.. Copyright 2003-2023 by Wilson Snyder. .. SPDX-License-Identifier: LGPL-3.0-only OR Artistic-2.0 .. _Simulating: ************************************ Simulating (Verilated-Model Runtime) ************************************ This section describes items related to simulating, that is, using a Verilated model's executable. For the runtime arguments to a simulated model, see :ref:`Simulation Runtime Arguments`. .. _Benchmarking & Optimization: Benchmarking & Optimization =========================== For best performance, run Verilator with the :vlopt:`-O3` :vlopt:`--x-assign fast <--x-assign>` :vlopt:`--x-initial fast <--x-initial>` :vlopt:`--noassert <--assert>` options. The :vlopt:`-O3` option will require a longer time to run Verilator, and :vlopt:`--x-assign fast <--x-assign>` :vlopt:`--x-initial fast <--x-assign>` may increase the risk of reset bugs in trade for performance; see the above documentation for these options. If using Verilated multithreaded, use ``numactl`` to ensure you use non-conflicting hardware resources. See :ref:`Multithreading`. Also, consider using profile-guided optimization; see :ref:`Thread PGO`. Minor Verilog code changes can also give big wins. You should not have any :option:`UNOPTFLAT` warnings from Verilator. Fixing these warnings can result in huge improvements; one user fixed their one UNOPTFLAT warning by making a simple change to a clocked latch used to gate clocks and gained a 60% performance improvement. Beyond that, the performance of a Verilated model depends primarily on your C++ compiler and the size of your CPU's caches. Experience shows that the instruction cache size often limits large models, and reducing code size, if possible, can be beneficial. The supplied $VERILATOR_ROOT/include/verilated.mk file uses the OPT, OPT_FAST, OPT_SLOW, and OPT_GLOBAL variables to control optimization. You can set these when compiling the output of Verilator with Make, for example: .. code-block:: bash make OPT_FAST="-Os -march=native" -f Vour.mk Vour__ALL.a OPT_FAST specifies optimization options for those parts of the model on the fast path. This is mostly code that is executed every cycle. OPT_SLOW applies to slow-path code, which rarely executes, often only once at the beginning or end of the simulation. OPT_SLOW is ignored if VM_PARALLEL_BUILDS is not 1, in which case all generated code will be compiled in a single compilation unit using OPT_FAST. See also the Verilator :vlopt:`--output-split` option. The OPT_GLOBAL variable applies to common code in the runtime library used by Verilated models (shipped in $VERILATOR_ROOT/include). Additional C++ files passed on the verilator command line use OPT_FAST. The OPT variable applies to all compilation units and the specific "OPT" variables described above. You can also use the :vlopt:`-CFLAGS` and/or :vlopt:`-LDFLAGS` options on the verilator command line to pass arguments directly to the compiler or linker. The default values of the "OPT" variables are chosen to yield good simulation speed with reasonable C++ compilation times. To this end, OPT_FAST is set to "-Os" by default. Higher optimization such as "-O2" or "-O3" may help (though often they provide only a minimal performance benefit), but compile times may be excessively large even with medium-sized designs. Compilation times can be improved at the expense of simulation speed by reducing optimization, for example, with OPT_FAST="-O0". Often good simulation speed can be achieved with OPT_FAST="-O1 -fstrict-aliasing" but with improved compilation times. Files controlled by OPT_SLOW have little effect on performance, and therefore OPT_SLOW is empty by default (equivalent to "-O0") for improved compilation speed. In common use cases, there should be little benefit in changing OPT_SLOW. OPT_GLOBAL is set to "-Os" by default, and there should rarely be a need to change it. As the runtime library is small compared to many Verilated models, disabling optimization on the runtime library should not seriously affect overall compilation time but may have a detrimental effect on simulation speed, especially with tracing. In addition to the above, for best results, use OPT="-march=native", the latest Clang compiler (about 10% faster than GCC), and link statically. Generally, the answer to which optimization level gives the best user experience depends on the use case, and some experimentation can pay dividends. For a speedy debug cycle during development, especially on large designs where C++ compilation speed can dominate, consider using lower optimization to get to an executable faster. For throughput-oriented use cases, for example, regressions, it is usually worth spending extra compilation time to reduce total CPU time. If you will be running many simulations on a single model, you can investigate profile-guided