Simulation::Automate - A Simulation Automation Tool
The set of modules is called Simulation::Automate.
The tool itself is called SynSim, the command synsim.
synsim
a unix-like system
perl 5
gnuplot for postprocessing (optional)
use Simulation::Automate; &synsim();
SynSim is a generic template-driven simulation automation tool. It works with any simulator that accepts text input files and generates text output (and even those that don't. See EXAMPLES for special cases). It executes thousands of simulations with different input files automatically, and processes the results. Postprocessing facilities include basic statistical analysis and automatic generation of PostScript plots with Gnuplot. SynSim is entirely modular, making it easy to add your own analysis and postprocessing routines.
tar -xvf Simulation-Automate-0.9.5.tar.gz
cd Simulation-Automate-0.9.5 perl Makefile.PL
make
make test
make install
make localinstall
or
perl -e "use Simulation::Automate;&Simulation::Automate::localinstall();"
make setup or perl -e "use Simulation::Automate;&Simulation::Automate::setup();"
The archive structure is as follows:
README Makefile.PL Automate.pm Automate/ Remote.pm PostProcLib.pm Analysis.pm Dictionary.pm PostProcessors.pm eg/ synsim synsim.data ErrorFlags.data Histogram.data SweepVar.data Expressions.data gnuplot.data SOURCES/ bufsim3.cc MersenneTwister.h TEMPLATES/ DEVTYPES/ SIMTYPES/ bufsim3.templ
SynSim must be configured for use with your simulator. This is done by providing template and source files, creating (or modifying) datafiles and (optionally) customizing some modules for postprocessing the simulation results. All files must be put in a particilar directory structure:
You can use "make setup" to create a SynSim directory structure. If you want to create it manually, this is the structure:
YourProject/ synsim YourDataFile.data [SOURCES/] TEMPLATES/ [DEVTYPES/] SIMTYPES/ YourSimTempl.templ [Simulation/SynSim/] [Dictionary.pm] [PostProcessors.pm]
The synsim script contains the 2 lines from the SYNOPSIS. The local Simulation/Automate modules are only required if you want to customize the postprocessing (highly recommended).
Copy all files which are needed "read-only" by your simulator (e.g. header files, library files) to SOURCES/. This directory is optional.
Template files are files in which simulation variables will be substituted by their values to create the input file for your simulator. SynSim can create an input file by combining two different template files, generally called device templates and simulation templates. This is useful in case you want to run different types of simulations on different devices, e.g. DC analysis, transient simulations, small-signal and noise analysis on 4 different types of operation amplifiers. In total, this requires 16 different input files, but only 8 different template files (4 for the simulation type, 4 for the device types).
To create a template file, start from an existing input file for your simulator. Replace the values of the variables to be modified by SynSim by a SynSim variable name (e.g. var1 = 2.5 => var1 = _VAR1).
Put the template files in TEMPLATES/SIMTYPES and TEMPLATES/DEVTYPES.
There must be at least one template file in SIMTYPES; files in DEVTYPES are optional. SynSim will check both directories for files as defined in the datafile. If a matching file is found in DEVTYPES, it will be prepended to the simulation template from SIMTYPES. This is useful if the datafile defines multiple simulation types on a particular device (See "DATAFILE DESCRIPTION" for more information).
NOTE:
SynSim creates a run directory ath the same level as the SOURCES and TEMPLATES directories. All commands (compilations etc.) are executed in that directory. As a consequence, paths to source files (e.g. header files) should be "../SOURCES/sourcefilename".
../SOURCES/
The datafile is the input file for synsim. It contains the list of simulation variables and their values to be substituted in the template files, as well as a number of configuration variables (See "DATAFILE DESCRIPTION" for more information).
The PostProcessing.pm module contains routines to perform postprocessing on the simulation results. A number of generic routines are provided, as well as a library of functions to make it easier to develop your own postprocessing routines. See POSTPROCESSING for a full description).
The Dictionary.pm module contains descriptions of the parameters used in the simulation. These descriptions are used by the postprocessing routines to make the simulation results more readable. See DICTIONARY for a full description).
The datafile defines which simulations to run, with which parameter values to use, and how to run the simulation. By convention, it has the extension .data.
