[LAD] JACK on Gigaport HD+ at 4800kHz (reduced channel count)

[Fons Adriaensen]

On Mon, Feb 24, 2014 at 07:53:58PM -0500, Paul Davis wrote:

> On Mon, Feb 24, 2014 at 6:19 PM, Fons Adriaensen <fons at linuxaudio.org>wrote:
> 
> >
> > The problem I guess is due to the way the HW related params
> > are set:
> >
> > 1.  Allocate a hw_params struct
> > 2.  Fill in the params using set_*** functions
> > 3.  Submit the hw_params struct.
> >
v> > In step 2, the set_*** functions are aware of the acceptable
> > ranges/values and will report error if anything is not possible.
> > The backend code sets the sample rate before anything related
> > to channel counts is set or tested. So if the ALSA code still
> > reports that 8 channels are available after the sample rate
> > is set to 48 kHz, that looks to me like a bug in the driver.
> >
> 
> No, this is based on a misconception.
> 
> The legal ALSA driver parameters form a sparse multi-dimensional space. The
> driver knows the legal intervals along each dimensional axis, but it does
> not know the shape of the intersecting space.

Depends on the driver. For the hdspm driver for example, once the sample rate
is set to 96 kHz in the hw_params (and thus before any real HW configuration
is done), both snd_pcm_hw_params_get_channels_max() and ..._min() return 32
instead of 64. This depends *only* on the contents of the partially filled-in
hw_params struct, not on the actual rate used at the time. And it is the only
way user code can find out the value to use (apart from trying a zillion
combinations). 

> Thus the only way to find out if a given combination of parameters is legal
> is to try to set them and see if the hardware accepts them.

That works only if the driver _lets you set the correct values_ in the first
place. If for example snd_hdspm would not work as described above, it would
be impossible to set the channel count to the correct value. Both min and max
would be 64, and snd_pcm_hw_params_set_channels() would refuse to accept the
correct value, 32.
 
> I only know the RME Hammerfall driver well, but in that case, where the
> channel count changes depending on whether you do <88.2kHz or >= 88.2kHz,
> there is no way that the driver can tell you that your attempt to use too
> many channels is wrong until you submit the entire h/w params struct.

Wrong, at least for snd_hdspm (and I did my homework before posting this
message and the previous one).

Ciao,

-- 
FA

A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)