[gmx-developers] Found 2 problems in gromacs 4.0.7

Ake Sandgren ake.sandgren at hpc2n.umu.se
Mon Feb 15 15:34:29 CET 2010


On Mon, 2010-02-15 at 14:37 +0100, Ake Sandgren wrote:
> Hi!
> 
> When building gromacs 4.0.7 with Pathscale i have found that for at
> least one set of input data there is an uninitialized variable in
> mdlib/clincs.c in lincs_rmsd. lincsd->rmsd_data[0] (and 1) is
> uninitialized.
> 
> The same input data will generate a second problem later in the code if
> i compile gromacs with pathscale's -zerouv flag (zero initializes
> uninitialized variables). In mdlib/nsgrid.c in fill_grid, cg_cm[cg][d]
> is NaN in the for loop after bUse=TRUE.
> 
> Do you want input data to test with?

The last problem also appears when built with intel. And i forgot to say
that this is an mpi version (openmpi if that matters).
That cg_cm[cg][d] being NaN later results in cell_index[cg] having the
value -2147483635 which is way outside its limits.

Building mdrun in double precision make this problem vanish (as far as i
have been able to test)

-- 
Ake Sandgren, HPC2N, Umea University, S-90187 Umea, Sweden
Internet: ake at hpc2n.umu.se   Phone: +46 90 7866134 Fax: +46 90 7866126
Mobile: +46 70 7716134 WWW: http://www.hpc2n.umu.se




More information about the gromacs.org_gmx-developers mailing list