[Mplayer-cvslog] CVS: main/libvo osd.c,1.8,1.9

[Nick Kurshev]

Hello, Michael!

On Sun, 11 Nov 2001 17:26:59 +0100 you wrote:

> On Sunday 11 November 2001 17:15, Nick Kurshev wrote:
> > Hello, Michael!
> >
> > On Sun, 11 Nov 2001 17:00:55 +0100 you wrote:
> > > Hi
> > >
> > > On Sunday 11 November 2001 15:42, Nick Kurshev wrote:
> > > > Update of /cvsroot/mplayer/main/libvo
> > > > In directory mplayer:/var/tmp.root/cvs-serv4910/main/libvo
> > > >
> > > > Modified Files:
> > > > 	osd.c
> > > > Log Message:
> > > > Enable MMX stuff
> > > > I don't know why Michael Niedermayer disabled it but:
> > >
> > > because its slower (on my comp at least, but perhaps my fonts are too
> > > large?)
> > >
> > > > my benchmarks of 25 frames for TIMER+SUBTITLE (measured by RDTSC):
> > > > Non-MMX stuff:
> > > >     total=306142159
> > > > MMX stuff:
> > > >     total=159534150
> > >
> > > ./mplayer -vo x11 -nosound
> > > ~/Lorenna_McKennit-Mummers_Dance-Mononoke_Hime-gabucino.avi
> > > Non-MMX
> > > total=432128286
> > > MMX
> > > total=529391629
> >
> > It's strange. Btw - I used my "own" font - it is on mplayer's download
> > page. (Arial 24)
> yes, very strange
> btw mmx is faster for me if i put "if(srca[x]!=0 && srca[x+1]!=0)" in front 
> of the asm block
> MMX
> 128814484
> non-mmx
> 139757888
> 
> perhaps my p3 can handle the if() faster, that would explain it perhaps, do u 
> know how the k7 or whatever cpu u have does branch prediction?
> the intel chips can predict simple repeating patterns since the ppro and the 
> transparent areas appear pretty predictable to me
> perhaps gcc did just generate bad code for u, (cmov?)
> 
According to AMD's manual 2207D (Athlon code optimization) page 49 there is only
suggestion to use CMOV instead of Jxx opcode if possible or such tricks:

CMP A, B
ADC X, Y

> Michael
> _______________________________________________
> Mplayer-cvslog mailing list
> Mplayer-cvslog at mplayerhq.hu
> http://mplayerhq.hu/mailman/listinfo/mplayer-cvslog
> 


Best regards! Nick