[FFmpeg-devel] [PATCH] Add AAC FATE encode tests.

[Reimar Döffinger]

On Fri, Apr 06, 2012 at 03:43:24PM -0400, Derek Buitenhuis wrote:
> On 06/04/2012 3:40 PM, Reimar Döffinger wrote:
> > Signed-off-by: Reimar Döffinger <Reimar.Doeffinger at gmx.de>
> > ---
> >  tests/Makefile     |    1 +
> >  tests/fate/aac.mak |   21 +++++++++++++++++++--
> >  2 files changed, 20 insertions(+), 2 deletions(-)

(Note: like the other test it needs the high-bitrate hang fix for the
encoder)

> Can you explain where the size tolerances come from? (I'm a bit uninformed.)

What exactly do you want to know? The tiny_psnr tool prints as last
statistics two size values, and the size tolerance is the difference
between these.
If you want to know why there is a difference in size between decoded
file and reference that is generally due to encoder/decoder delay
creating addition 0-samples in front, the encoder rounding the size
to whatever block size it needs and here also the "shift" value
influencing the "size" value that tiny_psnr calculates - which may
or may not make sense.
Large values tend to mean that the encoder/decoder aren't particularly
good, since I think for most formats that issue can be minimized.