Merge from master for release. (update)

Update CMakeList version
Merge from master for release.
2026-10-03 16:44:11 +02:00 · 2023-09-16 21:44:39 -04:00 · 2023-09-16 20:54:54 -04:00 · 2023-09-16 20:43:53 -04:00 · 2023-09-16 18:36:39 -04:00 · 2023-09-16 17:52:02 -04:00
763 changed files with 93019 additions and 6645 deletions
@@ -36,12 +36,14 @@ jobs:
        m32: [0, 1]
        exclude:
          # Build pull requests only with ubuntu-22.04 and without m32
-          - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-18.04' || 'do-not-exclude' }}
-          - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-20.04' || 'do-not-exclude' }}
-          - m32: ${{ github.event_name == 'pull_request' && 1              || 'do-not-exclude' }}
-          # Build -m32 only on ubuntu-22.04
+          # - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-18.04' || 'do-not-exclude' }}
+          # - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-20.04' || 'do-not-exclude' }}
+          # - m32: ${{ github.event_name == 'pull_request' && 1              || 'do-not-exclude' }}
+          # Build -m32 only on ubuntu-22.04 clang++
          - {os: ubuntu-18.04, m32: 1}
          - {os: ubuntu-20.04, m32: 1}
+          - compiler: { cc: gcc, cxx: g++ }
+            m32: 1
        include:
          # Build GCC 10 on ubuntu-20.04
          - os: ubuntu-20.04
@@ -105,12 +107,14 @@ jobs:
        suite: [dist-vlt-0, dist-vlt-1, dist-vlt-2, vltmt-0, vltmt-1]
        exclude:
          # Build pull requests only with ubuntu-22.04 and without m32
-          - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-18.04' || 'do-not-exclude' }}
-          - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-20.04' || 'do-not-exclude' }}
-          - m32: ${{ github.event_name == 'pull_request' && 1              || 'do-not-exclude' }}
+          # - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-18.04' || 'do-not-exclude' }}
+          # - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-20.04' || 'do-not-exclude' }}
+          # - m32: ${{ github.event_name == 'pull_request' && 1              || 'do-not-exclude' }}
          # Build -m32 only on ubuntu-22.04
          - {os: ubuntu-18.04, m32: 1}
          - {os: ubuntu-20.04, m32: 1}
+          - compiler: { cc: gcc, cxx: g++ }
+            m32: 1
        include:
          # Test with GCC 10 on ubuntu-20.04 without m32
          - {os: ubuntu-20.04, compiler: { cc: gcc-10, cxx: g++-10 }, m32: 0, suite: dist-vlt-0}
@@ -0,0 +1,85 @@
+# Build and push verilator docker image when tags are pushed to the repository.
+# The following variable(s) must be configured in the github repository:
+# DOCKER_HUB_NAMESPACE: docker hub namespace.
+# The following secrets must be configured in the github repository:
+# DOCKER_HUB_USER: user name for logging into docker hub
+# DOCKER_HUB_ACCESS_TOKEN: docker hub access token.
+name: Build Verilator Container
+
+on:
+  push:
+    tags: [ 'v*' ]
+  workflow_dispatch:
+    inputs:
+      manual_tag:
+        description: 'Git tag to use for image build'
+        required: true
+        type: string
+      add_latest_tag:
+        description: 'Tag workflow_dispatch docker image as "latest"'
+        required: true
+        type: boolean
+        default: false
+
+jobs:
+
+  build:
+
+    runs-on: ubuntu-22.04
+
+    strategy:
+      matrix:
+        contexts:
+          - "ci/docker/run:verilator"
+          # - "ci/docker/buildenv:verilator-buildenv"
+
+    steps:
+    - name: Checkout
+      uses: actions/checkout@v3
+
+    - name: Extract context variables
+      run: |
+        echo "${{ 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: 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 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: 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 }}
@@ -41,3 +41,5 @@ verilator-config-version.cmake
 **/_build/*
 **/obj_dir/*
 /.vscode/
+/.idea/
+/cmake-build-*/
@@ -15,7 +15,7 @@
 cmake_minimum_required(VERSION 3.15)
 cmake_policy(SET CMP0091 NEW) # Use MSVC_RUNTIME_LIBRARY to select the runtime
 project(Verilator
-    VERSION 5.010
+    VERSION 5.016
    HOMEPAGE_URL https://verilator.org
    LANGUAGES CXX
 )
@@ -8,6 +8,170 @@ The changes in each Verilator version are described below.  The
 contributors that suggested a given feature are shown in []. Thanks!


+Verilator 5.016 2023-09-16
+==========================
+
+**Minor:**
+
+* Add prepareClone and atClone APIs for Verilated models (#3503) (#4444). [Yinan Xu]
+* Add check for conflicting options e.g. binary and lint-only (#4409). [Ethan Sifferman]
+* Add --no-trace-top to not trace top signals (#4412) (#4422). [Frans Skarman]
+* Support recursive function calls (#3267).
+* Support assignments of packed values to stream expressions on queues (#4401). [Ryszard Rozak, Antmicro Ltd]
+* Support no-parentheses calls to static methods (#4432). [Krzysztof Boroński]
+* Support block_item_declaration in forks (#4455). [Krzysztof Boroński]
+* Support assignments of stream expressions on queues to packed values (#4458). [Ryszard Rozak, Antmicro Ltd]
+* Support function non-constant default arguments (#4470).
+* Support 'let'.
+* Optimize Verilator executable size by refactoring error reporting routines (#4446). [Anthony Donlon]
+* Optimize Verilation runtime pointers and graphs (#4396) (#4397) (#4398). [Krzysztof Bieganski, Antmicro Ltd]
+* Optimize preparations towards multithreaded Verilation (#4291) (#4463) (#4476) (#4477) (#4479).  [Kamil Rakoczy, Antmicro Ltd]
+* Fix Windows filename format, etc (#3873) (#4421). [Anthony Donlon].
+* Fix t_dist_cppstyle Perl performance issue (#4085). [Srinivasan Venkataramanan]
+* Fix using type in parameterized classes without #() (#4281) (#4440). [Anthony Donlon]
+* Fix false INFINITELOOP on forever..mailbox.get() (#4323). [Srinivasan Venkataramanan]
+* Fix data type of condition operation on class objects (#4345) (#4352). [Ryszard Rozak, Antmicro Ltd]
+* Fix variables mutated under fork..join_none/join_any blocks into anonymous objects (#4356). [Krzysztof Boroński]
+* Fix V3CUse, do not consider implementations (.cpp) at all (#4386). [Krzysztof Boroński]
+* Fix ++/-- under statements (#4399). [Aleksander Kiryk, Antmicro Ltd]
+* Fix detection of mixed blocking and nonblocking assignment in nested assignments (#4404). [Ryszard Rozak, Antmicro Ltd]
+* Fix jumping over object initialization (#4411). [Krzysztof Boroński]
+* Fix multiple issues towards short circuit support (#4413) (#4460). [Ryszard Rozak, Antmicro Ltd]
+* Fix variable lifetimes in extern methods (#4414). [Krzysztof Boroński]
+* Fix multiple function definitions in V3Sched (#4416). [Hennadii Chernyshchyk]
+* Fix false UNUSEDPARAM on generate localparam (#4427). [Bill Pringlemeir]
+* Fix checking for parameter and port connections in the wrong place (#4428). [Anthony Donlon]
+* Fix coroutine handle movement during queue manipulation (#4431). [Aleksander Kiryk, Antmicro Ltd]
+* Fix nested assignments on the LHS (#4435). [Ryszard Rozak, Antmicro Ltd]
+* Fix false MULTITOP on bound interfaces (#4438). [Alex Solomatnikov]
+* Fix internal error on real conversion (#4447). [vdhotre-ventana]
+* Fix lifetime unknown error on enum.name (#4448). [jwoutersymatra]
+* Fix unstable output of VHashSha256 (#4453). [Anthony Donlon]
+* Fix static cast from a stream type (#4469) (#4485). [Ryszard Rozak, Antmicro Ltd]
+* Fix error on enum with VARHIDDEN of cell (#4482). [Michail Rontionov]
+* Fix lint of case statements with enum and wildcard bits (#4464) (#4487). [Anthony Donlon]
+* Fix reference to extended class in parameterized class (#4466).
+* Fix recursive display causing segfault (#4480). [Kuoping Hsu]
+* Fix the error message when the type of ref argument is wrong (#4490). [Ryszard Rozak, Antmicro Ltd]
+* Fix display %x formatting of real.
+* Fix mis-warning on #() in classes' own functions.
+* Fix IGNOREDRETURN to not warn on void-cast static function calls.
+* Fix ZERODLY to not warn on 'wait(0)'.
+
+
+Verilator 5.014 2023-08-06
+==========================
+
+**Minor:**
+
+* Deprecation planned for 32-bit pointer -m32 mode (#4268).
+* Deprecate CMake config below version 3.13 (#4389) (#4390). [Vito Gamberini]
+* Support some stream operations on queues (#4292). [Ryszard Rozak, Antmicro Ltd]
+* Support property declaration with empty parentheses (#4313) (#4317). [Anthony Donlon]
+* Support locator methods with "with" on assoc arrays (#4335). [Ryszard Rozak, Antmicro Ltd]
+* Support string replication with variable (#4341). [Aleksander Kiryk, Antmicro Ltd]
+* Support more types in wait (#4374). [Aleksander Kiryk, Antmicro Ltd]
+* Support static method calls as default values of function arguments (#4378). [Ryszard Rozak, Antmicro Ltd]
+* Add GENUNNAMED lint warning. [Srinivasan Venkataramanan, Deepa Palaniappan]
+* Add MISINDENT lint warning for misleading indentation.
+* Fix 'VlForkSync' redeclaration (#4277). [Krzysztof Bieganski, Antmicro Ltd]
+* Fix processes that can outlive their parents (#4253). [Krzysztof Boronski, Antmicro Ltd]
+* Fix duplicate fork names (#4295). [Ryszard Rozak, Antmicro Ltd]
+* Fix splitting coroutines (#4297) (#4307). [Jiamin Zhu]
+* Fix error when multiple duplicate DPI exports (#4301).
+* Fix class reference assignment checking (#4296). [Ryszard Rozak, Antmicro Ltd]
+* Fix handling of ref types in initial values of type parameters (#4304). [Ryszard Rozak, Antmicro Ltd]
+* Fix comparison of string parameters (#4308). [Ryszard Rozak, Antmicro Ltd]
+* Fix state update for always processes (#4311). [Aleksander Kiryk, Antmicro Ltd]
+* Fix multiple edge timing controls in class methods (#4318) (#4320) (#4344). [Krzysztof Bieganski, Antmicro Ltd]
+* Fix implicit calls of base class constructors with optional arguments (#4319). [Ryszard Rozak, Antmicro Ltd]
+* Fix propagation of process requirement (#4321). [Krzysztof Boroński]
+* Fix unhandled overloads in V3InstrCount (#4324). [Krzysztof Boroński]
+* Fix selects of static members (#4326). [Ryszard Rozak, Antmicro Ltd]
+* Fix references to members of results of static methods (#4327). [Ryszard Rozak, Antmicro Ltd]
+* Fix unique..with method on queues of class objects (#4328). [Ryszard Rozak, Antmicro Ltd]
+* Fix queue slicing (#4329). [Aleksander Kiryk, Antmicro Ltd]
+* Fix wildcard referring types (#4336) (#4342). [Aleksander Kiryk, Antmicro Ltd]
+* Fix comparison of class objects (#4346). [Ryszard Rozak, Antmicro Ltd]
+* Fix unexpected RefDType on assoc arrays (#4337). [Aleksander Kiryk, Antmicro Ltd]
+* Fix cmake astgen for Rocky Linux 8.7 (#4343). [Julian Daube]
+* Fix class timescale in class packages (#4348). [Krzysztof Bieganski, Antmicro Ltd]
+* Fix string concatenations (#4354). [Ryszard Rozak, Antmicro Ltd]
+* Fix unlinked task error from broken context (#4355) (#4402). [Aleksander Kiryk, Antmicro Ltd]
+* Fix selects on unpacked structs (#4359). [Ryszard Rozak, Antmicro Ltd]
+* Fix select operation on assoc array with wide keys (#4360). [Ryszard Rozak, Antmicro Ltd]
+* Fix non-public methods with wide output (#4364). [Ryszard Rozak, Antmicro Ltd]
+* Fix handling of super.new calls (#4366). [Ryszard Rozak, Antmicro Ltd]
+* Fix assign to input var in methods (#4367). [Aleksander Kiryk, Antmicro Ltd]
+* Fix VlProcess not found (#4368). [Aleksander Kiryk, Antmicro Ltd]
+* Fix order of evaluation of function calls in statements (#4375). [Ryszard Rozak, Antmicro Ltd]
+* Fix config_build.h issues (#4380) (#4381). [Andrew Miloradovsky]
+
+
+Verilator 5.012 2023-06-13
+==========================
+
+**Major:**
+
+* With -j or --build-jobs, multithread Verilator's emit phase of Verilation. [Kamil Rakoczy, Antmicro Ltd]
+  Additional Verilator-internal stages will become multithreaded over time.
+
+**Minor:**
+
+* Add --main-top-name option for C main TOP name (#4235) (#4249). [Don Williamson]
+* Add creating __inputs.vpp file with --debug (#4177). [Tudor Timi]
+* Add NEWERSTD warning when using feature in newer language standard (#4168) (#4172). [Ethan Sifferman]
+* Add warning that timing controls in DPI exports are unsupported (#4238). [Krzysztof Bieganski, Antmicro Ltd]
+* Support std::process class (#4212). [Aleksander Kiryk, Antmicro Ltd]
+* Support inside expressions with strings and doubles (#4138) (#4139). [Krzysztof Boroński]
+* Support get_randstate/set_randstate class method functions.
+* Support for condition operator on class objects (#4214). [Ryszard Rozak, Antmicro Ltd]
+* Support array max (#4275). [Aleksander Kiryk, Antmicro Ltd]
+* Optimize VPI callValueCbs (#4155). [Hennadii Chernyshchyk]
+* Configure for faster C++ linking using 'mold', if it is installed.
+* Fix crash on duplicate imported modules (#3231). [Robert Balas]
+* Fix false WIDTHEXPAND on array declarations (#3959). [JOTEGO]
+* Fix marking overridden methods as coroutines (#4120) (#4169). [Krzysztof Bieganski, Antmicro Ltd]
+* Fix SystemC signal copy macro use (#4135). [Josep Sans]
+* Fix duplicate static names in blocks in functions (#4144) (#4160). [Stefan Wallentowitz]
+* Fix initialization order of initial static after function/task (#4159). [Kamil Rakoczy, Antmicro Ltd]
+* Fix linking AstRefDType if it has parameterized class ref (#4164) (#4170). [Ryszard Rozak, Antmicro Ltd]
+* Fix crash caused by $display() optimization (#4165) (#4166). [Tudor Timi]
+* Fix arrays of unpacked structs (#4173). [Risto Pejašinović]
+* Fix $fscanf of decimals overflowing variables (#4174). [Ahmed El-Mahmoudy]
+* Fix super.new missing data type (#4147). [Tudor Timi]
+* Fix missing class forward declarations (#4151). [Krzysztof Boroński]
+* Fix hashes of instances of parameterized classes (#4182). [Ryszard Rozak, Antmicro Ltd]
+* Fix forced assignments that override non-continuous assignments (#4183) (#4192). [Krzysztof Bieganski, Antmicro Ltd]
+* 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 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]
+* Fix deep traversal of class inheritance timing (#4216). [Krzysztof Boroński]
+* Fix class parameters of enum types (#4219). [Ryszard Rozak, Antmicro Ltd]
+* Fix static methods with prototypes (#4220). [Ryszard Rozak, Antmicro Ltd]
+* Fix LATCH warning on function local variables (#4221) (#4284) [Julien Margetts]
+* Fix VCD scope types (#4227) (#4282). [Àlex Torregrosa]
+* Fix incorrect multi-driven lint warning (#4231) (#4248). [Adrien Le Masle]
+* Fix missing assignment for wide unpacked structs (#4233). [Jiamin Zhu]
+* Fix unpacked struct == and != operators (#4234) (#4240). [Risto Pejašinović]
+* Fix AstStructSel clean when data type is structure (#4241) (#4244). [Risto Pejašinović]
+* Fix function calls in with statements (#4245). [Ryszard Rozak, Antmicro Ltd]
+* Fix operator == for unpacked struct, if elements are VlUnpacked arrays (#4247). [Risto Pejašinović]
+* Fix STATIC lifetime for variables created from clocking items (#4262). [Krzysztof Boroński]
+* Fix names of foreach blocks (#4264). [Ryszard Rozak, Antmicro Ltd]
+* Fix iterated variables in foreach loops to have VAUTOM lifetimes (#4265). [Krzysztof Boroński]
+* Fix missing assignment for wide class members (#4267). [Jiamin Zhu]
+* Fix the global uses timing flag when forks exist (#4274). [Krzysztof Bieganski, Antmicro Ltd]
+* Fix struct redefinition (#4276). [Aleksander Kiryk, Antmicro Ltd]
+* Fix detection of wire/reg duplicates.
+* Fix false IMPLICITSTATIC on package functions.
+* Fix method calls on function return values.
+
+
 Verilator 5.010 2023-04-30
 ==========================

@@ -424,17 +424,15 @@ YAPF_FLAGS = -i
 yapf:
 	$(YAPF) $(YAPF_FLAGS) $(PY_FILES)

-FLAKE8 = flake8
-FLAKE8_FLAGS = \
-	--extend-exclude=fastcov.py \
-	--ignore=E123,E129,E251,E402,E501,W503,W504,E701
-
 PYLINT = pylint
 PYLINT_FLAGS = --score=n --disable=R0801

+RUFF = ruff
+RUFF_FLAGS = check --ignore=E402,E501,E701
+
 lint-py:
-	-$(FLAKE8) $(FLAKE8_FLAGS) $(PY_PROGRAMS)
 	-$(PYLINT) $(PYLINT_FLAGS) $(PY_PROGRAMS)
+	-$(RUFF) $(RUFF_FLAGS) $(PY_PROGRAMS)

 format-pl-exec:
 	-chmod a+x test_regress/t/*.pl
@@ -451,8 +449,8 @@ install-msg:

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

-# autoheader might not change config_build.h.in, so touch it
-${srcdir}/config_build.h: ${srcdir}/config_build.h.in configure
+# autoheader might not change config_package.h.in, so touch it
+${srcdir}/config_package.h: ${srcdir}/config_package.h.in configure
 	cd ${srcdir} && autoheader
 	touch $@
 Makefile: Makefile.in config.status $(IN_WILD)
@@ -1,12 +1,16 @@
 .. 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
+    :target: https://verilator.org
 .. image:: https://img.shields.io/badge/License-LGPL%20v3-blue.svg
-    :target: https://www.gnu.org/licenses/lgpl-3.0]
+    :target: https://www.gnu.org/licenses/lgpl-3.0
 .. 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
    :target: https://repology.org/project/verilator/versions
+.. 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
    :target: https://www.codacy.com/gh/verilator/verilator
 .. image:: https://codecov.io/gh/verilator/verilator/branch/master/graph/badge.svg
@@ -57,17 +61,16 @@ files, the "Verilated" code.

 These Verilated C++/SystemC files are then compiled by a C++ compiler
 (gcc/clang/MSVC++), optionally along with a user's own C++/SystemC wrapper
-file to instantiate the Verilated model. Executing the resulting executable
-performs the design simulation. Verilator also supports linking Verilated
-generated libraries, optionally encrypted, into other simulators.
+file, to instantiate the Verilated model. Executing the resulting
+executable performs the design simulation. Verilator also supports linking
+Verilated generated libraries, optionally encrypted, into other simulators.

 Verilator may not be the best choice if you are expecting a full-featured
 replacement for a closed-source Verilog simulator, needs SDF annotation,
 mixed-signal simulation, or are doing a quick class project (we recommend
 `Icarus Verilog`_ for classwork.)  However, if you are looking for a path
 to migrate SystemVerilog to C++/SystemC, or want high-speed simulation of
-synthesizable designs containing limited verification constructs, Verilator
-is the tool for you.
+designs, Verilator is the tool for you.


 Performance
@@ -83,11 +86,11 @@ 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 the
-closed-source Verilog simulators (Carbon Design Systems Carbonator,
+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
-cycles/dollar.
+simulation cycles/dollar.


 Installation & Documentation
@@ -372,6 +372,7 @@ detailed descriptions of these arguments.
    --make <build-tool>         Generate scripts for specified build tool
     -MAKEFLAGS <flags>         Arguments to pass to make during --build
    --main                      Generate C++ main() file
+    --main-top-name             Specify top name passed to Verilated model in generated C++ main
    --max-num-width <value>     Maximum number width (default: 64K)
    --Mdir <directory>          Name of output object directory
    --MMD                       Create .d dependency files
@@ -439,6 +440,7 @@ detailed descriptions of these arguments.
    --trace-params              Enable tracing of parameters
    --trace-structs             Enable tracing structure names
    --trace-threads <threads>   Enable FST waveform creation on separate threads
+    --no-trace-top              Do not emit traces for signals in the top module generated by verilator
    --trace-underscore          Enable tracing of _signals
     -U<var>                    Undefine preprocessor define
    --no-unlimited-stack        Don't disable stack size limit
@@ -61,7 +61,7 @@ if [ "$CI_BUILD_STAGE_NAME" = "build" ]; then
    sudo apt-get update
    sudo apt-get install ccache help2man libfl-dev ||
    sudo apt-get install ccache help2man libfl-dev
-    if [ "$CI_RUNS_ON" != "ubuntu-22.04" ]; then
+    if [ "$CI_RUNS_ON" = "ubuntu-20.04" ]; then
      # Some conflict of libunwind verison on 22.04, can live without it for now
      sudo apt-get install libgoogle-perftools-dev ||
      sudo apt-get install libgoogle-perftools-dev
@@ -70,6 +70,10 @@ if [ "$CI_BUILD_STAGE_NAME" = "build" ]; then
      sudo apt-get install libsystemc libsystemc-dev ||
      sudo apt-get install libsystemc libsystemc-dev
    fi
+    if [ "$CI_RUNS_ON" = "ubuntu-22.04" ]; then
+      sudo apt-get install bear mold ||
+      sudo apt-get install bear mold
+    fi
    if [ "$COVERAGE" = 1 ]; then
      yes yes | sudo cpan -fi Parallel::Forker
    fi
@@ -102,8 +106,8 @@ elif [ "$CI_BUILD_STAGE_NAME" = "test" ]; then
    sudo apt-get install gdb gtkwave lcov libfl-dev ccache
    # Required for test_regress/t/t_dist_attributes.pl
    if [ "$CI_RUNS_ON" = "ubuntu-22.04" ]; then
-      sudo apt-get install libclang-dev ||
-      sudo apt-get install libclang-dev
+      sudo apt-get install libclang-dev mold ||
+      sudo apt-get install libclang-dev mold
      pip3 install clang==14.0
    fi
    if [ "$CI_RUNS_ON" = "ubuntu-20.04" ] || [ "$CI_RUNS_ON" = "ubuntu-22.04" ]; then
@@ -19,10 +19,16 @@ RUN apt-get update \
                        ccache \
                        flex \
                        git \
+                        help2man \
+                        libfl2 \
                        libfl-dev \
                        libgoogle-perftools-dev \
+                        numactl \
                        perl \
+                        perl-doc \
                        python3 \
+                        zlib1g \
+                        zlib1g-dev \
    && apt-get clean \
    && rm -rf /var/lib/apt/lists/*

@@ -39,7 +45,7 @@ RUN git clone "${REPO}" verilator && \
    git checkout "${SOURCE_COMMIT}" && \
    autoconf && \
    ./configure && \
-    make -j "$(nproc)" && \
+    make && \
    make install && \
    cd .. && \
    rm -r verilator
@@ -5,16 +5,16 @@
 # 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,
+# 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],[5.010 2023-04-30],
+AC_INIT([Verilator],[5.016 2023-09-16],
        [https://verilator.org],
        [verilator],[https://verilator.org])

-AC_CONFIG_HEADERS(src/config_build.h)
+AC_CONFIG_HEADERS(src/config_package.h)
 AC_CONFIG_FILES(Makefile src/Makefile src/Makefile_obj include/verilated.mk include/verilated_config.h verilator.pc verilator-config.cmake verilator-config-version.cmake)

 # Version
@@ -470,6 +470,10 @@ m4_foreach([ldflag], [
 AC_SUBST(CFG_LDLIBS_THREADS)
 AC_SUBST(CFG_LDFLAGS_THREADS_CMAKE)

+# If 'mold' is installed, use it to link for faster buildtimes
+_MY_LDLIBS_CHECK_OPT(CFG_LDFLAGS_SRC, -fuse-ld=mold)
+_MY_LDLIBS_CHECK_OPT(CFG_LDFLAGS_VERILATED, -fuse-ld=mold)
+
 # When linking partially statically
 if test "$CFG_ENABLE_PARTIAL_STATIC" = "yes"; then
  _MY_LDLIBS_CHECK_OPT(CFG_LDFLAGS_SRC, -static-libgcc)
@@ -480,6 +484,7 @@ else
  LTCMALLOC=-ltcmalloc_minimal
 fi
 AC_SUBST(CFG_LDFLAGS_SRC)
+AC_SUBST(CFG_LDFLAGS_VERILATED)

 # The pthread library is required by tcmalloc, so add it if it exists. If it
 # does not, the tcmalloc check below will fail anyway, and linking against
@@ -12,8 +12,12 @@ Alex Chadwick
 Aliaksei Chapyzhenka
 Ameya Vikram Singh
 Andreas Kuster
+Andrei Kostovski
+Andrew Miloradovsky
 Andrew Nolte
+Anthony Donlon
 Arkadiusz Kozdra
+Aylon Chaim Porat
 Cameron Kirk
 Chris Randall
 Chuxuan Wang
@@ -24,15 +28,20 @@ David Horton
 David Metz
 David Stanford
 David Turner
+Don Williamson
 Drew Ranck
 Drew Taussig
 Driss Hafdi
 Edgar E. Iglesias
 Eric Rippey
+Ethan Sifferman
 Eyck Jentzsch
 Fan Shupei
+february cozzocrea
 Felix Neumärker
 Felix Yan
+Frans Skarman
+G-A. Kamendje
 Garrett Smith
 Geza Lore
 Gianfranco Costamagna
@@ -41,16 +50,16 @@ Gökçe Aydos
 Graham Rushton
 Guokai Chen
 Gustav Svensk
-G-A. Kamendje
 Harald Heckmann
+Hennadii Chernyshchyk
 Howard Su
 Huang Rui
 Huanghuang Zhou
 HungMingWu
 HyungKi Jeong
+Ilya Barkov
 Iru Cai
 Ivan Vnučec
-Ilya Barkov
 Iztok Jeras
 Jake Merdich
 James Hanlon
@@ -65,14 +74,18 @@ Jeremy Bennett
 Jesse Taube
 Jevin Sweval
 Jiacheng Qian
+Jiamin Zhu
 Jiuyang Liu
 Joey Liu
 John Coiner
 John Demme
+John Wehle
 Jonathan Drolet
-Joseph Nwabueze
+Jose Loyola
 Josep Sans
+Joseph Nwabueze
 Josh Redford
+Julian Daube
 Julie Schwartz
 Julien Margetts
 Kaleb Barrett
@@ -82,6 +95,7 @@ Keith Colbert
 Kevin Kiningham
 Kritik Bhimani
 Krzysztof Bieganski
+Krzysztof Boronski
 Krzysztof Boroński
 Kuba Ober
 Larry Doolittle
@@ -108,6 +122,7 @@ Mostafa Gamal
 Nandu Raj
 Nathan Kohagen
 Nathan Myers
+Oleh Maksymenko
 Patrick Stewart
 Paul Wright
 Pawel Sagan
@@ -136,6 +151,7 @@ Stephen Henry
 Steven Hugg
 Sören Tempel
 Teng Huang
+Tim Hutt
 Tim Snyder
 Tobias Rosenkranz
 Tobias Wölfel
@@ -143,6 +159,7 @@ Todd Strader
 Tomasz Gorochowik
 Topa Topino
 Toru Niina
+Tudor Timi
 Tymoteusz Blazejczyk
 Udi Finkelstein
 Unai Martinez-Corral
@@ -150,15 +167,17 @@ Varun Koyyalagunta
 Vassilis Papaefstathiou
 Veripool API Bot
 Victor Besyakov
+Vito Gamberini
 William D. Jones
 Wilson Snyder
 Xi Zhang
 Yinan Xu
 Yoda Lee
 Yossi Nivin
+Yu-Sheng Lin
 Yuri Victorovich
 Yutetsu TAKATSUKASA
-Yu-Sheng Lin
 Yves Mathieu
 Zhanglei Wang
 Zixi Li
+أحمد المحمودي
@@ -18,12 +18,16 @@
 RST2HTML = rst2html
 PYTHON3 = python3
 DOXYGEN = doxygen
-SPHINXOPTS ?= -c guide
 SPHINXBUILD ?= sphinx-build

 SOURCEDIR = guide
 BUILDDIR = _build

+SPHINXOPTS ?= -c guide
+ifneq ($(VERILATOR_ANALYTICS_ID),)
+SPHINXOPTS += -D html_theme_options.analytics_id=$(VERILATOR_ANALYTICS_ID)
+endif
+
 ######################################################################

 .SUFFIXES:
@@ -31,7 +31,7 @@ def get_vlt_version():
            if match:
                try:
                    data = os.popen('git log -n 1 --pretty=%cs').read()
-                except Exception:
+                except Exception:  # pylint: disable=broad-except
                    data = ""  # fallback, and Sphinx will fill in today's date
                return "Devel " + match.group(1), data
    return "unknown", "unknown"
@@ -115,7 +115,6 @@ html_theme = 'sphinx_rtd_theme'
 html_theme_path = [sphinx_rtd_theme.get_html_theme_path()]

 html_theme_options = {
-    'analytics_id': 'G-D27B0C9QEB',
    'logo_only': True,
    'style_nav_header_background': "#45acf8",  # Default is #2980B9
    # 'canonical_url':
@@ -128,6 +128,43 @@ in the distribution.  These headers use Doxygen comments, `///` and `//<`,
 to indicate and document those functions that are part of the Verilated
 public API.

+Process-Level Clone APIs
+--------------------------
+
+Modern operating systems support process-level clone (a.k.a copying, forking)
+with system call interfaces in C/C++, e.g., :code:`fork()` in Linux.
+
+However, after cloning a parent process, some resources cannot be inherited
+in the child process. For example, in POSIX systems, when you fork a process,
+the child process inherits all the memory of the parent process. However,
+only the thread that called fork is replicated in the child process. Other
+threads are not.
+
+Therefore, to support the process-level clone mechanisms, Verilator supports
+:code:`prepareClone()` and :code:`atClone()` APIs to allow the user to manually
+re-construct the model in the child process. The two APIs handle all necessary
+resources required for releasing and re-initializing before and after cloning.
+
+The two APIs are supported in the verilated models. Here is an example of usage
+with Linux :code:`fork()` and :code:`pthread_atfork` APIs:
+
+.. code-block:: C++
+
+    // static function pointers to fit pthread_atfork
+    static auto prepareClone = [](){ topp->prepareClone(); };
+    static auto atClone = [](){ topp->atClone(); };
+
+    // in main function, register the handlers:
+    pthread_atfork(prepareClone, atClone, atClone);
+
+For better flexibility, you can also manually call the handlers before and
+after :code:`fork()`.
+
+With the process-level clone APIs, users can create process-level snapshots
+for the verilated models. While the Verilator save/restore option provides
+persistent and circuit-dependent snapshots, the process-level clone APIs
+enable in-memory, circuit-transparent, and highly efficient snapshots.
+

 Direct Programming Interface (DPI)
 ==================================
@@ -17,10 +17,10 @@ C++11 compiler support
  Verilator will require C++20 or newer compilers for both compiling
  Verilator and compiling all Verilated models no sooner than January 2025.

-Verilated_heavy.h
-  The legacy "verilated_heavy.h" include was replaced with just including
-  "verilated.h". Verilated_heavy.h is planned for removal no sooner than
-  July 2022.
+32-bit compiler support
+  Verilator currently regresses both 64-bit and 32-bit pointer modes (GCC's
+  `-m64` and `-m32`).  Support for 32-bit `-m32` mode will be deprecated no
+  sooner than January 2024.

 Option `-O<letter>`
  The debug `-O<letter>` options have been replaced with
@@ -91,9 +91,29 @@ associated programs.

 .. option:: VERILATOR_ROOT

-   Specifies the directory containing the distribution kit.  This is used
-   to find the executable, Perl library, and include files.  If not
-   specified, it will come from a default optionally specified at configure
-   time (before Verilator was compiled).  It should not be specified if
-   using a pre-compiled Verilator package as the hard-coded value should be
-   correct.  See :ref:`Installation`.
+   The ``VERILATOR_ROOT`` environment variable is used in several places:
+
+   * At ``./configure`` time: If set, it is embedded into the binary, and
+     at runtime if ``VERILATOR_ROOT`` is not set, the embedded value is
+     used for the runtime default.
+
+   * When ``verilator`` is run: If ``VERILATOR_ROOT`` is set it will be
+     used to find the ``verilator_bin`` executable (this is the actual
+     Verilator binary; ``verilator`` is a Perl wrapper). If not set, the
+     ``verilator`` script uses other methods to find ``verilator_bin``
+     (looking in the same directory and falling back to ``$PATH``).
+
+   * When ``make`` is run on the Makefile generated by ``verilator``: The
+     value of ``VERILATOR_ROOT`` (falling back to the value embedded in the
+     binary if not set) is used to find the include files
+     (``include/verilated.mk``).
+
+   If you are using a pre-compiled Verilator package, you should not need
+   to set ``VERILATOR_ROOT`` - the value embedded in the binary should be
+   correct.  In fact this option *does not work* with Verilator packages
+   that have been installed with ``make install``. If a Verilator package
+   has been installed using ``./configure --prefix=/some/path && make
+   install`` and then moved to another location, you cannot use
+   ``VERILATOR_ROOT`` to point to the new version.
+
+   See :ref:`Installation` for more details.
@@ -125,9 +125,12 @@ Summary:

   After generating the SystemC/C++ code, Verilator will invoke the
   toolchain to build the model library (and executable when :vlopt:`--exe`
-   is also used). Verilator manages the build itself, and for this --build
+   is also used).  Verilator manages the build itself, and for this --build
   requires GNU Make to be available on the platform.

+   :vlopt:`--build` cannot be specified when using :vlopt:`-E`,
+   :vlopt:`--dpi-hdr-only`, :vlopt:`--lint-only`, or :vlopt:`--xml-only`.
+
 .. option:: --build-dep-bin <filename>

   Rarely needed.  When a dependency (.d) file is created, this filename
@@ -808,6 +811,13 @@ Summary:

   See also :vlopt:`--binary`.

+.. option:: --main-top-name <string>
+
+   Specify the name passed to the Verilated model being constructed, in the
+   generated C++ main() function.
+
+   If the string ``"-"`` is used, no top level scope is added.
+
 .. option:: --max-num-width <value>

   Set the maximum number literal width (e.g., in 1024'd22 this
@@ -1307,6 +1317,7 @@ Summary:
   This is not needed with standard designs with only one top.  See also
   :option:`MULTITOP` warning.

+
 .. option:: --trace

   Adds waveform tracing code to the model using VCD format. This overrides
@@ -1385,6 +1396,17 @@ Summary:
   This option is accepted, but has absolutely no effect with
   :vlopt:`--trace`, which respects :vlopt:`--threads` instead.

+.. option:: --no-trace-top
+
+   Disables tracing for the input and output signals in the top wrapper which
+   Verilator adds to the design. The signals are still traced in the original
+   verilog top modules.
+
+   When combined with :option:`--main-top-name` set to "-" or when the name of
+   the top module is set to "" in its constructor, the generated trace file
+   will have the verilog top module as its root, rather than another module
+   added by Verilator.
+
 .. option:: --trace-underscore

   Enable tracing of signals or modules that start with an
@@ -1467,8 +1489,11 @@ Summary:
   them systematically.

   The generated file is in the Verilator Configuration format, see
-   :ref:`Configuration Files`, and can directly be consumed by
-   Verilator. The standard file extension is ".vlt".
+   :ref:`Configuration Files`. The standard file extension is ".vlt".
+   These files can directly be consumed by Verilator, typically by placing
+   the filename as part of the Verilator command line options. Waiver files
+   need to be listed on the command line before listing the files they are
+   waiving.

 .. option:: -Wall

@@ -1513,13 +1538,15 @@ Summary:

 .. option:: -Wno-lint

-   Disable all lint-related warning messages, and all style warnings.  This is
-   equivalent to ``-Wno-ALWCOMBORDER -Wno-ASCRANGE -Wno-BSSPACE -Wno-CASEINCOMPLETE
-   -Wno-CASEOVERLAP -Wno-CASEX -Wno-CASTCONST -Wno-CASEWITHX -Wno-CMPCONST -Wno-COLONPLUS
-   -Wno-IMPLICIT -Wno-IMPLICITSTATIC -Wno-PINCONNECTEMPTY
-   -Wno-PINMISSING -Wno-STATICVAR -Wno-SYNCASYNCNET -Wno-UNDRIVEN -Wno-UNSIGNED
-   -Wno-UNUSEDGENVAR -Wno-UNUSEDPARAM -Wno-UNUSEDSIGNAL
-   -Wno-WIDTH`` plus the list shown for Wno-style.
+   Disable all lint-related warning messages, and all style warnings.  This
+   is equivalent to ``-Wno-ALWCOMBORDER`` ``-Wno-ASCRANGE``
+   ``-Wno-BSSPACE`` ``-Wno-CASEINCOMPLETE`` ``-Wno-CASEOVERLAP``
+   ``-Wno-CASEX`` ``-Wno-CASTCONST`` ``-Wno-CASEWITHX`` ``-Wno-CMPCONST``
+   ``-Wno-COLONPLUS`` ``-Wno-IMPLICIT`` ``-Wno-IMPLICITSTATIC``
+   ``-Wno-PINCONNECTEMPTY`` ``-Wno-PINMISSING`` ``-Wno-STATICVAR``
+   ``-Wno-SYNCASYNCNET`` ``-Wno-UNDRIVEN`` ``-Wno-UNSIGNED``
+   ``-Wno-UNUSEDGENVAR`` ``-Wno-UNUSEDPARAM`` ``-Wno-UNUSEDSIGNAL``
+   ``-Wno-WIDTH``, plus the list shown for :vlopt:`-Wno-style`.

   It is strongly recommended that you clean up your code rather than using this
   option; it is only intended to be used when running test-cases of code
@@ -1527,12 +1554,13 @@ Summary:

 .. option:: -Wno-style

-   Disable all code style related warning messages (note that by default, they are
-   already disabled).  This is equivalent to ``-Wno-DECLFILENAME -Wno-DEFPARAM
-   -Wno-EOFNEWLINE -Wno-IMPORTSTAR -Wno-INCABSPATH -Wno-PINCONNECTEMPTY
-   -Wno-PINNOCONNECT -Wno-SYNCASYNCNET -Wno-UNDRIVEN
-   -Wno-UNUSEDGENVAR -Wno-UNUSEDPARAM -Wno-UNUSEDSIGNAL
-   -Wno-VARHIDDEN``.
+   Disable all code style related warning messages (note that by default,
+   they are already disabled).  This is equivalent to ``-Wno-DECLFILENAME``
+   ``-Wno-DEFPARAM`` ``-Wno-EOFNEWLINE`` ``-Wno-GENUNNAMED``
+   ``-Wno-IMPORTSTAR`` ``-Wno-INCABSPATH`` ``-Wno-PINCONNECTEMPTY``
+   ``-Wno-PINNOCONNECT`` ``-Wno-SYNCASYNCNET`` ``-Wno-UNDRIVEN``
+   ``-Wno-UNUSEDGENVAR`` ``-Wno-UNUSEDPARAM`` ``-Wno-UNUSEDSIGNAL``
+   ``-Wno-VARHIDDEN``.

 .. option:: -Wpedantic

@@ -1550,19 +1578,25 @@ Summary:

 .. option:: -Wwarn-lint

-   Enable all lint-related warning messages (note that by default, they are already
-   enabled), but do not affect style messages.  This is equivalent to
-   ``-Wwarn-ALWCOMBORDER -Wwarn-BSSPACE -Wwarn-CASEINCOMPLETE
-   -Wwarn-CASEOVERLAP -Wwarn-CASEX -Wwarn-CASTCONST -Wwarn-CASEWITHX -Wwarn-CMPCONST
-   -Wwarn-COLONPLUS -Wwarn-IMPLICIT -Wwarn-ASCRANGE
-   -Wwarn-PINMISSING -Wwarn-REALCVT -Wwarn-UNSIGNED -Wwarn-WIDTH``.
+   Enable all lint-related warning messages (note that by default, they are
+   already enabled), but do not affect style messages.  This is equivalent
+   to ``-Wwarn-ALWCOMBORDER`` ``-Wwarn-ASCRANGE`` ``-Wwarn-BSSPACE``
+   ``-Wwarn-CASEINCOMPLETE`` ``-Wwarn-CASEOVERLAP`` ``-Wwarn-CASEWITHX``
+   ``-Wwarn-CASEX`` ``-Wwarn-CASTCONST`` ``-Wwarn-CMPCONST``
+   ``-Wwarn-COLONPLUS`` ``-Wwarn-IMPLICIT`` ``-Wwarn-IMPLICITSTATIC``
+   ``-Wwarn-LATCH`` ``-Wwarn-MISINDENT`` ``-Wwarn-NEWERSTD``
+   ``-Wwarn-PINMISSING`` ``-Wwarn-REALCVT`` ``-Wwarn-STATICVAR``
+   ``-Wwarn-UNSIGNED`` ``-Wwarn-WIDTHTRUNC`` ``-Wwarn-WIDTHEXPAND``
+   ``-Wwarn-WIDTHXZEXPAND``.

 .. option:: -Wwarn-style

   Enable all code style-related warning messages.  This is equivalent to
-   ``-Wwarn ASSIGNDLY -Wwarn-DECLFILENAME -Wwarn-DEFPARAM -Wwarn-EOFNEWLINE
-   -Wwarn-INCABSPATH -Wwarn-PINNOCONNECT -Wwarn-SYNCASYNCNET -Wwarn-UNDRIVEN
-   -Wwarn-UNUSEDGENVAR -Wwarn-UNUSEDPARAM -Wwarn-UNUSEDSIGNAL -Wwarn-VARHIDDEN``.
+   ``-Wwarn-ASSIGNDLY`` ``-Wwarn-DECLFILENAME`` ``-Wwarn-DEFPARAM``
+   ``-Wwarn-EOFNEWLINE`` ``-Wwarn-GENUNNAMED`` ``-Wwarn-INCABSPATH``
+   ``-Wwarn-PINNOCONNECT`` ``-Wwarn-SYNCASYNCNET`` ``-Wwarn-UNDRIVEN``
+   ``-Wwarn-UNUSEDGENVAR`` ``-Wwarn-UNUSEDPARAM`` ``-Wwarn-UNUSEDSIGNAL``
+   ``-Wwarn-VARHIDDEN``.

 .. option:: --x-assign 0

@@ -356,8 +356,8 @@ Why do I get "undefined reference to \`VL_RAND_RESET_I' or \`Verilated::...'"?
 You need to link your compiled Verilated code against the
 :code:`verilated.cpp` file found in the include directory of the Verilator
 kit.  This is one target in the ``$(VK_GLOBAL_OBJS)`` make variable, which
-should be part of your Makefile's link rule.  If you use :vlopt:`--exe`,
-this is done for you.
+should be part of your Makefile's link rule.  If you use :vlopt:`--exe` or
+:vlopt:`--binary`, this is done for you.


 Is the PLI supported?
@@ -486,7 +486,7 @@ From the sc_main.cpp file, you'd then:

      #include "Vour.h"
      #include "Vour_our.h"
-      cout << "clock is " << top->our->clk << endl;
+      std::cout << "clock is " << top->our->clk << std::endl;


 In this example, clk is a bool you can read or set as any other variable.
@@ -126,8 +126,10 @@ In specific debug and other modes, it also creates:
     - Debugging graph files (from --debug)
   * - *{prefix}{misc}*\ .tree
     - Debugging files (from --debug)
-   * - {mod_prefix}_{each_verilog_base_filename}*\ .vpp
-     - Pre-processed verilog (from --debug)
+   * - *{prefix}*\ __inputs\ .vpp
+     - Pre-processed verilog for all files (from --debug)
+   * - *{prefix}*\ _ *{each_verilog_base_filename}*\ .vpp
+     - Pre-processed verilog for each file (from --debug)

 After running Make, the C++ compiler may produce the following:

@@ -39,7 +39,7 @@ In brief, to install from git:
 ::

   # Prerequisites:
-   #sudo apt-get install git perl python3 make autoconf g++ flex bison ccache
+   #sudo apt-get install git help2man perl python3 make autoconf g++ flex bison ccache
   #sudo apt-get install libgoogle-perftools-dev numactl perl-doc
   #sudo apt-get install libfl2  # Ubuntu only (ignore if gives error)
   #sudo apt-get install libfl-dev  # Ubuntu only (ignore if gives error)
@@ -104,6 +104,7 @@ for good performance:
 ::

   sudo apt-get install ccache  # If present at build, needed for run
+   sudo apt-get install mold  # If present at build, needed for run
   sudo apt-get install libgoogle-perftools-dev numactl

 The following is optional but is recommended for nicely rendered command line
@@ -124,9 +125,9 @@ Those developing Verilator itself may also want these (see internals.rst):

 ::

-   sudo apt-get install gdb graphviz cmake clang clang-format-14 gprof lcov
-   sudo apt-get install libclang-dev yapf3
-   sudo pip3 install clang sphinx sphinx_rtd_theme sphinxcontrib-spelling breathe
+   sudo apt-get install clang clang-format-14 cmake gdb gprof graphviz lcov
+   sudo apt-get install libclang-dev yapf3 bear
+   sudo pip3 install clang sphinx sphinx_rtd_theme sphinxcontrib-spelling breathe ruff
   cpan install Pod::Perldoc
   cpan install Parallel::Forker

@@ -201,8 +202,9 @@ These are the installation options:
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

 Our personal favorite is to always run Verilator in-place from its Git
-directory. This allows the easiest experimentation and upgrading, and
-allows many versions of Verilator to co-exist on a system.
+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.

 ::

@@ -211,17 +213,18 @@ allows many versions of Verilator to co-exist on a system.
   ./configure
   # Running will use files from $VERILATOR_ROOT, so no install needed

-Note after installing (below steps), a calling program or shell must set
-the environment variable :option:`VERILATOR_ROOT` to point to this Git
-directory, then execute ``$VERILATOR_ROOT/bin/verilator``, which will find
-the path to all needed files.
+Note after installing (see `Installation`_), a calling program or shell
+must set the environment variable :option:`VERILATOR_ROOT` to point to this
+Git directory, then execute ``$VERILATOR_ROOT/bin/verilator``, which will
+find the path to all needed files.


-2. Install into a specific location
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+2. Install into a Specific Prefix
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

-You may eventually be installing onto a project/company-wide "CAD" tools
-disk that may support multiple versions of every tool. Tell configure the
+You may be an OS package maintainer building a Verilator package, or you
+may eventually be installing onto a project/company-wide "CAD" tools disk
+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:

@@ -232,7 +235,8 @@ location include the Verilator version name:
   # For the tarball, use the version number instead of git describe
   ./configure --prefix /CAD_DISK/verilator/`git describe | sed "s/verilator_//"`

-Note after installing (below steps), if you use `modulecmd
+Note after installing (see `Installation`_), you need to add the path to
+the ``bin`` directory to your ``PATH``. Or, if you use `modulecmd
 <http://modules.sourceforge.net/>`__, you'll want a module file like the
 following:

@@ -245,23 +249,7 @@ following:
   prepend-path PKG_CONFIG_PATH $install_root/share/pkgconfig


-3. Install into a Specific Path
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
-
-You may eventually install Verilator into a specific installation prefix,
-as most GNU tools support:
-
-::
-
-   unset VERILATOR_ROOT      # if your shell is bash
-   unsetenv VERILATOR_ROOT   # if your shell is csh
-   ./configure --prefix /opt/verilator-VERSION
-
-Then after installing (below steps), you will need to add
-``/opt/verilator-VERSION/bin`` to your ``$PATH`` environment variable.
-
-
-4. Install System Globally
+3. Install System Globally
 ^^^^^^^^^^^^^^^^^^^^^^^^^^

 The final option is to eventually install Verilator globally, using
@@ -273,8 +261,8 @@ configure's default system paths:
   unsetenv VERILATOR_ROOT   # if your shell is csh
   ./configure

-Then after installing (below), the binaries should be in a location
-already in your ``$PATH`` environment variable.
+Then after installing (see `Installation`_), the binaries should be in a
+location already in your ``$PATH`` environment variable.


