Alpha-09 proj 20something of 56: Delay
Posted: Mon Apr 03, 2017 6:45 pm
I've spent far more time than I expected cleaning up the Delay module. I've actually made it 1.6x faster when you have 50 taps and think I can make that yet much better.
However I got bogged down with my automated test system. I have a "good example" of output from each of 700+ patches, and every day or two I run an automated test which makes one second of output with each patch and verifies the output is still good.
Well, the test system was saying EVERYTHING using Delay was bad... but I went back to the old version of Delay and it seems to make the same results, so I can't figure out what happened. If anything, the "good examples" seem to have lower delay output. After wasting two days trying to figure out why the examples don't match the current output, I just analyzed the current output sample by sample and decided it's good and the "good examples" are bad for some reason.
So anyway, tomorrow I'll increase performance far more for certain kinds of delays: ones where the tap delay in milliseconds is fixed, instead of variable. And ones where panning and gain are also fixed will get even more of a speed-up.
However I got bogged down with my automated test system. I have a "good example" of output from each of 700+ patches, and every day or two I run an automated test which makes one second of output with each patch and verifies the output is still good.
Well, the test system was saying EVERYTHING using Delay was bad... but I went back to the old version of Delay and it seems to make the same results, so I can't figure out what happened. If anything, the "good examples" seem to have lower delay output. After wasting two days trying to figure out why the examples don't match the current output, I just analyzed the current output sample by sample and decided it's good and the "good examples" are bad for some reason.
So anyway, tomorrow I'll increase performance far more for certain kinds of delays: ones where the tap delay in milliseconds is fixed, instead of variable. And ones where panning and gain are also fixed will get even more of a speed-up.