Verilator

I found a way to convert the VCD file output by Verilator to an LXT2 file during the simulation. This substantially reduced the size of the LXT file and used the multi-core capability of the CPU so that the conversion did not impact the Verilator simulation time.

1. Within the SystemPerl distribution, in src/SpTraceVcdC.cpp, add "|O_NONBLOCK" to the "m_fd = ::open" instruction.

2. If your VCD output file name is "vl_dump.vcd", type the following command (this only needs to be done once):

mkfifo vl_dump.vcd

3. Every time your run the simulation, type the following two commands:

vcd2lxt2 vl_dump.vcd vl_dump.lxt &
time obj_dir/Vxxx
ls -lh vl_dump.*

This uses a "named pipe", see "man mkfifo", so that the output to the vl_dump.vcd file from Verilator goes through a pipe to the vcd2lxt2 program instead of going to the disk. The O_NONBLOCK allows the simulation and vcd2lxt2 programs to run simultaneously at full speed. In practice, the vcd2lxt2 program uses 10% to 50% of one core while the Verilator simulation uses close to 100% of another core. Also, gtkwave reads the smaller LXT files faster than the VCD files and anecdotally crashes less often when the waveform is refreshed in the middle of a simulation.

vcd2lxt2 is part of gtkwave.