcplusplus documentation for the C++ developer
 documentation.
@note
 Some sections of this document are still to be completed. Please get in touch
@@ -108,8 +107,8 @@ function (see @ref axprogramexample) and whilst there are ways to achieve
 reductions, it is an example which would be better suited to a custom C++
 implementation.
@par
-For details of extending AX for custom geometry, see the @subpage axcplusplus
-developer documentation. Note that custom geometry support requires C++ extension.
+For details of extending AX for custom geometry, see the @ref axcplusplus
+developer documentation. Note that custom geometry support requires `C++` extension.


@subsection axprogramexample A Program Example
@@ -133,20 +132,20 @@ The following explains the above example in relation to OpenVDB data; OpenVDB
 points and OpenVDB volumes. The example demonstrates reading from a value on
 some input, either a point attribute or a voxel value, and storing the
 result in a @ref axdecls "local variable". Importantly, the attribute being
-accessed has the name @b `myattribute` and the type @b `float`. The first statement:
+accessed has the name `myattribute` and the type `float`. The first statement:
 <!-- A -->@code{.c}
 float temp = float@myattribute;
@endcode
 Invokes a @b read from this attribute (see @ref axattribaccess), retrieving the
-value from the geometry and storing it in the variable @b `temp` to be used in
+value from the geometry and storing it in the variable `temp` to be used in
 the AX program. The second statement:
@code{.c}
 if (temp > 0.0f) float@myattribute = 0.0f;
@endcode
-Performs a comparison of this value with @b `0.0f` and, if the value is greater
-than @b `0.0f`, performs a @b write to the geometry in the form of
-<b>`float@myattribute = 0.0f;`</b> which sets the value of this attribute on the
-geometry to @b `0.0f`.
+Performs a comparison of this value with `0.0f` and, if the value is greater
+than `0.0f`, performs a @b write to the geometry in the form of
+`float@myattribute = 0.0f;` which sets the value of this attribute on the
+geometry to `0.0f`.
@par
 For OpenVDB Points, this kernel is run over every point in an OpenVDB Points
 Grid. In simple pseudo code:
@@ -155,7 +154,7 @@ foreach(point in pointgrid)
    run_ax_kernel(point)
 done()
@endcode
-Where @b `pointgrid` is a single Point Data Grid. For OpenVDB volumes the kernel
+Where `pointgrid` is a single Point Data Grid. For OpenVDB volumes the kernel
 is run over each voxel in a provided grid:
@code{.unparsed}
 foreach(grid in grids)
@@ -165,8 +164,8 @@ foreach(grid in grids)
 done()
@endcode
@par
-Where @b `grids` can comprise of any number of OpenVDB volumes. In this program,
-only a floating point grid with the name @b `myattribute` will be updated. Each
+Where `grids` can comprise of any number of OpenVDB volumes. In this program,
+only a floating point grid with the name `myattribute` will be updated. Each
 voxel in this grid will have it's value compared in the same way as the points
 example given above.

@@ -282,9 +281,9 @@ missing types from OpenVDB.

@subsection axscalars Scalars
@par
-AX Supports boolean, integer and floating point scalar types. @b `int` and
-@b `int32` represent 32-bit integer values and @b `int64` represents a 64-bit
-value. @b `float` and @b `double` represent 32-bit and 64-bit floating point
+AX Supports boolean, integer and floating point scalar types. `int` and
+`int32` represent 32-bit integer values and `int64` represents a 64-bit
+value. `float` and `double` represent 32-bit and 64-bit floating point
 values respectively. All scalars, excluding bools, are signed; there are no
 other unsigned scalar types in AX.
@par Integer Overflow and Floating Point Truncation
@@ -309,10 +308,10 @@ int64 a = 2147483648l; // one more than can be held in an int (note the last let
 int b = a; // implicit conversion. "b" is set to -2147483648
@endcode
@par Infinite and NaN values
-Floating point values have two additional intrinsic states; @b `inf` and @b `nan`.
+Floating point values have two additional intrinsic states; `inf` and `nan`.
 These states can occur during invalid floating point arithmetic. AX performs no
-checks on arithmetic to catch these values. A typical example of both @b `inf`
-and @b `nan` values are division by zero expressions:
+checks on arithmetic to catch these values. A typical example of both `inf`
+and `nan` values are division by zero expressions:
@code
 float a = 1.0f/0.0f;
 print(a); // prints "inf"
@@ -337,16 +336,16 @@ There is currently support for vectors of sizes 2, 3 and 4 elements. The suffix
 letter on the vector types denotes the contained elements precision, with the
 number corresponding to the vector size. For example, @b `vec2i` denotes a vector
 of two 32-bit integer elements. There are no native types for vectors with
-@b `bool`, @b `int16` or @b `int64` elements, so these are not supported.
+`bool`, `int16` or `int64` elements, so these are not supported.
@par Matrices
 Matrices are stored in row major layout (lexographical access order) which
 matches the representation of a matrix in OpenVDB. Matrix support consists of
-@b `float` and @b `double` precision elements with dimensions either @b `3x3` or
-@b `4x4` (total size of 9 and 16 elements respectively). There are no integer or
+`float` and `double` precision elements with dimensions either `3x3` or
+`4x4` (total size of 9 and 16 elements respectively). There are no integer or
 boolean matrix types. Similiar to vectors, the suffix letter on the type denotes
 the contained elements precision, with the number corresponding to the matrix
-dimension. A matrix of type @b `mat3d` therefor corresponds to a @b `3x3` matrix
-(9 elements) with @b `double` precision elements.
+dimension. A matrix of type `mat3d` therefor corresponds to a `3x3` matrix
+(9 elements) with `double` precision elements.
@par Element Access
 As elements of all matrix and vector types are stored contiguously, they can be
 accessed by the @ref axvecaccessop "[]&nbsp;operator". Vector elements can also be
@@ -370,10 +369,10 @@ performing matrix projections (transformations) on vector arguments.
 Strings are an array of characters stored in a unique AX type which is
 incompatible with @ref axopaccess "container accesses". As such, string support
 is currently fairly limited. Characters themselves (users may know this as a
-@b `char` type) have no frontend type, so a @b `string` type must be used for
+`char` type) have no frontend type, so a `string` type must be used for
 any number of characters (including the empty string). String literals are
 represented by enclosing any number of @ref axtokens "supported AX characters"
-by quotes <b>\"&nbsp;\"</b>.
+by quotes&nbsp;`" "`.

@subsection axtypeprecedence Implicit conversion / type precedence
@par
@@ -441,9 +440,9 @@ depending on the operand types.
 <th colspan="9"> Operators
 </th></tr>
 <tr style="text-align:center">
-<td bgcolor="#ffc4c4" style="border-right-width: thick" colspan="4"> @ref axopbinary </td>
-<td bgcolor="#ffc4c4" style="border-right-width: thick" colspan="3"> @ref axopunary </td>
-<td bgcolor="#ffc4c4" colspan="2"> Other </td></tr>
+<td style="border-right-width: thick" colspan="4"> @ref axopbinary </td>
+<td style="border-right-width: thick" colspan="3"> @ref axopunary </td>
+<td colspan="2"> Other </td></tr>
 <tr style="text-align:center">
 <td> @ref axopassignment </td>
 <td> @ref axopbinarithmetic </td>
@@ -456,50 +455,33 @@ depending on the operand types.
 <td> @ref axopother </td></tr>
 <tr style="text-align:center">
 <td><p> <!-- *********************** COLUMN ASSIGNMENT *********************** -->
-<code>
-a = b<br/>a += b<br/>a -= b<br/>a *= b<br/>a /= b<br/>a %= b<br/>
-a &amp;= b<br/>a |= b<br/>a ^= b<br/>a &lt;&lt;= b<br/>a &gt;&gt;= b
-</code>
+`a = b`<br/>`a += b`<br/>`a -= b`<br/>`a *= b`<br/>`a /= b`<br/>`a %= b`<br/>
+`a &= b`<br/>`a |= b`<br/>`a ^= b`<br/>`a <<= b`<br/>`a >>= b`
 </p></td>
 <td><p> <!-- *********************** COLUMN ARITH *********************** -->
-<code>
-a + b<br/>a - b<br/>a * b<br/>a / b<br/>a % b<br/>a &amp; b<br/>a | b<br/>
-a ^ b<br/>a &lt;&lt; b<br/>a &gt;&gt; b
-</code></p></td>
+`a + b`<br/>`a - b`<br/>`a * b`<br/>`a / b`<br/>`a % b`<br/>`a & b`<br/>`a | b`<br/>
+`a ^ b`<br/>`a << b`<br/>`a >> b`
+</p></td>
 <td><p> <!-- *********************** COLUMN COMP *********************** -->
-<code>
-a == b<br/>a != b<br/>a &lt; b<br/>a &gt; b<br/>a &lt;= b<br/>a &gt;= b
-</code>
+`a == b`<br/>`a != b`<br/>`a < b`<br/>`a > b`<br/>`a <= b`<br/>`a >= b`
 </p></td>
 <td style="border-right-width: thick"><p> <!-- *********************** COLUMN LOGICAL *********************** -->
-<code>
-a&nbsp;&amp;&amp;&nbsp;b<br/>a&nbsp;||&nbsp;b
-</code>
+`a && b`<br/>`a || b`
 </p></td>
 <td><p> <!-- *********************** COLUMN ARITH *********************** -->
-<code>
-+a<br/>-a<br/>~a
-</code>
+`+a`<br/>`-a`<br/>`~a`
 </p></td>
 <td><p> <!-- *********************** COLUMN LOGICAL *********************** -->
-<code>
-!a
-</code>
+`!a`
 </p></td>
 <td style="border-right-width: thick"><p> <!-- *********************** COLUMN INC/DEC *********************** -->
-<code>
-++a<br/>\-\-a<br/>a++<br/>a\-\-
-</code>
+`++a`<br/>`--a`<br/>`a++`<br/>`a--`
 </p></td>
 <td><p> <!-- *********************** COLUMN OTHER *********************** -->
-<code>
-a[]<br/>a[,]<br/>a.b
-</code>
+`a[]`<br/>`a[,]`<br/>`a.b`
 </p></td>
 <td><p> <!-- *********************** COLUMN OTHER *********************** -->
-<code>
-a(...)<br/>a, b<br/>a ? b : c<br/>{&nbsp;a,b&nbsp;...&nbsp;}
-</code>
+`a(...)`<br/>`a, b`<br/>`a ? b : c`<br/>`a ?: c`<br/>`{ a,b ... }`
 </p></td>
 </tr>
 </table>
@@ -522,34 +504,34 @@ assignment operators have the following forms:
@par
 <table>
 <tr><th>Operator Name</th><th>Operator Syntax</th><th>Returns</th></tr>
-<tr><td>Simple assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Addition assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`+=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Subtraction assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`-=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Multiplication assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`*=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Division assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`/=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Modulo assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`%=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Bitwise AND assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`&amp;=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Bitwise OR assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`|=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Bitwise XOR assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`^=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Bitwise shift left assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`&lt;&lt;=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
-<tr><td>Bitwise shift right assignment</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`&gt;&gt;=`</b> @a `rhs` </td><td>Reference to @a `lhs`</td></tr>
+<tr><td>Simple assignment</td><td> @a lhs <b>`=`</b> @a rhs </td>
+<td rowspan="11">Reference to @a lhs</td></tr>
+<tr><td>Addition assignment</td><td> @a lhs <b>`+=`</b> @a rhs </td></tr>
+<tr><td>Subtraction assignment</td><td> @a lhs <b>`-=`</b> @a rhs </td></tr>
+<tr><td>Multiplication assignment</td><td> @a lhs <b>`*=`</b> @a rhs </td></tr>
+<tr><td>Division assignment</td><td> @a lhs <b>`/=`</b> @a rhs </td></tr>
+<tr><td>Modulo assignment</td><td> @a lhs <b>`%=`</b> @a rhs </td></tr>
+<tr><td>Bitwise AND assignment</td><td> @a lhs <b>`&=`</b> @a rhs </td></tr>
+<tr><td>Bitwise OR assignment</td><td> @a lhs <b>`|=`</b> @a rhs </td></tr>
+<tr><td>Bitwise XOR assignment</td><td> @a lhs <b>`^=`</b> @a rhs </td></tr>
+<tr><td>Bitwise shift left assignment</td><td> @a lhs <b>`<<=`</b> @a rhs </td></tr>
+<tr><td>Bitwise shift right assignment</td><td> @a lhs <b>`>>=`</b> @a rhs </td></tr>
 </table>
@par
 These can be further categorised into two types; direct assignment, which is
 comprised only of the <b>simple assignment</b> operator, and compound assignments,
 which is the set of all other assignment operators.
@par Direct Assignment
-Direct assignments replace the contents of the @b `lhs` with a copy of the @b
-`rhs`:
+Direct assignments replace the contents of the `lhs` with a copy of the `rhs`:
@par
 <table><tr>
-<td bgcolor="#ffc4c4"> @a `lhs` <b>= </b> @a `rhs` </td>
+<td> @a lhs <b>=</b> @a rhs </td>
 </tr></table>
@par
-The @b `rhs` side is not modified and the @b `lhs` is not read - only written
-to. If the @b `rhs` type does not match the @b `lhs` type, the @b `rhs` is first
-implicitly converted into a new temporary with the @b `lhs` type before being
-copied into the @b `lhs`. The following combination of AX type categories are
+The `rhs` side is not modified and the `lhs` is not read - only written
+to. If the `rhs` type does not match the `lhs` type, the `rhs` is first
+implicitly converted into a new temporary with the `lhs` type before being
+copied into the `lhs`. The following combination of AX type categories are
 supported for direct assignment  (if an operation is not listed, it is not
 supported):
@par
@@ -569,15 +551,14 @@ supported):
 <td style="border-top-width: thick">@ref axscalars "scalar"</td>
 <td style="border-top-width:thick;text-align:left">
 Each element of the vector is set to the right hand side scalar. i.e.
-<br/> `a[0]&nbsp;=&nbsp;b;&nbsp;...&nbsp;a[n-1]&nbsp;=&nbsp;b;` <br/>
-where `n` is the size of the vector. If the scalar type does not match
-element type of vector, the scalar is copied and @ref axtypeprecedence
-"implicitly cast" to that type.</td>
+`a[0] = b; ... a[n-1] = b;` where `n` is the size of the vector. If the
+scalar type does not match element type of vector, the scalar is copied and
+@ref axtypeprecedence "implicitly cast" to that type.</td>
 </tr>
 <tr style="text-align:center"><td>@ref axvecmats "vector"</td>
-<td style="text-align:left">Component wise assignment i.e.
-<br/> `a[0]&nbsp;=&nbsp;b;&nbsp;...&nbsp;a[n-1]&nbsp;=&nbsp;b;` <br/>
-where `n` is the size of the vector.
+<td style="text-align:left">
+Component wise assignment i.e. `a[0] = b; ... a[n-1] = b;` where `n` is the size
+of the vector.
 <!-- A -->@code{.c}
 vec3f a = 0, b = 1;
 a = b;
@@ -612,7 +593,7 @@ of matrix, the scalar is copied and @ref axtypeprecedence "implicitly cast" to
 that type.</td>
 <tr style="text-align:center"><td>@ref axvecmats "matrix"</td>
 <td style="text-align:left">Component wise assignment i.e.
-<br/> `a[0]&nbsp;=&nbsp;b[0];&nbsp;...&nbsp;a[n-1]&nbsp;=&nbsp;b[n-1];` <br/>
+<br/> `a[0] = b[0]; ... a[n-1] = b[n-1];` <br/>
 where `n` is the @b total size of the matrix.
 <!-- A -->@code{.c}
 mat4f a = 0, b = 1;
@@ -645,25 +626,27 @@ int a = 1; // assign a from an integer literal of value 1
 c = a; // assign components of c (x/y/z) to the result of float(a)
@endcode
@par Compound Assignment
-Compound assignments replace the the contents of the @b `lhs` with the result of
-a @ref axopbinarithmetic "binary arithmetic" operation of the @b `lhs` with the
-@b `rhs`:
+Compound assignments replace the the contents of the `lhs` with the result of
+a @ref axopbinarithmetic "binary arithmetic" operation of the `lhs` with the
+`rhs`:
@par
 <table>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b op @a `rhs` </td></tr>
+<tr><td> @a lhs @b op @a rhs </td></tr>
 </table>
-Where @b op is one of :
+
+Where @b op is one of:
+
 <table><tr>
-<td bgcolor="#ffc4c4"> <b>`+=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`-=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`*=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`/=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`%=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`&amp;=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`|=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`^=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`&lt;&lt;=`</b></td>
-<td bgcolor="#ffc4c4"> <b>`&gt;&gt;= `</b></td>
+<td> <b>`+=`</b></td>
+<td> <b>`-=`</b></td>
+<td> <b>`*=`</b></td>
+<td> <b>`/=`</b></td>
+<td> <b>`%=`</b></td>
+<td> <b>`&=`</b></td>
+<td> <b>`|=`</b></td>
+<td> <b>`^=`</b></td>
+<td> <b>`<<=`</b></td>
+<td> <b>`>>= `</b></td>
 </tr></table>
@par
 The behaviour of a given compound assignment (for example `a += b`) is similar
@@ -680,8 +663,8 @@ int a = 1;
 ++a += 1; // equal to a = ++a + 1 which is equal to 3
@endcode
@par
-The @b `rhs` side is not modified, however the @b `lhs` is both read from and
-written to. Note that the arithmetic operation may cast either @b `lhs` or @b
+The `rhs` side is not modified, however the `lhs` is both read from and
+written to. Note that the arithmetic operation may cast either `lhs` or
 `rhs` following the rules of AX's @ref axopbinarithmetic
 "arithmetic type precedence". See the @ref axopbinarithmetic
 "arithmetic operations" section for more information on arithmetic type
@@ -720,7 +703,7 @@ b = c;    // b becomes 4
 a = b;    // a becomes float(4)
@endcode
@par
-Importantly, @b `a` receives the result of any previous implicit casts to the
+Importantly, `a` receives the result of any previous implicit casts to the
 right hand side of its binary assignment.


@@ -737,20 +720,20 @@ container types). Binary arithmetic operations have the following forms:
@par
 <table>
 <tr><th>Operator&nbsp;Name</th><th>Operator&nbsp;Syntax</th><th>Returns</th></tr>
-<tr><td>Addition</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`+`</b> @a `rhs` </td><td>The sum of both operands</td></tr>
-<tr><td>Subtraction</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`-`</b> @a `rhs` </td><td>The first operand minus the second operand</td></tr>
-<tr><td>Multiplication</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`*`</b> @a `rhs` </td><td>The product of both operands</td></tr>
-<tr><td>Division</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`/`</b> @a `rhs` </td><td>The first operand divided by the second operand</td></tr>
-<tr><td>Modulo</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`%`</b> @a `rhs` </td><td>The @b floored modulo operator. See @ref axbinmultop "Multiplicative operands"</td></tr>
-<tr><td>Bitwise AND</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`&amp;`</b> @a `rhs` </td>
+<tr><td>Addition</td><td> @a lhs <b>`+`</b> @a rhs </td><td>The sum of both operands</td></tr>
+<tr><td>Subtraction</td><td> @a lhs <b>`-`</b> @a rhs </td><td>The first operand minus the second operand</td></tr>
+<tr><td>Multiplication</td><td> @a lhs <b>`*`</b> @a rhs </td><td>The product of both operands</td></tr>
+<tr><td>Division</td><td> @a lhs <b>`/`</b> @a rhs </td><td>The first operand divided by the second operand</td></tr>
+<tr><td>Modulo</td><td> @a lhs <b>`%`</b> @a rhs </td><td>The @b floored modulo operator. See @ref axbinmultop "Multiplicative operands"</td></tr>
+<tr><td>Bitwise AND</td><td> @a lhs <b>`&`</b> @a rhs </td>
 <td>The @b integral bitwise AND result of each bit in the first operand applied to the bit at the same location in the second operand</td></tr>
-<tr><td>Bitwise OR</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`|`</b> @a `rhs` </td>
+<tr><td>Bitwise OR</td><td> @a lhs <b>`|`</b> @a rhs </td>
 <td>The @b integral bitwise OR result of each bit in the first operand applied to the bit at the same location in the second operand</td></tr>
-<tr><td>Bitwise XOR</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`^`</b> @a `rhs` </td>
+<tr><td>Bitwise XOR</td><td> @a lhs <b>`^`</b> @a rhs </td>
 <td>The @b integral bitwise XOR result of each bit in the first operand applied to the bit at the same location in the second operand</td></tr>
-<tr><td>Bitwise shift left</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`&lt;&lt;`</b> @a `rhs` </td>
+<tr><td>Bitwise shift left</td><td> @a lhs <b>`<<`</b> @a rhs </td>
 <td>The @b integral bitwise left shift of the first operand by the second operand</td></tr>
-<tr><td>Bitwise shift right</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`&gt;&gt;`</b> @a `rhs` </td>
+<tr><td>Bitwise shift right</td><td> @a lhs <b>`>>`</b> @a rhs </td>
 <td>The @b integral bitwise right shift of the first operand by the second operand</td></tr>
 </table>
@anchor axbinadditiveop
@@ -759,13 +742,13 @@ Binary additive operations perform summations on the input arguments. They have
 the following forms:
@par
 <table>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `+` @a `rhs` </td></tr>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `-` @a `rhs` </td></tr>
+<tr><td> @a lhs @b + @a rhs </td></tr>
+<tr><td> @a lhs @b - @a rhs </td></tr>
 </table>
@par
 After @ref axtypeprecedence "implicit conversion", the result of the binary
-operation with the arithmetic @b `+` (@ref axtokens "plus token") is the sum of
-the operands, and the result of the binary operation with the arithmetic @b `-`
+operation with the arithmetic `+` (@ref axtokens "plus token") is the sum of
+the operands, and the result of the binary operation with the arithmetic `-`
 (@ref axtokens "minus token") token is the second operand subtracted from the
 first operand. The following combination of AX type categories are supported for
 additive operations (if an operation is not listed, it is not supported):
@@ -870,19 +853,19 @@ c - b; // valid vec - scalar
 The binary multiplicative operands have the following forms:
@par
 <table>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `*` @a `rhs` </td></tr>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `/` @a `rhs` </td></tr>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `%` @a `rhs` </td></tr>
+<tr><td> @a lhs `*` @a rhs </td></tr>
+<tr><td> @a lhs `/` @a rhs </td></tr>
+<tr><td> @a lhs `%` @a rhs </td></tr>
 </table>
@par
 After @ref axtypeprecedence "implicit conversion", the result of the binary
 operation:
- - with the arithmetic @b `*` (@ref axtokens "asterisk token") is:
+ - with the arithmetic `*` (@ref axtokens "asterisk token") is:
    - the product of the operands
    - @b OR performs matrix multiplication for vector and matrix types.
- - with the arithmetic @b `/` (@ref axtokens "forward slash token") is the first
+ - with the arithmetic `/` (@ref axtokens "forward slash token") is the first
 operand divided by the second operand.
- - with the arithmetic @b `%` (@ref axtokens "percentage token") is the
+ - with the arithmetic `%` (@ref axtokens "percentage token") is the
 @b floored modulo operations .i.e. the expression `d % D` returns the result
 `D - d * floor(D/d)`. This is in contrast to truncated modulo operations
 `D - d * (D/d)` where the division is truncated.
@@ -894,7 +877,7 @@ mean that the relationship properties between `%` and `/` differ when either
@b `d` or @b `D` is negative. i.e.:
 `(d/D)*D + (d%D) != d`
@note
-The multiplicative operation @b `*` has important behaviour for vectors and
+The multiplicative operation `*` has important behaviour for vectors and
 matrix types.
@par
 The following combination of AX type categories are supported for multiplicative
@@ -907,7 +890,7 @@ operations (if an operation is not listed, it is not supported):
 <tr style="text-align:center"><td rowspan="3">@ref axscalars "scalar"</td><td rowspan="2">`* / %`</td><td rowspan="1">@ref axscalars "scalar"</td>
 <td style="text-align:left">Returns the result of the scalar multiplication, division or remainder of the division respectively. </td>
 </tr>
-<tr style="text-align:center"><td>@ref axvecmats "vector"</td></td>
+<tr style="text-align:center"><td>@ref axvecmats "vector"</td>
 <td style="text-align:left" rowspan="1">
 Performs per component binary operations (after
@ref axtypeprecedence "implicit conversion") with the left hand side scalar to
@@ -1096,27 +1079,27 @@ Floating point multiplication is communicative (`a * b = b * a`) but is not
 Binary bitwise operations have the following forms:
@par
 <table>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `&` @a `rhs` </td></tr>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` <b> `|` </b> @a `rhs` </td></tr>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `^` @a `rhs` </td></tr>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `<<` @a `rhs` </td></tr>
-<tr><td bgcolor="#ffc4c4"> @a `lhs` @b `>>` @a `rhs` </td></tr>
+<tr><td> @a lhs <b> `&` </b> @a rhs </td></tr>
+<tr><td> @a lhs <b> `|` </b> @a rhs </td></tr>
+<tr><td> @a lhs <b> `^` </b> @a rhs </td></tr>
+<tr><td> @a lhs <b> `<<` </b> @a rhs </td></tr>
+<tr><td> @a lhs <b> `>>` </b> @a rhs </td></tr>
 </table>
@par
 These operations only run on @b integral operands. After @ref axtypeprecedence
 "implicit conversion", the result of the binary operation:
- - with the @b `&` (@ref axtokens "ampersand token") is the result of the
+ - with the @b & (@ref axtokens "ampersand token") is the result of the
  logical AND operation on each pair of the corresponding bits of the input
  operands.
- - with the <b> `|` </b> (@ref axtokens "pipe token") is the result of the
+ - with the <b> | </b> (@ref axtokens "pipe token") is the result of the
  logical OR operation on each pair of the corresponding bits of the input
  operands.
- - with the @b `^` (@ref axtokens "hat token") is the result of the
+ - with the @b ^ (@ref axtokens "hat token") is the result of the
  logical XOR operation on each pair of the corresponding bits of the input
  operands.
- - with the @b `<<` (@ref axtokens "less than tokens") is the value shifted left
+ - with the @b << (@ref axtokens "less than tokens") is the value shifted left
 with zeros shifted in on the right.
- - with the @b `>>` (@ref axtokens "greater than tokens") is the value shifted
+ - with the @b >> (@ref axtokens "greater than tokens") is the value shifted
 right with zeros shifted in on the left.
@par
 <table>
@@ -1144,17 +1127,17 @@ is of type @b `bool`. Binary comparison operations have the following forms:
@par
 <table>
 <tr><th>Operator&nbsp;Name</th><th>Operator&nbsp;Syntax</th><th>Returns</th></tr>
-<tr><td>Equal to</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`==`</b> @a `rhs` </td>
+<tr><td>Equal to</td><td> @a lhs <b>`==`</b> @a rhs </td>
 <td>Returns `true` if both operands are equal</td></tr>
-<tr><td>Not equal to</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`!=`</b> @a `rhs` </td>
+<tr><td>Not equal to</td><td> @a lhs <b>`!=`</b> @a rhs </td>
 <td>Returns `true` if operands are not equal</td></tr>
-<tr><td>Less than</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`<`</b> @a `rhs` </td>
+<tr><td>Less than</td><td> @a lhs <b>`<`</b> @a rhs </td>
 <td>Returns `true` the left hand side value is less than the right hand side</td></tr>
-<tr><td>Greater than</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`>`</b> @a `rhs` </td>
+<tr><td>Greater than</td><td> @a lhs <b>`>`</b> @a rhs </td>
 <td>Returns `true` the left hand side value is greater than the right hand side</td></tr>
-<tr><td>Less than or equal to</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`<=`</b> @a `rhs` </td>
+<tr><td>Less than or equal to</td><td> @a lhs <b>`<=`</b> @a rhs </td>
 <td>Returns `true` the left hand side value is less than or equal to the right hand side</td></tr>
-<tr><td>Greater than or equal to</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`>=`</b> @a `rhs` </td>
+<tr><td>Greater than or equal to</td><td> @a lhs <b>`>=`</b> @a rhs </td>
 <td>Returns `true` the left hand side value is greater than  or equal to the right hand side</td></tr>
 </table>
@par
@@ -1264,10 +1247,10 @@ They have the following forms:
@par
 <table>
 <tr><th>Operator&nbsp;Name</th><th>Operator&nbsp;Syntax</th><th>Returns</th></tr>
-<tr><td>AND</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`&&`</b> @a `rhs` </td>
+<tr><td>AND</td><td> @a lhs <b>`&&`</b> @a rhs </td>
 <td>Returns true if both operands are true. Otherwise, the result is false. This
 operator is short-circuiting.</td></tr>
-<tr><td>Inclusive OR</td><td bgcolor="#ffc4c4"> @a `lhs` <b>`||`</b> @a `rhs` </td>
+<tr><td>Inclusive OR</td><td> @a lhs <b>`||`</b> @a rhs </td>
 <td>Returns true if either the first or the second operand is true. This operator
 is short-circuiting.</td></tr>
 </table>
@@ -1300,14 +1283,14 @@ Arithmetic unary operators are comprised of arithemtic and bitwise operators.
@par
 <table>
 <tr><th>Operator Name</th><th>Operator Syntax</th><th>Returns</th></tr>
-<tr><td>Unary plus</td><td bgcolor="#ffc4c4"> <b>`+`</b>@a `a` </td>
-<td>Returns the value of its operand, @b `a`</td>
+<tr><td>Unary plus</td><td> `+a` </td>
+<td>Returns the value of its operand, `a`</td>
 </tr>
-<tr><td>Unary minus</td><td bgcolor="#ffc4c4"> <b>`-`</b>@a `a` </td>
-<td>Returns the negative representation of @b `a`</td>
+<tr><td>Unary minus</td><td> `-a` </td>
+<td>Returns the negative representation of `a`</td>
 </tr>
-<tr><td>Bitwise NOT</td><td bgcolor="#ffc4c4"> <b>`~`</b>@a `a` </td>
-<td>Returns the bitwise NOT (one's complement) value of @b `a`. This operator
+<tr><td>Bitwise NOT</td><td> `~a` </td>
+<td>Returns the bitwise NOT (one's complement) value of `a`. This operator
 is only valid on integral element types.</td>
 </tr>
 </table>
@@ -1365,7 +1348,7 @@ Logical unary operators consist of a single operand.
@par
 <table>
 <tr><th>Operator Name</th><th>Operator Syntax</th><th>Returns</th></tr>
-<tr><td>Logical NOT</td><td bgcolor="#ffc4c4"> <b>`!`</b>@a `a` </td>
+<tr><td>Logical NOT</td><td> `!a` </td>
 <td>Returns true if the operand is false. Otherwise, returns false.</td>
 </tr>
 </table>
@@ -1409,10 +1392,10 @@ each component of a vector or matrix. They have the following forms:
@par
 <table>
 <tr><th>Operator Name</th><th>Operator Syntax</th><th>Returns</th></tr>
-<tr><td>Pre-increment</td><td bgcolor="#ffc4c4"> <b>`++`</b>@a `a` </td><td>Returns a reference to the incremented result.</td>
-<tr><td>Post-increment</td><td bgcolor="#ffc4c4">@a `a`<b>`++`</b> </td><td>Returns a copy of the incremented result.</td>
-<tr><td>Pre-decrement</td><td bgcolor="#ffc4c4"> <b>`--`</b>@a `a` </td><td>Returns a reference to the decremented result.</td>
-<tr><td>Post-decrement</td><td bgcolor="#ffc4c4"> @a `a`<b>`--`</b> </td><td>Returns a copy of the decremented result.</td>
+<tr><td>Pre-increment</td><td> `++a` </td><td>Returns a reference to the incremented result.</td>
+<tr><td>Post-increment</td><td> `a++` </td><td>Returns a copy of the incremented result.</td>
+<tr><td>Pre-decrement</td><td> `--a` </td><td>Returns a reference to the decremented result.</td>
+<tr><td>Post-decrement</td><td> `a--` </td><td>Returns a copy of the decremented result.</td>
 </tr>
 </table>
@par
@@ -1444,20 +1427,20 @@ the following forms:
 <tr><th>Operator Name</th><th>Operator Syntax</th><th>Returns</th></tr>
 <tr style="text-align:left">
 <td>Dot&nbsp;Component&nbsp;Access</td>
-<td bgcolor="#ffc4c4"> <em>`vector`</em>&nbsp;<b>.</b>&nbsp;@a `component` </td>
-<td>Reference to component @a `component` of @a `vector`</td>
+<td> <em>`vector`</em>&nbsp;<b>.</b>&nbsp;`component` </td>
+<td>Reference to component `component` of `vector`</td>
 </tr>
 <tr style="text-align:left">
 <td>Container&nbsp;Component&nbsp;Access</td>
-<td bgcolor="#ffc4c4"> <em>`container`</em>&nbsp;<b>[</b>&nbsp;@a `exp` <b>]</b> </td>
-<td>Reference to component at index @a `exp` of @a `container`</td>
+<td> <em>`container`</em>&nbsp;<b>[</b>&nbsp;`exp` <b>]</b> </td>
+<td>Reference to component at index `exp` of `container`</td>
 </tr>
 <tr style="text-align:left">
 <td>Matrix&nbsp;Component&nbsp;Access</td>
-<td bgcolor="#ffc4c4"> <em>`matrix`</em>&nbsp;<b>[</b>&nbsp;@a `exp1`,&nbsp;<em>`exp2`</em>&nbsp;<b>]</b> </td>
+<td> <em>`matrix`</em>&nbsp;<b>[</b>&nbsp;`exp1`,&nbsp;<em>`exp2`</em>&nbsp;<b>]</b> </td>
 <td>Reference to component in row <b><em>`exp1`</em></b>, column <b><em>`exp2`</em></b>.
-Also equal to returning a reference to component at index: <b><em> `[exp1&nbsp;*&nbsp;dimension&nbsp;+&nbsp;exp2]` </em></b>
-of @a `matrix` where @a `dimension` is the the size of the matrix</td>
+Also equal to returning a reference to component at index: <b><em> `[exp1 * dimension + exp2]` </em></b>
+of `matrix` where `dimension` is the the size of the matrix</td>
 </tr>
 </table>
@par
@@ -1471,11 +1454,11 @@ container.
 The dot operator <b>.</b> is only valid for vector types. It has the form:
@par
 <table><tr>
-<td bgcolor="#ffc4c4">  @a `vector` <b>.</b> @a `component` </td>
+<td>  `vector` <b>.</b> `component` </td>
 </tr></table>
@par
 Where component is one of the below supported component letters. It only provides
-access for the first three elements of any vector (for @b `vec2` types, only the
+access for the first three elements of any vector (for `vec2` types, only the
 first 2 components are accessible). Accessing a vector using the dot operator is
 fundamentally the same as using the <b>`[]`</b> operator with the corresponding
 integer index.
@@ -1492,12 +1475,12 @@ a_{0}\quad a_{1}\quad a_{2}
@f]
 | Component   | Access Index   | Result        |
 |:-----------:|:--------------:|:-------------:|
-|  @b `A.x`   |       0        | @f$ a_{0} @f$ |
-|  @b `A.r`   |       0        | @f$ a_{0} @f$ |
-|  @b `A.y`   |       1        | @f$ a_{1} @f$ |
-|  @b `A.g`   |       1        | @f$ a_{1} @f$ |
-|  @b `A.z`   |       2        | @f$ a_{2} @f$ |
-|  @b `A.b`   |       2        | @f$ a_{2} @f$ |
+|     `A.x`   |       0        | @f$ a_{0} @f$ |
+|     `A.r`   |       0        | @f$ a_{0} @f$ |
+|     `A.y`   |       1        | @f$ a_{1} @f$ |
+|     `A.g`   |       1        | @f$ a_{1} @f$ |
+|     `A.z`   |       2        | @f$ a_{2} @f$ |
+|     `A.b`   |       2        | @f$ a_{2} @f$ |
 </center>
@par
 This operator can provide more immediately readability when working with
@@ -1507,12 +1490,12 @@ expressions that focus on colour or Cartesian coordinate systems.
 The <b>`[]`</b> operator is valid for both vectors and matrices. It has the form:
@par
 <table><tr>
-<td bgcolor="#ffc4c4"> @a `container` <b>[</b><em>`exp`</em><b>]</b> </td>
+<td> `container` <b>[</b><em>`exp`</em><b>]</b> </td>
 </tr></table>
@par
-@a `exp` is expected to evaluate to a value which is castable to an integer
+`exp` is expected to evaluate to a value which is castable to an integer
 following the @ref axopassignment "assignment casting" rules and falls within
-the valid size of the @a `container`. <b>No bounds checking is performed</b>.
+the valid size of the `container`. <b>No bounds checking is performed</b>.
@par
 As matrices are also an array of contigous values, they can be accessed in the
 same way. This can make some operations much simplier; a typical example being
@@ -1533,16 +1516,16 @@ for (int i = 0; i < 16; ++i) b[i] = i;
 The <b>`[,]`</b> operator is only valid for matrices. It has the form:
@par
 <table><tr>
-<td bgcolor="#ffc4c4"> @a `matrix` <b>[</b><em>`exp1`</em>, <em>`exp2`</em><b>]</b> </td>
+<td> `matrix` <b>[</b><em>`exp1`</em>, <em>`exp2`</em><b>]</b> </td>
 </tr></table>
@par
-@a `exp1` and @a `exp2` are expected to evaluate to a value which is castable to
+`exp1` and `exp2` are expected to evaluate to a value which is castable to
 an integer following the @ref axopassignment "assignment casting" rules and
-falls within the valid size of the @a `matrix` dimensions. <b>No bounds checking
+falls within the valid size of the `matrix` dimensions. <b>No bounds checking
 is performed</b>. In general it is more readable to use this operator when
-accessing specific elements, where @b `exp1` is the row index and @b `exp2` is
+accessing specific elements, where `exp1` is the row index and `exp2` is
 the column index. The tables below show the different access patterns for
-@b `3x3` and @b `4x4` matrices.
+`3x3` and `4x4` matrices.
@par
 <!-- A -->@code{.c}
 mat3f a;
@@ -1555,22 +1538,26 @@ for (int i = 0; i < 3; ++i)
 <tr style="text-align:center">
 <td style="border:none" colspan=3>
@f[
+{
 A=\begin{bmatrix}
 a_{00}\quad a_{01}\quad a_{02}\\
 a_{10}\quad a_{11}\quad a_{12}\\
 a_{20}\quad a_{21}\quad a_{22}\\
 \end{bmatrix}
+}
@f]
 </td>
 <td style="border:none"></td>
 <td style="border:none" colspan=3>
@f[
+{
 A=\begin{bmatrix}
 a_{00}\quad a_{01}\quad a_{02}\quad a_{03}\\
 a_{10}\quad a_{11}\quad a_{12}\quad a_{13}\\
 a_{20}\quad a_{21}\quad a_{22}\quad a_{23}\\
 a_{30}\quad a_{31}\quad a_{32}\quad a_{33}\\
 \end{bmatrix}
+}
@f]
 </td>
 </tr>
@@ -1584,112 +1571,62 @@ a_{30}\quad a_{31}\quad a_{32}\quad a_{33}\\
 <th>Result</th>
 </tr>
 <tr style="text-align:center">
-<td><code>A[0]</code></td>
-<td><code>A[0,0]</code></td>
-<td>@f$ a_{00} @f$</td>
-<td><code>A[0]</code></td>
-<td><code>A[0,0]</code></td>
-<td>@f$ a_{00} @f$</td>
+<td>`A[0]`</td><td>`A[0,0]`</td><td>@f$ a_{00} @f$</td>
+<td>`A[0]`</td><td>`A[0,0]`</td><td>@f$ a_{00} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[1]</code></td>
-<td><code>A[0,1]</code></td>
-<td>@f$ a_{01} @f$</td>
-<td><code>A[1]</code></td>
-<td><code>A[0,1]</code></td>
-<td>@f$ a_{01} @f$</td>
+<td>`A[1]`</td><td>`A[0,1]`</td><td>@f$ a_{01} @f$</td>
+<td>`A[1]`</td><td>`A[0,1]`</td><td>@f$ a_{01} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[2]</code></td>
-<td><code>A[0,2]</code></td>
-<td>@f$ a_{02} @f$</td>
-<td><code>A[2]</code></td>
-<td><code>A[0,2]</code></td>
-<td>@f$ a_{02} @f$</td>
+<td>`A[2]`</td><td>`A[0,2]`</td><td>@f$ a_{02} @f$</td>
+<td>`A[2]`</td><td>`A[0,2]`</td><td>@f$ a_{02} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[3]</code></td>
-<td><code>A[1,0]</code></td>
-<td>@f$ a_{10} @f$</td>
-<td><code>A[3]</code></td>
-<td><code>A[0,3]</code></td>
-<td>@f$ a_{03} @f$</td>
+<td>`A[3]`</td><td>`A[1,0]`</td><td>@f$ a_{10} @f$</td>
+<td>`A[3]`</td><td>`A[0,3]`</td><td>@f$ a_{03} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[4]</code></td>
-<td><code>A[1,1]</code></td>
-<td>@f$ a_{10} @f$</td>
-<td><code>A[4]</code></td>
-<td><code>A[1,0]</code></td>
-<td>@f$ a_{10} @f$</td>
+<td>`A[4]`</td><td>`A[1,1]`</td><td>@f$ a_{10} @f$</td>
+<td>`A[4]`</td><td>`A[1,0]`</td><td>@f$ a_{10} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[5]</code></td>
-<td><code>A[1,2]</code></td>
-<td>@f$ a_{12} @f$</td>
-<td><code>A[5]</code></td>
-<td><code>A[1,1]</code></td>
-<td>@f$ a_{11} @f$</td>
+<td>`A[5]`</td><td>`A[1,2]`</td><td>@f$ a_{12} @f$</td>
+<td>`A[5]`</td><td>`A[1,1]`</td><td>@f$ a_{11} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[6]</code></td>
-<td><code>A[2,0]</code></td>
-<td>@f$ a_{20} @f$</td>
-<td><code>A[6]</code></td>
-<td><code>A[1,2]</code></td>
-<td>@f$ a_{12} @f$</td>
+<td>`A[6]`</td><td>`A[2,0]`</td><td>@f$ a_{20} @f$</td>
+<td>`A[6]`</td><td>`A[1,2]`</td><td>@f$ a_{12} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[7]</code></td>
-<td><code>A[2,1]</code></td>
-<td>@f$ a_{21} @f$</td>
-<td><code>A[7]</code></td>
-<td><code>A[1,3]</code></td>
-<td>@f$ a_{13} @f$</td>
+<td>`A[7]`</td><td>`A[2,1]`</td><td>@f$ a_{21} @f$</td>
+<td>`A[7]`</td><td>`A[1,3]`</td><td>@f$ a_{13} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[8]</code></td>
-<td><code>A[2,2]</code></td>
-<td>@f$ a_{22} @f$</td>
-<td><code>A[8]</code></td>
-<td><code>A[2,0]</code></td>
-<td>@f$ a_{20} @f$</td>
+<td>`A[8]`</td><td>`A[2,2]`</td><td>@f$ a_{22} @f$</td>
+<td>`A[8]`</td><td>`A[2,0]`</td><td>@f$ a_{20} @f$</td>
 </tr> <!-- ******************** MAT4 DATA ONLY ******************** -->
 <tr style="text-align:center">
-<td style="border:none;" colspan="3" rowspan="7"></td>
-<td><code>A[9]</code></td>
-<td><code>A[2,1]</code></td>
-<td>@f$ a_{21} @f$</td>
+<td style="border:right;" colspan="4" rowspan="7"></td>
+<td>`A[9]`</td><td>`A[2,1]`</td><td>@f$ a_{21} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[10]</code></td>
-<td><code>A[2,2]</code></td>
-<td>@f$ a_{22} @f$</td>
+<td>`A[10]`</td><td>`A[2,2]`</td><td>@f$ a_{22} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[11]</code></td>
-<td><code>A[2,3]</code></td>
-<td>@f$ a_{23} @f$</td>
+<td>`A[11]`</td><td>`A[2,3]`</td><td>@f$ a_{23} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[12]</code></td>
-<td><code>A[3,0]</code></td>
-<td>@f$ a_{30} @f$</td>
+<td>`A[12]`</td><td>`A[3,0]`</td><td>@f$ a_{30} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[13]</code></td>
-<td><code>A[3,1]</code></td>
-<td>@f$ a_{31} @f$</td>
+<td>`A[13]`</td><td>`A[3,1]`</td><td>@f$ a_{31} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[14]</code></td>
-<td><code>A[3,2]</code></td>
-<td>@f$ a_{32} @f$</td>
+<td>`A[14]`</td><td>`A[3,2]`</td><td>@f$ a_{32} @f$</td>
 </tr>
 <tr style="text-align:center">
-<td><code>A[15]</code></td>
-<td><code>A[3,3]</code></td>
-<td>@f$ a_{33} @f$</td>
+<td>`A[15]`</td><td>`A[3,3]`</td><td>@f$ a_{33} @f$</td>
 </tr>
 </table>

