[LAD] Denormals / subnormals (again)

Paul Davis paul at linuxaudiosystems.com
Tue Jan 3 02:30:56 UTC 2012

On Mon, Jan 2, 2012 at 9:17 PM, Fons Adriaensen <fons at linuxaudio.org> wrote:

> I don't agree with blaming the Intel architecture. Denormals are
> just one tiny aspect of a much wider issue which is numerical
> precision and stability, and that affects all architectures.
> Anyone writing numerical or DSP code should be aware of this
> and be able to analyse the algorithm. You don't produce good
> numerical or DSP code by blindly implementing textbook equations.


but there are two aspects to this issue:

    (1) good numerical or DSP code
    (2) not slowing the processor down by a factor of 10 or more

non-Intel architectures took care of (2), while leaving (1) for the
programmer to tackle.

anyway, i'm not really arguing with you. from the initial tests with
carl's program, it appears that the advice for plugin authors could be
"If you or anyone else is going to distribute your plugin WITHOUT SSE
math, you should add offsets to take care of denormals. otherwise, you
can assume that the host will take care of it".  and meanwhile, for
host authors "enable DAZ if possible, which more or less eliminates
any cost from denormals, and if not possible enable FTZ".

if everyone does this, the reports we get repeatedly on #ardour about
plugin-related performance problems will become a thing of the past.
starting very soon, ardour will use DAZ and FTZ by default.


More information about the Linux-audio-dev mailing list