dmix bugs. sounds very broken up.

dmix bugs. sounds very broken up.

[James Courtier-Dutton]

I have found two problems with using the dmix alsa device name.
1) snd_pcm_hw_params_can_pause (params)  causes alsa-lib to assert!

2) Sound is broken up.
a) It can only function with 2 periods, why is that? Having 8 periods
might be better, although "front" works fine with 2 periods, but "dmix"
with 2 periods fails.
b) Of the 2 periods, it sounds like sound is only being played from one
of the periods, with silence for the other period.

Summary: -
If I use device name "front", there are no problems with sound output.
If I use device name "dmix", there are the above problems.

Cheers
James




-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf

Re: dmix bugs. sounds very broken up.

[Jaroslav Kysela]

On Mon, 1 Sep 2003, James Courtier-Dutton wrote:

> I have found two problems with using the dmix alsa device name.
> 1) snd_pcm_hw_params_can_pause (params)  causes alsa-lib to assert!
> 
> 2) Sound is broken up.
> a) It can only function with 2 periods, why is that? Having 8 periods
> might be better, although "front" works fine with 2 periods, but "dmix"
> with 2 periods fails.
> b) Of the 2 periods, it sounds like sound is only being played from one
> of the periods, with silence for the other period.
> 
> Summary: -
> If I use device name "front", there are no problems with sound output.
> If I use device name "dmix", there are the above problems.

Show me your code, please (compilable example).

						Jaroslav

-----
Jaroslav Kysela <perex@suse.cz>
Linux Kernel Sound Maintainer
ALSA Project, SuSE Labs



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf

Re: dmix bugs. sounds very broken up.

[James Courtier-Dutton]

[-- Attachment #1: Type: text/plain, Size: 999 bytes --]

Jaroslav Kysela wrote:
> On Mon, 1 Sep 2003, James Courtier-Dutton wrote:
> 
> 
>>I have found two problems with using the dmix alsa device name.
>>1) snd_pcm_hw_params_can_pause (params)  causes alsa-lib to assert!
>>
>>2) Sound is broken up.
>>a) It can only function with 2 periods, why is that? Having 8 periods
>>might be better, although "front" works fine with 2 periods, but "dmix"
>>with 2 periods fails.

See attachment dmix-fail.c
compile with
gcc -g -DDEBUG -lasound a.c

Shows problem (1) and (2)

>>b) Of the 2 periods, it sounds like sound is only being played from one
>>of the periods, with silence for the other period.
>>
I don't have a small compilable example for this yet. The problem 
application is the latest xine cvs. (xine.sf.net)

>>Summary: -
>>If I use device name "front", there are no problems with sound output.
>>If I use device name "dmix", there are the above problems.
> 
> 
> Show me your code, please (compilable example).
> 
> 						Jaroslav
> 

Cheers
James

[-- Attachment #2: dmix-fail.c --]
[-- Type: text/plain, Size: 10176 bytes --]

/* 
 * Copyright (C) 2000-2002 the xine project
 * 
 * This file is part of xine, a free video player.
 * 
 * xine is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 * 
 * xine is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 * 
 * You should have received a copy of the GNU General Public License
 * along with this program; if not, write to the Free Software
 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA  02111-1307, USA
 *
 * Credits go 
 * - for the SPDIF A/52 sync part
 * - frame size calculation added (16-08-2001)
 * (c) 2001 Andy Lo A Foe <andy@alsaplayer.org>
 * for initial ALSA 0.9.x support.
 *     adding MONO/STEREO/4CHANNEL/5CHANNEL/5.1CHANNEL analogue support.
 * (c) 2001 James Courtier-Dutton <James@superbug.demon.co.uk>
 *
 * 
 * $Id: audio_alsa_out.c,v 1.106 2003/09/01 04:08:41 jcdutton Exp $
 */

#ifdef HAVE_CONFIG_H
#include "config.h"
#endif

#include <stdio.h>
#include <stdarg.h>
#include <errno.h>
#include <string.h>
#include <unistd.h>
#include <stdlib.h>
#include <fcntl.h>
#include <math.h>
#include <alloca.h>

