[FFmpeg-devel] [PATCH] swscale-example: Don't test fast_bilinear.

Michael Niedermayer michaelni
Thu Aug 13 18:18:28 CEST 2009

On Thu, Aug 13, 2009 at 10:29:58AM -0300, Ramiro Polla wrote:
> On Wed, Aug 12, 2009 at 10:13 PM, Michael Niedermayer<michaelni at gmx.at> wrote:
> > On Wed, Aug 12, 2009 at 09:47:06PM -0300, Ramiro Polla wrote:
> >> Hi,
> >>
> >> As michael suggested on a thread in -soc, fast_bilinear should not be
> >> considered bitexact. This patch removes it from swscale-example.
> >>
> >> Currently fast_bilinear reads past the end and its output is not
> >> portable between architectures. Most of my attempts to make it more
> >> bitexact made it not fast.
> >
> > i think it should still be tested, one just must keep in mind that the
> > results are only compareable within C-C / MMX-MMX / MMX2-MMX2 ...
> It's not only C-C / MMX-MMX... It always overreads one byte so the
> output varies randomly. Fixing this would make a tiny little bit
> slower (and you've already said no slower code for fast_bilinear).

use a little fantasy ...
like just not comparing the rightmost pixels or making sure these extra
input values have a definit value like 123 

> And for regression testing swscale, swscale-example currently
> generates 6.2Mb output. We can probably trim some data by printing
> less chars, but it will still be in the many Mbs range. How do you
> suggest we compare against a reference? bzip the output (should cut it
> down to about ~250kb)? Or some other method besides swscale-example?

first and foremost i was thinking about a developer changing sws to
run it before and after a change (set), that way the size is not very
critical ...

for true regression tests with public checksums the fast bilinear code
would of course have to be disabled and we would need to make the output
much more compact, maybe a table with rows as input format and columns
as output and the actual entries the SSE values

Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

Opposition brings concord. Out of discord comes the fairest harmony.
-- Heraclitus
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/attachments/20090813/d3ac02af/attachment.pgp>

More information about the ffmpeg-devel mailing list