[FFmpeg-devel] [PATCH 4/4] avfilter/vf_colormatrix: add 10 & 12 bit depth support

[Ronald S. Bultje]

Hi,

On Sun, Mar 27, 2016 at 7:09 AM, Thomas Mundt <
loudmax-at-yahoo.de at ffmpeg.org> wrote:

> >>>Ronald S. Bultje <rsbultje at gmail.com> schrieb am Sa, 26.3.2016:
> > On Sat, Mar 26, 2016 at 10:04 AM, Thomas Mundt <loudmax-at-yahoo.de at
> ffmpeg.org> wrote:
> >
> >>>>>Kieran Kunhya <kierank at obe.tv> schrieb am Sa, 26.3.2016:
> >> >> On Fri, 25 Mar 2016 at 22:32 Thomas Mundt <loudmax-at-yahoo.de at>
> ffmpeg.org> wrote:
> >> >
> >> >> Signed-off-by: Thomas Mundt <loudmax at yahoo.de>
> >> >> ---
> >> >>  libavfilter/vf_colormatrix.c | 182
> >> >> ++++++++++++++++++++++++++++++++++++++++++-
> >> >>
> >>> >
> >> > These functions are basically the same, have you considered factoring
> the
> >> > code out?
> >> >
> >> > Kieran
> >>
> >> I thought keeping it seperate would be easier to review.
> >> But I can do that. Maybe as a subsequent patch?
> >
> >
> > I think he means templating out typelessly like h264/hevc/vp9 do. I agree
> > that would probably be nicer. Binary size is same but less duplicated
> code.
> >
> > Ronald
>
> Okay, I tried to solve this with the use of macros. The attached file is
> the result (interesting part starting at line 203).
> It works, but I´m not sure if this is what you expect.
> Also I didn´t find a way factoring out the two different av_clip.
> One could replace av_clip_uint8 by av_clip with limits 0 and 255. But this
> would have an impact on speed.
> If there is a more elegant way, could you please provide a short example?!


We don't like macros, but you're surprisingly close. Have a look at
libavcodec/bit_depth_template.c and grep for its usage in libavcodec.

You'll end up with a vf_colormatrixdsp_template.c file, which includes
bit_depth_template.c for types/clips etc., and it will then define full
functions per bit depth. This is included in vf_colormatrixdsp.c, and that
defines the actual dsp functions which do the core of what this filter does.

Since it's now in a dsp file, it'll be trivial to write x86 simd for it
(which I have also in the works).

Ronald