optimization. See :ref:`Compiler PGO`. Modern compilers also support link-time optimization (LTO), which can help, especially if you link in DPI code. To enable LTO on GCC, pass "-flto" in both compilation and link. Note that LTO may cause excessive compile times on large designs. Unfortunately, using the optimizer with SystemC files can result in compilation taking several minutes. (The SystemC libraries have many little inlined functions that drive the compiler nuts.) If using your own makefiles, you may want to compile the Verilated code with ``--MAKEFLAGS -DVL_INLINE_OPT=inline``. This will inline functions; however, this requires that all cpp files be compiled in a single compiler run. You may uncover further tuning possibilities by profiling the Verilog code. See :ref:`profiling`. When done optimizing, please let the author know the results. We like to keep tabs on how Verilator compares and may be able to suggest additional improvements. .. _Coverage Analysis: Coverage Analysis ================= Verilator supports adding code to the Verilated model to support SystemVerilog code coverage. With :vlopt:`--coverage`, Verilator enables all forms of coverage: * :ref:`User Coverage` * :ref:`Line Coverage` * :ref:`Toggle Coverage` When a model with coverage is executed, it will create a coverage file for collection and later analysis, see :ref:`Coverage Collection`. .. _User Coverage: Functional Coverage ------------------- With :vlopt:`--coverage` or :vlopt:`--coverage-user`, Verilator will translate functional coverage points the user has inserted manually win SystemVerilog code through into the Verilated model. Currently, all functional coverage points are specified using SystemVerilog assertion syntax, which must be separately enabled with :vlopt:`--assert`. For example, the following SystemVerilog statement will add a coverage point under the coverage name "DefaultClock": .. code-block:: sv DefaultClock: cover property (@(posedge clk) cyc==3); .. _Line Coverage: Line Coverage ------------- With :vlopt:`--coverage` or :vlopt:`--coverage-line`, Verilator will automatically add coverage analysis at each code flow change point (e.g., at branches). At each such branch, a counter is incremented. At the end of a test, the counters, filename, and line number corresponding to each counter are written into the coverage file. Verilator automatically disables coverage of branches with a $stop in them, as it is assumed that $stop branches contain an error check that should not occur. A :option:`/*verilator&32;coverage_block_off*/` metacomment will perform a similar function on any code in that block or below, or :option:`/*verilator&32;coverage_off*/` and :option:`/*verilator&32;coverage_on*/` will disable and enable coverage respectively around a block of code. Verilator may over-count combinatorial (non-clocked) blocks when those blocks receive signals which have had the :option:`UNOPTFLAT` warning disabled; for the most accurate results, do not disable this warning when using coverage. .. _Toggle Coverage: Toggle Coverage --------------- With :vlopt:`--coverage` or :vlopt:`--coverage-toggle`, Verilator will automatically add toggle coverage analysis into the Verilated model. Every bit of every signal in a module has a counter inserted, and the counter will increment on every edge change of the corresponding bit. Signals that are part of tasks or begin/end blocks are considered local variables and are not covered. Signals that begin with underscores (see :vlopt:`--coverage-underscore`), are integers, or are very wide (>256 bits total storage across all dimensions, see :vlopt:`--coverage-max-width`) are also not covered. Hierarchy is compressed, so if a module is instantiated multiple times, coverage will be summed for that bit across **all** instantiations of that module with the same parameter set. A module instantiated with different parameter values is considered a different module and will get counted separately. Verilator makes a minimally-intelligent decision about what clock domain the signal goes to, and only looks for edges in that clock domain. This means that edges may be ignored if it is known that the receiving logic could never see the edge. This algorithm may improve in the future. The net result is that coverage may be lower than what would be seen by looking at traces, but the coverage is a more accurate representation of the quality of stimulus into the design. There may be edges counted near time zero while the model stabilizes. It's a good practice to zero all coverage just before releasing reset to prevent counting such behavior. A :option:`/*verilator&32;coverage_off*/` :option:`/*verilator&32;coverage_on*/` metacomment pair can be used around signals that do not need toggle analysis, such as RAMs and register