lkml.org 
[lkml]   [2020]   [Jul]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 1/4] media: v4l2-ctrl: Add frame-skip std encoder control
From
Date
Le mercredi 15 juillet 2020 à 18:42 +0300, Stanimir Varbanov a écrit :
> Hi Nicolas,
>
> On 7/7/20 11:53 PM, Nicolas Dufresne wrote:
> > Le dimanche 05 juillet 2020 à 15:11 +0300, Stanimir Varbanov a écrit :
> > > Adds encoders standard v4l2 control for frame-skip. The control
> > > is a copy of a custom encoder control so that other v4l2 encoder
> > > drivers can use it.
> > >
> > > Signed-off-by: Stanimir Varbanov <stanimir.varbanov@linaro.org>
> > > ---
> > > .../media/v4l/ext-ctrls-codec.rst | 32 +++++++++++++++++++
> > > drivers/media/v4l2-core/v4l2-ctrls.c | 10 ++++++
> > > include/uapi/linux/v4l2-controls.h | 6 ++++
> > > 3 files changed, 48 insertions(+)
> > >
> > > diff --git a/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst b/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
> > > index d0d506a444b1..a8b4c0b40747 100644
> > > --- a/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
> > > +++ b/Documentation/userspace-api/media/v4l/ext-ctrls-codec.rst
> > > @@ -592,6 +592,38 @@ enum v4l2_mpeg_video_bitrate_mode -
> > > the average video bitrate. It is ignored if the video bitrate mode
> > > is set to constant bitrate.
> > >
> > > +``V4L2_CID_MPEG_VIDEO_FRAME_SKIP_MODE (enum)``
> > > +
> > > +enum v4l2_mpeg_video_frame_skip_mode -
> > > + Indicates in what conditions the encoder should skip frames. If
> > > + encoding a frame would cause the encoded stream to be larger then a
> > > + chosen data limit then the frame will be skipped. Possible values
> > > + are:
> >
> > I have nothing against this API, in fact it's really nice to generalize
> > as this is very common. Though, I think we are missing two things. This
> > documentation refer to the "chosen data limit". Is there controls to
> > configure these *chosen* limit ? The other issue is the vagueness of
> > the documented mode, see lower...
> >
> > > +
> > > +
> > > +.. tabularcolumns:: |p{9.2cm}|p{8.3cm}|
> > > +
> > > +.. raw:: latex
> > > +
> > > + \small
> > > +
> > > +.. flat-table::
> > > + :header-rows: 0
> > > + :stub-columns: 0
> > > +
> > > + * - ``V4L2_MPEG_FRAME_SKIP_MODE_DISABLED``
> > > + - Frame skip mode is disabled.
> > > + * - ``V4L2_MPEG_FRAME_SKIP_MODE_LEVEL_LIMIT``
> > > + - Frame skip mode enabled and buffer limit is set by the chosen
> > > + level and is defined by the standard.
> >
> > At least for H.264, a level is compose of 3 limits. One is the maximum
> > number of macroblocks, this is is evidently not use for frame skipping
> > and already constrained in V4L2 (assuming the driver does not ignore
> > the level control of course). The two other limits are decoded
> > macroblocks/s and encoded kbits/s. Both are measure over time, which
> > means the M2M encoder needs to be timing aware. I think the time source
> > should be documented. Perhaps it is mandatory to set a frame interval
> > for this to work ? Or we need some timestamp to allow variable frame
> > interval ? (I don't think the second is really an option without
> > extending the API again, and confusingly, since I think we have used
> > the timestamp for other purpose already)
>
> Do you want to say that the encoder input timestamp, bitrate control
> (V4L2_CID_MPEG_VIDEO_BITRATE) and S_PARM is not enough to describe
> FRAME_SKIP_MODE_LEVEL_LIMIT mode?

I don't think we have spec to give the input timestamp a meaning that
driver can interpret. In fact I think we gave it a meaning that the
driver must not interpret it (aka driver opaque). So remain S_PARM to
give a clue, but some stream don't have a framerate (like RTP streams,
unless written in bitstream).

>
> > > + * - ``V4L2_MPEG_FRAME_SKIP_MODE_BUF_LIMIT``
> > > + - Frame skip mode enabled and buffer limit is set by the VBV
> > > + (MPEG1/2/4) or CPB (H264) buffer size control.
> >
> > The notion of VBV an CPB is unlikely well known. If my memory is right,
> > these are constrained in buffering: in bytes (VBV) or bits per frame
> > over a window of n-frames (or the gop size for some less flexible
> > encoder) (CPB). I think these should be somehow chosen by application
> > (with controls), directly or indirectly, and documented here to ensure
> > we get consistent implementation across drivers.
>
> I guess you want me to add here references to the following controls:
>
> V4L2_CID_MPEG_VIDEO_VBV_SIZE
> V4L2_CID_MPEG_VIDEO_VBV_DELAY
> V4L2_CID_MPEG_VIDEO_H264_CPB_SIZE

Perhaps it would be helpful yes.

