[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <dc584b3a4eebdf259e555200dbc4bf8467d60591.camel@collabora.com>
Date: Thu, 25 Jun 2020 10:42:48 -0400
From: Nicolas Dufresne <nicolas.dufresne@...labora.com>
To: Ezequiel Garcia <ezequiel@...labora.com>,
linux-media@...r.kernel.org, linux-kernel@...r.kernel.org
Cc: Tomasz Figa <tfiga@...omium.org>, kernel@...labora.com,
Jonas Karlman <jonas@...boo.se>,
Hans Verkuil <hverkuil@...all.nl>,
Alexandre Courbot <acourbot@...omium.org>,
Jeffrey Kardatzke <jkardatzke@...omium.org>,
Philipp Zabel <p.zabel@...gutronix.de>,
Maxime Ripard <mripard@...nel.org>,
Paul Kocialkowski <paul.kocialkowski@...tlin.com>
Subject: Re: [RFC 0/7] media: Clean and make H264 stateless uAPI public
Le mardi 23 juin 2020 à 15:28 -0300, Ezequiel Garcia a écrit :
> The recent patch posted by Jernej (which I'm including for context),
> encouraged me to address all the known issues in the uAPI.
>
> I hope we can finally make this uAPI interface
> public; it would be nice to address the other codec
> interfaces so we can move the codec drivers out of staging.
>
> It should be noted that there is already GStreamer native
> support for this interface, which will be part of 1.18,
> once it's released [1], as well as support in Chromium [2].
Keep in mind, no one did a MultiView and/or SVC implementation in
userspace yet. Multiview seems to share a lot of interlacing, which
ffmpeg implementation covers, but some validation would be nice.
As for SVC, I only started looking at that this means for decoding. It
looks like buffer "holding" mechanism is mostly what we need, I'd would
simply like to verify, as I was told that scaling down might be
required when SVC stream is missing (but perhaps this is optional ?)
>
> The basic idea here is to sanitize the interface,
> making sure the structs are aligned to 64-bit,
> adding reserved fields for padding where suitable.
>
> These reserved fields can then be used to support future extensions,
> in case such need appears.
>
> In addition to this, moving the slice invariant fields
> to the per-frame control, makes the frame-mode driver
> implementation much nicer and the interface; see patch 6/7 for details.
>
> I'm not adding a MAINTAINERS entry, but I'd like to do so,
> so we make sure any uAPI changes are sent to those involved.
>
> Another potential change is the addition of a "Since:" tag to the
> control specification, so we can document which kernel version
> added the interface. This might prove useful if reserved
> fields are then used to extend the interface.
>
> Finally, I'm sneaking here a change from Philipp Zabel
> which apparently fell thru the cracks.
Thanks a lot of working on this.
>
> [1] https://gitlab.freedesktop.org/gstreamer/gst-plugins-bad/-/tree/master/sys/v4l2codecs
> [2] https://chromium.googlesource.com/chromium/src.git/+/refs/heads/master/media/gpu/v4l2/
>
> Ezequiel Garcia (5):
> fixup! media: uapi: h264: update reference lists
> media: uapi: h264: increase size of fields
> media: uapi: h264: pad v4l2_ctrl_h264_pps to 64-bit
> media: uapi: h264: Clean slice invariants syntax elements
> media: uapi: make H264 stateless codec interface public
>
> Jernej Skrabec (1):
> media: uapi: h264: update reference lists
>
> Philipp Zabel (1):
> media: uapi: h264: clarify pic_order_cnt_bit_size field
>
> .../media/v4l/ext-ctrls-codec.rst | 135 ++++++++++++------
> drivers/media/v4l2-core/v4l2-ctrls.c | 31 ++++
> drivers/media/v4l2-core/v4l2-h264.c | 8 +-
> .../staging/media/hantro/hantro_g1_h264_dec.c | 21 ++-
> drivers/staging/media/hantro/hantro_h264.c | 3 +-
> drivers/staging/media/hantro/hantro_hw.h | 5 +-
> drivers/staging/media/rkvdec/rkvdec-h264.c | 6 +-
> .../staging/media/sunxi/cedrus/cedrus_h264.c | 15 +-
> include/media/v4l2-ctrls.h | 3 +-
> include/media/v4l2-h264.h | 5 +-
> .../linux/v4l2-h264-ctrls.h} | 73 ++++++----
> 11 files changed, 194 insertions(+), 111 deletions(-)
> rename include/{media/h264-ctrls.h => uapi/linux/v4l2-h264-ctrls.h} (88%)
>
Download attachment "signature.asc" of type "application/pgp-signature" (196 bytes)
Powered by blists - more mailing lists