files. .. _Coverage Collection: Coverage Collection ------------------- When any coverage flag is used to Verilate, Verilator will add appropriate coverage point insertions into the model and collect the coverage data. To get the coverage data from the model, in the user wrapper code, typically at the end once a test passes, call :code:`Verilated::threadContextp()->coveragep()->write` with an argument of the filename for the coverage data file to write coverage data to (typically "logs/coverage.dat"). Run each of your tests in different directories, potentially in parallel. Each test will create a :file:`logs/coverage.dat` file. After running all of the tests, execute the :command:`verilator_coverage` command, passing arguments pointing to the filenames of all the individual coverage files. :command:`verilator_coverage` will read the :file:`logs/coverage.dat` file(s), and create an annotated source code listing showing code coverage details. :command:`verilator_coverage` may also be used for test grading, computing which tests are important to give full verification coverage on the design. For an example, see the :file:`examples/make_tracing_c/logs` directory. Grep for lines starting with '%' to see what lines Verilator believes need more coverage. Additional options of :command:`verilator_coverage` allow for the merging of coverage data files or other transformations. Info files can be written by verilator_coverage for import to :command:`lcov`. This enables using :command:`genhtml` for HTML reports and importing reports to sites such as `https://codecov.io `_. .. _Profiling: Code Profiling ============== The Verilated model may be code-profiled using GCC or Clang's C++ profiling mechanism. Verilator provides additional flags to help map the resulting C++ profiling results back to the original Verilog code responsible for the profiled C++ code functions. To use profiling: #. Use Verilator's :vlopt:`--prof-cfuncs`. #. Build and run the simulation model. #. The model will create gmon.out. #. Run :command:`gprof` to see where in the C++ code the time is spent. #. Run the gprof output through the :command:`verilator_profcfunc` program, and it will tell you what Verilog line numbers on which most of the time is being spent. .. _Execution Profiling: Execution Profiling =================== For performance optimization, it is helpful to see statistics and visualize how execution time is distributed in a verilated model. With the :vlopt:`--prof-exec` option, Verilator will: * Add code to the Verilated model to record execution flow. * Add code to save profiling data in non-human-friendly form to the file specified with :vlopt:`+verilator+prof+exec+file+\`. * In multithreaded models, add code to record each macro-task's start and end time across several calls to eval. (What is a macro-task? See the Verilator internals document (:file:`docs/internals.rst` in the distribution.) The :command:`verilator_gantt` program may then be run to transform the saved profiling file into a visual format and produce related statistics. .. figure:: figures/fig_gantt_min.png Example verilator_gantt output, as viewed with GTKWave. The measured_parallelism shows the number of CPUs being used at a given moment. The cpu_thread section shows which thread is executing on each physical CPU. The thread_mtask section shows which macro-task is running on a given thread. For more information, see :command:`verilator_gantt`. .. _Profiling ccache efficiency: Profiling ccache efficiency =========================== The Verilator-generated Makefile supports basic profiling of ccache behavior during the build. This can be used to track down files that might be unnecessarily rebuilt, though as of today, even minor code changes will usually require rebuilding a large number of files. Improving ccache efficiency during the edit/compile/test loop is an active development area. To get a basic report of how well ccache is doing, add the `ccache-report` target when invoking the generated Makefile: .. code-block:: bash make -C obj_dir -f Vout.mk Vout ccache-report This will print a report based on all executions of ccache during this invocation of Make. The report is also written to a file, in this example `obj_dir/Vout__cache_report.txt`. To use the `ccache-report` target, at least one other explicit build target must be specified, and OBJCACHE must be set to 'ccache'. This feature is currently experimental and might change in subsequent releases. .. _Save/Restore: Save/Restore ============ The intermediate state of a Verilated model may be saved so that it may later be restored. To enable this feature, use :vlopt:`--savable`. There are limitations in what language features are supported along with :vlopt:`--savable`; if you attempt to use an unsupported feature, Verilator will