[openbeos] Re: Resampling algorithms for audio mixer

2007/11/28, François Revol <revol@xxxxxxx>:
> > >>> Rather than put workarounds for the denormal problem in
> > >>> the resampling algorithm (thus slowing it down), I propose
> > >>> completely disabling denormals on the CPU across the entire
> > >>> OS as standard.
> > >> This is up to Haiku devs :-)
> > >
> > > At least I have no clue about this, but we don't have to decide
> > > this today, don't we? :-)
> > > Do we even have a choice with being BeOS compatible and all
> > > (as I said, I have no idea what I'm talking about, just in case
> > > you're wondering)?
> >
> > Denormals are very small floating numbers.
> > Calculation under this "subnormal" level often switch FPU into a
> > slower cycle
> > state, which could hit badly the whole system, especially on P4's.
> > While in scientific computing field it's obviously well tolerated,
> > when it comes
> > to near zero floating point audio samples, hit doesn't worth it and a
> > plain
> > silence is better than an near zero silent but slower system.
> >
> > CPUs can be set to consider denormals as zero, per SSE+ FTZ & DAZ
> > flags
> >
>
> Hmm what about fixed point ?
> Could be vectorized with mmx..

Weak choice for audio IMHO. For two reasons:
#1. Integer and fixed point DSP is more difficult to do than floating
point (it is easier to implement workarounds for denormals, in most
cases just adding a 1.0e-10 value is enough).
#2. Most people use floating point for audio DSP.

But "I have no voice on the matter", so you decide ;-)

Stefano

Other related posts: