[MPlayer-dev-eng] Re: Compile options

[Andrew Savchenko]

On 21 Sep 2006 04:22, Carl Eugen Hoyos wrote:
> Andrew Savchenko <Bircoph <at> list.ru> writes:
>
> [...]
>
> > Results for MPEG4:
> > ==> General -O4 build:
> >
> > vc: (8.835 \pm 0.019)s
> > vo: (176.051 \pm 0.070)s
> > user: 1m(45.945 \pm 0.152)s
> >
> > ==> General -O4 build, -fno-inline-functions for dsputil_mmx.o:
> >
> > vc: (8.974 \pm 0.029)s
> > vo: (176.092 \pm 0.142)s
> > user: 1m(46.189 \pm 0.143)s
>
> MPEG-4 tests with MPlayer dev-SVN-r19920-4.0.2:
> ../mplayer test.avi -benchmark -nosound -vo xv
> -O4:
> BENCHMARKs: VC:  14.133s VO:   4.587s A:   0.000s Sys:   1.171s =  
> 19.892s BENCHMARKs: VC:  14.148s VO:   4.602s A:   0.000s Sys:  
> 1.152s =   19.902s BENCHMARKs: VC:  14.207s VO:   4.511s A:   0.000s
> Sys:   1.167s =   19.886s
>
> -O4, -fno-inline-functions for dsputil_mmx.o:
> BENCHMARKs: VC:  13.790s VO:   4.554s A:   0.000s Sys:   1.148s =  
> 19.492s BENCHMARKs: VC:  13.715s VO:   4.523s A:   0.000s Sys:  
> 1.150s =   19.388s BENCHMARKs: VC:  13.750s VO:   4.587s A:   0.000s
> Sys:   1.113s =   19.450s
>
> Would it be possible for you to repeat your tests with -vo xv or -vo
> null?

Yes, but later, not today. This takes some time, I have other things to 
do. 

> I saw your options in an earlier email now (-vo x11 -zoom -sws 
> 10 -fs), and if I use them, real time playback isn't possible for any
> of my samples,

Intresting, as I remember you have Athlon 2600+, while mine is AthlonXP 
2500+ oc 3200+. While playing trailer by link you have provided earlier 
I have more than 40% CPU idle with mine standart options, upscaling was 
done to fit 1024x768 screen. And we have not big difference in CPU 
speed while system performance (if I understood you properly) differs 
perceptibly.

> so perhaps this is the reason why our results differ  
> so much.

For vo or user time it is possible, but, please, tell me _how_ different 
VO can affect VC time?

Maybe results also depends on type of movie content (a lot of small 
details and so on). My test material except link you provided earlier 
is anime.

> Anyway, can you confirm that -finline-functions produces the
> binary-identical dsputil_mmx.o

Yes (when using -O4 -finline-functions).

> and that -fno-inline-functions reduces 
> code size by ~2/3? (For me, dsputil_mmx.o shrinks from 536128 to
> 171840.)

Yes. With mplayer-r19918-4.0.2 and unstripped object file it shrinks:
536184 -> 171880. But smallest size doesn't mean fastest execution in 
general.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 191 bytes
Desc: not available
URL: <http://lists.mplayerhq.hu/pipermail/mplayer-dev-eng/attachments/20060921/679bde41/attachment.pgp>