#define ALSA_PCM_NEW_HW_PARAMS_API
#define ALSA_PCM_NEW_SW_PARAMS_API
#include <alsa/asoundlib.h>

#include <sys/ioctl.h>
#include <inttypes.h>
#include <pthread.h>

/*
#define ALSA_LOG
*/
/*
#define LOG_DEBUG
*/

#define AO_OUT_ALSA_IFACE_VERSION 7

#define BUFFER_TIME               1000*1000
#define PERIOD_TIME               100*1000
#define GAP_TOLERANCE             5000

#define MIXER_MASK_LEFT           (1 << 0)
#define MIXER_MASK_RIGHT          (1 << 1)
#define MIXER_MASK_STEREO         (MIXER_MASK_LEFT|MIXER_MASK_RIGHT)

typedef struct alsa_driver_s {

  snd_pcm_t         *audio_fd;
  int                capabilities;
  int                open_mode;
  int		     has_pause_resume;

  int32_t            output_sample_rate, input_sample_rate;
  double             sample_rate_factor;
  uint32_t           num_channels;
  uint32_t           bits_per_sample;
  uint32_t           bytes_per_frame;
  uint32_t           bytes_in_buffer;      /* number of bytes writen to audio hardware   */
  snd_pcm_uframes_t  buffer_size;
  int32_t            mmap; 

} alsa_driver_t;

static snd_output_t *jcd_out;

/*
 * open the audio device for writing to
 */
int main (int argc, char **argv) {
  
  alsa_driver_t         variables;
  alsa_driver_t         *this = &variables;
  char                 *pcm_device;
  snd_pcm_stream_t      direction = SND_PCM_STREAM_PLAYBACK; 
  snd_pcm_hw_params_t  *params;
  snd_pcm_sw_params_t  *swparams;
  snd_pcm_access_mask_t *mask;
  snd_pcm_uframes_t     period_size;
  uint32_t              periods;
  uint32_t              buffer_time=BUFFER_TIME;
  int                   err, dir;
  int                 open_mode=1; /* NONBLOCK */
  /* int                   open_mode=0;  BLOCK */
  int rate = 48000;
  int bits = 16;

  snd_pcm_hw_params_alloca(&params);
  snd_pcm_sw_params_alloca(&swparams);
  err = snd_output_stdio_attach(&jcd_out, stdout, 0);
 
  pcm_device = "dmix"; 

#ifdef ALSA_LOG
  printf("audio_alsa_out: Audio Device name = %s\n",pcm_device);
  printf("audio_alsa_out: Number of channels = %d\n",this->num_channels);
#endif

  this->audio_fd = NULL;

  this->open_mode              = open_mode;
  this->input_sample_rate      = rate;
  this->bits_per_sample        = bits;
  this->num_channels        = 2;
  this->bytes_in_buffer        = 0;
  /*
   * open audio device
   */
  err=snd_pcm_open(&this->audio_fd, pcm_device, direction, open_mode);      
  if(err <0 ) {                                                           
    printf ("audio_alsa_out: snd_pcm_open() of %s failed: %s\n", pcm_device, snd_strerror(err));               
    printf ("audio_alsa_out: >>> check if another program don't already use PCM <<<\n");     
    return 0;
  }
  /* printf ("audio_alsa_out: snd_pcm_open() opened %s\n", pcm_device); */ 
  /* We wanted non blocking open but now put it back to normal */
  //snd_pcm_nonblock(this->audio_fd, 0);
  snd_pcm_nonblock(this->audio_fd, 1);
  /*
   * configure audio device
   */
  err = snd_pcm_hw_params_any(this->audio_fd, params);
  if (err < 0) {
    printf ("audio_alsa_out: broken configuration for this PCM: no configurations available\n");
    goto __close;
  }
  /* set interleaved access */
  if (this->mmap != 0) {
    mask = alloca(snd_pcm_access_mask_sizeof());
    snd_pcm_access_mask_none(mask);
    snd_pcm_access_mask_set(mask, SND_PCM_ACCESS_MMAP_INTERLEAVED);
    snd_pcm_access_mask_set(mask, SND_PCM_ACCESS_MMAP_NONINTERLEAVED);
    snd_pcm_access_mask_set(mask, SND_PCM_ACCESS_MMAP_COMPLEX);
    err = snd_pcm_hw_params_set_access_mask(this->audio_fd, params, mask);
    if (err < 0) {
      printf ("audio_alsa_out: mmap not availiable, falling back to compatiblity mode\n");
      this->mmap=0;
      err = snd_pcm_hw_params_set_access(this->audio_fd, params,
                                     SND_PCM_ACCESS_RW_INTERLEAVED);
    }
  } else {
    err = snd_pcm_hw_params_set_access(this->audio_fd, params,
                                     SND_PCM_ACCESS_RW_INTERLEAVED);
  }
      
