[gmx-developers] The equestion on the implement of thread parallel in gromacs 4.5+

[Bogdan Costescu]

On Tue, May 1, 2012 at 5:34 PM, Yorquant Wang <wangykoo at gmail.com> wrote:
>     I recently want to add some myself pull code in the gromacs4.5.4 kernel.

The pull code is only called from mdrun, but mdrun is also able to run
on multiple nodes using MPI - so you should not rely on the code
always running multithreaded. The thread_mpi internal library
implements the subset of the MPI calls which are used in GROMACS, so
the function calls are the same and which ones are used (an external
MPI lib or thread_mpi) is decided at compile time. You don't need to
concern yourself with sharing memory, but use the MPI data model. If
you only have experience with OpenMP/threads programming, it's time to
look up some MPI tutorials.

Good luck!
Bogdan