[gmx-developers] different energies with different compilers in implicit solvent

[Berk Hess]

Hi,

There is still a bug in the GB code.
For now we'll assume that is the issue.

I fixed the upstring issue, the others will all fix at once before we
release a beta.

Berk

Servaas Michielssens wrote:
> Hi,
>
> I noticed a problem with the implicit solvent model but I am not
> sure if it is related to the compiler issues that were
> previously discussed on the list. I get different energies with
> different compilers. I  use a git version I pulled yesterday.
> I get an energy that looks correct with (-299.814087kj/mol):
> gcc version 4.3.2 (on ubuntu)
> icc Version 11.1 (SUSE)
> But different energies with(-583.496033kj/mol):
> gcc version 3.4.6 (SUSE)
> gcc version 4.1.2 (SUSE)
> With all of them I compile using single precision and without any MPI options (./configure --without-x).
> If needed you can always send me an email to provide you with my input files, I failed to attach them in a mail to the list. 
> It should be noted that in vacuum I get the same energies with all compilers.
>
> These are the warnings
> I get with gcc version 3.4.6  (no warnings with icc compiler, or gcc version 4.3.2)
> checkpoint.c: In function ‘read_checkpoint_simulation_part’:
> checkpoint.c:1932: warning: implicit declaration of function ‘upstring’
> statutil.c: In function ‘parse_common_args’:
> statutil.c:254: warning: ‘i’ may be used uninitialized in this function
> statutil.c:255: warning: ‘select’ may be used uninitialized in this function
> genborn.c: In function ‘table_log’:
> genborn.c:183: warning: dereferencing type-punned pointer will break
> strict-aliasing rules
> pppm.c: In function ‘convolution’:
> pppm.c:389: warning: dereferencing type-punned pointer will break
> strict-aliasing rules
> pme.c: In function ‘solve_pme’:
> pme.c:1015: warning: dereferencing type-punned pointer will break
> strict-aliasing rules
>
> kind
> regards,
>
> servaas
>