While parallel programming on distributed systems is difficult, making applications scale across multiple machines – or hybrid compute elements that mix CPUs with FPGAs, GPUs, DSPs, or other motors – ...