Can the epiphany cores do atomic operations on main memory. I recall using these on the CELL processor to help with SPU's collaborating (e.g. a SPU's could update an atomic counter for the next job index .. be that input or output. think of operations where the number of outputs is variable, and you want to accumulate a collection)
If it can't - what are the workarounds.
example use cases: parallellized for_each (several cores consult a single atomic counter for the 'next' index to process); and accumulating collision results (e.g. comparing a list of objects against itself, emitting a list of contact points which is variable - use an atomic counter for the 'output index')