Nbc rebase

[mdavezac]

see #591

[mdavezac]

This is just to see if I can use anything for the RBC branch

[rupertnash]

retest this please

[mdavezac]

Should we merge this then?

[rupertnash]

No, because I need to implement the switch from wait to test outlined in #590. The version as-is hurts single core and scaling performance.

[rupertnash]

This PR will close #647

[rupertnash]

@mdavezac I'd appreciate it if you could have a glance at the problem with MPI+cmake on Legion+GNU. It is certainly possible there's an issue in the mpi.cmake file that checks there is MPI 3.0 available. I'm just going to check this on ARCHER

[rupertnash]

For some reason this was stalled in Jenkins on the simulation shut down. Works fine on my MacBook, so retest this please to find out if this is just a one off.

[rupertnash]

Pleased to say that unit and regression tests are passing on ARCHER. Gonna test scaling now.

[rupertnash]

retest this please

[mdavezac]

@rupertnash, the hand-rolled MPI version thingie breaks on legion with openmpi 1.8.4 (and gcc). It corresponds to version 3.0. Are you sure you need to roll out your own FindMPI + extra script?

[mdavezac]

I'm not too keen on investigating why hand-rolled stuff breaks :(
Unless it's my own hand-rolled crap :)