Non blocking scatter and gather

[SStroteich]

Right now the redistribution function uses blocking send and receive commands. This makes the communication very inefficient and should be fixed by using the non blocking commands in redistribute.f90.

[d7919]

Yes, our experience in GS2 is that swapping for non-blocking communications (and ideally persistent versions of this) can give a good performance improvement in most cases. It's possible on some systems for some problems that posting all send/receive calls at once can cause hardware contention issues which degrade performance but such instances are rare. It may be relatively simple to port the changes from GS2/AGK here?