Re: [Audacity-devel] Fwd: [Hydrogen-devel] SVN changes: exponential envelopes

[<xip...@xi...>]

On Fri, Nov 18, 2005 at 03:36:52PM +0300, Alexandre Prokoudine wrote:
> Hi,
> 
> This "using precomputed tables and lookup/interpolation functions - so we get
> frame-precise smoothness with very little CPU drain" sounds very close
> to what we want for crossfades.

A crossfade, even done via the simplest, slowest code possible won't
sink any detectable amount of a modern CPU.  In fact, I'm going to
posit a sufficiently large lookup table may be slower as it will add
to L1/L2 cache pressure.  

I also recommend against interpolation.  There's simply no reason.
It's an unneccessary optimisation that will only add noise.  I don't
want 60dB of signal depth in Audacity, I want 140dB of signal depth.

Speed and performance are indeed valid considerations, but not in this
case.  If the realtime crossfade is being that slow, something else is
wrong.  My own theater software does realtime crossfades on hundreds
of layers at a time on a G3-300 entirely on demand and the processor
load is barely detectable.  Like Audacity, it tends to be disk bound
not CPU bound.

Also, the fancier the code, the more likely there's something minor
unexpected wrong with it.  Audacity is riddled with minor DSP bugs as
it is, let's not blithely add to the pile.

Monty