Verilating

Verilator may be used in five major ways:

  • With the --binary option, Verilator will translate the design into an executable, via generating C++ and compiling it. See Binary, C++ and SystemC Generation.

  • With the --cc or --sc options, Verilator will translate the design into C++ or SystemC code, respectively. See Binary, C++ and SystemC Generation.

  • With the --lint-only option, Verilator will lint the design to check for warnings but will not typically create any output files.

  • With the --xml-only option, Verilator will create XML output that may be used to feed into other user-designed tools. See docs/xml.rst in the distribution.

  • With the -E option, Verilator will preprocess the code according to IEEE preprocessing rules and write the output to standard out. This is useful to feed other tools and to debug how “`define” statements are expanded.

Binary, C++ and SystemC Generation

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

When using these options:

  1. Verilator reads the input Verilog code and determines all “top modules”, that is, modules or programs that are not used as instances under other cells. If --top-module is used, then that determines the top module, and all other top modules are removed; otherwise a MULTITOP warning is given.

  2. Verilator writes the C++/SystemC code to output files into the --Mdir option-specified directory, or defaults to “obj_dir”. The prefix is set with --prefix, or defaults to the name of the top module.

  3. If --binary or --main is used, Verilator creates a C++ top wrapper to read command line arguments, create the model, and execute the model.

  4. If --binary or --exe is used, Verilator creates makefiles to generate a simulation executable, otherwise, it creates makefiles to generate an archive (.a) containing the objects.

  5. If --binary or --build is used, it calls GNU Make or CMake to build the model.

Once a model is built, the next step is typically for the user to run it, see Simulating (Verilated-Model Runtime).

Hierarchical Verilation

Large designs may take long (e.g., 10+ minutes) and huge memory (e.g., 100+ GB) to Verilate. In hierarchical mode, the user manually selects some large lower-level hierarchy blocks to separate from the larger design. For example, a core may be the hierarchy block separated out of a multi-core SoC design.

Verilator is run in hierarchical mode on the whole SoC. Verilator will make two models, one for the CPU hierarchy block and one for the SoC. The Verilated code for the SoC will automatically call the CPU Verilated model.

The current hierarchical Verilation is based on --lib-create. Each hierarchy block is Verilated into a library. User modules of the hierarchy blocks will see a tiny wrapper generated by --lib-create.

Usage

Users need to mark one or more moderate-size modules as hierarchy block(s). There are two ways to mark a module:

Then pass the --hierarchical option to Verilator.

The compilation is the same as when not using hierarchical mode.

make -C obj_dir -f Vtop_module_name.mk

Limitations

Hierarchy blocks have some limitations, including:

  • The hierarchy block cannot be accessed using dot (.) from the upper module(s) or other hierarchy blocks.

  • Signals in the block cannot be traced.

  • Modport cannot be used at the hierarchical block boundary.

  • The simulation speed is likely not as fast as flat Verilation, in which all modules are globally scheduled.

  • Generated clocks may not work correctly if generated in the hierarchical model and passed into another hierarchical model or the top module.

  • Delays are not allowed in hierarchy blocks.

But, the following usage is supported:

  • Nested hierarchy blocks. A hierarchy block may instantiate other hierarchy blocks.

  • Parameterized hierarchy block. Parameters of a hierarchy block can be overridden using #(.param_name(value)) construct.

Overlapping Verilation and Compilation

Verilator needs to run 2 + N times in hierarchical Verilation, where N is the number of hierarchy blocks. One of the two is for the top module, which refers to the wrappers of all other hierarchy blocks. The second of the two is the initial run that searches modules marked with /*verilator hier_block*/ metacomment and creates a plan and write in prefix_hier.mk. This initial run internally invokes other N + 1 runs, so you don’t have to care about these N + 1 times of run. The additional N is the Verilator run for each hierarchical block.

If :-j {jobs} option is specified, Verilation for hierarchy blocks runs in parallel.

If --build option is specified, C++ compilation also runs as soon as a hierarchy block is Verilated. C++ compilation and Verilation for other hierarchy blocks run simultaneously.

Cross Compilation

Verilator supports cross-compiling Verilated code. This is generally used to run Verilator on a Linux system and produce C++ code that is then compiled on Windows.

Cross-compilation involves up to three different OSes. The build system is where you configure and compile Verilator, the host system is where you run Verilator, and the target system is where you compile the Verilated code and run the simulation.

Verilator requires the build and host system types to be the same, though the target system type may be different. To support this, ./configure and make Verilator on the build system. Then, run Verilator on the host system. Finally, the output of Verilator may be compiled on the different target system.

To support this, none of the files that Verilator produces will reference any configure-generated build-system-specific files, such as config.h (which is renamed in Verilator to config_package.h to reduce confusion.) The disadvantage of this approach is that include/verilatedos.h must self-detect the requirements of the target system, rather than using configure.

The target system may also require edits to the Makefiles, the simple Makefiles produced by Verilator presume the target system is the same type as the build system.

Multithreading

Verilator supports multithreaded simulation models.

With --threads 1, the generated model is single-threaded; however, the support libraries are multithread safe. This allows different instantiations of the model(s) to potentially each be run under a different thread. All threading is the responsibility of the user’s C++ testbench.

With --threads {N}, where N is at least 2, the generated model will be designed to run in parallel on N threads. The thread calling eval() provides one of those threads, and the generated model will create and manage the other N-1 threads. It’s the client’s responsibility not to oversubscribe the available CPU cores. Under CPU oversubscription, the Verilated model should not livelock nor deadlock; however, you can expect performance to be far worse than it would be with the proper ratio of threads and CPU cores.

The thread used for constructing a model must be the same thread that calls eval() into the model; this is called the “eval thread”. The thread used to perform certain global operations, such as saving and tracing, must be done by a “main thread”. In most cases, the eval thread and main thread are the same thread (i.e. the user’s top C++ testbench runs on a single thread), but this is not required.

When making frequent use of DPI imported functions in a multithreaded model, it may be beneficial to performance to adjust the --instr-count-dpi option based on some experimentation. This influences the partitioning of the model by adjusting the assumed execution time of DPI imports.

When using --trace to perform VCD tracing, the VCD trace construction is parallelized using the same number of threads as specified with --threads, and is executed on the same thread pool as the model.

The --trace-threads options can be used with --trace-fst to offload FST tracing using multiple threads. If --trace-threads is given without --threads, then --trace-threads will imply --threads 1, i.e., the support libraries will be thread safe.

With --trace-threads 0, trace dumps are produced on the main thread. This again gives the highest single-thread performance.

With --trace-threads {N}, where N is at least 1, up to N additional threads will be created and managed by the trace files (e.g., VerilatedFstC), to offload construction of the trace dump. The main thread will be released to proceed with execution as soon as possible, though some main thread blocking is still necessary while capturing the trace. FST tracing can utilize up to 2 offload threads, so there is no use of setting --trace-threads higher than 2 at the moment.

When running a multithreaded model, the default Linux task scheduler often works against the model by assuming short-lived threads and thus it often schedules threads using multiple hyperthreads within the same physical core. For best performance, use the numactl program to (when the threading count fits) select unique physical cores on the same socket. The same applies for --trace-threads as well.

As an example, if a model was Verilated with --threads 4, we consult:

egrep 'processor|physical id|core id' /proc/cpuinfo

To select cores 0, 1, 2, and 3 that are all located on the same socket (0) but have different physical cores. (Also useful is numactl --hardware, or lscpu, but those don’t show hyperthreading cores.) Then we execute:

numactl -m 0 -C 0,1,2,3 -- verilated_executable_name

This will limit memory to socket 0, and threads to cores 0, 1, 2, 3, (presumably on socket 0), optimizing performance. Of course, this must be adjusted if you want another simulator to use, e.g., socket 1, or if you Verilated with a different number of threads. To see what CPUs are actually used, use --prof-exec.

Multithreaded Verilog and Library Support

$display/$stop/$finish are delayed until the end of an eval() call to maintain ordering between threads. This may result in additional tasks completing after the $stop or $finish.

If using --coverage, the coverage routines are fully thread-safe.

If using the DPI, Verilator assumes pure DPI imports are thread-safe, balancing performance versus safety. See --threads-dpi.

If using --savable, the save/restore classes are not multithreaded and must be called only by the eval thread.

If using --sc, the SystemC kernel is not thread-safe; therefore, the eval thread and main thread must be the same.

If using --trace, the tracing classes must be constructed and called from the main thread.

If using --vpi, since SystemVerilog VPI was not architected by IEEE to be multithreaded, Verilator requires all VPI calls are only made from the main thread.

GNU Make

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

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

CMake

Verilator can be run using CMake, which takes care of both running Verilator and compiling the output. There is a CMake example in the examples/ directory. The following is a minimal CMakeLists.txt that would build the code listed in Example C++ Execution

project(cmake_example)
find_package(verilator HINTS $ENV{VERILATOR_ROOT})
add_executable(Vour sim_main.cpp)
verilate(Vour SOURCES our.v)

find_package will automatically find an installed copy of Verilator, or use a local build if VERILATOR_ROOT is set.

Using CMake >= 3.12 and the Ninja generator is recommended, though other combinations should work. To build with CMake, change to the folder containing CMakeLists.txt and run:

mkdir build
cd build
cmake -GNinja ..
ninja

Or to build with your system default generator:

mkdir build
cd build
cmake ..
cmake --build .

If you’re building the example, you should have an executable to run:

./Vour

The package sets the CMake variables verilator_FOUND, VERILATOR_ROOT, and VERILATOR_BIN to the appropriate values and creates a verilate() function. verilate() will automatically create custom commands to run Verilator and add the generated C++ sources to the target specified.

Verilate in CMake

verilate(target SOURCES source ... [TOP_MODULE top] [PREFIX name]
         [TRACE] [TRACE_FST] [SYSTEMC] [COVERAGE]
         [INCLUDE_DIRS dir ...] [OPT_SLOW ...] [OPT_FAST ...]
         [OPT_GLOBAL ..] [DIRECTORY dir] [THREADS num]
         [TRACE_THREADS num] [VERILATOR_ARGS ...])

Lowercase and … should be replaced with arguments; the uppercase parts delimit the arguments and can be passed in any order or left out entirely if optional.

verilate(target …) can be called multiple times to add other Verilog modules to an executable or library target.

When generating Verilated SystemC sources, you should list the SystemC include directories and link to the SystemC libraries.

target

Name of a target created by add_executable or add_library.

COVERAGE

Optional. Enables coverage if present, equivalent to “VERILATOR_ARGS –coverage”.

DIRECTORY

Optional. Set the verilator output directory. It is preferable to use the default, which will avoid collisions with other files.

INCLUDE_DIRS

Optional. Sets directories that Verilator searches (same as -y).

OPT_SLOW

Optional. Set compiler options for the slow path. You may want to reduce the optimization level to improve compile times with large designs.

OPT_FAST

Optional. Set compiler options for the fast path.

OPT_GLOBAL

Optional. Set compiler options for the common runtime library used by Verilated models.

PREFIX

Optional. Sets the Verilator output prefix. Defaults to the name of the first source file with a “V” prepended. It must be unique in each call to verilate(), so this is necessary if you build a module multiple times with different parameters. It must be a valid C++ identifier, i.e., it contains no white space and only characters A-Z, a-z, 0-9 or _.

SOURCES

List of Verilog files to Verilate. You must provide at least one file.

SYSTEMC

Optional. Enables SystemC mode, defaults to C++ if not specified.

When using Accellera’s SystemC with CMake support, a CMake target is available that simplifies the SystemC steps. This will only work if CMake can find the SystemC installation, and this can be configured by setting the CMAKE_PREFIX_PATH variable during CMake configuration.

Don’t forget to set the same C++ standard for the Verilated sources as the SystemC library. This can be specified using the SYSTEMC_CXX_FLAGS environment variable.

THREADS

Optional. Enable a multithreaded model; see --threads.

TRACE_THREADS

Optional. Enable multithreaded FST trace; see --trace-threads.

TOP_MODULE

Optional. Sets the name of the top module. Defaults to the name of the first file in the SOURCES array.

TRACE

Optional. Enables VCD tracing if present, equivalent to “VERILATOR_ARGS –trace”.

TRACE_FST

Optional. Enables FST tracing if present, equivalent to “VERILATOR_ARGS –trace-fst”.

VERILATOR_ARGS

Optional. Extra arguments to Verilator. Do not specify --Mdir or --prefix here; use DIRECTORY or PREFIX.

Verilation Summary Report

When Verilator generates code, unless --quiet-stats is used, it will print a report to stdout summarizing the build. For example:

- V e r i l a t i o n   R e p o r t: Verilator ....
- Verilator: Built from 354 MB sources in 247 modules,
    into 74 MB in 89 C++ files needing 0.192 MB
- Verilator: Walltime 26.580 s (elab=2.096, cvt=18.268,
    bld=2.100); cpu 26.548 s on 1 threads; alloced 2894.672 MB

The information in this report is:

"Verilator ..."

Program version.

"234 MB sources"

Characters of post-preprocessed text in all input Verilog and Verilator Control files in megabytes.

"247 modules"

Number of interfaces/modules/classes/packages in design before elaboration.

"into 74 MB"

Characters of output C++ code, including comments in megabytes.

"89 C++ files"

Number of .cpp files created.

"needing 192MB"

Verilation-time minimum-bound estimate of memory needed to run model in megabytes. (Expect to need significantly more.)

"Walltime 26.580 s"

Real elapsed wall time for Verilation and build.

"elab=2.096"

Wall time to read in files and complete elaboration.

"cvt=18.268"

Wall time for Verilator to process and write output.

"bld=2.1"

Wall time to compile gcc/clang (if using --build).

"cpu 22.548 s"

CPU time used, total across all CPU threads.

"4 threads"

Number of simultaneous threads used.

"alloced 123 MB"

Total memory used during build by Verilator executable (excludes --build compiler’s usage) in megabytes.