[FFmpeg-devel] [PATCH] lavfi: remove mp=denoise3d filter.

Loren Merritt lorenm at u.washington.edu
Tue Dec 4 20:02:39 CET 2012


On Tue, 4 Dec 2012, Reimar Döffinger wrote:
> On 4 Dec 2012, at 00:31, Stefano Sabatini <stefasab at gmail.com> wrote:
>> On date Monday 2012-12-03 22:24:40 +0100, Clément Bÿÿsch encoded:
>>
>>> hqdn3d was meant to be a more efficient denoise3d filter, is supported
>>> natively, and is faster on my computer thanks to the ASM optimisations.
>>> ---
>>> doc/filters.texi                       |   1 -
>>> libavfilter/Makefile                   |   1 -
>>> libavfilter/libmpcodecs/vf_denoise3d.c | 268 ---------------------------------
>>> libavfilter/version.h                  |   2 +-
>>> libavfilter/vf_mp.c                    |   2 -
>>> 5 files changed, 1 insertion(+), 273 deletions(-)
>>> delete mode 100644 libavfilter/libmpcodecs/vf_denoise3d.c
>>
>> I'd like a reply from some MPlayer devs/users. Is there any advantage
>> in having or using this filter, as opposed to just keep hqdn3d?
>>
>> Possible advantages: it may be simpler, faster (even if delivering
>> less quality)? If there are no clear advantages, then sure we should
>> remove it.
>
> I'm not sure if it's just a matter of configuring hqdn3d to be faster, but I expect it should be a good bit faster on architectures that have no asm optimization.
> I didn't actually benchmark it though.

I optimized the C version too. hqdn3d minus asm is still faster than
denoise3d on sandybridge.
If someone wanted to port those optimizations then denoise3d might be
improved too, but failing that it's obsolete.

--Loren Merritt


More information about the ffmpeg-devel mailing list