[FFmpeg-devel] [PATCH] pthread_frame: attempt to get frame to reduce latency

[Dai, Jianhui J]

It does not break FFmpeg output frames management logic (e.g. h264_select_output_frame), just enter that phase earlier.

Perhaps my understanding is not correct. 
In my opinions, render should be the one considering variable latency issue, instead of decoder.
The decoder should try best generating frame ASAP (MUST follow standard).

And some FFmpeg user will call libavcodec FF_THREAD_FRAME directly and handle the variable latency issue in their own render code.

Thanks,
Jianhui Dai

-----Original Message-----
From: ffmpeg-devel <ffmpeg-devel-bounces at ffmpeg.org> On Behalf Of Kieran Kunhya
Sent: Wednesday, March 11, 2020 10:41 AM
To: FFmpeg development discussions and patches <ffmpeg-devel at ffmpeg.org>
Subject: Re: [FFmpeg-devel] [PATCH] pthread_frame: attempt to get frame to reduce latency

On Wed, 11 Mar 2020 at 02:29, Dai, Jianhui J <jianhui.j.dai at intel.com>
wrote:

> Thanks, I will update the commit message.
>
> I test it with FFmpeg native sw H264 decoder.
> In previous FF_THREAD_FRAME,  the latency is constant as N ( = 
> thread_count - 1) frames.
> It won't sync thread state until no idle threads available, therefore 
> N frames are cached internal, even some frames are ready for output.
> E.g. in RTSP 30fps streaming playback, 16 frame threads (default), the 
> internal frame caching contributes 495ms(33ms x 15frame) latency.
> And the latency is the same, regardless the video resolution is 
> 640x480 or 4320x2160.
>
> In this patch, I attempt to check thread state and try best output 
> frame to reduce the constant frame caching latency.
>
> Thanks,
> Jianhui Dai
>

Unless I misunderstand, surely this outputs frames nondeterministically?
i.e you don't know that the thread which has finished is the next one that is due

Not to mention the variable latency problems.

Kieran
_______________________________________________
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".