[XviD-devel] [CVS commit] 1.0 -- fixed buglet

[jmtest at voila.fr]

Hello, 

I think there is another *small*  bug when you rebuild the matrix at the end of the function. the sum is set to zero, and you forget the last element of the trellis. But this bug is not visible either...

Another point is that this function is very, very CPU expensive, so we should call it only when the input sum>1 (coz if sum==1 there is only a 1 in the matrix, the trellis doesn't work). And maybe when sum is low (2 or 3?) there's only very little gain possible (but may be I'm wrong?)

Regards,

Jean-Marc

> Message du 14/04/04 à 21h48
> De : Edouard Gomez <ed.gomez at free.fr>
> A : xvid-devel at xvid.org
> Copie à : 
> Objet : [XviD-devel] [CVS commit] 1.0 -- fixed buglet
> 
> Edouard Gomez (ed.gomez at free.fr) wrote:
> > Seems reasonable. I'll do more testing tomorrow.
> 
> You were right. But as usual we're lucky hackers, and our bugs aren't
> impacting that much :-)
> 
> I commited a fix to the 1.0 branch, head merging will done as usual at
> a later time. BS version bumped to 31, just in case. I need some
> testing from users if we want to avoid another RC.
> 
> 2004-04-14 19:45:23 GMT	                                	patch-14
> 
>     Summary:
>       Fixed small bug in trellis code.
>     Revision:
>       xvidcore--stable--1.0--patch-14
> 
>     From ed.gomez (spotted by jean marc):
>      * Trellis optimization was computing the sum |coeffs| wrongly
>        because the Compute_sum function wasn't doing zigzag scanning
>        and stopped at last non zero coeffs in linear scanning...
>        The trivial fix could have been to pass the zigzag to the compute
>        sum function so it could have done its job right.
>        But... Why computing stuff that is already known in the upper layer ?
>        I removed the compute sum function, and just pass the original sum
>        value in trellis function parameters, just in case we have to return
>        it again because trellis failed optimizing the block coeffs.
>     
>        This fix the bug and saves cycles :-) (but should not be noticeable)
> 
>     modified files:
>      src/utils/mbtransquant.c src/xvid.h
> 
> -- 
> Edouard Gomez
> _______________________________________________
> XviD-devel mailing list
> XviD-devel at xvid.org
> http://list.xvid.org/mailman/listinfo/xvid-devel
> 
------------------------------------------

Faites un voeu et puis Voila ! www.voila.fr