Hi all! I intended to get rid of the "jack_transport; play; exit" phase of starting jackd and pulseaudio. But then I installed a realtime patched kernel and started to run jackd with very low latency settings, and found out that the jack sink module is unusable in such setup. So I had more to fix - the processing callback had to be made realtime-safe. The result is attached. This is the first patch I ever send anywhere, so it may be in a wrong format or anything. It's the output of "svn diff". What I did was adding internal buffering to the sink. That's implemented with Jack's utility ringbuffer. Now the processing function just reads from the buffer and tells the pulse core (correct terminology?) to render some more audio, but the processing function doesn't wait for that to happen. I guess that's about it. A new module parameter is added too. buffersize takes an unsigned integer as an argument, which is the amount of frames the user wishes to be in the ringbuffer, defaulting to 4096. The actual buffersize is rounded up to nearest power of two, and is always greater than the jack blocksize (frames per processing call). The blocksize change callback isn't tested, since it seems that my jack doesn't support dynamic changing of the blocksize. I believe I enabled it when configuring the jack build, but maybe it's just not yet implemented. Or maybe I didn't enable it. -- Tanu Kaskinen -------------- next part -------------- A non-text attachment was scrubbed... Name: more_buffering.patch Type: text/x-patch Size: 14043 bytes Desc: not available URL: <http://lists.freedesktop.org/archives/pulseaudio-discuss/attachments/20070508/0b89ad24/attachment.bin>
