Version bump

Commentary: Changes update
Fix $countbits in assert with non-tristates (#5566 ).
2026-10-02 16:13:34 +02:00 · 2024-10-27 09:38:06 -04:00 · 2024-10-27 09:33:54 -04:00 · 2024-10-27 09:30:54 -04:00 · 2024-10-25 18:54:39 -04:00 · 2024-10-25 18:49:45 -04:00
5990 changed files with 109533 additions and 121802 deletions
@@ -1,4 +1,4 @@
-Checks:          '*,-hicpp*,-android-cloexec-fopen,-cert-dcl50-cpp,-cert-env33-c,-cert-err34-c,-cert-err58-cpp,-clang-analyzer-core.UndefinedBinaryOperatorResult,-clang-analyzer-security*,-cppcoreguidelines-avoid-magic-numbers,-cppcoreguidelines-no-malloc,-cppcoreguidelines-owning-memory,-cppcoreguidelines-pro-bounds-array-to-pointer-decay,-cppcoreguidelines-pro-bounds-array-to-pointer-decay,-cppcoreguidelines-pro-bounds-constant-array-index,-cppcoreguidelines-pro-bounds-pointer-arithmetic,-cppcoreguidelines-pro-type-const-cast,-cppcoreguidelines-pro-type-reinterpret-cast,-cppcoreguidelines-pro-type-static-cast-downcast,-cppcoreguidelines-pro-type-union-access,-cppcoreguidelines-pro-type-vararg,-cppcoreguidelines-special-member-functions,-fuchsia-*,-google-default-arguments,-google-readability-todo,-google-runtime-references,-llvm-header-guard,-llvm-include-order,-misc-string-integer-assignment,-misc-string-literal-with-embedded-nul,-modernize-use-auto,-modernize-use-trailing-return-type,-readability-braces-around-statements,-readability-container-size-empty,-readability-delete-null-pointer,-readability-else-after-return,-readability-implicit-bool-conversion,-readability-named-parameter,-readability-static-accessed-through-instance'
+Checks:          '*,-hicpp*,-android-cloexec-fopen,-cert-dcl50-cpp,-cert-env33-c,-cert-err34-c,-cert-err58-cpp,-clang-analyzer-core.UndefinedBinaryOperatorResult,-clang-analyzer-security*,-cppcoreguidelines-avoid-magic-numbers,-cppcoreguidelines-no-malloc,-cppcoreguidelines-owning-memory,-cppcoreguidelines-pro-bounds-array-to-pointer-decay,-cppcoreguidelines-pro-bounds-array-to-pointer-decay,-cppcoreguidelines-pro-bounds-constant-array-index,-cppcoreguidelines-pro-bounds-pointer-arithmetic,-cppcoreguidelines-pro-type-const-cast,-cppcoreguidelines-pro-type-reinterpret-cast,-cppcoreguidelines-pro-type-static-cast-downcast,-cppcoreguidelines-pro-type-union-access,-cppcoreguidelines-pro-type-vararg,-cppcoreguidelines-special-member-functions,-fuchsia-*,-google-default-arguments,-google-readability-todo,-google-runtime-references,-llvm-header-guard,-llvm-include-order,-misc-string-integer-assignment,-misc-string-literal-with-embedded-nul,-modernize-use-auto,-modernize-use-trailing-return-type,-readability-braces-around-statements,-readability-container-size-empty,-readability-delete-null-pointer,-readability-else-after-return,-readability-implicit-bool-conversion,-readability-named-parameter,-readability-static-accessed-through-instance,-llvmlibc-*,-altera-*'
 WarningsAsErrors: ''
 HeaderFilterRegex: ''
 FormatStyle:     none
@@ -1,6 +1,6 @@
 ---
 exclude_paths:
-  - '.github/**'
-  - 'ci/build_verilator.sh'
-  - 'include/vltstd/**'
-  - 'nodist/fastcov.py'
+  - ".github/**"
+  - "ci/build_verilator.sh"
+  - "include/vltstd/**"
+  - "nodist/fastcov.py"
@@ -1,7 +1,7 @@
 {
-	"name": "Verilator Build Environment",
+        "name": "Verilator Build Environment",

-	"build": {
+        "build": {
        "dockerfile": "../ci/docker/buildenv/Dockerfile"
    }

@@ -18,3 +18,5 @@ What 'verilator --version' are you using?  Did you try it with the git master ve
 What OS and distribution are you using?

 May we assist you in trying to fix this in Verilator yourself?
+
+(Please avoid attaching screenshots that show text - you can convert images to text using e.g. https://ocr.space)
@@ -11,6 +11,8 @@ Thanks for taking the time to report this.

 What would you like added/supported?

+What 'verilator --version' are you using?  Did you try it with the git master version?
+
 Can you attach an example that runs on other simulators?  (Must be openly licensed, ideally in test_regress format.)

 May we assist you in trying to fix this in Verilator yourself?
@@ -8,3 +8,5 @@ assignees: ''
 ---

 How may we help - what is your question?
+
+(If reporting a bug or requesting a feature please use one of the other issue templates.)
@@ -1,3 +1,4 @@
+---
 # DESCRIPTION: Github actions config
 # This name is key to badges in README.rst, so we use the name build
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
@@ -9,7 +10,7 @@ on:
  pull_request:
  workflow_dispatch:
  schedule:
-  - cron: '0 0 * * 0' # weekly
+    - cron: '0 0 * * 0'  # weekly

 env:
  CI_OS_NAME: linux
@@ -35,14 +36,14 @@ jobs:
    strategy:
      fail-fast: false
      matrix:
-        os: [ubuntu-22.04, ubuntu-20.04]
+        os: [ubuntu-24.04, ubuntu-22.04, ubuntu-20.04]
        compiler:
-          - { cc: clang, cxx: clang++ }
-          - { cc: gcc,   cxx: g++     }
+          - {cc: clang, cxx: clang++}
+          - {cc: gcc, cxx: g++}
        include:
          # Build GCC 10 on ubuntu-20.04
          - os: ubuntu-20.04
-            compiler: { cc: gcc-10, cxx: g++-10 }
+            compiler: {cc: gcc-10, cxx: g++-10}
    runs-on: ${{ matrix.os }}
    name: Build | ${{ matrix.os }} | ${{ matrix.compiler.cc }}
    env:
@@ -51,41 +52,40 @@ jobs:
      CC: ${{ matrix.compiler.cc }}
      CXX: ${{ matrix.compiler.cxx }}
      CACHE_BASE_KEY: build-${{ matrix.os }}-${{ matrix.compiler.cc }}
-      CCACHE_MAXSIZE: 1000M # Per build matrix entry (* 5 = 5000M in total)
+      CCACHE_MAXSIZE: 1000M  # Per build matrix entry (* 5 = 5000M in total)
      VERILATOR_ARCHIVE: verilator-${{ github.sha }}-${{ matrix.os }}-${{ matrix.compiler.cc }}.tar.gz
    steps:

-    - name: Checkout
-      uses: actions/checkout@v4
-      with:
-        path: repo
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          path: repo

-    - name: Cache $CCACHE_DIR
-      uses: actions/cache@v4
-      env:
-        CACHE_KEY: ${{ env.CACHE_BASE_KEY }}-ccache
-      with:
-        path: ${{ env.CCACHE_DIR }}
-        key: ${{ env.CACHE_KEY }}-${{ github.sha }}
-        restore-keys: |
-          ${{ env.CACHE_KEY }}-
+      - name: Cache $CCACHE_DIR
+        uses: actions/cache@v4
+        env:
+          CACHE_KEY: ${{ env.CACHE_BASE_KEY }}-ccache
+        with:
+          path: ${{ env.CCACHE_DIR }}
+          key: ${{ env.CACHE_KEY }}-${{ github.sha }}
+          restore-keys: |
+            ${{ env.CACHE_KEY }}-

-    - name: Install packages for build
-      run: ./ci/ci-install.bash
+      - name: Install packages for build
+        run: ./ci/ci-install.bash

-    - name: Build
-      run: ./ci/ci-script.bash
+      - name: Build
+        run: ./ci/ci-script.bash

-    - name: Tar up repository
-      working-directory: ${{ github.workspace }}
-      run: tar --posix -c -z -f ${{ env.VERILATOR_ARCHIVE }} repo
-
-    - name: Upload tar archive
-      uses: actions/upload-artifact@v4
-      with:
-        path: ${{ github.workspace }}/${{ env.VERILATOR_ARCHIVE }}
-        name: ${{ env.VERILATOR_ARCHIVE }}
+      - name: Tar up repository
+        working-directory: ${{ github.workspace }}
+        run: tar --posix -c -z -f ${{ env.VERILATOR_ARCHIVE }} repo

+      - name: Upload tar archive
+        uses: actions/upload-artifact@v4
+        with:
+          path: ${{ github.workspace }}/${{ env.VERILATOR_ARCHIVE }}
+          name: ${{ env.VERILATOR_ARCHIVE }}

  test:
    needs: build
@@ -94,25 +94,25 @@ jobs:
      matrix:
        os: [ubuntu-22.04, ubuntu-20.04]
        compiler:
-          - { cc: clang, cxx: clang++ }
-          - { cc: gcc,   cxx: g++     }
+          - {cc: clang, cxx: clang++}
+          - {cc: gcc, cxx: g++}
        reloc: [0]
        suite: [dist-vlt-0, dist-vlt-1, dist-vlt-2, dist-vlt-3, vltmt-0, vltmt-1]
        include:
-          # Test with GCC 10 on ubuntu-20.04
-          - {os: ubuntu-20.04, compiler: { cc: gcc-10, cxx: g++-10 }, suite: dist-vlt-0}
-          - {os: ubuntu-20.04, compiler: { cc: gcc-10, cxx: g++-10 }, suite: dist-vlt-1}
-          - {os: ubuntu-20.04, compiler: { cc: gcc-10, cxx: g++-10 }, suite: dist-vlt-2}
-          - {os: ubuntu-20.04, compiler: { cc: gcc-10, cxx: g++-10 }, suite: dist-vlt-3}
-          - {os: ubuntu-20.04, compiler: { cc: gcc-10, cxx: g++-10 }, suite: vltmt-0}
-          - {os: ubuntu-20.04, compiler: { cc: gcc-10, cxx: g++-10 }, suite: vltmt-1}
-          # Test relocated installation - on most common platform only
-          - {os: ubuntu-22.04, compiler: { cc: gcc, cxx: g++ }, reloc: 1, suite: dist-vlt-0}
-          - {os: ubuntu-22.04, compiler: { cc: gcc, cxx: g++ }, reloc: 1, suite: dist-vlt-1}
-          - {os: ubuntu-22.04, compiler: { cc: gcc, cxx: g++ }, reloc: 1, suite: dist-vlt-2}
-          - {os: ubuntu-22.04, compiler: { cc: gcc, cxx: g++ }, reloc: 1, suite: dist-vlt-3}
-          - {os: ubuntu-22.04, compiler: { cc: gcc, cxx: g++ }, reloc: 1, suite: vltmt-0}
-          - {os: ubuntu-22.04, compiler: { cc: gcc, cxx: g++ }, reloc: 1, suite: vltmt-1}
+          # Test with GCC 10 on ubuntu-20.04, also test relocation
+          - {os: ubuntu-20.04, compiler: {cc: gcc-10, cxx: g++-10}, reloc: 1, suite: dist-vlt-0}
+          - {os: ubuntu-20.04, compiler: {cc: gcc-10, cxx: g++-10}, reloc: 1, suite: dist-vlt-1}
+          - {os: ubuntu-20.04, compiler: {cc: gcc-10, cxx: g++-10}, reloc: 1, suite: dist-vlt-2}
+          - {os: ubuntu-20.04, compiler: {cc: gcc-10, cxx: g++-10}, reloc: 1, suite: dist-vlt-3}
+          - {os: ubuntu-20.04, compiler: {cc: gcc-10, cxx: g++-10}, reloc: 1, suite: vltmt-0}
+          - {os: ubuntu-20.04, compiler: {cc: gcc-10, cxx: g++-10}, reloc: 1, suite: vltmt-1}
+          # Ubuntu 24.04 only on GCC; not passing on clang yet
+          - {os: ubuntu-24.04, compiler: {cc: gcc, cxx: g++}, suite: dist-vlt-0}
+          - {os: ubuntu-24.04, compiler: {cc: gcc, cxx: g++}, suite: dist-vlt-1}
+          - {os: ubuntu-24.04, compiler: {cc: gcc, cxx: g++}, suite: dist-vlt-2}
+          - {os: ubuntu-24.04, compiler: {cc: gcc, cxx: g++}, suite: dist-vlt-3}
+          - {os: ubuntu-24.04, compiler: {cc: gcc, cxx: g++}, suite: vltmt-0}
+          - {os: ubuntu-24.04, compiler: {cc: gcc, cxx: g++}, suite: vltmt-1}
    runs-on: ${{ matrix.os }}
    name: Test | ${{ matrix.os }} | ${{ matrix.compiler.cc }} | ${{ matrix.reloc && 'reloc | ' || '' }} ${{ matrix.suite }}
    env:
@@ -122,37 +122,37 @@ jobs:
      CC: ${{ matrix.compiler.cc }}
      CXX: ${{ matrix.compiler.cxx }}
      CACHE_BASE_KEY: test-${{ matrix.os }}-${{ matrix.compiler.cc }}-${{ matrix.reloc }}-${{ matrix.suite }}
-      CCACHE_MAXSIZE: 100M # Per build per suite (* 5 * 5 = 2500M in total)
+      CCACHE_MAXSIZE: 100M  # Per build per suite (* 5 * 5 = 2500M in total)
      VERILATOR_ARCHIVE: verilator-${{ github.sha }}-${{ matrix.os }}-${{ matrix.compiler.cc }}.tar.gz
    steps:

-    - name: Download tar archive
-      uses: actions/download-artifact@v4
-      with:
-        name: ${{ env.VERILATOR_ARCHIVE }}
-        path: ${{ github.workspace }}
+      - name: Download tar archive
+        uses: actions/download-artifact@v4
+        with:
+          name: ${{ env.VERILATOR_ARCHIVE }}
+          path: ${{ github.workspace }}

-    - name: Unpack tar archive
-      working-directory: ${{ github.workspace }}
-      run: tar -x -z -f ${{ env.VERILATOR_ARCHIVE }}
+      - name: Unpack tar archive
+        working-directory: ${{ github.workspace }}
+        run: tar -x -z -f ${{ env.VERILATOR_ARCHIVE }}

-    - name: Cache $CCACHE_DIR
-      uses: actions/cache@v4
-      env:
-        CACHE_KEY: ${{ env.CACHE_BASE_KEY }}-ccache2
-      with:
-        path: ${{ env.CCACHE_DIR }}
-        key: ${{ env.CACHE_KEY }}-${{ github.sha }}
-        restore-keys: |
-          ${{ env.CACHE_KEY }}-
+      - name: Cache $CCACHE_DIR
+        uses: actions/cache@v4
+        env:
+          CACHE_KEY: ${{ env.CACHE_BASE_KEY }}-ccache2
+        with:
+          path: ${{ env.CCACHE_DIR }}
+          key: ${{ env.CACHE_KEY }}-${{ github.sha }}
+          restore-keys: |
+            ${{ env.CACHE_KEY }}-

-    - name: Install test dependencies
-      run: ./ci/ci-install.bash
+      - name: Install test dependencies
+        run: ./ci/ci-install.bash

-    - name: Test
-      env:
-        TESTS: ${{ matrix.suite }}
-      run: ./ci/ci-script.bash
+      - name: Test
+        env:
+          TESTS: ${{ matrix.suite }}
+        run: ./ci/ci-script.bash

  lint-py:
    runs-on: ubuntu-22.04
@@ -161,21 +161,21 @@ jobs:
      CI_BUILD_STAGE_NAME: build
      CI_RUNS_ON: ubuntu-22.04
    steps:
-    - name: Checkout
-      uses: actions/checkout@v4
-      with:
-        path: repo
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          path: repo

-    - name: Install packages for build
-      run: ./ci/ci-install.bash
+      - name: Install packages for build
+        run: ./ci/ci-install.bash

    # We use specific version numbers, otherwise a Python package
    # update may add a warning and break our build
-    - name: Install packages for lint
-      run: sudo pip3 install pylint==3.0.2 ruff==0.1.3 clang sphinx sphinx_rtd_theme sphinxcontrib-spelling breathe ruff
+      - name: Install packages for lint
+        run: sudo pip3 install pylint==3.0.2 ruff==0.1.3 clang sphinx sphinx_rtd_theme sphinxcontrib-spelling breathe ruff

-    - name: Configure
-      run: autoconf && ./configure --enable-longtests --enable-ccwarn
+      - name: Configure
+        run: autoconf && ./configure --enable-longtests --enable-ccwarn

-    - name: Lint
-      run: make -k lint-py
+      - name: Lint
+        run: make -k lint-py
@@ -1,17 +1,16 @@
+---
 # DESCRIPTION: Github actions config
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
-
+#
 name: Contributor Agreement
-
 on:
  push:
  pull_request:
  workflow_dispatch:
-
 jobs:
  Test:
    name: "'docs/CONTRIBUTORS' was signed"
    runs-on: ubuntu-22.04
    steps:
-    - uses: actions/checkout@v4
-    - run: test_regress/t/t_dist_contributors.pl
+      - uses: actions/checkout@v4
+      - run: test_regress/t/t_dist_contributors.py
@@ -1,3 +1,4 @@
+---
 # DESCRIPTION: Github actions config
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

@@ -6,7 +7,7 @@ name: coverage
 on:
  workflow_dispatch:
  schedule:
-  - cron: '0 0 * * 0' # weekly
+    - cron: '0 0 * * 0'  # weekly

 env:
  CI_OS_NAME: linux
@@ -21,7 +22,6 @@ defaults:

 jobs:

-
  Build:
    runs-on: ubuntu-22.04
    env:
@@ -29,49 +29,36 @@ jobs:
      CI_RUNS_ON: ubuntu-22.04
    steps:

-    - name: Checkout
-      uses: actions/checkout@v4
-      with:
-        path: repo
+      - name: Checkout
+        uses: actions/checkout@v4
+        with:
+          path: repo

-    - name: Install packages for build
-      run: ./ci/ci-install.bash
+      - name: Install packages for build
+        run: ./ci/ci-install.bash

-    - name: Build
-      run: ./ci/ci-script.bash
+      - name: Build
+        run: ./ci/ci-script.bash

-    - name: Tar up repository
-      working-directory: ${{ github.workspace }}
-      run: tar --posix -c -z -f ${{ env.VERILATOR_ARCHIVE }} repo
-
-    - name: Upload tar archive
-      uses: actions/upload-artifact@v4
-      with:
-        path: ${{ github.workspace }}/${{ env.VERILATOR_ARCHIVE }}
-        name: ${{ env.VERILATOR_ARCHIVE }}
+      - name: Tar up repository
+        working-directory: ${{ github.workspace }}
+        run: tar --posix -c -z -f ${{ env.VERILATOR_ARCHIVE }} repo

+      - name: Upload tar archive
+        uses: actions/upload-artifact@v4
+        with:
+          path: ${{ github.workspace }}/${{ env.VERILATOR_ARCHIVE }}
+          name: ${{ env.VERILATOR_ARCHIVE }}

  Test:
    needs: Build
    strategy:
      fail-fast: false
      matrix:
-        test:
-        - vlt-
-        - vltmt-
-        num:
-        - 0
-        - 1
-        - 2
-        - 3
-        - 4
-        - 5
-        - 6
-        - 7
-        - 8
-        - 9
+        test: [vlt-, vltmt-]
+        num: [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
        include:
-        - { test: dist, num: '' }
+          - {test: dist, num: ''}
    runs-on: ubuntu-22.04
    name: test-${{ matrix.test }}${{ matrix.num }}
    env:
@@ -79,20 +66,27 @@ jobs:
      CI_RUNS_ON: ubuntu-22.04
    steps:

-    - name: Download tar archive
-      uses: actions/download-artifact@v4
-      with:
-        name: ${{ env.VERILATOR_ARCHIVE }}
-        path: ${{ github.workspace }}
+      - name: Download tar archive
+        uses: actions/download-artifact@v4
+        with:
+          name: ${{ env.VERILATOR_ARCHIVE }}
+          path: ${{ github.workspace }}

-    - name: Unpack tar archive
-      working-directory: ${{ github.workspace }}
-      run: tar -x -z -f ${{ env.VERILATOR_ARCHIVE }}
+      - name: Unpack tar archive
+        working-directory: ${{ github.workspace }}
+        run: tar -x -z -f ${{ env.VERILATOR_ARCHIVE }}

-    - name: Install test dependencies
-      run: ./ci/ci-install.bash
+      - name: Install test dependencies
+        run: ./ci/ci-install.bash

-    - name: Test
-      env:
-        TESTS: coverage-${{ matrix.test }}${{ matrix.num }}
-      run: ./ci/ci-script.bash
+      - name: Test
+        env:
+          TESTS: coverage-${{ matrix.test }}${{ matrix.num }}
+        run: ./ci/ci-script.bash
+
+      - name: Upload coverage data to Codecov
+        env:
+          CODECOV_TOKEN: ${{ secrets.CODECOV_TOKEN }}
+        run: |
+          find . -name '*.gcno' -exec rm {} \;
+          ./ci/codecov -v upload-process -Z --sha ${{ github.sha }} -f nodist/obj_dir/coverage/app_total.info
@@ -1,3 +1,4 @@
+---
 # Build and push verilator docker image when tags are pushed to the repository.
 # The following variable(s) must be configured in the github repository:
 # DOCKER_HUB_NAMESPACE: docker hub namespace.
@@ -8,7 +9,7 @@ name: Build Verilator Container

 on:
  push:
-    tags: [ 'v*' ]
+    tags: ['v*']
  workflow_dispatch:
    inputs:
      manual_tag:
@@ -34,52 +35,52 @@ jobs:
          # - "ci/docker/buildenv:verilator-buildenv"

    steps:
-    - name: Checkout
-      uses: actions/checkout@v4
+      - name: Checkout
+        uses: actions/checkout@v4

-    - name: Extract context variables
-      run: |
-        echo "${{ matrix.contexts }}" | sed -r 's/(.*):.*/build_context=\1/' >> "$GITHUB_ENV"
-        echo "${{ matrix.contexts }}" | sed -r 's/.*:(.*)/image_name=\1/' >> "$GITHUB_ENV"
-        echo "git_tag=${GITHUB_REF#refs/*/}" >> "$GITHUB_ENV"
+      - name: Extract context variables
+        run: |
+          echo "${{ matrix.contexts }}" | sed -r 's/(.*):.*/build_context=\1/' >> "$GITHUB_ENV"
+          echo "${{ matrix.contexts }}" | sed -r 's/.*:(.*)/image_name=\1/' >> "$GITHUB_ENV"
+          echo "git_tag=${GITHUB_REF#refs/*/}" >> "$GITHUB_ENV"

-    - name: Use manual tag
-      if: ${{ inputs.manual_tag }}
-      run: |
-        echo "git_tag=${{ inputs.manual_tag }}" >> "$GITHUB_ENV"
+      - name: Use manual tag
+        if: ${{ inputs.manual_tag }}
+        run: |
+          echo "git_tag=${{ inputs.manual_tag }}" >> "$GITHUB_ENV"

-    - name: Docker meta
-      id: docker_meta
-      uses: docker/metadata-action@v4
-      with:
-        images: |
-          ${{ vars.DOCKER_HUB_NAMESPACE }}/${{ env.image_name }}
-        tags: |
-          type=match,pattern=(v.*),group=1,enable=${{ startsWith(github.ref, 'refs/tags/v') }}
-          type=raw,value=${{ inputs.manual_tag }},enable=${{ inputs.manual_tag != '' }}
-          type=raw,value=latest,enable=${{ inputs.add_latest_tag == true }}
+      - name: Docker meta
+        id: docker_meta
+        uses: docker/metadata-action@v4
+        with:
+          images: |
+            ${{ vars.DOCKER_HUB_NAMESPACE }}/${{ env.image_name }}
+          tags: |
+            type=match,pattern=(v.*),group=1,enable=${{ startsWith(github.ref, 'refs/tags/v') }}
+            type=raw,value=${{ inputs.manual_tag }},enable=${{ inputs.manual_tag != '' }}
+            type=raw,value=latest,enable=${{ inputs.add_latest_tag == true }}

-    - name: Set up QEMU
-      uses: docker/setup-qemu-action@v2
+      - name: Set up QEMU
+        uses: docker/setup-qemu-action@v2

-    - name: Set up Docker Buildx
-      uses: docker/setup-buildx-action@v2
-      with:
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v2
+        with:
          buildkitd-flags: --debug

-    - name: Login to Docker Hub
-      uses: docker/login-action@v2
-      with:
-        username: ${{ secrets.DOCKER_HUB_USER }}
-        password: ${{ secrets.DOCKER_HUB_ACCESS_TOKEN }}
+      - name: Login to Docker Hub
+        uses: docker/login-action@v2
+        with:
+          username: ${{ secrets.DOCKER_HUB_USER }}
+          password: ${{ secrets.DOCKER_HUB_ACCESS_TOKEN }}

-    - name: Build and Push to Docker
-      uses: docker/build-push-action@v4
-      if: startsWith(github.ref, 'refs/tags/v')
-      with:
-        context: ${{ env.build_context }}
-        build-args: SOURCE_COMMIT=${{ env.git_tag }}
-        platforms: linux/arm64,linux/amd64
-        push: ${{ !env.ACT && startsWith(github.ref, 'refs/tags/v') }}
-        tags: ${{ steps.docker_meta.outputs.tags }}
-        labels: ${{ steps.docker_meta.outputs.labels }}
+      - name: Build and Push to Docker
+        uses: docker/build-push-action@v4
+        if: startsWith(github.ref, 'refs/tags/v')
+        with:
+          context: ${{ env.build_context }}
+          build-args: SOURCE_COMMIT=${{ env.git_tag }}
+          platforms: linux/arm64,linux/amd64
+          push: ${{ !env.ACT && startsWith(github.ref, 'refs/tags/v') }}
+          tags: ${{ steps.docker_meta.outputs.tags }}
+          labels: ${{ steps.docker_meta.outputs.labels }}
@@ -1,13 +1,12 @@
+---
 # DESCRIPTION: Github actions config
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
-
+#
 name: format
-
 on:
  push:
  pull_request_target:
  workflow_dispatch:
-
 jobs:
  format:
    runs-on: ubuntu-22.04
@@ -1,55 +1,50 @@
+---
 # DESCRIPTION: Github actions config
 # This name is key to badges in README.rst, so we use the name build
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
-
+#
 name: msbuild
-
 on:
  push:
  pull_request:
  workflow_dispatch:
  schedule:
-  - cron: '0 0 * * 0' # weekly
-
+    - cron: 0 0 * * 0  # weekly
 env:
  CI_OS_NAME: win
  CI_COMMIT: ${{ github.sha }}
  CCACHE_COMPRESS: 1
  CCACHE_DIR: ${{ github.workspace }}/.ccache
  CCACHE_LIMIT_MULTIPLE: 0.95
-
 defaults:
  run:
    working-directory: repo
-
 concurrency:
  group: ${{ github.workflow }}-${{ github.event_name == 'pull_request' && github.ref || github.run_id }}
  cancel-in-progress: true
-
 jobs:
-
-   windows:
+  windows:
    name: run on windows
    runs-on: windows-latest
    steps:
-    - uses: actions/checkout@v4
-      with:
-        path: repo
-    - name: Cache $CCACHE_DIR
-      uses: actions/cache@v4
-      with:
-        path: ${{ env.CCACHE_DIR }}
-        key: msbuild-msvc-cmake
-    - name: compile
-      env:
-        WIN_FLEX_BISON: ${{ github.workspace }}/.ccache
-      run: ./ci/ci-win-compile.ps1
-    - name: test build
-      run: ./ci/ci-win-test.ps1
-    - name: Zip up repository
-      run: Compress-Archive -LiteralPath install -DestinationPath verilator.zip
-    - name: Upload zip archive
-      uses: actions/upload-artifact@v4
-      with:
-        path: ${{ github.workspace }}/repo/verilator.zip
-        name: verilator-win.zip
+      - uses: actions/checkout@v4
+        with:
+          path: repo
+      - name: Cache $CCACHE_DIR
+        uses: actions/cache@v4
+        with:
+          path: ${{ env.CCACHE_DIR }}
+          key: msbuild-msvc-cmake
+      - name: compile
+        env:
+          WIN_FLEX_BISON: ${{ github.workspace }}/.ccache
+        run: ./ci/ci-win-compile.ps1
+      - name: test build
+        run: ./ci/ci-win-test.ps1
+      - name: Zip up repository
+        run: Compress-Archive -LiteralPath install -DestinationPath verilator.zip
+      - name: Upload zip archive
+        uses: actions/upload-artifact@v4
+        with:
+          path: ${{ github.workspace }}/repo/verilator.zip
+          name: verilator-win.zip
@@ -29,14 +29,15 @@ dddrun*
 doxygen-doc
 gdbrun*
 gmon.out
-ncverilog.history
 internals.txt
+ncverilog.history
+nohup.out
+verilator-config-version.cmake
+verilator-config.cmake
+verilator.pc
 verilator.txt
 verilator_bin*
 verilator_coverage_bin*
-verilator.pc
-verilator-config.cmake
-verilator-config-version.cmake
 **/__pycache__/*
 **/_build/*
 **/obj_dir/*
@@ -46,3 +47,6 @@ verilator-config-version.cmake
 /test_regress/snapshot/
 xmverilog.*
 xrun.history
+
+# Normal CMake build directory
+/build
@@ -0,0 +1,8 @@
+---
+- id: verilator
+  name: verilator-lint
+  description: Runs verilator Docker image to lint (System) Verilog designs
+  args: [--lint-only]
+  language: docker_image
+  entry: verilator/verilator:latest
+  types_or: [verilog, system-verilog]
@@ -0,0 +1,4 @@
+[style]
+based_on_style = pep8
+column_limit = 99
+#split_before_arithmetic_operator=True
@@ -0,0 +1,27 @@
+# See https://citation-file-format.github.io/
+cff-version: 1.2.0
+title: Verilator
+message: >-
+  If you use this software, please cite it using the
+  metadata from this file.
+type: software
+authors:
+  - given-names: Wilson
+    family-names: Snyder
+    email: [email protected]
+    affiliation: Veripool
+  - given-names: Paul
+    family-names: Wasson
+  - given-names: Duane
+    family-names: Galbi
+  - name: 'et al'
+repository-code: 'https://github.com/verilator/verilator'
+url: 'https://verilator.org'
+abstract: >-
+  The Verilator package converts Verilog and SystemVerilog hardware
+  description language (HDL) designs into a fast C++ or SystemC model
+  that, after compiling, can be executed.  Verilator is not a
+  traditional simulator but a compiler.
+license:
+  - LGPL-3.0-only
+  - Artistic-2.0
@@ -14,36 +14,44 @@

 cmake_minimum_required(VERSION 3.15)
 cmake_policy(SET CMP0091 NEW) # Use MSVC_RUNTIME_LIBRARY to select the runtime
-project(Verilator
-    VERSION 5.024
+project(
+    Verilator
+    VERSION 5.030
    HOMEPAGE_URL https://verilator.org
    LANGUAGES CXX
 )

-option(DEBUG_AND_RELEASE_AND_COVERAGE
-    "Builds both the debug and release binaries, overriding CMAKE_BUILD_TYPE. Not supported under MSBuild.")
+option(
+    DEBUG_AND_RELEASE_AND_COVERAGE
+    "Builds both the debug and release binaries, overriding CMAKE_BUILD_TYPE. Not supported under MSBuild."
+)

 find_package(Python3 COMPONENTS Interpreter)
-set(PYTHON3 Python3::Interpreter)
-set(CMAKE_INSTALL_DATADIR ${CMAKE_INSTALL_PREFIX})
+set(PYTHON3 ${Python3_EXECUTABLE})
+# See also CMake built-in; CMAKE_INSTALL_PREFIX is applied by the install command.
+set(CMAKE_INSTALL_DATADIR .)
 include(GNUInstallDirs)
 include(CMakePackageConfigHelpers)
 include(CheckStructHasMember)
 include(ExternalProject)
+include(FindThreads)

-if (NOT WIN32)
+if(NOT WIN32)
    message(WARNING "CMake support on Linux/OSX is experimental.")
 endif()

-if (WIN32)
-    if (DEFINED ENV{WIN_FLEX_BISON})
+if(WIN32)
+    if(DEFINED ENV{WIN_FLEX_BISON})
        set(WIN_FLEX_BISON "$ENV{WIN_FLEX_BISON}")
    endif()
-    if (EXISTS ${WIN_FLEX_BISON})
+    if(EXISTS ${WIN_FLEX_BISON})
        list(APPEND CMAKE_PREFIX_PATH ${WIN_FLEX_BISON})
    endif()
-    if (NOT WIN_FLEX_BISON)
-        message(FATAL_ERROR "Please install https://github.com/lexxmark/winflexbison and set WIN_FLEX_BISON environment variable. Please use install cmake target after a successful build.")
+    if(NOT WIN_FLEX_BISON)
+        message(
+            FATAL_ERROR
+            "Please install https://github.com/lexxmark/winflexbison and set WIN_FLEX_BISON environment variable. Please use install cmake target after a successful build."
+        )
    endif()
    set(CMAKE_CXX_STANDARD 20)
 endif()
@@ -51,20 +59,25 @@ endif()
 set(OBJCACHE "" CACHE STRING "Path for ccache, auto-detected if empty")
 option(OBJCACHE_ENABLED "Compile Verilator with ccache" ON)

-if (OBJCACHE_ENABLED)
-    if (OBJCACHE STREQUAL "")
+if(OBJCACHE_ENABLED)
+    if(OBJCACHE STREQUAL "")
        find_program(OBJCACHE_PATH ccache)
-        if (OBJCACHE_PATH STREQUAL "OBJCACHE_PATH-NOTFOUND")
+        if(OBJCACHE_PATH STREQUAL "OBJCACHE_PATH-NOTFOUND")
            set(OBJCACHE_PATH "")
        endif()
    else()
        set(OBJCACHE_PATH "${OBJCACHE}")
    endif()
-    if (NOT OBJCACHE_PATH STREQUAL "")
-        execute_process(COMMAND "${OBJCACHE_PATH}" --version
-                OUTPUT_VARIABLE objcache_version)
+    if(NOT OBJCACHE_PATH STREQUAL "")
+        execute_process(
+            COMMAND "${OBJCACHE_PATH}" --version
+            OUTPUT_VARIABLE objcache_version
+        )
        string(REGEX MATCH "[^\n\r]+" objcache_version "${objcache_version}")
-        message(STATUS "Found ccache: ${OBJCACHE_PATH} (\"${objcache_version}\")")
+        message(
+            STATUS
+            "Found ccache: ${OBJCACHE_PATH} (\"${objcache_version}\")"
+        )
        set(CMAKE_CXX_COMPILER_LAUNCHER "${OBJCACHE_PATH}")
    endif()
 endif()
@@ -75,8 +88,8 @@ find_package(FLEX)
 # Build
 #set_property(GLOBAL PROPERTY JOB_POOLS one_job=1)

-if (DEBUG_AND_RELEASE_AND_COVERAGE)
-    if (CMAKE_GENERATOR MATCHES "^Visual Studio ")
+if(DEBUG_AND_RELEASE_AND_COVERAGE)
+    if(CMAKE_GENERATOR MATCHES "^Visual Studio ")
        error("%Error: The DEBUG_AND_RELEASE_AND_COVERAGE option is not supported in MSBuild-based builds.")
    endif()
    set(saved_build_type ${CMAKE_BUILD_TYPE})
@@ -95,42 +108,69 @@ endif()

 set(PACKAGE_NAME ${PROJECT_NAME})
 set(PACKAGE_VERSION ${PROJECT_VERSION})
+set(CXX ${CMAKE_CXX_COMPILER})
+set(AR ${CMAKE_AR})

 configure_file(include/verilated_config.h.in include/verilated_config.h @ONLY)
+configure_file(include/verilated.mk.in include/verilated.mk @ONLY)

-install(FILES ${CMAKE_CURRENT_BINARY_DIR}/include/verilated_config.h DESTINATION ${CMAKE_INSTALL_PREFIX}/include)
-
-configure_package_config_file(verilator-config.cmake.in verilator-config.cmake
-    INSTALL_DESTINATION ${CMAKE_INSTALL_PREFIX}
+install(
+    FILES ${CMAKE_CURRENT_BINARY_DIR}/include/verilated_config.h
+    DESTINATION include
+)
+install(
+    FILES ${CMAKE_CURRENT_BINARY_DIR}/include/verilated.mk
+    DESTINATION include
 )

-install(FILES ${CMAKE_CURRENT_BINARY_DIR}/verilator-config.cmake DESTINATION ${CMAKE_INSTALL_PREFIX})
-
-configure_package_config_file(verilator-config-version.cmake.in verilator-config-version.cmake
-    INSTALL_DESTINATION ${CMAKE_INSTALL_PREFIX}
+configure_package_config_file(
+    verilator-config.cmake.in
+    verilator-config.cmake
+    INSTALL_DESTINATION .
 )

-install(FILES ${CMAKE_CURRENT_BINARY_DIR}/verilator-config-version.cmake DESTINATION ${CMAKE_INSTALL_PREFIX})
+install(
+    FILES ${CMAKE_CURRENT_BINARY_DIR}/verilator-config.cmake
+    DESTINATION .
+)

-foreach (program
+configure_package_config_file(
+    verilator-config-version.cmake.in
+    verilator-config-version.cmake
+    INSTALL_DESTINATION .
+)
+
+install(
+    FILES ${CMAKE_CURRENT_BINARY_DIR}/verilator-config-version.cmake
+    DESTINATION .
+)
+
+foreach(
+    program
    verilator
    verilator_gantt
    verilator_ccache_report
    verilator_difftree
    verilator_profcfunc
+    verilator_includer
 )
    install(PROGRAMS bin/${program} TYPE BIN)
 endforeach()

-install(DIRECTORY examples TYPE DATA FILES_MATCHING
+install(
+    DIRECTORY examples
+    TYPE DATA
+    FILES_MATCHING
    PATTERN "examples/*/*.[chv]*"
    PATTERN "examples/*/Makefile*"
    PATTERN "examples/*/CMakeLists.txt"
 )

-install(DIRECTORY include TYPE DATA FILES_MATCHING
+install(
+    DIRECTORY include
+    TYPE DATA
+    FILES_MATCHING
    PATTERN "include/verilated_config.h"
-    PATTERN "include/verilated.mk"
    PATTERN "include/*.[chv]"
    PATTERN "include/*.cpp"
    PATTERN "include/*.sv"
@@ -8,6 +8,255 @@ The changes in each Verilator version are described below.  The
 contributors that suggested a given feature are shown in []. Thanks!


+Verilator 5.030 2024-10-27
+==========================
+
+**Major:**
+
+* Add `-output-groups` to build with concatenated .cpp files (#5257). [Mariusz Glebocki]
+* Self-tests have been converted to Python, run `{test_name}.py` instead of `{test_name}.pl`.
+
+**Minor:**
+
+* Change .vlt config files to be read before .v files (#5185). [David Moberg]
+* Change to use maximum for cover point aggregation (#5402). [Andrew Nolte]
+* Change `--main` and `--binary` to use a TOP hierarchy name of "" (#5482).
+* Change install of public executables into bindir instead of pkgdatadir (#5140) (#5544). [Geza Lore]
+* Support IEEE-compliant intra-assign delays (#3711) (#5441). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support `wor`, `trior`, `wand`, `triand` (#5386) (#5496). [Zhou Shen]
+* Support unconstrained randomization for unions (#5395) (#5396). [Yilou Wang]
+* Support basic constrained queue randomization (#5413). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support packed/unpacked and dynamic array unconstrained randomization (#5414) (#5415). [Yilou Wang]
+* Support appending to queue via `[]` (#5421). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support named event locals (#5422). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support basic `dist` constraints (#5431). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support unpacked array constrained randomization (#5437) (#5489). [Yilou Wang]
+* Support inside array constraints (#5448). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support DPI imports and exports with double underscores (#5481).
+* Support ccache when compiling Verilated files with cmake.
+* Support `local` and `protected` on `typedef` (#5460).
+* Support unconstrained randomization for associative array and queue (#5515). [Yilou Wang]
+* Support `rand` dynamic arrays of objects (#5557) (#5564). [Ryszard Rozak, Antmicro Ltd.]
+* Add error on misused genvar (#408). [Alex Solomatnikov]
+* Add error on instances without parenthesis.
+* Add Docker pre-commit hook (#5238) (#5452). [Chris Bachhuber]
+* Add partial coverage symbol and branch data in lcov info files (#5388). [Andrew Nolte]
+* Add method to check if there are VPI callbacks of the given type (#5399). [Kaleb Barrett]
+* Remove warning on unsized numbers exceeding 32-bits.
+* Improve Verilation thread pool (#5161). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Improve performance of V3VariableOrder with parallelism (#5406). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Improve parser error handling (#5493). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Improve process trigger performance (#5483). [Geza Lore]
+* Fix suppression of WIDTH* warnings when immediately under a size cast (#3417).
+* Fix `$fatal` to not be affected by `+verilator+error+limit` (#5135). [Gökçe Aydos]
+* Fix equivalence checking when replacing type parameters (#5213) (#5255). [Han Qi]
+* Fix display with multiple string formats (#5311). [Luiza de Melo]
+* Fix performance of V3Trace when many activity blocks (#5372). [Deniz Güzel]
+* Fix REALCVT warning on integral timescale conversions (#5378). [Liam Braun]
+* Fix multidimensional function return value selects (#5382). [Gökçe Aydos]
+* Fix internal error in out-of-range select (#5393) (#5443). [Geza Lore]
+* Fix dot fallback finding wrong symbols (#5394). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix infinite recursion due to recursive functions/tasks (#5398). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix V3Randomize compile error on old GCC (#5403) (#5417). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix extra events in traces (#5405).
+* Fix empty `foreach` in `if` in constraints (#5408). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix queue `[$-i]` select as reference argument (#5411). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix `pre`/`post_randomize` on `randomize() with` (#5412). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix capturing params in `randomize() with` (#5416) (#5418). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix `sformatf` internal error on initial automatics (#5423). [Todd Strader]
+* Fix clearing trigger of events with no sensitivity trees (#5426). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix driving clocking block in reactive region (#5430). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix associative array next/prev/first/last mis-propagating constants (#5435). [Ethan Sifferman]
+* Fix randomize treated as std::randomize in classes (#5436). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix `foreach` colliding index names (#5444). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix fault on defparam with UNSUPPORTED ignored (#5450). [Luiza de Melo]
+* Fix class reference with pin that is a class reference (#5454).
+* Fix not reporting class reference with extra parameters (#5467).
+* Fix user-type parameter overlap (#5469). [Todd Strader]
+* Fix tracing when name() is empty (#5470). [Sam Shahrestani]
+* Fix timing mode not exiting on empty events (#5472).
+* Fix coverage counts missing due to table optimization (#5473) (#5474). [Vito Gamberini]
+* Fix `--binary` with .cpp PLI filenames under relative directory paths.
+* Fix extra dot in coverage point hierarchy when using name()=''.
+* Fix short-circuiting with associative array access (#5484). [Ethan Sifferman]
+* Fix short-circuiting on method calls (#5486). [Ethan Sifferman]
+* Fix exponential concatenate performance (#5488). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix V3Table trying to generate 'x' bits in the lookup table. (#5491). [Geza Lore]
+* Fix randomize with foreach constraints (#5492). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix explicit CMAKE_INSTALL_PREFIX usages (#5500). [Fabian Keßler]
+* Fix configure inserting absolute paths for Python and Perl (#5504) (#5505). [Nathan Graybeal]
+* Fix pattern initialization with typedef key (#5512). [Eugene Feinberg]
+* Fix `-j` option without argument in hierarchical Verilation (#5514). [Ryszard Rozak, Antmicro Ltd.]
+* Fix `foreach` with 2-D queues and dynamic arrays (#5525) (#5529). [Yilou Wang]
+* Fix struct array assignment (#5455) (#5537). [Yilou Wang]
+* Fix copy constructor of classes that use std::process (#5528). [Ryszard Rozak, Antmicro Ltd.]
+* Fix foreach on associative array (#5530). [Yilou Wang]
+* Fix multi-range indices assignment (#5534) (#5547). [Yilou Wang]
+* Fix static function wrappers (#5536). [Ryszard Rozak, Antmicro Ltd.]
+* Fix assignments of concatenation to queues and dynamic arrays (#5540). [Ryszard Rozak, Antmicro Ltd.]
+* Fix container reduction methods (#5542). [Krzysztof Boroński]
+* Fix complex user type problem with `--x-assign` (#5543). [Todd Strader]
+* Fix long module names crashing string handling (#5546). [Filip Badáň]
+* Fix array trace splitting (#5549). [Todd Strader]
+* Fix queue element access (#5551). [Ryszard Rozak, Antmicro Ltd.]
+* Fix struct literal on pattern assignment (#5552) (#5559). [Todd Strader]
+* Fix build on gcc when using the Spack wrapper (#5555). [Eric Müller]
+* Fix enum name method (#5563). [Todd Strader]
+* Fix `$countbits` in assert with non-tristates (#5566). [Shou-Li Hsu]
+
+
+Verilator 5.028 2024-08-21
+==========================
+
+**Minor:**
+
+* Support state-dependent constraints (#5217). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support cross-module clocking variable access (#5184). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support inline constraints for class randomization methods (#5234). [Krzysztof Boroński]
+* Support clocking blocks in virtual interfaces (#5235). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support `$assertcontrol` assertion_type (#5236). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Support conditional constraints (#5245). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support`--compiler-include` headers in user-supplied cpp files  (#5271). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Support `rand_mode` (#5273). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support `this.randomize with` (#5282). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support foreach constraints (#5302). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support `parameter type` in hierarchical blocks (#5309) (#5333). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Support assertcontrol directive type  (#5310). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Support inline random variable control (#5317). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support streaming operator on arrays and wide data (#5326). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support streams to/from arrays of wide data (#5334). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support `constraint_mode` (#5338). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support constraining AstSel (#5344). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support default value on module input (#5358) (#5373). [Drew Ranck]
+* Add `--compiler-include` for additional C++ includes (#5139) (#5202). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Add `--emit-accessors` (#5182) (#5227). [Ryan Ziegler]
+* Add suggestions on misspelled PLI functions.
+* Add warning on dist in constraints (#5264). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Add more `rand_mode` unsupported errors (#5329). [Krzysztof Bieganski, Antmicro Ltd.]
+* Add parsing but otherwise ignore std::randomize (#5354). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Add Verilated cc define when `--timing` used (#5383). [Kaleb Barrett]
+* Improve emitted code to use a reference for VlSelf (#5254). [Yangyu Chen]
+* Fix monitor block sensitivity items (#4400) (#5294). [Udaya Raj Subedi]
+* Fix fusing macro arguments to not ignore whitespace (#5061). [Tudor Timi]
+* Fix optimized-out sensitivity trees with `--timing` (#5080) (#5349). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix classes/modules of case-similar names (#5109). [Arkadiusz Kozdra]
+* Fix mis-removing $value$plusargs calls (#5127) (#5137). [Seth Pellegrino]
+* Fix incorrect result of width mismatch (#5186) (#5189). [Yutetsu TAKATSUKASA]
+* Fix compiler coroutine check (#5190) (#5300). [Ricardo Barbedo]
+* Fix shortened module names when searching for files (#5196) (#5246). [Tim Hutt]
+* Fix `--x-assign` to be independent from `+verilator+rand+reset` (#5214). [Andrew Nolte]
+* Fix splitting if statements with impure conditions (#5219). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Fix unknown conversion on queues (#5220). [Alex Solomatnikov]
+* Fix top-level unpacked structure resets (#5221).
+* Fix concurrency for mailbox and semaphores (#5222). [Liam Braun]
+* Fix forks capturing non-input ports in tasks (#5237) (#5343). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix toggle coverage aggregation on same line (#5248). [Krzysztof Obłonczek]
+* Fix error on empty generate with -O0 (#5250).
+* Fix unconstrained randomization of unpacked structs (#5252). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix inlining of variables driven from forced vars (#5259). [Geza Lore]
+* Fix tracing with `--main-top-name -` (#5261). [Ethan Sifferman]
+* Fix randomization when used with inheritance (#5268). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix inline constraints creating class random generator (#5280). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix WIDTHEXPAND on left shift of intuitive amount (#5284). [Greg Taylor]
+* Fix elaborating foreach loops (#5285). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix initializing static array in dynamic arrays and queues (#5287). [Baruch Sterin]
+* Fix static variable initializers in procedures (#5296). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Fix randomizing current object with `rand` class instance member (#5292). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix handling of rand fields not referenced in constraints (#5305). [Ryszard Rozak, Antmicro Ltd.]
+* Fix Python3 path discovery in make flows to avoid mixing system and user python interpreters (#5307) [Markus Krause]
+* Fix make flows to pass PYTHON3 (like PERL) (#5307) (#5308). [Markus Krause]
+* Fix assert on wide expression (#5319) (#5324). [Varun Koyyalagunta]
+* Fix output clock variable overwriting signal (#5320)  (#5347). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix stringify in nested preprocessor macros (#5323). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix $sformat with array arguments (#5330). [Abe Jordan]
+* Fix -Wunused-but-set-variable clang warning (#5331). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Fix purity of functions with AstJumpBlock or AstStmtExpr (#5332). [Ryszard Rozak, Antmicro Ltd.]
+* Fix compilation error on unreachable disable fork / wait fork (#5339). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix missing type coercion in 'inside {array}' (#5340). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix +: and -: unpacked array slicing when array has nonzero low index (#5345) (#5387). [James Bailey]
+* Fix tracing_{on,off} in the presence of non-inlined modules (#5346). [Geza Lore]
+* Fix NBAs in suspendables (#5348). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix lint_off on Errors (#5351) (#5363). [Ethan Sifferman]
+* Fix cache config file resolution performance (#5369). [Geza Lore]
+* Fix capturing fields from superclass in `randomize() with` (#5389). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix virtual interface null checks (#5391). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix ==? and !=? with X values.
+* Fix CPU time being zero.
+* Fix inline function ref port persistence.
+
+
+Verilator 5.026 2024-06-15
+==========================
+
+**Major:**
+
+* Support constrained randomization with external solvers (#4947). [Arkadiusz Kozdra, Antmicro Ltd.]
+
+**Minor:**
+
+* Support `$psprintf` system function (#4314) (#5169). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support 2D dynamic array initialization (#4700) (#5122). [Valentin Atepalikhin]
+* Support __en/__out signals on top level inout ports (#4812) (#4856). [Paul Wright]
+* Support empty queue as dynarray default value (#5055). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support vpiInertialDelay (#5087). [Todd Strader]
+* Support NBAs to arrays inside loops (#5092). [Geza Lore]
+* Support parsing and otherwise ignoring inline constraints (#5126). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support `inout` clocking items (#5160). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Support StructSel in unpacked array assignments (#5176). [Geza Lore]
+* Add error on zero width select (#5028).
+* Add CITATION.cff (#5057) (#5058). [Gijs Burghoorn]
+* Add VPI eval needed tracking (#5065). [Todd Strader]
+* Add `--localize-max-size` option and optimization (#5072).
+* Add parameterless assert control system tasks (#5010). [Bartłomiej Chmiel]
+* Add traceCapable indication to model header (#5053). [Vito Gamberini]
+* Add increasing of stack size when possible (#5071) (#5104). [Yinan Xu]
+* Add assertion on reusing VerilatedContext (#5167).
+* Add `--pins-sc-uint-bool` to force SystemC uint type (#5192). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Improve DFG regularization in cyclic graphs (#5142). [Geza Lore]
+* Improve VerilatedVpiPutHolder storage requirements (#5144). [Kaleb Barrett]
+* Fix coroutines without awaits to have a co_return (#4208) (#5175). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix missing flex include path variable (#4970) (#4971). [Christopher Taylor]
+* Fix missing parameters with comma to be errors (#4979) (#5012). [Paul Swirhun]
+* Fix 'experimental/coroutine' file not found on MacOS (#5030) (#5031) (#5151). [Paul Bowen-Huggett]
+* Fix bound queue printing (#5032). [Aleksander Kiryk, Antmicro Ltd.]
+* Fix consecutive zero-delays (#5038). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix attempted to destroy locked thread pool error (#5040). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Fix `$system` with string argument (#5042).
+* Fix width extension on delays (#5043).
+* Fix `$typename` on array.min and others (#5049). [Gökçe Aydos]
+* Fix `make $(info)` which cannot be silenced (#5059). [Gökçe Aydos]
+* Fix CMake builds to export VERILATOR_ROOT (#5063). [Michael Bikovitsky]
+* Fix false ASSIGNIN on functions with explicit port map (#5069).
+* Fix 4-state value support for $readmem (#5070) (#5078). [Ethan Sifferman]
+* Fix DFG assertion with SystemC (#5076). [Geza Lore]
+* Fix `$typename` string to be more standard (#5082) (#5083). [Andrew Nolte]
+* Fix missed optimization in V3Delayed (#5089). [Geza Lore]
+* Fix macro expansion in strings per 1800-2023 (#5094). [Geza Lore]
+* Fix width extension of unpacked array select (#5095). [Varun Koyyalagunta]
+* Fix MacOS missing <type_traits> header (#5096) (#5097). [Vito Gamberini]
+* Fix assertion failure in V3Gate (#5101). [Yutetsu TAKATSUKASA]
+* Fix aliases for forced port signals (#5105). [Geza Lore]
+* Fix tracing interface functions (#5108). [Alex Solomatnikov]
+* Fix method calls parsing in constraints (#5110). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix vpiInertialDelay for memories (#5113). [Todd Strader]
+* Fix hierarchical compilation with nested -F (#5114) (#5124). [Alex Solomatnikov]
+* Fix references to ports in forks (#5123). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix output C++ type error on change detect of I/O arrays (#5125) (#5131). [Pawel Jewstafjew]
+* Fix x-valued parameters with `--x-assign unique` (#5129). [Ethan Sifferman]
+* Fix overflow of string on VPI reads (#5145) (#5146). [Kaleb Barrett]
+* Fix VerilatedVpiPutHolder class (#5156). [Kaleb Barrett]
+* Fix extending out-of-range select (#5159) (#5164). [Geza Lore]
+* Fix radix in width warnings (#5166). [Geza Lore]
+* Fix SystemC BITS_PER_DIGIT in VL_ASSIGN_SBW (#5170). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Fix non-constant replication in concats (#5171). [Arkadiusz Kozdra, Antmicro Ltd.]
+* Fix table optimization when applied on real data type (#5172) (#5173). [Arthur Rosa]
+* Fix signed types emitted in hierarchical Verilation (#5178). [Bartłomiej Chmiel, Antmicro Ltd.]
+* Fix DPI import of null C-string (#5179).
+* Fix CMake installation missing verilated.mk (#5187) (#5188). [Philip Axer]
+* Fix linking with pthreads on CMake (#5194). [Tim Hutt]
+* Fix clang-17 coroutines configuration with -std=gnu++20 (#5200). [Gus Smith]
+
+
 Verilator 5.024 2024-04-05
 ==========================

@@ -56,9 +305,10 @@ Verilator 5.024 2024-04-05
 * Fix preprocessor to respect strings in joins (#5007).
 * Fix tracing class parameters (#5014).
 * Fix memory leaks (#5016). [Geza Lore]
-* Fix $readmem with missing newline (#5019). [Josse Van Delm]
+* Fix `$readmem` with missing newline (#5019). [Josse Van Delm]
 * Fix internal error on missing pattern key (#5023).
 * Fix tracing replicated hierarchical models (#5027).
+* Fix false LIFETIME warning on `repeat` in `fork-join` (#5456).