@@ -1703,22 +1640,27 @@ section. They have the following forms:
 <tr><th>Operator Name</th><th>Operator Syntax</th><th>Returns</th></tr>
 <tr style="text-align:left">
 <td>Call / Explicit Cast</td>
-<td bgcolor="#ffc4c4"> <em>`a`</em><b>`(...)`</b></td>
+<td> `a(...)` </td>
 <td>Returns the result of a function call or inbuilt explicit cast</td>
 </tr>
 <tr style="text-align:left">
 <td>Comma</td>
-<td bgcolor="#ffc4c4"> <em>`a`</em><b>,</b>&nbsp; <em>`b`</em></td>
-<td>Returns the value of @b `b` after chained evaluation</td>
+<td> `a, b, ...` </td>
+<td>Returns the value of the last expression after chained evaluation</td>
 </tr>
 <tr style="text-align:left">
 <td>Ternary Conditional</td>
-<td bgcolor="#ffc4c4"> <em>`a`</em>&nbsp;<b>`?`</b>&nbsp;<em>`b`</em>&nbsp;<b>`:`</b><em>`c`</em></td>
-<td>@b `b` if @b `a` is @b true, @b `c` otherwise.</td>
+<td> `a ? b : c` </td>
+<td> `b` if `a` is true, `c` otherwise.</td>
+</tr>
+<tr style="text-align:left">
+<td>Ternary Conditional</td>
+<td> `a ? : c` </td>
+<td> `a` if `a` is true, `c` otherwise.</td>
 </tr>
 <tr>
 <td>Vector/Matrix Initializer</td>
-<td bgcolor="#ffc4c4"> <b>`{`</b>&nbsp;<em> `a,&nbsp; `</em> `b&nbsp; `<b>`...&nbsp;}`</b> </td>
+<td>`{ a, b, ... }`</td>
 <td>Returns a temporary vector or matrix</td>
 </tr>
 </table>
@@ -1728,10 +1670,10 @@ The function call operator provides the ability to invoke functions @b or invoke
 a supported explicit cast. It has the form:
@par
 <table><tr>
-<td bgcolor="#ffc4c4"> <em> `func(a, b, c, ...)` </em></td>
+<td> `func(a, b, c, ...)` </td>
 </tr></table>
@par
-Where @b `func` is a function name @b or a valid explicit cast typename. Explicit
+Where `func` is a function name @b or a valid explicit cast typename. Explicit
 casts are only supported for @ref axscalars "scalar types". For example:
 <!-- A -->@code{.c}
 int a = int(1.1f);
@@ -1740,7 +1682,7 @@ a = int(b.x);
@endcode
@par
 Whilst explicit casts take only a single argument, the operator itself supports
-multiple or empty arguments for function calls. Whe multiple arguments are
+multiple or empty arguments for function calls. When multiple arguments are
 provided, they are parsed from first to last and processed in reverse order.
@b However, the order of runtime evaluation is @b unspecified i.e. they may
 evaluate in any order. Each function parameter is initialized with its
@@ -1753,10 +1695,10 @@ functions.
 The Comma operator has the form:
@par
 <table><tr>
-<td bgcolor="#ffc4c4"><em>`a`</em><b>`,` </b><em>`b`</em></td>
+<td>`a, b, ...`</td>
 </tr></table>
@par
-Where @b `a` and @b `b` are expressions. Chained expressions evaluate each
+Where `a` and `b` are expressions. Chained expressions evaluate each
 expression in syntactical order (first to last), discarding any return value
 except for the last expression, which is returned from the entire list. Note that
 although return values for any expression but the last are discarded, their side
@@ -1780,32 +1722,33 @@ The ternary conditional operator can be thought of as an in-line @ref axbranchin
 a single expression that will be returned. It has the form:
@par
 <table><tr>
-<td bgcolor="#ffc4c4"><em>`a`</em>&nbsp;<b>`?`</b>&nbsp;<em>`b`</em>&nbsp;<b>`:`</b><em>`c`</em></td>
+<td>`a ? b : c`</td>
 </tr></table>
-Where @b `a` is the condition, evaluated and converted to @b bool, and @b `b` and @b
-`c` are expressions of the same or implicit-castable types. @b `b` is evaluated and
-returned if the condition is @b true, @b `c` is evaluated and returned if the
-condition is @b false. Only the expression out of @b `b` and @b `c` that is
-returned will be evaluated. Expressions with no return value (a.k.a @b `void`)
-are supported as long as @b `b` and @b `c` are both of this type.
@par
-Also supported is the 'Elvis' operator, <b>`a` `?` `:` `c`</b> (or `?:`).
-Here, @b `a` is evaluated once, and if when converted to @b bool is @b true, @b `a` is
-returned, otherwise @b `c` is evaluated and returned. In this case, @b `a` and @b
+Where `a` is the condition, evaluated and converted to @b bool, and `b` and
+`c` are expressions of the same or implicit-castable types. `b` is evaluated and
+returned if the condition is @b true, `c` is evaluated and returned if the
+condition is @b false. Only the expression out of `b` and `c` that is
+returned will be evaluated. Expressions with no return value (a.k.a `void`)
+are supported as long as `b` and `c` are both of this type.
+@par
+Also supported is the 'Elvis' operator, `a ? : c` (or `?:`).
+Here, `a` is evaluated once, and if when converted to @b bool is @b true, `a` is
+returned, otherwise `c` is evaluated and returned. In this case, `a` and
 `c` must be the same or implicit-castable types, and both implicit-castable to @b bool.

@anchor axvecmatinit
@par Vector/Matrix Initializer
@ref axvecmats "Vectors and matrices" can be represented as local variables (e.g.
-@b `vec3f`), external attributes or parameters (e.g. @b `vec3f\@attrib`) or as
+`vec3f`), external attributes or parameters (e.g. `vec3f@attrib`) or as
 temporary values using the vector/matrix initializer syntax. This operator has
 the form:
@par
 <table><tr>
-<td bgcolor="#ffc4c4"> <b>`{`</b>&nbsp; <em>`a,&nbsp; b&nbsp; `</em><b>`...&nbsp;}`</b> </td>
+<td> `{ a, b, ... }`</td>
 </tr></table>
@par
-Where @b `a` and @b `b` are expressions returning a scalar value. When this
+Where `a` and `b` are expressions returning a scalar value. When this
 operator is invoked with a specific number of arguments, a vector or matrix is
 implicitly created with a copy of those arguments which can be assigned to an
 existing container of matching size. Expression in the operator are evaluated
@@ -1833,7 +1776,7 @@ The temporary container evaluates all expressions and uses the @b highest rankin
 type (as defined by @ref axtypeprecedence "arithmetic type precedence") to infer
 the element type of the container, implicitly casting all other types to that
 type where necessary. In the above example, the temporary created forms a
-@b `mat3d`, before being implicitly cast to a @b `mat3f` due to the assignment.
+`mat3d`, before being implicitly cast to a `mat3f` due to the assignment.


@subsection axopprecedence Operator Precedence
@@ -2004,10 +1947,10 @@ can be used to assign any scalar a value.
@par
 <table>
 <tr><th>Type</th><th>Literal Tokens</th></tr>
-<tr><td>`bool`</td><td>Tokens @b `true` and @b `false`</td></tr>
+<tr><td>`bool`</td><td>Tokens `true` and `false`</td></tr>
 <tr><td>`int32`</td><td>No suffix, automatically infered from integral literals</td></tr>
-<tr><td>`int64`</td><td>The letter @b `l` e.g. @code int64 a = 10l; @endcode</td></tr>
-<tr><td>`float`</td><td>The letter @b `f` e.g. @code float a = 0.0f; @endcode</td></tr>
+<tr><td>`int64`</td><td>The letter `l` e.g. @code int64 a = 10l; @endcode</td></tr>
+<tr><td>`float`</td><td>The letter `f` e.g. @code float a = 0.0f; @endcode</td></tr>
 <tr><td>`double`</td><td>No suffix, automatically infered from floating point literals.</td></tr>
 <tr><td>`string`</td><td>Character strings wrapped in double quotes @b \" @b \"</td></tr>
 </table>
@@ -2025,7 +1968,6 @@ can be used to assign any scalar a value.
@subsection axattribaccess Attribute Access
 <!-- codecs not supported, how are points and volumes sampled (point vs volume) -->

-<!--
 Further, the active states of
 OpenVDB voxels change which voxels AX will process. The defined behavior for all
 cases is shown below.
@@ -2064,8 +2006,6 @@ foreach(ActiveVoxel in surface)
    DensityIndexSpaceCoord = density.transformToIndex(SurfaceWorldSpaceCoord);
    ActiveVoxel = density.getValue(DensityIndexSpaceCoord);
@endcode
-done()
-->


@subsection axexternalaccess External Parameter Access
@@ -2108,6 +2048,4 @@ AX Roadmap as a near future feature.

@section axvexsupport AX VEX Support

-</div>
-
 */
@@ -1,9 +1,8 @@
 /**
-<div style="width:800px;text-align:justify;">

-@page axcplusplus OpenVDB AX
+@page axcplusplus AX C++ Documentation

-@section axcdocs AX C++ Documentation
+@section axcdocs Intro
@par
 The following documentation focuses on the C++ design and API of OpenVDB AX,
@b not the front end AX language. For the latter, see the
@@ -27,7 +26,6 @@ API doxygen comments or contact us for more information.
        - @ref vdbaxcompiler
    - @ref vdbaxbinary
 - @ref vdbaxtoaxtypes
- @ref vdbaxextend

 <hr>

@@ -118,7 +116,7 @@ parser is accessed through the @b ast module.
@par
 The @b ast module provides:
 - The complete implemenation of an @ref AST.h "AX abstract syntax tree"
- - The openvdb::ax::ast::parse() method
+ - The @vdblink::ax::ast::parse ax::ast::parse()@endlink method
 - The @vdblink::ax::ast::Visitor AST Visitor@endlink
 - AST scanners and tools
@par
@@ -128,7 +126,7 @@ The complete AST hierarchy is visualized by the @vdblink::ax::ast::Node doxygen
 class graph for an openvdb::ax::ast::Node@endlink, which represents the top most
 node in the AST. The AST API exists in a nested namespace (`openvdb::ax::ast`)
 and can be used by clients for more granular control and modifications to an AST.
-Importantly, the AST API contains the openvdb::ax::ast::parse() method, which
+Importantly, the AST API contains the `openvdb::ax::ast::parse()` method, which
 invokes the C/C++ functions built by the grammar to iteratively construct an
 AST from a provided character string.
@par
@@ -139,7 +137,7 @@ branches. All these methods use the @vdblink::ax::ast::Visitor AX visitor
 framework@endlink, a visitor pattern traversal class strongly influenced by
 <A HREF="https://clang.llvm.org/doxygen/classclang_1_1RecursiveASTVisitor.html">
 Clang's RecursiveVisitor</A>. Clients can use the AX visitor to implement their
-own AST analysis or modifications (see @ref vdbaxextend).
+own AST analysis or modifications.

@subsubsection vdbaxastscanners Scanners
@par
@@ -285,11 +283,11 @@ AX ASTs to verify access patterns of your data.
@subsubsection vdbaxcompilercustomdata Custom Data
@par
 The AX language provides two main tokens to access memory which has not been
-allocated by AX itself. The first token, @b \@ (the "at" symbol), is designed to
+allocated by AX itself. The first token, `@` (the "at" symbol), is designed to
 provide read/write access to geometry attributes. The second token is the dollar
-character @b $. This allows <b>read only</b> access to custom data that can be
+character `$`. This allows <b>read only</b> access to custom data that can be
 modified in C++ i.e. outside the scope of an AX program. In the same way as the
-@b \@attribute syntax, the value of this data can change without the need for AX
+`@attribute` syntax, the value of this data can change without the need for AX
 to re-compile the program. It is, however, solely on the C++ clients of AX to
 make sure that this data is setup correctly when integrating AX into their
 applications.
@@ -487,10 +485,9 @@ Vec4IGrid::registerGrid();
@endcode

 <hr>
-@section vdbaxextend Extending OpenVDB AX
+<!--@section vdbaxextend Extending OpenVDB AX-->
 <!--@subsection vdbaxcustomtraversal Custom AST Traversal-->
 <!--@subsection vdbaxcustomtraversal Designing an Executable-->
 <!--@subsection vdbaxcustomtraversal Geometry Callbacks-->
 <!--@subsection vdbaxcustomtraversal Custom Code Generation-->
-</div>
 */
@@ -1,5 +1,4 @@
 /**
-<div style="width:800px;text-align:justify;">

@page axexamples AX Code Examples

@@ -30,12 +29,12 @@ These examples demonstrate how to use AX on OpenVDB points grids.
@subsection axexamplepointbasic Basic point attributes
@par
 Below is a small example which demonstrates working with a few point attributes.
-The @ symbol is the identifier for an AX attribute. The type of each attribute
-is specified before the @ symbol and the name is specified afterwards. For
-example: @b `int\@count` would imply an @b integer attribute called @b count.
+The `@` symbol is the identifier for an AX attribute. The type of each attribute
+is specified before the `@` symbol and the name is specified afterwards. For
+example: `int@count` would imply an @b integer attribute called @b count.
@par
 In this example there are three point attributes: a @b float attribute @b speed,
-a @b vec3f float attribute @b velocity and a @b vec3f float attribute @b colour.
+a `vec3f` float attribute @b velocity and a `vec3f` float attribute @b colour.
@par
 This snippet uses the functions @ref axlength "length" and @ref axfit "fit" to
 give points a colour between black and white based on their speed.
@@ -68,8 +67,8 @@ special attribute.
@subsection axexamplepointdelete Deleting Points
@par
 This example demonstrates a simple way of removing points from a VDB points grid.
-It uses a combination of @ref axrand and @ref axdeletepoint to randomly delete
-roughly half of the points in the input VDB points grid.
+It uses a combination of @ref axrand "rand()" and @ref axdeletepoint "deletepoint()"
+to randomly delete roughly half of the points in the input VDB points grid.
@par
 <!-- P -->@code{.c}
 float threshold = 0.5f;
@@ -83,7 +82,7 @@ if (rand(int64@id) > threshold) {
@endcode
@par
 The @b id attribute in this example is assumed to be a unique @b integer for
-each point in the VDB points grid such that @ref axrand will receive a
+each point in the VDB points grid such that @ref axrand "rand()" will receive a
 different seed value per point producing a good distribution of random values.

@subsection axexamplepointdrag Applying Drag
@@ -114,7 +113,7 @@ v@P += v@v * dt;
@endcode
@par
 The final AX command, `v@P += v@v * dt;` writes to the position attribute of
-the points @b `v\@P`. Whenever position is written to in this way, the AX
+the points `v@P`. Whenever position is written to in this way, the AX
 `PointExecutable` will move (re-bucket) points in PointDataGrids.

@subsection axexamplepointcurlnoise Curl Noise
@@ -198,12 +197,12 @@ These examples demonstrate how to use AX on VDB volume grids.
@subsection axexamplevolumeclamp Volume Clamping
@par
 Below is a small example of reading adn writing to voxels within a VDB volume.
-As with points, the @ symbol specifies an access to a named volume, with the
+As with points, the `@` symbol specifies an access to a named volume, with the
 type of each volume preceding it and the name appearing as a suffix. For
 example: `float@density` would imply a @b float volume called @b density.
@par
-This snippet uses @ref axclamp function to constrain the @b density attribute
-between zero and one.
+This snippet uses @ref axclamp "clamp()" function to constrain the @b density
+attribute between zero and one.
@par
 <!-- V -->@code{.c}
 float@density = clamp(float@density, 0.f, 1.f);
@@ -234,10 +233,10 @@ vec3f@vel += dt * massInverse * vec3f@force;
@par
 Here we demonstrate how AX can be used to blend two volumes together. Note that
 only the volumes which are written to are executed over. In the below code,
-only @b `\@surface_a` is written to. This means that the final blended result,
+only `@surface_a` is written to. This means that the final blended result,
 stored in `surface_a`, will only be updated in `surface_a`'s topology. Should
-we want a combined blend of @b non @b overlapping areas of both surface a and b,
-we would need to activate a's topology with respect to b first.
+we want a combined blend of @b non @b overlapping areas of both surface `a` and
+`b`, we would need to activate `a`'s topology with respect to `b` first.
@par
 <!-- V -->@code{.c}
 // time constants for each run.
@@ -263,5 +262,4 @@ mask = clamp(mask, 0.0f, 1.0f);
@surface_a = a + (b-a) * mask;
@endcode

-</div>
 */
@@ -1,5 +1,4 @@
 /**
-<div style="width:800px;text-align:justify;">

@page axfunctionlist AX Supported Functions

@@ -721,7 +720,7 @@ double(int64 seed);
@anchor axrand32
@par rand32
 Creates a random number based on the provided 32 bit seed. The number will be in the range of 0 to
-1. The same number is produced for the same seed. NOTE: This function does not share the same
+1\. The same number is produced for the same seed. NOTE: This function does not share the same
 random number generator as rand(). rand32() may provide a higher throughput on some
 architectures, but will produce different results to rand(). NOTE: If rand32 is called without
 a seed the previous state of the random number generator is advanced for the currently
@@ -881,6 +880,4 @@ int64(int64 dividend; int64 divisor);
 int32(int32 dividend; int32 divisor);
@endcode

-
-</div>
 */
@@ -1,5 +1,4 @@
 /**
-<div style="width:800px;text-align:justify;">

@page openvdbax OpenVDB AX

@@ -41,7 +40,6 @@ provided:
@par
 - @subpage axexamples "Code Examples" for examples of OpenVDB AX in action.
 - @subpage axfunctionlist "AX Functions" for a list of all available AX functions.
- - @subpage build "Building OpenVDB" and @subpage dependencies "OpenVDB Dependencies" for information on building OpenVDB and <b>OpenVDB AX</b>.
+ - @ref build "Building OpenVDB" and @ref dependencies "OpenVDB Dependencies" for information on building OpenVDB and <b>OpenVDB AX</b>.

-</div>
 */
@@ -238,8 +238,8 @@ it's a good idea to read the above section on

 DCC      | Supported Version | OpenVDB ABI |
 -------- | ----------------- | ----------- |
-Houdini  | 20.0              | 10          |
 Houdini  | 20.5              | 11          |
+Houdini  | 21.0              | 12          |

@subsection buildBuildHou Building Against Houdini

@@ -257,12 +257,10 @@ CMake          | Cross-platform family of tools designed to help build software
 C++17 Compiler | Matching Houdini compiler and version                           | All                |
 Boost          | Components: iostreams                                           | All                |
 GoogleTest     | A unit testing framework module for C++                         | Unit Tests         |
-CppUnit        | A unit testing framework module for C++                         | Unit Tests (AX)    |
 GLFW           | Simple API for OpenGL development                               | OpenVDB View       |
 Doxygen        | Documentation generation from C++                               | Documentation      |
 Log4cplus      | An optional dependency for improved OpenVDB Logging             | Optional (All)     |
 nanobind       | C++/python bindings                                             | Optional (Python)  |
-NumPy          | Scientific computing with Python                                | Optional (Python)  |
 LLVM           | Target-independent code generation                              | OpenVDB AX         |

 At a minimum a matching C++17 compiler and CMake will be required. See
@@ -290,7 +288,7 @@ Required:

 Optional:

-  - @b [CMAKE_INSTALL_PREFIX](https://cmake.org/cmake/help/latest/variable/CMAKE_INSTALL_PREFIX.html)
+  - [CMAKE_INSTALL_PREFIX](https://cmake.org/cmake/help/latest/variable/CMAKE_INSTALL_PREFIX.html)
  =`/path/to/install/dir` # The location you wish to install OpenVDB to. See
 the link for default values.
  - @b OPENVDB_HOUDINI_INSTALL_PREFIX =`/path/to/install/dir` # The location to
@@ -349,14 +347,12 @@ Boost          | Components: iostreams
 ZLIB           | Compression library for disk serialization compression            | All                              |
 Blosc          | Recommended dependency for improved disk compression              | All*                             |
 GoogleTest     | A unit testing framework module for C++                           | Unit Tests                       |
-CppUnit        | A unit testing framework module for C++                           | Unit Tests (AX)                  |
 GLFW           | Simple API for OpenGL development                                 | OpenVDB View                     |
 Doxygen        | Documentation generation from C++                                 | Documentation                    |
 OpenEXR        | EXR serialization support                                         | OpenVDB Render                   |
 Imath          | Used half precision floating points and EXR serialization support | Optional (All)                   |
 Log4cplus      | An optional dependency for improved OpenVDB Logging               | Optional (All)                   |
 nanobind       | C++/python bindings                                               | Optional (Python)                |
-NumPy          | Scientific computing with Python                                  | Optional (Python)                |
 LLVM           | Target-independent code generation                                | OpenVDB AX                       |


@@ -385,7 +381,7 @@ Required:

 Optional:

-  - @b [CMAKE_INSTALL_PREFIX](https://cmake.org/cmake/help/latest/variable/CMAKE_INSTALL_PREFIX.html)
+  - [CMAKE_INSTALL_PREFIX](https://cmake.org/cmake/help/latest/variable/CMAKE_INSTALL_PREFIX.html)
  =`/path/to/install/dir` # The location you wish to install OpenVDB to. See
 the link for default values.
  - @b OPENVDB_MAYA_INSTALL_PREFIX =`/path/to/install/dir` # The location to
@@ -453,7 +449,7 @@ Now you can call CMake by providing the absolute or relative path to the root
 of the OpenVDB source tree. Below are some common options you may want to
 provide:

- @b [CMAKE_INSTALL_PREFIX](https://cmake.org/cmake/help/latest/variable/CMAKE_INSTALL_PREFIX.html)
+- [CMAKE_INSTALL_PREFIX](https://cmake.org/cmake/help/latest/variable/CMAKE_INSTALL_PREFIX.html)
 =`/path/to/install/dir` # The location you wish to install OpenVDB to. See
 the link for default values.
 - @b Xxx_ROOT =`/path/to/Xxx/` # Give CMake a path to where a package has been
@@ -512,7 +508,7 @@ You can add the below CMake snippet to your main `CMakeLists.txt` to bring in
 OpenVDB as a dependency:

@code{.cmake}
-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 list(APPEND CMAKE_MODULE_PATH "/location/of/openvdb/install/lib/cmake/OpenVDB")
 find_package(OpenVDB REQUIRED)
 target_link_libraries(myapp OpenVDB::openvdb)
@@ -583,8 +579,7 @@ bypass these checks with `-D DISABLE_DEPENDENCY_VERSION_CHECKS=ON` if necessary.
 Note that using older or untested dependencies may produce undesired behavior.
 Older versions in particular are not supported by OpenVDB.

-@b Note: Boost will produce a version error in the format `Detected version of
-Boost is too old.`
+@b Note: Boost will produce a version error in the format: `Detected version of Boost is too old.`

@subsection troubleshoot3 CMake warnings/errors in FindBoost.cmake

@@ -605,10 +600,12 @@ OpenVDB uses [imported targets](https://cmake.org/cmake/help/latest/command/add_
 for all its dependencies. For imported Boost compatibility, the following
 versions of CMake are required:

-  - Boost 1.80 requires CMake 3.24.2 or newer.
-  - Boost 1.81 requires CMake 3.25.2 or newer.
  - Boost 1.82 requires CMake 3.27.0 or newer.
  - Boost 1.83 requires CMake 3.27.4 or newer.
+  - Boost 1.84 requires CMake 3.28.2 or newer.
+  - Boost 1.85 requires CMake 3.29.3 or newer.
+  - Boost 1.86 requires CMake 3.30.6, 3.31.1, or newer.
+  - Boost 1.87 requires CMake 3.31.4 or newer.

 Use the above list to identify the version of CMake you require for your version
 of Boost. Note that this may be implemented into the OpenVDB CMake in the future.
@@ -634,9 +631,9 @@ Note that you can provide a value for `CMAKE_MSVC_RUNTIME_LIBRARY` which is a
 native CMake variable to override any CRT selection done by OpenVDB.

 For more information, see the following articles:
-Description of CRT selection: https://docs.microsoft.com/en-us/cpp/build/reference/md-mt-ld-use-run-time-library
-In depth descriptopn of the CRT on Windows: https://docs.microsoft.com/en-us/cpp/c-runtime-library/crt-library-features
-Related errors: https://docs.microsoft.com/en-us/cpp/error-messages/tool-errors/linker-tools-error-lnk2038
+- Description of CRT selection: https://docs.microsoft.com/en-us/cpp/build/reference/md-mt-ld-use-run-time-library
+- In depth descriptopn of the CRT on Windows: https://docs.microsoft.com/en-us/cpp/c-runtime-library/crt-library-features
+- Related errors: https://docs.microsoft.com/en-us/cpp/error-messages/tool-errors/linker-tools-error-lnk2038
 `CMAKE_MSVC_RUNTIME_LIBRARY` docs: https://cmake.org/cmake/help/latest/variable/CMAKE_MSVC_RUNTIME_LIBRARY.html

@subsection troubleshoot5 Unexpected value for the Windows CRT with target build artifacts.
@@ -658,9 +655,9 @@ usage, it is strongly encouraged to match the selection against your desired
 build type.

 For more information, see the following articles:
-Description of CRT selection: https://docs.microsoft.com/en-us/cpp/build/reference/md-mt-ld-use-run-time-library
-In depth descriptopn of the CRT on Windows: https://docs.microsoft.com/en-us/cpp/c-runtime-library/crt-library-features
-Related errors: https://docs.microsoft.com/en-us/cpp/error-messages/tool-errors/linker-tools-error-lnk2038
+- Description of CRT selection: https://docs.microsoft.com/en-us/cpp/build/reference/md-mt-ld-use-run-time-library
+- In depth descriptopn of the CRT on Windows: https://docs.microsoft.com/en-us/cpp/c-runtime-library/crt-library-features
+- Related errors: https://docs.microsoft.com/en-us/cpp/error-messages/tool-errors/linker-tools-error-lnk2038
 `CMAKE_MSVC_RUNTIME_LIBRARY` docs: https://cmake.org/cmake/help/latest/variable/CMAKE_MSVC_RUNTIME_LIBRARY.html

@subsection troubleshoot6 error LNK2038: mismatch detected for 'RuntimeLibrary'
@@ -2,9 +2,322 @@

@page changes Release Notes

-@htmlonly <a name="v12_0_2_changes"></a>@endhtmlonly
+@htmlonly <a name="v13_0_1_changes"></a>@endhtmlonly
@par
-<B>Version 12.0.2</B> - <I>In development</I>
+<B>Version 13.0.1</B> - <I>In Development</I>
+
+@htmlonly <a name="v13_0_0_changes"></a>@endhtmlonly
+@par
+<B>Version 13.0.0</B> - <I>November 3, 2025</I>
+
+@par
+<BLOCKQUOTE>
+This version introduces ABI changes relative to older major releases, so to
+preserve ABI compatibility it might be necessary to define the macro
+<TT>OPENVDB_ABI_VERSION_NUMBER=</TT><I>N</I>, where, for example,
+<I>N</I> is 11 for Houdini&nbsp;20.5 and 12 for Houdini&nbsp;21.0.
+</BLOCKQUOTE>
+
+@par
+<BLOCKQUOTE>
+Support for reading <TT>.vdb</TT> files written prior to VDB&nbsp;1.0 has been
+removed. To read older <TT>.vdb</TT> files, please use VDB&nbsp;12.x or older
+and re-write them, so that they use the newer file format and can be read in
+VDB&nbsp;13+.
+</BLOCKQUOTE>
+
+@par
+Highlights:
+- OpenVDB now includes in-memory support for Half Grids and the file format
+  has been extended to add I/O serialization for half grids. More work is
+  planned on adapting algorithms when using Half Grids for better performance
+  and improved precision.
+- NanoVDB is no longer limited to static applications, such as rendering, but
+  can increasingly be applied to problems that involve dynamic topology, such
+  as level set tracking, grid building, morphology operations and merging of
+  grids. This is due to the fact that we keep adding more and more tools to
+  NanoVDB that facilitates tree modification. However, unlike OpenVDB, these
+  operations generate new NanoVDB grids (as oppose to in-palace modifications
+  in OpenVDB), yet are typically much faster since they are performed on the
+  GPU. Many of these exciting new features are the foundation of fVDB.
+
+@par
+OpenVDB:
+- New features:
+  - Added native registration of HalfGrid type.
+  - Added customization of the output transformation attribute for
+    @vdblink::points::pca points::pca@endlink. Can be output as a quaternion
+    or as a combined affine transform with the scale/stretch ommitted.
+  - Added support for input quaternions attributes and affine transformations
+    (that include scale) in
+    @vdblink::points::rasterizeSdf points::rasterizeSdf@endlink for
+    ellipsoidal rasterizations.
+
+- API Changes:
+  - Added @c openvdb::tree::LeafBuffer::storage()
+    for direct const access to the raw data buffer when using the bool LeafNode.
+  - Removed Grid::evalMinMax(), use
+    @vdblink::tools::minMax tools::minMax@endlink for superior threaded
+    performance.
+  - Removed points::rasterize() that took a custom filter and interrupter.
+  - Removed LeafNodeBool TopologyCopy constructor in favor of using explicit
+    mask/buffer constructors.
+  - Removed @vdblink::tree::ValueAccessor ValueAccessor@endlink NodeT0,
+    NodeT1, NodeT2 aliases, use NodeTypeAtLevel instead.
+  - Add @vdblink::tree::InternalNode::setValueMaskUnsafe() InternalNode::setValueMaskUnsafe()@endlink
+    to write a value mask directly.
+  - LegacyFrustum and RootNodeMask have been removed. These classes were
+    only used for reading old .vdb files, which is no longer supported.
+  - @vdblink::io::writeCompressedValues() io::writeCompressedValues()@endlink
+    now takes a const ValueT* source buffer for improved const-correctness.
+
+- Bug Fixes:
+  - Fix a bug in the
+    @vdblink::tools::LevelSetRayIntersector LevelSetRayIntersector@endlink class
+    where two of the methods were incorrectly adjusting the time even when an
+    intersection was not found.
+  - Fixed an assertion in @vdblink::math::Quat math::Quat@endlink where a
+    vector length check was using too small a tolerance.
+  - Fixed a bug in the unit tests where detecting if assertions were enabled
+    was using @c NDEBUG rather than OpenVDB assertions.
+
+- Build:
+  - Removed the unnecessary @c USE_NUMPY build option for the OpenVDB Python
+    module. All functions which take numpy arrays are now always generated
+    (numpy is not required at compile time for this support).
+  - Bumped minimum supported CMake version to 3.24, and new future minimum to
+    4.0.0.
+  - Bumped minimum supported OpenEXR and Imath versions to 3.2, and new
+    future minimum to 3.3.
+  - Bumped minimum supported Boost version to 1.82, and new future minimum to
+    1.85.
+  - Bumped minimum supported Python version to 3.11
+  - Bumped minimum supported Log4cplus version to 2.0
+  - Bumped minimum supported Houdini version to 20.5
+  - Bumped minimum supported GLFW version to 3.3
+  - Bumped future minimum for TBB to 2021 (oneapi).
+  - Bumped future minimum for nanobind to 2.5.0.
+
+@par
+OpenVDB AX:
+- New:
+  - Added support for LLVM 21.
+
+@par
+NanoVDB:
+- Highlights:
+  - NanoVDB is no longer limited to static applications, such as rendering,
+    but can increasingly be applied to problems that involve dynamic topology,
+    such as level set tracking, grid building, morphology operations and
+    merging of grids. This is due to the fact that we keep adding more and
+    more tools to NanoVDB that facilitates tree modification. However, unlike
+    OpenVDB, these operations generate new NanoVDB grids (as oppose to
+    in-palace modifications in OpenVDB), yet are typically much faster since
+    they are performed on the GPU. Many of these exciting new features are the
+    foundation of fVDB.
+
+- New features:
+  - Added <TT>nanovdb/tools/cuda/DilateGrid.cuh</TT>, which performs topology
+    dilation on the GPU.
+  - Added <TT>nanovdb/tools/cuda/MergeGrids.cuh</TT>, which merges the topology
+    of two grids on the GPU.
+  - Added <TT>nanovdb/tools/cuda/CoarsenGrid.cuh</TT>, which performs topological
+    coarsening of grids on the GPU.
+  - Added <TT>nanovdb/tools/cuda/RefineGrid.cuh</TT>, which performs topological
+    refinement of grids on the GPU.
+  - Added <TT>nanovdb/tools/cuda/PruneGrid.cuh</TT>, which prunes the topology of
+    a grid using a sidecar mask on the GPU.
+  - Added <TT>nanovdb/util/cuda/Injection.cuh</TT>, which facilitates sidecar
+    data copying - on the GPU - in the overlap of two grids with different
+    topology.
+  - Added Mat classes and other math types to NanoVDB.
+  - Major rewrite of nanovdb::NanoToOpenVDB, which now allows
+    nanovdb::IndexGrids and blind (or sidecar) data to be converted into
+    an openvdb grid.
+  - Added support for LevelSet, FogVolume, and Staggered grid semantics.
+    Added toGridClass and toSemantic to map Grid Blind data semantic to
+    Grid class and vice versa.
+
+- Improvements:
+  - Improvements to mGPU grid building and introduction of a separate test
+    executable for mGPU tests
+  - Added stand-alone function nanovdb::writeUncompressedGrid that allows
+    nanovdb::IndexGrids and blind (or sidecar) data to be written directly
+    to a stream.
+  - @c nanovdb_print now (optionally) prints the blind data count per grid.
+  - Extended nanovdb::GridBlindDataSemantic with nanovdb::GridClass types
+    to support conversion between IndexGrid and regular grids.
+  - Improvements and bugfixes to PNanoVDB.
+
+- API Changes:
+  - Removed nanovdb::ValueIndexMask and nanovdb::ValueOnIndexMask since
+    they are no longer used by fVDB (or elsewhere).
+  - NanoVDB version is now 32.9.0, so the ABI of the grid is preserved
+    but not the API. However, the API changes are very minor!
+
+- Bug Fixes:
+  - Several fixes for compiler warnings.
+  - Fixed error in <TT>nanovdb/io/IO.h</TT> related to missing throws.
+
+@par
+Houdini:
+- Remove support for openvdb_houdini::Interrupter, use
+  openvdb_houdini::HoudiniInterrupter instead.
+- Remove input-stealing via duplicateSourceStealable(), SideFX have since
+  introduced verbification and compiled SOPs to provide zero-copy node
+  evaluation.
+
+
+@htmlonly <a name="v12_1_1_changes"></a>@endhtmlonly
+@par
+<B>Version 12.1.1</B> - <I>September 30, 2025</I>
+
+@par
+OpenVDB AX:
+- Fixes:
+  - Fixed an issue which was causing host specific optimization passes to be
+    omitted when using LLVM 16 and higher.
+
+@par
+Houdini:
+  - Fixed a compilation issue with the OpenVDB Activate SOP with Houdini 21
+  [Reported by \@callen128]
+
+
+@htmlonly <a name="v12_1_0_changes"></a>@endhtmlonly
+@par
+<B>Version 12.1.0</B> - <I>August 8, 2025</I>
+
+@par
+OpenVDB:
+- New features:
+  - New level set constructor tools::createLevelSetCapsule creates a grid
+    containing a narrow-band level set representation of a capsule (tube
+    with constant radius and sphere caps).
+  - New level set constructor tools::createLevelSetTaperedCapsule creates a
+    grid containing a narrow-band level set representation of a tapered
+    capsule (tube with sphere caps and different radii at both ends, or
+    equivalently the convex hull of two spheres with possibly different
+    centers and radii).
+  - New level set constructor tools::createLevelSetTubeComplex creates a grid
+    containing a narrow-band level set representation of a tube complex (a
+    collection of capsules or tapered capsules defined by endpoint coordinates
+    and segment indices).
+  - New level set constructor tools::createLevelSetDilatedMesh creates a grid
+    containing a narrow-band level set representation of a dilated surface
+    mesh (dilated by a radius in all directions).
+  - New class tools::ConvexVoxelizer, used by the above methods, which can be
+    inherited to implement a convex level set constructor.
+  - Introduced anisotropic surfacing tools in points/PointRasterizeSDF.h and
+    principal component analysis (PCA) methods in points/PrincipalComponentAnalysis.h.
+    The latter tool analyses point neighborhoods and computes affine
+    transformations representing elliptical distributions. The prior tool
+    takes rotational and stretch point attributes (optionally computed from
+      the prior PCA methods) and generates anisotropic surfaces.
+
+- Improvements:
+  - Significantly improved the performance of all LeafNode ValueIterators,
+    up to 5x on some platforms and up to 10x when delay loading is enabled.
+    Construction of a ValueIterator from a leaf node now requests the leaf
+    buffers ahead of iteration to avoid potentially expensive API calls.
+  - Improved the performance of PointRasterizeSDF.h tools with larger radii
+    with better bound computation approximations.
+  - Significant performance improvement to delay loaded files - the warning
+    logic which checks the underlying mapped file for the last modified time
+    has been removed, resulting in performance gains upwards of 10x when
+    loading in node data buffers. Note that the check that was removed is
+    entirely redundant on Windows and would only print a warning on other
+    system.
+    [Contributed by jjtParadox]
+  - Added OPENVDB_HAS_INVOKABLE_MEMBER_FUNCTION and OPENVDB_HAS_MEMBER_FUNCTION
+    macros to help determine the presence and compatibility of interface
+    member functions at compile time.
+  - Added OPENVDB_ASSUME macros to mimic builtin assume and C++23 assume
+    attributes.
+  - Improved read and write of PLY files with vdb_tool so they support Vec3d
+    and 64 bit attributes. Changed vdb_tool version number from 10.6.1 to
+    10.7.0.
+
+- API Changes:
+  - Mat3 and Mat4 functions have been marked [[nodiscard]] where it may be
+    confusing if they are operating in place or returning a result, in
+    particular tranpose, invert, adjoint, and timesDiagonal.
+  - Re-worked the API in PointRasterizeSDF.h with the addition of anisotropic
+    surfacing kernels. The new API exposes more concise free functions which
+    take structs of parameters. The old API remains supported but is
+    deprecated.
+  - Deprecated providing a filter or interrupter as arguments to
+    points::rasterize. These can instead be implemented on user transfer
+    schemes.
+  - Transfer schemes provided to points::rasterize can now optionally
+    implement a rasterizePoints callback that takes a range of points per
+    voxel.
+
+- Fixes:
+  - Fixed a bug in points::rasterizeSpheres and points::rasterizeSmoothSpheres
+    which could cause sections of the generated surface to incorrectly be
+    marked as negative interior values.
+  - Fixed a bug in points::rasterizeSmoothSpheres which could result in
+    incomplete surface reconstruction with significantly larger search
+    distances.
+  - Fixed some precision issues in various Matrix methods.
+  - Fixed bugs in vdb_tool that prevented build with newer namespaces in
+    nanovdb.
+
+@par
+OpenVDB AX:
+- Improvements:
+  - Added support for LLVM versions greater than 15. Upgraded CI to test with
+    LLVM versions 16, 17, 18, 19 and 20.
+  - As of this release, LLVM 15 is the minimum required version for OpenVDB
+    AX. Note that there is an issue with LLVM 17 symbols which makes it
+    incompatible with the OpenVDB Python plugin. AS such, LLVM version 18 and
+    greater are the recommended LLVM versions to use, with LLVM 18 being the
+    new future minimum version.
+  - Significant performance improvement to the compilation of AX kernels when
+    using LLVM 18 and greater.
+  - Significant backend refactoring to AX IR generation to support newer
+    versions of LLVM. In particular, parts of the codegen::Function and
+    codegen::FunctionBuilder API require new usage from LLVM 16 due to
+    the LLVM opaque pointer changes. If you're not using the OpenVDB AX
+    Function codegen API then no client changes should be necessary.
+  - Migrated OpenVDB AX unit tests from cppunit to gtest
+    [Contributed by Tim Straubinger]
+
+- Fixes:
+  - Fixed a case of undefined behaviour when explicitly casting variables to
+    the same type as their storage type e.g:
+      double a; double b = double(a);
+  - Fixed an AX compiler error which would produce invalid IR when using 64bit
+    integer variables for matrix array offsetting e.g:
+      int64 a = 0; mat4f b = 1; float c = b[a,a];
+
+@par
+NanoVDB:
+  - Added nanovdb::openToIndexVDB, which can convert an OpenVDB grid into a
+    NanoVDB IndexGrid.
+  - Added new options "--index" and "--onIndex" to the command-line tool
+    nanovdb_convert so it can produce IndexGrids.
+  - Fixed memory leak in DeviceBuffer.
+  - Added constructor for DeviceBuffer from an existing HostBuffer.
+  - Added VoxelBlockManager, an acceleration structure for sequential access
+    and stencil ops over active voxels.
+  - Added helper functions for kernel launches with shared memory.
+  - Added support for CUDA 13.0.
+  - Fix warnings and compilation errors for GCC and Clang.
+  - Fixed inconsistent usage of util::cuda::mallocAsync/freeAsync that
+    affected vGPUs.
+  - Added support for custom memory resources in PointsToGrid.
+  - Fixed race condition in split/mergeGridHandles.
+  - Fixed offset bug in nanovdb::tools::cuda::IndexToGrid related to non-POD
+    build types.
+    [Contributed by Andreas Henne]
+
+@par
+Houdini:
+  - Define OP_InputIdx and OP_OutputIdx for Houdini 21.0 compatibility
+  - Add Houdini 21 style camera interface to support LOP cameras
+  - Include env flag for avoiding stream writing (was backported to 19.5)


@htmlonly <a name="v12_0_1_changes"></a>@endhtmlonly
@@ -38,7 +351,7 @@ NanoVDB:
  - Fixed a call to a non-existent method in GridBuild.h


-@htmlonly <a name="v1_changes"></a>@endhtmlonly
+@htmlonly <a name="v12_0_0_changes"></a>@endhtmlonly
@par
 <B>Version 12.0.0</B> - <I>October 31, 2024</I>

