[FFmpeg-devel] [PATCH] Add and use cli options for v4l2

[Neal Gompa]

On Fri, Mar 25, 2022 at 10:40 PM <hydra3333 at gmail.com> wrote:
>
> Oops, accidentally submitted with the wrong title so not in the right thread.
> Is submitted with the right one now.
>
> A re-submit of a re-base of a formerly proposed patch to align with ffmpeg github
> commit 8ae15b565533944d042d3caf25f7262e002e8953
> as at 2022.03.26 ACST, after ffmpeg changes overlapped the formerly proposed patch.
>
> Add commandline options to v4l2_m2m_enc (h264_v4l2m2m only)
> and use those to configure options for the h264_v4l2m2m encoder.
> Uses AVOption options to filter for valid options per v4l2 spec.
> For h264 it adds spec-compliant:
> -profile <name> (high is max accepted by Raspberry Pi)
> -level <name>   (4.2 is max  accepted by Raspberry Pi)
> -rc <name>      (Bitrate mode, VBR or CBR or CQ)
> -shm <option>   (Sequence Header Mode, separate_buffer or joined_1st_frame)
> -rsh <boolean>  (Repeat Sequence Header 0(false) 1(true))
> -fsme           (Frame Skip Mode for encoder, rejected by Pi OS)
> -b:v <bps>      (Bit per second)
> -g <integer>    (pseudo GOP size, not an actual one)
> -iframe_period <integer> (the period between two I-frames)
> -qmin <integer> (Minimum quantization parameter for h264)
> -qmax <integer> (Maximum quantization parameter for h264)
>
> Patch does not address pre-existing quirks with h264_v4l2m2m,
> eg on a Raspberry Pi,
> - Bitrate mode VBR, file is reported by mediainfo as CBR
> - Bitrate mode CBR, encoder hangs and appears to
>   "lock" /dev/video11 until reboot
> - CFR input yields a VFR file reported by mediainfo (and an
>   odd framerate) whereas an equivalent libx264 commandline
>   yields expected CFR; tested on a Raspberry Pi4
> - Bitrate mode CBR, profile is limited to less than "high"
> - Bitrate mode VBR, only target bitrate option exposed to set
> - Bitrate mode CQ, is not exposed to set
>
> Notes:
> Patch arises from a desire to use ffmpeg on a Raspberry Pi (4 +).
> Fixes "--profile high" not working (required an integer).
> The Raspberry Pi OS does not expose a GOP size to set, so -g is
> used for backward compatibility with its value overriding
> the "close enough" effect of an "iframe_period" value.
> Hardware like Raspberry Pi 4 rejects some spec-compliant options
> beyond its capacity and/or not implemented by the Raspberry Pi OS.
> The Raspberry Pi OS repository for ffmpeg appears to have Repeat
> Sequence Header hard-coded as True, rather than a cli an option.
> Added more return value checking, AV_LOG_WARNING and a lot
> more AV_LOG_DEBUG code; one-time runtime cost of debug code
> during init phase is negligible.
> Intentionally left in //commented-out debug code.
>
> A working commandline using an interlaced source:
> /usr/local/bin/ffmpeg -hide_banner -nostats -v debug -i "~/Desktop/input_file_tiny.mp4" -vsync cfr -sws_flags
> lanczos+accurate_rnd+full_chroma_int+full_chroma_inp -strict experimental -filter_complex "[0:v]yadif=0:0:0,format=pix_fmts=yuv420p"
> -c:v h264_v4l2m2m -pix_fmt yuv420p -rc VBR -b:v 4000000 -qmin 10 -qmax 51 -profile:v high -level 4.2 -shm separate_buffer -rsh 0
> -g:v 25 -movflags +faststart+write_colr -an -y "./output_file_tiny_h264_VBR_g25.mp4"
>
> Signed-off-by: hydra3333 <hydra3333 at gmail.com <mailto:hydra3333 at gmail.com> >

This looks like a malformed statement?

> ---

This patch looks reasonably good, but it doesn't work on my computer
running Fedora Linux 36 on my Raspberry Pi 400 with Linux 5.17.0. Is
the necessary kernel functionality upstreamed yet? Taking a look at
the kernel code, I don't see a driver for it[1], not even in
staging[2].

[1]: https://github.com/torvalds/linux/tree/master/drivers/media/platform
[2]: https://github.com/torvalds/linux/tree/master/drivers/staging/media


-- 
真実はいつも一つ！/ Always, there's only one truth!