[x264-devel] bug in x264_validate_parameters(...) concerning the "--analyse" argument ?

[Mathieu Carnec]

Dear all,

I think I found a bug in function x264_validate_parameters(...) (in file
encoder.c) of x264.

When you put "--analyse all" in the arguments, the param.analyse.inter
becomes equal to ~0 (~ = complement to 1).

Then the function x264_validate_parameters(...) sets it to 305 using the
following lines :

    h->param.analyse.inter &=
X264_ANALYSE_PSUB16x16|X264_ANALYSE_PSUB8x8|X264_ANALYSE_BSUB16x16|X264_ANAL
YSE_I4x4|X264_ANALYSE_I8x8;

    if( !(h->param.analyse.inter & X264_ANALYSE_PSUB16x16) )
        h->param.analyse.inter &= ~X264_ANALYSE_PSUB8x8;


The bug shows up in function x264_mb_analyse_intra(...) (in file analyse.c).
In this function, I8x8 prediction is avoided while I4x4 prediction is
performed

    /* 8x8 prediction selection */
    if( flags & X264_ANALYSE_I8x8 ) ............WILL BE AVOIDED

    /* 4x4 prediction selection */
    if( flags & X264_ANALYSE_I4x4 ) ............WILL BE PERFORMED


To fix this bug, I think the lines that concern h->param.analyse.inter in
x264_validate_parameters(...) should be removed.

Indeed, initial value and parameters reading should be sufficient.

Can someone confirm that this is a bug and that the suggested modification
should fix it ?

Best regards,

Mathieu.

-----------------------------------------------------
Mathieu CARNEC
Research engineer
IVC (Images and Video Communications) team / IRCCyN lab.
Polytechnic School of the University of Nantes
La Chantrerie
rue Christian Pauc
BP 50609
44306 Nantes Cedex 3
France
Tel: +33 (0)2 40 68 32 07

-- 
This is the x264-devel mailing-list
To unsubscribe, go to: http://developers.videolan.org/lists.html