[gmx-users] Gromacs 2016.3 FLOATING-POINT EXECPTION/DIVIDE-BY_ZERO errors

[Rainer Rutka]

HI!.
My name is Rainer. I am one of the module-/SW-maintainers
for the bwHPC-C5-project. "http://www.bwhpc-c5.de/en/index.php"

Since a couple of time we can't run Gromacs 2016.n on
our clusters. Here's one of the errors we received from
one of the users:

<snip>
I am using Gromacs on BwUniCluster for few months and I performed 
molecular Dynamics Simulations without any problem until the end of 
July. Since August, I cannot prepare the input file and if I prepare 
them elsewhere, the simulation crash although it was the same that the 
ones which perfectly worked few days before. It seems that I have not 
such problem smaller systems with less atoms.

uc1n997:43275] *** Process received signal ***
[uc1n997:43275] Signal: Floating point exception (8)
[uc1n997:43275] Signal code: Integer divide-by-zero (1)
[uc1n997:43275] Failing at address: 0x42fd74
[uc1n997:43275] [ 0] /usr/lib64/libpthread.so.0(+0xf5e0)[0x2b65e89805e0]
[uc1n997:43275] [ 1] gmx_mpi_d(__svml_idiv4_h9+0x64)[0x42fd74]
[uc1n997:43275] [ 2] 
/pfs/data1/software_uc1/bwhpc/common/chem/gromacs/5.1.2-openmpi-1.8-intel-15.0/bin/../lib64/libgromacs_mpi_d.so.1(count_bonded_distances+0x43c)[0x2b65e6b1f72c]
[uc1n997:43275] [ 3] 
/pfs/data1/software_uc1/bwhpc/common/chem/gromacs/5.1.2-openmpi-1.8-intel-15.0/bin/../lib64/libgromacs_mpi_d.so.1(pme_load_estimate+0x20)[0x2b65e6b1ef70]
[uc1n997:43275] [ 4] 
/pfs/data1/software_uc1/bwhpc/common/chem/gromacs/5.1.2-openmpi-1.8-intel-15.0/bin/../lib64/libgromacs_mpi_d.so.1(gmx_grompp+0x257a)[0x2b65e639e84a]
[uc1n997:43275] [ 5] 
/pfs/data1/software_uc1/bwhpc/common/chem/gromacs/5.1.2-openmpi-1.8-intel-15.0/bin/../lib64/libgromacs_mpi_d.so.1(_ZN3gmx24CommandLineModuleManager3runEiPPc+0x267)[0x2b65e6127e97]
[uc1n997:43275] [ 6] gmx_mpi_d(main+0xbc)[0x40c13c]
[uc1n997:43275] [ 7] 
/usr/lib64/libc.so.6(__libc_start_main+0xf5)[0x2b65e8baec05]
[uc1n997:43275] [ 8] gmx_mpi_d[0x40bfb9]
[uc1n997:43275] *** End of error message ***
</snip>

Gromacs 2016.3 was build this way (excerpt):

[...]
# #(3) Load required modules for build process
module load compiler/intel/16.0
module load mpi/openmpi/2.1-intel-16.0
module load numlib/mkl/11.3.4
module load devel/cmake/3.3.2
[...]
# double precission
cmake -DCMAKE_VERBOSE_MAKEFILE=ON -DGMX_MPI=ON -DGMX_GPU=OFF 
-DGMX_DOUBLE=ON -DGMX_THREAD_MPI=OFF -DGMX_FFT_LIBRARY=mkl 
-DMPIEXEC=${MPI_BIN_DIR}/mpirun -DREGRESSIONTEST_DOWNLOAD=OFF 
-DCMAKE_INSTALL_PREFIX=${TARGET_DIR}
make 2>&1 | tee ${LOG_DIR}/make_double.out
make install 2>&1 | tee ${LOG_DIR}/make-install_double.out
[...]

ANY HELP IS MUCH APPRECIATED.

Thanx in advance.

.-)

-- 
Rainer Rutka
University of Konstanz
Communication, Information, Media Centre (KIM)
  * High-Performance-Computing (HPC)
  * KIM-Support and -Base-Services
Room: V511
78457 Konstanz, Germany
+49 7531 88-5413