NUMA emulation/conventional machines

[dobkeratops]

When working on the PS3 years ago, something I did to assist with porting was wrote a series of #defines / smart pointers that allowed a single codepath to be compiled & run on both a CPU and an SPU. Of course this wasn't an efficient way of using the SPU, but programming time was limited, and huge amounts of code overloaded the PS3 CPU, we needed a way of offloading *something* within tight programmer-time constraints.

Has anyone done anything similar for running similar code thats 'good enough' on conventional SMP and the Parallela accelerator?

I've even read that with an MMU you could actually emulate a NUMA setup on a cluster of SMP machines across a network (although it would have very different characteristics).. probably another question altogether .. is that another route toward getting code that could run well/acceptably on either architecture
googling around.. I guess i'm talking about something like this.
https://www.usenix.org/legacy/event/use ... index.html
perhaps such a setup with networked SMP machines could run code designed for an epiphany system with small changes