 Verilator 5.022 2024-02-24
@@ -180,6 +430,7 @@ Verilator 5.020 2024-01-01
 * Fix dynamic NBA conditions (#4773). [Krzysztof Bieganski, Antmicro Ltd.]
 * Fix `V3Fork` stage to run only if `--timing` is set (#4778). [Krzysztof Bieganski, Antmicro Ltd.]
 * Fix max multiply width and add runtime assertions if too small. (#4781)
+* Fix select value too wide (#5148) (#5153). [Dercury]


 Verilator 5.018 2023-10-30
@@ -396,7 +647,7 @@ Verilator 5.012 2023-06-13
 * Fix wide structure VL_TOSTRING_W generation (#4188) (#4189). [Aylon Chaim Porat]
 * Fix references to members of parameterized base classes (#4196). [Ryszard Rozak, Antmicro Ltd]
 * Fix tracing undefined alignment (#4201) (#4288) [John Wehle]
-* Fix class specific same methods for AstVarScope, AstVar, and AstScope (#4203) (#4250). [John Wehle]
+* Fix class-specific same methods for AstVarScope, AstVar, and AstScope (#4203) (#4250). [John Wehle]
 * Fix dotted references in parameterized classes (#4206). [Ryszard Rozak, Antmicro Ltd]
 * Fix bit selections under parameterized classes (#4210). [Ryszard Rozak, Antmicro Ltd]
 * Fix duplicate std:: declaration with -I (#4215). [Harald Pretl]
@@ -552,7 +803,7 @@ Verilator 5.006 2023-01-22
 **Minor:**

 * Support clocking blocks (#3674). [Krzysztof Bieganski, Antmicro Ltd]
-* Support packed structs (#3802). [Aleksander Kiryk, Antmicro Ltd]
+* Support unpacked structs (#3802). [Aleksander Kiryk, Antmicro Ltd]
 * Support Windows-native builds using cmake (#3814). [Kritik Bhimani]
 * Support p format for UnpackArray (#3877). [Aleksander Kiryk, Antmicro Ltd]
 * Support property calls without parenthesis (#3879) (#3893). [Ryszard Rozak, Antmicro Ltd]
@@ -1443,6 +1694,7 @@ Verilator 4.026 2020-01-11
 * Fix expand optimization slowing --lint-only. Closes #2091. [Thomas Watts]
 * Fix %{number}s with strings. #2093. [agrobman]
 * Fix shebang breaking some shells. Closes #2067. [zdave]
+* Fix errors on using string in incorrect format (#5240). [John Demme]


 Verilator 4.024 2019-12-08
@@ -52,6 +52,7 @@ INSTALL_PROGRAM = @INSTALL_PROGRAM@
 INSTALL_DATA = @INSTALL_DATA@
 MAKEINFO = makeinfo
 POD2TEXT = pod2text
+PYTHON3 = @PYTHON3@
 MKINSTALLDIRS = $(SHELL) $(srcdir)/src/mkinstalldirs

 # Version (for docs/guide/conf.py)
@@ -76,7 +77,7 @@ datadir = @datadir@
 # Directory in which to install documentation info files.
 infodir = @infodir@

-# Directory in which to install package specific files
+# Directory in which to install package-specific files
 # Generally ${prefix}/share/verilator
 pkgdatadir = @pkgdatadir@

@@ -91,10 +92,12 @@ datarootdir = @datarootdir@
 CFG_WITH_CCWARN = @CFG_WITH_CCWARN@
 CFG_WITH_DEFENV = @CFG_WITH_DEFENV@
 CFG_WITH_LONGTESTS = @CFG_WITH_LONGTESTS@
+CFG_WITH_SOLVER = @CFG_WITH_SOLVER@
 PACKAGE_VERSION = @PACKAGE_VERSION@

 #### End of system configuration section. ####
 ######################################################################
+# Main build targets

 .SUFFIXES:

@@ -145,6 +148,9 @@ verilator_exe verilator_bin$(EXEEXT) verilator_bin_dbg$(EXEEXT) verilator_covera
 	@echo "making verilator in src"
 	$(MAKE) -C src $(OBJCACHE_JOBS)

+######################################################################
+# Tests
+
 .PHONY:msg_test
 msg_test: all_nomsg
 	@echo "Build complete!"
@@ -155,7 +161,7 @@ msg_test: all_nomsg
 .PHONY: test
 ifeq ($(CFG_WITH_LONGTESTS),yes)  # Local... Else don't burden users
 test: smoke-test test_regress
-# examples is part of test_regress's test_regress/t/t_a2_examples.pl
+# examples is part of test_regress's test_regress/t/t_a2_examples.py
 # (because that allows it to run in parallel with other test_regress's)
 else
 test: smoke-test examples
@@ -167,8 +173,8 @@ endif
 	@echo

 smoke-test: all_nomsg
-	test_regress/t/t_a1_first_cc.pl
-	test_regress/t/t_a2_first_sc.pl
+	test_regress/t/t_a1_first_cc.py
+	test_regress/t/t_a2_first_sc.py

 test_regress: all_nomsg
 	$(MAKE) -C test_regress
@@ -182,6 +188,9 @@ examples: all_nomsg
 	  $(MAKE) -C $$p VERILATOR_ROOT=`pwd` || exit 10; \
 	done

+######################################################################
+# Docs
+
 .PHONY: docs
 docs: info

@@ -203,6 +212,20 @@ verilator.html:
 verilator.pdf: Makefile
 	$(MAKE) -C docs verilator.pdf

+TAGFILES=${srcdir}/*/*.cpp ${srcdir}/*/*.h ${srcdir}/*/*.in \
+	${srcdir}/*.in ${srcdir}/*.pod
+
+TAGS: $(TAGFILES)
+	etags $(TAGFILES)
+
+.PHONY: doxygen
+
+doxygen:
+	$(MAKE) -C docs doxygen
+
+######################################################################
+# Install
+
 # Public executables intended to be invoked directly by the user
 # Don't put wildcards in these variables, it might cause an uninstall of other stuff
 VL_INST_PUBLIC_SCRIPT_FILES = verilator \
@@ -243,11 +266,15 @@ mkbindirs:
 installbin: | mkbindirs
 	cd $(srcdir)/bin; \
 	for p in $(VL_INST_PUBLIC_SCRIPT_FILES) ; do \
-	  $(INSTALL_PROGRAM) $$p $(DESTDIR)$(pkgdatadir)/bin/$$p; \
+	  $(INSTALL_PROGRAM) $$p $(DESTDIR)$(bindir)/$$p; \
 	done
+	perl -p -i -e 'use File::Spec;' \
+	           -e' $$path = File::Spec->abs2rel("$(realpath $(DESTDIR)$(pkgdatadir))", "$(realpath $(DESTDIR)$(bindir))");' \
+	           -e 's/my \$$verilator_pkgdatadir_relpath = .*/my \$$verilator_pkgdatadir_relpath = "$$path";/g' \
+	           -- "$(DESTDIR)/$(bindir)/verilator"
 	cd bin; \
 	for p in $(VL_INST_PUBLIC_BIN_FILES) ; do \
-	  $(INSTALL_PROGRAM) $$p $(DESTDIR)$(pkgdatadir)/bin/$$p; \
+	  $(INSTALL_PROGRAM) $$p $(DESTDIR)$(bindir)/$$p; \
 	done
 	cd $(srcdir)/bin; \
 	for p in $(VL_INST_PRIVATE_SCRIPT_FILES) ; do \
@@ -257,11 +284,11 @@ installbin: | mkbindirs
 installredirect: installbin | mkbindirs
 	cp ${srcdir}/bin/redirect ${srcdir}/bin/redirect.tmp
 	perl -p -i -e 'use File::Spec;' \
-	           -e' $$path = File::Spec->abs2rel("$(realpath $(DESTDIR)$(pkgdatadir)/bin)", "$(realpath $(DESTDIR)$(bindir))");' \
+	           -e' $$path = File::Spec->abs2rel("$(realpath $(DESTDIR)$(bindir))", "$(realpath $(DESTDIR)$(pkgdatadir)/bin)");' \
 	           -e 's/RELPATH.*/"$$path";/g' -- "${srcdir}/bin/redirect.tmp"
 	cd $(srcdir)/bin; \
 	for p in $(VL_INST_PUBLIC_SCRIPT_FILES) $(VL_INST_PUBLIC_BIN_FILES) ; do \
-	  $(INSTALL_PROGRAM) redirect.tmp $(DESTDIR)$(bindir)/$$p; \
+	  $(INSTALL_PROGRAM) redirect.tmp $(DESTDIR)$(pkgdatadir)/bin/$$p; \
 	done
 	rm ${srcdir}/bin/redirect.tmp

@@ -343,6 +370,19 @@ install-all: installbin installredirect installman installdata install-msg

 install-here: installman info

+install-msg:
+	@echo
+	@echo "Installed binaries to $(DESTDIR)$(bindir)/verilator"
+	@echo "Installed man to $(DESTDIR)$(mandir)/man1"
+	@echo "Installed examples to $(DESTDIR)$(pkgdatadir)/examples"
+	@echo
+	@echo "For documentation see 'man verilator' or 'verilator --help'"
+	@echo "For forums and to report bugs see https://verilator.org"
+	@echo
+
+######################################################################
+# Format/Lint
+
 # Use --xml flag to see the cppcheck code to use for suppression
 CPPCHECK1_CPP = $(wildcard $(srcdir)/include/*.cpp)
 CPPCHECK2_CPP = $(wildcard $(srcdir)/examples/*/*.cpp)
@@ -410,7 +450,8 @@ analyzer-include:
 	-rm -rf examples/*/obj*
 	scan-build $(MAKE) -k examples

-format: clang-format yapf format-pl-exec
+format:
+	$(MAKE) -j 4 clang-format yapf format-exec

 CLANGFORMAT = clang-format-14
 CLANGFORMAT_FLAGS = -i
@@ -421,6 +462,20 @@ clang-format:
 		|| echo "*** You are not using clang-format-14, indents may differ from master's ***"
 	$(CLANGFORMAT) $(CLANGFORMAT_FLAGS) $(CLANGFORMAT_FILES)

+YAMLFIX = YAMLFIX_WHITELINES=1 YAMLFIX_LINE_LENGTH=130 YAMLFIX_preserve_quotes=true yamlfix
+
+yamlfix:
+	$(YAMLFIX) .
+
+# CMake files
+CMAKE_FILES = \
+	CMakeLists.txt \
+	examples/*/CMakeLists.txt \
+	src/CMakeLists.txt \
+	test_regress/CMakeLists.txt \
+	*.cmake.in \
+
+# Python programs, subject to format and lint
 PY_PROGRAMS = \
 	bin/verilator_ccache_report \
 	bin/verilator_difftree \
@@ -439,6 +494,7 @@ PY_PROGRAMS = \
 	src/flexfix \
 	src/vlcovgen \
 	src/.gdbinit.py \
+	test_regress/*.py \
 	test_regress/t/*.pf \
 	nodist/clang_check_attributes \
 	nodist/code_coverage \
@@ -446,45 +502,55 @@ PY_PROGRAMS = \
 	nodist/fuzzer/actual_fail \
 	nodist/fuzzer/generate_dictionary \
 	nodist/install_test \
+	nodist/log_changes \

+# Python files, subject to format but not lint
 PY_FILES = \
 	$(PY_PROGRAMS) \
 	nodist/code_coverage.dat \
+	test_regress/t/*.py \
+
+# Python files, test_regress tests
+PY_TEST_FILES = \
+	test_regress/t/*.py \

 YAPF = yapf3
-YAPF_FLAGS = -i
+YAPF_FLAGS = -i --parallel

 yapf:
 	$(YAPF) $(YAPF_FLAGS) $(PY_FILES)

+GERSEMI = gersemi
+GERSEMI_FLAGS = -i
+
+format-cmake:
+	$(GERSEMI) $(GERSEMI_FLAGS) $(CMAKE_FILES)
+
 PYLINT = pylint
-PYLINT_FLAGS = --score=n --disable=R0801
+PYLINT_FLAGS = --recursive=n --score=n --disable=R0801
+PYLINT_TEST_FLAGS = $(PYLINT_FLAGS) --disable=C0103,C0114,C0116,C0209,C0411,C0413,C0301,R0801,R0912,R0915,R0916,R1702,W0511,W0621

 RUFF = ruff
 RUFF_FLAGS = check --ignore=E402,E501,E701

 # "make -k" so can see all tool result errors
 lint-py:
-	$(MAKE) -k lint-py-pylint lint-py-ruff
+	$(MAKE) -k lint-py-pylint lint-py-pylint-tests lint-py-ruff

 lint-py-pylint:
 	$(PYLINT) $(PYLINT_FLAGS) $(PY_PROGRAMS)

+lint-py-pylint-tests:
+	$(PYLINT) $(PYLINT_TEST_FLAGS) $(PY_TEST_FILES) | $(PYTHON3) nodist/lint_py_test_filter
+
 lint-py-ruff:
 	$(RUFF) $(RUFF_FLAGS) $(PY_PROGRAMS)

-format-pl-exec:
-	-chmod a+x test_regress/t/*.pl
+format-exec:
+	-chmod a+x test_regress/t/*.py

-install-msg:
-	@echo
-	@echo "Installed binaries to $(DESTDIR)$(bindir)/verilator"
-	@echo "Installed man to $(DESTDIR)$(mandir)/man1"
-	@echo "Installed examples to $(DESTDIR)$(pkgdatadir)/examples"
-	@echo
-	@echo "For documentation see 'man verilator' or 'verilator --help'"
-	@echo "For forums and to report bugs see https://verilator.org"
-	@echo
+######################################################################
+# Configure

 IN_WILD := ${srcdir}/*.in ${srcdir}/*/*.in

@@ -505,6 +571,9 @@ else
 	autoconf
 endif

+######################################################################
+# Clean
+
 maintainer-clean::
 	@echo "This command is intended for maintainers to use;"
 	@echo "rebuilding the deleted files requires autoconf."
@@ -535,17 +604,6 @@ distclean maintainer-clean::
 	rm -f bin/verilator_bin* bin/verilator_coverage_bin*
 	rm -f include/verilated.mk include/verilated_config.h

-TAGFILES=${srcdir}/*/*.cpp ${srcdir}/*/*.h ${srcdir}/*/*.in \
-	${srcdir}/*.in ${srcdir}/*.pod
-
-TAGS: $(TAGFILES)
-	etags $(TAGFILES)
-
-.PHONY: doxygen
-
-doxygen:
-	$(MAKE) -C docs doxygen
-
 ######################################################################
 # Distributions

@@ -1,21 +1,23 @@
 .. Github doesn't render images unless absolute URL
 .. Do not know of a conditional tag, "only: github" nor "github display" works

-.. image:: https://img.shields.io/badge/Website-Verilator.org-181717.svg
+|badge1| |badge2| |badge3| |badge4| |badge5| |badge6| |badge7| |badge8|
+
+.. |badge1| image:: https://img.shields.io/badge/Website-Verilator.org-181717.svg
    :target: https://verilator.org
-.. image:: https://img.shields.io/badge/License-LGPL%20v3-blue.svg
+.. |badge2| image:: https://img.shields.io/badge/License-LGPL%20v3-blue.svg
    :target: https://www.gnu.org/licenses/lgpl-3.0
-.. image:: https://img.shields.io/badge/License-Artistic%202.0-0298c3.svg
+.. |badge3| image:: https://img.shields.io/badge/License-Artistic%202.0-0298c3.svg
    :target: https://opensource.org/licenses/Artistic-2.0
-.. image:: https://repology.org/badge/tiny-repos/verilator.svg?header=distro%20packages
+.. |badge4| image:: https://repology.org/badge/tiny-repos/verilator.svg?header=distro%20packages
    :target: https://repology.org/project/verilator/versions
-.. image:: https://img.shields.io/docker/pulls/verilator/verilator
+.. |badge5| image:: https://img.shields.io/docker/pulls/verilator/verilator
    :target: https://hub.docker.com/r/verilator/verilator
-.. image:: https://api.codacy.com/project/badge/Grade/fa78caa433c84a4ab9049c43e9debc6f
+.. |badge6| image:: https://api.codacy.com/project/badge/Grade/fa78caa433c84a4ab9049c43e9debc6f
    :target: https://www.codacy.com/gh/verilator/verilator
-.. image:: https://codecov.io/gh/verilator/verilator/branch/master/graph/badge.svg
+.. |badge7| image:: https://codecov.io/gh/verilator/verilator/branch/master/graph/badge.svg
    :target: https://codecov.io/gh/verilator/verilator
-.. image:: https://github.com/verilator/verilator/workflows/build/badge.svg
+.. |badge8| image:: https://github.com/verilator/verilator/workflows/build/badge.svg
    :target: https://github.com/verilator/verilator/actions?query=workflow%3Abuild


@@ -85,12 +87,11 @@ thread is about 100 times faster than interpreted Verilog simulators such
 as `Icarus Verilog`_. Another 2-10x speedup might be gained from
 multithreading (yielding 200-1000x total over interpreted simulators).

-Verilator has typically similar or better performance versus
-closed-source Verilog simulators (e.g., Carbon Design Systems Carbonator,
-Modelsim/Questa, Cadence Incisive/NC-Verilog, Synopsys VCS, VTOC, and
-Pragmatic CVer/CVC). But, Verilator is open-sourced, so you can spend on
-computes rather than licenses. Thus, Verilator gives you the best
-simulation cycles/dollar.
+Verilator has typically similar or better performance versus closed-source
+Verilog simulators (e.g., Aldec Riviera-Pro, Cadence Incisive/NC-Verilog,
+Mentor ModelSim/Questa, Synopsys VCS, VTOC, and Pragmatic CVer/CVC). But,
+Verilator is open-sourced, so you can spend on computes rather than
+licenses. Thus, Verilator gives you the best simulation cycles/dollar.


 Installation & Documentation
@@ -69,7 +69,9 @@ if (! GetOptions(
    pod2usage(-exitstatus => 2, -verbose => 0);
 }

-my $verilator_root = realpath("$RealBin/..");
+# WARNING: $verilator_pkgdatadir_relpath is substituted during Verilator 'make install'
+my $verilator_pkgdatadir_relpath = "..";
+my $verilator_root = realpath("$RealBin/$verilator_pkgdatadir_relpath");
 if (defined $ENV{VERILATOR_ROOT}) {
    if ((!-d $ENV{VERILATOR_ROOT}) || $verilator_root ne realpath($ENV{VERILATOR_ROOT})) {
        warn "%Error: verilator: VERILATOR_ROOT is set to inconsistent path. Suggest leaving it unset.\n";
@@ -324,6 +326,7 @@ detailed descriptions of these arguments.
    --clk <signal-name>         Mark specified signal as clock
    --no-clk <signal-name>      Prevent marking specified signal as clock
    --compiler <compiler-name>  Tune for specified C++ compiler
+    --compiler-include          Include additional header in the precompiled one
    --converge-limit <loops>    Tune convergence settle time
    --coverage                  Enable all coverage
    --coverage-line             Enable line coverage
@@ -356,6 +359,7 @@ detailed descriptions of these arguments.
    --dumpi-tree-json <level>   Enable dumping Ast .tree.json files at level
    --dumpi-<srcfile> <level>   Enable dumping everything in source file at level
     -E                         Preprocess, but do not compile
+    --emit-accessors            Emit getter and setter methods for model top class
    --error-limit <value>       Abort after this number of errors
    --exe                       Link to create executable
    --expand-limit <value>      Set expand optimization limit
@@ -375,6 +379,7 @@ detailed descriptions of these arguments.
    --get-supported <feature>   Get if feature is supported
    --help                      Show this help
    --hierarchical              Enable hierarchical Verilation
+    --hierarchical-params-file <name> Internal option that specifies parameters file for hier blocks
     -I<dir>                    Directory to search for includes
    --if-depth <value>          Tune IFDEPTH warning
     +incdir+<dir>              Directory to search for includes
@@ -387,6 +392,7 @@ detailed descriptions of these arguments.
    --lib-create <name>         Create a DPI library
     +libext+<ext>+[ext]...     Extensions for finding modules
    --lint-only                 Lint, but do not make output
+    --localize-max-size <value>  Tune localize optimization variable size
    --make <build-tool>         Generate scripts for specified build tool
     -MAKEFLAGS <flags>         Arguments to pass to make during --build
    --main                      Generate C++ main() file
@@ -401,13 +407,16 @@ detailed descriptions of these arguments.
     -O3                        High-performance optimizations
     -O<optimization-letter>    Selectable optimizations
     -o <executable>            Name of final executable
+    --output-groups <numfiles>  Group .cpp files into larger ones
    --output-split <statements>          Split .cpp files into pieces
    --output-split-cfuncs <statements>   Split model functions
    --output-split-ctrace <statements>   Split tracing functions
     -P                         Disable line numbers and blanks with -E
    --pins-bv <bits>            Specify types for top-level ports
+    --pins-inout-enables        Specify that __en and __out signals be created for inouts
    --pins-sc-biguint           Specify types for top-level ports
    --pins-sc-uint              Specify types for top-level ports
+    --pins-sc-uint-bool         Specify types for top-level ports
    --pins-uint8                Specify types for top-level ports
    --no-pins64                 Don't use uint64_t's for 33-64 bit sigs
    --pipe-filter <command>     Filter all input through a script
@@ -457,8 +466,8 @@ detailed descriptions of these arguments.
    --trace-coverage            Enable tracing of coverage
    --trace-depth <levels>      Depth of tracing
    --trace-fst                 Enable FST waveform creation
-    --trace-max-array <depth>   Maximum bit width for tracing
-    --trace-max-width <width>   Maximum array depth for tracing
+    --trace-max-array <depth>   Maximum array depth for tracing
+    --trace-max-width <width>   Maximum bit width for tracing
    --trace-params              Enable tracing of parameters
    --trace-structs             Enable tracing structure names
    --trace-threads <threads>   Enable FST waveform creation on separate threads
@@ -16,8 +16,7 @@ parser = argparse.ArgumentParser(

    For documentation see
    https://verilator.org/guide/latest/exe_verilator_ccache_report.html""",
-    epilog=
-    """Copyright 2002-2024 by Wilson Snyder. This program is free software; you
+    epilog="""Copyright 2002-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -67,30 +66,23 @@ else:
    wnames = max(len(_) for _ in results) + 1
    wresults = max(len(_) for _ in results.values()) + 1
    for k in sorted(results.keys()):
-        args.o.write("{:{wnames}} : {:{wresults}} : {}s\n".format(
-            k,
-            results[k],
-            elapsed[k].total_seconds(),
-            wnames=wnames,
-            wresults=wresults))
+        args.o.write("{:{wnames}} : {:{wresults}} : {}s\n".format(k,
+                                                                  results[k],
+                                                                  elapsed[k].total_seconds(),
+                                                                  wnames=wnames,
+                                                                  wresults=wresults))

    args.o.write("\nSummary:\n")
    counts = collections.Counter(_ for _ in results.values())
    total = sum(counts.values())
    for k in sorted(counts.keys()):
        c = counts[k]
-        args.o.write("{:{width}}| {} ({:.2%})\n".format(k,
-                                                        c,
-                                                        c / total,
-                                                        width=wresults))
+        args.o.write("{:{width}}| {} ({:.2%})\n".format(k, c, c / total, width=wresults))

    args.o.write("\nLongest:\n")
-    longest = sorted(list(elapsed.items()),
-                     key=lambda kv: -kv[1].total_seconds())
+    longest = sorted(list(elapsed.items()), key=lambda kv: -kv[1].total_seconds())
    for i, (k, v) in enumerate(longest):
-        args.o.write("{:{width}}| {}s\n".format(k,
-                                                v.total_seconds(),
-                                                width=wnames))
+        args.o.write("{:{width}}| {}s\n".format(k, v.total_seconds(), width=wnames))
        if i > 4:
            break

@@ -154,9 +154,9 @@ verilator_coverage - Verilator coverage analyzer

    verilator_coverage --annotate <obj>

-    verilator_coverage  -write merged.dat -read <datafiles>...
+    verilator_coverage  -write merged.dat <datafiles>...

-    verilator_coverage  -write-info merged.info -read <datafiles>...
+    verilator_coverage  -write-info merged.info <datafiles>...

 Verilator_coverage processes Verilated model-generated coverage reports.

@@ -47,8 +47,7 @@ def diff_dir(a, b):
        diff_file(a, b)
        anyfile = True
    if not anyfile:
-        sys.stderr.write(
-            "%Warning: No .tree files found that have similar base names\n")
+        sys.stderr.write("%Warning: No .tree files found that have similar base names\n")


 def diff_file(a, b):
@@ -109,18 +108,14 @@ parser = argparse.ArgumentParser(
 Verilator_difftree is used for debugging Verilator tree output files.
 It performs a diff between two files, or all files common between two
 directories, ignoring irrelevant pointer differences.""",
-    epilog=
-    """Copyright 2005-2024 by Wilson Snyder. This program is free software; you
+    epilog="""Copyright 2005-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.

 SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0""")

-parser.add_argument('--debug',
-                    action='store_const',
-                    const=9,
-                    help='enable debug')
+parser.add_argument('--debug', action='store_const', const=9, help='enable debug')
 parser.add_argument('--no-lineno',
                    action='store_false',
                    help='do not show differences in line numbering')
@@ -15,11 +15,7 @@ LongestVcdStrValueLength = 0
 Threads = collections.defaultdict(lambda: [])  # List of records per thread id
 Mtasks = collections.defaultdict(lambda: {'elapsed': 0, 'end': 0})
 Cpus = collections.defaultdict(lambda: {'mtask_time': 0})
-Global = {
-    'args': {},
-    'cpuinfo': collections.defaultdict(lambda: {}),
-    'stats': {}
-}
+Global = {'args': {}, 'cpuinfo': collections.defaultdict(lambda: {}), 'stats': {}}
 ElapsedTime = None  # total elapsed time
 ExecGraphTime = 0  # total elapsed time excuting an exec graph
 ExecGraphIntervals = []  # list of (start, end) pairs
@@ -31,8 +27,7 @@ def read_data(filename):
    with open(filename, "r", encoding="utf8") as fh:
        re_thread = re.compile(r'^VLPROFTHREAD (\d+)$')
        re_record = re.compile(r'^VLPROFEXEC (\S+) (\d+)(.*)$')
-        re_payload_mtaskBegin = re.compile(
-            r'id (\d+) predictStart (\d+) cpu (\d+)')
+        re_payload_mtaskBegin = re.compile(r'id (\d+) predictStart (\d+) cpu (\d+)')
        re_payload_mtaskEnd = re.compile(r'id (\d+) predictCost (\d+)')

        re_arg1 = re.compile(r'VLPROF arg\s+(\S+)\+([0-9.]*)\s*')
@@ -57,8 +52,7 @@ def read_data(filename):
                tick = int(tick)
                payload = payload.strip()
                if kind == "SECTION_PUSH":
-                    LongestVcdStrValueLength = max(LongestVcdStrValueLength,
-                                                   len(payload))
+                    LongestVcdStrValueLength = max(LongestVcdStrValueLength, len(payload))
                    SectionStack.append(payload)
                    Sections.append((tick, tuple(SectionStack)))
                elif kind == "SECTION_POP":
@@ -66,15 +60,13 @@ def read_data(filename):
                    SectionStack.pop()
                    Sections.append((tick, tuple(SectionStack)))
                elif kind == "MTASK_BEGIN":
-                    mtask, predict_start, ecpu = re_payload_mtaskBegin.match(
-                        payload).groups()
+                    mtask, predict_start, ecpu = re_payload_mtaskBegin.match(payload).groups()
                    mtask = int(mtask)
                    predict_start = int(predict_start)
                    ecpu = int(ecpu)
                    mTaskThread[mtask] = thread
                    records = Threads[thread]
-                    assert not records or records[-1]['start'] <= records[-1][
-                        'end'] <= tick
+                    assert not records or records[-1]['start'] <= records[-1]['end'] <= tick
                    records.append({
                        'start': tick,
                        'mtask': mtask,
@@ -85,8 +77,7 @@ def read_data(filename):
                    Mtasks[mtask]['thread'] = thread
                    Mtasks[mtask]['predict_start'] = predict_start
                elif kind == "MTASK_END":
-                    mtask, predict_cost = re_payload_mtaskEnd.match(
-                        payload).groups()
+                    mtask, predict_cost = re_payload_mtaskEnd.match(payload).groups()
                    mtask = int(mtask)
                    predict_cost = int(predict_cost)
                    begin = Mtasks[mtask]['begin']
@@ -163,8 +154,7 @@ def report():

    print("\nSummary:")
    print("  Total elapsed time = {} rdtsc ticks".format(ElapsedTime))
-    print("  Parallelized code  = {:.2%} of elapsed time".format(
-        ExecGraphTime / ElapsedTime))
+    print("  Parallelized code  = {:.2%} of elapsed time".format(ExecGraphTime / ElapsedTime))
    print("  Total threads      = %d" % nthreads)
    print("  Total CPUs used    = %d" % ncpus)
    print("  Total mtasks       = %d" % len(Mtasks))
@@ -176,15 +166,12 @@ def report():

    if nthreads > ncpus:
        print()
-        print("%%Warning: There were fewer CPUs (%d) than threads (%d)." %
-              (ncpus, nthreads))
+        print("%%Warning: There were fewer CPUs (%d) than threads (%d)." % (ncpus, nthreads))
        print("        : See docs on use of numactl.")
    else:
        if 'cpu_socket_cores_warning' in Global:
            print()
-            print(
-                "%Warning: Multiple threads scheduled on same hyperthreaded core."
-            )
+            print("%Warning: Multiple threads scheduled on same hyperthreaded core.")
            print("        : See docs on use of numactl.")
        if 'cpu_sockets_warning' in Global:
            print()
@@ -228,8 +215,7 @@ def report_mtasks():
    serialTime = ElapsedTime - ExecGraphTime

    def subReport(elapsed, work):
-        print("  Thread utilization = {:7.2%}".format(work /
-                                                      (elapsed * nthreads)))
+        print("  Thread utilization = {:7.2%}".format(work / (elapsed * nthreads)))
        print("  Speedup            = {:6.3}x".format(work / elapsed))

    print("\nParallelized code, measured:")
@@ -256,8 +242,7 @@ def report_mtasks():
        if Mtasks[mtask]['elapsed'] > 0:
            if Mtasks[mtask]['predict_cost'] == 0:
                Mtasks[mtask]['predict_cost'] = 1  # don't log(0) below
-            p2e_ratio = math.log(Mtasks[mtask]['predict_cost'] /
-                                 Mtasks[mtask]['elapsed'])
+            p2e_ratio = math.log(Mtasks[mtask]['predict_cost'] / Mtasks[mtask]['elapsed'])
            p2e_ratios.append(p2e_ratio)

            if p2e_ratio > max_p2e:
@@ -269,18 +254,14 @@ def report_mtasks():

    print("\nMTask statistics:")
    print("  Longest mtask id = {}".format(long_mtask))
-    print("  Longest mtask time = {:.2%} of time elapsed in parallelized code".
-          format(long_mtask_time / ExecGraphTime))
+    print("  Longest mtask time = {:.2%} of time elapsed in parallelized code".format(
+        long_mtask_time / ExecGraphTime))
    print("  min log(p2e) = %0.3f" % min_p2e, end="")

-    print("  from mtask %d (predict %d," %
-          (min_mtask, Mtasks[min_mtask]['predict_cost']),
-          end="")
+    print("  from mtask %d (predict %d," % (min_mtask, Mtasks[min_mtask]['predict_cost']), end="")
    print(" elapsed %d)" % Mtasks[min_mtask]['elapsed'])
    print("  max log(p2e) = %0.3f" % max_p2e, end="")
-    print("  from mtask %d (predict %d," %
-          (max_mtask, Mtasks[max_mtask]['predict_cost']),
-          end="")
+    print("  from mtask %d (predict %d," % (max_mtask, Mtasks[max_mtask]['predict_cost']), end="")
    print(" elapsed %d)" % Mtasks[max_mtask]['elapsed'])

    stddev = statistics.pstdev(p2e_ratios)
@@ -315,8 +296,8 @@ def report_cpus():
                model = cpuinfo['model_name']

        print("  {:3d} | {:7.2%} / {:16d} | {:>6s} | {:>4s} | {}".format(
-            cpu, Cpus[cpu]['mtask_time'] / ElapsedTime,
-            Cpus[cpu]['mtask_time'], socket, core, model))
+            cpu, Cpus[cpu]['mtask_time'] / ElapsedTime, Cpus[cpu]['mtask_time'], socket, core,
+            model))

    if len(Global['cpu_sockets']) > 1:
        Global['cpu_sockets_warning'] = True
@@ -366,8 +347,8 @@ def report_sections():

    def printTree(prefix, name, entries, tree):
        print("  {:7.2%} | {:7.2%} | {:8} | {:10.2f} | {}".format(
-            treeSum(tree) / ElapsedTime, tree[0] / ElapsedTime, tree[2],
-            tree[2] / entries, prefix + name))
+            treeSum(tree) / ElapsedTime, tree[0] / ElapsedTime, tree[2], tree[2] / entries,
+            prefix + name))
        for k in sorted(tree[1], key=lambda _: -treeSum(tree[1][_])):
            printTree(prefix + "  ", k, tree[2], tree[1][k])

@@ -438,10 +419,8 @@ def write_vcd(filename):
                addValue(code, start, mtask)
                addValue(code, end, None)

-        tStart = sorted(_['start'] for records in Threads.values()
-                        for _ in records)
-        tEnd = sorted(_['end'] for records in Threads.values()
-                      for _ in records)
+        tStart = sorted(_['start'] for records in Threads.values() for _ in records)
+        tEnd = sorted(_['end'] for records in Threads.values() for _ in records)

        # Predicted graph
        for start, end in ExecGraphIntervals:
@@ -455,11 +434,10 @@ def write_vcd(filename):
            # Predict mtasks that fill the time the execution occupied
            for mtask in Mtasks:
                thread = Mtasks[mtask]['thread']
-                pred_scaled_start = start + int(
-                    Mtasks[mtask]['predict_start'] * measured_scaling)
+                pred_scaled_start = start + int(Mtasks[mtask]['predict_start'] * measured_scaling)
                pred_scaled_end = start + int(
-                    (Mtasks[mtask]['predict_start'] +
-                     Mtasks[mtask]['predict_cost']) * measured_scaling)
+                    (Mtasks[mtask]['predict_start'] + Mtasks[mtask]['predict_cost']) *
+                    measured_scaling)
                if pred_scaled_start == pred_scaled_end:
                    continue

@@ -545,8 +523,7 @@ Verilator_gantt creates a visual representation to help analyze Verilator

 For documentation see
 https://verilator.org/guide/latest/exe_verilator_gantt.html""",
-    epilog=
-    """Copyright 2018-2024 by Wilson Snyder. This program is free software; you
+    epilog="""Copyright 2018-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -554,12 +531,8 @@ Version 2.0.
 SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0""")

 parser.add_argument('--debug', action='store_true', help='enable debug')
-parser.add_argument('--no-vcd',
-                    help='disable creating vcd',
-                    action='store_true')
-parser.add_argument('--vcd',
-                    help='filename for vcd outpue',
-                    default='profile_exec.vcd')
+parser.add_argument('--no-vcd', help='disable creating vcd', action='store_true')
+parser.add_argument('--vcd', help='filename for vcd outpue', default='profile_exec.vcd')
 parser.add_argument('filename',
                    help='input profile_exec.dat filename to process',
                    default='profile_exec.dat')
@@ -34,9 +34,8 @@ def profcfunc(filename):

            # Older gprofs have no call column for single-call functions
            #                  %time      cumesec   selfsec      {stuff}   name
-            match = re.match(
-                r'^\s*([0-9.]+)\s+[0-9.]+\s+([0-9.]+)\s+[^a-zA-Z_]*([a-zA-Z_].*)$',
-                line)
+            match = re.match(r'^\s*([0-9.]+)\s+[0-9.]+\s+([0-9.]+)\s+[^a-zA-Z_]*([a-zA-Z_].*)$',
+                             line)
            if match:
                pct = float(match.group(1))
                sec = float(match.group(2))
@@ -136,19 +135,15 @@ def profcfunc(filename):

    design_width = 1
    for func, func_item in vfuncs.items():
-        if design_width < len(func_item['design']):
-            design_width = len(func_item['design'])
+        design_width = max(design_width, len(func_item['design']))

    print("Verilog code profile:")
    print("   These are split into three categories:")
    print("      C++:     Time in non-Verilated C++ code")
    print("      Prof:    Time in profile overhead")
-    print("      VBlock:  Time attributable to a block in a" +
-          " Verilog file and line")
-    print("      VCommon: Time in a Verilated module," +
-          " due to all parts of the design")
-    print("      VLib:    Time in Verilated common libraries," +
-          " called by the Verilated code")
+    print("      VBlock:  Time attributable to a block in a" + " Verilog file and line")
+    print("      VCommon: Time in a Verilated module," + " due to all parts of the design")
+    print("      VLib:    Time in Verilated common libraries," + " called by the Verilated code")
    print()

    print("  %   cumulative   self              ")
@@ -156,13 +151,11 @@ def profcfunc(filename):
           "s  type      filename and line number") % "design")

    cume = 0
-    for func in sorted(vfuncs.keys(),
-                       key=lambda f: vfuncs[f]['sec'],
-                       reverse=True):
+    for func in sorted(vfuncs.keys(), key=lambda f: vfuncs[f]['sec'], reverse=True):
        cume += vfuncs[func]['sec']
        print(("%6.2f %9.2f %8.2f %10d   %-" + str(design_width) + "s  %s") %
-              (vfuncs[func]['pct'], cume, vfuncs[func]['sec'],
-               vfuncs[func]['calls'], vfuncs[func]['design'], func))
+              (vfuncs[func]['pct'], cume, vfuncs[func]['sec'], vfuncs[func]['calls'],
+               vfuncs[func]['design'], func))


 ######################################################################
@@ -180,18 +173,14 @@ in each Verilog block.

 For documentation see
 https://verilator.org/guide/latest/exe_verilator_profcfunc.html""",
-    epilog=
-    """Copyright 2002-2024 by Wilson Snyder. This program is free software; you
+    epilog="""Copyright 2002-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.

 SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0""")

-parser.add_argument('--debug',
-                    action='store_const',
-                    const=9,
-                    help='enable debug')
+parser.add_argument('--debug', action='store_const', const=9, help='enable debug')
 parser.add_argument('filename', help='input gprof output to process')

 Args = parser.parse_args()
@@ -60,17 +60,14 @@ if [ "$CI_BUILD_STAGE_NAME" = "build" ]; then
      sudo apt-get install libgoogle-perftools-dev ||
      sudo apt-get install libgoogle-perftools-dev
    fi
-    if [ "$CI_RUNS_ON" = "ubuntu-20.04" ] || [ "$CI_RUNS_ON" = "ubuntu-22.04" ]; then
+    if [ "$CI_RUNS_ON" = "ubuntu-20.04" ] || [ "$CI_RUNS_ON" = "ubuntu-22.04" ] || [ "$CI_RUNS_ON" = "ubuntu-24.04" ]; then
      sudo apt-get install libsystemc libsystemc-dev ||
      sudo apt-get install libsystemc libsystemc-dev
    fi
-    if [ "$CI_RUNS_ON" = "ubuntu-22.04" ]; then
+    if [ "$CI_RUNS_ON" = "ubuntu-22.04" ] || [ "$CI_RUNS_ON" = "ubuntu-24.04" ]; then
      sudo apt-get install bear mold ||
      sudo apt-get install bear mold
    fi
-    if [ "$COVERAGE" = 1 ]; then
-      yes yes | sudo cpan -fi Parallel::Forker
-    fi
  elif [ "$CI_OS_NAME" = "osx" ]; then
    brew update
    brew install ccache perl gperftools
@@ -92,32 +89,28 @@ elif [ "$CI_BUILD_STAGE_NAME" = "test" ]; then
    sudo apt-get update ||
    sudo apt-get update
    # libfl-dev needed for internal coverage's test runs
-    sudo apt-get install gdb gtkwave lcov libfl-dev ccache jq ||
-    sudo apt-get install gdb gtkwave lcov libfl-dev ccache jq
-    # Required for test_regress/t/t_dist_attributes.pl
-    if [ "$CI_RUNS_ON" = "ubuntu-22.04" ]; then
+    sudo apt-get install gdb gtkwave lcov libfl-dev ccache jq z3 ||
+    sudo apt-get install gdb gtkwave lcov libfl-dev ccache jq z3
+    # Required for test_regress/t/t_dist_attributes.py
+    if [ "$CI_RUNS_ON" = "ubuntu-22.04" ] || [ "$CI_RUNS_ON" = "ubuntu-24.04" ]; then
      sudo apt-get install python3-clang mold ||
      sudo apt-get install python3-clang mold
    fi
-    if [ "$CI_RUNS_ON" = "ubuntu-20.04" ] || [ "$CI_RUNS_ON" = "ubuntu-22.04" ]; then
+    if [ "$CI_RUNS_ON" = "ubuntu-20.04" ] || [ "$CI_RUNS_ON" = "ubuntu-22.04" ] || [ "$CI_RUNS_ON" = "ubuntu-24.04" ]; then
      sudo apt-get install libsystemc-dev ||
      sudo apt-get install libsystemc-dev
    fi
  elif [ "$CI_OS_NAME" = "osx" ]; then
    brew update
    # brew cask install gtkwave # fst2vcd hangs at launch, so don't bother
-    brew install ccache perl jq
+    brew install ccache perl jq z3
  elif [ "$CI_OS_NAME" = "freebsd" ]; then
    # fst2vcd fails with "Could not open '<input file>', exiting."
-    sudo pkg install -y ccache gmake perl5 python3 jq
+    sudo pkg install -y ccache gmake perl5 python3 jq z3
  else
    fatal "Unknown os: '$CI_OS_NAME'"
  fi
  # Common installs
-  if [ "$CI_RUNS_ON" != "ubuntu-14.04" ]; then
-    CI_CPAN_REPO=https://cpan.org
-  fi
-  yes yes | sudo cpan -M $CI_CPAN_REPO -fi Parallel::Forker
  install-vcddiff
  # Workaround -fsanitize=address crash
  sudo sysctl -w vm.mmap_rnd_bits=28
@@ -33,6 +33,7 @@ elif [ "$CI_OS_NAME" = "freebsd" ]; then
 else
  fatal "Unknown os: '$CI_OS_NAME'"
 fi
+NPROC=$(expr $NPROC '+' 1)

 if [ "$CI_BUILD_STAGE_NAME" = "build" ]; then
  ##############################################################################
@@ -61,6 +62,7 @@ elif [ "$CI_BUILD_STAGE_NAME" = "test" ]; then
  # Run tests

  export VERILATOR_TEST_NO_CONTRIBUTORS=1  # Separate workflow check
+  export VERILATOR_TEST_NO_LINT_PY=1  # Separate workflow check

  if [ "$CI_OS_NAME" = "osx" ]; then
    export VERILATOR_TEST_NO_GDB=1  # Pain to get GDB to work on OS X
@@ -85,7 +87,7 @@ elif [ "$CI_BUILD_STAGE_NAME" = "test" ]; then
  fi

  # Run sanitize on Ubuntu 22.04 only
-  [ "$CI_RUNS_ON" = 'ubuntu-22.04' ] && sanitize='--sanitize' || sanitize=''
+  ( [ "$CI_RUNS_ON" = 'ubuntu-22.04' ] || [ "$CI_RUNS_ON" = 'ubuntu-24.04' ] ) && sanitize='--sanitize' || sanitize=''

  TEST_REGRESS=test_regress
  if [ "$CI_RELOC" == 1 ]; then
@@ -192,13 +194,12 @@ elif [ "$CI_BUILD_STAGE_NAME" = "test" ]; then
      fatal "Unknown test: $TESTS"
      ;;
  esac
+
+  # To see load average (1 minute, 5 minute, 15 minute)
+  uptime
  # 22.04: ccache -s -v
  ccache -s

-  # Upload coverage data
-  if [[ $TESTS == coverage-* ]]; then
-    bash <(cat ci/coverage-upload.sh) -f nodist/obj_dir/coverage/app_total.info
-  fi
 else
  ##############################################################################
  # Unknown build stage
@@ -1,13 +1,26 @@
+# DESCRIPTION: Verilator: CI Windows Power Shell - Compile Verilator
+#
+# Copyright 2024 by Wilson Snyder. This program is free software; you
+# can redistribute it and/or modify it under the terms of either the GNU
+# Lesser General Public License Version 3 or the Perl Artistic License
+# Version 2.0.
+#
+# SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
+################################################################################
+
+Set-PSDebug -Trace 1
+
 if (-Not (Test-Path $PWD/../.ccache/win_bison.exe)) {
-	git clone --depth 1 https://github.com/lexxmark/winflexbison
-	cd winflexbison
-	mkdir build
-	cd build
-	cmake .. --install-prefix $PWD/../../../.ccache
-	cmake --build . --config Release -j 3
-	cmake --install . --prefix $PWD/../../../.ccache
-	cd ../..
+    git clone --depth 1 https://github.com/lexxmark/winflexbison
+    cd winflexbison
+    mkdir build
+    cd build
+    cmake .. --install-prefix $PWD/../../../.ccache
+    cmake --build . --config Release -j 3
+    cmake --install . --prefix $PWD/../../../.ccache
+    cd ../..
 }
+
 mkdir build
 cd build
 cmake .. --install-prefix $PWD/../install
@@ -1,3 +1,16 @@
+# DESCRIPTION: Verilator: CI Windows Power Shell - Verilate a test
+#
+# Copyright 2024 by Wilson Snyder. This program is free software; you
+# can redistribute it and/or modify it under the terms of either the GNU
+# Lesser General Public License Version 3 or the Perl Artistic License
+# Version 2.0.
+#
+# SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
+################################################################################
+
+
+Set-PSDebug -Trace 1
+
 cd install
 $Env:VERILATOR_ROOT=$PWD
 cd examples/cmake_tracing_c
@@ -5,6 +18,9 @@ mkdir build
 cd build
 cmake ..
 cmake --build . --config Release -j 3
-Release/example.exe
+
+# TODO put this back in, see issue# 5163
+# Release/example.exe
+
 cd ..
 Remove-Item -path build -recurse
@@ -43,6 +43,7 @@ RUN apt-get update \
                        perl \
                        python3 \
                        wget \
+                        z3 \
                        zlib1g \
                        zlib1g-dev \
    && apt-get clean \
@@ -50,8 +51,6 @@ RUN apt-get update \

 WORKDIR /tmp

-RUN cpan install -fi Parallel::Forker
-
 RUN git clone https://github.com/veripool/vcddiff.git && \
    make -C vcddiff && \
    cp -p vcddiff/vcddiff /usr/local/bin/vcddiff && \
@@ -1,6 +1,8 @@
 .. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

+.. _Verilator Build Docker Container:
+
 Verilator Build Docker Container
 ================================

@@ -11,10 +11,11 @@ easily as a docker image, e.g.:

   docker run -ti verilator/verilator:latest --version

-This will install the container, run the latest Verilator and print
+This will pull the container from `docker hub
+<https://hub.docker.com/r/verilator/verilator>`_, run the latest Verilator and print
 Verilator's version.