 Configure
@@ -44,7 +44,7 @@ The best place to get started is to try the :ref:`Examples`.
 .. [#] Verilog is defined by the `Institute of Electrical and Electronics
       Engineers (IEEE) Standard for Verilog Hardware Description
       Language`, Std. 1364, released in 1995, 2001, and 2005.  The
-       Verilator documentation uses the shorthand, e.g., "IEEE 1394-2005",
+       Verilator documentation uses the shorthand, e.g., "IEEE 1364-2005",
       to refer to the, e.g., 2005 version of this standard.

 .. [#] SystemVerilog is defined by the `Institute of Electrical and
@@ -143,7 +143,7 @@ Functional Coverage
 -------------------

 With :vlopt:`--coverage` or :vlopt:`--coverage-user`, Verilator will
-translate functional coverage points the user has inserted manually win
+translate functional coverage points the user has inserted manually in
 SystemVerilog code through into the Verilated model.

 Currently, all functional coverage points are specified using SystemVerilog
@@ -500,7 +500,7 @@ feedback-directed optimization.  See the appropriate compiler
 documentation.


-.. _Runtime Debugging
+.. _Runtime Debugging:

 Runtime Debugging
 =================
@@ -165,7 +165,7 @@ compiled on the different target system.

 To support this, none of the files that Verilator produces will reference
 any configure-generated build-system-specific files, such as
-:file:`config.h` (which is renamed in Verilator to :file:`config_build.h`
+:file:`config.h` (which is renamed in Verilator to :file:`config_package.h`
 to reduce confusion.)  The disadvantage of this approach is that
 :file:`include/verilatedos.h` must self-detect the requirements of the
 target system, rather than using configure.
@@ -670,6 +670,52 @@ List Of Warnings
   used as a clock.


+.. option:: GENUNNAMED
+
+   Warns that a generate block was unnamed and "genblk" will be used per
+   IEEE.
+
+   The potential issue is that adding additional generate blocks will
+   renumber the assigned names, which may cause eventual problems with
+   synthesis constraints or other tools that depend on hierarchical paths
+   remaining consistent.
+
+   Blocks that are empty may not be reported with this warning, as no
+   scopes are created for empty blocks, so there is no harm in having them
+   unnamed.
+
+   Disabled by default as this is a code-style warning; it will simulate
+   correctly.
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 2
+
+         generate
+            if (PARAM == 1) begin  //<--- Warning
+            end
+
+   Results in:
+
+   .. code-block::
+
+         %Warning-GENUNNAMED: example.v:2:9: Unnamed generate block (IEEE 1800-2017 27.6)
+
+   To fix this assign a label (often with the naming convention prefix of
+   :code:`gen_` or :code:`g_`), for example:
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 2
+
+         generate
+            if (PARAM == 1) begin : gen_param_1  //<--- Repaired
+            end
+
+   Other tools with similar warnings: Verible's generate-label, "All
+   generate block statements must have a label."
+
+
 .. option:: HIERBLOCK

   Warns that the top module is marked as a hierarchy block by the
@@ -769,11 +815,14 @@ List Of Warnings
   then automatic as static prevents the function from being reentrant,
   which may be a source of bugs, and/or performance issues.

-   If the function does not require static behavior, change it to "function
-   automatic".
+   If the function is in a module, and does not require static behavior,
+   change it to "function automatic".

-   If the function requires static behavior, change it to "function
-   static".
+   If the function is in a module, and requires static behavior, change it
+   to "function static".
+
+   If the function is in a package, it defaults to static, and label the
+   function's variables as static.

   Ignoring this warning will only suppress the lint check; it will
   simulate correctly.
@@ -873,6 +922,102 @@ List Of Warnings
   Ignoring this warning will only suppress the lint check; it will
   simulate correctly.

+.. option:: LIFETIME
+
+   Error when a variable is referenced in a process that can outlive the process
+   in which it was declared. This can happen when using 'fork..join_none' or
+   'fork..join_any' blocks, which spawn process that can outlive their parents.
+   This error occurs only when Verilator can't replace the reference with a
+   reference to copy of this variable, local to the forked process. For example:
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 3
+
+         task foo(int local_var);
+            fork
+               #10 local_var++;
+               #20 $display("local_var = %d", local_var);
+            join_none
+         endtask
+
+   In the example above 'local_var' exists only within scope of 'foo', once foo
+   finishes, the stack frame containing 'i' gets removed. However, the process
+   forked from foo continues, as it contains a delay. After 10 units of time
+   pass, this process attempts to modify 'local_var'. However, this variable no
+   longer exits. It can't be made local to the forked process upon spawning, because
+   it's modified and can be referenced somewhere else, for example in the other
+   forked process, that was delayed by 20 units of time in this example. Thus,
+   there's no viable stack allocation for it.
+
+   In order to fix it, if the intent is not to share the variable's state outside
+   of the process, then create a local copy of the variable.
+
+   For example:
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 4
+
+         task foo(int local_var);
+            fork
+               #10 begin
+                  int forked_var = local_var;
+                  forked_var++;
+               end
+               #20 begin
+                  // Note that we are going to print the original value here,
+                  // as `forked_var`is a local copy that was initialized while
+                  // `foo` was still alive.
+                  int forked_var = local_var;
+                  $display("forked_var = %d", forked_var)
+               end
+            join_none
+         endtask
+
+   If you need to share its state, another strategy is to ensure it's allocated
+   statically:
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 1
+
+         int static_var;
+
+         task foo();
+            fork
+               #10 static_var++;
+               #20 $display("static_var = %d", static_var);
+            join_none
+         endtask
+
+   However, if you need to be able to instantiate at runtime, the solution would be to
+   wrap it in an object, since the forked process can hold a reference to that object
+   and ensure that the variable stays alive this way:
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 2
+
+         class Wrapper;
+            int m_var;
+
+            // Here we implicitly hold a reference to `this`
+            task foo();
+               fork
+                  #10 m_var++;
+                  #20 $display("this.m_var = %d", m_var);
+               join_none
+            endtask
+         endclass
+
+         // Here we explicitly hold a handle to an object
+         task bar(Wrapper wrapper);
+            fork
+               #10 wrapper.m_var++;
+               #20 $display("wrapper.m_var = %d", wrapper.m_var);
+            join_none
+         endtask

 .. option:: LITENDIAN

@@ -894,6 +1039,51 @@ List Of Warnings
   unsupported. Verilator uses only the typical delay value.


+.. option:: MISINDENT
+
+   Warns that the indentation of a statement is misleading, suggesting the
+   statement is part of a previous :code:`if` or :code:`while` block while
+   it is not.
+
+   Verilator suppresses this check when there is an inconsistent mix of
+   spaces and tabs, as it cannot ensure the width of tabs.  Verilator also
+   ignores blocks with :code:`begin`/:code:`end`, as the :code:`end`
+   visually indicates the earlier statement's end.
+
+   Ignoring this warning will only suppress the lint check; it will
+   simulate correctly.
+
+   For example
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 3
+
+         if (something)
+            statement_in_if;
+            statement_not_in_if;  //<--- Warning
+
+   Results in:
+
+   .. code-block::
+
+         %Warning-MISINDENT: example.v:3:9: Misleading indentation
+
+   To fix this repair the indentation to match the correct earlier
+   statement, for example:
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 3
+
+         if (something)
+            statement_in_if;
+         statement_not_in_if;  //<--- Repaired
+
+   Other tools with similar warnings: GCC -Wmisleading-indentation,
+   clang-tidy readability-misleading-indentation.
+
+
 .. option:: MODDUP

   .. TODO better example
@@ -972,6 +1162,21 @@ List Of Warnings
   (neither :vlopt:`--timing` nor :vlopt:`--no-timing` option was provided).


+.. option:: NEWERSTD
+
+   Warns that a feature requires a newer standard of Verilog or SystemVerilog
+   than the one specified by the :vlopt:`--language` option. For example, unsized
+   unbased literals (`'0`, `'1`, `'z`, `'x`) require 1800-2005 or later.
+
+   To avoid this warning, use a Verilog or SystemVerilog standard that
+   supports the feature. Alternatively, modify your code to use a different
+   syntax that is supported by the Verilog/SystemVerilog standard specified
+   by the :vlopt:`--language` option.
+
+   Ignoring this warning will only suppress the lint check; it will
+   simulate correctly.
+
+
 .. option:: NOLATCH

   .. TODO better example
@@ -1785,11 +1990,15 @@ List Of Warnings

   .. code-block:: sv

-         wait(0);  // Blocks forever
+         wait(1);  // Blocks forever

   Warns that a `wait` statement awaits a constant condition, which means it
   either blocks forever or never blocks.

+   As a special case `wait(0)` with the literal constant `0` (as opposed to
+   something that elaborates to zero), does not warn, as it is presumed the
+   code is making the intent clear.
+

 .. option:: WIDTH

@@ -638,7 +638,22 @@ condition. See the `Timing Pass` section for more details.
 Timing Pass
 ~~~~~~~~~~~

-The visitor in ``V3Timing.cpp`` transforms each timing control into a ``co_await``.
+There are two visitors in ``V3Timing.cpp``.
+
+The first one, ``TimingSuspendableVisitor``, does not perform any AST
+transformations. It is responsible for marking processes and C++ functions that
+contain timing controls as suspendable. Processes that call suspendable
+functions are also marked as suspendable. Functions that call, are overridden
+by, or override suspendable functions are marked as suspendable as well.
+
+The visitor keeps a dependency graph of functions and processes to handle such
+cases. A function or process is dependent on a function if it calls it. A
+virtual class method is dependent on another class method if it calls it,
+overrides it, or is overriden by it.
+
+The second visitor in ``V3Timing.cpp``, ``TimingControlVisitor``, uses the
+information provided by ``TimingSuspendableVisitor`` and transforms each timing
+control into a ``co_await``.

 * event controls are turned into ``co_await`` on a trigger scheduler's
  ``trigger`` method. The awaited trigger scheduler is the one corresponding to
@@ -670,14 +685,9 @@ Each sub-statement of a ``fork`` is put in an ``AstBegin`` node for easier
 grouping. In a later step, each of these gets transformed into a new, separate
 function. See the `Forks` section for more detail.

-Processes that use awaits are marked as suspendable. Later, during ``V3Sched``,
-they are transformed into coroutines. Functions that use awaits get the return
-type of ``VlCoroutine``. This immediately makes them coroutines. Note that if a
-process calls a function that is a coroutine, the call gets wrapped in an
-await, which means the process itself will be marked as suspendable. A virtual
-function is a coroutine if any of its overriding or overridden functions are
-coroutines. The visitor keeps a dependency graph of functions and processes to
-handle such cases.
+Suspendable functions get the return type of ``VlCoroutine``, which makes them
+coroutines. Later, during ``V3Sched``, suspendable processes are also
+transformed into coroutines.

 Scheduling with timing
 ~~~~~~~~~~~~~~~~~~~~~~
@@ -1503,8 +1513,19 @@ debug level 5, with the V3Width.cpp file at level 9.
 --debug
 -------

-When you run with ``--debug``, there are two primary output file types
-placed into the obj_dir, .tree and .dot files.
+When you run with ``--debug``, there are three primary output file types
+placed into the obj_dir, .vpp, .tree and .dot files.
+
+.vpp Output
+-----------
+
+Verilator creates a *{mod_prefix}*\ __inputs\ .vpp file containing all the
+files that were read, filtered by preprocessing. This file can be fed back
+into Verilator, replacing on the command line all of the previous input
+files, to enable simplification of test cases.
+
+Verilator also creates .vpp files for each individual file passed on the
+command line.


 .dot Output
@@ -2102,7 +2123,7 @@ Or in a hand-written C++ wrapper:
 ::

   #ifdef TEST_VERBOSE
-      cout << "Read a=" << a << endl;
+      std::cout << "Read a=" << a << std::endl;
   #endif

 A filename that should be used to check the output results is given with
@@ -56,6 +56,7 @@ Das
 Dcache
 Deadman
 Debacker
+Deepa
 Defparams
 Denio
 Deprecations
@@ -213,6 +214,7 @@ Maupin
 Mdir
 Mednick
 Menküc
+Michail
 Michiels
 Microsystems
 Milanovic
@@ -246,6 +248,7 @@ Olof
 Olofsson
 Oyvind
 PLI
+Palaniappan
 Patricio
 Petr
 Piechotka
@@ -260,6 +263,8 @@ Prateek
 Pre
 Preprocess
 Pretet
+Pretl
+Pringlemeir
 Priyadharshini
 Pullup
 Pulver
@@ -274,6 +279,7 @@ Renga
 Requin
 Rodionov
 Rolfe
+Rontionov
 Roodselaar
 Runtime
 Ruud
@@ -375,6 +381,7 @@ Wfuture
 Whatson
 Wildman
 Wim
+Wmisleading
 Wno
 Wojciech
 Wolfel
@@ -409,7 +416,9 @@ architected
 args
 arrarys
 assertOn
+astgen
 async
+atClone
 ato
 atoi
 autoconf
@@ -472,6 +481,7 @@ constexpr
 constpool
 coredump
 coroutine
+coroutines
 countbits
 countones
 cout
@@ -602,6 +612,7 @@ genvar
 genvars
 getenv
 getline
+ggdb
 gmake
 gmon
 gotFinish
@@ -646,6 +657,7 @@ ish
 isunbounded
 isunknown
 jobserver
+jwoutersymatra
 killua
 lang
 lcov
@@ -765,6 +777,7 @@ pragmas
 pre
 precisions
 predefines
+prepareClone
 prepend
 prepended
 preprocess
@@ -791,6 +804,7 @@ qrq
 radix
 randc
 randcase
+randstate
 rarr
 rdtsc
 reStructuredText
@@ -896,10 +910,12 @@ typedefs
 typename
 uint
 un
+unbased
 undef
 undefineall
 undriven
 ungetc
+unhandled
 uniquified
 unistd
 unlink
@@ -920,7 +936,9 @@ vc
 vcd
 vcddiff
 vcoverage
+vdhotre
 vec
+ventana
 ver
 verFiles
 verible
@@ -2052,7 +2052,8 @@ if(xc && !xc->already_in_close && !xc->already_in_flush)
                int zfd;
                int fourpack_duo = 0;
 #ifndef __MINGW32__
-                char *fnam = (char *)malloc(strlen(xc->filename) + 5 + 1);
+		int fnam_len = strlen(xc->filename) + 5 + 1;
+                char *fnam = (char *)malloc(fnam_len);
 #endif

                fixup_offs = ftello(xc->handle);
@@ -2142,7 +2143,7 @@ if(xc && !xc->already_in_close && !xc->already_in_flush)
                fflush(xc->handle);

 #ifndef __MINGW32__
-                sprintf(fnam, "%s.hier", xc->filename);
+                snprintf(fnam, fnam_len, "%s.hier", xc->filename);
                unlink(fnam);
                free(fnam);
 #endif
@@ -2821,7 +2822,7 @@ if(ctx && name && literal_arr && val_arr && (elem_count != 0))
 	uint32_t i;

 	name_len = strlen(name);
-	elem_count_len = sprintf(elem_count_buf, "%" PRIu32, elem_count);
+	elem_count_len = snprintf(elem_count_buf, 16, "%" PRIu32, elem_count);

 	literal_lens = (unsigned int *)calloc(elem_count, sizeof(unsigned int));
 	val_lens = (unsigned int *)calloc(elem_count, sizeof(unsigned int));
@@ -3909,7 +3910,8 @@ int pass_status = 1;
 if(!xc->fh)
        {
        fst_off_t offs_cache = ftello(xc->f);
-        char *fnam = (char *)malloc(strlen(xc->filename) + 6 + 16 + 32 + 1);
+	int fnam_len = strlen(xc->filename) + 6 + 16 + 32 + 1;
+        char *fnam = (char *)malloc(fnam_len);
        unsigned char *mem = (unsigned char *)malloc(FST_GZIO_LEN);
        fst_off_t hl, uclen;
        fst_off_t clen = 0;
@@ -3928,7 +3930,7 @@ if(!xc->fh)
                htyp = xc->contains_hier_section_lz4duo ? FST_BL_HIER_LZ4DUO : FST_BL_HIER_LZ4;
                }

-        sprintf(fnam, "%s.hier_%d_%p", xc->filename, getpid(), (void *)xc);
+        snprintf(fnam, fnam_len, "%s.hier_%d_%p", xc->filename, getpid(), (void *)xc);
        fstReaderFseeko(xc, xc->f, xc->hier_pos, SEEK_SET);
        uclen = fstReaderUint64(xc->f);
 #ifndef __MINGW32__
@@ -4179,7 +4181,7 @@ if(!(isfeof=feof(xc->fh)))
                                if((xc->hier.u.attr.subtype == FST_MT_SOURCESTEM)||(xc->hier.u.attr.subtype == FST_MT_SOURCEISTEM))
                                        {
                                        int sidx_skiplen_dummy = 0;
-                                        xc->hier.u.attr.arg_from_name = fstGetVarint64((unsigned char *)xc->str_scope_nam, &sidx_skiplen_dummy);
+                                        xc->hier.u.attr.arg_from_name = fstGetVarint64((unsigned char *)xc->str_scope_attr, &sidx_skiplen_dummy);
                                        }
                                }
                        break;
@@ -4585,15 +4587,17 @@ if(sectype == FST_BL_ZWRAPPER)
        int zfd;
        int flen = strlen(xc->filename);
        char *hf;
+	int hf_len;

        seclen = fstReaderUint64(xc->f);
        uclen = fstReaderUint64(xc->f);

        if(!seclen) return(0); /* not finished compressing, this is a failed read */

-        hf = (char *)calloc(1, flen + 16 + 32 + 1);
+	hf_len = flen + 16 + 32 + 1;
+        hf = (char *)calloc(1, hf_len);

-        sprintf(hf, "%s.upk_%d_%p", xc->filename, getpid(), (void *)xc);
+        snprintf(hf, hf_len, "%s.upk_%d_%p", xc->filename, getpid(), (void *)xc);
        fcomp = fopen(hf, "w+b");
        if(!fcomp)
                {
@@ -5189,16 +5193,16 @@ for(;;)

                                if(beg_tim)
                                        {
-					if(dumpvars_state == 1) { wx_len = sprintf(wx_buf, "$end\n"); fstWritex(xc, wx_buf, wx_len); dumpvars_state = 2; }
-                                        wx_len = sprintf(wx_buf, "#%" PRIu64 "\n", beg_tim);
+					if(dumpvars_state == 1) { wx_len = snprintf(wx_buf, 32, "$end\n"); fstWritex(xc, wx_buf, wx_len); dumpvars_state = 2; }
+                                        wx_len = snprintf(wx_buf, 32, "#%" PRIu64 "\n", beg_tim);
                                        fstWritex(xc, wx_buf, wx_len);
-					if(!dumpvars_state) { wx_len = sprintf(wx_buf, "$dumpvars\n"); fstWritex(xc, wx_buf, wx_len); dumpvars_state = 1; }
+					if(!dumpvars_state) { wx_len = snprintf(wx_buf, 32, "$dumpvars\n"); fstWritex(xc, wx_buf, wx_len); dumpvars_state = 1; }
                                        }
                                if((xc->num_blackouts)&&(cur_blackout != xc->num_blackouts))
                                        {
                                        if(beg_tim == xc->blackout_times[cur_blackout])
                                                {
-                                                wx_len = sprintf(wx_buf, "$dump%s $end\n", (xc->blackout_activity[cur_blackout++]) ? "on" : "off");
+                                                wx_len = snprintf(wx_buf, 32, "$dump%s $end\n", (xc->blackout_activity[cur_blackout++]) ? "on" : "off");
                                                fstWritex(xc, wx_buf, wx_len);
                                                }
                                        }
@@ -5332,7 +5336,7 @@ for(;;)
                                                                                        clone_d[j] = srcdata[7-j];
                                                                                        }
                                                                                }
-                                                                        sprintf((char *)xc->temp_signal_value_buf, "%.16g", d);
+                                                                        snprintf((char *)xc->temp_signal_value_buf, xc->longest_signal_value_len + 1, "%.16g", d);
                                                                        value_change_callback(user_callback_data_pointer, beg_tim, idx+1, xc->temp_signal_value_buf);
                                                                        }
                                                                }
@@ -5360,7 +5364,7 @@ for(;;)
                                                                                }

                                                                        fstVcdID(vcdid_buf, idx+1);
-                                                                        wx_len = sprintf(wx_buf, "r%.16g %s\n", d, vcdid_buf);
+                                                                        wx_len = snprintf(wx_buf, 64, "r%.16g %s\n", d, vcdid_buf);
                                                                        fstWritex(xc, wx_buf, wx_len);
                                                                        }
                                                                }
@@ -5625,16 +5629,16 @@ for(;;)
                                                }
                                        }

-				if(dumpvars_state == 1) { wx_len = sprintf(wx_buf, "$end\n"); fstWritex(xc, wx_buf, wx_len); dumpvars_state = 2; }
-                                wx_len = sprintf(wx_buf, "#%" PRIu64 "\n", time_table[i]);
+				if(dumpvars_state == 1) { wx_len = snprintf(wx_buf, 32, "$end\n"); fstWritex(xc, wx_buf, wx_len); dumpvars_state = 2; }
+                                wx_len = snprintf(wx_buf, 32, "#%" PRIu64 "\n", time_table[i]);
                                fstWritex(xc, wx_buf, wx_len);
-				if(!dumpvars_state) { wx_len = sprintf(wx_buf, "$dumpvars\n"); fstWritex(xc, wx_buf, wx_len); dumpvars_state = 1; }
+				if(!dumpvars_state) { wx_len = snprintf(wx_buf, 32, "$dumpvars\n"); fstWritex(xc, wx_buf, wx_len); dumpvars_state = 1; }

                                if((xc->num_blackouts)&&(cur_blackout != xc->num_blackouts))
                                        {
                                        if(time_table[i] == xc->blackout_times[cur_blackout])
                                                {
-                                                wx_len = sprintf(wx_buf, "$dump%s $end\n", (xc->blackout_activity[cur_blackout++]) ? "on" : "off");
+                                                wx_len = snprintf(wx_buf, 32, "$dump%s $end\n", (xc->blackout_activity[cur_blackout++]) ? "on" : "off");
                                                fstWritex(xc, wx_buf, wx_len);
                                                }
                                        }
@@ -5884,7 +5888,7 @@ for(;;)
                                                                        clone_d[j] = srcdata[7-j];
                                                                        }
                                                                }
-                                                        sprintf((char *)xc->temp_signal_value_buf, "%.16g", d);
+                                                        snprintf((char *)xc->temp_signal_value_buf, xc->longest_signal_value_len + 1, "%.16g", d);
                                                        value_change_callback(user_callback_data_pointer, time_table[i], idx+1, xc->temp_signal_value_buf);
                                                        }
                                                }
@@ -5910,7 +5914,7 @@ for(;;)
                                                                        }
                                                                }

-                                                        wx_len = sprintf(wx_buf, "r%.16g", d);
+                                                        wx_len = snprintf(wx_buf, 32, "r%.16g", d);
                                                        fstWritex(xc, wx_buf, wx_len);
                                                        }
                                                }
@@ -6016,7 +6020,7 @@ if(xc->signal_lens[facidx] == 1)
                                }
                        }

-                sprintf((char *)buf, "%.16g", d);
+                snprintf((char *)buf, 32, "%.16g", d); /* this will write 18 bytes */
                }
        }

@@ -6608,7 +6612,7 @@ if(xc->signal_lens[facidx] == 1)
                                        }
                                }

-                        sprintf(buf, "r%.16g", d);
+                        snprintf(buf, 32, "r%.16g", d); /* this will write 19 bytes */
                        return(buf);
                        }
                }
@@ -315,15 +315,28 @@ void VlRNG::srandom(uint64_t n) VL_MT_UNSAFE {
    if (VL_COUNTONES_I(m_state[0]) < 10) m_state[0] = ~m_state[0];
    if (VL_COUNTONES_I(m_state[1]) < 10) m_state[1] = ~m_state[1];
 }
-// Unused: void VlRNG::set_randstate(const std::string& state) VL_MT_UNSAFE {
-// Unused:     if (VL_LIKELY(state.length() == sizeof(m_state))) {
-// Unused:         memcpy(m_state, state.data(), sizeof(m_state));
-// Unused:     }
-// Unused: }
-// Unused: std::string VlRNG::get_randstate() const VL_MT_UNSAFE {
-// Unused:     std::string result{reinterpret_cast<const char *>(&m_state), sizeof(m_state)};
-// Unused:     return result;
-// Unused: }
+std::string VlRNG::get_randstate() const VL_MT_UNSAFE {
+    // Though not stated in IEEE, assumption is the string must be printable
+    const char* const stateCharsp = reinterpret_cast<const char*>(&m_state);
+    static_assert(sizeof(m_state) == 16, "");
+    std::string result{"R00112233445566770011223344556677"};
+    for (size_t i = 0; i < sizeof(m_state); ++i) {
+        result[1 + i * 2] = 'a' + ((stateCharsp[i] >> 4) & 15);
+        result[1 + i * 2 + 1] = 'a' + (stateCharsp[i] & 15);
+    }
+    return result;
+}
+void VlRNG::set_randstate(const std::string& state) VL_MT_UNSAFE {
+    if (VL_UNLIKELY((state.length() != 1 + 2 * sizeof(m_state)) || (state[0] != 'R'))) {
+        VL_PRINTF_MT("%%Warning: set_randstate ignored as state string not from get_randstate\n");
+        return;
+    }
+    char* const stateCharsp = reinterpret_cast<char*>(&m_state);
+    for (size_t i = 0; i < sizeof(m_state); ++i) {
+        stateCharsp[i]
+            = (((state[1 + i * 2] - 'a') & 15) << 4) | ((state[1 + i * 2 + 1] - 'a') & 15);
+    }
+}

 static uint32_t vl_sys_rand32() VL_MT_SAFE {
    // Return random 32-bits using system library.
@@ -1360,16 +1373,18 @@ IData _vl_vsscanf(FILE* fp,  // If a fscanf
                    *p = t_tmp;
                } else if (obits <= VL_BYTESIZE) {
                    CData* const p = va_arg(ap, CData*);
-                    *p = owp[0];
+                    *p = VL_CLEAN_II(obits, obits, owp[0]);
                } else if (obits <= VL_SHORTSIZE) {
                    SData* const p = va_arg(ap, SData*);
-                    *p = owp[0];
+                    *p = VL_CLEAN_II(obits, obits, owp[0]);
                } else if (obits <= VL_IDATASIZE) {
                    IData* const p = va_arg(ap, IData*);
-                    *p = owp[0];
+                    *p = VL_CLEAN_II(obits, obits, owp[0]);
                } else if (obits <= VL_QUADSIZE) {
                    QData* const p = va_arg(ap, QData*);
-                    *p = VL_SET_QW(owp);
+                    *p = VL_CLEAN_QQ(obits, obits, VL_SET_QW(owp));
+                } else {
+                    _vl_clean_inplace_w(obits, owp);
                }
            }
            }  // switch
@@ -1881,6 +1896,12 @@ std::string VL_CVT_PACK_STR_NW(int lwords, const WDataInP lwp) VL_PURE {
    return std::string{destout, len};
 }

+std::string VL_CVT_PACK_STR_ND(const VlQueue<std::string>& q) VL_PURE {
+    std::string output;
+    for (const std::string& s : q) output += s;
+    return output;
+}
+
 std::string VL_PUTC_N(const std::string& lhs, IData rhs, CData ths) VL_PURE {
    std::string lstring = lhs;
    const int32_t rhs_s = rhs;  // To signed value
@@ -2587,6 +2608,14 @@ VerilatedVirtualBase* VerilatedContext::threadPoolp() {
    return m_threadPool.get();
 }

+void VerilatedContext::prepareClone() { delete m_threadPool.release(); }
+
+VerilatedVirtualBase* VerilatedContext::threadPoolpOnClone() {
+    if (VL_UNLIKELY(m_threadPool)) m_threadPool.release();
+    m_threadPool = std::unique_ptr<VlThreadPool>(new VlThreadPool{this, m_threads - 1});
+    return m_threadPool.get();
+}
+
 VerilatedVirtualBase*
 VerilatedContext::enableExecutionProfiler(VerilatedVirtualBase* (*construct)(VerilatedContext&)) {
    if (!m_executionProfiler) m_executionProfiler.reset(construct(*this));
@@ -3213,15 +3242,18 @@ void VerilatedAssertOneThread::fatal_different() VL_MT_SAFE {
 //===========================================================================
 // VlDeleter:: Methods

-void VlDeleter::deleteAll() {
+void VlDeleter::deleteAll() VL_EXCLUDES(m_mutex) VL_EXCLUDES(m_deleteMutex) VL_MT_SAFE {
    while (true) {
-        VerilatedLockGuard lock{m_mutex};
-        if (m_newGarbage.empty()) break;
-        VerilatedLockGuard deleteLock{m_deleteMutex};
-        std::swap(m_newGarbage, m_toDelete);
-        lock.unlock();  // So destructors can enqueue new objects
+        {
+            VerilatedLockGuard lock{m_mutex};
+            if (m_newGarbage.empty()) break;
+            m_deleteMutex.lock();
+            std::swap(m_newGarbage, m_toDelete);
+            // m_mutex is unlocked here, so destructors can enqueue new objects
+        }
        for (VlDeletable* const objp : m_toDelete) delete objp;
        m_toDelete.clear();
+        m_deleteMutex.unlock();
    }
 }

@@ -175,6 +175,7 @@ public:
    /// Construct mutex (without locking it)
    VerilatedMutex() = default;
    ~VerilatedMutex() = default;
+    VL_UNCOPYABLE(VerilatedMutex);
    const VerilatedMutex& operator!() const { return *this; }  // For -fthread_safety
    /// Acquire/lock mutex
    void lock() VL_ACQUIRE() VL_MT_SAFE {
@@ -218,23 +219,10 @@ public:
    /// Construct and hold given mutex lock until destruction or unlock()
    explicit VerilatedLockGuard(VerilatedMutex& mutexr) VL_ACQUIRE(mutexr) VL_MT_SAFE
        : m_mutexr(mutexr) {  // Need () or GCC 4.8 false warning
-        m_mutexr.lock();
+        mutexr.lock();
    }
    /// Destruct and unlock the mutex
    ~VerilatedLockGuard() VL_RELEASE() { m_mutexr.unlock(); }
-    /// Lock the mutex
-    void lock() VL_ACQUIRE() VL_MT_SAFE { m_mutexr.lock(); }
-    /// Unlock the mutex
-    void unlock() VL_RELEASE() VL_MT_SAFE { m_mutexr.unlock(); }
-    /// 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 {
-        m_mutexr.lockCheckStopRequest(checkStopRequestFunction);
-    }
 };

 // Internals: Remember the calling thread at construction time, and make
@@ -580,6 +568,8 @@ public:
    void addModel(VerilatedModel*);

    VerilatedVirtualBase* threadPoolp();
+    void prepareClone();
+    VerilatedVirtualBase* threadPoolpOnClone();
    VerilatedVirtualBase*
    enableExecutionProfiler(VerilatedVirtualBase* (*construct)(VerilatedContext&));

@@ -31,6 +31,8 @@ CFG_CXXFLAGS_NO_UNUSED = @CFG_CXXFLAGS_NO_UNUSED@
 CFG_CXXFLAGS_WEXTRA = @CFG_CXXFLAGS_WEXTRA@
 # Compiler flags that enable coroutine support
 CFG_CXXFLAGS_COROUTINES = @CFG_CXXFLAGS_COROUTINES@
+# Linker flags
+CFG_LDFLAGS_VERILATED = @CFG_LDFLAGS_VERILATED@
 # Linker libraries for multithreading
 CFG_LDLIBS_THREADS = @CFG_LDLIBS_THREADS@

@@ -77,6 +79,8 @@ VPATH += ..
 VPATH += $(VERILATOR_ROOT)/include
 VPATH += $(VERILATOR_ROOT)/include/vltstd

+LDFLAGS  += $(CFG_LDFLAGS_VERILATED)
+
 #OPT = -ggdb -DPRINTINITSTR -DDETECTCHANGE
 #OPT = -ggdb -DPRINTINITSTR
 CPPFLAGS += $(OPT)
@@ -210,8 +210,8 @@ private:

        std::string result = prefix + "*" + suffix;

-        // cout << "\nch pre="<<prefix<<"  s="<<suffix<<"\nch a="<<old<<"\nch b="<<add
-        // <<"\ncho="<<result<<endl;
+        // std::cout << "\nch pre=" << prefix << "  s=" << suffix << "\nch a="
+        // << old << "\nch b=" << add << "\ncho=" << result << std::endl;
        return result;
    }
    bool itemMatchesString(VerilatedCovImpItem* itemp, const std::string& match)
@@ -316,7 +316,7 @@ public:
        valps[2] = page_default.c_str();

        // Keys -> strings
-        std::string keys[VerilatedCovConst::MAX_KEYS];
+        std::array<std::string, VerilatedCovConst::MAX_KEYS> keys;
        for (int i = 0; i < VerilatedCovConst::MAX_KEYS; ++i) {
            if (ckeyps[i] && ckeyps[i][0]) keys[i] = ckeyps[i];
        }
@@ -337,7 +337,7 @@ public:
            const std::string key = keys[i];
            if (!keys[i].empty()) {
                const std::string val = valps[i];
-                // cout<<"   "<<__FUNCTION__<<"  "<<key<<" = "<<val<<endl;
+                // std::cout << "   " << __FUNCTION__ << "  " << key << " = " << val << std::endl;
                m_insertp->m_keys[addKeynum] = valueIndex(key);
                m_insertp->m_vals[addKeynum] = valueIndex(val);
                ++addKeynum;
@@ -152,9 +152,11 @@ extern const char* vl_mc_scan_plusargs(const char* prefixp) VL_MT_SAFE;  // PLIi
 // Base macros

 // Return true if data[bit] set; not 0/1 return, but 0/non-zero return.
+// Arguments must not have side effects
 #define VL_BITISSETLIMIT_W(data, width, bit) (((bit) < (width)) && VL_BITISSET_W(data, bit))

 // Shift appropriate word by bit. Does not account for wrapping between two words
+// Argument 'bit' must not have side effects
 #define VL_BITRSHIFT_W(data, bit) ((data)[VL_BITWORD_E(bit)] >> VL_BITBIT_E(bit))

 // Create two 32-bit words from quadword
@@ -758,6 +760,7 @@ static inline IData VL_ONEHOT0_W(int words, WDataInP const lwp) VL_PURE {

 static inline IData VL_CLOG2_I(IData lhs) VL_PURE {
    // There are faster algorithms, or fls GCC4 builtins, but rarely used
+    // In C++20 there will be std::bit_width(lhs) - 1
    if (VL_UNLIKELY(!lhs)) return 0;
    --lhs;
    int shifts = 0;
@@ -953,11 +956,19 @@ static inline void VL_NEGATE_INPLACE_W(int words, WDataOutP owp_lwp) VL_MT_SAFE
 // EMIT_RULE: VL_MUL:    oclean=dirty; lclean==clean; rclean==clean;
 // EMIT_RULE: VL_DIV:    oclean=dirty; lclean==clean; rclean==clean;
 // EMIT_RULE: VL_MODDIV: oclean=dirty; lclean==clean; rclean==clean;
-#define VL_DIV_III(lbits, lhs, rhs) (((rhs) == 0) ? 0 : (lhs) / (rhs))
-#define VL_DIV_QQQ(lbits, lhs, rhs) (((rhs) == 0) ? 0 : (lhs) / (rhs))
+static inline IData VL_DIV_III(int lbits, IData lhs, IData rhs) {
+    return (rhs == 0) ? 0 : lhs / rhs;
+}
+static inline QData VL_DIV_QQQ(int lbits, QData lhs, QData rhs) {
+    return (rhs == 0) ? 0 : lhs / rhs;
+}
 #define VL_DIV_WWW(lbits, owp, lwp, rwp) (_vl_moddiv_w(lbits, owp, lwp, rwp, 0))
-#define VL_MODDIV_III(lbits, lhs, rhs) (((rhs) == 0) ? 0 : (lhs) % (rhs))
-#define VL_MODDIV_QQQ(lbits, lhs, rhs) (((rhs) == 0) ? 0 : (lhs) % (rhs))
+static inline IData VL_MODDIV_III(int lbits, IData lhs, IData rhs) {
+    return (rhs == 0) ? 0 : lhs % rhs;
+}
+static inline QData VL_MODDIV_QQQ(int lbits, QData lhs, QData rhs) {
+    return (rhs == 0) ? 0 : lhs % rhs;
+}
 #define VL_MODDIV_WWW(lbits, owp, lwp, rwp) (_vl_moddiv_w(lbits, owp, lwp, rwp, 1))

 static inline WDataOutP VL_ADD_W(int words, WDataOutP owp, WDataInP const lwp,
@@ -1515,6 +1526,28 @@ 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);
+}
+
+template <typename T>
+static inline IData VL_DYN_TO_I(const VlQueue<T>& q, int elem_size) {
+    IData ret = 0;
+    for (int i = 0; i < q.size(); ++i) ret |= q.at(i) << (i * elem_size);
+    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;
+}
+
 // 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.
@@ -2185,6 +2218,7 @@ extern IData VL_DIST_UNIFORM(IData& seedr, IData ustart, IData uend) VL_MT_SAFE;
 // Conversion functions

 extern std::string VL_CVT_PACK_STR_NW(int lwords, const WDataInP lwp) VL_PURE;
+extern std::string VL_CVT_PACK_STR_ND(const VlQueue<std::string>& q) VL_PURE;
 inline std::string VL_CVT_PACK_STR_NQ(QData lhs) VL_PURE {
    VlWide<VL_WQ_WORDS_E> lw;
    VL_SET_WQ(lw, lhs);
@@ -1,31 +0,0 @@
-// -*- mode: C++; c-file-style: "cc-mode" -*-
-//*************************************************************************
-//
-// Code available from: https://verilator.org
-//
-// Copyright 2010-2023 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 old string and data-type header
-///
-/// This file is deprecated, and provided for backwards compatibility.
-/// Include verilated.h instead.
-///
-//*************************************************************************
-
-#ifndef VERILATOR_VERILATED_HEAVY_H_
-#define VERILATOR_VERILATED_HEAVY_H_
-
-#ifdef VL_NO_LEGACY
-#error "Include <verilated.h> instead of <verilated_heavy.h>"
-#endif
-
-#include "verilated.h"
-
-#endif  // Guard
@@ -26,10 +26,6 @@
 // verilator lint_off TIMESCALEMOD
 // verilator lint_off UNUSEDSIGNAL
 package std;
-   // The process class is not implemented, but it's predeclared here,
-   // so the linter accepts references to it.
-   typedef class process;
-
   class mailbox #(type T);
      protected int m_bound;
      protected T m_queue[$];
@@ -45,7 +41,7 @@ package std;
      task put(T message);
 `ifdef VERILATOR_TIMING
         if (m_bound != 0)
-           wait (m_queue.size() < m_bound);
+            wait (m_queue.size() < m_bound);
         m_queue.push_back(message);
 `endif
      endtask
@@ -117,43 +113,80 @@ package std;
   endclass

   class process;
-      typedef enum { FINISHED, RUNNING, WAITING, SUSPENDED, KILLED } state;
-      static process _s_global_process;
+      typedef enum {
+         FINISHED  = 0,
+         RUNNING   = 1,
+         WAITING   = 2,
+         SUSPENDED = 3,
+         KILLED    = 4
+      } state;
+
+`ifdef VERILATOR_TIMING
+      // Width visitor changes it to VlProcessRef
+      protected chandle m_process;
+`endif
+
      static function process self();
-         // Unsupported, emulating with single process' state
-         if (!_s_global_process) _s_global_process = new;
-         return _s_global_process;
+         process p = new;
+`ifdef VERILATOR_TIMING
+         $c(p.m_process, " = vlProcess;");
+`endif
+         return p;
      endfunction
+
+      protected function void set_status(state s);
+`ifdef VERILATOR_TIMING
+         $c(m_process, "->state(", s, ");");
+`endif
+      endfunction
+
      function state status();
-         // Unsupported, emulating with single process' state
+`ifdef VERILATOR_TIMING
+         return state'($c(m_process, "->state()"));
+`else
         return RUNNING;
+`endif
      endfunction
+
      function void kill();
-         $error("std::process::kill() not supported");
+         set_status(KILLED);
      endfunction
-      task await();
-         $error("std::process::await() not supported");
-      endtask
+
      function void suspend();
         $error("std::process::suspend() not supported");
      endfunction
+
      function void resume();
-         $error("std::process::resume() not supported");
+         set_status(RUNNING);
      endfunction
-      // When really implemented, srandom must operates on the process, but for
+
+      task await();
+`ifdef VERILATOR_TIMING
+         wait (status() == FINISHED || status() == KILLED);
+`endif
+      endtask
+
+      // When really implemented, srandom must operate on the process, but for
      // now rely on the srandom() that is automatically generated for all
      // classes.
+      //
      // function void srandom(int seed);
      // endfunction
+
+      // The methods below work only if set_randstate is never applied to
+      // a state string created before another such string. Full support
+      // could use VlRNG class to store the state per process in VlProcess
+      // objects.
      function string get_randstate();
-         // Could operate on all proceses for now
-         // No error, as harmless until set_randstate is called
-         return "NOT_SUPPORTED";
+         string s;
+
+         s.itoa($random);  // Get a random number
+         set_randstate(s);  // Pretend it's the state of RNG
+         return s;
      endfunction
-      function void set_randstate(string randstate);
-         $error("std::process::set_randstate() not supported");
-         // Could operate on all proceses for now
+
+      function void set_randstate(string s);
+         $urandom(s.atoi());  // Set the seed using a string
      endfunction
   endclass
-
 endpackage
@@ -173,7 +173,7 @@ public:
        VerilatedLockGuard lock{m_mutex};
        while (m_ready.empty()) {
            m_waiting = true;
-            m_cv.wait(lock);
+            m_cv.wait(m_mutex);
        }
        m_waiting = false;
        // As noted above this is inefficient if our ready list is ever
@@ -30,7 +30,16 @@ void VlCoroutineHandle::resume() {
    // main process
    if (VL_LIKELY(m_coro)) {
        VL_DEBUG_IF(VL_DBG_MSGF("             Resuming: "); dump(););
-        m_coro();
+        if (m_process) {  // If process state is managed with std::process
+            if (m_process->state() == VlProcess::KILLED) {
+                m_coro.destroy();
+            } else {
+                m_process->state(VlProcess::RUNNING);
+                m_coro();
+            }
+        } else {
+            m_coro();
+        }
        m_coro = nullptr;
    }
 }
@@ -95,8 +104,9 @@ void VlTriggerScheduler::resume(const char* eventDescription) {
    VL_DEBUG_IF(dump(eventDescription);
                VL_DBG_MSGF("         Resuming processes waiting for %s\n", eventDescription););
 #endif
-    for (auto& susp : m_ready) susp.resume();
-    m_ready.clear();
+    std::swap(m_ready, m_resumeQueue);
+    for (VlCoroutineHandle& coro : m_resumeQueue) coro.resume();
+    m_resumeQueue.clear();
    commit(eventDescription);
 }

@@ -96,33 +96,52 @@ class VlCoroutineHandle final {

    // MEMBERS
    std::coroutine_handle<> m_coro;  // The wrapped coroutine handle
+    VlProcessRef m_process;  // Data of the suspended process, null if not needed
    VlFileLineDebug m_fileline;

 public:
    // CONSTRUCTORS
    // Construct
-    VlCoroutineHandle()
-        : m_coro{nullptr} {}
-    VlCoroutineHandle(std::coroutine_handle<> coro, VlFileLineDebug fileline)
+    // non-explicit:
+    // cppcheck-suppress noExplicitConstructor
+    VlCoroutineHandle(VlProcessRef process)
+        : m_coro{nullptr}
+        , m_process{process} {
+        if (m_process) m_process->state(VlProcess::WAITING);
+    }
+    VlCoroutineHandle(std::coroutine_handle<> coro, VlProcessRef process, VlFileLineDebug fileline)
        : m_coro{coro}
-        , m_fileline{fileline} {}
+        , m_process{process}
+        , m_fileline{fileline} {
+        if (m_process) m_process->state(VlProcess::WAITING);
+    }
    // Move the handle, leaving a nullptr
+    // non-explicit:
+    // cppcheck-suppress noExplicitConstructor
    VlCoroutineHandle(VlCoroutineHandle&& moved)
        : m_coro{std::exchange(moved.m_coro, nullptr)}
+        , m_process{std::exchange(moved.m_process, nullptr)}
        , m_fileline{moved.m_fileline} {}
    // Destroy if the handle isn't null
    ~VlCoroutineHandle() {
        // Usually these coroutines should get resumed; we only need to clean up if we destroy a
        // model with some coroutines suspended
-        if (VL_UNLIKELY(m_coro)) m_coro.destroy();
+        if (VL_UNLIKELY(m_coro)) {
+            m_coro.destroy();
+            if (m_process && m_process->state() != VlProcess::KILLED) {
+                m_process->state(VlProcess::FINISHED);
+            }
+        }
    }
    // METHODS
    // Move the handle, leaving a null handle
    auto& operator=(VlCoroutineHandle&& moved) {
        m_coro = std::exchange(moved.m_coro, nullptr);
+        m_process = std::exchange(moved.m_process, nullptr);
+        m_fileline = moved.m_fileline;
        return *this;
    }
-    // Resume the coroutine if the handle isn't null
+    // Resume the coroutine if the handle isn't null and the process isn't killed
    void resume();
 #ifdef VL_DEBUG
    void dump() const;
@@ -173,21 +192,24 @@ public:
    void dump() const;
 #endif
    // Used by coroutines for co_awaiting a certain simulation time
-    auto delay(uint64_t delay, const char* filename = VL_UNKNOWN, int lineno = 0) {
+    auto delay(uint64_t delay, VlProcessRef process, const char* filename = VL_UNKNOWN,
+               int lineno = 0) {
        struct Awaitable {
+            VlProcessRef process;  // Data of the suspended process, null if not needed
            VlDelayedCoroutineQueue& queue;
            uint64_t delay;
            VlFileLineDebug fileline;

            bool await_ready() const { return false; }  // Always suspend
            void await_suspend(std::coroutine_handle<> coro) {
-                queue.push_back({delay, VlCoroutineHandle{coro, fileline}});
+                queue.push_back({delay, VlCoroutineHandle{coro, process, fileline}});
                // Move last element to the proper place in the max-heap
                std::push_heap(queue.begin(), queue.end());
            }
            void await_resume() const {}
        };
-        return Awaitable{m_queue, m_context.time() + delay, VlFileLineDebug{filename, lineno}};
+        return Awaitable{process, m_queue, m_context.time() + delay,
+                         VlFileLineDebug{filename, lineno}};
    }
 };

@@ -207,6 +229,10 @@ class VlTriggerScheduler final {
                                   // (not resumable)
    VlCoroutineVec m_ready;  // Coroutines that can be resumed (all coros from m_uncommitted are
                             // moved here in commit())
+    VlCoroutineVec m_resumeQueue;  // Coroutines being resumed by resume(); kept as a field to
+                                   // avoid reallocation. Resumed coroutines are moved to
+                                   // m_resumeQueue to allow adding coroutines to m_ready
+                                   // during resume(). Outside of resume() should always be empty.

 public:
    // METHODS
@@ -220,21 +246,23 @@ public:
    void dump(const char* eventDescription) const;
 #endif
    // Used by coroutines for co_awaiting a certain trigger
-    auto trigger(const char* eventDescription = VL_UNKNOWN, const char* filename = VL_UNKNOWN,
-                 int lineno = 0) {
+    auto trigger(bool commit, VlProcessRef process, const char* eventDescription = VL_UNKNOWN,
+                 const char* filename = VL_UNKNOWN, int lineno = 0) {
        VL_DEBUG_IF(VL_DBG_MSGF("         Suspending process waiting for %s at %s:%d\n",
                                eventDescription, filename, lineno););
        struct Awaitable {
            VlCoroutineVec& suspended;  // Coros waiting on trigger
+            VlProcessRef process;  // Data of the suspended process, null if not needed
            VlFileLineDebug fileline;

            bool await_ready() const { return false; }  // Always suspend
            void await_suspend(std::coroutine_handle<> coro) {
-                suspended.emplace_back(coro, fileline);
+                suspended.emplace_back(coro, process, fileline);
            }
            void await_resume() const {}
        };
-        return Awaitable{m_uncommitted, VlFileLineDebug{filename, lineno}};
+        return Awaitable{commit ? m_ready : m_uncommitted, process,
+                         VlFileLineDebug{filename, lineno}};
    }
 };

@@ -267,18 +295,19 @@ class VlDynamicTriggerScheduler final {
                            // with destructive post updates, e.g. named events)

