Version bump

Copyright year update
Tests: Fix last commit on width (#4781 ).
2026-10-04 17:13:40 +02:00 · 2024-01-01 03:42:12 -05:00 · 2024-01-01 03:19:59 -05:00 · 2023-12-31 13:28:49 -05:00 · 2023-12-31 11:42:06 -05:00 · 2023-12-31 11:35:11 -05:00
1135 changed files with 66556 additions and 41218 deletions
@@ -63,6 +63,8 @@ ForEachMacros:
 SortIncludes:    true
 IncludeBlocks:   Regroup
 IncludeCategories:
+  - Regex:           '"V3Pch.*\.h"'
+    Priority:        -2 # Precompiled headers
  - Regex:           '"(config_build|verilated_config|verilatedos)\.h"'
    Priority:        -1 # Sepecials before main header
  - Regex:           '(<|")verilated.*'
@@ -112,6 +114,9 @@ SpacesBeforeTrailingComments: 2
 SpacesInAngles:  false
 SpacesInContainerLiterals: true
 SpacesInCStyleCastParentheses: false
+SpacesInLineCommentPrefix:
+  Minimum: 0
+  Maximum: -1
 SpacesInParentheses: false
 SpacesInSquareBrackets: false
 Standard:        Cpp11
@@ -0,0 +1,8 @@
+{
+	"name": "Verilator Build Environment",
+
+	"build": {
+        "dockerfile": "../ci/docker/buildenv/Dockerfile"
+    }
+
+}
@@ -11,6 +11,8 @@ Thanks for taking the time to report this.

 Can you attach an example that shows the issue?  (Must be openly licensed, ideally in test_regress format.)

+What 'verilator' command line do we use to run your example?
+
 What 'verilator --version' are you using?  Did you try it with the git master version?

 What OS and distribution are you using?
@@ -23,6 +23,10 @@ defaults:
    shell: bash
    working-directory: repo

+concurrency:
+  group: ${{ github.workflow }}-${{ github.event_name == 'pull_request' && github.ref || github.run_id }}
+  cancel-in-progress: true
+
 jobs:

  build:
@@ -33,7 +37,8 @@ jobs:
        compiler:
          - { cc: clang, cxx: clang++ }
          - { cc: gcc,   cxx: g++     }
-        m32: [0, 1]
+        # m32 1 is deprecated, not here to speed up regressions
+        m32: [0]
        exclude:
          # Build pull requests only with ubuntu-22.04 and without m32
          # - os:  ${{ github.event_name == 'pull_request' && 'ubuntu-18.04' || 'do-not-exclude' }}
@@ -58,7 +63,7 @@ jobs:
      CC: ${{ matrix.compiler.cc }}
      CXX: ${{ matrix.compiler.cxx }}
      CACHE_BASE_KEY: build-${{ matrix.os }}-${{ matrix.compiler.cc }}-m32=${{ matrix.m32 }}
-      CCACHE_MAXSIZE: 250M # Per build matrix entry (2000M in total)
+      CCACHE_MAXSIZE: 1000M # Per build matrix entry (* 5 = 5000M in total)
      VERILATOR_ARCHIVE: verilator-${{ github.sha }}-${{ matrix.os }}-${{ matrix.compiler.cc }}${{ matrix.m32 && '-m32' || '' }}.tar.gz
    steps:

@@ -103,7 +108,7 @@ jobs:
        compiler:
          - { cc: clang, cxx: clang++ }
          - { cc: gcc,   cxx: g++     }
-        m32: [0, 1]
+        m32: [0]
        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
@@ -131,7 +136,7 @@ jobs:
      CC: ${{ matrix.compiler.cc }}
      CXX: ${{ matrix.compiler.cxx }}
      CACHE_BASE_KEY: test-${{ matrix.os }}-${{ matrix.compiler.cc }}-m32=${{ matrix.m32 }}-${{ matrix.suite }}
-      CCACHE_MAXSIZE: 64M # Per build matrix entry (2160M in total)
+      CCACHE_MAXSIZE: 100M # Per build per suite (* 5 * 5 = 2500M in total)
      VERILATOR_ARCHIVE: verilator-${{ github.sha }}-${{ matrix.os }}-${{ matrix.compiler.cc }}${{ matrix.m32 && '-m32' || '' }}.tar.gz
    steps:

@@ -162,3 +167,30 @@ jobs:
      env:
        TESTS: ${{ matrix.suite }}
      run: ./ci/ci-script.bash
+
+  lint-py:
+    runs-on: ubuntu-22.04
+    name: Lint Python
+    env:
+      CI_BUILD_STAGE_NAME: build
+      CI_RUNS_ON: ubuntu-22.04
+      CI_M32: 0
+    steps:
+    - name: Checkout
+      uses: actions/checkout@v3
+      with:
+        path: repo
+
+    - name: Install packages for build
+      run: ./ci/ci-install.bash
+
+    # We use specific version numbers, otherwise a Python package
+    # update may add a warning and break our build
+    - name: Install packages for lint
+      run: sudo pip3 install pylint==3.0.2 ruff==0.1.3 clang sphinx sphinx_rtd_theme sphinxcontrib-spelling breathe ruff
+
+    - name: Configure
+      run: autoconf && ./configure --enable-longtests --enable-ccwarn
+
+    - name: Lint
+      run: make -k lint-py
@@ -22,6 +22,10 @@ defaults:
  run:
    working-directory: repo

+concurrency:
+  group: ${{ github.workflow }}-${{ github.event_name == 'pull_request' && github.ref || github.run_id }}
+  cancel-in-progress: true
+
 jobs:

   windows:
@@ -43,3 +43,4 @@ verilator-config-version.cmake
 /.vscode/
 /.idea/
 /cmake-build-*/
+/test_regress/snapshot/
@@ -4,7 +4,7 @@
 #
 #*****************************************************************************
 #
-# Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+# Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 # can redistribute it and/or modify it under the terms of either the GNU
 # Lesser General Public License Version 3 or the Perl Artistic License
 # Version 2.0.
@@ -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 5.016
+    VERSION 5.020
    HOMEPAGE_URL https://verilator.org
    LANGUAGES CXX
 )
@@ -35,7 +35,7 @@ if (NOT WIN32)
    message(WARNING "CMake support on Linux/OSX is experimental.")
 endif()

-if (MSVC)
+if (WIN32)
    if (DEFINED ENV{WIN_FLEX_BISON})
        set(WIN_FLEX_BISON "$ENV{WIN_FLEX_BISON}")
    endif()
@@ -48,6 +48,27 @@ if (MSVC)
    set(CMAKE_CXX_STANDARD 20)
 endif()

+set(OBJCACHE "" CACHE STRING "Path for ccache, auto-detected if empty")
+option(OBJCACHE_ENABLED "Compile Verilator with ccache" ON)
+
+if (OBJCACHE_ENABLED)
+    if (OBJCACHE STREQUAL "")
+        find_program(OBJCACHE_PATH ccache)
+        if (OBJCACHE_PATH STREQUAL "OBJCACHE_PATH-NOTFOUND")
+            set(OBJCACHE_PATH "")
+        endif()
+    else()
+        set(OBJCACHE_PATH "${OBJCACHE}")
+    endif()
+    if (NOT OBJCACHE_PATH STREQUAL "")
+        execute_process(COMMAND "${OBJCACHE_PATH}" --version
+                OUTPUT_VARIABLE objcache_version)
+        string(REGEX MATCH "[^\n\r]+" objcache_version "${objcache_version}")
+        message(STATUS "Found ccache: ${OBJCACHE_PATH} (\"${objcache_version}\")")
+        set(CMAKE_CXX_COMPILER_LAUNCHER "${OBJCACHE_PATH}")
+    endif()
+endif()
+
 find_package(BISON)
 find_package(FLEX)

@@ -8,6 +8,147 @@ The changes in each Verilator version are described below.  The
 contributors that suggested a given feature are shown in []. Thanks!


+Verilator 5.020 2024-01-01
+==========================
+
+**Major:**
+
+* Support compilation with precompiled headers with Make and GCC or CLang.
+* Change include to systemc instead of systemc.h (#4622) (#4623). [Chih-Mao Chen]
+  This may require that SystemC programs add 'using namespace sc_core', 'using namespace sc_dt'.
+
+**Minor:**
+
+* Add devcontainer support (#4748). [Stefan Wallentowitz]
+* Support `iff` in sensitivity list (#1482) (#4626). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support parameterized virtual interfaces (#4047) (#4743). [Ryszard Rozak, Antmicro Ltd.]
+* Support --timing triggers for virtual interfaces (#4673). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support ccache when compiling Verilator with CMake (#4678). [Anthony Donlon]
+* Support passing constraints to --xml-only output (still otherwise unsupported) (#4683). [Shahid Ikram]
+* Support node memory usage information in --stats (#4684). [Geza Lore]
+* Support vpiConstType in vpi_get() (#4761). [Todd Strader]
+* Support vpi_iterate on packages with vpiInstance (#4726). [Todd Strader]
+* Support multiple parameters in virtual interfaces (#4745). [Ryszard Rozak, Antmicro Ltd.]
+* Support user C/C++ code in final archive, and make a lib{model}.a (#4749) (#4754). [Fan Shupei]
+* Support inside operator on unpacked arrays and queues (#4751). [Ryszard Rozak, Antmicro Ltd.]
+* Support VPI parameter iteration (#4765). [Todd Strader]
+* Support packages in vpi_handle_by_name() (#4768). [Todd Strader]
+* Support invoking interface methods on virtual interface variables (#4774) (#4775). [Jordan McConnon]
+* Remove deprecated options (#4663). [Geza Lore]
+* Remove older compiler support; require C++14 or newer (#4784) (#4786).
+* Optimize timing-delayed queue (#4584). [qrqiuren]
+* Optimize substitute optimization memory usage (#4687). [Geza Lore]
+* Optimize wide primitive operations with -Oz (#4733). [Geza Lore]
+* Optimize V3Premit performance etc. (#4736). [Geza Lore]
+* Fix VPI TOP level variable iteration (#3919) (#4618). [Marlon James]
+* Fix display with no % printing assoc array (#4376). [Alex Solomatnikov]
+* Fix scheduling of external force signals (#4577) (#4668). [Geza Lore]
+* Fix a memory leak in V3Fork (#4628). [Krzysztof Boroński]
+* Fix linking parameterized hierarchical blocks and recursive hierarchical blocks (#4654). [Anthony Donlon]
+* Fix identifiers that end with '_' on Windows (#4655). [Anthony Donlon]
+* Fix 'for' loop with outside variable reference (#4660). [David Harris]
+* Fix tracing FST enums (#4661) (#4756). [Todd Strader]
+* Fix interface parameters used in loop generate constructs (#4664) (#4665). [Anthony Donlon]
+* Fix C++20 compilation errors (#4670).
+* Fix deadlocks in error handler (#4672). [Mariusz Glebocki, Antmicro Ltd.]
+* Fix MingW compilation (#4675). [David Ledger]
+* Fix trace when using SystemC with certain configurations (#4676). [Anthony Donlon]
+* Fix range access to classes depending on parameter resolution (#4681). [Krzysztof Boroński]
+* Fix select into constant And/Or/Xor pattern (#4689). [Geza Lore]
+* Fix access type of function arguments (#4692) (#4694). [Ryszard Rozak, Antmicro Ltd.]
+* Fix dynamic NBAs with automatic vars (#4696). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix # 0 delays for process resumption, etc. (#4697). [Krzysztof Boroński]
+* Fix conflicted namespace for coroutines (#4701) (#4707). [Jinyan Xu]
+* Fix compilers seeing empty input due to file system races (#4708). [Flavien Solt]
+* Fix shift of > 32-bit number (#4719). [Flavien Solt]
+* Fix Windows include gates in filesystem Flush implementation. (#4720). [William D. Jones]
+* Fix power operator with wide numbers and constants (#4721) (#4763). [Flavien Solt]
+* Fix parameter passing to ports (#4723). [Ryszard Rozak, Antmicro Ltd.]
+* Fix block names of nested do..while loops (#4728). [Ryszard Rozak, Antmicro Ltd.]
+* Fix class name in error on 'new' on virtual class (#4739). [Ryszard Rozak, Antmicro Ltd.]
+* Fix typedefs pointing to parameterized classes (#4747). [Ryszard Rozak, Antmicro Ltd.]
+* Fix $finish twice to no longer exit (#4757). [Tim Hutt]
+* Fix dynamic NBA conditions (#4773). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix `V3Fork` stage to run only if `--timing` is set (#4778). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix max multiply width and add runtime assertions if too small. (#4781)
+
+
+Verilator 5.018 2023-10-30
+==========================
+
+**Major:**
+
+* Support compilation with precompiled headers with Make and GCC or CLang.
+* Change include of systemc instead of systemc.h (#4622) (#4623). [Chih-Mao Chen]
+  This may require that SystemC programs add 'using namespace sc_core', 'using namespace sc_dt'.
+
+**Minor:**
+
+* Add SIDEEFFECT warning on mishandled side effect cases.
+* Add trace() API even when Verilated without --trace (#4462). [phelter]
+* Add warning on interface instantiation without parens (#4094). [Gökçe Aydos]
+* Add sv_vpi_user.h from IEEE 1800-2017 Annex M (#4606). [Marlon James]
+* Support 'disable fork' (#4125) (#4569). [Aleksander Kiryk, Antmicro Ltd.]
+* Support 'wait fork' (#4586). [Aleksander Kiryk, Antmicro Ltd.]
+* Support 'randc' (#4349).
+* Support assigning events (#4403). [Krzysztof Boroński]
+* Support resizing function call inout arguments (#4467).
+* Support NBAs in non-inlined functions/tasks (#4496) (#4572). [Krzysztof Bieganski, Antmicro Ltd.]
+* Support converting parameters inside modules to localparams (#4511). [Anthony Donlon]
+* Support concatenation of unpacked arrays (#4558). [Yutetsu TAKATSUKASA]
+* Support Clang 16 (#4592). [Mariusz Glebocki]
+* Support VPI variables of real and string data types (#4594). [Marlon James]
+* Support making VL_LOCK_SPINS configurable (#4599). [Geza Lore]
+* Change code --stats output (#4597). [Geza Lore]
+* Change --prof-exec infrastructure and report (#4602). [Geza Lore]
+* Change lint_off to not propagate upwards to files including where the lint_off is.
+* Optimize empty expression statements (#4544).
+* Optimize trace internals (#4610) (#4612). [Geza Lore]
+* Optimize internal performance issues (#4638). [Geza Lore]
+* Fix conversion of impure logical expressions to bit expressions (#487 partial) (#4437). [Ryszard Rozak, Antmicro Ltd.]
+* Fix enum functions in localparams (#3999). [Andrew Nolte]
+* Fix passing arguments by reference (#3385 partial) (#4489). [Ryszard Rozak, Antmicro Ltd.]
+* Fix multithreading handling to separate by code units that use/never use it (#4228). [Mariusz Glebocki, Antmicro Ltd.]
+* Fix usage of annotation options (#4486) (#4504). [Michal Czyz]
+* Fix detecting local vars in nested forks (#4493) (#4506). [Kamil Rakoczy]
+* Fix handling input file path separator (#4515) (#4516). [Anthony Donlon]
+* Fix mis-support for parameterized UDPs (#4518). [Anthony Donlon]
+* Fix constant conversion of $realtobits, $bitstoreal (#4522).  [Andrew Nolte]
+* Fix conversion of integers in $display '%e' (#4528). [muzafferkal]
+* Fix non-inlined interface tracing (#3984) (#4530). [Todd Strader]
+* Fix stream operations with operands of struct type (#4531) (#4532). [Ryszard Rozak, Antmicro Ltd.]
+* Fix 'this' in a constructor (#4533). [Ryszard Rozak, Antmicro Ltd.]
+* Fix stream shift operator of 32 bits (#4536). [Julien Faucher]
+* Fix object destruction after a copy constructor (#4540) (#4541). [Ryszard Rozak, Antmicro Ltd.]
+* Fix inlining of real functions miscasting (#4543). [Andrew Nolte]
+* Fix broken link error for enum references (#4551). [Anthony Donlon]
+* Fix logical expressions with class objects - caching in v3Const (#4552). [Ryszard Rozak, Antmicro Ltd.]
+* Fix using functions/tasks following class definition inside module (#4553). [Anthony Donlon]
+* Fix large constant buffer overflow (#4556). [Varun Koyyalagunta]
+* Fix instance arrays connecting to array of structs (#4557). [raphmaster]
+* Fix error message for invalid parameter overrides (#4559). [Anthony Donlon]
+* Fix shift to remove operation side effects (#4563).
+* Fix compile warning on unused member function variable (#4567).
+* Fix method narrowing conversion compiler error (#4568).
+* Fix interface comparison (#4570). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix dynamic triggers for named events (#4571). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix dictionaries with keys of class types (#4576). [Ryszard Rozak, Antmicro Ltd.]
+* Fix to not remap local assign intervals in forks (#4583). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix display optimization ignoring side effects (#4585).
+* Fix PLI/DPI user defined system task/function grammar (#4587) (#4588). [Quentin Corradi]
+* Fix fault on empty clocking block (#4593). [Alex Mykyta]
+* Fix creating implicit nets for inputs of gate primitives (#4603). [Geza Lore]
+* Fix try_put method of unbounded mailbox (#4608). [Ryszard Rozak, Antmicro Ltd.]
+* Fix stable name generation in V3Fork (#4615) (#4624). [Krzysztof Boroński]
+* Fix virtual methods (#4616). [Ryszard Rozak, Antmicro Ltd.]
+* Fix insertion at queue end (#4619). [Krzysztof Boroński]
+* Fix rand fields of reference types (#4627). [Ryszard Rozak, Antmicro Ltd.]
+* Fix dynamic casts of null values (#4631). [Ryszard Rozak, Antmicro Ltd.]
+* Fix signals read via virtual interfaces being misoptimized (#4645). [Krzysztof Bieganski, Antmicro Ltd.]
+* Fix handling of static keyword in methods (#4649). [Ryszard Rozak, Antmicro Ltd.]
+* Fix preprocessor to show `line 2 on resumed file.
+
+
 Verilator 5.016 2023-09-16
 ==========================

@@ -434,7 +575,7 @@ Verilator 5.002 2022-10-29
 * Fix null access on optimized-out fork statements (#3658). [Krzysztof Bieganski, Antmicro Ltd]
 * Fix VPI inline module naming mismatch (#3690) (#3694). [Jiuyang Liu]
 * Fix deadlock in timeprecision when using SystemC (#3707). [Kamil Rakoczy, Antmicro Ltd]
-* Fix width mismatch on inside operator (#3714). [Alex Torregrosa]
+* Fix width mismatch on inside operator (#3714). [Àlex Torregrosa]


 Verilator 4.228 2022-10-01
@@ -664,7 +805,7 @@ Verilator 4.214 2021-10-17
 * Fix removing if statement with side effect in condition (#3131). [Alexander Grobman]
 * Fix --waiver-output for multiline warnings (#2429) (#3141). [Keith Colbert]
 * Fix internal error on bad widths (#3140) (#3145). [Zhanglei Wang]
-* Fix crash on clang 12/13 (#3148). [Kouping Hsu]
+* Fix crash on clang 12/13 (#3148). [Kuoping Hsu]
 * Fix cygwin compile error due to missing -std=gnu++14 (#3149). [Sun Kim]
 * Fix $urandom_range when the range is 0 ... UINT_MAX (#3161). [Iru Cai]
 * Fix constructor-parameter argument comma-separation in C++ (#3162). [Matthew Ballance]
@@ -779,12 +920,12 @@ Verilator 4.202 2021-04-24
 * Mark --no-relative-cfuncs as scheduled for deprecation.
 * Add --coverage-max-width (#2853). [xuejiazidi]
 * Add VerilatedCovContext::forcePerInstance (#2793). [Kevin Laeufer]
-* Add FST SystemC tracing (#2806). [Alex Torregrosa]
+* Add FST SystemC tracing (#2806). [Àlex Torregrosa]
 * Add PINNOTFOUND warning in place of error (#2868). [Udi Finkelstein]
 * Support overlaps in priority case statements (#2864). [Rupert Swarbrick]
 * Support for null ports (#2875). [Udi Finkelstein]
 * Fix class unpacked-array compile error (#2774). [Iru Cai]
-* Fix scope types in FST and VCD traces (#2805). [Alex Torregrosa]
+* Fix scope types in FST and VCD traces (#2805). [Àlex Torregrosa]
 * Fix exceeding command-line ar limit (#2834). [Yinan Xu]
 * Fix false $dumpfile warning on model save (#2834). [Yinan Xu]
 * Fix --timescale-override not suppressing TIMESCALEMOD (#2838). [Kaleb Barrett]
@@ -841,7 +982,7 @@ Verilator 4.110 2021-02-25
 **Minor:**

 * Support concat selection (#2721).
-* Support struct scopes when dumping structs to VCD (#2776) [Alex Torregrosa]
+* Support struct scopes when dumping structs to VCD (#2776) [Àlex Torregrosa]
 * Generate SELRANGE for potentially unreachable code (#2625) (#2754) [Pierre-Henri Horrein]
 * For --flatten, override inlining of public and no_inline modules (#2761) [James Hanlon]
 * Fix little endian interface pin swizzling (#2475). [Don Owen]
@@ -852,7 +993,7 @@ Verilator 4.110 2021-02-25
 * Fix class extends with VM_PARALLEL_BUILDS (#2775). [Iru Cai]
 * Fix shifts by > 32 bit values (#2785). [qrq992]
 * Fix examples not flushing vcd (#2787). [Richard E George]
-* Fix little endian packed array pattern assignment (#2795). [Alex Torregrosa]
+* Fix little endian packed array pattern assignment (#2795). [Àlex Torregrosa]


 Verilator 4.108 2021-01-10
@@ -1328,7 +1469,7 @@ Verilator 4.014 2019-05-08
 **Minor:**

 * Add --trace-fst-thread.
-* Support '#' comments in $readmem. (#1411) [Frederick Requin]
+* Support '#' comments in $readmem. (#1411) [Frédéric Requin]
 * Support "'dx" constants. (#1423) [Udi Finkelstein]
 * For FST tracing use LZ4 compression. [Tony Bybell]
 * Add error when use parameters without value. (#1424) [Peter Gerst]
@@ -1598,19 +1739,19 @@ Verilator 3.912 2017-09-23

 **Minor:**

-* Support or/and/xor array intrinsic methods. (#1210) [Mike Popoloski]
+* Support or/and/xor array intrinsic methods. (#1210) [Michael Popoloski]
 * Support package export. (#1217) [Usuario Eda]
-* Support module port parameters without defaults. (#1213) [Mike Popoloski]
+* Support module port parameters without defaults. (#1213) [Michael Popoloski]
 * Add performance information to --stats file.
 * Simplify VL_CONST_W macro generation for faster compiles.
 * Optimize improvements for Shift-And, and replication constructs.
-* Fix ordering of arrayed cell wide connections. (#1202 partial) [Mike Popoloski]
-* Fix LITENDIAN warning on arrayed cells. (#1202) [Mike Popoloski]
-* Fix enum ranges without colons. (#1204) [Mike Popoloski]
-* Fix GCC noreturn compile error. (#1209) [Mike Popoloski]
-* Fix constant function default parameters. (#1211) [Mike Popoloski]
-* Fix non-colon array of interface modports. (#1212) [Mike Popoloski]
-* Fix .name connections on interfaces. (#1214) [Mike Popoloski]
+* Fix ordering of arrayed cell wide connections. (#1202 partial) [Michael Popoloski]
+* Fix LITENDIAN warning on arrayed cells. (#1202) [Michael Popoloski]
+* Fix enum ranges without colons. (#1204) [Michael Popoloski]
+* Fix GCC noreturn compile error. (#1209) [Michael Popoloski]
+* Fix constant function default parameters. (#1211) [Michael Popoloski]
+* Fix non-colon array of interface modports. (#1212) [Michael Popoloski]
+* Fix .name connections on interfaces. (#1214) [Michael Popoloski]
 * Fix wide array indices causing compile error.


@@ -1732,8 +1873,8 @@ Verilator 3.888 2016-10-14

 **Minor:**

-* Add --no-decoration to remove output comments, msg2015. [Frederic Requin]
-* If VM_PARALLEL_BUILDS=1, use OPT_FAST and OPT_SLOW. [Frederic Requin]
+* Add --no-decoration to remove output comments, msg2015. [Frédéric Requin]
+* If VM_PARALLEL_BUILDS=1, use OPT_FAST and OPT_SLOW. [Frédéric Requin]
  Set VM_DEFAULT_RULES=0 for old behavior.
 * Add error on DPI functions > 32 bits. (#1898) [Elliot Mednick]
 * Improve Verilation performance on internal strings. (#1896) [Johan Bjork]
@@ -1769,7 +1910,7 @@ Verilator 3.884 2016-05-18
 * Fix --output-split of constructors. (#1035) [Johan Bjork]
 * Fix removal of empty packages, modules and cells. (#1034) [Johan Bjork]
 * Fix core dump on Arch Linux/GCC 6.1.1. (#1058) [Jannis Harder]
-* Fix $value$plusargs to string. (#1880) [Frederic Requin]
+* Fix $value$plusargs to string. (#1880) [Frédéric Requin]


 Verilator 3.882 2016-03-01
@@ -1848,13 +1989,13 @@ Verilator 3.876 2015-08-12

 **Minor:**

-* Add tracing_on, etc to vlt files. (#932) [Frederic Requin]
+* Add tracing_on, etc to vlt files. (#932) [Frédéric Requin]
 * Support extraction of enum bits. (#951) [Jonathon Donaldson]
 * Fix MinGW compiler error. (#927) (#929) [Hans Tichelaar]
 * Fix .c files to be treated as .cpp. (#930) [Jonathon Donaldson]
 * Fix string-to-int space conversion. (#931) [Fabrizio Ferrandi]
 * Fix dpi imports inside generates. [Michael Tresidder]
-* Fix rounding in trace $timescale. (#946) [Frederic Requin]
+* Fix rounding in trace $timescale. (#946) [Frédéric Requin]
 * Fix $fopen with SV string. (#947) [Sven Stucki]
 * Fix hashed error with typedef inside block. (#948) [Sven Stucki]
 * Fix makefile with --coverage. (#953) [Eivind Liland]
@@ -2094,7 +2235,7 @@ Verilator 3.852 2013-09-29

 * Support named function and task arguments. [Chris Randall]
 * Report SELRANGE warning for non-generate if. (#675) [Roland Kruse]
-* Fix ordering of $fgetc. (#1808) [Frederic Requin]
+* Fix ordering of $fgetc. (#1808) [Frédéric Requin]
 * Fix --output-split-cfunc to count internal functions. [Chris Randall]
 * Fix crash on 32-bit Ubuntu. (#670) [Mark Jackson Pulver]

@@ -2162,7 +2303,7 @@ Verilator 3.846 2013-03-09
 * Fix DETECTARRAY on packed structures. (#610) [Jeremy Bennett]
 * Fix LITENDIAN on unpacked structures. (#614) [Wai Sum Mong]
 * Fix 32-bit OS VPI scan issue. (#615)  [Jeremy Bennett, Rich Porter]
-* Fix opening a VerilatedVcdC file multiple times. (#1774) [Frederic Requin]
+* Fix opening a VerilatedVcdC file multiple times. (#1774) [Frédéric Requin]
 * Fix UNOPTFLAT circular array bounds crossing. (#630) [Jie Xu]


@@ -3474,7 +3615,7 @@ Verilator 3.460 2005-07-27 Beta
 * Fix false warning when a clock is constant.
 * Fix X/Z in decimal numbers. [Wim Michiels]
 * Fix genvar statements in non-named generate blocks.
-* Fix core dump when missing newline in `define. [David van der bokke]
+* Fix core dump when missing newline in `define. [David van der Bokke]


 Verilator 3.450 2005-07-12
@@ -4370,7 +4511,7 @@ Verilator 0.0 1994-07-08
 Copyright
 =========

-Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -7,7 +7,7 @@
 #
 #*****************************************************************************
 #
-# Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+# Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 # can redistribute it and/or modify it under the terms of either the GNU
 # Lesser General Public License Version 3 or the Perl Artistic License
 # Version 2.0.
@@ -173,6 +173,10 @@ smoke-test: all_nomsg
 test_regress: all_nomsg
 	$(MAKE) -C test_regress

+.PHONY: test-snap test-diff
+test-snap test-diff:
+	$(MAKE) -C test_regress $@
+
 examples: all_nomsg
 	for p in $(EXAMPLES) ; do \
 	  $(MAKE) -C $$p VERILATOR_ROOT=`pwd` || exit 10; \
@@ -430,9 +434,15 @@ PYLINT_FLAGS = --score=n --disable=R0801
 RUFF = ruff
 RUFF_FLAGS = check --ignore=E402,E501,E701

+# "make -k" so can see all tool result errors
 lint-py:
-	-$(PYLINT) $(PYLINT_FLAGS) $(PY_PROGRAMS)
-	-$(RUFF) $(RUFF_FLAGS) $(PY_PROGRAMS)
+	$(MAKE) -k lint-py-pylint lint-py-ruff
+
+lint-py-pylint:
+	$(PYLINT) $(PYLINT_FLAGS) $(PY_PROGRAMS)
+
+lint-py-ruff:
+	$(RUFF) $(RUFF_FLAGS) $(PY_PROGRAMS)

 format-pl-exec:
 	-chmod a+x test_regress/t/*.pl
@@ -142,7 +142,7 @@ Related Projects
 Open License
 ============

-Verilator is Copyright 2003-2023 by Wilson Snyder. (Report bugs to
+Verilator is Copyright 2003-2024 by Wilson Snyder. (Report bugs to
 `Verilator Issues <https://verilator.org/issues>`_.)

 Verilator is free software; you can redistribute it and/or modify it under
@@ -150,7 +150,7 @@ the terms of either the GNU Lesser General Public License Version 3 or the
 Perl Artistic License Version 2.0. See the documentation for more details.

 .. _CHIPS Alliance: https://chipsalliance.org
-.. _Icarus Verilog: http://iverilog.icarus.com
+.. _Icarus Verilog: https://steveicarus.github.io/iverilog
 .. _Linux Foundation: https://www.linuxfoundation.org
 .. |Logo| image:: https://www.veripool.org/img/verilator_256_200_min.png
 .. |verilator multithreaded performance| image:: https://www.veripool.org/img/verilator_multithreaded_performance_bg-min.png
@@ -1,7 +1,7 @@
 #!/usr/bin/env perl
 ######################################################################
 #
-# Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+# Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 # can redistribute it and/or modify it under the terms of either the GNU
 # Lesser General Public License Version 3 or the Perl Artistic License
 # Version 2.0.
@@ -501,7 +501,7 @@ description of these arguments.

 The latest version is available from L<https://verilator.org>.

-Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 redistribute it and/or modify the Verilator internals under the terms of
 either the GNU Lesser General Public License Version 3 or the Perl Artistic
 License Version 2.0.
@@ -17,7 +17,7 @@ parser = argparse.ArgumentParser(
    For documentation see
    https://verilator.org/guide/latest/exe_verilator_ccache_report.html""",
    epilog=
-    """Copyright 2002-2023 by Wilson Snyder. This program is free software; you
+    """Copyright 2002-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -1,7 +1,7 @@
 #!/usr/bin/env perl
 ######################################################################
 #
-# Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+# Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 # can redistribute it and/or modify it under the terms of either the GNU
 # Lesser General Public License Version 3 or the Perl Artistic License
 # Version 2.0.
@@ -188,7 +188,7 @@ L<https://verilator.org/guide/latest/exe_verilator_coverage.html>.

 The latest version is available from L<https://verilator.org>.

-Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify the Verilator internals under the terms
 of either the GNU Lesser General Public License Version 3 or the Perl
 Artistic License Version 2.0.
@@ -110,7 +110,7 @@ Verilator_difftree is used for debugging Verilator tree output files.
 It performs a diff between two files, or all files common between two
 directories, ignoring irrelevant pointer differences.""",
    epilog=
-    """Copyright 2005-2023 by Wilson Snyder. This program is free software; you
+    """Copyright 2005-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -1,33 +1,32 @@
 #!/usr/bin/env python3
-# pylint: disable=C0103,C0114,C0116,C0209,C0301,R0914,R0912,R0915,W0511,eval-used
+# pylint: disable=C0103,C0114,C0116,C0209,C0301,R0914,R0912,R0915,W0511,W0603,eval-used
 ######################################################################

 import argparse
+import bisect
 import collections
 import math
 import re
 import statistics
 # from pprint import pprint

-Threads = collections.defaultdict(lambda: collections.defaultdict(lambda: {}))
-Mtasks = collections.defaultdict(lambda: {})
-Evals = collections.defaultdict(lambda: {})
-EvalLoops = collections.defaultdict(lambda: {})
+Sections = []
+LongestVcdStrValueLength = 0
+Threads = collections.defaultdict(lambda: [])  # List of records per thread id
+Mtasks = collections.defaultdict(lambda: {'elapsed': 0, 'end': 0})
+Cpus = collections.defaultdict(lambda: {'mtask_time': 0})
 Global = {
    'args': {},
    'cpuinfo': collections.defaultdict(lambda: {}),
-    'rdtsc_cycle_time': 0,
    'stats': {}
 }
+ElapsedTime = None  # total elapsed time
+ExecGraphTime = 0  # total elapsed time excuting an exec graph
+ExecGraphIntervals = []  # list of (start, end) pairs

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


-def process(filename):
-    read_data(filename)
-    report()
-
-
 def read_data(filename):
    with open(filename, "r", encoding="utf8") as fh:
        re_thread = re.compile(r'^VLPROFTHREAD (\d+)$')
@@ -39,14 +38,17 @@ def read_data(filename):
        re_arg1 = re.compile(r'VLPROF arg\s+(\S+)\+([0-9.]*)\s*')
        re_arg2 = re.compile(r'VLPROF arg\s+(\S+)\s+([0-9.]*)\s*$')
        re_stat = re.compile(r'VLPROF stat\s+(\S+)\s+([0-9.]+)')
-        re_time = re.compile(r'rdtsc time = (\d+) ticks')
        re_proc_cpu = re.compile(r'VLPROFPROC processor\s*:\s*(\d+)\s*$')
        re_proc_dat = re.compile(r'VLPROFPROC ([a-z_ ]+)\s*:\s*(.*)$')
        cpu = None
        thread = None
+        execGraphStart = None

-        lastEvalBeginTick = None
-        lastEvalLoopBeginTick = None
+        global LongestVcdStrValueLength
+        global ExecGraphTime
+
+        SectionStack = []
+        mTaskThread = {}

        for line in fh:
            recordMatch = re_record.match(line)
@@ -54,29 +56,31 @@ def read_data(filename):
                kind, tick, payload = recordMatch.groups()
                tick = int(tick)
                payload = payload.strip()
-                if kind == "EVAL_BEGIN":
-                    Evals[tick]['start'] = tick
-                    lastEvalBeginTick = tick
-                elif kind == "EVAL_END":
-                    Evals[lastEvalBeginTick]['end'] = tick
-                    lastEvalBeginTick = None
-                elif kind == "EVAL_LOOP_BEGIN":
-                    EvalLoops[tick]['start'] = tick
-                    lastEvalLoopBeginTick = tick
-                elif kind == "EVAL_LOOP_END":
-                    EvalLoops[lastEvalLoopBeginTick]['end'] = tick
-                    lastEvalLoopBeginTick = None
+                if kind == "SECTION_PUSH":
+                    LongestVcdStrValueLength = max(LongestVcdStrValueLength,
+                                                   len(payload))
+                    SectionStack.append(payload)
+                    Sections.append((tick, tuple(SectionStack)))
+                elif kind == "SECTION_POP":
+                    assert SectionStack, "SECTION_POP without SECTION_PUSH"
+                    SectionStack.pop()
+                    Sections.append((tick, tuple(SectionStack)))
                elif kind == "MTASK_BEGIN":
                    mtask, predict_start, ecpu = re_payload_mtaskBegin.match(
                        payload).groups()
                    mtask = int(mtask)
                    predict_start = int(predict_start)
                    ecpu = int(ecpu)
-                    Threads[thread][tick]['mtask'] = mtask
-                    Threads[thread][tick]['predict_start'] = predict_start
-                    Threads[thread][tick]['cpu'] = ecpu
-                    if 'elapsed' not in Mtasks[mtask]:
-                        Mtasks[mtask] = {'end': 0, 'elapsed': 0}
+                    mTaskThread[mtask] = thread
+                    records = Threads[thread]
+                    assert not records or records[-1]['start'] <= records[-1][
+                        'end'] <= tick
+                    records.append({
+                        'start': tick,
+                        'mtask': mtask,
+                        'predict_start': predict_start,
+                        'cpu': ecpu
+                    })
                    Mtasks[mtask]['begin'] = tick
                    Mtasks[mtask]['thread'] = thread
                    Mtasks[mtask]['predict_start'] = predict_start
@@ -86,11 +90,18 @@ def read_data(filename):
                    mtask = int(mtask)
                    predict_cost = int(predict_cost)
                    begin = Mtasks[mtask]['begin']
-                    Threads[thread][begin]['end'] = tick
-                    Threads[thread][begin]['predict_cost'] = predict_cost
+                    record = Threads[mTaskThread[mtask]][-1]
+                    record['end'] = tick
+                    record['predict_cost'] = predict_cost
                    Mtasks[mtask]['elapsed'] += tick - begin
                    Mtasks[mtask]['predict_cost'] = predict_cost
                    Mtasks[mtask]['end'] = max(Mtasks[mtask]['end'], tick)
+                elif kind == "EXEC_GRAPH_BEGIN":
+                    execGraphStart = tick
+                elif kind == "EXEC_GRAPH_END":
+                    ExecGraphTime += tick - execGraphStart
+                    ExecGraphIntervals.append((execGraphStart, tick))
+                    execGraphStart = None
                elif Args.debug:
                    print("-Unknown execution trace record: %s" % line)
            elif re_thread.match(line):
@@ -109,7 +120,7 @@ def read_data(filename):
            elif re_proc_cpu.match(line):
                match = re_proc_cpu.match(line)
                cpu = int(match.group(1))
-            elif cpu and re_proc_dat.match(line):
+            elif cpu is not None and re_proc_dat.match(line):
                match = re_proc_dat.match(line)
                term = match.group(1)
                value = match.group(2)
@@ -121,11 +132,6 @@ def read_data(filename):
                pass
            elif Args.debug:
                print("-Unk: %s" % line)
-            # TODO -- this is parsing text printed by a client.
-            # Really, verilator proper should generate this
-            # if it's useful...
-            if re_time.match(line):
-                Global['rdtsc_cycle_time'] = re_time.group(1)


 def re_match_result(regexp, line, result_to):
@@ -144,125 +150,33 @@ def report():
        plus = "+" if re.match(r'^\+', arg) else " "
        print("  %s%s%s" % (arg, plus, Global['args'][arg]))

+    for records in Threads.values():
+        for record in records:
+            cpu = record['cpu']
+            elapsed = record['end'] - record['start']
+            Cpus[cpu]['mtask_time'] += elapsed
+
+    global ElapsedTime
+    ElapsedTime = int(Global['stats']['ticks'])
    nthreads = int(Global['stats']['threads'])
-    Global['cpus'] = {}
-    for thread in Threads:
-        # Make potentially multiple characters per column
-        for start in Threads[thread]:
-            if not Threads[thread][start]:
-                continue
-            cpu = Threads[thread][start]['cpu']
-            elapsed = Threads[thread][start]['end'] - start
-            if cpu not in Global['cpus']:
-                Global['cpus'][cpu] = {'cpu_time': 0}
-            Global['cpus'][cpu]['cpu_time'] += elapsed
+    ncpus = max(len(Cpus), 1)

-    measured_mt_mtask_time = 0
-    predict_mt_mtask_time = 0
-    long_mtask_time = 0
-    measured_last_end = 0
-    predict_last_end = 0
-    for mtask in Mtasks:
-        measured_mt_mtask_time += Mtasks[mtask]['elapsed']
-        predict_mt_mtask_time += Mtasks[mtask]['predict_cost']
-        measured_last_end = max(measured_last_end, Mtasks[mtask]['end'])
-        predict_last_end = max(
-            predict_last_end,
-            Mtasks[mtask]['predict_start'] + Mtasks[mtask]['predict_cost'])
-        long_mtask_time = max(long_mtask_time, Mtasks[mtask]['elapsed'])
-    Global['measured_last_end'] = measured_last_end
-    Global['predict_last_end'] = predict_last_end
-
-    # If we know cycle time in the same (rdtsc) units,
-    # this will give us an actual utilization number,
-    # (how effectively we keep the cores busy.)
-    #
-    # It also gives us a number we can compare against
-    # serial mode, to estimate the overhead of data sharing,
-    # which will show up in the total elapsed time. (Overhead
-    # of synchronization and scheduling should not.)
-    print("\nAnalysis:")
-    print("  Total threads             = %d" % nthreads)
-    print("  Total mtasks              = %d" % len(Mtasks))
-    ncpus = max(len(Global['cpus']), 1)
-    print("  Total cpus used           = %d" % ncpus)
-    print("  Total yields              = %d" %
-          int(Global['stats'].get('yields', 0)))
-    print("  Total evals               = %d" % len(Evals))
-    print("  Total eval loops          = %d" % len(EvalLoops))
-    if Mtasks:
-        print("  Total eval time           = %d rdtsc ticks" %
-              Global['measured_last_end'])
-        print("  Longest mtask time        = %d rdtsc ticks" % long_mtask_time)
-        print("  All-thread mtask time     = %d rdtsc ticks" %
-              measured_mt_mtask_time)
-        long_efficiency = long_mtask_time / (Global.get(
-            'measured_last_end', 1) or 1)
-        print("  Longest-thread efficiency = %0.1f%%" %
-              (long_efficiency * 100.0))
-        mt_efficiency = measured_mt_mtask_time / (
-            Global.get('measured_last_end', 1) * nthreads or 1)
-        print("  All-thread efficiency     = %0.1f%%" %
-              (mt_efficiency * 100.0))
-        print("  All-thread speedup        = %0.1f" %
-              (mt_efficiency * nthreads))
-        if Global['rdtsc_cycle_time'] > 0:
-            ut = measured_mt_mtask_time / Global['rdtsc_cycle_time']
-            print("tot_mtask_cpu=" + measured_mt_mtask_time + " cyc=" +
-                  Global['rdtsc_cycle_time'] + " ut=" + ut)
-
-        predict_mt_efficiency = predict_mt_mtask_time / (
-            Global.get('predict_last_end', 1) * nthreads or 1)
-        print("\nPrediction (what Verilator used for scheduling):")
-        print("  All-thread efficiency     = %0.1f%%" %
-              (predict_mt_efficiency * 100.0))
-        print("  All-thread speedup        = %0.1f" %
-              (predict_mt_efficiency * nthreads))
-
-        p2e_ratios = []
-        min_p2e = 1000000
-        min_mtask = None
-        max_p2e = -1000000
-        max_mtask = None
-
-        for mtask in sorted(Mtasks.keys()):
-            if Mtasks[mtask]['elapsed'] > 0:
-                if Mtasks[mtask]['predict_cost'] == 0:
-                    Mtasks[mtask]['predict_cost'] = 1  # don't log(0) below
-                p2e_ratio = math.log(Mtasks[mtask]['predict_cost'] /
-                                     Mtasks[mtask]['elapsed'])
-                p2e_ratios.append(p2e_ratio)
-
-                if p2e_ratio > max_p2e:
-                    max_p2e = p2e_ratio
-                    max_mtask = mtask
-                if p2e_ratio < min_p2e:
-                    min_p2e = p2e_ratio
-                    min_mtask = mtask
-
-        print("\nMTask statistics:")
-        print("  min log(p2e) = %0.3f" % min_p2e, end="")
-        print("  from mtask %d (predict %d," %
-              (min_mtask, Mtasks[min_mtask]['predict_cost']),
-              end="")
-        print(" elapsed %d)" % Mtasks[min_mtask]['elapsed'])
-        print("  max log(p2e) = %0.3f" % max_p2e, end="")
-        print("  from mtask %d (predict %d," %
-              (max_mtask, Mtasks[max_mtask]['predict_cost']),
-              end="")
-        print(" elapsed %d)" % Mtasks[max_mtask]['elapsed'])
-
-        stddev = statistics.pstdev(p2e_ratios)
-        mean = statistics.mean(p2e_ratios)
-        print("  mean = %0.3f" % mean)
-        print("  stddev = %0.3f" % stddev)
-        print("  e ^ stddev = %0.3f" % math.exp(stddev))
+    print("\nSummary:")
+    print("  Total elapsed time = {} rdtsc ticks".format(ElapsedTime))
+    print("  Parallelized code  = {:.2%} of elapsed time".format(
+        ExecGraphTime / ElapsedTime))
+    print("  Total threads      = %d" % nthreads)
+    print("  Total CPUs used    = %d" % ncpus)
+    print("  Total mtasks       = %d" % len(Mtasks))
+    print("  Total yields       = %d" % int(Global['stats'].get('yields', 0)))

+    report_mtasks()
    report_cpus()
+    report_sections()

    if nthreads > ncpus:
        print()
-        print("%%Warning: There were fewer CPUs (%d) then threads (%d)." %
+        print("%%Warning: There were fewer CPUs (%d) than threads (%d)." %
              (ncpus, nthreads))
        print("        : See docs on use of numactl.")
    else:
@@ -279,32 +193,130 @@ def report():
    print()


+def report_mtasks():
+    if not Mtasks:
+        return
+
+    nthreads = int(Global['stats']['threads'])
+
+    # If we know cycle time in the same (rdtsc) units,
+    # this will give us an actual utilization number,
+    # (how effectively we keep the cores busy.)
+    #
+    # It also gives us a number we can compare against
+    # serial mode, to estimate the overhead of data sharing,
+    # which will show up in the total elapsed time. (Overhead
+    # of synchronization and scheduling should not.)
+    total_mtask_time = 0
+    thread_mtask_time = collections.defaultdict(lambda: 0)
+    long_mtask_time = 0
+    long_mtask = None
+    predict_mtask_time = 0
+    predict_elapsed = 0
+    for mtaskId in Mtasks:
+        record = Mtasks[mtaskId]
+        predict_mtask_time += record['predict_cost']
+        total_mtask_time += record['elapsed']
+        thread_mtask_time[record['thread']] += record['elapsed']
+        predict_end = record['predict_start'] + record['predict_cost']
+        predict_elapsed = max(predict_elapsed, predict_end)
+        if record['elapsed'] > long_mtask_time:
+            long_mtask_time = record['elapsed']
+            long_mtask = mtaskId
+    Global['predict_last_end'] = predict_elapsed
+
+    serialTime = ElapsedTime - ExecGraphTime
+
+    def subReport(elapsed, work):
+        print("  Thread utilization = {:7.2%}".format(work /
+                                                      (elapsed * nthreads)))
+        print("  Speedup            = {:6.3}x".format(work / elapsed))
+
+    print("\nParallelized code, measured:")
+    subReport(ExecGraphTime, total_mtask_time)
+
+    print("\nParallelized code, predicted during static scheduling:")
+    subReport(predict_elapsed, predict_mtask_time)
+
+    print("\nAll code, measured:")
+    subReport(ElapsedTime, serialTime + total_mtask_time)
+
+    print("\nAll code, measured, scaled by predicted speedup:")
+    expectedParallelSpeedup = predict_mtask_time / predict_elapsed
+    scaledElapsed = serialTime + total_mtask_time / expectedParallelSpeedup
+    subReport(scaledElapsed, serialTime + total_mtask_time)
+
+    p2e_ratios = []
+    min_p2e = 1000000
+    min_mtask = None
+    max_p2e = -1000000
+    max_mtask = None
+
+    for mtask in sorted(Mtasks.keys()):
+        if Mtasks[mtask]['elapsed'] > 0:
+            if Mtasks[mtask]['predict_cost'] == 0:
+                Mtasks[mtask]['predict_cost'] = 1  # don't log(0) below
+            p2e_ratio = math.log(Mtasks[mtask]['predict_cost'] /
+                                 Mtasks[mtask]['elapsed'])
+            p2e_ratios.append(p2e_ratio)
+
+            if p2e_ratio > max_p2e:
+                max_p2e = p2e_ratio
+                max_mtask = mtask
+            if p2e_ratio < min_p2e:
+                min_p2e = p2e_ratio
+                min_mtask = mtask
+
+    print("\nMTask statistics:")
+    print("  Longest mtask id = {}".format(long_mtask))
+    print("  Longest mtask time = {:.2%} of time elapsed in parallelized code".
+          format(long_mtask_time / ExecGraphTime))
+    print("  min log(p2e) = %0.3f" % min_p2e, end="")
+
+    print("  from mtask %d (predict %d," %
+          (min_mtask, Mtasks[min_mtask]['predict_cost']),
+          end="")
+    print(" elapsed %d)" % Mtasks[min_mtask]['elapsed'])
+    print("  max log(p2e) = %0.3f" % max_p2e, end="")
+    print("  from mtask %d (predict %d," %
+          (max_mtask, Mtasks[max_mtask]['predict_cost']),
+          end="")
+    print(" elapsed %d)" % Mtasks[max_mtask]['elapsed'])
+
+    stddev = statistics.pstdev(p2e_ratios)
+    mean = statistics.mean(p2e_ratios)
+    print("  mean = %0.3f" % mean)
+    print("  stddev = %0.3f" % stddev)
+    print("  e ^ stddev = %0.3f" % math.exp(stddev))
+
+
 def report_cpus():
-    print("\nCPUs:")
+    print("\nCPU info:")

    Global['cpu_sockets'] = collections.defaultdict(lambda: 0)
    Global['cpu_socket_cores'] = collections.defaultdict(lambda: 0)

-    for cpu in sorted(Global['cpus'].keys()):
-        print("  cpu %d: " % cpu, end='')
-        print("cpu_time=%d" % Global['cpus'][cpu]['cpu_time'], end='')
-
-        socket = None
+    print("   Id | Time spent executing MTask | Socket | Core | Model")
+    print("      | % of elapsed ticks / ticks |        |      |")
+    print("  ====|============================|========|======|======")
+    for cpu in sorted(Cpus):
+        socket = ""
+        core = ""
+        model = ""
        if cpu in Global['cpuinfo']:
            cpuinfo = Global['cpuinfo'][cpu]
            if 'physical_id' in cpuinfo and 'core_id' in cpuinfo:
-                socket = int(cpuinfo['physical_id'])
+                socket = cpuinfo['physical_id']
                Global['cpu_sockets'][socket] += 1
-                print(" socket=%d" % socket, end='')
-
-                core = int(cpuinfo['core_id'])
-                Global['cpu_socket_cores'][str(socket) + "__" + str(core)] += 1
-                print(" core=%d" % core, end='')
+                core = cpuinfo['core_id']
+                Global['cpu_socket_cores'][socket + "__" + core] += 1

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

    if len(Global['cpu_sockets']) > 1:
        Global['cpu_sockets_warning'] = True
@@ -313,25 +325,68 @@ def report_cpus():
                Global['cpu_socket_cores_warning'] = True


+def report_sections():
+    if not Sections:
+        return
+    print("\nSection profile:")
+
+    totalTime = collections.defaultdict(lambda: 0)
+    selfTime = collections.defaultdict(lambda: 0)
+
+    sectionTree = [0, {}, 1]  # [selfTime, childTrees, numberOfTimesEntered]
+    prevTime = 0
+    prevStack = ()
+    for time, stack in Sections:
+        if len(stack) > len(prevStack):
+            scope = sectionTree
+            for item in stack:
+                scope = scope[1].setdefault(item, [0, {}, 0])
+            scope[2] += 1
+        dt = time - prevTime
+        scope = sectionTree
+        for item in prevStack:
+            scope = scope[1].setdefault(item, [0, {}, 0])
+        scope[0] += dt
+
+        if prevStack:
+            for name in prevStack:
+                totalTime[name] += dt
+            selfTime[prevStack[-1]] += dt
+        prevTime = time
+        prevStack = stack
+
+    def treeSum(tree):
+        n = tree[0]
+        for subTree in tree[1].values():
+            n += treeSum(subTree)
+        return n
+
+    # Make sure the tree sums to the elapsed time
+    sectionTree[0] += ElapsedTime - treeSum(sectionTree)
+
+    def printTree(prefix, name, entries, tree):
+        print("  {:7.2%} | {:7.2%} | {:8} | {:10.2f} | {}".format(
+            treeSum(tree) / ElapsedTime, tree[0] / ElapsedTime, tree[2],
+            tree[2] / entries, prefix + name))
+        for k in sorted(tree[1], key=lambda _: -treeSum(tree[1][_])):
+            printTree(prefix + "  ", k, tree[2], tree[1][k])
+
+    print(" Total    | Self    | Total    | Relative   | Section")
+    print(" time     | time    | entries  | entries    |  name  ")
+    print("==========|=========|==========|============|========")
+    printTree("", "*TOTAL*", 1, sectionTree)
+
+
 ######################################################################


 def write_vcd(filename):
    print("Writing %s" % filename)
    with open(filename, "w", encoding="utf8") as fh:
-        vcd = {
-            'values':
-            collections.defaultdict(lambda: {}),  # {<time>}{<code>} = value
-            'sigs': {
-                'predicted_threads': {},
-                'measured_threads': {},
-                'cpus': {},
-                'evals': {},
-                'mtasks': {},
-                'Stats': {}
-            }  # {<module>}{<sig}} = code
-        }
-        code = 0
+        # dict of dicts of hierarchy elements/signal name  -> (code, width)
+        scopeSigs = {}
+        # map from time -> code -> value
+        values = collections.defaultdict(lambda: {})

        parallelism = {
            'measured': collections.defaultdict(lambda: 0),
@@ -340,132 +395,139 @@ def write_vcd(filename):
        parallelism['measured'][0] = 0
        parallelism['predicted'][0] = 0

-        # Measured graph
-        for thread in sorted(Threads.keys()):
-            sig = "thread%d_mtask" % thread
-            if sig not in vcd['sigs']['measured_threads']:
-                vcd['sigs']['measured_threads'][sig] = code
-                code += 1
-            mcode = vcd['sigs']['measured_threads'][sig]
+        nextFreeCode = 0

-            for start in sorted(Threads[thread]):
-                mtask = Threads[thread][start]['mtask']
-                end = Threads[thread][start]['end']
-                cpu = Threads[thread][start]['cpu']
-                vcd['values'][start][mcode] = mtask
-                vcd['values'][end][mcode] = None
+        def getCode(width, *names):
+            nonlocal nextFreeCode
+            scope = scopeSigs
+            *path, name = names
+            for item in path:
+                scope = scope.setdefault(item, {})
+            code, oldWidth = scope.setdefault(name, (nextFreeCode, width))
+            assert oldWidth == width
+            if code == nextFreeCode:
+                nextFreeCode += 1
+            return code
+
+        def addValue(code, time, val):
+            if isinstance(val, str):
+                buf = ""
+                for c in val:
+                    buf += bin(ord(c))[2:].rjust(8, "0")
+                val = buf.ljust(LongestVcdStrValueLength * 8, "0")
+            values[time][code] = val
+
+        # Measured graph
+        for thread in sorted(Threads):
+            mcode = getCode(32, 'measured', 't%d_mtask' % thread)
+            for record in Threads[thread]:
+                start = record['start']
+                mtask = record['mtask']
+                end = record['end']
+                cpu = record['cpu']
+                addValue(mcode, start, mtask)
+                addValue(mcode, end, None)
                parallelism['measured'][start] += 1
                parallelism['measured'][end] -= 1

-                sig = "cpu%d_thread" % cpu
-                if sig not in vcd['sigs']['cpus']:
-                    vcd['sigs']['cpus'][sig] = code
-                    code += 1
-                ccode = vcd['sigs']['cpus'][sig]
-                vcd['values'][start][ccode] = thread
-                vcd['values'][end][ccode] = None
+                code = getCode(32, 'measured', 'cpu%d_thread' % cpu)
+                addValue(code, start, thread)
+                addValue(code, end, None)

-                sig = "mtask%d_cpu" % mtask
-                if sig not in vcd['sigs']['mtasks']:
-                    vcd['sigs']['mtasks'][sig] = code
-                    code += 1
-                ccode = vcd['sigs']['mtasks'][sig]
-                vcd['values'][start][ccode] = cpu
-                vcd['values'][end][ccode] = None
+                code = getCode(32, 'measured', 'cpu%d_mtask' % cpu)
+                addValue(code, start, mtask)
+                addValue(code, end, None)

-        # Eval graph
-        vcd['sigs']['evals']["eval"] = code
-        elcode = code
-        code += 1
-        n = 0
-        for eval_start in Evals:
-            eval_end = Evals[eval_start]['end']
-            n += 1
-            vcd['values'][eval_start][elcode] = n
-            vcd['values'][eval_end][elcode] = None
+        tStart = sorted(_['start'] for records in Threads.values()
+                        for _ in records)
+        tEnd = sorted(_['end'] for records in Threads.values()
+                      for _ in records)

-        # Eval_loop graph
-        vcd['sigs']['evals']["eval_loop"] = code
-        elcode = code
-        code += 1
-        n = 0
-        for eval_start in EvalLoops:
-            eval_end = EvalLoops[eval_start]['end']
-            n += 1
-            vcd['values'][eval_start][elcode] = n
-            vcd['values'][eval_end][elcode] = None
+        # Predicted graph
+        for start, end in ExecGraphIntervals:
+            # Find the earliest MTask start after the start point, and the
+            # latest MTask end before the end point, so we can scale to the
+            # same range
+            start = tStart[bisect.bisect_left(tStart, start)]
+            end = tEnd[bisect.bisect_right(tEnd, end) - 1]
+            # Compute scale so predicted graph is of same width as interval
+            measured_scaling = (end - start) / Global['predict_last_end']
+            # Predict mtasks that fill the time the execution occupied
+            for mtask in Mtasks:
+                thread = Mtasks[mtask]['thread']
+                pred_scaled_start = start + int(
+                    Mtasks[mtask]['predict_start'] * measured_scaling)
+                pred_scaled_end = start + int(
+                    (Mtasks[mtask]['predict_start'] +
+                     Mtasks[mtask]['predict_cost']) * measured_scaling)
+                if pred_scaled_start == pred_scaled_end:
+                    continue

-        if Mtasks:
-            # Predicted graph
-            for eval_start in EvalLoops:
-                eval_end = EvalLoops[eval_start]['end']
-                # Compute scale so predicted graph is of same width as eval
-                measured_scaling = (eval_end -
-                                    eval_start) / Global['predict_last_end']
-                # Predict mtasks that fill the time the eval occupied
-                for mtask in Mtasks:
-                    thread = Mtasks[mtask]['thread']
-                    pred_scaled_start = eval_start + int(
-                        Mtasks[mtask]['predict_start'] * measured_scaling)
-                    pred_scaled_end = eval_start + int(
-                        (Mtasks[mtask]['predict_start'] +
-                         Mtasks[mtask]['predict_cost']) * measured_scaling)
-                    if pred_scaled_start == pred_scaled_end:
-                        continue
+                mcode = getCode(32, 'predicted', 't%d_mtask' % thread)
+                addValue(mcode, pred_scaled_start, mtask)
+                addValue(mcode, pred_scaled_end, None)

-                    sig = "predicted_thread%d_mtask" % thread
-                    if sig not in vcd['sigs']['predicted_threads']:
-                        vcd['sigs']['predicted_threads'][sig] = code
-                        code += 1
-                    mcode = vcd['sigs']['predicted_threads'][sig]
+                parallelism['predicted'][pred_scaled_start] += 1
+                parallelism['predicted'][pred_scaled_end] -= 1

-                    vcd['values'][pred_scaled_start][mcode] = mtask
-                    vcd['values'][pred_scaled_end][mcode] = None
+        # Parallelism graph
+        for measpred in ('measured', 'predicted'):
+            pcode = getCode(32, 'stats', '%s_parallelism' % measpred)
+            value = 0
+            for time in sorted(parallelism[measpred].keys()):
+                value += parallelism[measpred][time]
+                addValue(pcode, time, value)

-                    parallelism['predicted'][pred_scaled_start] += 1
-                    parallelism['predicted'][pred_scaled_end] -= 1
-
-            # Parallelism graph
-            for measpred in ('measured', 'predicted'):
-                vcd['sigs']['Stats']["%s_parallelism" % measpred] = code
-                pcode = code
-                code += 1
-                value = 0
-                for time in sorted(parallelism[measpred].keys()):
-                    value += parallelism[measpred][time]
-                    vcd['values'][time][pcode] = value
+        # Section graph
+        if Sections:
+            scode = getCode(LongestVcdStrValueLength * 8, "section", "trace")
+            dcode = getCode(32, "section", "depth")
+            for time, stack in Sections:
+                addValue(scode, time, stack[-1] if stack else None)
+                addValue(dcode, time, len(stack))

        # Create output file
        fh.write("$version Generated by verilator_gantt $end\n")
        fh.write("$timescale 1ns $end\n")
        fh.write("\n")

-        all_codes = {}
+        all_codes = set()
+
+        def writeScope(scope):
+            assert isinstance(scope, dict)
+            for key in sorted(scope):
+                val = scope[key]
+                if isinstance(val, dict):
+                    fh.write("  $scope module %s $end\n" % key)
+                    writeScope(val)
+                    fh.write("  $upscope $end\n")
+                else:
+                    (code, width) = val
+                    all_codes.add(code)
+                    fh.write("   $var wire %d v%x %s [%d:0] $end\n" %
+                             (width, code, key, width - 1))
+
        fh.write(" $scope module gantt $end\n")
-        for module in sorted(vcd['sigs'].keys()):
-            fh.write("  $scope module %s $end\n" % module)
-            for sig in sorted(vcd['sigs'][module].keys()):
-                code = vcd['sigs'][module][sig]
-                fh.write("   $var wire 32 v%x %s [31:0] $end\n" % (code, sig))
-                all_codes[code] = 1
-            fh.write("  $upscope $end\n")
+        writeScope(scopeSigs)
        fh.write(" $upscope $end\n")
        fh.write("$enddefinitions $end\n")
        fh.write("\n")

        first = True
-        for time in sorted(vcd['values']):
+        for time in sorted(values):
            if first:
                first = False
                # Start with Z for any signals without time zero data
-                for code in sorted(all_codes.keys()):
-                    if code not in vcd['values'][time]:
-                        vcd['values'][time][code] = None
+                for code in sorted(all_codes):
+                    if code not in values[time]:
+                        addValue(code, time, None)
            fh.write("#%d\n" % time)
-            for code in sorted(vcd['values'][time].keys()):
-                value = vcd['values'][time][code]
+            for code in sorted(values[time]):
+                value = values[time][code]
                if value is None:
                    fh.write("bz v%x\n" % code)
+                elif isinstance(value, str):
+                    fh.write("b%s v%x\n" % (value, code))
                else:
                    fh.write("b%s v%x\n" % (format(value, 'b'), code))

@@ -484,7 +546,7 @@ Verilator_gantt creates a visual representation to help analyze Verilator
 For documentation see
 https://verilator.org/guide/latest/exe_verilator_gantt.html""",
    epilog=
-    """Copyright 2018-2023 by Wilson Snyder. This program is free software; you
+    """Copyright 2018-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -504,7 +566,8 @@ parser.add_argument('filename',

 Args = parser.parse_args()

-process(Args.filename)
+read_data(Args.filename)
+report()
 if not Args.no_vcd:
    write_vcd(Args.vcd)

@@ -1,7 +1,7 @@
 #!/usr/bin/env python3
 # pylint: disable=C0114,C0209
 #
-# Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+# Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 # can redistribute it and/or modify the Verilator internals under the terms
 # of either the GNU Lesser General Public License Version 3 or the Perl
 # Artistic License Version 2.0.
@@ -181,7 +181,7 @@ in each Verilog block.
 For documentation see
 https://verilator.org/guide/latest/exe_verilator_profcfunc.html""",
    epilog=
-    """Copyright 2002-2023 by Wilson Snyder. This program is free software; you
+    """Copyright 2002-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -26,4 +26,5 @@ if git log --format=%B -n 1 "$COMMIT" | grep -q -i '\[CI\s\+ccache\s\+clear\]';
 fi

 # Dump stats, then zero stats
+# 22.04: ccache -s -v -z
 ccache -s -z
@@ -49,6 +49,7 @@ if [ "$CI_BUILD_STAGE_NAME" = "build" ]; then
    ./configure --enable-longtests --enable-ccwarn ${CI_M32:+--enable-m32}
    ccache -z
    "$MAKE" -j "$NPROC" -k
+    # 22.04: ccache -s -v
    ccache -s
    if [ "$CI_OS_NAME" = "osx" ]; then
      file bin/verilator_bin
@@ -89,8 +90,8 @@ elif [ "$CI_BUILD_STAGE_NAME" = "test" ]; then
    export VERILATOR_TEST_NO_GPROF=1 # gprof is a bit different on FreeBSD, disable
  fi

-  # Run sanitize on Ubuntu 20.04 only
-  [ "$CI_RUNS_ON" = 'ubuntu-20.04' ] && [ "$CI_M32" = "" ] && sanitize='--sanitize' || sanitize=''
+  # Run sanitize on Ubuntu 22.04 only
+  [ "$CI_RUNS_ON" = 'ubuntu-22.04' ] && [ "$CI_M32" = "" ] && sanitize='--sanitize' || sanitize=''

  # Run the specified test
  ccache -z
@@ -180,6 +181,7 @@ elif [ "$CI_BUILD_STAGE_NAME" = "test" ]; then
      fatal "Unknown test: $TESTS"
      ;;
  esac
+  # 22.04: ccache -s -v
  ccache -s

  # Upload coverage data
@@ -8,6 +8,16 @@

 FROM ubuntu:22.04

+# Create the user
+RUN groupadd verilator \
+    && useradd -g verilator -m verilator -s /bin/bash \
+    && apt-get update \
+    && apt-get install --no-install-recommends -y sudo \
+    && apt-get clean \
+    && rm -rf /var/lib/apt/lists/* \
+    && echo verilator ALL=\(root\) NOPASSWD:ALL > /etc/sudoers.d/verilator \
+    && chmod 0440 /etc/sudoers.d/verilator
+
 RUN apt-get update \
    && DEBIAN_FRONTEND=noninteractive \
       apt-get install --no-install-recommends -y \
@@ -33,7 +43,6 @@ RUN apt-get update \
                        perl \
                        python3 \
                        wget \
-                        zlibc \
                        zlib1g \
                        zlib1g-dev \
    && apt-get clean \
@@ -52,6 +61,8 @@ COPY build.sh /tmp/build.sh

 ENV VERILATOR_AUTHOR_SITE=1

+USER verilator
+
 WORKDIR /work

 ENTRYPOINT [ "/tmp/build.sh" ]
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 Verilator Build Docker Container
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 Verilator Executable Docker Container
@@ -1,6 +1,6 @@
 # DESCRIPTION: codecov.io config
 #
-# Copyright 2020-2023 by Wilson Snyder. This program is free software; you
+# Copyright 2020-2024 by Wilson Snyder. This program is free software; you
 # can redistribute it and/or modify it under the terms of either the GNU
 # Lesser General Public License Version 3 or the Perl Artistic License
 # Version 2.0.
@@ -1,6 +1,6 @@
 # DESCRIPTION: Process this file with autoconf to produce a configure script.
 #
-# Copyright 2003-2023 by Wilson Snyder. Verilator is free software; you
+# Copyright 2003-2024 by Wilson Snyder. Verilator is free software; you
 # can redistribute it and/or modify it under the terms of either the GNU Lesser
 # General Public License Version 3 or the Perl Artistic License Version 2.0.
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
@@ -10,7 +10,7 @@
 # Then 'make maintainer-dist'
 #AC_INIT([Verilator],[#.### YYYY-MM-DD])
 #AC_INIT([Verilator],[#.### devel])
-AC_INIT([Verilator],[5.016 2023-09-16],
+AC_INIT([Verilator],[5.020 2024-01-01],
        [https://verilator.org],
        [verilator],[https://verilator.org])

@@ -24,6 +24,10 @@ AC_SUBST(PACKAGE_VERSION_NUMBER)
 AC_DEFINE_UNQUOTED([PACKAGE_VERSION_NUMBER_STRING],["$PACKAGE_VERSION_NUMBER"],[Package version as a number])
 VERILATOR_VERSION_INTEGER=`AS_ECHO("$PACKAGE_VERSION") | [sed 's/\([0-9]\)\.\([0-9][0-9][0-9]\) .*/\1\2000/g']`
 AC_SUBST(VERILATOR_VERSION_INTEGER)
+AC_DEFINE_UNQUOTED([PACKAGE_VERSION_STRING_CHAR],
+        [static const char* const PACKAGE_STRING_UNUSED = "$PACKAGE_STRING";],
+        [Package version as a number])
+AC_SUBST(PACKAGE_VERSION_STRING_CHAR)

 # Ignore automake flags passed by Ubuntu builds
 AC_ARG_ENABLE([dependency-tracking],
@@ -350,7 +354,7 @@ AC_SUBST(CFG_CXXFLAGS_PROFILE)
 # Macros work such that first option that passes is the one we take
 # Currently enable c++17/c++14 due to packaged SystemC dependency
 # c++17 is the newest that Verilator is regularly tested to support
-# c++11 is the oldest that Verilator supports
+# c++14 is the oldest that Verilator supports
 # gnu is required for Cygwin to compile verilated.h successfully
 #_MY_CXX_CHECK_SET(CFG_CXXFLAGS_STD_NEWEST,-std=gnu++20)
 #_MY_CXX_CHECK_SET(CFG_CXXFLAGS_STD_NEWEST,-std=c++20)
@@ -358,8 +362,6 @@ _MY_CXX_CHECK_SET(CFG_CXXFLAGS_STD_NEWEST,-std=gnu++17)
 _MY_CXX_CHECK_SET(CFG_CXXFLAGS_STD_NEWEST,-std=c++17)
 _MY_CXX_CHECK_SET(CFG_CXXFLAGS_STD_NEWEST,-std=gnu++14)
 _MY_CXX_CHECK_SET(CFG_CXXFLAGS_STD_NEWEST,-std=c++14)
-_MY_CXX_CHECK_SET(CFG_CXXFLAGS_STD_NEWEST,-std=gnu++11)
-_MY_CXX_CHECK_SET(CFG_CXXFLAGS_STD_NEWEST,-std=c++11)
 AC_SUBST(CFG_CXXFLAGS_STD_NEWEST)

 # Flags for compiling Verilator internals including parser, and Verilated files
@@ -371,10 +373,9 @@ _MY_CXX_CHECK_OPT(CFG_CXXFLAGS_WEXTRA,-Wthread-safety)
 AC_SUBST(CFG_CXXFLAGS_WEXTRA)

 # Flags for coroutine support for dynamic scheduling
-_MY_CXX_CHECK_IFELSE(
-  -fcoroutines-ts,
-  [CFG_CXXFLAGS_COROUTINES="-fcoroutines-ts"],
-  [CFG_CXXFLAGS_COROUTINES="-fcoroutines"])
+_MY_CXX_CHECK_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines-ts)
+_MY_CXX_CHECK_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines)
+_MY_CXX_CHECK_SET(CFG_CXXFLAGS_COROUTINES,-fcoroutines-ts -Wno-deprecated-experimental-coroutine)
 AC_SUBST(CFG_CXXFLAGS_COROUTINES)

 # HAVE_COROUTINES
@@ -442,14 +443,19 @@ m4_foreach([cflag],[
        [-mno-cet],
        [-Qunused-arguments],
        [-Wno-bool-operation],
-        [-Wno-tautological-bitwise-compare],
+        [-Wno-c++11-narrowing],
+        [-Wno-constant-logical-operand],
+        [-Wno-non-pod-varargs],
+        [-Wno-overloaded-virtual],
        [-Wno-parentheses-equality],
+        [-Wno-shadow],
        [-Wno-sign-compare],
+        [-Wno-tautological-bitwise-compare],
        [-Wno-uninitialized],
+        [-Wno-unused-but-set-parameter],
        [-Wno-unused-but-set-variable],
        [-Wno-unused-parameter],
-        [-Wno-unused-variable],
-        [-Wno-shadow]],[
+        [-Wno-unused-variable]],[
    _MY_CXX_CHECK_OPT(CFG_CXXFLAGS_NO_UNUSED,cflag)
    # CMake will test what flags work itself, so pass all flags through to it
    CFG_CXX_FLAGS_CMAKE="$CFG_CXX_FLAGS_CMAKE cflag"
@@ -513,14 +519,14 @@ _MY_LDLIBS_CHECK_IFELSE(
   fi])
 AC_SUBST(CFG_LIBS)

-# Need C++11 at least
-#Alternative: AX_CXX_COMPILE_STDCXX([11])
+# Need C++14 at least
+#Alternative: AX_CXX_COMPILE_STDCXX([14])
 AC_DEFUN([_MY_CXX_CHECK_CXX_VER],
-   [# _MY_CXX_CHECK_CXX_VER(flag) -- Check if compiler runs C++11
+   [# _MY_CXX_CHECK_CXX_VER(flag) -- Check if compiler runs C++14
    # Set $_my_result
    AC_LINK_IFELSE(
     [AC_LANG_PROGRAM([#include <thread>
-#if (__cplusplus < 201103L)
+#if (__cplusplus < 201402L)
 # error "Too old"
 #endif
    ], [[ ]])],
@@ -536,13 +542,13 @@ AC_DEFUN([_MY_CXX_CHECK_CXX_VER],
 # Add $CFG_CXXFLAGS_STD only if can't compile correctly otherwise,
 # as adding std= when not needed can cause errors with the C++ std library.
 CFG_CXXFLAGS_STD=$CFG_CXXFLAGS_STD_NEWEST
-AC_MSG_CHECKING(whether $CXX supports C++11)
+AC_MSG_CHECKING(whether $CXX supports C++14)
 _MY_CXX_CHECK_CXX_VER()
 AC_MSG_RESULT($_my_result)
 if test "$_my_result" = "no" ; then
   CXXFLAGS="$CXXFLAGS $CFG_CXXFLAGS_STD"
   CFG_CXX_FLAGS_CMAKE="$CFG_CXX_FLAGS_CMAKE $CFG_CXXFLAGS_STD"
-   AC_MSG_CHECKING(whether $CXX supports C++11 with $CFG_CXXFLAGS_STD)
+   AC_MSG_CHECKING(whether $CXX supports C++14 with $CFG_CXXFLAGS_STD)
   _MY_CXX_CHECK_CXX_VER()
   AC_MSG_RESULT($_my_result)
 else
@@ -552,13 +558,28 @@ else
 fi
 if test "$_my_result" = "no" ; then
   AC_MSG_NOTICE([[]])
-   AC_MSG_ERROR([[the $CXX compiler appears to not support C++11.
+   AC_MSG_ERROR([[the $CXX compiler appears to not support C++14.

