Abinit 8.0.8b kpoints - core dump in parallel mode
Posted: Tue Oct 25, 2016 11:54 am
Dear all,
I am experiencing difficulties with a recent compilation of Abinit 8.0.8b on a HPC (amd_opteron) with OpenMPI 1.8.8, MPI-IO and RDMA support against the Intel compiler suite. The code runs fine with a standard grid only defined by ngkpt, but dumps when running it with the nshiftk and shiftk parameters (even when they are set to the standard values). I also observed a difference when the flag autoparal is set (which only works with the standard settings) or unset (which appears to allow the use of nshiftk and shiftk in ground state calculations). This appears to me as if there may be a problem with parallelism in the k-point routine, which may come from MPI or abinit? I will try a couple of modifications (serial compilation, gcc) to trace down the problem, but I wonder if someone has encountered a similar problem?
Thanks and Best Wishes,
Joachim.
I am experiencing difficulties with a recent compilation of Abinit 8.0.8b on a HPC (amd_opteron) with OpenMPI 1.8.8, MPI-IO and RDMA support against the Intel compiler suite. The code runs fine with a standard grid only defined by ngkpt, but dumps when running it with the nshiftk and shiftk parameters (even when they are set to the standard values). I also observed a difference when the flag autoparal is set (which only works with the standard settings) or unset (which appears to allow the use of nshiftk and shiftk in ground state calculations). This appears to me as if there may be a problem with parallelism in the k-point routine, which may come from MPI or abinit? I will try a couple of modifications (serial compilation, gcc) to trace down the problem, but I wonder if someone has encountered a similar problem?
Thanks and Best Wishes,
Joachim.