@@ -427,10 +740,10 @@ OpenVDB:
  - Fixed a bug with LeafManager which wouldn't correctly
    initialize its LeafNode array for single level Tree configurations
    i.e. RootNode<LeafNode> (bug introduced in 7.2.0).
-    [Reported by @lanwatch]
+    [Reported by \@lanwatch]
  - Fixed a bug with LeafNodeBool Topology constructor with designated
    on/off values which wouldn't apply them correctly.
-    [Reported by @hozhaoea]
+    [Reported by \@hozhaoea]

@par
 OpenVDB AX:
@@ -2489,13 +2802,13 @@ Improvements:
  util::PagedArray@endlink.
 - Added an @c install_lib build target to the Makefile.
  <I>[Contributed&nbsp;by&nbsp;Double&nbsp;Negative]</I>
- Added @subpage points "documentation" and Cookbook
+- Added @ref points "documentation" and Cookbook
  @ref openvdbPointsHelloWorld "examples" for OpenVDB&nbsp;Points.
  <I>[Contributed&nbsp;by&nbsp;Double&nbsp;Negative]</I>
 - Registration of OpenVDB&nbsp;Points grid and attribute types is now
  handled in @vdblink::initialize() openvdb::initialize@endlink,
-  and @vdblink::points::initialize() points::initialize@endlink
-  and @vdblink::points::uninitialize() points::uninitialize@endlink
+  and @c openvdb::points::initialize()
+  and @c openvdb::points::uninitialize()
  are therefore deprecated.
 - Extended multi-pass I/O to handle a variable number of passes per leaf node.
  <I>[Contributed&nbsp;by&nbsp;Double&nbsp;Negative]</I>
@@ -3213,11 +3526,10 @@ Houdini:
 - @vdblink::tools::volumeToMesh() tools::volumeToMesh@endlink now compiles
  for all grid types but throws an exception if the input grid does not
  have a scalar value type.
- Added a
-  @vdblink::io::File::readGrid(const Name&, const BBoxd&) File::readGrid@endlink
-  overload and @vdblink::GridBase::readBuffers(std::istream&, const CoordBBox&)
-  readBuffers@endlink overloads to the grid, tree and node classes that allow
-  one to specify a bounding box against which to clip a grid while reading it.
+- Added an <TT>openvdb::io::File::readGrid(const Name&, const BBoxd&)</TT>
+  overload and <TT>openvdb::GridBase::readBuffers(std::istream&, const CoordBBox&)</TT>
+  overloads to the grid, tree and node classes that allow one to specify a
+  bounding box against which to clip a grid while reading it.
  For large grids, clipping while reading can result in significantly lower
  memory usage than clipping after reading.
 - Added @vdblink::GridBase::clipGrid() GridBase::clipGrid@endlink, which
@@ -3276,10 +3588,11 @@ Houdini:
 - Added constructors and methods to both
  @vdblink::math::RandInt RandInt@endlink and
  @vdblink::math::Rand01 Rand01@endlink to set and reset the random seed value.
- Added convenience methods for
-  @vdblink::math::Transform::indexToWorld(const BBoxd&) const transforming@endlink
-  @vdblink::math::Transform::worldToIndex(const BBoxd&) const bounding@endlink
-  @vdblink::math::Transform::worldToIndexCellCentered(const BBoxd&) const boxes@endlink
+- Added convenience methods
+  <TT>openvdb::math::Transform::indexToWorld(const BBoxd&)</TT>,
+  <TT>openvdb::math::Transform::worldToIndex(const BBoxd&)</TT> and
+  <TT>openvdb::math::Transform::worldToIndexCellCentered(const BBoxd&)</TT>
+  for transforming bounding boxes
  to @vdblink::math::Transform math::Transform@endlink.
 - @c vdb_view is now compatible with both GLFW&nbsp;2 and GLFW&nbsp;3.
 - Made many small changes to address type conversion and other warnings
@@ -3299,9 +3612,9 @@ Houdini:

@par
 API changes:
- The addition of a
-  @vdblink::GridBase::readBuffers(std::istream&, const CoordBBox&)
-  GridBase::readBuffers@endlink virtual function overload and the
+- The addition of an
+  <TT>openvdb::GridBase::readBuffers(std::istream&, const CoordBBox&)</TT>
+  virtual function overload and the
  @vdblink::GridBase::clip() GridBase::clip@endlink
  @vdblink::GridBase::readNonresidentBuffers()
  GridBase::readNonresidentBuffers@endlink and
@@ -3406,16 +3719,15 @@ Houdini:
  extractSparseTreeWithMask@endlink, which copies data from the index-space
  intersection of a sparse tree and a dense input grid.
 - Added copy constructors to the
-  @vdblink::Grid::Grid(const Grid<OtherTreeType>&) Grid@endlink,
-  @vdblink::tree::Tree::Tree(const Tree<OtherRootType>&) Tree@endlink,
-  @vdblink::tree::RootNode::RootNode(const RootNode<OtherChildType>&)
-  RootNode@endlink,
-  @vdblink::tree::InternalNode::InternalNode(const InternalNode<OtherChildNodeType, Log2Dim>&)
-  InternalNode@endlink and
-  @vdblink::tree::LeafNode::LeafNode(const LeafNode<OtherValueType, Log2Dim>&) LeafNode@endlink
+  <TT>openvdb::Grid::Grid(const Grid&lt;OtherTreeType&gt;&)</TT>,
+  <TT>openvdb::tree::Tree::Tree(const Tree&lt;OtherRootType&gt;&)</TT>,
+  <TT>openvdb::tree::RootNode::RootNode(const RootNode&lt;OtherChildType&gt;&)</TT>,
+  <TT>openvdb::tree::InternalNode::InternalNode(const InternalNode&lt;OtherChildNodeType, Log2Dim&gt;&)</TT>
+  and
+  <TT>openvdb::tree::LeafNode::LeafNode(const LeafNode&lt;OtherValueType, Log2Dim&gt;&)</TT>
  classes, and an assignment operator overload to
-  @vdblink::tree::RootNode::operator=(const RootNode<OtherChildType>&)
-  RootNode@endlink, that allow the source and destination to have different
+  <TT>openvdb::tree::RootNode::operator=(const RootNode&lt;OtherChildType&gt;&)</TT>,
+  that allow the source and destination to have different
  value types.
 - Modified @vdblink::tree::Tree::combine2() Tree::combine2@endlink to permit
  combination of trees with different value types.
@@ -3467,7 +3779,7 @@ API changes:
  LeafManager::LeafRange::Iterator@endlink,
  @vdblink::util::BaseMaskIterator::operator=() BaseMaskIterator@endlink,
  @vdblink::util::NodeMask::operator=() NodeMask@endlink and
-  @vdblink::util::RootNodeMask::operator=() RootNodeMask@endlink
+  @c openvdb::util::RootNodeMask::operator=()
  now return references to the respective objects.
 - Removed a number of methods that were deprecated in version&nbsp;2.0.0
  or earlier.
@@ -3933,7 +4245,7 @@ API changes:
 - @vdblink::tree::LeafManager tree::LeafManager@endlink can now be
  instantiated with a @const tree, although buffer swapping with @const trees
  is disabled.
- Added a @vdblink::Grid::signedFloodFill() Grid::signedFloodFill@endlink
+- Added an @c openvdb::Grid::signedFloodFill()
  overload that allows one to specify inside and outside values.
 - Fixed a bug in Grid::setBackground() so that now only the values of
  inactive voxels change.
@@ -3943,9 +4255,9 @@ API changes:
  that relied on input and output grids having compatible tree topology.
 - @vdblink::math::Transform math::Transform@endlink now includes an
  @vdblink::math::Transform::isIdentity() isIdentity@endlink predicate
-  and methods to @vdblink::math::Transform::preMult(const Mat4d&) pre-@endlink
-  and @vdblink::math::Transform::postMult(const Mat4d&) postmultiply@endlink
-  by a matrix.
+  and methods <TT>openvdb::math::Transform::preMult(const Mat4d&)</TT> and
+  <TT>openvdb::math::Transform::postMult(const Mat4d&)</TT>
+  to pre- and postmultiply by a matrix.
 - Modified the @link NodeMasks.h node mask@endlink classes to permit
  octree-like tree configurations (i.e., with a branching factor of two)
  and to use 64-bit operations instead of 32-bit operations.
@@ -4018,7 +4330,7 @@ API changes:
  faster for high-resolution meshes where the polygon/voxel size ratio
  is small.
 - Added @vdblink::createLevelSet() createLevelSet@endlink and
-  @vdblink::createLevelSetSphere() createLevelSetSphere@endlink
+  @vdblink::tools::createLevelSetSphere() createLevelSetSphere@endlink
  factory functions for level set grids.
 - @vdblink::tree::ValueAccessor tree::ValueAccessor@endlink is now faster
  for trees of height 2, 3 and 4 (the latter is the default), and it now
@@ -4131,7 +4443,7 @@ API changes:
 - Added @vdblink::tools::compReplace() tools::compReplace@endlink,
  which copies the active values of one grid into another, and added
  a "Replace A With Active B" mode to the Combine SOP.
- @vdblink::Grid::signedFloodFill() Grid::signedFloodFill@endlink
+- @c openvdb::Grid::signedFloodFill()
  no longer enters an infinite loop when filling an empty grid.
 - Fixed a bug in the particle to level set converter that sometimes
  produced level sets with holes, and fixed a bug in the SOP that
@@ -4394,8 +4706,8 @@ API changes:
  math library and a method to return the
  @vdblink::math::Mat3::trace trace@endlink of a @c Mat3.
 - Fixed a bug in the
-  @vdblink::tree::ValueAccessor::ValueAccessor(const ValueAccessor&)
-  ValueAccessor@endlink copy constructor that caused the copy to reference
+  <TT>openvdb::tree::ValueAccessor::ValueAccessor(const ValueAccessor&)</TT>
+  copy constructor that caused the copy to reference
  the original.
 - Fixed a bug in @vdblink::tree::RootNode::setActiveState()
  RootNode::setActiveState@endlink that caused a crash
@@ -4403,7 +4715,7 @@ API changes:
 - Added a @c Tree::pruneLevelSet method that is similar to but faster than
  Tree::pruneInactive() for level set grids.
 - Added fast leaf node voxel access
-  @vdblink::tree::LeafNode::getValue(Index) const methods@endlink
+  <TT>openvdb::tree::LeafNode::getValue(Index)</TT> methods
  that index by linear offset (as returned by
  @vdblink::tree::LeafNode::ValueOnIter::pos() ValueIter::pos@endlink)
  instead of by @ijk coordinates.
@@ -4456,7 +4768,7 @@ API changes:
  just the active states of voxels from one grid with those of
  another grid of a possibly different type.
 - Fixed an incorrect scale factor in the Laplacian diffusion
-  @vdblink::tools::Filter::laplacian() filter@endlink.
+  @c openvdb::tools::Filter::laplacian() filter.
 - Fixed a bug in @vdblink::tree::Tree::merge() Tree::merge@endlink
  that could leave a tree with invalid value accessors.
 - Added @vdblink::tree::TreeValueIteratorBase::setActiveState()
@@ -4491,7 +4803,7 @@ API changes:
@par
 <B>Version 0.92.0</B> - <I>August 23 2012</I>
 - Added a Laplacian diffusion
-  @vdblink::tools::Filter::laplacian() filter@endlink.
+  @c openvdb::tools::Filter::laplacian() filter.
 - Fixed a bug in the initialization of the sparse contour tracer
  that caused mesh-to-volume conversion to fail in certain cases.
 - Fixed a bug in the curvature stencil that caused mean curvature
@@ -4519,9 +4831,8 @@ API changes:
  method that replaces a grid&rsquo;s tree with a new, empty tree of the
  correct type.
 - Fixed a crash in
-  @vdblink::tree::Tree::setValueOff(const Coord& xyz, const ValueType& value)
-  Tree::setValueOff@endlink when the new value was equal to the
-  background value.
+  <TT>openvdb::tree::Tree::setValueOff(const Coord& xyz, const ValueType& value)</TT>
+  when the new value was equal to the background value.
 - Fixed bugs in Tree::prune() that could result in output tiles with
  incorrect active states.
 - Added @c librt to the link dependencies to address build failures
@@ -36,61 +36,69 @@ Reference Platform, but for those that do, their specified versions are

 Component               | Requirements                                    | Optional
 ----------------------- | ----------------------------------------------- | --------
-OpenVDB Core Library    | CMake, C++17 compiler, TBB::tbb                 | Blosc, ZLib, Log4cplus, Imath::Imath, Boost::iostream
+OpenVDB Core Library    | CMake, C++17 compiler, TBB::tbb                 | Blosc, ZLib, Log4cplus, Imath::Imath (included), Boost::iostream, VCL (included)
 OpenVDB Print           | Core Library dependencies                       | -
 OpenVDB LOD             | Core Library dependencies                       | -
 OpenVDB Render          | Core Library dependencies                       | OpenEXR, Imath::Imath, libpng
 OpenVDB View            | Core Library dependencies, OpenGL, GLFW3, GLEW* | -
-OpenVDB Python          | Core Library dependencies, Python, nanobind     | NumPy
+OpenVDB Python          | Core Library dependencies, Python, nanobind     | -
 OpenVDB AX              | Core Library dependencies, LLVM                 | Bison, Flex
 NanoVDB                 | -                                               | Core Library, CUDA, TBB, Blosc, ZLib
-OpenVDB UnitTests       | Core Library dependencies, GoogleTest           | CppUnit**
-OpenVDB Documentation   | Doxygen                                         | -
+OpenVDB UnitTests       | Core Library dependencies, GoogleTest, numpy*   | -
+OpenVDB Documentation   | Doxygen, Doxygen Awesome (included)             | -

 - @b * GLEW is only required for building the vdb_view binary on Windows.
- - @b ** CppUnit is only required for building the AX unit tests.
+ - @b * numpy only required for python binding unit tests
 - @b Note - Bison and Flex are only required for re-generation of the AX
 grammar which is not typically necessary for the majority of users.

+Note that OpenVDB comes bundled with the following dependencies. Please see
+their respective license documentation in `ext/THIRD-PARTY.md` for license
+information.
+
+ - Doxygen Awesome
+ - Imath::Imath
+ - VCL
+
@subsection depDependencyTable Dependency Table

-Package        | Minimum | Recommended | Description                                                       | apt-get | Homebrew | Source
-------------- | ------- | ----------- | ----------------------------------------------------------------- | ------- | -------- | ------
-CMake          | 3.20    | Latest      | Cross-platform family of tools designed to help build software    | Y       | Y        | https://cmake.org
-GCC            | 11.2.1  | 11.2.1      | C++ 17 Compiler: The GNU Compiler Collection                      | Y       | Y        | https://www.gnu.org/software/gcc
-Clang          | 5.0     | Latest      | C++ 17 Compiler: A C language family frontend for LLVM            | Y       | Y        | https://clang.llvm.org
-Intel ICC      | 19      | Latest      | C++ 17 Compiler: Intels C++ Compiler                              | Y       | Y        | https://software.intel.com/en-us/c-compilers
-MSVC           | 19.30   | 19.30       | C++ 17 Compiler: Microsoft Visual C++ Compiler                    | Y       | Y        | https://visualstudio.microsoft.com/vs
-Imath          | 3.1     | Latest      | Half precision floating points                                    | Y       | Y        | http://www.openexr.com
-OpenEXR        | 3.1     | Latest      | EXR serialization support                                         | Y       | Y        | http://www.openexr.com
-TBB            | 2020.3  | 2020.3      | Threading Building Blocks - template library for task parallelism | Y       | Y        | https://www.threadingbuildingblocks.org
-ZLIB           | 1.2.7   | Latest      | Compression library for disk serialization compression            | Y       | Y        | https://www.zlib.net
-Boost          | 1.80    | 1.82        | Components: iostreams                                             | Y       | Y        | https://www.boost.org
-LLVM           | 13.0.0  | 15.0.0*     | Target-independent code generation                                | Y       | Y        | https://llvm.org/
-Bison          | 3.7.0   | 3.7.0       | General-purpose parser generator                                  | Y       | Y        | https://www.gnu.org/software/gcc
-Flex           | 2.6.4   | 2.6.4       | Fast lexical analyzer generator                                   | Y       | Y        | https://github.com/westes/flex
-Python         | 3.10    | 3.11        | The python interpreter and libraries                              | Y       | Y        | https://www.python.org
-nanobind       | 2.0.0   | 2.1.0       | C++/python bindings                                               | Y       | Y        | https://nanobind.readthedocs.io
-NumPy          | 1.23.0  | 1.26.0      | Scientific computing with Python                                  | Y       | Y        | http://www.numpy.org
-GoogleTest     | 1.10    | Latest      | A unit testing framework module for C++                           | Y       | Y        | https://github.com/google/googletest
-CppUnit        | 1.10    | Latest      | A unit testing framework module for C++                           | N       | Y        | https://freedesktop.org/wiki/Software/cppunit
-Blosc          | 1.17.0* | 1.17.0      | Recommended dependency for improved disk compression              | Y       | Y        | https://github.com/Blosc/c-blosc/releases
-Log4cplus      | 1.1.2   | Latest      | An optional dependency for improved OpenVDB Logging               | Y       | Y        | https://github.com/log4cplus/log4cplus
-libpng         | -       | Latest      | Library for manipulating PNG images                               | Y       | Y        | http://www.libpng.org/pub/png/libpng.html
-GLFW           | 3.1     | > 3.3       | Simple API for OpenGL development                                 | Y       | Y        | https://www.glfw.org
-OpenGL         | 3.2     | Latest      | Environment for developing portable graphics applications         | Y       | Y        | https://www.opengl.org
-GLEW           | 1.0.0   | Latest      | A cross-platform OpenGL extension loading library.                | Y       | Y        | http://glew.sourceforge.net
-CUDA           | -       | Latest      | Parallel computing platform for graphical processing units.       | Y       | N        | https://developer.nvidia.com/cuda-downloads
-Doxygen        | 1.8.8   | 1.8.11      | Documentation generation from C++                                 | Y       | Y        | http://www.doxygen.nl
+Package         | Minimum | Recommended | Description                                                       | apt-get | Homebrew | Source
+--------------- | ------- | ----------- | ----------------------------------------------------------------- | ------- | -------- | ------
+CMake           | 3.24    | Latest      | Cross-platform family of tools designed to help build software    | Y       | Y        | https://cmake.org
+GCC             | 11.2.1  | 11.2.1      | C++ 17 Compiler: The GNU Compiler Collection                      | Y       | Y        | https://www.gnu.org/software/gcc
+Clang           | 5.0     | Latest      | C++ 17 Compiler: A C language family frontend for LLVM            | Y       | Y        | https://clang.llvm.org
+Intel ICC       | 19      | Latest      | C++ 17 Compiler: Intels C++ Compiler                              | Y       | Y        | https://software.intel.com/en-us/c-compilers
+MSVC            | 19.30   | 19.30       | C++ 17 Compiler: Microsoft Visual C++ Compiler                    | Y       | Y        | https://visualstudio.microsoft.com/vs
+Imath           | 3.2     | >= 3.3      | Half precision floating points                                    | Y       | Y        | http://www.openexr.com
+OpenEXR         | 3.2     | >= 3.3      | EXR serialization support                                         | Y       | Y        | http://www.openexr.com
+TBB             | 2020.3  | >= 2021     | Threading Building Blocks - template library for task parallelism | Y       | Y        | https://www.threadingbuildingblocks.org
+ZLIB            | 1.2.7   | Latest      | Compression library for disk serialization compression            | Y       | Y        | https://www.zlib.net
+Boost           | 1.82    | >= 1.85     | Components: iostreams                                             | Y       | Y        | https://www.boost.org
+VCL             | 2.02.0  | >= 2.02.0   | x86 intrinsics and SIMD class wrappers and                        | Y       | Y        | https://github.com/vectorclass/version2
+LLVM            | 15.0.0* | 18.0.0      | Target-independent code generation                                | Y       | Y        | https://llvm.org/
+Bison           | 3.7.0   | 3.7.0       | General-purpose parser generator                                  | Y       | Y        | https://www.gnu.org/software/gcc
+Flex            | 2.6.4   | 2.6.4       | Fast lexical analyzer generator                                   | Y       | Y        | https://github.com/westes/flex
+Python          | 3.10    | 3.11        | The python interpreter and libraries                              | Y       | Y        | https://www.python.org
+nanobind**      | 2.0.0   | >= 2.5.0    | C++/python bindings                                               | Y       | Y        | https://nanobind.readthedocs.io
+GoogleTest      | 1.10    | Latest      | A unit testing framework module for C++                           | Y       | Y        | https://github.com/google/googletest
+Blosc           | 1.17.0* | Latest      | Recommended dependency for improved disk compression              | Y       | Y        | https://github.com/Blosc/c-blosc/releases
+Log4cplus       | 1.1.2   | Latest      | An optional dependency for improved OpenVDB Logging               | Y       | Y        | https://github.com/log4cplus/log4cplus
+libpng          | -       | Latest      | Library for manipulating PNG images                               | Y       | Y        | http://www.libpng.org/pub/png/libpng.html
+GLFW            | 3.3     | Latest      | Simple API for OpenGL development                                 | Y       | Y        | https://www.glfw.org
+OpenGL          | 3.2     | Latest      | Environment for developing portable graphics applications         | Y       | Y        | https://www.opengl.org
+GLEW            | 1.0.0   | Latest      | A cross-platform OpenGL extension loading library.                | Y       | Y        | http://glew.sourceforge.net
+CUDA            | -       | Latest      | Parallel computing platform for graphical processing units.       | Y       | N        | https://developer.nvidia.com/cuda-downloads
+Doxygen         | 1.8.8   | <= 1.14.0   | Documentation generation from C++                                 | Y       | Y        | http://www.doxygen.nl
+Doxygen Awesome | 2.0.0   | >= 2.0.0    | Documentation generation from C++                                 | Y       | Y        | https://github.com/jothepro/doxygen-awesome-css

- - @b * See the note in [known issues](@ref depKnownIssues) regarding supported blosc/llvm versions.
+ - @b *  See the note in [known issues](@ref depKnownIssues) regarding supported blosc/llvm versions.
+ - @b ** See the note in [known issues](@ref depKnownIssues) regarding supported nanobind build issues.

@subsection depKnownIssues Known Issues

- - @b LLVM >= 15. LLVM has recently changed their IR pointer data types to be
- completely opaque. Some support is provided in LLVM 15 but tests with AX show
- that this is extremely unstable. AX currently only supports LLVM versions up
- until 14.X.X. See https://llvm.org/docs/OpaquePointers.html for more details.
+ - @b On Linux, LLVM == 17 has symbol issues which makes it incompatible when
+ used with the OpenVDB Python plugin. LLVM versions greater than 18 should be
+ preferred.

 - @b GLFW: Crash with GLFW 3.2.1 when used with EGL on Linux. OpenVDB View may
 segfault on closure. Fixed in GLFW 3.3
@@ -102,15 +110,18 @@ Doxygen        | 1.8.8   | 1.8.11      | Documentation generation from C++
 Additionally these versions of blosc may generally not be compatible with OpenVDB
 itself. Avoid using these versions and upgrade to the new minimum of 1.17.0 where
 possible.
-
- - All versions from 1.11.0 through to, but excluding, 1.14.0.
- - Versions 1.16.0 and 1.16.1.
-
+  - All versions from 1.11.0 through to, but excluding, 1.14.0.
+  - Versions 1.16.0 and 1.16.1.
+  .
 For more information see:
+  - https://www.blosc.org/posts/new-forward-compat-policy/
+  - https://github.com/Blosc/c-blosc/issues/260
+  - https://github.com/Blosc/c-blosc/issues/261

- - https://www.blosc.org/posts/new-forward-compat-policy/
- - https://github.com/Blosc/c-blosc/issues/260
- - https://github.com/Blosc/c-blosc/issues/261
+ - @b nanobind: Nanobind requires a fast-hash-map implementation library called
+ tsl-robin-map. If you're installing nanobind through homebrew, this library
+ may not be installed correctly and OpenVDB's CMake will only produce a warning.
+ Make sure to also: `brew install robin-map`.

@section depInstallingDependencies Installing Dependencies

@@ -148,14 +159,12 @@ apt-get install libblosc-dev            # Blosc
 apt-get install llvm-15-dev             # LLVM
 # Python
 apt-get install python-dev              # Python
-apt-get install python-numpy            # NumPy
 pip install nanobind                    # nanobind
 # Optional
 apt-get install libpng-dev              # libpng
 apt-get install libopenexr-dev          # OpenEXR
 apt-get install liblog4cplus-dev        # Log4cplus
 apt-get install googletest              # GoogleTest
-apt-get install libcppunit-dev          # CppUnit
 # vdb_view
 apt-get install libglfw3-dev            # GLFW
 # Documentation
@@ -181,11 +190,9 @@ brew install c-blosc                   # blosc
 brew install llvm                      # LLVM
 # Python
 brew install python                    # Python
-brew install numpy                     # NumPy
-brew install nanobind                  # nanobind
+brew install nanobind robin-map        # nanobind
 # Unit Tests
 brew install gtest                     # GoogleTest
-brew install cppunit                   # CppUnit
 # vdb_view
 brew install glfw                      # GLFW
 # vdb_render
@@ -22,44 +22,30 @@ primarily by
 - Mihai Ald&eacute;n
 - David Hill

-See the @subpage changes "Release Notes" for what&rsquo;s new in this version
-of @b OpenVDB.
-
-See the @subpage dependencies "Dependencies" page for @b OpenVDB requirements.
-
-See @subpage build "Building OpenVDB" for information on building @b OpenVDB.
-
-See the @subpage overview "Overview" for an introduction to the library.
-
-See @subpage transformsAndMaps "Transforms and Maps" for more
-discussion of the transforms used in @b OpenVDB.
-
-See the @subpage faq "FAQ" for frequently asked questions about @b OpenVDB.
-
-See the @subpage codeExamples "Cookbook" to get started using @b OpenVDB.
-
-See @subpage points "OpenVDB Points" to store point data using @b OpenVDB.
-
-See @subpage python "Using OpenVDB in Python" to get started with the
-@b OpenVDB Python module.
-
+@par Quick Links
+- @subpage changes "Release Notes"
+- @subpage dependencies "Dependencies"
+- @subpage build "Building OpenVDB"
+- @subpage overview "Overview"
+    - @ref transformsAndMaps "Transforms and Maps"
+    - @ref codeExamples "Cookbook"
+    - @ref faq "FAQ"
+- @subpage python "OpenVDB Python"
+- @subpage points "OpenVDB Points"
@if OPENVDB_AX
-See @subpage openvdbax "OpenVDB AX" for the AX documentation.
+- @subpage openvdbax "OpenVDB AX"
@endif
-
@if OPENVDB_HOUDINI
-See the @subpage houdini "Houdini Cookbook" for help on implementing Houdini nodes.
+- @subpage houdini "Houdini Cookbook"
@endif
-
@if NANOVDB
-See @subpage NanoVDB_MainPage "NanoVDB" for the NanoVDB documentation.
+- @subpage NanoVDB_MainPage "NanoVDB"
@endif
+- <a href="../code_coverage/index.html">Code Coverage</a>

 Contributors, please familiarize yourselves with our
@subpage codingStyle "coding standards".

-<a href="../code_coverage/index.html">Code Coverage</a>
-

@page overview OpenVDB Overview

@@ -81,6 +67,9 @@ Contributors, please familiarize yourselves with our
  - @ref subsecNodeIter
  - @ref subsecValueAccessor
  - @ref subsecTraversal
+- @subpage transformsAndMaps "Transforms and Maps"
+
+
 <!-- - @ref secRegistry -->


@@ -320,7 +309,7 @@ as level sets.
 (The lower-level @vdblink::tools::GridResampler GridResampler@endlink does not,
 but it is optimized for level set grids in that it transforms only voxels
 in the narrow band and relies on
-@vdblink::Grid::signedFloodFill() signed flood fill@endlink to reconstruct
+@vdblink::tools::signedFloodFill() signed flood fill@endlink to reconstruct
 the inside and outside regions.)
 Other tools whose behavior is affected by the grid class include the
@vdblink::tools::divergence() divergence@endlink operator (which has an
@@ -196,7 +196,7 @@ For multithreaded insertion operations we typically assign a separate grid
 to each thread and then merge the grids as threads terminate.
 This technique works remarkably well: because OpenVDB
 is sparse and hierarchical, merging is very efficient.
-For more details, please consult the @subpage codeExamples and
+For more details, please consult the @ref codeExamples and
 Appendix&nbsp;B in the <i>Transactions on Graphics</i> paper.

@section sMaxRes Is OpenVDB unbounded?
@@ -0,0 +1,88 @@
+<!-- HTML header for doxygen 1.15.0-->
+<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "https://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
+<html xmlns="http://www.w3.org/1999/xhtml" lang="$langISO">
+<head>
+<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
+<meta http-equiv="X-UA-Compatible" content="IE=11"/>
+<meta name="generator" content="Doxygen $doxygenversion"/>
+<meta name="viewport" content="width=device-width, initial-scale=1"/>
+<!--BEGIN PROJECT_NAME--><title>$projectname: $title</title><!--END PROJECT_NAME-->
+<!--BEGIN !PROJECT_NAME--><title>$title</title><!--END !PROJECT_NAME-->
+<!--BEGIN PROJECT_ICON-->
+<link rel="icon" href="$relpath^$projecticon" type="image/x-icon" />
+<!--END PROJECT_ICON-->
+<link href="$relpath^tabs.css" rel="stylesheet" type="text/css"/>
+<!--BEGIN FULL_SIDEBAR-->
+<script type="text/javascript">var page_layout=1;</script>
+<!--END FULL_SIDEBAR-->
+<script type="text/javascript" src="$relpath^jquery.js"></script>
+<script type="text/javascript" src="$relpath^dynsections.js"></script>
+<!--BEGIN COPY_CLIPBOARD-->
+<script type="text/javascript" src="$relpath^clipboard.js"></script>
+<!--END COPY_CLIPBOARD-->
+$treeview
+$search
+$mathjax
+$darkmode
+<link href="$relpath^$stylesheet" rel="stylesheet" type="text/css" />
+$extrastylesheet
+<script type="text/javascript" src="$relpath^doxygen-awesome-darkmode-toggle.js"></script>
+<script type="text/javascript">
+    DoxygenAwesomeDarkModeToggle.init()
+</script>
+<script type="text/javascript" src="$relpath^doxygen-awesome-fragment-copy-button.js"></script>
+<script type="text/javascript">
+    DoxygenAwesomeFragmentCopyButton.init()
+</script>
+<script type="text/javascript" src="$relpath^doxygen-awesome-paragraph-link.js"></script>
+<script type="text/javascript">
+    DoxygenAwesomeParagraphLink.init()
+</script>
+</head>
+<body>
+<!--BEGIN FULL_SIDEBAR-->
+<div id="side-nav" class="ui-resizable side-nav-resizable"><!-- do not remove this div, it is closed by doxygen! -->
+<!--END FULL_SIDEBAR-->
+
+<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
+
+<!--BEGIN TITLEAREA-->
+<div id="titlearea">
+<table cellspacing="0" cellpadding="0">
+ <tbody>
+ <tr id="projectrow">
+  <!--BEGIN PROJECT_LOGO-->
+  <td id="projectlogo"><img alt="Logo" src="$relpath^$projectlogo"$logosize/></td>
+  <!--END PROJECT_LOGO-->
+  <!--BEGIN PROJECT_NAME-->
+  <td id="projectalign">
+   <div id="projectname">$projectname<!--BEGIN PROJECT_NUMBER--><span id="projectnumber">&#160;$projectnumber</span><!--END PROJECT_NUMBER-->
+   </div>
+   <!--BEGIN PROJECT_BRIEF--><div id="projectbrief">$projectbrief</div><!--END PROJECT_BRIEF-->
+  </td>
+  <!--END PROJECT_NAME-->
+  <!--BEGIN !PROJECT_NAME-->
+   <!--BEGIN PROJECT_BRIEF-->
+    <td>
+    <div id="projectbrief">$projectbrief</div>
+    </td>
+   <!--END PROJECT_BRIEF-->
+  <!--END !PROJECT_NAME-->
+  <!--BEGIN DISABLE_INDEX-->
+   <!--BEGIN SEARCHENGINE-->
+     <!--BEGIN !FULL_SIDEBAR-->
+    <td>$searchbox</td>
+     <!--END !FULL_SIDEBAR-->
+   <!--END SEARCHENGINE-->
+  <!--END DISABLE_INDEX-->
+ </tr>
+  <!--BEGIN SEARCHENGINE-->
+   <!--BEGIN FULL_SIDEBAR-->
+   <tr><td colspan="2">$searchbox</td></tr>
+   <!--END FULL_SIDEBAR-->
+  <!--END SEARCHENGINE-->
+ </tbody>
+</table>
+</div>
+<!--END TITLEAREA-->
+<!-- end header part -->
@@ -26,7 +26,7 @@

 * **Does NanoVDB depend on and require CUDA?**

-  No, NanoVDB is by design agnostic and works with most graphics APIs, e.g. OpenGL, OpenCL, OptiX, CUDA, HLSL, GLSL, DirectX, and even the CPU! The data structure itself (defined in [NanoVDB.h](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/NanoVDB.h) and [PNanoVDB.h](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/PNanoVDB.h)) only depends on C++11 or C99. However, some of the auxiliary math classes are optionally optimized with intrinsic CUDA calls (e.g. fma in Vec3) and some of the [Allocator](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/util/CudaDeviceBuffer.h) uses CUDA for explicit memory management. It should be straightforward to replace (or remove) these CUDA dependencies if you so desire.
+  No, NanoVDB is by design agnostic and works with most graphics APIs, e.g. OpenGL, OpenCL, OptiX, CUDA, HLSL, GLSL, DirectX, and even the CPU! The data structure itself (defined in [NanoVDB.h](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/NanoVDB.h) and [PNanoVDB.h](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/PNanoVDB.h)) only depends on C++11 or C99. However, some of the auxiliary math classes are optionally optimized with intrinsic CUDA calls (e.g. fma in Vec3) and the [DeviceBuffer](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/cuda/DeviceBuffer.h) allocator uses CUDA for explicit memory management. It should be straightforward to replace (or remove) these CUDA dependencies if you so desire.

 * **How does the data structure of NanoVDB compare to that of OpenVDB?**

@@ -1,154 +1,38 @@
 # NanoVDB Hello World Examples {#NanoVDB_HelloWorld}

-[The first example](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/examples/ex_openvdb_to_nanovdb_accessor/openvdb_to_nanovdb_accessor.cc) shows how to convert an OpenVDB level set sphere into a NanoVDB level set, use accessors to print out multiple values from both grids, and save the NanoVDB grid to file. Note that **this example depends on both OpenVDB and NanoVDB**.
+The example snippets below are pulled directly from compiled example source
+files via Doxygen's `\snippet` mechanism, so the code shown here is always
+in sync with what is built and tested in CI. Enable
+`-DNANOVDB_BUILD_EXAMPLES=ON` to build them locally.

-```cpp
-#include <openvdb/tools/LevelSetSphere.h> // replace with your own dependencies for generating the OpenVDB grid
-#include <nanovdb/util/CreateNanoGrid.h> // converter from OpenVDB to NanoVDB (includes NanoVDB.h and GridManager.h)
-#include <nanovdb/util/IO.h>
+## Convert an OpenVDB grid to NanoVDB and use accessors

-// Convert an openvdb level set sphere into a nanovdb, use accessors to print out multiple values from both
-// grids and save the NanoVDB grid to file.
-// Note, main depends on BOTH OpenVDB and NanoVDB.
-int main()
-{
-    try {
-        // Create an OpenVDB grid (here a level set surface but replace this with your own code)
-        auto srcGrid = openvdb::tools::createLevelSetSphere<openvdb::FloatGrid>(100.0f, openvdb::Vec3f(0.0f), 1.0f);
+[The first example](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/examples/ex_openvdb_to_nanovdb_accessor/openvdb_to_nanovdb_accessor.cc)
+shows how to convert an OpenVDB level set sphere into a NanoVDB level set, use
+accessors to print out multiple values from both grids, and save the NanoVDB
+grid to file. Note that **this example depends on both OpenVDB and NanoVDB**.

-        // Convert the OpenVDB grid, srcGrid, into a NanoVDB grid handle.
-        auto handle = nanovdb::createNanoGrid(*srcGrid);
+\snippet nanovdb/nanovdb/examples/ex_openvdb_to_nanovdb_accessor/openvdb_to_nanovdb_accessor.cc openvdb_to_nanovdb_accessor

-        // Define a (raw) pointer to the NanoVDB grid on the host. Note we match the value type of the srcGrid!
-        auto* dstGrid = handle.grid<float>();
+## Read a NanoVDB grid from a file (host only)

-        if (!dstGrid)
-            throw std::runtime_error("GridHandle does not contain a grid with value type float");
+[The second example](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor/read_nanovdb_sphere_accessor.cc)
+reads a NanoVDB grid from a file (the one saved in the previous example) and
+prints out multiple values. Note that **this example does not depend on
+OpenVDB (nor CUDA), only NanoVDB**.

-        // Get accessors for the two grids. Note that accessors only accelerate repeated access!
-        auto dstAcc = dstGrid->getAccessor();
-        auto srcAcc = srcGrid->getAccessor();
+\snippet nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor/read_nanovdb_sphere_accessor.cc read_nanovdb_sphere_accessor

-        // Access and print out a cross-section of the narrow-band level set from the two grids
-        for (int i = 97; i < 104; ++i) {
-            printf("(%3i,0,0) OpenVDB cpu: % -4.2f, NanoVDB cpu: % -4.2f\n", i, srcAcc.getValue(openvdb::Coord(i, 0, 0)), dstAcc.getValue(nanovdb::Coord(i, 0, 0)));
-        }
+## Read a NanoVDB grid and access it on the CPU and GPU

-        nanovdb::io::writeGrid("data/sphere.nvdb", handle); // Write the NanoVDB grid to file and throw if writing fails
-    }
-    catch (const std::exception& e) {
-        std::cerr << "An exception occurred: \"" << e.what() << "\"" << std::endl;
-    }
-    return 0;
-}
-```
+[The third example](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor_cuda/read_nanovdb_sphere_accessor_cuda.cu)
+reads a NanoVDB grid from a file (the one saved in the first example) and
+prints out multiple values on both the CPU and GPU. Note that **this example
+does NOT depend on OpenVDB, only NanoVDB and CUDA**.

-[The second example](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor/read_nanovdb_sphere_accessor.cc) reads a NanoVDB grid from a file (the one saved in the previous example) and prints out multiple values. Note that **this example does not depend on OpenVDB (nor CUDA), only NanoVDB**.
+\snippet nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor_cuda/read_nanovdb_sphere_accessor_cuda.cu read_nanovdb_sphere_accessor_cuda

-```cpp
-#include <nanovdb/util/IO.h> // this is required to read (and write) NanoVDB files on the host
+NVCC requires the CUDA kernel to be defined in
+[a separate .cu file](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor_cuda/read_nanovdb_sphere_accessor_cuda_kernel.cu):

-/// @brief Read a NanoVDB grid from a file and print out multiple values.
-///
-/// @note Note This example does NOT depend on OpenVDB (nor CUDA), only NanoVDB.
-int main()
-{
-    try {
-        auto handle = nanovdb::io::readGrid("data/sphere.nvdb"); // reads first grid from file
-
-        auto* grid = handle.grid<float>(); // get a (raw) pointer to a NanoVDB grid of value type float
-
-        if (!grid)
-            throw std::runtime_error("File did not contain a grid with value type float");
-
-        auto acc = grid->getAccessor(); // create an accessor for fast access to multiple values
-        for (int i = 97; i < 104; ++i) {
-            printf("(%3i,0,0) NanoVDB cpu: % -4.2f\n", i, acc.getValue(nanovdb::Coord(i, 0, 0)));
-        }
-    }
-    catch (const std::exception& e) {
-        std::cerr << "An exception occurred: \"" << e.what() << "\"" << std::endl;
-    }
-    return 0;
-}
-```
-
-[The third example](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor_cuda/read_nanovdb_sphere_accessor_cuda.cc) reads a NanoVDB grid from a file (the one saved in the first example) and prints out multiple values on both the CPU and GPU. Note that **this example does NOT depend on OpenVDB, only NanoVDB and CUDA**.
-
-```cpp
-#include <nanovdb/util/IO.h> // this is required to read (and write) NanoVDB files on the host
-#include <nanovdb/util/cuda/CudaDeviceBuffer.h> // required for CUDA memory management
-
-extern "C" void launch_kernels(const nanovdb::NanoGrid<float>*,
-                               const nanovdb::NanoGrid<float>*,
-                               cudaStream_t stream);
-
-/// @brief Read a NanoVDB grid from a file and print out multiple values on both the cpu and gpu.
-///
-/// @note Note This example does NOT depend on OpenVDB, only NanoVDB and CUDA.
-int main()
-{
-    try {
-        // returns a GridHandle using CUDA for memory management.
-        auto handle = nanovdb::io::readGrid<nanovdb::CudaDeviceBuffer>("data/sphere.nvdb");
-
-        cudaStream_t stream; // Create a CUDA stream to allow for asynchronous copy of pinned CUDA memory.
-        cudaStreamCreate(&stream);
-
-        handle.deviceUpload(stream, false); // Copy the NanoVDB grid to the GPU asynchronously
-
-        auto* cpuGrid = handle.grid<float>(); // get a (raw) pointer to a NanoVDB grid of value type float on the CPU
-        auto* deviceGrid = handle.deviceGrid<float>(); // get a (raw) pointer to a NanoVDB grid of value type float on the GPU
-
-        if (!deviceGrid || !cpuGrid)
-            throw std::runtime_error("GridHandle did not contain a grid with value type float");
-
-        launch_kernels(deviceGrid, cpuGrid, stream); // Call a host method to print a grid values on both the CPU and GPU
-
-        cudaStreamDestroy(stream); // Destroy the CUDA stream
-    }
-    catch (const std::exception& e) {
-        std::cerr << "An exception occurred: \"" << e.what() << "\"" << std::endl;
-    }
-
-    return 0;
-}
-```
-
-NVCC requires the CUDA kernel to be defined in [a separate .cu file](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor_cuda/read_nanovdb_sphere_accessor_cuda.cu):
-
-```cpp
-#include <nanovdb/NanoVDB.h> // this defined the core tree data structure of NanoVDB accessible on both the host and device
-#include <stdio.h> // for printf
-
-// This is called by the host only
-void cpu_kernel(const nanovdb::NanoGrid<float>* cpuGrid)
-{
-    auto cpuAcc = cpuGrid->getAccessor();
-    for (int i = 97; i < 104; ++i) {
-        printf("(%3i,0,0) NanoVDB cpu: % -4.2f\n", i, cpuAcc.getValue(nanovdb::Coord(i, 0, 0)));
-    }
-}
-
-// This is called by the device only
-__global__ void gpu_kernel(const nanovdb::NanoGrid<float>* deviceGrid)
-{
-    if (threadIdx.x > 6)
-        return;
-    int  i = 97 + threadIdx.x;
-    auto gpuAcc = deviceGrid->getAccessor();
-    printf("(%3i,0,0) NanoVDB gpu: % -4.2f\n", i, gpuAcc.getValue(nanovdb::Coord(i, 0, 0)));
-}
-
-// This is called by the client code on the host
-extern "C" void launch_kernels(const nanovdb::NanoGrid<float>* deviceGrid,
-                               const nanovdb::NanoGrid<float>* cpuGrid,
-                               cudaStream_t                    stream)
-{
-    // Launch the device kernel asynchronously
-    gpu_kernel<<<1, 64, 0, stream>>>(deviceGrid);
-
-    // Launch the host "kernel" (synchronously)
-    cpu_kernel(cpuGrid);
-}
-```
+\snippet nanovdb/nanovdb/examples/ex_read_nanovdb_sphere_accessor_cuda/read_nanovdb_sphere_accessor_cuda_kernel.cu read_nanovdb_sphere_accessor_cuda_kernel
@@ -12,8 +12,8 @@
 * [math/HDDA.h](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/math/HDDA.h) HDDA related.
 * [math/SampleFromVoxels.h](https://github.com/AcademySoftwareFoundation/openvdb/blob/master/nanovdb/nanovdb/math/SampleFromVoxels.h) interpolation.

