[FFmpeg-devel] [PATCH] lavfi/af_anequalizer: remove cabs, cexp dependencies

[Ganesh Ajjanagadde]

On Mon, Dec 28, 2015 at 10:27 PM, Matt Oliver <protogonoi at gmail.com> wrote:
> On 29 December 2015 at 11:51, Ganesh Ajjanagadde <gajjanagadde at gmail.com>
> wrote:
>
>> On Mon, Dec 28, 2015 at 8:22 AM, Paul B Mahol <onemda at gmail.com> wrote:
>> > On 12/27/15, Ganesh Ajjanagadde <gajjanagadde at gmail.com> wrote:
>> >> Replaces by real arithmetic. Tested the validity of these
>> transformations
>> >> separately.
>> >>
>> >> Signed-off-by: Ganesh Ajjanagadde <gajjanagadde at gmail.com>
>> >> ---
>> >>  configure                    |  1 -
>> >>  libavfilter/af_anequalizer.c | 25 +++++++++++++++++--------
>> >>  2 files changed, 17 insertions(+), 9 deletions(-)
>> >>
>> >
>> > If those are mathematically equivalent ok.
>>
>> mathematically equivalent, tested offline, numerical differences of
>> the order of ~ 1e-15 or so, which is perfectly fine IMHO for double
>> precision especially since it is a priori not clear which is more
>> precise. Pushed, thanks.
>
>
> Wouldn't it also be a good idea to remove the COMPLEX_FUNCS list from
> configure as now neither cabs or cexp are used so configure can be
> simplified by removing the checks for them.

I asked Paul about this (when I removed them from some other filter).
That time he said he wanted them for another filter (anequalizer)
since the expressions were complicated. Judging from this commit, I
think that was a fair assessment.

In short, this is really a question for Paul (or others who use
complex numbers): if there are plans on doing something with complex
numbers that is not trivial to rewrite, it may be helpful not to
remove as you suggest. If not, then it is perhaps ok.

Personally, over the long term, I think there will be further
instances of complex number usage. If we remove it now, I suspect it
will get added back in at some point. As such, I won't write a patch
for this; really upto others especially Paul.

Maybe long run we can create an av_complex or something like that. A
few problems are (among possibly many others):
1. Syntactic sugar for *, / , +, - will be lost, and one will need to
create something like cmult, cdiv, and call at an increased verbosity
cost. In C++, these basic operators can be overloaded (but then again
stdc++ has a complex header); AFAIK there is no such mechanism that we
can use in FFmpeg's "least common denominator" C.
2. Natural promotion rules will be lost, e.g suppose one passes a
double to a complex function, say while multiplying a double complex
by a double as in anequalizer. With proper complex number support,
promotion rules kick in and everything works. With av_complex, this
does not work, and explicit setting of imaginary part to 0 will be
needed.

> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel