Support for MPI communicators which are provided externally

[aa-g]

Proposed Changes

Replace references to the global MPI communicator MPI_COMM_WORLD with SU2_MPI::GetComm() where appropriate.

This prevents deadlocking in cases where SU2 is given an MPI communicator other than MPI_COMM_WORLD, for example when creating an instance of CDriver through the Python interface.

Related Work

Issue #1179

PR Checklist

• I am submitting my contribution to the develop branch.
• My contribution generates no new compiler warnings (try with the '-Wall -Wextra -Wno-unused-parameter -Wno-empty-body' compiler flags, or simply --warnlevel=2 when using meson).
• My contribution is commented and consistent with SU2 style.
• I have added a test case that demonstrates my contribution, if necessary.
• I have updated appropriate documentation (Tutorials, Docs Page, config_template.cpp) , if necessary.

[pcarruscag]

Thank you for opening the PR!

Delete line 470 of mpi_structure.hpp please.

[aa-g]

Thank you for opening the PR!

Delete line 470 of mpi_structure.hpp please.

No worries. It took a while to get it going due to some issues on my side catching up with 7.1.0 (in particular, compiling against MPICH).

Regarding the deletion of the dummy MPI_COMM_WORLD from mpi_structure.hpp: the value is still referenced in several places, namely line 32 of mpi_structure.cpp, and in each of the SU2_CFD.cpp, SU2_DEF.cpp, SU2_DOT.cpp, SU2_GEO.cpp, SU2_SOL.cpp, where MPI_COMM_WORLD would be the default communicator.

[pcarruscag]

Add a method to the MPI wrappers to get the global communicator or something equivalent, the only way to make this kind of change future-compatible is to break compilation.

[pcarruscag]

I merged #1178 earlier and fixed the old and new conflicts, some MPI_COMM_WORLD's will have been re-introduced in CFVMFlowSolverBase.hpp and .inl because in my previous 2 PR's I moved a lot of code.

[aa-g]

I've removed the dummy MPI_COMM_WORLD altogether now. To keep the changes minimal, I've just changed the way that CBaseMPIWrapper::currentComm is initialised in mpi_structure.cpp, rather than introducing a new method (which would only be used once, at this location).

Now, the CBaseMPIWrapper::currentComm is initialised to MPI_COMM_WORLD if compiling with MPI, otherwise it is set to the dummy value 0, consistent with the previous behaviour.

[pcarruscag]

I like this solution.
The rank and size are updated by the SetComm function so everything should work ok when creating a driver from scratch.

[aa-g]

Glad to hear that. Is anything still missing ? Not sure how to go about the failing check for code complexity ...

[pcarruscag]

Not for me, I'll ask around the developers meeting tomorrow if anyone else has comments, codefactor glitches out sometimes no need to do anything.

By the way would you like to be added to su2code? You can then have branches here instead of in your fork, makes it simpler to manage future PR's you might open.

[aa-g]

Yes, I suppose it would be more convenient that way. Thanks !