[Mplayer-cvslog] CVS: main/libmpcodecs vd_ffmpeg.c,1.25,1.26

[Arpi]

Hi,

> > Modified Files:
> > 	vd_ffmpeg.c
> > Log Message:
> > direct rendering method 1 (disabled currently as its not bugfree / finished
> > yet, just set dr1=1 if u want to try it)
> hmm, I've tested it, but surprisingly the speedup is pretty small, ffdivx with 
no wonder. in best case it saves one memcpy, otherwise nothing.
if you have fast ram, the speedup is near zero :(

> dr1 wins less than a second decoding coyote ugly sample (1m26s), but divxds 
> wins more than three seconds, so it seems dr2 (used by dshow) is more 
> effective than dr1.
of course, especially as that sample file has 50% black bands, so uses half
video bandwith with dr2...

i think time to implement MB skipping to swscaler or at least write a
blitter (agpmemcpy) using that info, and copy only non-skipped MBs.
it's easy to export skip MB array from libmpeg2 and libavcodec, and sorenson
too, and some old win32 codec supports IF09 which has a 4th plane containing
this data (the new ones support dr2).

it's more or less the emulation of dr2, with the lack of in-l1cache csp
conversion. but, when no conversion is needed (for most cases, where codec's
yuv format is supported by vo) it's equal or better than dr2.


A'rpi / Astral & ESP-team

--
Developer of MPlayer, the Movie Player for Linux - http://www.MPlayerHQ.hu