[FFmpeg-devel] [PATCH] VC-1 MMX DSP functions

[Aurelien Jacobs]

Christophe GISQUET wrote:

> Michael Niedermayer a ?crit :
> > the code which is overall (whole decoder) fastest
> > and for cases where 2 are indistingishable the simpler one
> 
> Sorry for the delay in replying but it was somewhat worth it: testing on
> a P4 showed that at least one optimization was in fact degrading
> performance (special case in vc1_put_shift2_mmx when stride == offset).
> 
> Therefore, final (as far as I see) patch attached.
> 
> Summary:
> MMX version for VC-1 subpel motion compensation functions. 30% faster
> decoding.

I've just tested this patch on a HD-DVD sample.
First, it works fine on x86-64... great.

Now the bench on my Athlon64 (best of 3 runs):

$ time mplayer UNILOGO.EVO -nosound -benchmark -vo null -quiet

unpatched:

BENCHMARKs: VC:  26.387s VO:   0.002s A:   0.000s Sys:   0.289s =   26.678s
BENCHMARK%: VC: 98.9100% VO:  0.0067% A:  0.0000% Sys:  1.0833% = 100.0000%
real	0m26.722s
user	0m26.526s
sys	0m0.188s

patched:

BENCHMARKs: VC:  17.850s VO:   0.002s A:   0.000s Sys:   0.291s =   18.143s
BENCHMARK%: VC: 98.3881% VO:  0.0094% A:  0.0000% Sys:  1.6025% = 100.0000%
real	0m18.190s
user	0m18.053s
sys	0m0.132s

Very nice optimization !

Aurel