[MPlayer-dev-eng] Fast OpenGL playback

[The Wanderer]

Reimar ? wrote:

>>>> -vo X11:     5.4 sec
>>>> -vo gl2:     6.85 sec
>>>> -vo gl :    ~7 sec
>>>> -vo dga:     4.7 sec
>>>> -vo vesa:   ~3.7 sec
>>>> -vo gl_hack: 3.7 sec 
>>> 
>>> Those are all really slow vos. Much more interesting would be a
>>> comparison against -vo xv or vidix, I would consider same speed
>>> as vo_vesa not really good. You might also want to compare -vo
>>> gl:manyfmts (if  your MPlayer version is new enough).
>> 
>> -vo xv:          4.1 seconds
>> -vo vidix:       (doesn't work for me)
>> -vo gl:manyfmts  ~6.7 seconds
>> 
>> Unless I'm not understanding something right, 3.7 seconds is
>> actually pretty darned good.  According to the Mpalyer benchmark
>> outputs, I'm spending about 90% of the time in the VC module, about
>> 5% of the time in the VO module, and 5% of the time in Sys (not
>> sure what that accounts for).  So unless I'm forcing the decoder to
>> use a suboptimal format (Mpeg-2, and I'm using Planar YV12), I
>> don't think I have much room for improvement (the decoder alone
>> takes about 3.2 seconds)
> 
> Well, for me vo_vesa is actually quite slow, and xv should be
> significiantly faster... I find those values at least a bit weird...

Because I've seen discrepancies between my experience and claims made
for the speed of some VO methods (specifically, I've found VESA to be
the fastest available), I decided to run some benchmarks myself. The
results - for, again, 300 frames apiece - are below in increasing order
of speed, averaged from three runs each. (A couple of these were
included just for the heck of it.)

caca            ~3.910
gl:manyfmts     ~3.834
gl               3.812
gl2             ~3.430
x11              0.812
xvidix          ~0.662  (window displayed only solid green)
cvidix          ~0.635  (no display, 'DHA kernelhelper failed' messages)
sdl             ~0.428
xv               0.418
aa               0.408
vesa             0.002  (display appeared to have wrong aspect)
null             0.001

XVIDIX, CVIDIX and VESA work (or, at least, don't fail with no playback)
for me only when I run as root; this shouldn't skew the results AFAIK,
but is mentioned for the sake of full disclosure.

In line with what you say, the two GL forms are among the slowest,
although (contrary to what I think you imply) -vo gl:manyfmts is at
least as slow. X11 is not much faster for me than is XV, although
(amusingly enough) AAlib is about the same speed as the latter; in line
with my experience but *not* with the reports I've heard, VESA is by far
the fastest VO method, competitive even with no video output at all.

I'd like to test VESA against a few others on my brothers' systems,
since those are much slower and they (like me before I got better
hardware) have been unable to reliably avoid the "system too slow"
message with anything but VESA output; however, they likewise cannot get
VESA to work when not run as root, and I don't know their root
passwords. I'll inquire with them when they wake up/get back from
school, respectively.

-- 
       The Wanderer

Warning: Simply because I argue an issue does not mean I agree with any
side of it.

A government exists to serve its citizens, not to control them.