[Ffmpeg-devel] benchmark of different CABAC routines

Guillaume POIRIER poirierg
Tue Oct 10 13:35:58 CEST 2006


Hi,

On 10/10/06, Guillaume Poirier <gpoirier at mplayerhq.hu> wrote:

> With svn-r6623
>
> On P-M (2nd generation)
> vanilla:
> 3899 dezicycles in decode_residual, 2095922 runs, 1230 skipsbits/s
> dup=0 drop=0,
>
> with asm routine of renorm_cabac_decoder_once which has cmov in it
> 3897 dezicycles in decode_residual, 2095978 runs, 1174 skipsbits/s
> dup=0 drop=0
>
>
> with CMOV_IS_FAST
> 3825 dezicycles in decode_residual, 2096057 runs, 1095 skipsbits/s
> dup=0 drop=0
>
> with CMOV_IS_FAST + asm routine of renorm_cabac_decoder_once which has
> cmov in it
> 3807 dezicycles in decode_residual, 2096022 runs, 1130 skipsbits/s
> dup=0 drop=0
>
> So on P-M (which is a P3 variant), the winner is CMOV_IS_FAST + asm
> routine of renorm_cabac_decoder_once which has cmov in it (2.36% faster)

I forgot to mention which sample I used:
http://samples.mplayerhq.hu/V-codecs/AVC1/NeroRecode_STD_AVC_(ateme1022).mp4

Now on AMD-64 (32 bits mode), with same sample

vanilla:
3695 dezicycles in decode_residual, 2095058 runs, 2094 skips

asm routine of renorm_cabac_decoder_once which has cmov in it
3695 dezicycles in decode_residual, 2094994 runs, 2158 skips

CMOV_IS_FAST
3616 dezicycles in decode_residual, 2095104 runs, 2048 skips

CMOV_IS_FAST + asm routine of renorm_cabac_decoder_once which has cmov in it
3616 dezicycles in decode_residual, 2095091 runs, 2061 skips

Guillaume
-- 
With DADVSI (http://en.wikipedia.org/wiki/DADVSI), France finally has
a lead on USA on selling out individuals right to corporations!
Vive la France!




More information about the ffmpeg-devel mailing list