> Arg, I figured that was a longshot. You really need to get VIS working > for you, then you stand a chance. I know that some time ago, the guy > that did so much of the outstanding Linux work for PPC added the code > to xine to use the "Velocity Engine" or whatever of the G4 to get xine > to play mpegs reasonably on G4 apples. Possibly you could model some > modifications after that code to use VIS instead. If you wanted to "do > the right thing" you could write the VIS version of the mlib functions, > and a lot of cheering would be heard in the sparc/linux camp then! There are three possible approaches to this as I see it:
*. Add direct VIS support to mplayer - it will give you the quickest 'win' but it will be the most effort in the long run. *. Add (if it doesn't exist already as I suspect it does) generic vectorised[1] support in mplayer and wait for gcc to gain back end support for converting vector data types to VIS. I'm hoping to get time to work on this as part of a larger project in the next few weeks / months. As an UltraSPARC user I'm very keen to get this added. *. Wait for gcc to gain enough autovectorisation code to have it all done automatically. This is something we're working on but it's going to take a while... I'm not sure what solutions current programs / libraries have used for UltraSPARC / other architectures. You're most likely to see vector instructions turn up in PPC and x86 code (AltiVec, 3DNow! and SSE) Cheers, - Martin [1]. As in using the gcc vector data types. -- Martin [EMAIL PROTECTED] "Seasons change, things come to pass"
