> > v4l2 has buffer flags for even/odd field, v4l has not. Is there any way > > Does it make sense to do v4l2 ? When is it going to make it into the > kernel ? (I don't want to require people to patch the kernel in addition > to everything else..) including v4l2 is planned for 2.5.x. > > to capture both fields interleaved into one buffer? Using pitch = > > 2*scanlinelength or something like that? Applications expect the image > > interleaved in one buffer ... > > Not really. (As DMA granularity is 4096 bytes, but pitch is 640*2). I > could make it fake this during read but this requires doubling buffers. > (4 instead of 2, 1.2meg ram instead of 600k) Hmm, bad. > > Any v4l app should work (in theory ...). > > There is no test suite (at least I don't know any ...). > > Well, I just wanted to take the easy way and get this to work with some > existing apps instead of implementing ioctl's one by one ;) As you already have xawtv you can try this one ... > > Using more (than two) capture buffers basically reduces the risc that > > you lose frames on temporary CPU load peaks. > > Do you know of any apps that actually care about latency issues ? Videoconferencing is very sensible to latencies, you might have a loom at www.openh323.org / gnomemeeting. Gerd
