Parallelization (again) - MPI to the rescue!

Ha, you mean an ugly name of an ugly function until we have all language features to come up with a nice map function? Fine with me.

What is making our life difficult is the absence of a ragged data structure type. Having that would avoid those integer arrays for the parameters and the data. I am not sure how one could avoid the integer arrays of f_out though.

Your example with closures looks nice. Where I fail to get my head around for now is how we can map this onto the boost MPI commands scatter/gather which we should feed with essentially continuous chunks of memory to work efficiently. So the interface with boost::mpi is low-level.