You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
PR #106 Adds user-specified communicators. This is OK if we have Vernier in one coupled model component. If we wanted to run Vernier in more than one coupled model component, we would need to either use a broader-scale communicator or have different sets of output files for each sub-communicator / sub-component.
Perhaps we could add a tag - set during the call to vernier.init(...) - which ultimately sets a data member in the MPIContext class, readable via a get_tag() method. This tag would then be added to output filenames.
I haven't included this under #106 because the change was getting big without this. Would be useful to have some discussion, anyhow. Depends on #106.
Beware: The Fortran layer would need to add c_null_char to any strings passed from Fortran. There is already a subroutine to do this in the Fortran layer.
The text was updated successfully, but these errors were encountered:
Allow client codes to set the MPI communicator used within Vernier. This
is needed to allow one component of a model to use Vernier. [Multiple components using Vernier simultaneously would need further work, since they'd all be trying to write to the same output filenames. See #116 ]
Changes:
* Adds `MPIContext` class.
* Global Vernier object now has `init` and `finalize` methods.
* Most C++ and Fortran tests now use MPI, if that only means that MPI is
initialised.
Allow client codes to set the MPI communicator used within Vernier. This
is needed to allow one component of a model to use Vernier. [Multiple components using Vernier simultaneously would need further work, since they'd all be trying to write to the same output filenames. See #116 ]
Changes:
* Adds `MPIContext` class.
* Global Vernier object now has `init` and `finalize` methods.
* Most C++ and Fortran tests now use MPI, if that only means that MPI is
initialised.
PR #106 Adds user-specified communicators. This is OK if we have Vernier in one coupled model component. If we wanted to run Vernier in more than one coupled model component, we would need to either use a broader-scale communicator or have different sets of output files for each sub-communicator / sub-component.
Perhaps we could add a tag - set during the call to
vernier.init(...)
- which ultimately sets a data member in theMPIContext
class, readable via aget_tag()
method. This tag would then be added to output filenames.I haven't included this under #106 because the change was getting big without this. Would be useful to have some discussion, anyhow. Depends on #106.
Beware: The Fortran layer would need to add
c_null_char
to any strings passed from Fortran. There is already a subroutine to do this in the Fortran layer.The text was updated successfully, but these errors were encountered: