System Verilog & the VHDL user

[Niv]

Hi all, we currently use VHDL for all our design & verification via
testbenches.
I have been asked to look at alternative verification methodologies,
particularly SystemVerilog.

(I've been on a PSL course a few years ago, but until now never had
the tool(s) to exploit it's use, but
people seem to think SysVerilog may be a better approach anyway.

So, what is the best book(s) to learn SV for a VHDL conversant
audience, and possibly more important,
whare can we quickly learn what SV can do for us above and beyond VHDl
testbenches.

(Cross posted to Verilog group)

Regards, Niv.

 

[Dal]

Approaching a similar decision myself. Have you considered SystemC?

I would also like to know what SV can do above and beyond SC and VHDL.

Darrin

 

[HT-Lab]

Hi all, we currently use VHDL for all our design & verification via
testbenches.
I have been asked to look at alternative verification methodologies,
particularly SystemVerilog.

SystemVerilog is a language, not a verification methodology. Perhaps you
should focus on the required methodology first and then look at the
available languages and tools?

(I've been on a PSL course a few years ago, but until now never had
the tool(s) to exploit it's use,

Yes unfortunately the tools for PSL/SVA are quite expensive, lets hope
VHDL2007/8 will introduce more people to the power of assertions and as such
bring the price down.

but
people seem to think SysVerilog may be a better approach anyway.

Better approach for what, Assertion Based Verification(ABV), Testbench
Development, Constrained Random(CR), Transaction Level Modelling(TLM),
fixed/floating point modelling?

So, what is the best book(s) to learn SV for a VHDL conversant
audience, and possibly more important,
whare can we quickly learn what SV can do for us above and beyond VHDl
testbenches.

Probably less than what you would expect. As posted some time ago VHDL can
be used for TLM, CR, ABV (using PSL) but probably not as easy as with a
modern language like SystemVerilog/SystemC.

Another option as suggested by Darrin is to look at SystemC. The advantage
of SystemC is that you can add it to a "lowcost" simulator like Modelsim PE
(no need to go to SE/Questa) and it will give you access to TLM, CR, OO and
a very easy C/C++ interface to your PE VHDL.

Hans
www.ht-lab.com

 

[Niv]

SystemVerilog is a language, not a verification methodology. Perhaps you
should focus on the required methodology first and then look at the
available languages and tools?




Yes unfortunately the tools for PSL/SVA are quite expensive, lets hope
VHDL2007/8 will introduce more people to the power of assertions and as such
bring the price down.


Better approach for what, Assertion Based Verification(ABV), Testbench
Development, Constrained Random(CR), Transaction Level Modelling(TLM),
fixed/floating point modelling?


Probably less than what you would expect. As posted some time ago VHDL can
be used for TLM, CR, ABV (using PSL) but probably not as easy as with a
modern language like SystemVerilog/SystemC.

Another option as suggested by Darrin is to look at SystemC. The advantage
of SystemC is that you can add it to a "lowcost" simulator like Modelsim PE
(no need to go to SE/Questa) and it will give you access to TLM, CR, OO and
a very easy C/C++ interface to your PE VHDL.

Hanswww.ht-lab.com







- Show quoted text -

We do have ModelSim SE & Questa, so that's not an issue.
However, no-one is really using Questa at the moment due to lack of
understanding of what we can do with it.

We currently do large FPGA designs and test using VHDL testbenches
(quite a lot of VHDL ASSERT statements are used).
I've been led to believe (possibly erroneously) that with
SystemVeilog, the assertions are more comprehensive & simpler to
apply.
I realise that SV is a language & not a methodology; and there seems a
general consensus here that SV will "win" over SystemC;
although that could be rubbish.

We want to address our verification methodology to make it more
thorough & hopefully simpler, albeit having to learn SV on top of
VHDL. (Why no SystemVHDL?)

Niv.

 

[HT-Lab]

We do have ModelSim SE & Questa, so that's not an issue.
However, no-one is really using Questa at the moment due to lack of
understanding of what we can do with it.

In that case I would suggest studying the AVM/OVM which is quite a
comprehensive verification environment and supported by Questa. You can
download a free copy from the Mentor website

http://www.mentor.com/products/fv/_3b715c/

We currently do large FPGA designs and test using VHDL testbenches
(quite a lot of VHDL ASSERT statements are used).
I've been led to believe (possibly erroneously) that with
SystemVeilog, the assertions are more comprehensive & simpler to
apply.

I am not sure about more comprehensive but given that they are part of the
language you might be right with respect to applying them. However, as you
probably agree, adding PSL statements in comments (or link in a vunit) is
not such a big issue. In VHDL2007/8 comments are no longer required for PSL
and if I understand the P1076 reference manual correctly vunits can be used
as packages.

I realise that SV is a language & not a methodology; and there seems a
general consensus here that SV will "win" over SystemC;
although that could be rubbish.

I think that is indeed rubbish since we are not talking about HD-DVD versus
Blu-Ray, SV and SC have non-overlapping capabilities. SV is closer to the
hardware has build in assertions and is more capable in functional
verification (functional coverage is much easier in SV than SC) to name a
few. SystemC on the other hand is more capable in terms of high-level
modelling (better OO support), so if you want to model a complete SoC
including say an Operating System you would pick SystemC over SystemVerilog.

We want to address our verification methodology to make it more
thorough & hopefully simpler, albeit having to learn SV on top of
VHDL. (Why no SystemVHDL?)

It is always good to learn another language so your effort on SV will not be
wasted, just make sure you don't throw out the baby with the bathwater

Hans
www.ht-lab.com

 

[Mike Treseler]

Exactly.

We do have ModelSim SE & Questa, so that's not an issue.
However, no-one is really using Questa at the moment due to lack of
understanding of what we can do with it.

You may be able to do what you need with plain vhdl.
Make sure the testbench stim and response data is properly synchronized.
Use STA to verify timing, not the testbench.
Make a synchronous watcher process for global/mulicycle assertions.

We want to address our verification methodology to make it more
thorough & hopefully simpler, albeit having to learn SV on top of
VHDL. (Why no SystemVHDL?)

I agree with Hans.
Focus on what is missing in your existing testbench suites.
You may or may not need a new language.

-- Mike Treseler