a11y: howto reduce latency with alsa's dmix plugin?

a11y: howto reduce latency with alsa's dmix plugin?

[Halim Sahin]

Hello,
Is there a way to reduce latency on audiocard if the dmix plugin is in
use?
I am asking because there is a strange problem with asistive
technologies which are outputing through alsa.
E. G. speech-dispatcher takes almost a half sec to stop the
speech-synthesizer.
If the user navigates quickly through menus, there are simultane outputs
at a time.
(tested with an usb logitech headset, snd-intel8x0 based card,
snd-hda-intel).
The same problem can be reproduced by running mplayer with alsa output
driver and pressing
quickly left/right arrow.
You should hear several samples at a time.

Using an sblive card which doesn't use dmix doesn't have such problems.

Using the dmix plugin should audio much faster to avoid this problem.

Please help!!!
Regards
Halim

really no idea? Re: a11y: howto reduce latency with alsa's dmix plugin?

[Halim Sahin]

On So, Aug 23, 2009 at 09:55:18 +0000, Halim Sahin wrote:
> Hello,
> Is there a way to reduce latency on audiocard if the dmix plugin is in
> use?
> I am asking because there is a strange problem with asistive
> technologies which are outputing through alsa.
> E. G. speech-dispatcher takes almost a half sec to stop the
> speech-synthesizer.
> If the user navigates quickly through menus, there are simultane outputs
> at a time.
> (tested with an usb logitech headset, snd-intel8x0 based card,
> snd-hda-intel).
> The same problem can be reproduced by running mplayer with alsa output
> driver and pressing
> quickly left/right arrow.
> You should hear several samples at a time.
> 
> Using an sblive card which doesn't use dmix doesn't have such problems.
> 
> Using the dmix plugin should audio much faster to avoid this problem.
> 
> Please help!!!
> Regards
> Halim

-- 
Halim Sahin
E-Mail:				
halim.sahin (at) t-online.de

Re: a11y: howto reduce latency with alsa's dmix plugin?

[Takashi Iwai]

At Sun, 23 Aug 2009 09:55:18 +0000 (UTC),
Halim Sahin wrote:
> 
> Hello,
> Is there a way to reduce latency on audiocard if the dmix plugin is in
> use?
> I am asking because there is a strange problem with asistive
> technologies which are outputing through alsa.
> E. G. speech-dispatcher takes almost a half sec to stop the
> speech-synthesizer.
> If the user navigates quickly through menus, there are simultane outputs
> at a time.
> (tested with an usb logitech headset, snd-intel8x0 based card,
> snd-hda-intel).
> The same problem can be reproduced by running mplayer with alsa output
> driver and pressing
> quickly left/right arrow.
> You should hear several samples at a time.
> 
> Using an sblive card which doesn't use dmix doesn't have such problems.
> 
> Using the dmix plugin should audio much faster to avoid this problem.

It's a bit hard to guess.  This could be the difference of PCM drain
implementation between hw and dmix, but a half second is still too
long.

Could you prepare a small test case to reproduce the buggy behavior?


thanks,

Takashi

Re: a11y: howto reduce latency with alsa's dmix plugin?

[Halim Sahin]

On Mo, Aug 31 2009, Takashi Iwai wrote:

> At Sun, 23 Aug 2009 09:55:18 +0000 (UTC),
> Halim Sahin wrote:
>> 
>> Hello,
>> Is there a way to reduce latency on audiocard if the dmix plugin is in
>> use?
>> I am asking because there is a strange problem with asistive
>> technologies which are outputing through alsa.
>> E. G. speech-dispatcher takes almost a half sec to stop the
>> speech-synthesizer.
>> If the user navigates quickly through menus, there are simultane outputs
>> at a time.
>> (tested with an usb logitech headset, snd-intel8x0 based card,
>> snd-hda-intel).
>> The same problem can be reproduced by running mplayer with alsa output
>> driver and pressing
>> quickly left/right arrow.
>> You should hear several samples at a time.
>> 
>> Using an sblive card which doesn't use dmix doesn't have such problems.
>> 
>> Using the dmix plugin should audio much faster to avoid this problem.
>
> It's a bit hard to guess.  This could be the difference of PCM drain
> implementation between hw and dmix, but a half second is still too
> long.
>
> Could you prepare a small test case to reproduce the buggy behavior?

Yes that's easy:
The following should be tested with active dmix.
1. login to a textconsole
2. mplayer -ao alsa your_favourite_file
3. Press space during playback.

Result: it takes about an half sec to stop the playback.

The second test would be:
1. login to a console
2. mplayer -ao alsa:device=hw=0,0 your_file
3. press space during playback
Result:
The playback stops without delay

Hope you can help.
Thanks for your response
Kind regards
Halim

>
> thanks,
>
> Takashi

-- 
Halim Sahin
E-Mail:				
halim.sahin (at) t-online.de

Re: a11y: howto reduce latency with alsa's dmix plugin?

