[FFmpeg-devel] [PATCH] avcodec/nvenc: support dynamic resolution change

[Michael Fabian 'Xaymar' Dirks]

On 08/08/2020 17:22, Linjie Fu wrote:
> On Sat, Aug 8, 2020 at 7:21 PM Timo Rothenpieler <timo at rothenpieler.org> wrote:
>>
>> On 08.08.2020 02:48, Linjie Fu wrote:
>>> On Fri, Aug 7, 2020 at 4:19 PM leozhang <leozhang at qiyi.com> wrote:
>>>>
>>>> Allow dynamic resolution change, this is useful for real time video communication application.
>>>>
>>>> Use below commands to test it,
>>>> ffmpeg -i reinit-large_420_8-to-small_420_8.h264 -noautoscale -c:v hevc_nvenc out.265 -loglevel verbose -y
>>>> ffmpeg -i reinit-large_420_8-to-small_420_8.h264 -noautoscale -c:v h264_nvenc out.264 -loglevel verbose -y
>>>>
>>>> Signed-off-by: leozhang <leozhang at qiyi.com>
>>>> ---
>>>>    libavcodec/nvenc.c | 22 +++++++++++++++++++++-
>>>>    1 file changed, 21 insertions(+), 1 deletion(-)
>>>
>>> Hi leozhang,
>>>
>>> Adding dynamic resolution encoding support is the tendency, and there
>>> are some previous discussions[1] [2]
>>> about adding support for dynamic resolution encoding.
>>>
>>> And one conclusion is that we'd prefer to cope with this in a more
>>> general way, like recreating the
>>> encoder instance instead of modifying in specific codec.
>>>
>>> - Linjie
>>
>> The problem is, that specially in the case of nvenc, re-creating the
>> encoder can take several seconds.
> 
> There is once a concern about parameter changing in the AVCodecontext
> in the last discussion:
> [1] libavcodec encoders have always assumed that the stream parameters
> are set once and never change afterwards.
> 
>> Patch generally looks okay to me, and I'm fine with applying it.
> I'm ok with this if it didn't cause the side-effects you've mentioned.
> (since before we got a decent support, better ever than never)
> 
> - Linjie
> 
> [1] https://lists.ffmpeg.org/pipermail/ffmpeg-devel/2020-February/257377.html
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request at ffmpeg.org with subject "unsubscribe".
> 

As far as I am aware, the NVENC encoder requires that both 
maxEncodeWidth and maxEncodeHeight are greater or equal to the new 
resolution you are targetting. If they aren't, the resolution change 
will fail. It is most likely safe to assume that the encoder is unsafe 
to use if that change fails.

For example in live transcoding you would have to know the possible 
maximum resolution ahead of time - which unless documented or included 
in the container information is impossible. But it might be safe to just 
set them to the resolution that started the encode session, and only 
allow switching to smaller-than-original resolutions.

Note that dynamic resolution changes should be impossible while 
maxEncodeWidth and maxEncodeHeight are at 0, according to the 
nvEncodeAPI.h file:

 > Client should allocate output buffers according to this dimension for 
dynamic resolution change. If set to 0, Encoder will not allow dynamic 
resolution change.

-- 
Sincerely | Mit freundlichen Grüßen

Michael Fabian 'Xaymar' Dirks

PS: I apologize if I made any mistake replying to this list. I haven't 
actively used mailing lists for several years now.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_0xE3F6FC54A2BACD73.asc
Type: application/pgp-keys
Size: 3164 bytes
Desc: not available
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20200924/3d4edc4f/attachment.key>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature
Type: application/pgp-signature
Size: 840 bytes
Desc: OpenPGP digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20200924/3d4edc4f/attachment.sig>