-```bash
-foo@bar:~$ tree
+```
+$ tree
 .
 ├── CMakeLists.txt
 ├── cmd
@@ -175,4 +175,5 @@ foo@bar:~$ tree
    ├── SampleFromVoxels.h
    ├── Stencils.h
    ├── Timer.h
-    └── Util.h
+    └── Util.h
+```
@@ -0,0 +1,26 @@
+
+html {
+    /* override light-mode variables here */
+}
+
+html.dark-mode {
+    /* define dark-mode variable overrides here */
+    --openvdb-group-header-override: white;
+}
+
+/* Fix @name group sections in dark theme appearing strangely */
+div.groupHeader
+{
+    box-shadow: none;
+    color: var(--openvdb-group-header-override);
+}
+
+/* Apply invert to all images in tables, regardless of nesting, required to
+   properly handle latex formulas in dark theme (for some reason
+   doxygen-awesome only applies this to the first level of nested tables via:
+     html.dark-mode div.contents > table img)
+*/
+html.dark-mode div.contents table img
+{
+    filter: brightness(89%) hue-rotate(180deg) invert();
+}
@@ -21,7 +21,6 @@
  - @ref secPtIndexFilters
 - @ref secPtSpaceAndTrans

-
@section secPtOverview Introduction
@anchor openvdbPointsOverview

@@ -1,6 +1,6 @@
 /**

-@page python Using OpenVDB in Python
+@page python OpenVDB Python

 This section describes the OpenVDB Python module and includes Python code
 snippets and some complete programs that illustrate how to perform common tasks.
@@ -441,11 +441,9 @@ not on their active states or other attributes.

 Large data sets are often handled in Python using
 <a href="http://www.numpy.org/">NumPy</a>.
-The OpenVDB Python module can optionally be compiled with NumPy support.
-With NumPy enabled, the @b copyFromArray and @b copyToArray grid methods
-can be used to exchange data efficiently between scalar-valued grids
-and three-dimensional NumPy arrays and between vector-valued grids and
-four-dimensional NumPy arrays.
+The @b copyFromArray and @b copyToArray grid methods can be used to exchange
+data efficiently between scalar-valued grids and three-dimensional NumPy
+arrays and between vector-valued grids and four-dimensional NumPy arrays.
@code{.py}
 >>> import pyopenvdb as vdb
 >>> import numpy
@@ -0,0 +1,126 @@
+See [LICENSE](LICENSE) for the main open source license of original
+code written for the OpenVDB project.
+
+The remainder of this file reproduces the open source licensing details
+of other projects that have been imported, incorporated into, or derived
+into parts of OpenVDB.
+
+In no particular order:
+
+-------------------------------------------------------------------------
+
+BSD 3-Clause License
+SPDX-License-Identifier: BSD-3-Clause
+
+* Imath, Copyright (c) Contributors to the Imath Project.
+  https://github.com/AcademySoftwareFoundation/Imath
+
+  OpenVDB includes a copy of Half.h from the Imath project. This copy can
+  be disabled in favour of using Half.h from the Imath project - but one
+  must be provided either way.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+   - Redistributions of source code must retain the above copyright notice,
+     this list of conditions and the following disclaimer.
+
+   - Redistributions in binary form must reproduce the above copyright
+     notice, this list of conditions and the following disclaimer in the
+     documentation and/or other materials provided with the distribution.
+
+   - Neither the name of the copyright holder 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 THE COPYRIGHT OWNER 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.
+
+-------------------------------------------------------------------------
+
+MIT License
+
+* doxygen-awesome Copyright (c) 2021 - 2023 jothepro
+  https://github.com/jothepro/doxygen-awesome-css
+
+  OpenVDB uses doxygen-awesome for stylized html output with doxygen and
+  comes bundled with copies of the css provided by doxygen-awesome. Usage
+  of this css can be disabled.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+-------------------------------------------------------------------------
+
+Apache 2.0
+
+* VCL (c) Copyright 2012-2022 Agner Fog.
+  https://github.com/vectorclass/version2
+
+  OpenVDB includes a copy of vectorclass for x86 SIMD intrinsic usage. Usage
+  of VCL is disabled by default and requires the VCL headers to be shipped with
+  OpenVDB installations. This behaviour and usage of VCL can be controlled
+  during OpenVDB configuration. See the full LICENSE terms in:
+
+    vcl/openvdb/ext/vcl/LICENSE
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+   http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+
+-------------------------------------------------------------------------
+
+Public domain
+
+* The simplex noise implementation included in OpenVDB AX is Copyright (c)
+  Arthur Tombs. It bears the following notice:
+
+  This is a derivative work based on OpenSimplex by Kurt Spencer:
+    https://gist.github.com/KdotJPG/b1270127455a94ac5d19
+
+  Anyone is free to make use of this software in whatever way they want.
+  Attribution is appreciated, but not required.
+
+  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+  IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR
+  OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+  ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+  OTHER DEALINGS IN THE SOFTWARE.
+
+-------------------------------------------------------------------------
+
+If we have left anything out, it is unintentional. Please let us know.
@@ -0,0 +1,138 @@
+// SPDX-License-Identifier: MIT
+/**
+
+Doxygen Awesome
+https://github.com/jothepro/doxygen-awesome-css
+
+Copyright (c) 2021 - 2025 jothepro
+
+*/
+
+class DoxygenAwesomeDarkModeToggle extends HTMLElement {
+    // SVG icons from https://fonts.google.com/icons
+    // Licensed under the Apache 2.0 license:
+    // https://www.apache.org/licenses/LICENSE-2.0.html
+    static lightModeIcon = `<svg xmlns="http://www.w3.org/2000/svg" enable-background="new 0 0 24 24" height="24px" viewBox="0 0 24 24" width="24px" fill="#FCBF00"><rect fill="none" height="24" width="24"/><circle cx="12" cy="12" opacity=".3" r="3"/><path d="M12,9c1.65,0,3,1.35,3,3s-1.35,3-3,3s-3-1.35-3-3S10.35,9,12,9 M12,7c-2.76,0-5,2.24-5,5s2.24,5,5,5s5-2.24,5-5 S14.76,7,12,7L12,7z M2,13l2,0c0.55,0,1-0.45,1-1s-0.45-1-1-1l-2,0c-0.55,0-1,0.45-1,1S1.45,13,2,13z M20,13l2,0c0.55,0,1-0.45,1-1 s-0.45-1-1-1l-2,0c-0.55,0-1,0.45-1,1S19.45,13,20,13z M11,2v2c0,0.55,0.45,1,1,1s1-0.45,1-1V2c0-0.55-0.45-1-1-1S11,1.45,11,2z M11,20v2c0,0.55,0.45,1,1,1s1-0.45,1-1v-2c0-0.55-0.45-1-1-1C11.45,19,11,19.45,11,20z M5.99,4.58c-0.39-0.39-1.03-0.39-1.41,0 c-0.39,0.39-0.39,1.03,0,1.41l1.06,1.06c0.39,0.39,1.03,0.39,1.41,0s0.39-1.03,0-1.41L5.99,4.58z M18.36,16.95 c-0.39-0.39-1.03-0.39-1.41,0c-0.39,0.39-0.39,1.03,0,1.41l1.06,1.06c0.39,0.39,1.03,0.39,1.41,0c0.39-0.39,0.39-1.03,0-1.41 L18.36,16.95z M19.42,5.99c0.39-0.39,0.39-1.03,0-1.41c-0.39-0.39-1.03-0.39-1.41,0l-1.06,1.06c-0.39,0.39-0.39,1.03,0,1.41 s1.03,0.39,1.41,0L19.42,5.99z M7.05,18.36c0.39-0.39,0.39-1.03,0-1.41c-0.39-0.39-1.03-0.39-1.41,0l-1.06,1.06 c-0.39,0.39-0.39,1.03,0,1.41s1.03,0.39,1.41,0L7.05,18.36z"/></svg>`
+    static darkModeIcon = `<svg xmlns="http://www.w3.org/2000/svg" enable-background="new 0 0 24 24" height="24px" viewBox="0 0 24 24" width="24px" fill="#FE9700"><rect fill="none" height="24" width="24"/><path d="M9.37,5.51C9.19,6.15,9.1,6.82,9.1,7.5c0,4.08,3.32,7.4,7.4,7.4c0.68,0,1.35-0.09,1.99-0.27 C17.45,17.19,14.93,19,12,19c-3.86,0-7-3.14-7-7C5,9.07,6.81,6.55,9.37,5.51z" opacity=".3"/><path d="M9.37,5.51C9.19,6.15,9.1,6.82,9.1,7.5c0,4.08,3.32,7.4,7.4,7.4c0.68,0,1.35-0.09,1.99-0.27C17.45,17.19,14.93,19,12,19 c-3.86,0-7-3.14-7-7C5,9.07,6.81,6.55,9.37,5.51z M12,3c-4.97,0-9,4.03-9,9s4.03,9,9,9s9-4.03,9-9c0-0.46-0.04-0.92-0.1-1.36 c-0.98,1.37-2.58,2.26-4.4,2.26c-2.98,0-5.4-2.42-5.4-5.4c0-1.81,0.89-3.42,2.26-4.4C12.92,3.04,12.46,3,12,3L12,3z"/></svg>`
+    static title = "Toggle Light/Dark Mode"
+
+    static prefersLightModeInDarkModeKey = "prefers-light-mode-in-dark-mode"
+    static prefersDarkModeInLightModeKey = "prefers-dark-mode-in-light-mode"
+
+    static _staticConstructor = function() {
+        DoxygenAwesomeDarkModeToggle.enableDarkMode(DoxygenAwesomeDarkModeToggle.userPreference)
+        // Update the color scheme when the browsers preference changes
+        // without user interaction on the website.
+        window.matchMedia('(prefers-color-scheme: dark)').addEventListener('change', event => {
+            DoxygenAwesomeDarkModeToggle.onSystemPreferenceChanged()
+        })
+        // Update the color scheme when the tab is made visible again.
+        // It is possible that the appearance was changed in another tab 
+        // while this tab was in the background.
+        document.addEventListener("visibilitychange", visibilityState => {
+            if (document.visibilityState === 'visible') {
+                DoxygenAwesomeDarkModeToggle.onSystemPreferenceChanged()
+            }
+        });
+    }()
+
+    static init() {
+        $(function() {
+            $(document).ready(function() {
+                const toggleButton = document.createElement('doxygen-awesome-dark-mode-toggle')
+                toggleButton.title = DoxygenAwesomeDarkModeToggle.title
+                toggleButton.updateIcon()
+
+                window.matchMedia('(prefers-color-scheme: dark)').addEventListener('change', event => {
+                    toggleButton.updateIcon()
+                })
+                document.addEventListener("visibilitychange", visibilityState => {
+                    if (document.visibilityState === 'visible') {
+                        toggleButton.updateIcon()
+                    }
+                });
+
+                $(document).ready(function(){
+                    document.getElementById("MSearchBox").parentNode.appendChild(toggleButton)
+                })
+                $(window).resize(function(){
+                    document.getElementById("MSearchBox").parentNode.appendChild(toggleButton)
+                })
+            })
+        })
+    }
+
+    constructor() {
+        super();
+        this.onclick=this.toggleDarkMode
+    }
+
+    /**
+     * @returns `true` for dark-mode, `false` for light-mode system preference
+     */
+    static get systemPreference() {
+        return window.matchMedia('(prefers-color-scheme: dark)').matches
+    }
+
+    /**
+     * @returns `true` for dark-mode, `false` for light-mode user preference
+     */
+    static get userPreference() {
+        return (!DoxygenAwesomeDarkModeToggle.systemPreference && localStorage.getItem(DoxygenAwesomeDarkModeToggle.prefersDarkModeInLightModeKey)) || 
+        (DoxygenAwesomeDarkModeToggle.systemPreference && !localStorage.getItem(DoxygenAwesomeDarkModeToggle.prefersLightModeInDarkModeKey))
+    }
+
+    static set userPreference(userPreference) {
+        DoxygenAwesomeDarkModeToggle.darkModeEnabled = userPreference
+        if(!userPreference) {
+            if(DoxygenAwesomeDarkModeToggle.systemPreference) {
+                localStorage.setItem(DoxygenAwesomeDarkModeToggle.prefersLightModeInDarkModeKey, true)
+            } else {
+                localStorage.removeItem(DoxygenAwesomeDarkModeToggle.prefersDarkModeInLightModeKey)
+            }
+        } else {
+            if(!DoxygenAwesomeDarkModeToggle.systemPreference) {
+                localStorage.setItem(DoxygenAwesomeDarkModeToggle.prefersDarkModeInLightModeKey, true)
+            } else {
+                localStorage.removeItem(DoxygenAwesomeDarkModeToggle.prefersLightModeInDarkModeKey)
+            }
+        }
+        DoxygenAwesomeDarkModeToggle.onUserPreferenceChanged()
+    }
+
+    static enableDarkMode(enable) {
+        if(enable) {
+            DoxygenAwesomeDarkModeToggle.darkModeEnabled = true
+            document.documentElement.classList.add("dark-mode")
+            document.documentElement.classList.remove("light-mode")
+        } else {
+            DoxygenAwesomeDarkModeToggle.darkModeEnabled = false
+            document.documentElement.classList.remove("dark-mode")
+            document.documentElement.classList.add("light-mode")
+        }
+    }
+
+    static onSystemPreferenceChanged() {
+        DoxygenAwesomeDarkModeToggle.darkModeEnabled = DoxygenAwesomeDarkModeToggle.userPreference
+        DoxygenAwesomeDarkModeToggle.enableDarkMode(DoxygenAwesomeDarkModeToggle.darkModeEnabled)
+    }
+
+    static onUserPreferenceChanged() {
+        DoxygenAwesomeDarkModeToggle.enableDarkMode(DoxygenAwesomeDarkModeToggle.darkModeEnabled)
+    }
+
+    toggleDarkMode() {
+        DoxygenAwesomeDarkModeToggle.userPreference = !DoxygenAwesomeDarkModeToggle.userPreference
+        this.updateIcon()
+    }
+
+    updateIcon() {
+        if(DoxygenAwesomeDarkModeToggle.darkModeEnabled) {
+            this.innerHTML = DoxygenAwesomeDarkModeToggle.darkModeIcon
+        } else {
+            this.innerHTML = DoxygenAwesomeDarkModeToggle.lightModeIcon
+        }
+    }
+}
+
+customElements.define("doxygen-awesome-dark-mode-toggle", DoxygenAwesomeDarkModeToggle);
@@ -0,0 +1,66 @@
+// SPDX-License-Identifier: MIT
+/**
+
+Doxygen Awesome
+https://github.com/jothepro/doxygen-awesome-css
+
+Copyright (c) 2022 - 2025 jothepro
+
+*/
+
+class DoxygenAwesomeFragmentCopyButton extends HTMLElement {
+    constructor() {
+        super();
+        this.onclick=this.copyContent
+    }
+    static title = "Copy to clipboard"
+    static copyIcon = `<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" width="24" height="24"><path d="M0 0h24v24H0V0z" fill="none"/><path d="M23.04,10.322c0,-2.582 -2.096,-4.678 -4.678,-4.678l-6.918,-0c-2.582,-0 -4.678,2.096 -4.678,4.678c0,-0 0,8.04 0,8.04c0,2.582 2.096,4.678 4.678,4.678c0,-0 6.918,-0 6.918,-0c2.582,-0 4.678,-2.096 4.678,-4.678c0,-0 0,-8.04 0,-8.04Zm-2.438,-0l-0,8.04c-0,1.236 -1.004,2.24 -2.24,2.24l-6.918,-0c-1.236,-0 -2.239,-1.004 -2.239,-2.24l-0,-8.04c-0,-1.236 1.003,-2.24 2.239,-2.24c0,0 6.918,0 6.918,0c1.236,0 2.24,1.004 2.24,2.24Z"/><path d="M5.327,16.748c-0,0.358 -0.291,0.648 -0.649,0.648c0,0 0,0 0,0c-2.582,0 -4.678,-2.096 -4.678,-4.678c0,0 0,-8.04 0,-8.04c0,-2.582 2.096,-4.678 4.678,-4.678l6.918,0c2.168,0 3.994,1.478 4.523,3.481c0.038,0.149 0.005,0.306 -0.09,0.428c-0.094,0.121 -0.239,0.191 -0.392,0.191c-0.451,0.005 -1.057,0.005 -1.457,0.005c-0.238,0 -0.455,-0.14 -0.553,-0.357c-0.348,-0.773 -1.128,-1.31 -2.031,-1.31c-0,0 -6.918,0 -6.918,0c-1.236,0 -2.24,1.004 -2.24,2.24l0,8.04c0,1.236 1.004,2.24 2.24,2.24l0,-0c0.358,-0 0.649,0.29 0.649,0.648c-0,0.353 -0,0.789 -0,1.142Z" style="fill-opacity:0.6;"/></svg>`
+    static successIcon = `<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" width="24" height="24"><path d="M8.084,16.111c-0.09,0.09 -0.212,0.141 -0.34,0.141c-0.127,-0 -0.249,-0.051 -0.339,-0.141c-0.746,-0.746 -2.538,-2.538 -3.525,-3.525c-0.375,-0.375 -0.983,-0.375 -1.357,0c-0.178,0.178 -0.369,0.369 -0.547,0.547c-0.375,0.375 -0.375,0.982 -0,1.357c1.135,1.135 3.422,3.422 4.75,4.751c0.27,0.27 0.637,0.421 1.018,0.421c0.382,0 0.749,-0.151 1.019,-0.421c2.731,-2.732 10.166,-10.167 12.454,-12.455c0.375,-0.375 0.375,-0.982 -0,-1.357c-0.178,-0.178 -0.369,-0.369 -0.547,-0.547c-0.375,-0.375 -0.982,-0.375 -1.357,0c-2.273,2.273 -9.567,9.567 -11.229,11.229Z"/></svg>`
+    static successDuration = 980
+    static init() {
+        $(function() {
+            $(document).ready(function() {
+                if(navigator.clipboard) {
+                    const fragments = document.getElementsByClassName("fragment")
+                    for(const fragment of fragments) {
+                        const fragmentWrapper = document.createElement("div")
+                        fragmentWrapper.className = "doxygen-awesome-fragment-wrapper"
+                        const fragmentCopyButton = document.createElement("doxygen-awesome-fragment-copy-button")
+                        fragmentCopyButton.innerHTML = DoxygenAwesomeFragmentCopyButton.copyIcon
+                        fragmentCopyButton.title = DoxygenAwesomeFragmentCopyButton.title
+                
+                        fragment.parentNode.replaceChild(fragmentWrapper, fragment)
+                        fragmentWrapper.appendChild(fragment)
+                        fragmentWrapper.appendChild(fragmentCopyButton)
+            
+                    }
+                }
+            })
+        })
+    }
+
+
+    copyContent() {
+        const content = this.previousSibling.cloneNode(true)
+        // filter out line number from file listings
+        content.querySelectorAll(".lineno, .ttc").forEach((node) => {
+            node.remove()
+        })
+        let textContent = content.textContent
+        // remove trailing newlines that appear in file listings
+        let numberOfTrailingNewlines = 0
+        while(textContent.charAt(textContent.length - (numberOfTrailingNewlines + 1)) == '\n') {
+            numberOfTrailingNewlines++;
+        }
+        textContent = textContent.substring(0, textContent.length - numberOfTrailingNewlines)
+        navigator.clipboard.writeText(textContent);
+        this.classList.add("success")
+        this.innerHTML = DoxygenAwesomeFragmentCopyButton.successIcon
+        window.setTimeout(() => {
+            this.classList.remove("success")
+            this.innerHTML = DoxygenAwesomeFragmentCopyButton.copyIcon
+        }, DoxygenAwesomeFragmentCopyButton.successDuration);
+    }
+}
+
+customElements.define("doxygen-awesome-fragment-copy-button", DoxygenAwesomeFragmentCopyButton)
@@ -0,0 +1,32 @@
+// SPDX-License-Identifier: MIT
+/**
+
+Doxygen Awesome
+https://github.com/jothepro/doxygen-awesome-css
+
+Copyright (c) 2022 - 2025 jothepro
+
+*/
+
+class DoxygenAwesomeParagraphLink {
+    // Icon from https://fonts.google.com/icons
+    // Licensed under the Apache 2.0 license:
+    // https://www.apache.org/licenses/LICENSE-2.0.html
+    static icon = `<svg xmlns="http://www.w3.org/2000/svg" height="20px" viewBox="0 0 24 24" width="20px"><path d="M0 0h24v24H0V0z" fill="none"/><path d="M17 7h-4v2h4c1.65 0 3 1.35 3 3s-1.35 3-3 3h-4v2h4c2.76 0 5-2.24 5-5s-2.24-5-5-5zm-6 8H7c-1.65 0-3-1.35-3-3s1.35-3 3-3h4V7H7c-2.76 0-5 2.24-5 5s2.24 5 5 5h4v-2zm-3-4h8v2H8z"/></svg>`
+    static title = "Permanent Link"
+    static init() {
+        $(function() {
+            $(document).ready(function() {
+                document.querySelectorAll(".contents a.anchor[id], .contents .groupheader > a[id]").forEach((node) => {
+                    let anchorlink = document.createElement("a")
+                    anchorlink.setAttribute("href", `#${node.getAttribute("id")}`)
+                    anchorlink.setAttribute("title", DoxygenAwesomeParagraphLink.title)
+                    anchorlink.classList.add("anchorlink")
+                    node.classList.add("anchor")
+                    anchorlink.innerHTML = DoxygenAwesomeParagraphLink.icon
+                    node.parentElement.appendChild(anchorlink)
+                })
+            })
+        })
+    }
+}
@@ -0,0 +1,20 @@
+/* SPDX-License-Identifier: MIT */
+/**
+
+Doxygen Awesome
+https://github.com/jothepro/doxygen-awesome-css
+
+Copyright (c) 2021 - 2025 jothepro
+
+*/
+
+@media screen and (min-width: 768px) {
+
+    #MSearchBox {
+        width: calc(var(--side-nav-fixed-width) - calc(2 * var(--spacing-medium)) - var(--searchbar-height) - 1px);
+    }
+
+    #MSearchField {
+        width: calc(var(--side-nav-fixed-width) - calc(2 * var(--spacing-medium)) - 66px - var(--searchbar-height));
+    }
+}
@@ -0,0 +1,105 @@
+/* SPDX-License-Identifier: MIT */
+/**
+
+Doxygen Awesome
+https://github.com/jothepro/doxygen-awesome-css
+
+Copyright (c) 2021 - 2025 jothepro
+
+ */
+
+html {
+    /* side nav width. MUST be = `TREEVIEW_WIDTH`.
+     * Make sure it is wide enough to contain the page title (logo + title + version)
+     */
+    --side-nav-fixed-width: 335px;
+    --menu-display: none;
+
+    --top-height: 120px;
+    --toc-sticky-top: -25px;
+    --toc-max-height: calc(100vh - 2 * var(--spacing-medium) - 25px);
+}
+
+#projectname {
+    white-space: nowrap;
+}
+
+
+@media screen and (min-width: 768px) {
+    html {
+        --searchbar-background: var(--page-background-color);
+    }
+
+    #side-nav {
+        min-width: var(--side-nav-fixed-width);
+        max-width: var(--side-nav-fixed-width);
+        top: var(--top-height);
+        overflow: visible;
+    }
+
+    #nav-tree, #side-nav {
+        height: calc(100vh - var(--top-height)) !important;
+    }
+
+    #top {
+        display: block;
+        border-bottom: none;
+        height: var(--top-height);
+        margin-bottom: calc(0px - var(--top-height));
+        max-width: var(--side-nav-fixed-width);
+        overflow: hidden;
+        background: var(--side-nav-background);
+    }
+
+    #main-nav {
+        float: left;
+        padding-right: 0;
+    }
+
+    .ui-resizable-handle {
+        display: none;
+    }
+
+    .ui-resizable-e {
+        width: 0;
+    }
+
+    #nav-path {
+        position: fixed;
+        right: 0;
+        left: calc(var(--side-nav-fixed-width) + 1px);
+        bottom: 0;
+        width: auto;
+    }
+
+    #doc-content {
+        height: calc(100vh - 31px) !important;
+        padding-bottom: calc(3 * var(--spacing-large));
+        padding-top: calc(var(--top-height) - 80px);
+        box-sizing: border-box;
+        margin-left: var(--side-nav-fixed-width) !important;
+    }
+
+    #MSearchBox {
+        width: calc(var(--side-nav-fixed-width) - calc(2 * var(--spacing-medium)));
+    }
+
+    #MSearchField {
+        width: calc(var(--side-nav-fixed-width) - calc(2 * var(--spacing-medium)) - 65px);
+    }
+
+    #MSearchResultsWindow {
+        left: var(--spacing-medium) !important;
+        right: auto;
+    }
+
+    #nav-sync {
+        bottom: 4px;
+        right: auto;
+        left: 300px;
+        width: 35px;
+        top: auto !important;
+        user-select: none;
+        position: fixed
+    }
+}
@@ -12,7 +12,7 @@
 #ifndef OPENVDB_MATH_HALF_HAS_BEEN_INCLUDED
 #define OPENVDB_MATH_HALF_HAS_BEEN_INCLUDED

-/// @file half.h
+/// @file Half.h
 /// The half type is a 16-bit floating number, compatible with the
 /// IEEE 754-2008 binary16 type.
 ///
@@ -21,6 +21,7 @@
 /// We assume that a float, f, is an IEEE 754 single-precision
 /// floating point number, whose bits are arranged as follows:
 ///
+/// <pre>
 ///     31 (msb)
 ///     |
 ///     | 30     23
@@ -30,28 +31,36 @@
 ///     X XXXXXXXX XXXXXXXXXXXXXXXXXXXXXXX
 ///
 ///     s e        m
+/// </pre>
 ///
 /// S is the sign-bit, e is the exponent and m is the significand.
 ///
 /// If e is between 1 and 254, f is a normalized number:
 ///
+/// <pre>
 ///             s    e-127
 ///     f = (-1)  * 2      * 1.m
+/// </pre>
 ///
 /// If e is 0, and m is not zero, f is a denormalized number:
 ///
+/// <pre>
 ///             s    -126
 ///     f = (-1)  * 2      * 0.m
+/// </pre>
 ///
 /// If e and m are both zero, f is zero:
 ///
+/// <pre>
 ///     f = 0.0
+/// </pre>
 ///
 /// If e is 255, f is an "infinity" or "not a number" (NAN),
 /// depending on whether m is zero or not.
 ///
 /// Examples:
 ///
+/// <pre>
 ///     0 00000000 00000000000000000000000 = 0.0
 ///     0 01111110 00000000000000000000000 = 0.5
 ///     0 01111111 00000000000000000000000 = 1.0
@@ -62,11 +71,13 @@
 ///     1 11111111 00000000000000000000000 = -infinity
 ///     0 11111111 10000000000000000000000 = NAN
 ///     1 11111111 11111111111111111111111 = NAN
+/// </pre>
 ///
 /// **Representation of a 16-bit half:**
 ///
 /// Here is the bit-layout for a half number, h:
 ///
+/// <pre>
 ///     15 (msb)
 ///     |
 ///     | 14  10
@@ -76,28 +87,36 @@
 ///     X XXXXX XXXXXXXXXX
 ///
 ///     s e     m
+/// </pre>
 ///
 /// S is the sign-bit, e is the exponent and m is the significand.
 ///
 /// If e is between 1 and 30, h is a normalized number:
 ///
+/// <pre>
 ///             s    e-15
 ///     h = (-1)  * 2     * 1.m
+/// </pre>
 ///
 /// If e is 0, and m is not zero, h is a denormalized number:
 ///
+/// <pre>
 ///             S    -14
 ///     h = (-1)  * 2     * 0.m
+/// </pre>
 ///
 /// If e and m are both zero, h is zero:
 ///
+/// <pre>
 ///     h = 0.0
+/// </pre>
 ///
 /// If e is 31, h is an "infinity" or "not a number" (NAN),
 /// depending on whether m is zero or not.
 ///
 /// Examples:
 ///
+/// <pre>
 ///     0 00000 0000000000 = 0.0
 ///     0 01110 0000000000 = 0.5
 ///     0 01111 0000000000 = 1.0
@@ -108,13 +127,14 @@
 ///     1 11111 0000000000 = -infinity
 ///     0 11111 1000000000 = NAN
 ///     1 11111 1111111111 = NAN
+/// </pre>
 ///
 /// **Conversion via Lookup Table:**
 ///
 /// Converting from half to float is performed by default using a
 /// lookup table. There are only 65,536 different half numbers; each
 /// of these numbers has been converted and stored in a table pointed
-/// to by the ``imath_half_to_float_table`` pointer.
+/// to by the `imath_half_to_float_table` pointer.
 ///
 /// Prior to Imath v3.1, conversion from float to half was
 /// accomplished with the help of an exponent look table, but this is
@@ -135,30 +155,25 @@
 /// 65,536-entry lookup table.
 ///
 /// The lookup table symbol is included in the compilation even if
-/// ``IMATH_HALF_USE_LOOKUP_TABLE`` is false, because application code
-/// using the exported ``half.h`` may choose to enable the use of the table.
+/// `IMATH_HALF_USE_LOOKUP_TABLE` is false, because application code
+/// using the exported `half.h` may choose to enable the use of the table.
 ///
 /// An implementation can eliminate the table from compilation by
-/// defining the ``IMATH_HALF_NO_LOOKUP_TABLE`` preprocessor symbol.
-/// Simply add:
+/// defining the `IMATH_HALF_NO_LOOKUP_TABLE` preprocessor symbol.
+/// Simply add `#define IMATH_HALF_NO_LOOKUP_TABLE` before including
+/// `half.h`, or define the symbol on the compile command line.
 ///
-///     #define IMATH_HALF_NO_LOOKUP_TABLE
-///
-/// before including ``half.h``, or define the symbol on the compile
-/// command line.
-///
-/// Furthermore, an implementation wishing to receive ``FE_OVERFLOW``
-/// and ``FE_UNDERFLOW`` floating point exceptions when converting
+/// Furthermore, an implementation wishing to receive `FE_OVERFLOW`
+/// and `FE_UNDERFLOW` floating point exceptions when converting
 /// float to half by the bit-shift algorithm can define the
-/// preprocessor symbol ``IMATH_HALF_ENABLE_FP_EXCEPTIONS`` prior to
-/// including ``half.h``:
-///
-///     #define IMATH_HALF_ENABLE_FP_EXCEPTIONS
+/// preprocessor symbol `IMATH_HALF_ENABLE_FP_EXCEPTIONS` (that is,
+/// `#define IMATH_HALF_ENABLE_FP_EXCEPTIONS`) prior to including `half.h`.
 ///
 /// **Conversion Performance Comparison:**
 ///
 /// Testing on a Core i9, the timings are approximately:
 ///
+/// <pre>
 /// half to float
 /// - table: 0.71 ns / call
 /// - no table: 1.06 ns / call
@@ -168,6 +183,7 @@
 /// - original: 5.2 ns / call
 /// - no exp table + opt: 1.27 ns / call
 /// - f16c: 0.45 ns / call
+/// </pre>
 ///
 /// **Note:** the timing above depends on the distribution of the
 /// floats in question.
@@ -206,6 +222,12 @@ namespace internal {
 // Thus any attempt to link two versions of OpenVDB with different
 // namespaces will clash due to redefinition with a new type.
 // The default was not to use a lookup table.
+//
+/// @cond
+/// These internal half-float configuration and limit macros are excluded from
+/// the docs: the `half` type itself lives in the (undocumented) `internal`
+/// namespace, and documenting the `#define`s makes Doxygen auto-link the
+/// literal `#define` token (an unresolvable "link to 'define'" warning).
 #undef  IMATH_HALF_USE_LOOKUP_TABLE
 #define IMATH_HALF_NO_LOOKUP_TABLE

@@ -227,7 +249,7 @@ namespace internal {
 #define VDB_HALF_MIN 6.10351562e-05f
 /// Largest positive half
 #define VDB_HALF_MAX 65504.0f
-/// Smallest positive e for which ``half(1.0 + e) != half(1.0)``
+/// Smallest positive e for which `half(1.0 + e) != half(1.0)`
 #define VDB_HALF_EPSILON 0.00097656f
 #else
 /// Smallest positive denormalized half
@@ -238,7 +260,7 @@ namespace internal {
 #define VDB_HALF_MIN 6.10351562e-05f
 /// Largest positive half
 #define VDB_HALF_MAX 65504.0
-/// Smallest positive e for which ``half(1.0 + e) != half(1.0)``
+/// Smallest positive e for which `half(1.0 + e) != half(1.0)`
 #define VDB_HALF_EPSILON 0.00097656
 #endif

@@ -246,19 +268,19 @@ namespace internal {
 #define VDB_HALF_MANT_DIG 11
 /// Number of base 10 digits that can be represented without change:
 ///
-/// ``floor( (HALF_MANT_DIG - 1) * log10(2) ) => 3.01... -> 3``
+/// `floor( (HALF_MANT_DIG - 1) * log10(2) ) => 3.01... -> 3`
 #define VDB_HALF_DIG 3
 /// Number of base-10 digits that are necessary to uniquely represent
 /// all distinct values:
 ///
-/// ``ceil(HALF_MANT_DIG * log10(2) + 1) => 4.31... -> 5``
+/// `ceil(HALF_MANT_DIG * log10(2) + 1) => 4.31... -> 5`
 #define VDB_HALF_DECIMAL_DIG 5
 /// Base of the exponent
 #define VDB_HALF_RADIX 2
-/// Minimum negative integer such that ``HALF_RADIX`` raised to the power
+/// Minimum negative integer such that `HALF_RADIX` raised to the power
 /// of one less than that integer is a normalized half
 #define VDBB_HALF_DENORM_MIN_EXP -13
-/// Maximum positive integer such that ``HALF_RADIX`` raised to the power
+/// Maximum positive integer such that `HALF_RADIX` raised to the power
 /// of one less than that integer is a normalized half
 #define VDB_HALF_MAX_EXP 16
 /// Minimum positive integer such that 10 raised to that power is a
@@ -267,6 +289,7 @@ namespace internal {
 /// Maximum positive integer such that 10 raised to that power is a
 /// normalized half
 #define VDB_HALF_MAX_10_EXP 4
+/// @endcond

 /// a type for both C-only programs and C++ to use the same utilities
 typedef union imath_half_uif
@@ -0,0 +1,191 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+  
+   Copyright 2012-2019 Agner Fog.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.
@@ -0,0 +1,14 @@
+# version2
+Vector Class Library, latest version
+
+This is a C++ class library for using the Single Instruction Multiple Data (SIMD) instructions to improve performance on modern microprocessors with the x86 or x86/64 instruction set on Windows, Linux, and Mac platforms. There are no plans to support ARM or other instruction sets.
+
+[Latest release](https://github.com/vectorclass/version2/releases)
+
+[Download manual](https://github.com/vectorclass/manual/raw/master/vcl_manual.pdf)
+
+[Add-on packages for particular applications](https://github.com/vectorclass/add-on)
+
+[Getting-started video.](https://www.youtube.com/watch?v=TKjYdLIMTrI) Video blogger Christopher Rose has made this nice video telling how to get started with the Vector Class Library.
+
+**Help:** You may ask for programming help on [StackOverflow](https://stackoverflow.com) using the tag vector-class-library.
@@ -0,0 +1,244 @@
+Change log for Vector class library
+-----------------------------------
+
+2026-04-14 version 2.02.03
+  * conversions Vec8h <-> Vec8f and Vec16h <-> Vec16f converted to templates in vectorfp16e.h
+
+2023-07-04 version 2.02.02
+  * remove various MS compiler warnings
+
+2022-06-03 version 2.02.01
+  * minor bug fixes and updates
+
+2022-07-20 version 2.02.00
+  * support half precision floating point vectors
+  * new trigonometric functions sinpi, cospi, sincospi, tanpi
+  * floating point modulo and remainder functions
+  * functions extend_z and concatenate2
+  * functions compress_saturated with one integer parameter
+  * support for Intel compiler
+  * fix bug for NAN inputs in operator > and >=
+  * fix bug for underflow in pow function
+  * fix unsafe optimization in store_partial function
+  * new add-on package with container class and matrix class templates
+  * major improvements in add-on package 'complex'
+  * minor improvements in add-on packages 'random' and 'vector3d'
+  
+2021-08-18 version 2.01.04
+  * fix bug in elementtype for Vec8uq
+  * improved optimization of permute functions in MS compiler by using constexpr
+
+2020-11-04 version 2.01.03
+  * fix overflow in sin, cos, tan for large x
+  * fix bug in is_nan for instruction sets prior to AVX
+  * warning for MS compiler versions with poor support for AVX512
+
+2020-04-11 version 2.01.02
+  * only minor fixes
+
+2020-02-25 version 2.01.01
+  * added function store_nt
+  * New dispatch_example1.cpp, dispatch_example2.cpp
+
+2019-11-23 version 2.01.00
+  * problem with performance of permute and blend functions fixed by avoiding
+    unions in constexpr functions
+
+2019-10-31 version 2.00.02
+  * bug fix in permute function
+  * is_nan function improved
+  * templates constant4ui etc. improved
+
+2019-08-02 version 2.00
+  Derived from version 1.40
+  * use C++17
+  * use compact boolean vectors of all sizes if AVX512VL is enabled
+  * permute and blend functions improved, using C++17 metaprogramming features
+  * deprecated functions removed
+
+2019-08-02 version 1.40
+  * hosted on github
+  * license changed to Apache 2.0.
+  * added classes Vec64c, Vec64uc, Vec32s, Vec32us
+  * test bench and scripts for automatic testing of VCL
+  * new functions: maximum, minimum, to_float, to_double
+  * conversion of bitfields to boolean vectors with load_bits. This replaces to_vec4ib etc.
+  * shift_bytes_up/down functions changed to templates (old versions deprecated)
+  * removed VECTORMATH define. vectormath_lib.h rewritten. svmlpatch.lib added.
+  * many improvements and bug fixes
+  * renamed functions: round_to_int and round_to_int64 functions renamed to roundi,
+  * renamed functions: the type letter is removed from all permute and blend functions,
+    e.g. permute4f renamed to permute4.
+    These changes are made to facilitate generic template programming
+  * renamed functions: to_Vec.. Replaced by load_bits member functions
+  * deprecated functions: set_bit, get_bit
+  * deprecated bit vector classes: Vec128b, Vec256b, Vec512b
+
+2017-07-27 version 1.30
+  * fixed bug in permute8f for a particular combination of indexes
+
+2017-05-10 version 1.29
+  * Reversed Apple Clang patch in version 1.28 because the problem has reoccurred in
+    later versions of Clang
+
+2017-05-02 version 1.28
+  * Fixed problem with Apple Clang version 6.2 in vectorf128.h
+  * Fixed return type for Vec8sb operator > (Vec8us, Vec8us)
+  * cpuid function modified in instrset_detect.cpp
+
+2017-02-19 version 1.27
+  * fixed problem with scatter functions in MS Visual Studio
+
+2016-12-21 version 1.26
+  * added constant4ui template
+  * fixed error for complexvec.h with clang
+  * fixed error in vectormath_exp.h for MAX_VECTOR_SIZE < 512
+
+2016-11-25 version 1.25
+  * scatter functions
+  * new functions to_float for unsigned integer vectors
+  * instrset_detect function can detect AVX512VL, AVX512BW, AVX512DQ
+  * functions hasF16C and hasAVX512ER for detecting instruction set extensions
+  * fix bugs in horizontal_and and pow(0,0) for AVX512
+  * functions improved for AVX512 and AVX512VL: pow, approx_recipr,
+      approx_rsqrt
+  * functions improved for AVX512DQ: 64 bit multiplication, to_double,
+      32 and 64 bit rotate_left, round_to_int64, truncate_to_int64
+  * functions improved for AVX512ER: approx_recipr, approx_rsqrt,
+      exponential functions
+
+2016-10-31 version 1.24
+  * fix bug in Vec8uq constructor in vectori512e.h
+
+2016-09-27 version 1.23
+  * temporary fix of a problem in Clang version 3.9 inserted in vectorf128.h
+
+2016-05-03 version 1.22
+  * added optional namespace
+  * fixed problem with decimal.h
+
+2016-04-24 version 1.21
+  * fix problems with XOP option in gcc
+  * improved horizontal_and/or for sse2
+  * improved Vec2q and Vec4q constructor on Microsoft Visual Studio 2015
+  * removed warnings by gcc option -Wcast-qual
+
+2015-12-04 version 1.20
+  * round functions: suppress precision exception under SSE4.1 and higher
+  * fix compiler problems with AVX512 multiplication in gcc version 5.1
+  * fix compiler problems with pow function in Microsoft Visual Studio 2015
+
+2015-11-14 version 1.19
+  * fix various problems with Clang compiler
+
+2015-09-25 version 1.18
+  * fix compiler error for Vec8s divide_by_i(Vec8s const & x) under Clang compiler
+  * fix error in Vec4d::size() in vectorf256e.h
+
+2015-07-31 version 1.17
+  * improved operator > for Vec4uq
+  * more special cases in blend4q
+  * nan_code functions made static inline
+  * template parameter BTYPE renamed to BVTYPE in mathematical functions to avoid clash
+      with macro named BTYPE in winnt.h
+  * fixed bug in Vec4db constructor
+
+2014-10-24 version 1.16
+  * workaround for problem in Clang compiler extended to version 3.09 because not
+      fixed yet by Clang (vectorf128.h line 134)
+  * recognize problem with Apple version of Clang reporting wrong version number
+  * remove various minor problems with Clang
+  * function pow(vector, int) modified to strengthen type checking and avoid compiler warnings
+  * manual discusses dynamic allocation of arrays of vectors
+  * various minor changes
+
+2014-10-17 version 1.15
+  * added files ranvec1.h and ranvec1.cpp for random number generator
+  * constructors to make boolean vectors from their elements
+  * constructors and = operators to broadcast boolean scalar into boolean vectors
+  * various lookup functions improved
+  * operators &, |, ^, ~, etc. defined for various boolean vectors to avoid converson
+      to integer vectors
+  * nmul_add functions
+  * mul_add etc. moved to main header files
+  * explicit fused multiply-and-add used in math functions to improve performance
+      on compilers that don't automatically insert FMA
+
+2014-07-24 version 1.14
+  * support for AVX-512f instruction set and 512-bit vectors:
+      Vec16i, Vec16ui, Vec8q, Vec8uq, Vec16f, Vec8d, and corresponding boolean vectors
+  * new define MAX_VECTOR_SIZE, valid values are 128, 256 and 512
+  * added hyperbolic functions sinh, cosh, tanh, asinh, acosh, atanh
+  * size() member function on all vector classes returns the number of elements
+  * functions for conversion between boolean vectors and integer bitfields
+  * extracting an element from a boolean vector now returns a bool, not an int
+  * improved precision in exp2 and exp10 functions
+  * various bug fixes
+
+2014-05-11 version 1.13
+  * pow function improved
+  * mul_add, mul_sub, mul_sub_x functions
+  * propagation of error codes through nan_code function
+  * "denormal" renamed to "subnormal" everywhere, in accordance with IEEE 754-2008 standard
+
+2014-04-20 version 1.12
+  * inline implementation of mathematical functions added (vectormath_exp.h vectormath_trig.h
+      vectormath_common.h)
+  * vectormath.h renamed to vectormath_lib.h because a new alternative is added
+  * gather functions with constant indexes
+  * function sign_combine
+  * function pow_const(vector, const int)
+  * function pow_ratio(vector, const int, const int)
+  * functions horizontal_find_first, horizontal_count
+  * function recipr_sqrt removed
+  * functions round_to_int64_limited, truncate_to_int64_limited, to_double_limited
+  * function cubic_root renamed to cbrt
+  * function atan(vector,vector) renamed to atan2
+  * function if_mul
+  * function Vec4i round_to_int(Vec2d)
+  * operator & (float vector, boolean vector)
+  * operator &= (int vector, int vector)
+  * removed constructor Vec128b(int) and Vec256b(int) to avoid implicit conversion
+  * removed signalling nan function
+  * minor improvements in various blend and lookup functions
+
+2014-03-01 version 1.11
+  * fixed missing unsigned operators >>= in vectori256.h
+
+2013-10-04 version 1.10
+  * clear distinction between boolean vectors and integer vectors for the sake of
+      compatibility with mask registers in forthcoming AVX512 instruction set
+  * added function if_add
+  * tentative support for clang version 3.3 with workaround for bugs
+  * remove ambiguity for builtin m128i operator == in clang compiler.
+  * problems in clang compiler, bug reports filed at clang
+      (http://llvm.org/bugs/show_bug.cgi?id=17164, 17312)
+  * instrset.h fixes problem with macros named min and max in MS windows.h
+  * workaround problem in MS Visual Studio 11.0. Bug report 735861 and 804274
+  * minor bug fixes
+
+2013-03-31 version 1.03 beta
+  * bug fix for Vec2d cos (Vec2d const & x), VECTORMATH = 1
+
+2012-08-01 version 1.02 beta
+  * added file vector3d.h for 3-dimensional vectors
+  * added file complexvec.h for complex numbers and complex vectors
+  * added file quaternion.h for quaternions
+  * added function change_sign for floating point vectors
+  * added operators +, -, *, / between floating point vectors and scalars to remove
+      overloading ambiguity
+
+2012-07-08 version 1.01 beta
+  * added file decimal.h with Number <-> string conversion functions:
+      bin2bcd, bin2ascii, bin2hex_ascii, ascii2bin
+  * added andnot function for boolean vectors
+  * added functions shift_bytes_up and shift_bytes_down
+  * added operators for unsigned integer vector classes: >>=, &, &&, |, ||, ^, ~
+  * inteldispatchpatch.cpp removed. Use asmlib instead (www.agner.org/optimize/#asmlib)
+  * prefix ++ and -- operators now return a reference, postfix operators return a value
+  * various improvements in permute and blend functions
+  * minor improvement in abs function
+  * added version number to VECTORCLASS_H
+
+2012-05-30 version 1.00 beta
+  * first public release at www.agner.org
@@ -0,0 +1,828 @@
+/**************************  vector_convert.h   *******************************
+* Author:        Agner Fog
+* Date created:  2014-07-23
+* Last modified: 2022-07-20
+* Version:       2.02.00
+* Project:       vector class library
+* Description:
+* Header file for conversion between different vector classes with different
+* sizes. Also includes verious generic template functions.
+*
+* (c) Copyright 2012-2022 Agner Fog.
+* Apache License version 2.0 or later.
+*****************************************************************************/
+
+#ifndef VECTOR_CONVERT_H
+#define VECTOR_CONVERT_H
+
+#ifndef VECTORCLASS_H
+#include "vectorclass.h"
+#endif
+
+#if VECTORCLASS_H < 20200
+#error Incompatible versions of vector class library mixed
+#endif
+
+#ifdef VCL_NAMESPACE
+namespace VCL_NAMESPACE {
+#endif
+
+#if MAX_VECTOR_SIZE >= 256
+
+/*****************************************************************************
+*
+*          Extend from 128 to 256 bit vectors
+*
+*****************************************************************************/
+
+#if INSTRSET >= 8  // AVX2. 256 bit integer vectors
+
+// sign extend
+static inline Vec16s extend (Vec16c const a) {
+    return _mm256_cvtepi8_epi16(a);
+}
+
+// zero extend
+static inline Vec16us extend (Vec16uc const a) {
+    return _mm256_cvtepu8_epi16(a);
+}
+
+// sign extend
+static inline Vec8i extend (Vec8s const a) {
+    return _mm256_cvtepi16_epi32(a);
+}
+
+// zero extend
+static inline Vec8ui extend (Vec8us const a) {
+    return _mm256_cvtepu16_epi32(a);
+}
+
+// sign extend
+static inline Vec4q extend (Vec4i const a) {
+    return _mm256_cvtepi32_epi64(a);
+}
+
+// zero extend
+static inline Vec4uq extend (Vec4ui const a) {
+    return _mm256_cvtepu32_epi64(a);
+}
+
+
+#else  // no AVX2. 256 bit integer vectors are emulated
+
+// sign extend and zero extend functions:
+static inline Vec16s extend (Vec16c const a) {
+    return Vec16s(extend_low(a), extend_high(a));
+}
+
+static inline Vec16us extend (Vec16uc const a) {
+    return Vec16us(extend_low(a), extend_high(a));
+}
+
+static inline Vec8i extend (Vec8s const a) {
+    return Vec8i(extend_low(a), extend_high(a));
+}
+
+static inline Vec8ui extend (Vec8us const a) {
+    return Vec8ui(extend_low(a), extend_high(a));
+}
+
+static inline Vec4q extend (Vec4i const a) {
+    return Vec4q(extend_low(a), extend_high(a));
+}
+
+static inline Vec4uq extend (Vec4ui const a) {
+    return Vec4uq(extend_low(a), extend_high(a));
+}
+
+#endif  // AVX2
+
+/*****************************************************************************
+*
+*          Conversions between float and double
+*
+*****************************************************************************/
+#if INSTRSET >= 7  // AVX. 256 bit float vectors
+
+// float to double
+static inline Vec4d to_double (Vec4f const a) {
+    return _mm256_cvtps_pd(a);
+}
+
+// double to float
+static inline Vec4f to_float (Vec4d const a) {
+    return _mm256_cvtpd_ps(a);
+}
+
+#else  // no AVX2. 256 bit float vectors are emulated
+
+// float to double
+static inline Vec4d to_double (Vec4f const a) {
+    Vec2d lo = _mm_cvtps_pd(a);
+    Vec2d hi = _mm_cvtps_pd(_mm_movehl_ps(a, a));
+    return Vec4d(lo,hi);
+}
+
+// double to float
+static inline Vec4f to_float (Vec4d const a) {
+    Vec4f lo = _mm_cvtpd_ps(a.get_low());
+    Vec4f hi = _mm_cvtpd_ps(a.get_high());
+    return _mm_movelh_ps(lo, hi);
+}
+
+#endif
+
+/*****************************************************************************
+*
+*          Reduce from 256 to 128 bit vectors
+*
+*****************************************************************************/
+#if INSTRSET >= 10  // AVX512VL
+
+// compress functions. overflow wraps around
+static inline Vec16c compress (Vec16s const a) {
+    return _mm256_cvtepi16_epi8(a);
+}
+
+static inline Vec16uc compress (Vec16us const a) {
+    return _mm256_cvtepi16_epi8(a);
+}
+
+static inline Vec8s compress (Vec8i const a) {
+    return _mm256_cvtepi32_epi16(a);
+}
+
+static inline Vec8us compress (Vec8ui const a) {
+    return _mm256_cvtepi32_epi16(a);
+}
+
+static inline Vec4i compress (Vec4q const a) {
+    return _mm256_cvtepi64_epi32(a);
+}
+
+static inline Vec4ui compress (Vec4uq const a) {
+    return _mm256_cvtepi64_epi32(a);
+}
+
+// compress_saturated functions. overflow saturates
+static inline Vec16c compress_saturated (Vec16s const a) {
+    return _mm256_cvtsepi16_epi8(a);
+}
+
+static inline Vec16uc compress_saturated (Vec16us const a) {
+    return _mm256_cvtusepi16_epi8(a);
+}
+
+static inline Vec8s compress_saturated (Vec8i const a) {
+    return _mm256_cvtsepi32_epi16(a);
+}
+
+static inline Vec8us compress_saturated (Vec8ui const a) {
+    return _mm256_cvtusepi32_epi16(a);
+}
+
+static inline Vec4i compress_saturated (Vec4q const a) {
+    return _mm256_cvtsepi64_epi32(a);
+}
+
+static inline Vec4ui compress_saturated (Vec4uq const a) {
+    return _mm256_cvtusepi64_epi32(a);
+}
+
+
+#else  // no AVX512
+
+// compress functions. overflow wraps around
+static inline Vec16c compress (Vec16s const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec16uc compress (Vec16us const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec8s compress (Vec8i const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec8us compress (Vec8ui const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec4i compress (Vec4q const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec4ui compress (Vec4uq const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+// compress_saturated functions. overflow saturates
+static inline Vec16c compress_saturated (Vec16s const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec16uc compress_saturated (Vec16us const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec8s compress_saturated (Vec8i const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec8us compress_saturated (Vec8ui const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec4i compress_saturated (Vec4q const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec4ui compress_saturated (Vec4uq const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+#endif  // AVX512
+
+#endif // MAX_VECTOR_SIZE >= 256
+
+
+#if MAX_VECTOR_SIZE >= 512
+
+/*****************************************************************************
+*
+*          Reduce from 512 to 256 bit vectors
+*
+*****************************************************************************/
+#if INSTRSET >= 10  // AVX512VL
+
+// compress_saturated functions. overflow saturates
+static inline Vec32c compress_saturated (Vec32s const a) {
+    return _mm512_cvtsepi16_epi8(a);
+}
+
+static inline Vec32uc compress_saturated (Vec32us const a) {
+    return _mm512_cvtusepi16_epi8(a);
+}
+
+static inline Vec16s compress_saturated (Vec16i const a) {
+    return _mm512_cvtsepi32_epi16(a);
+}
+
+static inline Vec16us compress_saturated (Vec16ui const a) {
+    return _mm512_cvtusepi32_epi16(a);
+}
+
+static inline Vec8i compress_saturated (Vec8q const a) {
+    return _mm512_cvtsepi64_epi32(a);
+}
+
+static inline Vec8ui compress_saturated (Vec8uq const a) {
+    return _mm512_cvtusepi64_epi32(a);
+}
+
+#else  // no AVX512
+
+// compress_saturated functions. overflow saturates
+static inline Vec32c compress_saturated (Vec32s const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec32uc compress_saturated (Vec32us const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec16s compress_saturated (Vec16i const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec16us compress_saturated (Vec16ui const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec8i compress_saturated (Vec8q const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+static inline Vec8ui compress_saturated (Vec8uq const a) {
+    return compress_saturated(a.get_low(), a.get_high());
+}
+
+#endif  // AVX512
+
+/*****************************************************************************
+*
+*          Extend from 256 to 512 bit vectors
+*
+*****************************************************************************/
+
+#if INSTRSET >= 9  // AVX512. 512 bit integer vectors
+
+// sign extend
+static inline Vec32s extend (Vec32c const a) {
+#if INSTRSET >= 10
+    return _mm512_cvtepi8_epi16(a);
+#else
+    return Vec32s(extend_low(a), extend_high(a));
+#endif
+}
+
+// zero extend
+static inline Vec32us extend (Vec32uc const a) {
+#if INSTRSET >= 10
+    return _mm512_cvtepu8_epi16(a);
+#else
+    return Vec32us(extend_low(a), extend_high(a));
+#endif
+}
+
+// sign extend
+static inline Vec16i extend (Vec16s const a) {
+    return _mm512_cvtepi16_epi32(a);
+}
+
+// zero extend
+static inline Vec16ui extend (Vec16us const a) {
+    return _mm512_cvtepu16_epi32(a);
+}
+
+// sign extend
+static inline Vec8q extend (Vec8i const a) {
+    return _mm512_cvtepi32_epi64(a);
+}
+
+// zero extend
+static inline Vec8uq extend (Vec8ui const a) {
+    return _mm512_cvtepu32_epi64(a);
+}
+
+#else  // no AVX512. 512 bit vectors are emulated
+
+
+
+// sign extend
+static inline Vec32s extend (Vec32c const a) {
+    return Vec32s(extend_low(a), extend_high(a));
+}
+
+// zero extend
+static inline Vec32us extend (Vec32uc const a) {
+    return Vec32us(extend_low(a), extend_high(a));
+}
+
+// sign extend
+static inline Vec16i extend (Vec16s const a) {
+    return Vec16i(extend_low(a), extend_high(a));
+}
+
+// zero extend
+static inline Vec16ui extend (Vec16us const a) {
+    return Vec16ui(extend_low(a), extend_high(a));
+}
+
+// sign extend
+static inline Vec8q extend (Vec8i const a) {
+    return Vec8q(extend_low(a), extend_high(a));
+}
+
+// zero extend
+static inline Vec8uq extend (Vec8ui const a) {
+    return Vec8uq(extend_low(a), extend_high(a));
+}
+
+#endif  // AVX512
+
+
+/*****************************************************************************
+*
+*          Reduce from 512 to 256 bit vectors
+*
+*****************************************************************************/
+#if INSTRSET >= 9  // AVX512F
+
+// compress functions. overflow wraps around
+static inline Vec32c compress (Vec32s const a) {
+#if INSTRSET >= 10  // AVVX512BW
+    return _mm512_cvtepi16_epi8(a);
+#else
+    return compress(a.get_low(), a.get_high());
+#endif
+}
+
+static inline Vec32uc compress (Vec32us const a) {
+    return Vec32uc(compress(Vec32s(a)));
+}
+
+static inline Vec16s compress (Vec16i const a) {
+    return _mm512_cvtepi32_epi16(a);
+}
+
+static inline Vec16us compress (Vec16ui const a) {
+    return _mm512_cvtepi32_epi16(a);
+}
+
+static inline Vec8i compress (Vec8q const a) {
+    return _mm512_cvtepi64_epi32(a);
+}
+
+static inline Vec8ui compress (Vec8uq const a) {
+    return _mm512_cvtepi64_epi32(a);
+}
+
+#else  // no AVX512
+
+// compress functions. overflow wraps around
+static inline Vec32c compress (Vec32s const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec32uc compress (Vec32us const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec16s compress (Vec16i const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec16us compress (Vec16ui const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec8i compress (Vec8q const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+static inline Vec8ui compress (Vec8uq const a) {
+    return compress(a.get_low(), a.get_high());
+}
+
+#endif  // AVX512
+
+/*****************************************************************************
+*
+*          Conversions between float and double
+*
+*****************************************************************************/
+
+#if INSTRSET >= 9  // AVX512. 512 bit float vectors
+
+// float to double
+static inline Vec8d to_double (Vec8f const a) {
+    return _mm512_cvtps_pd(a);
+}
+
+// double to float
+static inline Vec8f to_float (Vec8d const a) {
+    return _mm512_cvtpd_ps(a);
+}
+
+#else  // no AVX512. 512 bit float vectors are emulated
+
+// float to double
+static inline Vec8d to_double (Vec8f const a) {
+    Vec4d lo = to_double(a.get_low());
+    Vec4d hi = to_double(a.get_high());
+    return Vec8d(lo,hi);
+}
+
+// double to float
+static inline Vec8f to_float (Vec8d const a) {
+    Vec4f lo = to_float(a.get_low());
+    Vec4f hi = to_float(a.get_high());
+    return Vec8f(lo, hi);
+}
+
+#endif
+
+#endif // MAX_VECTOR_SIZE >= 512
+
+// double to float
+static inline Vec4f to_float (Vec2d const a) {
+    return _mm_cvtpd_ps(a);
+}
+
+
+/*****************************************************************************
+*
+*          Generic template functions
+*
+*  These templates define functions for multiple vector types in one template
+*
+*****************************************************************************/
+
+// concatenate two vectors into one vector of double size
+template <typename T> auto concatenate2(T const a, T const b) {
+    static_assert(sizeof(T) * 8 < MAX_VECTOR_SIZE, "Maximum vector size exceeded");
+    return decltype(extend_z(a))(a, b);   // call constructor for double size vector type
+}
+
+
+// horizontal min/max of vector elements
+// implemented with universal template, works for all vector types:
+
+template <typename T> auto horizontal_min(T const x) {
+    if constexpr (T::elementtype() >= 15) {
+        // T is a float or double vector
+        if (horizontal_or(is_nan(x))) {
+            // check for NAN because min does not guarantee NAN propagation
+            return x[horizontal_find_first(is_nan(x))];
+        }
+    }
+    return horizontal_min1(x);
+}
+
+template <typename T> auto horizontal_min1(T const x) {
+    if constexpr (T::elementtype() <= 3) {       // boolean vector type
+        return horizontal_and(x);
+    }
+    else if constexpr (sizeof(T) >= 32) {
+        // split recursively into smaller vectors
+        return horizontal_min1(min(x.get_low(), x.get_high()));
+    }
+    else if constexpr (T::size() == 2) {
+        T a = permute2 <1, V_DC>(x);             // high half
+        T b = min(a, x);
+        return b[0];
+    }
+    else if constexpr (T::size() == 4) {
+        T a = permute4<2, 3, V_DC, V_DC>(x);     // high half
+        T b = min(a, x);
+        a = permute4<1, V_DC, V_DC, V_DC>(b);
+        b = min(a, b);
+        return b[0];
+    }
+    else if constexpr (T::size() == 8) {
+        T a = permute8<4, 5, 6, 7, V_DC, V_DC, V_DC, V_DC>(x);  // high half
+        T b = min(a, x);
+        a = permute8<2, 3, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = min(a, b);
+        a = permute8<1, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = min(a, b);
+        return b[0];
+    }
+    else {
+        static_assert(T::size() == 16);          // no other size is allowed
+        T a = permute16<8, 9, 10, 11, 12, 13, 14, 15, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC >(x);  // high half
+        T b = min(a, x);
+        a = permute16<4, 5, 6, 7, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = min(a, b);
+        a = permute16<2, 3, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = min(a, b);
+        a = permute16<1, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = min(a, b);
+        return b[0];
+    }
+}
+
+template <typename T> auto horizontal_max(T const x) {
+    if constexpr (T::elementtype() >= 15) {
+        // T is a float or double vector
+        if (horizontal_or(is_nan(x))) {
+            // check for NAN because max does not guarantee NAN propagation
+            return x[horizontal_find_first(is_nan(x))];
+        }
+    }
+    return horizontal_max1(x);
+}
+
+template <typename T> auto horizontal_max1(T const x) {
+    if constexpr (T::elementtype() <= 3) {       // boolean vector type
+        return horizontal_or(x);
+    }
+    else if constexpr (sizeof(T) >= 32) {
+        // split recursively into smaller vectors
+        return horizontal_max1(max(x.get_low(), x.get_high()));
+    }
+    else if constexpr (T::size() == 2) {
+        T a = permute2 <1, V_DC>(x);             // high half
+        T b = max(a, x);
+        return b[0];
+    }
+    else if constexpr (T::size() == 4) {
+        T a = permute4<2, 3, V_DC, V_DC>(x);     // high half
+        T b = max(a, x);
+        a = permute4<1, V_DC, V_DC, V_DC>(b);
+        b = max(a, b);
+        return b[0];
+    }
+    else if constexpr (T::size() == 8) {
+        T a = permute8<4, 5, 6, 7, V_DC, V_DC, V_DC, V_DC>(x);  // high half
+        T b = max(a, x);
+        a = permute8<2, 3, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = max(a, b);
+        a = permute8<1, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = max(a, b);
+        return b[0];
+    }
+    else {
+        static_assert(T::size() == 16);          // no other size is allowed
+        T a = permute16<8, 9, 10, 11, 12, 13, 14, 15, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC >(x);  // high half
+        T b = max(a, x);
+        a = permute16<4, 5, 6, 7, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = max(a, b);
+        a = permute16<2, 3, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = max(a, b);
+        a = permute16<1, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC, V_DC>(b);
+        b = max(a, b);
+        return b[0];
+    }
+}
+
+// Find first element that is true in a boolean vector
+template <typename V>
+static inline int horizontal_find_first(V const x) {
+    static_assert(V::elementtype() == 2 || V::elementtype() == 3, "Boolean vector expected");
+    auto bits = to_bits(x);                      // convert to bits
+    if (bits == 0) return -1;
+    if constexpr (V::size() < 32) {
+        return bit_scan_forward((uint32_t)bits);
+    }
+    else {
+        return bit_scan_forward(bits);
+    }
+}
+
+// Count the number of elements that are true in a boolean vector
+template <typename V>
+static inline int horizontal_count(V const x) {
+    static_assert(V::elementtype() == 2 || V::elementtype() == 3, "Boolean vector expected");
+    auto bits = to_bits(x);                      // convert to bits
+    if constexpr (V::size() < 32) {
+        return vml_popcnt((uint32_t)bits);
+    }
+    else {
+        return (int)vml_popcnt(bits);
+    }
+}
+
+// maximum and minimum functions. This version is sure to propagate NANs,
+// conforming to the new IEEE-754 2019 standard
+template <typename V>
+static inline V maximum(V const a, V const b) {
+    if constexpr (V::elementtype() < 15) {
+        return max(a, b);              // integer type
+    }
+    else {                             // float or double vector
+        V y = select(is_nan(a), a, max(a, b));
+#ifdef SIGNED_ZERO                     // pedantic about signed zero
+        y = select(a == b, a & b, y);  // maximum(+0, -0) = +0
+#endif
+        return y;
+    }
+}
+
+template <typename V>
+static inline V minimum(V const a, V const b) {
+    if constexpr (V::elementtype() < 15) {
+        return min(a, b);              // integer type
+    }
+    else {                             // float or double vector
+        V y = select(is_nan(a), a, min(a, b));
+#ifdef SIGNED_ZERO                     // pedantic about signed zero
+        y = select(a == b, a | b, y);  // minimum(+0, -0) = -0
+#endif
+        return y;
+    }
+}
+
+// floating point remainder
+// -denominator/2 <= result < denominator/2
+template <typename V>
+static inline V fremainder(V const numerator, double const denominator) {
+    // (Optimization notice: Calculation of 1/denominator and constants for extended precision reduction
+    // may be optimized by a compiler moving loop-invariant code. This is intended)
+    static_assert(V::elementtype() == 16 || V::elementtype() == 17, "wrong vector type"); // supports only float and double
+    if (denominator > 0.) {                                // denominator must be positive
+        if constexpr (V::elementtype() == 16) {            // float
+#ifdef __FMA__
+            float recipd = float(1.0 / denominator);       // reciprocal denominator
+            float fd = float(denominator);                 // denominator rounded to single precision
+            float d2 = float(denominator - fd);            // remaining bits for double precision
+            V q = round(numerator * recipd);               // divide and round
+            V m = nmul_add(q, d2, nmul_add(q, fd, numerator));// double precision reduction
+#else  // no FMA. Use extended precision reduction
+            union {
+                float f;
+                uint32_t i;
+            } u;
+            u.f = float(denominator);
+            u.i &= 0xFFFFF000;                             // remove 12 least significant bits for extended precision reduction
+            float d2 = denominator - u.f;                  // remaining bits
+            float recipd = float(1.0 / denominator);       // reciprocal
+            V q = round(numerator * recipd);               // divide and round
+            V m = nmul_add(q, d2, nmul_add(q, u.f, numerator));// extended precision reduction
+#endif  // FMA
+            if (true) { // Check that result is within desired interval. This may be omitted if not essential:
+                // This check may be needed in extreme cases of numerator > 1.E5 * denominator
+                auto too_high = m >= float( denominator * 0.5);
+                auto too_low  = m <  float(-denominator * 0.5);
+                m = if_sub(too_high, m, float(denominator));
+                m = if_add(too_low,  m, float(denominator));
+            }
+            return m;
+        }
+        else if constexpr (V::elementtype() == 17) {       // double precision
+#ifdef __FMA__
+            double recipd = 1.0 / denominator;             // reciprocal
+            V q = round(numerator * recipd);               // divide and round
+            V m = nmul_add(q, denominator, numerator);     // nmul_add has extended precision
+#else  // no FMA. Use extended precision reduction
+            union {
+                double f;
+                uint64_t i;
+            } u;
+            u.f = denominator;
+            u.i &= 0xFFFFFFFFFF000000;                     // remove 24 least significant bits for extended precision reduction
+            double d2 = denominator - u.f;                 // remaining bits
+            double recipd = 1.0 / denominator;             // reciprocal
+            V q = round(numerator * recipd);               // divide and round
+            V m = nmul_add(q, d2, nmul_add(q, u.f, numerator));// extended precision reduction
+#endif  // FMA
+            if (true) { // Check that result is within desired interval. This may be omitted if not essential:
+                // This check is rarely needed except in extreme cases of numerator > 1.E14 * denominator
+                auto too_high = m >=  denominator * 0.5;
+                auto too_low  = m <  -denominator * 0.5;
+                m = if_sub(too_high, m, denominator);
+                m = if_add(too_low,  m, denominator);
+            }
+            return m;
+        }
+    }
+    else {
+        return nan_vec<V>(1);                              //  denominator is not positive
+    }
+}
+
+// floating point modulo
+// 0 <= result < denominator
+template <typename V>
+static inline V fmodulo(V const numerator, double const denominator) {
+    // (Optimization notice: Calculation of 1/denominator and constants for extended precision reduction
+    // may be optimized by a compiler moving loop-invariant code. This is intended)
+    static_assert(V::elementtype() == 16 || V::elementtype() == 17, "wrong vector type"); // supports only float and double
+    if (denominator > 0.) {                                // denominator must be positive
+        if constexpr (V::elementtype() == 16) {            // float
+#ifdef __FMA__
+            float recipd = float(1.0 / denominator);       // reciprocal denominator
+            float fd = float(denominator);                 // denominator rounded to single precision
+            float d2 = float(denominator - fd);            // remaining bits for double precision
+            V q = floor(numerator * recipd);               // divide and floor
+            V m = nmul_add(q, d2, nmul_add(q, fd, numerator));// double precision reduction
+#else  // no FMA. Use extended precision reduction
+            union {
+                float f;
+                uint32_t i;
+            } u;
+            u.f = float(denominator);
+            u.i &= 0xFFFFF000;                             // remove 12 least significant bits for extended precision reduction
+            float d2 = denominator - u.f;                  // remaining bits
+            float recipd = float(1.0 / denominator);       // reciprocal
+            V q = floor(numerator * recipd);               // divide and floor
+            V m = nmul_add(q, d2, nmul_add(q, u.f, numerator));// extended precision reduction
+#endif  // FMA
+            if (true) { // Check that result is within desired interval. This may be omitted if not essential:
+                // This check may be needed in extreme cases of numerator > 1.E5 * denominator
+                auto too_high = m >= float(denominator);
+                auto too_low  = m <  0.f;
+                m = if_sub(too_high, m, float(denominator));
+                m = if_add(too_low,  m, float(denominator));
+            }
+            return m;
+        }
+        else if constexpr (V::elementtype() == 17) {       // double precision
+#ifdef __FMA__
+            double recipd = 1.0 / denominator;             // reciprocal
+            V q = floor(numerator * recipd);               // divide and floor
+            V m = nmul_add(q, denominator, numerator);     // nmul_add has extended precision
+#else  // no FMA. Use extended precision reduction
+            union {
+                double f;
+                uint64_t i;
+            } u;
+            u.f = denominator;
+            u.i &= 0xFFFFFFFFFF000000;                     // remove 24 least significant bits for extended precision reduction
+            double d2 = denominator - u.f;                 // remaining bits
+            double recipd = 1.0 / denominator;             // reciprocal
+            V q = floor(numerator * recipd);               // divide and floor
+            V m = nmul_add(q, d2, nmul_add(q, u.f, numerator));// extended precision reduction
+#endif  // FMA
+            if (true) { // Check that result is within desired interval. This may be omitted if not essential:
+                // This check is rarely needed except in extreme cases of numerator > 1.E14 * denominator
+                auto too_high = m >= denominator;
+                auto too_low  = m <  0.;
+                m = if_sub(too_high, m, denominator);
+                m = if_add(too_low,  m, denominator);
+            }
+            return m;
+        }
+    }
+    else {
+        return nan_vec<V>(1);                              //  denominator is not positive
+    }
+}
+
+#ifdef VCL_NAMESPACE
+}
+#endif
+
+#endif // VECTOR_CONVERT_H
@@ -0,0 +1,87 @@
+/****************************  vectorclass.h   ********************************
+* Author:        Agner Fog
+* Date created:  2012-05-30
+* Last modified: 2026-04-14
+* Version:       2.02.03
+* Project:       vector class library
+* Home:          https://github.com/vectorclass
+* Description:
+* Header file defining vector classes as interface to intrinsic functions
+* in x86 and x86-64 microprocessors with SSE2 and later instruction sets.
+*
+* Instructions:
+* Use Gnu, Clang, Microsoft, or Intel C++ compiler. Compile for the desired
+* instruction set, which must be at least SSE2. Specify the supported
+* instruction set by a command line define, e.g. __SSE4_1__ if the
+* compiler does not automatically do so.
+* For detailed instructions, see vcl_manual.pdf
+*
+* Each vector object is represented internally in the CPU as a vector
+* register with 128, 256 or 512 bits.
+*
+* This header file includes the appropriate header files depending on the
+* selected instruction set.
+*
+* (c) Copyright 2012-2026 Agner Fog.
+* Apache License version 2.0 or later.
+******************************************************************************/
+
+#ifndef VECTORCLASS_H
+#define VECTORCLASS_H  20203
+
+// Maximum vector size, bits. Allowed values are 128, 256, 512
+#ifndef MAX_VECTOR_SIZE
+#define MAX_VECTOR_SIZE 512
+#endif
+
+// Determine instruction set, and define platform-dependent functions
+#include "instrset.h"        // Select supported instruction set
+
+#if INSTRSET < 2             // instruction set SSE2 is the minimum
+#error Please compile for the SSE2 instruction set or higher
+#else
+
+// Select appropriate .h files depending on instruction set
+#include "vectori128.h"      // 128-bit integer vectors
+#include "vectorf128.h"      // 128-bit floating point vectors
+
+#if MAX_VECTOR_SIZE >= 256
+#if INSTRSET >= 8
+#include "vectori256.h"      // 256-bit integer vectors, requires AVX2 instruction set
+#else
+#include "vectori256e.h"     // 256-bit integer vectors, emulated
+#endif  // INSTRSET >= 8
+#if INSTRSET >= 7
+#include "vectorf256.h"      // 256-bit floating point vectors, requires AVX instruction set
+#else
+#include "vectorf256e.h"     // 256-bit floating point vectors, emulated
+#endif  //  INSTRSET >= 7
+#endif  //  MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+#if INSTRSET >= 9
+#include "vectori512.h"      // 512-bit vectors of 32 and 64 bit integers, requires AVX512F instruction set
+#include "vectorf512.h"      // 512-bit floating point vectors, requires AVX512F instruction set
+#else
+#include "vectori512e.h"     // 512-bit integer vectors, emulated
+#include "vectorf512e.h"     // 512-bit floating point vectors, emulated
+#endif  //  INSTRSET >= 9
+#if INSTRSET >= 10
+#include "vectori512s.h"     // 512-bit vectors of 8 and 16 bit integers, requires AVX512BW instruction set
+#else
+#include "vectori512se.h"    // 512-bit vectors of 8 and 16 bit integers, emulated
+#endif
+#endif  //  MAX_VECTOR_SIZE >= 512
+
+#include "vector_convert.h"  // conversion between different vector sizes, and common templates
+
+#endif  // INSTRSET >= 2
+
+
+#else   // VECTORCLASS_H
+
+#if VECTORCLASS_H < 20000
+#error Mixed versions of vector class library
+#endif
+
+#endif  // VECTORCLASS_H
@@ -0,0 +1,327 @@
+/***************************  vectormath_common.h   ****************************
+* Author:        Agner Fog
+* Date created:  2014-04-18
+* Last modified: 2022-07-20
+* Version:       2.02.00
+* Project:       vector classes
+* Description:
+* Header file containing common code for inline version of mathematical functions.
+*
+* For detailed instructions, see VectorClass.pdf
+*
+* (c) Copyright 2014-2022 Agner Fog.
+* Apache License version 2.0 or later.
+******************************************************************************/
+
+#ifndef VECTORMATH_COMMON_H
+#define VECTORMATH_COMMON_H  2
+
+#ifdef VECTORMATH_LIB_H
+#error conflicting header files. More than one implementation of mathematical functions included
+#endif
+
+#include <cmath>
+
+#ifndef VECTORCLASS_H
+#include "vectorclass.h"
+#endif
+
+#if VECTORCLASS_H < 20200
+#error Incompatible versions of vector class library mixed
+#endif
+
+
+/******************************************************************************
+                    Define NAN payload values
+******************************************************************************/
+#define NAN_LOG 0x101  // logarithm for x<0
+#define NAN_POW 0x102  // negative number raised to non-integer power
+#define NAN_HYP 0x104  // acosh for x<1 and atanh for abs(x)>1
+
+
+/******************************************************************************
+                    Define mathematical constants
+******************************************************************************/
+#define VM_PI       3.14159265358979323846           // pi
+#define VM_PI_2     1.57079632679489661923           // pi / 2
+#define VM_PI_4     0.785398163397448309616          // pi / 4
+#define VM_SQRT2    1.41421356237309504880           // sqrt(2)
+#define VM_LOG2E    1.44269504088896340736           // 1/log(2)
+#define VM_LOG10E   0.434294481903251827651          // 1/log(10)
+#define VM_LOG210   3.321928094887362347808          // log2(10)
+#define VM_LN2      0.693147180559945309417          // log(2)
+#define VM_LN10     2.30258509299404568402           // log(10)
+#define VM_SMALLEST_NORMAL  2.2250738585072014E-308  // smallest normal number, double
+#define VM_SMALLEST_NORMALF 1.17549435E-38f          // smallest normal number, float
+
+
+#ifdef VCL_NAMESPACE
+namespace VCL_NAMESPACE {
+#endif
+
+/******************************************************************************
+      templates for producing infinite and nan in desired vector type
+******************************************************************************/
+template <typename VTYPE>
+static inline VTYPE infinite_vec();
+
+template <>
+inline Vec2d infinite_vec<Vec2d>() {
+    return infinite2d();
+}
+
+template <>
+inline Vec4f infinite_vec<Vec4f>() {
+    return infinite4f();
+}
+
+#if MAX_VECTOR_SIZE >= 256
+
+template <>
+inline Vec4d infinite_vec<Vec4d>() {
+    return infinite4d();
+}
+
+template <>
+inline Vec8f infinite_vec<Vec8f>() {
+    return infinite8f();
+}
+
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+
+template <>
+inline Vec8d infinite_vec<Vec8d>() {
+    return infinite8d();
+}
+
+template <>
+inline Vec16f infinite_vec<Vec16f>() {
+    return infinite16f();
+}
+
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+
+/******************************************************************************
+*                 Detect NAN codes
+*
+* These functions return the code hidden in a NAN. The sign bit is ignored
+******************************************************************************/
+
+static inline Vec4ui nan_code(Vec4f const x) {
+    Vec4ui a = Vec4ui(reinterpret_i(x));
+    Vec4ui const n = 0x007FFFFF;
+    return select(Vec4ib(is_nan(x)), a & n, 0);
+}
+
+// This function returns the code hidden in a NAN. The sign bit is ignored
+static inline Vec2uq nan_code(Vec2d const x) {
+    Vec2uq a = Vec2uq(reinterpret_i(x));
+    return select(Vec2qb(is_nan(x)), a << 12 >> (12+29), 0);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+
+// This function returns the code hidden in a NAN. The sign bit is ignored
+static inline Vec8ui nan_code(Vec8f const x) {
+    Vec8ui a = Vec8ui(reinterpret_i(x));
+    Vec8ui const n = 0x007FFFFF;
+    return select(Vec8ib(is_nan(x)), a & n, 0);
+}
+
+// This function returns the code hidden in a NAN. The sign bit is ignored
+static inline Vec4uq nan_code(Vec4d const x) {
+    Vec4uq a = Vec4uq(reinterpret_i(x));
+    return select(Vec4qb(is_nan(x)), a << 12 >> (12+29), 0);
+}
+
+#endif // MAX_VECTOR_SIZE >= 256
+#if MAX_VECTOR_SIZE >= 512
+
+// This function returns the code hidden in a NAN. The sign bit is ignored
+static inline Vec16ui nan_code(Vec16f const x) {
+    Vec16ui a = Vec16ui(reinterpret_i(x));
+    Vec16ui const n = 0x007FFFFF;
+    return select(Vec16ib(is_nan(x)), a & n, 0);
+}
+
+// This function returns the code hidden in a NAN. The sign bit is ignored
+static inline Vec8uq nan_code(Vec8d const x) {
+    Vec8uq a = Vec8uq(reinterpret_i(x));
+    return select(Vec8qb(is_nan(x)), a << 12 >> (12+29), 0);
+}
+
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+/******************************************************************************
+                  templates for polynomials
+Using Estrin's scheme to make shorter dependency chains and use FMA, starting
+longest dependency chains first.
+******************************************************************************/
+
+// template <typedef VECTYPE, typedef CTYPE>
+template <class VTYPE, class CTYPE>
+static inline VTYPE polynomial_2(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2) {
+    // calculates polynomial c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    //return = x2 * c2 + (x * c1 + c0);
+    return mul_add(x2, c2, mul_add(x, c1, c0));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_3(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3) {
+    // calculates polynomial c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    //return (c2 + c3*x)*x2 + (c1*x + c0);
+    return mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_4(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4) {
+    // calculates polynomial c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    VTYPE x4 = x2 * x2;
+    //return (c2+c3*x)*x2 + ((c0+c1*x) + c4*x4);
+    return mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0) + c4*x4);
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_4n(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3) {
+    // calculates polynomial 1*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    VTYPE x4 = x2 * x2;
+    //return (c2+c3*x)*x2 + ((c0+c1*x) + x4);
+    return mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0) + x4);
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_5(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5) {
+    // calculates polynomial c5*x^5 + c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    VTYPE x4 = x2 * x2;
+    //return (c2+c3*x)*x2 + ((c4+c5*x)*x4 + (c0+c1*x));
+    return mul_add(mul_add(c3, x, c2), x2, mul_add(mul_add(c5, x, c4), x4, mul_add(c1, x, c0)));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_5n(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4) {
+    // calculates polynomial 1*x^5 + c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    VTYPE x4 = x2 * x2;
+    //return (c2+c3*x)*x2 + ((c4+x)*x4 + (c0+c1*x));
+    return mul_add(mul_add(c3, x, c2), x2, mul_add(c4 + x, x4, mul_add(c1, x, c0)));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_6(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5, CTYPE c6) {
+    // calculates polynomial c6*x^6 + c5*x^5 + c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    VTYPE x4 = x2 * x2;
+    //return  (c4+c5*x+c6*x2)*x4 + ((c2+c3*x)*x2 + (c0+c1*x));
+    return mul_add(mul_add(c6, x2, mul_add(c5, x, c4)), x4, mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0)));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_6n(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5) {
+    // calculates polynomial 1*x^6 + c5*x^5 + c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    VTYPE x4 = x2 * x2;
+    //return  (c4+c5*x+x2)*x4 + ((c2+c3*x)*x2 + (c0+c1*x));
+    return mul_add(mul_add(c5, x, c4 + x2), x4, mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0)));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_7(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5, CTYPE c6, CTYPE c7) {
+    // calculates polynomial c7*x^7 + c6*x^6 + c5*x^5 + c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x * x;
+    VTYPE x4 = x2 * x2;
+    //return  ((c6+c7*x)*x2 + (c4+c5*x))*x4 + ((c2+c3*x)*x2 + (c0+c1*x));
+    return mul_add(mul_add(mul_add(c7, x, c6), x2, mul_add(c5, x, c4)), x4, mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0)));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_8(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5, CTYPE c6, CTYPE c7, CTYPE c8) {
+    // calculates polynomial c8*x^8 + c7*x^7 + c6*x^6 + c5*x^5 + c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x  * x;
+    VTYPE x4 = x2 * x2;
+    VTYPE x8 = x4 * x4;
+    //return  ((c6+c7*x)*x2 + (c4+c5*x))*x4 + (c8*x8 + (c2+c3*x)*x2 + (c0+c1*x));
+    return mul_add(mul_add(mul_add(c7, x, c6), x2, mul_add(c5, x, c4)), x4,
+        mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0) + c8*x8));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_9(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5, CTYPE c6, CTYPE c7, CTYPE c8, CTYPE c9) {
+    // calculates polynomial c9*x^9 + c8*x^8 + c7*x^7 + c6*x^6 + c5*x^5 + c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x  * x;
+    VTYPE x4 = x2 * x2;
+    VTYPE x8 = x4 * x4;
+    //return  (((c6+c7*x)*x2 + (c4+c5*x))*x4 + (c8+c9*x)*x8) + ((c2+c3*x)*x2 + (c0+c1*x));
+    return mul_add(mul_add(c9, x, c8), x8, mul_add(
+        mul_add(mul_add(c7, x, c6), x2, mul_add(c5, x, c4)), x4,
+        mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0))));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_10(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5, CTYPE c6, CTYPE c7, CTYPE c8, CTYPE c9, CTYPE c10) {
+    // calculates polynomial c10*x^10 + c9*x^9 + c8*x^8 + c7*x^7 + c6*x^6 + c5*x^5 + c4*x^4 + c3*x^3 + c2*x^2 + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x  * x;
+    VTYPE x4 = x2 * x2;
+    VTYPE x8 = x4 * x4;
+    //return  (((c6+c7*x)*x2 + (c4+c5*x))*x4 + (c8+c9*x+c10*x2)*x8) + ((c2+c3*x)*x2 + (c0+c1*x));
+    return mul_add(mul_add(x2, c10, mul_add(c9, x, c8)), x8,
+        mul_add(mul_add(mul_add(c7, x, c6), x2, mul_add(c5, x, c4)), x4,
+            mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0))));
+}
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_13(VTYPE const x, CTYPE c0, CTYPE c1, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5, CTYPE c6, CTYPE c7, CTYPE c8, CTYPE c9, CTYPE c10, CTYPE c11, CTYPE c12, CTYPE c13) {
+    // calculates polynomial c13*x^13 + c12*x^12 + ... + c1*x + c0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x  * x;
+    VTYPE x4 = x2 * x2;
+    VTYPE x8 = x4 * x4;
+    return mul_add(
+        mul_add(
+            mul_add(c13, x, c12), x4,
+            mul_add(mul_add(c11, x, c10), x2, mul_add(c9, x, c8))), x8,
+        mul_add(
+            mul_add(mul_add(c7, x, c6), x2, mul_add(c5, x, c4)), x4,
+            mul_add(mul_add(c3, x, c2), x2, mul_add(c1, x, c0))));
+}
+
+
+template<class VTYPE, class CTYPE>
+static inline VTYPE polynomial_13m(VTYPE const x, CTYPE c2, CTYPE c3, CTYPE c4, CTYPE c5, CTYPE c6, CTYPE c7, CTYPE c8, CTYPE c9, CTYPE c10, CTYPE c11, CTYPE c12, CTYPE c13) {
+    // calculates polynomial c13*x^13 + c12*x^12 + ... + x + 0
+    // VTYPE may be a vector type, CTYPE is a scalar type
+    VTYPE x2 = x  * x;
+    VTYPE x4 = x2 * x2;
+    VTYPE x8 = x4 * x4;
+    // return  ((c8+c9*x) + (c10+c11*x)*x2 + (c12+c13*x)*x4)*x8 + (((c6+c7*x)*x2 + (c4+c5*x))*x4 + ((c2+c3*x)*x2 + x));
+    return mul_add(
+        mul_add(mul_add(c13, x, c12), x4, mul_add(mul_add(c11, x, c10), x2, mul_add(c9, x, c8))), x8,
+        mul_add(mul_add(mul_add(c7, x, c6), x2, mul_add(c5, x, c4)), x4, mul_add(mul_add(c3, x, c2), x2, x)));
+}
+
+#ifdef VCL_NAMESPACE
+}
+#endif
+
+#endif
@@ -0,0 +1,717 @@
+/****************************  vectormath_hyp.h   ******************************
+* Author:        Agner Fog
+* Date created:  2014-07-09
+* Last modified: 2022-07-20
+* Version:       2.02.00
+* Project:       vector class library
+* Description:
+* Header file containing inline vector functions of hyperbolic and inverse
+* hyperbolic functions:
+* sinh        hyperbolic sine
+* cosh        hyperbolic cosine
+* tanh        hyperbolic tangent
+* asinh       inverse hyperbolic sine
+* acosh       inverse hyperbolic cosine
+* atanh       inverse hyperbolic tangent
+*
+* Theory, methods and inspiration based partially on these sources:
+* > Moshier, Stephen Lloyd Baluk: Methods and programs for mathematical functions.
+*   Ellis Horwood, 1989.
+* > VDT library developed on CERN by Danilo Piparo, Thomas Hauth and
+*   Vincenzo Innocente, 2012, https://svnweb.cern.ch/trac/vdt
+* > Cephes math library by Stephen L. Moshier 1992,
+*   http://www.netlib.org/cephes/
+*
+* For detailed instructions, see vectormath_common.h and vcl_manual.pdf
+*
+* (c) Copyright 2014-2022 Agner Fog.
+* Apache License version 2.0 or later.
+******************************************************************************/
+
+#ifndef VECTORMATH_HYP_H
+#define VECTORMATH_HYP_H  202
+
+#include "vectormath_exp.h"
+
+#ifdef VCL_NAMESPACE
+namespace VCL_NAMESPACE {
+#endif
+
+/******************************************************************************
+*                 Hyperbolic functions
+******************************************************************************/
+
+// Template for sinh function, double precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE sinh_d(VTYPE const x0) {
+// The limit of abs(x) is 709.7, as defined by max_x in vectormath_exp.h for 0.5*exp(x).
+
+    // Coefficients
+    const double p0 = -3.51754964808151394800E5;
+    const double p1 = -1.15614435765005216044E4;
+    const double p2 = -1.63725857525983828727E2;
+    const double p3 = -7.89474443963537015605E-1;
+
+    const double q0 = -2.11052978884890840399E6;
+    const double q1 =  3.61578279834431989373E4;
+    const double q2 = -2.77711081420602794433E2;
+    const double q3 =  1.0;
+
+    // data vectors
+    VTYPE  x, x2, y1, y2;
+
+    x = abs(x0);
+    auto x_small = x <= 1.0;                     // use Pade approximation if abs(x) <= 1
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        x2 = x*x;
+        y1 = polynomial_3(x2, p0, p1, p2, p3) / polynomial_3(x2, q0, q1, q2, q3);
+        y1 = mul_add(y1, x*x2, x);               // y1 = x + x2*(x*y1);
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 =  exp_d<VTYPE, 0, 1>(x);             //   0.5 * exp(x)
+        y2 -= 0.25 / y2;                         // - 0.5 * exp(-x)
+    }
+    y1 = select(x_small, y1, y2);                // choose method
+    y1 = sign_combine(y1, x0);                   // get original sign
+    // you can avoid the sign_combine by replacing x by x0 above, but at a loss of precision
+
+    return y1;
+}
+
+// instances of sinh_d template
+static inline Vec2d sinh(Vec2d const x) {
+    return sinh_d(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec4d sinh(Vec4d const x) {
+    return sinh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec8d sinh(Vec8d const x) {
+    return sinh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for sinh function, single precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE sinh_f(VTYPE const x0) {
+// The limit of abs(x) is 89.0, as defined by max_x in vectormath_exp.h for 0.5*exp(x).
+
+    // Coefficients
+    const float r0 = 1.66667160211E-1f;
+    const float r1 = 8.33028376239E-3f;
+    const float r2 = 2.03721912945E-4f;
+
+    // data vectors
+    VTYPE x, x2, y1, y2;
+
+    x = abs(x0);
+    auto x_small = x <= 1.0f;                    // use polynomial approximation if abs(x) <= 1
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        x2 = x*x;
+        y1 = polynomial_2(x2, r0, r1, r2);
+        y1 = mul_add(y1, x2*x, x);               // y1 = x + x2*(x*y1);
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 =  exp_f<VTYPE, 0, 1>(x);             //   0.5 * exp(x)
+        y2 -= 0.25f / y2;                        // - 0.5 * exp(-x)
+    }
+    y1 = select(x_small, y1, y2);                // choose method
+    y1 = sign_combine(y1, x0);                   // get original sign
+    // you can avoid the sign_combine by replacing x by x0 above, but at a loss of precision
+
+    return y1;
+}
+
+// instances of sinh_f template
+static inline Vec4f sinh(Vec4f const x) {
+    return sinh_f(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec8f sinh(Vec8f const x) {
+    return sinh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec16f sinh(Vec16f const x) {
+    return sinh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for cosh function, double precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE cosh_d(VTYPE const x0) {
+// The limit of abs(x) is 709.7, as defined by max_x in vectormath_exp.h for 0.5*exp(x).
+
+    // data vectors
+    VTYPE x, y;
+    x  = abs(x0);
+    y  = exp_d<VTYPE, 0, 1>(x);                  //   0.5 * exp(x)
+    y += 0.25 / y;                               // + 0.5 * exp(-x)
+    return y;
+}
+
+// instances of sinh_d template
+static inline Vec2d cosh(Vec2d const x) {
+    return cosh_d(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec4d cosh(Vec4d const x) {
+    return cosh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec8d cosh(Vec8d const x) {
+    return cosh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for cosh function, single precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE cosh_f(VTYPE const x0) {
+// The limit of abs(x) is 89.0, as defined by max_x in vectormath_exp.h for 0.5*exp(x).
+
+    // data vectors
+    VTYPE x, y;
+    x  = abs(x0);
+    y  = exp_f<VTYPE, 0, 1>(x);                  //   0.5 * exp(x)
+    y += 0.25f / y;                              // + 0.5 * exp(-x)
+    return y;
+}
+
+// instances of sinh_d template
+static inline Vec4f cosh(Vec4f const x) {
+    return cosh_f(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec8f cosh(Vec8f const x) {
+    return cosh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec16f cosh(Vec16f const x) {
+    return cosh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for tanh function, double precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE tanh_d(VTYPE const x0) {
+
+    // Coefficients
+    const double p0 = -1.61468768441708447952E3;
+    const double p1 = -9.92877231001918586564E1;
+    const double p2 = -9.64399179425052238628E-1;
+
+    const double q0 =  4.84406305325125486048E3;
+    const double q1 =  2.23548839060100448583E3;
+    const double q2 =  1.12811678491632931402E2;
+    const double q3 =  1.0;
+
+    // data vectors
+    VTYPE  x, x2, y1, y2;
+
+    x = abs(x0);
+    auto x_small = x <= 0.625;                   // use Pade approximation if abs(x) <= 5/8
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        x2 = x*x;
+        y1 = polynomial_2(x2, p0, p1, p2) / polynomial_3(x2, q0, q1, q2, q3);
+        y1 = mul_add(y1, x2*x, x);               // y1 = x + x2*(x*y1);
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 = exp(x+x);                           // exp(2*x)
+        y2 = 1.0 - 2.0 / (y2 + 1.0);             // tanh(x)
+    }
+    auto x_big = x > 350.;
+    y1 = select(x_small, y1, y2);                // choose method
+    y1 = select(x_big,  1.0, y1);                // avoid overflow
+    y1 = sign_combine(y1, x0);                   // get original sign
+    return y1;
+}
+
+// instances of tanh_d template
+static inline Vec2d tanh(Vec2d const x) {
+    return tanh_d(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec4d tanh(Vec4d const x) {
+    return tanh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec8d tanh(Vec8d const x) {
+    return tanh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for tanh function, single precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE tanh_f(VTYPE const x0) {
+// The limit of abs(x) is 89.0, as defined by max_x in vectormath_exp.h for 0.5*exp(x).
+
+    // Coefficients
+    const float r0 = -3.33332819422E-1f;
+    const float r1 =  1.33314422036E-1f;
+    const float r2 = -5.37397155531E-2f;
+    const float r3 =  2.06390887954E-2f;
+    const float r4 = -5.70498872745E-3f;
+
+    // data vectors
+    VTYPE x, x2, y1, y2;
+
+    x = abs(x0);
+    auto x_small = x <= 0.625f;                  // use polynomial approximation if abs(x) <= 5/8
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        x2 = x*x;
+        y1 = polynomial_4(x2, r0, r1, r2, r3, r4);
+        y1 = mul_add(y1, x2*x, x);               // y1 = x + (x2*x)*y1;
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 = exp(x+x);                           // exp(2*x)
+        y2 = 1.0f - 2.0f / (y2 + 1.0f);          // tanh(x)
+    }
+    auto x_big = x > 44.4f;
+    y1 = select(x_small, y1, y2);                // choose method
+    y1 = select(x_big,  1.0f, y1);               // avoid overflow
+    y1 = sign_combine(y1, x0);                   // get original sign
+    return y1;
+}
+
+// instances of tanh_f template
+static inline Vec4f tanh(Vec4f const x) {
+    return tanh_f(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec8f tanh(Vec8f const x) {
+    return tanh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec16f tanh(Vec16f const x) {
+    return tanh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+
+/******************************************************************************
+*                 Inverse hyperbolic functions
+******************************************************************************/
+
+// Template for asinh function, double precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE asinh_d(VTYPE const x0) {
+
+    // Coefficients
+    const double p0 = -5.56682227230859640450E0;
+    const double p1 = -9.09030533308377316566E0;
+    const double p2 = -4.37390226194356683570E0;
+    const double p3 = -5.91750212056387121207E-1;
+    const double p4 = -4.33231683752342103572E-3;
+
+    const double q0 =  3.34009336338516356383E1;
+    const double q1 =  6.95722521337257608734E1;
+    const double q2 =  4.86042483805291788324E1;
+    const double q3 =  1.28757002067426453537E1;
+    const double q4 =  1.0;
+
+    // data vectors
+    VTYPE  x, x2, y1, y2;
+
+    x2 = x0 * x0;
+    x  = abs(x0);
+    auto x_small = x <= 0.533;                   // use Pade approximation if abs(x) <= 0.5
+    // Both methods give the highest error close to 0.5.
+    // This limit is adjusted for minimum error
+    auto x_huge  = x > 1.E20;                    // simple approximation, avoid overflow
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        y1 = polynomial_4(x2, p0, p1, p2, p3, p4) / polynomial_4(x2, q0, q1, q2, q3, q4);
+        y1 = mul_add(y1, x2*x, x);               // y1 = x + (x2*x)*y1;
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 = log(x + sqrt(x2 + 1.0));
+        if (horizontal_or(x_huge)) {
+            // At least one element needs huge method to avoid overflow
+            y2 = select(x_huge, log(x) + VM_LN2, y2);
+        }
+    }
+    y1 = select(x_small, y1, y2);                // choose method
+    y1 = sign_combine(y1, x0);                   // get original sign
+    return y1;
+}
+
+// instances of asinh_d template
+static inline Vec2d asinh(Vec2d const x) {
+    return asinh_d(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec4d asinh(Vec4d const x) {
+    return asinh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec8d asinh(Vec8d const x) {
+    return asinh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for asinh function, single precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE asinh_f(VTYPE const x0) {
+
+    // Coefficients
+    const float r0 = -1.6666288134E-1f;
+    const float r1 =  7.4847586088E-2f;
+    const float r2 = -4.2699340972E-2f;
+    const float r3 =  2.0122003309E-2f;
+
+    // data vectors
+    VTYPE  x, x2, y1, y2;
+
+    x2 = x0 * x0;
+    x  = abs(x0);
+    auto x_small = x <= 0.51f;                   // use polynomial approximation if abs(x) <= 0.5
+    auto x_huge  = x > 1.E10f;                   // simple approximation, avoid overflow
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        y1 = polynomial_3(x2, r0, r1, r2, r3);
+        y1 = mul_add(y1, x2*x, x);               // y1 = x + (x2*x)*y1;
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 = log(x + sqrt(x2 + 1.0f));
+        if (horizontal_or(x_huge)) {
+            // At least one element needs huge method to avoid overflow
+            y2 = select(x_huge, log(x) + (float)VM_LN2, y2);
+        }
+    }
+    y1 = select(x_small, y1, y2);                // choose method
+    y1 = sign_combine(y1, x0);                   // get original sign
+    return y1;
+}
+
+// instances of asinh_f template
+static inline Vec4f asinh(Vec4f const x) {
+    return asinh_f(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec8f asinh(Vec8f const x) {
+    return asinh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec16f asinh(Vec16f const x) {
+    return asinh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for acosh function, double precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE acosh_d(VTYPE const x0) {
+
+    // Coefficients
+    const double p0 = 1.10855947270161294369E5;
+    const double p1 = 1.08102874834699867335E5;
+    const double p2 = 3.43989375926195455866E4;
+    const double p3 = 3.94726656571334401102E3;
+    const double p4 = 1.18801130533544501356E2;
+
+    const double q0 = 7.83869920495893927727E4;
+    const double q1 = 8.29725251988426222434E4;
+    const double q2 = 2.97683430363289370382E4;
+    const double q3 = 4.15352677227719831579E3;
+    const double q4 = 1.86145380837903397292E2;
+    const double q5 = 1.0;
+
+    // data vectors
+    VTYPE  x1, y1, y2;
+
+    x1      = x0 - 1.0;
+    auto undef   = x0 < 1.0;                     // result is NAN
+    auto x_small = x1 < 0.49;                    // use Pade approximation if abs(x-1) < 0.5
+    auto x_huge  = x1 > 1.E20;                   // simple approximation, avoid overflow
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        y1 = sqrt(x1) * (polynomial_4(x1, p0, p1, p2, p3, p4) / polynomial_5(x1, q0, q1, q2, q3, q4, q5));
+        // x < 1 generates NAN
+        y1 = select(undef, nan_vec<VTYPE>(NAN_HYP), y1);
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 = log(x0 + sqrt(mul_sub(x0,x0,1.0)));
+        if (horizontal_or(x_huge)) {
+            // At least one element needs huge method to avoid overflow
+            y2 = select(x_huge, log(x0) + VM_LN2, y2);
+        }
+    }
+    y1 = select(x_small, y1, y2);                // choose method
+    return y1;
+}
+
+// instances of acosh_d template
+static inline Vec2d acosh(Vec2d const x) {
+    return acosh_d(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec4d acosh(Vec4d const x) {
+    return acosh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec8d acosh(Vec8d const x) {
+    return acosh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for acosh function, single precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE acosh_f(VTYPE const x0) {
+
+    // Coefficients
+    const float r0 =  1.4142135263E0f;
+    const float r1 = -1.1784741703E-1f;
+    const float r2 =  2.6454905019E-2f;
+    const float r3 = -7.5272886713E-3f;
+    const float r4 =  1.7596881071E-3f;
+
+    // data vectors
+    VTYPE  x1, y1, y2;
+
+    x1      = x0 - 1.0f;
+    auto undef   = x0 < 1.0f;                    // result is NAN
+    auto x_small = x1 < 0.49f;                   // use Pade approximation if abs(x-1) < 0.5
+    auto x_huge  = x1 > 1.E10f;                  // simple approximation, avoid overflow
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        y1 = sqrt(x1) * polynomial_4(x1, r0, r1, r2, r3, r4);
+        // x < 1 generates NAN
+        y1 = select(undef, nan_vec<VTYPE>(NAN_HYP), y1);
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 = log(x0 + sqrt(mul_sub(x0,x0,1.0)));
+        if (horizontal_or(x_huge)) {
+            // At least one element needs huge method to avoid overflow
+            y2 = select(x_huge, log(x0) + (float)VM_LN2, y2);
+        }
+    }
+    y1 = select(x_small, y1, y2);                // choose method
+    return y1;
+}
+
+// instances of acosh_f template
+static inline Vec4f acosh(Vec4f const x) {
+    return acosh_f(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec8f acosh(Vec8f const x) {
+    return acosh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec16f acosh(Vec16f const x) {
+    return acosh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for atanh function, double precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE atanh_d(VTYPE const x0) {
+
+    // Coefficients
+    const double p0 = -3.09092539379866942570E1;
+    const double p1 =  6.54566728676544377376E1;
+    const double p2 = -4.61252884198732692637E1;
+    const double p3 =  1.20426861384072379242E1;
+    const double p4 = -8.54074331929669305196E-1;
+
+    const double q0 = -9.27277618139601130017E1;
+    const double q1 =  2.52006675691344555838E2;
+    const double q2 = -2.49839401325893582852E2;
+    const double q3 =  1.08938092147140262656E2;
+    const double q4 = -1.95638849376911654834E1;
+    const double q5 =  1.0;
+
+    // data vectors
+    VTYPE  x, x2, y1, y2, y3;
+
+    x  = abs(x0);
+    auto x_small = x < 0.5;                      // use Pade approximation if abs(x) < 0.5
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        x2 = x * x;
+        y1 = polynomial_4(x2, p0, p1, p2, p3, p4) / polynomial_5(x2, q0, q1, q2, q3, q4, q5);
+        y1 = mul_add(y1, x2*x, x);
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 = log((1.0+x)/(1.0-x)) * 0.5;
+        // check if out of range
+        y3 = select(x == 1.0, infinite_vec<VTYPE>(), nan_vec<VTYPE>(NAN_HYP));
+        y2 = select(x >= 1.0, y3, y2);
+    }
+    y1 = select(x_small, y1, y2);                // choose method
+    y1 = sign_combine(y1, x0);                   // get original sign
+    return y1;
+}
+
+// instances of atanh_d template
+static inline Vec2d atanh(Vec2d const x) {
+    return atanh_d(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec4d atanh(Vec4d const x) {
+    return atanh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec8d atanh(Vec8d const x) {
+    return atanh_d(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+
+// Template for atanh function, single precision
+// This function does not produce denormals
+// Template parameters:
+// VTYPE:  double vector type
+template<typename VTYPE>
+static inline VTYPE atanh_f(VTYPE const x0) {
+
+    // Coefficients
+    const float r0 = 3.33337300303E-1f;
+    const float r1 = 1.99782164500E-1f;
+    const float r2 = 1.46691431730E-1f;
+    const float r3 = 8.24370301058E-2f;
+    const float r4 = 1.81740078349E-1f;
+
+    // data vectors
+    VTYPE  x, x2, y1, y2, y3;
+
+    x  = abs(x0);
+    auto x_small = x < 0.5f;                     // use polynomial approximation if abs(x) < 0.5
+
+    if (horizontal_or(x_small)) {
+        // At least one element needs small method
+        x2 = x * x;
+        y1 = polynomial_4(x2, r0, r1, r2, r3, r4);
+        y1 = mul_add(y1, x2*x, x);
+    }
+    if (!horizontal_and(x_small)) {
+        // At least one element needs big method
+        y2 = log((1.0f+x)/(1.0f-x)) * 0.5f;
+        // check if out of range
+        y3 = select(x == 1.0f, infinite_vec<VTYPE>(), nan_vec<VTYPE>(NAN_HYP));
+        y2 = select(x >= 1.0f, y3, y2);
+    }
+    y1 = select(x_small, y1, y2);                // choose method
+    y1 = sign_combine(y1, x0);                   // get original sign
+    return y1;
+}
+
+// instances of atanh_f template
+static inline Vec4f atanh(Vec4f const x) {
+    return atanh_f(x);
+}
+
+#if MAX_VECTOR_SIZE >= 256
+static inline Vec8f atanh(Vec8f const x) {
+    return atanh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 256
+
+#if MAX_VECTOR_SIZE >= 512
+static inline Vec16f atanh(Vec16f const x) {
+    return atanh_f(x);
+}
+#endif // MAX_VECTOR_SIZE >= 512
+
+#ifdef VCL_NAMESPACE
+}
+#endif
+
+#endif
@@ -7,7 +7,7 @@

