Zync to Epiphany: Use of COPRTHR SDK

[gordon]

[nickoppen]

How about you start here: . It's a simple vector adder.

Then, decide on what code you want to run on the epiphany and what goes on the host. Then, how you are going to split up your work (i.e. what core gets what part of the job), and that will give you the data structure that you'll need to communicate between the host and epiphany.

Liberal use of host_printf() when things don't work plus grim determination will get you home.

[jar]

Also, Epiphany-III does not support native double precision operations, so it would be prudent to use single precision (float). Parallelization of this code appears that it will be non-trivial. But if you intend to run this code independent of other cores (batch), you may not need to parallelize. The matrix and vector sizes are unusually small. If you want to parallelize it, matrices and vectors must be distributed evenly among all cores.

[gordon]

[nickoppen]

That's what the job is for the program designer. The simple answer is that you analyses your code, figure out the routines that do all the work and those are the one that you write for the Epiphany. You really need a fundamental understanding of the code and how the epiphany works. For this type of programming there is no other way.

If you have the time, I found this lecture series from MIT very helpful . It's from 2007 and the platform is the Playstation II but covers all of the fundamentals that you need to know in the first couple of sessions. Then have a read of my blog post which is specific for the Parallella. I used the OpenCL library for the examples but COPRTHR has equivalent calls and the same structure.

nick

[gordon]