  if (err < 0) {
    printf ("audio_alsa_out: access type not available\n");
    goto __close;
  }
  /* set the sample format ([SU]{8,16{LE,BE}})*/
  err = snd_pcm_hw_params_set_format(this->audio_fd, params, (bits == 16) ?
#ifdef WORDS_BIGENDIAN
		  SND_PCM_FORMAT_S16_BE
#else
		  SND_PCM_FORMAT_S16_LE
#endif
		  : SND_PCM_FORMAT_U8);
  if (err < 0) {
    printf ("audio_alsa_out: sample format non available\n");
    goto __close;
  }
  /* set the number of channels */
  err = snd_pcm_hw_params_set_channels(this->audio_fd, params, this->num_channels);
  if (err < 0) {
    printf ("audio_alsa_out: Cannot set number of channels to %d (err=%d)\n", this->num_channels, err);
    goto __close;
  }
  /* set the stream rate [Hz] */
  dir=0;
  err = snd_pcm_hw_params_set_rate_near(this->audio_fd, params, &rate, &dir);
  if (err < 0) {
    printf ("audio_alsa_out: rate not available\n");
    goto __close;
  }
  this->output_sample_rate = (uint32_t)rate;
  if (this->input_sample_rate != this->output_sample_rate) {
    printf ("audio_alsa_out: audio rate : %d requested, %d provided by device/sec\n",
	    this->input_sample_rate, this->output_sample_rate);
  }

  /* Set period to buffer size ratios at 8 periods to 1 buffer */
  dir=-1;
  periods=8;
  err = snd_pcm_hw_params_set_periods_near(this->audio_fd, params, &periods ,&dir);
  if (err < 0) {
    printf ("audio_alsa_out: unable to set any periods\n");
    goto __close;
  }
  printf ("audio_alsa_out: Requested 8 periods, got %d\n", periods);

  /* set the ring-buffer time [us] (large enough for x us|y samples ...) */
  dir=0;
  err = snd_pcm_hw_params_set_buffer_time_near(this->audio_fd, params, &buffer_time, &dir);
  if (err < 0) {
    printf ("audio_alsa_out: buffer time not available\n");
    goto __close;
  }
  err = snd_pcm_hw_params_get_buffer_size(params, &(this->buffer_size));

#if 0
  /* set the period time [us] (interrupt every x us|y samples ...) */
  dir=0;
  period_size=this->buffer_size/8;
  err = snd_pcm_hw_params_set_period_size_near(this->audio_fd, params, &period_size, &dir);
  if (err < 0) {
    printf ("audio_alsa_out: period time not available");
    goto __close;
  }
#endif
  dir=0;
  err = snd_pcm_hw_params_get_period_size(params, &period_size, &dir);
  if (2*period_size > this->buffer_size) {
    printf ("audio_alsa_out: buffer to small, could not use\n");
    goto __close;
  }
  