    // METHODS
-    auto awaitable(VlCoroutineVec& queue, const char* filename, int lineno) {
+    auto awaitable(VlProcessRef process, VlCoroutineVec& queue, const char* filename, int lineno) {
        struct Awaitable {
+            VlProcessRef process;  // Data of the suspended process, null if not needed
            VlCoroutineVec& suspended;  // Coros waiting on trigger
            VlFileLineDebug fileline;

            bool await_ready() const { return false; }  // Always suspend
            void await_suspend(std::coroutine_handle<> coro) {
-                suspended.emplace_back(coro, fileline);
+                suspended.emplace_back(coro, process, fileline);
            }
            void await_resume() const {}
        };
-        return Awaitable{queue, VlFileLineDebug{filename, lineno}};
+        return Awaitable{process, queue, VlFileLineDebug{filename, lineno}};
    }

 public:
@@ -292,23 +321,26 @@ public:
    void dump() const;
 #endif
    // Used by coroutines for co_awaiting trigger evaluation
-    auto evaluation(const char* eventDescription, const char* filename, int lineno) {
+    auto evaluation(VlProcessRef process, const char* eventDescription, const char* filename,
+                    int lineno) {
        VL_DEBUG_IF(VL_DBG_MSGF("         Suspending process waiting for %s at %s:%d\n",
                                eventDescription, filename, lineno););
-        return awaitable(m_suspended, filename, lineno);
+        return awaitable(process, m_suspended, filename, lineno);
    }
    // Used by coroutines for co_awaiting the trigger post update step
-    auto postUpdate(const char* eventDescription, const char* filename, int lineno) {
+    auto postUpdate(VlProcessRef process, const char* eventDescription, const char* filename,
+                    int lineno) {
        VL_DEBUG_IF(
            VL_DBG_MSGF("         Process waiting for %s at %s:%d awaiting the post update step\n",
                        eventDescription, filename, lineno););
-        return awaitable(m_post, filename, lineno);
+        return awaitable(process, m_post, filename, lineno);
    }
    // Used by coroutines for co_awaiting the resumption step (in 'act' eval)
-    auto resumption(const char* eventDescription, const char* filename, int lineno) {
+    auto resumption(VlProcessRef process, const char* eventDescription, const char* filename,
+                    int lineno) {
        VL_DEBUG_IF(VL_DBG_MSGF("         Process waiting for %s at %s:%d awaiting resumption\n",
                                eventDescription, filename, lineno););
-        return awaitable(m_triggered, filename, lineno);
+        return awaitable(process, m_triggered, filename, lineno);
    }
 };

@@ -338,24 +370,27 @@ class VlForkSync final {

 public:
    // Create the join object and set the counter to the specified number
-    void init(size_t count) { m_join.reset(new VlJoin{count, {}}); }
+    void init(size_t count, VlProcessRef process) { m_join.reset(new VlJoin{count, {process}}); }
    // Called whenever any of the forked processes finishes. If the join counter reaches 0, the
    // main process gets resumed
    void done(const char* filename = VL_UNKNOWN, int lineno = 0);
    // Used by coroutines for co_awaiting a join
-    auto join(const char* filename = VL_UNKNOWN, int lineno = 0) {
+    auto join(VlProcessRef process, const char* filename = VL_UNKNOWN, int lineno = 0) {
        assert(m_join);
        VL_DEBUG_IF(
            VL_DBG_MSGF("             Awaiting join of fork at: %s:%d\n", filename, lineno););
        struct Awaitable {
+            VlProcessRef process;  // Data of the suspended process, null if not needed
            const std::shared_ptr<VlJoin> join;  // Join to await on
            VlFileLineDebug fileline;

            bool await_ready() { return join->m_counter == 0; }  // Suspend if join still exists
-            void await_suspend(std::coroutine_handle<> coro) { join->m_susp = {coro, fileline}; }
+            void await_suspend(std::coroutine_handle<> coro) {
+                join->m_susp = {coro, process, fileline};
+            }
            void await_resume() const {}
        };
-        return Awaitable{m_join, VlFileLineDebug{filename, lineno}};
+        return Awaitable{process, m_join, VlFileLineDebug{filename, lineno}};
    }
 };

@@ -75,7 +75,7 @@ public:
    // Get an element from the front of the queue. Blocks if none available
    T get() VL_MT_SAFE_EXCLUDES(m_mutex) {
        VerilatedLockGuard lock{m_mutex};
-        m_cv.wait(lock, [this]() VL_REQUIRES(m_mutex) { return !m_queue.empty(); });
+        m_cv.wait(m_mutex, [this]() VL_REQUIRES(m_mutex) { return !m_queue.empty(); });
        assert(!m_queue.empty());
        T value = m_queue.front();
        m_queue.pop_front();
@@ -447,7 +447,9 @@ public:
        if (VL_UNLIKELY(diff)) fullIData(oldp, newval, bits);
    }
    VL_ATTR_ALWINLINE void chgQData(uint32_t* oldp, QData newval, int bits) {
-        const uint64_t diff = *reinterpret_cast<QData*>(oldp) ^ newval;
+        QData old;
+        std::memcpy(&old, oldp, sizeof(old));
+        const uint64_t diff = old ^ newval;
        if (VL_UNLIKELY(diff)) fullQData(oldp, newval, bits);
    }
    VL_ATTR_ALWINLINE void chgWData(uint32_t* oldp, const WData* newvalp, int bits) {
@@ -460,8 +462,9 @@ public:
    }
    VL_ATTR_ALWINLINE void chgEvent(uint32_t* oldp, VlEvent newval) { fullEvent(oldp, newval); }
    VL_ATTR_ALWINLINE void chgDouble(uint32_t* oldp, double newval) {
-        // cppcheck-suppress invalidPointerCast
-        if (VL_UNLIKELY(*reinterpret_cast<double*>(oldp) != newval)) fullDouble(oldp, newval);
+        double old;
+        std::memcpy(&old, oldp, sizeof(old));
+        if (VL_UNLIKELY(old != newval)) fullDouble(oldp, newval);
    }
 };

@@ -482,7 +482,7 @@ VL_ATTR_NOINLINE void VerilatedTrace<VL_SUB_T, VL_BUF_T>::ParallelWorkerData::wa
    // We have been spinning for a while, so yield the thread
    VerilatedLockGuard lock{m_mutex};
    m_waiting = true;
-    m_cv.wait(lock, [this] { return m_ready.load(std::memory_order_relaxed); });
+    m_cv.wait(m_mutex, [this] { return m_ready.load(std::memory_order_relaxed); });
    m_waiting = false;
 }

@@ -872,7 +872,7 @@ void VerilatedTraceBuffer<VL_BUF_T>::fullIData(uint32_t* oldp, IData newval, int
 template <>
 void VerilatedTraceBuffer<VL_BUF_T>::fullQData(uint32_t* oldp, QData newval, int bits) {
    const uint32_t code = oldp - m_sigs_oldvalp;
-    *reinterpret_cast<QData*>(oldp) = newval;
+    std::memcpy(oldp, &newval, sizeof(newval));
    if (VL_UNLIKELY(m_sigs_enabledp && !(VL_BITISSET_W(m_sigs_enabledp, code)))) return;
    emitQData(code, newval, bits);
 }
@@ -888,7 +888,7 @@ void VerilatedTraceBuffer<VL_BUF_T>::fullWData(uint32_t* oldp, const WData* newv
 template <>
 void VerilatedTraceBuffer<VL_BUF_T>::fullDouble(uint32_t* oldp, double newval) {
    const uint32_t code = oldp - m_sigs_oldvalp;
-    *reinterpret_cast<double*>(oldp) = newval;
+    std::memcpy(oldp, &newval, sizeof(newval));
    if (VL_UNLIKELY(m_sigs_enabledp && !(VL_BITISSET_W(m_sigs_enabledp, code)))) return;
    // cppcheck-suppress invalidPointerCast
    emitDouble(code, newval);
@@ -73,6 +73,36 @@ extern std::string VL_TO_STRING_W(int words, const WDataInP obj);
 #define VL_OUT(name, msb, lsb) IData name  ///< Declare output signal, 17-32 bits
 #define VL_OUTW(name, msb, lsb, words) VlWide<words> name  ///< Declare output signal, 65+ bits

+//===================================================================
+// VlProcess stores metadata of running processes
+
+class VlProcess final {
+    // MEMBERS
+    int m_state;  // Current state of the process
+
+public:
+    // TYPES
+    enum : int {  // Type int for compatibility with $c
+        FINISHED = 0,
+        RUNNING = 1,
+        WAITING = 2,
+        SUSPENDED = 3,
+        KILLED = 4,
+    };
+
+    // CONSTRUCTORS
+    VlProcess()
+        : m_state{RUNNING} {}
+
+    // METHODS
+    int state() { return m_state; }
+    void state(int s) { m_state = s; }
+};
+
+using VlProcessRef = std::shared_ptr<VlProcess>;
+
+inline std::string VL_TO_STRING(const VlProcessRef& p) { return std::string("process"); }
+
 //===================================================================
 // Activity trigger vector

@@ -161,8 +191,8 @@ public:
    VlRNG() VL_MT_SAFE;
    explicit VlRNG(uint64_t seed0) VL_MT_SAFE : m_state{0x12341234UL, seed0} {}
    void srandom(uint64_t n) VL_MT_UNSAFE;
-    // Unused: std::string get_randstate() const VL_MT_UNSAFE;
-    // Unused: void set_randstate(const std::string& state) VL_MT_UNSAFE;
+    std::string get_randstate() const VL_MT_UNSAFE;
+    void set_randstate(const std::string& state) VL_MT_UNSAFE;
    uint64_t rand64() VL_MT_UNSAFE;
    // Threadsafe, but requires use on vl_thread_rng
    static uint64_t vl_thread_rng_rand64() VL_MT_SAFE;
@@ -487,7 +517,7 @@ public:
    template <typename Func>
    VlQueue unique(Func with_func) const {
        VlQueue out;
-        std::set<T_Value> saw;
+        std::set<decltype(with_func(0, m_deque[0]))> saw;
        for (const auto& i : m_deque) {
            const auto i_mapped = with_func(0, i);
            const auto it = saw.find(i_mapped);
@@ -516,7 +546,7 @@ public:
    VlQueue<IData> unique_index(Func with_func) const {
        VlQueue<IData> out;
        IData index = 0;
-        std::unordered_set<T_Value> saw;
+        std::set<decltype(with_func(0, m_deque[0]))> saw;
        for (const auto& i : m_deque) {
            const auto i_mapped = with_func(index, i);
            auto it = saw.find(i_mapped);
@@ -830,6 +860,22 @@ public:
        }
        return out;
    }
+    template <typename Func>
+    VlQueue<T_Value> unique(Func with_func) const {
+        VlQueue<T_Value> out;
+        T_Key default_key;
+        using WithType = decltype(with_func(m_map.begin()->first, m_map.begin()->second));
+        std::set<WithType> saw;
+        for (const auto& i : m_map) {
+            const auto i_mapped = with_func(default_key, i.second);
+            const auto it = saw.find(i_mapped);
+            if (it == saw.end()) {
+                saw.insert(it, i_mapped);
+                out.push_back(i.second);
+            }
+        }
+        return out;
+    }
    VlQueue<T_Key> unique_index() const {
        VlQueue<T_Key> out;
        std::set<T_Key> saw;
@@ -843,6 +889,21 @@ public:
        return out;
    }
    template <typename Func>
+    VlQueue<T_Key> unique_index(Func with_func) const {
+        VlQueue<T_Key> out;
+        using WithType = decltype(with_func(m_map.begin()->first, m_map.begin()->second));
+        std::set<WithType> saw;
+        for (const auto& i : m_map) {
+            const auto i_mapped = with_func(i.first, i.second);
+            auto it = saw.find(i_mapped);
+            if (it == saw.end()) {
+                saw.insert(it, i_mapped);
+                out.push_back(i.first);
+            }
+        }
+        return out;
+    }
+    template <typename Func>
    VlQueue<T_Value> find(Func with_func) const {
        VlQueue<T_Value> out;
        for (const auto& i : m_map)
@@ -903,6 +964,16 @@ public:
            });
        return VlQueue<T_Value>::cons(it->second);
    }
+    template <typename Func>
+    VlQueue<T_Value> min(Func with_func) const {
+        if (m_map.empty()) return VlQueue<T_Value>();
+        const auto it = std::min_element(
+            m_map.begin(), m_map.end(),
+            [&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);
+    }
    VlQueue<T_Value> max() const {
        if (m_map.empty()) return VlQueue<T_Value>();
        const auto it = std::max_element(
@@ -912,6 +983,16 @@ public:
            });
        return VlQueue<T_Value>::cons(it->second);
    }
+    template <typename Func>
+    VlQueue<T_Value> max(Func with_func) const {
+        if (m_map.empty()) return VlQueue<T_Value>();
+        const auto it = std::max_element(
+            m_map.begin(), m_map.end(),
+            [&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);
+    }

    T_Value r_sum() const {
        T_Value out(0);  // Type must have assignment operator
@@ -1040,9 +1121,15 @@ void VL_WRITEMEM_N(bool hex, int bits, const std::string& filename,

 template <class T_Value, std::size_t T_Depth>
 struct VlUnpacked final {
+private:
+    // TYPES
+    using T_Key = IData;  // Index type, for uniformity with other containers
+    using Unpacked = T_Value[T_Depth];
+
+public:
    // MEMBERS
    // This should be the only data member, otherwise generated static initializers need updating
-    T_Value m_storage[T_Depth];  // Contents of the unpacked array
+    Unpacked m_storage;  // Contents of the unpacked array

    // CONSTRUCTORS
    // Default constructors and destructor are used. Note however that C++20 requires that
@@ -1053,6 +1140,7 @@ struct VlUnpacked final {
    // Default copy assignment operators are used.

    // METHODS
+public:
    // Raw access
    WData* data() { return &m_storage[0]; }
    const WData* data() const { return &m_storage[0]; }
@@ -1065,6 +1153,170 @@ struct VlUnpacked final {
    bool neq(const VlUnpacked<T_Value, T_Depth>& that) const { return neq(*this, that); }
    // 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); }
+
+    void sort() { std::sort(std::begin(m_storage), std::end(m_storage)); }
+    template <typename Func>
+    void sort(Func with_func) {
+        // with_func returns arbitrary type to use for the sort comparison
+        std::sort(std::begin(m_storage), std::end(m_storage),
+                  [=](const T_Value& a, const T_Value& b) {
+                      // index number is meaningless with sort, as it changes
+                      return with_func(0, a) < with_func(0, b);
+                  });
+    }
+    // std::rbegin/std::rend not available until C++14
+    void rsort() {
+        std::sort(std::begin(m_storage), std::end(m_storage), std::greater<T_Value>());
+    }
+    template <typename Func>
+    void rsort(Func with_func) {
+        // with_func returns arbitrary type to use for the sort comparison
+        // std::rbegin/std::rend not available until C++14, so using > below
+        std::sort(std::begin(m_storage), std::end(m_storage),
+                  [=](const T_Value& a, const T_Value& b) {
+                      // index number is meaningless with sort, as it changes
+                      return with_func(0, a) > with_func(0, b);
+                  });
+    }
+    void reverse() { std::reverse(std::begin(m_storage), std::end(m_storage)); }
+    void shuffle() { std::shuffle(std::begin(m_storage), std::end(m_storage), VlURNG{}); }
+    VlQueue<T_Value> unique() const {
+        VlQueue<T_Value> out;
+        std::set<T_Value> saw;
+        for (const auto& i : m_storage) {
+            const auto it = saw.find(i);
+            if (it == saw.end()) {
+                saw.insert(it, i);
+                out.push_back(i);
+            }
+        }
+        return out;
+    }
+    template <typename Func>
+    VlQueue<T_Value> unique(Func with_func) const {
+        VlQueue<T_Value> out;
+        std::set<T_Value> saw;
+        for (const auto& i : m_storage) {
+            const auto i_mapped = with_func(0, i);
+            const auto it = saw.find(i_mapped);
+            if (it == saw.end()) {
+                saw.insert(it, i_mapped);
+                out.push_back(i);
+            }
+        }
+        return out;
+    }
+    VlQueue<T_Key> unique_index() const {
+        VlQueue<T_Key> out;
+        IData index = 0;
+        std::set<T_Value> saw;
+        for (const auto& i : m_storage) {
+            const auto it = saw.find(i);
+            if (it == saw.end()) {
+                saw.insert(it, i);
+                out.push_back(index);
+            }
+            ++index;
+        }
+        return out;
+    }
+    template <typename Func>
+    VlQueue<T_Key> unique_index(Func with_func) const {
+        VlQueue<T_Key> out;
+        IData index = 0;
+        std::unordered_set<T_Value> saw;
+        for (const auto& i : m_storage) {
+            const auto i_mapped = with_func(index, i);
+            auto it = saw.find(i_mapped);
+            if (it == saw.end()) {
+                saw.insert(it, i_mapped);
+                out.push_back(index);
+            }
+            ++index;
+        }
+        return out;
+    }
+    template <typename Func>
+    VlQueue<T_Value> find(Func with_func) const {
+        VlQueue<T_Value> out;
+        IData index = 0;
+        for (const auto& i : m_storage) {
+            if (with_func(index, i)) out.push_back(i);
+            ++index;
+        }
+        return out;
+    }
+    template <typename Func>
+    VlQueue<T_Key> find_index(Func with_func) const {
+        VlQueue<T_Key> out;
+        IData index = 0;
+        for (const auto& i : m_storage) {
+            if (with_func(index, i)) out.push_back(index);
+            ++index;
+        }
+        return out;
+    }
+    template <typename Func>
+    VlQueue<T_Value> find_first(Func with_func) const {
+        // 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);
+            ++index;
+        }
+        return VlQueue<T_Value>{};
+    }
+    template <typename Func>
+    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);
+            ++index;
+        }
+        return VlQueue<T_Key>{};
+    }
+    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]);
+        }
+        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);
+        }
+        return VlQueue<T_Key>{};
+    }
+
+    // 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);
+    }
+    template <typename Func>
+    VlQueue<T_Value> min(Func with_func) const {
+        const auto it = std::min_element(std::begin(m_storage), std::end(m_storage),
+                                         [&with_func](const IData& a, const IData& b) {
+                                             return with_func(0, a) < with_func(0, b);
+                                         });
+        return VlQueue<T_Value>::cons(*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);
+    }
+    template <typename Func>
+    VlQueue<T_Value> max(Func with_func) const {
+        const auto it = std::max_element(std::begin(m_storage), std::end(m_storage),
+                                         [&with_func](const IData& a, const IData& b) {
+                                             return with_func(0, a) < with_func(0, b);
+                                         });
+        return VlQueue<T_Value>::cons(*it);
+    }

    // Dumping. Verilog: str = $sformatf("%p", assoc)
    std::string to_string() const {
@@ -1138,7 +1390,7 @@ public:
    }

    // Deletes all queued garbage objects.
-    void deleteAll() VL_MT_SAFE;
+    void deleteAll() VL_EXCLUDES(m_mutex) VL_EXCLUDES(m_deleteMutex) VL_MT_SAFE;
 };

 //===================================================================
@@ -1286,8 +1538,14 @@ public:
    // For 'if (ptr)...'
    operator bool() const { return m_objp; }
    // In SV A == B iff both are handles to the same object (IEEE 1800-2017 8.4)
-    bool operator==(const VlClassRef& rhs) const { return m_objp == rhs.m_objp; };
-    bool operator!=(const VlClassRef& rhs) const { return m_objp != rhs.m_objp; };
+    template <typename T_OtherClass>
+    bool operator==(const VlClassRef<T_OtherClass>& rhs) const {
+        return m_objp == rhs.m_objp;
+    };
+    template <typename T_OtherClass>
+    bool operator!=(const VlClassRef<T_OtherClass>& rhs) const {
+        return m_objp != rhs.m_objp;
+    };
 };

 template <typename T, typename U>
@@ -410,28 +410,14 @@ void VerilatedVcd::dumpHeader() {
            if (*np == ' ') np++;
            if (*np == '\t') break;  // tab means signal name starts
            printIndent(1);
-            // Find character after name end
-            const char* sp = np;
-            while (*sp && *sp != ' ' && *sp != '\t' && !(*sp & '\x80')) sp++;
-
-            printStr("$scope ");
-            if (*sp & '\x80') {
-                switch (*sp & 0x7f) {
-                case VLT_TRACE_SCOPE_STRUCT: printStr("struct "); break;
-                case VLT_TRACE_SCOPE_INTERFACE: printStr("interface "); break;
-                case VLT_TRACE_SCOPE_UNION: printStr("union "); break;
-                default: printStr("module ");
-                }
-            } else {
-                printStr("module ");
-            }
+            printStr("$scope module ");

            for (; *np && *np != ' ' && *np != '\t'; np++) {
                if (*np == '[') {
                    printStr("[");
                } else if (*np == ']') {
                    printStr("]");
-                } else if (!(*np & '\x80')) {
+                } else {
                    *m_writep++ = *np;
                }
            }
@@ -165,6 +165,7 @@ class VerilatedVpioVarBase VL_NOT_FINAL : public VerilatedVpio {
 protected:
    const VerilatedVar* m_varp = nullptr;
    const VerilatedScope* m_scopep = nullptr;
+    std::string m_fullname;
    const VerilatedRange& get_range() const {
        // Determine number of dimensions and return outermost
        return (m_varp->dims() > 1) ? m_varp->unpacked() : m_varp->packed();
@@ -173,11 +174,13 @@ protected:
 public:
    VerilatedVpioVarBase(const VerilatedVar* varp, const VerilatedScope* scopep)
        : m_varp{varp}
-        , m_scopep{scopep} {}
+        , m_scopep{scopep}
+        , m_fullname{std::string{m_scopep->name()} + '.' + m_varp->name()} {}
    explicit VerilatedVpioVarBase(const VerilatedVpioVarBase* varp) {
        if (varp) {
            m_varp = varp->m_varp;
            m_scopep = varp->m_scopep;
+            m_fullname = varp->m_fullname;
        }
    }
    static VerilatedVpioVarBase* castp(vpiHandle h) {
@@ -188,11 +191,7 @@ public:
    uint32_t size() const override { return get_range().elements(); }
    const VerilatedRange* rangep() const override { return &get_range(); }
    const char* name() const override { return m_varp->name(); }
-    const char* fullname() const override {
-        static thread_local std::string t_out;
-        t_out = std::string{m_scopep->name()} + "." + name();
-        return t_out.c_str();
-    }
+    const char* fullname() const override { return m_fullname.c_str(); }
 };

 class VerilatedVpioParam final : public VerilatedVpioVarBase {
@@ -646,23 +645,21 @@ public:
                continue;
            }
            VerilatedVpiCbHolder& ho = *it++;
-            if (VerilatedVpioVar* const varop = VerilatedVpioVar::castp(ho.cb_datap()->obj)) {
-                void* const newDatap = varop->varDatap();
-                void* const prevDatap
-                    = varop->prevDatap();  // Was malloced when we added the callback
-                VL_DEBUG_IF_PLI(VL_DBG_MSGF("- vpi: value_test %s v[0]=%d/%d %p %p\n",
-                                            varop->fullname(), *(static_cast<CData*>(newDatap)),
-                                            *(static_cast<CData*>(prevDatap)), newDatap,
-                                            prevDatap););
-                if (std::memcmp(prevDatap, newDatap, varop->entSize()) != 0) {
-                    VL_DEBUG_IF_PLI(VL_DBG_MSGF("- vpi: value_callback %" PRId64 " %s v[0]=%d\n",
-                                                ho.id(), varop->fullname(),
-                                                *(static_cast<CData*>(newDatap))););
-                    update.insert(varop);
-                    vpi_get_value(ho.cb_datap()->obj, ho.cb_datap()->value);
-                    (ho.cb_rtnp())(ho.cb_datap());
-                    called = true;
-                }
+            VerilatedVpioVar* const varop
+                = reinterpret_cast<VerilatedVpioVar*>(ho.cb_datap()->obj);
+            void* const newDatap = varop->varDatap();
+            void* const prevDatap = varop->prevDatap();  // Was malloced when we added the callback
+            VL_DEBUG_IF_PLI(VL_DBG_MSGF("- vpi: value_test %s v[0]=%d/%d %p %p\n",
+                                        varop->fullname(), *(static_cast<CData*>(newDatap)),
+                                        *(static_cast<CData*>(prevDatap)), newDatap, prevDatap););
+            if (std::memcmp(prevDatap, newDatap, varop->entSize()) != 0) {
+                VL_DEBUG_IF_PLI(VL_DBG_MSGF("- vpi: value_callback %" PRId64 " %s v[0]=%d\n",
+                                            ho.id(), varop->fullname(),
+                                            *(static_cast<CData*>(newDatap))););
+                update.insert(varop);
+                vpi_get_value(ho.cb_datap()->obj, ho.cb_datap()->value);
+                (ho.cb_rtnp())(ho.cb_datap());
+                called = true;
            }
            if (was_last) break;
        }
@@ -2310,7 +2307,7 @@ PLI_INT32 vpi_flush(void) {
    VerilatedVpiImp::assertOneCheck();
    VL_VPI_ERROR_RESET_();
    Verilated::runFlushCallbacks();
-    return 0;
+    return 0;  // Gcc coverage bug // LCOV_EXCL_LINE
 }

 PLI_INT32 vpi_mcd_flush(PLI_UINT32 mcd) {
@@ -37,8 +37,13 @@
 //=========================================================================
 // Compiler pragma abstraction

+#if defined(__clang__)
+# define VL_CLANG_ATTR(attr) __attribute__(( attr ))
+#else
+# define VL_CLANG_ATTR(attr)
+#endif
+
 #ifdef __GNUC__
-# define VL_ATTR_ALIGNED(alignment) __attribute__((aligned(alignment)))
 # define VL_ATTR_ALWINLINE __attribute__((always_inline)) inline
 # define VL_ATTR_NOINLINE __attribute__((noinline))
 # define VL_ATTR_COLD __attribute__((cold))
@@ -57,19 +62,6 @@
 // All VL_ATTR_WEAK symbols must be marked with the macOS -U linker flag in verilated.mk.in
 #  define VL_ATTR_WEAK __attribute__((weak))
 # endif
-# if defined(__clang__)
-#  define VL_ACQUIRE(...) __attribute__((annotate("ACQUIRE"))) __attribute__((acquire_capability(__VA_ARGS__)))
-#  define VL_ACQUIRE_SHARED(...) __attribute__((annotate("ACQUIRE_SHARED"))) __attribute__((acquire_shared_capability(__VA_ARGS__)))
-#  define VL_RELEASE(...) __attribute__((annotate("RELEASE"))) __attribute__((release_capability(__VA_ARGS__)))
-#  define VL_RELEASE_SHARED(...) __attribute__((annotate("RELEASE_SHARED"))) __attribute__((release_shared_capability(__VA_ARGS__)))
-#  define VL_TRY_ACQUIRE(...) __attribute__((try_acquire_capability(__VA_ARGS__)))
-#  define VL_TRY_ACQUIRE_SHARED(...) __attribute__((try_acquire_shared_capability(__VA_ARGS__)))
-#  define VL_CAPABILITY(x) __attribute__((capability(x)))
-#  define VL_REQUIRES(x) __attribute__((annotate("REQUIRES"))) __attribute__((requires_capability(x)))
-#  define VL_GUARDED_BY(x) __attribute__((annotate("GUARDED_BY"))) __attribute__((guarded_by(x)))
-#  define VL_EXCLUDES(x) __attribute__((annotate("EXCLUDES"))) __attribute__((locks_excluded(x)))
-#  define VL_SCOPED_CAPABILITY __attribute__((scoped_lockable))
-# endif
 # define VL_LIKELY(x) __builtin_expect(!!(x), 1)  // Prefer over C++20 [[likely]]
 # define VL_UNLIKELY(x) __builtin_expect(!!(x), 0)  // Prefer over C++20 [[unlikely]]
 # define VL_UNREACHABLE __builtin_unreachable()  // C++23 std::unreachable()
@@ -77,10 +69,74 @@
 # define VL_PREFETCH_RW(p) __builtin_prefetch((p), 1)
 #endif

-// Defaults for unsupported compiler features
-#ifndef VL_ATTR_ALIGNED
-# define VL_ATTR_ALIGNED(alignment)  ///< Attribute to align structure to byte alignment
+// Function acquires a capability/lock (-fthread-safety)
+#define VL_ACQUIRE(...) \
+        VL_CLANG_ATTR(annotate("ACQUIRE")) \
+        VL_CLANG_ATTR(acquire_capability(__VA_ARGS__))
+// Function acquires a shared capability/lock (-fthread-safety)
+#define VL_ACQUIRE_SHARED(...) \
+        VL_CLANG_ATTR(annotate("ACQUIRE_SHARED")) \
+        VL_CLANG_ATTR(acquire_shared_capability(__VA_ARGS__))
+// Function releases a capability/lock (-fthread-safety)
+#define VL_RELEASE(...) \
+        VL_CLANG_ATTR(annotate("RELEASE")) \
+        VL_CLANG_ATTR(release_capability(__VA_ARGS__))
+// Function releases a shared capability/lock (-fthread-safety)
+#define VL_RELEASE_SHARED(...) \
+        VL_CLANG_ATTR(annotate("RELEASE_SHARED")) \
+        VL_CLANG_ATTR(release_shared_capability(__VA_ARGS__))
+// Function returns bool if acquired a capability (-fthread-safety)
+#define VL_TRY_ACQUIRE(...) \
+        VL_CLANG_ATTR(try_acquire_capability(__VA_ARGS__))
+// Function returns bool if acquired shared (-fthread-safety)
+#define VL_TRY_ACQUIRE_SHARED(...) \
+        VL_CLANG_ATTR(try_acquire_shared_capability(__VA_ARGS__))
+// Function requires a capability inbound (-fthread-safety)
+#define VL_CAPABILITY(x) \
+        VL_CLANG_ATTR(capability(x))
+// Name of mutex protecting this variable (-fthread-safety)
+#define VL_EXCLUDES(x) \
+        VL_CLANG_ATTR(annotate("EXCLUDES")) \
+        VL_CLANG_ATTR(locks_excluded(x))
+// Scoped threaded capability/lock (-fthread-safety)
+#define VL_SCOPED_CAPABILITY \
+        VL_CLANG_ATTR(scoped_lockable)
+// Annotated function returns reference to the given capability.
+// Allowed on: function, method. (-fthread-safety)
+#define VL_RETURN_CAPABILITY(x) \
+        VL_CLANG_ATTR(lock_returned(x))
+// Assert that capability is already held.
+// Allowed on: function, method. (-fthread-safety)
+#define VL_ASSERT_CAPABILITY(x) \
+        VL_CLANG_ATTR(assert_capability(x))
+
+// Require mutex locks only in code units which work with enabled multi-threading.
+#if !defined(VL_MT_DISABLED_CODE_UNIT)
+// Function requires not having a capability inbound (-fthread-safety)
+# define VL_REQUIRES(x) \
+        VL_CLANG_ATTR(annotate("REQUIRES")) \
+        VL_CLANG_ATTR(requires_capability(x))
+// Name of capability/lock (-fthread-safety)
+# define VL_GUARDED_BY(x) \
+        VL_CLANG_ATTR(annotate("GUARDED_BY")) \
+        VL_CLANG_ATTR(guarded_by(x))
+// The data that the annotated pointer points to is protected by the given capability.
+// The pointer itself is not protected.
+// Allowed on: pointer data member. (-fthread-safety)
+# define VL_PT_GUARDED_BY(x) \
+        VL_CLANG_ATTR(annotate("PT_GUARDED_BY")) \
+        VL_CLANG_ATTR(pt_guarded_by(x))
+#else
+// Keep annotations for clang_check_attributes
+# define VL_REQUIRES(x) \
+        VL_CLANG_ATTR(annotate("REQUIRES"))
+# define VL_GUARDED_BY(x) \
+        VL_CLANG_ATTR(annotate("GUARDED_BY"))
+# define VL_PT_GUARDED_BY(x) \
+        VL_CLANG_ATTR(annotate("PT_GUARDED_BY"))
 #endif
+
+// Defaults for unsupported compiler features
 #ifndef VL_ATTR_ALWINLINE
 # define VL_ATTR_ALWINLINE  ///< Attribute to inline, even when not optimizing
 #endif
@@ -111,19 +167,6 @@
 #ifndef VL_ATTR_WEAK
 # define VL_ATTR_WEAK  ///< Attribute that function external that is optionally defined
 #endif
-#ifndef VL_CAPABILITY
-# define VL_ACQUIRE(...)  ///< Function acquires a capability/lock (-fthread-safety)
-# define VL_ACQUIRE_SHARED(...)  ///< Function acquires a shared capability/lock (-fthread-safety)
-# define VL_RELEASE(...)  ///< Function releases a capability/lock (-fthread-safety)
-# define VL_RELEASE_SHARED(...)  ///< Function releases a shared capability/lock (-fthread-safety)
-# define VL_TRY_ACQUIRE(...)  ///< Function returns bool if acquired a capability (-fthread-safety)
-# define VL_TRY_ACQUIRE_SHARED(...)  ///< Function returns bool if acquired shared (-fthread-safety)
-# define VL_REQUIRES(x)  ///< Function requires a capability inbound (-fthread-safety)
-# define VL_EXCLUDES(x)  ///< Function requires not having a capability inbound (-fthread-safety)
-# define VL_CAPABILITY(x)  ///< Name of capability/lock (-fthread-safety)
-# define VL_GUARDED_BY(x)  ///< Name of mutex protecting this variable (-fthread-safety)
-# define VL_SCOPED_CAPABILITY  ///< Scoped threaded capability/lock (-fthread-safety)
-#endif
 #ifndef VL_LIKELY
 # define VL_LIKELY(x) (!!(x))  ///< Return boolean expression that is more often true
 # define VL_UNLIKELY(x) (!!(x))  ///< Return boolean expression that is more often false
@@ -142,6 +185,7 @@


 #ifndef VL_NO_LEGACY
+# define VL_ATTR_ALIGNED(alignment)  // Deprecated
 # define VL_FUNC __func__  // Deprecated
 # define VL_THREAD  // Deprecated
 # define VL_THREAD_LOCAL thread_local  // Deprecated
@@ -149,56 +193,24 @@
 #endif

 // Comment tag that Function is pure (and thus also VL_MT_SAFE)
-#if defined(__clang__)
-# define VL_PURE __attribute__((annotate("PURE")))
-#else
-# define VL_PURE
-#endif
+#define VL_PURE VL_CLANG_ATTR(annotate("PURE"))
 // Comment tag that function is threadsafe
-#if defined(__clang__)
-# define VL_MT_SAFE __attribute__((annotate("MT_SAFE")))
-#else
-# define VL_MT_SAFE
-#endif
+#define VL_MT_SAFE VL_CLANG_ATTR(annotate("MT_SAFE"))
 // Comment tag that function is threadsafe, only if
 // other threads doesn't change tree topology
-#if defined(__clang__)
-# define VL_MT_STABLE __attribute__((annotate("MT_STABLE")))
-#else
-# define VL_MT_STABLE
-#endif
+#define VL_MT_STABLE VL_CLANG_ATTR(annotate("MT_STABLE"))
 // Comment tag that function is threadsafe, only
 // during normal operation (post-init)
-#if defined(__clang__)
-# define VL_MT_SAFE_POSTINIT __attribute__((annotate("MT_SAFE_POSTINIT")))
-#else
-# define VL_MT_SAFE_POSTINIT
-#endif
+#define VL_MT_SAFE_POSTINIT VL_CLANG_ATTR(annotate("MT_SAFE_POSTINIT"))
 // Attribute that function is clang threadsafe and uses given mutex
-#if defined(__clang__)
-# define VL_MT_SAFE_EXCLUDES(mutex) __attribute__((annotate("MT_SAFE_EXCLUDES"))) VL_EXCLUDES(mutex)
-#else
-# define VL_MT_SAFE_EXCLUDES(mutex) VL_EXCLUDES(mutex)
-#endif
+#define VL_MT_SAFE_EXCLUDES(mutex) VL_CLANG_ATTR(annotate("MT_SAFE_EXCLUDES")) VL_EXCLUDES(mutex)
 // Comment tag that function is not threadsafe
-#if defined(__clang__)
-# define VL_MT_UNSAFE __attribute__((annotate("MT_UNSAFE")))
-#else
-# define VL_MT_UNSAFE
-#endif
+#define VL_MT_UNSAFE VL_CLANG_ATTR(annotate("MT_UNSAFE"))
 // Comment tag that function is not threadsafe
 // protected to make sure single-caller
-#if defined(__clang__)
-# define VL_MT_UNSAFE_ONE __attribute__((annotate("MT_UNSAFE_ONE")))
-#else
-# define VL_MT_UNSAFE_ONE
-#endif
+#define VL_MT_UNSAFE_ONE VL_CLANG_ATTR(annotate("MT_UNSAFE_ONE"))
 // Comment tag that function is entry point of parallelization
-#if defined(__clang__)
-# define VL_MT_START __attribute__((annotate("MT_START")))
-#else
-# define VL_MT_START
-#endif
+#define VL_MT_START VL_CLANG_ATTR(annotate("MT_START"))

 #ifndef VL_NO_LEGACY
 # define VL_ULL(c) (c##ULL)  // Add appropriate suffix to 64-bit constant (deprecated)
@@ -324,6 +336,7 @@ extern "C" void __gcov_dump();
 // Now that C++ requires these standard types the vl types are deprecated
 #include <cstdint>
 #include <cinttypes>
+#include <cmath>

 #ifndef VL_NO_LEGACY
 using vluint8_t = uint8_t;  ///< 8-bit unsigned type (backward compatibility)
@@ -429,6 +442,11 @@ using ssize_t = uint32_t;  ///< signed size_t; returned from read()
    Type(const Type& other) = delete; \
    Type& operator=(const Type&) = delete

+// Declare a class as unmovable; put after a private:
+#define VL_UNMOVABLE(Type) \
+    Type(Type&& other) = delete; \
+    Type& operator=(Type&&) = delete
+
 //=========================================================================
 // Verilated function size macros

@@ -448,11 +466,14 @@ using ssize_t = uint32_t;  ///< signed size_t; returned from read()
 #define VL_SIZEBITS_E (VL_EDATASIZE - 1)  ///< Bit mask for bits in a quad

 /// Return mask for words with 1's where relevant bits are (0=all bits)
+/// Arguments must not have side effects
 #define VL_MASK_I(nbits) (((nbits) & VL_SIZEBITS_I) ? ((1U << ((nbits) & VL_SIZEBITS_I)) - 1) : ~0)
 /// Return mask for quads with 1's where relevant bits are (0=all bits)
+/// Arguments must not have side effects
 #define VL_MASK_Q(nbits) \
    (((nbits) & VL_SIZEBITS_Q) ? ((1ULL << ((nbits) & VL_SIZEBITS_Q)) - 1ULL) : ~0ULL)
 /// Return mask for EData with 1's where relevant bits are (0=all bits)
+/// Arguments must not have side effects
 #define VL_MASK_E(nbits) VL_MASK_I(nbits)

 #define VL_EUL(n) VL_UL(n)  // Make constant number EData sized
@@ -474,8 +495,12 @@ using ssize_t = uint32_t;  ///< signed size_t; returned from read()
 // #defines, to avoid requiring math.h on all compile runs

 #ifdef _MSC_VER
-# define VL_TRUNC(n) (((n) < 0) ? std::ceil((n)) : std::floor((n)))
-# define VL_ROUND(n) (((n) < 0) ? std::ceil((n)-0.5) : std::floor((n) + 0.5))
+static inline double VL_TRUNC(double n) {
+    return (n < 0) ? std::ceil(n) : std::floor(n);
+}
+static inline double VL_ROUND(double n) {
+    return (n < 0) ? std::ceil(n-0.5) : std::floor(n + 0.5);
+}
 #else
 # define VL_TRUNC(n) std::trunc(n)
 # define VL_ROUND(n) std::round(n)
@@ -569,8 +594,8 @@ using ssize_t = uint32_t;  ///< signed size_t; returned from read()
 //=========================================================================
 // Stringify macros

-#define VL_STRINGIFY(x) VL_STRINGIFY2(x)
-#define VL_STRINGIFY2(x) #x
+#define VL_STRINGIFY(...) VL_STRINGIFY2(__VA_ARGS__)
+#define VL_STRINGIFY2(...) #__VA_ARGS__

 //=========================================================================
 // Offset of field in type
@@ -10,16 +10,36 @@ import argparse
 import os
 import sys
 import shlex
-from typing import Callable, Iterable, Optional, Union
+from typing import Callable, Iterable, Optional, Union, TYPE_CHECKING
 import dataclasses
 from dataclasses import dataclass
 import enum
 from enum import Enum
 import multiprocessing
+import re
 import tempfile

 import clang.cindex
-from clang.cindex import CursorKind, Index, TranslationUnitSaveError, TranslationUnitLoadError
+from clang.cindex import (
+    Index,
+    TranslationUnitSaveError,
+    TranslationUnitLoadError,
+    CompilationDatabase,
+)
+
+if not TYPE_CHECKING:
+    from clang.cindex import CursorKind
+else:
+    # Workaround for missing support for members defined out-of-class in Pylance:
+    # https://github.com/microsoft/pylance-release/issues/2365#issuecomment-1035803067
+
+    class CursorKindMeta(type):
+
+        def __getattr__(cls, name: str) -> clang.cindex.CursorKind:
+            return getattr(clang.cindex.CursorKind, name)
+
+    class CursorKind(clang.cindex.CursorKind, metaclass=CursorKindMeta):
+        pass


 def fully_qualified_name(node):
@@ -33,6 +53,26 @@ def fully_qualified_name(node):
    return [node.displayname] if node.displayname else []


+# Returns True, if `class_node` contains node
+# that matches `member` spelling
+def check_class_member_exists(class_node, member):
+    for child in class_node.get_children():
+        if member.spelling == child.spelling:
+            return True
+    return False
+
+
+# Returns Base class (if found) of `class_node`
+# that is of type `base_type`
+def get_base_class(class_node, base_type):
+    for child in class_node.get_children():
+        if child.kind is CursorKind.CXX_BASE_SPECIFIER:
+            base_class = child.type
+            if base_type.spelling == base_class.spelling:
+                return base_class
+    return None
+
+
@dataclass
 class VlAnnotations:
    mt_start: bool = False
@@ -40,6 +80,7 @@ class VlAnnotations:
    stable_tree: bool = False
    mt_safe_postinit: bool = False
    mt_unsafe: bool = False
+    mt_disabled: bool = False
    mt_unsafe_one: bool = False
    pure: bool = False
    guarded: bool = False
@@ -61,7 +102,7 @@ class VlAnnotations:
        return self.stable_tree or self.mt_start

    def is_mt_unsafe_call(self):
-        return self.mt_unsafe or self.mt_unsafe_one
+        return self.mt_unsafe or self.mt_unsafe_one or self.mt_disabled

    def is_mt_safe_call(self):
        return (not self.is_mt_unsafe_call()
@@ -111,6 +152,8 @@ class VlAnnotations:
                    result.mt_unsafe = True
                elif node.displayname == "MT_UNSAFE_ONE":
                    result.mt_unsafe_one = True
+                elif node.displayname == "MT_DISABLED":
+                    result.mt_disabled = True
                elif node.displayname == "PURE":
                    result.pure = True
                elif node.displayname in ["ACQUIRE", "ACQUIRE_SHARED"]:
@@ -183,6 +226,19 @@ class FunctionInfo:
    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,
+                                          annotations: VlAnnotations):
+        file = os.path.abspath(file)
+        refd = refd.canonical
+        assert refd is not None
+        name_parts = fully_qualified_name(refd)
+        usr = refd.get_usr()
+        ftype = FunctionType.from_node(refd)
+
+        return FunctionInfo(name_parts, usr, file, line, annotations, ftype)
+
    @staticmethod
    def from_node(node: clang.cindex.Cursor,
                  refd: Optional[clang.cindex.Cursor] = None,
@@ -208,6 +264,7 @@ class DiagnosticKind(Enum):
    NON_PURE_CALL_IN_PURE_CTX = enum.auto()
    NON_MT_SAFE_CALL_IN_MT_SAFE_CTX = enum.auto()
    NON_STABLE_TREE_CALL_IN_STABLE_TREE_CTX = enum.auto()
+    MISSING_MT_DISABLED_ANNOTATION = enum.auto()

    def __lt__(self, other):
        return self.value < other.value
@@ -245,13 +302,109 @@ class CallAnnotationsValidator:

        self._index = Index.create()

-        self._processed_headers: set[str] = set()
+        # Map key represents translation unit initial defines
+        # (from command line and source's lines before any include)
+        self._processed_headers: dict[str, set[str]] = {}
+        self._external_decls: dict[str, set[tuple[str, int]]] = {}

        # Current context
+        self._main_source_file: str = ""
+        self._defines: dict[str, str] = {}
        self._call_location: Optional[FunctionInfo] = None
        self._caller: Optional[FunctionInfo] = None
+        self._constructor_context: list[clang.cindex.Cursor] = []
        self._level: int = 0
-        self._constructor_context: int = 0
+
+    def is_mt_disabled_code_unit(self):
+        return "VL_MT_DISABLED_CODE_UNIT" in self._defines
+
+    def is_constructor_context(self):
+        return len(self._constructor_context) > 0
+
+    # Parses all lines in a form: `#define KEY VALUE` located before any `#include` line.
+    # The parsing is very simple, there is no support for line breaks, etc.
+    @staticmethod
+    def parse_initial_defines(source_file: str) -> dict[str, str]:
+        defs: dict[str, str] = {}
+        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)
+                if match:
+                    if match.group(1).startswith("define"):
+                        key = match.group(2)
+                        value = match.groups("1")[2]
+                        defs[key] = value
+                    elif match.group(1).startswith("include"):
+                        break
+        return defs
+
+    @staticmethod
+    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")
+                                                or arg.endswith(".h")):
+                    continue
+
+                # Skipped options with separate value argument.
+                if arg in ["-o", "-T", "-MT", "-MQ", "-MF"
+                           "-L"]:
+                    next(args_iter)
+                    continue
+
+                # Skipped options without separate value argument.
+                if arg == "-c" or arg.startswith("-W") or arg.startswith("-L"):
+                    continue
+
+                # Preserved options with separate value argument.
+                if arg in [
+                        "-x"
+                        "-Xclang", "-I", "-isystem", "-iquote", "-include",
+                        "-include-pch"
+                ]:
+                    filtered_args += [arg, next(args_iter)]
+                    continue
+
+                kv_str = None
+                d_or_u = None
+                # Preserve define/undefine with separate value argument.
+                if arg in ["-D", "-U"]:
+                    filtered_args.append(arg)
+                    d_or_u = arg[1]
+                    kv_str = next(args_iter)
+                    filtered_args.append(kv_str)
+                # Preserve define/undefine without separate value argument.
+                elif arg[0:2] in ["-D", "-U"]:
+                    filtered_args.append(arg)
+                    kv_str = arg[2:]
+                    d_or_u = arg[1]
+                # Preserve everything else.
+                else:
+                    filtered_args.append(arg)
+                    continue
+
+                # Keep track of defines for class' internal purposes.
+                key_value = kv_str.split("=", 1)
+                key = key_value[0]
+                val = "1" if len(key_value) == 1 else key_value[1]
+
+                if d_or_u == "D":
+                    defines[key] = val
+                elif d_or_u == "U" and key in defines:
+                    del defines[key]
+
+        except StopIteration:
+            pass
+
+        return (filtered_args, defines)

    def compile_and_analyze_file(self, source_file: str,
                                 compiler_args: list[str],
@@ -259,18 +412,32 @@ class CallAnnotationsValidator:
        filename = os.path.abspath(source_file)
        initial_cwd = "."

