If the filter inputs are all dropped down towards zero the engine tends towards 50% or higher CPU load. Probable cause is denormals on the low signal level so some form of prevention needs to be applied.
The problem, reported by Colin Fletcher with a good description of the reproduction, is easy to demonstrate. An initial fix has been sent out where -84dB of noise is injected into the filter whenever the net input signal levels drops below -84dB. The S/N ratio at full mix to no mix is then actually better than -84dB since multiple inputs can be mixed, it resolves the denormals and additionally the noise ensures that the filter will 'self' oscillate, not previously possible - it always required an input signal to oscillate. OK, now it always has one so it is not really self oscillation but the net effect is.....
This should be closed and tracked as a duplicate of bug id 2665269 which requests denormal be addressed for all the filters rather than just for the ARP 2600.
Log in to post a comment.