Is using memcpy real time safe?

dealerpriest

Hello!
I was watching this very interesting talk:

The talk descbribes some things to look out for when doing audio dsp. Things that could risk having the audio thread have arbitrarily long processing time (potentially resulting in audio drop-outs).
He mentions that using new and delete should be avoided as they internally use locks (as I understood it...), and hence can introduce unexpected wait in the running code.
But what about memcpy? Can I safely use memcpy inside the render function? I want to prepare a float array for a processing function by copying.

Also, at the moment I use new and deletein some of my effect classes' constructors and destructors. This is to instantiate ringbuffers where I don't know the size beforehand. I'm only calling the constructors outside render. Is that ok, or is it better to always define a big enough array size in the declaration? If so, why?

Any thoughts and know how would be greatly appreciated!

adanlbenito

Hi dealerpriest ,!

AndyCap Is right.

It is safe to use memcpy inside the render() function as it only copies data from a source to a block of memory referenced by a pointer. Sometimes using memcpy could be even more efficient than manually copying arrays, as the standard library is often optimised.

Regarding the instantiation of classes using new/delete (do not forget to delete the object, as the resources are not automatically freed when your pointer goes out of scope) outside render() is also good practice. Allocating memory inside your audio callback is not real time safe and the operator new(http://www.cplusplus.com/reference/new/operator%20new/
) allocates a block of memory for your object so you should use it before (either in the global scope or in the setup() function).

Timur’s talk is quite nice, another great resource on do’s and dont’s re. real-time audio programming is this blog post by Ross Bencina, where he mentions the problem of using new/delete inside the audio callback:
http://www.rossbencina.com/code/real-time-audio-programming-101-time-waits-for-nothing

Hope this helps.
Adan

AndyCap

With Bela It's really only Render that you need to worry about for audio dropouts, code in constructors/destructors called outside of render is no problem at all, just don't create/destroy things in render.

Memcpy has good performance so no problem to call from render, why do you want to do this though rather than passing a pointer to the existing data?

dealerpriest

Thx guys! That was helpful.
@AndyCap
The reason for me to do a memcpy was for a windowed sinc interpolation where I want to apply a crossfade of two signals (actually same signal from same buffer but at different indices) prior to the interpolation. since the sinc interpolation operates on several samples I will need to add the two signals into a small buffer/array before I give it to the interpolation algorithm. If that makes sense... I'd like to do it for performance reasons in order to avoid running the interpolation twice.
The funny thing now is that I tried to replace it with linear interpolation, and I couldn't hear a difference. Will see if I'll end up using sinc interpolation after all. Depends on how much processing time I can spare in the end.