  /* write the parameters to device */
  err = snd_pcm_hw_params(this->audio_fd, params);
  if (err < 0) {
    printf ("audio_alsa_out: pcm hw_params failed: %s\n", snd_strerror(err));
    goto __close;
  }
  /* Check for pause/resume support */
  this->has_pause_resume = ( snd_pcm_hw_params_can_pause (params)
			    && snd_pcm_hw_params_can_resume (params) );
  this->sample_rate_factor = (double) this->output_sample_rate / (double) this->input_sample_rate;
  this->bytes_per_frame = snd_pcm_frames_to_bytes (this->audio_fd, 1);
  /*
   * audio buffer size handling
   */
  /* Copy current parameters into swparams */
  err = snd_pcm_sw_params_current(this->audio_fd, swparams);
  if (err < 0) {
    printf ("audio_alsa_out: Unable to determine current swparams: %s\n", snd_strerror(err));
    goto __close;
  }
  /* align all transfers to 1 sample */
  err = snd_pcm_sw_params_set_xfer_align(this->audio_fd, swparams, 1);
  if (err < 0) {
    printf ("audio_alsa_out: Unable to set transfer alignment: %s\n", snd_strerror(err));
    goto __close;
  }
  /* allow the transfer when at least period_size samples can be processed */
  err = snd_pcm_sw_params_set_avail_min(this->audio_fd, swparams, period_size);
  if (err < 0) {
    printf ("audio_alsa_out: Unable to set available min: %s\n", snd_strerror(err));
    goto __close;
  }
  /* start the transfer when the buffer contains at least period_size samples */
  err = snd_pcm_sw_params_set_start_threshold(this->audio_fd, swparams, period_size);
  if (err < 0) {
    printf ("audio_alsa_out: Unable to set start threshold: %s\n", snd_strerror(err));
    goto __close;
  }

  /* never stop the transfer, even on xruns */
  err = snd_pcm_sw_params_set_stop_threshold(this->audio_fd, swparams, this->buffer_size);
  if (err < 0) {
    printf ("audio_alsa_out: Unable to set stop threshold: %s\n", snd_strerror(err));
    goto __close;
  }

  /* Install swparams into current parameters */
  err = snd_pcm_sw_params(this->audio_fd, swparams);
  if (err < 0) {
    printf ("audio_alsa_out: Unable to set swparams: %s\n", snd_strerror(err));
    goto __close;
  }
#ifdef ALSA_LOG
  snd_pcm_dump_setup(this->audio_fd, jcd_out); 
  snd_pcm_sw_params_dump(swparams, jcd_out);
#endif
  
//  return this->output_sample_rate;

__close:
  snd_pcm_close (this->audio_fd);
  this->audio_fd=NULL;
  return 0;
}

Re: dmix bugs. sounds very broken up.

[James Courtier-Dutton]

Jaroslav Kysela wrote:
> On Tue, 2 Sep 2003, James Courtier-Dutton wrote:
> 
> 
>>Jaroslav Kysela wrote:
>>
>>>On Mon, 1 Sep 2003, James Courtier-Dutton wrote:
>>>
>>>
>>>
>>>>I have found two problems with using the dmix alsa device name.
>>>>1) snd_pcm_hw_params_can_pause (params)  causes alsa-lib to assert!
>>>>
>>>>2) Sound is broken up.
>>>>a) It can only function with 2 periods, why is that? Having 8 periods
>>>>might be better, although "front" works fine with 2 periods, but "dmix"
>>>>with 2 periods fails.
>>
>>See attachment dmix-fail.c
>>compile with
>>gcc -g -DDEBUG -lasound a.c
>>
>>Shows problem (1) and (2)
> 
> 
> (1)
> 
> The count of used periods is probably 2.73 for 64kB buffer, my setup is:
> 
> buffer_size  : 16384
> period_size  : 6000
> 
> If you want another setup, you must write your own configuration to
> ~/.asoundrc.
> 
> (2)
> 
> Problem with uninitialized pcm->info value is fixed in CVS.
> 
> 
>>>>b) Of the 2 periods, it sounds like sound is only being played from one
>>>>of the periods, with silence for the other period.
>>>>
>>
>>I don't have a small compilable example for this yet. The problem
>>application is the latest xine cvs. (xine.sf.net)
> 
> 
> Note that it might be caused that your code is broken. Please, use the
> correct transfer unit (period_size) and not the period count which might
> be a real not integer value, thus buffer_size / period_size !=
> integer(period_count).
> 
> 						Jaroslav
> 
> -----
> Jaroslav Kysela <perex@suse.cz>
> Linux Kernel Sound Maintainer
> ALSA Project, SuSE Labs
> 
> 
The aim of my code is to try for a buffer that is about 1 second long or 
less, and has hopefully 8 periods.
I had always assumed that the period_count had to be an integer.
If it is in fact possible to to have the period_count a non-interger on 
all sound cards, then I will change my code.

