Bela Gem multi-core processing, parallel and/or series documentation

resynth

I'm trying to implement some heavy multi-step audio processing on Bela Gem Stereo so am keen to ensure I'm utilising all cores. My current project would benefit greatly from series processing but I also have plans that would require parallel processing. The multiple cores are what I was actually really excited about with Bela Gem but I'm having problems knowing where/how to start!

I've tried search terms like render_pre(), render_post(), busses, RtLock, RtMsgFifo and RtThread in docs.bela.io no results are returned.

When I search the Bela Core code (as pulled from https://github.com/BelaPlatform/Bela) no results are returned there either.

It occurs to me that maybe the core code repo and docs.bela.io are not relevant to the Bela Gem?

Is anyone using this functionality? I'd be grateful for any hints of how to get started, where to find low level documentation or, failing docs, more examples that cover the various possibilities?

Also it would be great to know what is considered the definitive source for Bela Gem C++ documentation, class and function descriptions etc. That would be very helpful but for all C++ Gem development.

Thanks :^)

giuliomoro

Have a look here https://forum.bela.io/d/7550-gem-parallel-processing-syntax

I don't have more details at the moment unfortunately, let me know if you have specific questions.The most up to date doxygen documentation is on your board: go to the documentation tab and click launch full documentation.

resynth

Thanks Giulio, I'd seen that and have been playing with it but it has no info RE series processing. Interesting for parallel though and I have that working from your example there.

I wonder if there is any way to see the % of each core?

Do render_pre and render_post both strictly use core 0?

giuliomoro

Cores are assigned from 3 to 0, because 0 is more generally busy with linux irq servicing. The pre and post functions run on core 3. Look at the per-thread cpu usage running watch -n 1 evl ps -st in the console. See if any one thread is particularly high.

giuliomoro

Series processing is done by using an array to pass data from one thread to the next.
E.g.: thread 0 reads its input from the context (or it generates it algorithimically) and writes its output to array0. Thread 1 reads its output from array0 and writes to array1. Thread 2 reads its input from array1 and writes its output to array2. Thread 3 reads its input from array2 and writes its output to the context. You wouldn't normally need a render_pre or render_post in this case. Each additional series thread adds one blocksize of latency.

resynth

Really helpful Giulio, cheers!

giuliomoro

Each additional series thread adds one blocksize of latency.