parallel computing for real-time simulation in Xpc target computer

I want to acceleration the speed of real-time simulating in Xpc target computer. The main code is written in S-function and embeded in Simulink model. Whether any parallel computing method (parallel computing box, GPU acceleration or MPI) can be used in Xpc target computer? Thanks!

Answer 1

[Suneesh]

Parallel computing box, GPU acceleration etc., is not applicable to xPC. The approach you would have to take is by utilizing the multiple rates in your modelling if possible. For example, if the base rate of your model is T1 seconds then you could inspect if all the calculations have to be done at the rate. If there is possibility of performing some parts of the calculation "not as often" then you may set the rate of this part of the model to T2 seconds where T1 < T2. You may also partition using model referencing. See:

http://www.mathworks.com/help/simulink/ug/introduction_bs13l5v-1.html

You could also configure your model for execution on multi-core targets. This is a good starting point:

http://www.mathworks.com/help/xpc/examples/concurrent-execution-on-xpc-target.html

[Zhu]

Thanks for your reply. So what's the reason that why those ways (Parallel computing box, GPU acceleration etc.) are not applicable to xPC?

[Suneesh]

Parallel Computing toolbox is intended to spawn multiple threads on a (mostly) non realtime HOST platform running something like Windows 7, Debian Linux etc. Simulink Concurrent execution which I referred to provide better ways of doing the similar thing but for a variety of realtime targets including xPC. So it is just a different workflow (plus you don't need to buy an additional license).

Also, if you have a multirate model that you run on a multicore xPC target then the software automatically takes advantage of the cores by automatically creating tasks for each of the rates.