-Containers are automatically built for all released versions, so you may
+Containers are automatically built and pushed to docker hub for all released versions, so you may
 easily compare results across versions, e.g.:

 ::
@@ -1,3 +1,4 @@
+---
 # DESCRIPTION: codecov.io config
 #
 # Copyright 2020-2024 by Wilson Snyder. This program is free software; you
@@ -8,31 +9,28 @@
 ####################
 # Validate:
 #  curl --data-binary @codecov.yml https://codecov.io/validate
-
+#
 codecov:
-  require_ci_to_pass: no
-
+  require_ci_to_pass: false
 coverage:
  precision: 2
  round: down
-  range: "50...100"
+  range: 50...100
  ignore:
-    - "ci"
-    - "docs"
-    - "examples"
-    - "include/gtkwave"
-    - "include/vltstd"
-    - "test_regress"
-
+    - "ci"  #
+    - "docs"  #
+    - "examples"  #
+    - "include/gtkwave"  #
+    - "include/vltstd"  #
+    - "test_regress"  #
 parsers:
  gcov:
    branch_detection:
-      conditional: yes
-      loop: yes
-      method: no
-      macro: no
-
+      conditional: true
+      loop: true
+      method: false
+      macro: false
 comment:
  layout: "reach,diff,flags,tree"
  behavior: default
-  require_changes: yes
+  require_changes: true
@@ -5,12 +5,12 @@
 # General Public License Version 3 or the Perl Artistic License Version 2.0.
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

-# When releasing, also update header of Changes file, and CmakeLists.txt,
+# When releasing, also update header of Changes file, and CMakeLists.txt,
 # and commit using "devel release" or "Version bump" message
 # Then 'make maintainer-dist'
 #AC_INIT([Verilator],[#.### YYYY-MM-DD])
 #AC_INIT([Verilator],[#.### devel])
-AC_INIT([Verilator],[5.024 2024-04-05],
+AC_INIT([Verilator],[5.030 2024-10-27],
        [https://verilator.org],
        [verilator],[https://verilator.org])

@@ -29,6 +29,9 @@ AC_DEFINE_UNQUOTED([PACKAGE_VERSION_STRING_CHAR],
        [Package version as a number])
 AC_SUBST(PACKAGE_VERSION_STRING_CHAR)

+######################################################################
+## Arguments/flag checking
+
 # Ignore automake flags passed by Ubuntu builds
 AC_ARG_ENABLE([dependency-tracking],
              [AS_HELP_STRING([--disable-dependency-tracking], [ignored])])
@@ -134,6 +137,33 @@ AC_ARG_ENABLE([longtests],
 AC_SUBST(CFG_WITH_LONGTESTS)
 AC_MSG_RESULT($CFG_WITH_LONGTESTS)

+AC_CHECK_PROG(HAVE_Z3,z3,yes)
+AC_CHECK_PROG(HAVE_CVC5,cvc5,yes)
+AC_CHECK_PROG(HAVE_CVC4,cvc4,yes)
+
+# Special Substitutions - CFG_WITH_SOLVER
+AC_MSG_CHECKING(for SMT solver)
+AC_ARG_WITH([solver],
+              [AS_HELP_STRING([--with-solver='z3 --in'],
+                              [set default SMT solver for constrained randomization])],
+              [CFG_WITH_SOLVER="${withval}"],
+              [CFG_WITH_SOLVER=no
+               if test "x$HAVE_Z3" = "xyes"; then
+                CFG_WITH_SOLVER="z3 --in"
+               elif test "x$HAVE_CVC5" = "xyes"; then
+                CFG_WITH_SOLVER="cvc5 --incremental"
+               elif test "x$HAVE_CVC4" = "xyes"; then
+                CFG_WITH_SOLVER="cvc4 --lang=smt2 --incremental"
+               fi]
+              )
+AC_SUBST(CFG_WITH_SOLVER)
+AC_MSG_RESULT($CFG_WITH_SOLVER)
+
+######################################################################
+## Compiler checks
+
+AC_MSG_RESULT([compiler CXX inbound is set to... $CXX])
+
 # Compiler flags (ensure they are not empty to avoid configure defaults)
 CFLAGS="$CFLAGS "
 CPPFLAGS="$CPPFLAGS "
@@ -160,24 +190,26 @@ if test "x$AR" = "x" ; then
   AC_MSG_ERROR([Cannot find "ar" in your PATH, please install it])
 fi

-AC_PATH_PROG(PERL,perl)
+AC_CHECK_PROG(PERL,perl,perl)
 if test "x$PERL" = "x" ; then
   AC_MSG_ERROR([Cannot find "perl" in your PATH, please install it])
 fi

-AC_PATH_PROG(PYTHON3,python3)
+AC_CHECK_PROG(PYTHON3,python3,python3)
 if test "x$PYTHON3" = "x" ; then
   AC_MSG_ERROR([Cannot find "python3" in your PATH, please install it])
 fi
+python3_version=$($PYTHON3 --version | head -1)
+AC_MSG_RESULT([$PYTHON3 --version = $python3_version])

-AC_PATH_PROG(LEX,flex)
+AC_CHECK_PROG(LEX,flex,flex)
 if test "x$LEX" = "x" ; then
   AC_MSG_ERROR([Cannot find "flex" in your PATH, please install it])
 fi
 flex_version=$($LEX --version | head -1)
 AC_MSG_RESULT([$LEX --version = $flex_version])

-AC_PATH_PROG(YACC,bison)
+AC_CHECK_PROG(YACC,bison,bison)
 if test "x$YACC" = "x" ; then
   AC_MSG_ERROR([Cannot find "bison" in your PATH, please install it])
 fi
@@ -266,6 +298,43 @@ AC_DEFUN([_MY_CXX_CHECK_OPT],
    _MY_CXX_CHECK_IFELSE($2, $1="$$1 $2")
   ])

+AC_DEFUN([_MY_CXX_CHECK_CORO_SET],
+   [# _MY_CXX_CHECK_CORO_SET(variable,option)
+    # Check if compiler supports coroutines with specific option. If it does,
+    # set variable to option, only if not previously set.
+    # Define HAVE_COROUTINES if supported.
+    if test "$$1" = ""; then
+      if test "$2" != ""; then
+        _my_msg_opt=" with $2"
+      else
+        _my_msg_opt=""
+      fi
+      AC_MSG_CHECKING([whether coroutines are supported by $CXX$_my_msg_opt])
+      ACO_SAVE_CXXFLAGS="$CXXFLAGS"
+      CXXFLAGS="$CXXFLAGS $2"
+      AC_LINK_IFELSE(
+         [AC_LANG_PROGRAM([
+            #if defined(__clang__) && !defined(__cpp_impl_coroutine)
+            #define __cpp_impl_coroutine 1
+            #endif
+            #include <coroutine>
+          ],[[std::coroutine_handle<> h;]])],
+         [_my_result=yes],
+         [AC_LINK_IFELSE(
+             [AC_LANG_PROGRAM(
+                [#include <experimental/coroutine>],
+                [[std::experimental::coroutine_handle<> h;]])],
+             [_my_result=yes],
+             [_my_result=no])])
+      AC_MSG_RESULT($_my_result)
+      CXXFLAGS="$ACO_SAVE_CXXFLAGS"
+      if test "$_my_result" = "yes" ; then
+        $1="$2"
+        AC_DEFINE([HAVE_COROUTINES],[1],[Defined if coroutines are supported by $CXX])
+      fi
+    fi
+  ])
+
 AC_DEFUN([_MY_LDLIBS_CHECK_FLAG],
   [# _MY_LDLIBS_CHECK_FLAG(flag) -- Check if linker supports specific options
    # Set $_my_result appropriately
@@ -351,32 +420,12 @@ _MY_CXX_CHECK_OPT(CFG_CXXFLAGS_WEXTRA,-Wthread-safety)
 AC_SUBST(CFG_CXXFLAGS_WEXTRA)

 # Flags for coroutine support for dynamic scheduling
-_MY_CXX_CHECK_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines-ts)
-_MY_CXX_CHECK_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines)
-_MY_CXX_CHECK_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines-ts -Wno-deprecated-experimental-coroutine)
+_MY_CXX_CHECK_CORO_SET(CFG_CXXFLAGS_COROUTINES, "")
+_MY_CXX_CHECK_CORO_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines-ts)
+_MY_CXX_CHECK_CORO_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines)
+_MY_CXX_CHECK_CORO_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines-ts -Wno-deprecated-experimental-coroutine)
+_MY_CXX_CHECK_CORO_SET(CFG_CXXFLAGS_COROUTINES,-std=gnu++20)
 AC_SUBST(CFG_CXXFLAGS_COROUTINES)
-
-# HAVE_COROUTINES
-# Check if coroutines are supported at all
-AC_MSG_CHECKING([whether coroutines are supported by $CXX])
-ACO_SAVE_CXXFLAGS="$CXXFLAGS"
-CXXFLAGS="$CXXFLAGS $CFG_CXXFLAGS_COROUTINES"
-AC_LINK_IFELSE(
-    [AC_LANG_PROGRAM([
-#ifdef __clang__
-#define __cpp_impl_coroutine 1
-#endif
-#include <coroutine>
-    ],[[]])],
-    [_my_result=yes
-     AC_DEFINE([HAVE_COROUTINES],[1],[Defined if coroutines are supported by $CXX])],
-    [AC_LINK_IFELSE(
-        [AC_LANG_PROGRAM([#include <experimental/coroutine>],[[]])],
-        [_my_result=yes
-         AC_DEFINE([HAVE_COROUTINES],[1],[Defined if coroutines are supported by $CXX])],
-        [_my_result=no])])
-AC_MSG_RESULT($_my_result)
-CXXFLAGS="$ACO_SAVE_CXXFLAGS"
 AC_SUBST(HAVE_COROUTINES)

 # Flags for compiling Verilator internals including parser always
@@ -584,6 +633,9 @@ AC_SUBST(HAVE_SYSTEMC)

 # Checks for system services

+######################################################################
+## Output
+
 # Other install directories
 pkgdatadir=${datadir}/verilator
 AC_SUBST(pkgdatadir)
@@ -7,22 +7,28 @@ Adam Bagley
 Adrian Sampson
 Adrien Le Masle
 Ahmed El-Mahmoudy
+Aidan McNay
 Aleksander Kiryk
 Alex Chadwick
+Alex Solomatnikov
 Àlex Torregrosa
 Aliaksei Chapyzhenka
 Ameya Vikram Singh
+Andrea Calabrese
 Andreas Kuster
 Andrei Kostovski
 Andrew Miloradovsky
 Andrew Nolte
 Anthony Donlon
 Arkadiusz Kozdra
+Arthur Rosa
 Aylon Chaim Porat
 Bartłomiej Chmiel
 Cameron Kirk
 Chih-Mao Chen
+Chris Bachhuber
 Chris Randall
+Christopher Taylor
 Chuxuan Wang
 Chykon
 Conor McCullough
@@ -33,14 +39,18 @@ David Ledger
 David Metz
 David Stanford
 David Turner
+Dercury
 Don Williamson
 Drew Ranck
 Drew Taussig
 Driss Hafdi
 Edgar E. Iglesias
+Eric Müller
 Eric Rippey
 Ethan Sifferman
 Eyck Jentzsch
+Furqan Nadir
+Fabian Keßler
 Fan Shupei
 february cozzocrea
 Felix Neumärker
@@ -51,12 +61,14 @@ G-A. Kamendje
 Garrett Smith
 Geza Lore
 Gianfranco Costamagna
+Gijs Burghoorn
 Glen Gibb
 Gökçe Aydos
 Graham Rushton
 Guokai Chen
 Gus Smith
 Gustav Svensk
+Han Qi
 Harald Heckmann
 Hennadii Chernyshchyk
 Howard Su
@@ -70,6 +82,7 @@ Iru Cai
 Ivan Vnučec
 Iztok Jeras
 Jake Merdich
+James Bailey
 James Hanlon
 James Hutchinson
 James Pallister
@@ -112,8 +125,11 @@ Kritik Bhimani
 Krzysztof Bieganski
 Krzysztof Boronski
 Krzysztof Boroński
+Krzysztof Obłonczek
 Kuba Ober
 Larry Doolittle
+Liam Braun
+Luca Colagrande
 Ludwig Rogiers
 Lukasz Dalek
 Maarten De Braekeleer
@@ -127,6 +143,7 @@ Marshal Qiao
 Martin Schmidt
 Martin Stadler
 Matthew Ballance
+Michael Bikovitsky
 Michael Killough
 Michaël Lefebvre
 Michal Czyz
@@ -136,18 +153,22 @@ Mladen Slijepcevic
 Morten Borup Petersen
 Mostafa Gamal
 Nandu Raj
+Nathan Graybeal
 Nathan Kohagen
 Nathan Myers
 Nolan Poe
 Oleh Maksymenko
 Patrick Stewart
+Paul Bowen-Huggett
 Paul Swirhun
 Paul Wright
+Pawel Jewstafjew
 Pawel Sagan
 Pengcheng Xu
 Peter Debacker
 Peter Horvath
 Peter Monsson
+Philip Axer
 Philipp Wagner
 Pierre-Henri Horrein
 Pieter Kapsenberg
@@ -157,15 +178,18 @@ Quentin Corradi
 Rafal Kapuscik
 Rasfunk
 Raynard Qiao
+Ricardo Barbedo
 Richard Myers
 Risto Pejašinović
 Robert Balas
 Rupert Swarbrick
+Ryan Ziegler
 Ryszard Rozak
 Samuel Riedel
 Sean Cross
 Sebastien Van Cauwenberghe
 Sergi Granell
+Seth Pellegrino
 Srinivasan Venkataramanan
 Stefan Wallentowitz
 Stephen Henry
@@ -185,6 +209,8 @@ Tudor Timi
 Tymoteusz Blazejczyk
 Udi Finkelstein
 Unai Martinez-Corral
+Udaya Raj Subedi
+Valentin Atepalikhin
 Varun Koyyalagunta
 Vassilis Papaefstathiou
 Veripool API Bot
@@ -194,6 +220,8 @@ William D. Jones
 Wilson Snyder
 Xi Zhang
 Yan Xu
+Yangyu Chen
+Yilou Wang
 Yinan Xu
 Yoda Lee
 Yossi Nivin
@@ -205,3 +233,4 @@ Zhanglei Wang
 Zhou Shen
 Zixi Li
 أحمد المحمودي
+404allen404
@@ -23,7 +23,7 @@ SPHINXBUILD ?= sphinx-build
 SOURCEDIR = guide
 BUILDDIR = _build

-SPHINXOPTS ?= -c guide
+SPHINXOPTS ?= -c guide -j 4
 ifneq ($(VERILATOR_ANALYTICS_ID),)
 SPHINXOPTS += -D html_theme_options.analytics_id=$(VERILATOR_ANALYTICS_ID)
 endif
@@ -23,9 +23,7 @@ class VlSphinxExtract:
                    outname = match.group(1)
                    print("Writing %s" % outname)
                    fhw = open(outname, "w", encoding="utf8")  # pylint: disable=consider-using-with
-                    fhw.write(
-                        ".. comment: generated by vl_sphinx_extract from " +
-                        filename + "\n")
+                    fhw.write(".. comment: generated by vl_sphinx_extract from " + filename + "\n")
                    fhw.write(".. code-block::\n")
                elif re.match(r'^[=a-zA-Z0-9_]', line):
                    fhw = None
@@ -39,18 +37,14 @@ parser = argparse.ArgumentParser(
    allow_abbrev=False,
    formatter_class=argparse.RawDescriptionHelpFormatter,
    description="""Read a file and extract documentation data.""",
-    epilog=
-    """ Copyright 2021-2024 by Wilson Snyder.  This package is free software;
+    epilog=""" Copyright 2021-2024 by Wilson Snyder.  This package is free software;
 you can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.

 SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0""")

-parser.add_argument('--debug',
-                    action='store_const',
-                    const=9,
-                    help='enable debug')
+parser.add_argument('--debug', action='store_const', const=9, help='enable debug')
 parser.add_argument('path', help='path to extract from')
 Args = parser.parse_args()

@@ -18,8 +18,7 @@ class VlSphinxFix:
        if os.path.isdir(path):
            for basefile in os.listdir(path):
                file = os.path.join(path, basefile)
-                if ((basefile != ".") and (basefile != "..")
-                        and basefile not in self.SkipBasenames
+                if ((basefile != ".") and (basefile != "..") and basefile not in self.SkipBasenames
                        and not os.path.islink(file)):
                    self.process(file)
        elif re.search(r'\.(html|tex)$', path):
@@ -54,18 +53,14 @@ parser = argparse.ArgumentParser(
    allow_abbrev=False,
    formatter_class=argparse.RawDescriptionHelpFormatter,
    description="""Post-process Sphinx HTML.""",
-    epilog=
-    """ Copyright 2021-2024 by Wilson Snyder.  This package is free software;
+    epilog=""" Copyright 2021-2024 by Wilson Snyder.  This package is free software;
 you can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.

 SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0""")

-parser.add_argument('--debug',
-                    action='store_const',
-                    const=9,
-                    help='enable debug')
+parser.add_argument('--debug', action='store_const', const=9, help='enable debug')
 parser.add_argument('path', help='path to edit')
 Args = parser.parse_args()

@@ -0,0 +1,11 @@
+.. comment: generated by t_lint_pinmissing_bad
+.. code-block:: sv
+   :linenos:
+   :emphasize-lines: 2
+
+   module t;
+      sub sub();  // <--- Warning
+   endmodule
+   module sub
+     (output port);
+   endmodule
@@ -0,0 +1,4 @@
+.. comment: generated by t_lint_pinmissing_bad
+.. code-block::
+
+   %Warning-PINMISSING: example.v:1:8 Cell has missing pin: 'port'
@@ -0,0 +1,11 @@
+.. comment: generated by t_lint_pkgnodecl_bad
+.. code-block:: sv
+   :linenos:
+   :emphasize-lines: 2
+
+   module t;
+      initial Pkg::hello();  //<--- Warning
+   endmodule
+   package Pkg;
+      function void hello(); endfunction
+   endpackage
@@ -0,0 +1,4 @@
+.. comment: generated by t_lint_pkgnodecl_bad
+.. code-block::
+
+   %Error-PKGNODECL: example.v:1:12 Package/class 'Pkg' not found, and needs to be predeclared (IEEE 1800-2023 26.3)
@@ -23,8 +23,7 @@ def get_vlt_version():
    filename = "../../Makefile"
    with open(filename, "r", encoding="utf8") as fh:
        for line in fh:
-            match = re.search(r"PACKAGE_VERSION *= *([a-z0-9.]+) +([-0-9]+)",
-                              line)
+            match = re.search(r"PACKAGE_VERSION *= *([a-z0-9.]+) +([-0-9]+)", line)
            if match:
                return match.group(1), match.group(2)
            match = re.search(r"PACKAGE_VERSION *= *([a-z0-9.]+) +devel", line)
@@ -75,8 +74,7 @@ extensions = []
 # directories to ignore when looking for source files.
 # This pattern also affects html_static_path and html_extra_path.
 exclude_patterns = [
-    '_build', 'Thumbs.db', '.DS_Store', 'internals.rst', 'xml.rst', 'gen/ex_*',
-    'CONTRIBUTING.rst'
+    '_build', 'Thumbs.db', '.DS_Store', 'internals.rst', 'xml.rst', 'gen/ex_*', 'CONTRIBUTING.rst'
 ]

 # Warn about refs
@@ -407,6 +407,16 @@ only a couple of instructions.
 For signal callbacks to work the main loop of the program must call
 :code:`VerilatedVpi::callValueCbs()`.

+Verilator also tracks when the model state has been modified via the VPI with
+an :code:`evalNeeded` flag.  This flag can be checked with :code:`VerilatedVpi::evalNeeded()`
+and it can be cleared with :code:`VerilatedVpi::clearEvalNeeded()`.  Used together
+it is possible to skip :code:`eval()` calls if no model state has been changed
+since the last :code:`eval()`.
+
+Any data written via :code:`vpi_put_value` with :code:`vpiInertialDelay` will
+be deferred for later.  These delayed values can be flushed to the model with
+:code:`VerilatedVpi::doInertialPuts()`.
+

 .. _VPI Example:

@@ -23,24 +23,24 @@ Next, try the :vlopt:`--debug` option.  This will enable additional
 internal assertions, and may help identify the problem.

 Finally, reduce your code to the smallest possible routine that exhibits
-the bug.  Even better, create a test in the :file:`test_regress/t`
-directory, as follows:
+the bug (see: :ref:`Minimizing bug-inducing code`).  Even better, create
+a test in the :file:`test_regress/t` directory, as follows:

 .. code-block:: bash

       cd test_regress
-       cp -p t/t_EXAMPLE.pl t/t_BUG.pl
+       cp -p t/t_EXAMPLE.py t/t_BUG.py
       cp -p t/t_EXAMPLE.v t/t_BUG.v

 There are many hints on how to write a good test in the
-:file:`test_regress/driver.pl` documentation which can be seen by running:
+:file:`test_regress/driver.py` documentation which can be seen by running:

 .. code-block:: bash

       cd $VERILATOR_ROOT  # Need the original distribution kit
-       test_regress/driver.pl --help
+       test_regress/driver.py --help

-Edit :file:`t/t_BUG.pl` to suit your example; you can do anything you want
+Edit :file:`t/t_BUG.py` to suit your example; you can do anything you want
 in the Verilog code there; just make sure it retains the single clk input
 and no outputs.  Now, the following should fail:

@@ -48,20 +48,38 @@ and no outputs.  Now, the following should fail:

       cd $VERILATOR_ROOT  # Need the original distribution kit
       cd test_regress
-       t/t_BUG.pl  # Run on Verilator
-       t/t_BUG.pl --debug # Run on Verilator, passing --debug to Verilator
-       t/t_BUG.pl --vcs  # Run on VCS simulator
-       t/t_BUG.pl --nc|--iv|--ghdl  # Likewise on other simulators
+       t/t_BUG.py  # Run on Verilator
+       t/t_BUG.py --debug # Run on Verilator, passing --debug to Verilator
+       t/t_BUG.py --vcs  # Run on VCS simulator
+       t/t_BUG.py --nc|--iv|--ghdl  # Likewise on other simulators

 The test driver accepts a number of options, many of which mirror the main
 Verilator options. For example the previous test could have been run with
 debugging enabled.  The full set of test options can be seen by running
-:command:`driver.pl --help` as shown above.
+:command:`driver.py --help` as shown above.

 Finally, report the bug at `Verilator Issues
 <https://verilator.org/issues>`_.  The bug will become publicly visible; if
 this is unacceptable, mail the bug report to ``[email protected]``.

+.. _Minimizing bug-inducing code:
+
+Minimizing bug-inducing code
+============================
+
+In some cases, the part of the code that causes the bug is clearly visible
+and the design can be easily manually reduced. In other cases, the bug is
+caused by a complex interaction of many parts of the design, and it is not
+clear which parts are necessary to reproduce the bug. In these cases, an
+Open Source tool called `sv-bugpoint
+<https://github.com/antmicro/sv-bugpoint>_` can be used to automatically
+reduce a SystemVerilog design to the smallest possible reproducer.
+It can be used to automatically reduce a design with hundreds of thousands of
+lines to a minimal test case while preserving the bug-inducing behavior.
+
+Please refer to the `README
+<https://github.com/antmicro/sv-bugpoint/blob/main/README.md>`_ file for more
+information on how to use `sv-bugpoint`.

 .. Contributing
 .. ============
@@ -12,7 +12,7 @@ C++14 compiler support
  Verilated models with --no-timing.

  Verilator will require C++20 or newer compilers for both compiling
-  Verilator and compiling all Verilated models no sooner than January 2025.
+  Verilator and compiling all Verilated models no sooner than May 2025.

 XML output
  Verilator currently supports XML parser output (enabled with `--xml-only`).
@@ -118,6 +118,12 @@ associated programs.

   See :ref:`Installation` for more details.

+.. option:: VERILATOR_SOLVER
+
+   If set, the command to run as a constrained randomization backend, such
+   as :command:`cvc4 --lang=smt2 --incremental`.  If not specified, it will use
+   the one supplied or found during configure, or :command:`z3 --in` if empty.
+
 .. option:: VERILATOR_VALGRIND

   If set, the command to run when using the :vlopt:`--valgrind` option, such as
@@ -28,6 +28,7 @@ Now, let's create an example Verilog, and SystemC wrapper file:

     cat >sc_main.cpp <<'EOF'
       #include "Vour.h"
+       using namespace sc_core;
       int sc_main(int argc, char** argv) {
           Verilated::commandArgs(argc, argv);
           sc_clock clk{"clk", 10, SC_NS, 0.5, 3, SC_NS, true};
@@ -40,8 +40,8 @@ Summary:
 .. option:: +verilator+error+limit+<value>

   Set number of non-fatal errors (e.g. assertion failures) before exiting
-   simulation runtime. Also affects number of $stop calls needed before
-   exit. Defaults to 1.
+   simulation runtime. Also affects number of `$stop` calls needed before
+   exit. Does not affect `$fatal`. Defaults to 1.

 .. option:: +verilator+help

@@ -222,6 +222,13 @@ Summary:
     expressions into sub-expressions to avoid error C1009, and breaking
     deep blocks into functions to avoid error C1061.

+.. option:: --compiler-include <header-path>
+
+   Specifies additional headers to be included in the final PCH header.
+   It is required to add them to this header, due to compilers'
+   limitation that allow only one precompiled header per compilation.
+   Use this instead of ::vlopt:`-CFLAGS` with `-include <header-path>`.
+
 .. option:: --converge-limit <loops>

   Rarely needed.  Specifies the maximum number of runtime iterations
@@ -476,6 +483,12 @@ Summary:
   See also :vlopt:`--dump-defines`, :vlopt:`-P`, and
   :vlopt:`--pp-comments` options.

+.. option:: --emit-accessors
+
+   Emit getter and setter methods for each top-level signal in the
+   model top class. Signals are still available as public members,
+   but with the `__Vm_sig_` prefix.
+
 .. option:: --error-limit <value>

   After this number of errors are encountered during Verilator run, exit.
@@ -710,6 +723,13 @@ Summary:
   :option:`/*verilator&32;hier_block*/` metacomment is ignored.  See
   :ref:`Hierarchical Verilation`.

+.. option:: --hierarchical-params-file <filename>
+
+   Internal flag inserted used during :vlopt:`--hierarchical`; specifies
+   name of hierarchical parameters file for deparametrized modules with
+   :option:`/*verilator&32;hier_block*/` metacomment. See
+   :ref:`Hierarchical Verilation`.
+
 .. option:: -I<dir>

   See :vlopt:`-y`.
@@ -817,6 +837,11 @@ Summary:
   If the design is not to be completely Verilated, see also the
   :vlopt:`--bbox-sys` and :vlopt:`--bbox-unsup` options.

+.. option:: --localize-max-size <value>
+
+   Rarely needed.  Set the maximum variable size in bytes for it to be
+   subject to localizing-to-stack optimization.  Defaults to 1024.
+
 .. option:: --make <build-tool>

   Generates a script for the specified build tool.
@@ -933,6 +958,22 @@ Summary:
   delayed assignments.  This option should only be used when suggested by
   the developers.

+.. option:: --output-groups <numfiles>
+
+   Enables concatenating the output .cpp files into the given number of
+   effective output .cpp files.  This is useful if the compiler startup
+   overhead from compiling many small files becomes unacceptable,
+   which can happen in designs making extensive use of SystemVerilog classes,
+   templates or generate blocks.
+
+   Using :vlopt:`--output-groups` can adversely impact caching and stability
+   (as in reproducibility) of compiled code.  Compilation of larger .cpp
+   files also has higher memory requirements.  Too low values might result in
+   swap thrashing with large designs, high values give no benefits.  The
+   value should range from 2 to 20 for small to medium designs.
+
+   Default is zero, which disables this feature.
+
 .. option:: --output-split <statements>

   Enables splitting the output .cpp files into multiple outputs.  When a
@@ -985,6 +1026,13 @@ Summary:
   :option:`/*verilator&32;sc_bv*/` metacomment to select specific ports to
   be sc_bv.

+.. option:: --pins-inout-enables
+
+   Specifies that the __en and __out outputs will always be created for
+   inouts in the top-level module. The __en variable has a one in a bit
+   position to indicate the corresponding bit of the __out variable has
+   a value being driven from within the Verilated model.
+
 .. option:: --pins-sc-uint

   Specifies SystemC inputs/outputs greater than 2 bits wide should use
@@ -992,6 +1040,10 @@ Summary:
   :vlopt:`--pins-sc-biguint` combination, it results in sc_uint being used
   between 2 and 64 and sc_biguint being used between 65 and 512.

+.. option:: --pins-sc-uint-bool
+
+   Specifies SystemC inputs/outputs one bit wide should use sc_uint<1>.
+
 .. option:: --pins-sc-biguint

   Specifies SystemC inputs/outputs greater than 65 bits wide should use
@@ -1732,9 +1784,9 @@ Summary:

   .. note::

-      This option applies only to values explicitly written as X
-      in modules (not classes) in the Verilog source code. Initial values
-      of clocks are set to 0 unless `--x-initial-edge` is
+      This option applies only to values explicitly written as X in modules
+      (not classes, nor parameters) in the Verilog source code. Initial
+      values of clocks are set to 0 unless `--x-initial-edge` is
      specified. Initial values of all other state holding variables are
      controlled with `--x-initial`.

@@ -1873,8 +1925,12 @@ Configuration Files

 In addition to the command line, warnings and other features for the
 :command:`verilator` command may be controlled with configuration files,
-typically named with the .vlt extension (what makes it a configuration file
-is the :option:`\`verilator_config` directive). An example:
+typically named with the `.vlt` extension (what makes it a configuration
+file is the :option:`\`verilator_config` directive).  These files, when
+named `.vlt`, are read before source code files; if this behavior is
+undesired, name the config file with a `.v` suffix.
+
+An example:

 .. code-block:: sv

@@ -1967,6 +2023,13 @@ The grammar of configuration commands is as follows:
   that the setting is ignored unless the :vlopt:`--hierarchical` option is
   specified.  See :ref:`Hierarchical Verilation`.

+.. option:: hier_params -module "<modulename>"
+
+   Specifies that the module contains parameters a :vlopt:`--hierarchical` block. This option
+   is used internally to specify parameters for deparametrized hier block instances.
+   This option should not be used directly.
+   See :ref:`Hierarchical Verilation`.
+
 .. option:: inline -module "<modulename>"

   Specifies the module may be inlined into any modules that use this
@@ -37,9 +37,9 @@ verilator_coverage Example Usage

    verilator_coverage --annotate <obj>

-    verilator_coverage  -write merged.dat -read <datafiles>...
+    verilator_coverage  -write merged.dat <datafiles>...

-    verilator_coverage  -write-info merged.info -read <datafiles>...
+    verilator_coverage  -write-info merged.info <datafiles>...


 verilator_coverage Arguments
@@ -58,32 +58,36 @@ to read multiple inputs.  If no data file is specified, by default,
 Specifies the directory name to which source files with annotated coverage
 data should be written.

-Converting from the Verilator coverage data format to the info format is
-lossy; the info will have all forms of coverage merged line coverage, and
-if there are multiple coverage points on a single line they will merge.
-The minimum coverage across all merged points will be used to report
-coverage of the line.
+Points are children of each line coverage- branches or toggle points.
+When point counts are aggregated into a line, the minimum and maximum counts
+are used to determine the status of the line (complete, partial, failing)
+The count is equal to the maximum of the points.

 Coverage data is annotated at the beginning of the line and is formatted
 as a special character followed by the number of coverage hits. The special
-characters " ,%,+,-" indicate summary of the coverage, and allow use of grep
+characters " ,%,~,+,-" indicate summary of the coverage, and allow use of grep
 to filter the report.

-* " " (whitespace) indicates that all points on the line are above the coverage limit.
-* "%" indicates at least one point on the line was below the coverage limit.
-* "+" coverage point was at or above the limit. Only used with :option:`--annotate-points`.
-* "-" coverage point was below the limit.  Only used with :option:`--annotate-points`.
+* " " (whitespace) indicates that all points on the line are above the coverage min.
+* "%" indicates that all points on the line are below the coverage min.
+* "~" indicates that some points on the line are above the coverage min and some are below.
+* "+" coverage point was at or above the min. Only used with :option:`--annotate-points`.
+* "-" coverage point was below the min.  Only used with :option:`--annotate-points`.

 .. code-block::

-   100000  input logic a;   // Begins with whitespace, because
-                            // number of hits (100000) is above the limit.
-  +100000  point: comment=a // Begins with +, because
-                            // number of hits (100000) is above the limit.
-  %000000  input logic b;   // Begins with %, because
-                            // number of hits (0) is below the limit.
-  -000000  point: comment=b // Begins with -, because
-                            // number of hits (0) is below the limit.
+   100000  input logic a;      // Begins with whitespace, because
+                               // number of hits (100000) is above the min.
+  +100000  point: comment=a    // Begins with +, because
+                               // number of hits (100000) is above the min.
+  %000000  input logic b;      // Begins with %, because
+                               // number of hits (0) is below the min.
+  -000000  point: comment=b    // Begins with -, because
+                               // number of hits (0) is below the min.
+  ~000010  if (cyc!=0) begin   // Begins with ~, because
+                               // branches are below and above the min.
+  +000010  point: comment=if   // The if branch is above the min.
+  -000000  point: comment=else // The else branch is below the min.

 .. option:: --annotate-all

@@ -154,10 +158,7 @@ generated from random test runs) into one master coverage file.
 Specifies the aggregate coverage results, summed across all the files,
 should be written to the given filename in :command:`lcov` .info format.
 This may be used to feed into :command:`lcov` to aggregate or generate
-reports.
-
-Converting from the Verilator coverage data format to the info format is
-lossy; the info will have all forms of coverage merged line coverage, and
-if there are multiple coverage points on a single line they will merge.
-The minimum coverage across all merged points will be used to report
-coverage of the line.
+reports. This format lacks the comments for cover points that the
+verilator_coverage format has. It can be used with :command:`genhtml`
+to generate an HTML report. :command:`genhtml --branch-coverage` will
+also display the branch coverage, analogous to :option:`--annotate-points`
@@ -259,9 +259,11 @@ How do I view waveforms (aka dumps or traces)?
 """"""""""""""""""""""""""""""""""""""""""""""

 Verilator creates standard VCD (Value Change Dump) and FST files.  VCD
-files are viewable with the open-source GTKWave (recommended), or Dinotrace
-(legacy) programs, or any of the many closed-source offerings; FST is
-supported only by GTKWave.
+files are viewable with the open-source
+`GTKWave <http://gtkwave.sourceforge.net/>`_,
+`Surfer <https://surfer-project.org/>`_, Dinotrace (legacy), or any of the
+many closed-source viewer offerings; FST is supported only by GTKWave and
+Surfer.


 How do I speed up writing large waveform (trace) files?
@@ -48,12 +48,6 @@ For --cc/--sc, it creates:
     - Make include file for compiling (from --make gmake)
   * - *{prefix}*\ _classes.mk
     - Make include file with class names (from --make gmake)
-   * - *{prefix}*\ _hier.mk
-     - Make file for hierarchy blocks (from --make gmake)
-   * - *{prefix}*\ __hierMkArgs.f
-     - Arguments for hierarchical Verilation (from --make gmake)
-   * - *{prefix}*\ __hierCMakeArgs.f
-     - Arguments for hierarchical Verilation (from --make cmake)
   * - *{prefix}*\ .h
     - Model header
   * - *{prefix}*\ .cpp
@@ -95,16 +89,24 @@ For --cc/--sc, it creates:
   * - *{prefix}{each_verilog_module}{__DepSet_hash__n}*\ .cpp
     - Additional lower C++ files (hashed to reduce build times)

-For --hierarchy mode, it creates:
+For --hierarchical mode, it creates:

 .. list-table::

   * - V\ *{hier_block}*\ /
-     - Directory to Verilate each hierarchy block (from --hierarchy)
+     - Directory to Verilate each hierarchical block (from --hierarchical)
   * - *{prefix}*\ __hierVer.d
-     - Make dependencies of the top module (from --hierarchy)
+     - Make dependencies of the top module (from --hierarchical)
+   * - *{prefix}*\ _hier.mk
+     - Make file for hierarchical blocks (from --make gmake)
+   * - *{prefix}*\ __hierCMakeArgs.f
+     - Arguments for hierarchical Verilation (from --make cmake)
+   * - *{prefix}*\ __hierMkArgs.f
+     - Arguments for hierarchical Verilation (from --make gmake)
+   * - *{prefix}*\ __hierParameters.v
+     - Module parameters for hierarchical blocks
   * - *{prefix}*\ __hier.dir
-     - Directory to store .dot, .vpp, .tree of top module (from --hierarchy)
+     - Directory to store .dot, .vpp, .tree of top module (from --hierarchical)

 In specific debug and other modes, it also creates:

@@ -19,13 +19,33 @@ started. (Note packages are unlikely to have the most recent version, so
 :ref:`Git Install` might be a better alternative.) To install as a
 package:

-::
+.. code-block:: shell

   apt-get install verilator   # On Ubuntu

 For other distributions, refer to `Repology Verilator Distro Packages
 <https://repology.org/project/verilator>`__.

+.. _pre-commit Quick Install:
+
+pre-commit Quick Install
+=============================
+
+You can use Verilator's `pre-commit <https://pre-commit.com/>`__ hook to
+lint your code before committing it.  It encapsulates the :ref:`Verilator
+Build Docker Container`, so you need docker on your system to use it.  The
+verilator image will be downloaded automatically.
+
+To use the hook, add the following entry to your :code:`.pre-commit-config.yaml`:
+
+.. code-block:: yaml
+
+   repos:
+     - repo: https://github.com/verilator/verilator
+       rev: v5.026  # or later
+       hooks:
+         - id: verilator
+
 .. _Git Install:

 Git Quick Install
@@ -36,7 +56,7 @@ options and details, see :ref:`Detailed Build Instructions` below.

 In brief, to install from git:

-::
+.. code-block:: shell

   # Prerequisites:
   #sudo apt-get install git help2man perl python3 make autoconf g++ flex bison ccache
@@ -89,7 +109,7 @@ Install Prerequisites

 To build or run Verilator, you need these standard packages:

-::
+.. code-block:: shell

   sudo apt-get install git help2man perl python3 make
   sudo apt-get install g++  # Alternatively, clang
@@ -101,7 +121,7 @@ To build or run Verilator, you need these standard packages:
 To build or run Verilator, the following are optional but should be installed
 for good performance:

-::
+.. code-block:: shell

   sudo apt-get install ccache  # If present at build, needed for run
   sudo apt-get install mold  # If present at build, needed for run
@@ -110,27 +130,26 @@ for good performance:
 The following is optional but is recommended for nicely rendered command line
 help when running Verilator:

-::
+.. code-block:: shell

   sudo apt-get install perl-doc

 To build Verilator you will need to install these packages; these do not
 need to be present to run Verilator:

-::
+.. code-block:: shell

   sudo apt-get install git autoconf flex bison

 Those developing Verilator itself may also want these (see internals.rst):

-::
+.. code-block:: shell

   sudo apt-get install clang clang-format-14 cmake gdb gprof graphviz lcov
   sudo apt-get install python3-clang yapf3 bear jq
   sudo pip3 install sphinx sphinx_rtd_theme sphinxcontrib-spelling breathe ruff
   sudo pip3 install git+https://github.com/antmicro/astsee.git
   cpan install Pod::Perldoc
-   cpan install Parallel::Forker


 Install SystemC
@@ -152,6 +171,17 @@ To make use of Verilator FST tracing you will want `GTKwave
 required at Verilator build time.


+Install Z3
+^^^^^^^^^^
+
+In order to use constrained randomization the `Z3 Theorem Prover
+<https://github.com/z3prover/z3#readme>`__ must be installed, however this is
+not required at Verilator build time. There are other compatible SMT solvers,
+like CVC5/CVC4, but they are not guaranteed to work. Since different solvers are
+faster for different scenarios, the solver to use at run-time can be specified
+by the environment variable :option:`VERILATOR_SOLVER`.
+
+
 .. _Obtain Sources:

 Obtain Sources
@@ -160,14 +190,14 @@ Obtain Sources
 Get the sources from the git repository: (You need to do this only once,
 ever.)

-::
+.. code-block:: shell

   git clone https://github.com/verilator/verilator   # Only first time
   ## Note the URL above is not a page you can see with a browser; it's for git only

 Enter the checkout and determine what version/branch to use:

-::
+.. code-block:: shell

   cd verilator
   git pull        # Make sure we're up-to-date
@@ -182,7 +212,7 @@ Auto Configure

 Create the configuration script:

-::
+.. code-block:: shell

   autoconf        # Create ./configure script

@@ -193,8 +223,9 @@ Eventual Installation Options
 Before configuring the build, you must decide how you're going to
 eventually install Verilator onto your system. Verilator will be compiling
 the current value of the environment variables :option:`VERILATOR_ROOT`,
-:option:`SYSTEMC_INCLUDE`, and :option:`SYSTEMC_LIBDIR` as defaults into
-the executable, so they must be correct before configuring.
+:option:`VERILATOR_SOLVER`, :option:`SYSTEMC_INCLUDE`, and
+:option:`SYSTEMC_LIBDIR` as defaults into the executable, so they must be
+correct before configuring.

 These are the installation options:

@@ -207,7 +238,7 @@ directory (don't run ``make install``). This allows the easiest
 experimentation and upgrading, and allows many versions of Verilator to
 co-exist on a system.

-::
+.. code-block:: shell

   export VERILATOR_ROOT=`pwd`   # if your shell is bash
   setenv VERILATOR_ROOT `pwd`   # if your shell is csh
@@ -229,7 +260,7 @@ that may support multiple versions of every tool. Tell configure the
 eventual destination directory name.  We recommend that the destination
 location include the Verilator version name:

-::
+.. code-block:: shell

   unset VERILATOR_ROOT      # if your shell is bash
   unsetenv VERILATOR_ROOT   # if your shell is csh
@@ -241,7 +272,7 @@ the ``bin`` directory to your ``PATH``. Or, if you use `modulecmd
 <http://modules.sourceforge.net/>`__, you'll want a module file like the
 following:

-::
+.. code-block:: shell

   set install_root /CAD_DISK/verilator/{version-number-used-above}
   unsetenv VERILATOR_ROOT
@@ -256,7 +287,7 @@ following:
 The final option is to eventually install Verilator globally, using
 configure's default system paths:

-::
+.. code-block:: shell

   unset VERILATOR_ROOT      # if your shell is bash
   unsetenv VERILATOR_ROOT   # if your shell is csh
@@ -273,7 +304,7 @@ The command to configure the package was described in the previous step.
 Developers should configure to have more complete developer tests.
 Additional packages may be required for these tests.

-::
+.. code-block:: shell

   export VERILATOR_AUTHOR_SITE=1    # Put in your .bashrc
   ./configure --enable-longtests  ...above options...
@@ -284,7 +315,7 @@ Compile

 Compile Verilator:

-::
+.. code-block:: shell

   make -j `nproc`  # Or if error on `nproc`, the number of CPUs in system

@@ -294,7 +325,7 @@ Test

 Check the compilation by running self-tests:

-::
+.. code-block:: shell

   make test

@@ -306,7 +337,7 @@ If you used any install option other than the `1. Run-in-Place from
 VERILATOR_ROOT <#_1_run_in_place_from_verilator_root>`__ scheme, install
 the files:

-::
+.. code-block:: shell

   make install

@@ -18,7 +18,9 @@ Simulation Summary Report
 =========================

 When simulation finishes, it will print a report to stdout summarizing the
-simulation. This requires the model being Verilated with :vlopt:`--main`.
+simulation. This requires the model being Verilated with :vlopt:`--main`,
+or the user's `main()` calling `VerilatedContext->statsPrintSummary()`.
+
 The report may be disabled with :vlopt:`+verilator+quiet`.

 For example:
@@ -304,9 +304,9 @@ GNU Make
 ========

 Verilator defaults to creating GNU Make makefiles for the model.  Verilator
-will call make automatically when the :vlopt:'--build' option is used.
+will call make automatically when the :vlopt:`--build` option is used.

-If calling Verilator from a makefile, the :vlopt:'-MMD' option will create
+If calling Verilator from a makefile, the :vlopt:`--MMD` option will create
 a dependency file, allowing Make to only run Verilator if input Verilog
 files change.

@@ -210,22 +210,35 @@ List Of Warnings

 .. option:: BLKLOOPINIT

-   .. TODO better example
-
-   This indicates that the initialization of an array needs to use
-   non-delayed assignments.  This is done in the interest of speed; if
-   delayed assignments were used, the simulator would have to copy large
-   arrays every cycle.  (In smaller loops, loop unrolling allows the
-   delayed assignment to work, though it's a bit slower than a non-delayed
-   assignment.)  Here's an example
+   Indicates certain constructs where non-blocking assignments to unpacked
+   arrays (memories) are not supported inside loops. These typically appear in
+   initialization/reset code:

   .. code-block:: sv

         always @(posedge clk)
            if (~reset_l)
                for (i=0; i<`ARRAY_SIZE; i++)
-                    array[i] = 0;  // Non-delayed for verilator
+                    array[i] <= 0;  // Non-blocking assignment inside loop
+            else
+                array[address] <= data;

+   While this is supported in typical synthesizeable code (including the
+   example above), some complicated cases are not supported. Namely:
+
+   1. If the above loop is inside a suspendable process or fork statement.
+
+   2. If the variable is also the target of a '<=' non-blocking assignment
+   in a suspendable process or fork statement (in addition to a synthesizable
+   loop).
+
+   3. If the element type of the array is a compound type.
+
+   4. In versions before 5.026, any delayed assignment to an array.
+
+   It might slightly improve run-time performance if you change the
+   non-blocking assignment inside the loop into a blocking assignment
+   (that is: use '=' instead of '<='), if possible.

   This message is only seen on large or complicated loops because
   Verilator generally unrolls small loops.  You may want to try increasing
@@ -406,9 +419,9 @@ List Of Warnings

 .. option:: CONSTRAINTIGN

-   Warns that Verilator does not support :code:`constraint`,
-   :code:`constraint_mode`, or :code:`rand_mode`, and the construct was are
-   ignored.
+   Warns that Verilator does not support certain forms of
+   :code:`constraint`, :code:`constraint_mode`, or :code:`rand_mode`, and
+   the construct was are ignored.

   Ignoring this warning may make Verilator randomize() simulations differ
   from other simulators.
@@ -1204,6 +1217,14 @@ List Of Warnings
   :vlopt:`--no-timing` option.


+.. option:: NONSTD
+
+   Warns when a non-standard language feature is used that has a standard
+   equivalent, which might behave differently in corner cases. For example
+   :code:`$psprintf` system function is replaced by its standard equivalent
+   :code:`$sformatf`.
+
+
 .. option:: NULLPORT

   Warns that a null port was detected in the module definition port
@@ -1243,7 +1264,7 @@ List Of Warnings

 .. option:: PINMISSING

-   .. TODO better example
+   .. TODO better example (frequent)

   Warns that a module has a pin that is not mentioned in an instance.  If
   a pin is not missing it should still be specified on the instance
@@ -1252,6 +1273,22 @@ List Of Warnings
   Ignoring this warning will only suppress the lint check; it will
   simulate correctly.

+   Faulty example:
+
+   .. include:: ../../docs/gen/ex_PKGNODECL_faulty.rst
+
+   Results in:
+
+   .. include:: ../../docs/gen/ex_PKGNODECL_msg.rst
+
+   Repaired example:
+
+   .. code-block:: sv
+      :emphasize-lines: 2
+
+         sub sub (
+                  .port());
+
   Other tools with similar warnings: Icarus Verilog's portbind, "warning:
   Instantiating module ... with dangling input port (...)". Slang's
   unconnected-port, "port '...' has no connection".
@@ -1283,7 +1320,7 @@ List Of Warnings
         generate
            if (A==0) begin
               b b_inst1 (.x(1'b0));  //<--- error nonexistent port
-               b #(.PX(1'b0)) b_inst2 (); //<--- error nonexistent parameter
+               b #(.PX(1'b0)) b_inst2 ();  //<--- error nonexistent parameter
            end
          endgenerate
       endmodule
@@ -1325,11 +1362,22 @@ List Of Warnings

 .. option:: PKGNODECL

-   .. TODO better example
+   An error that a package/class appears to have been referenced that has
+   not yet been declared.  According to IEEE 1800-2023 26.3, all packages
+   must be declared before being used.

-   An error that a package/class appears to have been referenced that has not
-   yet been declared.  According to IEEE 1800-2023 26.3, all packages must
-   be declared before being used.
+   Faulty example:
+
+   .. include:: ../../docs/gen/ex_PKGNODECL_faulty.rst
+
+   Results in:
+
+   .. include:: ../../docs/gen/ex_PKGNODECL_msg.rst
+
+   Often the package is declared in its own header file.  In this case add
+   an include of that package header file to the referencing file.  (And
+   make sure you have header guards in the package's header file to prevent
+   multiple declarations of the package.)


 .. option:: PREPROCZERO
@@ -1354,7 +1402,7 @@ List Of Warnings
       `endif

   The portable way to suppress this warning is to use a define value other
-   than zero to when used in a preprocessor expression.
+   than zero, when it is to be used in a preprocessor expression.


 .. option:: PROCASSWIRE
@@ -1876,7 +1924,8 @@ List Of Warnings

 .. option:: UNPACKED

-   Warns that unpacked structs and unions are not supported.
+   Warns that unpacked structs and unions are not supported because
+   :vlopt:`--structs-packed` was used, or by up through version 5.004.

   Ignoring this warning will make Verilator treat the structure as packed,
   which may make Verilator simulations differ from other simulators. This
@@ -1927,14 +1976,16 @@ List Of Warnings

   .. TODO better example

-   Warns that the specified genvar is never used/consumed.
+   Warns that the specified genvar is never used/consumed. See similar
+   :option:`UNUSEDSIGNAL`.


 .. option:: UNUSEDPARAM

   .. TODO better example

-   Warns that the specified parameter is never used/consumed.
+   Warns that the specified parameter is never used/consumed. See similar
+   :option:`UNUSEDSIGNAL`.


 .. option:: UNUSEDSIGNAL
@@ -2057,10 +2108,6 @@ List Of Warnings

 .. option:: WAITCONST

-   .. code-block:: sv
-
-         wait(1);  // Blocks forever
-
   Warns that a `wait` statement awaits a constant condition, which means it
   either blocks forever or never blocks.

@@ -2068,6 +2115,12 @@ List Of Warnings
   something that elaborates to zero), does not warn, as it is presumed the
   code is making the intent clear.

+   Faulty example:
+
+   .. code-block:: sv
+
+         wait(1);  // Blocks forever
+

 .. option:: WIDTH

@@ -2104,7 +2157,7 @@ List Of Warnings

   .. include:: ../../docs/gen/ex_WIDTHEXPAND_1_faulty.rst

-   Results in a WIDTHEXPAND warning:
+   Results in a :option:`WIDTHEXPAND` warning:

   .. include:: ../../docs/gen/ex_WIDTHEXPAND_1_msg.rst

@@ -2115,15 +2168,18 @@ List Of Warnings

 .. option:: WIDTHTRUNC

-   A more granular WIDTH warning, for when a value is truncated.
+   A more granular :option:`WIDTH` warning, for when a value is
+   truncated. See :option:`WIDTH`.

 .. option:: WIDTHEXPAND

-   A more granular WIDTH warning, for when a value is zero expanded.
+   A more granular :option:`WIDTH` warning, for when a value is zero
+   expanded. See :option:`WIDTH`.

 .. option:: WIDTHXZEXPAND

-   A more granular WIDTH warning, for when a value is X/Z expanded.
+   A more granular :option:`WIDTH` warning, for when a value is X/Z
+   expanded. See :option:`WIDTH`.

 .. option:: WIDTHCONCAT

@@ -2172,7 +2228,7 @@ List Of Warnings
            #(parameter int MY_PARAM = 0);
           reg [7:0] data;
           always @* begin
-             data = {MY_PARAM{1'b1}}; //<--- WARNING
+             data = {MY_PARAM{1'b1}};  //<--- Warning
           end
         endmodule

@@ -2193,7 +2249,7 @@ List Of Warnings
      :emphasize-lines: 2

         module dut
-            #(parameter int MY_PARAM = 1); //<--- REPAIRED
+            #(parameter int MY_PARAM = 1);  //<--- REPAIRED
           reg [7:0] data;
           always @* begin
             data = {MY_PARAM{1'b1}};
@@ -1031,6 +1031,91 @@ the evaluation process records a bitmask of variables that might have
 changed; if clear, checking those signals for changes may be skipped.


