Am 19.02.22 um 16:07 schrieb Will Godfrey:
It looks like there is something screwy about my computer :(
I managed to get the old AMD running. This is a dual core Athlon - so you
can see how old it is! It had debian 'squeeze' on it which is too old for the
C++ stuff we're using now, so I upgraded it as far as 'buster'. The latest
version of padthread runs perfectly smoothly - not an Xrun in sight - and
that's on a 2.6G dual core whereas the Intel is 3.1G.
Am 22.02.22 um 09:39 schrieb Will Godfrey:> Well, first of all. Apologies for
this very major distraction!
On a personal note, I had an accident about a week ago and put my back out,
so at the moment I can't sit at the computer for long periods - most
frustrating :(
Meanwhile, with the system effectively proved for quite old machines as well
as newer ones, is it possible to look at the possibility of a smooth morph
from old to new wavetable. If so, I would suggest (separate to the autoapply
flag) an integer with zero representing off, then values representing (say)
100s of milliseconds.
Hi Will,
sad to hear about your accident, let's just hope you'll get well soon.
Regarding that Xrun investigation, I don't see that as distraction, as it helped
us to understand better what's going on, so no need to apologize. This
concurrency stuff can be tricky, and it's easy to make subtle mistakes,
so the more testing, the better.
Another thing we might consider is to provide for an option to run the wavetable
building not at all as parallel task, but rather (as it was before) blocking
in the event handling thread. So the setting might be a select box with three
choices: "blocking", "background", "auto-apply". This way, we'd retain an option
for those users with very limited hardware, while auto-apply obviously consumes
the most resources. Do you consider that might be an idea worth exploring?
Regarding the crossfade, based on my own experience, I'd expect good settings
to be somewhere between 50ms and 500ms, since that is some transitory region
where the human perception switches gears. Transients below 20ms are perceived
as colours, while changes above 0.2s start to be preceptible as a change over
time, so in-between is a good place to hide crossfades. Since that is not a
musical setting, I doubt we need a finer granularity than 1ms. Incidentally,
did you intend to make that a global setting, or a setting per part/instrument?
-- Hermann
PS: while this week was rather busy for me at the office, this evening
there was an surprise opportunity to do an organ recording, together with
a friend who works as composer, and wrote this score for a film sound track.
_______________________________________________
Yoshimi-devel mailing list
Yoshimi-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/yoshimi-devel