@ubii : Nice work, You've got it. My suggestion is : write an 'own' benchmark, or recompile an exists one.
@ysapir : You said : "...the host-device communication channel was horribly slow." Oops. ... But, luckily we have a new challenge. As usually at 'IO devices' --- oh No : Epiphany just as an IO device. YET. --- So, DMA and IRQ for her. (Is he or she, I think, she. We (males) are able to focus/do only one thing in one time period --- no parallelism ---.) A message passing method with polling technique is a good and safe starting point. [I|We] need some more efficient stuff. Share with us the details. I am sure, there is someone, who deliver a very nice solution, and an another one, and another one... Be open.
ed2k : Re to Your 'BTW note' : YES. This is it, this is the main point. Parallella Board : Everybody is on the board. I mean : here are these: processing system (CPU), programming logic (FPGA), supercomputing (CPU mesh). What premier league teams : ARM, Xilinx, Adapteva. Well, for Your keywords: - repo: under which umbrella? (where) ; - Core Team: agree. Let them work; - Community: Sorry, I do not understand You. What does it mean? : "Let the community take off your burden.".
Have a nice day! Gabor