[FFmpeg-devel] [PATCH] swscale: avoid overread in planar2x

[Michael Niedermayer]

On Sun, Sep 12, 2010 at 02:45:51PM -0300, Ramiro Polla wrote:
> On Sun, Sep 12, 2010 at 11:33 AM, Michael Niedermayer <michaelni at gmx.at> wrote:
> > On Sun, Sep 12, 2010 at 11:16:53AM -0300, Ramiro Polla wrote:
> >> On Sun, Sep 12, 2010 at 10:43 AM, Michael Niedermayer <michaelni at gmx.at> wrote:
> >> > On Sun, Sep 12, 2010 at 10:00:07AM -0300, Ramiro Polla wrote:
> >> >> On Sun, Sep 12, 2010 at 7:44 AM, Michael Niedermayer <michaelni at gmx.at> wrote:
> >> [...]
> >> >> > after a second look i see that the C code overrides the first pixel and thus
> >> ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> >> >> > the output should not be affected, still my suggestion though seems nicer than
> >> >> > requireing all buffers to have a byte allocated before them
> >> >>
> >> >> Thanks for the suggestion. Patch attached. No speed difference measured.
> >> >
> >> >> ?rgb2rgb_template.c | ? ?4 +++-
> >> >> ?1 file changed, 3 insertions(+), 1 deletion(-)
> >> >> 9c5b0f0bc55c99bcfbda486bff43d26c34cb96f5 ?planar2x_avoid_overread_2.diff
> >> >> Index: rgb2rgb_template.c
> >> >> ===================================================================
> >> >> --- rgb2rgb_template.c ? ? ? ?(revision 32166)
> >> >> +++ rgb2rgb_template.c ? ? ? ?(working copy)
> >> >> @@ -1773,12 +1773,13 @@ static inline void RENAME(planar2x)(const uint8_t
> >> >> ? ? ? ? ?const x86_reg mmxSize= srcWidth&~15;
> >> >> ? ? ? ? ?__asm__ volatile(
> >> >> ? ? ? ? ? ? ?"mov ? ? ? ? ? %4, %%"REG_a" ? ? ? ? ? ?\n\t"
> >> >> + ? ? ? ? ? ?"movq ? ? ? ? (%0, %%"REG_a"), %%mm4 ? ?\n\t"
> >> >> + ? ? ? ? ? ?"psllq ? ? ? ? ? ? ? ? ? ? $8, %%mm4 ? ?\n\t"
> >> >
> >> > the first value is then 0 which would lead to a darker border pixel
> >>
> >> Which is overwritten by the C code. Or did I miss something?
> >
> > dont you want to beat the c code speedwise?
> 
> I didn't at first, but thanks for the push =). Patch attached.
> 
> before:
> 63750690 dezicycles in yuv410p 2048x2048 --> yuv420p 2048x2048, 2048
> runs, 0 skips
> after:
> 62768947 dezicycles in yuv410p 2048x2048 --> yuv420p 2048x2048, 2048
> runs, 0 skips

>  rgb2rgb.c          |    1 +
>  rgb2rgb_template.c |   17 ++++++++++++++---
>  2 files changed, 15 insertions(+), 3 deletions(-)
> d4001ef2ab1fd368d66ea7f97223152d36dc6eed  planar2x_avoid_overread_3.diff

ok

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

Good people do not need laws to tell them to act responsibly, while bad
people will find a way around the laws. -- 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/20100913/fd21247d/attachment.pgp>