[Stk] fast time domain filter implementation
Stephen Sinclair
sinclair at music.mcgill.ca
Tue Apr 15 09:40:19 PDT 2008
Hi Gary!
Gary Scavone wrote:
> Hi Stephen,
>
> The vector [] operator does not do bounds checking and it _should_ be
> defined inline by the class.
It's good to know that it doesn't do bounds checking, I wasn't sure. I
would think the STL programmers were good enough to ensure that if
optimized correctly the Vector class would compile to the equivalent of
a normal array. But I wouldn't put it past the VS programmers to not
properly handle inline in Debug mode. I can say that in my experience,
the presence of a /O flag was the difference between my algorithm
running properly and not running at all. But this was only true when I
compiled it with Visual Studio, it ran fine on Linux.
The STK piano does use quite a few filters, and it uses the subclasses
rather than Filter directly, as you mention.
> I think the Windows stutter problem is more related to DirectSound ...
> it's a terrible API. Always use ASIO drivers on Windoze if possible.
Yeah, I definitely agree here. I've especially noticed that for example
with Pd I have to set the buffer size to like 100 ms to avoid stuttering
without ASIO.
>
> Note that there are currently inefficiencies with virtualized
> functions in many of the classes, though that shouldn't be the case
> when using the Filter class. It would be an issue with subclasses of
> Filter. This is something I plan to address this Spring/Summer.
Hm, that's interesting, I hadn't thought of that in relation to the
Filter subclasses. Do virtual functions really insert a non-negligible
amount of delay? I guess if you put several thousand of them together
under real-time constraints then every look-up operation counts.
Steve
More information about the Stk
mailing list