[mpg123-devel] [ mpg123-Bugs-1547470 ] ALSA output fails on x86-64 64-bit

SourceForge Headquarters 225 Broadway Suite 1600 San Diego, CA 92101 +1 (858) 454-5900

Bugs item #1547470, was opened at 2006-08-27 12:23
Message generated for change (Comment added) made by nobody
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=733194&aid=1547470&group_id=135704

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: mpg123
Group: 0.60-beta5
Status: Closed
Resolution: Fixed
Priority: 5
Private: No
Submitted By: Stanley (sw_stanley)
Assigned to: Nobody/Anonymous (nobody)
Summary: ALSA output fails on x86-64 64-bit

Initial Comment:
I compiled a 64-bit binary and a 32-bit binary of
mpg123 0.60-beta6 (using --with-cpu=generic_fpu       

                 for fair test)                       

When playing an mp3 file using 64-bit version on Gentoo
Linux, 
only OSS output works.
ALSA output is just silence, but decoder exits normally.
Using 32-bit binary, both output works. 
Is it something 64-bit related?

----------------------------------------------------------------------

Comment By: Nobody/Anonymous (nobody)
Date: 2011-03-28 02:48

Message:
Index.. Neat :)

----------------------------------------------------------------------

Comment By: Stanley (sw_stanley)
Date: 2006-11-10 12:37

Message:
Logged In: YES 
user_id=1584814

With ALSA 1.0.13, the output now works.
May someone make a note of this on the homepage?

----------------------------------------------------------------------

Comment By: Clemens Ladisch (cladisch)
Date: 2006-11-03 17:08

Message:
Logged In: YES 
user_id=240068

There was a bug in the handling of the threshold values in
the sample rate converter plugin.

Please try the latest version of alsa-lib.

----------------------------------------------------------------------

Comment By: Thomas Orgis (sobukus)
Date: 2006-10-20 11:57

Message:
Logged In: YES 
user_id=470743

I guess I'll ask Clemens personally...
meanwhile: any update with a possibly upgraded Alsa/Kernel?

----------------------------------------------------------------------

Comment By: Thomas Orgis (sobukus)
Date: 2006-09-04 04:34

Message:
Logged In: YES 
user_id=470743

Just to lend the word fro while from sw_stanley: Perhaps 
you've found an Alsa bug... but let's wait what the Alsa 
expert says. Clemens? 

----------------------------------------------------------------------

Comment By: Stanley (sw_stanley)
Date: 2006-09-02 10:32

Message:
Logged In: YES 
user_id=1584814

I was always too quick to draw a conclusion!
Here are what I actually found, hope it helps:

1)In the main code, detecting/recovering from underrun 
depends on the value of snd_pcm_delay. However, I never 
see delay becomes < 0 EVEN when written == -EPIPE.
(Possibly an ALSA bug/doc error/misunderstanding)
IMO, the xrun state is not recovered using snd_pcm_forward 
too.

2)Stop threshold is odd on my machine. When threshold == 
boundary, it works with 32-bit but no luck with 64-bit: 
After first successful write scm_pcm_writei keeps 
returning -EPIPE. Shouldn't be xrun prevented using 
boundary?

----------------------------------------------------------------------

Comment By: Stanley (sw_stanley)
Date: 2006-09-01 15:57

Message:
Logged In: YES 
user_id=1584814

I found it!
The error is not caused by stop thershold or anything else,
but the audio_play_samples function in audio_alsa.c!

Line 221:
written = snd_pcm_writei(ai->handle, buf, frames); 
if (written >= 0)
	return snd_pcm_frames_to_bytes(ai->handle, written);
else
	return written;

On my box, snd_pcm_writei always returns -32(-EPIPE) except
for the first time(returns 4096). According to ALSA
documentation, -ve value means 3 types of errors: -EBADFD,
-EPIPE and -ESTRPIPE. AFAIK, there is no proper handling of
these errors in the mpg123 code.

----------------------------------------------------------------------

Comment By: Stanley (sw_stanley)
Date: 2006-08-30 14:11

Message:
Logged In: YES 
user_id=1584814

I forgot to mention that the decoder exits after a short 
period of time, just like decode and output to /dev/null

----------------------------------------------------------------------

Comment By: Stanley (sw_stanley)
Date: 2006-08-30 13:46

Message:
Logged In: YES 
user_id=1584814

I'm using Gentoo Linux with gentoo-linux-2.6.17-r5 and 
ALSA v1.0.11, with snd-intel8x0 driver.
I understand that by setting stop_threshold to boundary, 
ALSA would not stop device on underrun. But, this does not 
work well in 64-bit mode. Maybe an ALSA driver bug?

----------------------------------------------------------------------

Comment By: Clemens Ladisch (cladisch)
Date: 2006-08-30 11:16

Message:
Logged In: YES 
user_id=240068

The stop threshold is set to the boundary value to prevent
underruns from stopping the device. (There's a comment right
in front of that line.)

The boundary value tells where the hardware/application
sample pointers wrap around.  It's a multiple of the buffer
size and chosen to be quite large.  On a 64-bit architecture
it's bigger because snd_pcm_(u)frames_t has more bits.

Neither of these has anything to do with the output being
silent.

sw_stanley, what is your kernel version and your ALSA version?

----------------------------------------------------------------------

Comment By: Stanley (sw_stanley)
Date: 2006-08-30 09:28

Message:
Logged In: YES 
user_id=1584814

32-bit:
[audio_alsa.c:147] debug: Alsa buffer_size 3763 vs. 
boundary 986447872

64-bit:
[audio_alsa.c:147] debug: Alsa buffer_size 3763 vs. 
boundary 4236761349448794112

----------------------------------------------------------------------

Comment By: Thomas Orgis (sobukus)
Date: 2006-08-30 02:43

Message:
Logged In: YES 
user_id=470743

Hm, reading the alsa doc, the hw buffer size kindof makes 
sense to me in that place. 
But I don't know if the guy who kindly made that alsa code 
for mpg123 had something in mind with it. 

Could you check out the current svn version and 
(after ./autogen.sh) build with debugging enabled? 

I added output of these two possible values, looking like 
this for my ia32 laptop: 

[audio_alsa.c:147] debug: Alsa buffer_size 22050 vs. 
boundary 1445068800 

It would be interesting to see if these values are indeed 
different for you in 32 or 64bit mode - or if it is even 
more subtle. 

----------------------------------------------------------------------

Comment By: Stanley (sw_stanley)
Date: 2006-08-29 03:48

Message:
Logged In: YES 
user_id=1584814

audio_alsa.c, line 144:
snd_pcm_sw_params_set_stop_threshold(ai->handle,sw,boundary)

I found that in ALSA aplay utility, the stop threshold is
NOT set to sw_params boundary BUT the HW buffer_size. I
don't know why setting to boundary doesn't give the same
effect for mpg123! However, change to HW buffer_size works
for me in either 32-bit or 64-bit mode. Maybe it is a bug in
my own ALSA?(If then, sorry for annoying)

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=733194&aid=1547470&group_id=135704