[alsa-devel] [PATCH v2] staging: bcm2835-audio: interpolate audio delay

[Takashi Iwai]

On Thu, 25 Oct 2018 00:02:34 +0200,
Kirill Marinushkin wrote:
> 
> >> When you play sound - the pointer increments.
> > 
> > Unfortunately, when you play sound, the pointer does not actually increment, for up to about 10 milliseconds. I know of no way to actually access the true “live” position of the frame that is being played at any instant; hence the desire to estimate it.
> > 
> 
> Your vision of situation in the opposite from my vision. What you see as a
> symptom - I see as a root cause. As I see, you should fix the
> pointer-not-incrementing. Why do you think that it's okay that the pointer is
> not updating during sound play? Why do you insist that there is a delay? I don't
> understand why we are so stuck here.

Well, in the API POV, it's nothing wrong to keep hwptr sticking while
updating only delay value.  It implies that the hardware chip doesn't
provide the hwptr update.

Though, usually the delay value is provided also from the hardware,
e.g. reading the link position or such.  It's a typical case like
USB-audio, where the hwptr gets updated and the delay indicates the
actual position *behind* hwptr.  That works because hwptr shows the
position in the ring buffer at which you can access the data.  And it
doesn't mean that hwptr is the actually playing position, but it can
be ahead of the current position, when many samples are queued on
FIFO.  The delay is provided to correct the position back to the
actual point.

But, this also doesn't mean that the delay shouldn't be used for the
purpose like this patchset, either.  OTOH, providing a finer hwptr
value would be likely more apps-friendly; there must be many programs
that don't evaluate the delay value properly.

So, I suppose that hwptr update might be a better option if the code
won't become too complex.  Let's see.


One another thing I'd like to point out is that the value given in the
patch is nothing but an estimated position, optimistically calculated
via the system timer.  Mike and I had already discussion in another
thread, and another possible option would be to provide the proper
timestamp-vs-hwptr pair, instead of updating the timestamp always at
the status read.

Maybe it's worth to have a module option to suppress this optimistic
hwptr update behavior, in case something went wrong with clocks?


thanks,

Takashi