+Constrained randomization
+-------------------------
+
+Because general constrained randomization is a co-NP-hard problem, not all
+cases are implemented in Verilator, and an external specialized SMT solver is
+used for any non-obvious ones.
+
+The ``randomize()`` method spawns an SMT solver in a sub-process. Then the
+solver gets a setup query, then the definition of variables, then all the
+constraints (SMT assertions) about the variables. Since the solver has no
+information about the class' PRNG state, if the problem is satisfiable,
+the solution space is further constrained by adding extra random constraints,
+and querying the values satisfying the problem statement.
+The constraint is currently constructed as fixing a simple xor of randomly
+chosen bits of the variables being randomized.
+
+The runtime classes used for handling the randomization are  defined in
+``verilated_random.h`` and ``verilated_random.cpp``.
+
+
+``VlSubprocess``
+~~~~~~~~~~~~~~~~
+
+Subprocess handle, responsible for keeping track of the resources like child
+PID, read and write file descriptors, and presenting them as a C++ iostream.
+
+
+``VlRandomizer``
+~~~~~~~~~~~~~~~~
+
+Randomizer class, responsible for keeping track of variables and constraints,
+and communicating with the solver subprocess.
+
+The solver gets the constraints in `SMT-LIB2
+<https://smtlib.cs.uiowa.edu/>`__ textual format in the following syntax:
+
+::
+
+    (set-info :smt-lib-version 2.0)
+    (set-option :produce-models true)
+    (set-logic QF_BV)
+
+    (declare-fun v () (_ BitVec 16))
+    (declare-fun w () (_ BitVec 64))
+    (declare-fun x () (_ BitVec 48))
+    (declare-fun z () (_ BitVec 24))
+    (declare-fun t () (_ BitVec 23))
+    (assert (or (= v #x0003) (= v #x0008)))
+    (assert (= w #x0000000000000009))
+    (assert (or (or (= x #x000000000001) (= x #x000000000002)) (or (= x #x000000000004) (= x #x000000000009))))
+    (assert (bvult ((_ zero_extend 8) z) #x00000015))
+    (assert (bvugt ((_ zero_extend 8) z) #x0000000d))
+
+    (check-sat)
+
+The solver responds with either ``sat`` or ``unsat``. Then the initial solution
+is queried with:
+
+::
+
+    (get-value (v w x z t ))
+
+The solver then responds with e.g.:
+
+::
+
+    ((v #x0008)
+     (w #x0000000000000005)
+     (x #x000000000002)
+     (z #x000010)
+     (t #b00000000000000000000000))
+
+And then a follow-up query (or a series thereof) is asked, and the solver gets
+reset, so that it can be reused by subsequent randomization attempts:
+
+::
+
+    (assert (= (bvxor (bvxor <...> (bvxor ((_ extract 21 21) z) ((_ extract 39 39) x)) ((_ extract 5 5) w)) <...> ((_ extract 10 10) w)) #b0))
+    (check-sat)
+    (get-value)
+    ...
+    (reset)
+
+
+
 Coding Conventions
 ==================

@@ -1387,9 +1472,6 @@ For all tests to pass, you must install the following packages:

 -  SystemC to compile the SystemC outputs, see http://systemc.org

-  Parallel::Forker from CPAN to run tests in parallel; you can install
-   this with e.g. "sudo cpan install Parallel::Forker".
-
 -  vcddiff to find differences in VCD outputs. See the readme at
   https://github.com/veripool/vcddiff

@@ -1399,7 +1481,7 @@ For all tests to pass, you must install the following packages:
 Controlling the Test Driver
 ---------------------------

-The test driver script `driver.pl` runs tests; see the `Test Driver`
+The test driver script `driver.py` runs tests; see the `Test Driver`
 section.  The individual test drivers are written in Perl; see `Test
 Language`.

@@ -1412,7 +1494,7 @@ A specific regression test can be executed manually. To start the

 ::

-   test_regress/t/t_EXAMPLE.pl
+   test_regress/t/t_EXAMPLE.py


 Regression Testing for Developers
@@ -1432,13 +1514,6 @@ Developers will also want to call ./configure with two extra flags:
   disabled by default, as SystemC installation problems would otherwise
   falsely indicate a Verilator problem.

-When enabling the long tests, some additional Perl modules are needed,
-which you can install using cpan.
-
-::
-
-   cpan install Parallel::Forker
-
 There are some traps to avoid when running regression tests

 - When checking the MANIFEST, the test will fail on unexpected code in the
@@ -1786,7 +1861,7 @@ represent the pointers (``op1p``, ``op2p``, etc) between the nodes.
 Debugging with GDB
 ------------------

-The `driver.pl` script accepts ``--debug --gdb`` to start
+The `driver.py` script accepts ``--debug --gdb`` to start
 Verilator under gdb and break when an error is hit, or the program is about
 to exit. You can also use ``--debug --gdbbt`` to just backtrace and then
 exit gdb. To debug the Verilated executable, use ``--gdbsim``.
@@ -1797,7 +1872,7 @@ can use ``--debug`` and look at the underlying invocation of

 ::

-   t/t_alw_dly.pl --debug
+   t/t_alw_dly.py --debug

 shows it invokes the command:

@@ -1894,7 +1969,7 @@ Generally, what would you do to add a new feature?
   Follow the convention described above about the AstNode type hierarchy.
   Ordering of definitions is enforced by ``astgen``.

-5. Now you can run ``test_regress/t/t_<newtestcase>.pl --debug`` and it'll
+5. Now you can run ``test_regress/t/t_<newtestcase>.py --debug`` and it'll
   probably fail, but you'll see a
   ``test_regress/obj_dir/t_<newtestcase>/*.tree`` file which you can examine
   to see if the parsing worked. See also the sections above on debugging.
@@ -1943,7 +2018,7 @@ IEEE 1800-2023 33 Config
 Test Driver
 ===========

-This section documents the test driver script, `driver.pl`.  driver.pl
+This section documents the test driver script, `driver.py`.  driver.py
 invokes Verilator or another simulator on each test file.  For test file
 contents description see `Test Language`.

@@ -1955,7 +2030,7 @@ the regression tests with OBJCACHE enabled and in parallel on a machine
 with many cores.  See the -j option and OBJCACHE environment variable.


-driver.pl Non-Scenario Arguments
+driver.py Non-Scenario Arguments
 --------------------------------

 --benchmark [<cycles>]
@@ -2025,13 +2100,13 @@ driver.pl Non-Scenario Arguments
  memory leaks.

 --site
-  Run site specific tests also.
+  Run site-specific tests also.

 --stop
  Stop on the first error.

 --trace
-  Set the simulator specific flags to request waveform tracing.
+  Set the simulator-specific flags to request waveform tracing.

 --valgrind
  Same as ``verilator --valgrind``: Run Verilator under `Valgrind <https://valgrind.org/>`_.
@@ -2044,7 +2119,7 @@ driver.pl Non-Scenario Arguments
  For tests using the standard C++ wrapper, enable runtime debug mode.


-driver.pl Scenario Arguments
+driver.py Scenario Arguments
 ----------------------------

 The following options control which simulator is used, and which tests are
@@ -2086,7 +2161,7 @@ simultaneously.
  Run Xilinx XSim simulator tests.


-driver.pl Environment
+driver.py Environment
 ---------------------

 HARNESS_UPDATE_GOLDEN
@@ -2125,6 +2200,10 @@ VERILATOR_ROOT
  Standard path to Verilator distribution root; see primary Verilator
  documentation.

+VERILATOR_SOLVER
+  SMT solver command for constrained randomization; see primary Verilator
+  documentation.
+
 VERILATOR_TESTS_SITE
  Used with ``--site``, a colon-separated list of directories with tests to
  be added to testlist.
@@ -2142,30 +2221,30 @@ VERILATOR_XVLOG
 Test Language
 =============

-This section describes the format of the ``test_regress/t/*.pl`` test
-language files, executed by `driver.pl`.
+This section describes the format of the ``test_regress/t/*.py`` test
+language files, executed by `driver.py`.

 Test Language Summary
 ---------------------

 For convenience, a summary of the most commonly used features is provided
 here, with a reference in a later section. All test files typically have a
-call to the ``lint`` or ``compile`` subroutine to compile the test. For
-run-time tests, this is followed by a call to the ``execute``
-subroutine. Both of these functions can optionally be provided with
-arguments specifying additional options.
+call to the ``test.lint`` or ``test.compile`` methods to compile the
+test. For run-time tests, this is followed by a call to the
+``test.execute`` method. Both of these functions can optionally be provided
+with arguments specifying additional options.

-If those complete, the script calls ``ok`` to increment the count of
-successful tests and then returns 1 as its result.
+If those complete, the script calls ``test.passes`` to increment the count
+of successful tests.

-The driver.pl script assumes by default that the source Verilog file name
+The driver.py script assumes by default that the source Verilog file name
 matches the test script name. So a test whose driver is
-``t/t_mytest.pl`` will expect a Verilog source file ``t/t_mytest.v``.
+``t/t_mytest.py`` will expect a Verilog source file ``t/t_mytest.v``.
 This can be changed using the ``top_filename`` subroutine, for example

 ::

-   top_filename("t/t_myothertest.v");
+   test.top_filename = "t/t_myothertest.v"

 By default, all tests will run with major simulators (Icarus Verilog, NC,
 VCS, ModelSim, etc.) as well as Verilator, to allow results to be
@@ -2174,26 +2253,25 @@ can use the following:

 ::

-   scenarios(vlt => 1);
+   test.scenarios('vlt')

-Of the many options that can be set through arguments to ``compiler`` and
-``execute``, the following are particularly useful:
+Of the many options that can be set through arguments to ``test.compiler``
+and ``test.execute``, the following are particularly useful:

 ``verilator_flags2``
  A list of flags to be passed to verilator when compiling.

 ``fails``
-  Set to 1 to indicate that the compilation or execution is intended to fail.
+  Set true to indicate that the compilation or execution is intended to fail.

 For example, the following would specify that compilation requires two
 defines and is expected to fail.

 ::

-   compile(
+   test.compile(
      verilator_flags2 => ["-DSMALL_CLOCK -DGATED_COMMENT"],
-      fails => 1,
-      );
+      fails = True)

 Hints On Writing Tests
 ----------------------
@@ -2206,10 +2284,10 @@ same name as the test, but with .cpp as suffix

 ::

-   compile(
-      make_top_shell   => 0,
-      make_main        => 0,
-      verilator_flags2 => ["--exe $Self->{t_dir}/$Self->{name}.cpp"], );
+   test.compile(
+      make_top_shell=False,
+      make_main=False,
+      verilator_flags2=["--exe", test.t_dir + "/" + test.name + ".cpp"])

 Tests should be self-checking, rather than producing lots of output. If a
 test succeeds it should print ``*-* All Finished *-*`` to standard output
@@ -2249,9 +2327,8 @@ compile time, it is the only option. For example:
 ::

   compile(
-      fails => 1,
-      expect_filename => $Self->{golden_filename},
-      );
+      fails=True,
+      expect_filename=test.golden_filename)

 Note ``expect_filename`` strips some debugging information from the logfile
 when comparing.
@@ -2260,9 +2337,9 @@ when comparing.
 Test Language Compile/Lint/Run Arguments
 ----------------------------------------

-This section describes common arguments to ``compile()``, ``lint()``, and
-``run()``.  The full list of arguments can be found by looking at the
-``driver.pl`` source code.
+This section describes common arguments to ``test.compile``, ``test.lint``,
+and ``test.run``.  The full list of arguments can be found by looking at
+the ``driver.py`` source code.

 all_run_flags
  A list of flags to be passed when running the simulator (Verilated model
@@ -2273,11 +2350,6 @@ check_finished
  string ``*-* All Finished *-*`` being printed on standard output. This is
  the normal way for successful tests to finish.

-expect
-  A quoted list of strings or regular expression to be matched in the
-  output. See `Hints On Writing Tests` for more detail on how this argument
-  should be used.
-
 fails
  True to indicate this step is expected to fail.  Tests that are expected
  to fail generally have _bad in their filename.
@@ -20,6 +20,7 @@ Ashutosh
 Ast
 Atmel
 Aurelien
+Badáň
 Bagri
 Balboni
 Baltazar
@@ -40,6 +41,7 @@ Brej
 Briquet
 Brownlee
 Buckenmaier
+Burghoorn
 Bybell
 CLang
 CMake
@@ -73,6 +75,7 @@ Deepa
 Defparams
 Delm
 Denio
+Deniz
 Deprecations
 Deroo
 Desai
@@ -108,8 +111,10 @@ Eugen
 Fabrizio
 Faucher
 Faure
+Feinberg
 Fekete
 Ferrandi
+Filip
 Flachs
 Flavien
 Florian
@@ -131,6 +136,7 @@ Gernot
 Gerst
 Gielda
 Gigerl
+Gijs
 Gileadi
 Giri
 Gisselquist
@@ -147,6 +153,7 @@ Grulfen
 Gu
 Gunter
 Guo
+Güzel
 Hameed
 Hao
 Haojin
@@ -164,6 +171,7 @@ Homebrew
 Hornung
 Hossell
 Hsu
+Huggett
 Hutt
 Hyperthreading
 Ibrahim
@@ -240,6 +248,7 @@ Liwei
 Lockhart
 Longo
 Luca
+Luiza
 Lussier
 Lübeck
 MMD
@@ -270,6 +279,8 @@ MinW
 Mindspeed
 MingW
 Miodrag
+Moberg
+Mobert
 ModelSim
 Modport
 Moinak
@@ -351,6 +362,7 @@ Sasselli
 Scharrer
 Seitz
 Shahid
+Shahrestani
 Shankar
 Shanshan
 Sharad
@@ -361,6 +373,7 @@ Shi
 Shinkarovsky
 Shinya
 Shirakawa
+Shou
 Shuba
 Shunyao
 Slager
@@ -372,10 +385,12 @@ Solaris
 Solomatnikov
 Solt
 Southwell
+Spack
 Srini
 Srinivasan
 Stamness
 Stephane
+Sterin
 Stoddard
 Stroebel
 Strouble
@@ -458,6 +473,7 @@ Wojciech
 Wolfel
 Wouters
 Wpedantic
+Wunused
 Wwarn
 XSim
 Xcelium
@@ -494,6 +510,7 @@ architected
 args
 arrarys
 assertOn
+assertcontrol
 astgen
 async
 atClone
@@ -510,6 +527,7 @@ basename
 bbox
 benchmarking
 biguint
+bindir
 biops
 bisonpre
 bitOpTree
@@ -579,6 +597,7 @@ datadir
 datafiles
 david
 ddd
+de
 deassign
 debugi
 defenv
@@ -586,6 +605,7 @@ defname
 defparam
 demangling
 dep
+deparametrized
 der
 dereference
 desassign
@@ -698,6 +718,7 @@ genvar
 genvars
 getenv
 getline
+getter
 ggdb
 glibc
 gmake
@@ -711,6 +732,7 @@ hh
 hier
 hierCMakeArgs
 hierMkArgs
+hierParameters
 hierVer
 hx
 hyperthreading
@@ -842,6 +864,8 @@ parallelized
 param
 parameterization
 parameterized
+parameterized
+parameterless
 params
 parens
 pawel
@@ -854,6 +878,7 @@ phelter
 picoChip
 pinIndex
 pinout
+pkgdatadir
 plusargs
 pmos
 poping
@@ -885,7 +910,9 @@ profcfunc
 profiler
 prototyptes
 ps
+psprintf
 pthread
+pthreads
 ptr
 pulldown
 pulldowns
@@ -916,6 +943,7 @@ redeclaring
 regs
 reloop
 replaceShiftOp
+reproducibility
 resetall
 respecified
 rodata
@@ -968,6 +996,9 @@ submakes
 submodule
 submodules
 substring
+superclass
+suspendable
+suspendables
 sv
 svBitVal
 svBitVecVal
@@ -978,6 +1009,7 @@ svLogicVal
 svdpi
 svgGetTimeUnit
 swrite
+synthesizeable
 sys
 systemc
 tenghtt
@@ -996,10 +1028,13 @@ toolchain
 topcell
 toplevel
 toupper
+traceCapable
 traceEverOn
 tran
 treei
 tri
+triand
+trior
 tristate
 tristates
 trunc
@@ -1060,6 +1095,7 @@ vluint
 vpi
 vpiConstType
 vpiDefName
+vpiInertialDelay
 vpiInstance
 vpiLeftRange
 vpiModule
@@ -1074,6 +1110,7 @@ warmup
 waveforms
 whitespace
 widthed
+wor
 wreal
 writeb
 writeme
@@ -24,8 +24,11 @@ cmake_policy(SET CMP0074 NEW)
 project(cmake_hello_c)

 find_package(verilator HINTS $ENV{VERILATOR_ROOT} ${VERILATOR_ROOT})
-if (NOT verilator_FOUND)
-  message(FATAL_ERROR "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable")
+if(NOT verilator_FOUND)
+    message(
+        FATAL_ERROR
+        "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable"
+    )
 endif()

 # Create a new executable target that will contain all your sources
@@ -35,4 +38,5 @@ target_compile_features(example PUBLIC cxx_std_14)
 # Add the Verilated circuit to the target
 verilate(example
  INCLUDE_DIRS "../make_hello_c"
-  SOURCES ../make_hello_c/top.v)
+  SOURCES ../make_hello_c/top.v
+)
@@ -24,8 +24,11 @@ cmake_policy(SET CMP0074 NEW)
 project(cmake_hello_sc CXX)

 find_package(verilator HINTS $ENV{VERILATOR_ROOT} ${VERILATOR_ROOT})
-if (NOT verilator_FOUND)
-  message(FATAL_ERROR "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable")
+if(NOT verilator_FOUND)
+    message(
+        FATAL_ERROR
+        "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable"
+    )
 endif()

 # SystemC dependencies
@@ -39,14 +42,12 @@ find_package(SystemCLanguage QUIET)
 add_executable(example ../make_hello_sc/sc_main.cpp)
 target_compile_features(example PUBLIC cxx_std_14)

-set_property(
-  TARGET example
-  PROPERTY CXX_STANDARD ${SystemC_CXX_STANDARD}
-  )
+set_property(TARGET example PROPERTY CXX_STANDARD ${SystemC_CXX_STANDARD})

 # Add the Verilated circuit to the target
 verilate(example SYSTEMC
  INCLUDE_DIRS "../make_hello_sc"
-  SOURCES ../make_hello_sc/top.v)
+  SOURCES ../make_hello_sc/top.v
+)

 verilator_link_systemc(example)
@@ -24,8 +24,11 @@ cmake_policy(SET CMP0074 NEW)
 project(cmake_protect_lib)

 find_package(verilator HINTS $ENV{VERILATOR_ROOT} ${VERILATOR_ROOT})
-if (NOT verilator_FOUND)
-  message(FATAL_ERROR "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable")
+if(NOT verilator_FOUND)
+    message(
+        FATAL_ERROR
+        "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable"
+    )
 endif()

 # Create the main executable target
@@ -55,9 +58,11 @@ verilate(verilated_secret
  VERILATOR_ARGS --protect-lib verilated_secret
                 --protect-key ${PROTECT_KEY}
  DIRECTORY ${CMAKE_CURRENT_BINARY_DIR}/verilated_secret
-  SOURCES ../make_protect_lib/secret_impl.v)
+  SOURCES ../make_protect_lib/secret_impl.v
+)

 # Include location of verilated_secret.sv wrapper
 verilate(example
  VERILATOR_ARGS "-I${CMAKE_CURRENT_BINARY_DIR}/verilated_secret"
-  SOURCES ../make_protect_lib/top.v)
+  SOURCES ../make_protect_lib/top.v
+)
@@ -24,8 +24,11 @@ cmake_policy(SET CMP0074 NEW)
 project(cmake_tracing_c)

 find_package(verilator HINTS $ENV{VERILATOR_ROOT} ${VERILATOR_ROOT})
-if (NOT verilator_FOUND)
-  message(FATAL_ERROR "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable")
+if(NOT verilator_FOUND)
+    message(
+        FATAL_ERROR
+        "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable"
+    )
 endif()

 # Create a new executable target that will contain all your sources
@@ -36,4 +39,5 @@ target_compile_features(example PUBLIC cxx_std_14)
 verilate(example COVERAGE TRACE
  INCLUDE_DIRS "../make_tracing_c"
  VERILATOR_ARGS -f ../make_tracing_c/input.vc -x-assign fast
-  SOURCES ../make_tracing_c/top.v)
+  SOURCES ../make_tracing_c/top.v
+)
@@ -25,8 +25,11 @@ cmake_policy(SET CMP0074 NEW)
 project(cmake_tracing_sc_example CXX)

 find_package(verilator HINTS $ENV{VERILATOR_ROOT} ${VERILATOR_ROOT})
-if (NOT verilator_FOUND)
-  message(FATAL_ERROR "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable")
+if(NOT verilator_FOUND)
+    message(
+        FATAL_ERROR
+        "Verilator was not found. Either install it, or set the VERILATOR_ROOT environment variable"
+    )
 endif()

 # SystemC dependencies
@@ -40,15 +43,13 @@ find_package(SystemCLanguage QUIET)
 add_executable(example ../make_tracing_sc/sc_main.cpp)
 target_compile_features(example PUBLIC cxx_std_14)

-set_property(
-  TARGET example
-  PROPERTY CXX_STANDARD ${SystemC_CXX_STANDARD}
-  )
+set_property(TARGET example PROPERTY CXX_STANDARD ${SystemC_CXX_STANDARD})

 # Add the Verilated circuit to the target
 verilate(example SYSTEMC COVERAGE TRACE
  INCLUDE_DIRS "../make_tracing_sc"
  VERILATOR_ARGS -f ../make_tracing_sc/input.vc -x-assign fast
-  SOURCES ../make_tracing_sc/top.v)
+  SOURCES ../make_tracing_sc/top.v
+)

 verilator_link_systemc(example)
@@ -13,6 +13,6 @@ module sub
    );

   // Some simple logic
-   always_comb out = ~ in;
+   always_comb out = ~in;

 endmodule
@@ -25,8 +25,7 @@ class VlFileCopy:

        self.debug = debug

-        with NamedTemporaryFile() as tree_temp, NamedTemporaryFile(
-        ) as meta_temp:
+        with NamedTemporaryFile() as tree_temp, NamedTemporaryFile() as meta_temp:
            vargs = [
                '--json-only-output',
                tree_temp.name,
@@ -61,8 +60,7 @@ class VlFileCopy:
            print("\t%s " % command)
        status = subprocess.call(command, shell=True)
        if status != 0:
-            raise RuntimeError("Command failed running Verilator with '" +
-                               command + "', stopped")
+            raise RuntimeError("Command failed running Verilator with '" + command + "', stopped")


 #######################################################################
@@ -71,8 +69,7 @@ if __name__ == '__main__':
    parser = argparse.ArgumentParser(
        allow_abbrev=False,
        formatter_class=argparse.RawTextHelpFormatter,
-        description=
-        """Example of using Verilator JSON output to copy a list of files to an
+        description="""Example of using Verilator JSON output to copy a list of files to an
 output directory (-odir, defaults to 'copied'), e.g. to easily create a
 tarball of the design to pass to others.

@@ -95,11 +92,7 @@ This file ONLY is placed under the Creative Commons Public Domain, for
 any use, without warranty, 2019 by Wilson Snyder.
 SPDX-License-Identifier: CC0-1.0
 """)
-    parser.add_argument('-debug',
-                        '--debug',
-                        action='store_const',
-                        const=9,
-                        help='enable debug')
+    parser.add_argument('-debug', '--debug', action='store_const', const=9, help='enable debug')
    parser.add_argument('-odir',
                        '--odir',
                        action='store',
@@ -108,9 +101,7 @@ SPDX-License-Identifier: CC0-1.0
                        help='target output directory')
    (args, rem) = parser.parse_known_args()

-    print(
-        "NOTE: vl_file_copy is only an example starting point for writing your own tool."
-    )
+    print("NOTE: vl_file_copy is only an example starting point for writing your own tool.")
    # That is:
    # 1. We will accept basic patches
    # 2. We are not expecting to make this globally useful. (e.g. we don't cleanup obj_dir)
@@ -24,8 +24,7 @@ class VlHierGraph:
        self.next_vertex_number = 0
        self.addr_to_number = {}

-        with NamedTemporaryFile() as tree_temp, NamedTemporaryFile(
-        ) as meta_temp:
+        with NamedTemporaryFile() as tree_temp, NamedTemporaryFile() as meta_temp:
            vargs = [
                '--json-only-output',
                tree_temp.name,
@@ -45,9 +44,7 @@ class VlHierGraph:
            fh.write("digraph {\n")
            fh.write("  dpi=300;\n")
            fh.write("  order=LR;\n")
-            fh.write(
-                "  node [fontsize=8 shape=\"box\" margin=0.01 width=0 height=0]"
-            )
+            fh.write("  node [fontsize=8 shape=\"box\" margin=0.01 width=0 height=0]")
            fh.write("  edge [fontsize=6]")
            # Find cells
            modules = self.flatten(self.tree, lambda n: n['type'] == "MODULE")
@@ -101,8 +98,7 @@ class VlHierGraph:
            print("\t%s " % command)
        status = subprocess.call(command, shell=True)
        if status != 0:
-            raise RuntimeError("Command failed running Verilator with '" +
-                               command + "', stopped")
+            raise RuntimeError("Command failed running Verilator with '" + command + "', stopped")


 #######################################################################
@@ -111,8 +107,7 @@ if __name__ == '__main__':
    parser = argparse.ArgumentParser(
        allow_abbrev=False,
        formatter_class=argparse.RawTextHelpFormatter,
-        description=
-        """Example of using Verilator JSON output to create a .dot file showing the
+        description="""Example of using Verilator JSON output to create a .dot file showing the
 design module hierarchy.

 Example usage:
@@ -134,11 +129,7 @@ This file ONLY is placed under the Creative Commons Public Domain, for
 any use, without warranty, 2019 by Wilson Snyder.
 SPDX-License-Identifier: CC0-1.0
 """)
-    parser.add_argument('-debug',
-                        '--debug',
-                        action='store_const',
-                        const=9,
-                        help='enable debug')
+    parser.add_argument('-debug', '--debug', action='store_const', const=9, help='enable debug')
    parser.add_argument('-o',
                        '--o',
                        action='store',
@@ -147,18 +138,14 @@ SPDX-License-Identifier: CC0-1.0
                        help='output filename')
    (args, rem) = parser.parse_known_args()

-    print(
-        "NOTE: vl_hier_graph is only an example starting point for writing your own tool."
-    )
+    print("NOTE: vl_hier_graph is only an example starting point for writing your own tool.")
    # That is:
    # 1. We will accept basic patches
    # 2. We are not expecting to make this globally useful. (e.g. we don't cleanup obj_dir)
    # 3. "make install" will not install this.
    # 4. This has not had production-worthy validation.

-    fc = VlHierGraph(output_filename=args.o,
-                     debug=args.debug,
-                     verilator_args=rem)
+    fc = VlHierGraph(output_filename=args.o, debug=args.debug, verilator_args=rem)

 ######################################################################
 # Local Variables:
@@ -47,12 +47,12 @@ int main(int argc, char** argv) {

    // Simulate until $finish
    while (!contextp->gotFinish()) {
-        contextp->timeInc(1);
        top->clk = ~top->clk & 0x1;
        top->eval();
 #if VM_TRACE
        if (tfp) tfp->dump(contextp->time());
 #endif
+        contextp->timeInc(1);
    }

    // Final model cleanup
@@ -13,7 +13,7 @@ module sub

   // Example counter/flop
   reg [31:0] count_c;
-   always_ff @ (posedge clk) begin
+   always_ff @(posedge clk) begin
      if (!reset_l) begin
         /*AUTORESET*/
         // Beginning of autoreset for uninitialized flops
@@ -32,11 +32,11 @@ module sub
   end

   // An example assertion
-   always_ff @ (posedge clk) begin
-      AssertionExample: assert (!reset_l || count_c<100);
+   always_ff @(posedge clk) begin
+      AssertionExample : assert (!reset_l || count_c < 100);
   end

   // And example coverage analysis
-   cover property (@(posedge clk) count_c==3);
+   cover property (@(posedge clk) count_c == 3);

 endmodule
@@ -14,7 +14,7 @@ module sub

   // Example counter/flop
   reg [31:0] count_f;
-   always_ff @ (posedge fastclk) begin
+   always_ff @(posedge fastclk) begin
      if (!reset_l) begin
         /*AUTORESET*/
         // Beginning of autoreset for uninitialized flops
@@ -28,7 +28,7 @@ module sub

   // Another example flop
   reg [31:0] count_c;
-   always_ff @ (posedge clk) begin
+   always_ff @(posedge clk) begin
      if (!reset_l) begin
         /*AUTORESET*/
         // Beginning of autoreset for uninitialized flops
@@ -38,8 +38,7 @@ module sub
      else begin
         count_c <= count_c + 1;
         if (count_c >= 3) begin
-            $display("[%0t] fastclk is %0d times faster than clk\n",
-                     $time, count_f/count_c);
+            $display("[%0t] fastclk is %0d times faster than clk\n", $time, count_f / count_c);
            // This write is a magic value the Makefile uses to make sure the
            // test completes successfully.
            $write("*-* All Finished *-*\n");
@@ -49,11 +48,11 @@ module sub
   end

   // An example assertion
-   always_ff @ (posedge clk) begin
-      AssertionExample: assert(!reset_l || count_c<100);
+   always_ff @(posedge clk) begin
+      AssertionExample : assert (!reset_l || count_c < 100);
   end

   // And example coverage analysis
-   cover property (@(posedge clk) count_c==3);
+   cover property (@(posedge clk) count_c == 3);

 endmodule
@@ -130,6 +130,7 @@ void **JenkinsIns(void *base_i, const unsigned char *mem, uint32_t length, uint3
 #define FST_HDR_TIMEZERO_SIZE           (8)
 #define FST_GZIO_LEN                    (32768)
 #define FST_HDR_FOURPACK_DUO_SIZE       (4*1024*1024)
+#define FST_ZWRAPPER_HDR_SIZE           (1+8+8)

 #if defined(__APPLE__) && defined(__MACH__)
 #define FST_MACOSX
@@ -347,8 +348,9 @@ return(NULL);

 static void *fstMmap2(size_t __len, int __fd, fst_off_t __off)
 {
+DWORD64 len64 = __len;  /* Must be 64-bit for shift below */
 HANDLE handle = CreateFileMapping((HANDLE)_get_osfhandle(__fd), NULL,
-				  PAGE_READWRITE, (DWORD)(__len >> 32),
+				  PAGE_READWRITE, (DWORD)(len64 >> 32),
 				  (DWORD)__len, NULL);
 if (!handle) { return NULL; }

@@ -548,8 +550,9 @@ return(rc);

 static uint32_t fstReaderVarint32(FILE *f)
 {
-int chk_len = 5; /* TALOS-2023-1783 */
-unsigned char buf[chk_len];
+const int chk_len_max = 5; /* TALOS-2023-1783 */
+int chk_len = chk_len_max;
+unsigned char buf[chk_len_max];
 unsigned char *mem = buf;
 uint32_t rc = 0;
 int ch;
@@ -580,8 +583,9 @@ return(rc);

 static uint32_t fstReaderVarint32WithSkip(FILE *f, uint32_t *skiplen)
 {
-int chk_len = 5; /* TALOS-2023-1783 */
-unsigned char buf[chk_len];
+const int chk_len_max = 5; /* TALOS-2023-1783 */
+int chk_len = chk_len_max;
+unsigned char buf[chk_len_max];
 unsigned char *mem = buf;
 uint32_t rc = 0;
 int ch;
@@ -613,8 +617,9 @@ return(rc);

 static uint64_t fstReaderVarint64(FILE *f)
 {
-int chk_len = 16; /* TALOS-2023-1783 */
-unsigned char buf[chk_len];
+const int chk_len_max = 16; /* TALOS-2023-1783 */
+int chk_len = chk_len_max;
+unsigned char buf[chk_len_max];
 unsigned char *mem = buf;
 uint64_t rc = 0;
 int ch;
@@ -4632,9 +4637,14 @@ if(sectype == FST_BL_ZWRAPPER)
                }
 #endif

-        fstReaderFseeko(xc, xc->f, 1+8+8, SEEK_SET);
+        fstReaderFseeko(xc, xc->f, FST_ZWRAPPER_HDR_SIZE, SEEK_SET);
 #ifndef __MINGW32__
        fflush(xc->f);
+#else
+	/* Windows UCRT runtime library reads one byte ahead in the file
+	   even with buffering disabled and does not synchronise the
+	   file position after fseek. */
+	_lseek(fileno(xc->f), FST_ZWRAPPER_HDR_SIZE, SEEK_SET);
 #endif

        zfd = dup(fileno(xc->f));
@@ -6184,7 +6194,9 @@ struct fstReaderContext *xc = (struct fstReaderContext *)ctx;
 fst_off_t blkpos = 0, prev_blkpos;
 uint64_t beg_tim, end_tim, beg_tim2, end_tim2;
 int sectype;
+#ifdef FST_DEBUG
 unsigned int secnum = 0;
+#endif
 uint64_t seclen;
 uint64_t tsec_uclen = 0, tsec_clen = 0;
 uint64_t tsec_nitems;
@@ -6276,7 +6288,9 @@ for(;;)
                }

        blkpos += seclen;
+#ifdef FST_DEBUG
        secnum++;
+#endif
        }

 xc->rvat_beg_tim = beg_tim;
@@ -82,6 +82,12 @@

 #include "verilated_trace.h"

+#ifdef VM_SOLVER_DEFAULT
+#define VL_SOLVER_DEFAULT VM_SOLVER_DEFAULT
+#else
+#define VL_SOLVER_DEFAULT "z3 --in"
+#endif
+
 // Max characters in static char string for VL_VALUE_STRING
 constexpr unsigned VL_VALUE_STRING_MAX_WIDTH = 8192;

@@ -122,6 +128,7 @@ void vl_finish(const char* filename, int linenum, const char* hier) VL_MT_UNSAFE

 #ifndef VL_USER_STOP  ///< Define this to override the vl_stop function
 void vl_stop(const char* filename, int linenum, const char* hier) VL_MT_UNSAFE {
+    // $stop or $fatal reporting; would break current API to add param as to which
    const char* const msg = "Verilog $stop";
    Verilated::threadContextp()->gotError(true);
    Verilated::threadContextp()->gotFinish(true);
@@ -166,6 +173,7 @@ void vl_fatal(const char* filename, int linenum, const char* hier, const char* m

 #ifndef VL_USER_STOP_MAYBE  ///< Define this to override the vl_stop_maybe function
 void vl_stop_maybe(const char* filename, int linenum, const char* hier, bool maybe) VL_MT_UNSAFE {
+    // $stop or $fatal
    Verilated::threadContextp()->errorCountInc();
    if (maybe
        && Verilated::threadContextp()->errorCount() < Verilated::threadContextp()->errorLimit()) {
@@ -406,6 +414,8 @@ IData VL_RAND_RESET_I(int obits) VL_MT_SAFE {
    data &= VL_MASK_I(obits);
    return data;
 }
+IData VL_RAND_RESET_ASSIGN_I(int obits) VL_MT_SAFE { return VL_RANDOM_I() & VL_MASK_I(obits); }
+
 QData VL_RAND_RESET_Q(int obits) VL_MT_SAFE {
    if (Verilated::threadContextp()->randReset() == 0) return 0;
    QData data = ~0ULL;
@@ -415,11 +425,19 @@ QData VL_RAND_RESET_Q(int obits) VL_MT_SAFE {
    data &= VL_MASK_Q(obits);
    return data;
 }
+
+QData VL_RAND_RESET_ASSIGN_Q(int obits) VL_MT_SAFE { return VL_RANDOM_Q() & VL_MASK_Q(obits); }
+
 WDataOutP VL_RAND_RESET_W(int obits, WDataOutP outwp) VL_MT_SAFE {
    for (int i = 0; i < VL_WORDS_I(obits) - 1; ++i) outwp[i] = VL_RAND_RESET_I(32);
    outwp[VL_WORDS_I(obits) - 1] = VL_RAND_RESET_I(32) & VL_MASK_E(obits);
    return outwp;
 }
+WDataOutP VL_RAND_RESET_ASSIGN_W(int obits, WDataOutP outwp) VL_MT_SAFE {
+    for (int i = 0; i < VL_WORDS_I(obits) - 1; ++i) outwp[i] = VL_RAND_RESET_ASSIGN_I(32);
+    outwp[VL_WORDS_I(obits) - 1] = VL_RAND_RESET_ASSIGN_I(32) & VL_MASK_E(obits);
+    return outwp;
+}
 WDataOutP VL_ZERO_RESET_W(int obits, WDataOutP outwp) VL_MT_SAFE {
    // Not inlined to speed up compilation of slowpath code
    return VL_ZERO_W(obits, outwp);
@@ -1171,8 +1189,8 @@ static char* _vl_vsss_read_bin(FILE* fp, int& floc, const WDataInP fromp, const
 static void _vl_vsss_setbit(WDataOutP iowp, int obits, int lsb, int nbits, IData ld) VL_MT_SAFE {
    for (; nbits && lsb < obits; nbits--, lsb++, ld >>= 1) VL_ASSIGNBIT_WI(lsb, iowp, ld & 1);
 }
-static void _vl_vsss_based(WDataOutP owp, int obits, int baseLog2, const char* strp,
-                           size_t posstart, size_t posend) VL_MT_SAFE {
+void _vl_vsss_based(WDataOutP owp, int obits, int baseLog2, const char* strp, size_t posstart,
+                    size_t posend) VL_MT_SAFE {
    // Read in base "2^^baseLog2" digits from strp[posstart..posend-1] into owp of size obits.
    VL_ZERO_W(obits, owp);
    int lsb = 0;
@@ -1747,9 +1765,11 @@ IData VL_SYSTEM_IQ(QData lhs) VL_MT_SAFE {
    return VL_SYSTEM_IW(VL_WQ_WORDS_E, lhsw);
 }
 IData VL_SYSTEM_IW(int lhswords, const WDataInP lhsp) VL_MT_SAFE {
-    char filenamez[VL_VALUE_STRING_MAX_CHARS + 1];
-    _vl_vint_to_string(lhswords * VL_EDATASIZE, filenamez, lhsp);
-    const int code = std::system(filenamez);  // Yes, std::system() is threadsafe
+    const std::string lhs = VL_CVT_PACK_STR_NW(lhswords, lhsp);
+    return VL_SYSTEM_IN(lhs);
+}
+IData VL_SYSTEM_IN(const std::string& lhs) VL_MT_SAFE {
+    const int code = std::system(lhs.c_str());  // Yes, std::system() is threadsafe
    return code >> 8;  // Want exit status
 }

@@ -1896,20 +1916,16 @@ std::string VL_TOUPPER_NN(const std::string& ld) VL_PURE {

 std::string VL_CVT_PACK_STR_NW(int lwords, const WDataInP lwp) VL_PURE {
    // See also _vl_vint_to_string
-    char destout[VL_VALUE_STRING_MAX_CHARS + 1];
+    std::string result;
+    result.reserve((lwords * VL_EDATASIZE) / 8 + 1);
    const int obits = lwords * VL_EDATASIZE;
    int lsb = obits - 1;
-    char* destp = destout;
-    size_t len = 0;
    for (; lsb >= 0; --lsb) {
        lsb = (lsb / 8) * 8;  // Next digit
        const IData charval = VL_BITRSHIFT_W(lwp, lsb) & 0xff;
-        if (charval) {
-            *destp++ = static_cast<char>(charval);
-            ++len;
-        }
+        if (charval) result += static_cast<char>(charval);
    }
-    return std::string{destout, len};
+    return result;
 }

 std::string VL_CVT_PACK_STR_ND(const VlQueue<std::string>& q) VL_PURE {
@@ -2037,22 +2053,26 @@ bool VlReadMem::get(QData& addrr, std::string& valuer) {
    if (VL_UNLIKELY(!m_fp)) return false;
    valuer = "";
    // Prep for reading
-    bool indata = false;
-    bool ignore_to_eol = false;
-    bool ignore_to_cmt = false;
-    bool reading_addr = false;
-    int lastc = ' ';
+    bool inData = false;
+    bool ignoreToEol = false;
+    bool ignoreToComment = false;
+    bool readingAddress = false;
+    int lastCh = ' ';
    // Read the data
    // We process a character at a time, as then we don't need to deal
    // with changing buffer sizes dynamically, etc.
    while (true) {
        int c = std::fgetc(m_fp);
        if (VL_UNLIKELY(c == EOF)) break;
+        const bool chIs4StateBin
+            = c == '0' || c == '1' || c == 'x' || c == 'X' || c == 'z' || c == 'Z';
+        const bool chIs2StateHex = std::isxdigit(c);
+        const bool chIs4StateHex = std::isxdigit(c) || chIs4StateBin;
        // printf("%d: Got '%c' Addr%lx IN%d IgE%d IgC%d\n",
-        //        m_linenum, c, m_addr, indata, ignore_to_eol, ignore_to_cmt);
+        //        m_linenum, c, m_addr, inData, ignoreToEol, ignoreToComment);
        // See if previous data value has completed, and if so return
        if (c == '_') continue;  // Ignore _ e.g. inside a number
-        if (indata && !std::isxdigit(c) && c != 'x' && c != 'X') {
+        if (inData && !chIs4StateHex) {
            // printf("Got data @%lx = %s\n", m_addr, valuer.c_str());
            ungetc(c, m_fp);
            addrr = m_addr;
@@ -2062,50 +2082,46 @@ bool VlReadMem::get(QData& addrr, std::string& valuer) {
        // Parse line
        if (c == '\n') {
            ++m_linenum;
-            ignore_to_eol = false;
-            reading_addr = false;
+            ignoreToEol = false;
+            readingAddress = false;
        } else if (c == '\t' || c == ' ' || c == '\r' || c == '\f') {
-            reading_addr = false;
+            readingAddress = false;
        }
        // Skip // comments and detect /* comments
-        else if (ignore_to_cmt && lastc == '*' && c == '/') {
-            ignore_to_cmt = false;
-            reading_addr = false;
-        } else if (!ignore_to_eol && !ignore_to_cmt) {
-            if (lastc == '/' && c == '*') {
-                ignore_to_cmt = true;
-            } else if (lastc == '/' && c == '/') {
-                ignore_to_eol = true;
+        else if (ignoreToComment && lastCh == '*' && c == '/') {
+            ignoreToComment = false;
+            readingAddress = false;
+        } else if (!ignoreToEol && !ignoreToComment) {
+            if (lastCh == '/' && c == '*') {
+                ignoreToComment = true;
+            } else if (lastCh == '/' && c == '/') {
+                ignoreToEol = true;
            } else if (c == '/') {  // Part of /* or //
            } else if (c == '#') {
-                ignore_to_eol = true;
+                ignoreToEol = true;
            } else if (c == '@') {
-                reading_addr = true;
+                readingAddress = true;
                m_anyAddr = true;
                m_addr = 0;
-            }
-            // Check for hex or binary digits as file format requests
-            else if (std::isxdigit(c) || (!reading_addr && (c == 'x' || c == 'X'))) {
+            } else if (readingAddress && chIs2StateHex) {
                c = std::tolower(c);
-                const int value
-                    = (c >= 'a' ? (c == 'x' ? VL_RAND_RESET_I(4) : (c - 'a' + 10)) : (c - '0'));
-                if (reading_addr) {
-                    // Decode @ addresses
-                    m_addr = (m_addr << 4) + value;
-                } else {
-                    indata = true;
-                    valuer += static_cast<char>(c);
-                    // printf(" Value width=%d  @%x = %c\n", width, m_addr, c);
-                    if (VL_UNLIKELY(value > 1 && !m_hex)) {
-                        VL_FATAL_MT(m_filename.c_str(), m_linenum, "",
-                                    "$readmemb (binary) file contains hex characters");
-                    }
+                const int addressValue = (c >= 'a') ? (c - 'a' + 10) : (c - '0');
+                m_addr = (m_addr << 4) + addressValue;
+            } else if (readingAddress && chIs4StateHex) {
+                VL_FATAL_MT(m_filename.c_str(), m_linenum, "",
+                            "$readmem address contains 4-state characters");
+            } else if (chIs4StateHex) {
+                inData = true;
+                valuer += static_cast<char>(c);
+                if (VL_UNLIKELY(!m_hex && !chIs4StateBin)) {
+                    VL_FATAL_MT(m_filename.c_str(), m_linenum, "",
+                                "$readmemb (binary) file contains hex characters");
                }
            } else {
                VL_FATAL_MT(m_filename.c_str(), m_linenum, "", "$readmem file syntax error");
            }
        }
-        lastc = c;
+        lastCh = c;
    }

    if (VL_UNLIKELY(m_end != ~0ULL && m_addr <= m_end && !m_anyAddr)) {
@@ -2114,7 +2130,7 @@ bool VlReadMem::get(QData& addrr, std::string& valuer) {
    }

    addrr = m_addr;
-    return indata;  // EOF
+    return inData;  // EOF
 }
 void VlReadMem::setData(void* valuep, const std::string& rhs) {
    const QData shift = m_hex ? 4ULL : 1ULL;
@@ -2122,8 +2138,9 @@ void VlReadMem::setData(void* valuep, const std::string& rhs) {
    // Shift value in
    for (const auto& i : rhs) {
        const char c = std::tolower(i);
-        const int value
-            = (c >= 'a' ? (c == 'x' ? VL_RAND_RESET_I(4) : (c - 'a' + 10)) : (c - '0'));
+        const int value = (c == 'x' || c == 'z') ? VL_RAND_RESET_I(m_hex ? 4 : 1)
+                          : (c >= 'a')           ? (c - 'a' + 10)
+                                                 : (c - '0');
        if (m_bits <= 8) {
            CData* const datap = reinterpret_cast<CData*>(valuep);
            if (!innum) *datap = 0;
@@ -2457,6 +2474,7 @@ VerilatedContext::VerilatedContext()
    m_ns.m_coverageFilename = "coverage.dat";
    m_ns.m_profExecFilename = "profile_exec.dat";
    m_ns.m_profVltFilename = "profile.vlt";
+    m_ns.m_solverProgram = VlOs::getenvStr("VERILATOR_SOLVER", VL_SOLVER_DEFAULT);
    m_fdps.resize(31);
    std::fill(m_fdps.begin(), m_fdps.end(), static_cast<FILE*>(nullptr));
    m_fdFreeMct.resize(30);
@@ -2483,9 +2501,47 @@ VerilatedContext::Serialized::Serialized() {
    m_timeprecision = picosecond;  // Initial value until overridden by _Vconfigure
 }

+bool VerilatedContext::assertOn() const VL_MT_SAFE { return m_s.m_assertOn; }
 void VerilatedContext::assertOn(bool flag) VL_MT_SAFE {
-    const VerilatedLockGuard lock{m_mutex};
-    m_s.m_assertOn = flag;
+    // Set all assert and directive types when true, clear otherwise.
+    m_s.m_assertOn = VL_MASK_I(ASSERT_ON_WIDTH) * flag;
+}
+bool VerilatedContext::assertOnGet(VerilatedAssertType_t type,
+                                   VerilatedAssertDirectiveType_t directive) const VL_MT_SAFE {
+    // Check if selected directive type bit in the assertOn is enabled for assertion type.
+    // Note: it is assumed that this is checked only for one type at the time.
+
+    // Flag unspecified assertion types as disabled.
+    if (type == 0) return false;
+
+    // Get index of 3-bit group guarding assertion type status.
+    // Since the assertOnGet is generated __always__ for a single assert type, we assume that only
+    // a single bit will be set. Thus, ceil log2 will work fine.
+    VL_DEBUG_IFDEF(assert((type & (type - 1)) == 0););
+    const IData typeMaskPosition = VL_CLOG2_I(type);
+
+    // Check if directive type bit is enabled in corresponding assertion type bits.
+    return m_s.m_assertOn & (directive << (typeMaskPosition * ASSERT_DIRECTIVE_TYPE_MASK_WIDTH));
+}
+void VerilatedContext::assertOnSet(VerilatedAssertType_t types,
+                                   VerilatedAssertDirectiveType_t directives) VL_MT_SAFE {
+    // For each assertion type, set directive bits.
+
+    // Iterate through all positions of assertion type bits. If bit for this assertion type is set,
+    // set directive type bits mask at this group index.
+    for (int i = 0; i < std::numeric_limits<VerilatedAssertType_t>::digits; ++i) {
+        if (VL_BITISSET_I(types, i))
+            m_s.m_assertOn |= directives << (i * ASSERT_DIRECTIVE_TYPE_MASK_WIDTH);
+    }
+}
+void VerilatedContext::assertOnClear(VerilatedAssertType_t types,
+                                     VerilatedAssertDirectiveType_t directives) VL_MT_SAFE {
+    // Iterate through all positions of assertion type bits. If bit for this assertion type is set,
+    // clear directive type bits mask at this group index.
+    for (int i = 0; i < std::numeric_limits<VerilatedAssertType_t>::digits; ++i) {
+        if (VL_BITISSET_I(types, i))
+            m_s.m_assertOn &= ~(directives << (i * ASSERT_DIRECTIVE_TYPE_MASK_WIDTH));
+    }
 }
 void VerilatedContext::calcUnusedSigs(bool flag) VL_MT_SAFE {
    const VerilatedLockGuard lock{m_mutex};
@@ -2567,6 +2623,14 @@ std::string VerilatedContext::profVltFilename() const VL_MT_SAFE {
    const VerilatedLockGuard lock{m_mutex};
    return m_ns.m_profVltFilename;
 }
+void VerilatedContext::solverProgram(const std::string& flag) VL_MT_SAFE {
+    const VerilatedLockGuard lock{m_mutex};
+    m_ns.m_solverProgram = flag;
+}
+std::string VerilatedContext::solverProgram() const VL_MT_SAFE {
+    const VerilatedLockGuard lock{m_mutex};
+    return m_ns.m_solverProgram;
+}
 void VerilatedContext::quiet(bool flag) VL_MT_SAFE {
    const VerilatedLockGuard lock{m_mutex};
    m_s.m_quiet = flag;
@@ -2645,8 +2709,16 @@ void VerilatedContext::addModel(VerilatedModel* modelp) {
        m_ns.m_cpuTimeStart.start();
        m_ns.m_wallTimeStart.start();
    }
-    threadPoolp();  // Ensure thread pool is created, so m_threads cannot change any more

+    // We look for time passing, as opposed to post-eval(), as embedded
+    // models might get added inside initial blocks.
+    if (VL_UNLIKELY(time()))
+        VL_FATAL_MT(
+            "", 0, "",
+            "Adding model when time is non-zero. ... Suggest check time(), or for restarting"
+            " model use a new VerilatedContext");
+
+    threadPoolp();  // Ensure thread pool is created, so m_threads cannot change any more
    m_threadsInModels += modelp->threads();
    if (VL_UNLIKELY(modelp->threads() > m_threads)) {
        std::ostringstream msg;
@@ -2882,7 +2954,7 @@ void VerilatedContext::statsPrintSummary() VL_MT_UNSAFE {
    VL_PRINTF("- Verilator: %s at %s; walltime %0.3f s; speed %s/s\n", endwhy.c_str(),
              simtime.c_str(), walltime, simtimePerf.c_str());
    const double modelMB = VlOs::memUsageBytes() / 1024.0 / 1024.0;
-    VL_PRINTF("- Verilator: cpu %0.3f s on %d threads; alloced %0.0f MB\n", cputime,
+    VL_PRINTF("- Verilator: cpu %0.3f s on %u threads; alloced %0.0f MB\n", cputime,
              threadsInModels(), modelMB);
 }

@@ -2927,13 +2999,12 @@ const VerilatedScopeNameMap* VerilatedContext::scopeNameMap() VL_MT_SAFE {
 //======================================================================
 // VerilatedContext:: Methods - trace

-void VerilatedContext::trace(VerilatedTraceBaseC* tfp, int levels, int options) VL_MT_SAFE {
+void VerilatedContext::trace(VerilatedTraceBaseC* tfp, int levels, int options) {
    VL_DEBUG_IF(VL_DBG_MSGF("+ VerilatedContext::trace\n"););
    if (tfp->isOpen()) {
        VL_FATAL_MT("", 0, "",
                    "Testbench C call to 'VerilatedContext::trace()' must not be called"
                    " after 'VerilatedTrace*::open()'\n");
-        return;
    }
    {
        // Legacy usage may call {modela}->trace(...) then {modelb}->trace(...)
@@ -3126,12 +3197,21 @@ void Verilated::stackCheck(QData needSize) VL_MT_UNSAFE {
        if (haveSize == RLIM_INFINITY) haveSize = 0;
    }
    // VL_PRINTF_MT("-Info: stackCheck(%" PRIu64 ") have %" PRIu64 "\n", needSize, haveSize);
-    // Check for 1.5x need, but suggest 2x so small model increase won't cause warning
-    // if the user follows the suggestions
-    if (VL_UNLIKELY(haveSize && needSize && haveSize < (needSize + needSize / 2))) {
-        VL_PRINTF_MT("%%Warning: System has stack size %" PRIu64 " kb"
-                     " which may be too small; suggest 'ulimit -s %" PRIu64 "' or larger\n",
-                     haveSize / 1024, (needSize * 2) / 1024);
+    // Check and request for 1.5x need. This is automated so the user doesn't need to do anything.
+    QData requestSize = needSize + needSize / 2;
+    if (VL_UNLIKELY(haveSize && needSize && haveSize < requestSize)) {
+        // Try to increase the stack limit to the requested size
+        rlim.rlim_cur = requestSize;
+        if (
+#ifdef _VL_TEST_RLIMIT_FAIL
+            true ||
+#endif
+            setrlimit(RLIMIT_STACK, &rlim)) {
+            VL_PRINTF_MT("%%Warning: System has stack size %" PRIu64 " kb"
+                         " which may be too small; failed to request more"
+                         " using 'ulimit -s %" PRIu64 "'\n",
+                         haveSize / 1024, requestSize);
+        }
    }
 #else
    (void)needSize;  // Unused argument
@@ -55,6 +55,7 @@
 #include <cstring>
 #include <deque>
 #include <functional>
+#include <limits>
 #include <map>
 #include <memory>
 #include <set>
@@ -153,6 +154,27 @@ enum VerilatedVarFlags {
    VLVF_DPI_CLAY = (1 << 10)  // DPI compatible C standard layout
 };