.data
The datafile is a case-sensitive text file with following syntax:
Comments are preceded by '#'. Comments, blanks and empty lines are ignored
Simulation variables ("parameters") are in UPPERCASE with a leading '_', and must be separated from their values with a '='.
Configuration variables ("keywords") are in UPPERCASE, and must be separated from their values with a ':'.
Lists of values have one or more items. Valid list separators are ',',';' and, for a 2-element list, '..'.
If a (','- or ';'-separated) list has 3 elements START,STOP,STEP |START|<|STOP| and |STEP|<|STOP-START|, then this list will be expanded as a for-loop from START to STOP with step STEP.
These must be lines containing only the simulation type
The main purpose of the datafile is to provide a list of all variables and their values to be substituted in the template files. The lists of values for the variables can be used in two different ways:
A simulation will be performed for every possible combination of the values for all parameters.
Example:
_PAR1 = 1,2 _PAR2 = 3,4,5
defines 6 simulations: (_PAR1,_PAR2)=(1,3),(1,4),(1,5),(2,3),(2,4),(2,5)
Simulation results for all values in ','-separated list are stored in a separate files.
If more than one ';'-separated list exists, they must have the same number of items. The values of all parameters at the same position in the list will be used.
_PAR1 = 0;1;2;4 _PAR2 = 3;4;5;6
defines 3 simulations: (_PAR1,_PAR2)=(0,3);(1,4);(2,5);(4,6)
Values from ';'-separated lists are processed one after another while are values for all others parameters are kept constant. In other words, the ';'-separated list is the innermost of all nested loops.
Simulation results for all values in the ';'-separated list are stored in a common file. For this reason, ';'-separated lists are preferred as sweep variables (X-axis values), whereas ','-separated lists are more suited for parameters (sets of curves).
Example: consider simulation of packet loss vs number of buffers with 3 types of buffer and 2 different traffic distributions.
_NBUFS = 4;8;16;32;64;128 _BUFTYPE = 1,2,3 _TRAFDIST = 1,2
This will produces 6 files, each file containing the simulation results for all values of _NBUFS. A plot of this simulation would show a set of 6 curves, with _NBUFS as X-axis variable.
This is a special case where the list has exactly three elements From;To;Step and following condition holds:
(|From|<|To|) AND (|Step|<|To-From|)
_NBUFS = 16;64;8 # from 16 to 64 in steps if 8: 16;24;32;40;48;56;64
This is a shorthand for a ';'-separated list if the value increase in steps of 1. E.g. 0..7 is equivalent to 0;1;2;3;4;5;6;7.
A number of variables are provided to configure SynSim's behaviour:
If the value of INCLUDE is an exisiting filename, this datafile will be included on the spot.
The program that runs the input file, i.e. the simulator command (default: perl).
Extension of input file (default: .pl)
Extension of template files (default: .templ)
The type of simulation to perform. This can be a ','-separated list. SynSim will look in TEMPLATES/SIMTYPES for a file with TEMPL and SIMTYPE
The name of the device on which to perform the simulation. If defined, SynSim will look in TEMPLATES/DEVTYPES for a file with TEMPL and DEVTYPE, and prepend this file to the simulation template before parsing.
A Perl regular expression to filter the output of the simulation (default : .*).
Name of the routine to be used for the result analysis (postprocessing). This routine must be defined in PostProcessors.pm (default: SweepVar, a generic routine which assumes a sweep for one variable and uses all others as parameters).
The name of the variable to be sweeped. Mandatory if the postprocessing routine is SweepVar.
The name of the variable to normalise the results with. The results will be divided by the corresponding value of the variable.
The number of times the simulation has to be performed. For statistical work.
The column in the output file which contains simulation results (default: 2). Mandatory if using any of the generic postprocessing routines.
String describing the simulation, for use in the postprocessing.
Variables to allow more flexibility in the customization of the plots. XLABEL and YLABEL are the X and Y axis labels. LOGSCALE is either X, Y or XY, and results in a logarithmic scale for the chosen axis. STYLE is the gnuplot plot style (lines, points etc)
Variables to allow more flexibility in the customization of the plots (not implemented yet).
