Feat: Add Wrappers for MPI_AllGatherv #124
Conversation
There was a problem hiding this comment.
Same concern I've with this PR as I mentioned in the other PR, i.e. we also need to make sure (by standalone tests) that the way the subroutine is used in pFUnit should be added as a standalone test case.
I see two uses of MPI_Allgather_v in pFUnit:
call Mpi_allGatherV( &
& values, size(values), MPI_INTEGER, &
& global_list, counts, displacements, MPI_INTEGER, &
& this%mpiCommunicator, ier)and the other is:
call Mpi_AllgatherV( &
& values_, size(values), MPI_LOGICAL, &
& list, counts, displacements, MPI_LOGICAL, &
& this%mpiCommunicator, ier)maybe we can ensure that we write test cases for different MPI_types and/or dimensions of array(s) (if any)
| c_sendbuf = c_MPI_IN_PLACE | ||
| else | ||
| c_sendbuf = c_loc(sendbuf) | ||
| end if |
There was a problem hiding this comment.
I think we should've a utility function for this now, which should do this work for us now.
There was a problem hiding this comment.
Yes we can add utility function like we have for comm_world and other things. Will make a subsequent PR for that.
|
Thanks for posting it's usecase in pFUnit. I will do this in evening!!(ASAP), for both PRs |
|
Why don't we do one thing, As you mentioned the use case for them it includes the MPI_LOGICAL and MPI_CHARACTER datatype to be implemented which i would like to handle in separate PR. Get CI fixed -> merge pending PRs -> Add bindings with LOGICAL and CHARACTER -> Add pfUnit like tests for both subroutines (which will invlove to add subroutine which can take character arrays as an input ,currently it takes int/real(8) as input for As doing all in these PRs would clutter things up |
Yes, that's the idea. I just wanted you to be aware that I can't merge PR's for now, and any subsequent PR's might cause merge conflicts, which can be annoying sometimes. |
|
Now the CI are fixed, Should we merge pending PRs (there are 3 of them ) now? Need your approval |
Towards #121
MPI_Allgatherv is similar to MPI_Gatherv, but instead of gathering data to a single root process, it gathers data to all processes in the communicator.