Hi,
Is there a way to disable the underrun detect: blocks dropped messages?
Also, does it make sense that I would be getting a constant stream of them? I'm using ~30% CPU, but that doesn't seem like it should be too taxing. I've tried everything I can think of to get that % down, but I'm still getting the errors. Maybe it's from the [netsend -u] usage...? Aside from that, I can't see what would be causing the drops.
The reason why you are getting the message is that the CPU load is not constant across different audio callbacks: [env~] has a default analysis size of 1024. Depending on how it is implemented, this may (and probably does) mean that when running with blocksizes smaller than 1024, most of the time the object is doing nothing, simply collecting samples and storing them in a buffer. Then, when the buffer is full, it is analyzed all in one go. Now, the average CPU load (which is the one you see in the IDE) is low, because most of the time the object is doing nothing. But when it does the analysis, then it overruns the time available in the audio callback, thus causing the dropout message you see (and also causing a glitch in the audio output!). You can disable the warning by commenting out the relevant lines in core/PRU.cpp, however that is not going to fix your audio glitch!
I think you cold re-implement [env~] using a Pd abstraction that makes the CPU load constant (e.g.: do the processing as the samples come in, instead of waiting for the whole buffer to be filled), but I'd need to have a look at its C code to see what it does exactly.
Right on. I changed the [env~] objects back to the default and increased the project block size to 128. This seems to have solved it.
Thank you yet again!
this will have just hidden it: you will still have spikes in the CPU load, so that when the average CPU usage is small but the peak usage may cause dropouts. If you do require a high CPU load (e.g.: because of other things going on in the patch), you may have to follow my advice above.
Did anyone re-implemented env~, fiddle~ or sigmund~ objects to avoid blocks dropped?
Or any advice or examples of pitch or envelop following with PD on bela?
Floris Did anyone re-implemented env~, fiddle~ or sigmund~ objects to avoid blocks dropped?
I attempted a couple of times with [sigmund~], had it partially working and gave up. It's so horrible to write Pd source C code ... anyhow the current solution is to increase the block size on Bela to a multiple of the step size.
will try the block size trick then!