+        filtered_args, defines = self.filter_out_unsupported_compiler_args(
+            compiler_args)
+        defines.update(self.parse_initial_defines(source_file))
+
        if build_dir:
            initial_cwd = os.getcwd()
            os.chdir(build_dir)
-        translation_unit = self._index.parse(filename, compiler_args)
-        has_errors = False
-        for diag in translation_unit.diagnostics:
-            if diag.severity > clang.cindex.Diagnostic.Error:
-                has_errors = True
-        if translation_unit and not has_errors:
+        try:
+            translation_unit = self._index.parse(filename, filtered_args)
+        except TranslationUnitLoadError:
+            translation_unit = None
+        errors = []
+        if translation_unit:
+            for diag in translation_unit.diagnostics:
+                if diag.severity >= clang.cindex.Diagnostic.Error:
+                    errors.append(str(diag))
+        if translation_unit and len(errors) == 0:
+            self._defines = defines
+            self._main_source_file = filename
            self.process_translation_unit(translation_unit)
+            self._main_source_file = ""
+            self._defines = {}
        else:
            print(f"%Error: parsing failed: {filename}", file=sys.stderr)
+            for error in errors:
+                print(f"        {error}", file=sys.stderr)
        if build_dir:
            os.chdir(initial_cwd)

@@ -341,7 +508,7 @@ class CallAnnotationsValidator:
                    return None

            refn = find_object_ref(node)
-            if self._constructor_context and not refn:
+            if self.is_constructor_context() and not refn:
                # we are in constructor and no object reference means
                # we are calling local method. It is MT safe
                # only if this method is also only calling local methods or
@@ -353,20 +520,44 @@ class CallAnnotationsValidator:
            elif refn and refn.kind == CursorKind.MEMBER_REF_EXPR and refn.referenced:
                refn = refn.referenced
                refna = VlAnnotations.from_nodes_list(refn.get_children())
-                if refna.guarded or self._constructor_context:
+                if refna.guarded:
                    is_mt_safe = True
+                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
+                    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)
+                        if base_class:
+                            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:
-                # This is probably a local or an argument.
-                # Calling methods on local pointers or references is MT-safe,
-                # but on argument pointers or references is not.
-                if "*" not in refn.type.spelling and "&" not in refn.type.spelling:
-                    is_mt_safe = True
-                # local variable
-                if refn.referenced.kind == CursorKind.VAR_DECL:
-                    is_mt_safe = True
+                if refn.get_definition():
+                    if refn.referenced.semantic_parent:
+                        if refn.referenced.semantic_parent.kind in [
+                                CursorKind.FUNCTION_DECL, CursorKind.CXX_METHOD
+                        ]:
+                            # This is a local or an argument.
+                            # Calling methods on local pointers or references is MT-safe,
+                            # but on argument pointers or references is not.
+                            if "*" not in refn.type.spelling and "&" not in refn.type.spelling:
+                                is_mt_safe = True
+                            # local variable
+                            if refn.referenced.kind == CursorKind.VAR_DECL:
+                                is_mt_safe = True
+                else:
+                    # Global variable in different translation unit, unsafe
+                    pass
            elif refn and refn.kind == CursorKind.CALL_EXPR:
-                if self._constructor_context:
+                if self.is_constructor_context():
                    # call to local function from constructor context
                    # safe if this function also calling local methods or
                    # MT-safe methods
@@ -442,18 +633,18 @@ class CallAnnotationsValidator:

        # Constructors are OK in MT-safe context
        # only if they call local methods or MT-safe functions.
-        if ctx.is_mt_safe_context() or self._constructor_context:
-            self._constructor_context += 1
+        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._constructor_context -= 1
+            self._constructor_context.pop()

        # stable tree context
        if ctx.is_stabe_tree_context():
-            self._constructor_context += 1
+            self._constructor_context.append(refd)
            self.iterate_children(refd.get_children(),
                                  self.dispatch_node_inside_definition)
-            self._constructor_context -= 1
+            self._constructor_context.pop()

        # pure context
        if ctx.is_pure_context():
@@ -516,6 +707,18 @@ class CallAnnotationsValidator:
                  f"    from: {node.location.file.name}:{node.location.line}")
        return True

+    def process_function_declaration(self, node: clang.cindex.Cursor):
+        # Ignore declarations in main .cpp file
+        if node.location.file.name != self._main_source_file:
+            children = list(node.get_children())
+            annotations = VlAnnotations.from_nodes_list(children)
+            if not annotations.mt_disabled:
+                self._external_decls.setdefault(node.get_usr(), set()).add(
+                    (str(node.location.file.name), int(node.location.line)))
+            return self.iterate_children(children, self.dispatch_node)
+
+        return self.iterate_children(node.get_children(), self.dispatch_node)
+
    # Definition handling

    def dispatch_node_inside_definition(self, node: clang.cindex.Cursor):
@@ -546,6 +749,18 @@ class CallAnnotationsValidator:
        assert refd is not None

        def_annotations = VlAnnotations.from_nodes_list(node_children)
+        # 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
+        # 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
+            if refd.location.file.name == self._main_source_file:
+                def_annotations.mt_disabled = True
+                def_annotations.requires = True

        if not (def_annotations.is_empty() or def_annotations == annotations):
            # Use definition's annotations for the diagnostic
@@ -558,12 +773,26 @@ class CallAnnotationsValidator:
                DiagnosticKind.ANNOTATIONS_DEF_DECL_MISMATCH)

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

+        if self.is_mt_disabled_code_unit():
+            # Report declarations of this functions that don't have MT_DISABLED annotation
+            # and are located in headers.
+            if node.location.file.name == self._main_source_file:
+                usr = node.get_usr()
+                declarations = self._external_decls.get(usr, set())
+                for file, line in declarations:
+                    self.emit_diagnostic(
+                        FunctionInfo.from_decl_file_line_and_refd_node(
+                            file, line, refd, def_annotations),
+                        DiagnosticKind.MISSING_MT_DISABLED_ANNOTATION)
+                if declarations:
+                    del self._external_decls[usr]
+
        self.iterate_children(node_children,
                              self.dispatch_node_inside_definition)

@@ -575,34 +804,36 @@ class CallAnnotationsValidator:
    # Nodes not located inside definition

    def dispatch_node(self, node: clang.cindex.Cursor):
-        if node.is_definition() and node.kind in [
+        if node.kind in [
                CursorKind.CXX_METHOD, CursorKind.FUNCTION_DECL,
                CursorKind.CONSTRUCTOR, CursorKind.CONVERSION_FUNCTION
        ]:
-            return self.process_function_definition(node)
-        if node.is_definition() and node.kind in [
-                CursorKind.NAMESPACE, CursorKind.STRUCT_DECL,
-                CursorKind.UNION_DECL, CursorKind.CLASS_DECL
-        ]:
-            return self.iterate_children(node.get_children(),
-                                         self.dispatch_node)
+            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)

    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())
        for child in translation_unit.cursor.get_children():
            if self._is_ignored_top_level(child):
                continue
-            if self._processed_headers:
+            if tu_processed_headers:
                filename = os.path.abspath(child.location.file.name)
-                if filename in self._processed_headers:
+                if filename in tu_processed_headers:
                    continue
            self.dispatch_node(child)
        self._level -= 1

-        self._processed_headers.update([
+        tu_processed_headers.update([
            os.path.abspath(str(hdr.source))
            for hdr in translation_unit.get_includes()
        ])
@@ -664,34 +895,40 @@ def get_filter_funcs(verilator_root: str):


 def precompile_header(compile_command: CompileCommand, tmp_dir: str) -> str:
+    initial_cwd = os.getcwd()
+    errors = []
    try:
-        initial_cwd = os.getcwd()
        os.chdir(compile_command.directory)

        index = Index.create()
        translation_unit = index.parse(compile_command.filename,
                                       compile_command.args)
        for diag in translation_unit.diagnostics:
-            if diag.severity > clang.cindex.Diagnostic.Error:
-                pch_file = None
-                break
-        else:
+            if diag.severity >= clang.cindex.Diagnostic.Error:
+                errors.append(str(diag))
+
+        if len(errors) == 0:
            pch_file = os.path.join(
                tmp_dir,
                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:
+        print(f"%Warning: {exception}", file=sys.stderr)
+
+    finally:
        os.chdir(initial_cwd)

-        if pch_file:
-            return pch_file
-
-    except (TranslationUnitSaveError, TranslationUnitLoadError, OSError):
-        pass
-
    print(
-        f"%Warning: Precompiling failed, skipping: {compile_command.filename}")
+        f"%Warning: Precompilation failed, skipping: {compile_command.filename}",
+        file=sys.stderr)
+    for error in errors:
+        print(f"          {error}", file=sys.stderr)
    return ""


@@ -714,7 +951,7 @@ def run_analysis(ccl: Iterable[CompileCommand], pccl: Iterable[CompileCommand],
                                       is_ignored_def, is_ignored_call)
        for compile_command in ccl:
            cav.compile_and_analyze_file(compile_command.filename,
-                                         compile_command.args + extra_args,
+                                         extra_args + compile_command.args,
                                         compile_command.directory)


@@ -834,6 +1071,10 @@ class TopDownSummaryPrinter():
                name += "is pure but calls non-pure function(s)"
            elif func.reason == DiagnosticKind.NON_STABLE_TREE_CALL_IN_STABLE_TREE_CTX:
                name += "is stable_tree but calls non-stable_tree or non-mtsafe"
+            elif func.reason == DiagnosticKind.MISSING_MT_DISABLED_ANNOTATION:
+                name += ("defined in a file marked as " +
+                         "VL_MT_DISABLED_CODE_UNIT has declaration(s) " +
+                         "without VL_MT_DISABLED annotation")
            else:
                name += "for unknown reason (please add description)"

@@ -900,10 +1141,15 @@ 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("--cxxflags",
                        type=str,
                        default=None,
-                        help="Flags passed to clang++.")
+                        help="Extra flags passed to clang++.")
    parser.add_argument(
        "--compilation-root",
        type=str,
@@ -928,34 +1174,52 @@ def main():
        cmdline.compilation_root = cmdline.verilator_root

    verilator_root = os.path.abspath(cmdline.verilator_root)
-    compilation_root = os.path.abspath(cmdline.compilation_root)
+    default_compilation_root = os.path.abspath(cmdline.compilation_root)
+
+    compdb: Optional[CompilationDatabase] = None
+    if cmdline.compile_commands_dir:
+        compdb = CompilationDatabase.fromDirectory(
+            cmdline.compile_commands_dir)

-    default_cxx_flags = [
-        f"-I{verilator_root}/src",
-        f"-I{verilator_root}/include",
-        f"-I{verilator_root}/src/obj_opt",
-        "-fcoroutines-ts",
-    ]
    if cmdline.cxxflags is not None:
-        cxxflags = shlex.split(cmdline.cxxflags)
+        common_cxxflags = shlex.split(cmdline.cxxflags)
    else:
-        cxxflags = default_cxx_flags
+        common_cxxflags = []

    precompile_commands_list = []

    if cmdline.precompile:
-        hdr_cxxflags = ['-xc++-header'] + cxxflags
+        hdr_cxxflags = ['-xc++-header'] + common_cxxflags
        for refid, file in enumerate(cmdline.precompile):
            filename = os.path.abspath(file)
            compile_command = CompileCommand(refid, filename, hdr_cxxflags,
-                                             compilation_root)
+                                             default_compilation_root)
            precompile_commands_list.append(compile_command)

    compile_commands_list = []
    for refid, file in enumerate(cmdline.file):
        filename = os.path.abspath(file)
-        compile_command = CompileCommand(refid, filename, cxxflags,
-                                         compilation_root)
+        root = default_compilation_root
+        cxxflags = []
+        if compdb:
+            entry = compdb.getCompileCommands(filename)
+            entry_list = list(entry)
+            # Compilation database can contain multiple entries for single file,
+            # e.g. when it has been updated by appending new entries.
+            # Use last entry for the file, if it exists, as it is the newest one.
+            if len(entry_list) > 0:
+                last_entry = entry_list[-1]
+                root = last_entry.directory
+                entry_args = list(last_entry.arguments)
+                # First argument in compile_commands.json arguments list is
+                # compiler executable name/path. CIndex (libclang) always
+                # implicitly prepends executable name, so it shouldn't be passed
+                # here.
+                cxxflags = common_cxxflags + entry_args[1:]
+        else:
+            cxxflags = common_cxxflags[:]
+
+        compile_command = CompileCommand(refid, filename, cxxflags, root)
        compile_commands_list.append(compile_command)

    summary_printer = TopDownSummaryPrinter()
@@ -1,6 +1,6 @@
 *~
 *.old
-config_build.h
+config_package.h
 Makefile
 Makefile_obj
 .objcache*
@@ -82,6 +82,7 @@ set(HEADERS
    V3File.h
    V3FileLine.h
    V3Force.h
+    V3Fork.h
    V3FunctionTraits.h
    V3Gate.h
    V3Global.h
@@ -150,6 +151,7 @@ set(HEADERS
    V3Table.h
    V3Task.h
    V3ThreadPool.h
+    V3ThreadSafety.h
    V3Timing.h
    V3Trace.h
    V3TraceDecl.h
@@ -227,6 +229,7 @@ set(COMMON_SOURCES
    V3File.cpp
    V3FileLine.cpp
    V3Force.cpp
+    V3Fork.cpp
    V3Gate.cpp
    V3Global.cpp
    V3Graph.cpp
@@ -322,7 +325,7 @@ file(TO_NATIVE_PATH ${srcdir}/../bin/verilator_includer VERILATOR_INCLUDER)
 file(TO_NATIVE_PATH ${srcdir}/vlcovgen VLCOVGEN)
 file(TO_NATIVE_PATH ${srcdir}/config_rev CONFIG_REV)

-configure_file(config_build.h.in config_build.h @ONLY)
+configure_file(config_package.h.in config_package.h @ONLY)

 add_custom_command(
    OUTPUT V3Ast__gen_forward_class_decls.h V3Dfg__gen_forward_class_decls.h
@@ -414,7 +417,7 @@ foreach(astgen_name ${ASTGENERATED_NAMES})
        MAIN_DEPENDENCY ${astgen_name}.cpp
        DEPENDS ${ASTGEN} V3Ast.h
        COMMAND ${PYTHON3} ARGS
-            ${ASTGEN} -I"${srcdir}" --astdef V3AstNodeDType.h --astdef V3AstNodeExpr.h --astdef V3AstNodeOther.h --dfgdef V3DfgVertices.h ${astgen_name}.cpp
+            ${ASTGEN} -I "${srcdir}" --astdef V3AstNodeDType.h --astdef V3AstNodeExpr.h --astdef V3AstNodeOther.h --dfgdef V3DfgVertices.h ${astgen_name}.cpp
    )
    list(APPEND GENERATED_FILES ${astgen_name}__gen.cpp)
 endforeach()
@@ -488,11 +491,13 @@ target_include_directories(${verilator}
 )

 if (WIN32)
+    target_compile_options(${verilator} PRIVATE /bigobj)
    target_compile_definitions(${verilator} PRIVATE
        YY_NO_UNISTD_H
    )
    target_include_directories(${verilator} PRIVATE ../platform/win32)
    target_link_libraries(${verilator} PRIVATE bcrypt psapi)
+    target_link_options(${verilator} PRIVATE /STACK:10000000)
 endif()

 install(TARGETS ${verilator})
@@ -18,14 +18,13 @@
 #### Start of system configuration section. ####

 srcdir = @srcdir@
-PYTHON3 = @PYTHON3@
 EXEEXT = @EXEEXT@
+PYTHON3 = @PYTHON3@
 # VPATH only does sources; fix install_test's not creating ../bin
 vpath %.h @srcdir@

 #### End of system configuration section. ####

-
 default: dbg opt
 debug: dbg
 optimize: opt
@@ -36,6 +35,26 @@ endif

 UNDER_GIT = $(wildcard ${srcdir}/../.git/logs/HEAD)

+ifeq (,$(wildcard obj_dbg/bear.o))
+  ifneq (, $(shell which bear 2>/dev/null))
+    BEAR := $(shell which bear)
+    ifeq (, $(shell $(BEAR) --output obj_dbg/comptest.json -- true))
+      $(shell which bear 2>/dev/null >obj_dbg/bear.o)
+    else
+      # unsupported version
+      BEAR :=
+    endif
+  endif
+else
+  BEAR := $(shell cat obj_dbg/bear.o)
+endif
+
+ifneq ($(BEAR),)
+BEAR_OBJ_OPT := $(BEAR) --append --output obj_dbg/compile_commands.json --
+else
+BEAR_OBJ_OPT :=
+endif
+
 #*********************************************************************

 obj_opt:
@@ -62,8 +81,8 @@ endif

 dbg: ../bin/verilator_bin_dbg$(EXEEXT) ../bin/verilator_coverage_bin_dbg$(EXEEXT)
 ../bin/verilator_bin_dbg$(EXEEXT): obj_dbg ../bin prefiles
-	$(MAKE) -C obj_dbg -j 1  TGT=../$@ VL_DEBUG=1 -f ../Makefile_obj serial
-	$(MAKE) -C obj_dbg       TGT=../$@ VL_DEBUG=1 -f ../Makefile_obj
+	$(BEAR_OBJ_OPT) $(MAKE) -C obj_dbg -j 1  TGT=../$@ VL_DEBUG=1 -f ../Makefile_obj serial
+	$(BEAR_OBJ_OPT) $(MAKE) -C obj_dbg       TGT=../$@ VL_DEBUG=1 -f ../Makefile_obj

 ../bin/verilator_coverage_bin_dbg$(EXEEXT): obj_dbg ../bin prefiles
 	$(MAKE) -C obj_dbg       TGT=../$@ VL_DEBUG=1 VL_VLCOV=1 -f ../Makefile_obj serial_vlcov
@@ -95,7 +114,7 @@ clean mostlyclean distclean maintainer-clean::
 	-rm -f .objcache*

 distclean maintainer-clean::
-	-rm -f Makefile Makefile_obj config_build.h
+	-rm -f Makefile Makefile_obj config_package.h

 maintainer-clean::
 	-rm -f config_rev.h
@@ -205,6 +205,7 @@ RAW_OBJS = \
 	V3File.o \
 	V3FileLine.o \
 	V3Force.o \
+	V3Fork.o \
 	V3Gate.o \
 	V3Global.o \
 	V3Graph.o \
@@ -288,6 +289,7 @@ NON_STANDALONE_HEADERS = \
 	V3AstNodeExpr.h  \
 	V3AstNodeOther.h  \
 	V3DfgVertices.h  \
+	V3ThreadPool.h \
 	V3WidthCommit.h  \

 AST_DEFS := \
@@ -47,6 +47,7 @@ VL_DEFINE_DEBUG_FUNCTIONS;
 // Extend V3GraphVertex class for use in latch detection graph

 class LatchDetectGraphVertex final : public V3GraphVertex {
+    VL_RTTI_IMPL(LatchDetectGraphVertex, V3GraphVertex)
 public:
    enum VertexType : uint8_t { VT_BLOCK, VT_BRANCH, VT_OUTPUT };

@@ -183,7 +184,7 @@ public:
                        << " (not all control paths of combinational always assign a value)\n"
                        << nodep->warnMore()
                        << "... Suggest use of always_latch for intentional latches");
-                if (dumpGraph() >= 9) dumpDotFilePrefixed("latch_" + vrp->name());
+                if (dumpGraphLevel() >= 9) dumpDotFilePrefixed("latch_" + vrp->name());
            }
            vertp->user(false);  // Clear again (see above)
            vrp->varp()->isLatched(latch_detected);
@@ -320,7 +321,8 @@ private:
    // VISITORS
    void visit(AstVarRef* nodep) override {
        const AstVar* const varp = nodep->varp();
-        if (nodep->access().isWriteOrRW() && varp->isSignal() && !varp->isUsedLoopIdx()) {
+        if (nodep->access().isWriteOrRW() && varp->isSignal() && !varp->isUsedLoopIdx()
+            && !varp->isFuncLocalSticky()) {
            m_graph.addAssignment(nodep);
        }
    }
@@ -333,6 +335,8 @@ private:
            m_graph.addPathVertex(branchp, "ELSE");
            iterateAndNextConstNull(nodep->elsesp());
            m_graph.currentp(parentp);
+        } else {
+            iterateChildrenConst(nodep);
        }
    }
    //--------------------
@@ -652,5 +656,5 @@ public:
 void V3Active::activeAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { ActiveVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("active", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("active", 0, dumpTreeLevel() >= 3);
 }
@@ -145,5 +145,5 @@ public:
 void V3ActiveTop::activeTopAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { ActiveTopVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("activetop", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("activetop", 0, dumpTreeLevel() >= 3);
 }
@@ -20,6 +20,7 @@
 #include "V3Assert.h"

 #include "V3Ast.h"
+#include "V3Error.h"
 #include "V3Global.h"
 #include "V3Stats.h"

@@ -156,7 +157,7 @@ private:
            sentreep->unlinkFrBack();
            if (m_procedurep) {
                // To support this need queue of asserts to activate
-                nodep->v3error("Unsupported: Procedural concurent assertion with"
+                nodep->v3error("Unsupported: Procedural concurrent assertion with"
                               " clocking event inside always (IEEE 1800-2917 16.14.6)");
            }
        }
@@ -255,7 +256,7 @@ private:
                ifp = nextifp;
            } while (ifp);

-            AstNode* const newifp = nodep->cloneTree(false);
+            AstIf* const newifp = nodep->cloneTree(false);
            const bool allow_none = nodep->unique0Pragma();

            // Empty case means no property
@@ -544,5 +545,5 @@ public:
 void V3Assert::assertAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { AssertVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("assert", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("assert", 0, dumpTreeLevel() >= 3);
 }
@@ -21,7 +21,6 @@
 #include "verilatedos.h"

 #include "V3Ast.h"
-#include "V3Error.h"

 //============================================================================

@@ -496,5 +496,5 @@ public:
 void V3AssertPre::assertPreAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { AssertPreVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("assertpre", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("assertpre", 0, dumpTreeLevel() >= 3);
 }
@@ -55,12 +55,28 @@ bool VNUser5InUse::s_userBusy = false;
 int AstNodeDType::s_uniqueNum = 0;

 //######################################################################
-// V3AstType
+// VNType
+
+const VNTypeInfo VNType::typeInfoTable[] = {
+#include "V3Ast__gen_type_info.h"  // From ./astgen
+};

 std::ostream& operator<<(std::ostream& os, VNType rhs);

 //######################################################################
-// Creators
+// VSelfPointerText
+
+const std::shared_ptr<const string> VSelfPointerText::s_emptyp = std::make_shared<string>("");
+const std::shared_ptr<const string> VSelfPointerText::s_thisp = std::make_shared<string>("this");
+
+string VSelfPointerText::protect(bool useSelfForThis, bool protect) const {
+    const string& sp
+        = useSelfForThis ? VString::replaceWord(asString(), "this", "vlSelf") : asString();
+    return VIdProtect::protectWordsIf(sp, protect);
+}
+
+//######################################################################
+// AstNode

 AstNode::AstNode(VNType t, FileLine* fl)
    : m_type{t}
@@ -125,9 +141,7 @@ string AstNode::encodeNumber(int64_t num) {
    }
 }

-string AstNode::nameProtect() const VL_MT_STABLE {
-    return VIdProtect::protectIf(name(), protect());
-}
+string AstNode::nameProtect() const { return VIdProtect::protectIf(name(), protect()); }
 string AstNode::origNameProtect() const { return VIdProtect::protectIf(origName(), protect()); }

 string AstNode::shortName() const {
@@ -285,7 +299,7 @@ string AstNode::vpiName(const string& namein) {
    return pretty;
 }

-string AstNode::prettyTypeName() const VL_MT_STABLE {
+string AstNode::prettyTypeName() const {
    if (name() == "") return typeName();
    return std::string{typeName()} + " '" + prettyName() + "'";
 }
@@ -1081,10 +1095,12 @@ bool AstNode::sameTreeIter(const AstNode* node1p, const AstNode* node2p, bool ig
    // private: Return true if the two trees are identical
    if (!node1p && !node2p) return true;
    if (!node1p || !node2p) return false;
-    if (node1p->type() != node2p->type() || node1p->dtypep() != node2p->dtypep()
-        || !node1p->same(node2p) || (gateOnly && !node1p->isGateOptimizable())) {
-        return false;
-    }
+    if (node1p->type() != node2p->type()) return false;
+    UASSERT_OBJ(
+        (!node1p->dtypep() && !node2p->dtypep()) || (node1p->dtypep() && node2p->dtypep()), node1p,
+        "Comparison of a node with dtypep() with a node without dtypep()\n-node2=" << node2p);
+    if (node1p->dtypep() && !node1p->dtypep()->similarDType(node2p->dtypep())) return false;
+    if (!node1p->same(node2p) || (gateOnly && !node1p->isGateOptimizable())) return false;
    return (sameTreeIter(node1p->m_op1p, node2p->m_op1p, false, gateOnly)
            && sameTreeIter(node1p->m_op2p, node2p->m_op2p, false, gateOnly)
            && sameTreeIter(node1p->m_op3p, node2p->m_op3p, false, gateOnly)
@@ -1098,9 +1114,57 @@ bool AstNode::sameTreeIter(const AstNode* node1p, const AstNode* node2p, bool ig
 void AstNode::checkTreeIter(const AstNode* prevBackp) const VL_MT_STABLE {
    // private: Check a tree and children
    UASSERT_OBJ(prevBackp == this->backp(), this, "Back node inconsistent");
-    switch (this->type()) {
-#include "V3Ast__gen_op_checks.h"
-    default: VL_UNREACHABLE;  // LCOV_EXCL_LINE
+    const VNTypeInfo& typeInfo = *type().typeInfo();
+    for (int i = 1; i <= 4; i++) {
+        AstNode* nodep = nullptr;
+        switch (i) {
+        case 1: nodep = op1p(); break;
+        case 2: nodep = op2p(); break;
+        case 3: nodep = op3p(); break;
+        case 4: nodep = op4p(); break;
+        default: this->v3fatalSrc("Bad case"); break;
+        }
+        const char* opName = typeInfo.m_opNamep[i - 1];
+        switch (typeInfo.m_opType[i - 1]) {
+        case VNTypeInfo::OP_UNUSED:
+            UASSERT_OBJ(!nodep, this, typeInfo.m_namep << " must not use " << opName << "()");
+            break;
+        case VNTypeInfo::OP_USED:
+            UASSERT_OBJ(nodep, this,
+                        typeInfo.m_namep << " must have non nullptr " << opName << "()");
+            UASSERT_OBJ(!nodep->nextp(), this,
+                        typeInfo.m_namep << "::" << opName
+                                         << "() cannot have a non nullptr nextp()");
+            nodep->checkTreeIter(this);
+            break;
+        case VNTypeInfo::OP_LIST:
+            if (const AstNode* const headp = nodep) {
+                const AstNode* backp = this;
+                const AstNode* tailp = headp;
+                const AstNode* opp = headp;
+                do {
+                    opp->checkTreeIter(backp);
+                    UASSERT_OBJ(opp == headp || !opp->nextp() || !opp->m_headtailp, opp,
+                                "Headtailp should be null in middle of lists");
+                    backp = tailp = opp;
+                    opp = opp->nextp();
+                } while (opp);
+                UASSERT_OBJ(headp->m_headtailp == tailp, headp,
+                            "Tail in headtailp is inconsistent");
+                UASSERT_OBJ(tailp->m_headtailp == headp, tailp,
+                            "Head in headtailp is inconsistent");
+            }
+            break;
+        case VNTypeInfo::OP_OPTIONAL:
+            if (nodep) {
+                UASSERT_OBJ(!nodep->nextp(), this,
+                            typeInfo.m_namep << "::" << opName
+                                             << "() cannot have a non-nullptr nextp()");
+                nodep->checkTreeIter(this);
+            }
+            break;
+        default: this->v3fatalSrc("Bad case"); break;
+        }
    }
 }

@@ -1217,7 +1281,7 @@ void AstNode::dumpTreeFile(const string& filename, bool append, bool doDump, boo
            if (logsp->fail()) v3fatal("Can't write " << filename);
            *logsp << "Verilator Tree Dump (format 0x3900) from <e" << std::dec << editCountLast();
            *logsp << "> to <e" << std::dec << editCountGbl() << ">\n";
-            if (editCountGbl() == editCountLast() && ::dumpTree() < 9) {
+            if (editCountGbl() == editCountLast() && ::dumpTreeLevel() < 9) {
                *logsp << '\n';
                *logsp << "No changes since last dump!\n";
            } else {
@@ -1227,7 +1291,7 @@ void AstNode::dumpTreeFile(const string& filename, bool append, bool doDump, boo
        }
    }
    if (doDump && v3Global.opt.debugEmitV()) V3EmitV::debugEmitV(filename + ".v");
-    if (doCheck && (v3Global.opt.debugCheck() || ::dumpTree())) {
+    if (doCheck && (v3Global.opt.debugCheck() || ::dumpTreeLevel())) {
        // Error check
        checkTree();
        // Broken isn't part of check tree because it can munge iterp's
@@ -1289,12 +1353,6 @@ bool AstNode::isTreePureRecurse() const {
    return true;
 }

-void AstNode::v3errorEndFatal(std::ostringstream& str) const VL_REQUIRES(V3Error::s().m_mutex) {
-    v3errorEnd(str);
-    assert(0);  // LCOV_EXCL_LINE
-    VL_UNREACHABLE;
-}
-
 string AstNode::instanceStr() const {
    // Max iterations before giving up on location search,
    // in case we have some circular reference bug.
@@ -1319,9 +1377,9 @@ string AstNode::instanceStr() const {

    return "";
 }
-void AstNode::v3errorEnd(std::ostringstream& str) const VL_REQUIRES(V3Error::s().m_mutex) {
+void AstNode::v3errorEnd(std::ostringstream& str) const VL_RELEASE(V3Error::s().m_mutex) {
    if (!m_fileline) {
-        V3Error::s().v3errorEnd(str, instanceStr());
+        V3Error::v3errorEnd(str, instanceStr());
    } else {
        std::ostringstream nsstr;
        nsstr << str.str();
@@ -1338,6 +1396,11 @@ void AstNode::v3errorEnd(std::ostringstream& str) const VL_REQUIRES(V3Error::s()
            nsstr, m_fileline->warnIsOff(V3Error::s().errorCode()) ? "" : instanceStr());
    }
 }
+void AstNode::v3errorEndFatal(std::ostringstream& str) const VL_RELEASE(V3Error::s().m_mutex) {
+    v3errorEnd(str);
+    assert(0);  // LCOV_EXCL_LINE
+    VL_UNREACHABLE;
+}

 //======================================================================
 // Data type conversion
@@ -1406,6 +1469,9 @@ AstNodeDType* AstNode::findQueueIndexDType() const {
 AstNodeDType* AstNode::findVoidDType() const {
    return v3Global.rootp()->typeTablep()->findVoidDType(fileline());
 }
+AstNodeDType* AstNode::findStreamDType() const {
+    return v3Global.rootp()->typeTablep()->findStreamDType(fileline());
+}

 //######################################################################
 // VNDeleter
@@ -82,12 +82,36 @@ using MTaskIdSet = std::set<int>;  // Set of mtaskIds for Var sorting
 // Ast<nodetypename>*' is returned.
 #define VN_AS(nodep, nodetypename) (AstNode::privateAs<Ast##nodetypename, decltype(nodep)>(nodep))

+// Same as VN_AS, but only checks the type in debug builds. Type checking is less critical in node
+// child getters (the strongly-typed functions that wrap op*p pointers). This is because the op*p
+// pointers are usually populated by code that already asserts the correct type. Having fewer type
+// assertions yields better performance in release builds.
+#ifdef VL_DEBUG
+#define VN_DBG_AS(nodep, nodetypename) VN_AS(nodep, nodetypename)
+#else
+#define VN_DBG_AS(nodep, nodetypename) \
+    (AstNode::unsafePrivateAs<Ast##nodetypename, decltype(nodep)>(nodep))
+#endif
+
 // (V)erilator (N)ode deleted: Pointer to deleted AstNode (for assertions only)
 #define VN_DELETED(nodep) VL_UNLIKELY((uint64_t)(nodep) == 0x1)

 //######################################################################

+struct VNTypeInfo {
+    const char* m_namep;
+    enum uint8_t {
+        OP_UNUSED,
+        OP_USED,
+        OP_LIST,
+        OP_OPTIONAL,
+    } m_opType[4];
+    const char* m_opNamep[4];
+};
+
 class VNType final {
+    static const VNTypeInfo typeInfoTable[];
+
 public:
 #include "V3Ast__gen_type_enum.h"  // From ./astgen
    // Above include has:
@@ -100,6 +124,7 @@ public:
    constexpr VNType(en _e) VL_MT_SAFE : m_e{_e} {}
    explicit VNType(int _e)
        : m_e(static_cast<en>(_e)) {}  // Need () or GCC 4.8 false warning
+    const VNTypeInfo* typeInfo() const VL_MT_SAFE { return &typeInfoTable[m_e]; }
    constexpr operator en() const VL_MT_SAFE { return m_e; }
 };
 constexpr bool operator==(const VNType& lhs, const VNType& rhs) VL_PURE {
@@ -461,6 +486,7 @@ public:
        TRIGGER_SCHEDULER,
        DYNAMIC_TRIGGER_SCHEDULER,
        FORK_SYNC,
+        PROCESS_REFERENCE,
        // Unsigned and two state; fundamental types
        UINT32,
        UINT64,
@@ -493,6 +519,7 @@ public:
                                            "VlTriggerScheduler",
                                            "VlDynamicTriggerScheduler",
                                            "VlFork",
+                                            "VlProcessRef",
                                            "IData",
                                            "QData",
                                            "LOGIC_IMPLICIT",
@@ -500,13 +527,20 @@ public:
        return names[m_e];
    }
    const char* dpiType() const {
-        static const char* const names[]
-            = {"%E-unk",       "svBit",         "char",          "void*",        "char",
-               "int",          "%E-integer",    "svLogic",       "long long",    "double",
-               "short",        "%E-time",       "const char*",   "%E-untyped",   "dpiScope",
-               "const char*",  "%E-mtaskstate", "%E-triggervec", "%E-dly-sched", "%E-trig-sched",
-               "%E-dyn-sched", "%E-fork",       "IData",         "QData",        "%E-logic-implct",
-               " MAX"};
+        static const char* const names[] = {"%E-unk",        "svBit",
+                                            "char",          "void*",
+                                            "char",          "int",
+                                            "%E-integer",    "svLogic",
+                                            "long long",     "double",
+                                            "short",         "%E-time",
+                                            "const char*",   "%E-untyped",
+                                            "dpiScope",      "const char*",
+                                            "%E-mtaskstate", "%E-triggervec",
+                                            "%E-dly-sched",  "%E-trig-sched",
+                                            "%E-dyn-sched",  "%E-fork",
+                                            "%E-proc-ref",   "IData",
+                                            "QData",         "%E-logic-implct",
+                                            " MAX"};
        return names[m_e];
    }
    static void selfTest() {
@@ -545,6 +579,7 @@ public:
        case TRIGGER_SCHEDULER: return 0;  // opaque
        case DYNAMIC_TRIGGER_SCHEDULER: return 0;  // opaque
        case FORK_SYNC: return 0;  // opaque
+        case PROCESS_REFERENCE: return 0;  // opaque
        case UINT32: return 32;
        case UINT64: return 64;
        default: return 0;
@@ -584,7 +619,7 @@ public:
        return (m_e == EVENT || m_e == STRING || m_e == SCOPEPTR || m_e == CHARPTR
                || m_e == MTASKSTATE || m_e == TRIGGERVEC || m_e == DELAY_SCHEDULER
                || m_e == TRIGGER_SCHEDULER || m_e == DYNAMIC_TRIGGER_SCHEDULER || m_e == FORK_SYNC
-                || m_e == DOUBLE || m_e == UNTYPED);
+                || m_e == PROCESS_REFERENCE || m_e == DOUBLE || m_e == UNTYPED);
    }
    bool isDouble() const VL_MT_SAFE { return m_e == DOUBLE; }
    bool isEvent() const { return m_e == EVENT; }
@@ -781,6 +816,7 @@ public:
            "BLOCKTEMP", "MODULETEMP", "STMTTEMP",     "XTEMP",   "IFACEREF", "MEMBER"};
        return names[m_e];
    }
+    bool isParam() const { return m_e == GPARAM || m_e == LPARAM; }
    bool isSignal() const {
        return (m_e == WIRE || m_e == WREAL || m_e == IMPLICITWIRE || m_e == TRIWIRE || m_e == TRI0
                || m_e == TRI1 || m_e == PORT || m_e == SUPPLY0 || m_e == SUPPLY1 || m_e == VAR);
@@ -1147,30 +1183,34 @@ inline std::ostream& operator<<(std::ostream& os, const VNumRange& rhs) {
 class VUseType final {
 public:
    enum en : uint8_t {
-        IMP_INCLUDE,  // Implementation (.cpp) needs an include
-        INT_INCLUDE,  // Interface (.h) needs an include
-        IMP_FWD_CLASS,  // Implementation (.cpp) needs a forward class declaration
-        INT_FWD_CLASS,  // Interface (.h) needs a forward class declaration
+        // Enum values are compared with <, so order matters
+        INT_FWD_CLASS = 1 << 0,  // Interface (.h) needs a forward class declaration
+        INT_INCLUDE = 1 << 1,  // Interface (.h) needs an include
    };
    enum en m_e;
    VUseType()
-        : m_e{IMP_FWD_CLASS} {}
+        : m_e{INT_FWD_CLASS} {}
    // cppcheck-suppress noExplicitConstructor
    constexpr VUseType(en _e)
        : m_e{_e} {}
    explicit VUseType(int _e)
        : m_e(static_cast<en>(_e)) {}  // Need () or GCC 4.8 false warning
-    bool isInclude() const { return m_e == IMP_INCLUDE || m_e == INT_INCLUDE; }
-    bool isFwdClass() const { return m_e == IMP_FWD_CLASS || m_e == INT_FWD_CLASS; }
    constexpr operator en() const { return m_e; }
+    bool containsAny(VUseType other) { return m_e & other.m_e; }
    const char* ascii() const {
-        static const char* const names[] = {"IMP_INC", "INT_INC", "IMP_FWD", "INT_FWD"};
-        return names[m_e];
+        static const char* const names[] = {"INT_FWD", "INT_INC", "INT_FWD_INC"};
+        return names[m_e - 1];
    }
 };
 constexpr bool operator==(const VUseType& lhs, const VUseType& rhs) { return lhs.m_e == rhs.m_e; }
 constexpr bool operator==(const VUseType& lhs, VUseType::en rhs) { return lhs.m_e == rhs; }
 constexpr bool operator==(VUseType::en lhs, const VUseType& rhs) { return lhs == rhs.m_e; }
+constexpr VUseType::en operator|(VUseType::en lhs, VUseType::en rhs) {
+    return VUseType::en((uint8_t)lhs | (uint8_t)rhs);
+}
+constexpr VUseType::en operator&(VUseType::en lhs, VUseType::en rhs) {
+    return VUseType::en((uint8_t)lhs & (uint8_t)rhs);
+}
 inline std::ostream& operator<<(std::ostream& os, const VUseType& rhs) {
    return os << rhs.ascii();
 }
@@ -1211,6 +1251,44 @@ public:
    ~VBasicTypeKey() = default;
 };

+// ######################################################################
+//  VSelfPointerText - Represents text to be emitted before a given var reference, call, etc. to
+//  serve as a pointer to a 'self' object. For example, it could be empty (no self pointer), or the
+//  string 'this', or 'vlSymsp->...'
+
+class VSelfPointerText final {
+private:
+    // STATIC MEMBERS
+    // Keep these in shared pointers to avoid branching for special cases
+    static const std::shared_ptr<const string> s_emptyp;  // Holds ""
+    static const std::shared_ptr<const string> s_thisp;  // Holds "this"
+
+    // MEMBERS
+    std::shared_ptr<const string> m_strp;
+
+public:
+    // CONSTRUCTORS
+    class Empty {};  // for creator type-overload selection
+    VSelfPointerText(Empty)
+        : m_strp{s_emptyp} {}
+    class This {};  // for creator type-overload selection
+    VSelfPointerText(This)
+        : m_strp{s_thisp} {}
+    VSelfPointerText(This, const string& field)
+        : m_strp{std::make_shared<const string>("this->" + field)} {}
+    class VlSyms {};  // for creator type-overload selection
+    VSelfPointerText(VlSyms, const string& field)
+        : m_strp{std::make_shared<const string>("(&vlSymsp->" + field + ')')} {}
+
+    // METHODS
+    bool isEmpty() const { return m_strp == s_emptyp; }
+    bool isVlSym() const { return m_strp->find("vlSymsp") != string::npos; }
+    bool hasThis() const { return m_strp == s_thisp || VString::startsWith(*m_strp, "this"); }
+    string protect(bool useSelfForThis, bool protect) const;
+    const std::string& asString() const { return *m_strp; }
+    bool operator==(const VSelfPointerText& other) const { return *m_strp == *other.m_strp; }
+};
+
 //######################################################################
 // AstNUser - Generic base class for AST User nodes.
 //          - Also used to allow parameter passing up/down iterate calls
@@ -1293,55 +1371,55 @@ protected:
 class VNUser1InUse final : VNUserInUseBase {
 protected:
    friend class AstNode;
-    static uint32_t     s_userCntGbl;   // Count of which usage of userp() this is
-    static bool         s_userBusy;     // Count is in use
+    static uint32_t s_userCntGbl;  // Count of which usage of userp() this is
+    static bool s_userBusy;  // Count is in use
 public:
-    VNUser1InUse()     { allocate(1, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
-    ~VNUser1InUse()    { free    (1, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    VNUser1InUse()      { allocate(1, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    ~VNUser1InUse()     { free    (1, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void clear() { clearcnt(1, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void check() { checkcnt(1, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
 };
 class VNUser2InUse final : VNUserInUseBase {
 protected:
    friend class AstNode;
-    static uint32_t     s_userCntGbl;   // Count of which usage of userp() this is
-    static bool         s_userBusy;     // Count is in use
+    static uint32_t s_userCntGbl;  // Count of which usage of userp() this is
+    static bool s_userBusy;  // Count is in use
 public:
-    VNUser2InUse()     { allocate(2, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
-    ~VNUser2InUse()    { free    (2, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    VNUser2InUse()      { allocate(2, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    ~VNUser2InUse()     { free    (2, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void clear() { clearcnt(2, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void check() { checkcnt(2, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
 };
 class VNUser3InUse final : VNUserInUseBase {
 protected:
    friend class AstNode;
-    static uint32_t     s_userCntGbl;   // Count of which usage of userp() this is
-    static bool         s_userBusy;     // Count is in use
+    static uint32_t s_userCntGbl;  // Count of which usage of userp() this is
+    static bool s_userBusy;  // Count is in use
 public:
-    VNUser3InUse()     { allocate(3, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
-    ~VNUser3InUse()    { free    (3, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    VNUser3InUse()      { allocate(3, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    ~VNUser3InUse()     { free    (3, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void clear() { clearcnt(3, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void check() { checkcnt(3, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
 };
 class VNUser4InUse final : VNUserInUseBase {
 protected:
    friend class AstNode;
-    static uint32_t     s_userCntGbl;   // Count of which usage of userp() this is
-    static bool         s_userBusy;     // Count is in use
+    static uint32_t s_userCntGbl;  // Count of which usage of userp() this is
+    static bool s_userBusy;  // Count is in use
 public:
-    VNUser4InUse()     { allocate(4, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
-    ~VNUser4InUse()    { free    (4, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    VNUser4InUse()      { allocate(4, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    ~VNUser4InUse()     { free    (4, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void clear() { clearcnt(4, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void check() { checkcnt(4, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
 };
 class VNUser5InUse final : VNUserInUseBase {
 protected:
    friend class AstNode;
-    static uint32_t     s_userCntGbl;   // Count of which usage of userp() this is
-    static bool         s_userBusy;     // Count is in use
+    static uint32_t s_userCntGbl;  // Count of which usage of userp() this is
+    static bool s_userBusy;  // Count is in use
 public:
-    VNUser5InUse()     { allocate(5, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
-    ~VNUser5InUse()    { free    (5, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    VNUser5InUse()      { allocate(5, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
+    ~VNUser5InUse()     { free    (5, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void clear() { clearcnt(5, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
    static void check() { checkcnt(5, s_userCntGbl/*ref*/, s_userBusy/*ref*/); }
 };
@@ -1624,7 +1702,7 @@ public:
    const char* typeName() const VL_MT_SAFE { return type().ascii(); }  // See also prettyTypeName
    AstNode* nextp() const VL_MT_STABLE { return m_nextp; }
    AstNode* backp() const VL_MT_STABLE { return m_backp; }
-    AstNode* abovep() const;  // Parent node above, only when no nextp() as otherwise slow
+    AstNode* abovep() const;  // Get parent node above, only for list head and tail
    AstNode* op1p() const VL_MT_STABLE { return m_op1p; }
    AstNode* op2p() const VL_MT_STABLE { return m_op2p; }
    AstNode* op3p() const VL_MT_STABLE { return m_op3p; }
@@ -1676,7 +1754,7 @@ public:
    static constexpr int INSTR_COUNT_PLI = 20;  // PLI routines

    // ACCESSORS
-    virtual string name() const VL_MT_STABLE { return ""; }
+    virtual string name() const { return ""; }
    virtual string origName() const { return ""; }
    virtual void name(const string& name) {
        this->v3fatalSrc("name() called on object without name() method");
@@ -1684,7 +1762,7 @@ public:
    virtual void tag(const string& text) {}
    virtual string tag() const { return ""; }
    virtual string verilogKwd() const { return ""; }
-    string nameProtect() const VL_MT_STABLE;  // Name with --protect-id applied
+    string nameProtect() const;  // Name with --protect-id applied
    string origNameProtect() const;  // origName with --protect-id applied
    string shortName() const;  // Name with __PVT__ removed for concatenating scopes
    static string dedotName(const string& namein);  // Name with dots removed
@@ -1697,10 +1775,10 @@ public:
    static string encodeName(const string& namein);
    static string encodeNumber(int64_t num);  // Encode number into internal C representation
    static string vcdName(const string& namein);  // Name for printing out to vcd files
-    string prettyName() const VL_MT_STABLE { return prettyName(name()); }
+    string prettyName() const { return prettyName(name()); }
    string prettyNameQ() const { return prettyNameQ(name()); }
    // "VARREF" for error messages (NOT dtype's pretty name)
-    string prettyTypeName() const VL_MT_STABLE;
+    string prettyTypeName() const;
    virtual string prettyOperatorName() const { return "operator " + prettyTypeName(); }
    FileLine* fileline() const VL_MT_SAFE { return m_fileline; }
    void fileline(FileLine* fl) { m_fileline = fl; }
@@ -1732,74 +1810,74 @@ public:
    inline bool isString() const VL_MT_STABLE;

    // clang-format off
-    VNUser      user1u() const VL_MT_STABLE {
+    VNUser user1u() const VL_MT_STABLE {
        // Slows things down measurably, so disabled by default
-        //UASSERT_STATIC(VNUser1InUse::s_userBusy, "userp set w/o busy");
-        return ((m_user1Cnt==VNUser1InUse::s_userCntGbl) ? m_user1u : VNUser{0});
+        //UASSERT_STATIC(VNUser1InUse::s_userBusy, "user1p used without AstUserInUse");
+        return ((m_user1Cnt == VNUser1InUse::s_userCntGbl) ? m_user1u : VNUser{0});
    }
-    AstNode*    user1p() const VL_MT_STABLE { return user1u().toNodep(); }
-    void        user1u(const VNUser& user) { m_user1u=user; m_user1Cnt=VNUser1InUse::s_userCntGbl; }
-    void        user1p(void* userp) { user1u(VNUser{userp}); }
-    int         user1() const { return user1u().toInt(); }
-    void        user1(int val) { user1u(VNUser{val}); }
-    int         user1Inc(int val=1) { int v=user1(); user1(v+val); return v; }
-    int         user1SetOnce() { int v=user1(); if (!v) user1(1); return v; }  // Better for cache than user1Inc()
+    AstNode* user1p() const VL_MT_STABLE { return user1u().toNodep(); }
+    void user1u(const VNUser& user) { m_user1u = user; m_user1Cnt = VNUser1InUse::s_userCntGbl; }
+    void user1p(void* userp) { user1u(VNUser{userp}); }
+    void user1(int val) { user1u(VNUser{val}); }
+    int user1() const { return user1u().toInt(); }
+    int user1Inc(int val = 1) { int v = user1(); user1(v + val); return v; }
+    int user1SetOnce() { int v = user1(); if (!v) user1(1); return v; }  // Better for cache than user1Inc()
    static void user1ClearTree() { VNUser1InUse::clear(); }  // Clear userp()'s across the entire tree

-    VNUser      user2u() const VL_MT_STABLE {
+    VNUser user2u() const VL_MT_STABLE {
        // Slows things down measurably, so disabled by default
-        //UASSERT_STATIC(VNUser2InUse::s_userBusy, "userp set w/o busy");
-        return ((m_user2Cnt==VNUser2InUse::s_userCntGbl) ? m_user2u : VNUser{0});
+        //UASSERT_STATIC(VNUser2InUse::s_userBusy, "user2p used without AstUserInUse");
+        return ((m_user2Cnt == VNUser2InUse::s_userCntGbl) ? m_user2u : VNUser{0});
    }
-    AstNode*    user2p() const VL_MT_STABLE { return user2u().toNodep(); }
-    void        user2u(const VNUser& user) { m_user2u=user; m_user2Cnt=VNUser2InUse::s_userCntGbl; }
-    void        user2p(void* userp) { user2u(VNUser{userp}); }
-    int         user2() const { return user2u().toInt(); }
-    void        user2(int val) { user2u(VNUser{val}); }
-    int         user2Inc(int val=1) { int v=user2(); user2(v+val); return v; }
-    int         user2SetOnce() { int v=user2(); if (!v) user2(1); return v; }  // Better for cache than user2Inc()
+    AstNode* user2p() const VL_MT_STABLE { return user2u().toNodep(); }
+    void user2u(const VNUser& user) { m_user2u = user; m_user2Cnt = VNUser2InUse::s_userCntGbl; }
+    void user2p(void* userp) { user2u(VNUser{userp}); }
+    void user2(int val) { user2u(VNUser{val}); }
+    int user2() const { return user2u().toInt(); }
+    int user2Inc(int val = 1) { int v = user2(); user2(v + val); return v; }
+    int user2SetOnce() { int v = user2(); if (!v) user2(1); return v; }  // Better for cache than user2Inc()
    static void user2ClearTree() { VNUser2InUse::clear(); }  // Clear userp()'s across the entire tree