[Takashi Iwai]

At Tue, 01 Sep 2009 12:32:27 +0200,
Halim Sahin wrote:
> 
> On Mo, Aug 31 2009, Takashi Iwai wrote:
> 
> > At Sun, 23 Aug 2009 09:55:18 +0000 (UTC),
> > Halim Sahin wrote:
> >> 
> >> Hello,
> >> Is there a way to reduce latency on audiocard if the dmix plugin is in
> >> use?
> >> I am asking because there is a strange problem with asistive
> >> technologies which are outputing through alsa.
> >> E. G. speech-dispatcher takes almost a half sec to stop the
> >> speech-synthesizer.
> >> If the user navigates quickly through menus, there are simultane outputs
> >> at a time.
> >> (tested with an usb logitech headset, snd-intel8x0 based card,
> >> snd-hda-intel).
> >> The same problem can be reproduced by running mplayer with alsa output
> >> driver and pressing
> >> quickly left/right arrow.
> >> You should hear several samples at a time.
> >> 
> >> Using an sblive card which doesn't use dmix doesn't have such problems.
> >> 
> >> Using the dmix plugin should audio much faster to avoid this problem.
> >
> > It's a bit hard to guess.  This could be the difference of PCM drain
> > implementation between hw and dmix, but a half second is still too
> > long.
> >
> > Could you prepare a small test case to reproduce the buggy behavior?
> 
> Yes that's easy:
> The following should be tested with active dmix.
> 1. login to a textconsole
> 2. mplayer -ao alsa your_favourite_file
> 3. Press space during playback.
> 
> Result: it takes about an half sec to stop the playback.

This isn't a test case to analyze.  If the app you mentioned is using
libao as the backend, then yes, it's the same issue.  If not, it just
appears same, but you don't know whether it's the same.

At best, we need a small C code that reproduces the behavior.

> The second test would be:
> 1. login to a console
> 2. mplayer -ao alsa:device=hw=0,0 your_file
> 3. press space during playback
> Result:
> The playback stops without delay

This could be also the difference whether the device supports
pause or not, too.


Takashi

Re: a11y: howto reduce latency with alsa's dmix plugin?

[Halim Sahin]

[-- Attachment #1: Type: text/plain, Size: 370 bytes --]

On Di, Sep 01 2009, Takashi Iwai wrote:

> At best, we need a small C code that reproduces the behavior.
Ok here is a small example.
It's a modified version of pcm_min.c.
Simply press ctrl+c to reproduce the problematic behaviour.
on my machine it takes about an half sek to stop with an usb head which
uses dmix.
Using an sblive card (without dmix) stopps immediately.

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: pcm_min.c --]
[-- Type: text/x-csrc, Size: 1703 bytes --]

/*
 *  This extra small demo sends a random samples to your speakers.
 */

#include <alsa/asoundlib.h>
#include <signal.h>
#include <unistd.h>
static char *device = "default";

snd_output_t *output = NULL;

unsigned char buffer[16 * 1024];        /* some random data */

snd_pcm_t *handle;

void suspend (int signum)
{
  int err;

  printf ("trying to stop\n");
  if ((err = snd_pcm_drop (handle)) < 0)
    printf ("cannot pause\n");

  sleep (1);
  snd_pcm_close (handle);

  exit (1);
}

int main (void)
{
  int err;

  unsigned int i;

  snd_pcm_sframes_t frames;

  signal (SIGINT, suspend);
  for (i = 0; i < sizeof (buffer); i++)
    buffer[i] = random () & 0xff;

  if ((err = snd_pcm_open (&handle, device, SND_PCM_STREAM_PLAYBACK, 0)) < 0)
   {
     printf ("Playback open error: %s\n", snd_strerror (err));
     exit (EXIT_FAILURE);
   }
  if ((err = snd_pcm_set_params (handle,
                                 SND_PCM_FORMAT_U8,
                                 SND_PCM_ACCESS_RW_INTERLEAVED,
                                 1, 48000, 1, 500000)) < 0)
   {                            /* 0.5sec */
     printf ("Playback open error: %s\n", snd_strerror (err));
     exit (EXIT_FAILURE);
   }

  for (i = 0; i < 16; i++)
   {
     frames = snd_pcm_writei (handle, buffer, sizeof (buffer));
     if (frames < 0)
       frames = snd_pcm_recover (handle, frames, 0);
     if (frames < 0)
      {
        printf ("snd_pcm_writei failed: %s\n", snd_strerror (err));
        break;
      }
     if (frames > 0 && frames < (long) sizeof (buffer))
       printf ("Short write (expected %li, wrote %li)\n",
               (long) sizeof (buffer), frames);
   }

  snd_pcm_close (handle);
  return 0;
}

[-- Attachment #3: Type: text/plain, Size: 38 bytes --]

Hope you can help.
Kind regards
Halim

[-- Attachment #4: Type: text/plain, Size: 160 bytes --]

_______________________________________________
Alsa-devel mailing list
Alsa-devel@alsa-project.org
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel

Re: a11y: howto reduce latency with alsa's dmix plugin?

[Takashi Iwai]

At Wed, 02 Sep 2009 10:56:17 +0200,
Halim Sahin wrote:
> 
> On Di, Sep 01 2009, Takashi Iwai wrote:
> 
> > At best, we need a small C code that reproduces the behavior.
> Ok here is a small example.
> It's a modified version of pcm_min.c.
> Simply press ctrl+c to reproduce the problematic behaviour.
> on my machine it takes about an half sek to stop with an usb head which
> uses dmix.

Hmm, I cannot reproduce the problem, at least, dmix + HD-audio.
After removing sleep(1), it quits immediately.

Doesn't this problem happen if you use usb-audio with "hw" PCM?

It's possible that the behavior of snd_pcm_drop() of usb-audio is
different from others because of URB handling.


Takashi

Re: a11y: howto reduce latency with alsa's dmix plugin?

[Halim Sahin]

On Fr, Sep 04 2009, Takashi Iwai wrote:

> At Wed, 02 Sep 2009 10:56:17 +0200,
> Halim Sahin wrote:
>> 
>> On Di, Sep 01 2009, Takashi Iwai wrote:
>> 
>> > At best, we need a small C code that reproduces the behavior.
>> Ok here is a small example.
>> It's a modified version of pcm_min.c.
>> Simply press ctrl+c to reproduce the problematic behaviour.
>> on my machine it takes about an half sek to stop with an usb head which
>> uses dmix.
>
> Hmm, I cannot reproduce the problem, at least, dmix + HD-audio.
> After removing sleep(1), it quits immediately.

Yes you can't, because the program doesn't run.
The sleep shouldn't be removed to see the problem.
When speech-dispatcher is running and someone tries to stop the output,
the programm doesn't exit.
Think about it like a pause key :-).

The sblive stopps emmidiately with inserted sleep the usb headset
doesn't.


> Doesn't this problem happen if you use usb-audio with "hw" PCM?

The pcm_min example of alsalib doesn't run this way.
--8<---------------cut here---------------start------------->8---

ALSA lib pcm.c:7125:(snd_pcm_set_params) Sample format not available for
PLAYBAC
K: Invalid argument
Playback open error: Invalid argument
--8<---------------cut here---------------end--------------->8---

Can you tell me which sampleformat works with this device???

> It's possible that the behavior of snd_pcm_drop() of usb-audio is
> different from others because of URB handling.

The stop problem also happends on my laptop with hd audio and ad1981
chip.

with kind regards
Halim

Re: a11y: howto reduce latency with alsa's dmix plugin?

[Takashi Iwai]

At Fri, 04 Sep 2009 11:09:17 +0200,
Halim Sahin wrote:
> 
> On Fr, Sep 04 2009, Takashi Iwai wrote:
> 
> > At Wed, 02 Sep 2009 10:56:17 +0200,
> > Halim Sahin wrote:
> >> 
> >> On Di, Sep 01 2009, Takashi Iwai wrote:
> >> 
> >> > At best, we need a small C code that reproduces the behavior.
> >> Ok here is a small example.
> >> It's a modified version of pcm_min.c.
> >> Simply press ctrl+c to reproduce the problematic behaviour.
> >> on my machine it takes about an half sek to stop with an usb head which
> >> uses dmix.
> >
> > Hmm, I cannot reproduce the problem, at least, dmix + HD-audio.
> > After removing sleep(1), it quits immediately.
> 
> Yes you can't, because the program doesn't run.
> The sleep shouldn't be removed to see the problem.

I don't understand here.
So, how is the problem exactly?  In which point is the delay expected
and where not expected?

You catch a signal and add a delay there -- so the program works, no?
If you are checking whether snd_pcm_drop() is executed immediately or
not, you should put a printf() after snd_pcm_drop() to indicate when
the function call finishes.

> When speech-dispatcher is running and someone tries to stop the output,
> the programm doesn't exit.

So, "doesn't exit" is the intended behavior?

> Think about it like a pause key :-).
> 
> The sblive stopps emmidiately with inserted sleep the usb headset
> doesn't.

In that sense, HD-audio + dmix works as well.  There is no 0.5 sec
delay in snd_pcm_drop() itself but the code to sleep immediately.

> > Doesn't this problem happen if you use usb-audio with "hw" PCM?
> 
> The pcm_min example of alsalib doesn't run this way.
> --8<---------------cut here---------------start------------->8---
> 
> ALSA lib pcm.c:7125:(snd_pcm_set_params) Sample format not available for
> PLAYBAC
> K: Invalid argument
> Playback open error: Invalid argument
> --8<---------------cut here---------------end--------------->8---
> 
> Can you tell me which sampleformat works with this device???

Use plughw instead.

> > It's possible that the behavior of snd_pcm_drop() of usb-audio is
> > different from others because of URB handling.
> 
> The stop problem also happends on my laptop with hd audio and ad1981
> chip.

Works fine on my machine.  There shouldn't be any difference with
the codec chip regarding this, and very unlikely due to the difference
of controller chip, too.  So, the hardware shouldn't be a problem as
long as you use the relatively new version of alsa-driver.

The question might be the alsa-lib version.  I've tested only with
the latest one, alsa-lib-1.0.21.


Takashi