Mixing coarrays and scalapack

[OP1]

Is it possible to create a code that would mix coarrays and calls to scalapack routines? In other words, assuming that a coarray program spawns 10 images (ideally, on 10 different cores), would it be possible to have one of these images (the "master" image for the purpose of clarity) call a scalapak routine on all 10 cores?

In pseudo code:

! ...
SYNC ALL
! ... all images do some kind of work.
SYNC ALL
IF (THIS_IMAGE()==1) THEN
    ! The "master" image collects data from other images' coarrays, then
    ! calls a scalapack routine where work is distributed on all resources,
    ! and redistribute the results onto each image local coarrays.
END IF
SYNC ALL
! ... more work is done by all images.

[Steven_L_Intel1]

Really your question is equivalent to "I have an MPI program with 10 ranks and I want to call SCALAPACK to compute on those same 10 ranks". It would probably work, but might not be the most effective thing to do. By 10 cores do you mean a single shared memory system with 10 cores, or 10 distributed systems?

[OP1]

This scenario would involve a distributed system. 10 'nodes' with 10 coarray images (assuming 1 image per node and no migration of images from node to node), each node having a certain number of cores that would be leveraged with OpenMP for tasks associated with each image. At some regular intervals, the master image would make a call to SCALAPACK (with a process grid distributed over all nodes, and while the other images would be idle at that time, waiting at the SYNC ALL barrier).

As you pointed out, besides the feasibility of it, I am wondering about the efficiency of having two processes coexisting on the same node, even though one of them may be idling. Maybe somebody has already some experience with the strategy outlined above?