[vlc-devel] audio output is starving problem - using libmad and alsa

[Celil Urgan]

Hi All

I sent this problem to forum so copying from there to here.

MSG1
Using alsa as audio output i am getting a error like "audio output is
starving (39734), playing silence".We know that mp3(libmad) decoding is
in mpga2fixed filter.My questions re "Which thread handle this
filter?Decoder or Alsa?" ,"What should i do for to fix this issue?"

MSG2
Update 

I also checked pthread code for priority settings. From
this code if you don't give "-rt-priority" option it seems all threads
will have same priority.
So all threads will be prioritized by their
own work. Input (disk access speed or network access or etc.) ,
Decoder(fixed-floating math time) , output(alsa buffersizes ,consume
speed)

Am i missing something?

int __vlc_thread_create( vlc_object_t *p_this, const char * psz_file, int i_line,
                         const char *psz_name, void * ( *func ) ( vlc_object_t * ),
                         int i_priority, bool b_wait )
{
    int i_ret;
    vlc_object_internals_t *p_priv = vlc_internals( p_this );

    struct vlc_thread_boot *boot = malloc (sizeof (*boot));
    if (boot == NULL)
        return errno;
    boot->entry = func;
    boot->object = p_this;

    vlc_object_lock( p_this );

    /* Make sure we don't re-create a thread if the object has already one */
    assert( !p_priv->b_thread );

#if defined( LIBVLC_USE_PTHREAD )
    pthread_attr_t attr;
    pthread_attr_init (&attr);

    /* Block the signals that signals interface plugin handles.
     * If the LibVLC caller wants to handle some signals by itself, it should
     * block these before whenever invoking LibVLC. And it must obviously not
     * start the VLC signals interface plugin.
     *
     * LibVLC will normally ignore any interruption caused by an asynchronous
     * signal during a system call. But there may well be some buggy cases
     * where it fails to handle EINTR (bug reports welcome). Some underlying
     * libraries might also not handle EINTR properly.
     */
    sigset_t set, oldset;
    sigemptyset (&set);
    sigdelset (&set, SIGHUP);
    sigaddset (&set, SIGINT);
    sigaddset (&set, SIGQUIT);
    sigaddset (&set, SIGTERM);

    sigaddset (&set, SIGPIPE); /* We don't want this one, really! */
    pthread_sigmask (SIG_BLOCK, &set, &oldset);

#ifndef __APPLE__
    if( config_GetInt( p_this, "rt-priority" ) > 0 )
#endif
    {
        struct sched_param p = { .sched_priority = i_priority, };
        int policy;

        /* Hack to avoid error msg */
        if( config_GetType( p_this, "rt-offset" ) )
            p.sched_priority += config_GetInt( p_this, "rt-offset" );
        if( p.sched_priority <= 0 )
            p.sched_priority += sched_get_priority_max (policy = SCHED_OTHER);
        else
            p.sched_priority += sched_get_priority_min (policy = SCHED_RR);

        pthread_attr_setschedpolicy (&attr, policy);
        pthread_attr_setschedparam (&attr, &p);
    }

    i_ret = pthread_create( &p_priv->thread_id, &attr, thread_entry, boot );
    pthread_sigmask (SIG_SETMASK, &oldset, NULL);
    pthread_attr_destroy (&attr)

MSG3
I know that my system is fast enough also libmad optimized for arch. Then what is your opinion about this problem?

MSG4
if i re-added "AOUT_PTS_TOLERANCE" to if block problem fixed while
playing song.But at start of song audio is choppy now. A old bug cause
a new bug.
So doing this switched to old bug a need a fix for it. Any comments?

vlc/src/audio_output.c line 320
    /* Here we suppose that all buffers have the same duration - this is
     * generally true, and anyway if it's wrong it won't be a disaster.
     */
    if ( p_buffer->start_date > start_date
                         + (p_buffer->end_date - p_buffer->start_date)+ AOUT_PTS_TOLERANCE  )
    /*
     *                   + AOUT_PTS_TOLERANCE )
     * There is no reason to want that, it just worsen the scheduling of
     * an audio sample after an output starvation (ie. on start or on resume)
     * --Gibalou
     */
    {
        const mtime_t i_delta = p_buffer->start_date - start_date;
        aout_unlock_output_fifo( p_aout );
 
        if ( !p_aout->output.b_starving )
            msg_Dbg( p_aout, "audio output is starving (%"PRId64"), "
                     "playing silence", i_delta );
        p_aout->output.b_starving = 1;
        return NULL;
    }



Any comments?

Cheers
curgan

_________________________________________________________________
More than messages–check out the rest of the Windows Live™.
http://www.microsoft.com/windows/windowslive/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/vlc-devel/attachments/20090911/865930ea/attachment.html>