Interleaving

jeremybernstein

Hi Bela,

The 'interleave' flag does not appear to do what it claims. I'm using the Bela API to play some existing audio code using the Bela hardware. Although I'm specifying BelaInitSettings.interleave = 0 (and verifying that the BELA_FLAG_INTERLEAVED is disabled in render(), the buffers are always interleaved. A peek in Utilities.h shows that the utility functions also assume interleaved buffers:

static inline float audioRead(BelaContext *context, int frame, int channel) { return context->audioIn[frame * context->audioInChannels + channel]; }

I guess that's fine, but it would be nice if the documentation were updated about this (better yet, it would be great if the functionality could be implemented).

Thanks, Jeremy

giuliomoro

Good point.
Added runtime error and warning in the Doxygen that this is not implemented.

If you need de-interleaved buffers, you can draw inspiration from core/libpd_default_render.cpp where the de-interleaving takes place.

More details here
https://github.com/BelaPlatform/Bela/issues/38
Note: feel free to add an issue on the repo for obvious software bugs

The idea is that once it is implemented, the utilities should transparently handle interleaved and de-interleaved buffers. Not sure, though, how to do this without a) the performance penalty of having an if() in a function that could potentially be called multiple times per sample and, b) compile-time flags, which make it difficult for make to check if a file is up-to-date

thetechnobear

giuliomoro Not sure, though, how to do this without a) the performance penalty of having an if() in a function that could potentially be called multiple times per sample and, b) compile-time flags, which make it difficult for make to check if a file is up-to-date

if you can structure the code, such that it is a constant parameter, and the compiler can determine that it can never be changed after initialisation... then the optimiser should remove the if clause.
(the 'tricky' bit, is coming up with an API that make this possible 😉 )

sletz

Has the Interleaving model ever be implemented ? What does the audio low-level layer? In the Faust architecture file, we need to deinterleave at input and interleave at output which seems quite stupid is this could be done only once in the underlying level.

giuliomoro

The de-interleaving mode has been implemented. Just set:

	settings->uniformSampleRate = 1;
	settings->interleave = 0;
	settings->analogOutputsPersist = 0;

where, in addition, uniformSampleRate resamples the analog channels in the backend to match the audio sample rate and analogOutputPersist saves some memory operations when the analog outputs are written continuously.

You can do the above either before calling Bela_initAudio(), example here, or - if that's inconvenient - you can implement a function Bela_userSettings(), example here.

Once the interleave has been applied correctly, you will see that the BELA_FLAG_INTERLEAVED bit will be cleared in context->flags : assert((context->flags & BELA_FLAG_INTERLEAVED) == 0);, once the uniformSampleRate has been applied correctly, you will see that context->audioSampleRate == context->analogSampleRate.

Also have a look here.