The problem with trying the period_size approach was: -
set_buffer_time_near(1 second)
size = get_buffer_size()  (e.g. this gets set to 16384)
set_period_size_near(size/8, dir=0)  (e.g. try to set period_size to 
2048) <- Failed on some cards/devices that did not like 8 periods.

I thought that set_period_size_near with a dir=0, should have resulted 
in a setting of 6000 for "dmix", but it in-fact failed, and I did not 
want to go round a loop trying different period sizes until one worked.

Does this point to a bug in the set_period_size_near function then?

Cheers
James



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf

Re: dmix bugs. sounds very broken up.

[Jaroslav Kysela]

On Tue, 2 Sep 2003, James Courtier-Dutton wrote:

> Jaroslav Kysela wrote:
> > On Tue, 2 Sep 2003, James Courtier-Dutton wrote:
> >
> >
> >>Jaroslav Kysela wrote:
> >>
> >>>On Mon, 1 Sep 2003, James Courtier-Dutton wrote:
> >>>
> >>>
> >>>
> >>>>I have found two problems with using the dmix alsa device name.
> >>>>1) snd_pcm_hw_params_can_pause (params)  causes alsa-lib to assert!
> >>>>
> >>>>2) Sound is broken up.
> >>>>a) It can only function with 2 periods, why is that? Having 8 periods
> >>>>might be better, although "front" works fine with 2 periods, but "dmix"
> >>>>with 2 periods fails.
> >>
> >>See attachment dmix-fail.c
> >>compile with
> >>gcc -g -DDEBUG -lasound a.c
> >>
> >>Shows problem (1) and (2)
> >
> >
> > (1)
> >
> > The count of used periods is probably 2.73 for 64kB buffer, my setup is:
> >
> > buffer_size  : 16384
> > period_size  : 6000
> >
> > If you want another setup, you must write your own configuration to
> > ~/.asoundrc.
> >
> > (2)
> >
> > Problem with uninitialized pcm->info value is fixed in CVS.
> >
> >
> >>>>b) Of the 2 periods, it sounds like sound is only being played from one
> >>>>of the periods, with silence for the other period.
> >>>>
> >>
> >>I don't have a small compilable example for this yet. The problem
> >>application is the latest xine cvs. (xine.sf.net)
> >
> >
> > Note that it might be caused that your code is broken. Please, use the
> > correct transfer unit (period_size) and not the period count which might
> > be a real not integer value, thus buffer_size / period_size !=
> > integer(period_count).
> >
> > 						Jaroslav
> >
> > -----
> > Jaroslav Kysela <perex@suse.cz>
> > Linux Kernel Sound Maintainer
> > ALSA Project, SuSE Labs
> >
> >
> The aim of my code is to try for a buffer that is about 1 second long or
> less, and has hopefully 8 periods.
> I had always assumed that the period_count had to be an integer.
> If it is in fact possible to to have the period_count a non-interger on
> all sound cards, then I will change my code.

Most cards does not have such constraint, so yes, this fact should be
handled. The dmix plugin comes with these "strange" defaults mainly for
the application developers to debug their "broken" code which does not
conform fully the ALSA semantics.

> The problem with trying the period_size approach was: -
> set_buffer_time_near(1 second)
> size = get_buffer_size()  (e.g. this gets set to 16384)
> set_period_size_near(size/8, dir=0)  (e.g. try to set period_size to
> 2048) <- Failed on some cards/devices that did not like 8 periods.
>
> I thought that set_period_size_near with a dir=0, should have resulted
> in a setting of 6000 for "dmix", but it in-fact failed, and I did not
> want to go round a loop trying different period sizes until one worked.
>
> Does this point to a bug in the set_period_size_near function then?

