Subject: Re: [alsa-devel] Re: [linux-audio-dev] laaga, round 2
From: Jörn Nettingsmeier (nettings_AT_folkwang-hochschule.de)
Date: Tue May 15 2001 - 20:49:53 EEST
Abramo Bagnara wrote:
>
> Paul Davis wrote:
> >
> > I wrote:
> >
> > >> context switch costs that we've been discussing, the cost of zeroing
> > >> memory goes down more or less linearly with processor speed (helped
> > >> along by RAM speed, which is increasing much more slowly). Because of
> > >> this, we should consider the cost to zero a buffer to be an
> > >> essentially vanishing item in the overall operation of an audio
> > >> system.
> >
> > Abramo replied:
> >
> > >I think you should have benchmarked this statement before to write it
> > >;-)
> >
> > I have. Re-read what I wrote. I know that there are real costs of
> > zeroing the buffer and using "+=". But that cost is almost linear with
> > CPU speed. You speak often of planning for the future. The cost right
> > now is quite manageable, and will only get smaller. The benefit of
> > adopting this model is immense in terms of simplicitly. To design a
> > system in which we create complexity to solve a *vanishing* problem
> > seems foolhardy to me.
> >
> > Contrast this to the costs of IPC, which do not scale linearly with
> > CPU speed, and will never really vanish until we are all using 64 bit
> > operating systems that use a single address space (i.e. the TLB is
> > never flushed). Such operating systems do exist, but only in CS and
> > R&D departments.
>
> I don't want you to think I disagree: I've only commented your statement
> with some actual numbers to know the cost on low end machine... and I've
> to say that I've still some doubts:
>
> If plugin is used outside a realtime constraint performance impact might
> become noticeable (suppose we want to handle 1E6 samples per second).
>
> what about normalization?
> who does it?
just for clarification, aren't you using floats ?
or has the design been generalized for all data types ???
if floats only, the only place i can see where you'd need to
normalize is before the output dacs or when writing a file.
jörn (trying *very* had to follow the discussion)
-- Jörn Nettingsmeier home://Kurfürstenstr.49.45138.Essen.Germany phone://+49.201.491621 http://icem-www.folkwang-hochschule.de/~nettings/ http://www.linuxdj.com/audio/lad/
This archive was generated by hypermail 2b28 : Tue May 15 2001 - 21:07:31 EEST