-    VNUser      user3u() const VL_MT_STABLE {
+    VNUser user3u() const VL_MT_STABLE {
        // Slows things down measurably, so disabled by default
-        //UASSERT_STATIC(VNUser3InUse::s_userBusy, "userp set w/o busy");
-        return ((m_user3Cnt==VNUser3InUse::s_userCntGbl) ? m_user3u : VNUser{0});
+        //UASSERT_STATIC(VNUser3InUse::s_userBusy, "user3p used without AstUserInUse");
+        return ((m_user3Cnt == VNUser3InUse::s_userCntGbl) ? m_user3u : VNUser{0});
    }
-    AstNode*    user3p() const VL_MT_STABLE { return user3u().toNodep(); }
-    void        user3u(const VNUser& user) { m_user3u=user; m_user3Cnt=VNUser3InUse::s_userCntGbl; }
-    void        user3p(void* userp) { user3u(VNUser{userp}); }
-    int         user3() const { return user3u().toInt(); }
-    void        user3(int val) { user3u(VNUser{val}); }
-    int         user3Inc(int val=1) { int v=user3(); user3(v+val); return v; }
-    int         user3SetOnce() { int v=user3(); if (!v) user3(1); return v; }  // Better for cache than user3Inc()
+    AstNode* user3p() const VL_MT_STABLE { return user3u().toNodep(); }
+    void user3u(const VNUser& user) { m_user3u = user; m_user3Cnt = VNUser3InUse::s_userCntGbl; }
+    void user3p(void* userp) { user3u(VNUser{userp}); }
+    void user3(int val) { user3u(VNUser{val}); }
+    int user3() const { return user3u().toInt(); }
+    int user3Inc(int val = 1) { int v = user3(); user3(v + val); return v; }
+    int user3SetOnce() { int v = user3(); if (!v) user3(1); return v; }  // Better for cache than user3Inc()
    static void user3ClearTree() { VNUser3InUse::clear(); }  // Clear userp()'s across the entire tree

-    VNUser      user4u() const VL_MT_STABLE {
+    VNUser user4u() const VL_MT_STABLE {
        // Slows things down measurably, so disabled by default
-        //UASSERT_STATIC(VNUser4InUse::s_userBusy, "userp set w/o busy");
-        return ((m_user4Cnt==VNUser4InUse::s_userCntGbl) ? m_user4u : VNUser{0});
+        //UASSERT_STATIC(VNUser4InUse::s_userBusy, "user4p used without AstUserInUse");
+        return ((m_user4Cnt == VNUser4InUse::s_userCntGbl) ? m_user4u : VNUser{0});
    }
-    AstNode*    user4p() const VL_MT_STABLE { return user4u().toNodep(); }
-    void        user4u(const VNUser& user) { m_user4u=user; m_user4Cnt=VNUser4InUse::s_userCntGbl; }
-    void        user4p(void* userp) { user4u(VNUser{userp}); }
-    int         user4() const { return user4u().toInt(); }
-    void        user4(int val) { user4u(VNUser{val}); }
-    int         user4Inc(int val=1) { int v=user4(); user4(v+val); return v; }
-    int         user4SetOnce() { int v=user4(); if (!v) user4(1); return v; }  // Better for cache than user4Inc()
+    AstNode* user4p() const VL_MT_STABLE { return user4u().toNodep(); }
+    void user4u(const VNUser& user) { m_user4u = user; m_user4Cnt = VNUser4InUse::s_userCntGbl; }
+    void user4p(void* userp) { user4u(VNUser{userp}); }
+    void user4(int val) { user4u(VNUser{val}); }
+    int user4() const { return user4u().toInt(); }
+    int user4Inc(int val = 1) { int v = user4(); user4(v + val); return v; }
+    int user4SetOnce() { int v = user4(); if (!v) user4(1); return v; }  // Better for cache than user4Inc()
    static void user4ClearTree() { VNUser4InUse::clear(); }  // Clear userp()'s across the entire tree

-    VNUser      user5u() const VL_MT_STABLE {
+    VNUser user5u() const VL_MT_STABLE {
        // Slows things down measurably, so disabled by default
-        //UASSERT_STATIC(VNUser5InUse::s_userBusy, "userp set w/o busy");
-        return ((m_user5Cnt==VNUser5InUse::s_userCntGbl) ? m_user5u : VNUser{0});
+        //UASSERT_STATIC(VNUser5InUse::s_userBusy, "user5p used without AstUserInUse");
+        return ((m_user5Cnt == VNUser5InUse::s_userCntGbl) ? m_user5u : VNUser{0});
    }
-    AstNode*    user5p() const VL_MT_STABLE { return user5u().toNodep(); }
-    void        user5u(const VNUser& user) { m_user5u=user; m_user5Cnt=VNUser5InUse::s_userCntGbl; }
-    void        user5p(void* userp) { user5u(VNUser{userp}); }
-    int         user5() const { return user5u().toInt(); }
-    void        user5(int val) { user5u(VNUser{val}); }
-    int         user5Inc(int val=1) { int v=user5(); user5(v+val); return v; }
-    int         user5SetOnce() { int v=user5(); if (!v) user5(1); return v; }  // Better for cache than user5Inc()
+    AstNode* user5p() const VL_MT_STABLE { return user5u().toNodep(); }
+    void user5u(const VNUser& user) { m_user5u = user; m_user5Cnt = VNUser5InUse::s_userCntGbl; }
+    void user5p(void* userp) { user5u(VNUser{userp}); }
+    void user5(int val) { user5u(VNUser{val}); }
+    int user5() const { return user5u().toInt(); }
+    int user5Inc(int val = 1) { int v = user5(); user5(v + val); return v; }
+    int user5SetOnce() { int v = user5(); if (!v) user5(1); return v; }  // Better for cache than user5Inc()
    static void user5ClearTree() { VNUser5InUse::clear(); }  // Clear userp()'s across the entire tree
    // clang-format on

@@ -1817,6 +1895,7 @@ public:

    // ACCESSORS for specific types
    // Alas these can't be virtual or they break when passed a nullptr
+    inline bool isClassHandleValue() const;
    inline bool isZero() const;
    inline bool isOne() const;
    inline bool isNeqZero() const;
@@ -1856,6 +1935,7 @@ public:
    void dtypeSetUInt64() { dtypep(findUInt64DType()); }  // Twostate
    void dtypeSetEmptyQueue() { dtypep(findEmptyQueueDType()); }
    void dtypeSetVoid() { dtypep(findVoidDType()); }
+    void dtypeSetStream() { dtypep(findStreamDType()); }

    // Data type locators
    AstNodeDType* findBitDType() const { return findBasicDType(VBasicDTypeKwd::LOGIC); }
@@ -1867,6 +1947,7 @@ public:
    AstNodeDType* findCHandleDType() const { return findBasicDType(VBasicDTypeKwd::CHANDLE); }
    AstNodeDType* findEmptyQueueDType() const;
    AstNodeDType* findVoidDType() const;
+    AstNodeDType* findStreamDType() const;
    AstNodeDType* findQueueIndexDType() const;
    AstNodeDType* findBitDType(int width, int widthMin, VSigning numeric) const;
    AstNodeDType* findLogicDType(int width, int widthMin, VSigning numeric) const;
@@ -1877,9 +1958,9 @@ public:
    static AstBasicDType* findInsertSameDType(AstBasicDType* nodep);

    // METHODS - dump and error
-    void v3errorEnd(std::ostringstream& str) const VL_REQUIRES(V3Error::s().m_mutex);
+    void v3errorEnd(std::ostringstream& str) const VL_RELEASE(V3Error::s().m_mutex);
    void v3errorEndFatal(std::ostringstream& str) const VL_ATTR_NORETURN
-        VL_REQUIRES(V3Error::s().m_mutex);
+        VL_RELEASE(V3Error::s().m_mutex);
    string warnContextPrimary() const VL_REQUIRES(V3Error::s().m_mutex) {
        return fileline()->warnContextPrimary();
    }
@@ -1915,14 +1996,8 @@ public:
    // Iterate and insert - assumes tree format
    virtual void addNextStmt(AstNode* newp,
                             AstNode* belowp);  // When calling, "this" is second argument
-    virtual void addBeforeStmt(AstNode* newp,
-                               AstNode* belowp);  // When calling, "this" is second argument

    // METHODS - Iterate on a tree
-    // Clone or return nullptr if nullptr
-    static AstNode* cloneTreeNull(AstNode* nodep, bool cloneNextLink) {
-        return nodep ? nodep->cloneTree(cloneNextLink) : nullptr;
-    }
    AstNode* cloneTree(bool cloneNextLink);  // Not const, as sets clonep() on original nodep
    bool gateTree() { return gateTreeIter(); }  // Is tree isGateOptimizable?
    inline bool sameTree(const AstNode* node2p) const;  // Does tree of this == node2p?
@@ -2056,24 +2131,34 @@ public:
        return nodep && privateTypeTest<T>(nodep) ? reinterpret_cast<const T*>(nodep) : nullptr;
    }

-    // For use via the VN_AS macro only
+    // For use via privateAs or the VN_DBG_AS macro only
    template <typename T, typename E>
-    static T* privateAs(AstNode* nodep) VL_PURE {
+    static T* unsafePrivateAs(AstNode* nodep) VL_PURE {
        static_assert(!uselessCast<T, E>(), "Unnecessary VN_AS, node known to have target type.");
        static_assert(!impossibleCast<T, E>(), "Unnecessary VN_AS, node cannot be this type.");
-        UASSERT_OBJ(!nodep || privateTypeTest<T>(nodep), nodep,
-                    "AstNode is not of expected type, but instead has type '" << nodep->typeName()
-                                                                              << "'");
        return reinterpret_cast<T*>(nodep);
    }
    template <typename T, typename E>
-    static const T* privateAs(const AstNode* nodep) VL_PURE {
+    static const T* unsafePrivateAs(const AstNode* nodep) VL_PURE {
        static_assert(!uselessCast<T, E>(), "Unnecessary VN_AS, node known to have target type.");
        static_assert(!impossibleCast<T, E>(), "Unnecessary VN_AS, node cannot be this type.");
+        return reinterpret_cast<const T*>(nodep);
+    }
+
+    // For use via the VN_AS macro only
+    template <typename T, typename E>
+    static T* privateAs(AstNode* nodep) VL_PURE {
        UASSERT_OBJ(!nodep || privateTypeTest<T>(nodep), nodep,
                    "AstNode is not of expected type, but instead has type '" << nodep->typeName()
                                                                              << "'");
-        return reinterpret_cast<const T*>(nodep);
+        return unsafePrivateAs<T, E>(nodep);
+    }
+    template <typename T, typename E>
+    static const T* privateAs(const AstNode* nodep) VL_PURE {
+        UASSERT_OBJ(!nodep || privateTypeTest<T>(nodep), nodep,
+                    "AstNode is not of expected type, but instead has type '" << nodep->typeName()
+                                                                              << "'");
+        return unsafePrivateAs<T, E>(nodep);
    }

    // Predicate that returns true if the given 'nodep' might have a descendant of type 'T_Node'.
@@ -41,6 +41,10 @@ bool AstNode::isString() const VL_MT_STABLE {
 }
 bool AstNode::isSigned() const VL_MT_STABLE { return dtypep() && dtypep()->isSigned(); }

+bool AstNode::isClassHandleValue() const {
+    return (VN_IS(this, Const) && VN_AS(this, Const)->num().isNull())
+           || VN_IS(dtypep(), ClassRefDType);
+}
 bool AstNode::isZero() const {
    return (VN_IS(this, Const) && VN_AS(this, Const)->num().isEqZero());
 }
@@ -150,12 +154,15 @@ AstCExpr::AstCExpr(FileLine* fl, const string& textStmt, int setwidth, bool clea
 }

 AstVarRef::AstVarRef(FileLine* fl, AstVar* varp, const VAccess& access)
-    : ASTGEN_SUPER_VarRef(fl, varp->name(), varp, access) {}
+    : ASTGEN_SUPER_VarRef(fl, varp, access) {}
 // This form only allowed post-link (see above)
 AstVarRef::AstVarRef(FileLine* fl, AstVarScope* varscp, const VAccess& access)
-    : ASTGEN_SUPER_VarRef(fl, varscp->varp()->name(), varscp->varp(), access) {
+    : ASTGEN_SUPER_VarRef(fl, varscp->varp(), access) {
    varScopep(varscp);
 }
+
+string AstVarRef::name() const { return varp() ? varp()->name() : "<null>"; }
+
 bool AstVarRef::same(const AstVarRef* samep) const {
    if (varScopep()) {
        return (varScopep() == samep->varScopep() && access() == samep->access());
@@ -169,13 +176,14 @@ bool AstVarRef::sameNoLvalue(AstVarRef* samep) const {
        return (varScopep() == samep->varScopep());
    } else {
        return (selfPointer() == samep->selfPointer()
-                && (!selfPointer().empty() || !samep->selfPointer().empty())
+                && (!selfPointer().isEmpty() || !samep->selfPointer().isEmpty())
                && varp()->name() == samep->varp()->name());
    }
 }

 AstVarXRef::AstVarXRef(FileLine* fl, AstVar* varp, const string& dotted, const VAccess& access)
-    : ASTGEN_SUPER_VarXRef(fl, varp->name(), varp, access)
+    : ASTGEN_SUPER_VarXRef(fl, varp, access)
+    , m_name{varp->name()}
    , m_dotted{dotted} {
    dtypeFrom(varp);
 }
@@ -126,13 +126,13 @@ public:
    const char* charIQWN() const {
        return (isString() ? "N" : isWide() ? "W" : isQuad() ? "Q" : "I");
    }
-    string cType(const string& name, bool forFunc, bool isRef) const VL_MT_STABLE;
+    string cType(const string& name, bool forFunc, bool isRef) const;
    // Represents a C++ LiteralType? (can be constexpr)
    bool isLiteralType() const VL_MT_STABLE;

 private:
    class CTypeRecursed;
-    CTypeRecursed cTypeRecurse(bool compound) const VL_MT_STABLE;
+    CTypeRecursed cTypeRecurse(bool compound) const;
 };
 class AstNodeArrayDType VL_NOT_FINAL : public AstNodeDType {
    // Array data type, ie "some_dtype var_name [2:0]"
@@ -199,12 +199,9 @@ class AstNodeUOrStructDType VL_NOT_FINAL : public AstNodeDType {
    // A struct or union; common handling
    // @astgen op1 := membersp : List[AstMemberDType]
 private:
-    // TYPES
-    using MemberNameMap = std::map<const std::string, AstMemberDType*>;
    // MEMBERS
    string m_name;  // Name from upper typedef, if any
-    AstNodeModule* m_classOrPackagep = nullptr;  // Package hierarchy
-    MemberNameMap m_members;
+    AstNodeModule* m_classOrPackagep = nullptr;  // Package it will be emitted with
    const int m_uniqueNum;
    bool m_packed;
    bool m_isFourstate = false;  // V3Width computes
@@ -250,12 +247,6 @@ public:
    static bool packedUnsup() { return true; }
    void isFourstate(bool flag) { m_isFourstate = flag; }
    bool isFourstate() const override VL_MT_SAFE { return m_isFourstate; }
-    void clearCache() { m_members.clear(); }
-    void repairMemberCache();
-    AstMemberDType* findMember(const string& name) const {
-        const auto it = m_members.find(name);
-        return (it == m_members.end()) ? nullptr : it->second;
-    }
    static int lo() { return 0; }
    int hi() const { return dtypep()->width() - 1; }  // Packed classes look like arrays
    VNumRange declRange() const { return VNumRange{hi(), lo()}; }
@@ -438,6 +429,7 @@ public:
    bool isEvent() const VL_MT_SAFE { return keyword() == VBasicDTypeKwd::EVENT; }
    bool isTriggerVec() const VL_MT_SAFE { return keyword() == VBasicDTypeKwd::TRIGGERVEC; }
    bool isForkSync() const VL_MT_SAFE { return keyword() == VBasicDTypeKwd::FORK_SYNC; }
+    bool isProcessRef() const VL_MT_SAFE { return keyword() == VBasicDTypeKwd::PROCESS_REFERENCE; }
    bool isDelayScheduler() const VL_MT_SAFE {
        return keyword() == VBasicDTypeKwd::DELAY_SCHEDULER;
    }
@@ -887,6 +879,7 @@ public:
        if (m_refDTypep && m_refDTypep->clonep()) m_refDTypep = m_refDTypep->clonep();
    }
    AstNodeDType* getChildDTypep() const override { return childDTypep(); }
+    AstNodeUOrStructDType* getChildStructp() const;
    AstNodeDType* subDTypep() const override VL_MT_STABLE {
        return m_refDTypep ? m_refDTypep : childDTypep();
    }
@@ -1060,7 +1053,7 @@ private:
    // Post-width typedefs are removed and point to type directly
    AstNodeDType* m_refDTypep = nullptr;  // data type pointed to, BELOW the AstTypedef
    string m_name;  // Name of an AstTypedef
-    AstNodeModule* m_classOrPackagep = nullptr;  // Package hierarchy
+    AstNodeModule* m_classOrPackagep = nullptr;  // Class/package in which it was defined
 public:
    AstRefDType(FileLine* fl, const string& name)
        : ASTGEN_SUPER_RefDType(fl)
@@ -1075,6 +1068,7 @@ public:
    AstRefDType(FileLine* fl, FlagTypeOfExpr, AstNode* typeofp)
        : ASTGEN_SUPER_RefDType(fl) {
        this->typeofp(typeofp);
+        if (AstNodeDType* const dtp = VN_CAST(typeofp, NodeDType)) refDTypep(dtp);
    }
    ASTGEN_MEMBERS_AstRefDType;
    // METHODS
@@ -1186,6 +1180,34 @@ public:
    int widthTotalBytes() const override { return sizeof(std::map<std::string, std::string>); }
    bool isCompound() const override { return true; }
 };
+class AstStreamDType final : public AstNodeDType {
+    // Stream data type, used only as data type of stream operations
+    // Should behave like AstPackArrayDType, but it doesn't have a size
+public:
+    explicit AstStreamDType(FileLine* fl)
+        : ASTGEN_SUPER_StreamDType(fl) {
+        dtypep(this);
+    }
+    ASTGEN_MEMBERS_AstStreamDType;
+    void dumpSmall(std::ostream& str) const override;
+    bool hasDType() const override { return true; }
+    bool maybePointedTo() const override { return true; }
+    bool undead() const override { return true; }
+    AstNodeDType* subDTypep() const override VL_MT_SAFE { return nullptr; }
+    AstNodeDType* virtRefDTypep() const override { return nullptr; }
+    void virtRefDTypep(AstNodeDType* nodep) override {}
+    bool similarDType(const AstNodeDType* samep) const override { return this == samep; }
+    AstBasicDType* basicp() const override VL_MT_STABLE { return nullptr; }
+    // cppcheck-suppress csyleCast
+    AstNodeDType* skipRefp() const override VL_MT_STABLE { return (AstNodeDType*)this; }
+    // cppcheck-suppress csyleCast
+    AstNodeDType* skipRefToConstp() const override { return (AstNodeDType*)this; }
+    // cppcheck-suppress csyleCast
+    AstNodeDType* skipRefToEnump() const override { return (AstNodeDType*)this; }
+    int widthAlignBytes() const override { return 1; }
+    int widthTotalBytes() const override { return 1; }
+    bool isCompound() const override { return false; }
+};
 class AstUnsizedArrayDType final : public AstNodeDType {
    // Unsized/open-range Array data type, ie "some_dtype var_name []"
    // @astgen op1 := childDTypep : Optional[AstNodeDType] // moved to refDTypep() in V3Width
@@ -214,7 +214,7 @@ class AstNodeFTaskRef VL_NOT_FINAL : public AstNodeExpr {
    // @astgen op3 := pinsp : List[AstNodeExpr]
    // @astgen op4 := scopeNamep : Optional[AstScopeName]
    AstNodeFTask* m_taskp = nullptr;  // [AfterLink] Pointer to task referenced
-    AstNodeModule* m_classOrPackagep = nullptr;  // Package hierarchy
+    AstNodeModule* m_classOrPackagep = nullptr;  // Class/package of the task
    string m_name;  // Name of variable
    string m_dotted;  // Dotted part of scope the name()ed task/func is under or ""
    string m_inlinedDots;  // Dotted hierarchy flattened out
@@ -360,14 +360,8 @@ class AstNodeCond VL_NOT_FINAL : public AstNodeTriop {
    // @astgen alias op2 := thenp
    // @astgen alias op3 := elsep
 protected:
-    AstNodeCond(VNType t, FileLine* fl, AstNodeExpr* condp, AstNodeExpr* thenp, AstNodeExpr* elsep)
-        : AstNodeTriop{t, fl, condp, thenp, elsep} {
-        if (thenp) {
-            dtypeFrom(thenp);
-        } else if (elsep) {
-            dtypeFrom(elsep);
-        }
-    }
+    AstNodeCond(VNType t, FileLine* fl, AstNodeExpr* condp, AstNodeExpr* thenp,
+                AstNodeExpr* elsep);

 public:
    ASTGEN_MEMBERS_AstNodeCond;
@@ -449,21 +443,19 @@ class AstNodeVarRef VL_NOT_FINAL : public AstNodeExpr {
    VAccess m_access;  // Left hand side assignment
    AstVar* m_varp;  // [AfterLink] Pointer to variable itself
    AstVarScope* m_varScopep = nullptr;  // Varscope for hierarchy
-    AstNodeModule* m_classOrPackagep = nullptr;  // Package hierarchy
-    string m_name;  // Name of variable
-    string m_selfPointer;  // Output code object pointer (e.g.: 'this')
-
+    AstNodeModule* m_classOrPackagep = nullptr;  // Class/package of the variable
+    VSelfPointerText m_selfPointer
+        = VSelfPointerText{VSelfPointerText::Empty()};  // Output code object
+                                                        // pointer (e.g.: 'this')
 protected:
-    AstNodeVarRef(VNType t, FileLine* fl, const string& name, const VAccess& access)
+    AstNodeVarRef(VNType t, FileLine* fl, const VAccess& access)
        : AstNodeExpr{t, fl}
-        , m_access{access}
-        , m_name{name} {
+        , m_access{access} {
        varp(nullptr);
    }
-    AstNodeVarRef(VNType t, FileLine* fl, const string& name, AstVar* varp, const VAccess& access)
+    AstNodeVarRef(VNType t, FileLine* fl, AstVar* varp, const VAccess& access)
        : AstNodeExpr{t, fl}
-        , m_access{access}
-        , m_name{name} {
+        , m_access{access} {
        // May have varp==nullptr
        this->varp(varp);
    }
@@ -474,8 +466,6 @@ public:
    const char* broken() const override;
    int instrCount() const override { return widthInstrs(); }
    void cloneRelink() override;
-    string name() const override VL_MT_STABLE { return m_name; }  // * = Var name
-    void name(const string& name) override { m_name = name; }
    VAccess access() const { return m_access; }
    void access(const VAccess& flag) { m_access = flag; }  // Avoid using this; Set in constructor
    AstVar* varp() const { return m_varp; }  // [After Link] Pointer to variable
@@ -485,9 +475,11 @@ public:
    }
    AstVarScope* varScopep() const { return m_varScopep; }
    void varScopep(AstVarScope* varscp) { m_varScopep = varscp; }
-    string selfPointer() const { return m_selfPointer; }
-    void selfPointer(const string& value) { m_selfPointer = value; }
-    string selfPointerProtect(bool useSelfForThis) const;
+    const VSelfPointerText& selfPointer() const { return m_selfPointer; }
+    void selfPointer(const VSelfPointerText& selfPointer) { m_selfPointer = selfPointer; }
+    string selfPointerProtect(bool useSelfForThis) const {
+        return selfPointer().protect(useSelfForThis, protect());
+    }
    AstNodeModule* classOrPackagep() const { return m_classOrPackagep; }
    void classOrPackagep(AstNodeModule* nodep) { m_classOrPackagep = nodep; }
    // Know no children, and hot function, so skip iterator for speed
@@ -593,21 +585,13 @@ class AstCMethodHard final : public AstNodeExpr {
    string m_name;  // Name of method
    bool m_pure = false;  // Pure optimizable
 public:
-    AstCMethodHard(FileLine* fl, AstNodeExpr* fromp, VFlagChildDType, const string& name,
-                   AstNodeExpr* pinsp = nullptr)
-        : ASTGEN_SUPER_CMethodHard(fl)
-        , m_name{name} {
-        // TODO: this constructor is exactly the same as the other, bar the ignored tag argument
-        this->fromp(fromp);
-        this->addPinsp(pinsp);
-        dtypep(nullptr);  // V3Width will resolve
-    }
    AstCMethodHard(FileLine* fl, AstNodeExpr* fromp, const string& name,
                   AstNodeExpr* pinsp = nullptr)
        : ASTGEN_SUPER_CMethodHard(fl)
        , m_name{name} {
        this->fromp(fromp);
        this->addPinsp(pinsp);
+        setPurity();
    }
    ASTGEN_MEMBERS_AstCMethodHard;
    string name() const override VL_MT_STABLE { return m_name; }  // * = Var name
@@ -617,11 +601,13 @@ public:
        return (m_name == asamep->m_name);
    }
    bool isPure() const override { return m_pure; }
-    void pure(bool flag) { m_pure = flag; }
    int instrCount() const override;
    string emitVerilog() override { V3ERROR_NA_RETURN(""); }
    string emitC() override { V3ERROR_NA_RETURN(""); }
    bool cleanOut() const override { return true; }
+
+private:
+    void setPurity();
 };
 class AstCast final : public AstNodeExpr {
    // Cast to appropriate data type
@@ -721,7 +707,7 @@ class AstClassOrPackageRef final : public AstNodeExpr {
 private:
    string m_name;
    // Node not NodeModule to appease some early parser usage
-    AstNode* m_classOrPackageNodep;  // Package hierarchy
+    AstNode* m_classOrPackageNodep;  // Pointer to class/package referenced
 public:
    AstClassOrPackageRef(FileLine* fl, const string& name, AstNode* classOrPackageNodep,
                         AstPin* paramsp)
@@ -975,6 +961,20 @@ public:
        , m_num(this, 1, on) {
        dtypeSetBit();
    }
+    class All0 {};
+    AstConst(FileLine* fl, All0)
+        : ASTGEN_SUPER_Const(fl)
+        , m_num(this, "'0") {
+        initWithNumber();
+        fl->warnOff(V3ErrorCode::NEWERSTD, true);
+    }
+    class All1 {};
+    AstConst(FileLine* fl, All1)
+        : ASTGEN_SUPER_Const(fl)
+        , m_num(this, "'1") {
+        initWithNumber();
+        fl->warnOff(V3ErrorCode::NEWERSTD, true);
+    }
    class Null {};
    AstConst(FileLine* fl, Null)
        : ASTGEN_SUPER_Const(fl)
@@ -1010,6 +1010,34 @@ public:
    // May return nullptr on parse failure.
    static AstConst* parseParamLiteral(FileLine* fl, const string& literal);
 };
+class AstCvtDynArrayToPacked final : public AstNodeExpr {
+    // Cast from dynamic queue data type to packed array
+    // @astgen op1 := fromp : AstNodeExpr
+public:
+    AstCvtDynArrayToPacked(FileLine* fl, AstNodeExpr* fromp, AstNodeDType* dtp)
+        : ASTGEN_SUPER_CvtDynArrayToPacked(fl) {
+        this->fromp(fromp);
+        dtypeFrom(dtp);
+    }
+    ASTGEN_MEMBERS_AstCvtDynArrayToPacked;
+    string emitVerilog() override { V3ERROR_NA_RETURN(""); }
+    string emitC() override { V3ERROR_NA_RETURN(""); }
+    bool cleanOut() const override { return true; }
+};
+class AstCvtPackedToDynArray final : public AstNodeExpr {
+    // Cast from packed array to dynamic queue data type
+    // @astgen op1 := fromp : AstNodeExpr
+public:
+    AstCvtPackedToDynArray(FileLine* fl, AstNodeExpr* fromp, AstNodeDType* dtp)
+        : ASTGEN_SUPER_CvtPackedToDynArray(fl) {
+        this->fromp(fromp);
+        dtypeFrom(dtp);
+    }
+    ASTGEN_MEMBERS_AstCvtPackedToDynArray;
+    string emitVerilog() override { V3ERROR_NA_RETURN(""); }
+    string emitC() override { V3ERROR_NA_RETURN(""); }
+    bool cleanOut() const override { return true; }
+};
 class AstDot final : public AstNodeExpr {
    // A dot separating paths in an AstVarXRef, AstFuncRef or AstTaskRef
    // These are eliminated in the link stage
@@ -1048,7 +1076,7 @@ public:
 };
 class AstEnumItemRef final : public AstNodeExpr {
    AstEnumItem* m_itemp;  // [AfterLink] Pointer to item
-    AstNodeModule* m_classOrPackagep;  // Package hierarchy
+    AstNodeModule* m_classOrPackagep;  // Class/package in which it was defined
 public:
    AstEnumItemRef(FileLine* fl, AstEnumItem* itemp, AstNodeModule* classOrPackagep)
        : ASTGEN_SUPER_EnumItemRef(fl)
@@ -1092,7 +1120,8 @@ public:
    // METHODS
    string emitVerilog() override { V3ERROR_NA_RETURN(""); }
    string emitC() override { V3ERROR_NA_RETURN(""); }
-    bool cleanOut() const override { return false; }
+    bool cleanOut() const override { return true; }
+    bool isPure() const override { return false; }
    bool same(const AstNode*) const override { return true; }
 };
 class AstFError final : public AstNodeExpr {
@@ -1299,7 +1328,9 @@ public:
    bool cleanOut() const override { return true; }
 };
 class AstImplication final : public AstNodeExpr {
-    // Verilog |-> |=>
+    // Verilog Implication Operator
+    // Nonoverlapping "|=>"
+    // Overlapping "|->" (doesn't currently use this - might make new Ast type)
    // @astgen op1 := lhsp : AstNodeExpr
    // @astgen op2 := rhsp : AstNodeExpr
    // @astgen op3 := sentreep : Optional[AstSenTree]
@@ -1414,6 +1445,7 @@ class AstMemberSel final : public AstNodeExpr {
    // @astgen op1 := fromp : AstNodeExpr
    // Don't need the class we are extracting from, as the "fromp()"'s datatype can get us to it
    string m_name;
+    VAccess m_access;  // Read or write, as in AstNodeVarRef
    AstVar* m_varp = nullptr;  // Post link, variable within class that is target of selection
 public:
    AstMemberSel(FileLine* fl, AstNodeExpr* fromp, VFlagChildDType, const string& name)
@@ -1434,10 +1466,12 @@ public:
    void dump(std::ostream& str) const override;
    string name() const override VL_MT_STABLE { return m_name; }
    void name(const string& name) override { m_name = name; }
+    VAccess access() const { return m_access; }
+    void access(const VAccess& flag) { m_access = flag; }
    string emitVerilog() override { V3ERROR_NA_RETURN(""); }
    string emitC() override { V3ERROR_NA_RETURN(""); }
-    bool cleanOut() const override { return false; }
-    bool same(const AstNode* samep) const override { return true; }  // dtype comparison does it
+    bool cleanOut() const override { return true; }
+    bool same(const AstNode* samep) const override;
    int instrCount() const override { return widthInstrs(); }
    AstVar* varp() const { return m_varp; }
    void varp(AstVar* nodep) { m_varp = nodep; }
@@ -1897,7 +1931,7 @@ public:
    bool same(const AstNode* /*samep*/) const override { return true; }
 };
 class AstStructSel final : public AstNodeExpr {
-    // Unpacked struct member access
+    // Unpacked struct/union member access
    // Parents: math|stmt
    // Children: varref, math
    // @astgen op1 := fromp : AstNodeExpr
@@ -1915,7 +1949,10 @@ public:
    void name(const string& name) override { m_name = name; }
    string emitVerilog() override { V3ERROR_NA_RETURN(""); }
    string emitC() override { V3ERROR_NA_RETURN(""); }
-    bool cleanOut() const override { return false; }
+    bool cleanOut() const override {
+        // Not a union
+        return VN_IS(fromp()->dtypep()->skipRefp(), StructDType);
+    }
    bool same(const AstNode* samep) const override {
        const AstStructSel* const sp = static_cast<const AstStructSel*>(samep);
        return m_name == sp->m_name;
@@ -2108,14 +2145,14 @@ class AstWith final : public AstNodeExpr {
    // Children: expression (equation establishing the with)
    // @astgen op1 := indexArgRefp : AstLambdaArgRef
    // @astgen op2 := valueArgRefp : AstLambdaArgRef
-    // @astgen op3 := exprp : AstNodeExpr
+    // @astgen op3 := exprp : List[AstNode]
 public:
    AstWith(FileLine* fl, AstLambdaArgRef* indexArgRefp, AstLambdaArgRef* valueArgRefp,
            AstNodeExpr* exprp)
        : ASTGEN_SUPER_With(fl) {
        this->indexArgRefp(indexArgRefp);
        this->valueArgRefp(valueArgRefp);
-        this->exprp(exprp);
+        this->addExprp(exprp);
    }
    ASTGEN_MEMBERS_AstWith;
    bool same(const AstNode* /*samep*/) const override { return true; }
@@ -4008,16 +4045,19 @@ public:
 // === AstNodeCCall ===
 class AstCCall final : public AstNodeCCall {
    // C++ function call
-    string m_selfPointer;  // Output code object pointer (e.g.: 'this')
-
+    VSelfPointerText m_selfPointer
+        = VSelfPointerText{VSelfPointerText::Empty()};  // Output code object
+                                                        // pointer (e.g.: 'this')
 public:
    AstCCall(FileLine* fl, AstCFunc* funcp, AstNodeExpr* argsp = nullptr)
        : ASTGEN_SUPER_CCall(fl, funcp, argsp) {}
    ASTGEN_MEMBERS_AstCCall;

-    string selfPointer() const { return m_selfPointer; }
-    void selfPointer(const string& value) { m_selfPointer = value; }
-    string selfPointerProtect(bool useSelfForThis) const;
+    const VSelfPointerText& selfPointer() const { return m_selfPointer; }
+    void selfPointer(const VSelfPointerText& selfPointer) { m_selfPointer = selfPointer; }
+    string selfPointerProtect(bool useSelfForThis) const {
+        return selfPointer().protect(useSelfForThis, protect());
+    }
 };
 class AstCMethodCall final : public AstNodeCCall {
    // C++ method call
@@ -4657,7 +4697,7 @@ public:
    ASTGEN_MEMBERS_AstCvtPackString;
    void numberOperate(V3Number& out, const V3Number& lhs) override { out.opAssign(lhs); }
    string emitVerilog() override { return "%f$_CAST(%l)"; }
-    string emitC() override { return "VL_CVT_PACK_STR_N%lq(%lW, %li)"; }
+    string emitC() override { V3ERROR_NA_RETURN(""); }
    bool cleanOut() const override { return true; }
    bool cleanLhs() const override { return true; }
    bool sizeMattersLhs() const override { return false; }
@@ -5335,15 +5375,15 @@ public:
 class AstVarRef final : public AstNodeVarRef {
    // A reference to a variable (lvalue or rvalue)
 public:
-    AstVarRef(FileLine* fl, const string& name, const VAccess& access)
-        : ASTGEN_SUPER_VarRef(fl, name, nullptr, access) {}
    // This form only allowed post-link because output/wire compression may
    // lead to deletion of AstVar's
    inline AstVarRef(FileLine* fl, AstVar* varp, const VAccess& access);
    // This form only allowed post-link (see above)
    inline AstVarRef(FileLine* fl, AstVarScope* varscp, const VAccess& access);
    ASTGEN_MEMBERS_AstVarRef;
+    inline string name() const override;  // * = Var name
    void dump(std::ostream& str) const override;
+    const char* broken() const override;
    bool same(const AstNode* samep) const override;
    inline bool same(const AstVarRef* samep) const;
    inline bool sameNoLvalue(AstVarRef* samep) const;
@@ -5355,14 +5395,17 @@ public:
 class AstVarXRef final : public AstNodeVarRef {
    // A VarRef to something in another module before AstScope.
    // Includes pin on a cell, as part of a ASSIGN statement to connect I/Os until AstScope
+    string m_name;
    string m_dotted;  // Dotted part of scope the name()'ed reference is under or ""
    string m_inlinedDots;  // Dotted hierarchy flattened out
 public:
    AstVarXRef(FileLine* fl, const string& name, const string& dotted, const VAccess& access)
-        : ASTGEN_SUPER_VarXRef(fl, name, nullptr, access)
+        : ASTGEN_SUPER_VarXRef(fl, nullptr, access)
+        , m_name{name}
        , m_dotted{dotted} {}
    inline AstVarXRef(FileLine* fl, AstVar* varp, const string& dotted, const VAccess& access);
    ASTGEN_MEMBERS_AstVarXRef;
+    string name() const override VL_MT_STABLE { return m_name; }  // * = Var name
    void dump(std::ostream& str) const override;
    string dotted() const { return m_dotted; }
    void dotted(const string& dotted) { m_dotted = dotted; }
@@ -32,7 +32,7 @@

 class AstNodeBlock VL_NOT_FINAL : public AstNode {
    // A Begin/fork block
-    // @astgen op1 := stmtsp : List[AstNode]
+    // @astgen op2 := stmtsp : List[AstNode]
    // Parents: statement
 private:
    string m_name;  // Name of block
@@ -81,11 +81,12 @@ private:
    bool m_isConstructor : 1;  // Class constructor
    bool m_isHideLocal : 1;  // Verilog local
    bool m_isHideProtected : 1;  // Verilog protected
-    bool m_pure : 1;  // DPI import pure (vs. virtual pure)
+    bool m_dpiPure : 1;  // DPI import pure (vs. virtual pure)
    bool m_pureVirtual : 1;  // Pure virtual
    bool m_recursive : 1;  // Recursive or part of recursion
    bool m_underGenerate : 1;  // Under generate (for warning)
    bool m_virtual : 1;  // Virtual method in class
+    bool m_needProcess : 1;  // Implements part of a process that allocates std::process
    VLifetime m_lifetime;  // Lifetime
 protected:
    AstNodeFTask(VNType t, FileLine* fl, const string& name, AstNode* stmtsp)
@@ -106,21 +107,25 @@ protected:
        , m_isConstructor{false}
        , m_isHideLocal{false}
        , m_isHideProtected{false}
-        , m_pure{false}
+        , m_dpiPure{false}
        , m_pureVirtual{false}
        , m_recursive{false}
        , m_underGenerate{false}
-        , m_virtual{false} {
+        , m_virtual{false}
+        , m_needProcess{false} {
        addStmtsp(stmtsp);
        cname(name);  // Might be overridden by dpi import/export
    }

 public:
    ASTGEN_MEMBERS_AstNodeFTask;
+    virtual AstNodeFTask* cloneType(const string& name) = 0;
    void dump(std::ostream& str = std::cout) const override;
    string name() const override VL_MT_STABLE { return m_name; }  // * = Var name
    bool maybePointedTo() const override { return true; }
-    bool isGateOptimizable() const override { return !((m_dpiExport || m_dpiImport) && !m_pure); }
+    bool isGateOptimizable() const override {
+        return !((m_dpiExport || m_dpiImport) && !m_dpiPure);
+    }
    // {AstFunc only} op1 = Range output variable
    void name(const string& name) override { m_name = name; }
    string cname() const { return m_cname; }
@@ -160,8 +165,8 @@ public:
    void isHideLocal(bool flag) { m_isHideLocal = flag; }
    bool isHideProtected() const { return m_isHideProtected; }
    void isHideProtected(bool flag) { m_isHideProtected = flag; }
-    void pure(bool flag) { m_pure = flag; }
-    bool pure() const { return m_pure; }
+    void dpiPure(bool flag) { m_dpiPure = flag; }
+    bool dpiPure() const { return m_dpiPure; }
    void pureVirtual(bool flag) { m_pureVirtual = flag; }
    bool pureVirtual() const { return m_pureVirtual; }
    void recursive(bool flag) { m_recursive = flag; }
@@ -170,9 +175,20 @@ public:
    bool underGenerate() const { return m_underGenerate; }
    void isVirtual(bool flag) { m_virtual = flag; }
    bool isVirtual() const { return m_virtual; }
+    void setNeedProcess() { m_needProcess = true; }
+    bool needProcess() const { return m_needProcess; }
    void lifetime(const VLifetime& flag) { m_lifetime = flag; }
    VLifetime lifetime() const { return m_lifetime; }
    bool isFirstInMyListOfStatements(AstNode* n) const override { return n == stmtsp(); }
+    void propagateAttrFrom(const AstNodeFTask* fromp) {
+        // Creating a wrapper with e.g. cloneType(); preserve some attributes
+        classMethod(fromp->classMethod());
+        isHideLocal(fromp->isHideLocal());
+        isHideProtected(fromp->isHideProtected());
+        isVirtual(fromp->isVirtual());
+        lifetime(fromp->lifetime());
+        underGenerate(fromp->underGenerate());
+    }
 };
 class AstNodeFile VL_NOT_FINAL : public AstNode {
    // Emitted Output file
@@ -272,10 +288,13 @@ public:
 class AstNodeProcedure VL_NOT_FINAL : public AstNode {
    // IEEE procedure: initial, final, always
    // @astgen op2 := stmtsp : List[AstNode] // Note: op1 is used in some sub-types only
-    bool m_suspendable = false;  // Is suspendable by a Delay, EventControl, etc.
+    bool m_suspendable : 1;  // Is suspendable by a Delay, EventControl, etc.
+    bool m_needProcess : 1;  // Implements part of a process that allocates std::process
 protected:
    AstNodeProcedure(VNType t, FileLine* fl, AstNode* stmtsp)
        : AstNode{t, fl} {
+        m_needProcess = false;
+        m_suspendable = false;
        addStmtsp(stmtsp);
    }

@@ -286,6 +305,8 @@ public:
    bool isJustOneBodyStmt() const { return stmtsp() && !stmtsp()->nextp(); }
    bool isSuspendable() const { return m_suspendable; }
    void setSuspendable() { m_suspendable = true; }
+    bool needProcess() const { return m_needProcess; }
+    void setNeedProcess() { m_needProcess = true; }
 };
 class AstNodeRange VL_NOT_FINAL : public AstNode {
    // A range, sized or unsized
@@ -308,8 +329,6 @@ public:
    // METHODS
    void addNextStmt(AstNode* newp,
                     AstNode* belowp) override;  // Stop statement searchback here
-    void addBeforeStmt(AstNode* newp,
-                       AstNode* belowp) override;  // Stop statement searchback here
    void dump(std::ostream& str = std::cout) const override;
 };
 class AstNodeAssign VL_NOT_FINAL : public AstNodeStmt {
@@ -571,13 +590,14 @@ private:
    bool m_isInline : 1;  // Inline function
    bool m_isVirtual : 1;  // Virtual function
    bool m_entryPoint : 1;  // User may call into this top level function
-    bool m_pure : 1;  // Pure function
+    bool m_dpiPure : 1;  // Pure DPI function
    bool m_dpiContext : 1;  // Declared as 'context' DPI import/export function
    bool m_dpiExportDispatcher : 1;  // This is the DPI export entry point (i.e.: called by user)
    bool m_dpiExportImpl : 1;  // DPI export implementation (called from DPI dispatcher via lookup)
    bool m_dpiImportPrototype : 1;  // This is the DPI import prototype (i.e.: provided by user)
    bool m_dpiImportWrapper : 1;  // Wrapper for invoking DPI import prototype from generated code
    bool m_dpiTraceInit : 1;  // DPI trace_init
+    bool m_needProcess : 1;  // Implements part of a process that allocates std::process
 public:
    AstCFunc(FileLine* fl, const string& name, AstScope* scopep, const string& rtnType = "")
        : ASTGEN_SUPER_CFunc(fl) {
@@ -597,8 +617,9 @@ public:
        m_isLoose = false;
        m_isInline = false;
        m_isVirtual = false;
+        m_needProcess = false;
        m_entryPoint = false;
-        m_pure = false;
+        m_dpiPure = false;
        m_dpiContext = false;
        m_dpiExportDispatcher = false;
        m_dpiExportImpl = false;
@@ -665,10 +686,12 @@ public:
    void isInline(bool flag) { m_isInline = flag; }
    bool isVirtual() const { return m_isVirtual; }
    void isVirtual(bool flag) { m_isVirtual = flag; }
+    bool needProcess() const { return m_needProcess; }
+    void setNeedProcess() { m_needProcess = true; }
    bool entryPoint() const { return m_entryPoint; }
    void entryPoint(bool flag) { m_entryPoint = flag; }
-    bool pure() const { return m_pure; }
-    void pure(bool flag) { m_pure = flag; }
+    bool dpiPure() const { return m_dpiPure; }
+    void dpiPure(bool flag) { m_dpiPure = flag; }
    bool dpiContext() const { return m_dpiContext; }
    void dpiContext(bool flag) { m_dpiContext = flag; }
    bool dpiExportDispatcher() const VL_MT_SAFE { return m_dpiExportDispatcher; }
@@ -688,6 +711,16 @@ public:
               && finalsp() == nullptr;
    }
 };
+class AstCLocalScope final : public AstNode {
+    // Pack statements into an unnamed scope when generating C++
+    // @astgen op1 := stmtsp : List[AstNode]
+public:
+    AstCLocalScope(FileLine* fl, AstNode* stmtsp)
+        : ASTGEN_SUPER_CLocalScope(fl) {
+        this->addStmtsp(stmtsp);
+    }
+    ASTGEN_MEMBERS_AstCLocalScope;
+};
 class AstCUse final : public AstNode {
    // C++ use of a class or #include; indicates need of forward declaration
    // Parents:  NODEMODULE
@@ -809,6 +842,8 @@ class AstClassExtends final : public AstNode {
    // @astgen op1 := childDTypep : Optional[AstNodeDType]
    // @astgen op2 := classOrPkgsp : Optional[AstNode]
    const bool m_isImplements = false;  // class implements
+    bool m_parameterized = false;  // has parameters in its statement
+
 public:
    AstClassExtends(FileLine* fl, AstNode* classOrPkgsp, bool isImplements)
        : ASTGEN_SUPER_ClassExtends(fl)
@@ -819,8 +854,12 @@ public:
    void dump(std::ostream& str) const override;
    bool hasDType() const override { return true; }
    string verilogKwd() const override { return isImplements() ? "implements" : "extends"; }
-    AstClass* classp() const;  // Class being extended (after link)
+    // Class being extended (after link and instantiation if needed)
+    AstClass* classOrNullp() const;
+    AstClass* classp() const;  // Like above, but throws error if nulll
    bool isImplements() const { return m_isImplements; }
+    void parameterized(bool flag) { m_parameterized = flag; }
+    bool parameterized() const { return m_parameterized; }
 };
 class AstClocking final : public AstNode {
    // Parents:  MODULE
@@ -1359,8 +1398,8 @@ public:
    string name() const override VL_MT_STABLE { return m_name; }  // * = Scope name
    void name(const string& name) override { m_name = name; }
    void dump(std::ostream& str) const override;
+    bool same(const AstNode* samep) const override;
    string nameDotless() const;
-    string nameVlSym() const { return string{"vlSymsp->"} + nameDotless(); }
    AstNodeModule* modp() const { return m_modp; }
    //
    AstScope* aboveScopep() const VL_MT_SAFE { return m_aboveScopep; }
@@ -1495,6 +1534,7 @@ class AstTypeTable final : public AstNode {
    AstEmptyQueueDType* m_emptyQueuep = nullptr;
    AstQueueDType* m_queueIndexp = nullptr;
    AstVoidDType* m_voidp = nullptr;
+    AstStreamDType* m_streamp = nullptr;
    AstBasicDType* m_basicps[VBasicDTypeKwd::_ENUM_MAX]{};
    //
    using DetailedMap = std::map<VBasicTypeKey, AstBasicDType*>;
@@ -1520,6 +1560,7 @@ public:
    AstEmptyQueueDType* findEmptyQueueDType(FileLine* fl);
    AstQueueDType* findQueueIndexDType(FileLine* fl);
    AstVoidDType* findVoidDType(FileLine* fl);
+    AstStreamDType* findStreamDType(FileLine* fl);
    void clearCache();
    void repairCache();
    void dump(std::ostream& str = std::cout) const override;
@@ -1628,6 +1669,7 @@ class AstVar final : public AstNode {
    bool m_usedLoopIdx : 1;  // Variable subject of for unrolling
    bool m_usedVirtIface : 1;  // Signal used through a virtual interface
    bool m_funcLocal : 1;  // Local variable for a function
+    bool m_funcLocalSticky : 1;  // As m_funcLocal but remains set if var is moved to a static
    bool m_funcReturn : 1;  // Return variable for a function
    bool m_attrScBv : 1;  // User force bit vector attribute
    bool m_attrIsolateAssign : 1;  // User isolate_assignments attribute
@@ -1671,6 +1713,7 @@ class AstVar final : public AstNode {
        m_sigUserRdPublic = false;
        m_sigUserRWPublic = false;
        m_funcLocal = false;
+        m_funcLocalSticky = false;
        m_funcReturn = false;
        m_attrScBv = false;
        m_attrIsolateAssign = false;
@@ -1758,6 +1801,7 @@ public:
    }
    ASTGEN_MEMBERS_AstVar;
    void dump(std::ostream& str) const override;
+    bool same(const AstNode* samep) const override;
    string name() const override VL_MT_STABLE VL_MT_SAFE { return m_name; }  // * = Var name
    bool hasDType() const override { return true; }
    bool maybePointedTo() const override { return true; }
@@ -1789,7 +1833,7 @@ public:
    string dpiTmpVarType(const string& varName) const;
    // Return Verilator internal type for argument: CData, SData, IData, WData
    string vlArgType(bool named, bool forReturn, bool forFunc, const string& namespc = "",
-                     bool asRef = false) const VL_MT_STABLE;
+                     bool asRef = false) const;
    string vlEnumType() const;  // Return VerilatorVarType: VLVT_UINT32, etc
    string vlEnumDir() const;  // Return VerilatorVarDir: VLVD_INOUT, etc
    string vlPropDecl(const string& propName) const;  // Return VerilatorVarProps declaration
@@ -1833,7 +1877,10 @@ public:
    void isContinuously(bool flag) { m_isContinuously = flag; }
    void isStatic(bool flag) { m_isStatic = flag; }
    void isIfaceParent(bool flag) { m_isIfaceParent = flag; }
-    void funcLocal(bool flag) { m_funcLocal = flag; }
+    void funcLocal(bool flag) {
+        m_funcLocal = flag;
+        if (flag) m_funcLocalSticky = true;
+    }
    void funcReturn(bool flag) { m_funcReturn = flag; }
    void hasStrengthAssignment(bool flag) { m_hasStrengthAssignment = flag; }
    bool hasStrengthAssignment() { return m_hasStrengthAssignment; }
@@ -1886,9 +1933,7 @@ public:
    bool isClassMember() const { return varType() == VVarType::MEMBER; }
    bool isStatementTemp() const { return (varType() == VVarType::STMTTEMP); }
    bool isXTemp() const { return (varType() == VVarType::XTEMP); }
-    bool isParam() const VL_MT_SAFE {
-        return (varType() == VVarType::LPARAM || varType() == VVarType::GPARAM);
-    }
+    bool isParam() const { return varType().isParam(); }
    bool isGParam() const { return (varType() == VVarType::GPARAM); }
    bool isGenVar() const { return (varType() == VVarType::GENVAR); }
    bool isBitLogic() const {
@@ -1915,6 +1960,7 @@ public:
    bool isStatic() const VL_MT_SAFE { return m_isStatic; }
    bool isLatched() const { return m_isLatched; }
    bool isFuncLocal() const { return m_funcLocal; }
+    bool isFuncLocalSticky() const { return m_funcLocalSticky; }
    bool isFuncReturn() const { return m_funcReturn; }
    bool isPullup() const { return m_isPullup; }
    bool isPulldown() const { return m_isPulldown; }
@@ -1928,18 +1974,25 @@ public:
    string verilogKwd() const override;
    void lifetime(const VLifetime& flag) { m_lifetime = flag; }
    VLifetime lifetime() const { return m_lifetime; }
-    void propagateAttrFrom(AstVar* fromp) {
+    void propagateAttrFrom(const AstVar* fromp) {
        // This is getting connected to fromp; keep attributes
        // Note the method below too
        if (fromp->attrFileDescr()) attrFileDescr(true);
        if (fromp->attrIsolateAssign()) attrIsolateAssign(true);
        if (fromp->isContinuously()) isContinuously(true);
    }
+    void propagateWrapAttrFrom(const AstVar* fromp) {
+        // Creating a function wrapper; keep attributes
+        propagateAttrFrom(fromp);
+        direction(fromp->direction());
+        declDirection(fromp->declDirection());
+        lifetime(fromp->lifetime());
+    }
    bool gateMultiInputOptimizable() const {
        // Ok to gate optimize; must return false if propagateAttrFrom would do anything
        return !isUsedClock();
    }
-    void combineType(AstVar* typevarp) {
+    void combineType(const AstVar* typevarp) {
        // This is same as typevarp (for combining input & reg decls)
        // "this" is the input var. typevarp is the reg var.
        propagateAttrFrom(typevarp);
@@ -2000,6 +2053,7 @@ public:
    bool maybePointedTo() const override { return true; }
    string name() const override VL_MT_STABLE { return scopep()->name() + "->" + varp()->name(); }
    void dump(std::ostream& str) const override;
+    bool same(const AstNode* samep) const override;
    bool hasDType() const override { return true; }
    AstVar* varp() const VL_MT_STABLE { return m_varp; }  // [After Link] Pointer to variable
    AstScope* scopep() const VL_MT_STABLE { return m_scopep; }  // Pointer to scope it's under
@@ -2012,7 +2066,7 @@ public:
 class AstBegin final : public AstNodeBlock {
    // A Begin/end named block, only exists shortly after parsing until linking
    // Parents: statement
-    // @astgen op2 := genforp : Optional[AstNode]
+    // @astgen op1 := genforp : Optional[AstNode]

    bool m_generate;  // Underneath a generate
    const bool m_implied;  // Not inserted by user
@@ -2031,6 +2085,7 @@ public:
 };
 class AstFork final : public AstNodeBlock {
    // A fork named block
+    // @astgen op1 := initsp : List[AstNode]
    // Parents: statement
    // Children: statements
 private:
@@ -2056,6 +2111,24 @@ public:
    }
    ASTGEN_MEMBERS_AstFunc;
    bool hasDType() const override { return true; }
+    AstNodeFTask* cloneType(const string& name) override {
+        return new AstFunc{fileline(), name, nullptr, nullptr};
+    }
+};
+class AstLet final : public AstNodeFTask {
+    // Verilog "let" statement
+    // Parents: MODULE
+    // stmtp is always a StmtExpr as Let always returns AstNodeExpr
+public:
+    AstLet(FileLine* fl, const string& name)
+        : ASTGEN_SUPER_Let(fl, name, nullptr) {}
+    ASTGEN_MEMBERS_AstLet;
+    bool hasDType() const override { return true; }
+    const char* broken() const override {
+        BROKEN_RTN(!VN_IS(stmtsp(), StmtExpr));
+        return nullptr;
+    }
+    AstNodeFTask* cloneType(const string& name) override { return new AstLet{fileline(), name}; }
 };
 class AstProperty final : public AstNodeFTask {
    // A property inside a module
@@ -2064,6 +2137,9 @@ public:
        : ASTGEN_SUPER_Property(fl, name, stmtp) {}
    ASTGEN_MEMBERS_AstProperty;
    bool hasDType() const override { return true; }
+    AstNodeFTask* cloneType(const string& name) override {
+        return new AstProperty{fileline(), name, nullptr};
+    }
 };
 class AstTask final : public AstNodeFTask {
    // A task inside a module
@@ -2071,6 +2147,9 @@ public:
    AstTask(FileLine* fl, const string& name, AstNode* stmtp)
        : ASTGEN_SUPER_Task(fl, name, stmtp) {}
    ASTGEN_MEMBERS_AstTask;
+    AstNodeFTask* cloneType(const string& name) override {
+        return new AstTask{fileline(), name, nullptr};
+    }
 };

