I'm CCing alsa-devel because this discussion might be interesting for other developers, too.
On Mon, 28 May 2007, Maarten Lankhorst wrote:
Jaroslav Kysela schreef:
On Mon, 28 May 2007, Maarten Lankhorst wrote:
After playing around a little more with alsa I tried to do manipulation of pointer using the instructions, I wrote some simple code either does rewind or forward to get to the right point, taking in respect the amount of delay and where to go.
However it doesn't seem to work: Looking at the source code it seems rewinding isn't implemented, using the #if 0'd code there gives problems when doing snd_pcm_forward later: It then ends up at the pointer where the sound is currently being played at.
Not all alsa-lib plugins are capable or have implemented this fine position control (check error value).
For a plugin like dmix, I would expect it to be implemented, especially since most cards sold are integrated cards that don't have multiple sound buffers.
Go ahead and send us patches. We're doing open source :-) Anyway any rewind operation mens "undo" - it will eat extra CPU ticks for nothing.
As workaround I used the following in my own code: typeof(snd_pcm_forward) snd_pcm_mmap_appl_forward; #define snd_pcm_forward snd_pcm_mmap_appl_forward
typeof(snd_pcm_rewind) snd_pcm_mmap_appl_backward; #define snd_pcm_rewind snd_pcm_mmap_appl_backward
This works, but I wonder if there is a cleaner way to do it.
It won't work with all plugins and mentioned functions are not part of public API.
As I work in wine on it's DirectSound implementation, I can tell it relies heavily on the fact that it can lock a buffer at an arbitrary position, and that it works with absolute pointers, rather then relative. So if snd_pcm_rewind/snd_pcm_forward don't work as expected, an alsa driver for it would become nearly useless, I don't expect it to always work, but I would expect it to work for a common plugin like dmix at least.
Also, I noticed some audio glitches when using this in combination with alsa-lib's resampling, is that because of the way it is done or is something else wrong?
You should not use mentioned functions.
You can try to use the direct hardware access (hw:CARD_NAME) devices (you'll lose all stream conversions from the ALSA side) or use an intermediate buffer for fine position control and configure ALSA to use smallest ring buffer as possible (to minimalize latency).
JaroslaI wish that would work, directsound expects to have access to a fixed size buffer, and if I go for direct hardware access, I lose the ability to work properly with most cheap cards on the market, since people want to be able to listen to music on background while gaming. For now it seems I have to use the internal calls in case of dmix, or modify already committed data, which is what winealsa's current driver does. I don't know what is the least of those 2 evils, or if there is a third option. I'm open for advice.
I'm afraid, that you're trying to exactly copy DirectSound API. Playing with the buffer position in this way is always dangerous, because it's really hard to predict the final result in the multitasking environment.
Think about these proposals:
1) what sample (ring buffer) rewritting is supposed to do - maybe you can change only not yet commited samples with satisfactory results 2) use a small ALSA ring buffer and use own "big" ring buffer for changes before you commit samples to ALSA's ring buffer - you'll increase a bit sample flow latency, but you can do anything you want (you just add extra buffering to avoid rewind / forward for ALSA API) 3) fix (or better implement) dmix plugin rewind function
Jaroslav
----- Jaroslav Kysela perex@suse.cz Linux Kernel Sound Maintainer ALSA Project, SUSE Labs