Hi Kassen! I feel your pain! We'll definitely get on it. There are actually macro's throughout to combat floating point denormalization (for example, measures are taken in the general UGen architecture and also strategically added to all of STK in ChucK where there is feedback that could lead to denormalization), but perhaps they have not yet made their way into LiSa. Do you have a minimal example of the feedback loop that leads to floating-point denormalization? ASIO is long due and I am sorry it's not part of the canonical ChucK release yet. Spencer has actually been be gearing up for massive ChucK/miniAudicle development here at CCRMA. We will priority boost ASIO support! Rock on, Ge! On Wed, 3 Nov 2010, Kassen wrote:
Fellow ChucKists, I'm rather displeased right now, which is a bit of a euphemism for "I feel like breaking stuff".
I have a issue in my sequencer that I'm sure relates to floating-point denormalisation (involving a feedback loop on a LiSa) as I have CPU spikes that occur after a while and stay present after re-starting ChucK. These are quite hard to pinpoint as they take a while to surface. Even if I would find the culprit I'd then have to see whether it still exists in the current version of ChucK because I'm running into this on the latest ASIO version that we have around, which is downright out of date.
All of this is very, very frustrating, frustrating to the tune of looking for what's wrong for three days on end. It's probably fixable but it's frustrating that there is no canonical ASIO version of ChucK and that there hasn't been more attention to floating-point denormalisation. One of the main advantages over SC is that we can do sample-accurate feedback in CK, yet we throw much of that away if we risk this. There is also no Windows version out of the box that can use multi-channel audio and have acceptable latency.
Both seem quite serious to me. Could something be done about this? I've been thinking about how hard it would be to port my setup to a hardware PIC or even SC. None of this should be necessary at all; a out-of-the-box ASIO version is quite reasonable and I believe that from the time denormalisation has been a issue there have been instructions to simply round to 0.
Yours, Kas.