The SynSim datafile has support for expressions, i.e. it is possible to express the value list of a variable in terms of the values of other variables.
# average packet length for IP dist _MEANPL = ((_AGGREGATE==0)?2784:9120) # average gap width _MEANGW= int(_MEANPL*(1/_LOAD-1)) # average load _LOAD = 0.1;0.2;0.3;0.4;0.5;0.6;0.7;0.8;0.9 # aggregate _AGGREGATE = 0,12000
The variables used in the expressions must be defined in the datafile, although not upfront. Using circular references will not work. The expression syntax is Perl syntax, so any Perl function can be used. Due to the binding rules, it is necessary to enclose expressions using the ternary operator ?: with brackets (see example).
The next sections (DICTIONARY and POSTPROCESSING) are optional. For instructions on how to run SynSim, go to "RUNNING SYNSIM".
The Dictionary.pm module contains descriptions of the parameters used in the simulation. These descriptions are used by the postprocessing routines to make the simulation results more readable. The dictionary is stored in an associative array called make_nice. The description of the variable is stored in a field called 'title'; Descriptions of values are stored in fields indexed by the values.
make_nice
Following example illustrates the syntax:
# Translate the parameter names and values into something meaningful %Dictionary::make_nice=( _BUFTYPE => { title=>'Buffer type', 0=>'Adjustable', 1=>'Fixed-length', 2=>'Multi-exit', }, _YOURVAR1 => { title=>'Your description for variable 1', }, _YOURVAR2 => { title=>'Your description for variable 2', 'val1' => 'First value of _YOURVAR2', 'val3' => 'Second value of _YOURVAR2', }, );
Postprocessing of the simulation results is handled by routines in the PostProcessors.pm module. This module uses the PostProcLib.pm and Analysis.pm.
PostProcessors.pm
PostProcLib.pm
Analysis.pm
Routines to perform analysis on the simulation results in the PostProcessors module. In general you will have to create your own routines, but the version of PostProcessors.pm in the distribution contains a number of more or less generic postprocessing routines. All of these have hooks for simple functions that modify a file in-place. To call these functions, include them in the datafile with the PREPROCESSOR variable. e.g:
PREPROCESSOR
PREPROCESSOR : modify_results_file
All functions must be put in PostProcessors.pm, and the template could be like this:
sub modify_results_file { my $resultsfile=shift; open(IN,"<$resultsfile"); open(TMP,">$resultsfile.tmp"); while(<IN>) { #Do whatever is required } close IN; close TMP; rename "$resultsfile.tmp","$resultsfile" or die $!; }
Required configuration variables: SWEEPVAR
SWEEPVAR
Creates a plot using SWEEPVAR as X-axis and all other variables as parameters. This routine is completely generic. The SWEEPVAR value list must be semicolon-separated.
Required configuration variables: SETVAR,SWEEPVAR and CONDITION.
SETVAR
CONDITION
Creates a plot using SETVAR as X-axis; SWEEPVAR is checked against the condition COND (or CONDITION). The first value of SWEEPVAR that meets the condition is plotted. All other variables are parameters. This routine is completely generic. The SWEEPVAR value list must be semicolon-separated.
COND
Required configuration variables: SWEEPVAR, NRUNS
NRUNS
Optional configuration variables: NSIGMAS
NSIGMAS
Creates a plot using SWEEPVAR as X-axis and all other variables as paramters. Calculates average and 95% confidence intervals for NRUNS simulation runs and plots error flags. This routine is fully generic, the confidence interval (95% by default) can be set with NSIGMAS. See eg/ErrorFlags.data for an example datafile. The SWEEPVAR value list must be semicolon-separated.
Required configuration variables: NBINS
NBINS
Optional configuration variables: BINWIDTH, OUTPUT_FILTER_PATTERN, NSIGMAS
BINWIDTH
OUTPUT_FILTER_PATTERN
Creates a histogram of the simulation results. This requires the simulator to produce raw data for the histograms in a tabular format. The configuration variable OUTPUT_FILTER_PATTERN can be used to 'grep' the simulator output. When specifying logscale X or XY for the plot, the histogram bins will be logarithmic. See eg/Histogram.data for an example. The number of bins in the histogram must be specified via NBINS. The width of the bins can be set with BINWIDTH.
