[FFmpeg-devel] [RFC]Don't use round when encoding png
michaelni at gmx.at
Sun Mar 1 22:18:03 CET 2015
On Sun, Mar 01, 2015 at 09:34:46PM +0100, wm4 wrote:
> On Sun, 1 Mar 2015 20:34:57 +0100
> Michael Niedermayer <michaelni at gmx.at> wrote:
> > On Sun, Mar 01, 2015 at 07:57:56PM +0100, wm4 wrote:
> > > On Sun, 1 Mar 2015 19:48:20 +0100
> > > Nicolas George <george at nsup.org> wrote:
> > >
> > > > Le primidi 11 ventôse, an CCXXIII, wm4 a écrit :
> > > > > I prefer the color of fuck broken platforms nobody uses.
> > > >
> > > > Well, obviously someone uses a platform where it makes a difference,
> > > > otherwise nobody would have noticed.
> > > >
> > > > Regards,
> > > >
> > >
> > > If they really care, get them to fix the platform (as long as our use
> > > of the function is actually guaranteed by the standard).
> > the reason why it fails is probably that the code in pngenc.c tries
> > to write a double precission value by using AV_W* macros
> > this is a bad idea and likely not what the author intended
> (Maybe then it shouldn't have been merged without review and without
> presence on the mailing list?)
I reviewed it, and i missed that bug.
So has everyone else on the ML who looked at carls fix
> I'm pretty sure the author expected the values to be correctly rounded,
> and that the AV_WB32() macro casts its float argument to an integer. Is
> that not the case?
the macro does not cast, and adding such casts could cause a slowdown
as it adds a operation which compilers might fail to remove
what do you suggest to solve this bug ?
its easy to solve technically the hard part is finding a solution
Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB
While the State exists there can be no freedom; when there is freedom there
will be no State. -- Vladimir Lenin
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Size: 181 bytes
Desc: Digital signature
More information about the ffmpeg-devel