The Designer's Guide Community Forum

The Designer's Guide Community Forum
https://designers-guide.org/forum/YaBB.pl
Simulators >> Circuit Simulators >> Getting convergence info from Spectre
https://designers-guide.org/forum/YaBB.pl?num=1190297332

Message started by Marq Kole on Sep 20^th, 2007, 7:08am

Title: Getting convergence info from Spectre
Post by Marq Kole on Sep 20^th, 2007, 7:08am

Hi,

I would like to verify the performance of a behavioral model in transient by checking out how much time the simulation that I use it in needs for the implicit DC, as well as how many Newton iterations it performs during each transient step. Is there a way to get that information from Spectre? I tried to search the documentation that comes through spectre -h, but couldn't find any reference.

Marq

Title: Re: Getting convergence info from Spectre
Post by Geoffrey_Coram on Sep 24^th, 2007, 9:02am

Hi, Marq -
I've had success in some simulators using the $debug system task, which prints something for every newton iteration. Pipe the output to a file and use the unix "wc" to count the lines that contain your magic string. $strobe prints on convergence; you'd probably need a perl script to count how many $debugs happen between successive $strobes.

-Geoffrey

Title: Re: Getting convergence info from Spectre
Post by Marq Kole on Sep 24^th, 2007, 11:53pm

Hi Geoffrey,

Thanks - I had already considered that approach, but I had hoped there might be some built-in debug output from Spectre that could help me. The Verilog-A $debug approach will obviously have an effect on throughput time. I guess I will have to try the $debug approach.

Marq.

Title: Re: Getting convergence info from Spectre
Post by Geoffrey_Coram on Sep 25^th, 2007, 4:18am

You could try "audit=full" on the options line. Do spectre -help options and scroll down to the Annotation parameters section.

Title: Re: Getting convergence info from Spectre
Post by Marq Kole on Sep 26^th, 2007, 1:18am

OK, for the benefit of the community here is the $debug Verilog-A code as well as a very kludgy shell script that does the extraction of the iteration data. It now returns the average number of iterations and the maximum number of iterations. With a little more work the variation (in terms of a Poisson distribution, of course) could be extracted as well.

This is the Verilog-A code of a simple module that writes the data out to the logfile in case of Spectre.

Code:

module stepcount;

integer count;

analog
begin
if (analysis("static")) begin
$debug("MINOR (DC)");
$strobe("MAJOR (DC) %d", count);
end
else
if (analysis("tran") && !analysis("static")) begin
$debug("MINOR (TRAN)");
$strobe("MAJOR (TRAN) %d", count);
end
count = count + 1;
end

endmodule // stepcount

And this is a function in Korn shell (/bin/ksh) style that extracts the data from the log file and creates a number of local variables that can be written out at a later stage. With some minor effort this can be turned into a more efficient Perl or Python script that does all of it and more.

Code:

IterationCount() { # count the number of major and minor steps in
# the logfile $2 and number of accepted steps $3
# if the flag $1 is 1.
if [ $1 -gt 0 ]
then \
(( hasstepcount=1-$(grep -q -e '^MAJOR ' $2 ; echo $?) ))
if [ ${hasstepcount} -gt 0 ]
then \
(( nminordc=$(grep -c -e '^MINOR (DC)' $2) ))
(( nmajordc=$(grep -c -e '^MAJOR (DC)' $2) ))

(( nminortr=$(grep -c -e '^MINOR (TRAN)' $2) ))
(( nmajortr=$(grep -c -e '^MAJOR (TRAN)' $2) ))
(( aveminortr=${nminortr}*1.0/${nmajortr} ))
(( maxminortr=0 ))
linetext0=$(grep -n -m 1 -e '^MINOR (TRAN)' $2)
(( linenr0=$(python -c 'import sys ; s=sys.argv[1] ; r1=s.split(":")[0] ; print r1,' ${linetext0})-1 ))
for (( count=1 ; $3-${count} ; count=${count}+1 ))
do \
findtext="^MAJOR (TRAN) ${count}"
linetext1=$(grep -n -m 1 -e "${findtext}" $2)
(( linenr1=$(python -c 'import sys ; s=sys.argv[1] ; r1=s.split(":")[0] ; print r1,' ${linetext1}) ))
(( nlines=${linenr1}-${linenr0}-1 ))
if [ ${maxminortr} -lt ${nlines} ]
then \
(( maxminortr=${nlines} ))
fi
(( linenr0=${linenr1} ))
done
else \
IterationCount 0 $2 $3
fi
else \
(( nminordc=0 ))
(( nmajordc=1 ))
(( nminortr=0 ))
(( nmajortr=$3 ))
(( aveminortr=0.0 ))
(( maxminortr=0 ))
fi
}

Title: Re: Getting convergence info from Spectre
Post by jorobins on Sep 29^th, 2007, 6:00am

Thanks Marq. This looks fantastic. I'll try it out and see how it works for me.

Jose [smiley=thumbsup.gif]