In a lot of cases you will want to create your own postprocessing routines. To make this easier, a library of functions is at your disposal. This library resides in the PostProcLib.pm module.
Following functions are exported:
&prepare_plot # does what it says. see example below &gnuplot # idem. Just pipes the first argument string to gnuplot. The option -persist can be added to keep the plot window after gnuplot exits. &gnuplot_combined # See example, most useful to create nice plots. Looks for all files matching ${simtempl}-${anatempl}-*.res, and creates a line in the gnuplot script based on a template you provide.
Following variables are exported (see PostProcLib.pm for a full list):
%simdata # contains all simulation variables and their value lists @results # an array of all results for a sweep (i.e. a var with a ';'-sep. value list $sweepvar # SWEEPVAR $normvar # NORMVAR $sweepvals # string containing all names and values of parameters for the sweep, joined with '-' $datacol # DATACOL $count # cfr. OUTPUT FILES section $simtempl # SIMTYPE $anatempl # ANALYSIS_TEMPLATE $dirname # name of run directory. cfr. OUTPUT FILES section $last # indicates end of a sweep $verylast # indicates end of all simulations $sweepvartitle # title for SWEEPVAR (from Dictionary.pm) $title # TITLE $legend # plot legend (uses Dictionary.pm) $legendtitle # plot legend title (uses Dictionary.pm) $xlabel # XLABEL $ylabel # YLABEL $logscale # LOGSCALE $plot # corresponds to -p flag $interactive # corresponds to -i flag
An example of how all this is used:
sub YourRoutine { ## Unless you want to dig really deep into the code, start all your routines like this: ## Get all arguments, to whit: $datafilename,$count,$dataref,$flagsref,$returnvalue my @args=@_; ## But don't bother with these, following function does all the processing for you: &prepare_plot(@args); ## this makes all above-listed variables available ## Define your own variables. ## As every variable can have a list of values, ## $simdata{'_YOURVAR1'} is an array reference. my $yourvar=${$simdata{'_YOURVAR1'}}[0]; my @sweepvarvals=@{$simdata{$sweepvar}}; ## $verylast indicates the end of all simulations if($verylast==0) { ## what to do for all simulations ## example: parse SynSim .res file and put into final files for gnuplot open(HEAD,">${simtempl}-${anatempl}-${sweepvals}.res"); open(IN,"<${simtempl}_C$count.res"); while(<IN>) { /\#/ && !/Parameters|$sweepvar/ && do { ## do something with $_ print HEAD $_ }; } close IN; close HEAD; my $i=0; foreach my $sweepvarval ( @sweepvarvals ) { open(RES,">>${simtempl}-${anatempl}-${sweepvals}.res"); print RES "$sweepvarval\t$results[$i]"; close RES; $i++; } ## $last indicates the end of a sweep if($last) { ## $interactive corresponds to the -i flag if($interactive) { ## do something, typically plot intermediate results my $gnuplotscript=<<"ENDS"; # your gnuplot script here ENDS &gnuplot($gnuplotscript); } # if interactive } # if last } else { ## On the very last run, collect the results into one nice plot ## You must provide a template line for gnuplot. Next line is a good working example. ## This line will be eval()'ed by the &gnuplot_combined() routine. ## This means the variables $filename and $legend are defined in the scope of this routine. ## Don't locally scoped put variables in there, use the substitution trick as below or some other way. #this is very critical. The quotes really matter! # as a rule, quotes inside gnuplot commands must be escaped my $plotlinetempl=q["\'$filename\' using (\$1*1):(\$_DATACOL) title \"$legend\" with lines"]; $plotlinetempl=~s/_DATACOL/$datacol/; ##this is a trick, you might try to eval() the previous line or something. TIMTOWDI :-) my $firstplotline=<<"ENDH"; # header for your gnuplot script here ENDH &gnuplot_combined($firstplotline,$plotlinetempl); } } #END of YourRoutine()
A module for basic statistical analysis is also available (Analysis.pm). Currently, the module provides 2 routines:
To calculate average, standard deviation, min. and max. of a set of values.
