Okay... so finally I got another hand on my side. : )
Another problem I found in XviD function interpolate8x8_quarterpel as
described in
the beginning of this thread. Did anyone verify this ?
--
Jeanson Hung
BU2, ADMtek Inc.
Phone: 03-5788879-521
Christoph
Lampert 收件人： xvid-devel@xvid.org
<chl@math.uni-b 副本抄送：
onn.de> 主旨： Re: [XviD-devel] Quarterpel problem.
寄件人：
xvid-devel-admi
n@xvid.org
2002/10/08
04:50 PM
請回信 給
xvid-devel
On Tue, 8 Oct 2002 jeansonh@admtek.com.tw wrote:
> Hi,
>
> In standard 7.6.9.5.3, it did say that. But it's describing the DIRECT
> mode, isn't it?
> Anyone has MomuSys reference you may look at function
B_MotionCompensateMB.
> It does InterpolateBlock with 16x16 blocks... As the standard section
> 7.6.9.2 ~ 7.6.9.4,
> the luminance MC are all in 16x16 blocks.
Yes, of course you are right. And since qpel-sections 7.6.2.2 defines
QPel for block of arbitrary size MxN, it's completely logical that it's
possible to use it for 16x16 block, too, no only for 8x8.
So maybe this is bug in DivX5 :-(
Now what? For decoding we could detect DivX5.0x (x<=2) and add a faulty
"DivX5 compatability mode", but for encoding? There's not much we can
do, I guess... If DivXnetworks doesn't know already, we should tell them
to bugfix their encoder in time, before QPel hardware are on the market.
gruel