[gmx-developers] grompp and vsites

[Michel Cuendet]

Hi all,

The last CVS version of grompp seems not to be working properly on 32bit 
machines in conjunction with vsites. I haven't tried to download the 
current GIT version yet, please tell me if there have been any updates. 
In the following, I use the gmx CVS version downloaded on 12.6.2009. 
Compilers and architecture are indicated below.

On a 32bit machine :
    $there/i686_gcc4.2.4/grompp -p $base_run/topol.top -c 
$base_run/equil/confout.gro -t $base_run/equil/traj.trr  -f 
$base_run/prod/grompp.mdp -o topol32.tpr >& grompp32.log

On a 64 bit machine :
    $there/em64t_gcc3.4.6/bin/grompp -p $base_run/topol.top -c 
$base_run/equil/confout.gro -t $base_run/equil/traj.trr  -f 
$base_run/prod/grompp.mdp -o topol64.tpr >& grompp64.log

grompp64.log is similar to what the 4.0.4 version gives. But 
grompp32.log shows incorrect behavior. It lacks the "Cleaning up 
constraints and constant bonded interactions with virtual sites" section 
for chains B and C (there are 3 chains in my system, A, B, C).

Also the tpr file produced on a 32 bit machine is not correct. Comparing 
topol64.tpr with the tpr file created with earlier versions of grompp 
indicates no differences. However,

    gmxcheck -s1 topol32.tpr -s2 topol64.tpr
[... usual output ...]
comparing inputrec
inputrec->rgbradii (1.000000e+00 - 2.000000e+00)
inputrec->gb_dielectric_offset (9.000000e-03 - 9.000000e-02)
inputrec->sa_algorithm (0 - 1)
inputrec->nstdhdl (10 - 1)
comparing top
comparing idef
idef->ntypes (1343 - 1316)
idef->functype[850][850] (13 - 4)
idef->iparam[850]1: theta= 1.09500000e+02, ktheta= 3.68191986e+02, r13= 
0.00000000e+00, kUB= 0.00000000e+00
idef->iparam[850]2: theta= 0.00000000e+00, ktheta= 0.00000000e+00, r13= 
0.00000000e+00, kUB= 0.00000000e+00
idef->iparam[851]1: theta= 1.09500000e+02, ktheta= 2.51039993e+02, r13= 
2.07399994e-01, kUB= 1.67360000e+04
idef->iparam[851]2: theta= 1.07500000e+02, ktheta= 4.30951996e+02, r13= 
0.00000000e+00, kUB= 0.00000000e+00
[... lots of output ...]


Furthermore, if you try to read an energy file with the 32bit version of 
grompp, you get a Segmentation Fault, whereas the 64bit version of 
grompp is fine.

    grompp -p $base_run/topol.top -c $base_run/equil/confout.gro -t 
$base_run/equil/traj.trr -e $base_run/equil/ener.edr -f 
$base_run/prod/grompp.mdp -o topol32bis.tpr
[ ... usual output ... ]
Reading Coordinates, Velocities and Box size from old trajectory
Will read whole trajectory
trn version: GMX_trn_file (single precision)
Last frame          4 time 1200.000
Using frame at t = 1200 ps
Starting time for run is 0 ps
Note: enx file_version 1, software version 2
Opened ../../dyna/AA_1//equil/ener.edr as single precision energy file
Note: enx file_version 1, software version 2
Segmentation fault

Cheers,
Michel