( ESNUG 492 Item 2 ) -------------------------------------------- [05/26/11]
Subject: ModelSim, Aldec, Verilator, Icarus, and SynaptiCAD VeriLogger
> I've used Verilator extensively. It is fast, stable with good support.
>
> Our main usage model was to have some $$$ commercial simulators (VCS,
> ModelSim, NC-Sim) for their debugging, System Verilog etc. and next to
> that Verilator was used to run parallel regressions. We then convert
> our Verilog code into C++ to run in Matlab, Simulink and SystemC benches.
>
> - Johan Wouters
> Ymatra BVBA Ham, Belgium
From: [ The Mouse That Roared ]
Hi, John,
I very much identified with the original engineer trying to find a cheap
Verilog simulator for multi-simulation regression.
I work for a small company that can't afford to buy 10+ licenses of Incisive
(NC-Verilog), VCS, or Modelsim SE for regressions.
We use Modelsim PE for the majority of our interactive simulation/debugging
work. We had been trying to get by Modelsim PE a regression solution, but
it took about 5 days to complete our basic regression runs.
We had evaluated Modelsim SE, and Aldec Riviera-Pro. We were able to drop
these into our sim scripts pretty quick, as both have good compatibility
with our existing Modelsim PE flow. Both were on the order of 3X faster
than PE but the cost was much too high for us.
I looked into the free tools. I was very impressed with Verilator's speed,
but it didn't work well for us, since it involved converting our testbench
to C. (Our Verilog testbenches were too large and complicated for that to
work for us.)
The Verilator web site listed Icarus Verilog as an option. With Icarus,
our initial test cases ran quite fast, and the simulator had good language
support, but it couldn't do parameter overrides -- which our regressions
needed. We tried getting support with Icarus, but they have only a small
team of volunteers, and their time is very limited -- so we couldn't get
the support we needed in a short time frame.
We ended up evaluating VeriLogger (from SynaptiCAD). At first we ran into
issues with some of our testbench code being incompatible. Surprisingly,
their support guys debugged and fixed these issues and turned around a new
version of VeriLogger within a couple of days.
We had to make some tweaks to our regression environment, because VeriLogger
didn't support compiling source into a work library with a separate compile
for different sources. One other engineer and I had the regression scripts
working with VeriLogger after a couple of hours work. By running 12 sims in
parallel, we can now run our regression in hours instead of days. We have
run into a couple simulator problems in the last year, and each time
SynaptiCAD has fixed these for us within a couple of days.
BE WARNED -- You can't run sims with encrypted IP models in SynaptiCAD, and
System Verilog is not supported. Also, VeriLogger waveform debugging is a
mess. We end up dumping a VCD and using gtkwave, or convert to Modelsim PE.
But for cheap ($1 K per license) pure Verilog 2001-compliant RTL simulation
for regressions, it works great.
Anyway if you do decide to post any of this, John, please keep me anonymous
as we can't afford to tick off the larger Verilog sim vendors.
- [ The Mouse That Roared ]
Join
Index
Next->Item
|
|