+// IEEE 1800-2023 Table 20-6
+enum class VerilatedAssertType : uint8_t {
+    ASSERT_TYPE_CONCURRENT = (1 << 0),
+    ASSERT_TYPE_SIMPLE_IMMEDIATE = (1 << 1),
+    ASSERT_TYPE_OBSERVED_DEFERRED_IMMEDIATE = (1 << 2),
+    ASSERT_TYPE_FINAL_DEFERRED_IMMEDIATE = (1 << 3),
+    ASSERT_TYPE_EXPECT = (1 << 4),
+    ASSERT_TYPE_UNIQUE = (1 << 5),
+    ASSERT_TYPE_UNIQUE0 = (1 << 6),
+    ASSERT_TYPE_PRIORITY = (1 << 7),
+};
+
+// IEEE 1800-2023 Table 20-7
+enum class VerilatedAssertDirectiveType : uint8_t {
+    DIRECTIVE_TYPE_ASSERT = (1 << 0),
+    DIRECTIVE_TYPE_COVER = (1 << 1),
+    DIRECTIVE_TYPE_ASSUME = (1 << 2),
+};
+using VerilatedAssertType_t = std::underlying_type<VerilatedAssertType>::type;
+using VerilatedAssertDirectiveType_t = std::underlying_type<VerilatedAssertDirectiveType>::type;
+
 //=============================================================================
 // Utility functions

@@ -199,19 +221,6 @@ public:
    void unlock() VL_RELEASE() VL_MT_SAFE { m_mutex.unlock(); }
    /// Try to acquire mutex.  Returns true on success, and false on failure.
    bool try_lock() VL_TRY_ACQUIRE(true) VL_MT_SAFE { return m_mutex.try_lock(); }
-    /// Acquire/lock mutex and check for stop request
-    /// It tries to lock the mutex and if it fails, it check if stop request was send.
-    /// It returns after locking mutex.
-    /// This function should be extracted to V3ThreadPool, but due to clang thread-safety
-    /// limitations it needs to be placed here.
-    void lockCheckStopRequest(std::function<void()> checkStopRequestFunction)
-        VL_ACQUIRE() VL_MT_SAFE {
-        while (true) {
-            checkStopRequestFunction();
-            if (m_mutex.try_lock()) return;
-            VL_CPU_RELAX();
-        }
-    }
 };

 /// Lock guard for mutex (ala std::unique_lock), wrapped to allow -fthread_safety checks
@@ -347,6 +356,19 @@ public:
 class VerilatedContext VL_NOT_FINAL {
    friend class VerilatedContextImp;

+private:
+    // MEMBERS
+    // Numer of assertion directive type members. Then each of them will represented as 1-bit in a
+    // mask.
+    static constexpr size_t ASSERT_DIRECTIVE_TYPE_MASK_WIDTH = 3;
+    // Specifies how many groups of directive type bit groups there are based on a number of
+    // assertion types.
+    // Note: we add one bit to store information whether Verilator's internal
+    // directive types are enabled, for example `violation if`s.
+    static constexpr size_t ASSERT_ON_WIDTH
+        = ASSERT_DIRECTIVE_TYPE_MASK_WIDTH * std::numeric_limits<VerilatedAssertType_t>::digits
+          + 1;
+
 protected:
    // TYPES
    using traceBaseModelCb_t
@@ -360,7 +382,11 @@ protected:
        // No std::strings or pointers or will serialize badly!
        // Fast path
        uint64_t m_time = 0;  // Current $time (unscaled), 0=at zero, or legacy
-        bool m_assertOn = true;  // Assertions are enabled
+        std::atomic<uint32_t> m_assertOn{
+            std::numeric_limits<uint32_t>::max()};  // Enabled assertions,
+                                                    // for each VerilatedAssertType we store
+                                                    // 3-bits, one for each directive type. Last
+                                                    // bit guards internal directive types.
        bool m_calcUnusedSigs = false;  // Waves file on, need all signals calculated
        bool m_fatalOnError = true;  // Fatal on $stop/non-fatal error
        bool m_fatalOnVpiError = true;  // Fatal on vpi error/unsupported
@@ -396,6 +422,7 @@ protected:
        std::string m_coverageFilename;  // +coverage+file filename
        std::string m_profExecFilename;  // +prof+exec+file filename
        std::string m_profVltFilename;  // +prof+vlt filename
+        std::string m_solverProgram;  // SMT solver program
        VlOs::DeltaCpuTime m_cpuTimeStart{false};  // CPU time, starts when create first model
        VlOs::DeltaWallTime m_wallTimeStart{false};  // Wall time, starts when create first model
        std::vector<traceBaseModelCb_t> m_traceBaseModelCbs;  // Callbacks to traceRegisterModel
@@ -447,14 +474,23 @@ public:

    // METHODS - User called

-    /// Enable assertions
-    void assertOn(bool flag) VL_MT_SAFE;
    /// Return if assertions enabled
-    bool assertOn() const VL_MT_SAFE { return m_s.m_assertOn; }
-    /// Enable calculation of unused signals (for traces)
-    void calcUnusedSigs(bool flag) VL_MT_SAFE;
+    bool assertOn() const VL_MT_SAFE;
+    /// Enable all assertion types
+    void assertOn(bool flag) VL_MT_SAFE;
+    /// Get enabled status for given assertion types
+    bool assertOnGet(VerilatedAssertType_t type,
+                     VerilatedAssertDirectiveType_t directive) const VL_MT_SAFE;
+    /// Set enabled status for given assertion types
+    void assertOnSet(VerilatedAssertType_t types,
+                     VerilatedAssertDirectiveType_t directives) VL_MT_SAFE;
+    /// Clear enabled status for given assertion types
+    void assertOnClear(VerilatedAssertType_t types,
+                       VerilatedAssertDirectiveType_t directives) VL_MT_SAFE;
    /// Return if calculating of unused signals (for traces)
    bool calcUnusedSigs() const VL_MT_SAFE { return m_s.m_calcUnusedSigs; }
+    /// Enable calculation of unused signals (for traces)
+    void calcUnusedSigs(bool flag) VL_MT_SAFE;
    /// Record command-line arguments, for retrieval by $test$plusargs/$value$plusargs,
    /// and for parsing +verilator+ run-time arguments.
    /// This should be called before the first model is created.
@@ -469,52 +505,51 @@ public:
    /// Return VerilatedCovContext, allocate if needed
    /// Note if get unresolved reference then likely forgot to link verilated_cov.cpp
    VerilatedCovContext* coveragep() VL_MT_SAFE;
-    /// Set debug level
+    /// Return debug level
+    static inline int debug() VL_MT_SAFE;  /// Set debug level
    /// Debug is currently global, but for forward compatibility have a per-context method
    static inline void debug(int val) VL_MT_SAFE;
-    /// Return debug level
-    static inline int debug() VL_MT_SAFE;
+    /// Return current number of errors/assertions
+    int errorCount() const VL_MT_SAFE { return m_s.m_errorCount; }
    /// Set current number of errors/assertions
    void errorCount(int val) VL_MT_SAFE;
    /// Increment current number of errors/assertions
    void errorCountInc() VL_MT_SAFE;
-    /// Return current number of errors/assertions
-    int errorCount() const VL_MT_SAFE { return m_s.m_errorCount; }
-    /// Set number of errors/assertions before stop
-    void errorLimit(int val) VL_MT_SAFE;
    /// Return number of errors/assertions before stop
    int errorLimit() const VL_MT_SAFE { return m_s.m_errorLimit; }
-    /// Set to throw fatal error on $stop/non-fatal error
-    void fatalOnError(bool flag) VL_MT_SAFE;
+    /// Set number of errors/assertions before stop
+    void errorLimit(int val) VL_MT_SAFE;
    /// Return if to throw fatal error on $stop/non-fatal
    bool fatalOnError() const VL_MT_SAFE { return m_s.m_fatalOnError; }
-    /// Set to throw fatal error on VPI errors
-    void fatalOnVpiError(bool flag) VL_MT_SAFE;
+    /// Set to throw fatal error on $stop/non-fatal error
+    void fatalOnError(bool flag) VL_MT_SAFE;
    /// Return if to throw fatal error on VPI errors
    bool fatalOnVpiError() const VL_MT_SAFE { return m_s.m_fatalOnVpiError; }
-    /// Set if got a $stop or non-fatal error
-    void gotError(bool flag) VL_MT_SAFE;
+    /// Set to throw fatal error on VPI errors
+    void fatalOnVpiError(bool flag) VL_MT_SAFE;
    /// Return if got a $stop or non-fatal error
    bool gotError() const VL_MT_SAFE { return m_s.m_gotError; }
-    /// Set if got a $finish or $stop/error
-    void gotFinish(bool flag) VL_MT_SAFE;
+    /// Set if got a $stop or non-fatal error
+    void gotError(bool flag) VL_MT_SAFE;
    /// Return if got a $finish or $stop/error
    bool gotFinish() const VL_MT_SAFE { return m_s.m_gotFinish; }
-    /// Enable quiet (also prevents need for OS calls to get CPU time)
-    void quiet(bool flag) VL_MT_SAFE;
+    /// Set if got a $finish or $stop/error
+    void gotFinish(bool flag) VL_MT_SAFE;
    /// Return if quiet enabled
    bool quiet() const VL_MT_SAFE { return m_s.m_quiet; }
+    /// Enable quiet (also prevents need for OS calls to get CPU time)
+    void quiet(bool flag) VL_MT_SAFE;
+    /// Return randReset value
+    int randReset() VL_MT_SAFE { return m_s.m_randReset; }
    /// Select initial value of otherwise uninitialized signals.
    /// 0 = Set to zeros
    /// 1 = Set all bits to one
    /// 2 = Randomize all bits
    void randReset(int val) VL_MT_SAFE;
-    /// Return randReset value
-    int randReset() VL_MT_SAFE { return m_s.m_randReset; }
-    /// Return default random seed
-    void randSeed(int val) VL_MT_SAFE;
    /// Set default random seed, 0 = seed it automatically
    int randSeed() const VL_MT_SAFE { return m_s.m_randSeed; }
+    /// Return default random seed
+    void randSeed(int val) VL_MT_SAFE;

    /// Return statistic: CPU time delta from model created until now
    double statCpuTimeSinceStart() const VL_MT_SAFE_EXCLUDES(m_mutex);
@@ -606,23 +641,27 @@ public:
    enableExecutionProfiler(VerilatedVirtualBase* (*construct)(VerilatedContext&));

    // Internal: coverage
-    void coverageFilename(const std::string& flag) VL_MT_SAFE;
    std::string coverageFilename() const VL_MT_SAFE;
+    void coverageFilename(const std::string& flag) VL_MT_SAFE;

    // Internal: $dumpfile
-    void dumpfile(const std::string& flag) VL_MT_SAFE_EXCLUDES(m_timeDumpMutex);
    std::string dumpfile() const VL_MT_SAFE_EXCLUDES(m_timeDumpMutex);
+    void dumpfile(const std::string& flag) VL_MT_SAFE_EXCLUDES(m_timeDumpMutex);
    std::string dumpfileCheck() const VL_MT_SAFE_EXCLUDES(m_timeDumpMutex);

    // Internal: --prof-exec related settings
-    void profExecStart(uint64_t flag) VL_MT_SAFE;
    uint64_t profExecStart() const VL_MT_SAFE { return m_ns.m_profExecStart; }
-    void profExecWindow(uint64_t flag) VL_MT_SAFE;
+    void profExecStart(uint64_t flag) VL_MT_SAFE;
    uint32_t profExecWindow() const VL_MT_SAFE { return m_ns.m_profExecWindow; }
-    void profExecFilename(const std::string& flag) VL_MT_SAFE;
+    void profExecWindow(uint64_t flag) VL_MT_SAFE;
    std::string profExecFilename() const VL_MT_SAFE;
-    void profVltFilename(const std::string& flag) VL_MT_SAFE;
+    void profExecFilename(const std::string& flag) VL_MT_SAFE;
    std::string profVltFilename() const VL_MT_SAFE;
+    void profVltFilename(const std::string& flag) VL_MT_SAFE;
+
+    // Internal: SMT solver program
+    std::string solverProgram() const VL_MT_SAFE;
+    void solverProgram(const std::string& flag) VL_MT_SAFE;

    // Internal: Find scope
    const VerilatedScope* scopeFind(const char* namep) const VL_MT_SAFE;
@@ -757,8 +796,6 @@ class Verilated final {
 public:
    // METHODS - User called

-    /// Enable debug of internal verilated code
-    static void debug(int level) VL_MT_SAFE;
 #ifdef VL_DEBUG
    /// Return debug level
    /// When multithreaded this may not immediately react to another thread
@@ -768,6 +805,8 @@ public:
    /// Return constant 0 debug level, so C++'s optimizer rips up
    static constexpr int debug() VL_PURE { return 0; }
 #endif
+    /// Enable debug of internal verilated code
+    static void debug(int level) VL_MT_SAFE;

    /// Set the last VerilatedContext accessed
    /// Generally threadContextp(value) should be called instead
@@ -799,18 +838,18 @@ public:
    }

 #ifndef VL_NO_LEGACY
-    /// Call VerilatedContext::assertOn using current thread's VerilatedContext
-    static void assertOn(bool flag) VL_MT_SAFE { Verilated::threadContextp()->assertOn(flag); }
    /// Return VerilatedContext::assertOn() using current thread's VerilatedContext
    static bool assertOn() VL_MT_SAFE { return Verilated::threadContextp()->assertOn(); }
-    /// Call VerilatedContext::calcUnusedSigs using current thread's VerilatedContext
-    static void calcUnusedSigs(bool flag) VL_MT_SAFE {
-        Verilated::threadContextp()->calcUnusedSigs(flag);
-    }
+    /// Call VerilatedContext::assertOn using current thread's VerilatedContext
+    static void assertOn(bool flag) VL_MT_SAFE { Verilated::threadContextp()->assertOn(flag); }
    /// Return VerilatedContext::calcUnusedSigs using current thread's VerilatedContext
    static bool calcUnusedSigs() VL_MT_SAFE {
        return Verilated::threadContextp()->calcUnusedSigs();
    }
+    /// Call VerilatedContext::calcUnusedSigs using current thread's VerilatedContext
+    static void calcUnusedSigs(bool flag) VL_MT_SAFE {
+        Verilated::threadContextp()->calcUnusedSigs(flag);
+    }
    /// Call VerilatedContext::commandArgs using current thread's VerilatedContext
    static void commandArgs(int argc, const char** argv) VL_MT_SAFE {
        Verilated::threadContextp()->commandArgs(argc, argv);
@@ -826,44 +865,44 @@ public:
    static const char* commandArgsPlusMatch(const char* prefixp) VL_MT_SAFE {
        return Verilated::threadContextp()->commandArgsPlusMatch(prefixp);
    }
-    /// Call VerilatedContext::errorLimit using current thread's VerilatedContext
-    static void errorLimit(int val) VL_MT_SAFE { Verilated::threadContextp()->errorLimit(val); }
    /// Return VerilatedContext::errorLimit using current thread's VerilatedContext
    static int errorLimit() VL_MT_SAFE { return Verilated::threadContextp()->errorLimit(); }
+    /// Call VerilatedContext::errorLimit using current thread's VerilatedContext
+    static void errorLimit(int val) VL_MT_SAFE { Verilated::threadContextp()->errorLimit(val); }
+    /// Return VerilatedContext::fatalOnError using current thread's VerilatedContext
+    static bool fatalOnError() VL_MT_SAFE { return Verilated::threadContextp()->fatalOnError(); }
    /// Call VerilatedContext::fatalOnError using current thread's VerilatedContext
    static void fatalOnError(bool flag) VL_MT_SAFE {
        Verilated::threadContextp()->fatalOnError(flag);
    }
-    /// Return VerilatedContext::fatalOnError using current thread's VerilatedContext
-    static bool fatalOnError() VL_MT_SAFE { return Verilated::threadContextp()->fatalOnError(); }
-    /// Call VerilatedContext::fatalOnVpiError using current thread's VerilatedContext
-    static void fatalOnVpiError(bool flag) VL_MT_SAFE {
-        Verilated::threadContextp()->fatalOnVpiError(flag);
-    }
    /// Return VerilatedContext::fatalOnVpiError using current thread's VerilatedContext
    static bool fatalOnVpiError() VL_MT_SAFE {
        return Verilated::threadContextp()->fatalOnVpiError();
    }
-    /// Call VerilatedContext::gotError using current thread's VerilatedContext
-    static void gotError(bool flag) VL_MT_SAFE { Verilated::threadContextp()->gotError(flag); }
+    /// Call VerilatedContext::fatalOnVpiError using current thread's VerilatedContext
+    static void fatalOnVpiError(bool flag) VL_MT_SAFE {
+        Verilated::threadContextp()->fatalOnVpiError(flag);
+    }
    /// Return VerilatedContext::gotError using current thread's VerilatedContext
    static bool gotError() VL_MT_SAFE { return Verilated::threadContextp()->gotError(); }
-    /// Call VerilatedContext::gotFinish using current thread's VerilatedContext
-    static void gotFinish(bool flag) VL_MT_SAFE { Verilated::threadContextp()->gotFinish(flag); }
+    /// Call VerilatedContext::gotError using current thread's VerilatedContext
+    static void gotError(bool flag) VL_MT_SAFE { Verilated::threadContextp()->gotError(flag); }
    /// Return VerilatedContext::gotFinish using current thread's VerilatedContext
    static bool gotFinish() VL_MT_SAFE { return Verilated::threadContextp()->gotFinish(); }
-    /// Call VerilatedContext::randReset using current thread's VerilatedContext
-    static void randReset(int val) VL_MT_SAFE { Verilated::threadContextp()->randReset(val); }
+    /// Call VerilatedContext::gotFinish using current thread's VerilatedContext
+    static void gotFinish(bool flag) VL_MT_SAFE { Verilated::threadContextp()->gotFinish(flag); }
    /// Return VerilatedContext::randReset using current thread's VerilatedContext
    static int randReset() VL_MT_SAFE { return Verilated::threadContextp()->randReset(); }
-    /// Call VerilatedContext::randSeed using current thread's VerilatedContext
-    static void randSeed(int val) VL_MT_SAFE { Verilated::threadContextp()->randSeed(val); }
+    /// Call VerilatedContext::randReset using current thread's VerilatedContext
+    static void randReset(int val) VL_MT_SAFE { Verilated::threadContextp()->randReset(val); }
    /// Return VerilatedContext::randSeed using current thread's VerilatedContext
    static int randSeed() VL_MT_SAFE { return Verilated::threadContextp()->randSeed(); }
-    /// Call VerilatedContext::time using current thread's VerilatedContext
-    static void time(uint64_t val) VL_MT_SAFE { Verilated::threadContextp()->time(val); }
+    /// Call VerilatedContext::randSeed using current thread's VerilatedContext
+    static void randSeed(int val) VL_MT_SAFE { Verilated::threadContextp()->randSeed(val); }
    /// Return VerilatedContext::time using current thread's VerilatedContext
    static uint64_t time() VL_MT_SAFE { return Verilated::threadContextp()->time(); }
+    /// Call VerilatedContext::time using current thread's VerilatedContext
+    static void time(uint64_t val) VL_MT_SAFE { Verilated::threadContextp()->time(val); }
    /// Call VerilatedContext::timeInc using current thread's VerilatedContext
    static void timeInc(uint64_t add) VL_MT_UNSAFE { Verilated::threadContextp()->timeInc(add); }
    // Deprecated
@@ -953,8 +992,8 @@ public:

    // Internal: Set the mtaskId, called when an mtask starts
    // Per thread, so no need to be in VerilatedContext
-    static void mtaskId(uint32_t id) VL_MT_SAFE { t_s.t_mtaskId = id; }
    static uint32_t mtaskId() VL_MT_SAFE { return t_s.t_mtaskId; }
+    static void mtaskId(uint32_t id) VL_MT_SAFE { t_s.t_mtaskId = id; }
    static void endOfEvalReqdInc() VL_MT_SAFE { ++t_s.t_endOfEvalReqd; }
    static void endOfEvalReqdDec() VL_MT_SAFE { --t_s.t_endOfEvalReqd; }

@@ -80,6 +80,7 @@ VK_CPPFLAGS_ALWAYS += \
 		-I$(VERILATOR_ROOT)/include/vltstd \
 		-DVM_COVERAGE=$(VM_COVERAGE) \
 		-DVM_SC=$(VM_SC) \
+		-DVM_TIMING=$(VM_TIMING) \
 		-DVM_TRACE=$(VM_TRACE) \
 		-DVM_TRACE_FST=$(VM_TRACE_FST) \
 		-DVM_TRACE_VCD=$(VM_TRACE_VCD) \
@@ -225,7 +226,6 @@ endif
 #   1. Extract object files from .a
 #   2. Create a new archive from extracted .o and given .o
 %.a: | %.verilator_deplist.tmp
-	$(info Archive $(AR) -rcs $@ $^)
 	$(foreach L, $(filter-out %.a,$^), $(shell echo $L >>[email protected]_deplist.tmp))
 	@if test $(words $(filter %.a,$^)) -eq 0; then \
 		$(RM) -f $@; \
@@ -93,8 +93,8 @@ public:
    /// Make all data per_instance, overriding point's per_instance
    void forcePerInstance(bool flag) VL_MT_SAFE;
    /// Write all coverage data to a file
-    void write(const std::string& filename) VL_MT_SAFE;
    void write() VL_MT_SAFE { write(defaultFilename()); }
+    void write(const std::string& filename) VL_MT_SAFE;
    /// Clear coverage points (and call delete on all items)
    void clear() VL_MT_SAFE;
    /// Clear items not matching the provided string
@@ -165,8 +165,8 @@ public:
    /// Return default filename for the current thread
    static std::string defaultFilename() VL_MT_SAFE { return threadCovp()->defaultFilename(); }
    /// Write all coverage data to a file for the current thread
-    static void write(const std::string& filename) VL_MT_SAFE { threadCovp()->write(filename); }
    static void write() VL_MT_SAFE { write(defaultFilename()); }
+    static void write(const std::string& filename) VL_MT_SAFE { threadCovp()->write(filename); }
    /// Clear coverage points (and call delete on all items) for the current thread
    static void clear() VL_MT_SAFE { threadCovp()->clear(); }
    /// Clear items not matching the provided string for the current thread
@@ -127,6 +127,7 @@ void VerilatedFst::declDTypeEnum(int dtypenum, const char* name, uint32_t elemen
 // TODO: should return std::optional<fstScopeType>, but I can't have C++17
 static std::pair<bool, fstScopeType> toFstScopeType(VerilatedTracePrefixType type) {
    switch (type) {
+    case VerilatedTracePrefixType::ROOTIO_MODULE: return {true, FST_ST_VCD_MODULE};
    case VerilatedTracePrefixType::SCOPE_MODULE: return {true, FST_ST_VCD_MODULE};
    case VerilatedTracePrefixType::SCOPE_INTERFACE: return {true, FST_ST_VCD_INTERFACE};
    case VerilatedTracePrefixType::STRUCT_PACKED:
@@ -137,7 +138,20 @@ static std::pair<bool, fstScopeType> toFstScopeType(VerilatedTracePrefixType typ
 }

 void VerilatedFst::pushPrefix(const std::string& name, VerilatedTracePrefixType type) {
-    const std::string newPrefix = m_prefixStack.back().first + name;
+    assert(!m_prefixStack.empty());  // Constructor makes an empty entry
+    std::string pname = name;
+    // An empty name means this is the root of a model created with name()=="".  The
+    // tools get upset if we try to pass this as empty, so we put the signals under a
+    // new scope, but the signals further down will be peers, not children (as usual
+    // for name()!="")
+    // Terminate earlier $root?
+    if (m_prefixStack.back().second == VerilatedTracePrefixType::ROOTIO_MODULE) popPrefix();
+    if (pname.empty()) {  // Start new temporary root
+        pname = "$rootio";  // VCD names are not backslash escaped
+        m_prefixStack.emplace_back("", VerilatedTracePrefixType::ROOTIO_WRAPPER);
+        type = VerilatedTracePrefixType::ROOTIO_MODULE;
+    }
+    const std::string newPrefix = m_prefixStack.back().first + pname;
    const auto pair = toFstScopeType(type);
    const bool properScope = pair.first;
    const fstScopeType scopeType = pair.second;
@@ -149,10 +163,11 @@ void VerilatedFst::pushPrefix(const std::string& name, VerilatedTracePrefixType
 }

 void VerilatedFst::popPrefix() {
+    assert(!m_prefixStack.empty());
    const bool properScope = toFstScopeType(m_prefixStack.back().second).first;
    if (properScope) fstWriterSetUpscope(m_fst);
    m_prefixStack.pop_back();
-    assert(!m_prefixStack.empty());
+    assert(!m_prefixStack.empty());  // Always one left, the constructor's initial one
 }

 void VerilatedFst::declare(uint32_t code, const char* name, int dtypenum,
@@ -294,7 +309,7 @@ void VerilatedFst::configure(const VerilatedTraceConfig& config) {
 // so always inline them.

 VL_ATTR_ALWINLINE
-void VerilatedFstBuffer::emitEvent(uint32_t code, const VlEventBase* newval) {
+void VerilatedFstBuffer::emitEvent(uint32_t code) {
    VL_DEBUG_IFDEF(assert(m_symbolp[code]););
    fstWriterEmitValueChange(m_fst, m_symbolp[code], "1");
 }
@@ -47,7 +47,7 @@ private:
    friend VerilatedFstBuffer;  // Give the buffer access to the private bits

    //=========================================================================
-    // FST specific internals
+    // FST-specific internals

    void* m_fst = nullptr;
    std::map<uint32_t, vlFstHandle> m_code2symbol;
@@ -177,7 +177,7 @@ class VerilatedFstBuffer VL_NOT_FINAL {

    // Implementations of duck-typed methods for VerilatedTraceBuffer. These are
    // called from only one place (the full* methods), so always inline them.
-    VL_ATTR_ALWINLINE void emitEvent(uint32_t code, const VlEventBase* newval);
+    VL_ATTR_ALWINLINE void emitEvent(uint32_t code);
    VL_ATTR_ALWINLINE void emitBit(uint32_t code, CData newval);
    VL_ATTR_ALWINLINE void emitCData(uint32_t code, CData newval, int bits);
    VL_ATTR_ALWINLINE void emitSData(uint32_t code, SData newval, int bits);
@@ -42,19 +42,20 @@ extern void vl_finish(const char* filename, int linenum, const char* hier) VL_MT
 /// Routine to call for $stop and non-fatal error
 /// User code may wish to replace this function, to do so, define VL_USER_STOP.
 /// This code does not have to be thread safe.
-/// Verilator internal code must call VL_FINISH_MT instead, which eventually calls this.
+/// Verilator internal code must call VL_STOP_MT instead, which eventually calls this.
 extern void vl_stop(const char* filename, int linenum, const char* hier) VL_MT_UNSAFE;

 /// Routine to call for fatal messages
 /// User code may wish to replace this function, to do so, define VL_USER_FATAL.
 /// This code does not have to be thread safe.
-/// Verilator internal code must call VL_FINISH_MT instead, which eventually calls this.
+/// Verilator internal code must call VL_FATAL_MT instead, which eventually calls this.
 extern void vl_fatal(const char* filename, int linenum, const char* hier,
                     const char* msg) VL_MT_UNSAFE;

 /// Routine to call for warning messages
 /// User code may wish to replace this function, to do so, define VL_USER_WARN.
 /// This code does not have to be thread safe.
+/// Verilator internal code must call VL_WARN_MT instead, which eventually calls this.
 extern void vl_warn(const char* filename, int linenum, const char* hier,
                    const char* msg) VL_MT_UNSAFE;

@@ -106,6 +107,14 @@ extern IData VL_RAND_RESET_I(int obits) VL_MT_SAFE;
 extern QData VL_RAND_RESET_Q(int obits) VL_MT_SAFE;
 /// Random reset a signal of given width (init time only)
 extern WDataOutP VL_RAND_RESET_W(int obits, WDataOutP outwp) VL_MT_SAFE;
+
+/// Random reset a signal of given width (assign time only)
+extern IData VL_RAND_RESET_ASSIGN_I(int obits) VL_MT_SAFE;
+/// Random reset a signal of given width (assign time only)
+extern QData VL_RAND_RESET_ASSIGN_Q(int obits) VL_MT_SAFE;
+/// Random reset a signal of given width (assign time only)
+extern WDataOutP VL_RAND_RESET_ASSIGN_W(int obits, WDataOutP outwp) VL_MT_SAFE;
+
 /// Zero reset a signal (slow - else use VL_ZERO_W)
 extern WDataOutP VL_ZERO_RESET_W(int obits, WDataOutP outwp) VL_MT_SAFE;

@@ -115,6 +124,9 @@ extern void VL_PRINTTIMESCALE(const char* namep, const char* timeunitp,
 extern WDataOutP _vl_moddiv_w(int lbits, WDataOutP owp, WDataInP const lwp, WDataInP const rwp,
                              bool is_modulus) VL_MT_SAFE;

+extern void _vl_vsss_based(WDataOutP owp, int obits, int baseLog2, const char* strp,
+                           size_t posstart, size_t posend) VL_MT_SAFE;
+
 extern IData VL_FGETS_IXI(int obits, void* destp, IData fpi) VL_MT_SAFE;

 extern void VL_FFLUSH_I(IData fdi) VL_MT_SAFE;
@@ -152,6 +164,7 @@ extern std::string VL_STACKTRACE_N() VL_MT_SAFE;
 extern IData VL_SYSTEM_IW(int lhswords, WDataInP const lhsp) VL_MT_SAFE;
 extern IData VL_SYSTEM_IQ(QData lhs) VL_MT_SAFE;
 inline IData VL_SYSTEM_II(IData lhs) VL_MT_SAFE { return VL_SYSTEM_IQ(lhs); }
+extern IData VL_SYSTEM_IN(const std::string& lhs) VL_MT_SAFE;

 extern IData VL_TESTPLUSARGS_I(const std::string& format) VL_MT_SAFE;
 extern const char* vl_mc_scan_plusargs(const char* prefixp) VL_MT_SAFE;  // PLIish
@@ -215,6 +228,10 @@ static inline QData VL_CVT_Q_D(double lhs) VL_PURE {
    return u.q;
 }
 // clang-format on
+// Return string from DPI char*
+static inline std::string VL_CVT_N_CSTR(const char* lhsp) VL_PURE {
+    return lhsp ? std::string{lhsp} : ""s;
+}

 // Return double from lhs (numeric) unsigned
 double VL_ITOR_D_W(int lbits, WDataInP const lwp) VL_PURE;
@@ -238,7 +255,7 @@ static inline double VL_ISTOR_D_Q(int lbits, QData lhs) VL_MT_SAFE {
    VL_SET_WQ(lwp, lhs);
    return VL_ISTOR_D_W(lbits, lwp);
 }
-// Return QData from double (numeric)
+// Return IData truncated from double (numeric)
 static inline IData VL_RTOI_I_D(double lhs) VL_PURE { return static_cast<int32_t>(VL_TRUNC(lhs)); }

 // Sign extend such that if MSB set, we get ffff_ffff, else 0s
@@ -504,19 +521,18 @@ static inline void VL_ASSIGNBIT_WO(int bit, WDataOutP owp) VL_MT_SAFE {
    { (svar).write(rd); }
 #define VL_ASSIGN_SBQ(obits, svar, rd) \
    { (svar).write(rd); }
-#define VL_SC_BITS_PER_DIGIT 30  // This comes from sc_nbdefs.h BITS_PER_DIGIT
 #define VL_ASSIGN_SBW(obits, svar, rwp) \
    { \
        sc_dt::sc_biguint<(obits)> _butemp; \
        int32_t lsb = 0; \
        uint32_t* chunkp = _butemp.get_raw(); \
-        while (lsb + VL_SC_BITS_PER_DIGIT < (obits)) { \
+        while (lsb + BITS_PER_DIGIT < (obits)) { \
            static_assert(std::is_same<IData, EData>::value, "IData and EData mismatch"); \
-            const uint32_t data = VL_SEL_IWII(lsb + VL_SC_BITS_PER_DIGIT + 1, (rwp).data(), lsb, \
-                                              VL_SC_BITS_PER_DIGIT); \
-            *chunkp = data & VL_MASK_E(VL_SC_BITS_PER_DIGIT); \
+            const uint32_t data \
+                = VL_SEL_IWII(lsb + BITS_PER_DIGIT + 1, (rwp).data(), lsb, BITS_PER_DIGIT); \
+            *chunkp = data & VL_MASK_E(BITS_PER_DIGIT); \
            ++chunkp; \
-            lsb += VL_SC_BITS_PER_DIGIT; \
+            lsb += BITS_PER_DIGIT; \
        } \
        if (lsb < (obits)) { \
            const uint32_t msb_data = VL_SEL_IWII((obits) + 1, (rwp).data(), lsb, (obits)-lsb); \
@@ -1570,49 +1586,184 @@ static inline WDataOutP VL_STREAML_WWI(int lbits, WDataOutP owp, WDataInP const
    return owp;
 }

-template <typename T>
-static inline void VL_ASSIGN_DYN_Q(VlQueue<T>& q, int elem_size, int lbits, QData from) {
-    const int size = (lbits + elem_size - 1) / elem_size;
-    q.renew(size);
-    const QData mask = VL_MASK_Q(elem_size);
-    for (int i = 0; i < size; ++i) q.at(i) = (T)((from >> (i * elem_size)) & mask);
+static inline IData VL_PACK_II(int obits, int lbits, const VlQueue<CData>& q) {
+    IData ret = 0;
+    for (size_t i = 0; i < q.size(); ++i) ret |= static_cast<IData>(q.at(i)) << (i * lbits);
+    return ret;
 }

-template <typename T, std::size_t T_Depth>
-static inline void VL_ASSIGN_UNPACK_Q(VlUnpacked<T, T_Depth>& q, size_t elem_size, QData from) {
-    const QData mask = VL_MASK_Q(elem_size);
+static inline IData VL_PACK_II(int obits, int lbits, const VlQueue<SData>& q) {
+    IData ret = 0;
+    for (size_t i = 0; i < q.size(); ++i) ret |= static_cast<IData>(q.at(i)) << (i * lbits);
+    return ret;
+}
+
+static inline IData VL_PACK_II(int obits, int lbits, const VlQueue<IData>& q) {
+    IData ret = 0;
+    for (size_t i = 0; i < q.size(); ++i) ret |= q.at(i) << (i * lbits);
+    return ret;
+}
+
+template <std::size_t T_Depth>
+static inline IData VL_PACK_II(int obits, int lbits, const VlUnpacked<CData, T_Depth>& q) {
+    IData ret = 0;
    for (size_t i = 0; i < T_Depth; ++i)
-        q[i] = (T)((from >> ((T_Depth - 1 - i) * elem_size)) & mask);
+        ret |= static_cast<IData>(q[T_Depth - 1 - i]) << (i * lbits);
+    return ret;
 }

-template <typename T>
-static inline IData VL_DYN_TO_I(const VlQueue<T>& q, int elem_size) {
+template <std::size_t T_Depth>
+static inline IData VL_PACK_II(int obits, int lbits, const VlUnpacked<SData, T_Depth>& q) {
    IData ret = 0;
-    for (int i = 0; i < q.size(); ++i) ret |= q.at(i) << (i * elem_size);
+    for (size_t i = 0; i < T_Depth; ++i)
+        ret |= static_cast<IData>(q[T_Depth - 1 - i]) << (i * lbits);
    return ret;
 }

-template <typename T>
-static inline QData VL_DYN_TO_Q(const VlQueue<T>& q, int elem_size) {
-    QData ret = 0;
-    for (int i = 0; i < q.size(); ++i) ret |= q.at(i) << (i * elem_size);
-    return ret;
-}
-
-template <typename T, std::size_t T_Depth>
-static inline IData VL_UNPACK_TO_I(const VlUnpacked<T, T_Depth>& q, size_t elem_size) {
+template <std::size_t T_Depth>
+static inline IData VL_PACK_II(int obits, int lbits, const VlUnpacked<IData, T_Depth>& q) {
    IData ret = 0;
-    for (size_t i = 0; i < T_Depth; ++i) ret |= q[T_Depth - 1 - i] << (i * elem_size);
+    for (size_t i = 0; i < T_Depth; ++i) ret |= q[T_Depth - 1 - i] << (i * lbits);
    return ret;
 }

-template <typename T, std::size_t T_Depth>
-static inline QData VL_UNPACK_TO_Q(const VlUnpacked<T, T_Depth>& q, size_t elem_size) {
+static inline QData VL_PACK_QI(int obits, int lbits, const VlQueue<CData>& q) {
    QData ret = 0;
-    for (size_t i = 0; i < T_Depth; ++i) ret |= q[T_Depth - 1 - i] << (i * elem_size);
+    for (size_t i = 0; i < q.size(); ++i) ret |= static_cast<QData>(q.at(i)) << (i * lbits);
    return ret;
 }

+static inline QData VL_PACK_QI(int obits, int lbits, const VlQueue<SData>& q) {
+    QData ret = 0;
+    for (size_t i = 0; i < q.size(); ++i) ret |= static_cast<QData>(q.at(i)) << (i * lbits);
+    return ret;
+}
+
+static inline QData VL_PACK_QI(int obits, int lbits, const VlQueue<IData>& q) {
+    QData ret = 0;
+    for (size_t i = 0; i < q.size(); ++i) ret |= static_cast<QData>(q.at(i)) << (i * lbits);
+    return ret;
+}
+
+template <std::size_t T_Depth>
+static inline QData VL_PACK_QI(int obits, int lbits, const VlUnpacked<CData, T_Depth>& q) {
+    QData ret = 0;
+    for (size_t i = 0; i < T_Depth; ++i)
+        ret |= static_cast<QData>(q[T_Depth - 1 - i]) << (i * lbits);
+    return ret;
+}
+
+template <std::size_t T_Depth>
+static inline QData VL_PACK_QI(int obits, int lbits, const VlUnpacked<SData, T_Depth>& q) {
+    QData ret = 0;
+    for (size_t i = 0; i < T_Depth; ++i)
+        ret |= static_cast<QData>(q[T_Depth - 1 - i]) << (i * lbits);
+    return ret;
+}
+
+template <std::size_t T_Depth>
+static inline QData VL_PACK_QI(int obits, int lbits, const VlUnpacked<IData, T_Depth>& q) {
+    QData ret = 0;
+    for (size_t i = 0; i < T_Depth; ++i)
+        ret |= static_cast<QData>(q[T_Depth - 1 - i]) << (i * lbits);
+    return ret;
+}
+
+static inline QData VL_PACK_QQ(int obits, int lbits, const VlQueue<QData>& q) {
+    QData ret = 0;
+    for (size_t i = 0; i < q.size(); ++i) ret |= q.at(i) << (i * lbits);
+    return ret;
+}
+
+template <std::size_t T_Depth>
+static inline QData VL_PACK_QQ(int obits, int lbits, const VlUnpacked<QData, T_Depth>& q) {
+    QData ret = 0;
+    for (size_t i = 0; i < T_Depth; ++i) ret |= q[T_Depth - 1 - i] << (i * lbits);
+    return ret;
+}
+
+static inline WDataOutP VL_PACK_WI(int obits, int lbits, WDataOutP owp, const VlQueue<CData>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < q.size(); ++i)
+        _vl_insert_WI(owp, q.at(i), i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+static inline WDataOutP VL_PACK_WI(int obits, int lbits, WDataOutP owp, const VlQueue<SData>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < q.size(); ++i)
+        _vl_insert_WI(owp, q.at(i), i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+static inline WDataOutP VL_PACK_WI(int obits, int lbits, WDataOutP owp, const VlQueue<IData>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < q.size(); ++i)
+        _vl_insert_WI(owp, q.at(i), i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+template <std::size_t T_Depth>
+static inline WDataOutP VL_PACK_WI(int obits, int lbits, WDataOutP owp,
+                                   const VlUnpacked<CData, T_Depth>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < T_Depth; ++i)
+        _vl_insert_WI(owp, q[T_Depth - 1 - i], i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+template <std::size_t T_Depth>
+static inline WDataOutP VL_PACK_WI(int obits, int lbits, WDataOutP owp,
+                                   const VlUnpacked<SData, T_Depth>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < T_Depth; ++i)
+        _vl_insert_WI(owp, q[T_Depth - 1 - i], i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+template <std::size_t T_Depth>
+static inline WDataOutP VL_PACK_WI(int obits, int lbits, WDataOutP owp,
+                                   const VlUnpacked<IData, T_Depth>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < T_Depth; ++i)
+        _vl_insert_WI(owp, q[T_Depth - 1 - i], i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+static inline WDataOutP VL_PACK_WQ(int obits, int lbits, WDataOutP owp, const VlQueue<QData>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < q.size(); ++i)
+        _vl_insert_WQ(owp, q.at(i), i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+template <std::size_t T_Depth>
+static inline WDataOutP VL_PACK_WQ(int obits, int lbits, WDataOutP owp,
+                                   const VlUnpacked<QData, T_Depth>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < T_Depth; ++i)
+        _vl_insert_WQ(owp, q[T_Depth - 1 - i], i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+template <std::size_t N>
+static inline WDataOutP VL_PACK_WW(int obits, int lbits, WDataOutP owp,
+                                   const VlQueue<VlWide<N>>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < q.size(); ++i)
+        _vl_insert_WW(owp, q.at(i), i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
+template <std::size_t T_Depth, std::size_t N>
+static inline WDataOutP VL_PACK_WW(int obits, int lbits, WDataOutP owp,
+                                   const VlUnpacked<VlWide<N>, T_Depth>& q) {
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    for (size_t i = 0; i < T_Depth; ++i)
+        _vl_insert_WW(owp, q[T_Depth - 1 - i], i * lbits + lbits - 1, i * lbits);
+    return owp;
+}
+
 // Because concats are common and wide, it's valuable to always have a clean output.
 // Thus we specify inputs must be clean, so we don't need to clean the output.
 // Note the bit shifts are always constants, so the adds in these constify out.
@@ -2052,6 +2203,181 @@ static inline WDataOutP VL_SEL_WWII(int obits, int lbits, WDataOutP owp, WDataIn
 //======================================================================
 // Expressions needing insert/select

+static inline void VL_UNPACK_II(int lbits, int rbits, VlQueue<CData>& q, IData from) {
+    const size_t size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) q.atWrite(i) = (from >> (i * lbits)) & mask;
+}
+
+static inline void VL_UNPACK_II(int lbits, int rbits, VlQueue<SData>& q, IData from) {
+    const size_t size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) q.atWrite(i) = (from >> (i * lbits)) & mask;
+}
+
+static inline void VL_UNPACK_II(int lbits, int rbits, VlQueue<IData>& q, IData from) {
+    const size_t size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) q.atWrite(i) = (from >> (i * lbits)) & mask;
+}
+
+static inline void VL_UNPACK_IQ(int lbits, int rbits, VlQueue<CData>& q, QData from) {
+    const size_t size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) q.atWrite(i) = (from >> (i * lbits)) & mask;
+}
+
+static inline void VL_UNPACK_IQ(int lbits, int rbits, VlQueue<SData>& q, QData from) {
+    const size_t size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) q.atWrite(i) = (from >> (i * lbits)) & mask;
+}
+
+static inline void VL_UNPACK_IQ(int lbits, int rbits, VlQueue<IData>& q, QData from) {
+    const size_t size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) q.atWrite(i) = (from >> (i * lbits)) & mask;
+}
+
+static inline void VL_UNPACK_QQ(int lbits, int rbits, VlQueue<QData>& q, QData from) {
+    const size_t size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const QData mask = VL_MASK_Q(lbits);
+    for (size_t i = 0; i < size; ++i) q.atWrite(i) = (from >> (i * lbits)) & mask;
+}
+
+static inline void VL_UNPACK_IW(int lbits, int rbits, VlQueue<CData>& q, WDataInP rwp) {
+    const int size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) {
+        q.atWrite(i) = VL_SEL_IWII(rbits, rwp, i * lbits, lbits) & mask;
+    }
+}
+
+static inline void VL_UNPACK_IW(int lbits, int rbits, VlQueue<SData>& q, WDataInP rwp) {
+    const int size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) {
+        q.atWrite(i) = VL_SEL_IWII(rbits, rwp, i * lbits, lbits) & mask;
+    }
+}
+
+static inline void VL_UNPACK_IW(int lbits, int rbits, VlQueue<IData>& q, WDataInP rwp) {
+    const int size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < size; ++i) {
+        q.atWrite(i) = VL_SEL_IWII(rbits, rwp, i * lbits, lbits) & mask;
+    }
+}
+
+static inline void VL_UNPACK_QW(int lbits, int rbits, VlQueue<QData>& q, WDataInP rwp) {
+    const int size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    const QData mask = VL_MASK_Q(lbits);
+    for (size_t i = 0; i < size; ++i) {
+        q.atWrite(i) = VL_SEL_QWII(rbits, rwp, i * lbits, lbits) & mask;
+    }
+}
+
+template <std::size_t N>
+static inline void VL_UNPACK_WW(int lbits, int rbits, VlQueue<VlWide<N>>& q, WDataInP rwp) {
+    const int size = (rbits + lbits - 1) / lbits;
+    q.renew(size);
+    for (size_t i = 0; i < size; ++i) {
+        VL_SEL_WWII(lbits, rbits, q.atWrite(i), rwp, i * lbits, lbits);
+    }
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_II(int lbits, int rbits, VlUnpacked<CData, T_Depth>& q, IData from) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i) q[i] = (from >> ((T_Depth - 1 - i) * lbits)) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_II(int lbits, int rbits, VlUnpacked<SData, T_Depth>& q, IData from) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i) q[i] = (from >> ((T_Depth - 1 - i) * lbits)) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_II(int lbits, int rbits, VlUnpacked<IData, T_Depth>& q, IData from) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i) q[i] = (from >> ((T_Depth - 1 - i) * lbits)) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_IQ(int lbits, int rbits, VlUnpacked<CData, T_Depth>& q, QData from) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i) q[i] = (from >> ((T_Depth - 1 - i) * lbits)) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_IQ(int lbits, int rbits, VlUnpacked<SData, T_Depth>& q, QData from) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i) q[i] = (from >> ((T_Depth - 1 - i) * lbits)) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_IQ(int lbits, int rbits, VlUnpacked<IData, T_Depth>& q, QData from) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i) q[i] = (from >> ((T_Depth - 1 - i) * lbits)) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_QQ(int lbits, int rbits, VlUnpacked<QData, T_Depth>& q, QData from) {
+    const QData mask = VL_MASK_Q(lbits);
+    for (size_t i = 0; i < T_Depth; ++i) q[i] = (from >> ((T_Depth - 1 - i) * lbits)) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_IW(int lbits, int rbits, VlUnpacked<CData, T_Depth>& q,
+                                WDataInP rwp) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i)
+        q[i] = VL_SEL_IWII(rbits, rwp, (T_Depth - 1 - i) * lbits, lbits) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_IW(int lbits, int rbits, VlUnpacked<SData, T_Depth>& q,
+                                WDataInP rwp) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i)
+        q[i] = VL_SEL_IWII(rbits, rwp, (T_Depth - 1 - i) * lbits, lbits) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_IW(int lbits, int rbits, VlUnpacked<IData, T_Depth>& q,
+                                WDataInP rwp) {
+    const IData mask = VL_MASK_I(lbits);
+    for (size_t i = 0; i < T_Depth; ++i)
+        q[i] = VL_SEL_IWII(rbits, rwp, (T_Depth - 1 - i) * lbits, lbits) & mask;
+}
+
+template <std::size_t T_Depth>
+static inline void VL_UNPACK_QW(int lbits, int rbits, VlUnpacked<QData, T_Depth>& q,
+                                WDataInP rwp) {
+    const QData mask = VL_MASK_Q(lbits);
+    for (size_t i = 0; i < T_Depth; ++i)
+        q[i] = VL_SEL_QWII(rbits, rwp, (T_Depth - 1 - i) * lbits, lbits) & mask;
+}
+
+template <std::size_t T_Depth, std::size_t N>
+static inline void VL_UNPACK_WW(int lbits, int rbits, VlUnpacked<VlWide<N>, T_Depth>& q,
+                                WDataInP rwp) {
+    for (size_t i = 0; i < T_Depth; ++i)
+        VL_SEL_WWII(lbits, rbits, q[i], rwp, (T_Depth - 1 - i) * lbits, lbits);
+}
+
 // Return QData from double (numeric)
 // EMIT_RULE: VL_RTOIROUND_Q_D:  oclean=dirty; lclean==clean/real
 static inline QData VL_RTOIROUND_Q_D(double lhs) VL_PURE {
@@ -12,13 +12,13 @@
 //*************************************************************************
 ///
 /// \file
-/// \brief Verilator common target specific intrinsics header
+/// \brief Verilator common target-specific intrinsics header
 ///
 /// This file is not part of the Verilated public-facing API.
 ///
-/// It is only for internal use; code using machine specific intrinsics for
+/// It is only for internal use; code using machine-specific intrinsics for
 /// optimization should include this header rather than directly including
-/// he target specific headers. We provide macros to check for availability
+/// the target-specific headers. We provide macros to check for availability
 /// of instruction sets, and a common mechanism to disable them.
 ///
 //*************************************************************************
@@ -0,0 +1,506 @@
+// -*- mode: C++; c-file-style: "cc-mode" -*-
+//*************************************************************************
+//
+// Code available from: https://verilator.org
+//
+// Copyright 2024 by Wilson Snyder.  This program is free software; you can
+// redistribute it and/or modify it under the terms of either the GNU Lesser
+// General Public License Version 3 or the Perl Artistic License Version 2.0.
+// SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
+//
+//=========================================================================
+///
+/// \file
+/// \brief Verilated randomization implementation code
+///
+/// This file must be compiled and linked against all Verilated objects
+/// that use randomization features.
+///
+/// See the internals documentation docs/internals.rst for details.
+///
+//=========================================================================
+
+#include "verilated_random.h"
+
+#include <iostream>
+#include <sstream>
+#include <streambuf>
+
+#define _VL_SOLVER_HASH_LEN 1
+#define _VL_SOLVER_HASH_LEN_TOTAL 4
+
+// clang-format off
+#if defined(__unix__) || defined(__unix) || (defined(__APPLE__) && defined(__MACH__))
+# define _VL_SOLVER_PIPE  // Allow pipe SMT solving.  Needs fork()
+#endif
+
+#ifdef _VL_SOLVER_PIPE
+# include <sys/wait.h>
+# include <fcntl.h>
+#endif
+
+#if defined(_WIN32) || defined(__MINGW32__)
+# include <io.h>  // open, read, write, close
+#endif
+// clang-format on
+
+class Process final : private std::streambuf, public std::iostream {
+    static constexpr int BUFFER_SIZE = 4096;
+    const char* const* m_cmd = nullptr;  // fork() process argv
+#ifdef _VL_SOLVER_PIPE
+    pid_t m_pid = 0;  // fork() process id
+#else
+    int m_pid = 0;  // fork() process id - always zero as disabled
+#endif
+    bool m_pidExited = true;  // If subprocess has exited and can be opened
+    int m_pidStatus = 0;  // fork() process exit status, valid if m_pidExited
+    int m_writeFd = -1;  // File descriptor TO subprocess
+    int m_readFd = -1;  // File descriptor FROM subprocess
+    char m_readBuf[BUFFER_SIZE];
+    char m_writeBuf[BUFFER_SIZE];
+
+public:
+    typedef std::streambuf::traits_type traits_type;
+
+protected:
+    int overflow(int c = traits_type::eof()) override {
+        char c2 = static_cast<char>(c);
+        if (pbase() == pptr()) return 0;
+        size_t size = pptr() - pbase();
+        ssize_t n = ::write(m_writeFd, pbase(), size);
+        if (n == -1) perror("write");
+        if (n <= 0) {
+            wait_report();
+            return traits_type::eof();
+        }
+        if (n == size)
+            setp(m_writeBuf, m_writeBuf + sizeof(m_writeBuf));
+        else
+            setp(m_writeBuf + n, m_writeBuf + sizeof(m_writeBuf));
+        if (c != traits_type::eof()) sputc(c2);
+        return 0;
+    }
+    int underflow() override {
+        sync();
+        ssize_t n = ::read(m_readFd, m_readBuf, sizeof(m_readBuf));
+        if (n == -1) perror("read");
+        if (n <= 0) {
+            wait_report();
+            return traits_type::eof();
+        }
+        setg(m_readBuf, m_readBuf, m_readBuf + n);
+        return traits_type::to_int_type(m_readBuf[0]);
+    }
+    int sync() override {
+        overflow();
+        return 0;
+    }
+
+public:
+    explicit Process(const char* const* const cmd = nullptr)
+        : std::streambuf{}
+        , std::iostream{this}
+        , m_cmd{cmd} {
+        open(cmd);
+    }
+
+    void wait_report() {
+        if (m_pidExited) return;
+#ifdef _VL_SOLVER_PIPE
+        if (waitpid(m_pid, &m_pidStatus, 0) != m_pid) return;
+        if (m_pidStatus) {
+            std::stringstream msg;
+            msg << "Subprocess command `" << m_cmd[0];
+            for (const char* const* arg = m_cmd + 1; *arg; arg++) msg << ' ' << *arg;
+            msg << "' failed: ";
+            if (WIFSIGNALED(m_pidStatus))
+                msg << strsignal(WTERMSIG(m_pidStatus))
+                    << (WCOREDUMP(m_pidStatus) ? " (core dumped)" : "");
+            else if (WIFEXITED(m_pidStatus))
+                msg << "exit status " << WEXITSTATUS(m_pidStatus);
+            const std::string str = msg.str();
+            VL_WARN_MT("", 0, "Process", str.c_str());
+        }
+#endif
+        m_pidExited = true;
+        m_pid = 0;
+        closeFds();
+    }
+
+    void closeFds() {
+        if (m_writeFd != -1) {
+            close(m_writeFd);
+            m_writeFd = -1;
+        }
+        if (m_readFd != -1) {
+            close(m_readFd);
+            m_readFd = -1;
+        }
+    }
+
+    bool open(const char* const* const cmd) {
+        setp(std::begin(m_writeBuf), std::end(m_writeBuf));
+        setg(m_readBuf, m_readBuf, m_readBuf);
+#ifdef _VL_SOLVER_PIPE
+        if (!cmd || !cmd[0]) return false;
+        m_cmd = cmd;
+        int fd_stdin[2];  // Can't use std::array
+        int fd_stdout[2];  // Can't use std::array
+        constexpr int P_RD = 0;
+        constexpr int P_WR = 1;
+
+        if (pipe(fd_stdin) != 0) {
+            perror("Process::open: pipe");
+            return false;
+        }
+        if (pipe(fd_stdout) != 0) {
+            perror("Process::open: pipe");
+            close(fd_stdin[P_RD]);
+            close(fd_stdin[P_WR]);
+            return false;
+        }
+
+        if (fd_stdin[P_RD] <= 2 || fd_stdin[P_WR] <= 2 || fd_stdout[P_RD] <= 2
+            || fd_stdout[P_WR] <= 2) {
+            // We'd have to rearrange all of the FD usages in this case.
+            // Too unlikely; verilator isn't a daemon.
+            fprintf(stderr, "stdin/stdout closed before pipe opened\n");
+            close(fd_stdin[P_RD]);
+            close(fd_stdin[P_WR]);
+            close(fd_stdout[P_RD]);
+            close(fd_stdout[P_WR]);
+            return false;
+        }
+
+        const pid_t pid = fork();
+        if (pid < 0) {
+            perror("Process::open: fork");
+            close(fd_stdin[P_RD]);
+            close(fd_stdin[P_WR]);
+            close(fd_stdout[P_RD]);
+            close(fd_stdout[P_WR]);
+            return false;
+        }
+        if (pid == 0) {
+            // Child
+            close(fd_stdin[P_WR]);
+            dup2(fd_stdin[P_RD], STDIN_FILENO);
+            close(fd_stdout[P_RD]);
+            dup2(fd_stdout[P_WR], STDOUT_FILENO);
+            execvp(cmd[0], const_cast<char* const*>(cmd));
+            std::stringstream msg;
+            msg << "Process::open: execvp(" << cmd[0] << ")";
+            const std::string str = msg.str();
+            perror(str.c_str());
+            _exit(127);
+        }
+        // Parent
+        m_pid = pid;
+        m_pidExited = false;
+        m_pidStatus = 0;
+        m_readFd = fd_stdout[P_RD];
+        m_writeFd = fd_stdin[P_WR];
+
+        close(fd_stdin[P_RD]);
+        close(fd_stdout[P_WR]);
+
+        return true;
+#else
+        return false;
+#endif
+    }
+};
+
+static Process& getSolver() {
+    static Process s_solver;
+    static bool s_done = false;
+    if (s_done) return s_solver;
+    s_done = true;
+
+    static std::vector<const char*> s_argv;
+    static std::string s_program = Verilated::threadContextp()->solverProgram();
+    s_argv.emplace_back(&s_program[0]);
+    for (char* arg = &s_program[0]; *arg; arg++) {
+        if (*arg == ' ') {
+            *arg = '\0';
+            s_argv.emplace_back(arg + 1);
+        }
+    }
+    s_argv.emplace_back(nullptr);
+
+    const char* const* const cmd = &s_argv[0];
+    s_solver.open(cmd);
+    s_solver << "(set-logic QF_ABV)\n";
+    s_solver << "(check-sat)\n";
+    s_solver << "(reset)\n";
+    std::string s;
+    getline(s_solver, s);
+    if (s == "sat") return s_solver;
+
+    std::stringstream msg;
+    msg << "Unable to communicate with SAT solver, please check its installation or specify a "
+           "different one in VERILATOR_SOLVER environment variable.\n";
+    msg << " ... Tried: $";
+    for (const char* const* arg = cmd; *arg; arg++) msg << ' ' << *arg;
+    msg << '\n';
+    const std::string str = msg.str();
+    VL_WARN_MT("", 0, "randomize", str.c_str());
+
+    while (getline(s_solver, s)) {}
+    return s_solver;
+}
+
+std::string readUntilBalanced(std::istream& stream) {
+    std::string result;
+    std::string token;
+    int parenCount = 1;
+    while (stream >> token) {
+        for (const char c : token) {
+            if (c == '(') {
+                ++parenCount;
+            } else if (c == ')') {
+                --parenCount;
+            }
+        }
+        result += token + " ";
+        if (parenCount == 0) break;
+    }
+    return result;
+}
+
+std::string parseNestedSelect(const std::string& nested_select_expr,
+                              std::vector<std::string>& indices) {
+    std::istringstream nestedStream(nested_select_expr);
+    std::string name, idx;
+    nestedStream >> name;
+    if (name == "(select") {
+        const std::string further_nested_expr = readUntilBalanced(nestedStream);
+        name = parseNestedSelect(further_nested_expr, indices);
+    }
+    std::getline(nestedStream, idx, ')');
+    indices.push_back(idx);
+    return name;
+}
+
+std::string flattenIndices(const std::vector<std::string>& indices, const VlRandomVar* const var) {
+    int flattenedIndex = 0;
+    int multiplier = 1;
+    for (int i = indices.size() - 1; i >= 0; --i) {
+        int indexValue = 0;
+        std::string trimmedIndex = indices[i];
+
+        trimmedIndex.erase(0, trimmedIndex.find_first_not_of(" \t"));
+        trimmedIndex.erase(trimmedIndex.find_last_not_of(" \t") + 1);
+
+        if (trimmedIndex.find("#x") == 0) {
+            indexValue = std::strtoul(trimmedIndex.substr(2).c_str(), nullptr, 16);
+        } else if (trimmedIndex.find("#b") == 0) {
+            indexValue = std::strtoul(trimmedIndex.substr(2).c_str(), nullptr, 2);
+        } else {
+            indexValue = std::strtoul(trimmedIndex.c_str(), nullptr, 10);
+        }
+        const int length = var->getLength(i);
+        if (length == -1) {
+            VL_WARN_MT(__FILE__, __LINE__, "randomize",
+                       "Internal: Wrong Call: Only RandomArray can call getLength()");
+            break;
+        }
+        flattenedIndex += indexValue * multiplier;
+        multiplier *= length;
+    }
+    std::string hexString = std::to_string(flattenedIndex);
+    while (hexString.size() < 8) { hexString.insert(0, "0"); }
+    return "#x" + hexString;
+}
+//======================================================================
+// VlRandomizer:: Methods
+
+void VlRandomVar::emitGetValue(std::ostream& s) const { s << ' ' << m_name; }
+void VlRandomVar::emitExtract(std::ostream& s, int i) const {
+    s << " ((_ extract " << i << ' ' << i << ") " << m_name << ')';
+}
+void VlRandomVar::emitType(std::ostream& s) const { s << "(_ BitVec " << width() << ')'; }
+int VlRandomVar::totalWidth() const { return m_width; }
+static bool parseSMTNum(int obits, WDataOutP owp, const std::string& val) {
+    int i;
+    for (i = 0; val[i] && val[i] != '#'; i++) {}
+    if (val[i++] != '#') return false;
+    switch (val[i++]) {
+    case 'b': _vl_vsss_based(owp, obits, 1, &val[i], 0, val.size() - i); break;
+    case 'o': _vl_vsss_based(owp, obits, 3, &val[i], 0, val.size() - i); break;
+    case 'h':  // FALLTHRU
+    case 'x': _vl_vsss_based(owp, obits, 4, &val[i], 0, val.size() - i); break;
+    default:
+        VL_WARN_MT(__FILE__, __LINE__, "randomize",
+                   "Internal: Unable to parse solver's randomized number");
+        return false;
+    }
+    return true;
+}
+bool VlRandomVar::set(const std::string& idx, const std::string& val) const {
+    VlWide<VL_WQ_WORDS_E> qowp;
+    VL_SET_WQ(qowp, 0ULL);
+    WDataOutP owp = qowp;
+    const int obits = width();
+    VlWide<VL_WQ_WORDS_E> qiwp;
+    VL_SET_WQ(qiwp, 0ULL);
+    if (!idx.empty() && !parseSMTNum(64, qiwp, idx)) return false;
+    const int nidx = qiwp[0];
+    if (obits > VL_QUADSIZE) owp = reinterpret_cast<WDataOutP>(datap(nidx));
+    if (!parseSMTNum(obits, owp, val)) return false;
+
+    if (obits <= VL_BYTESIZE) {
+        CData* const p = static_cast<CData*>(datap(nidx));
+        *p = VL_CLEAN_II(obits, obits, owp[0]);
+    } else if (obits <= VL_SHORTSIZE) {
+        SData* const p = static_cast<SData*>(datap(nidx));
+        *p = VL_CLEAN_II(obits, obits, owp[0]);
+    } else if (obits <= VL_IDATASIZE) {
+        IData* const p = static_cast<IData*>(datap(nidx));
+        *p = VL_CLEAN_II(obits, obits, owp[0]);
+    } else if (obits <= VL_QUADSIZE) {
+        QData* const p = static_cast<QData*>(datap(nidx));
+        *p = VL_CLEAN_QQ(obits, obits, VL_SET_QW(owp));
+    } else {
+        _vl_clean_inplace_w(obits, owp);
+    }
+    return true;
+}
+
+void VlRandomizer::randomConstraint(std::ostream& os, VlRNG& rngr, int bits) {
+    const IData hash = VL_RANDOM_RNG_I(rngr) & ((1 << bits) - 1);
+    int varBits = 0;
+    for (const auto& var : m_vars) varBits += var.second->totalWidth();
+    os << "(= #b";
+    for (int i = bits - 1; i >= 0; i--) os << (VL_BITISSET_I(hash, i) ? '1' : '0');
+    if (bits > 1) os << " (concat";
+    for (int i = 0; i < bits; i++) {
+        IData varBitsLeft = varBits;
+        IData varBitsWant = (varBits + 1) / 2;
+        if (varBits > 2) os << " (bvxor";
+        for (const auto& var : m_vars) {
+            for (int j = 0; j < var.second->totalWidth(); j++, varBitsLeft--) {
+                const bool doEmit = (VL_RANDOM_RNG_I(rngr) % varBitsLeft) < varBitsWant;
+                if (doEmit) {
+                    var.second->emitExtract(os, j);
+                    if (--varBitsWant == 0) break;
+                }
+            }
+            if (varBitsWant == 0) break;
+        }
+        if (varBits > 2) os << ')';
+    }
+    if (bits > 1) os << ')';
+    os << ')';
+}
+
+bool VlRandomizer::next(VlRNG& rngr) {
+    if (m_vars.empty()) return true;
+    std::iostream& f = getSolver();
+    if (!f) return false;
+
+    f << "(set-option :produce-models true)\n";
+    f << "(set-logic QF_ABV)\n";
+    f << "(define-fun __Vbv ((b Bool)) (_ BitVec 1) (ite b #b1 #b0))\n";
+    f << "(define-fun __Vbool ((v (_ BitVec 1))) Bool (= #b1 v))\n";
+    for (const auto& var : m_vars) {
+        f << "(declare-fun " << var.second->name() << " () ";
+        var.second->emitType(f);
+        f << ")\n";
+    }
+    for (const std::string& constraint : m_constraints) {
+        f << "(assert (= #b1 " << constraint << "))\n";
+    }
+    f << "(check-sat)\n";
+
+    bool sat = parseSolution(f);
+    if (!sat) {
+        f << "(reset)\n";
+        return false;
+    }
+    for (int i = 0; i < _VL_SOLVER_HASH_LEN_TOTAL && sat; i++) {
+        f << "(assert ";
+        randomConstraint(f, rngr, _VL_SOLVER_HASH_LEN);
+        f << ")\n";
+        f << "\n(check-sat)\n";
+        sat = parseSolution(f);
+    }
+
+    f << "(reset)\n";
+    return true;
+}
+
+bool VlRandomizer::parseSolution(std::iostream& f) {
+    std::string sat;
+    do { std::getline(f, sat); } while (sat == "");
+
+    if (sat == "unsat") return false;
+    if (sat != "sat") {
+        std::stringstream msg;
+        msg << "Internal: Solver error: " << sat;
+        const std::string str = msg.str();
+        VL_WARN_MT(__FILE__, __LINE__, "randomize", str.c_str());
+        return false;
+    }
+
+    f << "(get-value (";
+    for (const auto& var : m_vars) var.second->emitGetValue(f);
+    f << "))\n";
+
+    // Quasi-parse S-expression of the form ((x #xVALUE) (y #bVALUE) (z #xVALUE))
+    char c;
+    f >> c;
+    if (c != '(') {
+        VL_WARN_MT(__FILE__, __LINE__, "randomize",
+                   "Internal: Unable to parse solver's response: invalid S-expression");
+        return false;
+    }
+
+    while (true) {
+        f >> c;
+        if (c == ')') break;
+        if (c != '(') {
+            VL_WARN_MT(__FILE__, __LINE__, "randomize",
+                       "Internal: Unable to parse solver's response: invalid S-expression");
+            return false;
+        }
+        std::string name, idx, value;
+        std::vector<std::string> indices;
+        f >> name;
+        indices.clear();
+        if (name == "(select") {
+            const std::string selectExpr = readUntilBalanced(f);
+            name = parseNestedSelect(selectExpr, indices);
+            idx = indices[0];
+        }
+        std::getline(f, value, ')');
+        const auto it = m_vars.find(name);
+        if (it == m_vars.end()) continue;
+        const VlRandomVar& varr = *it->second;
+        if (m_randmode && !varr.randModeIdxNone()) {
+            if (!(m_randmode->at(varr.randModeIdx()))) continue;
+        }
+        if (indices.size() > 1) {
+            const std::string flattenedIndex = flattenIndices(indices, &varr);
+            varr.set(flattenedIndex, value);
+        } else {
+            varr.set(idx, value);
+        }
+    }
+    return true;
+}
+
+void VlRandomizer::hard(std::string&& constraint) {
+    m_constraints.emplace_back(std::move(constraint));
+}
+
+void VlRandomizer::clear() { m_constraints.clear(); }
+
+#ifdef VL_DEBUG
+void VlRandomizer::dump() const {
+    for (const auto& var : m_vars) {
+        VL_PRINTF("Variable (%d): %s\n", var.second->width(), var.second->name());
+    }
+    for (const std::string& c : m_constraints) VL_PRINTF("Constraint: %s\n", c.c_str());
+}
+#endif
@@ -0,0 +1,239 @@
+// -*- mode: C++; c-file-style: "cc-mode" -*-
+//*************************************************************************
+//
+// Code available from: https://verilator.org
+//
+// Copyright 2024 by Wilson Snyder.  This program is free software; you can
+// redistribute it and/or modify it under the terms of either the GNU Lesser
+// General Public License Version 3 or the Perl Artistic License Version 2.0.
+// SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
+//
+//*************************************************************************
+///
+/// \file
+/// \brief Verilated randomization header
+///
+/// This file is included automatically by Verilator in some of the C++ files
+/// it generates if randomization features are used.
+///
+/// This file is not part of the Verilated public-facing API.
+/// It is only for internal use.
+///
+/// See the internals documentation docs/internals.rst for details.
+///
+//*************************************************************************
+#ifndef VERILATOR_VERILATED_RANDOM_H_
+#define VERILATOR_VERILATED_RANDOM_H_
+
+#include "verilated.h"
+
+#include <ostream>
+
+//=============================================================================
+// VlRandomExpr and subclasses represent expressions for the constraint solver.
+
+class VlRandomVar VL_NOT_FINAL {
+    const char* const m_name;  // Variable name
+    void* const m_datap;  // Reference to variable data
+    const int m_width;  // Variable width in bits
+    const int m_dimension;  //Variable dimension, default is 0
+    const std::uint32_t m_randModeIdx;  // rand_mode index
+
+public:
+    VlRandomVar(const char* name, int width, void* datap, int dimension, std::uint32_t randModeIdx)
+        : m_name{name}
+        , m_datap{datap}
+        , m_width{width}
+        , m_dimension{dimension}
+        , m_randModeIdx{randModeIdx} {}
+    virtual ~VlRandomVar() = default;
+    const char* name() const { return m_name; }
+    int width() const { return m_width; }
+    int dimension() const { return m_dimension; }
+    virtual void* datap(int idx) const { return m_datap; }
+    std::uint32_t randModeIdx() const { return m_randModeIdx; }
+    bool randModeIdxNone() const { return randModeIdx() == std::numeric_limits<unsigned>::max(); }
+    bool set(const std::string& idx, const std::string& val) const;
+    virtual void emitGetValue(std::ostream& s) const;
+    virtual void emitExtract(std::ostream& s, int i) const;
+    virtual void emitType(std::ostream& s) const;
+    virtual int totalWidth() const;
+    virtual int getLength(int dimension) const { return -1; }
+};
+
+template <typename T>
+class VlRandomQueueVar final : public VlRandomVar {
+public:
+    VlRandomQueueVar(const char* name, int width, void* datap, int dimension,
+                     std::uint32_t randModeIdx)
+        : VlRandomVar{name, width, datap, dimension, randModeIdx} {}
+    void* datap(int idx) const override {
+        return &static_cast<T*>(VlRandomVar::datap(idx))->atWrite(idx);
+    }
+    void emitSelect(std::ostream& s, int i) const {
+        s << " (select " << name() << " #x";
+        for (int j = 28; j >= 0; j -= 4) s << "0123456789abcdef"[(i >> j) & 0xf];
+        s << ')';
+    }
+    void emitGetValue(std::ostream& s) const override {
+        const int length = static_cast<T*>(VlRandomVar::datap(0))->size();
+        for (int i = 0; i < length; i++) emitSelect(s, i);
+    }
+    void emitType(std::ostream& s) const override {
+        s << "(Array (_ BitVec 32) (_ BitVec " << width() << "))";
+    }
+    int totalWidth() const override {
+        const int length = static_cast<T*>(VlRandomVar::datap(0))->size();
+        return width() * length;
+    }
+    void emitExtract(std::ostream& s, int i) const override {
+        const int j = i / width();
+        i = i % width();
+        s << " ((_ extract " << i << ' ' << i << ')';
+        emitSelect(s, j);
+        s << ')';
+    }
+};
+
+template <typename T>
+class VlRandomArrayVar final : public VlRandomVar {
+public:
+    VlRandomArrayVar(const char* name, int width, void* datap, int dimension,
+                     std::uint32_t randModeIdx)
+        : VlRandomVar{name, width, datap, dimension, randModeIdx} {}
+
+    void* datap(int idx) const override {
+        if (idx < 0) return &static_cast<T*>(VlRandomVar::datap(0))->operator[](0);
+        std::vector<size_t> indices(dimension());
+        for (int dim = dimension() - 1; dim >= 0; --dim) {
+            const int length = getLength(dim);
+            indices[dim] = idx % length;
+            idx /= length;
+        }
+        return &static_cast<T*>(VlRandomVar::datap(0))->find_element(indices);
+    }
+
+    void emitSelect(std::ostream& s, const std::vector<int>& indices) const {
+        for (size_t idx = 0; idx < indices.size(); ++idx) s << "(select ";
+        s << name();
+        for (size_t idx = 0; idx < indices.size(); ++idx) {
+            s << " #x";
+            for (int j = 28; j >= 0; j -= 4) {
+                s << "0123456789abcdef"[(indices[idx] >> j) & 0xf];
+            }
+            s << ")";
+        }
+    }
+
+    int getLength(int dimension) const override {
+        const auto var = static_cast<const T*>(datap(-1));
+        const int lenth = var->find_length(dimension);
+        return lenth;
+    }
+
+    void emitGetValue(std::ostream& s) const override {
+        const int total_dimensions = dimension();
+        std::vector<int> lengths;
+        for (int dim = 0; dim < total_dimensions; dim++) {
+            const int len = getLength(dim);
+            lengths.push_back(len);
+        }
+        std::vector<int> indices(total_dimensions, 0);
+        while (true) {
+            emitSelect(s, indices);
+            int currentDimension = total_dimensions - 1;
+            while (currentDimension >= 0
+                   && ++indices[currentDimension] >= lengths[currentDimension]) {
+                indices[currentDimension] = 0;
+                --currentDimension;
+            }
+            if (currentDimension < 0) break;
+        }
+    }
+
+    void emitType(std::ostream& s) const override {
+        if (dimension() > 0) {
+            for (int i = 0; i < dimension(); ++i) s << "(Array (_ BitVec 32) ";
+            s << "(_ BitVec " << width() << ")";
+            for (int i = 0; i < dimension(); ++i) s << ")";
+        }
+    }
+
+    int totalWidth() const override {
+        int totalLength = 1;
+        for (int dim = 0; dim < dimension(); ++dim) {
+            const int length = getLength(dim);
+            if (length == -1) return 0;
+            totalLength *= length;
+        }
+        return width() * totalLength;
+    }
+
+    void emitExtract(std::ostream& s, int i) const override {
+        const int j = i / width();
+        i = i % width();
+        std::vector<int> indices(dimension());
+        int idx = j;
+        for (int dim = dimension() - 1; dim >= 0; --dim) {
+            int length = getLength(dim);
+            indices[dim] = idx % length;
+            idx /= length;
+        }
+        s << " ((_ extract " << i << ' ' << i << ')';
+        emitSelect(s, indices);
+        s << ')';
+    }
+};
+//=============================================================================
+// VlRandomizer is the object holding constraints and variable references.
+
+class VlRandomizer final {
+    // MEMBERS
+    std::vector<std::string> m_constraints;  // Solver-dependent constraints
+    std::map<std::string, std::shared_ptr<const VlRandomVar>> m_vars;  // Solver-dependent
+                                                                       // variables
+    const VlQueue<CData>* m_randmode;  // rand_mode state;
+
+    // PRIVATE METHODS
+    void randomConstraint(std::ostream& os, VlRNG& rngr, int bits);
+    bool parseSolution(std::iostream& file);
+
+public:
+    // CONSTRUCTORS
+    VlRandomizer() = default;
+    ~VlRandomizer() = default;
+
+    // METHODS
+    // Finds the next solution satisfying the constraints
+    bool next(VlRNG& rngr);
+    template <typename T>
+    void write_var(T& var, int width, const char* name, int dimension,
+                   std::uint32_t randmodeIdx = std::numeric_limits<std::uint32_t>::max()) {
+        if (m_vars.find(name) != m_vars.end()) return;
+        // TODO: make_unique once VlRandomizer is per-instance not per-ref
+        m_vars[name]
+            = std::make_shared<const VlRandomVar>(name, width, &var, dimension, randmodeIdx);
+    }
+    template <typename T>
+    void write_var(VlQueue<T>& var, int width, const char* name, int dimension,
+                   std::uint32_t randmodeIdx = std::numeric_limits<std::uint32_t>::max()) {
+        if (m_vars.find(name) != m_vars.end()) return;
+        m_vars[name] = std::make_shared<const VlRandomQueueVar<VlQueue<T>>>(
+            name, width, &var, dimension, randmodeIdx);
+    }
+    template <typename T, std::size_t N>
+    void write_var(VlUnpacked<T, N>& var, int width, const char* name, int dimension,
+                   std::uint32_t randmodeIdx = std::numeric_limits<std::uint32_t>::max()) {
+        if (m_vars.find(name) != m_vars.end()) return;
+        m_vars[name] = std::make_shared<const VlRandomArrayVar<VlUnpacked<T, N>>>(
+            name, width, &var, dimension, randmodeIdx);
+    }
+    void hard(std::string&& constraint);
+    void clear();
+    void set_randmode(const VlQueue<CData>& randmode) { m_randmode = &randmode; }
+#ifdef VL_DEBUG
+    void dump() const;
+#endif
+};
+
+#endif  // Guard
@@ -95,8 +95,7 @@ void VerilatedDeserialize::header() VL_MT_UNSAFE_ONE {
    if (VL_UNLIKELY(os.readDiffers(VLTSAVE_HEADER_STR, std::strlen(VLTSAVE_HEADER_STR)))) {
        const std::string fn = filename();
        const std::string msg
-            = std::
-                  string{"Can't deserialize; file has wrong header signature, or file not found: "}
+            = "Can't deserialize; file has wrong header signature, or file not found: "s
              + filename();
        VL_FATAL_MT(fn.c_str(), 0, "", msg.c_str());
        // Die before we close() as close would infinite loop
@@ -40,7 +40,7 @@ package std;

      task put(T message);
 `ifdef VERILATOR_TIMING
-         if (m_bound != 0)
+         while (m_bound != 0 && m_queue.size() >= m_bound)
            wait (m_queue.size() < m_bound);
         m_queue.push_back(message);
 `endif
@@ -56,7 +56,9 @@ package std;

      task get(ref T message);
 `ifdef VERILATOR_TIMING
-         wait (m_queue.size() > 0);
+         while (m_queue.size() == 0) begin
+            wait (m_queue.size() > 0);
+         end
         message = m_queue.pop_front();
 `endif
      endtask
@@ -71,7 +73,9 @@ package std;

      task peek(ref T message);
 `ifdef VERILATOR_TIMING
-         wait (m_queue.size() > 0);
+         while (m_queue.size() == 0) begin
+            wait (m_queue.size() > 0);
+         end
         message = m_queue[0];
 `endif
      endtask
@@ -98,7 +102,9 @@ package std;

      task get(int keyCount = 1);
 `ifdef VERILATOR_TIMING
-         wait (m_keyCount >= keyCount);
+         while (m_keyCount < keyCount) begin
+            wait (m_keyCount >= keyCount);
+         end
         m_keyCount -= keyCount;
 `endif
      endtask
@@ -189,4 +195,7 @@ package std;
         $urandom(s.atoi());  // Set the seed using a string
      endfunction
   endclass
+   function int randomize();
+      randomize = 0;
+   endfunction
 endpackage
@@ -67,9 +67,16 @@ void VlDelayScheduler::resume() {
    }

    if (!m_zeroDelayed.empty()) {
-        for (auto&& handle : m_zeroDelayed) handle.resume();
-        m_zeroDelayed.clear();
+        // First, we need to move the coroutines out of the queue, as a resumed coroutine can
+        // suspend on #0 again, adding itself to the queue, which can result in reallocating the
+        // queue mid-iteration.
+        // We swap with the m_zeroDlyResumed field to keep the allocated buffer.
+        m_zeroDlyResumed.swap(m_zeroDelayed);
+        for (auto&& handle : m_zeroDlyResumed) handle.resume();
+        m_zeroDlyResumed.clear();
        resumed = true;
+        // We are now in the Active region, so any coroutines added to m_zeroDelayed in the
+        // meantime will have to wait until the next Inactive region.
    }

    if (!resumed) {
@@ -32,14 +32,18 @@
 // clang-format off
 // Some preprocessor magic to support both Clang and GCC coroutines with both libc++ and libstdc++
 #if defined _LIBCPP_VERSION  // libc++
-# if __clang_major__ > 13  // Clang > 13 warns that coroutine types in std::experimental are deprecated
-#  pragma clang diagnostic push
-#  pragma clang diagnostic ignored "-Wdeprecated-experimental-coroutine"
+# if defined(__has_include) && !__has_include(<coroutine>) && __has_include(<experimental/coroutine>)
+#  if __clang_major__ > 13  // Clang > 13 warns that coroutine types in std::experimental are deprecated
+#   pragma clang diagnostic push
+#   pragma clang diagnostic ignored "-Wdeprecated-experimental-coroutine"
+#  endif
+#  include <experimental/coroutine>
+   namespace std {
+       using namespace experimental;  // Bring std::experimental into the std namespace
+   }
+# else
+#  include <coroutine>
 # endif
-# include <experimental/coroutine>
-  namespace std {
-      using namespace experimental;  // Bring std::experimental into the std namespace
-  }
 #else
 # if defined __clang__ && defined __GLIBCXX__ && !defined __cpp_impl_coroutine
 #  define __cpp_impl_coroutine 1  // Clang doesn't define this, but it's needed for libstdc++
@@ -165,6 +169,9 @@ class VlDelayScheduler final {
    VerilatedContext& m_context;
    VlDelayedCoroutineQueue m_queue;  // Coroutines to be restored at a certain simulation time
    std::vector<VlCoroutineHandle> m_zeroDelayed;  // Coroutines waiting for #0
+    std::vector<VlCoroutineHandle> m_zeroDlyResumed;  // Coroutines that waited for #0 and are
+                                                      // to be resumed. Kept as a field to avoid
+                                                      // reallocation.

 public:
    // CONSTRUCTORS
@@ -193,9 +200,9 @@ public:
            VlProcessRef process;  // Data of the suspended process, null if not needed
            VlDelayedCoroutineQueue& queue;
            std::vector<VlCoroutineHandle>& queueZeroDelay;
-            uint64_t delay;
-            VlDelayPhase phase;
-            VlFileLineDebug fileline;
+            const uint64_t delay;
+            const VlDelayPhase phase;
+            const VlFileLineDebug fileline;

            bool await_ready() const { return false; }  // Always suspend
            void await_suspend(std::coroutine_handle<> coro) {
@@ -208,7 +215,7 @@ public:
            void await_resume() const {}
        };

-        VlDelayPhase phase = (delay == 0) ? VlDelayPhase::INACTIVE : VlDelayPhase::ACTIVE;
+        const VlDelayPhase phase = (delay == 0) ? VlDelayPhase::INACTIVE : VlDelayPhase::ACTIVE;
 #ifdef VL_DEBUG
        if (phase == VlDelayPhase::INACTIVE) {
            VL_WARN_MT(filename, lineno, VL_UNKNOWN,
@@ -30,8 +30,9 @@
 #include <condition_variable>
 #include <memory>
 #include <string>
-#include <unordered_map>
-#include <unordered_set>
+#include <type_traits>
+#include <map>
+#include <set>
 #include <vector>

 #include <deque>
@@ -48,10 +49,12 @@ class VerilatedTraceOffloadBuffer;
 //=============================================================================
 // Common enumerations

-enum class VerilatedTracePrefixType : uint32_t {
+enum class VerilatedTracePrefixType : uint8_t {
    // Note: Entries must match VTracePrefixType (by name, not necessarily by value)
    ARRAY_PACKED,
    ARRAY_UNPACKED,
+    ROOTIO_MODULE,  // $rootio, used when name()=="", other modules become peers
+    ROOTIO_WRAPPER,  // "Above" ROOTIO_MODULE
    SCOPE_MODULE,
    SCOPE_INTERFACE,
    STRUCT_PACKED,
@@ -60,7 +63,7 @@ enum class VerilatedTracePrefixType : uint32_t {
 };

 // Direction attribute for ports
-enum class VerilatedTraceSigDirection : uint32_t {
+enum class VerilatedTraceSigDirection : uint8_t {
    NONE,
    INPUT,
    OUTPUT,
@@ -68,7 +71,7 @@ enum class VerilatedTraceSigDirection : uint32_t {
 };

 // Kind of signal. Similar to nettype but with a few more alternatives
-enum class VerilatedTraceSigKind : uint32_t {
+enum class VerilatedTraceSigKind : uint8_t {
    PARAMETER,
    SUPPLY0,
    SUPPLY1,
@@ -80,7 +83,7 @@ enum class VerilatedTraceSigKind : uint32_t {
 };

 // Base data type of signal
-enum class VerilatedTraceSigType : uint32_t {
+enum class VerilatedTraceSigType : uint8_t {
    DOUBLE,
    INTEGER,
    BIT,
@@ -197,8 +200,8 @@ public:
 //=============================================================================
 // VerilatedTrace

-// T_Trace is the format specific subclass of VerilatedTrace.
-// T_Buffer is the format specific base class of VerilatedTraceBuffer.
+// T_Trace is the format-specific subclass of VerilatedTrace.
+// T_Buffer is the format-specific base class of VerilatedTraceBuffer.
 template <class T_Trace, class T_Buffer>
 class VerilatedTrace VL_NOT_FINAL {
 public:
@@ -294,7 +297,7 @@ private:
    uint64_t m_timeLastDump = 0;  // Last time we did a dump
    bool m_didSomeDump = false;  // Did at least one dump (i.e.: m_timeLastDump is valid)
    VerilatedContext* m_contextp = nullptr;  // The context used by the traced models
-    std::unordered_set<const VerilatedModel*> m_models;  // The collection of models being traced
+    std::set<const VerilatedModel*> m_models;  // The collection of models being traced

    void addCallbackRecord(std::vector<CallbackRecord>& cbVec, CallbackRecord&& cbRec)
        VL_MT_SAFE_EXCLUDES(m_mutex);
@@ -347,7 +350,7 @@ private:

 protected:
    //=========================================================================
-    // Internals available to format specific implementations
+    // Internals available to format-specific implementations

    mutable VerilatedMutex m_mutex;  // Ensure dump() etc only called from single thread

@@ -380,7 +383,7 @@ protected:
    }

    //=========================================================================
-    // Virtual functions to be provided by the format specific implementation
+    // Virtual functions to be provided by the format-specific implementation

    // Called when the trace moves forward to a new time point
    virtual void emitTimeChange(uint64_t timeui) = 0;
@@ -437,7 +440,7 @@ public:
 //=============================================================================
 // VerilatedTraceBuffer

-// T_Buffer is the format specific base class of VerilatedTraceBuffer.
+// T_Buffer is the format-specific base class of VerilatedTraceBuffer.
 // The format-specific hot-path methods use duck-typing via T_Buffer for performance.
 template <class T_Buffer>
 class VerilatedTraceBuffer VL_NOT_FINAL : public T_Buffer {
@@ -463,7 +466,7 @@ public:
    // Hot path internal interface to Verilator generated code

    // Implementation note: We rely on the following duck-typed implementations
-    // in the derived class T_Derived. These emit* functions record a format
+    // in the derived class T_Derived. These emit* functions record a format-
    // specific trace entry. Normally one would use pure virtual functions for
    // these here, but we cannot afford dynamic dispatch for calling these as
    // this is very hot code during tracing.
@@ -486,7 +489,8 @@ public:
    void fullQData(uint32_t* oldp, QData newval, int bits);
    void fullWData(uint32_t* oldp, const WData* newvalp, int bits);
    void fullDouble(uint32_t* oldp, double newval);
-    void fullEvent(uint32_t* oldp, const VlEventBase* newval);
+    void fullEvent(uint32_t* oldp, const VlEventBase* newvalp);
+    void fullEventTriggered(uint32_t* oldp);

    // In non-offload mode, these are called directly by the trace callbacks,
    // and are called chg*. In offload mode, they are called by the worker
@@ -523,9 +527,10 @@ public:
            }
        }
    }
-    VL_ATTR_ALWINLINE void chgEvent(uint32_t* oldp, const VlEventBase* newval) {
-        fullEvent(oldp, newval);
+    VL_ATTR_ALWINLINE void chgEvent(uint32_t* oldp, const VlEventBase* newvalp) {
+        if (newvalp->isTriggered()) fullEvent(oldp, newvalp);
    }
+    VL_ATTR_ALWINLINE void chgEventTriggered(uint32_t* oldp) { fullEventTriggered(oldp); }
    VL_ATTR_ALWINLINE void chgDouble(uint32_t* oldp, double newval) {
        double old;
        std::memcpy(&old, oldp, sizeof(old));
@@ -536,7 +541,7 @@ public:
 //=============================================================================
 // VerilatedTraceOffloadBuffer

-// T_Buffer is the format specific base class of VerilatedTraceBuffer.
+// T_Buffer is the format-specific base class of VerilatedTraceBuffer.
 // The format-specific hot-path methods use duck-typing via T_Buffer for performance.
 template <class T_Buffer>
 class VerilatedTraceOffloadBuffer final : public VerilatedTraceBuffer<T_Buffer> {
@@ -604,7 +609,10 @@ public:
        m_offloadBufferWritep += 4;
        VL_DEBUG_IF(assert(m_offloadBufferWritep <= m_offloadBufferEndp););
    }
-    void chgEvent(uint32_t code, const VlEventBase* newval) {
+    void chgEvent(uint32_t code, const VlEventBase* newvalp) {
+        if (newvalp->isTriggered()) chgEventTriggered(code);
+    }
+    void chgEventTriggered(uint32_t code) {
        m_offloadBufferWritep[0] = VerilatedTraceOffloadCommand::CHG_EVENT;
        m_offloadBufferWritep[1] = code;
        m_offloadBufferWritep += 2;
@@ -12,7 +12,7 @@
 //=============================================================================
 //
 // Verilated tracing implementation code template common to all formats.
-// This file is included by the format specific implementations and
+// This file is included by the format-specific implementations and
 // should not be used otherwise.
 //
 //=============================================================================
@@ -169,7 +169,7 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::offloadWorkerThreadMain() {
                continue;
            case VerilatedTraceOffloadCommand::CHG_EVENT:
                VL_TRACE_OFFLOAD_DEBUG("Command CHG_EVENT " << top);
-                traceBufp->chgEvent(oldp, reinterpret_cast<const VlEventBase*>(readp));
+                traceBufp->chgEventTriggered(oldp);
                continue;

                //===
@@ -299,7 +299,7 @@ VerilatedTrace<VL_SUB_T, VL_BUF_T>::~VerilatedTrace() {
 }

 //=========================================================================
-// Internals available to format specific implementations
+// Internals available to format-specific implementations

 template <>
 void VerilatedTrace<VL_SUB_T, VL_BUF_T>::traceInit() VL_MT_UNSAFE {
@@ -401,7 +401,7 @@ bool VerilatedTrace<VL_SUB_T, VL_BUF_T>::declCode(uint32_t code, const std::stri
 }

 //=========================================================================
-// Internals available to format specific implementations
+// Internals available to format-specific implementations

 template <>
 std::string VerilatedTrace<VL_SUB_T, VL_BUF_T>::timeResStr() const {
@@ -543,7 +543,7 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::dump(uint64_t timeui) VL_MT_SAFE_EXCLUD

    Verilated::quiesce();

-    // Call hook for format specific behaviour
+    // Call hook for format-specific behaviour
    if (VL_UNLIKELY(m_fullDump)) {
        if (!preFullDump()) return;
    } else {
@@ -633,12 +633,13 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addModel(VerilatedModel* modelp)

    // Validate
    if (!newModel) {  // LCOV_EXCL_START
-        VL_FATAL_MT(__FILE__, __LINE__, "",
-                    "The same model has already been added to this trace file");
+        VL_FATAL_MT(
+            __FILE__, __LINE__, "",
+            "The same model has already been added to this trace file or VerilatedContext");
    }
    if (VL_UNCOVERABLE(m_contextp && contextp != m_contextp)) {
        VL_FATAL_MT(__FILE__, __LINE__, "",
-                    "A trace file instance can only handle models from the same context");
+                    "A trace file instance can only handle models from the same VerilatedContext");
    }
    if (VL_UNCOVERABLE(m_didSomeDump)) {
        VL_FATAL_MT(__FILE__, __LINE__, "",
@@ -666,7 +667,7 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addModel(VerilatedModel* modelp)
        VL_FATAL_MT(__FILE__, __LINE__, "", "Cannot use parallel tracing with offloading");
    }  // LCOV_EXCL_STOP

-    // Configure format specific sub class
+    // Configure format-specific sub class
    configure(*(configp.get()));
 }

@@ -817,8 +818,8 @@ VerilatedTraceBuffer<VL_BUF_T>::VerilatedTraceBuffer(Trace& owner)
    , m_sigs_enabledp{owner.m_sigs_enabledp} {}

 // These functions must write the new value back into the old value store,
-// and subsequently call the format specific emit* implementations. Note
-// that this file must be included in the format specific implementation, so
+// and subsequently call the format-specific emit* implementations. Note
+// that this file must be included in the format-specific implementation, so
 // the emit* functions can be inlined for performance.

 template <>
@@ -830,10 +831,17 @@ void VerilatedTraceBuffer<VL_BUF_T>::fullBit(uint32_t* oldp, CData newval) {
 }

 template <>
-void VerilatedTraceBuffer<VL_BUF_T>::fullEvent(uint32_t* oldp, const VlEventBase* newval) {
+void VerilatedTraceBuffer<VL_BUF_T>::fullEvent(uint32_t* oldp, const VlEventBase* newvalp) {
    const uint32_t code = oldp - m_sigs_oldvalp;
-    *oldp = 1;  // Do we really store an "event" ?
-    emitEvent(code, newval);
+    // No need to update *oldp
+    if (newvalp->isTriggered()) emitEvent(code);
+}
+
+template <>
+void VerilatedTraceBuffer<VL_BUF_T>::fullEventTriggered(uint32_t* oldp) {
+    const uint32_t code = oldp - m_sigs_oldvalp;
+    // No need to update *oldp
+    emitEvent(code);
 }

 template <>
@@ -36,7 +36,6 @@
 #include <memory>
 #include <set>
 #include <string>
-#include <unordered_set>
 #include <utility>

 //=========================================================================
@@ -415,8 +414,20 @@ public:

 static int _vl_cmp_w(int words, WDataInP const lwp, WDataInP const rwp) VL_PURE;

+template <std::size_t T_Words>
+struct VlWide;
+
+// Type trait to check if a type is VlWide
+template <typename>
+struct VlIsVlWide : public std::false_type {};
+
+template <std::size_t T_Words>
+struct VlIsVlWide<VlWide<T_Words>> : public std::true_type {};
+
 template <std::size_t T_Words>
 struct VlWide final {
+    static constexpr size_t Words = T_Words;
+
    // MEMBERS
    // This should be the only data member, otherwise generated static initializers need updating
    EData m_storage[T_Words];  // Contents of the packed array
@@ -474,6 +485,8 @@ private:

 public:
    using const_iterator = typename Deque::const_iterator;
+    template <class Func>
+    using WithFuncReturnType = decltype(std::declval<Func>()(0, std::declval<T_Value>()));

 private:
    // MEMBERS
@@ -501,28 +514,29 @@ public:
        return *this;
    }

-    static VlQueue cons(const T_Value& lhs) {
+    // Construct new object from _V_alue and/or _C_ontainer child objects
+    static VlQueue consV(const T_Value& lhs) {
        VlQueue out;
        out.push_back(lhs);
        return out;
    }
-    static VlQueue cons(const T_Value& lhs, const T_Value& rhs) {
+    static VlQueue consVV(const T_Value& lhs, const T_Value& rhs) {
        VlQueue out;
        out.push_back(rhs);
        out.push_back(lhs);
        return out;
    }
-    static VlQueue cons(const VlQueue& lhs, const T_Value& rhs) {
+    static VlQueue consCV(const VlQueue& lhs, const T_Value& rhs) {
        VlQueue out = lhs;
        out.push_front(rhs);
        return out;
    }
-    static VlQueue cons(const T_Value& lhs, const VlQueue& rhs) {
+    static VlQueue consVC(const T_Value& lhs, const VlQueue& rhs) {
        VlQueue out = rhs;
        out.push_back(lhs);
        return out;
    }
-    static VlQueue cons(const VlQueue& lhs, const VlQueue& rhs) {
+    static VlQueue consCC(const VlQueue& lhs, const VlQueue& rhs) {
        VlQueue out = rhs;
        for (const auto& i : lhs.m_deque) out.push_back(i);
        return out;
@@ -581,18 +595,28 @@ public:
        return v;
    }

-    // Setting. Verilog: assoc[index] = v
-    // Can't just overload operator[] or provide a "at" reference to set,
-    // because we need to be able to insert only when the value is set
-    T_Value& at(int32_t index) {
+    // Setting. Verilog: assoc[index] = v (should only be used by dynamic arrays)
+    T_Value& atWrite(int32_t index) {
+        // cppcheck-suppress variableScope
        static thread_local T_Value t_throwAway;
        // Needs to work for dynamic arrays, so does not use T_MaxSize
        if (VL_UNLIKELY(index < 0 || index >= m_deque.size())) {
            t_throwAway = atDefault();
            return t_throwAway;
-        } else {
-            return m_deque[index];
        }
+        return m_deque[index];
+    }
+    // Setting. Verilog: assoc[index] = v (should only be used by queues)
+    T_Value& atWriteAppend(int32_t index) {
+        // cppcheck-suppress variableScope
+        static thread_local T_Value t_throwAway;
+        if (VL_UNLIKELY(index < 0 || index > m_deque.size())) {
+            t_throwAway = atDefault();
+            return t_throwAway;
+        } else if (VL_UNLIKELY(index == m_deque.size())) {
+            push_back(atDefault());
+        }
+        return m_deque[index];
    }
    // Accessing. Verilog: v = assoc[index]
    const T_Value& at(int32_t index) const {
@@ -604,7 +628,7 @@ public:
        }
    }
    // Access with an index counted from end (e.g. q[$])
-    T_Value& atBack(int32_t index) { return at(m_deque.size() - 1 - index); }
+    T_Value& atWriteAppendBack(int32_t index) { return atWriteAppend(m_deque.size() - 1 - index); }
    const T_Value& atBack(int32_t index) const { return at(m_deque.size() - 1 - index); }

    // function void q.insert(index, value);
@@ -740,7 +764,7 @@ public:
        // Can't use std::find_if as need index number
        IData index = 0;
        for (const auto& i : m_deque) {
-            if (with_func(index, i)) return VlQueue::cons(i);
+            if (with_func(index, i)) return VlQueue::consV(i);
            ++index;
        }
        return VlQueue{};
@@ -749,7 +773,7 @@ public:
    VlQueue<IData> find_first_index(Func with_func) const {
        IData index = 0;
        for (const auto& i : m_deque) {
-            if (with_func(index, i)) return VlQueue<IData>::cons(index);
+            if (with_func(index, i)) return VlQueue<IData>::consV(index);
            ++index;
        }
        return VlQueue<IData>{};
@@ -758,7 +782,7 @@ public:
    VlQueue find_last(Func with_func) const {
        IData index = m_deque.size() - 1;
        for (auto& item : vlstd::reverse_view(m_deque)) {
-            if (with_func(index, item)) return VlQueue::cons(item);
+            if (with_func(index, item)) return VlQueue::consV(item);
            --index;
        }
        return VlQueue{};
@@ -767,7 +791,7 @@ public:
    VlQueue<IData> find_last_index(Func with_func) const {
        IData index = m_deque.size() - 1;
        for (auto& item : vlstd::reverse_view(m_deque)) {
-            if (with_func(index, item)) return VlQueue<IData>::cons(index);
+            if (with_func(index, item)) return VlQueue<IData>::consV(index);
            --index;
        }
        return VlQueue<IData>{};
@@ -777,7 +801,7 @@ public:
    VlQueue min() const {
        if (m_deque.empty()) return VlQueue{};
        const auto it = std::min_element(m_deque.cbegin(), m_deque.cend());
-        return VlQueue::cons(*it);
+        return VlQueue::consV(*it);
    }
    template <typename Func>
    VlQueue min(Func with_func) const {
@@ -786,12 +810,12 @@ public:
                                         [&with_func](const IData& a, const IData& b) {
                                             return with_func(0, a) < with_func(0, b);
                                         });
-        return VlQueue::cons(*it);
+        return VlQueue::consV(*it);
    }
    VlQueue max() const {
        if (m_deque.empty()) return VlQueue{};
        const auto it = std::max_element(m_deque.cbegin(), m_deque.cend());
-        return VlQueue::cons(*it);
+        return VlQueue::consV(*it);
    }
    template <typename Func>
    VlQueue max(Func with_func) const {
@@ -800,7 +824,7 @@ public:
                                         [&with_func](const IData& a, const IData& b) {
                                             return with_func(0, a) < with_func(0, b);
                                         });
-        return VlQueue::cons(*it);
+        return VlQueue::consV(*it);
    }

    T_Value r_sum() const {
@@ -809,70 +833,63 @@ public:
        return out;
    }
    template <typename Func>
-    T_Value r_sum(Func with_func) const {
-        T_Value out(0);  // Type must have assignment operator
+    WithFuncReturnType<Func> r_sum(Func with_func) const {
+        WithFuncReturnType<Func> out = WithFuncReturnType<Func>(0);
        IData index = 0;
        for (const auto& i : m_deque) out += with_func(index++, i);
        return out;
    }
    T_Value r_product() const {
-        if (m_deque.empty()) return T_Value(0);
-        auto it = m_deque.cbegin();
-        T_Value out{*it};
-        ++it;
-        for (; it != m_deque.cend(); ++it) out *= *it;
+        if (m_deque.empty()) return T_Value(0);  // The big three do it this way
+        T_Value out = T_Value(1);
+        for (const auto& i : m_deque) out *= i;
        return out;
    }
    template <typename Func>
-    T_Value r_product(Func with_func) const {
-        if (m_deque.empty()) return T_Value(0);
-        auto it = m_deque.cbegin();
+    WithFuncReturnType<Func> r_product(Func with_func) const {
+        if (m_deque.empty()) return WithFuncReturnType<Func>(0);  // The big three do it this way
+        WithFuncReturnType<Func> out = WithFuncReturnType<Func>(1);
        IData index = 0;
-        T_Value out{with_func(index, *it)};
-        ++it;
-        ++index;
-        for (; it != m_deque.cend(); ++it) out *= with_func(index++, *it);
+        for (const auto& i : m_deque) out *= with_func(index++, i);
        return out;
    }
    T_Value r_and() const {
-        if (m_deque.empty()) return T_Value(0);
-        auto it = m_deque.cbegin();
-        T_Value out{*it};
-        ++it;
-        for (; it != m_deque.cend(); ++it) out &= *it;
+        if (m_deque.empty()) return T_Value(0);  // The big three do it this way
+        T_Value out = ~T_Value(0);
+        for (const auto& i : m_deque) out &= i;
        return out;
    }
    template <typename Func>
-    T_Value r_and(Func with_func) const {
-        if (m_deque.empty()) return T_Value(0);
-        auto it = m_deque.cbegin();
+    WithFuncReturnType<Func> r_and(Func with_func) const {
+        if (m_deque.empty()) return WithFuncReturnType<Func>(0);  // The big three do it this way
        IData index = 0;
-        T_Value out{with_func(index, *it)};
-        ++it;
-        ++index;
-        for (; it != m_deque.cend(); ++it) out &= with_func(index, *it);
+        WithFuncReturnType<Func> out = ~WithFuncReturnType<Func>(0);
+        for (const auto& i : m_deque) out &= with_func(index++, i);
        return out;
    }
    T_Value r_or() const {
-        T_Value out(0);  // Type must have assignment operator
+        T_Value out = T_Value(0);
        for (const auto& i : m_deque) out |= i;
        return out;
    }
    template <typename Func>
-    T_Value r_or(Func with_func) const {
-        T_Value out(0);  // Type must have assignment operator
+    WithFuncReturnType<Func> r_or(Func with_func) const {
+        WithFuncReturnType<Func> out = WithFuncReturnType<Func>(0);
        IData index = 0;
        for (const auto& i : m_deque) out |= with_func(index++, i);
        return out;
    }
    T_Value r_xor() const {
-        T_Value out(0);  // Type must have assignment operator
+#ifdef VERILATOR_BIG3_NULLARY_ARITHMETICS_QUIRKS
+        if (m_deque.empty()) return T_Value(0);
+#endif
+        T_Value out = T_Value(0);
        for (const auto& i : m_deque) out ^= i;
        return out;
    }
    template <typename Func>
-    T_Value r_xor(Func with_func) const {
-        T_Value out(0);  // Type must have assignment operator
+    WithFuncReturnType<Func> r_xor(Func with_func) const {
+        WithFuncReturnType<Func> out = WithFuncReturnType<Func>(0);
        IData index = 0;
        for (const auto& i : m_deque) out ^= with_func(index++, i);
        return out;
@@ -891,8 +908,8 @@ public:
    }
 };

