[MPlayer-users] Bicubic scaling in vo_gl - does it ever work?

Vladimir Mosgalin mosgalin at VM10124.spb.edu
Tue Jun 20 02:00:27 CEST 2006


Hi Reimar D?ffinger!

 On 2006.06.19 at 22:21:16 +0200, Reimar D?ffinger wrote next:

> > notice that shader is "slow"? I thought that it's the matter of just
> > getting desired fps so there are no slowdowns, you shouldn't even be
> > able to notice if the shader is faster than that. Am I missing
> > something?
> 
> Well, I can make a 1024-instruction shader. That is if you don't mind
> watching the video in a 1x1 sized window. So making the shader faster
> means on the same video card I can still play videos at a higher
> (fullscreen) resolution.

Sure.. I forgot that you need more resources as resolution increases.

> Also for those who absolutely want to minimize A-V delay it relates more
> directly to CPU usage: If you enable glfinish you see exactly how much
> time the shader uses.

Hm. Well, triple buffering should almost get rid of this effect.. at the
price of having shown video frame to lag from rendered one by two
frames. If a-v sync code allows to insert constant 2/fps compensation,
it may be useful.

> And in theory there is also a difference in power consumption, though I
> doubt it matters (well, actually as soon as there are GPUs that reduce
> frequency dependant on load it might become relevant).

It's not only about reducing frequency, stock cooling system on some
cards becomes much too loud when the card is hot. The gl driver has this
downside - you are watching something and you hear the card becomes
louder in the middle of it.

Well anyway. Both version of the patch don't help at all, so it must be
something else. Two facts:
1) -vf-add eq2 get rids of b/c/s changing problems, of course
2) when not running mplayer for a minute or so, chances to get no
   defects on either first or second run are very high on any video.
3) did I mention that once appeared, defects won't go way no matter
   what?

Also maybe it's just me, but chances to get no defects are higher with
yuv=2 than with yuv=3. yuv=4 doesn't report any problem with texture
units but is definitely broken when used together with lscale/cscale.

When using only yuv=2 and 2) rule, I can increase success rate for every
file. A few still don't need these tricks to display correctly, though.

PS er.. there is no such thing as undefined value in register or
something in video card programming, is it?


It isn't overheating problem or something like that too. Also I'm sure
ATI Powerplay or whatever that feature is called doesn't work for me,
the card always work in 2D profile mode with slower frequencies. Not
that it causes any problems, of course.

-- 

Vladimir



More information about the MPlayer-users mailing list