>
> ?
>
> > > +
> > > +.. raw:: latex
> > > +
> > > + \normalsize
> > > +
> > > ``V4L2_CID_MPEG_VIDEO_TEMPORAL_DECIMATION (integer)``
> > > For every captured frame, skip this many subsequent frames (default
> > > 0).
> > > diff --git a/drivers/media/v4l2-core/v4l2-ctrls.c b/drivers/media/v4l2-core/v4l2-ctrls.c
> > > index 3f3fbcd60cc6..d088acfa6dd8 100644
> > > --- a/drivers/media/v4l2-core/v4l2-ctrls.c
> > > +++ b/drivers/media/v4l2-core/v4l2-ctrls.c
> > > @@ -590,6 +590,12 @@ const char * const *v4l2_ctrl_get_menu(u32 id)
> > > "External",
> > > NULL,
> > > };
> > > + static const char * const mpeg_video_frame_skip[] = {
> > > + "Disabled",
> > > + "Level Limit",
> > > + "VBV/CPB Limit",
> > > + NULL,
> > > + };
> > >
> > > switch (id) {
> > > case V4L2_CID_MPEG_AUDIO_SAMPLING_FREQ:
> > > @@ -651,6 +657,8 @@ const char * const *v4l2_ctrl_get_menu(u32 id)
> > > return flash_strobe_source;
> > > case V4L2_CID_MPEG_VIDEO_HEADER_MODE:
> > > return header_mode;
> > > + case V4L2_CID_MPEG_VIDEO_FRAME_SKIP_MODE:
> > > + return mpeg_video_frame_skip;
> > > case V4L2_CID_MPEG_VIDEO_MULTI_SLICE_MODE:
> > > return multi_slice;
> > > case V4L2_CID_MPEG_VIDEO_H264_ENTROPY_MODE:
> > > @@ -844,6 +852,7 @@ const char *v4l2_ctrl_get_name(u32 id)
> > > case V4L2_CID_MPEG_VIDEO_MB_RC_ENABLE: return "H264 MB Level Rate Control";
> > > case V4L2_CID_MPEG_VIDEO_HEADER_MODE: return "Sequence Header Mode";
> > > case V4L2_CID_MPEG_VIDEO_MAX_REF_PIC: return "Max Number of Reference Pics";
> > > + case V4L2_CID_MPEG_VIDEO_FRAME_SKIP_MODE: return "Frame Skip Mode";
> > > case V4L2_CID_MPEG_VIDEO_H263_I_FRAME_QP: return "H263 I-Frame QP Value";
> > > case V4L2_CID_MPEG_VIDEO_H263_P_FRAME_QP: return "H263 P-Frame QP Value";
> > > case V4L2_CID_MPEG_VIDEO_H263_B_FRAME_QP: return "H263 B-Frame QP Value";
> > > @@ -1265,6 +1274,7 @@ void v4l2_ctrl_fill(u32 id, const char **name, enum v4l2_ctrl_type *type,
> > > case V4L2_CID_FLASH_LED_MODE:
> > > case V4L2_CID_FLASH_STROBE_SOURCE:
> > > case V4L2_CID_MPEG_VIDEO_HEADER_MODE:
> > > + case V4L2_CID_MPEG_VIDEO_FRAME_SKIP_MODE:
> > > case V4L2_CID_MPEG_VIDEO_MULTI_SLICE_MODE:
> > > case V4L2_CID_MPEG_VIDEO_H264_ENTROPY_MODE:
> > > case V4L2_CID_MPEG_VIDEO_H264_LEVEL:
> > > diff --git a/include/uapi/linux/v4l2-controls.h b/include/uapi/linux/v4l2-controls.h
> > > index 62271418c1be..4e1526175a4c 100644
> > > --- a/include/uapi/linux/v4l2-controls.h
> > > +++ b/include/uapi/linux/v4l2-controls.h
> > > @@ -742,6 +742,12 @@ enum v4l2_cid_mpeg_video_hevc_size_of_length_field {
> > > #define V4L2_CID_MPEG_VIDEO_HEVC_HIER_CODING_L6_BR (V4L2_CID_MPEG_BASE + 642)
> > > #define V4L2_CID_MPEG_VIDEO_REF_NUMBER_FOR_PFRAMES (V4L2_CID_MPEG_BASE + 643)
> > > #define V4L2_CID_MPEG_VIDEO_PREPEND_SPSPPS_TO_IDR (V4L2_CID_MPEG_BASE + 644)
> > > +#define V4L2_CID_MPEG_VIDEO_FRAME_SKIP_MODE (V4L2_CID_MPEG_BASE + 645)
> > > +enum v4l2_mpeg_video_frame_skip_mode {
> > > + V4L2_MPEG_VIDEO_FRAME_SKIP_MODE_DISABLED = 0,
> > > + V4L2_MPEG_VIDEO_FRAME_SKIP_MODE_LEVEL_LIMIT = 1,
> > > + V4L2_MPEG_VIDEO_FRAME_SKIP_MODE_BUF_LIMIT = 2,
> > > +};
> > >
> > > /* MPEG-class control IDs specific to the CX2341x driver as defined by V4L2 */
> > > #define V4L2_CID_MPEG_CX2341X_BASE (V4L2_CTRL_CLASS_MPEG | 0x1000)

\
 
 \ /
  Last update: 2020-07-15 20:13    [W:0.767 / U:0.092 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site