 // === AstNodeFile ===
@@ -2109,17 +2188,13 @@ public:
 // === AstNodeModule ===
 class AstClass final : public AstNodeModule {
    // @astgen op4 := extendsp : List[AstClassExtends]
-    // TYPES
-    using MemberNameMap = std::map<const std::string, AstNode*>;
    // MEMBERS
-    MemberNameMap m_members;  // Members or method children
-    AstClassPackage* m_classOrPackagep = nullptr;  // Class package this is under
+    AstClassPackage* m_classOrPackagep = nullptr;  // Package it will be emitted with
    bool m_extended = false;  // Is extension or extended by other classes
    bool m_interfaceClass = false;  // Interface class
    bool m_needRNG = false;  // Need RNG, uses srandom/randomize
    bool m_virtual = false;  // Virtual class
    bool m_parameterized = false;  // Parameterized class
-    void insertCache(AstNode* nodep);

 public:
    AstClass(FileLine* fl, const string& name)
@@ -2134,16 +2209,7 @@ public:
    AstClassPackage* classOrPackagep() const VL_MT_SAFE { return m_classOrPackagep; }
    void classOrPackagep(AstClassPackage* classpackagep) { m_classOrPackagep = classpackagep; }
    AstNode* membersp() const { return stmtsp(); }
-    void addMembersp(AstNode* nodep) {
-        insertCache(nodep);
-        addStmtsp(nodep);
-    }
-    void clearCache() { m_members.clear(); }
-    void repairCache();
-    AstNode* findMember(const string& name) const {
-        const auto it = m_members.find(name);
-        return (it == m_members.end()) ? nullptr : it->second;
-    }
+    void addMembersp(AstNode* nodep) { addStmtsp(nodep); }
    bool isExtended() const { return m_extended; }
    void isExtended(bool flag) { m_extended = flag; }
    bool isInterfaceClass() const { return m_interfaceClass; }
@@ -2159,6 +2225,7 @@ public:
    static bool isClassExtendedFrom(const AstClass* refClassp, const AstClass* baseClassp);
    // Return the lowest class extended from, or this class
    AstClass* baseMostClassp();
+    static bool isCacheableChild(const AstNode* nodep);
 };
 class AstClassPackage final : public AstNodeModule {
    // The static information portion of a class (treated similarly to a package)
@@ -3240,8 +3307,6 @@ public:
    int instrCount() const override { return INSTR_COUNT_BRANCH; }
    bool same(const AstNode* /*samep*/) const override { return true; }
    // Stop statement searchback here
-    void addBeforeStmt(AstNode* newp, AstNode* belowp) override;
-    // Stop statement searchback here
    void addNextStmt(AstNode* newp, AstNode* belowp) override;
    bool isFirstInMyListOfStatements(AstNode* n) const override { return n == stmtsp(); }
 };
@@ -25,6 +25,7 @@
 #include "V3Hasher.h"
 #include "V3PartitionGraph.h"  // Just for mtask dumping
 #include "V3String.h"
+#include "V3Width.h"

 #include "V3Ast__gen_macros.h"  // Generated by 'astgen'

@@ -63,7 +64,7 @@ void AstNodeFTaskRef::cloneRelink() {
 bool AstNodeFTaskRef::isPure() const {
    // TODO: For non-DPI functions we could traverse the AST of function's body to determine
    // pureness.
-    return this->taskp() && this->taskp()->dpiImport() && this->taskp()->pure();
+    return this->taskp() && this->taskp()->dpiImport() && this->taskp()->dpiPure();
 }

 bool AstNodeFTaskRef::isGateOptimizable() const { return m_taskp && m_taskp->isGateOptimizable(); }
@@ -83,12 +84,6 @@ void AstNodeVarRef::cloneRelink() {
    }
 }

-string AstNodeVarRef::selfPointerProtect(bool useSelfForThis) const {
-    const string& sp
-        = useSelfForThis ? VString::replaceWord(selfPointer(), "this", "vlSelf") : selfPointer();
-    return VIdProtect::protectWordsIf(sp, protect());
-}
-
 void AstAddrOfCFunc::cloneRelink() {
    if (m_funcp && m_funcp->clonep()) m_funcp = m_funcp->clonep();
 }
@@ -103,29 +98,8 @@ int AstNodeSel::bitConst() const {
    return (constp ? constp->toSInt() : 0);
 }

-void AstNodeUOrStructDType::repairMemberCache() {
-    clearCache();
-    for (AstMemberDType* itemp = membersp(); itemp; itemp = VN_AS(itemp->nextp(), MemberDType)) {
-        if (m_members.find(itemp->name()) != m_members.end()) {
-            itemp->v3error("Duplicate declaration of member name: " << itemp->prettyNameQ());
-        } else {
-            m_members.emplace(itemp->name(), itemp);
-        }
-    }
-}
-
 const char* AstNodeUOrStructDType::broken() const {
    BROKEN_RTN(m_classOrPackagep && !m_classOrPackagep->brokeExists());
-    std::unordered_set<AstMemberDType*> exists;
-    for (AstMemberDType* itemp = membersp(); itemp; itemp = VN_AS(itemp->nextp(), MemberDType)) {
-        exists.insert(itemp);
-    }
-    for (MemberNameMap::const_iterator it = m_members.begin(); it != m_members.end(); ++it) {
-        if (VL_UNCOVERABLE(exists.find(it->second) == exists.end())) {
-            this->v3error("Internal: Structure member broken: " << it->first);
-            return "member broken";
-        }
-    }
    return nullptr;
 }

@@ -147,14 +121,22 @@ const char* AstNodeCCall::broken() const {
    BROKEN_RTN(m_funcp && !m_funcp->brokeExists());
    return nullptr;
 }
-bool AstNodeCCall::isPure() const { return funcp()->pure(); }
+bool AstNodeCCall::isPure() const { return funcp()->dpiPure(); }

-string AstCCall::selfPointerProtect(bool useSelfForThis) const {
-    const string& sp
-        = useSelfForThis ? VString::replaceWord(selfPointer(), "this", "vlSelf") : selfPointer();
-    return VIdProtect::protectWordsIf(sp, protect());
+AstNodeCond::AstNodeCond(VNType t, FileLine* fl, AstNodeExpr* condp, AstNodeExpr* thenp,
+                         AstNodeExpr* elsep)
+    : AstNodeTriop{t, fl, condp, thenp, elsep} {
+    UASSERT_OBJ(thenp, this, "No thenp expression");
+    UASSERT_OBJ(elsep, this, "No elsep expression");
+    if (thenp->isClassHandleValue() && elsep->isClassHandleValue()) {
+        // Get the most-deriving class type that both arguments can be casted to.
+        AstNodeDType* const commonClassTypep = V3Width::getCommonClassTypep(thenp, elsep);
+        UASSERT_OBJ(commonClassTypep, this, "No common base class exists");
+        dtypep(commonClassTypep);
+    } else {
+        dtypeFrom(thenp);
+    }
 }
-
 void AstNodeCond::numberOperate(V3Number& out, const V3Number& lhs, const V3Number& rhs,
                                const V3Number& ths) {
    if (lhs.isNeqZero()) {
@@ -391,13 +373,15 @@ string AstVar::verilogKwd() const {
        return "wreal";
    } else if (varType() == VVarType::IFACEREF) {
        return "ifaceref";
-    } else {
+    } else if (dtypep()) {
        return dtypep()->name();
+    } else {
+        return "UNKNOWN";
    }
 }

 string AstVar::vlArgType(bool named, bool forReturn, bool forFunc, const string& namespc,
-                         bool asRef) const VL_MT_STABLE {
+                         bool asRef) const {
    UASSERT_OBJ(!forReturn, this,
                "Internal data is never passed as return, but as first argument");
    string ostatic;
@@ -714,12 +698,12 @@ public:
    }
 };

-string AstNodeDType::cType(const string& name, bool /*forFunc*/, bool isRef) const VL_MT_STABLE {
+string AstNodeDType::cType(const string& name, bool /*forFunc*/, bool isRef) const {
    const CTypeRecursed info = cTypeRecurse(false);
    return info.render(name, isRef);
 }

-AstNodeDType::CTypeRecursed AstNodeDType::cTypeRecurse(bool compound) const VL_MT_STABLE {
+AstNodeDType::CTypeRecursed AstNodeDType::cTypeRecurse(bool compound) const {
    // Legacy compound argument currently just passed through and unused
    CTypeRecursed info;

@@ -782,6 +766,8 @@ AstNodeDType::CTypeRecursed AstNodeDType::cTypeRecurse(bool compound) const VL_M
            info.m_type = "VlDynamicTriggerScheduler";
        } else if (bdtypep->isForkSync()) {
            info.m_type = "VlForkSync";
+        } else if (bdtypep->isProcessRef()) {
+            info.m_type = "VlProcessRef";
        } else if (bdtypep->isEvent()) {
            info.m_type = "VlEvent";
        } else if (dtypep->widthMin() <= 8) {  // Handle unpacked arrays; not bdtypep->width
@@ -1067,6 +1053,15 @@ AstVoidDType* AstTypeTable::findVoidDType(FileLine* fl) {
    return m_voidp;
 }

+AstStreamDType* AstTypeTable::findStreamDType(FileLine* fl) {
+    if (VL_UNLIKELY(!m_streamp)) {
+        AstStreamDType* const newp = new AstStreamDType{fl};
+        addTypesp(newp);
+        m_streamp = newp;
+    }
+    return m_streamp;
+}
+
 AstQueueDType* AstTypeTable::findQueueIndexDType(FileLine* fl) {
    if (VL_UNLIKELY(!m_queueIndexp)) {
        AstQueueDType* const newp = new AstQueueDType{fl, AstNode::findUInt32DType(), nullptr};
@@ -1269,44 +1264,17 @@ AstVarScope* AstConstPool::findConst(AstConst* initp, bool mergeDType) {
 //======================================================================
 // Special walking tree inserters

-void AstNode::addBeforeStmt(AstNode* newp, AstNode*) {
-    UASSERT_OBJ(backp(), newp, "Can't find current statement to addBeforeStmt");
-    // Look up; virtual call will find where to put it
-    this->backp()->addBeforeStmt(newp, this);
-}
 void AstNode::addNextStmt(AstNode* newp, AstNode*) {
    UASSERT_OBJ(backp(), newp, "Can't find current statement to addNextStmt");
    // Look up; virtual call will find where to put it
    this->backp()->addNextStmt(newp, this);
 }

-void AstNodeStmt::addBeforeStmt(AstNode* newp, AstNode*) {
-    // Insert newp before current node
-    this->addHereThisAsNext(newp);
-}
 void AstNodeStmt::addNextStmt(AstNode* newp, AstNode*) {
    // Insert newp after current node
    this->addNextHere(newp);
 }

-void AstWhile::addBeforeStmt(AstNode* newp, AstNode* belowp) {
-    // Special, as statements need to be put in different places
-    // Belowp is how we came to recurse up to this point
-    // Preconditions insert first just before themselves (the normal rule
-    // for other statement types)
-    if (belowp == precondsp()) {
-        // Must have been first statement in precondsp list, so newp is new first statement
-        belowp->addHereThisAsNext(newp);
-    } else if (belowp == condp()) {
-        // Goes before condition, IE in preconditions
-        addPrecondsp(newp);
-    } else if (belowp == stmtsp()) {
-        // Was first statement in body, so new front
-        belowp->addHereThisAsNext(newp);
-    } else {
-        belowp->v3fatalSrc("Doesn't look like this was really under the while");
-    }
-}
 void AstWhile::addNextStmt(AstNode* newp, AstNode* belowp) {
    // Special, as statements need to be put in different places
    // Belowp is how we came to recurse up to this point
@@ -1377,6 +1345,7 @@ void AstNode::dump(std::ostream& str) const {
 void AstNodeProcedure::dump(std::ostream& str) const {
    this->AstNode::dump(str);
    if (isSuspendable()) str << " [SUSP]";
+    if (needProcess()) str << " [NPRC]";
 }

 void AstAlways::dump(std::ostream& str) const {
@@ -1440,29 +1409,10 @@ const char* AstClassPackage::broken() const {
 void AstClassPackage::cloneRelink() {
    if (m_classp && m_classp->clonep()) m_classp = m_classp->clonep();
 }
-void AstClass::insertCache(AstNode* nodep) {
-    const bool doit = (VN_IS(nodep, Var) || VN_IS(nodep, EnumItemRef)
-                       || (VN_IS(nodep, NodeFTask) && !VN_AS(nodep, NodeFTask)->isExternProto())
-                       || VN_IS(nodep, CFunc));
-    if (doit) {
-        if (m_members.find(nodep->name()) != m_members.end()) {
-            nodep->v3error("Duplicate declaration of member name: " << nodep->prettyNameQ());
-        } else {
-            m_members.emplace(nodep->name(), nodep);
-        }
-    }
-}
-void AstClass::repairCache() {
-    clearCache();
-    for (auto* itemp = membersp(); itemp; itemp = itemp->nextp()) {
-        if (const auto* const scopep = VN_CAST(itemp, Scope)) {
-            for (auto* blockp = scopep->blocksp(); blockp; blockp = blockp->nextp()) {
-                insertCache(blockp);
-            }
-        } else {
-            insertCache(itemp);
-        }
-    }
+bool AstClass::isCacheableChild(const AstNode* nodep) {
+    return (VN_IS(nodep, Var) || VN_IS(nodep, EnumItemRef)
+            || (VN_IS(nodep, NodeFTask) && !VN_AS(nodep, NodeFTask)->isExternProto())
+            || VN_IS(nodep, CFunc));
 }
 AstClass* AstClass::baseMostClassp() {
    AstClass* basep = this;
@@ -1499,13 +1449,20 @@ void AstClassExtends::dump(std::ostream& str) const {
    this->AstNode::dump(str);
    if (isImplements()) str << " [IMPLEMENTS]";
 }
-AstClass* AstClassExtends::classp() const {
-    const AstClassRefDType* refp = VN_CAST(dtypep(), ClassRefDType);
-    if (VL_UNLIKELY(!refp)) {  // LinkDot uses this for 'super.'
-        refp = VN_AS(childDTypep(), ClassRefDType);
+AstClass* AstClassExtends::classOrNullp() const {
+    const AstNodeDType* const dtp = dtypep() ? dtypep() : childDTypep();
+    const AstClassRefDType* const refp = VN_CAST(dtp, ClassRefDType);
+    if (refp && !refp->paramsp()) {
+        // Class already resolved
+        return refp->classp();
+    } else {
+        return nullptr;
    }
-    UASSERT_OBJ(refp, this, "class extends non-ref");
-    return refp->classp();
+}
+AstClass* AstClassExtends::classp() const {
+    AstClass* const clsp = classOrNullp();
+    UASSERT_OBJ(clsp, this, "Extended class is unresolved");
+    return clsp;
 }
 void AstClassRefDType::dump(std::ostream& str) const {
    this->AstNodeDType::dump(str);
@@ -1669,10 +1626,25 @@ void AstLogOr::dump(std::ostream& str) const {
    this->AstNodeExpr::dump(str);
    if (sideEffect()) str << " [SIDE]";
 }
+
 void AstMemberDType::dumpSmall(std::ostream& str) const {
    this->AstNodeDType::dumpSmall(str);
    str << "member";
 }
+AstNodeUOrStructDType* AstMemberDType::getChildStructp() const {
+    AstNodeDType* subdtp = skipRefp();
+    while (AstNodeArrayDType* const asubdtp = VN_CAST(subdtp, NodeArrayDType)) {
+        subdtp = asubdtp->subDTypep();
+    }
+    return VN_CAST(subdtp, NodeUOrStructDType);  // Maybe nullptr
+}
+
+bool AstMemberSel::same(const AstNode* samep) const {
+    const AstMemberSel* const sp = static_cast<const AstMemberSel*>(samep);
+    return sp != nullptr && access() == sp->access() && fromp()->same(sp->fromp())
+           && name() == sp->name() && varp()->same(sp->varp());
+}
+
 void AstMemberSel::dump(std::ostream& str) const {
    this->AstNodeExpr::dump(str);
    str << " -> ";
@@ -1829,6 +1801,7 @@ void AstNodeUOrStructDType::dump(std::ostream& str) const {
    this->AstNodeDType::dump(str);
    if (packed()) str << " [PACKED]";
    if (isFourstate()) str << " [4STATE]";
+    if (classOrPackagep()) str << " pkg=" << nodeAddr(classOrPackagep());
 }
 void AstNodeDType::dump(std::ostream& str) const {
    this->AstNode::dump(str);
@@ -2060,6 +2033,10 @@ void AstVoidDType::dumpSmall(std::ostream& str) const {
    this->AstNodeDType::dumpSmall(str);
    str << "void";
 }
+void AstStreamDType::dumpSmall(std::ostream& str) const {
+    this->AstNodeDType::dumpSmall(str);
+    str << "stream";
+}
 void AstVarScope::dump(std::ostream& str) const {
    this->AstNode::dump(str);
    if (isTrace()) str << " [T]";
@@ -2071,6 +2048,10 @@ void AstVarScope::dump(std::ostream& str) const {
        str << " ->UNLINKED";
    }
 }
+bool AstVarScope::same(const AstNode* samep) const {
+    const AstVarScope* const asamep = static_cast<const AstVarScope*>(samep);
+    return varp()->same(asamep->varp()) && scopep()->same(asamep->scopep());
+}
 void AstNodeVarRef::dump(std::ostream& str) const {
    this->AstNodeExpr::dump(str);
    if (classOrPackagep()) str << " pkg=" << nodeAddr(classOrPackagep());
@@ -2098,6 +2079,10 @@ void AstVarRef::dump(std::ostream& str) const {
        str << "UNLINKED";
    }
 }
+const char* AstVarRef::broken() const {
+    BROKEN_RTN(!varp());
+    return AstNodeVarRef::broken();
+}
 bool AstVarRef::same(const AstNode* samep) const {
    return same(static_cast<const AstVarRef*>(samep));
 }
@@ -2134,12 +2119,23 @@ void AstVar::dump(std::ostream& str) const {
    if (!lifetime().isNone()) str << " [" << lifetime().ascii() << "] ";
    str << " " << varType();
 }
+bool AstVar::same(const AstNode* samep) const {
+    const AstVar* const asamep = static_cast<const AstVar*>(samep);
+    return name() == asamep->name() && varType() == asamep->varType();
+}
 void AstScope::dump(std::ostream& str) const {
    this->AstNode::dump(str);
    str << " [abovep=" << reinterpret_cast<const void*>(aboveScopep()) << "]";
    str << " [cellp=" << reinterpret_cast<const void*>(aboveCellp()) << "]";
    str << " [modp=" << reinterpret_cast<const void*>(modp()) << "]";
 }
+bool AstScope::same(const AstNode* samep) const {
+    const AstScope* const asamep = static_cast<const AstScope*>(samep);
+    return name() == asamep->name()
+           && ((!aboveScopep() && !asamep->aboveScopep())
+               || (aboveScopep() && asamep->aboveScopep()
+                   && aboveScopep()->name() == asamep->aboveScopep()->name()));
+}
 void AstScopeName::dump(std::ostream& str) const {
    this->AstNodeExpr::dump(str);
    if (dpiExport()) str << " [DPIEX]";
@@ -2291,7 +2287,7 @@ void AstCFile::dump(std::ostream& str) const {
 void AstCFunc::dump(std::ostream& str) const {
    this->AstNode::dump(str);
    if (slow()) str << " [SLOW]";
-    if (pure()) str << " [PURE]";
+    if (dpiPure()) str << " [DPIPURE]";
    if (isStatic()) str << " [STATIC]";
    if (dpiExportDispatcher()) str << " [DPIED]";
    if (dpiExportImpl()) str << " [DPIEI]";
@@ -2302,6 +2298,7 @@ void AstCFunc::dump(std::ostream& str) const {
    if (isDestructor()) str << " [DTOR]";
    if (isVirtual()) str << " [VIRT]";
    if (isCoroutine()) str << " [CORO]";
+    if (needProcess()) str << " [NPRC]";
 }
 const char* AstCAwait::broken() const {
    BROKEN_RTN(m_sensesp && !m_sensesp->brokeExists());
@@ -2320,7 +2317,7 @@ void AstCAwait::dump(std::ostream& str) const {
 int AstCMethodHard::instrCount() const {
    if (AstBasicDType* const basicp = fromp()->dtypep()->basicp()) {
        // TODO: add a more structured description of library methods, rather than using string
-        //       matching. See #3715.
+        //       matching. See issue #3715.
        if (basicp->isTriggerVec() && m_name == "word") {
            // This is an important special case for scheduling so we compute it precisely,
            // it is simply a load.
@@ -2329,6 +2326,74 @@ int AstCMethodHard::instrCount() const {
    }
    return 0;
 }
+void AstCMethodHard::setPurity() {
+    static const std::map<std::string, bool> isPureMethod{{"andNot", false},
+                                                          {"any", true},
+                                                          {"assign", false},
+                                                          {"at", true},
+                                                          {"atBack", true},
+                                                          {"awaitingCurrentTime", true},
+                                                          {"clear", false},
+                                                          {"clearFired", false},
+                                                          {"commit", false},
+                                                          {"delay", false},
+                                                          {"done", false},
+                                                          {"erase", false},
+                                                          {"evaluate", false},
+                                                          {"evaluation", false},
+                                                          {"exists", true},
+                                                          {"find", true},
+                                                          {"find_first", true},
+                                                          {"find_first_index", true},
+                                                          {"find_index", true},
+                                                          {"find_last", true},
+                                                          {"find_last_index", true},
+                                                          {"fire", false},
+                                                          {"first", false},
+                                                          {"init", false},
+                                                          {"insert", false},
+                                                          {"isFired", true},
+                                                          {"isTriggered", true},
+                                                          {"join", false},
+                                                          {"last", false},
+                                                          {"max", true},
+                                                          {"min", true},
+                                                          {"neq", true},
+                                                          {"next", false},
+                                                          {"pop", false},
+                                                          {"pop_back", false},
+                                                          {"pop_front", false},
+                                                          {"prev", false},
+                                                          {"push", false},
+                                                          {"push_back", false},
+                                                          {"push_front", false},
+                                                          {"r_and", true},
+                                                          {"r_or", true},
+                                                          {"r_product", true},
+                                                          {"r_sum", true},
+                                                          {"r_xor", true},
+                                                          {"renew", false},
+                                                          {"renew_copy", false},
+                                                          {"resume", false},
+                                                          {"reverse", false},
+                                                          {"rsort", false},
+                                                          {"set", false},
+                                                          {"shuffle", false},
+                                                          {"size", true},
+                                                          {"slice", true},
+                                                          {"sliceBackBack", true},
+                                                          {"sliceFrontBack", true},
+                                                          {"sort", false},
+                                                          {"thisOr", false},
+                                                          {"trigger", false},
+                                                          {"unique", true},
+                                                          {"unique_index", true},
+                                                          {"word", true}};
+
+    auto isPureIt = isPureMethod.find(name());
+    UASSERT_OBJ(isPureIt != isPureMethod.end(), this, "Unknown purity of method " + name());
+    m_pure = isPureIt->second;
+}
 const char* AstCFunc::broken() const {
    BROKEN_RTN((m_scopep && !m_scopep->brokeExists()));
    return nullptr;
@@ -72,12 +72,12 @@ private:
    string m_unnamedScope;  // Name of begin blocks, including unnamed blocks
    int m_ifDepth = 0;  // Current if depth
    bool m_keepBegins = false;  // True if begins should not be inlined
-    std::set<AstVar*> m_staticFuncVars;  // Static variables from m_ftaskp

    // METHODS

    string dot(const string& a, const string& b) {
        if (a == "") return b;
+        if (b == "") return a;
        return a + "__DOT__" + b;
    }

@@ -123,25 +123,6 @@ private:
        }
    }

-    void renameAndStaticsRecurse(AstNode* const nodep) {
-        // Rename references and move InitialStatic items
-        if (AstVarRef* const varrefp = VN_CAST(nodep, VarRef)) {
-            const auto it = m_staticFuncVars.find(varrefp->varp());
-            if (it != m_staticFuncVars.end()) varrefp->name((*it)->name());
-        }
-
-        if (nodep->op1p()) renameAndStaticsRecurse(nodep->op1p());
-        if (nodep->op2p()) renameAndStaticsRecurse(nodep->op2p());
-        if (nodep->op3p()) renameAndStaticsRecurse(nodep->op3p());
-        if (nodep->op4p()) renameAndStaticsRecurse(nodep->op4p());
-        if (nodep->nextp()) renameAndStaticsRecurse(nodep->nextp());
-
-        if (VN_IS(nodep, InitialStatic)) {
-            nodep->unlinkFrBack();
-            m_ftaskp->addHereThisAsNext(nodep);
-        }
-    }
-
    // VISITORS
    void visit(AstFork* nodep) override {
        // Keep begins in forks to group their statements together
@@ -197,7 +178,10 @@ private:
            m_ftaskp = nodep;
            m_liftedp = nullptr;
            iterateChildren(nodep);
-            renameAndStaticsRecurse(nodep);
+            nodep->foreach([&](AstInitialStatic* const initp) {
+                initp->unlinkFrBack();
+                m_ftaskp->addHereThisAsNext(initp);
+            });
            if (m_liftedp) {
                // Place lifted nodes at beginning of stmtsp, so Var nodes appear before referenced
                if (AstNode* const stmtsp = nodep->stmtsp()) {
@@ -207,7 +191,6 @@ private:
                nodep->addStmtsp(m_liftedp);
                m_liftedp = nullptr;
            }
-            m_staticFuncVars.clear();
            m_ftaskp = nullptr;
        }
    }
@@ -246,11 +229,11 @@ private:
    void visit(AstVar* nodep) override {
        // If static variable, move it outside a function.
        if (nodep->lifetime().isStatic() && m_ftaskp) {
-            const std::string newName = m_ftaskp->name() + "__Vstatic__" + nodep->name();
+            const std::string newName
+                = m_ftaskp->name() + "__Vstatic__" + dot(m_unnamedScope, nodep->name());
            nodep->name(newName);
            nodep->unlinkFrBack();
            m_ftaskp->addHereThisAsNext(nodep);
-            m_staticFuncVars.insert(nodep);
            nodep->funcLocal(false);
        } else if (m_unnamedScope != "") {
            // Rename it
@@ -361,7 +344,6 @@ private:
    void visit(AstVarRef* nodep) override {
        if (nodep->varp()->user1()) {  // It was converted
            UINFO(9, "    relinVarRef " << nodep << endl);
-            nodep->name(nodep->varp()->name());
        }
        iterateChildren(nodep);
    }
@@ -392,5 +374,5 @@ void V3Begin::debeginAll(AstNetlist* nodep) {
        { BeginVisitor{nodep, &state}; }
        if (state.anyFuncInBegin()) { BeginRelinkVisitor{nodep, &state}; }
    }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("begin", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("begin", 0, dumpTreeLevel() >= 3);
 }
@@ -45,10 +45,12 @@ private:
    //  AstFTask::user1()       -> int.  Number of references
    const VNUser1InUse m_inuser1;

-    // STATE
+    // STATE - across all visitors
+    std::vector<AstCFunc*> m_cfuncsp;  // List of all tasks
+
+    // STATE - for current visit position (use VL_RESTORER)
    int m_likely = false;  // Excuses for branch likely taken
    int m_unlikely = false;  // Excuses for branch likely not taken
-    std::vector<AstCFunc*> m_cfuncsp;  // List of all tasks

    // METHODS

@@ -147,22 +147,25 @@ bool V3Broken::isLinkable(const AstNode* nodep) { return s_linkableTable.isLinka
 // Check every node in tree

 class BrokenCheckVisitor final : public VNVisitorConst {
-    bool m_inScope = false;  // Under AstScope
-
    // Constants for marking we are under/not under a node
    const uint8_t m_brokenCntCurrentNotUnder = s_brokenCntGlobal.get();  // Top bit is clear
    const uint8_t m_brokenCntCurrentUnder = m_brokenCntCurrentNotUnder | 0x80;  // Top bit is set

-    // Current CFunc, if any
-    const AstCFunc* m_cfuncp = nullptr;
+    // STATE - across all visitors
    // All local variables declared in current function
-    std::unordered_set<const AstVar*> m_localVars;
+    std::set<const AstVar*> m_localVars;
    // Variable references in current function that do not reference an in-scope local
-    std::unordered_map<const AstVar*, const AstNodeVarRef*> m_suspectRefs;
+    std::map<const AstVar*, const AstNodeVarRef*> m_suspectRefs;
    // Local variables declared in the scope of the current statement
    std::vector<std::unordered_set<const AstVar*>> m_localsStack;

+    // STATE - for current visit position (use VL_RESTORER)
+    const AstCFunc* m_cfuncp = nullptr;  // Current CFunc, if any
+    bool m_inScope = false;  // Under AstScope
+    std::set<std::string> m_cFuncNames;  // CFunc by name in current class/module
+
 private:
+    // METHODS
    static void checkWidthMin(const AstNode* nodep) {
        UASSERT_OBJ(nodep->width() == nodep->widthMin()
                        || v3Global.widthMinUsage() != VWidthMinUsage::MATCHES_WIDTH,
@@ -220,6 +223,7 @@ private:
        }
        return false;
    }
+    // VISITORS
    void visit(AstNodeAssign* nodep) override {
        processAndIterate(nodep);
        UASSERT_OBJ(!(v3Global.assertDTypesResolved() && nodep->brokeLhsMustBeLvalue()
@@ -235,10 +239,20 @@ private:
    }
    void visit(AstScope* nodep) override {
        VL_RESTORER(m_inScope);
-        {
-            m_inScope = true;
-            processAndIterate(nodep);
-        }
+        m_inScope = true;
+        VL_RESTORER(m_cFuncNames);
+        m_cFuncNames.clear();
+        processAndIterate(nodep);
+    }
+    void visit(AstNodeModule* nodep) override {
+        VL_RESTORER(m_cFuncNames);
+        m_cFuncNames.clear();
+        processAndIterate(nodep);
+    }
+    void visit(AstNodeUOrStructDType* nodep) override {
+        VL_RESTORER(m_cFuncNames);
+        m_cFuncNames.clear();
+        processAndIterate(nodep);
    }
    void visit(AstNodeVarRef* nodep) override {
        processAndIterate(nodep);
@@ -260,12 +274,19 @@ private:
    }
    void visit(AstCFunc* nodep) override {
        UASSERT_OBJ(!m_cfuncp, nodep, "Nested AstCFunc");
+        VL_RESTORER(m_cfuncp);
        m_cfuncp = nodep;
        m_localVars.clear();
        m_suspectRefs.clear();
        m_localsStack.clear();
        pushLocalScope();

+        // Check for duplicate names, otherwise linker might just ignore them
+        string nameArgs = nodep->name();
+        if (!nodep->argTypes().empty()) nameArgs += "(" + nodep->argTypes() + ")";
+        UASSERT_OBJ(m_cFuncNames.emplace(nameArgs).second, nodep,
+                    "Duplicate cfunc name: '" << nameArgs << "'");
+
        processAndIterate(nodep);

        // Check suspect references are all to non-locals
@@ -273,8 +294,6 @@ private:
            UASSERT_OBJ(m_localVars.count(pair.first) == 0, pair.second,
                        "Local variable not in scope where referenced: " << pair.first);
        }
-
-        m_cfuncp = nullptr;
    }
    void visit(AstNodeIf* nodep) override {
        // Each branch is a separate local variable scope
@@ -117,7 +117,7 @@ public:
                    callp->argTypes("vlSymsp");
                } else {
                    if (m_type.isCoverage()) callp->argTypes("first");
-                    callp->selfPointer("this");
+                    callp->selfPointer(VSelfPointerText{VSelfPointerText::This()});
                }
                rootFuncp->addStmtsp(callp->makeStmt());
            }
@@ -229,7 +229,7 @@ void V3CCtors::evalAsserts() {
                        // if (signal & CONST(upper_non_clean_mask)) { fail; }
                        AstVarRef* const vrefp
                            = new AstVarRef{varp->fileline(), varp, VAccess::READ};
-                        vrefp->selfPointer("this");
+                        vrefp->selfPointer(VSelfPointerText{VSelfPointerText::This()});
                        AstNodeExpr* newp = vrefp;
                        if (varp->isWide()) {
                            newp = new AstWordSel{
@@ -257,5 +257,5 @@ void V3CCtors::cctorsAll() {
    UINFO(2, __FUNCTION__ << ": " << endl);
    evalAsserts();
    { CCtorsVisitor{v3Global.rootp()}; }
-    V3Global::dumpCheckGlobalTree("cctors", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("cctors", 0, dumpTreeLevel() >= 3);
 }
@@ -28,9 +28,11 @@
 #include "V3CUse.h"

 #include "V3Ast.h"
+#include "V3FileLine.h"
 #include "V3Global.h"

-#include <set>
+#include <map>
+#include <utility>

 VL_DEFINE_DEBUG_FUNCTIONS;

@@ -45,32 +47,32 @@ class CUseVisitor final : public VNVisitor {
    const VNUser1InUse m_inuser1;

    // MEMBERS
-    bool m_impOnly = false;  // In details needed only for implementation
    AstNodeModule* const m_modp;  // Current module
-    std::set<std::pair<VUseType, std::string>> m_didUse;  // What we already used
+    std::map<std::string, std::pair<FileLine*, VUseType>> m_didUse;  // What we already used

    // METHODS
    void addNewUse(AstNode* nodep, VUseType useType, const string& name) {
-        if (m_didUse.emplace(useType, name).second) {
-            AstCUse* const newp = new AstCUse{nodep->fileline(), useType, name};
-            m_modp->addStmtsp(newp);
-            UINFO(8, "Insert " << newp << endl);
+        auto e = m_didUse.emplace(name, std::make_pair(nodep->fileline(), useType));
+        if (e.second || ((e.first->second.second & useType) != useType)) {
+            e.first->second.second = e.first->second.second | useType;
        }
    }

    // VISITORS
    void visit(AstClassRefDType* nodep) override {
-        if (nodep->user1SetOnce()) return;  // Process once
-        if (!m_impOnly) addNewUse(nodep, VUseType::INT_FWD_CLASS, nodep->classp()->name());
-        // Need to include extends() when we implement, but no need for pointers to know
-        VL_RESTORER(m_impOnly);
-        {
-            m_impOnly = true;
-            iterateChildren(nodep->classp());  // This also gets all extend classes
-        }
+        addNewUse(nodep, VUseType::INT_FWD_CLASS, nodep->classp()->name());
+    }
+    void visit(AstCFunc* nodep) override {
+        if (nodep->user1SetOnce()) return;
+        iterateAndNextNull(nodep->argsp());
+        iterateAndNextNull(nodep->stmtsp());
+    }
+    void visit(AstCCall* nodep) override { return; }
+    void visit(AstCReturn* nodep) override {
+        UASSERT(!nodep->user1SetOnce(), "Visited same return twice.");
+        iterate(nodep->lhsp()->dtypep());
    }
    void visit(AstNodeDType* nodep) override {
-        if (nodep->user1SetOnce()) return;  // Process once
        if (nodep->virtRefDTypep()) iterate(nodep->virtRefDTypep());
        if (nodep->virtRefDType2p()) iterate(nodep->virtRefDType2p());

@@ -79,11 +81,13 @@ class CUseVisitor final : public VNVisitor {
        if (stypep && stypep->classOrPackagep()) {
            addNewUse(nodep, VUseType::INT_INCLUDE, stypep->classOrPackagep()->name());
            iterateChildren(stypep);
+        } else if (AstClassRefDType* const classp = VN_CAST(nodep->skipRefp(), ClassRefDType)) {
+            addNewUse(nodep, VUseType::INT_FWD_CLASS, classp->name());
        }
    }
    void visit(AstNode* nodep) override {
        if (nodep->user1SetOnce()) return;  // Process once
-        if (nodep->dtypep() && !nodep->dtypep()->user1()) iterate(nodep->dtypep());
+        if (nodep->dtypep()) iterate(nodep->dtypep());
        iterateChildren(nodep);
    }
    void visit(AstCell* nodep) override {
@@ -98,6 +102,12 @@ public:
    explicit CUseVisitor(AstNodeModule* modp)
        : m_modp(modp) {
        iterate(modp);
+
+        for (auto& used : m_didUse) {
+            AstCUse* const newp = new AstCUse{used.second.first, used.second.second, used.first};
+            m_modp->addStmtsp(newp);
+            UINFO(8, "Insert " << newp << endl);
+        }
    }
    ~CUseVisitor() override = default;
    VL_UNCOPYABLE(CUseVisitor);
@@ -115,5 +125,5 @@ void V3CUse::cUseAll() {
        // for each output file and put under that
        CUseVisitor{modp};
    }
-    V3Global::dumpCheckGlobalTree("cuse", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("cuse", 0, dumpTreeLevel() >= 3);
 }
@@ -139,32 +139,42 @@ private:
    std::array<AstNode*, 1 << CASE_OVERLAP_WIDTH> m_valueItem;

    // METHODS
-    bool caseIsEnumComplete(AstCase* nodep, uint32_t numCases) {
-        // Return true if case is across an enum, and every value in the case
-        // statement corresponds to one of the enum values
-        if (!nodep->uniquePragma() && !nodep->unique0Pragma()) return false;
-        AstEnumDType* const enumDtp
+    //! Determine whether we should check case items are complete
+    //! @return  Enum's dtype if should check, nullptr if shouldn't
+    const AstEnumDType* getEnumCompletionCheckDType(const AstCase* const nodep) {
+        if (!nodep->uniquePragma() && !nodep->unique0Pragma()) return nullptr;
+        const AstEnumDType* const enumDtp
            = VN_CAST(nodep->exprp()->dtypep()->skipRefToEnump(), EnumDType);
-        if (!enumDtp) return false;  // Case isn't enum
-        AstBasicDType* const basicp = enumDtp->subDTypep()->basicp();
-        if (!basicp) return false;  // Not simple type (perhaps IEEE illegal)
-        if (basicp->width() > 32) return false;
-        // Find all case values into a set
-        std::set<uint32_t> caseSet;
-        for (uint32_t i = 0; i < numCases; ++i) {  // All case items
-            if (m_valueItem[i]) caseSet.emplace(i);
-        }
-        // Find all enum values into a set
-        std::set<uint32_t> enumSet;
-        for (AstEnumItem* itemp = enumDtp->itemsp(); itemp;
+        if (!enumDtp) return nullptr;  // Case isn't enum
+        const AstBasicDType* const basicp = enumDtp->subDTypep()->basicp();
+        if (!basicp) return nullptr;  // Not simple type (perhaps IEEE illegal)
+        if (basicp->width() > 32) return nullptr;
+        return enumDtp;
+    }
+    //! @return  True if case items are complete, false if there are uncovered enums
+    bool checkCaseEnumComplete(const AstCase* const nodep, const AstEnumDType* const dtype) {
+        const uint32_t numCases = 1UL << m_caseWidth;
+        for (AstEnumItem* itemp = dtype->itemsp(); itemp;
             itemp = VN_AS(itemp->nextp(), EnumItem)) {
            AstConst* const econstp = VN_AS(itemp->valuep(), Const);
-            const uint32_t val = econstp->toUInt();
-            // UINFO(9, "Complete enum item " << val << ": " << itemp << endl);
-            enumSet.emplace(val);
+            V3Number nummask{itemp, econstp->width()};
+            nummask.opBitsNonX(econstp->num());
+            const uint32_t mask = nummask.toUInt();
+            V3Number numval{itemp, econstp->width()};
+            numval.opBitsOne(econstp->num());
+            const uint32_t val = numval.toUInt();
+
+            for (uint32_t i = 0; i < numCases; ++i) {
+                if ((i & mask) == val) {
+                    if (!m_valueItem[i]) {
+                        nodep->v3warn(CASEINCOMPLETE, "Enum item " << itemp->prettyNameQ()
+                                                                   << " not covered by case\n");
+                        return false;  // enum has uncovered value by case items
+                    }
+                }
+            }
        }
-        // If sets match, all covered
-        return (caseSet == enumSet);
+        return true;  // enum is fully covered
    }
    bool isCaseTreeFast(AstCase* nodep) {
        int width = 0;
@@ -193,6 +203,8 @@ private:
        // We can cheat and use uint32_t's because we only support narrow case's
        bool reportedOverlap = false;
        bool reportedSubcase = false;
+        bool hasDefaultCase = false;
+        std::map<AstNode*, AstCaseItem*> caseItemMap;  // case condition -> case item
        for (AstCaseItem* itemp = nodep->itemsp(); itemp;
             itemp = VN_AS(itemp->nextp(), CaseItem)) {
            for (AstNode* icondp = itemp->condsp(); icondp; icondp = icondp->nextp()) {
@@ -202,6 +214,7 @@ private:
                if (neverItem(nodep, iconstp)) {
                    // X in casez can't ever be executed
                } else {
+                    const bool isCondWildcard = iconstp->num().isAnyXZ();
                    V3Number nummask{itemp, iconstp->width()};
                    nummask.opBitsNonX(iconstp->num());
                    const uint32_t mask = nummask.toUInt();
@@ -210,16 +223,17 @@ private:
                    const uint32_t val = numval.toUInt();

                    uint32_t firstOverlap = 0;
-                    bool foundOverlap = false;
+                    AstNode* overlappedCondp = nullptr;
                    bool foundHit = false;
                    for (uint32_t i = 0; i < numCases; ++i) {
                        if ((i & mask) == val) {
                            if (!m_valueItem[i]) {
-                                m_valueItem[i] = itemp;
+                                m_valueItem[i] = icondp;
+                                caseItemMap[icondp] = itemp;
                                foundHit = true;
-                            } else if (!foundOverlap) {
+                            } else if (!overlappedCondp) {
                                firstOverlap = i;
-                                foundOverlap = true;
+                                overlappedCondp = m_valueItem[i];
                                m_caseNoOverlapsAllCovered = false;
                            }
                        }
@@ -227,9 +241,19 @@ private:
                    if (!nodep->priorityPragma()) {
                        // If this case statement doesn't have the priority
                        // keyword, we want to warn on any overlap.
-                        if (!reportedOverlap && foundOverlap) {
-                            icondp->v3warn(CASEOVERLAP, "Case values overlap (example pattern 0x"
-                                                            << std::hex << firstOverlap << ")");
+                        if (!reportedOverlap && overlappedCondp) {
+                            std::ostringstream examplePattern;
+                            if (isCondWildcard) {
+                                examplePattern << " (example pattern 0x" << std::hex
+                                               << firstOverlap << ")";
+                            }
+                            icondp->v3warn(CASEOVERLAP,
+                                           "Case conditions overlap"
+                                               << examplePattern.str() << "\n"
+                                               << icondp->warnContextPrimary() << '\n'
+                                               << overlappedCondp->warnOther()
+                                               << "... Location of overlapping condition\n"
+                                               << overlappedCondp->warnContextSecondary());
                            reportedOverlap = true;
                        }
                    } else {
@@ -240,7 +264,11 @@ private:
                        if (!reportedSubcase && !foundHit) {
                            icondp->v3warn(CASEOVERLAP,
                                           "Case item ignored: every matching value is covered "
-                                           "by an earlier item");
+                                           "by an earlier condition\n"
+                                               << icondp->warnContextPrimary() << '\n'
+                                               << overlappedCondp->warnOther()
+                                               << "... Location of previous condition\n"
+                                               << overlappedCondp->warnContextPrimary());
                            reportedSubcase = true;
                        }
                    }
@@ -251,17 +279,28 @@ private:
                for (uint32_t i = 0; i < numCases; ++i) {
                    if (!m_valueItem[i]) m_valueItem[i] = itemp;
                }
+                caseItemMap[itemp] = itemp;
+                hasDefaultCase = true;
            }
        }
-        if (!caseIsEnumComplete(nodep, numCases)) {
-            for (uint32_t i = 0; i < numCases; ++i) {
-                if (!m_valueItem[i]) {
-                    nodep->v3warn(CASEINCOMPLETE, "Case values incompletely covered "
-                                                  "(example pattern 0x"
-                                                      << std::hex << i << ")");
+        if (!hasDefaultCase) {
+            const AstEnumDType* const dtype = getEnumCompletionCheckDType(nodep);
+            if (dtype) {
+                if (!checkCaseEnumComplete(nodep, dtype)) {
+                    // checkCaseEnumComplete has already warned of incompletion
                    m_caseNoOverlapsAllCovered = false;
                    return false;
                }
+            } else {
+                for (uint32_t i = 0; i < numCases; ++i) {
+                    if (!m_valueItem[i]) {  // has uncovered case
+                        nodep->v3warn(CASEINCOMPLETE, "Case values incompletely covered "
+                                                      "(example pattern 0x"
+                                                          << std::hex << i << ")");
+                        m_caseNoOverlapsAllCovered = false;
+                        return false;
+                    }
+                }
            }
        }

@@ -274,8 +313,10 @@ private:
        // Convert valueItem from AstCaseItem* to the expression
        // Not done earlier, as we may now have a nullptr because it's just a ";" NOP branch
        for (uint32_t i = 0; i < numCases; ++i) {
-            if (AstCaseItem* const itemp = VN_AS(m_valueItem[i], CaseItem)) {
-                m_valueItem[i] = itemp->stmtsp();
+            if (AstNode* const condp = m_valueItem[i]) {
+                AstCaseItem* caseItemp = caseItemMap[condp];
+                UASSERT(caseItemp, "caseItemp should exist");
+                m_valueItem[i] = caseItemp->stmtsp();
            }
        }
        return true;  // All is fine
@@ -543,10 +584,12 @@ private:
        }
    }
    //--------------------
-    void visit(AstNode* nodep) override {
-        if (VN_IS(nodep, Always)) m_alwaysp = nodep;
+    void visit(AstAlways* nodep) override {
+        VL_RESTORER(m_alwaysp)
+        m_alwaysp = nodep;
        iterateChildren(nodep);
    }
+    void visit(AstNode* nodep) override { iterateChildren(nodep); }

 public:
    // CONSTRUCTORS
@@ -566,7 +609,7 @@ public:
 void V3Case::caseAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { CaseVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("case", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("case", 0, dumpTreeLevel() >= 3);
 }
 void V3Case::caseLint(AstNodeCase* nodep) {
    UINFO(4, __FUNCTION__ << ": " << endl);
@@ -122,6 +122,31 @@ private:
        if (nodep->sizeMattersLhs()) ensureCast(nodep->lhsp());
        if (nodep->sizeMattersRhs()) ensureCast(nodep->rhsp());
    }
+    void visit(AstNodeCond* nodep) override {
+        // All class types are castable to each other. If they are of different types,
+        // a compilation error will be thrown, so an explicit cast is required. Types were
+        // already checked by V3Width and dtypep of a condition operator is a type of their
+        // common base class, so both classes can be safetly casted.
+        const AstClassRefDType* const thenClassDtypep
+            = VN_CAST(nodep->thenp()->dtypep(), ClassRefDType);
+        const AstClassRefDType* const elseClassDtypep
+            = VN_CAST(nodep->elsep()->dtypep(), ClassRefDType);
+        const bool castRequired = thenClassDtypep && elseClassDtypep
+                                  && (thenClassDtypep->classp() != elseClassDtypep->classp());
+        if (castRequired) {
+            const AstClass* const commonBaseClassp
+                = VN_AS(nodep->dtypep(), ClassRefDType)->classp();
+            if (thenClassDtypep->classp() != commonBaseClassp) {
+                AstNodeExpr* thenp = nodep->thenp()->unlinkFrBack();
+                nodep->thenp(new AstCCast{thenp->fileline(), thenp, nodep});
+            }
+            if (elseClassDtypep->classp() != commonBaseClassp) {
+                AstNodeExpr* elsep = nodep->elsep()->unlinkFrBack();
+                nodep->elsep(new AstCCast{elsep->fileline(), elsep, nodep});
+            }
+        }
+        visit(static_cast<AstNodeTriop*>(nodep));
+    }
    void visit(AstNodeTriop* nodep) override {
        iterateChildren(nodep);
        nodep->user1(nodep->lhsp()->user1() | nodep->rhsp()->user1() | nodep->thsp()->user1());
@@ -161,7 +186,8 @@ private:
            && !VN_IS(backp, NodeCCall) && !VN_IS(backp, CMethodHard) && !VN_IS(backp, SFormatF)
            && !VN_IS(backp, ArraySel) && !VN_IS(backp, StructSel) && !VN_IS(backp, RedXor)
            && (nodep->varp()->basicp() && !nodep->varp()->basicp()->isTriggerVec()
-                && !nodep->varp()->basicp()->isForkSync())
+                && !nodep->varp()->basicp()->isForkSync()
+                && !nodep->varp()->basicp()->isProcessRef())
            && backp->width() && castSize(nodep) != castSize(nodep->varp())) {
            // Cast vars to IData first, else below has upper bits wrongly set
            //  CData x=3; out = (QData)(x<<30);
@@ -208,5 +234,5 @@ public:
 void V3Cast::castAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { CastVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("cast", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("cast", 0, dumpTreeLevel() >= 3);
 }
@@ -27,6 +27,7 @@

 #include "V3Ast.h"
 #include "V3Global.h"
+#include "V3UniqueNames.h"

 VL_DEFINE_DEBUG_FUNCTIONS;

@@ -41,6 +42,7 @@ private:

    // MEMBERS
    string m_prefix;  // String prefix to add to name based on hier
+    V3UniqueNames m_names;  // For unique naming of structs and unions
    AstNodeModule* m_modp = nullptr;  // Current module
    AstNodeModule* m_classPackagep = nullptr;  // Package moving into
    const AstScope* m_classScopep = nullptr;  // Package moving scopes into
@@ -62,8 +64,10 @@ private:
        AstClassPackage* const packagep
            = new AstClassPackage{nodep->fileline(), nodep->origName()};
        packagep->name(nodep->name() + "__Vclpkg");
+        nodep->editCountInc();
        nodep->classOrPackagep(packagep);
        packagep->classp(nodep);
+        packagep->timeunit(nodep->timeunit());
        v3Global.rootp()->addModulesp(packagep);
        // Add package to hierarchy
        AstCell* const cellp = new AstCell{packagep->fileline(),
@@ -102,7 +106,6 @@ private:
            m_prefix = nodep->name() + "__02e";  // .
            iterateChildren(nodep);
        }
-        nodep->repairCache();
    }
    void visit(AstNodeModule* nodep) override {
        // Visit for NodeModules that are not AstClass (AstClass is-a AstNodeModule)
@@ -177,14 +180,15 @@ private:
    }

    void setStructModulep(AstNodeUOrStructDType* const dtypep) {
-        // Give it a pointer to its package and a final name
+        // Give struct a pointer to its package and a final name
+        dtypep->editCountInc();
        dtypep->classOrPackagep(m_classPackagep ? m_classPackagep : m_modp);
-        dtypep->name(dtypep->name() + (VN_IS(dtypep, UnionDType) ? "__union" : "__struct")
-                     + cvtToStr(dtypep->uniqueNum()));
+        dtypep->name(
+            m_names.get(dtypep->name() + (VN_IS(dtypep, UnionDType) ? "__union" : "__struct")));

        for (const AstMemberDType* itemp = dtypep->membersp(); itemp;
             itemp = VN_AS(itemp->nextp(), MemberDType)) {
-            AstNodeUOrStructDType* const subp = VN_CAST(itemp->skipRefp(), NodeUOrStructDType);
+            AstNodeUOrStructDType* const subp = itemp->getChildStructp();
            // Recurse only into anonymous unpacked structs inside this definition,
            // other unpacked structs will be reached from another typedefs
            if (subp && !subp->packed() && subp->name().empty()) setStructModulep(subp);
@@ -244,5 +248,5 @@ public:
 void V3Class::classAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { ClassVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("class", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("class", 0, dumpTreeLevel() >= 3);
 }
@@ -95,6 +95,7 @@ private:
                || VN_IS(nodep->dtypep()->skipRefp(), DynArrayDType)
                || VN_IS(nodep->dtypep()->skipRefp(), ClassRefDType)
                || VN_IS(nodep->dtypep()->skipRefp(), QueueDType)
+                || VN_IS(nodep->dtypep()->skipRefp(), StreamDType)
                || VN_IS(nodep->dtypep()->skipRefp(), UnpackArrayDType)
                || VN_IS(nodep->dtypep()->skipRefp(), VoidDType)) {
            } else {
@@ -298,7 +299,11 @@ private:
    }
    void visit(AstWith* nodep) override {
        iterateChildren(nodep);
-        ensureCleanAndNext(nodep->exprp());
+        setClean(nodep, true);
+    }
+    void visit(AstCReturn* nodep) override {
+        iterateChildren(nodep);
+        ensureClean(nodep->lhsp());
        setClean(nodep, true);
    }
    void visit(AstIntfRef* nodep) override {
@@ -325,5 +330,5 @@ public:
 void V3Clean::cleanAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { CleanVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("clean", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("clean", 0, dumpTreeLevel() >= 3);
 }
@@ -224,5 +224,5 @@ public:
 void V3Clock::clockAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { ClockVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("clock", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("clock", 0, dumpTreeLevel() >= 3);
 }
@@ -238,5 +238,5 @@ public:
 void V3Combine::combineAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    CombineVisitor::apply(nodep);
-    V3Global::dumpCheckGlobalTree("combine", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("combine", 0, dumpTreeLevel() >= 3);
 }
@@ -63,6 +63,7 @@ static void makeVlToString(AstIface* nodep) {
 }
 static void makeVlToString(AstNodeUOrStructDType* nodep) {
    AstNodeModule* const modp = nodep->classOrPackagep();
+    UASSERT_OBJ(modp, nodep, "Unlinked struct package");
    AstCFunc* const funcp
        = new AstCFunc{nodep->fileline(), "VL_TO_STRING", nullptr, "std::string"};
    funcp->argTypes("const " + EmitCBase::prefixNameProtect(nodep) + "& obj");
@@ -81,11 +82,13 @@ static void makeVlToString(AstNodeUOrStructDType* nodep) {
        }
        stmt += VIdProtect::protect(itemp->prettyName()) + ":\" + ";
        if (VN_IS(itemp->dtypep()->skipRefp(), BasicDType) && itemp->isWide()) {
-            stmt += "VL_TO_STRING_W";
+            stmt += "VL_TO_STRING_W(";
+            stmt += cvtToStr(itemp->widthWords());
+            stmt += ", ";
        } else {
-            stmt += "VL_TO_STRING";
+            stmt += "VL_TO_STRING(";
        }
-        stmt += "(obj." + itemp->nameProtect() + ");\n";
+        stmt += "obj." + itemp->nameProtect() + ");\n";
        funcp->addStmtsp(new AstCStmt{nodep->fileline(), stmt});
    }
    funcp->addStmtsp(new AstCStmt{nodep->fileline(), "out += \"}\";\n"});
@@ -172,5 +175,5 @@ void V3Common::commonAll() {
            if (!dtypep->packed()) makeVlToString(dtypep);
        }
    }
-    V3Global::dumpCheckGlobalTree("common", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("common", 0, dumpTreeLevel() >= 3);
 }
@@ -1799,7 +1799,7 @@ private:
        UASSERT_OBJ((rstart + rwidth) == lstart, nodep,
                    "tried to merge two selects which are not adjacent");
        AstSel* const newselp = new AstSel{
-            lselp->fromp()->fileline(), rselp->fromp()->cloneTree(false), rstart, lwidth + rwidth};
+            lselp->fromp()->fileline(), rselp->fromp()->unlinkFrBack(), rstart, lwidth + rwidth};
        UINFO(5, "merged two adjacent sel " << lselp << " and " << rselp << " to one " << newselp
                                            << endl);

@@ -1809,6 +1809,7 @@ private:
        VL_DO_DANGLING(nodep->deleteTree(), nodep);
    }
    void replaceConcatMerge(AstConcat* nodep) {
+        // {llp OP lrp, rlp OP rrp} => {llp, rlp} OP {lrp, rrp}, where OP = AND/OR/XOR
        AstNodeBiop* const lp = VN_AS(nodep->lhsp(), NodeBiop);
        AstNodeBiop* const rp = VN_AS(nodep->rhsp(), NodeBiop);
        AstNodeExpr* const llp = lp->lhsp()->cloneTree(false);
@@ -2052,7 +2053,7 @@ private:
                VL_DO_DANGLING(nodep->unlinkFrBack()->deleteTree(), nodep);
                return true;
            }
-        } else if (m_doV && VN_IS(nodep->lhsp(), Concat)) {
+        } else if (m_doV && VN_IS(nodep->lhsp(), Concat) && nodep->isTreePureRecurse()) {
            bool need_temp = false;
            if (m_warn && !VN_IS(nodep, AssignDly)) {  // Is same var on LHS and RHS?
                // Note only do this (need user4) when m_warn, which is
@@ -2145,33 +2146,42 @@ private:
            return true;
        } else if (m_doV && VN_IS(nodep->rhsp(), StreamR)) {
            // The right-streaming operator on rhs of assignment does not
-            // change the order of bits. Eliminate stream but keep its lhsp
-            // Unlink the stuff
-            AstNodeExpr* const srcp = VN_AS(nodep->rhsp(), StreamR)->lhsp()->unlinkFrBack();
-            AstNode* const sizep = VN_AS(nodep->rhsp(), StreamR)->rhsp()->unlinkFrBack();
-            AstNodeExpr* const streamp = VN_AS(nodep->rhsp(), StreamR)->unlinkFrBack();
+            // change the order of bits. Eliminate stream but keep its lhsp.
+            // Add a cast if needed.
+            AstStreamR* const streamp = VN_AS(nodep->rhsp(), StreamR)->unlinkFrBack();
+            AstNodeExpr* srcp = streamp->lhsp()->unlinkFrBack();
+            AstNodeDType* const srcDTypep = srcp->dtypep();
+            if (VN_IS(srcDTypep, QueueDType) || VN_IS(srcDTypep, DynArrayDType)) {
+                if (nodep->lhsp()->widthMin() > 64) {
+                    nodep->v3warn(E_UNSUPPORTED, "Unsupported: Assignment of stream of dynamic "
+                                                 "array to a variable of size greater than 64");
+                }
+                srcp = new AstCvtDynArrayToPacked{srcp->fileline(), srcp, srcDTypep};
+            }
            nodep->rhsp(srcp);
-            // Cleanup
-            VL_DO_DANGLING(sizep->deleteTree(), sizep);
            VL_DO_DANGLING(streamp->deleteTree(), streamp);
            // Further reduce, any of the nodes may have more reductions.
            return true;
        } else if (m_doV && VN_IS(nodep->lhsp(), StreamL)) {
            // Push the stream operator to the rhs of the assignment statement
-            const int dWidth = VN_AS(nodep->lhsp(), StreamL)->lhsp()->width();
-            const int sWidth = nodep->rhsp()->width();
-            // Unlink the stuff
-            AstNodeExpr* const dstp = VN_AS(nodep->lhsp(), StreamL)->lhsp()->unlinkFrBack();
-            AstNodeExpr* streamp = VN_AS(nodep->lhsp(), StreamL)->unlinkFrBack();
+            AstNodeExpr* streamp = nodep->lhsp()->unlinkFrBack();
+            AstNodeExpr* const dstp = VN_AS(streamp, StreamL)->lhsp()->unlinkFrBack();
            AstNodeExpr* const srcp = nodep->rhsp()->unlinkFrBack();
+            const int sWidth = srcp->width();
+            const int dWidth = dstp->width();
            // Connect the rhs to the stream operator and update its width
            VN_AS(streamp, StreamL)->lhsp(srcp);
-            streamp->dtypeSetLogicUnsized(srcp->width(), srcp->widthMin(), VSigning::UNSIGNED);
-            // Shrink the RHS if necessary
-            if (sWidth > dWidth) {
+            if (VN_IS(srcp->dtypep(), DynArrayDType) || VN_IS(srcp->dtypep(), QueueDType)
+                || VN_IS(srcp->dtypep(), UnpackArrayDType)) {
+                streamp->dtypeSetStream();
+            } else {
+                streamp->dtypeSetLogicUnsized(srcp->width(), srcp->widthMin(), VSigning::UNSIGNED);
+            }
+            if (dWidth == 0) {
+                streamp = new AstCvtPackedToDynArray{nodep->fileline(), streamp, dstp->dtypep()};
+            } else if (sWidth > dWidth) {
                streamp = new AstSel{streamp->fileline(), streamp, sWidth - dWidth, dWidth};
            }
-            // Link the nodes back in
            nodep->lhsp(dstp);
            nodep->rhsp(streamp);
            return true;
@@ -2179,23 +2189,35 @@ private:
            // The right stream operator on lhs of assignment statement does
            // not reorder bits. However, if the rhs is wider than the lhs,
            // then we select bits from the left-most, not the right-most.
-            const int dWidth = VN_AS(nodep->lhsp(), StreamR)->lhsp()->width();
-            const int sWidth = nodep->rhsp()->width();
-            // Unlink the stuff
-            AstNodeExpr* const dstp = VN_AS(nodep->lhsp(), StreamR)->lhsp()->unlinkFrBack();
-            AstNode* const sizep = VN_AS(nodep->lhsp(), StreamR)->rhsp()->unlinkFrBack();
-            AstNodeExpr* const streamp = VN_AS(nodep->lhsp(), StreamR)->unlinkFrBack();
+            AstNodeExpr* const streamp = nodep->lhsp()->unlinkFrBack();
+            AstNodeExpr* const dstp = VN_AS(streamp, StreamR)->lhsp()->unlinkFrBack();
            AstNodeExpr* srcp = nodep->rhsp()->unlinkFrBack();
-            if (sWidth > dWidth) {
+            const int sWidth = srcp->width();
+            const int dWidth = dstp->width();
+            if (dWidth == 0) {
+                srcp = new AstCvtPackedToDynArray{nodep->fileline(), srcp, dstp->dtypep()};
+            } else if (sWidth > dWidth) {
                srcp = new AstSel{streamp->fileline(), srcp, sWidth - dWidth, dWidth};
            }
            nodep->lhsp(dstp);
            nodep->rhsp(srcp);
-            // Cleanup
-            VL_DO_DANGLING(sizep->deleteTree(), sizep);
            VL_DO_DANGLING(streamp->deleteTree(), streamp);
            // Further reduce, any of the nodes may have more reductions.
            return true;
+        } else if (m_doV && VN_IS(nodep->rhsp(), StreamL)) {
+            AstNodeDType* const lhsDtypep = nodep->lhsp()->dtypep();
+            AstStreamL* streamp = VN_AS(nodep->rhsp(), StreamL);
+            AstNodeExpr* const srcp = streamp->lhsp();
+            const AstNodeDType* const srcDTypep = srcp->dtypep();
+            if (VN_IS(srcDTypep, QueueDType) || VN_IS(srcDTypep, DynArrayDType)) {
+                if (lhsDtypep->widthMin() > 64) {
+                    nodep->v3warn(E_UNSUPPORTED, "Unsupported: Assignment of stream of dynamic "
+                                                 "array to a variable of size greater than 64");
+                }
+                srcp->unlinkFrBack();
+                streamp->lhsp(new AstCvtDynArrayToPacked{srcp->fileline(), srcp, lhsDtypep});
+                streamp->dtypeFrom(lhsDtypep);
+            }
        } else if (m_doV && replaceAssignMultiSel(nodep)) {
            return true;
        }
@@ -2283,6 +2305,13 @@ private:
            iterateChildren(nodep);
        }
    }
+    void visit(AstCLocalScope* nodep) override {
+        iterateChildren(nodep);
+        if (!nodep->stmtsp()) {
+            nodep->unlinkFrBack();
+            VL_DO_DANGLING(nodep->deleteTree(), nodep);
+        }
+    }
    void visit(AstScope* nodep) override {
        // No ASSIGNW removals under scope, we've long eliminated INITIALs
        VL_RESTORER(m_wremove);
@@ -3064,7 +3093,7 @@ private:
                  && nodep->displayType() == VDisplayType::DT_WRITE)))
            return false;
        if ((prevp->filep() && !nodep->filep()) || (!prevp->filep() && nodep->filep())
-            || !prevp->filep()->sameTree(nodep->filep()))
+            || (prevp->filep() && nodep->filep() && !prevp->filep()->sameTree(nodep->filep())))
            return false;
        if (!prevp->fmtp() || prevp->fmtp()->nextp() || !nodep->fmtp() || nodep->fmtp()->nextp())
            return false;
@@ -3175,8 +3204,17 @@ private:
        iterateChildren(nodep);
    }

-    void visit(AstFuncRef* nodep) override {
+    void visit(AstNodeCCall* nodep) override {
        iterateChildren(nodep);
+        m_hasJumpDelay = true;  // As don't analyze inside tasks for timing controls
+    }
+    void visit(AstNodeFTaskRef* nodep) override {
+        // Note excludes AstFuncRef as other visitor below
+        iterateChildren(nodep);
+        m_hasJumpDelay = true;  // As don't analyze inside tasks for timing controls
+    }
+    void visit(AstFuncRef* nodep) override {
+        visit(static_cast<AstNodeFTaskRef*>(nodep));
        if (m_params) {  // Only parameters force us to do constant function call propagation
            replaceWithSimulation(nodep);
        }
@@ -3405,9 +3443,9 @@ private:
    TREEOPS("AstCond {$lhsp.isNeqZero}",        "replaceWIteratedRhs(nodep)");
    TREEOP ("AstCond{$condp.castNot,       $thenp, $elsep}", "AstCond{$condp->castNot()->lhsp(), $elsep, $thenp}");
    TREEOP ("AstNodeCond{$condp.width1, $thenp.width1,   $thenp.isAllOnes, $elsep}", "AstLogOr {$condp, $elsep}");  // a?1:b == a||b
-    TREEOP ("AstNodeCond{$condp.width1, $thenp.width1,   $thenp,    $elsep.isZero}", "AstLogAnd{$condp, $thenp}");  // a?b:0 == a&&b
+    TREEOP ("AstNodeCond{$condp.width1, $thenp.width1,   $thenp,    $elsep.isZero, !$elsep.isClassHandleValue}", "AstLogAnd{$condp, $thenp}");  // a?b:0 == a&&b
    TREEOP ("AstNodeCond{$condp.width1, $thenp.width1,   $thenp, $elsep.isAllOnes}", "AstLogOr {AstNot{$condp}, $thenp}");  // a?b:1 == ~a||b
-    TREEOP ("AstNodeCond{$condp.width1, $thenp.width1,   $thenp.isZero,    $elsep}", "AstLogAnd{AstNot{$condp}, $elsep}");  // a?0:b == ~a&&b
+    TREEOP ("AstNodeCond{$condp.width1, $thenp.width1,   $thenp.isZero, !$thenp.isClassHandleValue,   $elsep}", "AstLogAnd{AstNot{$condp}, $elsep}");  // a?0:b == ~a&&b
    TREEOP ("AstNodeCond{!$condp.width1, operandBoolShift(nodep->condp())}", "replaceBoolShift(nodep->condp())");
    // Prefer constants on left, since that often needs a shift, it lets
    // constant red remove the shift
@@ -3650,6 +3688,7 @@ private:
 public:
    // Processing Mode Enum
    enum ProcMode : uint8_t {
+        PROC_PARAMS_NOWARN,
        PROC_PARAMS,
        PROC_GENERATE,
        PROC_LIVE,
@@ -3665,16 +3704,18 @@ public:
        , m_concswapNames{globalPass ? ("__Vconcswap_" + cvtToStr(s_globalPassNum++)) : ""} {
        // clang-format off
        switch (pmode) {
-        case PROC_PARAMS:       m_doV = true;  m_doNConst = true; m_params = true;
-                                m_required = true; break;
-        case PROC_GENERATE:     m_doV = true;  m_doNConst = true; m_params = true;
-                                m_required = true; m_doGenerate = true; break;
-        case PROC_LIVE:         break;
-        case PROC_V_WARN:       m_doV = true;  m_doNConst = true; m_warn = true; break;
-        case PROC_V_NOWARN:     m_doV = true;  m_doNConst = true; break;
-        case PROC_V_EXPENSIVE:  m_doV = true;  m_doNConst = true; m_doExpensive = true; break;
-        case PROC_CPP:          m_doV = false; m_doNConst = true; m_doCpp = true; break;
-        default:                v3fatalSrc("Bad case"); break;
+        case PROC_PARAMS_NOWARN:  m_doV = true;  m_doNConst = true; m_params = true;
+                                  m_required = false; break;
+        case PROC_PARAMS:         m_doV = true;  m_doNConst = true; m_params = true;
+                                  m_required = true; break;
+        case PROC_GENERATE:       m_doV = true;  m_doNConst = true; m_params = true;
+                                  m_required = true; m_doGenerate = true; break;
+        case PROC_LIVE:           break;
+        case PROC_V_WARN:         m_doV = true;  m_doNConst = true; m_warn = true; break;
+        case PROC_V_NOWARN:       m_doV = true;  m_doNConst = true; break;
+        case PROC_V_EXPENSIVE:    m_doV = true;  m_doNConst = true; m_doExpensive = true; break;
+        case PROC_CPP:            m_doV = false; m_doNConst = true; m_doCpp = true; break;
+        default:                  v3fatalSrc("Bad case"); break;
        }
        // clang-format on
    }
@@ -3722,6 +3763,29 @@ AstNode* V3Const::constifyParamsEdit(AstNode* nodep) {
    return nodep;
 }

+//! Constify this cell node's parameter list if possible
+//! @return  Pointer to the edited node.
+AstNode* V3Const::constifyParamsNoWarnEdit(AstNode* nodep) {
+    // if (debug() > 0) nodep->dumpTree("-  forceConPRE : ");
+    // Resize even if the node already has a width, because buried in the tree
+    // we may have a node we just created with signing, etc, that isn't sized yet.
+
+    // Make sure we've sized everything first
+    nodep = V3Width::widthParamsEdit(nodep);
+    ConstVisitor visitor{ConstVisitor::PROC_PARAMS_NOWARN, /* globalPass: */ false};
+    if (AstVar* const varp = VN_CAST(nodep, Var)) {
+        // If a var wants to be constified, it's really a param, and
+        // we want the value to be constant.  We aren't passed just the
+        // init value because we need widthing above to handle the var's type.
+        if (varp->valuep()) visitor.mainAcceptEdit(varp->valuep());
+    } else {
+        nodep = visitor.mainAcceptEdit(nodep);
+    }
+    // Because we do edits, nodep links may get trashed and core dump this.
+    // if (debug() > 0) nodep->dumpTree("-  forceConDONE: ");
+    return nodep;
+}
+
 //! Force this cell node's parameter list to become a constant inside generate.
 //! If we are inside a generated "if", "case" or "for", we don't want to
 //! trigger warnings when we deal with the width. It is possible that these
@@ -3759,7 +3823,7 @@ void V3Const::constifyAllLint(AstNetlist* nodep) {
        ConstVisitor visitor{ConstVisitor::PROC_V_WARN, /* globalPass: */ true};
        (void)visitor.mainAcceptEdit(nodep);
    }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("const", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("const", 0, dumpTreeLevel() >= 3);
 }

 void V3Const::constifyCpp(AstNetlist* nodep) {
@@ -3768,7 +3832,7 @@ void V3Const::constifyCpp(AstNetlist* nodep) {
        ConstVisitor visitor{ConstVisitor::PROC_CPP, /* globalPass: */ true};
        (void)visitor.mainAcceptEdit(nodep);
    }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("const_cpp", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("const_cpp", 0, dumpTreeLevel() >= 3);
 }

 AstNode* V3Const::constifyEdit(AstNode* nodep) {
@@ -3792,7 +3856,7 @@ void V3Const::constifyAllLive(AstNetlist* nodep) {
        ConstVisitor visitor{ConstVisitor::PROC_LIVE, /* globalPass: */ true};
        (void)visitor.mainAcceptEdit(nodep);
    }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("const", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("const", 0, dumpTreeLevel() >= 3);
 }

 void V3Const::constifyAll(AstNetlist* nodep) {
@@ -3802,7 +3866,7 @@ void V3Const::constifyAll(AstNetlist* nodep) {
        ConstVisitor visitor{ConstVisitor::PROC_V_EXPENSIVE, /* globalPass: */ true};
        (void)visitor.mainAcceptEdit(nodep);
    }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("const", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("const", 0, dumpTreeLevel() >= 3);
 }

 AstNode* V3Const::constifyExpensiveEdit(AstNode* nodep) {
@@ -30,6 +30,10 @@ public:
    static AstNodeExpr* constifyParamsEdit(AstNodeExpr* exprp) {
        return VN_AS(constifyParamsEdit(static_cast<AstNode*>(exprp)), NodeExpr);
    }
+    static AstNode* constifyParamsNoWarnEdit(AstNode* nodep);
+    static AstNodeExpr* constifyParamsNoWarnEdit(AstNodeExpr* exprp) {
+        return VN_AS(constifyParamsNoWarnEdit(static_cast<AstNode*>(exprp)), NodeExpr);
+    }
    static AstNode* constifyGenerateParamsEdit(AstNode* nodep);
    // Only do constant pushing, without removing dead logic
    static void constifyAllLive(AstNetlist* nodep);
@@ -567,5 +567,5 @@ public:
 void V3Coverage::coverage(AstNetlist* rootp) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { CoverageVisitor{rootp}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("coverage", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("coverage", 0, dumpTreeLevel() >= 3);
 }
@@ -117,5 +117,5 @@ public:
 void V3CoverageJoin::coverageJoin(AstNetlist* rootp) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { CoverageJoinVisitor{rootp}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("coveragejoin", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("coveragejoin", 0, dumpTreeLevel() >= 3);
 }
@@ -62,9 +62,10 @@ private:
    // TYPES
    using AssignMap = std::multimap<AstVarScope*, AstNodeAssign*>;

-    // STATE
-    AstNodeModule* m_modp = nullptr;  // Current module
-    AstSelLoopVars* m_selloopvarsp = nullptr;  // Current loop vars
+    // STATE - across all visitors
+    const bool m_elimUserVars;  // Allow removal of user's vars
+    const bool m_elimDTypes;  // Allow removal of DTypes
+    const bool m_elimCells;  // Allow removal of Cells
    // List of all encountered to avoid another loop through tree
    std::vector<AstVar*> m_varsp;
    std::vector<AstNode*> m_dtypesp;
@@ -73,13 +74,14 @@ private:
    std::vector<AstCell*> m_cellsp;
    std::vector<AstClass*> m_classesp;
    std::vector<AstTypedef*> m_typedefsp;
-
    AssignMap m_assignMap;  // List of all simple assignments for each variable
-    const bool m_elimUserVars;  // Allow removal of user's vars
-    const bool m_elimDTypes;  // Allow removal of DTypes
-    const bool m_elimCells;  // Allow removal of Cells
    bool m_sideEffect = false;  // Side effects discovered in assign RHS

+    // STATE - for current visit position (use VL_RESTORER)
+    bool m_inAssign = false;  // Currently in an assign
+    AstNodeModule* m_modp = nullptr;  // Current module
+    AstSelLoopVars* m_selloopvarsp = nullptr;  // Current loop vars
+
    // METHODS

    void checkAll(AstNode* nodep) {
@@ -107,18 +109,16 @@ private:
    void visit(AstNodeModule* nodep) override {
        if (m_modp) m_modp->user1Inc();  // e.g. Class under Package
        VL_RESTORER(m_modp);
-        {
-            m_modp = nodep;
-            if (!nodep->dead()) {
-                iterateChildren(nodep);
-                checkAll(nodep);
-                if (AstClass* const classp = VN_CAST(nodep, Class)) {
-                    if (classp->extendsp()) classp->extendsp()->user1Inc();
-                    if (classp->classOrPackagep()) classp->classOrPackagep()->user1Inc();
-                    m_classesp.push_back(classp);
-                    // TODO we don't reclaim dead classes yet - graph implementation instead?
-                    classp->user1Inc();
-                }
+        m_modp = nodep;
+        if (!nodep->dead()) {
+            iterateChildren(nodep);
+            checkAll(nodep);
+            if (AstClass* const classp = VN_CAST(nodep, Class)) {
+                if (classp->extendsp()) classp->extendsp()->user1Inc();
+                if (classp->classOrPackagep()) classp->classOrPackagep()->user1Inc();
+                m_classesp.push_back(classp);
+                // TODO we don't reclaim dead classes yet - graph implementation instead?
+                classp->user1Inc();
            }
        }
    }
@@ -282,8 +282,11 @@ private:
        // See if simple assignments to variables may be eliminated because
        // that variable is never used.
        // Similar code in V3Life
-        VL_RESTORER(m_sideEffect);
+        const bool assignInAssign = m_inAssign;  // Might be Assign(..., ExprStmt(Assign), ...)
        {
+            VL_RESTORER(m_inAssign);
+            VL_RESTORER(m_sideEffect);
+            m_inAssign = true;
            m_sideEffect = false;
            iterateAndNextNull(nodep->rhsp());
            checkAll(nodep);
@@ -299,6 +302,7 @@ private:
            }
            iterateNull(nodep->timingControlp());
        }
+        if (assignInAssign) m_sideEffect = true;  // Parent assign shouldn't optimize
    }

    //-----
@@ -539,29 +543,29 @@ void V3Dead::deadifyModules(AstNetlist* nodep) {
    {
        DeadVisitor{nodep, false, false, false, false, !v3Global.opt.topIfacesSupported()};
    }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("deadModules", 0, dumpTree() >= 6);
+    V3Global::dumpCheckGlobalTree("deadModules", 0, dumpTreeLevel() >= 6);
 }

 void V3Dead::deadifyDTypes(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { DeadVisitor{nodep, false, true, false, false, false}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("deadDtypes", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("deadDtypes", 0, dumpTreeLevel() >= 3);
 }

 void V3Dead::deadifyDTypesScoped(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { DeadVisitor{nodep, false, true, true, false, false}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("deadDtypesScoped", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("deadDtypesScoped", 0, dumpTreeLevel() >= 3);
 }

 void V3Dead::deadifyAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { DeadVisitor{nodep, true, true, false, true, false}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("deadAll", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("deadAll", 0, dumpTreeLevel() >= 3);
 }

 void V3Dead::deadifyAllScoped(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { DeadVisitor{nodep, true, true, true, true, false}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("deadAllScoped", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("deadAllScoped", 0, dumpTreeLevel() >= 3);
 }
@@ -92,21 +92,23 @@ private:
    const VNUser4InUse m_inuser4;
    const VNUser5InUse m_inuser5;

-    // STATE
+    // STATE - across all visitors
+    std::unordered_map<const AstVarScope*, int> m_scopeVecMap;  // Next var number for each scope
+    std::set<AstSenTree*> m_timingDomains;  // Timing resume domains
+    using VarMap = std::map<const std::pair<AstNodeModule*, std::string>, AstVar*>;
+    VarMap m_modVarMap;  // Table of new var names created under module
+    VDouble0 m_statSharedSet;  // Statistic tracking
+
+    // STATE - for current visit position (use VL_RESTORER)
    AstActive* m_activep = nullptr;  // Current activate
    const AstCFunc* m_cfuncp = nullptr;  // Current public C Function
    AstAssignDly* m_nextDlyp = nullptr;  // Next delayed assignment in a list of assignments
    AstNodeProcedure* m_procp = nullptr;  // Current process
-    std::set<AstSenTree*> m_timingDomains;  // Timing resume domains
    bool m_inDly = false;  // True in delayed assignments
    bool m_inLoop = false;  // True in for loops
    bool m_inInitial = false;  // True in static initializers and initial blocks
    bool m_inSuspendableOrFork = false;  // True in suspendable processes and forks
    bool m_ignoreBlkAndNBlk = false;  // Suppress delayed assignment BLKANDNBLK
-    using VarMap = std::map<const std::pair<AstNodeModule*, std::string>, AstVar*>;
-    VarMap m_modVarMap;  // Table of new var names created under module
-    VDouble0 m_statSharedSet;  // Statistic tracking
-    std::unordered_map<const AstVarScope*, int> m_scopeVecMap;  // Next var number for each scope

    // METHODS

@@ -424,10 +426,8 @@ private:
    }
    void visit(AstCFunc* nodep) override {
        VL_RESTORER(m_cfuncp);
-        {
-            m_cfuncp = nodep;
-            iterateChildren(nodep);
-        }
+        m_cfuncp = nodep;
+        iterateChildren(nodep);
    }
    void visit(AstActive* nodep) override {
        m_activep = nodep;
@@ -443,10 +443,12 @@ private:
    void visit(AstNodeProcedure* nodep) override {
        VL_RESTORER(m_inSuspendableOrFork);
        m_inSuspendableOrFork = nodep->isSuspendable();
-        m_procp = nodep;
-        m_timingDomains.clear();
-        iterateChildren(nodep);
-        m_procp = nullptr;
+        {
+            VL_RESTORER(m_procp);
+            m_procp = nodep;
+            m_timingDomains.clear();
+            iterateChildren(nodep);
+        }
        if (m_timingDomains.empty()) return;
        if (auto* const actp = VN_AS(nodep->user3p(), Active)) {
            // Merge all timing domains (and possibly the active's domain) to create a sentree for
@@ -549,7 +551,9 @@ private:
            }
            if (!lhsp->backp()) VL_DO_DANGLING(pushDeletep(lhsp), lhsp);
        } else {
-            iterateChildren(nodep);
+            iterate(nodep->lhsp());
+            m_inDly = false;
+            iterate(nodep->rhsp());
        }
    }

@@ -618,10 +622,14 @@ private:
    }
    void visit(AstWhile* nodep) override {
        VL_RESTORER(m_inLoop);
-        {
-            m_inLoop = true;
-            iterateChildren(nodep);
-        }
+        m_inLoop = true;
+        iterateChildren(nodep);
+    }
+    void visit(AstExprStmt* nodep) override {
+        VL_RESTORER(m_inDly);
+        // Restoring is needed, because AstExprStmt may contain assignments
+        m_inDly = false;
+        iterateChildren(nodep);
    }

    //--------------------
@@ -641,5 +649,5 @@ public:
 void V3Delayed::delayedAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { DelayedVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("delayed", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("delayed", 0, dumpTreeLevel() >= 3);
 }
@@ -166,5 +166,5 @@ public:
 void V3Depth::depthAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { DepthVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("depth", 0, dumpTree() >= 6);
+    V3Global::dumpCheckGlobalTree("depth", 0, dumpTreeLevel() >= 6);
 }
@@ -127,5 +127,5 @@ public:
 void V3DepthBlock::depthBlockAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { DepthBlockVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("deepblock", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("deepblock", 0, dumpTreeLevel() >= 3);
 }
@@ -46,6 +46,10 @@ private:

    // TYPES
    using FuncMmap = std::multimap<std::string, AstCFunc*>;
+    struct ScopeSelfPtr final {
+        VSelfPointerText thisPtr = VSelfPointerText{VSelfPointerText::Empty()};
+        VSelfPointerText vlSymsPtr = VSelfPointerText{VSelfPointerText::Empty()};
+    };

    // STATE
    AstNodeModule* m_modp = nullptr;  // Current module
@@ -53,6 +57,7 @@ private:
    const AstCFunc* m_funcp = nullptr;  // Current function
    bool m_modSingleton = false;  // m_modp is only instantiated once
    FuncMmap m_modFuncs;  // Name of public functions added
+    std::map<const AstScope*, ScopeSelfPtr> m_scopeToSelf;  // Scope to self pointers

    // METHODS

@@ -68,12 +73,32 @@ private:
        return (instances == 1);
    }

+    // Construct a 'this' self pointer for the given scope
+    VSelfPointerText scopeThis(const AstScope* scopep) {
+        auto& ret = m_scopeToSelf[scopep];
+        if (ret.thisPtr.isEmpty()) {
+            string name = scopep->name();
+            string::size_type pos;
+            if ((pos = name.rfind('.')) != string::npos) name.erase(0, pos + 1);
+            ret.thisPtr = VSelfPointerText{VSelfPointerText::This(), name};
+        }
+        return ret.thisPtr;
+    }
+    // Construct a 'vlSyms' self pointer for the given scope
+    VSelfPointerText scopeVlSyms(const AstScope* scopep) {
+        auto& ret = m_scopeToSelf[scopep];
+        if (ret.vlSymsPtr.isEmpty()) {
+            ret.vlSymsPtr = VSelfPointerText{VSelfPointerText::VlSyms(), scopep->nameDotless()};
+        }
+        return ret.vlSymsPtr;
+    }
+
    // Construct the best self pointer to reference an object in 'scopep'  from a CFunc in
    // 'm_scopep'. Result may be relative ("this->[...]") or absolute ("vlSyms->[...]").
    //
    // Using relative references allows V3Combine'ing code across multiple instances of the same
    // module.
-    string descopedSelfPointer(const AstScope* scopep) {
+    VSelfPointerText descopedSelfPointer(const AstScope* scopep) {
        UASSERT(scopep, "Var/Func not scoped");
        // Static functions can't use relative references via 'this->'
        const bool relativeRefOk = !m_funcp->isStatic();
@@ -85,23 +110,20 @@ private:
        if (VN_IS(scopep->modp(), Class)) {
            // Direct reference to class members are from within the class itself, references from
            // outside the class must go via AstMemberSel
-            return "this";
+            return VSelfPointerText{VSelfPointerText::This()};
        } else if (relativeRefOk && scopep == m_scopep) {
-            return "this";
+            return VSelfPointerText{VSelfPointerText::This()};
        } else if (relativeRefOk && !m_modSingleton && scopep->aboveScopep() == m_scopep
                   && VN_IS(scopep->modp(), Module)) {
            // Reference to scope of instance directly under this module, can just "this->cell",
            // which can potentially be V3Combined, but note this requires one extra pointer
            // dereference which is slower, so we only use it if the source scope is not a
            // singleton.
-            string name = scopep->name();
-            string::size_type pos;
-            if ((pos = name.rfind('.')) != string::npos) name.erase(0, pos + 1);
-            return "this->" + name;
+            return scopeThis(scopep);
        } else {
            // Reference to something elsewhere, or relative references are disabled. Use global
            // variable
-            return "(&" + scopep->nameVlSym() + ")";
+            return scopeVlSyms(scopep);
        }
    }

@@ -163,11 +185,10 @@ private:
                    if (moreOfSame) {
                        AstIf* const ifp = new AstIf{
                            funcp->fileline(),
-                            new AstEq{
-                                funcp->fileline(), new AstCExpr{funcp->fileline(), "this", 64},
-                                new AstCExpr{funcp->fileline(),
-                                             string{"&("} + funcp->scopep()->nameVlSym() + ")",
-                                             64}},
+                            new AstEq{funcp->fileline(),
+                                      new AstCExpr{funcp->fileline(), "this", 64},
+                                      new AstCExpr{funcp->fileline(),
+                                                   scopeVlSyms(funcp->scopep()).asString(), 64}},
                            returnp};
                        newfuncp->addStmtsp(ifp);
                    } else {
@@ -225,15 +246,15 @@ private:
        const AstScope* const scopep = nodep->varScopep()->scopep();
        if (varp->isFuncLocal()) {
            // Reference to function locals need no self pointer
-            nodep->selfPointer("");
+            nodep->selfPointer(VSelfPointerText{VSelfPointerText::Empty()});
        } else if (scopep->modp() == v3Global.rootp()->constPoolp()->modp()) {
            // Reference to constant pool value need no self pointer
-            nodep->selfPointer("");
+            nodep->selfPointer(VSelfPointerText{VSelfPointerText::Empty()});
        } else {
            nodep->selfPointer(descopedSelfPointer(scopep));
        }
        nodep->varScopep(nullptr);
-        UINFO(9, "  refout " << nodep << " selfPtr=" << nodep->selfPointer() << endl);
+        UINFO(9, "  refout " << nodep << " selfPtr=" << nodep->selfPointer().asString() << endl);
    }
    void visit(AstCCall* nodep) override {
        // UINFO(9, "       " << nodep << endl);
@@ -285,5 +306,5 @@ public:
 void V3Descope::descopeAll(AstNetlist* nodep) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    { DescopeVisitor{nodep}; }  // Destruct before checking
-    V3Global::dumpCheckGlobalTree("descope", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("descope", 0, dumpTreeLevel() >= 3);
 }
@@ -503,11 +503,11 @@ public:
    inline bool inlined() const;

    // Methods that allow DfgVertex to participate in error reporting/messaging
-    void v3errorEnd(std::ostringstream& str) const VL_REQUIRES(V3Error::s().m_mutex) {
+    void v3errorEnd(std::ostringstream& str) const VL_RELEASE(V3Error::s().m_mutex) {
        m_filelinep->v3errorEnd(str);
    }
    void v3errorEndFatal(std::ostringstream& str) const VL_ATTR_NORETURN
-        VL_REQUIRES(V3Error::s().m_mutex) {
+        VL_RELEASE(V3Error::s().m_mutex) {
        m_filelinep->v3errorEndFatal(str);
    }
    string warnContextPrimary() const VL_REQUIRES(V3Error::s().m_mutex) {
@@ -334,17 +334,20 @@ class AstToDfgVisitor final : public VNVisitor {
                    const uint32_t bEnd = b.m_lsb + bWidth;
                    const uint32_t overlapEnd = std::min(aEnd, bEnd) - 1;

-                    varp->varp()->v3warn(  //
-                        MULTIDRIVEN,
-                        "Bits ["  //
-                            << overlapEnd << ":" << b.m_lsb << "] of signal "
-                            << varp->varp()->prettyNameQ()
-                            << " have multiple combinational drivers\n"
-                            << a.m_fileline->warnOther() << "... Location of first driver\n"
-                            << a.m_fileline->warnContextPrimary() << '\n'
-                            << b.m_fileline->warnOther() << "... Location of other driver\n"
-                            << b.m_fileline->warnContextSecondary() << varp->varp()->warnOther()
-                            << "... Only the first driver will be respected");
+                    if (a.m_fileline->operatorCompare(*b.m_fileline) != 0) {
+                        varp->varp()->v3warn(  //
+                            MULTIDRIVEN,
+                            "Bits ["  //
+                                << overlapEnd << ":" << b.m_lsb << "] of signal "
+                                << varp->varp()->prettyNameQ()
+                                << " have multiple combinational drivers\n"
+                                << a.m_fileline->warnOther() << "... Location of first driver\n"
+                                << a.m_fileline->warnContextPrimary() << '\n'
+                                << b.m_fileline->warnOther() << "... Location of other driver\n"
+                                << b.m_fileline->warnContextSecondary()
+                                << varp->varp()->warnOther()
+                                << "... Only the first driver will be respected");
+                    }

                    // If the first driver completely covers the range of the second driver,
                    // we can just delete the second driver completely, otherwise adjust the
@@ -437,6 +440,7 @@ class AstToDfgVisitor final : public VNVisitor {
    void visit(AstCell* nodep) override { markReferenced(nodep); }
    void visit(AstNodeProcedure* nodep) override { markReferenced(nodep); }
    void visit(AstVar* nodep) override {
+        if (nodep->isSc()) return;
        // No need to (and in fact cannot) handle variables with unsupported dtypes
        if (!DfgVertex::isSupportedDType(nodep->dtypep())) return;
        // Mark ports as having external references
@@ -237,7 +237,7 @@ void V3DfgOptimizer::extract(AstNetlist* netlistp) {
    UINFO(2, __FUNCTION__ << ": " << endl);
    // Extract more optimization candidates
    DataflowExtractVisitor::apply(netlistp);
-    V3Global::dumpCheckGlobalTree("dfg-extract", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("dfg-extract", 0, dumpTreeLevel() >= 3);
 }

 void V3DfgOptimizer::optimize(AstNetlist* netlistp, const string& label) {
@@ -267,7 +267,7 @@ void V3DfgOptimizer::optimize(AstNetlist* netlistp, const string& label) {

