[XviD-devel] [CVS commits] BVOP ref bidirectional reconstruction rounding error for both head and 1.0 branch

Edouard Gomez ed.gomez at free.fr
Sat Jul 24 13:55:11 CEST 2004


Here goes the fix on the 1.0 branch:

--------------------8<-------------------------------------------
2004-07-24 11:33:57 GMT	patch-50

    Summary:
      BVOP direct/interpolated ref block rounding fix.
    Revision:
      xvidcore--stable--1.0--patch-50

    From ed.gomez:
     * BVOP direct/interpolated ref block rounding fix.
       It's been using rounding=1 for averaging stage since ever.
       The standard says it's rounding=0.
    
       See standard clause 7.6.9.4 for explicit code and Section 6.3.5
       that says "rounding=0" in bframes as they don't set the
       vop_rounding_type in VOP header. Both sections match, xvid was
       wrong

    modified files:
     src/decoder.c
--------------------8<-------------------------------------------

And the merge on head branch:

--------------------8<-------------------------------------------
2004-07-24 11:39:57 GMT	Edouard Gomez <ed.gomez at free.fr>	patch-56

    Summary:
      Important bugfix from stable
    Revision:
      xvidcore--head--0.0--patch-56

    Important bugfix from stable
    
    Patches applied:
    
     * ed.gomez at free.fr--2004-1/xvidcore--stable--1.0--patch-50
       BVOP direct/interpolated ref block rounding fix.
    

    modified files:
     src/decoder.c

    new patches:
     ed.gomez at free.fr--2004-1/xvidcore--stable--1.0--patch-50
--------------------8<-------------------------------------------

New "faster bvops" patch updated at:
http://ed.gomez.free.fr/vrac/xvid-patches/faster-bvop-interpolation.diff

Now 1.0.1+bugfix, 1.1 C backend, and 1.1 skal asm code doe
match perfectly during decoding.

sysLin did look at the way this rounding could impact encoder
code, and he found no errors so far as encoder was using
rounding=0 for bvops. So at least our encoder was producing
good streams, only the decoder was borked.

-- 
Edouard Gomez


More information about the XviD-devel mailing list