This PR removes the early return in the distributed::Matrix::communicate. It is not possible to return early there because the MPI calls have to be executed by all processes, even if a process might not send anything.
I've also removed an if in the matrix apply, since IMO it complicated the control flow a bit.
This PR removes the early return in the
distributed::Matrix::communicate. It is not possible to return early there because the MPI calls have to be executed by all processes, even if a process might not send anything.I've also removed an
ifin the matrix apply, since IMO it complicated the control flow a bit.