[FFmpeg-devel] [PATCH 2/3] simple_idct12: align C and x86

Christophe Gisquet christophe.gisquet at gmail.com
Wed Oct 14 08:46:22 CEST 2015


2015-10-14 0:04 GMT+02:00 Michael Niedermayer <michael at niedermayer.cc>:
> the ome and syserr values worsen by this

I have mixed feelings about this, too.

On the one hand, omse in any case says anyway none of those idcts are
accurate enough to some sense (inter error propagation? / per the mpeg
specs), as there will be issues. Worse than that, this metric is
unable to catch overflows: the 14,17 shift produces omse values
clearly better (a third of the current ones), but with overflows...
Meanwhile, the 2 12bpp samples I tested are actually bitexact to

On the other hand, being more "error-prone" for the sake of
bitexactness is not so nice.

> iam not objecting to this if thats what people want, just want to
> make sure its not missed

There is not a lot of actual users for those (DNxHR and rare jpg?), so
I'm not sure there is enough people that matters to draw a
statistically significant conclusion. :D

But I'd guess they probably prefer safety over speed.

> also IIUC this is just to make C and x86 match, so it could just be
> skiped with no ill effects except that tnen x86 and C would not be
> bitexact matches ?

Yes, and the 12bpp.jpg fate should not be added then, because it'd
test just the C idct.


More information about the ffmpeg-devel mailing list