        // Build the DFG of this module
        const std::unique_ptr<DfgGraph> dfg{V3DfgPasses::astToDfg(*modp, ctx)};
-        if (dumpDfg() >= 8) dfg->dumpDotFilePrefixed(ctx.prefix() + "whole-input");
+        if (dumpDfgLevel() >= 8) dfg->dumpDotFilePrefixed(ctx.prefix() + "whole-input");

        // Extract the cyclic sub-graphs. We do this because a lot of the optimizations assume a
        // DAG, and large, mostly acyclic graphs could not be optimized due to the presence of
@@ -284,7 +284,7 @@ void V3DfgOptimizer::optimize(AstNetlist* netlistp, const string& label) {

        // For each cyclic component
        for (auto& component : cyclicComponents) {
-            if (dumpDfg() >= 7) component->dumpDotFilePrefixed(ctx.prefix() + "source");
+            if (dumpDfgLevel() >= 7) component->dumpDotFilePrefixed(ctx.prefix() + "source");
            // TODO: Apply optimizations safe for cyclic graphs
            // Add back under the main DFG (we will convert everything back in one go)
            dfg->addGraph(*component);
@@ -292,7 +292,7 @@ void V3DfgOptimizer::optimize(AstNetlist* netlistp, const string& label) {

        // For each acyclic component
        for (auto& component : acyclicComponents) {
-            if (dumpDfg() >= 7) component->dumpDotFilePrefixed(ctx.prefix() + "source");
+            if (dumpDfgLevel() >= 7) component->dumpDotFilePrefixed(ctx.prefix() + "source");
            // Optimize the component
            V3DfgPasses::optimize(*component, ctx);
            // Add back under the main DFG (we will convert everything back in one go)
@@ -300,9 +300,9 @@ void V3DfgOptimizer::optimize(AstNetlist* netlistp, const string& label) {
        }

        // Convert back to Ast
-        if (dumpDfg() >= 8) dfg->dumpDotFilePrefixed(ctx.prefix() + "whole-optimized");
+        if (dumpDfgLevel() >= 8) dfg->dumpDotFilePrefixed(ctx.prefix() + "whole-optimized");
        AstModule* const resultModp = V3DfgPasses::dfgToAst(*dfg, ctx);
        UASSERT_OBJ(resultModp == modp, modp, "Should be the same module");
    }
-    V3Global::dumpCheckGlobalTree("dfg-optimize", 0, dumpTree() >= 3);
+    V3Global::dumpCheckGlobalTree("dfg-optimize", 0, dumpTreeLevel() >= 3);
 }
@@ -285,7 +285,7 @@ void V3DfgPasses::optimize(DfgGraph& dfg, V3DfgOptimizationContext& ctx) {

    const auto apply = [&](int dumpLevel, const string& name, std::function<void()> pass) {
        pass();
-        if (dumpDfg() >= dumpLevel) {
+        if (dumpDfgLevel() >= dumpLevel) {
            const string strippedName = VString::removeWhitespace(name);
            const string label
                = ctx.prefix() + "pass-" + cvtToStr(passNumber) + "-" + strippedName;
@@ -294,7 +294,7 @@ void V3DfgPasses::optimize(DfgGraph& dfg, V3DfgOptimizationContext& ctx) {
        ++passNumber;
    };

-    if (dumpDfg() >= 8) dfg.dumpDotAllVarConesPrefixed(ctx.prefix() + "input");
+    if (dumpDfgLevel() >= 8) dfg.dumpDotAllVarConesPrefixed(ctx.prefix() + "input");
    apply(3, "input           ", [&]() {});
    apply(4, "inlineVars      ", [&]() { inlineVars(dfg); });
    apply(4, "cse0            ", [&]() { cse(dfg, ctx.m_cseContext0); });
@@ -304,5 +304,5 @@ void V3DfgPasses::optimize(DfgGraph& dfg, V3DfgOptimizationContext& ctx) {
        apply(4, "cse1            ", [&]() { cse(dfg, ctx.m_cseContext1); });
    }
    apply(4, "removeVars      ", [&]() { removeVars(dfg, ctx.m_removeVarsContext); });
-    if (dumpDfg() >= 8) dfg.dumpDotAllVarConesPrefixed(ctx.prefix() + "optimized");
+    if (dumpDfgLevel() >= 8) dfg.dumpDotAllVarConesPrefixed(ctx.prefix() + "optimized");
 }
@@ -123,9 +123,9 @@ struct V3DfgPeepholeContext final {
    const std::string m_label;  // Label to apply to stats

    // Enable flags for each optimization
-    bool m_enabled[VDfgPeepholePattern::_ENUM_END];
+    std::array<bool, VDfgPeepholePattern::_ENUM_END> m_enabled;
    // Count of applications for each optimization (for statistics)
-    VDouble0 m_count[VDfgPeepholePattern::_ENUM_END];
+    std::array<VDouble0, VDfgPeepholePattern::_ENUM_END> m_count;

    explicit V3DfgPeepholeContext(const std::string& label);
    ~V3DfgPeepholeContext();
@@ -100,5 +100,5 @@ void V3DupFinder::dumpFile(const string& filename, bool tree) {
 }

 void V3DupFinder::dumpFilePrefixed(const string& nameComment, bool tree) {
-    if (dump()) dumpFile(v3Global.debugFilename(nameComment) + ".hash", tree);
+    if (dumpLevel()) dumpFile(v3Global.debugFilename(nameComment) + ".hash", tree);
 }
@@ -79,6 +79,10 @@ string EmitCBaseVisitorConst::cFuncArgs(const AstCFunc* nodep) {
        args += prefixNameProtect(EmitCParentModule::get(nodep));
        args += "* vlSelf";
    }
+    if (nodep->needProcess()) {
+        if (!args.empty()) args += ", ";
+        args += "VlProcessRef vlProcess";
+    }
    if (!nodep->argTypes().empty()) {
        if (!args.empty()) args += ", ";
        args += nodep->argTypes();
@@ -224,21 +228,12 @@ void EmitCBaseVisitorConst::emitVarDecl(const AstVar* nodep, bool asRef) {
 }

 void EmitCBaseVisitorConst::emitModCUse(const AstNodeModule* modp, VUseType useType) {
-    string nl;
-    for (AstNode* itemp = modp->stmtsp(); itemp; itemp = itemp->nextp()) {
-        if (AstCUse* const usep = VN_CAST(itemp, CUse)) {
-            if (usep->useType() == useType) {
-                if (usep->useType().isInclude()) {
-                    puts("#include \"" + prefixNameProtect(usep) + ".h\"\n");
-                }
-                if (usep->useType().isFwdClass()) {
-                    puts("class " + prefixNameProtect(usep) + ";\n");
-                }
-                nl = "\n";
-            }
-        }
-    }
-    puts(nl);
+    bool nl = false;
+    forModCUse(modp, useType, [&](string entry) {
+        puts(entry);
+        nl = true;
+    });
+    if (nl) puts("\n");
 }

 void EmitCBaseVisitorConst::emitTextSection(const AstNodeModule* modp, VNType type) {
@@ -62,7 +62,7 @@ public:
    static string symClassAssign() {
        return symClassName() + "* const __restrict vlSymsp VL_ATTR_UNUSED = vlSelf->vlSymsp;\n";
    }
-    static string prefixNameProtect(const AstNode* nodep) VL_MT_STABLE {  // C++ name with prefix
+    static string prefixNameProtect(const AstNode* nodep) {  // C++ name with prefix
        return v3Global.opt.modPrefix() + "_" + VIdProtect::protect(nodep->name());
    }
    static bool isAnonOk(const AstVar* varp) {
@@ -113,6 +113,22 @@ public:
    void emitCFuncHeader(const AstCFunc* funcp, const AstNodeModule* modp, bool withScope);
    void emitCFuncDecl(const AstCFunc* funcp, const AstNodeModule* modp, bool cLinkage = false);
    void emitVarDecl(const AstVar* nodep, bool asRef = false);
+    template <typename F>
+    static void forModCUse(const AstNodeModule* modp, VUseType useType, F action) {
+        for (AstNode* itemp = modp->stmtsp(); itemp; itemp = itemp->nextp()) {
+            if (AstCUse* const usep = VN_CAST(itemp, CUse)) {
+                if (usep->useType().containsAny(useType)) {
+                    if (usep->useType().containsAny(VUseType::INT_INCLUDE)) {
+                        action("#include \"" + prefixNameProtect(usep) + ".h\"\n");
+                        continue;  // Forward declaration is not necessary
+                    }
+                    if (usep->useType().containsAny(VUseType::INT_FWD_CLASS)) {
+                        action("class " + prefixNameProtect(usep) + ";\n");
+                    }
+                }
+            }
+        }
+    }
    void emitModCUse(const AstNodeModule* modp, VUseType useType);
    void emitTextSection(const AstNodeModule* modp, VNType type);

@@ -108,7 +108,7 @@ void EmitCFunc::emitOpName(AstNode* nodep, const string& format, AstNode* lhsp,
                    UASSERT_OBJ(m_wideTempRefp, nodep,
                                "Wide Op w/ no temp, perhaps missing op in V3EmitC?");
                    COMMA;
-                    if (!m_wideTempRefp->selfPointer().empty()) {
+                    if (!m_wideTempRefp->selfPointer().isEmpty()) {
                        emitDereference(m_wideTempRefp->selfPointerProtect(m_useSelfForThis));
                    }
                    puts(m_wideTempRefp->varp()->nameProtect());
@@ -314,6 +314,7 @@ void EmitCFunc::displayNode(AstNode* nodep, AstScopeName* scopenamep, const stri

    // Convert Verilog display to C printf formats
    //          "%0t" becomes "%d"
+    VL_RESTORER(m_emitDispState);
    m_emitDispState.clear();
    string vfmt;
    string::const_iterator pos = vformat.begin();
@@ -417,6 +418,15 @@ void EmitCFunc::emitCCallArgs(const AstNodeCCall* nodep, const string& selfPoint
        puts(selfPointer);
        comma = true;
    }
+    if (nodep->funcp()->needProcess()) {
+        if (comma) puts(", ");
+        if (VN_IS(nodep->backp(), CAwait) || !nodep->funcp()->isCoroutine()) {
+            puts("vlProcess");
+        } else {
+            puts("std::make_shared<VlProcess>()");
+        }
+        comma = true;
+    }
    if (!nodep->argTypes().empty()) {
        if (comma) puts(", ");
        puts(nodep->argTypes());
@@ -443,6 +453,10 @@ void EmitCFunc::emitCvtPackStr(AstNode* nodep) {
        putbs("std::string{");
        putsQuoted(constp->num().toString());
        puts("}");
+    } else if (VN_IS(nodep->dtypep(), StreamDType)) {
+        putbs("VL_CVT_PACK_STR_ND(");
+        iterateAndNextConstNull(nodep);
+        puts(")");
    } else {
        putbs("VL_CVT_PACK_STR_N");
        emitIQW(nodep);
@@ -503,7 +517,7 @@ void EmitCFunc::emitConstant(AstConst* nodep, AstVarRef* assigntop, const string
            if (!assigntop) {
                puts(assignString);
            } else {
-                if (!assigntop->selfPointer().empty()) {
+                if (!assigntop->selfPointer().isEmpty()) {
                    emitDereference(assigntop->selfPointerProtect(m_useSelfForThis));
                }
                puts(assigntop->varp()->nameProtect());
@@ -525,7 +539,7 @@ void EmitCFunc::emitConstant(AstConst* nodep, AstVarRef* assigntop, const string
            if (!assigntop) {
                puts(assignString);
            } else {
-                if (!assigntop->selfPointer().empty()) {
+                if (!assigntop->selfPointer().isEmpty()) {
                    emitDereference(assigntop->selfPointerProtect(m_useSelfForThis));
                }
                puts(assigntop->varp()->nameProtect());
@@ -695,6 +709,8 @@ string EmitCFunc::emitVarResetRecurse(const AstVar* varp, const string& varNameP
        return "";
    } else if (basicp && basicp->isForkSync()) {
        return "";
+    } else if (basicp && basicp->isProcessRef()) {
+        return "";
    } else if (basicp && basicp->isDelayScheduler()) {
        return "";
    } else if (basicp && basicp->isTriggerScheduler()) {
@@ -22,6 +22,7 @@

 #include "V3EmitCConstInit.h"
 #include "V3Global.h"
+#include "V3MemberMap.h"

 #include <algorithm>
 #include <map>
@@ -115,6 +116,7 @@ public:

 class EmitCFunc VL_NOT_FINAL : public EmitCConstInit {
 private:
+    VMemberMap m_memberMap;
    AstVarRef* m_wideTempRefp = nullptr;  // Variable that _WW macros should be setting
    int m_labelNum = 0;  // Next label number
    int m_splitSize = 0;  // # of cfunc nodes placed into output file
@@ -148,6 +150,22 @@ protected:
    bool m_useSelfForThis = false;  // Replace "this" with "vlSelf"
    const AstNodeModule* m_modp = nullptr;  // Current module being emitted
    const AstCFunc* m_cfuncp = nullptr;  // Current function being emitted
+    bool m_instantiatesOwnProcess = false;
+
+    bool constructorNeedsProcess(const AstClass* const classp) {
+        const AstNode* const newp = m_memberMap.findMember(classp, "new");
+        if (!newp) return false;
+        const AstCFunc* const ctorp = VN_CAST(newp, CFunc);
+        if (!ctorp) return false;
+        UASSERT_OBJ(ctorp->isConstructor(), ctorp, "`new` is not a constructor!");
+        return ctorp->needProcess();
+    }
+
+    bool constructorNeedsProcess(const AstNodeDType* const dtypep) {
+        if (const AstClassRefDType* const crefdtypep = VN_CAST(dtypep, ClassRefDType))
+            return constructorNeedsProcess(crefdtypep->classp());
+        return false;
+    }

 public:
    // METHODS
@@ -174,12 +192,10 @@ public:
    }
    void emitScIQW(AstVar* nodep) {
        UASSERT_OBJ(nodep->isSc(), nodep, "emitting SystemC operator on non-SC variable");
-        // clang-format off
        puts(nodep->isScBigUint() ? "SB"
             : nodep->isScUint()  ? "SU"
             : nodep->isScBv()    ? "SW"
-             : (nodep->isScQuad() ? "SQ" : "SI"));
-        // clang-format on
+                                  : (nodep->isScQuad() ? "SQ" : "SI"));
    }
    void emitDatap(AstNode* nodep) {
        // When passing to a function with va_args the compiler doesn't
@@ -212,13 +228,34 @@ public:
            comma = true;
        }
    }
+    template <typename T>
+    string optionalProcArg(const T* const nodep) {
+        return (nodep && constructorNeedsProcess(nodep)) ? "vlProcess, " : "";
+    }
+    const AstCNew* getSuperNewCallRecursep(AstNode* const nodep) {
+        // Get the super.new call
+        if (!nodep) return nullptr;
+        if (const AstCNew* const cnewp = VN_CAST(nodep, CNew)) return cnewp;
+        if (const AstStmtExpr* const stmtp = VN_CAST(nodep, StmtExpr)) {
+            if (const AstCNew* const cnewp = VN_CAST(stmtp->exprp(), CNew)) return cnewp;
+        }
+        if (const AstJumpBlock* const blockp = VN_CAST(nodep, JumpBlock)) {
+            if (const AstCNew* const cnewp = getSuperNewCallRecursep(blockp->stmtsp())) {
+                return cnewp;
+            }
+        }
+        if (const AstCNew* const cnewp = getSuperNewCallRecursep(nodep->nextp())) return cnewp;
+        return nullptr;
+    }

    // VISITORS
    using EmitCConstInit::visit;
    void visit(AstCFunc* nodep) override {
        VL_RESTORER(m_useSelfForThis);
        VL_RESTORER(m_cfuncp);
+        VL_RESTORER(m_instantiatesOwnProcess)
        m_cfuncp = nodep;
+        m_instantiatesOwnProcess = false;

        splitSizeInc(nodep);

@@ -231,20 +268,16 @@ public:
        if (!nodep->baseCtors().empty()) {
            puts(": ");
            puts(nodep->baseCtors());
-            puts("(vlSymsp");
+            const AstClass* const classp = VN_CAST(nodep->scopep()->modp(), Class);
            // Find call to super.new to get the arguments
-            for (AstNode* stmtp = nodep->stmtsp(); stmtp; stmtp = stmtp->nextp()) {
-                AstNode* exprp;
-                if (VN_IS(stmtp, StmtExpr)) {
-                    exprp = VN_CAST(stmtp, StmtExpr)->exprp();
-                } else {
-                    exprp = stmtp;
-                }
-                if (AstCNew* const newRefp = VN_CAST(exprp, CNew)) {
-                    putCommaIterateNext(newRefp->argsp(), true);
-                    break;
-                }
+            if (classp && constructorNeedsProcess(classp)) {
+                puts("(vlProcess, vlSymsp");
+            } else {
+                puts("(vlSymsp");
            }
+            const AstCNew* const superNewCallp = getSuperNewCallRecursep(nodep->stmtsp());
+            UASSERT_OBJ(superNewCallp, nodep, "super.new call not found");
+            putCommaIterateNext(superNewCallp->argsp(), true);
            puts(")");
        }
        puts(" {\n");
@@ -265,6 +298,23 @@ public:
        puts(nodep->isLoose() ? "__" : "::");
        puts(nodep->nameProtect() + "\\n\"); );\n");

+        // Instantiate a process class if it's going to be needed somewhere later
+        nodep->forall([&](const AstNodeCCall* ccallp) -> bool {
+            if (ccallp->funcp()->needProcess()
+                && (ccallp->funcp()->isCoroutine() == VN_IS(ccallp->backp(), CAwait))) {
+                if (!nodep->needProcess() && !m_instantiatesOwnProcess) {
+                    m_instantiatesOwnProcess = true;
+                    return false;
+                }
+            }
+            return true;
+        });
+        if (m_instantiatesOwnProcess) {
+            AstNode* const vlprocp = new AstCStmt{
+                nodep->fileline(), "VlProcessRef vlProcess = std::make_shared<VlProcess>();\n"};
+            nodep->stmtsp()->addHereThisAsNext(vlprocp);
+        }
+
        for (AstNode* subnodep = nodep->argsp(); subnodep; subnodep = subnodep->nextp()) {
            if (AstVar* const varp = VN_CAST(subnodep, Var)) {
                if (varp->isFuncReturn()) emitVarDecl(varp);
@@ -295,6 +345,19 @@ public:
        emitVarDecl(nodep);
    }

+    void visit(AstCvtDynArrayToPacked* nodep) override {
+        puts("VL_DYN_TO_");
+        emitIQW(nodep);
+        puts("<");
+        const AstNodeDType* const elemDTypep = nodep->fromp()->dtypep()->subDTypep();
+        putbs(elemDTypep->cType("", false, false));
+        puts(">(");
+        iterateAndNextConstNull(nodep->fromp());
+        puts(", ");
+        puts(cvtToStr(elemDTypep->widthMin()));
+        puts(")");
+    }
+
    void visit(AstNodeAssign* nodep) override {
        bool paren = true;
        bool decind = false;
@@ -349,12 +412,28 @@ public:
            puts(cvtToStr(nodep->widthMin()) + ",");
            iterateAndNextConstNull(nodep->lhsp());
            puts(", ");
+        } else if (const AstCvtPackedToDynArray* const castp
+                   = VN_CAST(nodep->rhsp(), CvtPackedToDynArray)) {
+            puts("VL_ASSIGN_DYN_Q<");
+            putbs(castp->dtypep()->subDTypep()->cType("", false, false));
+            puts(">(");
+            iterateAndNextConstNull(nodep->lhsp());
+            puts(", ");
+            puts(cvtToStr(castp->dtypep()->subDTypep()->widthMin()));
+            puts(", ");
+            puts(cvtToStr(castp->fromp()->widthMin()));
+            puts(", ");
+            rhs = false;
+            iterateAndNextConstNull(castp->fromp());
        } else if (nodep->isWide() && VN_IS(nodep->lhsp(), VarRef)  //
                   && !VN_IS(nodep->rhsp(), CExpr)  //
                   && !VN_IS(nodep->rhsp(), CMethodHard)  //
                   && !VN_IS(nodep->rhsp(), VarRef)  //
                   && !VN_IS(nodep->rhsp(), AssocSel)  //
-                   && !VN_IS(nodep->rhsp(), ArraySel)) {
+                   && !VN_IS(nodep->rhsp(), MemberSel)  //
+                   && !VN_IS(nodep->rhsp(), StructSel)  //
+                   && !VN_IS(nodep->rhsp(), ArraySel)  //
+                   && !VN_IS(nodep->rhsp(), ExprStmt)) {
            // Wide functions assign into the array directly, don't need separate assign statement
            m_wideTempRefp = VN_AS(nodep->lhsp(), VarRef);
            paren = false;
@@ -381,19 +460,17 @@ public:
    void visit(AstAssocSel* nodep) override {
        iterateAndNextConstNull(nodep->fromp());
        putbs(".at(");
-        AstAssocArrayDType* const adtypep = VN_AS(nodep->fromp()->dtypep(), AssocArrayDType);
+        AstAssocArrayDType* const adtypep
+            = VN_AS(nodep->fromp()->dtypep()->skipRefp(), AssocArrayDType);
        UASSERT_OBJ(adtypep, nodep, "Associative select on non-associative type");
-        if (adtypep->keyDTypep()->isWide()) {
-            emitCvtWideArray(nodep->bitp(), nodep->fromp());
-        } else {
-            iterateAndNextConstNull(nodep->bitp());
-        }
+        iterateAndNextConstNull(nodep->bitp());
        puts(")");
    }
    void visit(AstWildcardSel* nodep) override {
        iterateAndNextConstNull(nodep->fromp());
        putbs(".at(");
-        AstWildcardArrayDType* const adtypep = VN_AS(nodep->fromp()->dtypep(), WildcardArrayDType);
+        AstWildcardArrayDType* const adtypep
+            = VN_AS(nodep->fromp()->dtypep()->skipRefp(), WildcardArrayDType);
        UASSERT_OBJ(adtypep, nodep, "Wildcard select on non-wildcard-associative type");
        iterateAndNextConstNull(nodep->bitp());
        puts(")");
@@ -417,7 +494,7 @@ public:
            puts(funcNameProtect(funcp));
        } else {
            // Calling regular method/function
-            if (!nodep->selfPointer().empty()) {
+            if (!nodep->selfPointer().isEmpty()) {
                emitDereference(nodep->selfPointerProtect(m_useSelfForThis));
            }
            puts(funcp->nameProtect());
@@ -441,8 +518,9 @@ public:
            // super.new case
            return;
        }
-        // assignment case
-        puts("VL_NEW(" + prefixNameProtect(nodep->dtypep()) + ", vlSymsp");
+        // assignment case;
+        puts("VL_NEW(" + prefixNameProtect(nodep->dtypep()) + ", "
+             + optionalProcArg(nodep->dtypep()) + "vlSymsp");
        putCommaIterateNext(nodep->argsp(), true);
        puts(")");
    }
@@ -467,7 +545,7 @@ public:
    void visit(AstLambdaArgRef* nodep) override { putbs(nodep->nameProtect()); }
    void visit(AstWith* nodep) override {
        // With uses a C++11 lambda
-        putbs("[=](");
+        putbs("[&](");
        if (auto* const argrefp = nodep->indexArgRefp()) {
            putbs(argrefp->dtypep()->cType(argrefp->nameProtect(), false, false));
            puts(",");
@@ -475,10 +553,9 @@ public:
        if (auto* const argrefp = nodep->valueArgRefp()) {
            putbs(argrefp->dtypep()->cType(argrefp->nameProtect(), false, false));
        }
-        // Probably fragile, V3Task may need to convert to a AstCReturn
-        puts(") { return ");
+        puts(") {\n");
        iterateAndNextConstNull(nodep->exprp());
-        puts("; }\n");
+        puts("}\n");
    }
    void visit(AstNodeCase* nodep) override {  // LCOV_EXCL_LINE
        // In V3Case...
@@ -792,7 +869,7 @@ public:
    void visit(AstSysFuncAsTask* nodep) override {
        if (!nodep->lhsp()->isWide()) puts("(void)");
        iterateAndNextConstNull(nodep->lhsp());
-        if (!nodep->lhsp()->isWide()) puts(";");
+        if (!nodep->lhsp()->isWide()) puts(";\n");
    }
    void visit(AstStackTraceF* nodep) override { puts("VL_STACKTRACE_N()"); }
    void visit(AstStackTraceT* nodep) override { puts("VL_STACKTRACE();\n"); }
@@ -831,6 +908,11 @@ public:
        iterateAndNextConstNull(nodep->endStmtsp());
        puts("}\n");
    }
+    void visit(AstCLocalScope* nodep) override {
+        puts("{\n");
+        iterateAndNextConstNull(nodep->stmtsp());
+        puts("}\n");
+    }
    void visit(AstJumpGo* nodep) override {
        puts("goto __Vlabel" + cvtToStr(nodep->labelp()->blockp()->labelNum()) + ";\n");
    }
@@ -1025,6 +1107,7 @@ public:
        UASSERT_OBJ(!emitSimpleOk(nodep), nodep, "Triop cannot be described in a simple way");
        emitOpName(nodep, nodep->emitC(), nodep->lhsp(), nodep->rhsp(), nodep->thsp());
    }
+    void visit(AstCvtPackString* nodep) override { emitCvtPackStr(nodep->lhsp()); }
    void visit(AstRedXor* nodep) override {
        if (nodep->lhsp()->isWide()) {
            visit(static_cast<AstNodeUniop*>(nodep));
@@ -1043,7 +1126,9 @@ public:
    }
    void visit(AstCCast* nodep) override {
        // Extending a value of the same word width is just a NOP.
-        if (nodep->size() <= VL_IDATASIZE) {
+        if (const AstClassRefDType* const classDtypep = VN_CAST(nodep->dtypep(), ClassRefDType)) {
+            puts("(" + classDtypep->cType("", false, false) + ")(");
+        } else if (nodep->size() <= VL_IDATASIZE) {
            puts("(IData)(");
        } else {
            puts("(QData)(");
@@ -1085,7 +1170,8 @@ public:
        puts(")");
    }
    void visit(AstNewCopy* nodep) override {
-        puts("VL_NEW(" + prefixNameProtect(nodep->dtypep()) + ", ");
+        puts("VL_NEW(" + prefixNameProtect(nodep->dtypep()) + ", "
+             + optionalProcArg(nodep->dtypep()));
        puts("*");  // i.e. make into a reference
        iterateAndNextConstNull(nodep->rhsp());
        puts(")");
@@ -1179,7 +1265,7 @@ public:
        } else if (varp->isIfaceRef()) {
            puts(nodep->selfPointerProtect(m_useSelfForThis));
            return;
-        } else if (!nodep->selfPointer().empty()) {
+        } else if (!nodep->selfPointer().isEmpty()) {
            emitDereference(nodep->selfPointerProtect(m_useSelfForThis));
        }
        puts(nodep->varp()->nameProtect());
@@ -17,11 +17,15 @@
 #include "config_build.h"
 #include "verilatedos.h"

+#include "V3Ast.h"
 #include "V3EmitC.h"
 #include "V3EmitCConstInit.h"
 #include "V3Global.h"

 #include <algorithm>
+#include <cstdint>
+#include <set>
+#include <string>
 #include <vector>

 VL_DEFINE_DEBUG_FUNCTIONS;
@@ -219,7 +223,7 @@ class EmitCHeader final : public EmitCConstInit {
        emitted.insert(sdtypep);
        for (const AstMemberDType* itemp = sdtypep->membersp(); itemp;
             itemp = VN_AS(itemp->nextp(), MemberDType)) {
-            AstNodeUOrStructDType* subp = VN_CAST(itemp->skipRefp(), NodeUOrStructDType);
+            AstNodeUOrStructDType* const subp = itemp->getChildStructp();
            if (subp && !subp->packed()) {
                // Recurse if it belongs to the current module
                if (subp->classOrPackagep() == modp) {
@@ -235,6 +239,20 @@ class EmitCHeader final : public EmitCConstInit {
            puts(itemp->dtypep()->cType(itemp->nameProtect(), false, false));
            puts(";\n");
        }
+
+        puts("\nbool operator==(const " + EmitCBase::prefixNameProtect(sdtypep)
+             + "& rhs) const {\n");
+        puts("return ");
+        for (const AstMemberDType* itemp = sdtypep->membersp(); itemp;
+             itemp = VN_AS(itemp->nextp(), MemberDType)) {
+            if (itemp != sdtypep->membersp()) puts("\n    && ");
+            puts(itemp->nameProtect() + " == " + "rhs." + itemp->nameProtect());
+        }
+        puts(";\n");
+        puts("}\n");
+        puts("bool operator!=(const " + EmitCBase::prefixNameProtect(sdtypep)
+             + "& rhs) const {\n");
+        puts("return !(*this == rhs);\n}\n");
        puts("};\n");
    }
    void emitStructs(const AstNodeModule* modp) {
@@ -287,11 +305,9 @@ class EmitCHeader final : public EmitCConstInit {
                     + ".h\"\n");
            }
        }
-        emitModCUse(modp, VUseType::INT_INCLUDE);

        // Forward declarations required by this AstNodeModule
        puts("\nclass " + symClassName() + ";\n");
-        emitModCUse(modp, VUseType::INT_FWD_CLASS);

        // From `systemc_header
        emitTextSection(modp, VNType::atScHdr);
@@ -300,6 +316,7 @@ class EmitCHeader final : public EmitCConstInit {

        // Open class body {{{
        puts("\nclass ");
+        if (!VN_IS(modp, Class)) puts("alignas(VL_CACHE_LINE_BYTES) ");
        puts(prefixNameProtect(modp));
        if (const AstClass* const classp = VN_CAST(modp, Class)) {
            puts(" : public ");
@@ -329,11 +346,7 @@ class EmitCHeader final : public EmitCConstInit {
        emitTextSection(modp, VNType::atScInt);

        // Close class body
-        if (!VN_IS(modp, Class)) {
-            puts("} VL_ATTR_ALIGNED(VL_CACHE_LINE_BYTES);\n");
-        } else {
-            puts("};\n");
-        }
+        puts("};\n");
        // }}}

        // Emit out of class function declarations
@@ -364,6 +377,18 @@ class EmitCHeader final : public EmitCConstInit {
        if (v3Global.opt.coverage()) puts("#include \"verilated_cov.h\"\n");
        if (v3Global.usesTiming()) puts("#include \"verilated_timing.h\"\n");

+        std::set<string> cuse_set;
+        auto add_to_cuse_set = [&](string s) { cuse_set.insert(s); };
+
+        forModCUse(modp, VUseType::INT_FWD_CLASS | VUseType::INT_INCLUDE, add_to_cuse_set);
+        if (const AstClassPackage* const packagep = VN_CAST(modp, ClassPackage)) {
+            forModCUse(packagep->classp(), VUseType::INT_INCLUDE | VUseType::INT_FWD_CLASS,
+                       add_to_cuse_set);
+        }
+
+        for (const string& s : cuse_set) puts(s);
+        puts("\n");
+
        emitAll(modp);

        if (const AstClassPackage* const packagep = VN_CAST(modp, ClassPackage)) {
@@ -22,6 +22,7 @@
 #include "V3EmitCFunc.h"
 #include "V3Global.h"
 #include "V3String.h"
+#include "V3ThreadPool.h"
 #include "V3UniqueNames.h"

 #include <map>
@@ -53,22 +54,23 @@ class EmitCGatherDependencies final : VNVisitorConst {
        } else if (const AstNodeUOrStructDType* const dtypep
                   = VN_CAST(nodep, NodeUOrStructDType)) {
            if (!dtypep->packed()) {
+                UASSERT_OBJ(dtypep->classOrPackagep(), nodep, "Unlinked struct package");
                m_dependencies.insert(EmitCBase::prefixNameProtect(dtypep->classOrPackagep()));
            }
        }
    }
-    void addSelfDependency(const string& selfPointer, AstNode* nodep) {
-        if (selfPointer.empty()) {
+    void addSelfDependency(VSelfPointerText selfPointer, AstNode* nodep) {
+        if (selfPointer.isEmpty()) {
            // No self pointer (e.g.: function locals, const pool values, loose static methods),
            // so no dependency
-        } else if (VString::startsWith(selfPointer, "this")) {
+        } else if (selfPointer.hasThis()) {
            // Dereferencing 'this', we need the definition of this module, which is also the
            // module that contains the variable.
            addModDependency(EmitCParentModule::get(nodep));
        } else {
            // Must be an absolute reference
-            UASSERT_OBJ(selfPointer.find("vlSymsp") != string::npos, nodep,
-                        "Unknown self pointer: '" << selfPointer << "'");
+            UASSERT_OBJ(selfPointer.isVlSym(), nodep,
+                        "Unknown self pointer: '" << selfPointer.asString() << "'");
            // Dereferencing vlSymsp, so we need it's definition...
            addSymsDependency();
        }
@@ -675,7 +677,7 @@ class EmitCTrace final : EmitCFunc {
            string fstvt;
            // Doubles have special decoding properties, so must indicate if a double
            if (nodep->dtypep()->basicp()->isDouble()) {
-                if (vartype == VVarType::GPARAM || vartype == VVarType::LPARAM) {
+                if (vartype.isParam()) {
                    fstvt = "FST_VT_VCD_REAL_PARAMETER";
                } else {
                    fstvt = "FST_VT_VCD_REAL";
@@ -908,7 +910,7 @@ class EmitCTrace final : EmitCFunc {
    ~EmitCTrace() override = default;

 public:
-    static void main(AstNodeModule* modp, bool slow, std::deque<AstCFile*>& cfilesr) {
+    static void main(AstNodeModule* modp, bool slow, std::deque<AstCFile*>& cfilesr) VL_MT_STABLE {
        EmitCTrace{modp, slow, cfilesr};
    }
 };
@@ -921,24 +923,37 @@ void V3EmitC::emitcImp() {
    // Make parent module pointers available.
    const EmitCParentModule emitCParentModule;
    std::list<std::deque<AstCFile*>> cfiles;
+    std::list<std::future<void>> futures;

    // Process each module in turn
    for (const AstNode* nodep = v3Global.rootp()->modulesp(); nodep; nodep = nodep->nextp()) {
        if (VN_IS(nodep, Class)) continue;  // Imped with ClassPackage
        const AstNodeModule* const modp = VN_AS(nodep, NodeModule);
        cfiles.emplace_back();
-        EmitCImp::main(modp, /* slow: */ true, cfiles.back());
+        auto& slowCfilesr = cfiles.back();
+        futures.push_back(V3ThreadPool::s().enqueue(
+            [modp, &slowCfilesr]() { EmitCImp::main(modp, /* slow: */ true, slowCfilesr); }));
        cfiles.emplace_back();
-        EmitCImp::main(modp, /* slow: */ false, cfiles.back());
+        auto& fastCfilesr = cfiles.back();
+        futures.push_back(V3ThreadPool::s().enqueue(
+            [modp, &fastCfilesr]() { EmitCImp::main(modp, /* slow: */ false, fastCfilesr); }));
    }

    // Emit trace routines (currently they can only exist in the top module)
    if (v3Global.opt.trace() && !v3Global.opt.lintOnly()) {
        cfiles.emplace_back();
-        EmitCTrace::main(v3Global.rootp()->topModulep(), /* slow: */ true, cfiles.back());
+        auto& slowCfilesr = cfiles.back();
+        futures.push_back(V3ThreadPool::s().enqueue([&slowCfilesr]() {
+            EmitCTrace::main(v3Global.rootp()->topModulep(), /* slow: */ true, slowCfilesr);
+        }));
        cfiles.emplace_back();
-        EmitCTrace::main(v3Global.rootp()->topModulep(), /* slow: */ false, cfiles.back());
+        auto& fastCfilesr = cfiles.back();
+        futures.push_back(V3ThreadPool::s().enqueue([&fastCfilesr]() {
+            EmitCTrace::main(v3Global.rootp()->topModulep(), /* slow: */ false, fastCfilesr);
+        }));
    }
+    // Wait for futures
+    V3ThreadPool::waitForFutures(futures);
    for (const auto& collr : cfiles) {
        for (const auto cfilep : collr) v3Global.rootp()->addFilesp(cfilep);
    }
@@ -51,6 +51,14 @@ private:
        // Not defining main_time/vl_time_stamp, so
        v3Global.opt.addCFlags("-DVL_TIME_CONTEXT");  // On MSVC++ anyways

+        // Optional main top name argument, with empty string replacement
+        string topArg;
+        string topName = v3Global.opt.mainTopName();
+        if (!topName.empty()) {
+            if (topName == "-") topName = "";
+            topArg = ", \"" + topName + "\"";
+        }
+
        // Heavily commented output, as users are likely to look at or copy this code
        ofp()->putsHeader();
        puts("// DESCRIPTION: main() calling loop, created with Verilator --main\n");
@@ -71,7 +79,7 @@ private:

        puts("// Construct the Verilated model, from Vtop.h generated from Verilating\n");
        puts("const std::unique_ptr<" + topClassName() + "> topp{new " + topClassName()
-             + "{contextp.get()}};\n");
+             + "{contextp.get()" + topArg + "}};\n");
        puts("\n");

        puts("// Simulate until $finish\n");
@@ -89,7 +89,7 @@ class EmitCModel final : public EmitCFunc {
        puts("\n");

        puts("// This class is the main interface to the Verilated model\n");
-        puts("class " + topClassName() + " VL_NOT_FINAL : ");
+        puts("class alignas(VL_CACHE_LINE_BYTES) " + topClassName() + " VL_NOT_FINAL : ");
        if (optSystemC()) {
            // SC_MODULE, but with multiple-inheritance of VerilatedModel
            puts("public ::sc_core::sc_module, ");
@@ -234,11 +234,17 @@ class EmitCModel final : public EmitCFunc {
        puts("const char* hierName() const override final;\n");
        puts("const char* modelName() const override final;\n");
        puts("unsigned threads() const override final;\n");
+        puts("/// Prepare for cloning the model at the process level (e.g. fork in Linux)\n");
+        puts("/// Release necessary resources. Called before cloning.\n");
+        puts("void prepareClone() const;\n");
+        puts("/// Re-init after cloning the model at the process level (e.g. fork in Linux)\n");
+        puts("/// Re-allocate necessary resources. Called after cloning.\n");
+        puts("void atClone() const;\n");
        if (v3Global.opt.trace()) {
            puts("std::unique_ptr<VerilatedTraceConfig> traceConfig() const override final;\n");
        }

-        puts("} VL_ATTR_ALIGNED(VL_CACHE_LINE_BYTES);\n");
+        puts("};\n");

        ofp()->putsEndGuard();

@@ -479,6 +485,15 @@ class EmitCModel final : public EmitCFunc {
             + "\"; }\n");
        puts("unsigned " + topClassName() + "::threads() const { return "
             + cvtToStr(std::max(1, v3Global.opt.threads())) + "; }\n");
+        puts("void " + topClassName()
+             + "::prepareClone() const { contextp()->prepareClone(); }\n");
+        puts("void " + topClassName() + "::atClone() const {\n");
+        if (v3Global.opt.threads() > 1) {
+            puts("vlSymsp->__Vm_threadPoolp = static_cast<VlThreadPool*>(");
+        }
+        puts("contextp()->threadPoolpOnClone()");
+        if (v3Global.opt.threads() > 1) puts(")");
+        puts(";\n}\n");

        if (v3Global.opt.trace()) {
            puts("std::unique_ptr<VerilatedTraceConfig> " + topClassName()
@@ -440,7 +440,8 @@ void EmitCSyms::emitSymHdr() {
    }

    puts("\n// SYMS CLASS (contains all model state)\n");
-    puts("class " + symClassName() + " final : public VerilatedSyms {\n");
+    puts("class alignas(VL_CACHE_LINE_BYTES)" + symClassName()
+         + " final : public VerilatedSyms {\n");
    ofp()->putsPrivate(false);  // public:

    puts("// INTERNAL STATE\n");
@@ -467,7 +468,7 @@ void EmitCSyms::emitSymHdr() {

    if (v3Global.opt.mtasks()) {
        puts("\n// MULTI-THREADING\n");
-        puts("VlThreadPool* const __Vm_threadPoolp;\n");
+        puts("VlThreadPool* __Vm_threadPoolp;\n");
        puts("bool __Vm_even_cycle__ico = false;\n");
        puts("bool __Vm_even_cycle__act = false;\n");
        puts("bool __Vm_even_cycle__nba = false;\n");
@@ -554,7 +555,7 @@ void EmitCSyms::emitSymHdr() {
        puts("void " + protect("__Vserialize") + "(VerilatedSerialize& os);\n");
        puts("void " + protect("__Vdeserialize") + "(VerilatedDeserialize& os);\n");
    }
-    puts("} VL_ATTR_ALIGNED(VL_CACHE_LINE_BYTES);\n");
+    puts("};\n");

    ofp()->putsEndGuard();
    VL_DO_CLEAR(delete m_ofp, m_ofp = nullptr);
@@ -424,6 +424,12 @@ class EmitVBaseVisitorConst VL_NOT_FINAL : public EmitCBaseVisitorConst {
        iterateAndNextConstNull(nodep->exprsp());
        puts(")");
    }
+    void visit(AstExprStmt* nodep) override {
+        putfs(nodep, "$_EXPRSTMT(\n");
+        iterateAndNextConstNull(nodep->stmtsp());
+        putbs(", ");
+        puts(");\n");
+    }

    void visit(AstCMethodHard* nodep) override {
        iterateConst(nodep->fromp());
@@ -434,6 +440,15 @@ class EmitVBaseVisitorConst VL_NOT_FINAL : public EmitCBaseVisitorConst {
        }
        puts(")");
    }
+    void visit(AstCMethodCall* nodep) override {
+        iterateConst(nodep->fromp());
+        puts("." + nodep->name() + "(");
+        for (AstNode* pinp = nodep->argsp(); pinp; pinp = pinp->nextp()) {
+            if (pinp != nodep->argsp()) puts(", ");
+            iterateConst(pinp);
+        }
+        puts(")");
+    }

    // Operators
    virtual void emitVerilogFormat(AstNode* nodep, const string& format, AstNode* lhsp = nullptr,
@@ -649,6 +664,12 @@ class EmitVBaseVisitorConst VL_NOT_FINAL : public EmitCBaseVisitorConst {
        iterateAndNextConstNull(nodep->pinsp());
        puts(")");
    }
+    void visit(AstCCall* nodep) override {
+        puts(nodep->funcp()->name());
+        puts("(");
+        iterateAndNextConstNull(nodep->argsp());
+        puts(")");
+    }
    void visit(AstArg* nodep) override { iterateAndNextConstNull(nodep->exprp()); }
    void visit(AstPrintTimeScale* nodep) override {
        puts(nodep->verilogKwd());
@@ -661,10 +682,11 @@ class EmitVBaseVisitorConst VL_NOT_FINAL : public EmitCBaseVisitorConst {
            putfs(nodep, nodep->varScopep()->prettyName());
        } else {
            if (nodep->varp()) {
-                if (nodep->selfPointer().empty()) {
+                if (nodep->selfPointer().isEmpty()) {
                    putfs(nodep, nodep->varp()->prettyName());
                } else {
-                    putfs(nodep, nodep->selfPointer() + "->");
+                    putfs(nodep, nodep->selfPointer().asString());
+                    putfs(nodep, "->");
                    puts(nodep->varp()->prettyName());
                }
            } else {
@@ -188,7 +188,7 @@ class EmitXmlFileVisitor final : public VNVisitorConst {
    void visit(AstVar* nodep) override {
        const VVarType typ = nodep->varType();
        const string kw = nodep->verilogKwd();
-        const string vt = nodep->dtypep()->name();
+        const string vt = nodep->dtypep() ? nodep->dtypep()->name() : "";
        outputTag(nodep, "");
        if (nodep->isIO()) {
            puts(" dir=");
@@ -225,7 +225,7 @@ class EmitXmlFileVisitor final : public VNVisitorConst {
    void visit(AstPin* nodep) override {
        // What we call a pin in verilator is a port in the IEEE spec.
        outputTag(nodep, "port");  // IEEE: vpiPort
-        if (nodep->modVarp()->isIO()) {
+        if (nodep->modVarp() && nodep->modVarp()->isIO()) {
            puts(" direction=\"" + nodep->modVarp()->direction().xmlKwd() + "\"");
        }
        puts(" portIndex=\"" + cvtToStr(nodep->pinNum()) + "\"");  // IEEE: vpiPortIndex
@@ -254,7 +254,7 @@ class EmitXmlFileVisitor final : public VNVisitorConst {
    void visit(AstNodeCCall* nodep) override {
        outputTag(nodep, "");
        puts(" func=");
-        putsQuoted(nodep->funcp()->name());
+        putsQuoted(nodep->funcp() ? nodep->funcp()->name() : nodep->name());
        outputChildrenEnd(nodep, "");
    }

@@ -401,7 +401,7 @@ private:
        }
    }
    void visit(AstCell* nodep) override {
-        if (nodep->modp()->dead()) return;
+        if (nodep->modp() && nodep->modp()->dead()) return;
        if (!m_hasChildren) m_os << ">\n";
        m_os << "<cell " << nodep->fileline()->xmlDetailedLocation() << " name=\"" << nodep->name()
             << "\""
@@ -220,22 +220,20 @@ void V3ErrorGuarded::v3errorEnd(std::ostringstream& sstr, const string& extra)
                    tellManual(2);
                }
 #ifndef V3ERROR_NO_GLOBAL_
-                if (dumpTree() || debug()) {
+                if (dumpTreeLevel() || debug()) {
                    V3Broken::allowMidvisitorCheck(true);
-                    V3ThreadPool::s().requestExclusiveAccess([&]() VL_REQUIRES(m_mutex) {
-                        if (dumpTree()) {
-                            v3Global.rootp()->dumpTreeFile(
-                                v3Global.debugFilename("final.tree", 990));
-                        }
-                        if (debug()) {
-                            execErrorExitCb();
-                            V3Stats::statsFinalAll(v3Global.rootp());
-                            V3Stats::statsReport();
-                        }
-                        // Abort in exclusive access to make sure other threads
-                        // don't change error code
-                        vlAbortOrExit();
-                    });
+                    const V3ThreadPool::ScopedExclusiveAccess exclusiveAccess;
+                    if (dumpTreeLevel()) {
+                        v3Global.rootp()->dumpTreeFile(v3Global.debugFilename("final.tree", 990));
+                    }
+                    if (debug()) {
+                        execErrorExitCb();
+                        V3Stats::statsFinalAll(v3Global.rootp());
+                        V3Stats::statsReport();
+                    }
+                    // Abort in exclusive access to make sure other threads
+                    // don't change error code
+                    vlAbortOrExit();
                }
 #endif
            }
@@ -293,3 +291,35 @@ void V3Error::vlAbort() {
    VL_GCOV_DUMP();
    std::abort();
 }
+void V3Error::v3errorAcquireLock() VL_ACQUIRE(s().m_mutex) {
+#ifndef V3ERROR_NO_GLOBAL_
+    V3Error::s().m_mutex.lockCheckStopRequest(
+        []() -> void { V3ThreadPool::s().waitIfStopRequested(); });
+#else
+    V3Error::s().m_mutex.lock();
+#endif
+}
+std::ostringstream& V3Error::v3errorPrep(V3ErrorCode code) VL_ACQUIRE(s().m_mutex) {
+    v3errorAcquireLock();
+    s().v3errorPrep(code);
+    return v3errorStr();
+}
+std::ostringstream& V3Error::v3errorPrepFileLine(V3ErrorCode code, const char* file, int line)
+    VL_ACQUIRE(s().m_mutex) {
+    v3errorPrep(code) << file << ":" << std::dec << line << ": ";
+    return v3errorStr();
+}
+std::ostringstream& V3Error::v3errorStr() VL_REQUIRES(s().m_mutex) { return s().v3errorStr(); }
+void V3Error::v3errorEnd(std::ostringstream& sstr, const string& extra) VL_RELEASE(s().m_mutex) {
+    s().v3errorEnd(sstr, extra);
+    V3Error::s().m_mutex.unlock();
+}
+
+void v3errorEnd(std::ostringstream& sstr) VL_RELEASE(V3Error::s().m_mutex) {
+    V3Error::v3errorEnd(sstr);
+}
+void v3errorEndFatal(std::ostringstream& sstr) VL_RELEASE(V3Error::s().m_mutex) {
+    V3Error::v3errorEnd(sstr);
+    assert(0);  // LCOV_EXCL_LINE
+    VL_UNREACHABLE;
+}
--- a/Show More
+++ b/Show More