-Verilator requires a C++11 or newer compiler.]])
+Verilator requires a C++14 or newer compiler.]])

 fi
 AC_SUBST(CFG_CXXFLAGS_STD)

+# Compiler precompiled header options (assumes either gcc or clang++)
+AC_MSG_CHECKING([for $CXX precompile header include option])
+if $CXX --help | grep include-pch >/dev/null 2>/dev/null ; then
+   # clang
+   CFG_CXXFLAGS_PCH_I=-include-pch
+   CFG_GCH_IF_CLANG=.gch
+else
+   # GCC
+   CFG_CXXFLAGS_PCH_I=-include
+   CFG_GCH_IF_CLANG=
+fi
+AC_MSG_RESULT($CFG_CXXFLAGS_PCH_I)
+AC_SUBST(CFG_CXXFLAGS_PCH_I)
+AC_SUBST(CFG_GCH_IF_CLANG)
+
 # Checks for library functions.
 AC_CHECK_MEMBER([struct stat.st_mtim.tv_nsec],
    [AC_DEFINE([HAVE_STAT_NSEC],[1],[Defined if struct stat has st_mtim.tv_nsec])],
@@ -573,7 +594,7 @@ ACO_SAVE_LIBS="$LIBS"
 LIBS="$LIBS -lsystemc"
 AC_LINK_IFELSE(
    [AC_LANG_PROGRAM([[#include <systemc.h>
-                     extern "C" int sc_main(int argc, char* argv[]) {}
+                     extern "C" int sc_main(int argc, char* argv[]) { return 0; }
                     ]],[[sc_version()]])],
    [_my_result=yes
     AC_DEFINE([HAVE_SYSTEMC],[1],[Defined if have SystemC library])],
@@ -19,12 +19,14 @@ Anthony Donlon
 Arkadiusz Kozdra
 Aylon Chaim Porat
 Cameron Kirk
+Chih-Mao Chen
 Chris Randall
 Chuxuan Wang
 Conor McCullough
 Dan Petrisko
 Daniel Bates
 David Horton
+David Ledger
 David Metz
 David Stanford
 David Turner
@@ -49,6 +51,7 @@ Glen Gibb
 Gökçe Aydos
 Graham Rushton
 Guokai Chen
+Gus Smith
 Gustav Svensk
 Harald Heckmann
 Hennadii Chernyshchyk
@@ -57,6 +60,7 @@ Huang Rui
 Huanghuang Zhou
 HungMingWu
 HyungKi Jeong
+Ícaro Lima
 Ilya Barkov
 Iru Cai
 Ivan Vnučec
@@ -75,12 +79,14 @@ Jesse Taube
 Jevin Sweval
 Jiacheng Qian
 Jiamin Zhu
+Jinyan Xu
 Jiuyang Liu
 Joey Liu
 John Coiner
 John Demme
 John Wehle
 Jonathan Drolet
+Jordan McConnon
 Jose Loyola
 Josep Sans
 Joseph Nwabueze
@@ -88,6 +94,7 @@ Josh Redford
 Julian Daube
 Julie Schwartz
 Julien Margetts
+Justin Thiel
 Kaleb Barrett
 Kamil Rakoczy
 Kanad Kanhere
@@ -114,6 +121,7 @@ Martin Stadler
 Matthew Ballance
 Michael Killough
 Michaël Lefebvre
+Michal Czyz
 Mike Popoloski
 Miodrag Milanović
 Mladen Slijepcevic
@@ -134,7 +142,9 @@ Pierre-Henri Horrein
 Pieter Kapsenberg
 Piotr Binkowski
 Qingyao Sun
+Quentin Corradi
 Rafal Kapuscik
+Rasfunk
 Raynard Qiao
 Richard Myers
 Risto Pejašinović
@@ -179,5 +189,6 @@ Yuri Victorovich
 Yutetsu TAKATSUKASA
 Yves Mathieu
 Zhanglei Wang
+Zhou Shen
 Zixi Li
 أحمد المحمودي
@@ -5,7 +5,7 @@
 #
 # Code available from: https://verilator.org
 #
-# Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+# Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 # can redistribute it and/or modify it under the terms of either the GNU
 # Lesser General Public License Version 3 or the Perl Artistic License
 # Version 2.0.
@@ -0,0 +1,21 @@
+Verilator Documentation
+=======================
+
+This folder contains sources for Verilator documentation.
+
+For formatted documentation see:
+
+- `Verilator README <https://github.com/verilator/verilator>`_
+
+- `Verilator installation and package directory structure
+  <https://verilator.org/install>`_
+
+- `Verilator manual (HTML) <https://verilator.org/verilator_doc.html>`_,
+  or `Verilator manual (PDF) <https://verilator.org/verilator_doc.pdf>`_
+
+- `Subscribe to Verilator announcements
+  <https://github.com/verilator/verilator-announce>`_
+
+- `Verilator forum <https://verilator.org/forum>`_
+
+- `Verilator issues <https://verilator.org/issues>`_
@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-# pylint: disable=C0112,C0114,C0115,C0116,C0209,C0301,R0201,R0903
+# pylint: disable=C0112,C0114,C0115,C0116,C0209,C0301,R0903
 # -*- Python -*- See copyright, etc below
 ######################################################################

@@ -40,7 +40,7 @@ parser = argparse.ArgumentParser(
    formatter_class=argparse.RawDescriptionHelpFormatter,
    description="""Read a file and extract documentation data.""",
    epilog=
-    """ Copyright 2021-2023 by Wilson Snyder.  This package is free software;
+    """ Copyright 2021-2024 by Wilson Snyder.  This package is free software;
 you can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -55,7 +55,7 @@ parser = argparse.ArgumentParser(
    formatter_class=argparse.RawDescriptionHelpFormatter,
    description="""Post-process Sphinx HTML.""",
    epilog=
-    """ Copyright 2021-2023 by Wilson Snyder.  This package is free software;
+    """ Copyright 2021-2024 by Wilson Snyder.  This package is free software;
 you can redistribute it and/or modify it under the terms of either the GNU
 Lesser General Public License Version 3 or the Perl Artistic License
 Version 2.0.
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 ****************
@@ -1,7 +1,7 @@
 # pylint: disable=C0103,C0114,C0116,C0301,E0402,W0622
 #
 # Configuration file for Verilator's Sphinx documentation builder.
-# Copyright 2003-2023 by Wilson Snyder.
+# Copyright 2003-2024 by Wilson Snyder.
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
 #
 # This file only contains overridden options. For a full list:
@@ -230,4 +230,4 @@ spelling_ignore_contributor_names = True

 breathe_projects = {"verilated": "_build/doxygen/verilated/xml/"}
 breathe_default_project = "verilated"
-breathe_default_members = ('members')
+breathe_default_members = 'members'
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Connecting:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 *******************************
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 ************************
@@ -27,107 +27,137 @@ Alliance <https://chipsalliance.org>`_, `Antmicro Ltd
 <https://antmicro.com>`_ and `Shunyao CAD <https://shunyaocad.com>`_.

 Previous major corporate sponsors of Verilator, by providing significant
-contributions of time or funds include: Atmel Corporation, Cavium
-Inc., Compaq Corporation, Digital Equipment Corporation, Embecosm Ltd.,
-Hicamp Systems, Intel Corporation, Mindspeed Technologies Inc., MicroTune
-Inc., picoChip Designs Ltd., Sun Microsystems Inc., Nauticus Networks Inc.,
-SiCortex Inc, and Shunyao CAD.
+contributions of time or funds include: Antmicro Ltd., Atmel Corporation,
+Compaq Corporation, Digital Equipment Corporation, Embecosm Ltd., Hicamp
+Systems, Intel Corporation, Marvell Inc., Mindspeed Technologies Inc.,
+MicroTune Inc., picoChip Designs Ltd., Sun Microsystems Inc., Nauticus
+Networks Inc., SiCortex Inc, Shunyao CAD, and Western Digital Inc.

-The contributors of major functionality are: Krzysztof
-Bieganski, Byron Bradley, Jeremy Bennett, Lane Brooks, John Coiner, Duane
-Galbi, Geza Lore, Todd Strader, Stefan Wallentowitz, Paul Wasson, Jie Xu,
-and Wilson Snyder.  Major testers included Jeff Dutton, Jonathon Donaldson,
-Ralf Karge, David Hewson, Iztok Jeras, Wim Michiels, Alex Solomatnikov,
-Sebastien Van Cauwenberghe, Gene Weber, and Clifford Wolf.
+The contributors of major functionality are: Jeremy Bennett, Krzysztof
+Bieganski, Byron Bradley, Lane Brooks, John Coiner, Duane Galbi, Geza Lore,
+Todd Strader, Yutetsu Takatsukasa, Stefan Wallentowitz, Paul Wasson, Jie
+Xu, and Wilson Snyder.

 Some of the people who have provided ideas, and feedback for Verilator
 include:

-David Addison, Tariq B. Ahmad, Nikana Anastasiadis, Vasu Arasanipalai, Jens
-Arm, Sharad Bagri, Matthew Ballance, Andrew Bardsley, Matthew Barr, Geoff
-Barrett, Kaleb Barrett, Julius Baxter, Jeremy Bennett, Michael Berman, Jean
-Berniolles, Victor Besyakov, Moinak Bhattacharyya, Krzysztof Bieganski,
-David Binderman, Piotr Binkowski, Johan Bjork, David Black, Tymoteusz
-Blazejczyk, Daniel Bone, Gregg Bouchard, Christopher Boumenot, Nick Bowler,
-Byron Bradley, Bryan Brady, Maarten De Braekeleer, Charlie Brej, J Briquet,
-Lane Brooks, John Brownlee, Jeff Bush, Lawrence Butcher, Tony Bybell, Iru
-Cai, Ted Campbell, Chris Candler, Lauren Carlson, Donal Casey, Alex
-Chadwick, Marcel Chang, Aliaksei Chapyzhenka, Guokai Chen, Terry Chen,
-Yi-Chung Chen, Enzo Chi, Robert A. Clark, Ryan Clarke, Allan Cochrane, John
-Coiner, Keith Colbert, Gianfranco Costamagna, Sean Cross, George Cuan, Joe
-DErrico, Lukasz Dalek, Gunter Dannoritzer, Ashutosh Das, Bernard Deadman,
-John Demme, Mike Denio, John Deroo, Philip Derrick, John Dickol, Ruben
-Diez, Danny Ding, Jacko Dirks, Ivan Djordjevic, Jonathon Donaldson, Larry
-Doolittle, Sebastian Dressler, Jonathan Drolet, Alex Duller, Jeff Dutton,
-Tomas Dzetkulic, Usuario Eda, Charles Eddleston, Chandan Egbert, Joe Eiler,
-Ahmed El-Mahmoudy, Trevor Elbourne, Mats Engstrom, Robert Farrell, Eugen
-Fekete, Fabrizio Ferrandi, Udi Finkelstein, Brian Flachs, Bill Flynn,
-Andrea Foletto, Bob Fredieu, Duane Galbi, Mostafa Gamal, Benjamin Gartner,
-Christian Gelinek, Richard E George, Peter Gerst, Glen Gibb, Michael
-Gielda, Barbara Gigerl, Shankar Giri, Dan Gisselquist, Petr Gladkikh, Sam
-Gladstone, Mariusz Glebocki, Andrew Goessling, Amir Gonnen, Chitlesh
-Goorah, Tomasz Gorochowik, Kai Gossner, Sergi Granell, Al Grant, Nathan
-Graybeal, Alexander Grobman, Graham Rushton, Xuan Guo, Driss Hafdi, Neil
-Hamilton, James Hanlon, Oyvind Harboe, Jannis Harder, Junji Hashimoto,
-Thomas Hawkins, Mitch Hayenga, Harald Heckmann, Robert Henry, Stephen
-Henry, David Hewson, Jamey Hicks, Joel Holdsworth, Andrew Holme, Hiroki
-Honda, Alex Hornung, Pierre-Henri Horrein, David Horton, Peter Horvath, Jae
-Hossell, Kuoping Hsu, Teng Huang, Steven Hugg, Alan Hunter, James
-Hutchinson, Ehab Ibrahim, Edgar E. Iglesias, Jamie Iles, Vighnesh Iyer, Ben
-Jackson, Shareef Jalloq, Marlon James, Krzysztof Jankowski, HyungKi Jeong,
-Iztok Jeras, Alexandre Joannou, James Johnson, Christophe Joly, Franck
-Jullien, James Jung, Mike Kagen, Arthur Kahlich, Kaalia Kahn, Guy-Armand
+David Addison, Tariq B. Ahmad, Nikana Anastasiadis, John David Anglin,
+Frederic Antonin, Hans Van Antwerpen, Vasu Arasanipalai, Jens Arm, Rohan
+Arshid, Gökçe Aydos, Adam Bagley, Sharad Bagri, Robert Balas, Marco
+Balboni, Matthew Ballance, Andrew Bardsley, Ilya Barkov, Matthew Barr,
+Geoff Barrett, Kaleb Barrett, Daniel Bates, Julius Baxter, Michael Berman,
+Jean Berniolles, Victor Besyakov, Narayan Bhagavatula, Moinak
+Bhattacharyya, Kritik Bhimani, David Biancolin, David Binderman, Piotr
+Binkowski, Johan Björk, David Black, Tymoteusz Blazejczyk, Scott Bleiweiss,
+David van der Bokke, Daniel Bone, Guy Bonneau, Krzysztof Boroński, Gregg
+Bouchard, Christopher Boumenot, Nick Bowler, Bryan Brady, Maarten De
+Braekeleer, Charlie Brej, J Briquet, John Brownlee, KC Buckenmaier, Jeff
+Bush, Lawrence Butcher, Tony Bybell, Iru Cai, Ted Campbell, Anthony Campos,
+Chris Candler, Lauren Carlson, Gregory Carver, Donal Casey, Sebastien Van
+Cauwenberghe, Alex Chadwick, Greg Chadwick, Marcel Chang, Aliaksei
+Chapyzhenka, Chih-Mao Chen, Guokai Chen, Terry Chen, Yi-Chung Chen, Yurii
+Cherkasov, Hennadii Chernyshchyk, Enzo Chi, Robert A. Clark, Ryan Clarke,
+Allan Cochrane, Keith Colbert, Quentin Corradi, Nassim Corteggiani,
+Gianfranco Costamagna, February Cozzocrea, Sean Cross, George Cuan, Michal
+Czyz, Joe DErrico, Jim Dai, Lukasz Dalek, Laurens van Dam, Gunter
+Dannoritzer, Ashutosh Das, Julian Daube, Bernard Deadman, Peter Debacker,
+John Demme, Mike Denio, John Deroo, Philip Derrick, Aadi Desai, John
+Dickol, Ruben Diez, Danny Ding, Jacko Dirks, Ivan Djordjevic, Brad Dobbie,
+Paul Donahue, Jonathon Donaldson, Anthony Donlon, Caleb Donovick, Larry
+Doolittle, Leendert van Doorn, Sebastian Dressler, Jonathan Drolet, Maciej
+Dudek, Alex Duller, Jeff Dutton, Tomas Dzetkulic, Usuario Eda, Charles
+Eddleston, Chandan Egbert, Joe Eiler, Ahmed El-Mahmoudy, Trevor Elbourne,
+Mats Engstrom, Robert Farrell, Julien Faucher, Olivier Faure, Eugen Fekete,
+Fabrizio Ferrandi, Udi Finkelstein, Brian Flachs, Bill Flynn, Andrea
+Foletto, Alex Forencich, Aurelien Francillon, Bob Fredieu, Manuel
+Freiberger, Mostafa Gamal, Vito Gamberini, Mostafa Garnal, Benjamin
+Gartner, Christian Gelinek, Richard E George, Peter Gerst, Glen Gibb,
+Michael Gielda, Barbara Gigerl, Nimrod Gileadi, Shankar Giri, Dan
+Gisselquist, Petr Gladkikh, Sam Gladstone, Mariusz Glebocki, Embedded Go,
+Andrew Goessling, Amir Gonnen, Chitlesh Goorah, Tomasz Gorochowik, Kai
+Gossner, Tarik Graba, Sergi Granell, Al Grant, Nathan Graybeal, Alexander
+Grobman, Qian Gu, Xuan Guo, Prabhat Gupta, Driss Hafdi, Neil Hamilton,
+James Hanlon, Tang Haojin, Øyvind Harboe, Jannis Harder, David Harris,
+Junji Hashimoto, Thomas Hawkins, Mitch Hayenga, Harald Heckmann, Robert
+Henry, Stephen Henry, Sebastian Hesselbarth, David Hewson, Jamey Hicks,
+Joel Holdsworth, Andrew Holme, Peter Holmes, Hiroki Honda, Alex Hornung,
+Pierre-Henri Horrein, David Horton, Peter Horvath, Jae Hossell, Kuoping
+Hsu, Teng Huang, Steven Hugg, Huanghuang Zhou, Alan Hunter, James
+Hutchinson, Tim Hutt, Ehab Ibrahim, Edgar E. Iglesias, Shahid Ikram, Jamie
+Iles, Vighnesh Iyer, Ben Jackson, Daniel Jacques, Shareef Jalloq, Marlon
+James, Krzysztof Jankowski, Eyck Jentzsch, HyungKi Jeong, Iztok Jeras,
+Alexandre Joannou, James Johnson, Christophe Joly, Justin Jones,
+William D. Jones, Larry Darryl Lee Jr., Franck Jullien, James Jung,
+Yoshitomo Kaneda, Mike Kagen, Arthur Kahlich, Kaalia Kahn, Guy-Armand
 Kamendje, Vasu Kandadi, Kanad Kanhere, Patricio Kaplan, Pieter Kapsenberg,
 Rafal Kapuscik, Ralf Karge, Per Karlsson, Dan Katz, Sol Katzman, Ian
-Kennedy, Michael Killough, Sun Kim, Jonathan Kimmitt, Olof Kindgren, Kevin
-Kiningham, Dan Kirkham, Aleksander Kiryk, Sobhan Klnv, Gernot Koch, Jack
-Koenig, Soon Koh, Nathan Kohagen, Steve Kolecki, Brett Koonce, Will
-Korteland, Wojciech Koszek, Varun Koyyalagunta, Arkadiusz Kozdra, Markus
-Krause, David Kravitz, Roland Kruse, Andreas Kuster, Sergey Kvachonok,
-Charles Eric LaForest, Ed Lander, Steve Lang, Stephane Laurent, Walter
-Lavino, Christian Leber, Larry Lee, Yoda Lee, Michaël Lefebvre, Igor Lesik,
-John Li, Eivind Liland, Yu Sheng Lin, Charlie Lind, Andrew Ling, Jiuyang
-Liu, Paul Liu, Derek Lockhart, Jake Longo, Geza Lore, Arthur Low, Stefan
-Ludwig, Dan Lussier, Fred Ma, Duraid Madina, Affe Mao, Julien Margetts,
+Kennedy, Ami Keren, Michael Killough, Sun Kim, Jonathan Kimmitt, Olof
+Kindgren, Kevin Kiningham, Cameron Kirk, Dan Kirkham, Aleksander Kiryk,
+Sobhan Klnv, Gernot Koch, Jack Koenig, Soon Koh, Nathan Kohagen, Steve
+Kolecki, Brett Koonce, Will Korteland, Andrei Kostovski, Wojciech Koszek,
+Varun Koyyalagunta, Arkadiusz Kozdra, Markus Krause, David Kravitz, Adam
+Krolnik, Roland Kruse, Mahesh Kumashikar, Andreas Kuster, Sergey Kvachonok,
+Charles Eric LaForest, Kevin Laeufer, Ed Lander, Steve Lang, Pierre
+Laroche, Stephane Laurent, Walter Lavino, Christian Leber, David Ledger,
+Alex Lee, Larry Lee, Yoda Lee, Michaël Lefebvre, Dag Lem, Igor Lesik, John
+Li, Kay Li, Zixi Li, Davide Libenzi, Nandor Licker, Eivind Liland, Ícaro
+Lima, Kevin Lin, Yu-Sheng Lin, Charlie Lind, Andrew Ling, Jiuyang Liu, Joey
+Liu, Paul Liu, Derek Lockhart, Jake Longo, Arthur Low, Jose Loyola, Stefan
+Ludwig, Dan Lussier, Konstantin Lübeck, Fred Ma, Liwei Ma, Duraid Madina,
+Oleh Maksymenko, Affe Mao, Julien Margetts, Chick Markley, Alexis Marquet,
 Mark Marshall, Alfonso Martinez, Unai Martinez-Corral, Adrien Le Masle,
-Yves Mathieu, Patrick Maupin, Conor McCullough, Jason McMullan, Elliot
-Mednick, David Metz, Wim Michiels, Miodrag Milanovic, Kevin Millis, Wai Sum
-Mong, Peter Monsson, Sean Moore, Dennis Muhlestein, John Murphy, Matt
-Myers, Nathan Myers, Richard Myers, Dimitris Nalbantis, Peter Nelson, Bob
-Newgard, Rachit Nigam, Paul Nitza, Yossi Nivin, Pete Nixon, Lisa Noack,
-Mark Nodine, Kuba Ober, Andreas Olofsson, Baltazar Ortiz, Aleksander Osman,
-Don Owen, James Pallister, Vassilis Papaefstathiou, Brad Parker, Morten
-Borup Petersen, Dan Petrisko, Maciej Piechotka, David Pierce, Cody
-Piersall, Michael Platzer, Dominic Plunkett, David Poole, Mike Popoloski,
-Roman Popov, Rich Porter, Stefan Post, Niranjan Prabhu, Damien Pretet, Usha
-Priyadharshini, Mark Jackson Pulver, Prateek Puri, Marshal Qiao, Nandu Raj,
-Kamil Rakoczy, Danilo Ramos, Drew Ranck, Chris Randall, Anton Rapp, Josh
-Redford, Odd Magne Reitan, Frederic Requin, Dustin Richmond, Samuel Riedel,
-Alberto Del Rio, Eric Rippey, Oleg Rodionov, Ludwig Rogiers, Paul Rolfe,
-Arjen Roodselaar, Tobias Rosenkranz, Ryszard Rozak, Huang Rui, Graham
-Rushton, Jan Egil Ruud, Denis Rystsov, John Sanguinetti, Martin Schmidt,
-Julie Schwartz, Galen Seitz, Joseph Shaker, Salman Sheikh, Hao Shi, Mike
-Shinkarovsky, Rafael Shirakawa, Jeffrey Short, Fan Shupei, Anderson Ignacio
-da Silva, Rodney Sinclair, Ameya Vikram Singh, Steven Slatter, Mladen
-Slijepcevic, Brian Small, Garrett Smith, Tim Snyder, Wilson Snyder, Maciej
+Yves Mathieu, Vladimir Matveyenko, Patrick Maupin, Stan Mayer, Conor
+McCullough, Jason McMullan, Elliot Mednick, Yuan Mei, Andy Meier,
+Rodrigo A. Melo, Benjamin Menküc, Jake Merdich, David Metz, Wim Michiels,
+Miodrag Milanović, Darryl Miles, Kevin Millis, Andrew Miloradovsky, Wai Sum
+Mong, Peter Monsson, Sean Moore, Stuart Morris, Dennis Muhlestein, John
+Murphy, Matt Myers, Nathan Myers, Richard Myers, Alex Mykyta, Dimitris
+Nalbantis, Peter Nelson, Felix Neumärker, Bob Newgard, Cong Van Nguyen,
+Rachit Nigam, Toru Niina, Paul Nitza, Yossi Nivin, Pete Nixon, Lisa Noack,
+Mark Nodine, Michael Nolan, Andrew Nolte, Joseph Nwabueze, Kuba Ober,
+Andreas Olofsson, Baltazar Ortiz, Aleksander Osman, Don Owen, Tim Paine,
+Deepa Palaniappan, James Pallister, Vassilis Papaefstathiou, Sanggyu Park,
+Brad Parker, Risto Pejašinović, Morten Borup Petersen, Dan Petrisko, Wesley
+Piard, Maciej Piechotka, David Pierce, Cody Piersall, T. Platz, Michael
+Platzer, Dominic Plunkett, David Poole, Michael Popoloski, Roman Popov,
+Aylon Chaim Porat, Oron Port, Rich Porter, Rick Porter, Stefan Post,
+Niranjan Prabhu, Damien Pretet, Harald Pretl, Bill Pringlemeir, Usha
+Priyadharshini, Mark Jackson Pulver, Prateek Puri, Jiacheng Qian, Marshal
+Qiao, Raynard Qiao, Yujia Qiao, Jasen Qin, Frank Qiu, Nandu Raj, Kamil
+Rakoczy, Danilo Ramos, Drew Ranck, Chris Randall, Anton Rapp, Josh Redford,
+Odd Magne Reitan, Frédéric Requin, Dustin Richmond, Samuel Riedel, Alberto
+Del Rio, Eric Rippey, Oleg Rodionov, Ludwig Rogiers, Paul Rolfe, Michail
+Rontionov, Arjen Roodselaar, Tobias Rosenkranz, Yernagula Roshit, Ryszard
+Rozak, Huang Rui, Graham Rushton, Jan Egil Ruud, Denis Rystsov, Pawel
+Sagan, Robert Sammelson, John Sanguinetti, Josep Sans, Luca Sasselli,
+Martin Scharrer, Martin Schmidt, Julie Schwartz, Galen Seitz, Joseph
+Shaker, Mark Shaw, Salman Sheikh, Zhou Shen, Hao Shi, James Shi, Michael
+Shinkarovsky, Rafael Shirakawa, Jeffrey Short, S Shuba, Fan Shupei, Ethan
+Sifferman, Anderson Ignacio da Silva, Rodney Sinclair, Ameya Vikram Singh,
+Sanjay Singh, Frans Skarman, Nate Slager, Steven Slatter, Mladen
+Slijepcevic, Brian Small, Garrett Smith, Gus Smith, Tim Snyder, Maciej
 Sobkowski, Stan Sokorac, Alex Solomatnikov, Flavien Solt, Wei Song, Trefor
 Southwell, Martin Stadler, Art Stamness, David Stanford, John Stevenson,
-Pete Stevenson, Patrick Stewart, Rob Stoddard, Todd Strader, John Stroebel,
-Sven Stucki, Howard Su, Emerson Suguimoto, Gene Sullivan, Qingyao Sun,
-Renga Sundararajan, Gustav Svensk, Rupert Swarbrick, Yutetsu Takatsukasa,
-Thierry Tambe, Drew Taussig, Jose Tejada, Peter Tengstrand, Wesley
-Terpstra, Rui Terra, Stefan Thiede, Gary Thomas, Ian Thompson, Kevin
-Thompson, Mike Thyer, Hans Tichelaar, Viktor Tomov, Steve Tong, Topa
-Topino, Alex Torregrosa, Michael Tresidder, David Turner, Neil Turton, Mike
-Urbach, Hans Van Antwerpen, Sebastien Van Cauwenberghe, Laurens van Dam,
-Leendert van Doorn, Cong Van Nguyen, Jan Van Winkel, Srini Vemuri, Yuri
-Victorovich, Bogdan Vukobratovic, Holger Waechtler, Philipp Wagner, Stefan
-Wallentowitz, Shawn Wang, Zhanglei Wang, Paul Wasson, Greg Waters, Thomas
-Watts, Eugene Weber, David Welch, Thomas J Whatson, Martin Whitaker, Marco
-Widmer, Leon Wildman, Daniel Wilkerson, Gerald Williams, Trevor Williams,
-Jeff Winston, Joshua Wise, Clifford Wolf, Tobias Wolfel, Johan Wouters,
-Paul Wright, Junyi Xi, Ding Xiaoliang, Jie Xu, Mandy Xu, Yinan Xu, Luke
-Yang, Amir Yazdanbakhsh, Keyi Zhang, and Xi Zhang.
+Pete Stevenson, Patrick Stewart, Rob Stoddard, Tood Strader, John Stroebel,
+Ray Strouble, Sven Stucki, Howard Su, Emerson Suguimoto, Gene Sullivan,
+Qingyao Sun, Renga Sundararajan, Kuba Sunderland-Ober, Gustav Svensk,
+Rupert Swarbrick, Jevin Sweval, Shinya T-Y, Thierry Tambe, Jesse Taube,
+Drew Taussig, Jose Tejada, Sören Tempel, Peter Tengstrand, Wesley Terpstra,
+Rui Terra, Stefan Thiede, Justin Thiel, Gary Thomas, Ian Thompson, Kevin
+Thompson, Mike Thyer, Hans Tichelaar, Tudor Timi, Viktor Tomov, Steve Tong,
+Topa Topino, Àlex Torregrosa, Topa Tota, Michael Tresidder, Lenny Truong,
+David Turner, Neil Turton, Hideto Ueno, Mike Urbach, Joel Vandergriendt,
+Srini Vemuri, Srinivasan Venkataramanan, Yuri Victorovich, Ivan Vnučec,
+Bogdan Vukobratovic, Holger Waechtler, Philipp Wagner, Johannes Walter, CY
+Wang, Chuxuan Wang, Shawn Wang, Zhanglei Wang, Greg Waters, Thomas Watts,
+Eugene Weber, John Wehle, Tianrui Wei, David Welch, Thomas J Whatson,
+Martin Whitaker, Marco Widmer, Leon Wildman, Daniel S. Wilkerson, Daniel
+Wilkerson, Gerald Williams, Trevor Williams, Don Williamson, Jan Van
+Winkel, Jeff Winston, Joshua Wise, Clifford Wolf, Johan Wouters, Paul
+Wright, Tobias Wölfel, Junyi Xi, Ding Xiaoliang, Liu Xiaoyi, Mandy Xu,
+Shanshan Xu, Yinan Xu, SU YANG, Felix Yan, Luke Yang, Amir Yazdanbakhsh,
+Chentai (Seven) Yuan, Florian Zaruba, Mat Zeno, Keyi Zhang, Xi Zhang, Yike
+Zhou, Jiamin Zhu.

 Thanks to them, and all those we've missed mentioning above, and to those
 whom have wished to remain anonymous.
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 *********
@@ -8,7 +8,7 @@ Copyright
 The latest version of Verilator is available from `https://verilator.org
 <https://verilator.org>`_.

-Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 can redistribute it and/or modify the Verilator internals under the terms
 of either the GNU Lesser General Public License Version 3 or the Perl
 Artistic License Version 2.0.
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 Deprecations
@@ -8,11 +8,8 @@ The following deprecated items are scheduled for future removal:

 C++11 compiler support
  Verilator currently requires a C++20 or newer compiler for timing, and a
-  C++11 or newer compiler for no-timing.
-
-  Verilator will require C++14 or newer compilers for both compiling
-  Verilator and compiling Verilated models with --no-timing no sooner than
-  January 2023.
+  C++14 or newer compiler for both compiling Verilator and compiling
+  Verilated models with --no-timing.

  Verilator will require C++20 or newer compilers for both compiling
  Verilator and compiling all Verilated models no sooner than January 2025.
@@ -21,19 +18,3 @@ C++11 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
-  `-fno-<optimization>` debug options to match GCC. The old options are
-  planned for removal no sooner than June 2023.
-
-Option `--prof-threads`
-  The `--prof-threads` option has been superseded by the `--prof-exec` and
-  `--prof-pgo` options and is planned for removal no sooner than April 2023.
-
-Verilated model options `+verilator+prof+threads+*`
-  The `+verilator+prof+threads+start`, `+verilator+prof+threads+window` and
-  `+verilator+prof+threads+file` options have been superseded by the
-  `+verilator+prof+exec+start`, `+verilator+prof+exec+window` and
-  `+verilator+prof+exec+file` options respectively and are planned for removal
-  no sooner than April 2023.
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 Environment
@@ -29,7 +29,7 @@ associated programs.
 .. option:: OBJCACHE

   Optionally specifies a caching or distribution program to place in front
-   of all runs of the C++ compiler.  For example, "ccache".  If using
+   of all runs of the C++ compiler.  For example, "ccache" or "sccache".  If using
   :command:`distcc` or :command:`icecc`/:command:`icecream`, they would
   generally be run under :command:`ccache`; see the documentation for
   those programs.  If OBJCACHE is not set, and at configure time ccache
@@ -117,3 +117,46 @@ associated programs.
   ``VERILATOR_ROOT`` to point to the new version.

   See :ref:`Installation` for more details.
+
+
+Make Variables
+==============
+
+This section describes the make variables used by Verilator. These may be
+set by passing them to make e.g. ``make CXX=my-gcc ...``.
+
+.. option:: AR
+
+   Optionally overrides the default :command:`ar` (archive) binary used by
+   the Verilated makefiles. If AR is not set, the version found at
+   configure time is used.
+
+.. option:: CXX
+
+   Optionally overrides the default compiler binary used by the Verilated
+   makefiles. If CXX is not set, the version found at configure time is
+   used.  Note the default flags passed to the compiler are determined at
+   configuration time, so changing the CXX compiler version using this
+   variable, as opposed to passing it at configuration time, may not give
+   desired results.
+
+.. option:: LINK
+
+   Optionally overrides the default linker binary used by the Verilated
+   makefiles. If LINK is not set, the version found at configure time is
+   used.  Note the default flags passed to the linker are determined at
+   configuration time, so changing the LINK version using this variable, as
+   opposed to passing it at configuration time, may not give desired
+   results.
+
+.. option:: PERL
+
+   Optionally overrides the default :command:`perl` binary used by the
+   Verilated makefiles. If PERL is not set, the version found at configure
+   time, and compiled into the Verilator binary, is used.
+
+.. option:: PYTHON3
+
+   Optionally overrides the default :command:`python3` binary used by the
+   Verilated makefiles. If PYTHON3 is not set, the version found at
+   configure time is used.
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Example Create-Binary Execution:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Example C++ Execution:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 First you need Verilator installed, see :ref:`Installation`.  In brief, if
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Examples in the Distribution:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Example SystemC Execution:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Examples:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Simulation Runtime Arguments:
@@ -64,15 +64,18 @@ Summary:

 .. option:: +verilator+prof+threads+file+<filename>

-   Deprecated. Alias for :vlopt:`+verilator+prof+exec+file+\<filename\>`
+   Removed in 5.020. Was an alias for
+   :vlopt:`+verilator+prof+exec+file+\<filename\>`

 .. option:: +verilator+prof+threads+start+<value>

-   Deprecated. Alias for :vlopt:`+verilator+prof+exec+start+\<value\>`
+   Removed in 5.020. Was an alias for
+   :vlopt:`+verilator+prof+exec+start+\<value\>`

 .. option:: +verilator+prof+threads+window+<value>

-   Deprecated. Alias for :vlopt:`+verilator+prof+exec+window+\<value\>`
+   Removed in 5.020. Was an alias for
+   :vlopt:`+verilator+prof+exec+window+\<value\>`

 .. option:: +verilator+prof+vlt+file+<filename>

@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 verilator Arguments
@@ -1036,7 +1036,7 @@ Summary:

 .. option:: --prof-threads

-   Deprecated. Same as --prof-exec and --prof-pgo together.
+   Removed in 5.020. Was an alias for --prof-exec and --prof-pgo together.

 .. option:: --protect-ids

@@ -1363,13 +1363,13 @@ Summary:
   Enable FST waveform tracing in the model. This overrides
   :vlopt:`--trace`.  See also :vlopt:`--trace-threads` option.

-.. option:: --trace-max-array *depth*
+.. option:: --trace-max-array <depth>

   Rarely needed.  Specify the maximum array depth of a signal that may be
   traced.  Defaults to 32, as tracing large arrays may greatly slow traced
   simulations.

-.. option:: --trace-max-width *width*
+.. option:: --trace-max-width <width>

   Rarely needed.  Specify the maximum bit width of a signal that may be
   traced.  Defaults to 256, as tracing large vectors may greatly slow
@@ -1386,7 +1386,7 @@ Summary:
   format constraints, this may result in significantly slower trace times
   and larger trace files.

-.. option:: --trace-threads *threads*
+.. option:: --trace-threads <threads>

   Enable waveform tracing using separate threads. This is typically faster
   in simulation runtime but uses more total compute. This option only
@@ -1427,12 +1427,12 @@ Summary:
   Rarely needed.  Specifies the maximum number of loop iterations that may be
   unrolled.  See also :option:`BLKLOOPINIT` warning.

-.. option:: --unroll-stmts *statements*
+.. option:: --unroll-stmts <statements>

   Rarely needed.  Specifies the maximum number of statements in a loop for
   that loop to be unrolled. See also :option:`BLKLOOPINIT` warning.

-.. option:: --unused-regexp *regexp*
+.. option:: --unused-regexp <regexp>

   Rarely needed.  Specifies a simple regexp with \* and ? that, if a signal
   name matches, will suppress the :option:`UNUSED` warning.  Defaults to
@@ -1444,7 +1444,7 @@ Summary:
   into Verilator.  (Similar to :command:`perl -V`.)  See also
   :vlopt:`--getenv` option.

-.. option:: -v *filename*
+.. option:: -v <filename>

   Read the filename as a Verilog library.  Any modules in the file may be
   used to resolve instances in the top-level module, otherwise, they are
@@ -1481,7 +1481,7 @@ Summary:

   Enable the use of VPI and linking against the :file:`verilated_vpi.cpp` files.

-.. option:: --waiver-output *filename*
+.. option:: --waiver-output <filename>

   Generate a waiver file that contains all waiver statements to suppress
   the warnings emitted during this Verilator run. This, in particular, is
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 verilator_coverage
@@ -64,30 +64,59 @@ if there are multiple coverage points on a single line they will merge.
 The minimum coverage across all merged points will be used to report
 coverage of the line.

-The first character of the line shows a summary of the coverage; this
-allows use of grep to filter the report.  `%` indicates at least one point
-on the line was below the coverage limit.  `+` indicates an
-:option:`--annotate-points` point was at or above the limit, and `-` that
-the point was below the limit.
+Coverage data is annotated at the beginning of the line and is formatted
+as a special character followed by the number of coverage hits. The special
+characters " ,%,+,-" indicate summary of the coverage, and allow use of grep
+to filter the report.
+
+* " " (whitespace) indicates that all points on the line are above the coverage limit.
+* "%" indicates at least one point on the line was below the coverage limit.
+* "+" coverage point was at or above the limit. Only used with :option:`--annotate-points`.
+* "-" coverage point was below the limit.  Only used with :option:`--annotate-points`.
+
+.. code-block::
+
+   100000  input logic a;   // Begins with whitespace, because
+                            // number of hits (100000) is above the limit.
+  +100000  point: comment=a // Begins with +, because
+                            // number of hits (100000) is above the limit.
+  %000000  input logic b;   // Begins with %, because
+                            // number of hits (0) is below the limit.
+  -000000  point: comment=b // Begins with -, because
+                            // number of hits (0) is below the limit.

 .. option:: --annotate-all

 Specifies all files should be shown.  By default, only those source files
 with low coverage are written to the output directory.

+This option should be used together with :option:`--annotate`.
+
 .. option:: --annotate-min <count>

-Specifies if the coverage point does not include the count number of
-coverage hits, then the coverage point will be considered above the
-threshold, and the coverage report will put a "%" to indicate the coverage
-is insufficient.  Defaults to 10.
+Specifies the threshold (<count>) below which coverage point is considered
+sufficient. If the threshold is not exceeded, then the annotation will begin
+with a "%" symbol to indicate the coverage is insufficient.
+
+The <count> threshold defaults to 10.
+
+This option should be used together with :option:`--annotate`.
+

 .. option:: --annotate-points

 Specifies all coverage points should be shown after each line of text.  By
 default, only source lines are shown.

-with low coverage are written to the output directory.
+.. code-block::
+
+  100000  input logic a, b, c;
+ +100000 point: comment=a // These lines are only shown
+ +200000 point: comment=b // with option --annotate-points
+ +300000 point: comment=c // enabled.
+
+
+This option should be used together with :option:`--annotate`.

 .. option:: --help

@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 verilator_gantt
@@ -30,15 +30,13 @@ Gantt Chart VCD Signals
 In waveforms, there are the following signals. In GTKWave, use "decimal"
 data format to remove the leading zeros and make the traces easier to read.

-evals
-  Increments each time when eval_step was measured to be active.  This
-  allow visualization of how much time eval_step was active.
+trace/section
+  Shows the name of the current top of the execution section stack.
+  Set GTKWave data format to "ASCII".

-eval_loop
-  Increments each time when the evaluation loop within eval_step was
-  measured to be active.  For best performance, there is only a single
-  evaluation loop within each eval_step call; that is, the eval_loop
-  waveform looks identical to the evals waveform.
+trace/depth
+  Shows the depth of the execution section stack.
+  Set GTKWave data format to "Analog".

 measured_parallelism
  The number of mtasks active at this time, for best performance, this will
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 verilator_profcfunc
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 *********************************
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 *******************
@@ -17,13 +17,21 @@ or "`ifdef`"'s may break other tools.

 .. option:: `__LINE__

-   The :option:`\`__LINE__` define expands to the current filename as a
-   string, like C++'s __LINE__.  This Verilator feature added in 2006 was
-   incorporated into IEEE 1800-2009.
+   The :option:`\`__LINE__` define expands to the current line number like
+   C++'s __LINE__.  This Verilator feature added in 2006 was incorporated
+   into IEEE 1800-2009.

 .. option:: `error [string]

-   This will report an error when encountered, like C++'s #error.
+   This will report an error when the preprocessor emits it, similar to
+   C++'s #error.
+
+.. option:: `line
+
+   As a special case `\`line \`__LINE__ "filename"` allows setting the
+   filename, without changing the line number. This is used for some
+   internal tests, so that debugging can leave the line numbers correctly
+   referring to the test file's line numbers.

 .. option:: """ [string] """

@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 ******************************
@@ -220,12 +220,6 @@ This is done for you if you are using the Verilator :vlopt:`--binary` or
 You can call :code:`->trace()` on multiple Verilated objects with the same
 trace file if you want all data to land in the same output file.

-When using SystemC 2.3, the SystemC library must have been built with the
-experimental simulation phase callback-based tracing disabled. This is
-disabled by default when building SystemC with its configure based build
-system, but when building SystemC with CMake, you must pass
-``-DENABLE_PHASE_CALLBACKS_TRACING=OFF`` to disable this feature.
-

 How do I generate FST waveforms (traces) in C++ or SystemC?
 """""""""""""""""""""""""""""""""""""""""""""""""""""""""""
@@ -401,7 +395,7 @@ How do I get faster build times?
 * Use a recent compiler.  Newer compilers tend to be faster.

 * Compile in parallel on many machines and use caching; see the web for the
-  ccache, distcc, and icecream packages. ccache will skip GCC runs between
+  ccache, sccache, distcc, or icecream packages. ccache will skip GCC runs between
  identical source builds, even across different users.  If ccache was
  installed when Verilator was built, it is used, or see OBJCACHE
  environment variable to override this. Also see the
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 *****
@@ -50,9 +50,9 @@ For --cc/--sc, it creates:
     - Make include file with class names (from --make gmake)
   * - *{prefix}*\ _hier.mk
     - Make file for hierarchy blocks (from --make gmake)
-   * - *{prefix}*\ _hierMkArgs.f
+   * - *{prefix}*\ __hierMkArgs.f
     - Arguments for hierarchical Verilation (from --make gmake)
-   * - *{prefix}*\ _hierCMakeArgs.f
+   * - *{prefix}*\ __hierCMakeArgs.f
     - Arguments for hierarchical Verilation (from --make cmake)
   * - *{prefix}*\ .h
     - Model header
@@ -145,6 +145,10 @@ After running Make, the C++ compiler may produce the following:
     - Intermediate objects
   * - *{prefix}*\
     - Final executable (from --exe)
+   * - lib\ *{prefix}*\ .a
+     - Final archive (default lib mode)
+   * - libverilated.a
+     - Runtime for verilated model (default lib mode)
   * - *{prefix}*\ __ALL.a
     - Library of all Verilated objects
   * - *{prefix}*\ __ALL.cpp
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 ######################
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _CMakeInstallation:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Installation:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 ***************
@@ -237,11 +237,9 @@ optimizations will be disabled around the latch.
 Structures and Unions
 ---------------------

-Presently Verilator only supports packed structs and packed unions.  Rand
-and randc tags on members are ignored.  All structures and unions are
-represented as a single vector, which means that generating one member of a
-structure from blocking, and another from non-blocking assignments is
-unsupported.
+All structures and unions are scheduled together, which means that
+generating one member of a structure from blocking, and another from
+non-blocking assignments is unsupported.


 .. _Unknown States:
@@ -320,20 +318,6 @@ there will be considered a two-state variable that is read and written
 instead of a four-state variable.


-Functions & Tasks
-----------------
-
-All functions and tasks will be inlined (will not become functions in C.)
-The only support provided is simple statements in tasks (which may
-affect global variables).
-
-Recursive functions and tasks are not supported.  All inputs and outputs
-are automatic as if they had the Verilog 2001 "automatic" keyword
-prepended.  (If you don't know what this means, Verilator will do what you
-probably expect, what C does. The default behavior of Verilog is
-different.)
-
-
 Gate Primitives
 ---------------

@@ -383,24 +367,6 @@ Verilator does not support SEREs yet.  All assertion and coverage
 statements must be simple expressions that complete in one cycle.


-Force statement
---------------
-
-Verilator supports the procedural `force` (and corresponding `release`)
-statement. However, the behavior of the `force` statement does not entirely
-comply with IEEE 1800. According to the standard,
-when a procedural statement of the form `force a = b;` is executed, the
-simulation should behave as if, from that point forwards, a continuous
-assignment `assign a = b;` has been added to override the drivers of `a`.
-More specifically: the value of `a` should be updated whenever the value of
-`b` changes, until a `release a;` statement is executed.
-Verilator instead evaluates the current value of `b` when the `force`
-statement is executed, and forces `a` to that value, without updating it
-until a new `force` or `release` statement is encountered that applies to
-`a`. This non-standard behavior is nevertheless consistent with some other
-simulators.
-
-
 Encrypted Verilog
 -----------------

@@ -471,6 +437,21 @@ disable
  provide loop break and continue functionality before SystemVerilog added
  the break and continue keywords.

+force, release
+  Verilator supports the procedural `force` (and corresponding `release`)
+  statement. However, the behavior of the `force` statement does not
+  entirely comply with IEEE 1800. According to the standard, when a
+  procedural statement of the form `force a = b;` is executed, the
+  simulation should behave as if, from that point forwards, a continuous
+  assignment `assign a = b;` has been added to override the drivers of `a`.
+  More specifically: the value of `a` should be updated whenever the value
+  of `b` changes, until a `release a;` statement is executed.  Verilator
+  instead evaluates the current value of `b` when the `force` statement is
+  executed, and forces `a` to that value, without updating it until a new
+  `force` or `release` statement is encountered that applies to `a`. This
+  non-standard behavior is nevertheless consistent with some other
+  simulators.
+
 inside
  Inside expressions may not include unpacked array traversal or $ as an
  upper bound.  Case inside and case matches are also unsupported.
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 ********
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 .. _Simulating:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 **********
@@ -7,9 +7,13 @@ Verilating

 Verilator may be used in five major ways:

+* With the :vlopt:`--binary` option, Verilator will translate the design
+  into an executable, via generating C++ and compiling it.  See
+  :ref:`Binary, C++ and SystemC Generation`.
+
 * With the :vlopt:`--cc` or :vlopt:`--sc` options, Verilator will translate
-  the design into C++ or SystemC code, respectively.  See :ref:`C++ and
-  SystemC Generation`.
+  the design into C++ or SystemC code, respectively.  See :ref:`Binary, C++
+  and SystemC Generation`.

 * With the :vlopt:`--lint-only` option, Verilator will lint the design to
  check for warnings but will not typically create any output files.
@@ -24,13 +28,15 @@ Verilator may be used in five major ways:
  expanded.


-.. _C++ and SystemC Generation:
+.. _Binary, C++ and SystemC Generation:

-C++ and SystemC Generation
-==========================
+Binary, C++ and SystemC Generation
+==================================

 Verilator will translate a SystemVerilog design into C++ with the
-:vlopt:`--cc` option, or into SystemC with the :vlopt:`--sc` option.
+:vlopt:`--cc` option, or into SystemC with the :vlopt:`--sc` option.  It
+will translate into C++ and compile it into an executable binary with the
+:vlopt:`--binary` option.

 When using these options:

@@ -45,14 +51,19 @@ When using these options:
   The prefix is set with :vlopt:`--prefix`, or defaults to the name of the
   top module.

-#. If :vlopt:`--exe` is used, Verilator creates makefiles to generate a
-   simulation executable, otherwise, it creates makefiles to generate an
-   archive (.a) containing the objects.
+#. If :vlopt:`--binary` or :vlopt:`--main` is used, Verilator creates a C++
+   top wrapper to read command line arguments, create the model, and
+   execute the model.

-#. If :vlopt:`--build` option was used, it calls :ref:`GNU Make` or
-   :ref:`CMake` to build the model.
+#. If :vlopt:`--binary` or :vlopt:`--exe` is used, Verilator creates
+   makefiles to generate a simulation executable, otherwise, it creates
+   makefiles to generate an archive (.a) containing the objects.

-Once a model is built, it is then typically run, see :ref:`Simulating`.
+#. If :vlopt:`--binary` or :vlopt:`--build` is used, it calls :ref:`GNU
+   Make` or :ref:`CMake` to build the model.
+
+Once a model is built, the next step is typically for the user to run it,
+see :ref:`Simulating`.


 .. _Hierarchical Verilation:
@@ -1,4 +1,4 @@
-.. Copyright 2003-2023 by Wilson Snyder.
+.. Copyright 2003-2024 by Wilson Snyder.
 .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0

 *******************
@@ -12,10 +12,18 @@ Disabling Warnings

 Warnings may be disabled in multiple ways:

-#. Disable the warning in the source code.  When the warning is printed, it
-   will include a warning code.  Surround the offending line with a
-   :code:`/*verilator&32;lint_off*/` and :code:`/*verilator&32;lint_on*/`
-   metacomment pair:
+#. Disable the warning globally by invoking Verilator with the
+   :code:`-Wno-{warning-code}` option.
+
+   Global disables should be avoided, as they removes all checking across
+   the source files, and prevents other users from compiling the sources
+   without knowing the magic set of disables needed to compile those
+   sources successfully.
+
+#. Disable the warning in the design source code.  When the warning is
+   printed, it will include a warning code.  Surround the offending line
+   with a :code:`/*verilator&32;lint_off*/` and
+   :code:`/*verilator&32;lint_on*/` metacomment pair:

   .. code-block:: sv

@@ -23,6 +31,12 @@ Warnings may be disabled in multiple ways:
         if (`DEF_THAT_IS_EQ_ZERO <= 3) $stop;
         // verilator lint_on UNSIGNED

+
+   A lint_off in the design source code will propagate down to any child
+   files (files later included by the file with the lint_off), but will not
+   propagate upwards to any parent file (file that included the file with
+   the lint_off).
+
 #. Disable the warning using :ref:`Configuration Files` with a
   :option:`lint_off` command.  This is useful when a script suppresses
   warnings, and the Verilog source should not be changed.  This method also
@@ -32,12 +46,6 @@ Warnings may be disabled in multiple ways:

         lint_off -rule UNSIGNED -file "*/example.v" -line 1

-#. Disable the warning globally invoking Verilator with the
-   :code:`-Wno-{warning-code}` option.  This should be avoided, as it
-   removes all checking across the designs, and prevents other users from
-   compiling your code without knowing the magic set of disables needed to
-   compile your design successfully.
-

 Error And Warning Format
 ========================
@@ -1166,7 +1174,7 @@ List Of Warnings

   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.
+   unbased literals (`'0`, `'1`, `'z`, `'x`) require IEEE 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
@@ -1360,7 +1368,10 @@ List Of Warnings

 .. option:: RANDC

-   Warns that the :code:`randc` keyword is unsupported and being converted
+   Historical, never issued since version 5.018, when :code:`randc` became
+   fully supported.
+
+   Warned that the :code:`randc` keyword was unsupported and was converted
   to :code:`rand`.


@@ -1503,6 +1514,39 @@ List Of Warnings
   modeled values, or DPI calls.


+.. option:: SIDEEFFECT
+
+   Warns that an expression has a side effect that might not properly be
+   executed by Verilator.
+
+   This often represents a bug in Verilator, as opposed to a bad code
+   construct, however the Verilog code can typically be changed to avoid
+   the warning.
+
+   Faulty example:
+
+   .. code-block:: sv
+      :linenos:
+      :emphasize-lines: 1
+
+         x = y[a++];
+
+   This example warns because Verilator does not currently handle side
+   effects inside array subscripts; the a++ may be executed multiple times.
+
+   Rewrite the code to avoid expression side effects, typically by using a
+   temporary:
+
+   .. code-block:: sv
+      :linenos:
+
+         temp = a++;
+         x = y[temp];
+
+   Ignoring this warning may make Verilator simulations differ from other
+   simulators.
+
+
 .. option:: SPLITVAR

   Warns that a variable with a :option:`/*verilator&32;split_var*/`
@@ -2043,17 +2087,18 @@ List Of Warnings

   .. include:: ../../docs/gen/ex_WIDTHEXPAND_1_fixed.rst

+
 .. option:: WIDTHTRUNC

-   A more granular WIDTH warning, for when a value is truncated
+   A more granular WIDTH warning, for when a value is truncated.

 .. option:: WIDTHEXPAND

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

 .. option:: WIDTHXZEXPAND

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

 .. option:: WIDTHCONCAT

@@ -1035,11 +1035,11 @@ Coding Conventions
 ==================


-Compiler Version and C++11
+Compiler Version and C++14
 --------------------------

-Verilator requires C11. Verilator does not require any newer versions, but
-is maintained to build successfully with C14/C17/C20.
+Verilator requires C14. Verilator does not require any newer versions, but
+is maintained to build successfully with C17/C20.


 Indentation and Naming Style
@@ -1222,9 +1222,9 @@ There are three ways data is passed between visitor functions.

 2. User attributes. Each ``AstNode`` (**Note.** The AST node, not the
   visitor) has five user attributes, which may be accessed as an
-   integer using the ``user1()`` through ``user5()`` methods, or as a
+   integer using the ``user1()`` through ``user4()`` methods, or as a
   pointer (of type ``AstNUser``) using the ``user1p()`` through
-   ``user5p()`` methods (a common technique lifted from graph traversal
+   ``user4p()`` methods (a common technique lifted from graph traversal
   packages).

   A visitor first clears the one it wants to use by calling
@@ -1400,7 +1400,7 @@ Controlling the Test Driver
 ---------------------------

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


@@ -1432,7 +1432,7 @@ Developers will also want to call ./configure with two extra flags:
   disabled by default, as SystemC installation problems would otherwise
   falsely indicate a Verilator problem.

-When enabling the long tests, some additional PERL modules are needed,
+When enabling the long tests, some additional Perl modules are needed,
 which you can install using cpan.

 ::
@@ -1462,6 +1462,19 @@ There are some traps to avoid when running regression tests
  system-dependent, but on Fedora Linux it would require editing the
  ``/etc/security/limits.conf`` file as root.

+Diffing generated code after changes
+------------------------------------
+
+When making a change in the code generation area that should not change the
+actual emitted code, it is useful to perform a diff to make sure the emitted
+code really did not change. To do this, the top level Makefile provides the
+*test-snap* and *test-diff* targets:
+
+- Run the test suite with ``make test``
+- Take a snapshot with ``make test-snap``
+- Apply your changes
+- Run the test suite again with ``make test``
+- See the changes in the output with ``make test-diff``

 Continuous Integration
 ----------------------
@@ -1640,7 +1653,7 @@ Source file and line
   file is ``a``, the 26th is ``z``, the 27th is ``aa``, and so on.

 User pointers
-   Shows the value of the node's user1p...user5p, if non-NULL.
+   Shows the value of the node's user1p...user4p, if non-NULL.

 Data type
   Many nodes have an explicit data type. "@dt=0x..." indicates the
@@ -2224,7 +2237,7 @@ xsim_flags / xsim_flags2 / xsim_run_flags
 Distribution
 ============

-Copyright 2008-2023 by Wilson Snyder. Verilator is free software; you can
+Copyright 2008-2024 by Wilson Snyder. Verilator is free software; you can
 redistribute it and/or modify it under the terms of either the GNU Lesser
 General Public License Version 3 or the Perl Artistic License Version 2.0.

@@ -1,4 +1,5 @@
 ABCp
+Aadi
 Accellera
 Affe
 Aleksander
@@ -13,21 +14,26 @@ Antonin
 Antwerpen
 Arasanipalai
 Arjen
+Arshid
 Asciidoc
 Ashutosh
 Ast
 Atmel
 Aurelien
 Bagri
+Balboni
 Baltazar
 Bardsley
 Benchmarking
 Berman
 Bhagavatula
 Bhattacharyya
+Biancolin
 Binderman
+Björk
 Bleiweiss
 Bogdan
+Bonneau
 Bouchard
 Boumenot
 Brej
@@ -35,24 +41,31 @@ Briquet
 Brownlee
 Buckenmaier
 Bybell
+CLang
 CMake
 Cai
 Cavium
 Cfuncs
 Chandan
+Chentai
+Cherkasov
+Chih
 Chitlesh
 Christophe
 Cochrane
 Conda
 Corteggiani
+Cozzocrea
 Cuan
 Cygwin
 DErrico
 DFG
 Da
+Dai
 Danilo
 Dannoritzer
 Das
+Davide
 Dcache
 Deadman
 Debacker
@@ -61,6 +74,7 @@ Defparams
 Denio
 Deprecations
 Deroo
+Desai
 Dickol
 Diez
 Dimitris
@@ -69,9 +83,11 @@ Djordjevic
 Dobbie
 Dockerfile
 Donal
+Donovick
 Doorn
 Doxygen
 Dressler
+Dudek
 Duraid
 Dutton
 Dzetkulic
@@ -89,14 +105,20 @@ Engstrom
 Enzo
 Eugen
 Fabrizio
+Faucher
+Faure
 Fekete
 Ferrandi
 Flachs
 Flavien
+Florian
 Foletto
+Forencich
 Forker
 Francillon
 Fredieu
+Freiberger
+Frédéric
 GTKWave
 Galbi
 Gantt
@@ -116,18 +138,23 @@ Goessling
 Gonnen
 Goorah
 Gossner
+Graba
 Graphviz
 Graybeal
 Grobman
 Grulfen
+Gu
 Gunter
 Guo
 Hao
+Haojin
 Harboe
 Hashimoto
 Hayenga
+Hesselbarth
 Hewson
 Hicamp
+Hideto
 Hiroki
 Holdsworth
 Holme
@@ -135,8 +162,10 @@ Homebrew
 Hornung
 Hossell
 Hsu
+Hutt
 Hyperthreading
 Ibrahim
+Ikram
 Iles
 Inlines
 Inout
@@ -148,6 +177,7 @@ Jae
 Jalloq
 Jankowski
 Jannis
+Jasen
 Jens
 Jeras
 Jiuyang
@@ -163,6 +193,7 @@ Kahn
 Kai
 Kamendje
 Kandadi
+Kaneda
 Kaplan
 Karge
 Karlsson
@@ -178,6 +209,7 @@ Klnv
 Koenig
 Koh
 Kolecki
+Konstantin
 Koonce
 Korteland
 Koszek
@@ -189,17 +221,23 @@ Kumashikar
 Kuoping
 Kurachi
 Laeufer
+Laroche
 Laurens
 Lavino
 Leber
 Leendert
 Lem
 Lesik
+Libenzi
+Licker
 Liland
 Liu
+Liwei
 Lockhart
 Longo
+Luca
 Lussier
+Lübeck
 MMD
 MODDUP
 MTasks
@@ -209,10 +247,15 @@ Mahesh
 Mahmoudy
 Makefile
 Makefiles
+Maksymenko
+Markley
+Marquet
 Matveyenko
 Maupin
 Mdir
 Mednick
+Mei
+Melo
 Menküc
 Michail
 Michiels
@@ -230,9 +273,11 @@ Mong
 Muhlestein
 Multithreaded
 Multithreading
+Mykyta
 NOUNOPTFLAT
 NaN
 Nalbantis
+Nandor
 Narayan
 Nassim
 Nauticus
@@ -243,16 +288,20 @@ Niranjan
 Nitza
 Noack
 Nodine
+Ober
 Oleg
 Olof
 Olofsson
+Oron
 Oyvind
 PLI
 Palaniappan
 Patricio
 Petr
+Piard
 Piechotka
 Piersall
+Platz
 Platzer
 Plunkett
 Popolon
@@ -269,6 +318,8 @@ Priyadharshini
 Pullup
 Pulver
 Puri
+Qin
+Qiu
 Questa
 Rachit
 Ralf
@@ -278,28 +329,37 @@ Reitan
 Renga
 Requin
 Rodionov
+Rohan
 Rolfe
 Rontionov
 Roodselaar
+Roshit
 Runtime
 Ruud
 Rystsov
 STandarD
 Salman
+Sammelson
+Sanggyu
 Sanguinetti
 Sanjay
+Sasselli
 Scharrer
 Seitz
+Shahid
 Shankar
 Shanshan
 Sharad
 Shareef
+Shen
 Sheng
 Shi
 Shinkarovsky
+Shinya
 Shirakawa
 Shuba
 Shunyao
+Slager
 Slatter
 SoC
 Sobhan
@@ -319,6 +379,7 @@ Stucki
 Su
 Suguimoto
 Sundararajan
+Sunderland
 Suse
 Syms
 Synopsys
@@ -326,6 +387,7 @@ SystemC
 SystemVerilog
 Takatsukasa
 Tambe
+Tarik
 Tariq
 Tejada
 Tengstrand
@@ -333,6 +395,7 @@ Terpstra
 Thiede
 Thierry
 Thyer
+Tianrui
 Tichelaar
 Timi
 Tomov
@@ -343,9 +406,11 @@ Trefor
 Tresidder
 Tri
 Tristate
+Truong
 Turton
 UNOPTFLAT
 Ubixum
+Ueno
 Uints
 Undefines
 Unsized
@@ -354,6 +419,7 @@ Uselib
 Usha
 Usuario
 VERILATOR
+Vandergriendt
 Vasu
 Vdeeptemp
 Vdly
@@ -391,14 +457,20 @@ Wwarn
 XSim
 Xcelium
 Xiaoliang
+Xiaoyi
 Xuan
 Xuanqi
 Yazdanbakhsh
+Yernagula
 Yi
 Yike
 Yinan
+Yoshitomo
 Yosys
 Yu
+Yujia
+Yurii
+Zaruba
 Zhang
 abirkmanis
 accessor
@@ -486,6 +558,7 @@ countbits
 countones
 cout
 cpp
+cppstyle
 cpu
 ctor
 ctrace
@@ -512,6 +585,7 @@ desassign
 destructor
 detections
 dev
+devcontainer
 devel
 dir
 displayb
@@ -520,6 +594,7 @@ doxygen
 dpiGetContext
 dpic
 dsvf
+dt
 dtor
 dumpall
 dumpfile
@@ -669,6 +744,7 @@ libext
 libgoogle
 libsystemc
 libtcmalloc
+libverilated
 linkers
 linter
 linux
@@ -697,6 +773,7 @@ miree
 mis
 misconnected
 misconversion
+misoptimized
 missized
 mk
 mno
@@ -721,6 +798,7 @@ multithreading
 musl
 mutexes
 mux
+muzafferkal
 myftptoyman
 mysignal
 namespace
@@ -759,6 +837,7 @@ pdf
 perf
 perftools
 pgo
+phelter
 picoChip
 pinIndex
 pinout
@@ -776,6 +855,7 @@ pragma
 pragmas
 pre
 precisions
+precompiled
 predefines
 prepareClone
 prepend
@@ -801,10 +881,12 @@ pvalue
 pwd
 py
 qrq
+qrqiuren
 radix
 randc
 randcase
 randstate
+raphmaster
 rarr
 rdtsc
 reStructuredText
@@ -814,6 +896,7 @@ readmemb
 readmemh
 realpath
 realtime
+realtobits
 recoding
 recrem
 redeclaring
@@ -832,6 +915,7 @@ sVerilator
 sawatzke
 sc
 scalared
+sccache
 sccanf
 seg
 setuphold
@@ -954,7 +1038,9 @@ vlt
 vltstd
 vluint
 vpi
+vpiConstType
 vpiDefName
+vpiInstance
 vpiLeftRange
 vpiModule
 vpiSize
@@ -984,3 +1070,5 @@ ypq
 yurivict
 zdave
 Øyvind
+Алексеевич
+Исаак
@@ -70,7 +70,7 @@ The XML document consists of 4 sections within the top level
 Distribution
 ============

-Copyright 2020-2023 by Wilson Snyder. Verilator is free software; you can
+Copyright 2020-2024 by Wilson Snyder. Verilator is free software; you can
 redistribute it and/or modify it under the terms of either the GNU Lesser
 General Public License Version 3 or the Perl Artistic License Version 2.0.

@@ -24,6 +24,7 @@
 #include "Vtop.h"

 using namespace sc_core;
+using namespace sc_dt;

 int sc_main(int argc, char* argv[]) {
    // This is a more complicated example, please also see the simpler examples/make_hello_c.
@@ -52,7 +53,7 @@ int sc_main(int argc, char* argv[]) {
    Verilated::commandArgs(argc, argv);

    // General logfile
-    ios::sync_with_stdio();
+    std::ios::sync_with_stdio();

    // Define clocks
    sc_clock clk{"clk", 10, SC_NS, 0.5, 3, SC_NS, true};
@@ -60,8 +60,8 @@ class VlFileCopy:
            print("\t%s " % command)
        status = subprocess.call(command, shell=True)
        if status != 0:
-            raise Exception("Command failed running Verilator with '" +
-                            command + "', stopped")
+            raise RuntimeError("Command failed running Verilator with '" +
+                               command + "', stopped")


 #######################################################################
@@ -83,8 +83,8 @@ class VlHierGraph:
            print("\t%s " % command)
        status = subprocess.call(command, shell=True)
        if status != 0:
-            raise Exception("Command failed running Verilator with '" +
-                            command + "', stopped")
+            raise RuntimeError("Command failed running Verilator with '" +
+                               command + "', stopped")


 #######################################################################
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -107,16 +107,9 @@ thread_local Verilated::ThreadLocal Verilated::t_s;

 #ifndef VL_USER_FINISH  ///< Define this to override the vl_finish function
 void vl_finish(const char* filename, int linenum, const char* hier) VL_MT_UNSAFE {
-    if (false && hier) {}
+    if (false && hier) {}  // Unused argument
    VL_PRINTF(  // Not VL_PRINTF_MT, already on main thread
        "- %s:%d: Verilog $finish\n", filename, linenum);
-    if (Verilated::threadContextp()->gotFinish()) {
-        VL_PRINTF(  // Not VL_PRINTF_MT, already on main thread
-            "- %s:%d: Second verilog $finish, exiting\n", filename, linenum);
-        Verilated::runFlushCallbacks();
-        Verilated::runExitCallbacks();
-        std::exit(0);
-    }
    Verilated::threadContextp()->gotFinish(true);
 }
 #endif
@@ -451,6 +444,8 @@ WDataOutP _vl_moddiv_w(int lbits, WDataOutP owp, const WDataInP lwp, const WData

    const int uw = VL_WORDS_I(umsbp1);  // aka "m" in the algorithm
    const int vw = VL_WORDS_I(vmsbp1);  // aka "n" in the algorithm
+    VL_DEBUG_IFDEF(assert(uw <= VL_MULS_MAX_WORDS););
+    VL_DEBUG_IFDEF(assert(vw <= VL_MULS_MAX_WORDS););

    if (vw == 1) {  // Single divisor word breaks rest of algorithm
        uint64_t k = 0;
@@ -472,27 +467,25 @@ WDataOutP _vl_moddiv_w(int lbits, WDataOutP owp, const WDataInP lwp, const WData

    // Zero for ease of debugging and to save having to zero for shifts
    // Note +1 as loop will use extra word
-    for (int i = 0; i < words + 1; ++i) { un[i] = vn[i] = 0; }
+    for (int i = 0; i < words + 1; ++i) un[i] = vn[i] = 0;

    // Algorithm requires divisor MSB to be set
    // Copy and shift to normalize divisor so MSB of vn[vw-1] is set
    const int s = 31 - VL_BITBIT_I(vmsbp1 - 1);  // shift amount (0...31)
-    const uint32_t shift_mask = s ? 0xffffffff : 0;  // otherwise >> 32 won't mask the value
-    for (int i = vw - 1; i > 0; --i) {
-        vn[i] = (rwp[i] << s) | (shift_mask & (rwp[i - 1] >> (32 - s)));
-    }
-    vn[0] = rwp[0] << s;
-
    // Copy and shift dividend by same amount; may set new upper word
    if (s) {
+        for (int i = vw - 1; i > 0; --i) vn[i] = (rwp[i] << s) | (rwp[i - 1] >> (32 - s));
+        vn[0] = rwp[0] << s;
        un[uw] = lwp[uw - 1] >> (32 - s);
+        for (int i = uw - 1; i > 0; --i) un[i] = (lwp[i] << s) | (lwp[i - 1] >> (32 - s));
+        un[0] = lwp[0] << s;
    } else {
+        for (int i = vw - 1; i > 0; --i) vn[i] = rwp[i];
+        vn[0] = rwp[0];
        un[uw] = 0;
+        for (int i = uw - 1; i > 0; --i) un[i] = lwp[i];
+        un[0] = lwp[0];
    }
-    for (int i = uw - 1; i > 0; --i) {
-        un[i] = (lwp[i] << s) | (shift_mask & (lwp[i - 1] >> (32 - s)));
-    }
-    un[0] = lwp[0] << s;

    // Main loop
    for (int j = uw - vw; j >= 0; --j) {
@@ -536,8 +529,10 @@ WDataOutP _vl_moddiv_w(int lbits, WDataOutP owp, const WDataInP lwp, const WData

    if (is_modulus) {  // modulus
        // Need to reverse normalization on copy to output
-        for (int i = 0; i < vw; ++i) {
-            owp[i] = (un[i] >> s) | (shift_mask & (un[i + 1] << (32 - s)));
+        if (s) {
+            for (int i = 0; i < vw; ++i) owp[i] = (un[i] >> s) | (un[i + 1] << (32 - s));
+        } else {
+            for (int i = 0; i < vw; ++i) owp[i] = un[i];
        }
        for (int i = vw; i < words; ++i) owp[i] = 0;
        return owp;
@@ -549,6 +544,8 @@ WDataOutP _vl_moddiv_w(int lbits, WDataOutP owp, const WDataInP lwp, const WData
 WDataOutP VL_POW_WWW(int obits, int, int rbits, WDataOutP owp, const WDataInP lwp,
                     const WDataInP rwp) VL_MT_SAFE {
    // obits==lbits, rbits can be different
+    const int owords = VL_WORDS_I(obits);
+    VL_DEBUG_IFDEF(assert(owords <= VL_MULS_MAX_WORDS););
    owp[0] = 1;
    for (int i = 1; i < VL_WORDS_I(obits); i++) owp[i] = 0;
    // cppcheck-has-bug-suppress variableScope
@@ -560,11 +557,11 @@ WDataOutP VL_POW_WWW(int obits, int, int rbits, WDataOutP owp, const WDataInP lw
    for (int bit = 0; bit < rbits; bit++) {
        if (bit > 0) {  // power = power*power
            VL_ASSIGN_W(obits, lastpowstore, powstore);
-            VL_MUL_W(VL_WORDS_I(obits), powstore, lastpowstore, lastpowstore);
+            VL_MUL_W(owords, powstore, lastpowstore, lastpowstore);
        }
        if (VL_BITISSET_W(rwp, bit)) {  // out *= power
            VL_ASSIGN_W(obits, lastoutstore, owp);
-            VL_MUL_W(VL_WORDS_I(obits), owp, lastoutstore, powstore);
+            VL_MUL_W(owords, owp, lastoutstore, powstore);
        }
    }
    return owp;
@@ -576,7 +573,10 @@ WDataOutP VL_POW_WWQ(int obits, int lbits, int rbits, WDataOutP owp, const WData
    return VL_POW_WWW(obits, lbits, rbits, owp, lwp, rhsw);
 }
 QData VL_POW_QQW(int, int, int rbits, QData lhs, const WDataInP rwp) VL_MT_SAFE {
-    // Skip check for rhs == 0, as short-circuit doesn't save time
+    const int rwords = VL_WORDS_I(rbits);
+    EData rnz = rwp[0];
+    for (int w = 1; w < rwords; ++w) rnz |= rwp[w];
+    if (!rnz) return 1;  // rwp == 0
    if (VL_UNLIKELY(lhs == 0)) return 0;
    QData power = lhs;
    QData result = 1ULL;
@@ -657,8 +657,10 @@ double VL_ITOR_D_W(int lbits, const WDataInP lwp) VL_PURE {
 }
 double VL_ISTOR_D_W(int lbits, const WDataInP lwp) VL_MT_SAFE {
    if (!VL_SIGN_W(lbits, lwp)) return VL_ITOR_D_W(lbits, lwp);
+    const int words = VL_WORDS_I(lbits);
+    VL_DEBUG_IFDEF(assert(words <= VL_MULS_MAX_WORDS););
    uint32_t pos[VL_MULS_MAX_WORDS + 1];  // Fixed size, as MSVC++ doesn't allow [words] here
-    VL_NEGATE_W(VL_WORDS_I(lbits), pos, lwp);
+    VL_NEGATE_W(words, pos, lwp);
    _vl_clean_inplace_w(lbits, pos);
    return -VL_ITOR_D_W(lbits, pos);
 }
@@ -2094,7 +2096,7 @@ bool VlReadMem::get(QData& addrr, std::string& valuer) {

    if (VL_UNLIKELY(m_end != ~0ULL && m_addr <= m_end && !m_anyAddr)) {
        VL_WARN_MT(m_filename.c_str(), m_linenum, "",
-                   "$readmem file ended before specified final address (IEEE 2017 21.4)");
+                   "$readmem file ended before specified final address (IEEE 1800-2017 21.4)");
    }

    return false;  // EOF
@@ -2710,14 +2712,11 @@ void VerilatedContextImp::commandArgVl(const std::string& arg) {
                        "Exiting due to command line argument (not an error)");
        } else if (arg == "+verilator+noassert") {
            assertOn(false);
-        } else if (commandArgVlUint64(arg, "+verilator+prof+exec+start+", u64)
-                   || commandArgVlUint64(arg, "+verilator+prof+threads+start+", u64)) {
+        } else if (commandArgVlUint64(arg, "+verilator+prof+exec+start+", u64)) {
            profExecStart(u64);
-        } else if (commandArgVlUint64(arg, "+verilator+prof+exec+window+", u64, 1)
-                   || commandArgVlUint64(arg, "+verilator+prof+threads+window+", u64, 1)) {
+        } else if (commandArgVlUint64(arg, "+verilator+prof+exec+window+", u64, 1)) {
            profExecWindow(u64);
-        } else if (commandArgVlString(arg, "+verilator+prof+exec+file+", str)
-                   || commandArgVlString(arg, "+verilator+prof+threads+file+", str)) {
+        } else if (commandArgVlString(arg, "+verilator+prof+exec+file+", str)) {
            profExecFilename(str);
        } else if (commandArgVlString(arg, "+verilator+prof+vlt+file+", str)) {
            profVltFilename(str);
@@ -3024,6 +3023,7 @@ void Verilated::endOfEval(VerilatedEvalMsgQueue* evalMsgQp) VL_MT_SAFE {
    // It should be ok to call into endOfEvalGuts, it returns immediately
    // if there are no transactions.
    VL_DEBUG_IF(VL_DBG_MSGF("End-of-eval cleanup\n"););
+    VerilatedThreadMsgQueue::flush(evalMsgQp);
    evalMsgQp->process();
 }

@@ -3248,11 +3248,11 @@ void VlDeleter::deleteAll() VL_EXCLUDES(m_mutex) VL_EXCLUDES(m_deleteMutex) VL_M
            VerilatedLockGuard lock{m_mutex};
            if (m_newGarbage.empty()) break;
            m_deleteMutex.lock();
-            std::swap(m_newGarbage, m_toDelete);
+            std::swap(m_newGarbage, m_deleteNow);
            // m_mutex is unlocked here, so destructors can enqueue new objects
        }
-        for (VlDeletable* const objp : m_toDelete) delete objp;
-        m_toDelete.clear();
+        for (VlDeletable* const objp : m_deleteNow) delete objp;
+        m_deleteNow.clear();
        m_deleteMutex.unlock();
    }
 }
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -133,7 +133,8 @@ enum VerilatedVarType : uint8_t {
    VLVT_UINT32,  // AKA IData
    VLVT_UINT64,  // AKA QData
    VLVT_WDATA,  // AKA WData
-    VLVT_STRING  // C++ string
+    VLVT_STRING,  // C++ string
+    VLVT_REAL  // AKA double
 };

 enum VerilatedVarFlags {
@@ -164,7 +165,9 @@ inline constexpr size_t roundUpToMultipleOf(size_t value) {
 // Return current thread ID (or 0), not super fast, cache if needed
 extern uint32_t VL_THREAD_ID() VL_MT_SAFE;

+#ifndef VL_LOCK_SPINS
 #define VL_LOCK_SPINS 50000  /// Number of times to spin for a mutex before yielding
+#endif

 /// Mutex, wrapped to allow -fthread_safety checks
 class VL_CAPABILITY("mutex") VerilatedMutex final {
@@ -624,8 +627,9 @@ class VerilatedScope final {
 public:
    enum Type : uint8_t {
        SCOPE_MODULE,
-        SCOPE_OTHER
-    };  // Type of a scope, currently module is only interesting
+        SCOPE_OTHER,
+        SCOPE_PACKAGE
+    };  // Type of a scope, currently only module and package are interesting
 private:
    // Fastpath:
    VerilatedSyms* m_symsp = nullptr;  // Symbol table
@@ -919,6 +923,7 @@ public:

    // Internal: Called at end of each thread mtask, before finishing eval
    static void endOfThreadMTask(VerilatedEvalMsgQueue* evalMsgQp) VL_MT_SAFE {
+        mtaskId(0);
        if (VL_UNLIKELY(t_s.t_endOfEvalReqd)) endOfThreadMTaskGuts(evalMsgQp);
    }
    // Internal: Called at end of eval loop
@@ -2,13 +2,16 @@
 ######################################################################
 # DESCRIPTION: Makefile commands for all verilated target files
 #
-# Copyright 2003-2023 by Wilson Snyder. This program is free software; you
+# Copyright 2003-2024 by Wilson Snyder. This program is free software; you
 # can redistribute it and/or modify it under the terms of either the GNU
 # Lesser General Public License Version 3 or the Perl Artistic License
 # Version 2.0.
 # SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
 ######################################################################

+# Tool names.
+# These are computed at configuration time, and most are not ?=
+# to avoid picking up potentially incorrect Make implicit variables
 AR = @AR@
 CXX = @CXX@
 LINK = @CXX@
@@ -16,6 +19,7 @@ OBJCACHE ?= @OBJCACHE@
 PERL = @PERL@
 PYTHON3 = @PYTHON3@

+# Configuration time options
 CFG_WITH_CCWARN = @CFG_WITH_CCWARN@
 CFG_WITH_LONGTESTS = @CFG_WITH_LONGTESTS@

@@ -31,6 +35,12 @@ 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@
+# Compiler flags when creating a precompiled header
+CFG_CXXFLAGS_PCH = -x c++-header
+# Compiler option to put in front of filename to read precompiled header
+CFG_CXXFLAGS_PCH_I = @CFG_CXXFLAGS_PCH_I@
+# Compiler's filename prefix for precompiled headers, .gch if clang, empty if GCC
+CFG_GCH_IF_CLANG = @CFG_GCH_IF_CLANG@
 # Linker flags
 CFG_LDFLAGS_VERILATED = @CFG_LDFLAGS_VERILATED@
 # Linker libraries for multithreading
@@ -43,6 +53,12 @@ VERILATOR_COVERAGE = $(PERL) $(VERILATOR_ROOT)/bin/verilator_coverage
 VERILATOR_INCLUDER = $(PYTHON3) $(VERILATOR_ROOT)/bin/verilator_includer
 VERILATOR_CCACHE_REPORT = $(PYTHON3) $(VERILATOR_ROOT)/bin/verilator_ccache_report

+######################################################################
+# CCACHE flags (via environment as no command line option available)
+
+CCACHE_SLOPPINESS ?= pch_defines,time_macros
+export CCACHE_SLOPPINESS
+
 ######################################################################
 # Make checks

@@ -138,7 +154,7 @@ ifeq ($(VM_SC),1)
  LDFLAGS  += $(SYSTEMC_CXX_FLAGS) $(addprefix -L, $(SYSTEMC_LIBDIR))
  SC_LIBS   = -lsystemc
 ifneq ($(wildcard $(SYSTEMC_LIBDIR)/*numeric_bit*),)
-  # Systemc 1.2.1beta
+  # SystemC 1.2.1beta
  SC_LIBS   += -lnumeric_bit -lqt
 endif
 endif
@@ -162,11 +178,17 @@ endif
 VM_FAST += $(VM_CLASSES_FAST) $(VM_SUPPORT_FAST)
 VM_SLOW += $(VM_CLASSES_SLOW) $(VM_SUPPORT_SLOW)

+# Precompiled header filename
+VK_PCH_H = $(VM_PREFIX)__pch.h
+# Compiler read-a-precompiled-header option for precompiled header filename
+VK_PCH_I_FAST = $(CFG_CXXFLAGS_PCH_I) $(VM_PREFIX)__pch.h.fast$(CFG_GCH_IF_CLANG)
+VK_PCH_I_SLOW = $(CFG_CXXFLAGS_PCH_I) $(VM_PREFIX)__pch.h.slow$(CFG_GCH_IF_CLANG)
+
 #######################################################################
 ### Overall Objects Linking

-VK_FAST_OBJS = $(addsuffix .o, $(VM_FAST))
-VK_SLOW_OBJS = $(addsuffix .o, $(VM_SLOW))
+VK_OBJS_FAST = $(addsuffix .o, $(VM_FAST))
+VK_OBJS_SLOW = $(addsuffix .o, $(VM_SLOW))

 VK_USER_OBJS = $(addsuffix .o, $(VM_USER_CLASSES))

@@ -193,7 +215,7 @@ else
  # Parallel build: Each .cpp file by itself. This can be somewhat slower for
  # very small designs and examples, but is a lot faster for large designs.

-  VK_OBJS += $(VK_FAST_OBJS) $(VK_SLOW_OBJS)
+  VK_OBJS += $(VK_OBJS_FAST) $(VK_OBJS_SLOW)
 endif

 # When archiving just objects (.o), use single $(AR) run
@@ -237,18 +259,32 @@ $(VM_PREFIX)__ALL.a: $(VK_OBJS) $(VM_HIER_LIBS)
 ### Compile rules

 ifneq ($(VM_DEFAULT_RULES),0)
-# Anything not in $(VK_SLOW_OBJS) or $(VK_GLOBAL_OBJS), including verilated.o
-# and user files passed on the Verilator command line use this rule.
-# We put OPT_FAST/OPT_SLOW/OPT_GLOBAL before the other flags to
+# Compilation rule for anything not in $(VK_OBJS_FAST), $(VK_OBJS_SLOW), or
+# $(VK_GLOBAL_OBJS) including verilated.o.  This typically means user files
+# passed on the Verilator command line.
+#
+# These rules put OPT_FAST/OPT_SLOW/OPT_GLOBAL before the other flags to
 # allow USER_CPPFLAGS to override them
 %.o: %.cpp
 	$(OBJCACHE) $(CXX) $(OPT_FAST) $(CXXFLAGS) $(CPPFLAGS) -c -o $@ $<

-$(VK_SLOW_OBJS): %.o: %.cpp
-	$(OBJCACHE) $(CXX) $(OPT_SLOW) $(CXXFLAGS) $(CPPFLAGS) -c -o $@ $<
+$(VK_OBJS_FAST): %.o: %.cpp $(VK_PCH_H).fast.gch
+	$(OBJCACHE) $(CXX) $(OPT_FAST) $(CXXFLAGS) $(CPPFLAGS) $(VK_PCH_I_FAST) -c -o $@ $<
+
+$(VK_OBJS_SLOW): %.o: %.cpp $(VK_PCH_H).slow.gch
+	$(OBJCACHE) $(CXX) $(OPT_SLOW) $(CXXFLAGS) $(CPPFLAGS) $(VK_PCH_I_SLOW) -c -o $@ $<

 $(VK_GLOBAL_OBJS): %.o: %.cpp
 	$(OBJCACHE) $(CXX) $(OPT_GLOBAL) $(CXXFLAGS) $(CPPFLAGS) -c -o $@ $<
+
+# Precompile a header file
+# PCH's compiler flags must match exactly the rules' above FAST/SLOW
+# arguments used for the .cpp files, or the PCH file won't be used.
+%.fast.gch: %
+	$(OBJCACHE) $(CXX) $(OPT_FAST) $(CXXFLAGS) $(CPPFLAGS) $(CFG_CXXFLAGS_PCH) $< -o $@
+%.slow.gch: %
+	$(OBJCACHE) $(CXX) $(OPT_SLOW) $(CXXFLAGS) $(CPPFLAGS) $(CFG_CXXFLAGS_PCH) $< -o $@
+
 endif

 #Default rule embedded in make:
@@ -296,19 +332,21 @@ endif

 debug-make::
 	@echo
-	@echo CXXFLAGS: $(CXXFLAGS)
 	@echo CPPFLAGS: $(CPPFLAGS)
+	@echo CXXFLAGS: $(CXXFLAGS)
 	@echo OPT_FAST: $(OPT_FAST)
 	@echo OPT_SLOW: $(OPT_SLOW)
-	@echo VM_PREFIX:  $(VM_PREFIX)
-	@echo VM_PARALLEL_BUILDS:  $(VM_PARALLEL_BUILDS)
+	@echo VK_OBJS: $(VK_OBJS)
+	@echo VK_OBJS_FAST: $(VK_OBJS_FAST)
+	@echo VK_OBJS_SLOW: $(VK_OBJS_SLOW)
 	@echo VM_CLASSES_FAST: $(VM_CLASSES_FAST)
 	@echo VM_CLASSES_SLOW: $(VM_CLASSES_SLOW)
-	@echo VM_SUPPORT_FAST: $(VM_SUPPORT_FAST)
-	@echo VM_SUPPORT_SLOW: $(VM_SUPPORT_SLOW)
 	@echo VM_GLOBAL_FAST: $(VM_GLOBAL_FAST)
 	@echo VM_GLOBAL_SLOW: $(VM_GLOBAL_SLOW)
-	@echo VK_OBJS: $(VK_OBJS)
+	@echo VM_PARALLEL_BUILDS:  $(VM_PARALLEL_BUILDS)
+	@echo VM_PREFIX:  $(VM_PREFIX)
+	@echo VM_SUPPORT_FAST: $(VM_SUPPORT_FAST)
+	@echo VM_SUPPORT_SLOW: $(VM_SUPPORT_SLOW)
 	@echo

 ######################################################################
@@ -2,7 +2,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -211,7 +211,7 @@ private:
        std::string result = prefix + "*" + suffix;

        // std::cout << "\nch pre=" << prefix << "  s=" << suffix << "\nch a="
-        // << old << "\nch b=" << add << "\ncho=" << result << std::endl;
+        // << old << "\nch b=" << add << "\ncho=" << result << "\n";
        return result;
    }
    bool itemMatchesString(VerilatedCovImpItem* itemp, const std::string& match)
@@ -337,7 +337,7 @@ public:
            const std::string key = keys[i];
            if (!keys[i].empty()) {
                const std::string val = valps[i];
-                // std::cout << "   " << __FUNCTION__ << "  " << key << " = " << val << std::endl;
+                // std::cout << "   " << __FUNCTION__ << "  " << key << " = " << val << "\n";
                m_insertp->m_keys[addKeynum] = valueIndex(key);
                m_insertp->m_vals[addKeynum] = valueIndex(val);
                ++addKeynum;
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2009-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2009-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -23,7 +23,6 @@

 // clang-format off

-#define __STDC_LIMIT_MACROS  // UINT64_MAX
 #include "verilated.h"
 #include "verilated_fst_c.h"

@@ -40,6 +39,7 @@
 #include <algorithm>
 #include <iterator>
 #include <sstream>
+#include <type_traits>

 #if defined(_WIN32) && !defined(__MINGW32__) && !defined(__CYGWIN__)
 # include <io.h>
@@ -50,33 +50,10 @@
 // clang-format on

 //=============================================================================
-// Check that vltscope_t matches fstScopeType
+// Check that forward declared types matches the FST API types

-static_assert(static_cast<int>(FST_ST_VCD_MODULE) == static_cast<int>(VLT_TRACE_SCOPE_MODULE),
-              "VLT_TRACE_SCOPE_MODULE mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_TASK) == static_cast<int>(VLT_TRACE_SCOPE_TASK),
-              "VLT_TRACE_SCOPE_TASK mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_FUNCTION) == static_cast<int>(VLT_TRACE_SCOPE_FUNCTION),
-              "VLT_TRACE_SCOPE_FUNCTION mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_BEGIN) == static_cast<int>(VLT_TRACE_SCOPE_BEGIN),
-              "VLT_TRACE_SCOPE_BEGIN mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_FORK) == static_cast<int>(VLT_TRACE_SCOPE_FORK),
-              "VLT_TRACE_SCOPE_FORK mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_GENERATE) == static_cast<int>(VLT_TRACE_SCOPE_GENERATE),
-              "VLT_TRACE_SCOPE_GENERATE mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_STRUCT) == static_cast<int>(VLT_TRACE_SCOPE_STRUCT),
-              "VLT_TRACE_SCOPE_STRUCT mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_UNION) == static_cast<int>(VLT_TRACE_SCOPE_UNION),
-              "VLT_TRACE_SCOPE_UNION mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_CLASS) == static_cast<int>(VLT_TRACE_SCOPE_CLASS),
-              "VLT_TRACE_SCOPE_CLASS mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_INTERFACE)
-                  == static_cast<int>(VLT_TRACE_SCOPE_INTERFACE),
-              "VLT_TRACE_SCOPE_INTERFACE mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_PACKAGE) == static_cast<int>(VLT_TRACE_SCOPE_PACKAGE),
-              "VLT_TRACE_SCOPE_PACKAGE mismatches");
-static_assert(static_cast<int>(FST_ST_VCD_PROGRAM) == static_cast<int>(VLT_TRACE_SCOPE_PROGRAM),
-              "VLT_TRACE_SCOPE_PROGRAM mismatches");
+static_assert(std::is_same<vlFstHandle, fstHandle>::value, "vlFstHandle mismatch");
+static_assert(std::is_same<vlFstEnumHandle, fstEnumHandle>::value, "vlFstHandle mismatch");

 //=============================================================================
 // Specialization of the generics for this trace format
@@ -104,19 +81,11 @@ void VerilatedFst::open(const char* filename) VL_MT_SAFE_EXCLUDES(m_mutex) {
    fstWriterSetPackType(m_fst, FST_WR_PT_LZ4);
    fstWriterSetTimescaleFromString(m_fst, timeResStr().c_str());  // lintok-begin-on-ref
    if (m_useFstWriterThread) fstWriterSetParallelMode(m_fst, 1);
+    constDump(true);  // First dump must contain the const signals
    fullDump(true);  // First dump must be full for fst

-    m_curScope.clear();
-
    Super::traceInit();

-    // Clear the scope stack
-    auto it = m_curScope.begin();
-    while (it != m_curScope.end()) {
-        fstWriterSetUpscope(m_fst);
-        it = m_curScope.erase(it);
-    }
-
    // convert m_code2symbol into an array for fast lookup
    if (!m_symbolp) {
        m_symbolp = new fstHandle[nextCode()]{0};
@@ -154,102 +123,142 @@ void VerilatedFst::declDTypeEnum(int dtypenum, const char* name, uint32_t elemen
    m_local2fstdtype[dtypenum] = enumNum;
 }

-void VerilatedFst::declare(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                           fstVarType vartype, bool array, int arraynum, bool bussed, int msb,
-                           int lsb) {
+// TODO: should return std::optional<fstScopeType>, but I can't have C++17
+static std::pair<bool, fstScopeType> toFstScopeType(VerilatedTracePrefixType type) {
+    switch (type) {
+    case VerilatedTracePrefixType::SCOPE_MODULE: return {true, FST_ST_VCD_MODULE};
+    case VerilatedTracePrefixType::SCOPE_INTERFACE: return {true, FST_ST_VCD_INTERFACE};
+    case VerilatedTracePrefixType::STRUCT_PACKED:
+    case VerilatedTracePrefixType::STRUCT_UNPACKED: return {true, FST_ST_VCD_STRUCT};
+    case VerilatedTracePrefixType::UNION_PACKED: return {true, FST_ST_VCD_UNION};
+    default: return {false, /* unused so whatever, just need a value */ FST_ST_VCD_SCOPE};
+    }
+}
+
+void VerilatedFst::pushPrefix(const std::string& name, VerilatedTracePrefixType type) {
+    const std::string newPrefix = m_prefixStack.back().first + name;
+    const auto pair = toFstScopeType(type);
+    const bool properScope = pair.first;
+    const fstScopeType scopeType = pair.second;
+    m_prefixStack.emplace_back(newPrefix + (properScope ? " " : ""), type);
+    if (properScope) {
+        const std::string scopeName = lastWord(newPrefix);
+        fstWriterSetScope(m_fst, scopeType, scopeName.c_str(), nullptr);
+    }
+}
+
+void VerilatedFst::popPrefix() {
+    const bool properScope = toFstScopeType(m_prefixStack.back().second).first;
+    if (properScope) fstWriterSetUpscope(m_fst);
+    m_prefixStack.pop_back();
+    assert(!m_prefixStack.empty());
+}
+
+void VerilatedFst::declare(uint32_t code, const char* name, int dtypenum,
+                           VerilatedTraceSigDirection direction, VerilatedTraceSigKind kind,
+                           VerilatedTraceSigType type, bool array, int arraynum, bool bussed,
+                           int msb, int lsb) {
    const int bits = ((msb > lsb) ? (msb - lsb) : (lsb - msb)) + 1;

-    const bool enabled = Super::declCode(code, name, bits, false);
+    const std::string hierarchicalName = m_prefixStack.back().first + name;
+
+    const bool enabled = Super::declCode(code, hierarchicalName, bits);
    if (!enabled) return;

-    std::string nameasstr = namePrefix() + name;
-    std::istringstream nameiss{nameasstr};
-    std::istream_iterator<std::string> beg(nameiss);
-    std::istream_iterator<std::string> end;
-    std::list<std::string> tokens(beg, end);  // Split name
-    std::string symbol_name{tokens.back()};
-    tokens.pop_back();  // Remove symbol name from hierarchy
-    std::string tmpModName;
-
-    // Find point where current and new scope diverge
-    auto cur_it = m_curScope.begin();
-    auto new_it = tokens.begin();
-    while (cur_it != m_curScope.end() && new_it != tokens.end()) {
-        if (*cur_it != *new_it) break;
-        ++cur_it;
-        ++new_it;
-    }
-
-    // Go back to the common point
-    while (cur_it != m_curScope.end()) {
-        fstWriterSetUpscope(m_fst);
-        cur_it = m_curScope.erase(cur_it);
-    }
-
-    // Follow the hierarchy of the new variable from the common scope point
-    while (new_it != tokens.end()) {
-        if ((new_it->back() & 0x80)) {
-            tmpModName = *new_it;
-            tmpModName.pop_back();
-            // If the scope ends with a non-ASCII character, it will be 0x80 + fstScopeType
-            fstWriterSetScope(m_fst, static_cast<fstScopeType>(new_it->back() & 0x7f),
-                              tmpModName.c_str(), nullptr);
-        } else {
-            fstWriterSetScope(m_fst, FST_ST_VCD_SCOPE, new_it->c_str(), nullptr);
-        }
-        m_curScope.push_back(*new_it);
-        new_it = tokens.erase(new_it);
-    }
-
+    assert(hierarchicalName.rfind(' ') != std::string::npos);
    std::stringstream name_ss;
-    name_ss << symbol_name;
+    name_ss << lastWord(hierarchicalName);
    if (array) name_ss << "[" << arraynum << "]";
    if (bussed) name_ss << " [" << msb << ":" << lsb << "]";
-    std::string name_str = name_ss.str();
+    const std::string name_str = name_ss.str();

-    if (dtypenum > 0) {
-        const fstEnumHandle enumNum = m_local2fstdtype[dtypenum];
-        fstWriterEmitEnumTableRef(m_fst, enumNum);
+    if (dtypenum > 0) fstWriterEmitEnumTableRef(m_fst, m_local2fstdtype[dtypenum]);
+
+    fstVarDir varDir;
+    switch (direction) {
+    case VerilatedTraceSigDirection::INOUT: varDir = FST_VD_INOUT; break;
+    case VerilatedTraceSigDirection::OUTPUT: varDir = FST_VD_OUTPUT; break;
+    case VerilatedTraceSigDirection::INPUT: varDir = FST_VD_INPUT; break;
+    case VerilatedTraceSigDirection::NONE: varDir = FST_VD_IMPLICIT; break;
    }

+    fstVarType varType;
+    // Doubles have special decoding properties, so must indicate if a double
+    if (type == VerilatedTraceSigType::DOUBLE) {
+        if (kind == VerilatedTraceSigKind::PARAMETER) {
+            varType = FST_VT_VCD_REAL_PARAMETER;
+        } else {
+            varType = FST_VT_VCD_REAL;
+        }
+    }
+    // clang-format off
+    else if (kind == VerilatedTraceSigKind::PARAMETER) varType = FST_VT_VCD_PARAMETER;
+    else if (kind == VerilatedTraceSigKind::SUPPLY0) varType = FST_VT_VCD_SUPPLY0;
+    else if (kind == VerilatedTraceSigKind::SUPPLY1) varType = FST_VT_VCD_SUPPLY1;
+    else if (kind == VerilatedTraceSigKind::TRI) varType = FST_VT_VCD_TRI;
+    else if (kind == VerilatedTraceSigKind::TRI0) varType = FST_VT_VCD_TRI0;
+    else if (kind == VerilatedTraceSigKind::TRI1) varType = FST_VT_VCD_TRI1;
+    else if (kind == VerilatedTraceSigKind::WIRE) varType = FST_VT_VCD_WIRE;
+    //
+    else if (type == VerilatedTraceSigType::INTEGER) varType = FST_VT_VCD_INTEGER;
+    else if (type == VerilatedTraceSigType::BIT) varType = FST_VT_SV_BIT;
+    else if (type == VerilatedTraceSigType::LOGIC) varType = FST_VT_SV_LOGIC;
+    else if (type == VerilatedTraceSigType::INT) varType = FST_VT_SV_INT;
+    else if (type == VerilatedTraceSigType::SHORTINT) varType = FST_VT_SV_SHORTINT;
+    else if (type == VerilatedTraceSigType::LONGINT) varType = FST_VT_SV_LONGINT;
+    else if (type == VerilatedTraceSigType::BYTE) varType = FST_VT_SV_BYTE;
+    else if (type == VerilatedTraceSigType::EVENT) varType = FST_VT_VCD_EVENT;
+    else if (type == VerilatedTraceSigType::TIME) varType = FST_VT_VCD_TIME;
+    else { assert(0); /* Unreachable */ }
+    // clang-format on
+
    const auto it = vlstd::as_const(m_code2symbol).find(code);
    if (it == m_code2symbol.end()) {  // New
        m_code2symbol[code]
-            = fstWriterCreateVar(m_fst, vartype, vardir, bits, name_str.c_str(), 0);
+            = fstWriterCreateVar(m_fst, varType, varDir, bits, name_str.c_str(), 0);
    } else {  // Alias
-        fstWriterCreateVar(m_fst, vartype, vardir, bits, name_str.c_str(), it->second);
+        fstWriterCreateVar(m_fst, varType, varDir, bits, name_str.c_str(), it->second);
    }
 }

-void VerilatedFst::declEvent(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                             fstVarType vartype, bool array, int arraynum) {
-    declare(code, name, dtypenum, vardir, vartype, array, arraynum, false, 0, 0);
+void VerilatedFst::declEvent(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                             VerilatedTraceSigDirection direction, VerilatedTraceSigKind kind,
+                             VerilatedTraceSigType type, bool array, int arraynum) {
+    declare(code, name, dtypenum, direction, kind, type, array, arraynum, false, 0, 0);
 }
-void VerilatedFst::declBit(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                           fstVarType vartype, bool array, int arraynum) {
-    declare(code, name, dtypenum, vardir, vartype, array, arraynum, false, 0, 0);
+void VerilatedFst::declBit(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                           VerilatedTraceSigDirection direction, VerilatedTraceSigKind kind,
+                           VerilatedTraceSigType type, bool array, int arraynum) {
+    declare(code, name, dtypenum, direction, kind, type, array, arraynum, false, 0, 0);
 }
-void VerilatedFst::declBus(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                           fstVarType vartype, bool array, int arraynum, int msb, int lsb) {
-    declare(code, name, dtypenum, vardir, vartype, array, arraynum, true, msb, lsb);
+void VerilatedFst::declBus(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                           VerilatedTraceSigDirection direction, VerilatedTraceSigKind kind,
+                           VerilatedTraceSigType type, bool array, int arraynum, int msb,
+                           int lsb) {
+    declare(code, name, dtypenum, direction, kind, type, array, arraynum, true, msb, lsb);
 }
-void VerilatedFst::declQuad(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                            fstVarType vartype, bool array, int arraynum, int msb, int lsb) {
-    declare(code, name, dtypenum, vardir, vartype, array, arraynum, true, msb, lsb);
+void VerilatedFst::declQuad(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                            VerilatedTraceSigDirection direction, VerilatedTraceSigKind kind,
+                            VerilatedTraceSigType type, bool array, int arraynum, int msb,
+                            int lsb) {
+    declare(code, name, dtypenum, direction, kind, type, array, arraynum, true, msb, lsb);
 }
-void VerilatedFst::declArray(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                             fstVarType vartype, bool array, int arraynum, int msb, int lsb) {
-    declare(code, name, dtypenum, vardir, vartype, array, arraynum, true, msb, lsb);
+void VerilatedFst::declArray(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                             VerilatedTraceSigDirection direction, VerilatedTraceSigKind kind,
+                             VerilatedTraceSigType type, bool array, int arraynum, int msb,
+                             int lsb) {
+    declare(code, name, dtypenum, direction, kind, type, array, arraynum, true, msb, lsb);
 }
-void VerilatedFst::declDouble(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                              fstVarType vartype, bool array, int arraynum) {
-    declare(code, name, dtypenum, vardir, vartype, array, arraynum, false, 63, 0);
+void VerilatedFst::declDouble(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                              VerilatedTraceSigDirection direction, VerilatedTraceSigKind kind,
+                              VerilatedTraceSigType type, bool array, int arraynum) {
+    declare(code, name, dtypenum, direction, kind, type, array, arraynum, false, 63, 0);
 }

 //=============================================================================
 // Get/commit trace buffer

-VerilatedFst::Buffer* VerilatedFst::getTraceBuffer() {
+VerilatedFst::Buffer* VerilatedFst::getTraceBuffer(uint32_t fidx) {
    if (offload()) return new OffloadBuffer{*this};
    return new Buffer{*this};
 }
@@ -284,7 +293,7 @@ void VerilatedFst::configure(const VerilatedTraceConfig& config) {
 // so always inline them.

