[gmx-developers] Domain decomposition and consistency of state and state_global

Berk Hess hess at kth.se
Wed Mar 21 22:55:10 CET 2018


I don't understand what you mean with global coordinates. The global 
coordinates are only available on the master rank and then only on steps 
where they have been collected. You should only be checking the local 



On 03/21/2018 08:05 PM, Vedran Miletić wrote:
> Hello,
> I'm developing the conditional stop feature (proposal at [1], a somewhat
> working (but not yet mergeable) code dump at [2]; it's using 2016, but
> will be rebased to 2018 and master when done). Non-DD code is rather
> easy to get up and running, but I'm struggling with understanding how am
> I supposed to use state and state_global with DD.
> I figured out how ga2la_get() is intended to be used, but the local atom
> coordinates I get (relevant part of the log file is in [3]) are
> consistent with global atom coordinates in 0 (modulo 8x8x8 box PBC,
> which is taken into account when computing distances), but not in step 1000.
> I assume that new coordinates have not been communicated yet and that
> I'm placing the code somewhere wrongly. Aside from the placement in [4],
> I tried placing it in both places after dd_collect_state() in the main
> while loop, but it did not help.
> Ideas are welcome.
> Thanks,
> Vedran
> [1] https://redmine.gromacs.org/issues/2224
> [2]
> https://github.com/hits-mbm/gromacs-developments/tree/conditional-stop-release-2016
> [3]
> https://vedran.miletic.net/projects/gromacs/conditional-stop-gmx-debug.log
> [4]
> https://github.com/HITS-MBM/gromacs-developments/blob/conditional-stop-release-2016/src/programs/mdrun/md.cpp#L1276

More information about the gromacs.org_gmx-developers mailing list