Arguments:
$file: name of the results file. The routine requires the data to be in whitespace-separated columns. $par: Determines if the data will be differentiated before processing ($par='DIFF') or not (any other value for $par). Differentiation is defined as subtracting the previous value in the array form the current value. A '0' is prepended to the array to avoid an undefined first point. $datacol: column to use for data $title: optional, a title for the histogram $log: optional, log of values before calculating histogram or not ('LOG' or '')
Use: my $file="your_results_file.res"; my $par='YOURPAR'; my $datacol=2; my %stats=%{&calc_statistics($file,[$par, $datacol])};
my $avg=$stats{$par}{AVG}; # average my $stdev=$stats{$par}{STDEV}; # standard deviation my $min=$stats{$par}{MIN}; # min. value in set my $max=$stats{$par}{MAX}; # max. value in set
To build histograms. There are 3 extra arguments:
$nbins: number of bins in the histogram $min: force the value of the smallest bin (optional) $max: force the value of the largest bin (optional)
use: my $par='DATA'; my %hists=%{&build_histograms("your_results_file.res",[$par,$datacol],$title,$log,$nbins,$min,$max)};
NOTE: Because the extra arguments are last, the $title and $log arguments can not be omitted. If not needed, supply ''.
The SynSim script must be executed in a subdirectory of the SynSim directory which contains the TEMPLATES subdir and the datafile (like the Example directory in the distribution).
The command line is as follows:
./synsim [-h -i -p -w -v -N -P -f] [datafile] [remote hostname]
The synsim script supports following command line options:
none: defaults to -f synsim.data -f [filename]: 'file input'. Expects a file containing info about simulation and device type. -p : plot. This enables generation of postscript plots via gnuplot. A postprocessing routine is required to generate the plots. -i : interactive. Enables generation of a plot on the screen after every iteration. Assumes -p -v. A postprocessing routine is required to generate the plots. -v : 'verbose'. Sends simulator output to STDOUT, otherwise to the [rundir]/simlog file -w : 'warn'. Show warnings about undefined variables -N : 'No simulations'. Perform only postprocessing. -P : 'Plot only'. Only plots the generated PostScript file. -h, -? : short help message
If [remote hostname] is provided, SynSim will try to run the simulation on the remote host.
The current implementation requires:
-ssh access to remote host
-scp access to remote host
-rsync server on the local host
-or,alternatively, an NFS mounted home directory
-as such, it will (probably) only work on Linux and similar systems
SynSim creates a run directory {SIMTYPE}-[datafile without .data]. It copies all necessary template files and source files to this directory; all output files are generated in this directory.
{SIMTYPE}-
SynSim generates following files:
Output files for all simulation runs.
The names of these files are are {SIMTYPE}_C[counter]_[simulation number].out
{SIMTYPE}_C
.out
counter is increased with every new combination of variables in ','-separated lists
simulation number is the position of the value in the ';'-separated list.
Combined output file for all values in a ';'-separated list.
The names of these files are are {SIMTYPE}_C[counter]_.out
_.out
counter is increased with every new combination of variables in ','-separated lists.
Only the lines matching /OUTPUT_FILTER_PATTERN/ (treated as a Perl regular expression) are put in this file.
/OUTPUT_FILTER_PATTERN/
Combined output file for all values in a ';'-separated list, with a header detailing all values for all variables.
The names of these files are are {SIMTYPE}_C[counter].res,
.res
Only the lines in the .out files matching /OUTPUT_FILTER_PATTERN/ (treated as a Perl regular expression) are put in this file.
Separate input files for every item in a ';'-separated list.
The names of these files are are {SIMTYPE}_[simulation number].{EXT}
{SIMTYPE}_
.{EXT}
simulation number is the position of the value in the list.
These files are overwritten for every combination of variables in ','-separated lists.
Here are some examples of how to use SynSim for different types of simulators.
