[pulseaudio-discuss] Pulseaudio fails when using VDPAU

[David Hagood]

I have been having a problem for some time that I finally traced down 
last night: it seems that the version 1.1 of Pulseaudio and the VDPAU 
library from nVidia as of 310.19 don't like each other when using 
digital output (but NOT analog).

Test case:
Clevo laptop:
* SPDIF connected to a stereo receiver that decodes SPDIF
* analog audio to internal speakers
* both made available to Pulseaudio clients a separate sinks.
* Intel(R) Core(TM)2 Duo CPU     P8700  @ 2.53GHz
* Linux version 3.7.0  SMP PREEMPT, NOHZ, CONFIG_HIGH_RES_TIMERS=y

Test 1: Run Audacious with the Pulseaudio output plugin. Play an MP3.
Results: Plays on both analog and digital with no problems.
Test 2: Run mplayer, with -ao pulse, and play the same MP3.
Results: Plays on both analog and digital with no problems.
Test 3: Run mplayer, with -ao pulse -vo vdpau, play a video file with audio.
Results: Plays on analog. Digital doesn't work - the receiver never 
locks onto the digital signal. Occasional popping heard.
Test 4: Run mplayer, with -ao pulse -vo gl2, play the same video file.
Results: Plays on both analog and digital with no problems.
Test 5: Run Audacious as above, with output to digital via Pulse. Start 
mplayer on a video file with "-ao null -vo vdpau".
Results: Audio fails as soon as mplayer starts playing (as above). Audio 
resumes as soon as mplayer stops.

It seems that merely accessing the vdpau library is enough to cause 
issues. I would guess that there may be some issue with vdpau blocking 
Pulseaudio long enough to cause gapping and disrupt the audio stream.

I'm not saying this is Pulseaudio's issue per se, but rather I am 
calling this to everybody's attention to see if
1) anybody else is seeing this behavior
2) is there anything that can be done to Pulseaudio to work around this?
(because I have little reason to believe that nVidia would change their 
code if they are indeed at fault in a timely manner).