Using a sander.LES.MPI compiled on 2010-03-02 I get errors for the two
segfaulting tests, but no errors for the others:
cd evb/malon_dbonds_umb_dg_UFF_3DG_qi_full_2D-PMF && ./Run.evb [aborts]
cd evb/malon_dbonds_umb_dg_UFF_3DG_qi_full_corrF && ./Run.evb [aborts]
I was able to track down the segfault to the pimd_part_spring_force()
subroutine; there is a do loop that begins around line 162 in pimd_force.f
that makes use of the array dmdlm(:), which is part of the pimd_vars
module. According to the description, "dmdlm = d mass / d lambda / mass
(for TI w.r.t. mass)". Further searching shows that this array is allocated
and initialized in the rdparm2() subroutine (rdparm.f) around line 557.
However, it is only allocated and set up if itimass is greater than 0, and
according to the nxtsec() call this requires a TI_MASS section in the prmtop
file (which doesn't appear to be present, at least in the 2D-PMF test case).
Setting itimass equal to 1 just causes sander.LES.MPI to complain about the
missing TI_MASS section. There is this comment:
! JVAN: dmdlm must be allocated here, not in pimd_init.
which seems to indicate at some point initialization used to be inside
pimd_init, but I have no idea how it used to be done and whether it required
the itimass variable to be set before.
Does anyone more familiar with EVB stuff have any idea what might be the
right way to fix this?
For the time being I will try and track down the other EVB errors I have
been seeing in the release candidate that don't seem present in the
2010-03-02 version.
-Dan
On Tue, Mar 30, 2010 at 8:52 AM, case <case.biomaps.rutgers.edu> wrote:
> On Mon, Mar 29, 2010, Daniel Roe wrote:
>
> > So on two separate platforms I get similar errors, all for EVB tests with
> > sander.LES.MPI:
>
> There is a good chance that this is related to the "state.f" stuff Mark put
> in. Someone who has some experience with this (or not) could look at the
> evb code for places where the old ener() arrays are being used.
>
> Another step would be to go back in CVS to before March 19, and see if the
> EVB
> was working at that point. That would help to pin the blame on (or
> absolve)
> the state.f code.
>
> ...thx...dac
>
>
> _______________________________________________
> AMBER-Developers mailing list
> AMBER-Developers.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber-developers
>
--
-------------------------
Daniel R. Roe
Postdoctoral Associate
SAS - Chemistry & Chemical Biology
610 Taylor Road
Piscataway, NJ 08854
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Tue Mar 30 2010 - 08:00:04 PDT