 VL_ATTR_ALWINLINE
-void VerilatedFstBuffer::emitEvent(uint32_t code, VlEvent newval) {
+void VerilatedFstBuffer::emitEvent(uint32_t code, const VlEventBase* newval) {
    VL_DEBUG_IFDEF(assert(m_symbolp[code]););
    fstWriterEmitValueChange(m_fst, m_symbolp[code], "1");
 }
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -24,13 +24,14 @@
 #include "verilated.h"
 #include "verilated_trace.h"

-#include "gtkwave/fstapi.h"
-
 #include <list>
 #include <map>
 #include <string>
 #include <vector>

+typedef uint32_t vlFstHandle;
+typedef uint32_t vlFstEnumHandle;
+
 class VerilatedFstBuffer;

 //=============================================================================
@@ -49,18 +50,22 @@ private:
    // FST specific internals

    void* m_fst = nullptr;
-    std::map<uint32_t, fstHandle> m_code2symbol;
-    std::map<int, fstEnumHandle> m_local2fstdtype;
-    std::list<std::string> m_curScope;
-    fstHandle* m_symbolp = nullptr;  // same as m_code2symbol, but as an array
+    std::map<uint32_t, vlFstHandle> m_code2symbol;
+    std::map<int, vlFstEnumHandle> m_local2fstdtype;
+    vlFstHandle* m_symbolp = nullptr;  // same as m_code2symbol, but as an array
    char* m_strbufp = nullptr;  // String buffer long enough to hold maxBits() chars

    bool m_useFstWriterThread = false;  // Whether to use the separate FST writer thread

+    // Prefixes to add to signal names/scope types
+    std::vector<std::pair<std::string, VerilatedTracePrefixType>> m_prefixStack{
+        {"", VerilatedTracePrefixType::SCOPE_MODULE}};
+
    // CONSTRUCTORS
    VL_UNCOPYABLE(VerilatedFst);
-    void declare(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                 fstVarType vartype, bool array, int arraynum, bool bussed, int msb, int lsb);
+    void declare(uint32_t code, const char* name, int dtypenum, VerilatedTraceSigDirection,
+                 VerilatedTraceSigKind, VerilatedTraceSigType, bool array, int arraynum,
+                 bool bussed, int msb, int lsb);

 protected:
    //=========================================================================
@@ -74,7 +79,7 @@ protected:
    bool preChangeDump() override { return isOpen(); }

    // Trace buffer management
-    Buffer* getTraceBuffer() override;
+    Buffer* getTraceBuffer(uint32_t fidx) override;
    void commitTraceBuffer(Buffer*) override;

    // Configure sub-class
@@ -101,18 +106,27 @@ public:
    //=========================================================================
    // Internal interface to Verilator generated code

-    void declEvent(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                   fstVarType vartype, bool array, int arraynum);
-    void declBit(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                 fstVarType vartype, bool array, int arraynum);
-    void declBus(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                 fstVarType vartype, bool array, int arraynum, int msb, int lsb);
-    void declQuad(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                  fstVarType vartype, bool array, int arraynum, int msb, int lsb);
-    void declArray(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                   fstVarType vartype, bool array, int arraynum, int msb, int lsb);
-    void declDouble(uint32_t code, const char* name, int dtypenum, fstVarDir vardir,
-                    fstVarType vartype, bool array, int arraynum);
+    void pushPrefix(const std::string&, VerilatedTracePrefixType);
+    void popPrefix();
+
+    void declEvent(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                   VerilatedTraceSigDirection, VerilatedTraceSigKind, VerilatedTraceSigType,
+                   bool array, int arraynum);
+    void declBit(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                 VerilatedTraceSigDirection, VerilatedTraceSigKind, VerilatedTraceSigType,
+                 bool array, int arraynum);
+    void declBus(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                 VerilatedTraceSigDirection, VerilatedTraceSigKind, VerilatedTraceSigType,
+                 bool array, int arraynum, int msb, int lsb);
+    void declQuad(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                  VerilatedTraceSigDirection, VerilatedTraceSigKind, VerilatedTraceSigType,
+                  bool array, int arraynum, int msb, int lsb);
+    void declArray(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                   VerilatedTraceSigDirection, VerilatedTraceSigKind, VerilatedTraceSigType,
+                   bool array, int arraynum, int msb, int lsb);
+    void declDouble(uint32_t code, uint32_t fidx, const char* name, int dtypenum,
+                    VerilatedTraceSigDirection, VerilatedTraceSigKind, VerilatedTraceSigType,
+                    bool array, int arraynum);

    void declDTypeEnum(int dtypenum, const char* name, uint32_t elements, unsigned int minValbits,
                       const char** itemNamesp, const char** itemValuesp);
@@ -149,7 +163,7 @@ class VerilatedFstBuffer VL_NOT_FINAL {
    // The FST file handle
    void* const m_fst = m_owner.m_fst;
    // code to fstHande map, as an array
-    const fstHandle* const m_symbolp = m_owner.m_symbolp;
+    const vlFstHandle* const m_symbolp = m_owner.m_symbolp;
    // String buffer long enough to hold maxBits() chars
    char* const m_strbufp = m_owner.m_strbufp;

@@ -163,7 +177,7 @@ class VerilatedFstBuffer VL_NOT_FINAL {

    // Implementations of duck-typed methods for VerilatedTraceBuffer. These are
    // called from only one place (the full* methods), so always inline them.
-    VL_ATTR_ALWINLINE void emitEvent(uint32_t code, VlEvent newval);
+    VL_ATTR_ALWINLINE void emitEvent(uint32_t code, const VlEventBase* newval);
    VL_ATTR_ALWINLINE void emitBit(uint32_t code, CData newval);
    VL_ATTR_ALWINLINE void emitCData(uint32_t code, CData newval, int bits);
    VL_ATTR_ALWINLINE void emitSData(uint32_t code, SData newval, int bits);
@@ -196,7 +210,7 @@ public:
    /// Return if file is open
    bool isOpen() const VL_MT_SAFE { return m_sptrace.isOpen(); }
    /// Open a new FST file
-    void open(const char* filename) VL_MT_SAFE { m_sptrace.open(filename); }
+    virtual void open(const char* filename) VL_MT_SAFE { m_sptrace.open(filename); }
    /// Close dump
    void close() VL_MT_SAFE { m_sptrace.close(); }
    /// Flush dump
@@ -216,12 +230,12 @@ public:

    // Set time units (s/ms, defaults to ns)
    // Users should not need to call this, as for Verilated models, these
-    // propage from the Verilated default timeunit
+    // propagate from the Verilated default timeunit
    void set_time_unit(const char* unitp) VL_MT_SAFE { m_sptrace.set_time_unit(unitp); }
    void set_time_unit(const std::string& unit) VL_MT_SAFE { m_sptrace.set_time_unit(unit); }
    // Set time resolution (s/ms, defaults to ns)
    // Users should not need to call this, as for Verilated models, these
-    // propage from the Verilated default timeprecision
+    // propagate from the Verilated default timeprecision
    void set_time_resolution(const char* unitp) VL_MT_SAFE {
        m_sptrace.set_time_resolution(unitp);
    }
@@ -3,7 +3,7 @@
 //
 // THIS MODULE IS PUBLICLY LICENSED
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -1,7 +1,7 @@
 // -*- mode: C++; c-file-style: "cc-mode" -*-
 //=============================================================================
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -12,8 +12,7 @@
 /// \file
 /// \brief Verilated tracing in FST format for SystemC header
 ///
-/// User wrapper code should use this header when creating FST SystemC
-/// traces.
+/// User wrapper code should use this header when creating FST SystemC traces.
 ///
 /// This class is not threadsafe, as the SystemC kernel is not threadsafe.
 ///
@@ -25,115 +24,33 @@
 #include "verilatedos.h"

 #include "verilated_fst_c.h"
-#include "verilated_sc.h"
-
-#include <string>
+#include "verilated_sc_trace.h"

 //=============================================================================
 // VerilatedFstSc
-///
-/// Class representing a Verilator-friendly FST trace format registered
-/// with the SystemC simulation kernel, just like a SystemC-documented
-/// trace format.
+/// Trace file used to create FST dump for SystemC version of Verilated models. It's very similar
+/// to its C version (see the class VerilatedFstC)

-class VerilatedFstSc final : sc_core::sc_trace_file, public VerilatedFstC {
+class VerilatedFstSc final : VerilatedScTraceBase, public VerilatedFstC {
    // CONSTRUCTORS
    VL_UNCOPYABLE(VerilatedFstSc);

 public:
-    /// Construct a SC trace object, and register with the SystemC kernel
    VerilatedFstSc() {
-        sc_core::sc_get_curr_simcontext()->add_trace_file(this);
-        // We want to avoid a depreciated warning, but still be back compatible.
-        // Turning off the message just for this still results in an
-        // annoying "to turn off" message.
-        const sc_core::sc_time t1sec{1, sc_core::SC_SEC};
-        if (t1sec.to_default_time_units() != 0) {
-            const sc_core::sc_time tunits{1.0 / t1sec.to_default_time_units(), sc_core::SC_SEC};
-            spTrace()->set_time_unit(tunits.to_string());
-        }
-        spTrace()->set_time_resolution(sc_core::sc_get_time_resolution().to_string());
-    }
-    /// Destruct, flush, and close the dump
-    ~VerilatedFstSc() override { close(); }
-
-    // METHODS - for SC kernel
-    // Called by SystemC simulate()
-    void cycle(bool delta_cycle) override {
-        if (!delta_cycle) this->dump(sc_core::sc_time_stamp().to_double());
+        spTrace()->set_time_unit(VerilatedScTraceBase::getScTimeUnit());
+        spTrace()->set_time_resolution(VerilatedScTraceBase::getScTimeResolution());
    }

+    // METHODS
    // Override VerilatedFstC. Must be called after starting simulation.
-    // Note: this is not a virtual function in the base class, so no 'override'
-    virtual void open(const char* filename) VL_MT_SAFE {
-        if (VL_UNLIKELY(!sc_core::sc_get_curr_simcontext()->elaboration_done())) {
-            Verilated::scTraceBeforeElaborationError();
-        }
+    void open(const char* filename) override VL_MT_SAFE {
+        VerilatedScTraceBase::checkScElaborationDone();
        VerilatedFstC::open(filename);
    }

-private:
-    // METHODS - Fake outs for linker
-
-#ifdef NC_SYSTEMC
-    // Cadence Incisive has these as abstract functions so we must create them
-    void set_time_unit(int exponent10_seconds) override {}  // deprecated
-#endif
-    void set_time_unit(double v, sc_core::sc_time_unit tu) override {}  // LCOV_EXCL_LINE
-
-    //--------------------------------------------------
-    // SystemC 2.1.v1
-
-    void write_comment(const std::string&) override {}
-    void trace(const unsigned int&, const std::string&, const char**) override {}
-
-#define DECL_TRACE_METHOD_A(tp) \
-    void trace(const tp& object, const std::string& name) override {}
-#define DECL_TRACE_METHOD_B(tp) \
-    void trace(const tp& object, const std::string& name, int width) override {}
-
-    // clang-format off
-    // Formatting matches that of sc_trace.h
-    // LCOV_EXCL_START
-#if (SYSTEMC_VERSION >= 20171012)
-    DECL_TRACE_METHOD_A( sc_core::sc_event )
-    DECL_TRACE_METHOD_A( sc_core::sc_time )
-#endif
-
-    DECL_TRACE_METHOD_A( bool )
-    DECL_TRACE_METHOD_A( sc_dt::sc_bit )
-    DECL_TRACE_METHOD_A( sc_dt::sc_logic )
-
-    DECL_TRACE_METHOD_B( unsigned char )
-    DECL_TRACE_METHOD_B( unsigned short )
-    DECL_TRACE_METHOD_B( unsigned int )
-    DECL_TRACE_METHOD_B( unsigned long )
-    DECL_TRACE_METHOD_B( char )
-    DECL_TRACE_METHOD_B( short )
-    DECL_TRACE_METHOD_B( int )
-    DECL_TRACE_METHOD_B( long )
-    DECL_TRACE_METHOD_B( sc_dt::int64 )
-    DECL_TRACE_METHOD_B( sc_dt::uint64 )
-
-    DECL_TRACE_METHOD_A( float )
-    DECL_TRACE_METHOD_A( double )
-    DECL_TRACE_METHOD_A( sc_dt::sc_int_base )
-    DECL_TRACE_METHOD_A( sc_dt::sc_uint_base )
-    DECL_TRACE_METHOD_A( sc_dt::sc_signed )
-    DECL_TRACE_METHOD_A( sc_dt::sc_unsigned )
-
-    DECL_TRACE_METHOD_A( sc_dt::sc_fxval )
-    DECL_TRACE_METHOD_A( sc_dt::sc_fxval_fast )
-    DECL_TRACE_METHOD_A( sc_dt::sc_fxnum )
-    DECL_TRACE_METHOD_A( sc_dt::sc_fxnum_fast )
-
-    DECL_TRACE_METHOD_A( sc_dt::sc_bv_base )
-    DECL_TRACE_METHOD_A( sc_dt::sc_lv_base )
-    // LCOV_EXCL_STOP
-    // clang-format on
-
-#undef DECL_TRACE_METHOD_A
-#undef DECL_TRACE_METHOD_B
+    // METHODS - for SC kernel
+    // Called from SystemC kernel
+    void cycle() override { VerilatedFstC::dump(sc_core::sc_time_stamp().to_double()); }
 };

 #endif  // Guard
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -284,6 +284,8 @@ inline uint64_t vl_time_stamp64() VL_MT_SAFE {
 # endif
 #endif

+// clang-format on
+
 uint64_t VerilatedContext::time() const VL_MT_SAFE {
    // When using non-default context, fastest path is return time
    if (VL_LIKELY(m_s.m_time)) return m_s.m_time;
@@ -311,25 +313,6 @@ double vl_time_multiplier(int scale) VL_PURE;
 // Return power of 10. e.g. returns 100 if n==2
 uint64_t vl_time_pow10(int n) VL_PURE;

-#ifdef VL_DEBUG
-/// Evaluate statement if VL_DEBUG defined
-# define VL_DEBUG_IFDEF(stmt) \
-    do { \
-        stmt \
-    } while (false)
-/// Evaluate statement if VL_DEBUG defined and Verilated::debug() enabled
-# define VL_DEBUG_IF(stmt) \
-    do { \
-        if (VL_UNLIKELY(Verilated::debug())) {stmt} \
-    } while (false)
-#else
-// We intentionally do not compile the stmt to improve compile speed
-# define VL_DEBUG_IFDEF(stmt) do {} while (false)
-# define VL_DEBUG_IF(stmt) do {} while (false)
-#endif
-
-// clang-format on
-
 //=========================================================================
 // Functional macros/routines
 // These all take the form
@@ -342,10 +325,23 @@ uint64_t vl_time_pow10(int n) VL_PURE;
 //===================================================================
 // SETTING OPERATORS

+VL_ATTR_ALWINLINE
+static WDataOutP VL_MEMSET_ZERO_W(WDataOutP owp, int words) VL_MT_SAFE {
+    return static_cast<WDataOutP>(std::memset(owp, 0, words * sizeof(EData)));
+}
+VL_ATTR_ALWINLINE
+static WDataOutP VL_MEMSET_ONES_W(WDataOutP owp, int words) VL_MT_SAFE {
+    return static_cast<WDataOutP>(std::memset(owp, 0xff, words * sizeof(EData)));
+}
+VL_ATTR_ALWINLINE
+static WDataOutP VL_MEMCPY_W(WDataOutP owp, WDataInP const iwp, int words) VL_MT_SAFE {
+    return static_cast<WDataOutP>(std::memcpy(owp, iwp, words * sizeof(EData)));
+}
+
 // Output clean
 // EMIT_RULE: VL_CLEAN:  oclean=clean; obits=lbits;
-#define VL_CLEAN_II(obits, lbits, lhs) ((lhs)&VL_MASK_I(obits))
-#define VL_CLEAN_QQ(obits, lbits, lhs) ((lhs)&VL_MASK_Q(obits))
+#define VL_CLEAN_II(obits, lbits, lhs) ((lhs) & (VL_MASK_I(obits)))
+#define VL_CLEAN_QQ(obits, lbits, lhs) ((lhs) & (VL_MASK_Q(obits)))

 // EMIT_RULE: VL_ASSIGNCLEAN:  oclean=clean; obits==lbits;
 #define VL_ASSIGNCLEAN_W(obits, owp, lwp) VL_CLEAN_WW((obits), (owp), (lwp))
@@ -356,18 +352,16 @@ static inline WDataOutP _vl_clean_inplace_w(int obits, WDataOutP owp) VL_MT_SAFE
 }
 static inline WDataOutP VL_CLEAN_WW(int obits, WDataOutP owp, WDataInP const lwp) VL_MT_SAFE {
    const int words = VL_WORDS_I(obits);
-    for (int i = 0; (i < (words - 1)); ++i) owp[i] = lwp[i];
+    VL_MEMCPY_W(owp, lwp, words - 1);
    owp[words - 1] = lwp[words - 1] & VL_MASK_E(obits);
    return owp;
 }
 static inline WDataOutP VL_ZERO_W(int obits, WDataOutP owp) VL_MT_SAFE {
-    const int words = VL_WORDS_I(obits);
-    for (int i = 0; i < words; ++i) owp[i] = 0;
-    return owp;
+    return VL_MEMSET_ZERO_W(owp, VL_WORDS_I(obits));
 }
 static inline WDataOutP VL_ALLONES_W(int obits, WDataOutP owp) VL_MT_SAFE {
    const int words = VL_WORDS_I(obits);
-    for (int i = 0; i < (words - 1); ++i) owp[i] = ~VL_EUL(0);
+    VL_MEMSET_ONES_W(owp, words - 1);
    owp[words - 1] = VL_MASK_E(obits);
    return owp;
 }
@@ -376,9 +370,7 @@ static inline WDataOutP VL_ALLONES_W(int obits, WDataOutP owp) VL_MT_SAFE {
 // For now, we always have a clean rhs.
 // Note: If a ASSIGN isn't clean, use VL_ASSIGNCLEAN instead to do the same thing.
 static inline WDataOutP VL_ASSIGN_W(int obits, WDataOutP owp, WDataInP const lwp) VL_MT_SAFE {
-    const int words = VL_WORDS_I(obits);
-    for (int i = 0; i < words; ++i) owp[i] = lwp[i];
-    return owp;
+    return VL_MEMCPY_W(owp, lwp, VL_WORDS_I(obits));
 }

 // EMIT_RULE: VL_ASSIGNBIT:  rclean=clean;
@@ -449,7 +441,7 @@ static inline void VL_ASSIGNBIT_WO(int bit, WDataOutP owp) VL_MT_SAFE {
 #define VL_ASSIGN_WSB(obits, owp, svar) \
    { \
        const int words = VL_WORDS_I(obits); \
-        sc_biguint<(obits)> _butemp = (svar).read(); \
+        sc_dt::sc_biguint<(obits)> _butemp = (svar).read(); \
        uint32_t* chunkp = _butemp.get_raw(); \
        int32_t lsb = 0; \
        while (lsb < obits - BITS_PER_DIGIT) { \
@@ -475,20 +467,20 @@ static inline void VL_ASSIGNBIT_WO(int bit, WDataOutP owp) VL_MT_SAFE {

 #define VL_ASSIGN_SWI(obits, svar, rd) \
    { \
-        sc_bv<(obits)> _bvtemp; \
+        sc_dt::sc_bv<(obits)> _bvtemp; \
        _bvtemp.set_word(0, (rd)); \
        (svar).write(_bvtemp); \
    }
 #define VL_ASSIGN_SWQ(obits, svar, rd) \
    { \
-        sc_bv<(obits)> _bvtemp; \
+        sc_dt::sc_bv<(obits)> _bvtemp; \
        _bvtemp.set_word(0, static_cast<IData>(rd)); \
        _bvtemp.set_word(1, static_cast<IData>((rd) >> VL_IDATASIZE)); \
        (svar).write(_bvtemp); \
    }
 #define VL_ASSIGN_SWW(obits, svar, rwp) \
    { \
-        sc_bv<(obits)> _bvtemp; \
+        sc_dt::sc_bv<(obits)> _bvtemp; \
        for (int i = 0; i < VL_WORDS_I(obits); ++i) _bvtemp.set_word(i, (rwp)[i]); \
        (svar).write(_bvtemp); \
    }
@@ -504,7 +496,7 @@ static inline void VL_ASSIGNBIT_WO(int bit, WDataOutP owp) VL_MT_SAFE {
 #define VL_SC_BITS_PER_DIGIT 30  // This comes from sc_nbdefs.h BITS_PER_DIGIT
 #define VL_ASSIGN_SBW(obits, svar, rwp) \
    { \
-        sc_biguint<(obits)> _butemp; \
+        sc_dt::sc_biguint<(obits)> _butemp; \
        int32_t lsb = 0; \
        uint32_t* chunkp = _butemp.get_raw(); \
        while (lsb + VL_SC_BITS_PER_DIGIT < (obits)) { \
@@ -536,19 +528,20 @@ static inline void VL_ASSIGNBIT_WO(int bit, WDataOutP owp) VL_MT_SAFE {
 static inline WDataOutP VL_EXTEND_WI(int obits, int, WDataOutP owp, IData ld) VL_MT_SAFE {
    // Note for extracts that obits != lbits
    owp[0] = ld;
-    for (int i = 1; i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
    return owp;
 }
 static inline WDataOutP VL_EXTEND_WQ(int obits, int, WDataOutP owp, QData ld) VL_MT_SAFE {
    VL_SET_WQ(owp, ld);
-    for (int i = VL_WQ_WORDS_E; i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    VL_MEMSET_ZERO_W(owp + VL_WQ_WORDS_E, VL_WORDS_I(obits) - VL_WQ_WORDS_E);
    return owp;
 }
 static inline WDataOutP VL_EXTEND_WW(int obits, int lbits, WDataOutP owp,
                                     WDataInP const lwp) VL_MT_SAFE {
-    for (int i = 0; i < VL_WORDS_I(lbits); ++i) owp[i] = lwp[i];
-    for (int i = VL_WORDS_I(lbits); i < VL_WORDS_I(obits); ++i) owp[i] = 0;
-    return owp;
+    const int lwords = VL_WORDS_I(lbits);
+    VL_PREFETCH_RD(lwp);
+    VL_MEMSET_ZERO_W(owp + lwords, VL_WORDS_I(obits) - lwords);
+    return VL_MEMCPY_W(owp, lwp, lwords);
 }

 // EMIT_RULE: VL_EXTENDS:  oclean=*dirty*; obits=lbits;
@@ -564,26 +557,37 @@ static inline QData VL_EXTENDS_QQ(int, int lbits, QData lhs) VL_PURE {
 }

 static inline WDataOutP VL_EXTENDS_WI(int obits, int lbits, WDataOutP owp, IData ld) VL_MT_SAFE {
-    const EData sign = VL_SIGNONES_E(lbits, static_cast<EData>(ld));
-    owp[0] = ld | (sign & ~VL_MASK_E(lbits));
-    for (int i = 1; i < VL_WORDS_I(obits); ++i) owp[i] = sign;
+    owp[0] = ld;
+    if (VL_SIGN_E(lbits, owp[0])) {
+        owp[0] |= ~VL_MASK_E(lbits);
+        VL_MEMSET_ONES_W(owp + 1, VL_WORDS_I(obits) - 1);
+    } else {
+        VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
+    }
    return owp;
 }
 static inline WDataOutP VL_EXTENDS_WQ(int obits, int lbits, WDataOutP owp, QData ld) VL_MT_SAFE {
    VL_SET_WQ(owp, ld);
-    const EData sign = VL_SIGNONES_E(lbits, owp[1]);
-    owp[1] |= sign & ~VL_MASK_E(lbits);
-    for (int i = VL_WQ_WORDS_E; i < VL_WORDS_I(obits); ++i) owp[i] = sign;
+    if (VL_SIGN_E(lbits, owp[1])) {
+        owp[1] |= ~VL_MASK_E(lbits);
+        VL_MEMSET_ONES_W(owp + VL_WQ_WORDS_E, VL_WORDS_I(obits) - VL_WQ_WORDS_E);
+    } else {
+        VL_MEMSET_ZERO_W(owp + VL_WQ_WORDS_E, VL_WORDS_I(obits) - VL_WQ_WORDS_E);
+    }
    return owp;
 }
 static inline WDataOutP VL_EXTENDS_WW(int obits, int lbits, WDataOutP owp,
                                      WDataInP const lwp) VL_MT_SAFE {
-    for (int i = 0; i < VL_WORDS_I(lbits) - 1; ++i) owp[i] = lwp[i];
-    const int lmsw = VL_WORDS_I(lbits) - 1;
-    const EData sign = VL_SIGNONES_E(lbits, lwp[lmsw]);
-    owp[lmsw] = lwp[lmsw] | (sign & ~VL_MASK_E(lbits));
-    for (int i = VL_WORDS_I(lbits); i < VL_WORDS_I(obits); ++i) owp[i] = sign;
-    return owp;
+    const int lwords = VL_WORDS_I(lbits);
+    VL_PREFETCH_RD(lwp);
+    owp[lwords - 1] = lwp[lwords - 1];
+    if (VL_SIGN_E(lbits, lwp[lwords - 1])) {
+        owp[lwords - 1] |= ~VL_MASK_E(lbits);
+        VL_MEMSET_ONES_W(owp + lwords, VL_WORDS_I(obits) - lwords);
+    } else {
+        VL_MEMSET_ZERO_W(owp + lwords, VL_WORDS_I(obits) - lwords);
+    }
+    return VL_MEMCPY_W(owp, lwp, lwords - 1);
 }

 //===================================================================
@@ -1028,6 +1032,7 @@ static inline QData VL_MULS_QQQ(int lbits, QData lhs, QData rhs) VL_PURE {
 static inline WDataOutP VL_MULS_WWW(int lbits, WDataOutP owp, WDataInP const lwp,
                                    WDataInP const rwp) VL_MT_SAFE {
    const int words = VL_WORDS_I(lbits);
+    VL_DEBUG_IFDEF(assert(words <= VL_MULS_MAX_WORDS););
    // cppcheck-suppress variableScope
    WData lwstore[VL_MULS_MAX_WORDS];  // Fixed size, as MSVC++ doesn't allow [words] here
    // cppcheck-suppress variableScope
@@ -1098,21 +1103,22 @@ static inline QData VL_MODDIVS_QQQ(int lbits, QData lhs, QData rhs) VL_PURE {

