Resgrp:utilities

This is a share place for utilities, scripts, and Linux tricks (DMT).

Bash shell

jump command (DMT)

The space quota for the $HOME directory is considerably smaller that $WORK. So, one should save input files in $HOME and save all large files (i.e. output, checkpoint files,...) in $WORK. Then, it is quite convenient to branch the folder directories identically in $HOME and $WORK, so they mirror each other.

The following command "jump" changes directory from $HOME to $WORK and vice-versa assuming that these are identically branched (use dirsync for this). Just add the following lines to your .bashrc profile.

#! /bin/bash

jump () {
 unset dname
unset jname
dname=$(pwd)
jname=$(echo $dname | cut -f 2 -d '/')
if [ "$jname" = "work" ] ; then
   jname="${dname/work/home}"
   echo "cd $jname"
else
   jname="${dname/home/work}"
   echo "cd $jname"
fi
cd $jname
}

opt command

Allows you to view convergence criteria and step number from a Gaussian Log file.

#!/bin/bash

egrep ' out of| SCF Don| Converged| NO | YES| exceeded' $1 | grep -v '\\\\'

freq command

Prints out the first few frequencies, both in the 3N coordinates and the 3N-6 coordinates.

#!/bin/bash

egrep ' Low frequencies --| Frequencies --' $1 | grep -v '\\\\' | head -5

dirsync command

Makes two directories, one in home, one in work. Useful to ensure your two directory trees are equivalent (use jump to go between them). if either directory tree is missing a portion you will get an error from the shell. you can also add commands to automatically generate files in a particular folder, here I create a file to add notes and the jobscript file I use in the home directory.

  
#!/bin/bash

# creates directory in both home and work and copies gdvh13_118.sh to home dir

var1=$1;
name=$(pwd)
mkdir $name/${var1}

dname=$(pwd)
jname=$(echo $dname | cut -f 2 -d '/')
if [ "$jname" = "work" ] ; then
   jname="${dname/work/home}"
   mkdir $jname/${var1}
else
   jname="${dname/home/work}"
   mkdir $jname/${var1}
fi
path=$(echo $dname | cut -f 4- -d '/')

cp /home/lmt09/bin/gdvh13_118.sh /home/lmt09/$path/${var1}/gdvh13_118.sh
touch /home/lmt09/$path/${var1}/${var1}notes.txt

cas command

View the MCSCF iterations to check convergence and also check any orbital rotations.

#! /bin/awk -f

/ Step number/{
        print;
}
/WARNING/{
        line=$0;
        getline;
        line=line $0 ;
        print line;
}
/ITN/{
        print;
}

C shell

jump command

Same as the bash version but works for the c shell.

#!/bin/csh

unset dname
unset jname
set dname="$PWD"
set jname=(`echo $dname | cut -f 2 -d '/'`)
switch ( $jname )
   case work:
      set jname=(`echo $dname | sed 's/work/home/'`)
      echo "cd $jname"
      breaksw
   case home:
      set jname=(`echo $dname | sed 's/home/work/'`)
      echo "cd $jname"
      breaksw
endsw
cd $jname
exit

Perl script

gregap

graphically displays gaussian results from the log file using gnuplot (so you will need to be using X forwarding).

1. Download file

2. tar -xvzf gregap.tar.gz

3. ./install.sh (ensure this is executable -> chmod 755 install.sh)

gregap.tar.gz

Python

checkparallel (G16)

This script will print out the total time that Gaussian 16 spends on a particular link within a file, as well as the average CPU usage. This allows you to find bottlenecks in calculations and quickly determine how well parallelised the link is.

Note that additional printing (#p) is required for this to work, and it is only available for G16. Numerical errors are expected for short running times: the measurements for parallelisation become more precise for longer running times.

#!/usr/bin/python

import numpy as np
import sys

def check_parallel(filename):
    link_dict = {}
    with open(filename, 'r') as f:
        for line in f:
            l = line.strip()

            #Data is printed when a link is left
            if l.startswith("Leave Link"):
                l = l.split()   
                link = int(l[2])
                cpu = float(l[12])
                elap = float(l[14])

                #Avoid division and numerical errors
                if elap != 0:   
                    if link not in link_dict:
                        link_dict[link] = []    
                    link_dict[link].append([cpu, elap])

            #Get number of processors
            elif l.startswith("Will use up to"):
                l = l.split()   
                procs = int(l[4])

    print("Processors: {:d}".format(procs))

    for link in sorted(link_dict.keys()):
        para = link_dict[link]
        cpu = [p[0] for p in para]
        elap = [p[1] for p in para]

        cpu_ave = np.average(cpu)
        elap_ave = np.average(elap)

        para_ave = cpu_ave / elap_ave

        print("Link: {:4d}   Total Time: {:12.1f}   CPU/Elap: {:6.2f}".format(link, sum(elap), para_ave))

if __name__ == "__main__":
    args = sys.argv[1:]
    check_parallel(args[0])

Contacts

Here there is useful information and people to get in touch with:

Chief Service Technician (Mr. Peter Sulsh): p.sulsh@imperial.ac.uk‎

ICT Service Desk: service.desk@imperial.ac.uk‎

Allowances (Mrs. Althea Hartley-Forbes): a.hartley-forbes@imperial.ac.uk

HPC Systems Support Specialist (Mr. Matt Harvey): m.j.harvey@imperial.ac.uk

Printing posters: get in touch with ‎Dr. Ian R. Gould (i.gould@imperial.ac.uk). More information: http://www3.imperial.ac.uk/people/i.gould/poster%20printing1