The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Modeling >> Behavioral Models >> gates.va and transition()
https://designers-guide.org/forum/YaBB.pl?num=1220537539

Message started by Geoffrey_Coram on Sep 4^th, 2008, 7:12am

Title: gates.va and transition()
Post by Geoffrey_Coram on Sep 4^th, 2008, 7:12am

I was running the Verilog-A logic gates from:
http://www.designers-guide.org/VerilogAMS/functional-blocks/gates/gates.va

in a commercial simulator, and I got this warning:
VERILOG-A PERFORMANCE WARNING:
"gates.va", line 125: The first argument to the transition filter is a
continuous signal, which might slow down the simulation.
Further occurrences of this warning will be suppressed.

If I change, for example,
V(out) <+ transition( !((V(in1) > vth) && (V(in2) > vth)) ? vh : vl, td, tt );
to
vout = ( !((V(in1) > vth) && (V(in2) > vth))) ? vh : vl;
V(out) <+ transition( vout, td, tt );

then I don't get the warning, and the simulation runs 40x faster!

Title: Re: gates.va and transition()
Post by jbdavid on Sep 4^th, 2008, 10:49am

you would think Ken would know better..
But maybe SOME commercial simulators figure out that the expression is results in a logical value, and some don't..
I normally only evaluate the logic output on the cross event.. not every time step -- unless I need to support spectreRF.. (and i think the transition statement itself is a problem there)
..
jbd

Title: Re: gates.va and transition()
Post by Andrew Beckett on Jan 5^th, 2009, 8:29am

Hmm, spectre doesn't seem to be that clever about figuring this out. Even just reformulating the same expression as:

Code:

vout = ( !((V(in1) > vth) && (V(in2) > vth))) ? vh : vl;
V(out) <+ transition( vout, td, tt );

stops the warning. I'll file a CCR to see if we can improve it.

Regards,

Andrew.

Title: Re: gates.va and transition()
Post by Dan Clement on Jan 25^th, 2009, 9:40pm

Can you please explain why there is a speed difference in the original question?

Thanks,
Dan

Title: Re: gates.va and transition()
Post by Andrew Beckett on Jan 27^th, 2009, 12:37pm

Dan,

No, but I didn't do any significant tests - mine was just to see if it compiled rather than giving it anything sensible to simulate.

Perhaps Geoffrey can comment - I'm sure he'll take this up with Cadence ;)

Regards,

Andrew.

Title: Re: gates.va and transition()
Post by Dan Clement on Jan 28^th, 2009, 6:10am

Hi Andrew,

That's cool. I seem to keep running into speed issues when I try to use the transition and slew operators. So if anyone can explain the right and wrong way to use these functions, I would truly appreciate more guidance.

I am a spectre user.

Best Regards,
Dan

Title: Re: gates.va and transition()
Post by Geoffrey_Coram on Feb 2^nd, 2009, 10:38am

I think the issue is, if you were to do

V(out) <+ transition( V(in1), td, tt);

then every time V(in1) changed even the slightest, the simulator would have to schedule a timepoint at t+td to propagate this change to the output (and further timepoints up to t+td+tt to finish the transition).

I guess the simulator sees the V(in1) in the original argument and schedules these timepoints, even at timepoints where the Boolean expression isn't changing.

Title: Re: gates.va and transition()
Post by Andrew Beckett on Feb 3^rd, 2009, 9:07pm

That's a good point. Merely fixing the warning is not enough - the warning is reasonable, if that's the behaviour. We need to fix it to prevent the transition being unnecessarily triggered by what it sees as a continuous signal. I'll append this to the CCR I filed.

Regards,

Andrew.

Title: Re: gates.va and transition()
Post by DmitTrix on Feb 8^th, 2011, 5:08am

(sorry for resurrecting an old thread, but it seems to me the appropriate place to ask...)

I've run into this warning ("The first argument to the transition filter is a continuous signal, which might slow down the simulation") while using a simple Sample-and-Hold implemented as follows:

parameter real Vt = 0.5 from (-inf:inf);

parameter real Td = 0 from [0:inf);
parameter real Tr = 100p from (0:inf);
parameter real Tf = 100p from (0:inf);

real vout_val;
analog begin
@(cross(V(vclk) - Vt, 1)) vout_val = V(vin);
V(vout) <+ transition( vout_val, Td, Tr, Tf );
end

Seems that Spectre refers to a real value as to a value that can change continuously (although it doesn't), and issues the warning.

I don't see any way to implement it without using a real value, or without using the "transition" filter (removing it will cause too abrupt changes at the output).

Any suggestions?

Thanx,
Dmitry.

P.S. I'm running MMSim 10.1

Title: Re: gates.va and transition()
Post by Geoffrey_Coram on Feb 8^th, 2011, 11:24am

Quote:

which might slow down

One question might be: does it seem to be slowing down the simulation?

Title: Re: gates.va and transition()
Post by naderi on Feb 24^th, 2011, 5:43pm

Hi,

I also have the same problem. The block is basically a sample and hold. The output signal is not continuous time, despite the warning.
I am not sure about the correct method of speed test. However, it seems to me the simulation time increases specially when it is used with other blocks.

Ali