x264 bug or feature ?

[audyovydeo]

Hello Dark Shikari


I have devised a quick set of tests which I run against new releases of bobor's builds (as that's what I use), mainly to have a quick & dirty measure of speed improvements of a build over the previous.
For one of the tests I also record ssim and kbps, as a further check.
Proverbial word-saving picture :

http://mapage.noos.fr/manamba/x264/x264_speed_tests.jpg

I notice that ssim and bitrate since releases >851 have changed. Is this a natural side-effect of the latest speed improvements, or a symptom of something borked ?
I don't normally expect these two metrics to remain constant as x264 is improved, it's merely that I've become used to them being constant across this range of versions.

Note that in 815 and 816 something was borked, and was corrected.

Also note that --aq-mode = 0 to allow comparison with <806 releases.

Thanks for your thoughts.

cheers
audyovydeo

[Dark Shikari]

The main change from 859->869 there is probably the Noboru patch for RD.

Not sure about 851->859.

[TheRyuu]

Are you using --rdcmp ssd with the latest version since it contains the psy RDO patch which butchers SSIM/PSNR. And since --rdcmp psy is on by default, this may cause your problems?

And don't use a number like that to judge how well something is. Visually look at it to truly compare the difference.

[Dark Shikari]

Are you using --rdcmp ssd with the latest version since it contains the psy RDO patch which butchers SSIM/PSNR. And since --rdcmp psy is on by default, this may cause your problems?

And don't use a number like that to judge how well something is. Visually look at it to truly compare the difference.If he was using psy RDO there would be a much much much larger difference than the one he is experiencing ;)

[audyovydeo]

Right

echo .................... encoding TEST1 ..........
%%~z --crf 25 -r 3 --me hex -b 16 -m 5 -t 1 -A all -8 --direct temporal --b-pyramid -w --bime --aq-mode 0 --no-fast-pskip --no-dct-decimate --no-psnr --threads auto --progress -o NUL real.avs
echo.
echo .................... encoding TEST2 ..........
%%~z --crf 25 -r 4 --me umh -b 16 -m 5 -t 1 -A all -8 --direct temporal --b-pyramid -w --bime --aq-mode 0 --no-fast-pskip --no-dct-decimate --no-psnr --threads auto --progress -o NUL real.avs
echo.
echo .................... encoding TEST3 ..........
%%~z --crf 25 -r 1 --me esa -b 16 -m 5 -t 1 -A all -8 --direct temporal --b-pyramid -w --bime --aq-mode 0 --no-fast-pskip --no-dct-decimate --no-psnr --threads auto --progress -o NUL real.avs

My main point is to measure speed improvements.
I know the ssim/visual debate off by heart, but I still rely on ssim as a quick compare of different encodes of same source.

Within this thread's context, my question is in my first post :
given the data since r745, is the change of ssim and bitrate a symptom of something ?

cheers
a/v