throw an error. To use save/restore, the user wrapper code must create a VerilatedSerialize or VerilatedDeserialze object and then call the :code:`<<` or :code:`>>` operators on the generated model and any other data the process needs to be saved/restored. These functions are not thread-safe and are typically called only by a main thread. For example: .. code-block:: C++ void save_model(const char* filenamep) { VerilatedSave os; os.open(filenamep); os << main_time; // user code must save the timestamp os << *topp; } void restore_model(const char* filenamep) { VerilatedRestore os; os.open(filenamep); os >> main_time; os >> *topp; } Profile-Guided Optimization =========================== Profile-guided optimization is the technique where profiling data is collected by running your simulation executable; then this information is used to guide the next Verilation or compilation. There are two forms of profile-guided optimizations. Unfortunately, for best results, they must each be performed from the highest level code to the lowest, which means performing them separately and in this order: * :ref:`Thread PGO` * :ref:`Compiler PGO` Other forms of PGO may be supported in the future, such as clock and reset toggle rate PGO, branch prediction PGO, statement execution time PGO, or others, as they prove beneficial. .. _Thread PGO: Thread Profile-Guided Optimization ---------------------------------- Verilator supports profile-guided optimization (Verilation) of multithreaded models (Thread PGO) to improve performance. When using multithreading, Verilator computes how long macro tasks take and tries to balance those across threads. (What is a macro-task? See the Verilator internals document (:file:`docs/internals.rst` in the distribution.) If the estimations are incorrect, the threads will not be balanced, leading to decreased performance. Thread PGO allows collecting profiling data to replace the estimates and better optimize these decisions. To use Thread PGO, Verilate the model with the :vlopt:`--prof-pgo` option. This will code to the verilated model to save profiling data for profile-guided optimization. Run the model executable. When the executable exits, it will create a profile.vlt file. Rerun Verilator, optionally omitting the :vlopt:`--prof-pgo` option and adding the :file:`profile.vlt` generated earlier to the command line. Note there is no Verilator equivalent to GCC's --fprofile-use. Verilator's profile data file (:file:`profile.vlt`) can be placed directly on the verilator command line without any option prefix. If results from multiple simulations are to be used in generating the optimization, multiple simulation's profile.vlt may be concatenated externally, or each file may be fed as separate command line options into Verilator. Verilator will sum the profile results, so a long-running test will have more weight for optimization proportionally than a shorter-running test. If you provide any profile feedback data to Verilator and it cannot use it, it will issue the :option:`PROFOUTOFDATE` warning that threads were scheduled using estimated costs. This usually indicates that the profile data was generated from a different Verilog source code than Verilator is currently running against. Therefore, repeat the data collection phase to create new profiling data, then rerun Verilator with the same input source files and that new profiling data. .. _Compiler PGO: Compiler Profile-Guided Optimization ------------------------------------ GCC and Clang support compiler profile-guided optimization (PGO). This optimizes any C/C++ program, including Verilated code. Using compiler PGO typically yields improvements of 5-15% on both single-threaded and multithreaded models. Please see the appropriate compiler documentation to use PGO with GCC or Clang. The process in GCC 10 was as follows: 1. Compile the Verilated model with the compiler's "-fprofile-generate" flag: .. code-block:: bash verilator [whatever_flags] --make \ -CFLAGS -fprofile-generate -LDFLAGS -fprofile-generate Or, if calling make yourself, add -fprofile-generate appropriately to your Makefile. 2. Run your simulation. This will create \*.gcda file(s) in the same directory as the source files. 3. Recompile the model with -fprofile-use. The compiler will read the \*.gcda file(s). For GCC: .. code-block:: bash verilator [whatever_flags] --build \ -CFLAGS "-fprofile-use -fprofile-correction" For Clang: .. code-block:: bash llvm-profdata merge -output default.profdata *.profraw verilator [whatever_flags] --build \ -CFLAGS "-fprofile-use -fprofile-correction" or, if calling make yourself, add these CFLAGS switches appropriately to your Makefile. Clang and GCC also support -fauto-profile, which uses sample-based feedback-directed optimization. See the appropriate compiler documentation.