 #]=======================================================================]

-cmake_minimum_required(VERSION 3.20)
+cmake_minimum_required(VERSION 3.24)
 project(NanoVDB LANGUAGES C CXX)

 include(GNUInstallDirs)
@@ -31,6 +31,7 @@ option(NANOVDB_BUILD_PYTHON_MODULE "Build the nanovdb Python module" OFF)
 option(NANOVDB_USE_INTRINSICS "Build with hardware intrinsics support" OFF)
 option(NANOVDB_USE_CUDA "Build with CUDA support" OFF)
 option(NANOVDB_CUDA_KEEP_PTX "Keep CUDA PTX" OFF)
+option(NANOVDB_USE_OLD_ACCESSOR "Use the old (buggy) ReadAccessor instead of the new fixed one" OFF)

 option(NANOVDB_USE_OPENVDB "Build with OpenVDB support" OFF)
 option(NANOVDB_USE_BLOSC "Build with BLOSC support" ${USE_BLOSC})
@@ -115,6 +116,18 @@ if(NANOVDB_USE_CUDA)
  endif()

  find_package(CUDAToolkit)
+
+
+  include(../../cmake/CPM.cmake)
+
+  # This will automatically clone CCCL from GitHub and make the exported cmake targets available
+  CPMAddPackage(
+      NAME CCCL
+      GITHUB_REPOSITORY nvidia/cccl
+      GIT_TAG main # Fetches the latest commit on the main branch
+  )
+
+  find_package(NCCL)
 endif()

 if(NANOVDB_USE_OPENVDB)
@@ -171,9 +184,12 @@ set(NANOVDB_INCLUDE_FILES
 # NanoVDB cuda header files
 set(NANOVDB_INCLUDE_CUDA_FILES
  cuda/DeviceBuffer.h
+  cuda/DeviceMesh.h
+  cuda/DeviceResource.h
  cuda/DeviceStreamMap.h
  cuda/GridHandle.cuh
  cuda/NodeManager.cuh
+  cuda/TempPool.h
  cuda/UnifiedBuffer.h
 )

@@ -202,6 +218,7 @@ set(NANOVDB_INCLUDE_TOOLS_FILES
  tools/GridStats.h
  tools/GridValidator.h
  tools/NanoToOpenVDB.h
+  tools/VoxelBlockManager.h
 )

 # NanoVDB tools/cuda header files
@@ -213,6 +230,14 @@ set(NANOVDB_INCLUDE_TOOLS_CUDA_FILES
  tools/cuda/IndexToGrid.cuh
  tools/cuda/PointsToGrid.cuh
  tools/cuda/SignedFloodFill.cuh
+  tools/cuda/MergeGrids.cuh
+  tools/cuda/TopologyBuilder.cuh
+  tools/cuda/RefineGrid.cuh
+  tools/cuda/PruneGrid.cuh
+  tools/cuda/DistributedPointsToGrid.cuh
+  tools/cuda/DilateGrid.cuh
+  tools/cuda/CoarsenGrid.cuh
+  tools/cuda/VoxelBlockManager.cuh
 )

 # NanoVDB util header files
@@ -241,6 +266,7 @@ set(NANOVDB_INCLUDE_UTIL_FILES
  util/Stencils.h
  util/Timer.h
  util/Util.h
+  util/MorphologyHelpers.h
 )

 # NanoVDB util/cuda header files
@@ -259,6 +285,10 @@ set(NANOVDB_INCLUDE_UTIL_CUDA_FILES
  util/cuda/CudaGridValidator.cuh
  util/cuda/CudaPointsToGrid.cuh
  util/cuda/GpuTimer.h
+  util/cuda/Morphology.cuh
+  util/cuda/MorphologyHelpers.cuh
+  util/cuda/DeviceGridTraits.cuh
+  util/cuda/Injection.cuh
 )

 add_library(nanovdb INTERFACE)
@@ -275,6 +305,10 @@ if(NANOVDB_USE_INTRINSICS)
  target_compile_definitions(nanovdb INTERFACE -DNANOVDB_USE_INTRINSICS)
 endif()

+if(NANOVDB_USE_OLD_ACCESSOR)
+  target_compile_definitions(nanovdb INTERFACE -DNANOVDB_USE_OLD_ACCESSOR)
+endif()
+
 if(NANOVDB_USE_OPENVDB)
  if(NOT OPENVDB_BUILD_CORE)
    target_link_libraries(nanovdb INTERFACE OpenVDB::openvdb)
@@ -311,8 +345,12 @@ if(NANOVDB_USE_CUDA)
  endif()

  # Add CUDA includes to any C++ units which use NanoVDB
-  target_include_directories(nanovdb INTERFACE
-    ${CMAKE_CUDA_TOOLKIT_INCLUDE_DIRECTORIES})
+  target_link_libraries(nanovdb INTERFACE CCCL::CCCL CUDA::cudart)
+
+  if(NCCL_FOUND)
+    target_compile_definitions(nanovdb INTERFACE -DNANOVDB_USE_NCCL)
+    target_link_libraries(nanovdb INTERFACE NCCL::NCCL)
+  endif()
 endif()

 if(NANOVDB_USE_MAGICAVOXEL)
@@ -8,8 +8,12 @@
 #ifndef __CNANOVDB__
 #define __CNANOVDB__

-#define CNANOVDB_DATA_ALIGNMENT 32
-#define CNANOVDB_ALIGNMENT_PADDING(x, n) (-(x) & ((n)-1))
+
+// We compute alignment, but if it evaluates to 0
+// it will be invalid C99, allowing pedantic compilers to
+// crash and claim they remain standards conforming.
+//#define CNANOVDB_DATA_ALIGNMENT 32
+//#define CNANOVDB_ALIGNMENT_PADDING(x, n) (-(x) & ((n)-1))

 #define USE_SINGLE_ROOT_KEY

@@ -50,6 +54,21 @@ enum cnanovdb_GridType
    cnanovdb_GridType_End     = 10
 };

+// #define versions of the grid type to allow pre-processor
+// tests.
+#define CNANOVDB_GRIDTYPE_UNKNOWN 0
+#define CNANOVDB_GRIDTYPE_FLOAT   1
+#define CNANOVDB_GRIDTYPE_DOUBLE  2
+#define CNANOVDB_GRIDTYPE_INT16   3
+#define CNANOVDB_GRIDTYPE_INT32   4
+#define CNANOVDB_GRIDTYPE_INT64   5
+#define CNANOVDB_GRIDTYPE_VEC3F   6
+#define CNANOVDB_GRIDTYPE_VEC3D   7
+#define CNANOVDB_GRIDTYPE_MASK    8
+#define CNANOVDB_GRIDTYPE_FP16    9
+#define CNANOVDB_GRIDTYPE_END     10
+
+
 #define ROOT_LEVEL 3

 #define DEFINEMASK_int(LOG2DIM, SIZE) \
@@ -85,7 +104,7 @@ typedef struct
    double mInvMatD[9]; // r,c = 3*r + c
    double mVecD[3];
    double mTaperD;
-} cnanovdb_map;
+} cnanovdb_map;         // Size 264

 typedef struct
 {
@@ -201,12 +220,13 @@ typedef struct
 {
    int64_t     mByteOffset;   // byte offset to the blind data, relative to the GridData.
    uint64_t    mElementCount; // number of elements, e.g. point count
-    uint32_t    mFlags;        // flags
+    uint32_t    mValueSize;    // byte size of each value, etc.
    uint32_t    mSemantic;     // semantic meaning of the data.
    uint32_t    mDataClass;    // 4 bytes
    uint32_t    mDataType;     // 4 bytes
    char        mName[256];
-    uint8_t     _reserved[CNANOVDB_ALIGNMENT_PADDING(sizeof(int64_t)+sizeof(uint64_t)+2*sizeof(uint32_t)+2*sizeof(uint32_t)+256*sizeof(char), CNANOVDB_DATA_ALIGNMENT)];
+    // 2*8+4*4+256 : aligned, so don't include.
+    // uint8_t     _reserved[CNANOVDB_ALIGNMENT_PADDING(sizeof(int64_t)+sizeof(uint64_t)+2*sizeof(uint32_t)+2*sizeof(uint32_t)+256*sizeof(char), CNANOVDB_DATA_ALIGNMENT)];
 } cnanovdb_gridblindmetadata;

 typedef struct
@@ -224,9 +244,12 @@ typedef struct
    double           mVoxelSize[3]; // 24B. size of a voxel in world units
    uint32_t         mGridClass; // 4B.
    uint32_t         mGridType; // 4B.
-    uint64_t         mBlindMetadataOffset; // 8B. offset of GridBlindMetaData structures.
-    int32_t          mBlindMetadataCount; // 4B. count of GridBlindMetaData structures.
-    uint32_t         _reserved[CNANOVDB_ALIGNMENT_PADDING(8 + 8 + 4 + 4 + 4 + 4 + 8 + 256 + 24 + 24 + sizeof(cnanovdb_map) + 24 + 4 + 4 + 8 + 4, CNANOVDB_DATA_ALIGNMENT) / 4];
+    int64_t          mBlindMetadataOffset; // 8B. offset of GridBlindMetaData structures.
+    uint32_t         mBlindMetadataCount; // 4B. count of GridBlindMetaData structures.
+
+    // Total 652: misaligned by 20.
+    // uint32_t         _reserved[CNANOVDB_ALIGNMENT_PADDING(8 + 8 + 4 + 4 + 4 + 4 + 8 + 256 + 24 + 24 + sizeof(cnanovdb_map) + 24 + 4 + 4 + 8 + 4, CNANOVDB_DATA_ALIGNMENT) / 4];
+    uint32_t         _reserved[5];
 } cnanovdb_griddata;

 static void
@@ -265,7 +288,8 @@ typedef struct
    uint32_t mNodeCount[ROOT_LEVEL];
    uint32_t mTileCount[ROOT_LEVEL];
    uint64_t mVoxelCount;
-    uint8_t  _reserved[CNANOVDB_ALIGNMENT_PADDING(4*sizeof(uint64_t)+(3+3)*sizeof(uint32_t)+sizeof(uint64_t), CNANOVDB_DATA_ALIGNMENT)];
+    // Total 64, so aligned, so do not include:
+    // uint8_t  _reserved[CNANOVDB_ALIGNMENT_PADDING(4*sizeof(uint64_t)+(3+3)*sizeof(uint32_t)+sizeof(uint64_t), CNANOVDB_DATA_ALIGNMENT)];
 } cnanovdb_treedata;

 static const CNANOVDB_GLOBAL cnanovdb_treedata *
@@ -298,6 +322,17 @@ cnanovdb_readaccessor_insert(cnanovdb_readaccessor *RESTRICT acc, int childlevel
    acc->mKey.mVec[2] = ijk->mVec[2];
 }

+
+// The padding of nodes varies on the signature, as the value type may
+// result in aligned or misaligned...
+#define LEAFDATA_PAD_F
+#define LEAFDATA_PAD_F3 uint32_t _reserved[4];
+#define LEAFDATA_PAD_I
+
+#define NODEDATA_PAD_F uint8_t _reserved[16];
+#define NODEDATA_PAD_F3
+#define NODEDATA_PAD_I uint8_t _reserved[16];
+
 #define CREATE_LEAF_NODE_int(LEVEL, LOG2DIM, CHILDTOTAL, TOTAL, MASK, VALUETYPE, STATSTYPE, SUFFIX) \
 typedef struct \
 { \
@@ -309,7 +344,8 @@ typedef struct \
    VALUETYPE                   mMaximum; \
    STATSTYPE                   mAverage; \
    STATSTYPE                   mStdDevi; \
-    uint32_t                    _reserved[ CNANOVDB_ALIGNMENT_PADDING(sizeof(cnanovdb_mask##LOG2DIM)+2*sizeof(VALUETYPE)+2*sizeof(STATSTYPE)+sizeof(cnanovdb_coord)+sizeof(uint8_t[3])+sizeof(uint8_t), CNANOVDB_DATA_ALIGNMENT)/4]; \
+    LEAFDATA_PAD_##SUFFIX \
+    /* uint32_t                    _reserved[ CNANOVDB_ALIGNMENT_PADDING(sizeof(cnanovdb_mask##LOG2DIM)+2*sizeof(VALUETYPE)+2*sizeof(STATSTYPE)+sizeof(cnanovdb_coord)+sizeof(uint8_t[3])+sizeof(uint8_t), CNANOVDB_DATA_ALIGNMENT)/4]; */ \
    VALUETYPE                   mVoxels[1u << (3*LOG2DIM)]; \
 } cnanovdb_node##LEVEL##SUFFIX; \
 \
@@ -337,6 +373,23 @@ cnanovdb_node##LEVEL##SUFFIX##_getValueAndCache(const CNANOVDB_GLOBAL cnanovdb_n
 } \
 \
 static bool \
+cnanovdb_node##LEVEL##SUFFIX##_setValue(CNANOVDB_GLOBAL cnanovdb_node##LEVEL##SUFFIX *RESTRICT node, const cnanovdb_coord *RESTRICT ijk, VALUETYPE value) \
+{ \
+    uint32_t n = cnanovdb_node##LEVEL##SUFFIX##_CoordToOffset(ijk); \
+    node->mVoxels[n] = value;  \
+    return true; \
+} \
+\
+static bool \
+cnanovdb_node##LEVEL##SUFFIX##_setValueAndCache(CNANOVDB_GLOBAL cnanovdb_node##LEVEL##SUFFIX *RESTRICT node, const cnanovdb_coord *RESTRICT ijk, VALUETYPE value,cnanovdb_readaccessor *RESTRICT /* DO NOT REMOVE: Required for C99 compliance */ acc) \
+{ \
+    (void)(acc); \
+    uint32_t n = cnanovdb_node##LEVEL##SUFFIX##_CoordToOffset(ijk); \
+    node->mVoxels[n] = value;  \
+    return true; \
+} \
+\
+static bool \
 cnanovdb_node##LEVEL##SUFFIX##_isActive(const CNANOVDB_GLOBAL cnanovdb_node##LEVEL##SUFFIX *RESTRICT node, const cnanovdb_coord *RESTRICT ijk) \
 { \
    uint32_t n = cnanovdb_node##LEVEL##SUFFIX##_CoordToOffset(ijk); \
@@ -376,7 +429,7 @@ typedef struct \
    cnanovdb_mask##LOG2DIM       mValueMask, mChildMask; \
    VALUETYPE                    mMinimum, mMaximum; \
    STATSTYPE                    mAverage, mStdDevi; \
-    uint8_t                      _reserved[CNANOVDB_ALIGNMENT_PADDING(sizeof(cnanovdb_mask##LOG2DIM)+sizeof(VALUETYPE)*2+sizeof(STATSTYPE)*2+sizeof(cnanovdb_coord)*2+sizeof(int32_t)+sizeof(uint32_t), CNANOVDB_DATA_ALIGNMENT)]; \
+    NODEDATA_PAD_##SUFFIX \
    cnanovdb_tileentry##SUFFIX   mTable[1u << (3*LOG2DIM)]; \
 } cnanovdb_node##LEVEL##SUFFIX; \
 \
@@ -395,6 +448,13 @@ cnanovdb_node##LEVEL##SUFFIX##_getChild(const CNANOVDB_GLOBAL cnanovdb_node##LEV
    return childnode; \
 } \
 \
+static CNANOVDB_GLOBAL cnanovdb_node##CHILDLEVEL##SUFFIX * \
+cnanovdb_node##LEVEL##SUFFIX##_getChildNC(CNANOVDB_GLOBAL cnanovdb_node##LEVEL##SUFFIX *RESTRICT node, uint32_t n) \
+{ \
+    CNANOVDB_GLOBAL cnanovdb_node##CHILDLEVEL##SUFFIX *childnode = (CNANOVDB_GLOBAL cnanovdb_node##CHILDLEVEL##SUFFIX *)( ((CNANOVDB_GLOBAL uint8_t *)node) + node->mTable[n].child); \
+    return childnode; \
+} \
+\
 static VALUETYPE \
 cnanovdb_node##LEVEL##SUFFIX##_getValue(const CNANOVDB_GLOBAL cnanovdb_node##LEVEL##SUFFIX *RESTRICT node, const cnanovdb_coord *RESTRICT ijk) \
 { \
@@ -421,6 +481,31 @@ cnanovdb_node##LEVEL##SUFFIX##_getValueAndCache(const CNANOVDB_GLOBAL cnanovdb_n
 } \
 \
 static bool \
+cnanovdb_node##LEVEL##SUFFIX##_setValue(CNANOVDB_GLOBAL cnanovdb_node##LEVEL##SUFFIX *RESTRICT node, const cnanovdb_coord *RESTRICT ijk, VALUETYPE value) \
+{ \
+    uint32_t n = cnanovdb_node##LEVEL##SUFFIX##_CoordToOffset(ijk); \
+    if (cnanovdb_mask##LOG2DIM##_isOn(&node->mChildMask, n)) \
+    { \
+        CNANOVDB_GLOBAL cnanovdb_node##CHILDLEVEL##SUFFIX *child = cnanovdb_node##LEVEL##SUFFIX##_getChildNC(node, n); \
+        return cnanovdb_node##CHILDLEVEL##SUFFIX##_setValue(child, ijk, value); \
+    } \
+    return false; \
+} \
+\
+static bool \
+cnanovdb_node##LEVEL##SUFFIX##_setValueAndCache(CNANOVDB_GLOBAL cnanovdb_node##LEVEL##SUFFIX *RESTRICT node, const cnanovdb_coord *RESTRICT ijk, VALUETYPE value, cnanovdb_readaccessor *RESTRICT acc) \
+{ \
+    uint32_t n = cnanovdb_node##LEVEL##SUFFIX##_CoordToOffset(ijk); \
+    if (cnanovdb_mask##LOG2DIM##_isOn(&node->mChildMask, n)) \
+    { \
+        CNANOVDB_GLOBAL cnanovdb_node##CHILDLEVEL##SUFFIX *child = cnanovdb_node##LEVEL##SUFFIX##_getChildNC(node, n); \
+        cnanovdb_readaccessor_insert(acc, CHILDLEVEL, child, ijk); \
+        return cnanovdb_node##CHILDLEVEL##SUFFIX##_setValueAndCache(child, ijk, value, acc); \
+    } \
+    return false; \
+} \
+\
+static bool \
 cnanovdb_node##LEVEL##SUFFIX##_isActive(const CNANOVDB_GLOBAL cnanovdb_node##LEVEL##SUFFIX *RESTRICT node, const cnanovdb_coord *RESTRICT ijk) \
 { \
    uint32_t n = cnanovdb_node##LEVEL##SUFFIX##_CoordToOffset(ijk); \
@@ -459,10 +544,33 @@ CREATE_INTERNAL_NODE_int(CHILDLEVEL, LEVEL, LOG2DIM, (TOTAL-LOG2DIM), TOTAL, ((1


 #ifdef USE_SINGLE_ROOT_KEY
+
+// Whether this is aligned varies with the signature...
+#define ROOTDATATILE_PAD_F uint8_t _reserved[8];
+#define ROOTDATATILE_PAD_I uint8_t _reserved[8];
+#define ROOTDATATILE_PAD_F3
+
 #define DEFINE_KEY(KEY) \
        uint64_t        KEY;
 #define KEYSIZE sizeof(uint64_t)

+#ifdef CNANOVDB_FULL_ROOT_SEARCH
+#define KEYSEARCH(SUFFIX) \
+    uint64_t                key; \
+    const CNANOVDB_GLOBAL cnanovdb_rootdata_tile##SUFFIX *result = 0; \
+    key = cnanovdb_coord_to_key(ijk); \
+\
+    (void)high; \
+    /* Re-read tablesize for compiler bug on intel windows driver */ \
+    for (int idx = low; idx < rootdata->mTableSize; idx++) \
+    { \
+        const CNANOVDB_GLOBAL cnanovdb_rootdata_tile##SUFFIX   *tile = tiles + idx; \
+        if (tile->key == key) \
+            result = tile; \
+    } \
+    return result; \
+/**/
+#else
 #define KEYSEARCH(SUFFIX) \
    uint64_t                key; \
    key = cnanovdb_coord_to_key(ijk); \
@@ -474,7 +582,14 @@ CREATE_INTERNAL_NODE_int(CHILDLEVEL, LEVEL, LOG2DIM, (TOTAL-LOG2DIM), TOTAL, ((1
            return tile; \
    } \
 /**/
+#endif
 #else
+
+// Whether this is aligned varies with the signature...
+#define ROOTDATATILE_PAD_F
+#define ROOTDATATILE_PAD_I
+#define ROOTDATATILE_PAD_F3 uint8_t _reserved[24];
+
 #define DEFINE_KEY(KEY) \
        cnanovdb_coord   KEY;
 #define KEYSIZE sizeof(cnanovdb_coord)
@@ -501,6 +616,9 @@ CREATE_INTERNAL_NODE_int(CHILDLEVEL, LEVEL, LOG2DIM, (TOTAL-LOG2DIM), TOTAL, ((1
 /**/
 #endif

+#define ROOTDATA_PAD_F uint32_t reserved[4];
+#define ROOTDATA_PAD_F3 uint32_t reserved[6];
+#define ROOTDATA_PAD_I uint32_t reserved[4];

