Is the spectreVerilog simulator much faster with IC/5.1.41_u

[Frank Buergel]

Hello,

for our last project I was running a mixed signal simulation using the
spectreVerilog simulator. The software versions used were IC/5.0.33 and
IUS/5.4. Such a simulation took typically almost 2 days and most
simulation time was consumed by the verilog process (~110000 seconds).

After the project was finished I reran all my scripts with IC/5.1.41_usr2
to see if there are any major differences. I was very surpised that my
simulation took only ~2 hours now and was giving me the same as with
the older software. Now it was the spectre process which took most of the
simulation time, the verilog process consumed only ~660 seconds.

Did anyone else make similar experience ? What was changed in the software
that makes verilog running 200 times faster in this mixed signal
simulation ?

Frank

PS: The computer used for these simulations was running und Linux.

 

[Andrew Beckett]

Hi Frank,

On Fri, 07 Oct 2005 08:40:21 +0200, Frank Buergel <don.not.use@nospam.com>
wrote:

Hello,

for our last project I was running a mixed signal simulation using the
spectreVerilog simulator. The software versions used were IC/5.0.33 and
IUS/5.4. Such a simulation took typically almost 2 days and most
simulation time was consumed by the verilog process (~110000 seconds).

After the project was finished I reran all my scripts with IC/5.1.41_usr2
to see if there are any major differences. I was very surpised that my
simulation took only ~2 hours now and was giving me the same as with
the older software. Now it was the spectre process which took most of the
simulation time, the verilog process consumed only ~660 seconds.

Did anyone else make similar experience ? What was changed in the software
that makes verilog running 200 times faster in this mixed signal
simulation ?

Frank

PS: The computer used for these simulations was running und Linux.

Do you have Verilog-A models in your simulation, using the cross() function -
perhaps effectively doing some kind of A to D conversion? If so, the
transient option fastcross might have helped (see "spectre -h tran"). If you
try setting fastcross=no on the transient options, see if that makes it slower
again?

This option did not exist in IC5033 - it was specifically added to help improve
the performance of cross() when used in AMS Designer connect modules - but
it would also help in other cases where cross is used in Verilog-A, particularly
where the input signal to the cross has not gone through some kind of transition
filter.

Regards,

Andrew.

 

[Frank Buergel]

On Mon, 10 Oct 2005 18:32:49 +0100, Andrew Beckett wrote:

Do you have Verilog-A models in your simulation, using the cross() function -
perhaps effectively doing some kind of A to D conversion? If so, the
transient option fastcross might have helped (see "spectre -h tran"). If you
try setting fastcross=no on the transient options, see if that makes it slower
again?

Hi Andrew,

thanks for your reply. I tried to rerun the simulation using
IC/5.1.41_usr2 and the setting fastcross=no and it was still as fast as
before.

But I'm not sure if I did this setting at the right location...
Originally I ran my simulation from Analog Artist GUI. But there I didn't
find a switch fastcross. So I went to the simulation directory, added this
option at the end of the netlist input.scs and restarted the simulation
using the script runSimulation.

How can I check whether this switch was passed correctly to the spectre
simulator ? Where should I set this switch ?

Where can I find a useful description of this switch ? In the Spectre
Reference Manual it is mentioned twice and the onliest informations I
found about it are which values of fastcross are allowed. I found no
description what these options mean.

 

[Andrew Beckett]

On Wed, 12 Oct 2005 09:43:13 +0200, Frank Buergel <don.not.use@nospam.com>
wrote:

On Mon, 10 Oct 2005 18:32:49 +0100, Andrew Beckett wrote:

Do you have Verilog-A models in your simulation, using the cross() function -
perhaps effectively doing some kind of A to D conversion? If so, the
transient option fastcross might have helped (see "spectre -h tran"). If you
try setting fastcross=no on the transient options, see if that makes it slower
again?

Hi Andrew,

thanks for your reply. I tried to rerun the simulation using
IC/5.1.41_usr2 and the setting fastcross=no and it was still as fast as
before.

But I'm not sure if I did this setting at the right location...
Originally I ran my simulation from Analog Artist GUI. But there I didn't
find a switch fastcross. So I went to the simulation directory, added this
option at the end of the netlist input.scs and restarted the simulation
using the script runSimulation.

How can I check whether this switch was passed correctly to the spectre
simulator ? Where should I set this switch ?

Where can I find a useful description of this switch ? In the Spectre
Reference Manual it is mentioned twice and the onliest informations I
found about it are which values of fastcross are allowed. I found no
description what these options mean.

Hi Frank,

I'm out of the office at the moment and so didn't have a chance to check this (I
read news groups in Windows, but run the software under Linux), but this option
would have been on the _transient_ Options form.

There's not a huge amount about it in the documentation as I mentioned. It may
be covered more in the AMS Designer documentation...

Mind you, it sounds as if you've probably tried it with and without, and it
didn't make much difference. Perhaps this is a red herring? (sorry if that
doesn't translate - it means "false lead").

Regards,

Andrew.