 static inline WDataOutP VL_DIVS_WWW(int lbits, WDataOutP owp, WDataInP const lwp,
                                    WDataInP const rwp) VL_MT_SAFE {
-    const int words = VL_WORDS_I(lbits);
-    const EData lsign = VL_SIGN_E(lbits, lwp[words - 1]);
-    const EData rsign = VL_SIGN_E(lbits, rwp[words - 1]);
+    const int lwords = VL_WORDS_I(lbits);
+    const EData lsign = VL_SIGN_E(lbits, lwp[lwords - 1]);
+    const EData rsign = VL_SIGN_E(lbits, rwp[lwords - 1]);
+    VL_DEBUG_IFDEF(assert(lwords <= VL_MULS_MAX_WORDS););
    // cppcheck-suppress variableScope
    WData lwstore[VL_MULS_MAX_WORDS];  // Fixed size, as MSVC++ doesn't allow [words] here
    // cppcheck-suppress variableScope
    WData rwstore[VL_MULS_MAX_WORDS];
    WDataInP ltup = lwp;
    WDataInP rtup = rwp;
-    if (lsign) ltup = _vl_clean_inplace_w(lbits, VL_NEGATE_W(VL_WORDS_I(lbits), lwstore, lwp));
-    if (rsign) rtup = _vl_clean_inplace_w(lbits, VL_NEGATE_W(VL_WORDS_I(lbits), rwstore, rwp));
+    if (lsign) ltup = _vl_clean_inplace_w(lbits, VL_NEGATE_W(lwords, lwstore, lwp));
+    if (rsign) rtup = _vl_clean_inplace_w(lbits, VL_NEGATE_W(lwords, rwstore, rwp));
    if ((lsign && !rsign) || (!lsign && rsign)) {
        WData qNoSign[VL_MULS_MAX_WORDS];
        VL_DIV_WWW(lbits, qNoSign, ltup, rtup);
-        _vl_clean_inplace_w(lbits, VL_NEGATE_W(VL_WORDS_I(lbits), owp, qNoSign));
+        _vl_clean_inplace_w(lbits, VL_NEGATE_W(lwords, owp, qNoSign));
        return owp;
    } else {
        return VL_DIV_WWW(lbits, owp, ltup, rtup);
@@ -1120,21 +1126,22 @@ static inline WDataOutP VL_DIVS_WWW(int lbits, WDataOutP owp, WDataInP const lwp
 }
 static inline WDataOutP VL_MODDIVS_WWW(int lbits, WDataOutP owp, WDataInP const lwp,
                                       WDataInP const rwp) VL_MT_SAFE {
-    const int words = VL_WORDS_I(lbits);
-    const EData lsign = VL_SIGN_E(lbits, lwp[words - 1]);
-    const EData rsign = VL_SIGN_E(lbits, rwp[words - 1]);
+    const int lwords = VL_WORDS_I(lbits);
+    const EData lsign = VL_SIGN_E(lbits, lwp[lwords - 1]);
+    const EData rsign = VL_SIGN_E(lbits, rwp[lwords - 1]);
+    VL_DEBUG_IFDEF(assert(lwords <= VL_MULS_MAX_WORDS););
    // cppcheck-suppress variableScope
    WData lwstore[VL_MULS_MAX_WORDS];  // Fixed size, as MSVC++ doesn't allow [words] here
    // cppcheck-suppress variableScope
    WData rwstore[VL_MULS_MAX_WORDS];
    WDataInP ltup = lwp;
    WDataInP rtup = rwp;
-    if (lsign) ltup = _vl_clean_inplace_w(lbits, VL_NEGATE_W(VL_WORDS_I(lbits), lwstore, lwp));
-    if (rsign) rtup = _vl_clean_inplace_w(lbits, VL_NEGATE_W(VL_WORDS_I(lbits), rwstore, rwp));
+    if (lsign) ltup = _vl_clean_inplace_w(lbits, VL_NEGATE_W(lwords, lwstore, lwp));
+    if (rsign) rtup = _vl_clean_inplace_w(lbits, VL_NEGATE_W(lwords, rwstore, rwp));
    if (lsign) {  // Only dividend sign matters for modulus
        WData qNoSign[VL_MULS_MAX_WORDS];
        VL_MODDIV_WWW(lbits, qNoSign, ltup, rtup);
-        _vl_clean_inplace_w(lbits, VL_NEGATE_W(VL_WORDS_I(lbits), owp, qNoSign));
+        _vl_clean_inplace_w(lbits, VL_NEGATE_W(lwords, owp, qNoSign));
        return owp;
    } else {
        return VL_MODDIV_WWW(lbits, owp, ltup, rtup);
@@ -1437,7 +1444,7 @@ static inline IData VL_STREAML_FAST_III(int lbits, IData ld, IData rd_log2) VL_P
    if (rd_log2) {
        const uint32_t lbitsFloor = lbits & ~VL_MASK_I(rd_log2);  // max multiple of rd <= lbits
        const uint32_t lbitsRem = lbits - lbitsFloor;  // number of bits in most-sig slice (MSS)
-        const IData msbMask = VL_MASK_I(lbitsRem) << lbitsFloor;  // mask to sel only bits in MSS
+        const IData msbMask = lbitsFloor == 32 ? 0UL : VL_MASK_I(lbitsRem) << lbitsFloor;
        ret = (ret & ~msbMask) | ((ret & msbMask) << ((VL_UL(1) << rd_log2) - lbitsRem));
    }
    switch (rd_log2) {
@@ -1566,63 +1573,66 @@ static inline QData VL_DYN_TO_Q(const VlQueue<T>& q, int elem_size) {
 static inline WDataOutP VL_CONCAT_WII(int obits, int lbits, int rbits, WDataOutP owp, IData ld,
                                      IData rd) VL_MT_SAFE {
    owp[0] = rd;
-    for (int i = 1; i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
    _vl_insert_WI(owp, ld, rbits + lbits - 1, rbits);
    return owp;
 }
 static inline WDataOutP VL_CONCAT_WWI(int obits, int lbits, int rbits, WDataOutP owp,
                                      WDataInP const lwp, IData rd) VL_MT_SAFE {
    owp[0] = rd;
-    for (int i = 1; i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
    _vl_insert_WW(owp, lwp, rbits + lbits - 1, rbits);
    return owp;
 }
 static inline WDataOutP VL_CONCAT_WIW(int obits, int lbits, int rbits, WDataOutP owp, IData ld,
                                      WDataInP const rwp) VL_MT_SAFE {
-    for (int i = 0; i < VL_WORDS_I(rbits); ++i) owp[i] = rwp[i];
-    for (int i = VL_WORDS_I(rbits); i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    const int rwords = VL_WORDS_I(rbits);
+    VL_MEMCPY_W(owp, rwp, rwords);
+    VL_MEMSET_ZERO_W(owp + rwords, VL_WORDS_I(obits) - rwords);
    _vl_insert_WI(owp, ld, rbits + lbits - 1, rbits);
    return owp;
 }
 static inline WDataOutP VL_CONCAT_WIQ(int obits, int lbits, int rbits, WDataOutP owp, IData ld,
                                      QData rd) VL_MT_SAFE {
    VL_SET_WQ(owp, rd);
-    for (int i = VL_WQ_WORDS_E; i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    VL_MEMSET_ZERO_W(owp + VL_WQ_WORDS_E, VL_WORDS_I(obits) - VL_WQ_WORDS_E);
    _vl_insert_WI(owp, ld, rbits + lbits - 1, rbits);
    return owp;
 }
 static inline WDataOutP VL_CONCAT_WQI(int obits, int lbits, int rbits, WDataOutP owp, QData ld,
                                      IData rd) VL_MT_SAFE {
    owp[0] = rd;
-    for (int i = 1; i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    VL_MEMSET_ZERO_W(owp + 1, VL_WORDS_I(obits) - 1);
    _vl_insert_WQ(owp, ld, rbits + lbits - 1, rbits);
    return owp;
 }
 static inline WDataOutP VL_CONCAT_WQQ(int obits, int lbits, int rbits, WDataOutP owp, QData ld,
                                      QData rd) VL_MT_SAFE {
    VL_SET_WQ(owp, rd);
-    for (int i = VL_WQ_WORDS_E; i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    VL_MEMSET_ZERO_W(owp + VL_WQ_WORDS_E, VL_WORDS_I(obits) - VL_WQ_WORDS_E);
    _vl_insert_WQ(owp, ld, rbits + lbits - 1, rbits);
    return owp;
 }
 static inline WDataOutP VL_CONCAT_WWQ(int obits, int lbits, int rbits, WDataOutP owp,
                                      WDataInP const lwp, QData rd) VL_MT_SAFE {
    VL_SET_WQ(owp, rd);
-    for (int i = VL_WQ_WORDS_E; i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    VL_MEMSET_ZERO_W(owp + VL_WQ_WORDS_E, VL_WORDS_I(obits) - VL_WQ_WORDS_E);
    _vl_insert_WW(owp, lwp, rbits + lbits - 1, rbits);
    return owp;
 }
 static inline WDataOutP VL_CONCAT_WQW(int obits, int lbits, int rbits, WDataOutP owp, QData ld,
                                      WDataInP const rwp) VL_MT_SAFE {
-    for (int i = 0; i < VL_WORDS_I(rbits); ++i) owp[i] = rwp[i];
-    for (int i = VL_WORDS_I(rbits); i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    const int rwords = VL_WORDS_I(rbits);
+    VL_MEMCPY_W(owp, rwp, rwords);
+    VL_MEMSET_ZERO_W(owp + rwords, VL_WORDS_I(obits) - rwords);
    _vl_insert_WQ(owp, ld, rbits + lbits - 1, rbits);
    return owp;
 }
 static inline WDataOutP VL_CONCAT_WWW(int obits, int lbits, int rbits, WDataOutP owp,
                                      WDataInP const lwp, WDataInP const rwp) VL_MT_SAFE {
-    for (int i = 0; i < VL_WORDS_I(rbits); ++i) owp[i] = rwp[i];
-    for (int i = VL_WORDS_I(rbits); i < VL_WORDS_I(obits); ++i) owp[i] = 0;
+    const int rwords = VL_WORDS_I(rbits);
+    VL_MEMCPY_W(owp, rwp, rwords);
+    VL_MEMSET_ZERO_W(owp + rwords, VL_WORDS_I(obits) - rwords);
    _vl_insert_WW(owp, lwp, rbits + lbits - 1, rbits);
    return owp;
 }
@@ -1647,6 +1657,24 @@ static inline void _vl_shiftl_inplace_w(int obits, WDataOutP iowp,
 // EMIT_RULE: VL_SHIFTL:  oclean=lclean; rclean==clean;
 // Important: Unlike most other funcs, the shift might well be a computed
 // expression.  Thus consider this when optimizing.  (And perhaps have 2 funcs?)
+// If RHS (rd/rwp) is larger than the output, zeros (or all ones for >>>) must be returned
+// (This corresponds to AstShift*Ovr Ast nodes)
+static inline IData VL_SHIFTL_III(int obits, int, int, IData lhs, IData rhs) VL_MT_SAFE {
+    if (VL_UNLIKELY(rhs >= VL_IDATASIZE)) return 0;
+    return lhs << rhs;  // Small is common so not clean return
+}
+static inline IData VL_SHIFTL_IIQ(int obits, int, int, IData lhs, QData rhs) VL_MT_SAFE {
+    if (VL_UNLIKELY(rhs >= VL_IDATASIZE)) return 0;
+    return VL_CLEAN_II(obits, obits, lhs << rhs);
+}
+static inline QData VL_SHIFTL_QQI(int obits, int, int, QData lhs, IData rhs) VL_MT_SAFE {
+    if (VL_UNLIKELY(rhs >= VL_QUADSIZE)) return 0;
+    return lhs << rhs;  // Small is common so not clean return
+}
+static inline QData VL_SHIFTL_QQQ(int obits, int, int, QData lhs, QData rhs) VL_MT_SAFE {
+    if (VL_UNLIKELY(rhs >= VL_QUADSIZE)) return 0;
+    return VL_CLEAN_QQ(obits, obits, lhs << rhs);
+}
 static inline WDataOutP VL_SHIFTL_WWI(int obits, int, int, WDataOutP owp, WDataInP const lwp,
                                      IData rd) VL_MT_SAFE {
    const int word_shift = VL_BITWORD_E(rd);
@@ -1684,11 +1712,7 @@ static inline IData VL_SHIFTL_IIW(int obits, int, int rbits, IData lhs,
            return 0;
        }
    }
-    return VL_CLEAN_II(obits, obits, lhs << rwp[0]);
-}
-static inline IData VL_SHIFTL_IIQ(int obits, int, int, IData lhs, QData rhs) VL_MT_SAFE {
-    if (VL_UNLIKELY(rhs >= VL_IDATASIZE)) return 0;
-    return VL_CLEAN_II(obits, obits, lhs << rhs);
+    return VL_SHIFTL_III(obits, obits, 32, lhs, rwp[0]);
 }
 static inline QData VL_SHIFTL_QQW(int obits, int, int rbits, QData lhs,
                                  WDataInP const rwp) VL_MT_SAFE {
@@ -1698,16 +1722,28 @@ static inline QData VL_SHIFTL_QQW(int obits, int, int rbits, QData lhs,
        }
    }
    // Above checks rwp[1]==0 so not needed in below shift
-    return VL_CLEAN_QQ(obits, obits, lhs << (static_cast<QData>(rwp[0])));
-}
-static inline QData VL_SHIFTL_QQQ(int obits, int, int, QData lhs, QData rhs) VL_MT_SAFE {
-    if (VL_UNLIKELY(rhs >= VL_QUADSIZE)) return 0;
-    return VL_CLEAN_QQ(obits, obits, lhs << rhs);
+    return VL_SHIFTL_QQI(obits, obits, 32, lhs, rwp[0]);
 }

 // EMIT_RULE: VL_SHIFTR:  oclean=lclean; rclean==clean;
 // Important: Unlike most other funcs, the shift might well be a computed
 // expression.  Thus consider this when optimizing.  (And perhaps have 2 funcs?)
+static inline IData VL_SHIFTR_III(int obits, int, int, IData lhs, IData rhs) VL_PURE {
+    if (VL_UNLIKELY(rhs >= VL_IDATASIZE)) return 0;
+    return lhs >> rhs;  // Small is common so assumed not clean
+}
+static inline IData VL_SHIFTR_IIQ(int obits, int, int, IData lhs, QData rhs) VL_PURE {
+    if (VL_UNLIKELY(rhs >= VL_IDATASIZE)) return 0;
+    return VL_CLEAN_QQ(obits, obits, lhs >> rhs);
+}
+static inline QData VL_SHIFTR_QQI(int obits, int, int, QData lhs, IData rhs) VL_PURE {
+    if (VL_UNLIKELY(rhs >= VL_QUADSIZE)) return 0;
+    return lhs >> rhs;  // Small is common so assumed not clean
+}
+static inline QData VL_SHIFTR_QQQ(int obits, int, int, QData lhs, QData rhs) VL_PURE {
+    if (VL_UNLIKELY(rhs >= VL_QUADSIZE)) return 0;
+    return VL_CLEAN_QQ(obits, obits, lhs >> rhs);
+}
 static inline WDataOutP VL_SHIFTR_WWI(int obits, int, int, WDataOutP owp, WDataInP const lwp,
                                      IData rd) VL_MT_SAFE {
    const int word_shift = VL_BITWORD_E(rd);  // Maybe 0
@@ -1751,29 +1787,16 @@ static inline WDataOutP VL_SHIFTR_WWQ(int obits, int lbits, int rbits, WDataOutP
 static inline IData VL_SHIFTR_IIW(int obits, int, int rbits, IData lhs,
                                  WDataInP const rwp) VL_PURE {
    for (int i = 1; i < VL_WORDS_I(rbits); ++i) {
-        if (VL_UNLIKELY(rwp[i])) {  // Huge shift 1>>32 or more
-            return 0;
-        }
+        if (VL_UNLIKELY(rwp[i])) return 0;  // Huge shift 1>>32 or more
    }
-    return VL_CLEAN_II(obits, obits, lhs >> rwp[0]);
+    return VL_SHIFTR_III(obits, obits, 32, lhs, rwp[0]);
 }
 static inline QData VL_SHIFTR_QQW(int obits, int, int rbits, QData lhs,
                                  WDataInP const rwp) VL_PURE {
    for (int i = 1; i < VL_WORDS_I(rbits); ++i) {
-        if (VL_UNLIKELY(rwp[i])) {  // Huge shift 1>>32 or more
-            return 0;
-        }
+        if (VL_UNLIKELY(rwp[i])) return 0;  // Huge shift 1>>32 or more
    }
-    // Above checks rwp[1]==0 so not needed in below shift
-    return VL_CLEAN_QQ(obits, obits, lhs >> (static_cast<QData>(rwp[0])));
-}
-static inline IData VL_SHIFTR_IIQ(int obits, int, int, IData lhs, QData rhs) VL_PURE {
-    if (VL_UNLIKELY(rhs >= VL_IDATASIZE)) return 0;
-    return VL_CLEAN_QQ(obits, obits, lhs >> rhs);
-}
-static inline QData VL_SHIFTR_QQQ(int obits, int, int, QData lhs, QData rhs) VL_PURE {
-    if (VL_UNLIKELY(rhs >= VL_QUADSIZE)) return 0;
-    return VL_CLEAN_QQ(obits, obits, lhs >> rhs);
+    return VL_SHIFTR_QQI(obits, obits, 32, lhs, rwp[0]);
 }

 // EMIT_RULE: VL_SHIFTRS:  oclean=false; lclean=clean, rclean==clean;
@@ -1783,11 +1806,13 @@ static inline IData VL_SHIFTRS_III(int obits, int lbits, int, IData lhs, IData r
    // must use lbits for sign; lbits might != obits,
    // an EXTEND(SHIFTRS(...)) can became a SHIFTRS(...) within same 32/64 bit word length
    const IData sign = -(lhs >> (lbits - 1));  // ffff_ffff if negative
+    if (VL_UNLIKELY(rhs >= VL_IDATASIZE)) return sign & VL_MASK_I(obits);
    const IData signext = ~(VL_MASK_I(lbits) >> rhs);  // One with bits where we've shifted "past"
    return (lhs >> rhs) | (sign & VL_CLEAN_II(obits, obits, signext));
 }
 static inline QData VL_SHIFTRS_QQI(int obits, int lbits, int, QData lhs, IData rhs) VL_PURE {
    const QData sign = -(lhs >> (lbits - 1));
+    if (VL_UNLIKELY(rhs >= VL_QUADSIZE)) return sign & VL_MASK_Q(obits);
    const QData signext = ~(VL_MASK_Q(lbits) >> rhs);
    return (lhs >> rhs) | (sign & VL_CLEAN_QQ(obits, obits, signext));
 }
@@ -1831,10 +1856,13 @@ static inline WDataOutP VL_SHIFTRS_WWW(int obits, int lbits, int rbits, WDataOut
    EData overshift = 0;  // Huge shift 1>>32 or more
    for (int i = 1; i < VL_WORDS_I(rbits); ++i) overshift |= rwp[i];
    if (VL_UNLIKELY(overshift || rwp[0] >= static_cast<IData>(obits))) {
-        const int lmsw = VL_WORDS_I(obits) - 1;
-        const EData sign = VL_SIGNONES_E(lbits, lwp[lmsw]);
-        for (int j = 0; j <= lmsw; ++j) owp[j] = sign;
-        owp[lmsw] &= VL_MASK_E(lbits);
+        const int owords = VL_WORDS_I(obits);
+        if (VL_SIGN_E(lbits, lwp[owords - 1])) {
+            VL_MEMSET_ONES_W(owp, owords);
+            owp[owords - 1] &= VL_MASK_E(lbits);
+        } else {
+            VL_MEMSET_ZERO_W(owp, owords);
+        }
        return owp;
    }
    return VL_SHIFTRS_WWI(obits, lbits, 32, owp, lwp, rwp[0]);
@@ -2045,9 +2073,7 @@ static inline void VL_ASSIGNSEL_WW(int rbits, int obits, int lsb, WDataOutP iowp

 static inline WDataOutP VL_COND_WIWW(int obits, WDataOutP owp, int cond, WDataInP const w1p,
                                     WDataInP const w2p) VL_MT_SAFE {
-    const int words = VL_WORDS_I(obits);
-    for (int i = 0; i < words; ++i) owp[i] = cond ? w1p[i] : w2p[i];
-    return owp;
+    return VL_MEMCPY_W(owp, cond ? w1p : w2p, VL_WORDS_I(obits));
 }

 //======================================================================
@@ -2060,7 +2086,7 @@ static inline WDataOutP VL_COND_WIWW(int obits, WDataOutP owp, int cond, WDataIn
 // If changing the number of functions here, also change EMITCINLINES_NUM_CONSTW

 #define VL_C_END_(obits, wordsSet) \
-    for (int i = (wordsSet); i < VL_WORDS_I(obits); ++i) o[i] = 0; \
+    VL_MEMSET_ZERO_W(o + (wordsSet), VL_WORDS_I(obits) - (wordsSet)); \
    return o

 // clang-format off
@@ -2111,60 +2137,60 @@ static inline WDataOutP VL_CONST_W_8X(int obits, WDataOutP o,
    VL_C_END_(obits, 8);
 }
 //
-static inline WDataOutP VL_CONSTHI_W_1X(int obits, int lsb, WDataOutP obase,
+static inline WDataOutP VL_CONSTHI_W_1X(int obits, int lsb, WDataOutP o,
                                        EData d0) VL_MT_SAFE {
-    WDataOutP o = obase + VL_WORDS_I(lsb);
-    o[0] = d0;
+    WDataOutP ohi = o + VL_WORDS_I(lsb);
+    ohi[0] = d0;
    VL_C_END_(obits, VL_WORDS_I(lsb) + 1);
 }
-static inline WDataOutP VL_CONSTHI_W_2X(int obits, int lsb, WDataOutP obase,
+static inline WDataOutP VL_CONSTHI_W_2X(int obits, int lsb, WDataOutP o,
                                        EData d1, EData d0) VL_MT_SAFE {
-    WDataOutP o = obase + VL_WORDS_I(lsb);
-    o[0] = d0;  o[1] = d1;
+    WDataOutP ohi = o + VL_WORDS_I(lsb);
+    ohi[0] = d0;  ohi[1] = d1;
    VL_C_END_(obits, VL_WORDS_I(lsb) + 2);
 }
-static inline WDataOutP VL_CONSTHI_W_3X(int obits, int lsb, WDataOutP obase,
+static inline WDataOutP VL_CONSTHI_W_3X(int obits, int lsb, WDataOutP o,
                                        EData d2, EData d1, EData d0) VL_MT_SAFE {
-    WDataOutP o = obase + VL_WORDS_I(lsb);
-    o[0] = d0;  o[1] = d1;  o[2] = d2;
+    WDataOutP ohi = o + VL_WORDS_I(lsb);
+    ohi[0] = d0;  ohi[1] = d1;  ohi[2] = d2;
    VL_C_END_(obits, VL_WORDS_I(lsb) + 3);
 }
-static inline WDataOutP VL_CONSTHI_W_4X(int obits, int lsb, WDataOutP obase,
+static inline WDataOutP VL_CONSTHI_W_4X(int obits, int lsb, WDataOutP o,
                                        EData d3, EData d2, EData d1, EData d0) VL_MT_SAFE {
-    WDataOutP o = obase + VL_WORDS_I(lsb);
-    o[0] = d0;  o[1] = d1;  o[2] = d2;  o[3] = d3;
+    WDataOutP ohi = o + VL_WORDS_I(lsb);
+    ohi[0] = d0;  ohi[1] = d1;  ohi[2] = d2;  ohi[3] = d3;
    VL_C_END_(obits, VL_WORDS_I(lsb) + 4);
 }
-static inline WDataOutP VL_CONSTHI_W_5X(int obits, int lsb, WDataOutP obase,
+static inline WDataOutP VL_CONSTHI_W_5X(int obits, int lsb, WDataOutP o,
                                        EData d4,
                                        EData d3, EData d2, EData d1, EData d0) VL_MT_SAFE {
-    WDataOutP o = obase + VL_WORDS_I(lsb);
-    o[0] = d0;  o[1] = d1;  o[2] = d2;  o[3] = d3;
-    o[4] = d4;
+    WDataOutP ohi = o + VL_WORDS_I(lsb);
+    ohi[0] = d0;  ohi[1] = d1;  ohi[2] = d2;  ohi[3] = d3;
+    ohi[4] = d4;
    VL_C_END_(obits, VL_WORDS_I(lsb) + 5);
 }
-static inline WDataOutP VL_CONSTHI_W_6X(int obits, int lsb, WDataOutP obase,
+static inline WDataOutP VL_CONSTHI_W_6X(int obits, int lsb, WDataOutP o,
                                        EData d5, EData d4,
                                        EData d3, EData d2, EData d1, EData d0) VL_MT_SAFE {
-    WDataOutP o = obase + VL_WORDS_I(lsb);
-    o[0] = d0;  o[1] = d1;  o[2] = d2;  o[3] = d3;
-    o[4] = d4;  o[5] = d5;
+    WDataOutP ohi = o + VL_WORDS_I(lsb);
+    ohi[0] = d0;  ohi[1] = d1;  ohi[2] = d2;  ohi[3] = d3;
+    ohi[4] = d4;  ohi[5] = d5;
    VL_C_END_(obits, VL_WORDS_I(lsb) + 6);
 }
-static inline WDataOutP VL_CONSTHI_W_7X(int obits, int lsb, WDataOutP obase,
+static inline WDataOutP VL_CONSTHI_W_7X(int obits, int lsb, WDataOutP o,
                                        EData d6, EData d5, EData d4,
                                        EData d3, EData d2, EData d1, EData d0) VL_MT_SAFE {
-    WDataOutP o = obase + VL_WORDS_I(lsb);
-    o[0] = d0;  o[1] = d1;  o[2] = d2;  o[3] = d3;
-    o[4] = d4;  o[5] = d5;  o[6] = d6;
+    WDataOutP ohi = o + VL_WORDS_I(lsb);
+    ohi[0] = d0;  ohi[1] = d1;  ohi[2] = d2;  ohi[3] = d3;
+    ohi[4] = d4;  ohi[5] = d5;  ohi[6] = d6;
    VL_C_END_(obits, VL_WORDS_I(lsb) + 7);
 }
-static inline WDataOutP VL_CONSTHI_W_8X(int obits, int lsb, WDataOutP obase,
+static inline WDataOutP VL_CONSTHI_W_8X(int obits, int lsb, WDataOutP o,
                                        EData d7, EData d6, EData d5, EData d4,
                                        EData d3, EData d2, EData d1, EData d0) VL_MT_SAFE {
-    WDataOutP o = obase + VL_WORDS_I(lsb);
-    o[0] = d0;  o[1] = d1;  o[2] = d2;  o[3] = d3;
-    o[4] = d4;  o[5] = d5;  o[6] = d6;  o[7] = d7;
+    WDataOutP ohi = o + VL_WORDS_I(lsb);
+    ohi[0] = d0;  ohi[1] = d1;  ohi[2] = d2;  ohi[3] = d3;
+    ohi[4] = d4;  ohi[5] = d5;  ohi[6] = d6;  ohi[7] = d7;
    VL_C_END_(obits, VL_WORDS_I(lsb) + 8);
 }

@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2009-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2009-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2012-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2012-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -161,12 +161,18 @@ void VlExecutionProfiler::dump(const char* filenamep, uint64_t tickEnd)

    // TODO Perhaps merge with verilated_coverage output format, so can
    // have a common merging and reporting tool, etc.
-    fprintf(fp, "VLPROFVERSION 2.0 # Verilator execution profile version 2.0\n");
+    fprintf(fp, "VLPROFVERSION 2.1 # Verilator execution profile version 2.1\n");
    fprintf(fp, "VLPROF arg +verilator+prof+exec+start+%" PRIu64 "\n",
            Verilated::threadContextp()->profExecStart());
    fprintf(fp, "VLPROF arg +verilator+prof+exec+window+%u\n",
            Verilated::threadContextp()->profExecWindow());
-    const unsigned threads = static_cast<unsigned>(m_traceps.size());
+    // Note that VerilatedContext will by default create as many threads as there are hardware
+    // processors, but not all of them might be utilized. Report the actual number that has trace
+    // entries to avoid over-counting.
+    unsigned threads = 0;
+    for (const auto& pair : m_traceps) {
+        if (!pair.second->empty()) ++threads;
+    }
    fprintf(fp, "VLPROF stat threads %u\n", threads);
    fprintf(fp, "VLPROF stat yields %" PRIu64 "\n", VlMTaskVertex::yields());

@@ -183,6 +189,7 @@ void VlExecutionProfiler::dump(const char* filenamep, uint64_t tickEnd)
    for (const auto& pair : m_traceps) {
        const uint32_t threadId = pair.first;
        ExecutionTrace* const tracep = pair.second;
+        if (tracep->empty()) continue;
        fprintf(fp, "VLPROFTHREAD %" PRIu32 "\n", threadId);

        for (const VlExecutionRecord& er : *tracep) {
@@ -191,10 +198,9 @@ void VlExecutionProfiler::dump(const char* filenamep, uint64_t tickEnd)
            fprintf(fp, "VLPROFEXEC %s %" PRIu64, name, time);

            switch (er.m_type) {
-            case VlExecutionRecord::Type::EVAL_BEGIN:
-            case VlExecutionRecord::Type::EVAL_END:
-            case VlExecutionRecord::Type::EVAL_LOOP_BEGIN:
-            case VlExecutionRecord::Type::EVAL_LOOP_END:
+            case VlExecutionRecord::Type::SECTION_POP:
+            case VlExecutionRecord::Type::EXEC_GRAPH_BEGIN:
+            case VlExecutionRecord::Type::EXEC_GRAPH_END:
                // No payload
                fprintf(fp, "\n");
                break;
@@ -209,6 +215,11 @@ void VlExecutionProfiler::dump(const char* filenamep, uint64_t tickEnd)
                fprintf(fp, " id %u predictCost %u\n", payload.m_id, payload.m_predictCost);
                break;
            }
+            case VlExecutionRecord::Type::SECTION_PUSH: {
+                const auto& payload = er.m_payload.sectionPush;
+                fprintf(fp, " %s\n", payload.m_name);
+                break;
+            }
            default: abort();  // LCOV_EXCL_LINE
            }
        }
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2012-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2012-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -58,12 +58,12 @@ VL_ATTR_ALWINLINE QData VL_CPU_TICK() {

 // clang-format off
 #define FOREACH_VlExecutionRecord_TYPE(macro) \
-    _VL_FOREACH_APPLY(macro, EVAL_BEGIN) \
-    _VL_FOREACH_APPLY(macro, EVAL_END) \
-    _VL_FOREACH_APPLY(macro, EVAL_LOOP_BEGIN) \
-    _VL_FOREACH_APPLY(macro, EVAL_LOOP_END) \
+    _VL_FOREACH_APPLY(macro, SECTION_PUSH) \
+    _VL_FOREACH_APPLY(macro, SECTION_POP) \
    _VL_FOREACH_APPLY(macro, MTASK_BEGIN) \
-    _VL_FOREACH_APPLY(macro, MTASK_END)
+    _VL_FOREACH_APPLY(macro, MTASK_END) \
+    _VL_FOREACH_APPLY(macro, EXEC_GRAPH_BEGIN) \
+    _VL_FOREACH_APPLY(macro, EXEC_GRAPH_END)
 // clang-format on

 class VlExecutionRecord final {
@@ -83,6 +83,9 @@ class VlExecutionRecord final {
    };

    union Payload {
+        struct {
+            const char* m_name;  // Name of section being entered
+        } sectionPush;
        struct {
            uint32_t m_id;  // MTask id
            uint32_t m_predictStart;  // Time scheduler predicted would start
@@ -109,10 +112,11 @@ public:
    VlExecutionRecord() = default;

    // METHODS
-    void evalBegin() { m_type = Type::EVAL_BEGIN; }
-    void evalEnd() { m_type = Type::EVAL_END; }
-    void evalLoopBegin() { m_type = Type::EVAL_LOOP_BEGIN; }
-    void evalLoopEnd() { m_type = Type::EVAL_LOOP_END; }
+    void sectionPush(const char* name) {
+        m_payload.sectionPush.m_name = name;
+        m_type = Type::SECTION_PUSH;
+    }
+    void sectionPop() { m_type = Type::SECTION_POP; }
    void mtaskBegin(uint32_t id, uint32_t predictStart) {
        m_payload.mtaskBegin.m_id = id;
        m_payload.mtaskBegin.m_predictStart = predictStart;
@@ -124,6 +128,8 @@ public:
        m_payload.mtaskEnd.m_predictCost = predictCost;
        m_type = Type::MTASK_END;
    }
+    void execGraphBegin() { m_type = Type::EXEC_GRAPH_BEGIN; }
+    void execGraphEnd() { m_type = Type::EXEC_GRAPH_END; }
 };

 static_assert(std::is_trivially_destructible<VlExecutionRecord>::value,
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2000-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2000-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2009-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2009-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -0,0 +1,212 @@
+// -*- mode: C++; c-file-style: "cc-mode" -*-
+//=============================================================================
+//
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you can
+// redistribute it and/or modify it under the terms of either the GNU
+// Lesser General Public License Version 3 or the Perl Artistic License
+// Version 2.0.
+// SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
+//
+//=============================================================================
+///
+/// \file
+/// \brief Verilated tracing for SystemC implementation code
+///
+///
+///
+//=============================================================================
+
+#ifndef VERILATOR_VERILATED_SC_TRACE_H_
+#define VERILATOR_VERILATED_SC_TRACE_H_
+
+#include "verilatedos.h"
+
+#include "verilated.h"
+#include "verilated_sc.h"
+
+#if SYSTEMC_VERSION < 20140417  // SystemC's simulation phase callback introduced in 2.3.1
+#define _VERILATOR_SYSTEMC_NO_PHASE_CALLBACK
+#endif
+
+//=============================================================================
+// VerilatedScTraceBase
+// Base class for VCD/FST trace format on SystemC
+// This is an internally used class - see VerilatedVcdSc and VerilatedFstSc for what to call from
+// applications
+//
+/// This class utilizes SystemC's callbacks, which allows to dump signals inside the Verilated
+/// module automatically as time advances.
+///
+/// For SystemC prior to 2.3.1, the only approach for being notified after each update is by adding
+/// a trace file (sc_trace_file) to the simulation context. And after this version the simulation
+/// phase callback approach has been introduced (sc_trace_file also utilizes this), which is
+/// presented only if it's enabled with the `--enable-phase-callbacks` option. However, when it's
+/// enabled with `--enable-phase-callbacks=tracing`, trace files will be therefore disabled, thus
+/// failing to provide its functionality.
+///
+/// To provide a universal way for tracing, the class attempts to register a phase callback first.
+/// If it fails (proving that the feature has been disabled), it'll use the trace file approach
+/// instead.
+
+class VerilatedScTraceBase VL_NOT_FINAL : private sc_core::sc_object,
+                                          private sc_core::sc_trace_file {
+    bool m_enableDeltaCycles = false;
+    bool m_traceFileAdded = false;
+    static void stubReportHandler(const sc_core::sc_report&, const sc_core::sc_actions&){};
+
+public:
+    void enableDeltaCycles(bool flag = true) {
+#ifndef _VERILATOR_SYSTEMC_NO_PHASE_CALLBACK
+        using namespace sc_core;
+        // Save old report handler before overriding it
+        const auto oldHandler = sc_report_handler::get_handler();
+        // Override the old handler to hide 'phase callbacks not enabled' message
+        sc_report_handler::set_handler(&stubReportHandler);
+        if (flag) {
+            // Register simulation phase callback for delta cycles
+            sc_object::register_simulation_phase_callback(SC_END_OF_UPDATE);
+        } else {
+            sc_object::unregister_simulation_phase_callback(SC_END_OF_UPDATE);
+        }
+        // Restore the old handler
+        sc_report_handler::set_handler(oldHandler);
+#endif
+        m_enableDeltaCycles = flag;
+    }
+
+protected:
+    VerilatedScTraceBase()
+        : sc_object(sc_core::sc_gen_unique_name("$$$$verilator_sc_trace$$$$"))
+        , sc_trace_file() {
+        registerTraceCallback();
+    };
+    ~VerilatedScTraceBase() override {
+#if SYSTEMC_VERSION >= 20140417  // remove_trace_file added in 2.3.1
+        // Phase callback is automatically unregistered in ~sc_object(). Only the trace file is
+        // needed to be removed here
+        if (m_traceFileAdded) simcontext()->remove_trace_file(this);
+#endif
+    };
+    void registerTraceCallback() {
+#ifndef _VERILATOR_SYSTEMC_NO_PHASE_CALLBACK
+        using namespace sc_core;
+        // Save old report handler before overriding it
+        const auto oldHandler = sc_report_handler::get_handler();
+        // Override the old handler to hide 'phase callbacks not enabled' message
+        sc_report_handler::set_handler(&stubReportHandler);
+        // Register regular simulation phase (non-delta cycle) callback
+        phase_cb_mask cb_mask = sc_object::register_simulation_phase_callback(SC_BEFORE_TIMESTEP);
+        if (cb_mask == SC_UNITIALIZED) {
+#endif
+            // Phase callback not enabled, use trace file instead
+            simcontext()->add_trace_file(this);
+            m_traceFileAdded = true;
+#ifndef _VERILATOR_SYSTEMC_NO_PHASE_CALLBACK
+        }
+        // Restore the old handler
+        sc_report_handler::set_handler(oldHandler);
+#endif
+    }
+    static std::string getScTimeUnit() {
+        // We want to avoid a depreciated warning, but still be back compatible.
+        // Turning off the message just for this still results in an
+        // annoying "to turn off" message.
+        const sc_core::sc_time t1sec{1, sc_core::SC_SEC};
+        if (t1sec.to_default_time_units() == 0) {
+            VL_FATAL_MT(__FILE__, __LINE__, "",  // LCOV_EXCL_LINE
+                        "Cannot to get valid SystemC default time unit for trace file");
+        }
+        const sc_core::sc_time tunits{1.0 / t1sec.to_default_time_units(), sc_core::SC_SEC};
+        return tunits.to_string();
+    }
+    static std::string getScTimeResolution() {
+        return sc_core::sc_get_time_resolution().to_string();
+    }
+    static void checkScElaborationDone() {
+        if (!sc_core::sc_get_curr_simcontext()->elaboration_done()) {
+            Verilated::scTraceBeforeElaborationError();
+        }
+    }
+
+    // METHODS - for SC kernel
+#ifndef _VERILATOR_SYSTEMC_NO_PHASE_CALLBACK
+    // Override sc_object. Called if using phase callback
+    void simulation_phase_callback() final { cycle(); }
+#endif
+    // Override sc_trace_file. Called if using trace file
+    void cycle(bool delta_cycle) final {
+        if (!delta_cycle || m_enableDeltaCycles) cycle();
+    }
+    // METHODS - callbacks
+    // Subclasses should implement this callback method
+    virtual void cycle() = 0;
+
+private:
+    // METHODS - Fake outs for linker
+    // LCOV_EXCL_START
+
+#ifdef NC_SYSTEMC
+    // Cadence Incisive has these as abstract functions so we must create them
+    void set_time_unit(int exponent10_seconds) override {}  // deprecated
+#endif
+    void set_time_unit(double v, sc_core::sc_time_unit tu) override {}  // LCOV_EXCL_LINE
+
+    //--------------------------------------------------
+    // SystemC 2.1.v1
+
+    void write_comment(const std::string&) override {}
+    void trace(const unsigned int&, const std::string&, const char**) override {}
+
+#define DECL_TRACE_METHOD_A(tp) \
+    void trace(const tp& object, const std::string& name) override {}
+#define DECL_TRACE_METHOD_B(tp) \
+    void trace(const tp& object, const std::string& name, int width) override {}
+
+    // clang-format off
+    // Formatting matches that of sc_trace.h
+#if SYSTEMC_VERSION >= 20171012  // SystemC >= 2.3.2
+    DECL_TRACE_METHOD_A( sc_core::sc_event )
+    DECL_TRACE_METHOD_A( sc_core::sc_time )
+#endif
+
+    DECL_TRACE_METHOD_A( bool )
+    DECL_TRACE_METHOD_A( sc_dt::sc_bit )
+    DECL_TRACE_METHOD_A( sc_dt::sc_logic )
+
+    DECL_TRACE_METHOD_B( unsigned char )
+    DECL_TRACE_METHOD_B( unsigned short )
+    DECL_TRACE_METHOD_B( unsigned int )
+    DECL_TRACE_METHOD_B( unsigned long )
+    DECL_TRACE_METHOD_B( char )
+    DECL_TRACE_METHOD_B( short )
+    DECL_TRACE_METHOD_B( int )
+    DECL_TRACE_METHOD_B( long )
+    DECL_TRACE_METHOD_B( sc_dt::int64 )
+    DECL_TRACE_METHOD_B( sc_dt::uint64 )
+
+    DECL_TRACE_METHOD_A( float )
+    DECL_TRACE_METHOD_A( double )
+    DECL_TRACE_METHOD_A( sc_dt::sc_int_base )
+    DECL_TRACE_METHOD_A( sc_dt::sc_uint_base )
+    DECL_TRACE_METHOD_A( sc_dt::sc_signed )
+    DECL_TRACE_METHOD_A( sc_dt::sc_unsigned )
+
+    DECL_TRACE_METHOD_A( sc_dt::sc_fxval )
+    DECL_TRACE_METHOD_A( sc_dt::sc_fxval_fast )
+    DECL_TRACE_METHOD_A( sc_dt::sc_fxnum )
+    DECL_TRACE_METHOD_A( sc_dt::sc_fxnum_fast )
+
+    DECL_TRACE_METHOD_A( sc_dt::sc_bv_base )
+    DECL_TRACE_METHOD_A( sc_dt::sc_lv_base )
+    // LCOV_EXCL_STOP
+    // clang-format on
+
+#undef DECL_TRACE_METHOD_A
+#undef DECL_TRACE_METHOD_B
+};
+
+#ifdef _VERILATOR_SYSTEMC_NO_PHASE_CALLBACK
+#undef _VERILATOR_SYSTEMC_NO_PHASE_CALLBACK
+#endif
+
+#endif  // Guard
@@ -4,7 +4,7 @@
 //
 //*************************************************************************
 //
-// Copyright 2022-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2022-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU Lesser
 // General Public License Version 3 or the Perl Artistic License Version 2.0.
 // SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0
@@ -47,7 +47,7 @@ package std;
      endtask

      function int try_put(T message);
-         if (num() < m_bound) begin
+         if (m_bound == 0 || num() < m_bound) begin
            m_queue.push_back(message);
            return 1;
         end
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2003-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2003-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2012-2023 by Wilson Snyder. This program is free software; you can
+// Copyright 2012-2024 by Wilson Snyder. This program is free software; you can
 // redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2012-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2012-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -53,36 +53,37 @@ void VlCoroutineHandle::dump() const {
 //======================================================================
 // VlDelayScheduler:: Methods

