[FFmpeg-devel] [PATCH] avcodec/utvideoenc: switch to planar RGB formats

Paul B Mahol onemda at gmail.com
Sun Dec 31 16:08:16 EET 2017


On 12/31/17, Carl Eugen Hoyos <ceffmpeg at gmail.com> wrote:
> 2017-12-31 14:49 GMT+01:00 Paul B Mahol <onemda at gmail.com>:
>> On 12/31/17, Carl Eugen Hoyos <ceffmpeg at gmail.com> wrote:
>>> 2017-12-31 10:48 GMT+01:00 Paul B Mahol <onemda at gmail.com>:
>>>> Signed-off-by: Paul B Mahol <onemda at gmail.com>
>>>> ---
>>>>  libavcodec/utvideoenc.c               |  47 +++++++++-------
>>>>  tests/ref/fate/utvideoenc_rgb_left    | 100
>>>> +++++++++++++++++-----------------
>>>>  tests/ref/fate/utvideoenc_rgb_median  | 100
>>>> +++++++++++++++++-----------------
>>>>  tests/ref/fate/utvideoenc_rgb_none    | 100
>>>> +++++++++++++++++-----------------
>>>>  tests/ref/fate/utvideoenc_rgba_left   | 100
>>>> +++++++++++++++++-----------------
>>>>  tests/ref/fate/utvideoenc_rgba_median | 100
>>>> +++++++++++++++++-----------------
>>>>  tests/ref/fate/utvideoenc_rgba_none   | 100
>>>> +++++++++++++++++-----------------
>>>>  7 files changed, 327 insertions(+), 320 deletions(-)
>>>
>>> Is there a speed impact?
>>> (Or actually: How much faster is gbr encoding, how much
>>> slower rgb encoding?)
>>
>> Very very fast, very very slow.
>
> This is not helpful;-(
> Is it so unlikely that the patch has small gain for gbr
> (theoretically compensated by existing fast conversion
> from gbr to rgb) but large impact for rgb (with slow
> conversion from rgb into gbr)?

No.

>
>>>> diff --git a/tests/ref/fate/utvideoenc_rgb_left
>>>> b/tests/ref/fate/utvideoenc_rgb_left
>>>> index a1d200096a..1ee7c58564 100644
>>>> --- a/tests/ref/fate/utvideoenc_rgb_left
>>>> +++ b/tests/ref/fate/utvideoenc_rgb_left
>>>
>>> Why do they change?
>>> Do I understand correctly that the files get bigger (~5%)?
>>> If yes, wouldn't that indicate that the patch is not a good idea?
>>
>> Its because of different scaling path. Have nothing to do with
>> good or bad idea.
>
> Thank you, imo this indicates the utvideo rgb tests
> should be fixed to use rgb input.

Patch welcome.


More information about the ffmpeg-devel mailing list