Normal use: spice -b circuit.sp > circuit.out
With SynSim:
Copy circuit.sp to TEMPLATES/SIMTYPE/circuit.templ Replace all variable values with SynSim variable names.
e.g. a MOS device line in SPICE:
M1 VD VG VS VB nch w=10u l=10u
becomes
M1 VD VG VS VB _MODEL w=_WIDTH l=_LENGTH
TITLE: MOS drain current vs. length SIMTYPE : circuit COMMAND : spice -b inputfile > outputfile # Required for postprocessing OUTPUT_FILTER_PATTERN : id # keep only the drain current on the output file ANALYSIS_TEMPLATE : SweepVar # default template for simple sweep SWEEPVAR : _L # we sweep the length, the other variables are parameters DATACOL: 2 # first col is the name _L = 1u;2u;5u;10u;20u;50u _W = 10u,100u _MODEL = nch
There are more possible keywords, cf. "DATAFILE DESCRIPTION".
./synsim -p -i -v -f IDvsL.data -p to create plots -i means interactive, so the plots are displayed during simulation -v for verbose output -f because the filename is not the default name
SynSim will run 12 SPICE simulations and produce 1 plot with all results.
All results are stored in the run directory, in this case:
circuit-IDvsL
Normal use: simplesim -a50 -b100 -c0.7
Output is saved in out.txt.
As simplesim does not take an input file, we create a wrapper simplesim.templ in TEMPLATES/SIMTYPE. This file is actually a template for a simple perl script:
system("simplesim -a_VAR1 -b_VAR2 -c_VAR3"); system("cp out.txt $ARGV[0]");
TITLE: simplesim test SIMTYPE : simplesim COMMAND : perl inputfile outputfile
./synsim -f test.data
SynSim will run without any messages and produce no plots.
simplesim-test
Normal use: Modify values for #if and #ifdef constants in the header file; then compile and run.
e.g.:
vi bufsim3.h g++ -o bufsim3 bufsim3.cc ./bufsim3 > outputfile
As bufsim3 does not take an input file, we create a wrapper bufsim3.templ in TEMPLATES/SIMTYPE. This file is actually a template for a perl script that writes the header file, compiles and runs the code:
open(HEADER,">bufsim3.h"); print HEADER <<"ENDH"; #define NBUFS _NBUFS #define NPACKETS _NPACK #AGGREGATE _AGGREGATE ENDH close HEADER; system("g++ -o bufsim3 bufsim3.cc"); system("./bufsim3 $ARGV[0]");
TITLE: bufsim3 test (_NBUFS, _NPACK) # will be substituted by the values SIMTYPE : bufsim3 COMMAND : perl inputfile outputfile
./synsim -w -v -f Aggregate.data
SynSim will run verbose and flag all variables not defined in the datafile.
bufsim3-Aggregate
Normal use: spectre circuit.scs -raw circuit.raw
Copy circuit.scs to TEMPLATES/SIMTYPE/circuit.templ Replace all variable values with SynSim variable names.
The .raw file is a binary file, so it should not be touched. SynSim creates output files with extension .out, and combines these with the headers etc. (cf. "OUTPUT FILES"). By keeping the extension .raw, the simulator output files will not be touched.
In the datafile:
TITLE: Spectre simulation with SPF output EXT: .scs COMMAND: spectre inputfile -raw outputfile.raw > outputfile
SynSim will process outputfile, but not outputfile.raw.
outputfile
outputfile.raw
To access the binary files, you'll have to write your own postprocessing routines. Most likely they will rely on an external tool to process the binary data. The files will be found in the run directory, and have names as described in "OUTPUT FILES", with the extra extension .raw.
This module is still Alpha, a lot of work remains to be done to make it more user-friendly. The main tasks is to add a GUI. A prototype can be found on my web site, it is already useful but too early to include here. The next version will also make it easier to create your own postprocessing routines.
Wim Vanderbauwhede <wim\x40motherearth.org>
Copyright (c) 2000,2002-2003 Wim Vanderbauwhede. All rights reserved. This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.
gnuplot http://www.ucc.ie/gnuplot/gnuplot.html
To install Simulation::Automate, copy and paste the appropriate command in to your terminal.
cpanm
cpanm Simulation::Automate
CPAN shell
perl -MCPAN -e shell install Simulation::Automate
For more information on module installation, please visit the detailed CPAN module installation guide.