 #define CREATE_ROOTDATA(VALUETYPE, STATSTYPE, SUFFIX) \
 typedef struct \
@@ -509,7 +627,11 @@ typedef struct \
    int64_t             child; \
    uint32_t            state; \
    VALUETYPE           value; \
-    uint8_t             _reserved[CNANOVDB_ALIGNMENT_PADDING(sizeof(KEYSIZE)+sizeof(VALUETYPE)+sizeof(int64_t)+sizeof(uint32_t), CNANOVDB_DATA_ALIGNMENT)]; \
+    /* Key is either 8 or 3*4, coord will round to 4*4 due to alignment */ \
+    /* of child, so 8+8 or 4*4+8.  state+value is either 4+4 or 4+3*4 */ \
+    /* 8+8+4+4:pad, 8+8+4*4:align, 4*4+8+2*4:align, 4*4+8+4*4:pad */ \
+    ROOTDATATILE_PAD_##SUFFIX \
+    /* uint8_t             _reserved[CNANOVDB_ALIGNMENT_PADDING(sizeof(KEYSIZE)+sizeof(VALUETYPE)+sizeof(int64_t)+sizeof(uint32_t), CNANOVDB_DATA_ALIGNMENT)]; */ \
 } cnanovdb_rootdata_tile##SUFFIX; \
 \
 typedef struct \
@@ -519,7 +641,10 @@ typedef struct \
    VALUETYPE      mBackground; \
    VALUETYPE      mMinimum, mMaximum; \
    STATSTYPE      mAverage, mStdDevi; \
-    uint32_t       _reserved[CNANOVDB_ALIGNMENT_PADDING(sizeof(cnanovdb_coord)*2+sizeof(uint32_t)+sizeof(VALUETYPE)*3+sizeof(STATSTYPE)*2, CNANOVDB_DATA_ALIGNMENT)/4]; \
+    /* 2*3*4 + 4 + 3*4 + 2*4: 12*4, unaligned */ \
+    /* 2*3*4 + 4 + 3*3*4 + 2*4: 18*4, unaligned */ \
+    ROOTDATA_PAD_##SUFFIX \
+    /* uint32_t       _reserved[CNANOVDB_ALIGNMENT_PADDING(sizeof(cnanovdb_coord)*2+sizeof(uint32_t)+sizeof(VALUETYPE)*3+sizeof(STATSTYPE)*2, CNANOVDB_DATA_ALIGNMENT)/4]; */ \
 } cnanovdb_rootdata##SUFFIX; \
 \
 static const CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX * \
@@ -537,6 +662,13 @@ cnanovdb_rootdata##SUFFIX##_getTile(const CNANOVDB_GLOBAL cnanovdb_rootdata##SUF
 \
 static const CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX * \
 cnanovdb_rootdata##SUFFIX##_getChild(const CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *RESTRICT rootdata, const CNANOVDB_GLOBAL cnanovdb_rootdata_tile##SUFFIX *RESTRICT tile) \
+{ \
+    CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *basenode = (CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *) (((CNANOVDB_GLOBAL uint8_t *) rootdata) + tile->child); \
+    return basenode; \
+} \
+ \
+static CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX * \
+cnanovdb_rootdata##SUFFIX##_getChildNC(CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *RESTRICT rootdata, CNANOVDB_GLOBAL cnanovdb_rootdata_tile##SUFFIX *RESTRICT tile) \
 { \
    CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *basenode = (CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *) (((CNANOVDB_GLOBAL uint8_t *) rootdata) + tile->child); \
    return basenode; \
@@ -552,6 +684,13 @@ cnanovdb_rootdata##SUFFIX##_findTile(const CNANOVDB_GLOBAL cnanovdb_rootdata##SU
    return 0; \
 } \
 \
+static CNANOVDB_GLOBAL cnanovdb_rootdata_tile##SUFFIX * \
+cnanovdb_rootdata##SUFFIX##_findTileNC(CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *RESTRICT rootdata, const cnanovdb_coord *RESTRICT ijk) \
+{ \
+    return (CNANOVDB_GLOBAL cnanovdb_rootdata_tile##SUFFIX *) \
+            cnanovdb_rootdata##SUFFIX##_findTile(rootdata, ijk); \
+} \
+ \
 static VALUETYPE \
 cnanovdb_rootdata##SUFFIX##_getValue(const CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *RESTRICT rootdata, const cnanovdb_coord *RESTRICT ijk) \
 { \
@@ -574,6 +713,30 @@ cnanovdb_rootdata##SUFFIX##_getValueAndCache(const CNANOVDB_GLOBAL cnanovdb_root
    const CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *child = cnanovdb_rootdata##SUFFIX##_getChild(rootdata, tile); \
    cnanovdb_readaccessor_insert(acc, 2, child, ijk); \
    return cnanovdb_node2##SUFFIX##_getValueAndCache( child, ijk, acc ); \
+} \
+ \
+static bool \
+cnanovdb_rootdata##SUFFIX##_setValue(CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *RESTRICT rootdata, const cnanovdb_coord *RESTRICT ijk, VALUETYPE value) \
+{ \
+    CNANOVDB_GLOBAL cnanovdb_rootdata_tile##SUFFIX *tile = cnanovdb_rootdata##SUFFIX##_findTileNC(rootdata, ijk); \
+    if (!tile) \
+        return false; \
+    if (tile->child == 0) \
+        return false; \
+    return cnanovdb_node2##SUFFIX##_setValue( cnanovdb_rootdata##SUFFIX##_getChildNC(rootdata, tile), ijk, value ); \
+} \
+ \
+static bool \
+cnanovdb_rootdata##SUFFIX##_setValueAndCache(CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *RESTRICT rootdata, const cnanovdb_coord *RESTRICT ijk, VALUETYPE value, cnanovdb_readaccessor *RESTRICT acc) \
+{ \
+    CNANOVDB_GLOBAL cnanovdb_rootdata_tile##SUFFIX *tile = cnanovdb_rootdata##SUFFIX##_findTileNC(rootdata, ijk); \
+    if (!tile) \
+        return false; \
+    if (tile->child == 0) \
+        return false; \
+    CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *child = cnanovdb_rootdata##SUFFIX##_getChildNC(rootdata, tile); \
+    cnanovdb_readaccessor_insert(acc, 2, child, ijk); \
+    return cnanovdb_node2##SUFFIX##_setValueAndCache( child, ijk, value, acc );\
 } \
 \
 static bool \
@@ -602,16 +765,19 @@ cnanovdb_rootdata##SUFFIX##_isActiveAndCache(const CNANOVDB_GLOBAL cnanovdb_root
 /**/


-inline void
+static inline void
 cnanovdb_readaccessor_init(cnanovdb_readaccessor *RESTRICT acc,
                    const CNANOVDB_GLOBAL void /*cnanovdb_rootdata* */ *RESTRICT rootdata)
 {
+    acc->mKey.mVec[0] = 0;
+    acc->mKey.mVec[1] = 0;
+    acc->mKey.mVec[2] = 0;
    acc->mNode[0] = acc->mNode[1] = acc->mNode[2] = 0;
    acc->mNode[3] = rootdata;
 }

 #define DEFINE_ISCACHED(LEVEL, MASK) \
-inline bool \
+static inline bool \
 cnanovdb_readaccessor_isCached##LEVEL(cnanovdb_readaccessor *RESTRICT acc, int32_t dirty) \
 { \
    if (!acc->mNode[LEVEL]) \
@@ -629,7 +795,7 @@ DEFINE_ISCACHED(0, ((1u <<  3) - 1u) )
 DEFINE_ISCACHED(1, ((1u <<  7) - 1u) )
 DEFINE_ISCACHED(2, ((1u << 12) - 1u) )

-inline int32_t
+static inline int32_t
 cnanovdb_readaccessor_computeDirty(const cnanovdb_readaccessor *RESTRICT acc, const cnanovdb_coord *RESTRICT ijk)
 {
    return (ijk->mVec[0] ^ acc->mKey.mVec[0]) |
@@ -638,35 +804,102 @@ cnanovdb_readaccessor_computeDirty(const cnanovdb_readaccessor *RESTRICT acc, co
 }

 #define CREATE_ACCESSOR(VALUETYPE, SUFFIX) \
-inline VALUETYPE \
+static inline VALUETYPE \
 cnanovdb_readaccessor_getValue##SUFFIX(cnanovdb_readaccessor *RESTRICT acc, const cnanovdb_coord *RESTRICT ijk) \
 { \
    int32_t dirty = cnanovdb_readaccessor_computeDirty(acc, ijk); \
 \
    if (cnanovdb_readaccessor_isCached0(acc, dirty)) \
-        return cnanovdb_node0##SUFFIX##_getValue( ((CNANOVDB_GLOBAL cnanovdb_node0##SUFFIX *) acc->mNode[0]), ijk); \
+        return cnanovdb_node0##SUFFIX##_getValue( ((const CNANOVDB_GLOBAL cnanovdb_node0##SUFFIX *) acc->mNode[0]), ijk); \
    if (cnanovdb_readaccessor_isCached1(acc, dirty)) \
-        return cnanovdb_node1##SUFFIX##_getValueAndCache( ((CNANOVDB_GLOBAL cnanovdb_node1##SUFFIX *) acc->mNode[1]), ijk, acc); \
+        return cnanovdb_node1##SUFFIX##_getValueAndCache( ((const CNANOVDB_GLOBAL cnanovdb_node1##SUFFIX *) acc->mNode[1]), ijk, acc); \
    if (cnanovdb_readaccessor_isCached2(acc, dirty)) \
-        return cnanovdb_node2##SUFFIX##_getValueAndCache( ((CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *) acc->mNode[2]), ijk, acc); \
+        return cnanovdb_node2##SUFFIX##_getValueAndCache( ((const CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *) acc->mNode[2]), ijk, acc); \
 \
-    return cnanovdb_rootdata##SUFFIX##_getValueAndCache( ((CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *)acc->mNode[3]), ijk, acc); \
+    return cnanovdb_rootdata##SUFFIX##_getValueAndCache( ((const CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *)acc->mNode[3]), ijk, acc); \
 } \
 \
-inline bool \
+static inline VALUETYPE \
+cnanovdb_readaccessor_getIJKValue##SUFFIX(cnanovdb_readaccessor *RESTRICT acc, int i, int j, int k) \
+{ \
+    cnanovdb_coord coord; \
+    coord.mVec[0] = i; coord.mVec[1] = j; coord.mVec[2] = k; \
+    return cnanovdb_readaccessor_getValue##SUFFIX(acc, &coord); \
+} \
+\
+static inline CNANOVDB_GLOBAL VALUETYPE * \
+cnanovdb_readaccessor_getLeaf##SUFFIX(cnanovdb_readaccessor *RESTRICT acc, const cnanovdb_coord *RESTRICT ijk) \
+{ \
+    int32_t dirty = cnanovdb_readaccessor_computeDirty(acc, ijk); \
+ \
+    if (cnanovdb_readaccessor_isCached0(acc, dirty)) \
+        return ((CNANOVDB_GLOBAL cnanovdb_node0##SUFFIX *) acc->mNode[0])->mVoxels; \
+    else if (cnanovdb_readaccessor_isCached1(acc, dirty)) \
+        cnanovdb_node1##SUFFIX##_getValueAndCache( ((const CNANOVDB_GLOBAL cnanovdb_node1##SUFFIX *) acc->mNode[1]), ijk, acc); \
+    else if (cnanovdb_readaccessor_isCached2(acc, dirty)) \
+        cnanovdb_node2##SUFFIX##_getValueAndCache( ((const CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *) acc->mNode[2]), ijk, acc); \
+    else \
+        cnanovdb_rootdata##SUFFIX##_getValueAndCache( ((const CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *)acc->mNode[3]), ijk, acc); \
+    if (acc->mNode[0]) \
+        return ((CNANOVDB_GLOBAL cnanovdb_node0##SUFFIX *) acc->mNode[0])->mVoxels; \
+    return 0; \
+} \
+\
+static inline CNANOVDB_GLOBAL VALUETYPE * \
+cnanovdb_readaccessor_getIJKLeaf##SUFFIX(cnanovdb_readaccessor *RESTRICT acc, int i, int j, int k) \
+{ \
+    cnanovdb_coord coord; \
+    coord.mVec[0] = i; coord.mVec[1] = j; coord.mVec[2] = k; \
+    return cnanovdb_readaccessor_getLeaf##SUFFIX(acc, &coord); \
+} \
+\
+static inline bool \
 cnanovdb_readaccessor_isActive##SUFFIX(cnanovdb_readaccessor *RESTRICT acc, const cnanovdb_coord *RESTRICT ijk) \
 { \
    int32_t dirty = cnanovdb_readaccessor_computeDirty(acc, ijk); \
 \
    if (cnanovdb_readaccessor_isCached0(acc, dirty)) \
-        return cnanovdb_node0##SUFFIX##_isActive( ((CNANOVDB_GLOBAL cnanovdb_node0##SUFFIX *) acc->mNode[0]), ijk); \
+        return cnanovdb_node0##SUFFIX##_isActive( ((const CNANOVDB_GLOBAL cnanovdb_node0##SUFFIX *) acc->mNode[0]), ijk); \
    if (cnanovdb_readaccessor_isCached1(acc, dirty)) \
-        return cnanovdb_node1##SUFFIX##_isActiveAndCache( ((CNANOVDB_GLOBAL cnanovdb_node1##SUFFIX *) acc->mNode[1]), ijk, acc); \
+        return cnanovdb_node1##SUFFIX##_isActiveAndCache( ((const CNANOVDB_GLOBAL cnanovdb_node1##SUFFIX *) acc->mNode[1]), ijk, acc); \
    if (cnanovdb_readaccessor_isCached2(acc, dirty)) \
-        return cnanovdb_node2##SUFFIX##_isActiveAndCache( ((CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *) acc->mNode[2]), ijk, acc); \
+        return cnanovdb_node2##SUFFIX##_isActiveAndCache( ((const CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *) acc->mNode[2]), ijk, acc); \
 \
-    return cnanovdb_rootdata##SUFFIX##_isActiveAndCache( ((CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *)acc->mNode[3]), ijk, acc); \
+    return cnanovdb_rootdata##SUFFIX##_isActiveAndCache( ((const CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *)acc->mNode[3]), ijk, acc); \
 } \
+\
+static inline bool \
+cnanovdb_readaccessor_isIJKActive##SUFFIX(cnanovdb_readaccessor *RESTRICT acc, int i, int j, int k) \
+{ \
+    cnanovdb_coord coord; \
+    coord.mVec[0] = i; coord.mVec[1] = j; coord.mVec[2] = k; \
+    return cnanovdb_readaccessor_isActive##SUFFIX(acc, &coord); \
+} \
+\
+static inline bool \
+cnanovdb_readaccessor_setValue##SUFFIX(cnanovdb_readaccessor *RESTRICT acc, const cnanovdb_coord *RESTRICT ijk, VALUETYPE value) \
+{ \
+    int32_t dirty = cnanovdb_readaccessor_computeDirty(acc, ijk); \
+ \
+    if (cnanovdb_readaccessor_isCached0(acc, dirty)) \
+        return cnanovdb_node0##SUFFIX##_setValue( ((CNANOVDB_GLOBAL cnanovdb_node0##SUFFIX *) acc->mNode[0]), ijk, value); \
+    if (cnanovdb_readaccessor_isCached1(acc, dirty)) \
+        return cnanovdb_node1##SUFFIX##_setValueAndCache( ((CNANOVDB_GLOBAL cnanovdb_node1##SUFFIX *) acc->mNode[1]), ijk, value, acc); \
+    if (cnanovdb_readaccessor_isCached2(acc, dirty)) \
+        return cnanovdb_node2##SUFFIX##_setValueAndCache( ((CNANOVDB_GLOBAL cnanovdb_node2##SUFFIX *) acc->mNode[2]), ijk, value, acc); \
+ \
+    return cnanovdb_rootdata##SUFFIX##_setValueAndCache( ((CNANOVDB_GLOBAL cnanovdb_rootdata##SUFFIX *)acc->mNode[3]), ijk, value, acc); \
+} \
+\
+static inline bool \
+cnanovdb_readaccessor_setIJKValue##SUFFIX(cnanovdb_readaccessor *RESTRICT acc, int i, int j, int k, VALUETYPE value) \
+{ \
+    cnanovdb_coord coord; \
+    coord.mVec[0] = i; coord.mVec[1] = j; coord.mVec[2] = k; \
+    return cnanovdb_readaccessor_setValue##SUFFIX(acc, &coord, value); \
+} \
+\
+\
 /**/


@@ -681,6 +914,7 @@ CREATE_ACCESSOR(VALUETYPE, SUFFIX) \

 CREATE_GRIDTYPE(float, float, F)
 CREATE_GRIDTYPE(cnanovdb_Vec3F, float, F3)
+CREATE_GRIDTYPE(int, int, I)

 static int
 cnanovdb_griddata_valid(const CNANOVDB_GLOBAL cnanovdb_griddata *RESTRICT grid)
@@ -712,4 +946,13 @@ cnanovdb_griddata_validF3(const CNANOVDB_GLOBAL cnanovdb_griddata *RESTRICT grid
    return 1;
 }

+static int
+cnanovdb_griddata_validI(const CNANOVDB_GLOBAL cnanovdb_griddata *RESTRICT grid)
+{
+    if (!cnanovdb_griddata_valid(grid))
+        return 0;
+    if (grid->mGridType != cnanovdb_GridType_Int32)
+        return 0;
+    return 1;
+}
 #endif
@@ -111,6 +111,9 @@ public:
    typename util::enable_if<BufferTraits<U>::hasDeviceDual, const void*>::type
    deviceData() const { return mBuffer.deviceData(); }
    template<typename U = BufferT>
+    typename util::enable_if<BufferTraits<U>::hasDeviceDual, const void*>::type
+    deviceData(int device) const { return mBuffer.deviceData(device); }
+    template<typename U = BufferT>
    typename util::enable_if<BufferTraits<U>::hasDeviceDual, void*>::type
    deviceData() { return mBuffer.deviceData(); }
    template<typename U = BufferT>
@@ -160,7 +163,6 @@ public:
    /// @brief Return a const pointer to the @a n'th grid encoded in this GridHandle on the device, e.g. GPU
    /// @tparam ValueT Value type of the grid point to be returned
    /// @param n Index of the grid pointer to be returned
-    /// @param verbose if non-zero error messages will be printed in case something failed
    /// @warning Note that the return pointer can be NULL if the GridHandle was not initialized, @a n is invalid,
    ///          or if the template parameter does not match the specified grid.
    template<typename ValueT, typename U = BufferT>
@@ -173,7 +175,7 @@ public:
    typename util::enable_if<BufferTraits<U>::hasDeviceDual, void>::type
    deviceUpload(void* stream, bool sync = true) { mBuffer.deviceUpload(stream, sync); }

-    /// @brief Upload the host buffer to a specefic device buffer. It device buffer doesn't exist it's created first
+    /// @brief Upload the host buffer to a specific device buffer. It device buffer doesn't exist it's created first
    /// @param device Device to upload host data to
    /// @param stream cuda stream
    /// @param sync if false the memory copy is asynchronous
@@ -212,7 +214,7 @@ public:
        return sum;
    }

-    /// @brief compute the size of unusedstorage in this buffer
+    /// @brief compute the size of unused storage in this buffer
    /// @return the number of unused bytes in this buffer.
    uint64_t freeSize() const {return mBuffer.size() - this->totalGridSize();}

@@ -286,12 +288,12 @@ public:
    /// @param is input stream containing a raw grid buffer
    /// @param gridName string name of the grid to be read
    /// @param pool optional pool from which to allocate the new grid buffer
-    /// @throw Will throw a std::logic_error if the stream does not contain a valid raw grid with the speficied name
+    /// @throw Will throw a std::logic_error if the stream does not contain a valid raw grid with the specified name
    void read(std::istream& is, const std::string &gridName, const BufferT& pool = BufferT());

    /// @brief Read a raw grid buffer from a file
-    /// @param filename string name of the input file containing a raw grid buffer
-    /// @param pool optional pool from which to allocate the new grid buffe
+    /// @param fileName string name of the input file containing a raw grid buffer
+    /// @param pool optional pool from which to allocate the new grid buffer
    void read(const std::string &fileName, const BufferT& pool = BufferT()) {
        std::ifstream is(fileName, std::ios::in | std::ios::binary);
        if (!is.is_open()) throw std::ios_base::failure("Unable to open file named \"" + fileName + "\" for input");
@@ -299,7 +301,7 @@ public:
    }

    /// @brief Read a specific grid from a file containing a raw grid buffer
-    /// @param filename string name of the input file containing a raw grid buffer
+    /// @param fileName string name of the input file containing a raw grid buffer
    /// @param n zero-based index of the grid to be read
    /// @param pool optional pool from which to allocate the new grid buffer
    /// @throw Will throw a std::ios_base::failure if the file does not exist and a
@@ -311,7 +313,7 @@ public:
    }

    /// @brief Read a specific grid from a file containing a raw grid buffer
-    /// @param filename string name of the input file containing a raw grid buffer
+    /// @param fileName string name of the input file containing a raw grid buffer
    /// @param gridName string name of the grid to be read
    /// @param pool optional pool from which to allocate the new grid buffer
    /// @throw Will throw a std::ios_base::failure if the file does not exist and a
@@ -397,6 +399,7 @@ GridHandle<BufferT>::deviceGrid(uint32_t n) const
 template<typename BufferT>
 void GridHandle<BufferT>::read(std::istream& is, const BufferT& pool)
 {
+    const std::streampos start = is.tellg();// remember where the raw buffer begins
    GridData data;
    is.read((char*)&data, sizeof(GridData));
    if (data.isValid()) {
@@ -407,7 +410,7 @@ void GridHandle<BufferT>::read(std::istream& is, const BufferT& pool)
            sum += data.mGridSize;
        }
        auto buffer = BufferT::create(size + sum, &pool);
-        is.seekg(-int64_t(sum + sizeof(GridData)), std::ios::cur);// rewind to start
+        is.seekg(start);// rewind to the start of the raw buffer
        is.read((char*)(buffer.data()), buffer.size());
        *this = GridHandle(std::move(buffer));
    } else {
@@ -467,7 +470,7 @@ void GridHandle<BufferT>::read(std::istream& is, const std::string &gridName, co
 /// @brief Split all grids in a single GridHandle into a vector of multiple GridHandles each with a single grid
 /// @tparam BufferT Type of the input and output grid buffers
 /// @param handle GridHandle with grids that will be slip into individual GridHandles
-/// @param pool optional pool used for allocation of output GridHandle
+/// @param other optional pool used for allocation of output GridHandle
 /// @return Vector of GridHandles each containing a single grid
 template<typename BufferT, template <class, class...> class VectorT = std::vector>
 inline VectorT<GridHandle<BufferT>>
@@ -487,7 +490,7 @@ splitGrids(const GridHandle<BufferT> &handle, const BufferT* other = nullptr)
        h = HandleT(std::move(buffer));
        ptr = util::PtrAdd(ptr, src->mGridSize);
    }
-    return std::move(handles);
+    return handles;
 }// splitGrids

 /// @brief Combines (or merges) multiple GridHandles into a single GridHandle containing all grids
@@ -192,6 +192,13 @@ public:
    void* data() { return mData; }
    //@}

+    /// @brief Returns an offset pointer of a specific type from the allocated host memory
+    /// @tparam T Type of the pointer returned
+    /// @param count Numbers of elements of @c parameter type T to skip
+    /// @warning might return NULL
+    template <typename T>
+    T* data(ptrdiff_t count = 0) const {return mData ? reinterpret_cast<T*>(mData) + count : nullptr;}
+
    //@{
    /// @brief Returns the size in bytes associated with this buffer.
    uint64_t bufferSize() const { return mSize; }
@@ -316,7 +323,6 @@ struct HostBuffer::Pool
               << mSize << " bytes of which\n\t" << (util::PtrDiff(alignedFree, mData) - mPadding)
               << " bytes are used by " << mRegister.size() << " other buffer(s). "
               << "Pool is " << (mManaged ? "internally" : "externally") << " managed.\n";
-            //std::cerr << ss.str();
            throw std::runtime_error(ss.str());
        }
        buffer->mSize = size;
@@ -389,7 +395,6 @@ struct HostBuffer::Pool
            }

            for (HostBuffer* buffer : mRegister) { // update registered buffers
-                //buffer->mData = paddedData + ptrdiff_t(buffer->mData - (mData + mPadding));
                buffer->mData = util::PtrAdd(paddedData, util::PtrDiff(buffer->mData, util::PtrAdd(mData, mPadding)));
            }
            mFree = util::PtrAdd(paddedData, memUsage); // update the free pointer
@@ -414,12 +419,8 @@ private:

    static void* alloc(uint64_t size)
    {
-//#if (__cplusplus >= 201703L)
-//    return std::aligned_alloc(NANOVDB_DATA_ALIGNMENT, size);//C++17 or newer
-//#else
    // make sure we alloc enough space to align the result
    return std::malloc(size + NANOVDB_DATA_ALIGNMENT);
-//#endif
    }

    static void* realloc(void* const origData,
@@ -42,8 +42,8 @@ NodeManagerHandle<BufferT> createNodeManager(const NanoGrid<BuildT> &grid,
                                             const BufferT& buffer = BufferT());

 struct NodeManagerData
-{// 48B = 6*8B
-    uint64_t        mMagic;// 8B
+{// 40B = 5*8B
+    __hostdev__ NodeManagerData(void *grid) : mPadding{0}, mGrid(grid), mPtr{0,0,0}{}
    union {int64_t  mPadding; uint8_t mLinear;};// 8B of which 1B is used for a binary flag
    void           *mGrid;//  8B pointer to either host or device grid
    union {int64_t *mPtr[3], mOff[3];};// 24B, use mOff if mLinear!=0
@@ -288,11 +288,7 @@ NodeManagerHandle<BufferT> createNodeManager(const NanoGrid<BuildT> &grid,
    auto *data = reinterpret_cast<NodeManagerData*>(handle.data());
    NANOVDB_ASSERT(data && isAligned(data));
    NANOVDB_ASSERT(toGridType<BuildT>() == grid.gridType());
-#ifdef NANOVDB_USE_NEW_MAGIC_NUMBERS
-    *data = NodeManagerData{NANOVDB_MAGIC_NODE, {0u}, (void*)&grid, {{0u,0u,0u}}};
-#else
-    *data = NodeManagerData{NANOVDB_MAGIC_NUMB, {0u}, (void*)&grid, {{0u,0u,0u}}};
-#endif
+    *data = NodeManagerData((void*)&grid);

    if (NodeManager<BuildT>::isLinear(grid)) {
        data->mLinear = uint8_t(1u);
@@ -198,47 +198,110 @@ PNANOVDB_BUF_FORCE_INLINE void pnanovdb_buf_write_uint64(pnanovdb_buf_t buf, uin
 typedef uint32_t pnanovdb_grid_type_t;
 #define PNANOVDB_GRID_TYPE_GET(grid_typeIn, nameIn) pnanovdb_grid_type_constants[grid_typeIn].nameIn
 #elif defined(PNANOVDB_BUF_HLSL)
-#if defined(PNANOVDB_ADDRESS_32)
+#if defined(PNANOVDB_BUF_HLSL_RW)
+#if defined(PNANOVDB_BUF_HLSL_64)
+#define pnanovdb_buf_t RWStructuredBuffer<uint2>
+#else
+#define pnanovdb_buf_t RWStructuredBuffer<uint>
+#endif
+#else
+#if defined(PNANOVDB_BUF_HLSL_64)
+#define pnanovdb_buf_t StructuredBuffer<uint2>
+#else
 #define pnanovdb_buf_t StructuredBuffer<uint>
+#endif
+#endif
+#if defined(PNANOVDB_ADDRESS_32)
 uint pnanovdb_buf_read_uint32(pnanovdb_buf_t buf, uint byte_offset)
 {
+#if defined(PNANOVDB_BUF_HLSL_64)
+    uint2 val = buf[(byte_offset >> 3u)];
+    return ((byte_offset & 4u) == 0u) ? val.x : val.y;
+#else
    return buf[(byte_offset >> 2u)];
+#endif
 }
 uint2 pnanovdb_buf_read_uint64(pnanovdb_buf_t buf, uint byte_offset)
 {
    uint2 ret;
+#if defined(PNANOVDB_BUF_HLSL_64)
+    ret = buf[(byte_offset >> 3u)];
+#else
    ret.x = pnanovdb_buf_read_uint32(buf, byte_offset + 0u);
    ret.y = pnanovdb_buf_read_uint32(buf, byte_offset + 4u);
+#endif
    return ret;
 }
 void pnanovdb_buf_write_uint32(pnanovdb_buf_t buf, uint byte_offset, uint value)
 {
    // NOP, by default no write in HLSL
+#if defined(PNANOVDB_BUF_HLSL_RW)
+#if defined(PNANOVDB_BUF_HLSL_64)
+    if ((byte_offset & 4u) == 0u) {buf[(byte_offset >> 3u)].x = value;}
+    else {buf[(byte_offset >> 3u)].y = value;}
+#else
+    buf[(byte_offset >> 2u)] = value;
+#endif
+#endif
 }
 void pnanovdb_buf_write_uint64(pnanovdb_buf_t buf, uint byte_offset, uint2 value)
 {
    // NOP, by default no write in HLSL
+#if defined(PNANOVDB_BUF_HLSL_RW)
+#if defined(PNANOVDB_BUF_HLSL_64)
+    buf[(byte_offset >> 3u)] = value;
+#else
+    pnanovdb_buf_write_uint32(buf, byte_offset + 0u, value.x);
+    pnanovdb_buf_write_uint32(buf, byte_offset + 4u, value.y);
+#endif
+#endif
 }
 #elif defined(PNANOVDB_ADDRESS_64)
-#define pnanovdb_buf_t StructuredBuffer<uint>
 uint pnanovdb_buf_read_uint32(pnanovdb_buf_t buf, uint64_t byte_offset)
 {
+#if defined(PNANOVDB_BUF_HLSL_64)
+    uint2 val = buf[uint(byte_offset >> 3u)];
+    return ((uint(byte_offset) & 4u) == 0u) ? val.x : val.y;
+#else
    return buf[uint(byte_offset >> 2u)];
+#endif
 }
 uint64_t pnanovdb_buf_read_uint64(pnanovdb_buf_t buf, uint64_t byte_offset)
 {
    uint64_t ret;
+#if defined(PNANOVDB_BUF_HLSL_64)
+    uint2 raw = buf[uint(byte_offset >> 3u)];
+    ret = uint64_t(raw.x) | (uint64_t(raw.y) << 32u);
+#else
    ret = pnanovdb_buf_read_uint32(buf, byte_offset + 0u);
-    ret = ret + (uint64_t(pnanovdb_buf_read_uint32(buf, byte_offset + 4u)) << 32u);
+    ret = ret | (uint64_t(pnanovdb_buf_read_uint32(buf, byte_offset + 4u)) << 32u);
+#endif
    return ret;
 }
 void pnanovdb_buf_write_uint32(pnanovdb_buf_t buf, uint64_t byte_offset, uint value)
 {
    // NOP, by default no write in HLSL
+#if defined(PNANOVDB_BUF_HLSL_RW)
+#if defined(PNANOVDB_BUF_HLSL_64)
+    if ((byte_offset & 4u) == 0u) {buf[uint(byte_offset >> 3u)].x = value;}
+    else {buf[uint(byte_offset >> 3u)].y = value;}
+#else
+    buf[uint(byte_offset >> 2u)] = value;
+#endif
+#endif
 }
 void pnanovdb_buf_write_uint64(pnanovdb_buf_t buf, uint64_t byte_offset, uint64_t value)
 {
    // NOP, by default no write in HLSL
+#if defined(PNANOVDB_BUF_HLSL_RW)
+#if defined(PNANOVDB_BUF_HLSL_64)
+    uint2 raw = uint2(uint(value), uint(value >> 32u));
+    buf[uint(byte_offset >> 3u)] = raw;
+#else
+    pnanovdb_buf_write_uint32(buf, byte_offset + 0u, uint(value));
+    pnanovdb_buf_write_uint32(buf, byte_offset + 4u, uint(value >> 32u));
+#endif
+#endif
 }
 #endif
 #define pnanovdb_grid_type_t uint
@@ -343,14 +406,26 @@ typedef struct pnanovdb_vec3_t
 {
    float x, y, z;
 }pnanovdb_vec3_t;
+typedef struct pnanovdb_uvec4_t
+{
+    pnanovdb_uint32_t x, y, z, w;
+}pnanovdb_uvec4_t;
+typedef struct pnanovdb_ivec4_t
+{
+    pnanovdb_int32_t x, y, z, w;
+}pnanovdb_ivec4_t;
+typedef struct pnanovdb_vec4_t
+{
+    float x, y, z, w;
+}pnanovdb_vec4_t;
 PNANOVDB_FORCE_INLINE pnanovdb_int32_t pnanovdb_uint32_as_int32(pnanovdb_uint32_t v) { return (pnanovdb_int32_t)v; }
 PNANOVDB_FORCE_INLINE pnanovdb_int64_t pnanovdb_uint64_as_int64(pnanovdb_uint64_t v) { return (pnanovdb_int64_t)v; }
 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_int64_as_uint64(pnanovdb_int64_t v) { return (pnanovdb_uint64_t)v; }
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_int32_as_uint32(pnanovdb_int32_t v) { return (pnanovdb_uint32_t)v; }
 PNANOVDB_FORCE_INLINE float pnanovdb_uint32_as_float(pnanovdb_uint32_t v) { float vf; pnanovdb_memcpy(&vf, &v, sizeof(vf)); return vf; }
-PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_float_as_uint32(float v) { return *((pnanovdb_uint32_t*)(&v)); }
+PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_float_as_uint32(float v) { pnanovdb_uint32_t vu; pnanovdb_memcpy(&vu, &v, sizeof(vu)); return vu; }
 PNANOVDB_FORCE_INLINE double pnanovdb_uint64_as_double(pnanovdb_uint64_t v) { double vf; pnanovdb_memcpy(&vf, &v, sizeof(vf)); return vf; }
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_double_as_uint64(double v) { return *((pnanovdb_uint64_t*)(&v)); }
+PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_double_as_uint64(double v) { pnanovdb_uint64_t vu; pnanovdb_memcpy(&vu, &v, sizeof(vu)); return vu; }
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_uint64_low(pnanovdb_uint64_t v) { return (pnanovdb_uint32_t)v; }
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_uint64_high(pnanovdb_uint64_t v) { return (pnanovdb_uint32_t)(v >> 32u); }
 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_uint32_as_uint64(pnanovdb_uint32_t x, pnanovdb_uint32_t y) { return ((pnanovdb_uint64_t)x) | (((pnanovdb_uint64_t)y) << 32u); }
@@ -373,6 +448,9 @@ typedef bool pnanovdb_bool_t;
 #define PNANOVDB_TRUE true
 typedef int3 pnanovdb_coord_t;
 typedef float3 pnanovdb_vec3_t;
+typedef uint4 pnanovdb_uvec4_t;
+typedef int4 pnanovdb_ivec4_t;
+typedef float4 pnanovdb_vec4_t;
 pnanovdb_int32_t pnanovdb_uint32_as_int32(pnanovdb_uint32_t v) { return int(v); }
 pnanovdb_uint32_t pnanovdb_int32_as_uint32(pnanovdb_int32_t v) { return uint(v); }
 float pnanovdb_uint32_as_float(pnanovdb_uint32_t v) { return asfloat(v); }
@@ -420,6 +498,9 @@ bool pnanovdb_int64_is_zero(pnanovdb_int64_t a) { return a == 0; }
 #define pnanovdb_int64_t ivec2
 #define pnanovdb_coord_t ivec3
 #define pnanovdb_vec3_t vec3
+#define pnanovdb_uvec4_t uvec4
+#define pnanovdb_ivec4_t ivec4
+#define pnanovdb_vec4_t vec4
 pnanovdb_int32_t pnanovdb_uint32_as_int32(pnanovdb_uint32_t v) { return int(v); }
 pnanovdb_int64_t pnanovdb_uint64_as_int64(pnanovdb_uint64_t v) { return ivec2(v); }
 pnanovdb_uint64_t pnanovdb_int64_as_uint64(pnanovdb_int64_t v) { return uvec2(v); }
@@ -525,6 +606,35 @@ PNANOVDB_FORCE_INLINE pnanovdb_coord_t pnanovdb_coord_add(pnanovdb_coord_t a, pn
    v.z = a.z + b.z;
    return v;
 }
+PNANOVDB_FORCE_INLINE pnanovdb_uvec4_t pnanovdb_uvec4_add(pnanovdb_uvec4_t a, pnanovdb_uvec4_t b)
+{
+    pnanovdb_uvec4_t ret;
+    ret.x = a.x + b.x;
+    ret.y = a.y + b.y;
+    ret.z = a.z + b.z;
+    ret.w = a.w + b.w;
+    return ret;
+}
+
+PNANOVDB_FORCE_INLINE pnanovdb_uvec4_t pnanovdb_uvec4_sub(pnanovdb_uvec4_t a, pnanovdb_uvec4_t b)
+{
+    pnanovdb_uvec4_t ret;
+    ret.x = a.x - b.x;
+    ret.y = a.y - b.y;
+    ret.z = a.z - b.z;
+    ret.w = a.w - b.w;
+    return ret;
+}
+
+PNANOVDB_FORCE_INLINE pnanovdb_vec4_t pnanovdb_vec4_add(pnanovdb_vec4_t a, pnanovdb_vec4_t b)
+{
+    pnanovdb_vec4_t ret;
+    ret.x = a.x + b.x;
+    ret.y = a.y + b.y;
+    ret.z = a.z + b.z;
+    ret.w = a.w + b.w;
+    return ret;
+}
 #elif defined(PNANOVDB_HLSL)
 pnanovdb_vec3_t pnanovdb_vec3_uniform(float a) { return float3(a, a, a); }
 pnanovdb_vec3_t pnanovdb_vec3_add(pnanovdb_vec3_t a, pnanovdb_vec3_t b) { return a + b; }
@@ -536,6 +646,9 @@ pnanovdb_vec3_t pnanovdb_vec3_max(pnanovdb_vec3_t a, pnanovdb_vec3_t b) { return
 pnanovdb_vec3_t pnanovdb_coord_to_vec3(pnanovdb_coord_t coord) { return float3(coord); }
 pnanovdb_coord_t pnanovdb_coord_uniform(pnanovdb_int32_t a) { return int3(a, a, a); }
 pnanovdb_coord_t pnanovdb_coord_add(pnanovdb_coord_t a, pnanovdb_coord_t b) { return a + b; }
+pnanovdb_uvec4_t pnanovdb_uvec4_add(pnanovdb_uvec4_t a, pnanovdb_uvec4_t b) { return a + b; }
+pnanovdb_uvec4_t pnanovdb_uvec4_sub(pnanovdb_uvec4_t a, pnanovdb_uvec4_t b) { return a - b; }
+pnanovdb_vec4_t pnanovdb_vec4_add(pnanovdb_vec4_t a, pnanovdb_vec4_t b) { return a + b; }
 #elif defined(PNANOVDB_GLSL)
 pnanovdb_vec3_t pnanovdb_vec3_uniform(float a) { return vec3(a, a, a); }
 pnanovdb_vec3_t pnanovdb_vec3_add(pnanovdb_vec3_t a, pnanovdb_vec3_t b) { return a + b; }
@@ -547,6 +660,9 @@ pnanovdb_vec3_t pnanovdb_vec3_max(pnanovdb_vec3_t a, pnanovdb_vec3_t b) { return
 pnanovdb_vec3_t pnanovdb_coord_to_vec3(const pnanovdb_coord_t coord) { return vec3(coord); }
 pnanovdb_coord_t pnanovdb_coord_uniform(pnanovdb_int32_t a) { return ivec3(a, a, a); }
 pnanovdb_coord_t pnanovdb_coord_add(pnanovdb_coord_t a, pnanovdb_coord_t b) { return a + b; }
+pnanovdb_uvec4_t pnanovdb_uvec4_add(pnanovdb_uvec4_t a, pnanovdb_uvec4_t b) { return a + b; }
+pnanovdb_uvec4_t pnanovdb_uvec4_sub(pnanovdb_uvec4_t a, pnanovdb_uvec4_t b) { return a - b; }
+pnanovdb_vec4_t pnanovdb_vec4_add(pnanovdb_vec4_t a, pnanovdb_vec4_t b) { return a + b; }
 #endif

 // ------------------------------------------------ Uint64 Utils -----------------------------------------------------------
@@ -589,6 +705,19 @@ PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_uint64_offset(pnanovdb_uint64_t
    return pnanovdb_uint32_as_uint64(low, high);
 }

+PNANOVDB_FORCE_INLINE pnanovdb_int64_t pnanovdb_uint64_diff(pnanovdb_uint64_t a, pnanovdb_uint64_t b)
+{
+    pnanovdb_uint32_t low = pnanovdb_uint64_low(a);
+    pnanovdb_uint32_t high = pnanovdb_uint64_high(a);
+    low -= pnanovdb_uint64_low(b);
+    if (low > pnanovdb_uint64_low(a))
+    {
+        high -= 1u;
+    }
+    high -= pnanovdb_uint64_high(b);
+    return pnanovdb_uint64_as_int64(pnanovdb_uint32_as_uint64(low, high));
+}
+
 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_uint64_dec(pnanovdb_uint64_t a)
 {
    pnanovdb_uint32_t low = pnanovdb_uint64_low(a);
@@ -636,6 +765,11 @@ PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_uint64_offset(pnanovdb_uint64_t
    return a + b;
 }

+PNANOVDB_FORCE_INLINE pnanovdb_int64_t pnanovdb_uint64_diff(pnanovdb_uint64_t a, pnanovdb_uint64_t b)
+{
+    return pnanovdb_uint64_as_int64(a - b);
+}
+
 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_uint64_dec(pnanovdb_uint64_t a)
 {
    return a - 1u;
@@ -702,6 +836,10 @@ PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_address_offset64_product(pnano
    ret.byte_offset += pnanovdb_uint64_low(byte_offset) * multiplier;
    return ret;
 }
+PNANOVDB_FORCE_INLINE pnanovdb_int64_t pnanovdb_address_diff(pnanovdb_address_t a, pnanovdb_address_t b)
+{
+    return pnanovdb_uint64_diff(pnanovdb_uint32_as_uint64_low(a.byte_offset),pnanovdb_uint32_as_uint64_low(b.byte_offset));
+}
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_address_mask(pnanovdb_address_t address, pnanovdb_uint32_t mask)
 {
    return address.byte_offset & mask;
@@ -725,6 +863,12 @@ PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_address_in_interval(pnanovdb_addr
 {
    return address.byte_offset >= min_address.byte_offset && address.byte_offset < max_address.byte_offset;
 }
+PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_address_add(pnanovdb_address_t a, pnanovdb_address_t b)
+{
+    pnanovdb_address_t sum = a;
+    sum.byte_offset += b.byte_offset;
+    return sum;
+}
 #elif defined(PNANOVDB_ADDRESS_64)
 struct pnanovdb_address_t
 {
@@ -762,6 +906,10 @@ PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_address_offset64_product(pnano
    ret.byte_offset += byte_offset * pnanovdb_uint32_as_uint64_low(multiplier);
    return ret;
 }
+PNANOVDB_FORCE_INLINE pnanovdb_int64_t pnanovdb_address_diff(pnanovdb_address_t a, pnanovdb_address_t b)
+{
+    return pnanovdb_uint64_diff(a.byte_offset, b.byte_offset);
+}
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_address_mask(pnanovdb_address_t address, pnanovdb_uint32_t mask)
 {
    return pnanovdb_uint64_low(address.byte_offset) & mask;
@@ -785,6 +933,12 @@ PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_address_in_interval(pnanovdb_addr
 {
    return address.byte_offset >= min_address.byte_offset && address.byte_offset < max_address.byte_offset;
 }
+PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_address_add(pnanovdb_address_t a, pnanovdb_address_t b)
+{
+    pnanovdb_address_t sum = a;
+    sum.byte_offset += b.byte_offset;
+    return sum;
+}
 #endif

 // ------------------------------------------------ High Level Buffer Read -----------------------------------------------------------
@@ -933,8 +1087,8 @@ PNANOVDB_FORCE_INLINE void pnanovdb_write_vec3(pnanovdb_buf_t buf, pnanovdb_addr
 #define PNANOVDB_MAGIC_FILE   0x324244566f6e614eUL// "NanoVDB2" in hex - little endian (uint64_t)

 #define PNANOVDB_MAJOR_VERSION_NUMBER 32// reflects changes to the ABI
-#define PNANOVDB_MINOR_VERSION_NUMBER  7// reflects changes to the API but not ABI
-#define PNANOVDB_PATCH_VERSION_NUMBER  0// reflects bug-fixes with no ABI or API changes
+#define PNANOVDB_MINOR_VERSION_NUMBER  9// reflects changes to the API but not ABI
+#define PNANOVDB_PATCH_VERSION_NUMBER  2// reflects bug-fixes with no ABI or API changes

