waveform dump problem using SystemVerilog

Hi ravurig,

I think you may try to use latest IUS583-S003. Because Verilog and SV co-simulation is not stable in old IUS version.

Regards,
Davy

Hi Ravi,
   
This is a classical problem when simulators move onto becoming full
fledged verification platforms. Honestly I'm surprised to see this with NCSIM though. 
As a quick fix, try:

$recordvars(top.dut)


This would make sure you dump your DUT signals without bothering about TB stuff. 
I've seen similar issues with other tool(s), but given the success of Specman's 
amazing dumping capability I thought Cadence would have done a better job here.

Also check with CDN help desk on when they would support this TB dumping. 
I've an article in making on "Debugging Testbenches in SystemVerilog" - this will 
be little tool specific and will require some methodology help etc. It is not in 
a shape to be shared right now, but if you are based in Bangalore drop me an 
email ajeetha <> gmail.com and we can come and talk to you in person on this 
topic.

Regards
Ajeetha, CVC
www.noveldv.com    <br>

I would suggest a couple of things.

1) As indicated, move to the latest Incisive (NC-Verilog) release which is currently 5.8 available on downloads.cadence.com
2) File a sourcelink help request or send email to support@cadence.com - most likely support will want a test case which
reproduces the error.
3) try switching to shm_open() and shm_dump()
4) Contact your local Incisive AE who is always more than willing to provide some advice
5) If you can identify the construct causing the problem, you can always dump "around it".

I would suspect maybe the dumping of some of the SV TB constructs lagged the implementation a bit and the 5.6 release is
about 2 years old now. The latest releases have a boatload more field testing these days, are far more stable and offer
way more features.

Best regards,

Jim McG.

Hi All,
Thank you very much for your suggestions.
I will try to move to newer version of IUS.

For time being I have integrated Novas Debussy for waveform dumping using fsdbdumpvars() and fsdbduumpfiles.
I am able to get my waveform.

I have used DPI to call C-function in my SV testbench along with my Verilog module.

I am comparing the simulation speed of SV testbench with Verilog'95 & DPI calling C-function
and another Verilog testbench with same C-function implemented as behavioral model(Verilog-2001).
I started the simulation with a view that SV testbench with DPI is going to give me better solution.
But behavioral model is faster than C-model testbench.

Case 1:-
Testbench:- SV
Verilog module:- '95
C-model :- Called by SV-DPI.

Case 2:-
Testbench:- Verilog'95
Verilog module:- '95
C-function as behavioral model:- Verilog 2001(to use its real and other functions).

I thought Case1 is faster than Case2.

But my analysis shows Case2 is faster than Case1.

Can any body help me out.

Hi Rav,
  I'm not surprised - SV support is still new and hence may not enjoy all the performance optimizations that good old Verilog 95 has. One suggestion - try using separate ncelab commands - if possible. i.e. in your Case1:

******
Case 1:-
Testbench:- SV
Verilog module:- '95
C-model :- Called by SV-DPI.
*********

Compile 95 module using ncvlog -f v95.f
             ncelab v95_top
       ncvlog -sv31a -f sv.f
       ncelab sv_tb_top v95_top

Also, since you believe the DPI will give you the boost - do you know how much is the time spent on the C-Verilog/SV intreaction? A profile run should tell you that - unless that PLI is a significant portion you won't see any boost.

More than this I believe you need to work with your AE to get it better. If you need more help on profiling and general optimization enhancements etc. (not necessariy tied to NC), contact me separately if you wish.

Good Luck
Ajeetha, CVC
www.noveldv.com