Creates a visual representation to help analyze Verilator multithreaded simulation performance, by showing when each macro-task starts and ends, and showing when each thread is busy or idle.
The generated Gantt chart has time on the X-axis. Times shown are to the scale printed, i.e. a certain about of time for each character width. The Y-axis shows threads, each thread's execution is shown on one line. That line shows “[” at the position in time when it executes.
Following the “[” is the cpu number the task executed on, followed by zero or more “-” to make the width of the characters match the scaled execution time, followed by a “]”. If the scale is too small, the cpu number and mtask number will not be printed. If the scale is very small, a “&” indicates multiple mtasks started at that time position.
Also creates a value change dump (VCD) format dump file which may be viewed in a waveform viewer (e.g.
GTKWave). See below.
Build with --prof-threads. Run a sim with +verilator+prof+threads+window 2. This will create profile_threads.dat. Then run: verilator_gantt profile_threads.dat The report will be printed on standard output, this also generates profile_threads.vcd View profile_threads.vcd in a waveform viewer.
In waveforms there are the following signals. Most signals the “decimal” format will remove the leading zeros and make the traces easier to read.
parallelism: The number of mtasks active at this time, for best performance this will match the thread count. You may want to use an “analog step” format to view this signal.
cpu#_thread: For the given CPU number, the thread number executing.
mtask#_cpu; For the given mtask id, the CPU it is executing on.
thread#_mtask: For the given thread number, the mtask id executing.
The filename to read data from, defaults to “profile_threads.dat”.
Displays this message and program version and exits.
- --scale n
On the X-axis of the generated Gantt chart, each character represents this many time units. (On x86, time units are rdtsc ticks.) Defaults to 0, which will automatically compute a reasonable scale where no two mtasks need to fit into same character width's worth of scaled time.
- --vcd filename
Set output filename for vcd dump, or disable. Default is verilator_gantt.vcd.
The latest version is available from <https://verilator.org>.
Copyright 2018-2021 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
Wilson Snyder <firstname.lastname@example.org>