[MPlayer-dev-eng] Fw: bug? which leads to vo/encoder reinit during playback / transcode

[Bohdan Horst]

On Fri, Aug 13, 2004 at 03:57:58PM -0500, Zoltan Hidvegi wrote:
> > Well, some time ago only libmpeg2 was suggested for any inverse telecine
> > filters, since ffmpeg wasn't exporting some flags that help the
> > proccess. I wonder whether it changed now.
> 
> The change to ffmpeg12 to set the field flags has been added several
> months ago.  In my experience ffmpeg12 is now much more reliable than
> libmpeg2, especially over corrupted mpeg streams (happens in broadcast
> a lot).  Actually, is there any reason why libmpeg2 is still the
> default decoder for mpeg?

it WAS faster before upgrade to latest libmpeg2 :)


my small bechmarks (mailed to mplayer-users a few days ago):


VIDEO:  MPEG1  352x288  (aspect 8)  25,000 fps  2400,0 kbps (300,0 kbyte/s)
FreeBSD 5.2.1
gcc version 3.3.3 [FreeBSD] 20031106
CPU: Intel Celeron 2/Pentium III Coppermine,Geyserville 498.8 MHz (Family: 6, Stepping: 1)
Detected cache-line size is 32 bytes
CPUflags:  MMX: 1 MMX2: 1 3DNow: 0 3DNow2: 0 SSE: 1 SSE2: 0
Compiled for x86 CPU with extensions: MMX MMX2 SSE


old libmpeg
BENCHMARKs: VC:  24,912s VO:  15,056s A:   0,000s Sys:   8,018s = 47,987s
BENCHMARKs: VC:  25,021s VO:  14,963s A:   0,000s Sys:   8,007s = 47,991s
BENCHMARKs: VC:  25,041s VO:  15,077s A:   0,000s Sys:   7,945s = 48,063s

new libmpeg
BENCHMARKs: VC:  28,754s VO:  15,502s A:   0,000s Sys:   8,018s = 52,275s
BENCHMARKs: VC:  28,687s VO:  15,518s A:   0,000s Sys:   8,022s = 52,227s
BENCHMARKs: VC:  28,704s VO:  15,534s A:   0,000s Sys:   7,954s = 52,192s

ffmpeg
BENCHMARKs: VC:  27,985s VO:  13,668s A:   0,000s Sys:   8,074s = 49,727s
BENCHMARKs: VC:  28,332s VO:  13,535s A:   0,000s Sys:   7,960s = 49,827s
BENCHMARKs: VC:  28,242s VO:  13,460s A:   0,000s Sys:   8,048s = 49,750s

regards

-- 
Bohdan 'Nexus' Horst