 #define PNANOVDB_GRID_TYPE_UNKNOWN 0
 #define PNANOVDB_GRID_TYPE_FLOAT 1
@@ -957,14 +1111,16 @@ PNANOVDB_FORCE_INLINE void pnanovdb_write_vec3(pnanovdb_buf_t buf, pnanovdb_addr
 #define PNANOVDB_GRID_TYPE_VEC4D 18
 #define PNANOVDB_GRID_TYPE_INDEX 19
 #define PNANOVDB_GRID_TYPE_ONINDEX 20
-#define PNANOVDB_GRID_TYPE_INDEXMASK 21
-#define PNANOVDB_GRID_TYPE_ONINDEXMASK 22
+//#define PNANOVDB_GRID_TYPE_INDEXMASK 21   // retired - available for future use
+//#define PNANOVDB_GRID_TYPE_ONINDEXMASK 22 // retired - available for future use
 #define PNANOVDB_GRID_TYPE_POINTINDEX 23
 #define PNANOVDB_GRID_TYPE_VEC3U8 24
 #define PNANOVDB_GRID_TYPE_VEC3U16 25
 #define PNANOVDB_GRID_TYPE_UINT8 26
 #define PNANOVDB_GRID_TYPE_END 27

+#define PNANOVDB_GRID_TYPE_CAP 32
+
 #define PNANOVDB_GRID_CLASS_UNKNOWN 0
 #define PNANOVDB_GRID_CLASS_LEVEL_SET 1     // narrow band level set, e.g. SDF
 #define PNANOVDB_GRID_CLASS_FOG_VOLUME 2    // fog volume, e.g. density
@@ -974,8 +1130,9 @@ PNANOVDB_FORCE_INLINE void pnanovdb_write_vec3(pnanovdb_buf_t buf, pnanovdb_addr
 #define PNANOVDB_GRID_CLASS_TOPOLOGY 6      // grid with active states only (no values)
 #define PNANOVDB_GRID_CLASS_VOXEL_VOLUME 7  // volume of geometric cubes, e.g. minecraft
 #define PNANOVDB_GRID_CLASS_INDEX_GRID 8    // grid whose values are offsets, e.g. into an external array
-#define PNANOVDB_GRID_CLASS_TENSOR_GRID 9 // grid which can have extra metadata and features
-#define PNANOVDB_GRID_CLASS_END 10
+#define PNANOVDB_GRID_CLASS_TENSOR_GRID 9   // grid which can have extra metadata and features
+#define PNANOVDB_GRID_CLASS_VOXEL_BVH 10    // grid where each voxel points to list of primitive ids
+#define PNANOVDB_GRID_CLASS_END 11

 #define PNANOVDB_GRID_FLAGS_HAS_LONG_GRID_NAME (1 << 0)
 #define PNANOVDB_GRID_FLAGS_HAS_BBOX (1 << 1)
@@ -989,22 +1146,22 @@ PNANOVDB_FORCE_INLINE void pnanovdb_write_vec3(pnanovdb_buf_t buf, pnanovdb_addr
 #define PNANOVDB_LEAF_TYPE_LITE 1
 #define PNANOVDB_LEAF_TYPE_FP 2
 #define PNANOVDB_LEAF_TYPE_INDEX 3
-#define PNANOVDB_LEAF_TYPE_INDEXMASK 4
+//#define PNANOVDB_LEAF_TYPE_INDEXMASK 4 // retired - available for future use
 #define PNANOVDB_LEAF_TYPE_POINTINDEX 5

 // BuildType = Unknown, float, double, int16_t, int32_t, int64_t, Vec3f, Vec3d, Mask, ...
 // bit count of values in leaf nodes, i.e. 8*sizeof(*nanovdb::LeafNode<BuildType>::mValues) or zero if no values are stored
-PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_value_strides_bits[PNANOVDB_GRID_TYPE_END]  = {  0, 32, 64, 16, 32, 64,  96, 192,  0, 16, 32,  1, 32,  4,  8, 16,  0, 128, 256,  0,  0,  0,  0, 16, 24, 48,  8 };
+PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_value_strides_bits[PNANOVDB_GRID_TYPE_CAP]  = {  0, 32, 64, 16, 32, 64,  96, 192,  0, 16, 32,  1, 32,  4,  8, 16,  0, 128, 256,  0,  0,  0,  0, 16, 24, 48,  8,  0 };
 // bit count of the Tile union in InternalNodes, i.e. 8*sizeof(nanovdb::InternalData::Tile)
-PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_table_strides_bits[PNANOVDB_GRID_TYPE_END]  = { 64, 64, 64, 64, 64, 64, 128, 192, 64, 64, 64, 64, 64, 64, 64, 64, 64, 128, 256, 64, 64, 64, 64, 64, 64, 64, 64 };
+PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_table_strides_bits[PNANOVDB_GRID_TYPE_CAP]  = { 64, 64, 64, 64, 64, 64, 128, 192, 64, 64, 64, 64, 64, 64, 64, 64, 64, 128, 256, 64, 64, 64, 64, 64, 64, 64, 64, 64 };
 // bit count of min/max values, i.e. 8*sizeof(nanovdb::LeafData::mMinimum) or zero if no min/max exists
-PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_minmax_strides_bits[PNANOVDB_GRID_TYPE_END] = {  0, 32, 64, 16, 32, 64,  96, 192,  8, 16, 32,  8, 32, 32, 32, 32, 32, 128, 256, 64, 64, 64, 64, 64, 24, 48,  8 };
+PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_minmax_strides_bits[PNANOVDB_GRID_TYPE_CAP] = {  0, 32, 64, 16, 32, 64,  96, 192,  8, 16, 32,  8, 32, 32, 32, 32, 32, 128, 256, 64, 64,  0,  0, 64, 24, 48,  8,  0 };
 // bit alignment of the value type, controlled by the smallest native type, which is why it is always 0, 8, 16, 32, or 64, e.g. for Vec3f it is 32
-PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_minmax_aligns_bits[PNANOVDB_GRID_TYPE_END]  = {  0, 32, 64, 16, 32, 64,  32,  64,  8, 16, 32,  8, 32, 32, 32, 32, 32,  32,  64, 64, 64, 64, 64, 64,  8, 16,  8 };
+PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_minmax_aligns_bits[PNANOVDB_GRID_TYPE_CAP]  = {  0, 32, 64, 16, 32, 64,  32,  64,  8, 16, 32,  8, 32, 32, 32, 32, 32,  32,  64, 64, 64,  0,  0, 64,  8, 16,  8,  0 };
 // bit alignment of the stats (avg/std-dev) types, e.g. 8*sizeof(nanovdb::LeafData::mAverage)
-PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_stat_strides_bits[PNANOVDB_GRID_TYPE_END]   = {  0, 32, 64, 32, 32, 64,  32,  64,  8, 32, 32,  8, 32, 32, 32, 32, 32,  32,  64, 64, 64, 64, 64, 64, 32, 32, 32 };
+PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_stat_strides_bits[PNANOVDB_GRID_TYPE_CAP]   = {  0, 32, 64, 32, 32, 64,  32,  64,  8, 32, 32,  8, 32, 32, 32, 32, 32,  32,  64, 64, 64,  0,  0, 64, 32, 32, 32,  0 };
 // one of the 4 leaf types defined above, e.g. PNANOVDB_LEAF_TYPE_INDEX = 3
-PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_leaf_type[PNANOVDB_GRID_TYPE_END]           = {  0,  0,  0,  0,  0,  0,  0,    0,  1,  0,  0,  1,  0,  2,  2,  2,  2,   0,   0,  3,  3,  4,  4,  5,  0,  0,  0 };
+PNANOVDB_STATIC_CONST pnanovdb_uint32_t pnanovdb_grid_type_leaf_type[PNANOVDB_GRID_TYPE_CAP]           = {  0,  0,  0,  0,  0,  0,  0,    0,  1,  0,  0,  1,  0,  2,  2,  2,  2,   0,   0,  3,  3,  0,  0,  5,  0,  0,  0,  0 };

 struct pnanovdb_map_t
 {
@@ -1099,7 +1256,9 @@ struct pnanovdb_grid_t
    pnanovdb_uint32_t grid_type;                // 4 bytes,     636
    pnanovdb_int64_t blind_metadata_offset;     // 8 bytes,     640
    pnanovdb_uint32_t blind_metadata_count;     // 4 bytes,     648
-    pnanovdb_uint32_t pad[5];                   // 20 bytes,    652
+    pnanovdb_uint32_t data0;                    // 4 bytes,     652
+    pnanovdb_uint64_t data1;                    // 8 bytes,     656
+    pnanovdb_uint64_t data2;                    // 8 bytes,     664
 };
 PNANOVDB_STRUCT_TYPEDEF(pnanovdb_grid_t)
 struct pnanovdb_grid_handle_t { pnanovdb_address_t address; };
@@ -1122,6 +1281,9 @@ PNANOVDB_STRUCT_TYPEDEF(pnanovdb_grid_handle_t)
 #define PNANOVDB_GRID_OFF_GRID_TYPE 636
 #define PNANOVDB_GRID_OFF_BLIND_METADATA_OFFSET 640
 #define PNANOVDB_GRID_OFF_BLIND_METADATA_COUNT 648
+#define PNANOVDB_GRID_OFF_DATA0 652
+#define PNANOVDB_GRID_OFF_DATA1 656
+#define PNANOVDB_GRID_OFF_DATA2 664

 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_grid_get_magic(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p) {
    return pnanovdb_read_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_MAGIC));
@@ -1170,6 +1332,15 @@ PNANOVDB_FORCE_INLINE pnanovdb_int64_t pnanovdb_grid_get_blind_metadata_offset(p
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_grid_get_blind_metadata_count(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p) {
    return pnanovdb_read_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_BLIND_METADATA_COUNT));
 }
+PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_grid_get_data0(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p) {
+    return pnanovdb_read_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_DATA0));
+}
+PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_grid_get_data1(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p) {
+    return pnanovdb_read_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_DATA1));
+}
+PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_grid_get_data2(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p) {
+    return pnanovdb_read_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_DATA2));
+}

 PNANOVDB_FORCE_INLINE void pnanovdb_grid_set_magic(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p, pnanovdb_uint64_t magic) {
    pnanovdb_write_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_MAGIC), magic);
@@ -1213,6 +1384,15 @@ PNANOVDB_FORCE_INLINE void pnanovdb_grid_set_blind_metadata_offset(pnanovdb_buf_
 PNANOVDB_FORCE_INLINE void pnanovdb_grid_set_blind_metadata_count(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p, pnanovdb_uint32_t metadata_count) {
    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_BLIND_METADATA_COUNT), metadata_count);
 }
+PNANOVDB_FORCE_INLINE void pnanovdb_grid_set_data0(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p, pnanovdb_uint32_t data0) {
+    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_DATA0), data0);
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_grid_set_data1(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p, pnanovdb_uint64_t data1) {
+    pnanovdb_write_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_DATA1), data1);
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_grid_set_data2(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p, pnanovdb_uint64_t data2) {
+    pnanovdb_write_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRID_OFF_DATA2), data2);
+}

 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_make_version(pnanovdb_uint32_t major, pnanovdb_uint32_t minor, pnanovdb_uint32_t patch_num)
 {
@@ -1232,6 +1412,38 @@ PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_version_get_patch(pnanovdb_uint
    return version & ((1u << 10u) - 1u);
 }

+#define PNANOVDB_GRIDBLINDMETADATA_CLASS_UNKNOWN 0
+#define PNANOVDB_GRIDBLINDMETADATA_CLASS_INDEX_ARRAY 1      // indices typically used for mapping into other arrays
+#define PNANOVDB_GRIDBLINDMETADATA_CLASS_ATTRIBUTE_ARRAY 2  // attributes typically associated with points
+#define PNANOVDB_GRIDBLINDMETADATA_CLASS_GRID_NAME 3        // grid names of length longer than 256 characters
+#define PNANOVDB_GRIDBLINDMETADATA_CLASS_CHANNEL_ARRAY 4    // channel of values typically used by index grids
+#define PNANOVDB_GRIDBLINDMETADATA_CLASS_END 5
+
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_UNKNOWN 0
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_POSITION 1  // 3D coordinates in an unknown space
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_COLOR 2     // color associated with point
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_NORMAL 3    // normal associated with point
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_RADIUS 4    // radius of point
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_VELOCITY 5  // velocity associated with point
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_ID 6        // integer ID of point
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_WORLD_COORDS 7    // 3D coordinates in world space, e.g. (0.056, 0.8, 1,8)
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_GRID_COORDS 8     // 3D coordinates in grid space, e.g. (1.2, 4.0, 5.7), aka index-space
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_VOXEL_COORDS 9    // 3D coordinates in voxel space, e.g. (0.2, 0.0, 0.7)
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_LEVEL_SET 10      // narrow band level set, e.g. SDF
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_FOG_VOLUME 11     // fog volume, e.g. density
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_STAGGERED 12      // staggered MAC grid, e.g. velocity
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_OPACITY 13  // opacity associated with point
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_QUAT 14     // quaternion rotation, wxyz convention
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_SCALE 15    // vec3 scale
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_SH0 16      // spherical harmonics, DC component
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_POINT_SHN 17      // spherical haromnics, AC components
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_LINE_ID 18        // integer ID of line
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_TRIANGLE_ID 19    // integer ID of triangle
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_GAUSSIAN_ID 20    // integer ID of Gaussian
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_RANGE 21          // begin/end pair of indices
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_VOXEL_BVH 22      // voxelbvh 64-bit voxel value
+#define PNANOVDB_GRIDBLINDMETADATA_SEMANTIC_END 23
+
 struct pnanovdb_gridblindmetadata_t
 {
    pnanovdb_int64_t data_offset;       // 8 bytes,     0
@@ -1278,6 +1490,28 @@ PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_gridblindmetadata_get_name(pnan
    return pnanovdb_read_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRIDBLINDMETADATA_OFF_NAME + 4u * index));
 }

+PNANOVDB_FORCE_INLINE void pnanovdb_gridblindmetadata_set_data_offset(pnanovdb_buf_t buf, pnanovdb_gridblindmetadata_handle_t p, pnanovdb_int64_t data_offset) {
+    pnanovdb_write_int64(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRIDBLINDMETADATA_OFF_DATA_OFFSET), data_offset);
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_gridblindmetadata_set_value_count(pnanovdb_buf_t buf, pnanovdb_gridblindmetadata_handle_t p, pnanovdb_uint64_t value_count) {
+    pnanovdb_write_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRIDBLINDMETADATA_OFF_VALUE_COUNT), value_count);
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_gridblindmetadata_set_value_size(pnanovdb_buf_t buf, pnanovdb_gridblindmetadata_handle_t p, pnanovdb_uint32_t value_size) {
+    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRIDBLINDMETADATA_OFF_VALUE_SIZE), value_size);
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_gridblindmetadata_set_semantic(pnanovdb_buf_t buf, pnanovdb_gridblindmetadata_handle_t p, pnanovdb_uint32_t semantic) {
+    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRIDBLINDMETADATA_OFF_SEMANTIC), semantic);
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_gridblindmetadata_set_data_class(pnanovdb_buf_t buf, pnanovdb_gridblindmetadata_handle_t p, pnanovdb_uint32_t data_class) {
+    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRIDBLINDMETADATA_OFF_DATA_CLASS), data_class);
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_gridblindmetadata_set_data_type(pnanovdb_buf_t buf, pnanovdb_gridblindmetadata_handle_t p, pnanovdb_uint32_t data_type) {
+    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRIDBLINDMETADATA_OFF_DATA_TYPE), data_type);
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_gridblindmetadata_set_name(pnanovdb_buf_t buf, pnanovdb_gridblindmetadata_handle_t p, pnanovdb_uint32_t index, pnanovdb_uint32_t name) {
+    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_GRIDBLINDMETADATA_OFF_NAME + 4u * index), name);
+}
+
 struct pnanovdb_tree_t
 {
    pnanovdb_uint64_t node_offset_leaf;
@@ -1287,9 +1521,9 @@ struct pnanovdb_tree_t
    pnanovdb_uint32_t node_count_leaf;
    pnanovdb_uint32_t node_count_lower;
    pnanovdb_uint32_t node_count_upper;
-    pnanovdb_uint32_t tile_count_leaf;
    pnanovdb_uint32_t tile_count_lower;
    pnanovdb_uint32_t tile_count_upper;
+    pnanovdb_uint32_t tile_count_root;
    pnanovdb_uint64_t voxel_count;
 };
 PNANOVDB_STRUCT_TYPEDEF(pnanovdb_tree_t)
@@ -1305,9 +1539,9 @@ PNANOVDB_STRUCT_TYPEDEF(pnanovdb_tree_handle_t)
 #define PNANOVDB_TREE_OFF_NODE_COUNT_LEAF 32
 #define PNANOVDB_TREE_OFF_NODE_COUNT_LOWER 36
 #define PNANOVDB_TREE_OFF_NODE_COUNT_UPPER 40
-#define PNANOVDB_TREE_OFF_TILE_COUNT_LEAF 44
-#define PNANOVDB_TREE_OFF_TILE_COUNT_LOWER 48
-#define PNANOVDB_TREE_OFF_TILE_COUNT_UPPER 52
+#define PNANOVDB_TREE_OFF_TILE_COUNT_LOWER 44
+#define PNANOVDB_TREE_OFF_TILE_COUNT_UPPER 48
+#define PNANOVDB_TREE_OFF_TILE_COUNT_ROOT 52
 #define PNANOVDB_TREE_OFF_VOXEL_COUNT 56

 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_tree_get_node_offset_leaf(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p) {
@@ -1331,15 +1565,15 @@ PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_tree_get_node_count_lower(pnano
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_tree_get_node_count_upper(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p) {
    return pnanovdb_read_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_NODE_COUNT_UPPER));
 }
-PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_tree_get_tile_count_leaf(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p) {
-    return pnanovdb_read_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_TILE_COUNT_LEAF));
-}
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_tree_get_tile_count_lower(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p) {
    return pnanovdb_read_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_TILE_COUNT_LOWER));
 }
 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_tree_get_tile_count_upper(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p) {
    return pnanovdb_read_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_TILE_COUNT_UPPER));
 }
+PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_tree_get_tile_count_root(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p) {
+    return pnanovdb_read_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_TILE_COUNT_ROOT));
+}
 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_tree_get_voxel_count(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p) {
    return pnanovdb_read_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_VOXEL_COUNT));
 }
@@ -1365,15 +1599,15 @@ PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_node_count_lower(pnanovdb_buf_t buf
 PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_node_count_upper(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_uint32_t node_count_upper) {
    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_NODE_COUNT_UPPER), node_count_upper);
 }
-PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_tile_count_leaf(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_uint32_t tile_count_leaf) {
-    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_TILE_COUNT_LEAF), tile_count_leaf);
-}
 PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_tile_count_lower(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_uint32_t tile_count_lower) {
    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_TILE_COUNT_LOWER), tile_count_lower);
 }
 PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_tile_count_upper(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_uint32_t tile_count_upper) {
    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_TILE_COUNT_UPPER), tile_count_upper);
 }
+PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_tile_count_root(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_uint32_t tile_count_root) {
+    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_TILE_COUNT_ROOT), tile_count_root);
+}
 PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_voxel_count(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_uint64_t voxel_count) {
    pnanovdb_write_uint64(buf, pnanovdb_address_offset(p.address, PNANOVDB_TREE_OFF_VOXEL_COUNT), voxel_count);
 }
@@ -1501,6 +1735,13 @@ PNANOVDB_FORCE_INLINE void pnanovdb_upper_set_bbox_min(pnanovdb_buf_t buf, pnano
 PNANOVDB_FORCE_INLINE void pnanovdb_upper_set_bbox_max(pnanovdb_buf_t buf, pnanovdb_upper_handle_t p, PNANOVDB_IN(pnanovdb_coord_t) bbox_max) {
    pnanovdb_write_coord(buf, pnanovdb_address_offset(p.address, PNANOVDB_UPPER_OFF_BBOX_MAX), bbox_max);
 }
+PNANOVDB_FORCE_INLINE void pnanovdb_upper_set_value_mask(pnanovdb_buf_t buf, pnanovdb_upper_handle_t p, pnanovdb_uint32_t bit_index, pnanovdb_bool_t value) {
+    pnanovdb_address_t addr = pnanovdb_address_offset(p.address, PNANOVDB_UPPER_OFF_VALUE_MASK + 4u * (bit_index >> 5u));
+    pnanovdb_uint32_t valueMask = pnanovdb_read_uint32(buf, addr);
+    if (!value) { valueMask &= ~(1u << (bit_index & 31u)); }
+    if (value) valueMask |= (1u << (bit_index & 31u));
+    pnanovdb_write_uint32(buf, addr, valueMask);
+}
 PNANOVDB_FORCE_INLINE void pnanovdb_upper_set_child_mask(pnanovdb_buf_t buf, pnanovdb_upper_handle_t p, pnanovdb_uint32_t bit_index, pnanovdb_bool_t value) {
    pnanovdb_address_t addr = pnanovdb_address_offset(p.address, PNANOVDB_UPPER_OFF_CHILD_MASK + 4u * (bit_index >> 5u));
    pnanovdb_uint32_t valueMask = pnanovdb_read_uint32(buf, addr);
@@ -1556,6 +1797,13 @@ PNANOVDB_FORCE_INLINE void pnanovdb_lower_set_bbox_min(pnanovdb_buf_t buf, pnano
 PNANOVDB_FORCE_INLINE void pnanovdb_lower_set_bbox_max(pnanovdb_buf_t buf, pnanovdb_lower_handle_t p, PNANOVDB_IN(pnanovdb_coord_t) bbox_max) {
    pnanovdb_write_coord(buf, pnanovdb_address_offset(p.address, PNANOVDB_LOWER_OFF_BBOX_MAX), bbox_max);
 }
+PNANOVDB_FORCE_INLINE void pnanovdb_lower_set_value_mask(pnanovdb_buf_t buf, pnanovdb_lower_handle_t p, pnanovdb_uint32_t bit_index, pnanovdb_bool_t value) {
+    pnanovdb_address_t addr = pnanovdb_address_offset(p.address, PNANOVDB_LOWER_OFF_VALUE_MASK + 4u * (bit_index >> 5u));
+    pnanovdb_uint32_t valueMask = pnanovdb_read_uint32(buf, addr);
+    if (!value) { valueMask &= ~(1u << (bit_index & 31u)); }
+    if (value) valueMask |= (1u << (bit_index & 31u));
+    pnanovdb_write_uint32(buf, addr, valueMask);
+}
 PNANOVDB_FORCE_INLINE void pnanovdb_lower_set_child_mask(pnanovdb_buf_t buf, pnanovdb_lower_handle_t p, pnanovdb_uint32_t bit_index, pnanovdb_bool_t value) {
    pnanovdb_address_t addr = pnanovdb_address_offset(p.address, PNANOVDB_LOWER_OFF_CHILD_MASK + 4u * (bit_index >> 5u));
    pnanovdb_uint32_t valueMask = pnanovdb_read_uint32(buf, addr);
@@ -1604,6 +1852,13 @@ PNANOVDB_FORCE_INLINE void pnanovdb_leaf_set_bbox_min(pnanovdb_buf_t buf, pnanov
 PNANOVDB_FORCE_INLINE void pnanovdb_leaf_set_bbox_dif_and_flags(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t p, pnanovdb_uint32_t bbox_dif_and_flags) {
    pnanovdb_write_uint32(buf, pnanovdb_address_offset(p.address, PNANOVDB_LEAF_OFF_BBOX_DIF_AND_FLAGS), bbox_dif_and_flags);
 }
+PNANOVDB_FORCE_INLINE void pnanovdb_leaf_set_value_mask(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t p, pnanovdb_uint32_t bit_index, pnanovdb_bool_t value) {
+    pnanovdb_address_t addr = pnanovdb_address_offset(p.address, PNANOVDB_LEAF_OFF_VALUE_MASK + 4u * (bit_index >> 5u));
+    pnanovdb_uint32_t valueMask = pnanovdb_read_uint32(buf, addr);
+    if (!value) { valueMask &= ~(1u << (bit_index & 31u)); }
+    if (value) valueMask |= (1u << (bit_index & 31u));
+    pnanovdb_write_uint32(buf, addr, valueMask);
+}

 struct pnanovdb_grid_type_constants_t
 {
@@ -1639,7 +1894,7 @@ struct pnanovdb_grid_type_constants_t
 PNANOVDB_STRUCT_TYPEDEF(pnanovdb_grid_type_constants_t)

 // The following table with offsets will nedd to be updates as new GridTypes are added in NanoVDB.h
-PNANOVDB_STATIC_CONST pnanovdb_grid_type_constants_t pnanovdb_grid_type_constants[PNANOVDB_GRID_TYPE_END] =
+PNANOVDB_STATIC_CONST pnanovdb_grid_type_constants_t pnanovdb_grid_type_constants[PNANOVDB_GRID_TYPE_CAP] =
 {
 {28, 28, 28, 28, 28, 32,  0, 8, 20, 32,  8224, 8224, 8224, 8224, 8224, 270368,  1056, 1056, 1056, 1056, 1056, 33824,  80, 80, 80, 80, 96, 96},
 {28, 32, 36, 40, 44, 64,  32, 8, 20, 32,  8224, 8228, 8232, 8236, 8256, 270400,  1056, 1060, 1064, 1068, 1088, 33856,  80, 84, 88, 92, 96, 2144},
@@ -1662,12 +1917,13 @@ PNANOVDB_STATIC_CONST pnanovdb_grid_type_constants_t pnanovdb_grid_type_constant
 {32, 64, 96, 128, 136, 160,  256, 32, 24, 64,  8224, 8256, 8288, 8296, 8320, 1056896,  1056, 1088, 1120, 1128, 1152, 132224,  80, 112, 144, 152, 160, 16544},
 {32, 40, 48, 56, 64, 96,  0, 8, 24, 32,  8224, 8232, 8240, 8248, 8256, 270400,  1056, 1064, 1072, 1080, 1088, 33856,  80, 80, 80, 80, 80, 96},
 {32, 40, 48, 56, 64, 96,  0, 8, 24, 32,  8224, 8232, 8240, 8248, 8256, 270400,  1056, 1064, 1072, 1080, 1088, 33856,  80, 80, 80, 80, 80, 96},
-{32, 40, 48, 56, 64, 96,  0, 8, 24, 32,  8224, 8232, 8240, 8248, 8256, 270400,  1056, 1064, 1072, 1080, 1088, 33856,  80, 80, 80, 80, 80, 160},
-{32, 40, 48, 56, 64, 96,  0, 8, 24, 32,  8224, 8232, 8240, 8248, 8256, 270400,  1056, 1064, 1072, 1080, 1088, 33856,  80, 80, 80, 80, 80, 160},
+{28, 28, 28, 28, 28, 32,  0, 8, 20, 32,  8224, 8224, 8224, 8224, 8224, 270368,  1056, 1056, 1056, 1056, 1056, 33824,  80, 80, 80, 80, 96, 96},
+{28, 28, 28, 28, 28, 32,  0, 8, 20, 32,  8224, 8224, 8224, 8224, 8224, 270368,  1056, 1056, 1056, 1056, 1056, 33824,  80, 80, 80, 80, 96, 96},
 {32, 40, 48, 56, 64, 96,  16, 8, 24, 32,  8224, 8232, 8240, 8248, 8256, 270400,  1056, 1064, 1072, 1080, 1088, 33856,  80, 88, 96, 96, 96, 1120},
 {28, 31, 34, 40, 44, 64,  24, 8, 20, 32,  8224, 8227, 8232, 8236, 8256, 270400,  1056, 1059, 1064, 1068, 1088, 33856,  80, 83, 88, 92, 96, 1632},
 {28, 34, 40, 48, 52, 64,  48, 8, 20, 32,  8224, 8230, 8236, 8240, 8256, 270400,  1056, 1062, 1068, 1072, 1088, 33856,  80, 86, 92, 96, 128, 3200},
 {28, 29, 30, 32, 36, 64,  8, 8, 20, 32,  8224, 8225, 8228, 8232, 8256, 270400,  1056, 1057, 1060, 1064, 1088, 33856,  80, 81, 84, 88, 96, 608},
+{28, 28, 28, 28, 28, 32,  0, 8, 20, 32,  8224, 8224, 8224, 8224, 8224, 270368,  1056, 1056, 1056, 1056, 1056, 33824,  80, 80, 80, 80, 96, 96},
 };

 // ------------------------------------------------ Basic Lookup -----------------------------------------------------------
@@ -1719,6 +1975,11 @@ PNANOVDB_FORCE_INLINE pnanovdb_root_tile_handle_t pnanovdb_root_get_tile_zero(pn
    return tile;
 }

+PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_root_tile_get_child_mask(pnanovdb_buf_t buf, pnanovdb_root_tile_handle_t tile)
+{
+    return !pnanovdb_int64_is_zero(pnanovdb_root_tile_get_child(buf, tile));
+}
+
 PNANOVDB_FORCE_INLINE pnanovdb_upper_handle_t pnanovdb_root_get_child(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_root_handle_t root, pnanovdb_root_tile_handle_t tile)
 {
    pnanovdb_upper_handle_t upper = { root.address };
@@ -1726,6 +1987,11 @@ PNANOVDB_FORCE_INLINE pnanovdb_upper_handle_t pnanovdb_root_get_child(pnanovdb_g
    return upper;
 }

+PNANOVDB_FORCE_INLINE void pnanovdb_root_set_child(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_root_handle_t root, pnanovdb_root_tile_handle_t tile, pnanovdb_upper_handle_t upper)
+{
+    pnanovdb_root_tile_set_child(buf, tile, pnanovdb_address_diff(upper.address, root.address));
+}
+
 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_coord_to_key(PNANOVDB_IN(pnanovdb_coord_t) ijk)
 {
 #if defined(PNANOVDB_NATIVE_64)
@@ -1760,6 +2026,27 @@ PNANOVDB_FORCE_INLINE pnanovdb_root_tile_handle_t pnanovdb_root_find_tile(pnanov
    return null_handle;
 }

+// ----------------------------- Grid Type Safe Set ---------------------------------------
+
+PNANOVDB_FORCE_INLINE void pnanovdb_grid_set_first_gridblindmetadata(pnanovdb_buf_t buf, pnanovdb_grid_handle_t p, pnanovdb_gridblindmetadata_handle_t gridblindmetadata) {
+    pnanovdb_grid_set_blind_metadata_offset(buf, p, pnanovdb_address_diff(gridblindmetadata.address, p.address));
+}
+
+// ----------------------------- Tree Type Safe Set ---------------------------------------
+
+PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_first_leaf(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_leaf_handle_t leaf) {
+    pnanovdb_tree_set_node_offset_leaf(buf, p, pnanovdb_address_diff(leaf.address, p.address));
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_first_lower(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_lower_handle_t lower) {
+    pnanovdb_tree_set_node_offset_lower(buf, p, pnanovdb_address_diff(lower.address, p.address));
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_first_upper(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_upper_handle_t upper) {
+    pnanovdb_tree_set_node_offset_upper(buf, p, pnanovdb_address_diff(upper.address, p.address));
+}
+PNANOVDB_FORCE_INLINE void pnanovdb_tree_set_first_root(pnanovdb_buf_t buf, pnanovdb_tree_handle_t p, pnanovdb_root_handle_t root) {
+    pnanovdb_tree_set_node_offset_root(buf, p, pnanovdb_address_diff(root.address, p.address));
+}
+
 // ----------------------------- Leaf Node ---------------------------------------

 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_leaf_coord_to_offset(PNANOVDB_IN(pnanovdb_coord_t) ijk)
@@ -1880,57 +2167,6 @@ PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_index_get_value_index(pnan
    return pnanovdb_uint64_offset(offset, n);
 }

-// ----------------------------- Leaf IndexMask Specialization ---------------------------------------
-
-PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_leaf_indexmask_has_stats(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf)
-{
-    return pnanovdb_leaf_index_has_stats(buf, leaf);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_indexmask_get_min_index(pnanovdb_buf_t buf, pnanovdb_address_t min_address)
-{
-    return pnanovdb_leaf_index_get_min_index(buf, min_address);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_indexmask_get_max_index(pnanovdb_buf_t buf, pnanovdb_address_t max_address)
-{
-    return pnanovdb_leaf_index_get_max_index(buf, max_address);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_indexmask_get_ave_index(pnanovdb_buf_t buf, pnanovdb_address_t ave_address)
-{
-    return pnanovdb_leaf_index_get_ave_index(buf, ave_address);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_indexmask_get_dev_index(pnanovdb_buf_t buf, pnanovdb_address_t dev_address)
-{
-    return pnanovdb_leaf_index_get_dev_index(buf, dev_address);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_indexmask_get_value_index(pnanovdb_buf_t buf, pnanovdb_address_t value_address, PNANOVDB_IN(pnanovdb_coord_t) ijk)
-{
-    return pnanovdb_leaf_index_get_value_index(buf, value_address, ijk);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_leaf_indexmask_get_mask_bit(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf, pnanovdb_uint32_t n)
-{
-    pnanovdb_uint32_t word_idx = n >> 5;
-    pnanovdb_uint32_t bit_idx = n & 31;
-    pnanovdb_uint32_t val_mask =
-        pnanovdb_read_uint32(buf, pnanovdb_address_offset(leaf.address, 96u + 4u * word_idx));
-    return (val_mask & (1u << bit_idx)) != 0u;
-}
-PNANOVDB_FORCE_INLINE void pnanovdb_leaf_indexmask_set_mask_bit(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf, pnanovdb_uint32_t n, pnanovdb_bool_t v)
-{
-    pnanovdb_uint32_t word_idx = n >> 5;
-    pnanovdb_uint32_t bit_idx = n & 31;
-    pnanovdb_uint32_t val_mask =
-        pnanovdb_read_uint32(buf, pnanovdb_address_offset(leaf.address, 96u + 4u * word_idx));
-    if (v)
-    {
-        val_mask = val_mask | (1u << bit_idx);
-    }
-    else
-    {
-        val_mask = val_mask & ~(1u << bit_idx);
-    }
-    pnanovdb_write_uint32(buf, pnanovdb_address_offset(leaf.address, 96u + 4u * word_idx), val_mask);
-}
-
 // ----------------------------- Leaf OnIndex Specialization ---------------------------------------

 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_leaf_onindex_get_value_count(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf)
@@ -2022,65 +2258,6 @@ PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_onindex_get_value_index(pn
    return value_index;
 }

-// ----------------------------- Leaf OnIndexMask Specialization ---------------------------------------
-
-PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_leaf_onindexmask_get_value_count(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf)
-{
-    return pnanovdb_leaf_onindex_get_value_count(buf, leaf);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_onindexmask_get_last_offset(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf)
-{
-    return pnanovdb_leaf_onindex_get_last_offset(buf, leaf);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_leaf_onindexmask_has_stats(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf)
-{
-    return pnanovdb_leaf_onindex_has_stats(buf, leaf);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_onindexmask_get_min_index(pnanovdb_buf_t buf, pnanovdb_address_t min_address)
-{
-    return pnanovdb_leaf_onindex_get_min_index(buf, min_address);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_onindexmask_get_max_index(pnanovdb_buf_t buf, pnanovdb_address_t max_address)
-{
-    return pnanovdb_leaf_onindex_get_max_index(buf, max_address);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_onindexmask_get_ave_index(pnanovdb_buf_t buf, pnanovdb_address_t ave_address)
-{
-    return pnanovdb_leaf_onindex_get_ave_index(buf, ave_address);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_onindexmask_get_dev_index(pnanovdb_buf_t buf, pnanovdb_address_t dev_address)
-{
-    return pnanovdb_leaf_onindex_get_dev_index(buf, dev_address);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_onindexmask_get_value_index(pnanovdb_buf_t buf, pnanovdb_address_t value_address, PNANOVDB_IN(pnanovdb_coord_t) ijk)
-{
-    return pnanovdb_leaf_onindex_get_value_index(buf, value_address, ijk);
-}
-PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_leaf_onindexmask_get_mask_bit(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf, pnanovdb_uint32_t n)
-{
-    pnanovdb_uint32_t word_idx = n >> 5;
-    pnanovdb_uint32_t bit_idx = n & 31;
-    pnanovdb_uint32_t val_mask =
-        pnanovdb_read_uint32(buf, pnanovdb_address_offset(leaf.address, 96u + 4u * word_idx));
-    return (val_mask & (1u << bit_idx)) != 0u;
-}
-PNANOVDB_FORCE_INLINE void pnanovdb_leaf_onindexmask_set_mask_bit(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf, pnanovdb_uint32_t n, pnanovdb_bool_t v)
-{
-    pnanovdb_uint32_t word_idx = n >> 5;
-    pnanovdb_uint32_t bit_idx = n & 31;
-    pnanovdb_uint32_t val_mask =
-        pnanovdb_read_uint32(buf, pnanovdb_address_offset(leaf.address, 96u + 4u * word_idx));
-    if (v)
-    {
-        val_mask = val_mask | (1u << bit_idx);
-    }
-    else
-    {
-        val_mask = val_mask & ~(1u << bit_idx);
-    }
-    pnanovdb_write_uint32(buf, pnanovdb_address_offset(leaf.address, 96u + 4u * word_idx), val_mask);
-}
-
 // ----------------------------- Leaf PointIndex Specialization ---------------------------------------

 PNANOVDB_FORCE_INLINE pnanovdb_uint64_t pnanovdb_leaf_pointindex_get_offset(pnanovdb_buf_t buf, pnanovdb_leaf_handle_t leaf)
@@ -2210,6 +2387,17 @@ PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_lower_get_value_address(pnanov
    return pnanovdb_lower_get_value_address_and_level(grid_type, buf, lower, ijk, PNANOVDB_REF(level));
 }

+PNANOVDB_FORCE_INLINE void pnanovdb_lower_set_table_child(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_lower_handle_t node, pnanovdb_uint32_t n, pnanovdb_int64_t child)
+{
+    pnanovdb_address_t table_address = pnanovdb_lower_get_table_address(grid_type, buf, node, n);
+    pnanovdb_write_int64(buf, table_address, child);
+}
+
+PNANOVDB_FORCE_INLINE void pnanovdb_lower_set_child(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_lower_handle_t lower, pnanovdb_uint32_t n, pnanovdb_leaf_handle_t leaf)
+{
+    pnanovdb_lower_set_table_child(grid_type, buf, lower, n, pnanovdb_address_diff(leaf.address, lower.address));
+}
+
 // ------------------------------------------------ Upper Node -----------------------------------------------------------

 PNANOVDB_FORCE_INLINE pnanovdb_uint32_t pnanovdb_upper_coord_to_offset(PNANOVDB_IN(pnanovdb_coord_t) ijk)
@@ -2291,8 +2479,19 @@ PNANOVDB_FORCE_INLINE void pnanovdb_upper_set_table_child(pnanovdb_grid_type_t g
    pnanovdb_write_int64(buf, bufAddress, child);
 }

+PNANOVDB_FORCE_INLINE void pnanovdb_upper_set_child(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_upper_handle_t upper, pnanovdb_uint32_t n, pnanovdb_lower_handle_t lower)
+{
+    pnanovdb_upper_set_table_child(grid_type, buf, upper, n, pnanovdb_address_diff(lower.address, upper.address));
+}
+
 // ------------------------------------------------ Root -----------------------------------------------------------

+PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_root_get_background_address(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_root_handle_t root)
+{
+    pnanovdb_uint32_t byte_offset = PNANOVDB_GRID_TYPE_GET(grid_type, root_off_background);
+    return pnanovdb_address_offset(root.address, byte_offset);
+}
+
 PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_root_get_min_address(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_root_handle_t root)
 {
    pnanovdb_uint32_t byte_offset = PNANOVDB_GRID_TYPE_GET(grid_type, root_off_min);
@@ -2591,12 +2790,6 @@ PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_lower_get_value_address_and_ca
    return pnanovdb_lower_get_value_address_and_level_and_cache(grid_type, buf, lower, ijk, acc, PNANOVDB_REF(level));
 }

-PNANOVDB_FORCE_INLINE void pnanovdb_lower_set_table_child(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_lower_handle_t node, pnanovdb_uint32_t n, pnanovdb_int64_t child)
-{
-    pnanovdb_address_t table_address = pnanovdb_lower_get_table_address(grid_type, buf, node, n);
-    pnanovdb_write_int64(buf, table_address, child);
-}
-
 PNANOVDB_FORCE_INLINE pnanovdb_address_t pnanovdb_upper_get_value_address_and_level_and_cache(pnanovdb_grid_type_t grid_type, pnanovdb_buf_t buf, pnanovdb_upper_handle_t upper, PNANOVDB_IN(pnanovdb_coord_t) ijk, PNANOVDB_INOUT(pnanovdb_readaccessor_t) acc, PNANOVDB_INOUT(pnanovdb_uint32_t) level)
 {
    pnanovdb_uint32_t n = pnanovdb_upper_coord_to_offset(ijk);
@@ -3227,6 +3420,19 @@ PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_hdda_update(PNANOVDB_INOUT(pnanov
    {
        return PNANOVDB_FALSE;
    }
+
+    // compute valid voxel range
+    pnanovdb_coord_t voxel_max = {
+        (PNANOVDB_DEREF(hdda).voxel.x + PNANOVDB_DEREF(hdda).dim - 1) & (~(dim - 1)),
+        (PNANOVDB_DEREF(hdda).voxel.y + PNANOVDB_DEREF(hdda).dim - 1) & (~(dim - 1)),
+        (PNANOVDB_DEREF(hdda).voxel.z + PNANOVDB_DEREF(hdda).dim - 1) & (~(dim - 1))
+    };
+    pnanovdb_coord_t voxel_min = {
+        PNANOVDB_DEREF(hdda).voxel.x & (~(dim - 1)),
+        PNANOVDB_DEREF(hdda).voxel.y & (~(dim - 1)),
+        PNANOVDB_DEREF(hdda).voxel.z & (~(dim - 1))
+    };
+
    PNANOVDB_DEREF(hdda).dim = dim;

    pnanovdb_vec3_t pos = pnanovdb_vec3_add(
@@ -3237,6 +3443,14 @@ PNANOVDB_FORCE_INLINE pnanovdb_bool_t pnanovdb_hdda_update(PNANOVDB_INOUT(pnanov

    PNANOVDB_DEREF(hdda).voxel = pnanovdb_hdda_pos_to_voxel(PNANOVDB_REF(pos), dim);

+    // clamp voxel to valid range
+    PNANOVDB_DEREF(hdda).voxel.x = PNANOVDB_DEREF(hdda).voxel.x < voxel_max.x ? PNANOVDB_DEREF(hdda).voxel.x : voxel_max.x;
+    PNANOVDB_DEREF(hdda).voxel.y = PNANOVDB_DEREF(hdda).voxel.y < voxel_max.y ? PNANOVDB_DEREF(hdda).voxel.y : voxel_max.y;
+    PNANOVDB_DEREF(hdda).voxel.z = PNANOVDB_DEREF(hdda).voxel.z < voxel_max.z ? PNANOVDB_DEREF(hdda).voxel.z : voxel_max.z;
+    PNANOVDB_DEREF(hdda).voxel.x = PNANOVDB_DEREF(hdda).voxel.x > voxel_min.x ? PNANOVDB_DEREF(hdda).voxel.x : voxel_min.x;
+    PNANOVDB_DEREF(hdda).voxel.y = PNANOVDB_DEREF(hdda).voxel.y > voxel_min.y ? PNANOVDB_DEREF(hdda).voxel.y : voxel_min.y;
+    PNANOVDB_DEREF(hdda).voxel.z = PNANOVDB_DEREF(hdda).voxel.z > voxel_min.z ? PNANOVDB_DEREF(hdda).voxel.z : voxel_min.z;
+
    if (PNANOVDB_DEREF(hdda).step.x != 0)
    {
        PNANOVDB_DEREF(hdda).next.x = PNANOVDB_DEREF(hdda).tmin + (PNANOVDB_DEREF(hdda).voxel.x - pos.x) * dir_inv.x;
--- a/Show More
+++ b/Show More