Maybe. I used your test code and set_period_size_near() works as expected
with the trident driver. Which cards are failing?

						Jaroslav

-----
Jaroslav Kysela <perex@suse.cz>
Linux Kernel Sound Maintainer
ALSA Project, SuSE Labs


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf

Safe to call snd_pcm_close after failed snd_pcm_open?

[Arve Knudsen]

Hi

Is it safe to call snd_pcm_close for a pcm which failed to open (assuming 
the pcm handle is !NULL)? Not that important, but it might simplify my 
code somewhat.

Regards

Arve Knudsen


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf

Re: Safe to call snd_pcm_close after failed snd_pcm_open?

[Jaroslav Kysela]

On Tue, 9 Sep 2003, Arve Knudsen wrote:

> Hi
>
> Is it safe to call snd_pcm_close for a pcm which failed to open (assuming
> the pcm handle is !NULL)? Not that important, but it might simplify my
> code somewhat.

No, if open failed, the pcm handle is unitialized, thus
the snd_pcm_close() call is not allowed.

						Jaroslav

-----
Jaroslav Kysela <perex@suse.cz>
Linux Kernel Sound Maintainer
ALSA Project, SuSE Labs


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf

Re: dmix bugs. sounds very broken up.

[James Courtier-Dutton]

Jaroslav Kysela wrote:
> 
> (2)
> 
> Problem with uninitialized pcm->info value is fixed in CVS.
> 
> 
It seems to be impossible to get anything from the sf.net anonymous cvs. 
Just try to follow the "CVS Web interface" link from the alsa web page. 
I never appears. That is the same interface the anonymouse cvs user 
would use.

I use kernel 2.6test4.
Would it be possible to create a nightly patch file, that would compare 
the alsa cvs, with the kernel 2.6test4, and create a patch file for us 
to download. It would be ideal if the patch was a progressive patch, in 
that with each change to the cvs, it just adds a new section of the 
patch file, instead of creating a unified patch of all the cvs checkins.
In this way, if I used the first patch, and then the cvs gets updated, I 
could download the latest patch, do a diff between it and the previous 
one I applied, and be left with just the new patches I needed.

Cheers
James



-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf

Re: dmix bugs. sounds very broken up.

[Jaroslav Kysela]

On Tue, 2 Sep 2003, James Courtier-Dutton wrote:

> Jaroslav Kysela wrote:
> >
> > (2)
> >
> > Problem with uninitialized pcm->info value is fixed in CVS.
> >
> >
> It seems to be impossible to get anything from the sf.net anonymous cvs.
> Just try to follow the "CVS Web interface" link from the alsa web page.
> I never appears. That is the same interface the anonymouse cvs user
> would use.

Yes, I hate the SF latest policy to update anonymous CVS trees once per
day. But CVS Web interface works for me. I am seriously thinking to move
the master CVS repository back to the ALSA server after next software
update.

> I use kernel 2.6test4.
> Would it be possible to create a nightly patch file, that would compare
> the alsa cvs, with the kernel 2.6test4, and create a patch file for us
> to download. It would be ideal if the patch was a progressive patch, in
> that with each change to the cvs, it just adds a new section of the
> patch file, instead of creating a unified patch of all the cvs checkins.
> In this way, if I used the first patch, and then the cvs gets updated, I
> could download the latest patch, do a diff between it and the previous
> one I applied, and be left with just the new patches I needed.

Please, subscribe our cvslog mailing list. There are full patches up to
20000 lines now (plus commit logs).

						Jaroslav

-----
Jaroslav Kysela <perex@suse.cz>
Linux Kernel Sound Maintainer
ALSA Project, SuSE Labs


-------------------------------------------------------
This sf.net email is sponsored by:ThinkGeek
Welcome to geek heaven.
http://thinkgeek.com/sf