-#ifdef VL_DEBUG
-void VlDelayScheduler::VlDelayedCoroutine::dump() const {
-    VL_DBG_MSGF("             Awaiting time %" PRIu64 ": ", m_timestep);
-    m_handle.dump();
-}
-#endif
-
 void VlDelayScheduler::resume() {
 #ifdef VL_DEBUG
    VL_DEBUG_IF(dump(); VL_DBG_MSGF("         Resuming delayed processes\n"););
 #endif
-    while (awaitingCurrentTime()) {
-        if (m_queue.front().m_timestep != m_context.time()) {
-            VL_FATAL_MT(__FILE__, __LINE__, "",
-                        "%Error: Encountered process that should've been resumed at an "
-                        "earlier simulation time. Missed a time slot?");
-        }
-        // Move max element in the heap to the end
-        std::pop_heap(m_queue.begin(), m_queue.end());
-        VlCoroutineHandle handle = std::move(m_queue.back().m_handle);
-        m_queue.pop_back();
+    bool resumed = false;
+
+    while (!m_queue.empty() && (m_queue.begin()->first == m_context.time())) {
+        VlCoroutineHandle handle = std::move(m_queue.begin()->second);
+        m_queue.erase(m_queue.begin());
        handle.resume();
+        resumed = true;
+    }
+
+    if (!m_zeroDelayed.empty()) {
+        for (auto&& handle : m_zeroDelayed) handle.resume();
+        m_zeroDelayed.clear();
+        resumed = true;
+    }
+
+    if (!resumed) {
+        VL_FATAL_MT(__FILE__, __LINE__, "",
+                    "%Error: Encountered process that should've been resumed at an "
+                    "earlier simulation time. Missed a time slot?\n");
    }
 }

 uint64_t VlDelayScheduler::nextTimeSlot() const {
-    if (empty()) {
+    if (!m_queue.empty()) return m_queue.begin()->first;
+    if (m_zeroDelayed.empty())
        VL_FATAL_MT(__FILE__, __LINE__, "", "%Error: There is no next time slot scheduled");
-    }
-    return m_queue.front().m_timestep;
+    return m_context.time();
 }

 #ifdef VL_DEBUG
@@ -91,7 +92,16 @@ void VlDelayScheduler::dump() const {
        VL_DBG_MSGF("         No delayed processes:\n");
    } else {
        VL_DBG_MSGF("         Delayed processes:\n");
-        for (const auto& susp : m_queue) susp.dump();
+        for (auto& susp : m_zeroDelayed) {
+            VL_DBG_MSGF("             Awaiting #0-delayed resumption, "
+                        "time () %" PRIu64 ": ",
+                        m_context.time());
+            susp.dump();
+        }
+        for (const auto& susp : m_queue) {
+            VL_DBG_MSGF("             Awaiting time %" PRIu64 ": ", susp.first);
+            susp.second.dump();
+        }
    }
 }
 #endif
@@ -153,11 +163,12 @@ void VlTriggerScheduler::dump(const char* eventDescription) const {
 // VlDynamicTriggerScheduler:: Methods

 bool VlDynamicTriggerScheduler::evaluate() {
+    m_anyTriggered = false;
    VL_DEBUG_IF(dump(););
    std::swap(m_suspended, m_evaluated);
    for (auto& coro : m_evaluated) coro.resume();
    m_evaluated.clear();
-    return !m_triggered.empty();
+    return m_anyTriggered;
 }

 void VlDynamicTriggerScheduler::doPostUpdates() {
@@ -27,6 +27,8 @@

 #include "verilated.h"

+#include <vector>
+
 // clang-format off
 // Some preprocessor magic to support both Clang and GCC coroutines with both libc++ and libstdc++
 #if defined _LIBCPP_VERSION  // libc++
@@ -36,15 +38,15 @@
 # endif
 # include <experimental/coroutine>
  namespace std {
-      using namespace experimental; // Bring std::experimental into the std namespace
+      using namespace experimental;  // Bring std::experimental into the std namespace
  }
 #else
-# if defined __clang__ && defined __GLIBCXX__
+# if defined __clang__ && defined __GLIBCXX__ && !defined __cpp_impl_coroutine
 #  define __cpp_impl_coroutine 1  // Clang doesn't define this, but it's needed for libstdc++
 # endif
 # include <coroutine>
 # if __clang_major__ < 14
-   namespace std { // Bring coroutine library into std::experimental, as Clang < 14 expects it to be there
+   namespace std {  // Bring coroutine library into std::experimental, as Clang < 14 expects it to be there
       namespace experimental {
           using namespace std;
       }
@@ -148,29 +150,21 @@ public:
 #endif
 };

+enum class VlDelayPhase : bool { ACTIVE, INACTIVE };
+
 //=============================================================================
 // VlDelayScheduler stores coroutines to be resumed at a certain simulation time. If the current
 // time is equal to a coroutine's resume time, the coroutine gets resumed.

 class VlDelayScheduler final {
    // TYPES
-    struct VlDelayedCoroutine {
-        uint64_t m_timestep;  // Simulation time when the coroutine should be resumed
-        VlCoroutineHandle m_handle;  // The suspended coroutine to be resumed
-
-        // Comparison operator for std::push_heap(), std::pop_heap()
-        bool operator<(const VlDelayedCoroutine& other) const {
-            return m_timestep > other.m_timestep;
-        }
-#ifdef VL_DEBUG
-        void dump() const;
-#endif
-    };
-    using VlDelayedCoroutineQueue = std::vector<VlDelayedCoroutine>;
+    // Time-sorted queue of timestamps and handles
+    using VlDelayedCoroutineQueue = std::multimap<uint64_t, VlCoroutineHandle>;

    // MEMBERS
    VerilatedContext& m_context;
    VlDelayedCoroutineQueue m_queue;  // Coroutines to be restored at a certain simulation time
+    std::vector<VlCoroutineHandle> m_zeroDelayed;  // Coroutines waiting for #0

 public:
    // CONSTRUCTORS
@@ -183,10 +177,11 @@ public:
    // coroutines)
    uint64_t nextTimeSlot() const;
    // Are there no delayed coroutines awaiting?
-    bool empty() const { return m_queue.empty(); }
+    bool empty() const { return m_queue.empty() && m_zeroDelayed.empty(); }
    // Are there coroutines to resume at the current simulation time?
    bool awaitingCurrentTime() const {
-        return !empty() && m_queue.front().m_timestep <= m_context.time();
+        return (!m_queue.empty() && (m_queue.begin()->first <= m_context.time()))
+               || !m_zeroDelayed.empty();
    }
 #ifdef VL_DEBUG
    void dump() const;
@@ -197,19 +192,34 @@ public:
        struct Awaitable {
            VlProcessRef process;  // Data of the suspended process, null if not needed
            VlDelayedCoroutineQueue& queue;
+            std::vector<VlCoroutineHandle>& queueZeroDelay;
            uint64_t delay;
+            VlDelayPhase phase;
            VlFileLineDebug fileline;

            bool await_ready() const { return false; }  // Always suspend
            void await_suspend(std::coroutine_handle<> coro) {
-                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());
+                if (phase == VlDelayPhase::ACTIVE) {
+                    queue.emplace(delay, VlCoroutineHandle{coro, process, fileline});
+                } else {
+                    queueZeroDelay.emplace_back(VlCoroutineHandle{coro, process, fileline});
+                }
            }
            void await_resume() const {}
        };
-        return Awaitable{process, m_queue, m_context.time() + delay,
-                         VlFileLineDebug{filename, lineno}};
+
+        VlDelayPhase phase = (delay == 0) ? VlDelayPhase::INACTIVE : VlDelayPhase::ACTIVE;
+#ifdef VL_DEBUG
+        if (phase == VlDelayPhase::INACTIVE) {
+            VL_WARN_MT(filename, lineno, VL_UNKNOWN,
+                       "Encountered #0 delay. #0 scheduling support is incomplete and the "
+                       "process will be resumed before combinational logic evaluation.");
+        }
+#endif
+
+        return Awaitable{process,       m_queue,
+                         m_zeroDelayed, m_context.time() + delay,
+                         phase,         VlFileLineDebug{filename, lineno}};
    }
 };

@@ -275,6 +285,7 @@ public:
 //         co_await __VdynSched.evaluation();
 //         <pre updates>;
 //         __Vtrigger = <trigger eval>;
+//         __VdynShed.anyTriggered(__Vtrigger);
 //         [optionally] co_await __VdynSched.postUpdate();
 //         <post updates>;
 //     }
@@ -288,6 +299,7 @@ class VlDynamicTriggerScheduler final {
    using VlCoroutineVec = std::vector<VlCoroutineHandle>;

    // MEMBERS
+    bool m_anyTriggered = false;  // If true, at least one trigger was set
    VlCoroutineVec m_suspended;  // Suspended coroutines awaiting trigger evaluation
    VlCoroutineVec m_evaluated;  // Coroutines currently being evaluated (for evaluate())
    VlCoroutineVec m_triggered;  // Coroutines whose triggers were set, and are awaiting resumption
@@ -313,6 +325,8 @@ class VlDynamicTriggerScheduler final {
 public:
    // Evaluates all dynamic triggers (resumed coroutines that co_await evaluation())
    bool evaluate();
+    // Called by coroutines that evaluate triggers to notify the scheduler if any triggers were set
+    void anyTriggered(bool triggered) { m_anyTriggered = m_anyTriggered || triggered; }
    // Runs post updates for all dynamic triggers (resumes coroutines that co_await postUpdate())
    void doPostUpdates();
    // Resumes all coroutines whose triggers are set (those that co_await resumption())
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -25,7 +25,6 @@
 // clang-format off

 #include "verilated.h"
-#include "verilated_trace_defs.h"

 #include <bitset>
 #include <condition_variable>
@@ -46,6 +45,54 @@ class VerilatedTraceBuffer;
 template <class T_Buffer>
 class VerilatedTraceOffloadBuffer;

+//=============================================================================
+// Common enumerations
+
+enum class VerilatedTracePrefixType : uint32_t {
+    // Note: Entries must match VTracePrefixType (by name, not necessarily by value)
+    ARRAY_PACKED,
+    ARRAY_UNPACKED,
+    SCOPE_MODULE,
+    SCOPE_INTERFACE,
+    STRUCT_PACKED,
+    STRUCT_UNPACKED,
+    UNION_PACKED
+};
+
+// Direction attribute for ports
+enum class VerilatedTraceSigDirection : uint32_t {
+    NONE,
+    INPUT,
+    OUTPUT,
+    INOUT,
+};
+
+// Kind of signal. Similar to nettype but with a few more alternatives
+enum class VerilatedTraceSigKind : uint32_t {
+    PARAMETER,
+    SUPPLY0,
+    SUPPLY1,
+    TRI,
+    TRI0,
+    TRI1,
+    WIRE,
+    VAR,
+};
+
+// Base data type of signal
+enum class VerilatedTraceSigType : uint32_t {
+    DOUBLE,
+    INTEGER,
+    BIT,
+    LOGIC,
+    INT,
+    SHORTINT,
+    LONGINT,
+    BYTE,
+    EVENT,
+    TIME,
+};
+
 //=============================================================================
 // Offloaded tracing

@@ -162,24 +209,29 @@ private:
        // (the one in Ubuntu 14.04 with GCC 4.8.4 in particular) use the
        // assignment operator on inserting into collections, so they don't work
        // with const fields...
-        union {  // The callback
+        const union {  // The callback
            initCb_t m_initCb;
            dumpCb_t m_dumpCb;
            dumpOffloadCb_t m_dumpOffloadCb;
            cleanupCb_t m_cleanupCb;
        };
-        void* m_userp;  // The user pointer to pass to the callback (the symbol table)
+        const uint32_t m_fidx;  // The index of the tracing function
+        void* const m_userp;  // The user pointer to pass to the callback (the symbol table)
        CallbackRecord(initCb_t cb, void* userp)
            : m_initCb{cb}
+            , m_fidx{0}
            , m_userp{userp} {}
-        CallbackRecord(dumpCb_t cb, void* userp)
+        CallbackRecord(dumpCb_t cb, uint32_t fidx, void* userp)
            : m_dumpCb{cb}
+            , m_fidx{fidx}
            , m_userp{userp} {}
-        CallbackRecord(dumpOffloadCb_t cb, void* userp)
+        CallbackRecord(dumpOffloadCb_t cb, uint32_t fidx, void* userp)
            : m_dumpOffloadCb{cb}
+            , m_fidx{fidx}
            , m_userp{userp} {}
        CallbackRecord(cleanupCb_t cb, void* userp)
            : m_cleanupCb{cb}
+            , m_fidx{0}
            , m_userp{userp} {}
    };

@@ -212,18 +264,20 @@ protected:
 private:
    std::vector<bool> m_sigs_enabledVec;  // Staging for m_sigs_enabledp
    std::vector<CallbackRecord> m_initCbs;  // Routines to initialize tracing
+    std::vector<CallbackRecord> m_constCbs;  // Routines to perform const dump
+    std::vector<CallbackRecord> m_constOffloadCbs;  // Routines to perform offloaded const dump
    std::vector<CallbackRecord> m_fullCbs;  // Routines to perform full dump
    std::vector<CallbackRecord> m_fullOffloadCbs;  // Routines to perform offloaded full dump
    std::vector<CallbackRecord> m_chgCbs;  // Routines to perform incremental dump
    std::vector<CallbackRecord> m_chgOffloadCbs;  // Routines to perform offloaded incremental dump
    std::vector<CallbackRecord> m_cleanupCbs;  // Routines to call at the end of dump
+    bool m_constDump = true;  // Whether a const dump is required on the next call to 'dump'
    bool m_fullDump = true;  // Whether a full dump is required on the next call to 'dump'
    uint32_t m_nextCode = 0;  // Next code number to assign
    uint32_t m_numSignals = 0;  // Number of distinct signals
    uint32_t m_maxBits = 0;  // Number of bits in the widest signal
-    std::vector<std::string> m_namePrefixStack{""};  // Path prefixes to add to signal names
+    // TODO: Should keep this as a Trie, that is how it's accessed all the time.
    std::vector<std::pair<int, std::string>> m_dumpvars;  // dumpvar() entries
-    char m_scopeEscape = '.';
    double m_timeRes = 1e-9;  // Time resolution (ns/ms etc)
    double m_timeUnit = 1e-0;  // Time units (ns/ms etc)
    uint64_t m_timeLastDump = 0;  // Last time we did a dump
@@ -289,6 +343,7 @@ protected:
    uint32_t nextCode() const { return m_nextCode; }
    uint32_t numSignals() const { return m_numSignals; }
    uint32_t maxBits() const { return m_maxBits; }
+    void constDump(bool value) { m_constDump = value; }
    void fullDump(bool value) { m_fullDump = value; }

    double timeRes() const { return m_timeRes; }
@@ -298,14 +353,7 @@ protected:
    void traceInit() VL_MT_UNSAFE;

    // Declare new signal and return true if enabled
-    bool declCode(uint32_t code, const char* namep, uint32_t bits, bool tri);
-
-    // Is this an escape?
-    bool isScopeEscape(char c) { return std::isspace(c) || c == m_scopeEscape; }
-    // Character that splits scopes.  Note whitespace are ALWAYS escapes.
-    char scopeEscape() { return m_scopeEscape; }
-    // Prefix to assume in signal declarations
-    const std::string& namePrefix() const { return m_namePrefixStack.back(); }
+    bool declCode(uint32_t code, const std::string& declName, uint32_t bits);

    void closeBase();
    void flushBase();
@@ -313,6 +361,13 @@ protected:
    bool offload() const { return m_offload; }
    bool parallel() const { return m_parallel; }

+    // Return last ' ' separated word. Assumes string does not end in ' '.
+    static std::string lastWord(const std::string& str) {
+        const size_t idx = str.rfind(' ');
+        if (idx == std::string::npos) return str;
+        return str.substr(idx + 1);
+    }
+
    //=========================================================================
    // Virtual functions to be provided by the format specific implementation

@@ -325,7 +380,7 @@ protected:
    virtual bool preChangeDump() = 0;

    // Trace buffer management
-    virtual Buffer* getTraceBuffer() = 0;
+    virtual Buffer* getTraceBuffer(uint32_t fidx) = 0;
    virtual void commitTraceBuffer(Buffer*) = 0;

    // Configure sub-class
@@ -359,16 +414,13 @@ public:

    void addModel(VerilatedModel*) VL_MT_SAFE_EXCLUDES(m_mutex);
    void addInitCb(initCb_t cb, void* userp) VL_MT_SAFE;
-    void addFullCb(dumpCb_t cb, void* userp) VL_MT_SAFE;
-    void addFullCb(dumpOffloadCb_t cb, void* userp) VL_MT_SAFE;
-    void addChgCb(dumpCb_t cb, void* userp) VL_MT_SAFE;
-    void addChgCb(dumpOffloadCb_t cb, void* userp) VL_MT_SAFE;
+    void addConstCb(dumpCb_t cb, uint32_t fidx, void* userp) VL_MT_SAFE;
+    void addConstCb(dumpOffloadCb_t cb, uint32_t fidx, void* userp) VL_MT_SAFE;
+    void addFullCb(dumpCb_t cb, uint32_t fidx, void* userp) VL_MT_SAFE;
+    void addFullCb(dumpOffloadCb_t cb, uint32_t fidx, void* userp) VL_MT_SAFE;
+    void addChgCb(dumpCb_t cb, uint32_t fidx, void* userp) VL_MT_SAFE;
+    void addChgCb(dumpOffloadCb_t cb, uint32_t fidx, void* userp) VL_MT_SAFE;
    void addCleanupCb(cleanupCb_t cb, void* userp) VL_MT_SAFE;
-
-    void scopeEscape(char flag) { m_scopeEscape = flag; }
-
-    void pushNamePrefix(const std::string&);
-    void popNamePrefix(unsigned count = 1);
 };

 //=============================================================================
@@ -423,7 +475,7 @@ public:
    void fullQData(uint32_t* oldp, QData newval, int bits);
    void fullWData(uint32_t* oldp, const WData* newvalp, int bits);
    void fullDouble(uint32_t* oldp, double newval);
-    void fullEvent(uint32_t* oldp, VlEvent newval);
+    void fullEvent(uint32_t* oldp, const VlEventBase* newval);

    // In non-offload mode, these are called directly by the trace callbacks,
    // and are called chg*. In offload mode, they are called by the worker
@@ -460,7 +512,9 @@ public:
            }
        }
    }
-    VL_ATTR_ALWINLINE void chgEvent(uint32_t* oldp, VlEvent newval) { fullEvent(oldp, newval); }
+    VL_ATTR_ALWINLINE void chgEvent(uint32_t* oldp, const VlEventBase* newval) {
+        fullEvent(oldp, newval);
+    }
    VL_ATTR_ALWINLINE void chgDouble(uint32_t* oldp, double newval) {
        double old;
        std::memcpy(&old, oldp, sizeof(old));
@@ -539,7 +593,7 @@ public:
        m_offloadBufferWritep += 4;
        VL_DEBUG_IF(assert(m_offloadBufferWritep <= m_offloadBufferEndp););
    }
-    void chgEvent(uint32_t code, VlEvent newval) {
+    void chgEvent(uint32_t code, const VlEventBase* newval) {
        m_offloadBufferWritep[0] = VerilatedTraceOffloadCommand::CHG_EVENT;
        m_offloadBufferWritep[1] = code;
        m_offloadBufferWritep += 2;
@@ -1,42 +0,0 @@
-// -*- mode: C++; c-file-style: "cc-mode" -*-
-//=============================================================================
-//
-// Code available from: https://verilator.org
-//
-// Copyright 2001-2023 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 tracing types
-///
-/// This file is not part of the Verilated public-facing API.
-/// It is only for internal use by Verilated tracing routines.
-///
-//=============================================================================
-#ifndef VERILATOR_VERILATED_TRACE_DEFS_H_
-#define VERILATOR_VERILATED_TRACE_DEFS_H_
-
-// Verilator tracing scope types:
-// The values should match FST_ST_VCD_* from fstScopeType in gtkwave/fstapi.h
-// verilated_fst_c.cpp contains assertions to enforce this
-enum VltTraceScope {
-    VLT_TRACE_SCOPE_MODULE = 0,
-    VLT_TRACE_SCOPE_TASK = 1,
-    VLT_TRACE_SCOPE_FUNCTION = 2,
-    VLT_TRACE_SCOPE_BEGIN = 3,
-    VLT_TRACE_SCOPE_FORK = 4,
-    VLT_TRACE_SCOPE_GENERATE = 5,
-    VLT_TRACE_SCOPE_STRUCT = 6,
-    VLT_TRACE_SCOPE_UNION = 7,
-    VLT_TRACE_SCOPE_CLASS = 8,
-    VLT_TRACE_SCOPE_INTERFACE = 9,
-    VLT_TRACE_SCOPE_PACKAGE = 10,
-    VLT_TRACE_SCOPE_PROGRAM = 11
-};
-
-#endif  // guard
@@ -3,7 +3,7 @@
 //
 // Code available from: https://verilator.org
 //
-// Copyright 2001-2023 by Wilson Snyder. This program is free software; you
+// Copyright 2001-2024 by Wilson Snyder. This program is free software; you
 // can redistribute it and/or modify it under the terms of either the GNU
 // Lesser General Public License Version 3 or the Perl Artistic License
 // Version 2.0.
@@ -31,7 +31,7 @@

 #if 0
 # include <iostream>
-# define VL_TRACE_OFFLOAD_DEBUG(msg) std::cout << "TRACE OFFLOAD THREAD: " << msg << std::endl
+# define VL_TRACE_OFFLOAD_DEBUG(msg) std::cout << "TRACE OFFLOAD THREAD: " << msg << "\n"
 #else
 # define VL_TRACE_OFFLOAD_DEBUG(msg)
 #endif
@@ -64,9 +64,9 @@ static std::string doubleToTimescale(double value) VL_PURE {
    const char* suffixp = "s";
    // clang-format off
    if      (value >= 1e0)   { suffixp = "s"; value *= 1e0; }
-    else if (value >= 1e-3 ) { suffixp = "ms"; value *= 1e3; }
-    else if (value >= 1e-6 ) { suffixp = "us"; value *= 1e6; }
-    else if (value >= 1e-9 ) { suffixp = "ns"; value *= 1e9; }
+    else if (value >= 1e-3)  { suffixp = "ms"; value *= 1e3; }
+    else if (value >= 1e-6)  { suffixp = "us"; value *= 1e6; }
+    else if (value >= 1e-9)  { suffixp = "ns"; value *= 1e9; }
    else if (value >= 1e-12) { suffixp = "ps"; value *= 1e12; }
    else if (value >= 1e-15) { suffixp = "fs"; value *= 1e15; }
    else if (value >= 1e-18) { suffixp = "as"; value *= 1e18; }
@@ -185,7 +185,7 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::offloadWorkerThreadMain() {
                continue;
            case VerilatedTraceOffloadCommand::CHG_EVENT:
                VL_TRACE_OFFLOAD_DEBUG("Command CHG_EVENT " << top);
-                traceBufp->chgEvent(oldp, *reinterpret_cast<const VlEvent*>(readp));
+                traceBufp->chgEvent(oldp, reinterpret_cast<const VlEventBase*>(readp));
                continue;

                //===
@@ -328,10 +328,7 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::traceInit() VL_MT_UNSAFE {
    // Call all initialize callbacks, which will:
    // - Call decl* for each signal (these eventually call ::declCode)
    // - Store the base code
-    for (uint32_t i = 0; i < m_initCbs.size(); ++i) {
-        const CallbackRecord& cbr = m_initCbs[i];
-        cbr.m_initCb(cbr.m_userp, self(), nextCode());
-    }
+    for (const CallbackRecord& cbr : m_initCbs) cbr.m_initCb(cbr.m_userp, self(), nextCode());

    if (expectedCodes && nextCode() != expectedCodes) {
        VL_FATAL_MT(__FILE__, __LINE__, "",
@@ -378,13 +375,12 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::traceInit() VL_MT_UNSAFE {
 }

 template <>
-bool VerilatedTrace<VL_SUB_T, VL_BUF_T>::declCode(uint32_t code, const char* namep, uint32_t bits,
-                                                  bool tri) {
+bool VerilatedTrace<VL_SUB_T, VL_BUF_T>::declCode(uint32_t code, const std::string& declName,
+                                                  uint32_t bits) {
    if (VL_UNCOVERABLE(!code)) {
        VL_FATAL_MT(__FILE__, __LINE__, "", "Internal: internal trace problem, code 0 is illegal");
    }
    // To keep it simple, this is O(enables * signals), but we expect few enables
-    std::string declName = namePrefix() + namep;
    bool enabled = false;
    if (m_dumpvars.empty()) enabled = true;
    for (const auto& item : m_dumpvars) {
@@ -410,10 +406,7 @@ bool VerilatedTrace<VL_SUB_T, VL_BUF_T>::declCode(uint32_t code, const char* nam
        break;
    }

-    // Note: The tri-state flag is not used by Verilator, but is here for
-    // compatibility with some foreign code.
    int codesNeeded = VL_WORDS_I(bits);
-    if (tri) codesNeeded *= 2;
    m_nextCode = std::max(m_nextCode, code + codesNeeded);
    ++m_numSignals;
    m_maxBits = std::max(m_maxBits, bits);
@@ -457,7 +450,7 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::dumpvars(int level, const std::string&
        for (auto& i : hierSpaced) {
            if (i == '.') i = ' ';
        }
-        m_dumpvars.push_back(std::make_pair(level, hierSpaced));
+        m_dumpvars.emplace_back(level, hierSpaced);
    }
 }

@@ -498,16 +491,15 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::runCallbacks(const std::vector<Callback
        // Main thread executes all jobs with index % threads == 0
        std::vector<ParallelWorkerData*> mainThreadWorkerData;
        // Enqueue all the jobs
-        for (unsigned i = 0; i < cbVec.size(); ++i) {
-            const CallbackRecord& cbr = cbVec[i];
+        for (const CallbackRecord& cbr : cbVec) {
            // Always get the trace buffer on the main thread
-            Buffer* const bufp = getTraceBuffer();
+            Buffer* const bufp = getTraceBuffer(cbr.m_fidx);
            // Create new work item
            workerData.emplace_back(cbr.m_dumpCb, cbr.m_userp, bufp);
            // Grab the new work item
            ParallelWorkerData* const itemp = &workerData.back();
            // Enqueue task to thread pool, or main thread
-            if (unsigned rem = i % threads) {
+            if (unsigned rem = cbr.m_fidx % threads) {
                threadPoolp->workerp(rem - 1)->addTask(parallelWorkerTask, itemp);
            } else {
                mainThreadWorkerData.push_back(itemp);
@@ -530,7 +522,7 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::runCallbacks(const std::vector<Callback
    }
    // Fall back on sequential execution
    for (const CallbackRecord& cbr : cbVec) {
-        Buffer* const traceBufferp = getTraceBuffer();
+        Buffer* const traceBufferp = getTraceBuffer(cbr.m_fidx);
        cbr.m_dumpCb(cbr.m_userp, traceBufferp);
        commitTraceBuffer(traceBufferp);
    }
@@ -541,7 +533,7 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::runOffloadedCallbacks(
    const std::vector<CallbackRecord>& cbVec) {
    // Fall back on sequential execution
    for (const CallbackRecord& cbr : cbVec) {
-        Buffer* traceBufferp = getTraceBuffer();
+        Buffer* traceBufferp = getTraceBuffer(cbr.m_fidx);
        cbr.m_dumpOffloadCb(cbr.m_userp, static_cast<OffloadBuffer*>(traceBufferp));
        commitTraceBuffer(traceBufferp);
    }
@@ -610,11 +602,17 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::dump(uint64_t timeui) VL_MT_SAFE_EXCLUD
        }
    }

-    for (uint32_t i = 0; i < m_cleanupCbs.size(); ++i) {
-        const CallbackRecord& cbr = m_cleanupCbs[i];
-        cbr.m_cleanupCb(cbr.m_userp, self());
+    if (VL_UNLIKELY(m_constDump)) {
+        m_constDump = false;
+        if (offload()) {
+            runOffloadedCallbacks(m_constOffloadCbs);
+        } else {
+            runCallbacks(m_constCbs);
+        }
    }

+    for (const CallbackRecord& cbr : m_cleanupCbs) cbr.m_cleanupCb(cbr.m_userp, self());
+
    if (offload() && VL_LIKELY(bufferp)) {
        // Mark end of the offload buffer we just filled
        *m_offloadBufferWritep++ = VerilatedTraceOffloadCommand::END;
@@ -695,37 +693,40 @@ void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addInitCb(initCb_t cb, void* userp) VL_
    addCallbackRecord(m_initCbs, CallbackRecord{cb, userp});
 }
 template <>
-void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addFullCb(dumpCb_t cb, void* userp) VL_MT_SAFE {
-    addCallbackRecord(m_fullCbs, CallbackRecord{cb, userp});
+void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addConstCb(dumpCb_t cb, uint32_t fidx,
+                                                    void* userp) VL_MT_SAFE {
+    addCallbackRecord(m_constCbs, CallbackRecord{cb, fidx, userp});
 }
 template <>
-void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addFullCb(dumpOffloadCb_t cb, void* userp) VL_MT_SAFE {
-    addCallbackRecord(m_fullOffloadCbs, CallbackRecord{cb, userp});
+void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addConstCb(dumpOffloadCb_t cb, uint32_t fidx,
+                                                    void* userp) VL_MT_SAFE {
+    addCallbackRecord(m_constOffloadCbs, CallbackRecord{cb, fidx, userp});
 }
 template <>
-void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addChgCb(dumpCb_t cb, void* userp) VL_MT_SAFE {
-    addCallbackRecord(m_chgCbs, CallbackRecord{cb, userp});
+void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addFullCb(dumpCb_t cb, uint32_t fidx,
+                                                   void* userp) VL_MT_SAFE {
+    addCallbackRecord(m_fullCbs, CallbackRecord{cb, fidx, userp});
 }
 template <>
-void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addChgCb(dumpOffloadCb_t cb, void* userp) VL_MT_SAFE {
-    addCallbackRecord(m_chgOffloadCbs, CallbackRecord{cb, userp});
+void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addFullCb(dumpOffloadCb_t cb, uint32_t fidx,
+                                                   void* userp) VL_MT_SAFE {
+    addCallbackRecord(m_fullOffloadCbs, CallbackRecord{cb, fidx, userp});
+}
+template <>
+void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addChgCb(dumpCb_t cb, uint32_t fidx,
+                                                  void* userp) VL_MT_SAFE {
+    addCallbackRecord(m_chgCbs, CallbackRecord{cb, fidx, userp});
+}
+template <>
+void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addChgCb(dumpOffloadCb_t cb, uint32_t fidx,
+                                                  void* userp) VL_MT_SAFE {
+    addCallbackRecord(m_chgOffloadCbs, CallbackRecord{cb, fidx, userp});
 }
 template <>
 void VerilatedTrace<VL_SUB_T, VL_BUF_T>::addCleanupCb(cleanupCb_t cb, void* userp) VL_MT_SAFE {
    addCallbackRecord(m_cleanupCbs, CallbackRecord{cb, userp});
 }

-template <>
-void VerilatedTrace<VL_SUB_T, VL_BUF_T>::pushNamePrefix(const std::string& prefix) {
-    m_namePrefixStack.push_back(m_namePrefixStack.back() + prefix);
-}
-
-template <>
-void VerilatedTrace<VL_SUB_T, VL_BUF_T>::popNamePrefix(unsigned count) {
-    while (count--) m_namePrefixStack.pop_back();
-    assert(!m_namePrefixStack.empty());
-}
-
 //=========================================================================
 // Primitives converting binary values to strings...

@@ -839,7 +840,7 @@ void VerilatedTraceBuffer<VL_BUF_T>::fullBit(uint32_t* oldp, CData newval) {
 }

 template <>
-void VerilatedTraceBuffer<VL_BUF_T>::fullEvent(uint32_t* oldp, VlEvent newval) {
+void VerilatedTraceBuffer<VL_BUF_T>::fullEvent(uint32_t* oldp, const VlEventBase* newval) {
    const uint32_t code = oldp - m_sigs_oldvalp;
    *oldp = 1;  // Do we really store an "event" ?
    emitEvent(code, newval);
--- a/Show More
+++ b/Show More