On Monday 28 August 2006 16:38, Timothy Miller wrote: > On 8/27/06, Dieter <[EMAIL PROTECTED]> wrote: > > I've been researching this a bit. Preliminary conclusion is that > > the OGC card needs to provide XvMC. > > Any patents apply here? > > Anyhow, I don't want to support any one specific codec. If we can > provide something that's as useful for Theora as it is for MPEG1, 2, > and 4, then I'll consider it, depending on how much logic is > required.
Without having looked at it in detail, some thoughts. We need to do two things: motion compensation, and iDCT. MC could be done using the drawing engine: basically we're rendering squares from one or two buffers to another. Perhaps there could be a special command format that allows sending the coordinates in a more compact way. XvMC doesn't have a codec-independent interface, instead it just lets you inform the application that you support MPEG-1, 2, 4, and/or H263. I haven't looked at it in detail, but it seems that they all do block-based motion compensation. Looking at the XvMC API 1.0 spec at http://www.xfree86.org/~mvojkovi/XvMC_API.txt it seems that XvMC only supports 8x8 blocks. I understand that some newer codecs use variable block size MC. Perhaps we should check first how relevant XvMC is, and whether we can devise something better and/or more flexible. For iDCT support, we have to have adders and multipliers. I found a page (http://f-cpu.seul.org/whygee/dct_fc0/dct_fc0.html) that discusses some algorithms and tradeoffs here. I reckon there are some people at the Theora lists that know much more about this stuff than we do. Perhaps we should ask them what they would like to have? Lourens
pgpO3I79ElEwz.pgp
Description: PGP signature
_______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