-template <class T_Value>
-std::string VL_TO_STRING(const VlQueue<T_Value>& obj) {
+template <class T_Value, size_t T_MaxSize>
+std::string VL_TO_STRING(const VlQueue<T_Value, T_MaxSize>& obj) {
    return obj.to_string();
 }

@@ -909,6 +926,9 @@ private:

 public:
    using const_iterator = typename Map::const_iterator;
+    template <class Func>
+    using WithFuncReturnType
+        = decltype(std::declval<Func>()(std::declval<T_Key>(), std::declval<T_Value>()));

 private:
    // MEMBERS
@@ -1081,7 +1101,7 @@ public:
                  return with_func(i.first, i.second);
              });
        if (it == m_map.end()) return VlQueue<T_Value>{};
-        return VlQueue<T_Value>::cons(it->second);
+        return VlQueue<T_Value>::consV(it->second);
    }
    template <typename Func>
    VlQueue<T_Key> find_first_index(Func with_func) const {
@@ -1090,7 +1110,7 @@ public:
                  return with_func(i.first, i.second);
              });
        if (it == m_map.end()) return VlQueue<T_Value>{};
-        return VlQueue<T_Key>::cons(it->first);
+        return VlQueue<T_Key>::consV(it->first);
    }
    template <typename Func>
    VlQueue<T_Value> find_last(Func with_func) const {
@@ -1098,7 +1118,7 @@ public:
            m_map.crbegin(), m_map.crend(),
            [=](const std::pair<T_Key, T_Value>& i) { return with_func(i.first, i.second); });
        if (it == m_map.rend()) return VlQueue<T_Value>{};
-        return VlQueue<T_Value>::cons(it->second);
+        return VlQueue<T_Value>::consV(it->second);
    }
    template <typename Func>
    VlQueue<T_Key> find_last_index(Func with_func) const {
@@ -1106,7 +1126,7 @@ public:
            m_map.crbegin(), m_map.crend(),
            [=](const std::pair<T_Key, T_Value>& i) { return with_func(i.first, i.second); });
        if (it == m_map.rend()) return VlQueue<T_Value>{};
-        return VlQueue<T_Key>::cons(it->first);
+        return VlQueue<T_Key>::consV(it->first);
    }

    // Reduction operators
@@ -1117,7 +1137,7 @@ public:
            [](const std::pair<T_Key, T_Value>& a, const std::pair<T_Key, T_Value>& b) {
                return a.second < b.second;
            });
-        return VlQueue<T_Value>::cons(it->second);
+        return VlQueue<T_Value>::consV(it->second);
    }
    template <typename Func>
    VlQueue<T_Value> min(Func with_func) const {
@@ -1127,7 +1147,7 @@ public:
            [&with_func](const std::pair<T_Key, T_Value>& a, const std::pair<T_Key, T_Value>& b) {
                return with_func(a.first, a.second) < with_func(b.first, b.second);
            });
-        return VlQueue<T_Value>::cons(it->second);
+        return VlQueue<T_Value>::consV(it->second);
    }
    VlQueue<T_Value> max() const {
        if (m_map.empty()) return VlQueue<T_Value>();
@@ -1136,7 +1156,7 @@ public:
            [](const std::pair<T_Key, T_Value>& a, const std::pair<T_Key, T_Value>& b) {
                return a.second < b.second;
            });
-        return VlQueue<T_Value>::cons(it->second);
+        return VlQueue<T_Value>::consV(it->second);
    }
    template <typename Func>
    VlQueue<T_Value> max(Func with_func) const {
@@ -1146,7 +1166,7 @@ public:
            [&with_func](const std::pair<T_Key, T_Value>& a, const std::pair<T_Key, T_Value>& b) {
                return with_func(a.first, a.second) < with_func(b.first, b.second);
            });
-        return VlQueue<T_Value>::cons(it->second);
+        return VlQueue<T_Value>::consV(it->second);
    }

    T_Value r_sum() const {
@@ -1155,64 +1175,56 @@ public:
        return out;
    }
    template <typename Func>
-    T_Value r_sum(Func with_func) const {
-        T_Value out(0);  // Type must have assignment operator
+    WithFuncReturnType<Func> r_sum(Func with_func) const {
+        WithFuncReturnType<Func> out = WithFuncReturnType<Func>(0);
        for (const auto& i : m_map) out += with_func(i.first, i.second);
        return out;
    }
    T_Value r_product() const {
-        if (m_map.empty()) return T_Value(0);
-        auto it = m_map.cbegin();
-        T_Value out{it->second};
-        ++it;
-        for (; it != m_map.cend(); ++it) out *= it->second;
+        if (m_map.empty()) return T_Value(0);  // The big three do it this way
+        T_Value out = T_Value(1);
+        for (const auto& i : m_map) out *= i.second;
        return out;
    }
    template <typename Func>
-    T_Value r_product(Func with_func) const {
-        if (m_map.empty()) return T_Value(0);
-        auto it = m_map.cbegin();
-        T_Value out{with_func(it->first, it->second)};
-        ++it;
-        for (; it != m_map.cend(); ++it) out *= with_func(it->first, it->second);
+    WithFuncReturnType<Func> r_product(Func with_func) const {
+        if (m_map.empty()) return WithFuncReturnType<Func>(0);  // The big three do it this way
+        WithFuncReturnType<Func> out = WithFuncReturnType<Func>(1);
+        for (const auto& i : m_map) out *= with_func(i.first, i.second);
        return out;
    }
    T_Value r_and() const {
-        if (m_map.empty()) return T_Value(0);
-        auto it = m_map.cbegin();
-        T_Value out{it->second};
-        ++it;
-        for (; it != m_map.cend(); ++it) out &= it->second;
+        if (m_map.empty()) return T_Value(0);  // The big three do it this way
+        T_Value out = ~T_Value(0);
+        for (const auto& i : m_map) out &= i.second;
        return out;
    }
    template <typename Func>
-    T_Value r_and(Func with_func) const {
-        if (m_map.empty()) return T_Value(0);
-        auto it = m_map.cbegin();
-        T_Value out{with_func(it->first, it->second)};
-        ++it;
-        for (; it != m_map.cend(); ++it) out &= with_func(it->first, it->second);
+    WithFuncReturnType<Func> r_and(Func with_func) const {
+        if (m_map.empty()) return WithFuncReturnType<Func>(0);  // The big three do it this way
+        WithFuncReturnType<Func> out = ~WithFuncReturnType<Func>(0);
+        for (const auto& i : m_map) out &= with_func(i.first, i.second);
        return out;
    }
    T_Value r_or() const {
-        T_Value out(0);  // Type must have assignment operator
+        T_Value out = T_Value(0);
        for (const auto& i : m_map) out |= i.second;
        return out;
    }
    template <typename Func>
    T_Value r_or(Func with_func) const {
-        T_Value out(0);  // Type must have assignment operator
+        T_Value out = T_Value(0);
        for (const auto& i : m_map) out |= with_func(i.first, i.second);
        return out;
    }
    T_Value r_xor() const {
-        T_Value out(0);  // Type must have assignment operator
+        T_Value out = T_Value(0);
        for (const auto& i : m_map) out ^= i.second;
        return out;
    }
    template <typename Func>
-    T_Value r_xor(Func with_func) const {
-        T_Value out(0);  // Type must have assignment operator
+    WithFuncReturnType<Func> r_xor(Func with_func) const {
+        WithFuncReturnType<Func> out = WithFuncReturnType<Func>(0);
        for (const auto& i : m_map) out ^= with_func(i.first, i.second);
        return out;
    }
@@ -1300,6 +1312,43 @@ public:
    WData* data() { return &m_storage[0]; }
    const WData* data() const { return &m_storage[0]; }

+    std::size_t size() const { return T_Depth; }
+    // To fit C++14
+    template <std::size_t CurrentDimension = 0, typename U = T_Value>
+    int find_length(int dimension, std::false_type) const {
+        return size();
+    }
+
+    template <std::size_t CurrentDimension = 0, typename U = T_Value>
+    int find_length(int dimension, std::true_type) const {
+        if (dimension == CurrentDimension) {
+            return size();
+        } else {
+            return m_storage[0].template find_length<CurrentDimension + 1>(dimension);
+        }
+    }
+
+    template <std::size_t CurrentDimension = 0>
+    int find_length(int dimension) const {
+        return find_length<CurrentDimension>(dimension, std::is_class<T_Value>{});
+    }
+
+    template <std::size_t CurrentDimension = 0, typename U = T_Value>
+    auto& find_element(const std::vector<size_t>& indices, std::false_type) {
+        return m_storage[indices[CurrentDimension]];
+    }
+
+    template <std::size_t CurrentDimension = 0, typename U = T_Value>
+    auto& find_element(const std::vector<size_t>& indices, std::true_type) {
+        return m_storage[indices[CurrentDimension]].template find_element<CurrentDimension + 1>(
+            indices);
+    }
+
+    template <std::size_t CurrentDimension = 0>
+    auto& find_element(const std::vector<size_t>& indices) {
+        return find_element<CurrentDimension>(indices, std::is_class<T_Value>{});
+    }
+
    T_Value& operator[](size_t index) { return m_storage[index]; }
    const T_Value& operator[](size_t index) const { return m_storage[index]; }

@@ -1309,7 +1358,11 @@ public:
    // Similar to 'neq' above, *this = that used for change detection
    void assign(const VlUnpacked<T_Value, T_Depth>& that) { *this = that; }
    bool operator==(const VlUnpacked<T_Value, T_Depth>& that) const { return !neq(that); }
-    bool operator!=(const VlUnpacked<T_Value, T_Depth>& that) { return neq(that); }
+    bool operator!=(const VlUnpacked<T_Value, T_Depth>& that) const { return neq(that); }
+    // interface to C style arrays (used in ports), see issue #5125
+    bool neq(const T_Value that[T_Depth]) const { return neq(*this, that); }
+    void assign(const T_Value that[T_Depth]) { std::copy_n(that, T_Depth, m_storage); }
+    void operator=(const T_Value that[T_Depth]) { assign(that); }

    // inside (set membership operator)
    bool inside(const T_Value& value) const {
@@ -1386,7 +1439,7 @@ public:
    VlQueue<T_Key> unique_index(Func with_func) const {
        VlQueue<T_Key> out;
        IData index = 0;
-        std::unordered_set<T_Value> saw;
+        std::set<T_Value> saw;
        for (const auto& i : m_storage) {
            const auto i_mapped = with_func(index, i);
            auto it = saw.find(i_mapped);
@@ -1423,7 +1476,7 @@ public:
        // Can't use std::find_if as need index number
        IData index = 0;
        for (const auto& i : m_storage) {
-            if (with_func(index, i)) return VlQueue<T_Value>::cons(i);
+            if (with_func(index, i)) return VlQueue<T_Value>::consV(i);
            ++index;
        }
        return VlQueue<T_Value>{};
@@ -1432,7 +1485,7 @@ public:
    VlQueue<T_Key> find_first_index(Func with_func) const {
        IData index = 0;
        for (const auto& i : m_storage) {
-            if (with_func(index, i)) return VlQueue<IData>::cons(index);
+            if (with_func(index, i)) return VlQueue<IData>::consV(index);
            ++index;
        }
        return VlQueue<T_Key>{};
@@ -1440,14 +1493,14 @@ public:
    template <typename Func>
    VlQueue<T_Value> find_last(Func with_func) const {
        for (int i = T_Depth - 1; i >= 0; i--) {
-            if (with_func(i, m_storage[i])) return VlQueue<T_Value>::cons(m_storage[i]);
+            if (with_func(i, m_storage[i])) return VlQueue<T_Value>::consV(m_storage[i]);
        }
        return VlQueue<T_Value>{};
    }
    template <typename Func>
    VlQueue<T_Key> find_last_index(Func with_func) const {
        for (int i = T_Depth - 1; i >= 0; i--) {
-            if (with_func(i, m_storage[i])) return VlQueue<IData>::cons(i);
+            if (with_func(i, m_storage[i])) return VlQueue<IData>::consV(i);
        }
        return VlQueue<T_Key>{};
    }
@@ -1455,7 +1508,7 @@ public:
    // Reduction operators
    VlQueue<T_Value> min() const {
        const auto it = std::min_element(std::begin(m_storage), std::end(m_storage));
-        return VlQueue<T_Value>::cons(*it);
+        return VlQueue<T_Value>::consV(*it);
    }
    template <typename Func>
    VlQueue<T_Value> min(Func with_func) const {
@@ -1463,11 +1516,11 @@ public:
                                         [&with_func](const IData& a, const IData& b) {
                                             return with_func(0, a) < with_func(0, b);
                                         });
-        return VlQueue<T_Value>::cons(*it);
+        return VlQueue<T_Value>::consV(*it);
    }
    VlQueue<T_Value> max() const {
        const auto it = std::max_element(std::begin(m_storage), std::end(m_storage));
-        return VlQueue<T_Value>::cons(*it);
+        return VlQueue<T_Value>::consV(*it);
    }
    template <typename Func>
    VlQueue<T_Value> max(Func with_func) const {
@@ -1475,7 +1528,7 @@ public:
                                         [&with_func](const IData& a, const IData& b) {
                                             return with_func(0, a) < with_func(0, b);
                                         });
-        return VlQueue<T_Value>::cons(*it);
+        return VlQueue<T_Value>::consV(*it);
    }

    // Dumping. Verilog: str = $sformatf("%p", assoc)
@@ -1499,6 +1552,15 @@ private:
        return false;
    }

+    template <typename T_Val, std::size_t T_Dep>
+    static bool neq(const VlUnpacked<T_Val, T_Dep>& a, const T_Val b[T_Dep]) {
+        for (size_t i = 0; i < T_Dep; ++i) {
+            // Recursive 'neq', in case T_Val is also a VlUnpacked<_, _>
+            if (neq(a.m_storage[i], b[i])) return true;
+        }
+        return false;
+    }
+
    template <typename T_Other>  //
    static bool neq(const T_Other& a, const T_Other& b) {
        // Base case (T_Other is not VlUnpacked<_, _>), fall back on !=
@@ -1511,6 +1573,181 @@ std::string VL_TO_STRING(const VlUnpacked<T_Value, T_Depth>& obj) {
    return obj.to_string();
 }

+//===================================================================
+// Helper to apply the given indices to a target expression
+
+template <size_t Curr, size_t Rank, typename T_Target>
+struct VlApplyIndices final {
+    VL_ATTR_ALWINLINE
+    static auto& apply(T_Target& target, const size_t* indicesp) {
+        return VlApplyIndices<Curr + 1, Rank, decltype(target[indicesp[Curr]])>::apply(
+            target[indicesp[Curr]], indicesp);
+    }
+};
+
+template <size_t Rank, typename T_Target>
+struct VlApplyIndices<Rank, Rank, T_Target> final {
+    VL_ATTR_ALWINLINE
+    static T_Target& apply(T_Target& target, const size_t*) { return target; }
+};
+
+//===================================================================
+// Commit queue for NBAs - currently only for unpacked arrays
+//
+// This data-structure is used to handle non-blocking assignments
+// that might execute a variable number of times in a single
+// evaluation. It has 2 operations:
+// - 'enqueue' will add an update to the queue
+// - 'commit' will apply all enqueued updates to the target variable,
+//   in the order they were enqueued. This ensures the last NBA
+//   takes effect as it is expected.
+// There are 2 specializations of this class below:
+// - A version when a partial element update is not required,
+//   e.g, to handle:
+//      logic [31:0] array[N];
+//      for (int i = 0 ; i < N ; ++i) array[i] <= x;
+//   Here 'enqueue' takes the RHS ('x'), and the array indices ('i')
+//   as arguments.
+// - A different version when a partial element update is required,
+//   e.g. for:
+//      logic [31:0] array[N];
+//      for (int i = 0 ; i < N ; ++i) array[i][3:1] <= y;
+//   Here 'enqueue' takes one additional argument, which is a bitmask
+//   derived from the bit selects (_[3:1]), which masks the bits that
+//   need to be updated, and additionally the RHS is widened to a full
+//   element size, with the bits inserted into the masked region.
+template <typename T_Target,  // Type of the variable this commit queue updates
+          bool Partial,  // Whether partial element updates are necessary
+          // The following we could figure out from 'T_Target using type traits, but passing
+          // explicitly to avoid template expansion, as Verilator already knows them
+          typename T_Element,  // Non-array leaf element type of T_Target array
+          std::size_t T_Rank  // Rank of T_Target (i.e.: how many dimensions it has)
+          >
+class VlNBACommitQueue;
+
+// Specialization for whole element updates only
+template <typename T_Target, typename T_Element, std::size_t T_Rank>
+class VlNBACommitQueue<T_Target, /* Partial: */ false, T_Element, T_Rank> final {
+    // TYPES
+    struct Entry final {
+        T_Element value;
+        size_t indices[T_Rank];
+    };
+
+    // STATE
+    std::vector<Entry> m_pending;  // Pending updates, in program order
+
+public:
+    // CONSTRUCTOR
+    VlNBACommitQueue() = default;
+    VL_UNCOPYABLE(VlNBACommitQueue);
+
+    // METHODS
+    template <typename... Args>
+    void enqueue(const T_Element& value, Args... indices) {
+        m_pending.emplace_back(Entry{value, {indices...}});
+    }
+
+    // Note: T_Commit might be different from T_Target. Specifically, when the signal is a
+    // top-level IO port, T_Commit will be a native C array, while T_Target, will be a VlUnpacked
+    template <typename T_Commit>
+    void commit(T_Commit& target) {
+        if (m_pending.empty()) return;
+        for (const Entry& entry : m_pending) {
+            VlApplyIndices<0, T_Rank, T_Commit>::apply(target, entry.indices) = entry.value;
+        }
+        m_pending.clear();
+    }
+};
+
+// With partial element updates
+template <typename T_Target, typename T_Element, std::size_t T_Rank>
+class VlNBACommitQueue<T_Target, /* Partial: */ true, T_Element, T_Rank> final {
+    // TYPES
+    struct Entry final {
+        T_Element value;
+        T_Element mask;
+        size_t indices[T_Rank];
+    };
+
+    // STATE
+    std::vector<Entry> m_pending;  // Pending updates, in program order
+
+    // STATIC METHODS
+
+    // Binary & | ~ for elements to use for masking in partial updates. Sorry for the templates.
+    template <typename T>
+    VL_ATTR_ALWINLINE static typename std::enable_if<!VlIsVlWide<T>::value, T>::type
+    bAnd(const T& a, const T& b) {
+        return a & b;
+    }
+
+    template <typename T>
+    VL_ATTR_ALWINLINE static typename std::enable_if<VlIsVlWide<T>::value, T>::type
+    bAnd(const T& a, const T& b) {
+        T result;
+        for (size_t i = 0; i < T::Words; ++i) {
+            result.m_storage[i] = a.m_storage[i] & b.m_storage[i];
+        }
+        return result;
+    }
+
+    template <typename T>
+    VL_ATTR_ALWINLINE static typename std::enable_if<!VlIsVlWide<T>::value, T>::type
+    bOr(const T& a, const T& b) {
+        return a | b;
+    }
+
+    template <typename T>
+    VL_ATTR_ALWINLINE static typename std::enable_if<VlIsVlWide<T>::value, T>::type  //
+    bOr(const T& a, const T& b) {
+        T result;
+        for (size_t i = 0; i < T::Words; ++i) {
+            result.m_storage[i] = a.m_storage[i] | b.m_storage[i];
+        }
+        return result;
+    }
+
+    template <typename T>
+    VL_ATTR_ALWINLINE static typename std::enable_if<!VlIsVlWide<T>::value, T>::type
+    bNot(const T& a) {
+        return ~a;
+    }
+
+    template <typename T>
+    VL_ATTR_ALWINLINE static typename std::enable_if<VlIsVlWide<T>::value, T>::type
+    bNot(const T& a) {
+        T result;
+        for (size_t i = 0; i < T::Words; ++i) result.m_storage[i] = ~a.m_storage[i];
+        return result;
+    }
+
+public:
+    // CONSTRUCTOR
+    VlNBACommitQueue() = default;
+    VL_UNCOPYABLE(VlNBACommitQueue);
+
+    // METHODS
+    template <typename... Args>
+    void enqueue(const T_Element& value, const T_Element& mask, Args... indices) {
+        m_pending.emplace_back(Entry{value, mask, {indices...}});
+    }
+
+    // Note: T_Commit might be different from T_Target. Specifically, when the signal is a
+    // top-level IO port, T_Commit will be a native C array, while T_Target, will be a VlUnpacked
+    template <typename T_Commit>
+    void commit(T_Commit& target) {
+        if (m_pending.empty()) return;
+        for (const Entry& entry : m_pending) {  //
+            auto& ref = VlApplyIndices<0, T_Rank, T_Commit>::apply(target, entry.indices);
+            // Maybe inefficient, but it works for now ...
+            const auto oldValue = ref;
+            ref = bOr(bAnd(entry.value, entry.mask), bAnd(oldValue, bNot(entry.mask)));
+        }
+        m_pending.clear();
+    }
+};
+
 //===================================================================
 // Object that VlDeleter is capable of deleting

@@ -1587,12 +1824,16 @@ public:
 };

 //===================================================================
-// Represents the null pointer. Used for setting VlClassRef to null instead of
-// via nullptr_t, to prevent the implicit conversion of 0 to nullptr.
+// Represents the null pointer. Used for:
+// * setting VlClassRef to null instead of via nullptr_t, to prevent the implicit conversion of 0
+//   to nullptr,
+// * comparing interface pointers to null.

 struct VlNull final {
    operator bool() const { return false; }
+    bool operator==(const void* ptr) const { return !ptr; }
 };
+inline bool operator==(const void* ptr, VlNull) { return !ptr; }

 //===================================================================
 // Verilog class reference container
@@ -305,8 +305,22 @@ void VerilatedVcd::printIndent(int level_change) {
 }

 void VerilatedVcd::pushPrefix(const std::string& name, VerilatedTracePrefixType type) {
-    std::string newPrefix = m_prefixStack.back().first + name;
+    assert(!m_prefixStack.empty());  // Constructor makes an empty entry
+    std::string pname = name;
+    // An empty name means this is the root of a model created with name()=="".  The
+    // tools get upset if we try to pass this as empty, so we put the signals under a
+    // new scope, but the signals further down will be peers, not children (as usual
+    // for name()!="")
+    // Terminate earlier $root?
+    if (m_prefixStack.back().second == VerilatedTracePrefixType::ROOTIO_MODULE) popPrefix();
+    if (pname.empty()) {  // Start new temporary root
+        pname = "$rootio";  // VCD names are not backslash escaped
+        m_prefixStack.emplace_back("", VerilatedTracePrefixType::ROOTIO_WRAPPER);
+        type = VerilatedTracePrefixType::ROOTIO_MODULE;
+    }
+    std::string newPrefix = m_prefixStack.back().first + pname;
    switch (type) {
+    case VerilatedTracePrefixType::ROOTIO_MODULE:
    case VerilatedTracePrefixType::SCOPE_MODULE:
    case VerilatedTracePrefixType::SCOPE_INTERFACE:
    case VerilatedTracePrefixType::STRUCT_PACKED:
@@ -326,7 +340,9 @@ void VerilatedVcd::pushPrefix(const std::string& name, VerilatedTracePrefixType
 }

 void VerilatedVcd::popPrefix() {
+    assert(!m_prefixStack.empty());
    switch (m_prefixStack.back().second) {
+    case VerilatedTracePrefixType::ROOTIO_MODULE:
    case VerilatedTracePrefixType::SCOPE_MODULE:
    case VerilatedTracePrefixType::SCOPE_INTERFACE:
    case VerilatedTracePrefixType::STRUCT_PACKED:
@@ -338,7 +354,7 @@ void VerilatedVcd::popPrefix() {
    default: break;
    }
    m_prefixStack.pop_back();
-    assert(!m_prefixStack.empty());
+    assert(!m_prefixStack.empty());  // Always one left, the constructor's initial one
 }

 void VerilatedVcd::declare(uint32_t code, const char* name, const char* wirep, bool array,
@@ -573,16 +589,11 @@ void VerilatedVcdBuffer::finishLine(uint32_t code, char* writep) {
 // so always inline them.

 VL_ATTR_ALWINLINE
-void VerilatedVcdBuffer::emitEvent(uint32_t code, const VlEventBase* newval) {
-    const bool triggered = newval->isTriggered();
-    // TODO : It seems that untriggered events are not filtered
-    // should be tested before this last step
-    if (triggered) {
-        // Don't prefetch suffix as it's a bit too late;
-        char* wp = m_writep;
-        *wp++ = '1';
-        finishLine(code, wp);
-    }
+void VerilatedVcdBuffer::emitEvent(uint32_t code) {
+    // Don't prefetch suffix as it's a bit too late;
+    char* wp = m_writep;
+    *wp++ = '1';
+    finishLine(code, wp);
 }

 VL_ATTR_ALWINLINE
@@ -43,7 +43,7 @@ private:
    friend VerilatedVcdBuffer;  // Give the buffer access to the private bits

    //=========================================================================
-    // VCD specific internals
+    // VCD-specific internals

    VerilatedVcdFile* m_filep;  // File we're writing to
    bool m_fileNewed;  // m_filep needs destruction
@@ -214,7 +214,7 @@ class VerilatedVcdBuffer VL_NOT_FINAL {
    // Implementation of VerilatedTraceBuffer interface
    // Implementations of duck-typed methods for VerilatedTraceBuffer. These are
    // called from only one place (the full* methods), so always inline them.
-    VL_ATTR_ALWINLINE void emitEvent(uint32_t code, const VlEventBase* newval);
+    VL_ATTR_ALWINLINE void emitEvent(uint32_t code);
    VL_ATTR_ALWINLINE void emitBit(uint32_t code, CData newval);
    VL_ATTR_ALWINLINE void emitCData(uint32_t code, CData newval, int bits);
    VL_ATTR_ALWINLINE void emitSData(uint32_t code, SData newval, int bits);
@@ -31,11 +31,14 @@
 #include "verilated.h"
 #include "verilated_imp.h"

+#include "vltstd/vpi_user.h"
+
+#include <cstdarg>
+#include <cstdio>
 #include <list>
 #include <map>
 #include <set>
 #include <string>
-#include <unordered_set>
 #include <utility>
 #include <vector>

@@ -630,6 +633,146 @@ public:
    void invalidate() { m_id = 0; }
 };

+class VerilatedVpiPutHolder final {
+    VerilatedVpioVar m_var;
+    s_vpi_value m_value;
+    union Storage {
+        char init = 0;  // to ensure trivial constructor
+        std::string str;
+        std::vector<s_vpi_vecval> vec;
+        ~Storage() noexcept {/* handled by VerilatedVpiPutHolder */};
+    } m_storage{};
+
+public:
+    VerilatedVpiPutHolder(const VerilatedVpioVar* vop, p_vpi_value valuep)
+        : m_var{vop} {
+        m_value.format = valuep->format;
+        switch (valuep->format) {
+        case vpiBinStrVal:  // FALLTHRU
+        case vpiOctStrVal:  // FALLTHRU
+        case vpiDecStrVal:  // FALLTHRU
+        case vpiHexStrVal:  // FALLTHRU
+        case vpiStringVal: {
+            new (&m_storage.str) std::string{valuep->value.str};
+            m_value.value.str = const_cast<char*>(m_storage.str.c_str());
+            break;
+        }
+        case vpiScalarVal: {
+            m_value.value.scalar = valuep->value.scalar;
+            break;
+        }
+        case vpiIntVal: {
+            m_value.value.integer = valuep->value.integer;
+            break;
+        }
+        case vpiRealVal: {
+            m_value.value.real = valuep->value.real;
+            break;
+        }
+        case vpiVectorVal: {
+            size_t words = 0;
+            switch (vop->varp()->vltype()) {
+            case VLVT_UINT8:
+            case VLVT_UINT16:
+            case VLVT_UINT32: {
+                words = 1;
+                break;
+            }
+            case VLVT_UINT64: {
+                words = 2;
+                break;
+            }
+            case VLVT_WDATA: {
+                words = VL_WORDS_I(vop->varp()->packed().elements());
+                break;
+            }
+            default: break;
+            }
+            new (&m_storage.vec)
+                std::vector<s_vpi_vecval>{valuep->value.vector, &valuep->value.vector[words]};
+            m_value.value.vector = m_storage.vec.data();
+            break;
+        }
+        }
+    }
+
+    VerilatedVpiPutHolder(VerilatedVpiPutHolder const& o)
+        : m_var{o.m_var}
+        , m_value{o.m_value} {
+        switch (m_value.format) {
+        case vpiBinStrVal:  // FALLTHRU
+        case vpiOctStrVal:  // FALLTHRU
+        case vpiDecStrVal:  // FALLTHRU
+        case vpiHexStrVal:  // FALLTHRU
+        case vpiStringVal: {
+            new (&m_storage.str) std::string{o.m_storage.str};
+            break;
+        }
+        case vpiVectorVal: {
+            new (&m_storage.vec) std::vector<s_vpi_vecval>{o.m_storage.vec};
+            break;
+        }
+        }
+    }
+
+    VerilatedVpiPutHolder(VerilatedVpiPutHolder&& o) noexcept
+        : m_var{std::move(o.m_var)}
+        , m_value{std::move(o.m_value)} {
+        switch (m_value.format) {
+        case vpiBinStrVal:  // FALLTHRU
+        case vpiOctStrVal:  // FALLTHRU
+        case vpiDecStrVal:  // FALLTHRU
+        case vpiHexStrVal:  // FALLTHRU
+        case vpiStringVal: {
+            new (&m_storage.str) std::string{std::move(o.m_storage.str)};
+            break;
+        }
+        case vpiVectorVal: {
+            new (&m_storage.vec) std::vector<s_vpi_vecval>{std::move(o.m_storage.vec)};
+            break;
+        }
+        }
+    }
+
+    ~VerilatedVpiPutHolder() noexcept {
+        switch (m_value.format) {
+        case vpiBinStrVal:  // FALLTHRU
+        case vpiOctStrVal:  // FALLTHRU
+        case vpiDecStrVal:  // FALLTHRU
+        case vpiHexStrVal:  // FALLTHRU
+        case vpiStringVal: m_storage.str.~basic_string(); break;
+        case vpiVectorVal: m_storage.vec.~vector(); break;
+        }
+    }
+
+    VerilatedVpioVar* varp() { return &m_var; }
+    p_vpi_value valuep() { return &m_value; }
+
+    static bool canInertialDelay(p_vpi_value valuep) {
+        switch (valuep->format) {
+        case vpiBinStrVal:  // FALLTHRU
+        case vpiOctStrVal:  // FALLTHRU
+        case vpiDecStrVal:  // FALLTHRU
+        case vpiHexStrVal:  // FALLTHRU
+        case vpiStringVal: {
+            if (VL_UNLIKELY(!valuep->value.str)) return false;
+            break;
+        }
+        case vpiScalarVal:  // FALLTHRU
+        case vpiIntVal:  // FALLTHRU
+        case vpiRealVal: break;
+        case vpiVectorVal: {
+            if (VL_UNLIKELY(!valuep->value.vector)) return false;
+            break;
+        }
+        default: {
+            return false;
+        }
+        }
+        return true;
+    }
+};
+
 struct VerilatedVpiTimedCbsCmp final {
    // Ordering sets keyed by time, then callback unique id
    bool operator()(const std::pair<QData, uint64_t>& a,
@@ -652,9 +795,11 @@ class VerilatedVpiImp final {
    std::array<VpioCbList, CB_ENUM_MAX_VALUE> m_cbCurrentLists;
    VpioFutureCbs m_futureCbs;  // Time based callbacks for future timestamps
    VpioFutureCbs m_nextCbs;  // cbNextSimTime callbacks
+    std::list<VerilatedVpiPutHolder> m_inertialPuts;  // Pending vpi puts due to vpiInertialDelay
    VerilatedVpiError* m_errorInfop = nullptr;  // Container for vpi error info
    VerilatedAssertOneThread m_assertOne;  // Assert only called from single thread
    uint64_t m_nextCallbackId = 1;  // Id to identify callback
+    bool m_evalNeeded = false;  // Model has had signals updated via vpi_put_value()

    static VerilatedVpiImp& s() {  // Singleton
        static VerilatedVpiImp s_s;
@@ -751,6 +896,9 @@ public:
        if (VL_LIKELY(it != s().m_futureCbs.cend())) return it->first.first;
        return ~0ULL;  // maxquad
    }
+    static bool hasCbs(const uint32_t reason) VL_MT_UNSAFE_ONE {
+        return !s().m_cbCurrentLists[reason].empty();
+    }
    static bool callCbs(const uint32_t reason) VL_MT_UNSAFE_ONE {
        VL_DEBUG_IF_PLI(VL_DBG_MSGF("- vpi: callCbs reason=%u\n", reason););
        assertOneCheck();
@@ -776,7 +924,7 @@ public:
        assertOneCheck();
        VpioCbList& cbObjList = s().m_cbCurrentLists[cbValueChange];
        bool called = false;
-        std::unordered_set<VerilatedVpioVar*> update;  // set of objects to update after callbacks
+        std::set<VerilatedVpioVar*> update;  // set of objects to update after callbacks
        if (cbObjList.empty()) return called;
        const auto last = std::prev(cbObjList.end());  // prevent looping over newly added elements
        for (auto it = cbObjList.begin(); true;) {
@@ -813,6 +961,17 @@ public:
    }
    static void dumpCbs() VL_MT_UNSAFE_ONE;
    static VerilatedVpiError* error_info() VL_MT_UNSAFE_ONE;  // getter for vpi error info
+    static bool evalNeeded() { return s().m_evalNeeded; }
+    static void evalNeeded(bool evalNeeded) { s().m_evalNeeded = evalNeeded; }
+    static void inertialDelay(const VerilatedVpioVar* vop, p_vpi_value valuep) {
+        s().m_inertialPuts.emplace_back(vop, valuep);
+    }
+    static void doInertialPuts() {
+        for (auto& it : s().m_inertialPuts) {
+            vpi_put_value(it.varp()->castVpiHandle(), it.valuep(), nullptr, vpiNoDelay);
+        }
+        s().m_inertialPuts.clear();
+    }
 };

 //======================================================================
@@ -900,6 +1059,10 @@ bool VerilatedVpi::callCbs(uint32_t reason) VL_MT_UNSAFE_ONE {
    return VerilatedVpiImp::callCbs(reason);
 }

+bool VerilatedVpi::hasCbs(uint32_t reason) VL_MT_UNSAFE_ONE {
+    return VerilatedVpiImp::hasCbs(reason);
+}
+
 // Historical, before we had multiple kinds of timed callbacks
 void VerilatedVpi::callTimedCbs() VL_MT_UNSAFE_ONE { VerilatedVpiImp::callCbs(cbAfterDelay); }

@@ -915,6 +1078,11 @@ PLI_INT32 VerilatedVpioReasonCb::dovpi_remove_cb() {
    return 1;
 }

+void VerilatedVpi::clearEvalNeeded() VL_MT_UNSAFE_ONE { VerilatedVpiImp::evalNeeded(false); }
+bool VerilatedVpi::evalNeeded() VL_MT_UNSAFE_ONE { return VerilatedVpiImp::evalNeeded(); }
+
+void VerilatedVpi::doInertialPuts() VL_MT_UNSAFE_ONE { VerilatedVpiImp::doInertialPuts(); }
+
 //======================================================================
 // VerilatedVpiImp implementation

@@ -2194,13 +2362,28 @@ bool vl_check_format(const VerilatedVar* varp, const p_vpi_value valuep, const c
    return status;
 }

+static void vl_strprintf(std::string& buffer, char const* fmt, ...) {
+    va_list args, args_copy;
+    va_start(args, fmt);
+    buffer.clear();
+    // Make copy of args since we may need to call VL_VSNPRINTF more than once
+    va_copy(args_copy, args);
+    // Try VL_VSNPRINTF in existing buffer
+    const int result
+        = VL_VSNPRINTF(const_cast<char*>(buffer.data()), buffer.capacity(), fmt, args_copy);
+    va_end(args_copy);
+    const int required = result + 1;  // Returned size doesn't include NUL terminator
+    // If there wasn't enough space, reallocate and try again
+    if (buffer.capacity() < required) {
+        buffer.reserve(required * 2);
+        VL_VSNPRINTF(const_cast<char*>(buffer.data()), buffer.capacity(), fmt, args);
+    }
+    va_end(args);
+}
+
 void vl_get_value(const VerilatedVar* varp, void* varDatap, p_vpi_value valuep,
                  const char* fullname) {
    if (!vl_check_format(varp, valuep, fullname, true)) return;
-    // Maximum required size is for binary string, one byte per bit plus null termination
-    static thread_local char t_outStr[VL_VALUE_STRING_MAX_WORDS * VL_EDATASIZE + 1];
-    // cppcheck-suppress variableScope
-    static const thread_local int t_outStrSz = sizeof(t_outStr) - 1;
    // string data type is dynamic and may vary in size during simulation
    static thread_local std::string t_outDynamicStr;
    // We used to presume vpiValue.format = vpiIntVal or if single bit vpiScalarVal
@@ -2244,43 +2427,21 @@ void vl_get_value(const VerilatedVar* varp, void* varDatap, p_vpi_value valuep,
            return;
        }
    } else if (valuep->format == vpiBinStrVal) {
-        valuep->value.str = t_outStr;
        int bits = varp->packed().elements();
+        t_outDynamicStr.resize(bits);
        const CData* datap = (reinterpret_cast<CData*>(varDatap));
-        int i;
-        if (bits > t_outStrSz) {
-            // limit maximum size of output to size of buffer to prevent overrun.
-            VL_VPI_WARNING_(
-                __FILE__, __LINE__,
-                "%s: Truncating string value of %s for %s"
-                " as buffer size (%d, VL_VALUE_STRING_MAX_WORDS=%d) is less than required (%d)",
-                __func__, VerilatedVpiError::strFromVpiVal(valuep->format), fullname, t_outStrSz,
-                VL_VALUE_STRING_MAX_WORDS, bits);
-            bits = t_outStrSz;
-        }
-        for (i = 0; i < bits; ++i) {
+        for (size_t i = 0; i < bits; ++i) {
            const char val = (datap[i >> 3] >> (i & 7)) & 1;
-            t_outStr[bits - i - 1] = val ? '1' : '0';
+            t_outDynamicStr[bits - i - 1] = val ? '1' : '0';
        }
-        t_outStr[i] = '\0';
+        valuep->value.str = const_cast<PLI_BYTE8*>(t_outDynamicStr.c_str());
        return;
    } else if (valuep->format == vpiOctStrVal) {
-        valuep->value.str = t_outStr;
        int chars = (varp->packed().elements() + 2) / 3;
+        t_outDynamicStr.resize(chars);
        const int bytes = VL_BYTES_I(varp->packed().elements());
        const CData* datap = (reinterpret_cast<CData*>(varDatap));
-        int i;
-        if (chars > t_outStrSz) {
-            // limit maximum size of output to size of buffer to prevent overrun.
-            VL_VPI_WARNING_(
-                __FILE__, __LINE__,
-                "%s: Truncating string value of %s for %s"
-                " as buffer size (%d, VL_VALUE_STRING_MAX_WORDS=%d) is less than required (%d)",
-                __func__, VerilatedVpiError::strFromVpiVal(valuep->format), fullname, t_outStrSz,
-                VL_VALUE_STRING_MAX_WORDS, chars);
-            chars = t_outStrSz;
-        }
-        for (i = 0; i < chars; ++i) {
+        for (size_t i = 0; i < chars; ++i) {
            const div_t idx = div(i * 3, 8);
            int val = datap[idx.quot];
            if ((idx.quot + 1) < bytes) {
@@ -2299,46 +2460,31 @@ void vl_get_value(const VerilatedVar* varp, void* varDatap, p_vpi_value valuep,
                    val &= (1 << rem) - 1;
                }
            }
-            t_outStr[chars - i - 1] = '0' + (val & 7);
+            t_outDynamicStr[chars - i - 1] = '0' + (val & 7);
        }
-        t_outStr[i] = '\0';
+        valuep->value.str = const_cast<PLI_BYTE8*>(t_outDynamicStr.c_str());
        return;
    } else if (valuep->format == vpiDecStrVal) {
-        valuep->value.str = t_outStr;
-        // outStrSz does not include nullptr termination so add one
        if (varp->vltype() == VLVT_UINT8) {
-            VL_SNPRINTF(t_outStr, t_outStrSz + 1, "%hhu",
-                        static_cast<unsigned char>(*(reinterpret_cast<CData*>(varDatap))));
-            return;
+            vl_strprintf(t_outDynamicStr, "%hhu",
+                         static_cast<unsigned char>(*(reinterpret_cast<CData*>(varDatap))));
        } else if (varp->vltype() == VLVT_UINT16) {
-            VL_SNPRINTF(t_outStr, t_outStrSz + 1, "%hu",
-                        static_cast<unsigned short>(*(reinterpret_cast<SData*>(varDatap))));
-            return;
+            vl_strprintf(t_outDynamicStr, "%hu",
+                         static_cast<unsigned short>(*(reinterpret_cast<SData*>(varDatap))));
        } else if (varp->vltype() == VLVT_UINT32) {
-            VL_SNPRINTF(t_outStr, t_outStrSz + 1, "%u",
-                        static_cast<unsigned int>(*(reinterpret_cast<IData*>(varDatap))));
-            return;
+            vl_strprintf(t_outDynamicStr, "%u",
+                         static_cast<unsigned int>(*(reinterpret_cast<IData*>(varDatap))));
        } else if (varp->vltype() == VLVT_UINT64) {
-            VL_SNPRINTF(t_outStr, t_outStrSz + 1, "%llu",
-                        static_cast<unsigned long long>(*(reinterpret_cast<QData*>(varDatap))));
-            return;
+            vl_strprintf(t_outDynamicStr, "%llu",  // lintok-format-ll
+                         static_cast<unsigned long long>(*(reinterpret_cast<QData*>(varDatap))));
        }
+        valuep->value.str = const_cast<PLI_BYTE8*>(t_outDynamicStr.c_str());
+        return;
    } else if (valuep->format == vpiHexStrVal) {
-        valuep->value.str = t_outStr;
        int chars = (varp->packed().elements() + 3) >> 2;
+        t_outDynamicStr.resize(chars);
        const CData* datap = (reinterpret_cast<CData*>(varDatap));
-        int i;
-        if (chars > t_outStrSz) {
-            // limit maximum size of output to size of buffer to prevent overrun.
-            VL_VPI_WARNING_(
-                __FILE__, __LINE__,
-                "%s: Truncating string value of %s for %s"
-                " as buffer size (%d, VL_VALUE_STRING_MAX_WORDS=%d) is less than required (%d)",
-                __func__, VerilatedVpiError::strFromVpiVal(valuep->format), fullname, t_outStrSz,
-                VL_VALUE_STRING_MAX_WORDS, chars);
-            chars = t_outStrSz;
-        }
-        for (i = 0; i < chars; ++i) {
+        for (size_t i = 0; i < chars; ++i) {
            char val = (datap[i >> 1] >> ((i & 1) << 2)) & 15;
            if (i == (chars - 1)) {
                // most significant char, mask off nonexistent bits when vector
@@ -2349,9 +2495,9 @@ void vl_get_value(const VerilatedVar* varp, void* varDatap, p_vpi_value valuep,
                    val &= (1 << rem) - 1;
                }
            }
-            t_outStr[chars - i - 1] = "0123456789abcdef"[static_cast<int>(val)];
+            t_outDynamicStr[chars - i - 1] = "0123456789abcdef"[static_cast<int>(val)];
        }
-        t_outStr[i] = '\0';
+        valuep->value.str = const_cast<PLI_BYTE8*>(t_outDynamicStr.c_str());
        return;
    } else if (valuep->format == vpiStringVal) {
        if (varp->vltype() == VLVT_STRING) {
@@ -2364,26 +2510,15 @@ void vl_get_value(const VerilatedVar* varp, void* varDatap, p_vpi_value valuep,
                return;
            }
        } else {
-            valuep->value.str = t_outStr;
            int bytes = VL_BYTES_I(varp->packed().elements());
+            t_outDynamicStr.resize(bytes);
            const CData* datap = (reinterpret_cast<CData*>(varDatap));
-            int i;
-            if (bytes > t_outStrSz) {
-                // limit maximum size of output to size of buffer to prevent overrun.
-                VL_VPI_WARNING_(__FILE__, __LINE__,
-                                "%s: Truncating string value of %s for %s"
-                                " as buffer size (%d, VL_VALUE_STRING_MAX_WORDS=%d) is less than "
-                                "required (%d)",
-                                __func__, VerilatedVpiError::strFromVpiVal(valuep->format),
-                                fullname, t_outStrSz, VL_VALUE_STRING_MAX_WORDS, bytes);
-                bytes = t_outStrSz;
-            }
-            for (i = 0; i < bytes; ++i) {
+            for (size_t i = 0; i < bytes; ++i) {
                const char val = datap[bytes - i - 1];
                // other simulators replace [leading?] zero chars with spaces, replicate here.
-                t_outStr[i] = val ? val : ' ';
+                t_outDynamicStr[i] = val ? val : ' ';
            }
-            t_outStr[i] = '\0';
+            valuep->value.str = const_cast<PLI_BYTE8*>(t_outDynamicStr.c_str());
            return;
        }
    } else if (valuep->format == vpiIntVal) {
@@ -2432,7 +2567,7 @@ void vpi_get_value(vpiHandle object, p_vpi_value valuep) {
 }

 vpiHandle vpi_put_value(vpiHandle object, p_vpi_value valuep, p_vpi_time /*time_p*/,
-                        PLI_INT32 /*flags*/) {
+                        PLI_INT32 flags) {
    VL_DEBUG_IF_PLI(VL_DBG_MSGF("- vpi: vpi_put_value %p %p\n", object, valuep););
    VerilatedVpiImp::assertOneCheck();
    VL_VPI_ERROR_RESET_();
@@ -2440,11 +2575,24 @@ vpiHandle vpi_put_value(vpiHandle object, p_vpi_value valuep, p_vpi_time /*time_
        VL_VPI_WARNING_(__FILE__, __LINE__, "Ignoring vpi_put_value with nullptr value pointer");
        return nullptr;
    }
+    PLI_INT32 delay_mode = flags & 0xfff;
    if (const VerilatedVpioVar* const vop = VerilatedVpioVar::castp(object)) {
+        if (delay_mode == vpiInertialDelay) {
+            if (!VerilatedVpiPutHolder::canInertialDelay(valuep)) {
+                VL_VPI_WARNING_(
+                    __FILE__, __LINE__,
+                    "%s: Unsupported p_vpi_value as requested for '%s' with vpiInertialDelay",
+                    __func__, vop->fullname());
+                return nullptr;
+            }
+            VerilatedVpiImp::inertialDelay(vop, valuep);
+            return object;
+        }
        VL_DEBUG_IF_PLI(
            VL_DBG_MSGF("- vpi:   vpi_put_value name=%s fmt=%d vali=%d\n", vop->fullname(),
                        valuep->format, valuep->value.integer);
            VL_DBG_MSGF("- vpi:   varp=%p  putatp=%p\n", vop->varp()->datap(), vop->varDatap()););
+        VerilatedVpiImp::evalNeeded(true);

        if (VL_UNLIKELY(!vop->varp()->isPublicRW())) {
            VL_VPI_WARNING_(__FILE__, __LINE__,
@@ -2549,7 +2697,8 @@ vpiHandle vpi_put_value(vpiHandle object, p_vpi_value valuep, p_vpi_time /*time_
        } else if (valuep->format == vpiDecStrVal) {
            char remainder[16];
            unsigned long long val;
-            const int success = std::sscanf(valuep->value.str, "%30llu%15s", &val, remainder);
+            const int success = std::sscanf(valuep->value.str, "%30llu%15s",  // lintok-format-ll
+                                            &val, remainder);
            if (success < 1) {
                VL_VPI_ERROR_(__FILE__, __LINE__, "%s: Parsing failed for '%s' as value %s for %s",
                              __func__, valuep->value.str,
@@ -49,11 +49,21 @@ public:
    /// Call callbacks of arbitrary types.
    /// User wrapper code should call this from their main loops.
    static bool callCbs(uint32_t reason) VL_MT_UNSAFE_ONE;
+    /// Returns true if there are callbacks of the given reason registered.
+    /// User wrapper code should call this from their main loops.
+    static bool hasCbs(uint32_t reason) VL_MT_UNSAFE_ONE;
    /// Returns time of the next registered VPI callback, or
    /// ~(0ULL) if none are registered
    static QData cbNextDeadline() VL_MT_UNSAFE_ONE;
    /// Debug dump of callbacks
    static void dumpCbs() VL_MT_UNSAFE_ONE;
+    /// Checks VPI dirty state (i.e. whether vpi_put_value() has
+    /// been called since the last clearEvalNeeded())
+    static bool evalNeeded() VL_MT_UNSAFE_ONE;
+    /// Clears VPI dirty state (see evalNeeded())
+    static void clearEvalNeeded() VL_MT_UNSAFE_ONE;
+    /// Perform inertially delayed puts
+    static void doInertialPuts() VL_MT_UNSAFE_ONE;

    // Self test, for internal use only
    static void selfTest() VL_MT_UNSAFE_ONE;
@@ -200,6 +200,11 @@

 // Comment tag that Function is pure (and thus also VL_MT_SAFE)
 #define VL_PURE VL_CLANG_ATTR(annotate("PURE"))
+// Annotated function can be called only in MT_DISABLED context, i.e. either in a code unit
+// compiled with VL_MT_DISABLED_CODE_UNIT preprocessor definition, or in the main thread.
+#define VL_MT_DISABLED \
+    VL_CLANG_ATTR(annotate("MT_DISABLED")) \
+    VL_EXCLUDES(VlOs::MtScopeMutex::s_haveThreadScope)
 // Comment tag that function is threadsafe
 #define VL_MT_SAFE VL_CLANG_ATTR(annotate("MT_SAFE"))
 // Comment tag that function is threadsafe, only if
@@ -216,7 +221,7 @@
 // protected to make sure single-caller
 #define VL_MT_UNSAFE_ONE VL_CLANG_ATTR(annotate("MT_UNSAFE_ONE"))
 // Comment tag that function is entry point of parallelization
-#define VL_MT_START VL_CLANG_ATTR(annotate("MT_START"))
+#define VL_MT_START VL_CLANG_ATTR(annotate("MT_START")) VL_REQUIRES(VlOs::MtScopeMutex::s_haveThreadScope)

 #ifndef VL_NO_LEGACY
 # define VL_ULL(c) (c##ULL)  // Add appropriate suffix to 64-bit constant (deprecated)
@@ -463,8 +468,6 @@ using ssize_t = uint32_t;  ///< signed size_t; returned from read()
    #define VL_VALUE_STRING_MAX_WORDS 64  ///< Max size in words of String conversion operation
 #endif

-#define VL_VALUE_STRING_MAX_CHARS (VL_VALUE_STRING_MAX_WORDS * VL_EDATASIZE / VL_BYTESIZE)
-
 //=========================================================================
 // Base macros

@@ -588,9 +591,9 @@ static inline double VL_ROUND(double n) {
 #endif

 //=========================================================================
-// Macros controlling target specific optimizations
+// Macros controlling target-specific optimizations

-// Define VL_PORTABLE_ONLY to disable all target specific optimizations
+// Define VL_PORTABLE_ONLY to disable all target-specific optimizations
 #ifndef VL_PORTABLE_ONLY
 # ifdef __x86_64__
 #  define VL_X86_64 1
@@ -614,8 +617,13 @@ static inline double VL_ROUND(double n) {
 //=========================================================================
 // Time and performance

+#include <string>
+
 namespace VlOs {

+/// Get environment variable
+extern std::string getenvStr(const std::string& envvar,
+                             const std::string& defaultValue) VL_MT_SAFE;
 extern uint64_t memUsageBytes() VL_MT_SAFE;  ///< Return memory usage in bytes, or 0 if unknown

 // Internal: Record CPU time, starting point on construction, and current delta from that
@@ -648,6 +656,14 @@ public:
        return (m_start == 0.0) ? 0.0 : gettime() - m_start;
    }
 };
+
+// Used by clang's -fthread-safety, ensures that only one instance of V3ThreadScope
+// is created at a time
+class VL_CAPABILITY("mutex") MtScopeMutex final {
+public:
+    static MtScopeMutex s_haveThreadScope;
+};
+
 }  //namespace VlOs

 //=========================================================================
@@ -51,10 +51,10 @@ double DeltaCpuTime::gettime() VL_MT_SAFE {
 #else
    // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
    timespec ts;
-    if (0 != clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts))  // MT-Safe
+    if (0 == clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &ts))  // MT-Safe  // LCOV_EXCL_BR_LINE
        return ts.tv_sec + ts.tv_nsec * 1e-9;
 #endif
-    return 0.0;
+    return 0.0;  // LCOV_EXCL_LINE
 }
 double DeltaWallTime::gettime() VL_MT_SAFE {
 #if defined(_WIN32) || defined(__MINGW32__)
@@ -66,9 +66,9 @@ double DeltaWallTime::gettime() VL_MT_SAFE {
 #else
    // NOLINTNEXTLINE(cppcoreguidelines-pro-type-member-init)
    timespec ts;
-    if (0 == clock_gettime(CLOCK_MONOTONIC, &ts))  // MT-Safe
+    if (0 == clock_gettime(CLOCK_MONOTONIC, &ts))  // MT-Safe  // LCOV_EXCL_BR_LINE
        return ts.tv_sec + ts.tv_nsec * 1e-9;
-    return 0.0;
+    return 0.0;  // LCOV_EXCL_LINE
 #endif
 }

@@ -99,5 +99,31 @@ uint64_t memUsageBytes() VL_MT_SAFE {
 #endif
 }

+//=========================================================================
+// VlOs::getenvStr implementation
+
+std::string getenvStr(const std::string& envvar, const std::string& defaultValue) VL_MT_SAFE {
+    std::string ret;
+#if defined(_MSC_VER)
+    // Note: MinGW does not offer _dupenv_s
+    const char* envvalue = nullptr;
+    _dupenv_s((char**)&envvalue, nullptr, envvar.c_str());
+    if (envvalue != nullptr) {
+        const std::string result{envvalue};
+        free((void*)envvalue);
+        ret = result;
+    } else {
+        ret = defaultValue;
+    }
+#else
+    if (const char* const envvalue = getenv(envvar.c_str())) {
+        ret = envvalue;
+    } else {
+        ret = defaultValue;
+    }
+#endif
+    return ret;
+}
+
 //=========================================================================
 }  //namespace VlOs
@@ -48,9 +48,11 @@ def fully_qualified_name(node):
    if node.kind == CursorKind.TRANSLATION_UNIT:
        return []
    res = fully_qualified_name(node.semantic_parent)
+    displayname = node.displayname
+    displayname = [displayname] if displayname else []
    if res:
-        return res + ([node.displayname] if node.displayname else [])
-    return [node.displayname] if node.displayname else []
+        return res + displayname
+    return displayname


 # Returns True, if `class_node` contains node
@@ -106,9 +108,8 @@ class VlAnnotations:

    def is_mt_safe_call(self):
        return (not self.is_mt_unsafe_call()
-                and (self.mt_safe or self.mt_safe_postinit or self.pure
-                     or self.requires or self.excludes or self.acquire
-                     or self.release))
+                and (self.mt_safe or self.mt_safe_postinit or self.pure or self.requires
+                     or self.excludes or self.acquire or self.release))

    def is_pure_call(self):
        return self.pure
@@ -140,31 +141,32 @@ class VlAnnotations:
        result = VlAnnotations()
        for node in nodes:
            if node.kind == CursorKind.ANNOTATE_ATTR:
-                if node.displayname == "MT_START":
+                displayname = node.displayname
+                if displayname == "MT_START":
                    result.mt_start = True
-                elif node.displayname == "MT_SAFE":
+                elif displayname == "MT_SAFE":
                    result.mt_safe = True
-                elif node.displayname == "MT_STABLE":
+                elif displayname == "MT_STABLE":
                    result.stable_tree = True
-                elif node.displayname == "MT_SAFE_POSTINIT":
+                elif displayname == "MT_SAFE_POSTINIT":
                    result.mt_safe_postinit = True
-                elif node.displayname == "MT_UNSAFE":
+                elif displayname == "MT_UNSAFE":
                    result.mt_unsafe = True
-                elif node.displayname == "MT_UNSAFE_ONE":
+                elif displayname == "MT_UNSAFE_ONE":
                    result.mt_unsafe_one = True
-                elif node.displayname == "MT_DISABLED":
+                elif displayname == "MT_DISABLED":
                    result.mt_disabled = True
-                elif node.displayname == "PURE":
+                elif displayname == "PURE":
                    result.pure = True
-                elif node.displayname in ["ACQUIRE", "ACQUIRE_SHARED"]:
+                elif displayname in ["ACQUIRE", "ACQUIRE_SHARED"]:
                    result.acquire = True
-                elif node.displayname in ["RELEASE", "RELEASE_SHARED"]:
+                elif displayname in ["RELEASE", "RELEASE_SHARED"]:
                    result.release = True
-                elif node.displayname == "REQUIRES":
+                elif displayname == "REQUIRES":
                    result.requires = True
-                elif node.displayname in ["EXCLUDES", "MT_SAFE_EXCLUDES"]:
+                elif displayname in ["EXCLUDES", "MT_SAFE_EXCLUDES"]:
                    result.excludes = True
-                elif node.displayname == "GUARDED_BY":
+                elif displayname == "GUARDED_BY":
                    result.guarded = True
            # Attributes are always at the beginning
            elif not node.kind.is_attribute():
@@ -203,9 +205,7 @@ class FunctionInfo:
    annotations: VlAnnotations
    ftype: FunctionType

-    _hash: Optional[int] = dataclasses.field(default=None,
-                                             init=False,
-                                             repr=False)
+    _hash: Optional[int] = dataclasses.field(default=None, init=False, repr=False)

    @property
    def name(self):
@@ -220,15 +220,13 @@ class FunctionInfo:
        return self._hash

    def __eq__(self, other):
-        return (self.usr == other.usr and self.file == other.file
-                and self.line == other.line)
+        return (self.usr == other.usr and self.file == other.file and self.line == other.line)

    def copy(self, /, **changes):
        return dataclasses.replace(self, **changes)

    @staticmethod
-    def from_decl_file_line_and_refd_node(file: str, line: int,
-                                          refd: clang.cindex.Cursor,
+    def from_decl_file_line_and_refd_node(file: str, line: int, refd: clang.cindex.Cursor,
                                          annotations: VlAnnotations):
        file = os.path.abspath(file)
        refd = refd.canonical
@@ -277,14 +275,11 @@ class Diagnostic:
    source_ctx: FunctionInfo
    kind: DiagnosticKind

-    _hash: Optional[int] = dataclasses.field(default=None,
-                                             init=False,
-                                             repr=False)
+    _hash: Optional[int] = dataclasses.field(default=None, init=False, repr=False)

    def __hash__(self):
        if not self._hash:
-            self._hash = hash(
-                hash(self.target) ^ hash(self.source_ctx) ^ hash(self.kind))
+            self._hash = hash(hash(self.target) ^ hash(self.source_ctx) ^ hash(self.kind))
        return self._hash


@@ -292,9 +287,9 @@ class CallAnnotationsValidator:

    def __init__(self, diagnostic_cb: Callable[[Diagnostic], None],
                 is_ignored_top_level: Callable[[clang.cindex.Cursor], bool],
-                 is_ignored_def: Callable[
-                     [clang.cindex.Cursor, clang.cindex.Cursor], bool],
-                 is_ignored_call: Callable[[clang.cindex.Cursor], bool]):
+                 is_ignored_def: Callable[[clang.cindex.Cursor, clang.cindex.Cursor],
+                                          bool], is_ignored_call: Callable[[clang.cindex.Cursor],
+                                                                           bool]):
        self._diagnostic_cb = diagnostic_cb
        self._is_ignored_top_level = is_ignored_top_level
        self._is_ignored_call = is_ignored_call
@@ -312,6 +307,7 @@ class CallAnnotationsValidator:
        self._defines: dict[str, str] = {}
        self._call_location: Optional[FunctionInfo] = None
        self._caller: Optional[FunctionInfo] = None
+        self._base_func_declarations: dict[str, clang.cindex.Cursor] = {}
        self._constructor_context: list[clang.cindex.Cursor] = []
        self._level: int = 0

@@ -329,8 +325,7 @@ class CallAnnotationsValidator:
        with open(source_file, "r", encoding="utf-8") as file:
            for line in file:
                line = line.strip()
-                match = re.fullmatch(
-                    r"^#\s*(define\s+(\w+)(?:\s+(.*))?|include\s+.*)$", line)
+                match = re.fullmatch(r"^#\s*(define\s+(\w+)(?:\s+(.*))?|include\s+.*)$", line)
                if match:
                    if match.group(1).startswith("define"):
                        key = match.group(2)
@@ -341,16 +336,14 @@ class CallAnnotationsValidator:
        return defs

    @staticmethod
-    def filter_out_unsupported_compiler_args(
-            args: list[str]) -> tuple[list[str], dict[str, str]]:
+    def filter_out_unsupported_compiler_args(args: list[str]) -> tuple[list[str], dict[str, str]]:
        filtered_args = []
        defines = {}
        args_iter = iter(args)
        try:
            while arg := next(args_iter):
                # Skip positional arguments (input file name).
-                if not arg.startswith("-") and (arg.endswith(".cpp")
-                                                or arg.endswith(".c")
+                if not arg.startswith("-") and (arg.endswith(".cpp") or arg.endswith(".c")
                                                or arg.endswith(".h")):
                    continue

@@ -367,8 +360,7 @@ class CallAnnotationsValidator:
                # Preserved options with separate value argument.
                if arg in [
                        "-x"
-                        "-Xclang", "-I", "-isystem", "-iquote", "-include",
-                        "-include-pch"
+                        "-Xclang", "-I", "-isystem", "-iquote", "-include", "-include-pch"
                ]:
                    filtered_args += [arg, next(args_iter)]
                    continue
@@ -406,14 +398,12 @@ class CallAnnotationsValidator:

        return (filtered_args, defines)

-    def compile_and_analyze_file(self, source_file: str,
-                                 compiler_args: list[str],
+    def compile_and_analyze_file(self, source_file: str, compiler_args: list[str],
                                 build_dir: Optional[str]):
        filename = os.path.abspath(source_file)
        initial_cwd = "."

-        filtered_args, defines = self.filter_out_unsupported_compiler_args(
-            compiler_args)
+        filtered_args, defines = self.filter_out_unsupported_compiler_args(compiler_args)
        defines.update(self.parse_initial_defines(source_file))

        if build_dir:
@@ -451,8 +441,7 @@ class CallAnnotationsValidator:
            self._diagnostic_cb(Diagnostic(target, source, source_ctx, kind))
        else:
            self._diagnostic_cb(
-                Diagnostic(FunctionInfo.from_node(target), source, source_ctx,
-                           kind))
+                Diagnostic(FunctionInfo.from_node(target), source, source_ctx, kind))

    def iterate_children(self, children: Iterable[clang.cindex.Cursor],
                         handler: Callable[[clang.cindex.Cursor], None]):
@@ -465,8 +454,7 @@ class CallAnnotationsValidator:
    @staticmethod
    def get_referenced_node_info(
        node: clang.cindex.Cursor
-    ) -> tuple[bool, Optional[clang.cindex.Cursor], VlAnnotations,
-               Iterable[clang.cindex.Cursor]]:
+    ) -> tuple[bool, Optional[clang.cindex.Cursor], VlAnnotations, Iterable[clang.cindex.Cursor]]:
        if not node.spelling and not node.displayname:
            return (False, None, VlAnnotations(), [])

@@ -480,8 +468,7 @@ class CallAnnotationsValidator:
        annotations = VlAnnotations.from_nodes_list(children)
        return (True, refd, annotations, children)

-    def check_mt_safe_call(self, node: clang.cindex.Cursor,
-                           refd: clang.cindex.Cursor,
+    def check_mt_safe_call(self, node: clang.cindex.Cursor, refd: clang.cindex.Cursor,
                           annotations: VlAnnotations):
        is_mt_safe = False

@@ -513,8 +500,7 @@ class CallAnnotationsValidator:
                # we are calling local method. It is MT safe
                # only if this method is also only calling local methods or
                # MT-safe methods
-                self.iterate_children(refd.get_children(),
-                                      self.dispatch_node_inside_definition)
+                self.iterate_children(refd.get_children(), self.dispatch_node_inside_definition)
                is_mt_safe = True
            # class/struct member
            elif refn and refn.kind == CursorKind.MEMBER_REF_EXPR and refn.referenced:
@@ -525,18 +511,15 @@ class CallAnnotationsValidator:
                if self.is_constructor_context() and refn.semantic_parent:
                    # we are in constructor, so calling local members is MT_SAFE,
                    # make sure object that we are calling is local to the constructor
-                    constructor_class = self._constructor_context[
-                        -1].semantic_parent
+                    constructor_class = self._constructor_context[-1].semantic_parent
                    if refn.semantic_parent.spelling == constructor_class.spelling:
                        if check_class_member_exists(constructor_class, refn):
                            is_mt_safe = True
                    else:
                        # check if this class inherits from some base class
-                        base_class = get_base_class(constructor_class,
-                                                    refn.semantic_parent)
+                        base_class = get_base_class(constructor_class, refn.semantic_parent)
                        if base_class:
-                            if check_class_member_exists(
-                                    base_class.get_declaration(), refn):
+                            if check_class_member_exists(base_class.get_declaration(), refn):
                                is_mt_safe = True
            # variable
            elif refn and refn.kind == CursorKind.DECL_REF_EXPR and refn.referenced:
@@ -567,8 +550,7 @@ class CallAnnotationsValidator:

    # Call handling

-    def process_method_call(self, node: clang.cindex.Cursor,
-                            refd: clang.cindex.Cursor,
+    def process_method_call(self, node: clang.cindex.Cursor, refd: clang.cindex.Cursor,
                            annotations: VlAnnotations):
        assert self._call_location
        ctx = self._call_location.annotations
@@ -576,58 +558,48 @@ class CallAnnotationsValidator:
        # MT-safe context
        if ctx.is_mt_safe_context():
            if not self.check_mt_safe_call(node, refd, annotations):
-                self.emit_diagnostic(
-                    FunctionInfo.from_node(refd, refd, annotations),
-                    DiagnosticKind.NON_MT_SAFE_CALL_IN_MT_SAFE_CTX)
+                self.emit_diagnostic(FunctionInfo.from_node(refd, refd, annotations),
+                                     DiagnosticKind.NON_MT_SAFE_CALL_IN_MT_SAFE_CTX)

        # stable tree context
        if ctx.is_stabe_tree_context():
            if annotations.is_mt_unsafe_call() or not (
-                    annotations.is_stabe_tree_call()
-                    or annotations.is_pure_call()
+                    annotations.is_stabe_tree_call() or annotations.is_pure_call()
                    or self.check_mt_safe_call(node, refd, annotations)):
-                self.emit_diagnostic(
-                    FunctionInfo.from_node(refd, refd, annotations),
-                    DiagnosticKind.NON_STABLE_TREE_CALL_IN_STABLE_TREE_CTX)
+                self.emit_diagnostic(FunctionInfo.from_node(refd, refd, annotations),
+                                     DiagnosticKind.NON_STABLE_TREE_CALL_IN_STABLE_TREE_CTX)

        # pure context
        if ctx.is_pure_context():
            if not annotations.is_pure_call():
-                self.emit_diagnostic(
-                    FunctionInfo.from_node(refd, refd, annotations),
-                    DiagnosticKind.NON_PURE_CALL_IN_PURE_CTX)
+                self.emit_diagnostic(FunctionInfo.from_node(refd, refd, annotations),
+                                     DiagnosticKind.NON_PURE_CALL_IN_PURE_CTX)

-    def process_function_call(self, refd: clang.cindex.Cursor,
-                              annotations: VlAnnotations):
+    def process_function_call(self, refd: clang.cindex.Cursor, annotations: VlAnnotations):
        assert self._call_location
        ctx = self._call_location.annotations

        # MT-safe context
        if ctx.is_mt_safe_context():
            if not annotations.is_mt_safe_call():
-                self.emit_diagnostic(
-                    FunctionInfo.from_node(refd, refd, annotations),
-                    DiagnosticKind.NON_MT_SAFE_CALL_IN_MT_SAFE_CTX)
+                self.emit_diagnostic(FunctionInfo.from_node(refd, refd, annotations),
+                                     DiagnosticKind.NON_MT_SAFE_CALL_IN_MT_SAFE_CTX)

        # stable tree context
        if ctx.is_stabe_tree_context():
-            if annotations.is_mt_unsafe_call() or not (
-                    annotations.is_pure_call()
-                    or annotations.is_mt_safe_call()
-                    or annotations.is_stabe_tree_call()):
-                self.emit_diagnostic(
-                    FunctionInfo.from_node(refd, refd, annotations),
-                    DiagnosticKind.NON_STABLE_TREE_CALL_IN_STABLE_TREE_CTX)
+            if annotations.is_mt_unsafe_call() or not (annotations.is_pure_call()
+                                                       or annotations.is_mt_safe_call()
+                                                       or annotations.is_stabe_tree_call()):
+                self.emit_diagnostic(FunctionInfo.from_node(refd, refd, annotations),
+                                     DiagnosticKind.NON_STABLE_TREE_CALL_IN_STABLE_TREE_CTX)

        # pure context
        if ctx.is_pure_context():
            if not annotations.is_pure_call():
-                self.emit_diagnostic(
-                    FunctionInfo.from_node(refd, refd, annotations),
-                    DiagnosticKind.NON_PURE_CALL_IN_PURE_CTX)
+                self.emit_diagnostic(FunctionInfo.from_node(refd, refd, annotations),
+                                     DiagnosticKind.NON_PURE_CALL_IN_PURE_CTX)

-    def process_constructor_call(self, refd: clang.cindex.Cursor,
-                                 annotations: VlAnnotations):
+    def process_constructor_call(self, refd: clang.cindex.Cursor, annotations: VlAnnotations):
        assert self._call_location
        ctx = self._call_location.annotations

@@ -635,31 +607,26 @@ class CallAnnotationsValidator:
        # only if they call local methods or MT-safe functions.
        if ctx.is_mt_safe_context() or self.is_constructor_context():
            self._constructor_context.append(refd)
-            self.iterate_children(refd.get_children(),
-                                  self.dispatch_node_inside_definition)
+            self.iterate_children(refd.get_children(), self.dispatch_node_inside_definition)
            self._constructor_context.pop()

        # stable tree context
        if ctx.is_stabe_tree_context():
            self._constructor_context.append(refd)
-            self.iterate_children(refd.get_children(),
-                                  self.dispatch_node_inside_definition)
+            self.iterate_children(refd.get_children(), self.dispatch_node_inside_definition)
            self._constructor_context.pop()

        # pure context
        if ctx.is_pure_context():
-            if not annotations.is_pure_call(
-            ) and not refd.is_default_constructor():
-                self.emit_diagnostic(
-                    FunctionInfo.from_node(refd, refd, annotations),
-                    DiagnosticKind.NON_PURE_CALL_IN_PURE_CTX)
+            if not annotations.is_pure_call() and not refd.is_default_constructor():
+                self.emit_diagnostic(FunctionInfo.from_node(refd, refd, annotations),
+                                     DiagnosticKind.NON_PURE_CALL_IN_PURE_CTX)

    def dispatch_call_node(self, node: clang.cindex.Cursor):
        [supported, refd, annotations, _] = self.get_referenced_node_info(node)

        if not supported:
-            self.iterate_children(node.get_children(),
-                                  self.dispatch_node_inside_definition)
+            self.iterate_children(node.get_children(), self.dispatch_node_inside_definition)
            return True

        assert refd is not None
@@ -676,19 +643,14 @@ class CallAnnotationsValidator:

        assert self._call_location is not None
        node_file = os.path.abspath(node.location.file.name)
-        self._call_location = self._call_location.copy(file=node_file,
-                                                       line=node.location.line)
+        self._call_location = self._call_location.copy(file=node_file, line=node.location.line)

        # Standalone functions and static class methods
        if (refd.kind == CursorKind.FUNCTION_DECL
-                or refd.kind == CursorKind.CXX_METHOD
-                and refd.is_static_method()):
+                or refd.kind == CursorKind.CXX_METHOD and refd.is_static_method()):
            self.process_function_call(refd, annotations)
        # Function pointer
-        elif refd.kind in [
-                CursorKind.VAR_DECL, CursorKind.FIELD_DECL,
-                CursorKind.PARM_DECL
-        ]:
+        elif refd.kind in [CursorKind.VAR_DECL, CursorKind.FIELD_DECL, CursorKind.PARM_DECL]:
            self.process_function_call(refd, annotations)
        # Non-static class methods
        elif refd.kind == CursorKind.CXX_METHOD:
@@ -726,18 +688,25 @@ class CallAnnotationsValidator:
            if self.dispatch_call_node(node) is False:
                return None
        elif node.is_definition() and node.kind in [
-                CursorKind.CXX_METHOD, CursorKind.FUNCTION_DECL,
-                CursorKind.CONSTRUCTOR, CursorKind.CONVERSION_FUNCTION
+                CursorKind.CXX_METHOD, CursorKind.FUNCTION_DECL, CursorKind.CONSTRUCTOR,
+                CursorKind.CONVERSION_FUNCTION
        ]:
            self.process_function_definition(node)
            return None

-        return self.iterate_children(node.get_children(),
-                                     self.dispatch_node_inside_definition)
+        return self.iterate_children(node.get_children(), self.dispatch_node_inside_definition)

    def process_function_definition(self, node: clang.cindex.Cursor):
        [supported, refd, annotations, _] = self.get_referenced_node_info(node)

+        # Fetch virtual annotations from base class.
+        # Set refd to virtual definition if present.
+        signature = node.displayname
+        if signature in self._base_func_declarations:
+            refd = self._base_func_declarations[signature]
+            virtual_annotations = VlAnnotations.from_nodes_list(refd.get_children())
+            annotations = annotations | virtual_annotations
+
        if refd and self._is_ignored_def(node, refd):
            return None

@@ -752,30 +721,29 @@ class CallAnnotationsValidator:
        # Implicitly mark definitions in VL_MT_DISABLED_CODE_UNIT .cpp files as
        # VL_MT_DISABLED. Existence of the annotation on declarations in .h
        # files is verified below.
-        # Also sets VL_REQUIRES, as this annotation is added together with
+        # Also sets VL_EXCLUDES, as this annotation is added together with
        # explicit VL_MT_DISABLED.
        if self.is_mt_disabled_code_unit():
            if node.location.file.name == self._main_source_file:
                annotations.mt_disabled = True
-                annotations.requires = True
+                annotations.excludes = True
            if refd.location.file.name == self._main_source_file:
                def_annotations.mt_disabled = True
-                def_annotations.requires = True
+                def_annotations.excludes = True

-        if not (def_annotations.is_empty() or def_annotations == annotations):
+        if def_annotations != annotations:
            # Use definition's annotations for the diagnostic
            # source (i.e. the definition)
            self._caller = FunctionInfo.from_node(node, refd, def_annotations)
            self._call_location = self._caller

-            self.emit_diagnostic(
-                FunctionInfo.from_node(refd, refd, annotations),
-                DiagnosticKind.ANNOTATIONS_DEF_DECL_MISMATCH)
+            self.emit_diagnostic(FunctionInfo.from_node(refd, refd, annotations),
+                                 DiagnosticKind.ANNOTATIONS_DEF_DECL_MISMATCH)

-        # Use concatenation of definition and declaration annotations
-        # for calls validation.
-        self._caller = FunctionInfo.from_node(node, refd,
-                                              def_annotations | annotations)
+        else:
+            # Use concatenation of definition and declaration annotations
+            # for calls validation.
+            self._caller = FunctionInfo.from_node(node, refd, def_annotations | annotations)
        prev_call_location = self._call_location
        self._call_location = self._caller

@@ -793,8 +761,7 @@ class CallAnnotationsValidator:
                if declarations:
                    del self._external_decls[usr]

-        self.iterate_children(node_children,
-                              self.dispatch_node_inside_definition)
+        self.iterate_children(node_children, self.dispatch_node_inside_definition)

        self._call_location = prev_call_location
        self._caller = prev_call_location
@@ -804,25 +771,37 @@ class CallAnnotationsValidator:
    # Nodes not located inside definition

    def dispatch_node(self, node: clang.cindex.Cursor):
-        if node.kind in [
-                CursorKind.CXX_METHOD, CursorKind.FUNCTION_DECL,
-                CursorKind.CONSTRUCTOR, CursorKind.CONVERSION_FUNCTION
+        kind = node.kind
+        if kind is CursorKind.CXX_BASE_SPECIFIER:
+            # Get referenced virtual declarations from base class.
+            for base in node.get_children():
+                if base.referenced:
+                    for declaration in base.referenced.get_children():
+                        self._base_func_declarations[declaration.displayname] = declaration
+        elif kind in [
+                CursorKind.CXX_METHOD, CursorKind.FUNCTION_DECL, CursorKind.CONSTRUCTOR,
+                CursorKind.CONVERSION_FUNCTION
        ]:
            if node.is_definition():
                return self.process_function_definition(node)
-            # else:
            return self.process_function_declaration(node)

-        return self.iterate_children(node.get_children(), self.dispatch_node)
+        result = self.iterate_children(node.get_children(), self.dispatch_node)

-    def process_translation_unit(
-            self, translation_unit: clang.cindex.TranslationUnit):
+        # Clean declarations if class declaration processing is finished.
+        if kind in [
+                CursorKind.CLASS_DECL, CursorKind.STRUCT_DECL, CursorKind.UNION_DECL,
+                CursorKind.ENUM_DECL, CursorKind.UNEXPOSED_DECL
+        ]:
+            self._base_func_declarations = {}
+        return result
+
+    def process_translation_unit(self, translation_unit: clang.cindex.TranslationUnit):
        self._level += 1
        kv_defines = sorted([f"{k}={v}" for k, v in self._defines.items()])
        concat_defines = '\n'.join(kv_defines)
        # List of headers already processed in a TU with specified set of defines.
-        tu_processed_headers = self._processed_headers.setdefault(
-            concat_defines, set())
+        tu_processed_headers = self._processed_headers.setdefault(concat_defines, set())
        for child in translation_unit.cursor.get_children():
            if self._is_ignored_top_level(child):
                continue
@@ -833,10 +812,8 @@ class CallAnnotationsValidator:
            self.dispatch_node(child)
        self._level -= 1

-        tu_processed_headers.update([
-            os.path.abspath(str(hdr.source))
-            for hdr in translation_unit.get_includes()
-        ])
+        tu_processed_headers.update(
+            [os.path.abspath(str(hdr.source)) for hdr in translation_unit.get_includes()])


@dataclass
@@ -857,8 +834,7 @@ def get_filter_funcs(verilator_root: str):
        filename = os.path.abspath(node.location.file.name)
        return not filename.startswith(verilator_root)

-    def is_ignored_def(node: clang.cindex.Cursor,
-                       refd: clang.cindex.Cursor) -> bool:
+    def is_ignored_def(node: clang.cindex.Cursor, refd: clang.cindex.Cursor) -> bool:
        # __*
        if str(refd.spelling).startswith("__"):
            return True
@@ -901,8 +877,7 @@ def precompile_header(compile_command: CompileCommand, tmp_dir: str) -> str:
        os.chdir(compile_command.directory)

        index = Index.create()
-        translation_unit = index.parse(compile_command.filename,
-                                       compile_command.args)
+        translation_unit = index.parse(compile_command.filename, compile_command.args)
        for diag in translation_unit.diagnostics:
            if diag.severity >= clang.cindex.Diagnostic.Error:
                errors.append(str(diag))
@@ -910,23 +885,20 @@ def precompile_header(compile_command: CompileCommand, tmp_dir: str) -> str:
        if len(errors) == 0:
            pch_file = os.path.join(
                tmp_dir,
-                f"{compile_command.refid:02}_{os.path.basename(compile_command.filename)}.pch"
-            )
+                f"{compile_command.refid:02}_{os.path.basename(compile_command.filename)}.pch")
            translation_unit.save(pch_file)

            if pch_file:
                return pch_file

-    except (TranslationUnitSaveError, TranslationUnitLoadError,
-            OSError) as exception:
+    except (TranslationUnitSaveError, TranslationUnitLoadError, OSError) as exception:
        print(f"%Warning: {exception}", file=sys.stderr)

    finally:
        os.chdir(initial_cwd)

-    print(
-        f"%Warning: Precompilation failed, skipping: {compile_command.filename}",
-        file=sys.stderr)
+    print(f"%Warning: Precompilation failed, skipping: {compile_command.filename}",
+          file=sys.stderr)
    for error in errors:
        print(f"          {error}", file=sys.stderr)
    return ""
@@ -934,10 +906,8 @@ def precompile_header(compile_command: CompileCommand, tmp_dir: str) -> str:

 # Compile and analyze inputs in a single process.
 def run_analysis(ccl: Iterable[CompileCommand], pccl: Iterable[CompileCommand],
-                 diagnostic_cb: Callable[[Diagnostic],
-                                         None], verilator_root: str):
-    (is_ignored_top_level, is_ignored_def,
-     is_ignored_call) = get_filter_funcs(verilator_root)
+                 diagnostic_cb: Callable[[Diagnostic], None], verilator_root: str):
+    (is_ignored_top_level, is_ignored_def, is_ignored_call) = get_filter_funcs(verilator_root)

    prefix = "verilator_clang_check_attributes_"
    with tempfile.TemporaryDirectory(prefix=prefix) as tmp_dir:
@@ -947,8 +917,8 @@ def run_analysis(ccl: Iterable[CompileCommand], pccl: Iterable[CompileCommand],
            if pch_file:
                extra_args += ["-include-pch", pch_file]

-        cav = CallAnnotationsValidator(diagnostic_cb, is_ignored_top_level,
-                                       is_ignored_def, is_ignored_call)
+        cav = CallAnnotationsValidator(diagnostic_cb, is_ignored_top_level, is_ignored_def,
+                                       is_ignored_call)
        for compile_command in ccl:
            cav.compile_and_analyze_file(compile_command.filename,
                                         extra_args + compile_command.args,
@@ -963,12 +933,11 @@ class ParallelAnalysisProcess:

    @staticmethod
    def init_data(verilator_root: str, tmp_dir: str):
-        (is_ignored_top_level, is_ignored_def,
-         is_ignored_call) = get_filter_funcs(verilator_root)
+        (is_ignored_top_level, is_ignored_def, is_ignored_call) = get_filter_funcs(verilator_root)

        ParallelAnalysisProcess.cav = CallAnnotationsValidator(
-            ParallelAnalysisProcess._diagnostic_handler, is_ignored_top_level,
-            is_ignored_def, is_ignored_call)
+            ParallelAnalysisProcess._diagnostic_handler, is_ignored_top_level, is_ignored_def,
+            is_ignored_call)
        ParallelAnalysisProcess.tmp_dir = tmp_dir

    @staticmethod
@@ -979,31 +948,27 @@ class ParallelAnalysisProcess:
    def analyze_cpp_file(compile_command: CompileCommand) -> set[Diagnostic]:
        ParallelAnalysisProcess.diags = set()
        assert ParallelAnalysisProcess.cav is not None
-        ParallelAnalysisProcess.cav.compile_and_analyze_file(
-            compile_command.filename, compile_command.args,
-            compile_command.directory)
+        ParallelAnalysisProcess.cav.compile_and_analyze_file(compile_command.filename,
+                                                             compile_command.args,
+                                                             compile_command.directory)
        return ParallelAnalysisProcess.diags

    @staticmethod
    def precompile_header(compile_command: CompileCommand) -> str:
-        return precompile_header(compile_command,
-                                 ParallelAnalysisProcess.tmp_dir)
+        return precompile_header(compile_command, ParallelAnalysisProcess.tmp_dir)


 # Compile and analyze inputs in multiple processes.
-def run_parallel_analysis(ccl: Iterable[CompileCommand],
-                          pccl: Iterable[CompileCommand],
-                          diagnostic_cb: Callable[[Diagnostic], None],
-                          jobs_count: int, verilator_root: str):
+def run_parallel_analysis(ccl: Iterable[CompileCommand], pccl: Iterable[CompileCommand],
+                          diagnostic_cb: Callable[[Diagnostic],
+                                                  None], jobs_count: int, verilator_root: str):
    prefix = "verilator_clang_check_attributes_"
    with tempfile.TemporaryDirectory(prefix=prefix) as tmp_dir:
-        with multiprocessing.Pool(
-                processes=jobs_count,
-                initializer=ParallelAnalysisProcess.init_data,
-                initargs=[verilator_root, tmp_dir]) as pool:
+        with multiprocessing.Pool(processes=jobs_count,
+                                  initializer=ParallelAnalysisProcess.init_data,
+                                  initargs=[verilator_root, tmp_dir]) as pool:
            extra_args = []
-            for pch_file in pool.imap_unordered(
-                    ParallelAnalysisProcess.precompile_header, pccl):
+            for pch_file in pool.imap_unordered(ParallelAnalysisProcess.precompile_header, pccl):
                if pch_file:
                    extra_args += ["-include-pch", pch_file]

@@ -1011,8 +976,7 @@ def run_parallel_analysis(ccl: Iterable[CompileCommand],
                for compile_command in ccl:
                    compile_command.args = compile_command.args + extra_args

-            for diags in pool.imap_unordered(
-                    ParallelAnalysisProcess.analyze_cpp_file, ccl, 1):
+            for diags in pool.imap_unordered(ParallelAnalysisProcess.analyze_cpp_file, ccl, 1):
                for diag in diags:
                    diagnostic_cb(diag)

@@ -1057,8 +1021,7 @@ class TopDownSummaryPrinter():
        row_groups: dict[str, list[list[str]]] = {}
        column_widths = [0, 0]
        for func in sorted(self._funcs.values(),
-                           key=lambda func:
-                           (func.info.file, func.info.line, func.info.usr)):
+                           key=lambda func: (func.info.file, func.info.line, func.info.usr)):
            func_info = func.info
            relfile = os.path.relpath(func_info.file, root_dir)

@@ -1082,31 +1045,23 @@ class TopDownSummaryPrinter():
            if func.mismatch:
                mrelfile = os.path.relpath(func.mismatch.file, root_dir)
                row_group.append([
-                    f"{mrelfile}:{func.mismatch.line}:",
-                    f"[{func.mismatch.annotations}]",
+                    f"{mrelfile}:{func.mismatch.line}:", f"[{func.mismatch.annotations}]",
                    func.mismatch.name + " [declaration]"
                ])

-            row_group.append([
-                f"{relfile}:{func_info.line}:", f"[{func_info.annotations}]",
-                func_info.name
-            ])
+            row_group.append(
+                [f"{relfile}:{func_info.line}:", f"[{func_info.annotations}]", func_info.name])

-            for callee in sorted(func.calees,
-                                 key=lambda func:
-                                 (func.file, func.line, func.usr)):
+            for callee in sorted(func.calees, key=lambda func: (func.file, func.line, func.usr)):
                crelfile = os.path.relpath(callee.file, root_dir)
-                row_group.append([
-                    f"{crelfile}:{callee.line}:", f"[{callee.annotations}]",
-                    "  " + callee.name
-                ])
+                row_group.append(
+                    [f"{crelfile}:{callee.line}:", f"[{callee.annotations}]", "  " + callee.name])

            row_groups[name] = row_group

            for row in row_group:
                for row_id, value in enumerate(row[0:-1]):
-                    column_widths[row_id] = max(column_widths[row_id],
-                                                len(value))
+                    column_widths[row_id] = max(column_widths[row_id], len(value))

        for label, rows in sorted(row_groups.items(), key=lambda kv: kv[0]):
            self.begin_group(label)
@@ -1114,21 +1069,17 @@ class TopDownSummaryPrinter():
                print(f"{row[0]:<{column_widths[0]}}  "
                      f"{row[1]:<{column_widths[1]}}    "
                      f"{row[2]}")
-        print(
-            f"Number of functions reported unsafe: {len(self._unsafe_in_safe)}"
-        )
+        print(f"Number of functions reported unsafe: {len(self._unsafe_in_safe)}")


 def main():
-    default_verilator_root = os.path.abspath(
-        os.path.join(os.path.dirname(__file__), ".."))
+    default_verilator_root = os.path.abspath(os.path.join(os.path.dirname(__file__), ".."))

    parser = argparse.ArgumentParser(
        allow_abbrev=False,
        formatter_class=argparse.RawDescriptionHelpFormatter,
        description="""Check function annotations for correctness""",
-        epilog=
-        """Copyright 2022-2024 by Wilson Snyder. Verilator is free software;
+        epilog="""Copyright 2022-2024 by Wilson Snyder. Verilator is free software;
    you can redistribute it and/or modify it under the terms of either the GNU
    Lesser General Public License Version 3 or the Apache License 2.0.
    SPDX-License-Identifier: LGPL-3.0-only OR Apache-2.0""")
@@ -1142,29 +1093,23 @@ def main():
                        type=int,
                        default=0,
                        help="Number of parallel jobs to use.")
-    parser.add_argument(
-        "--compile-commands-dir",
-        type=str,
-        default=None,
-        help="Path to directory containing compile_commands.json.")
+    parser.add_argument("--compile-commands-dir",
+                        type=str,
+                        default=None,
+                        help="Path to directory containing compile_commands.json.")
    parser.add_argument("--cxxflags",
                        type=str,
                        default=None,
                        help="Extra flags passed to clang++.")
-    parser.add_argument(
-        "--compilation-root",
-        type=str,
-        default=os.getcwd(),
-        help="Directory used as CWD when compiling source files.")
-    parser.add_argument(
-        "-c",
-        "--precompile",
-        action="append",
-        help="Header file to be precompiled and cached at the start.")
-    parser.add_argument("file",
+    parser.add_argument("--compilation-root",
                        type=str,
-                        nargs="+",
-                        help="Source file to analyze.")
+                        default=os.getcwd(),
+                        help="Directory used as CWD when compiling source files.")
+    parser.add_argument("-c",
+                        "--precompile",
+                        action="append",
+                        help="Header file to be precompiled and cached at the start.")
+    parser.add_argument("file", type=str, nargs="+", help="Source file to analyze.")

    cmdline = parser.parse_args()

@@ -1179,8 +1124,7 @@ def main():

    compdb: Optional[CompilationDatabase] = None
    if cmdline.compile_commands_dir:
-        compdb = CompilationDatabase.fromDirectory(
-            cmdline.compile_commands_dir)
+        compdb = CompilationDatabase.fromDirectory(cmdline.compile_commands_dir)

    if cmdline.cxxflags is not None:
        common_cxxflags = shlex.split(cmdline.cxxflags)
@@ -1230,8 +1174,7 @@ def main():
                     summary_printer.handle_diagnostic, verilator_root)
    else:
        run_parallel_analysis(compile_commands_list, precompile_commands_list,
-                              summary_printer.handle_diagnostic, cmdline.jobs,
-                              verilator_root)
+                              summary_printer.handle_diagnostic, cmdline.jobs, verilator_root)

    summary_printer.print_summary(verilator_root)

@@ -58,15 +58,12 @@ def test():
            if not Args.scenarios or re.match('dist', Args.scenarios):
                run("make examples VERILATOR_NO_OPT_BUILD=1")
            run("make test_regress VERILATOR_NO_OPT_BUILD=1" +
-                (" SCENARIOS='" + Args.scenarios +
-                 "'" if Args.scenarios else "") +
-                (" DRIVER_HASHSET='--hashset=" + Args.hashset +
-                 "'" if Args.hashset else "") +
+                (" SCENARIOS='" + Args.scenarios + "'" if Args.scenarios else "") +
+                (" DRIVER_HASHSET='--hashset=" + Args.hashset + "'" if Args.hashset else "") +
                ('' if Args.stop else ' || true'))
        else:
            for test in Args.tests:
-                if not os.path.exists(test) and os.path.exists(
-                        "test_regress/t/" + test):
+                if not os.path.exists(test) and os.path.exists("test_regress/t/" + test):
                    test = "test_regress/t/" + test
                run(test)
        ci_fold_end()
@@ -78,8 +75,7 @@ def test():
        os.makedirs(cc_dir, exist_ok=True)
        os.makedirs(cc_dir + "/info", exist_ok=True)

-        with subprocess.Popen("find . -print | grep .gcda",
-                              shell=True,
+        with subprocess.Popen("find . -print | grep .gcda", shell=True,
                              stdout=subprocess.PIPE) as sp:
            datout = sp.stdout.read()

@@ -98,8 +94,7 @@ def test():
                    del dats[dat]
                    break

-        with subprocess.Popen("find . -print | grep .gcno",
-                              shell=True,
+        with subprocess.Popen("find . -print | grep .gcno", shell=True,
                              stdout=subprocess.PIPE) as sp:
            datout = sp.stdout.read()

@@ -116,8 +111,7 @@ def test():
                if gbase in gcnos:
                    os.symlink(gcnos[gbase], gcno)
                else:
-                    print("MISSING .gcno for a .gcda: " + gcno,
-                          file=sys.stderr)
+                    print("MISSING .gcno for a .gcda: " + gcno, file=sys.stderr)
        ci_fold_end()

    if Args.stage_enabled[5]:
@@ -142,8 +136,7 @@ def test():
    if Args.stage_enabled[11]:
        ci_fold_start("dirs")
        print("Stage 11: Cleanup paths")
-        cleanup_abs_paths_info(cc_dir, cc_dir + "/app_total.info",
-                               cc_dir + "/app_total.info")
+        cleanup_abs_paths_info(cc_dir, cc_dir + "/app_total.info", cc_dir + "/app_total.info")
        ci_fold_end()

    if Args.stage_enabled[12]:
@@ -164,34 +157,38 @@ def test():
            inc = "--include " + inc
        if exc != '':
            exc = "--exclude " + exc
-        run("cd " + cc_dir + " ; " + RealPath +
-            "/fastcov.py -C app_total.info " + inc + " " + exc +
-            " -x --lcov -o app_total_f.info")
+        run("cd " + cc_dir + " ; " + RealPath + "/fastcov.py -C app_total.info " + inc + " " +
+            exc + " -x --lcov -o app_total_f.info")
        ci_fold_end()

    if Args.stage_enabled[17]:
        ci_fold_start("report")
        print("Stage 17: Create HTML")
        run("cd " + cc_dir + " ; genhtml app_total_f.info --demangle-cpp" +
-            " --rc lcov_branch_coverage=1 --rc genhtml_hi_limit=100 --output-directory html"
-            )
+            " --rc lcov_branch_coverage=1 --rc genhtml_hi_limit=100 --output-directory html")
        ci_fold_end()

    if Args.stage_enabled[18]:
        ci_fold_start("upload")
        print("Stage 18: Upload")
-        cmd = "bash <(cat ci/coverage-upload.sh) -f " + cc_dir + "/app_total.info"
-        print("print: Not running: export CODECOV_TOKEN=<hidden>")
-        print("print: Not running: " + cmd)
+        # curl -Os https://cli.codecov.io/latest/linux/codecov ; sudo chmod +x codecov
+        # --disable-search does not seem to work
+        # -d with false directory does not seem to work
+        # So, remove gcno files before calling codecov
+        upload_dir = "nodist/obj_dir/upload"
+        os.makedirs(upload_dir, exist_ok=True)
+        cmd = "ci/codecov -v upload-process -Z" + " -f " + cc_dir + "/app_total.info )"
+        print("print: Not running:")
+        print("  export CODECOV_TOKEN=<hidden>")
+        print("  find . -name '*.gcno' -exec rm {} \\;")
+        print("  " + cmd)
        ci_fold_end()

    if Args.stage_enabled[19]:
        print("*-* All Finished *-*")
        print("")
        print("* See report in " + cc_dir + "/html/index.html")
-        print(
-            "* Remember to make distclean && ./configure before working on non-coverage"
-        )
+        print("* Remember to make distclean && ./configure before working on non-coverage")


 def clone_sources(cc_dir):
@@ -200,9 +197,8 @@ def clone_sources(cc_dir):
    for globf in Source_Globs:
        for infile in glob.glob(globf):
            if re.match(r'^/', infile):
-                sys.exit(
-                    "%Error: source globs should be relative not absolute filenames, "
-                    + infile)
+                sys.exit("%Error: source globs should be relative not absolute filenames, " +
+                         infile)
            outfile = cc_dir + "/" + infile
            outpath = re.sub(r'/[^/]*$', '', outfile, count=1)
            os.makedirs(outpath, exist_ok=True)
@@ -243,22 +239,20 @@ def clone_sources(cc_dir):
                                done = True

                        ofh.write(line + "\n")
-    print("Number of source lines automatically LCOV_EXCL_LINE'ed: %d" %
-          excluded_lines)
-    print("Number of source lines automatically LCOV_EXCL_BR_LINE'ed: %d" %
-          excluded_br_lines)
+    print("Number of source lines automatically LCOV_EXCL_LINE'ed: %d" % excluded_lines)
+    print("Number of source lines automatically LCOV_EXCL_BR_LINE'ed: %d" % excluded_br_lines)


 def cleanup_abs_paths_info(cc_dir, infile, outfile):
    lines = []
    with open(infile, "r", encoding="utf8") as fh:
        for line in fh:
-            if re.search(r'^SF:', line):
-                line = re.sub(os.environ['VERILATOR_ROOT'] + '/',
-                              '',
-                              line,
-                              count=1)
+            if re.search(r'^SF:', line) and not re.search(r'^SF:/usr/', line):
+                line = re.sub(os.environ['VERILATOR_ROOT'] + '/', '', line, count=1)
                line = re.sub(cc_dir + '/', '', line, count=1)
+                line = re.sub(r'^SF:.*?/include/', 'SF:include/', line, count=1)
+                line = re.sub(r'^SF:.*?/src/', 'SF:src/', line, count=1)
+                line = re.sub(r'^SF:.*?/test_regress/', 'SF:test_regress/', line, count=1)
                line = re.sub(r'obj_dbg/verilog.y$', 'verilog.y', line)
                # print("Remaining SF: "+line)
            lines.append(line)
@@ -340,15 +334,13 @@ def ci_fold_end():
 parser = argparse.ArgumentParser(
    allow_abbrev=False,
    formatter_class=argparse.RawDescriptionHelpFormatter,
-    description=
-    """code_coverage builds Verilator with C++ coverage support and runs
+    description="""code_coverage builds Verilator with C++ coverage support and runs
 tests with coverage enabled. This will rebuild the current object
 files. Run as:

  cd $VERILATOR_ROOT
  nodist/code_coverage""",
-    epilog=
-    """Copyright 2019-2024 by Wilson Snyder. This program is free software; you
+    epilog="""Copyright 2019-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -358,22 +350,20 @@ SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0""")
 parser.add_argument('--debug', action='store_true', help='enable debug')
 parser.add_argument('--hashset',
                    action='store',
-                    help='pass test hashset onto driver.pl test harness')
+                    help='pass test hashset onto driver.py test harness')
 parser.add_argument('--scenarios',
                    action='store',
-                    help='pass test scenarios onto driver.pl test harness')
-parser.add_argument(
-    '--stages',
-    action='store',
-    help='runs a specific stage or range of stages (see the script)')
+                    help='pass test scenarios onto driver.py test harness')
+parser.add_argument('--stages',
+                    '--stage',
+                    action='store',
+                    help='runs a specific stage or range of stages (see the script)')
 parser.add_argument(
+    '--tests',
    '--test',
    action='append',
-    dest='tests',
    default=[],
-    help=
-    'Instead of normal regressions, run the specified test(s), may be used multiple times'
-)
+    help='Instead of normal regressions, run the specified test(s), may be used multiple times')
 parser.add_argument('--no-stop',
                    dest='stop',
                    action='store_false',
@@ -20,8 +20,7 @@ def dotread(filename):
        vnum = 0

        vertex_re = re.compile(r'^\t([a-zA-Z0-9_]+)\t(.*)$')
-        edge_re = re.compile(
-            r'^\t([a-zA-Z0-9_]+)\s+->\s+([a-zA-Z0-9_]+)\s*(.*)$')
+        edge_re = re.compile(r'^\t([a-zA-Z0-9_]+)\s+->\s+([a-zA-Z0-9_]+)\s*(.*)$')

        for line in fh:
            vertex_match = re.search(vertex_re, line)
@@ -29,11 +28,7 @@ def dotread(filename):
            if vertex_match:
                if vertex_match.group(1) != 'nTITLE':
                    header = False
-                    Vertexes.append({
-                        'num': vnum,
-                        'line': line,
-                        'name': vertex_match.group(1)
-                    })
+                    Vertexes.append({'num': vnum, 'line': line, 'name': vertex_match.group(1)})
                    vnum += 1
            elif edge_match:
                fromv = edge_match.group(1)
@@ -65,14 +60,13 @@ def cwrite(filename):
        fh.write("void V3GraphTestImport::dotImport() {\n")
        fh.write("    auto* gp = &m_graph;\n")
        for ver in sorted(Vertexes, key=lambda ver: ver['num']):
-            fh.write(
-                "    auto* %s = new V3GraphTestVertex{gp, \"%s\"};  if (%s) {}\n"
-                % (ver['name'], ver['name'], ver['name']))
+            fh.write("    auto* %s = new V3GraphTestVertex{gp, \"%s\"};  if (%s) {}\n" %
+                     (ver['name'], ver['name'], ver['name']))
        fh.write("\n")
        for edge in Edges:
-            fh.write("    new V3GraphEdge{gp, %s, %s, %s, %s};\n" %
-                     (edge['from'], edge['to'], edge['weight'],
-                      "true" if edge['cutable'] else "false"))
+            fh.write(
+                "    new V3GraphEdge{gp, %s, %s, %s, %s};\n" %
+                (edge['from'], edge['to'], edge['weight'], "true" if edge['cutable'] else "false"))
        fh.write("}\n")


@@ -82,22 +76,17 @@ def cwrite(filename):
 parser = argparse.ArgumentParser(
    allow_abbrev=False,
    formatter_class=argparse.RawDescriptionHelpFormatter,
-    description=
-    """dot_importer takes a graphvis .dot file and converts into .cpp file.
+    description="""dot_importer takes a graphvis .dot file and converts into .cpp file.
 This x.cpp file is then manually included in V3GraphTest.cpp to verify
 various xsub-algorithms.""",
-    epilog=
-    """Copyright 2005-2024 by Wilson Snyder. This program is free software; you
+    epilog="""Copyright 2005-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.

 SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0""")

-parser.add_argument('--debug',
-                    action='store_const',
-                    const=9,
-                    help='enable debug')
+parser.add_argument('--debug', action='store_const', const=9, help='enable debug')
 parser.add_argument('filename', help='input .dot filename to process')

 Args = parser.parse_args()
--- a/Show More
+++ b/Show More