[XviD-devel] Slice rendering : 2 edition

[Alban Bedel]

Hi peter ross,

on Wed, 10 Jul 2002 22:29:45 +1000 you wrote:

> >From: Alban Bedel <albeu@free.fr>
> >Reply-To: xvid-devel@xvid.org
> >To: Xvid Dev <xvid-devel@xvid.org>
> >Subject: [XviD-devel] Slice rendering : 2 edition
> >Date: Wed, 10 Jul 2002 02:13:16 +0200
> >
> >
> >Hi all,
> >
> >I redone my patch. The main the change is to move the copy function to 
> >image.c as
> >requested. I also added the possiblity to render at the MB level. Here 
> >rendering
> >at the MB a level is a little bit slower (~1 sec more to decode 2500 frame) 
> >but
> >ymmv. As usual comments are awaited with impatience ;)
> >	Albeu
> ><< slices4 >>
> 
> i dont have a unix box to test test this on (well i do, but its 486).
> just looking at the .patch, my comments are:
> 
> - why do you need stride copy?
> simply call output_mb() at the end of the decoder_mbinter() and 
> decoder_mbintra() functions.
Not 100% sure but I think that wich method (output_mb or or output_slice) will be
the fastest depend of the arch/cpu. That's why I allowed the 2.

> - output_mb performance could be improved by using xvid's transfer8x8_copy 
> funptr instead of the for-memcpy-loop's. we dont have a 
> transfer16x16_mmx/sse2, yet.
I would have use it if that was possible but these functions assume that
input stride == output stride wich is not always the case here.

> this patch receives my blessing.
> 
> OT: btw, is anyone maintaining a win32 mplayer build?
A Cygwin port is maintained but not fully ready afaik. There is only sdl as video output
and no support for ds/vfw codecs but it work. Any patch to improve it is welcome ;)
	Albeu