[FFmpeg-devel] [RFC] How to fix DR+lavfi+vflip crash

[Michael Niedermayer]

On Mon, Dec 27, 2010 at 10:43:32PM -0500, Ronald S. Bultje wrote:
> Hi,
> 
> On Mon, Dec 27, 2010 at 10:41 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> > On Mon, Dec 27, 2010 at 11:39:53PM +0100, Stefano Sabatini wrote:
> >> On date Monday 2010-12-27 20:14:00 +0100, Michael Niedermayer encoded:
> >> > On Mon, Dec 27, 2010 at 07:07:53PM +0100, Stefano Sabatini wrote:
> >> > > On date Monday 2010-12-27 18:14:26 +0100, Stefano Sabatini encoded:
> >> > > > On date Monday 2010-12-27 16:13:14 +0100, Michael Niedermayer encoded:
> >> > > [...]
> >> > > > > applied (without typo)
> >> > > >
> >> > > > Unfortunately can't work with pad, I was too quick to push the button.
> >> > >
> >> > > Check attached, works for me, but since I've been confused tons of
> >> > > times by this problem I prefer to get another look by someone else.
> >> >
> >> > pad crashes in put_pixels16_sse2() for me on its own and ironically not with
> >> > vflip but there are vissual artifacts on the left side
> >>
> >> put_pixels16_sse2() is assuming some kind of 16-bit alignment for the
> >> value in pixels, it doesn't crash when the y pad value is a multiple
> >> of 16.
> >
> > do you plan to put this in the documentation? ;)
> >
> >
> >>
> >> > Ill look into this once you fixed pad working without vflip and if its then
> >> > still not working with vflip
> >>
> >> Uh what's this a sort of bargain? The problems look completely
> >> unrelated.
> >
> > It doesnt crash for me with vflip, just without :)
> > and you provide no gdb output so theres not much i can do about vflip+pad.
> > I do need a reproduceable testcase & gdb output first.
> > I can fix pad and alignment if you dont know how to do it ...
> 
> Can we please force 16-byte alignment on buffers allocated within lav*?

We already have that except when pad/crop is used.
for them 16byte alignment means memcpy if their parameters arenr multiple of
16
we should be able to handle this easily with adding some flag in line of
negative linesizes thus avoiding the copy when the next filter doesnt need
alignment.
But its not the only possible solution ...

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Those who are too smart to engage in politics are punished by being
governed by those who are dumber. -- Plato 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20101228/cfcfc7fc/attachment.pgp>