linux-arm-msm.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: dikshita@codeaurora.org
To: linux-media@vger.kernel.org, stanimir.varbanov@linaro.org,
	Hans Verkuil <hverkuil@xs4all.nl>,
	sakari.ailus@linux.intel.com
Cc: linux-kernel@vger.kernel.org, linux-arm-msm@vger.kernel.org,
	vgarodia@codeaurora.org, majja@codeaurora.org,
	jdas@codeaurora.org, linux-media-owner@vger.kernel.org
Subject: Re: [RFC PATCH 0/3] METADATA design using V4l2 Request API
Date: Tue, 26 May 2020 12:01:10 +0530	[thread overview]
Message-ID: <66208196d7668fe184f6c9d8c6c69c8b@codeaurora.org> (raw)
In-Reply-To: <1588918890-673-1-git-send-email-dikshita@codeaurora.org>

Hi,

A gentle reminder for the review.

On 2020-05-08 11:51, Dikshita Agarwal wrote:
> There are many commercialized video use cases which needs metadata info
> to be circulated between v4l2 client and v4l2 driver.
> 
> METADATA has following requirements associated:
> •Metadata is an optional info available for a buffer. It is not
> mandatorily for every buffer.
>  For ex. consider metadata ROI (Region Of Interest). ROI is specified
> by clients to indicate
>  the region where enhanced quality is desired. This metadata is given
> as an input information
>  to encoder output plane. Client may or may not specify the ROI for a
> frame during encode as
>  an input metadata. Also if the client has not provided ROI metadata
> for a given frame,
>  it would be incorrect to take the metadata from previous frame. If the 
> data and
>  metadata is asynchronous, it would be difficult for hardware to decide 
> if it
>  needs to wait for metadata buffer or not before processing the input
> frame for encoding.
> •Synchronize the buffer requirement across both the video node/session
>  (incase metadata is being processed as a separate v4l2 video 
> node/session).
>  This is to avoid buffer starvation.
> •Associate the metadata buffer with the data buffer without adding any
> pipeline delay
>  in waiting for each other. This is applicable both at the hardware
> side (the processing end)
>  and client side (the receiving end).
> •Low latency usecases like WFD/split rendering/game streaming/IMS have
> sub-50ms e2e latency
>  requirements, and it is not practical to stall the pipeline due to
> inherent framework latencies.
>  High performance usecase like high-frame rate playback/record can
> lead to frame loss during any pipeline latency.
> 
> To address all above requirements, we used v4l2 Request API as 
> interlace.
> 
> As an experiment, We have introduced new control
> V4L2_CID_MPEG_VIDEO_VENUS_METADATA
> to contain the METADATA info. Exact controls can be finalized once the
> interface is discussed.
> 
> For setting metadata from userspace to kernel, let say on encode output 
> plane,
> following code sequence was followed
> 1. Video driver is registering for media device and creating a media
> node in /dev
> 2. Request fd is allocated by calling MEDIA_IOC_REQUEST_ALLOC IOCTL on 
> media fd.
> 3. METADATA configuration is being applied on request fd using
> VIDIOC_S_EXT_CTRLS IOCTL
>    and the same request fd is added to buf structure structure before
> calling VIDIOC_QBUF on video fd.
> 4. The same request is queued through MEDIA_REQUEST_IOC_QUEUE IOCTL to
> driver then, as a result
>    to which METADATA control will be applied to buffer through S_CTRL.
> 5. Once control is applied and request is completed,
> MEDIA_REQUEST_IOC_REINIT IOCTL is called
>    to re-initialize the request.
> 
> We could achieve the same on capture plane as well by removing few
> checks present currently
> in v4l2 core which restrict the implementation to only output plane.
> 
> We profiled below data with this implementation :
> 1. Total time taken ( round trip ) for setting up control data on video 
> driver
>    with VIDIOC_S_EXT_CTRLS, QBUF and Queue Request: 737us
> 2. Time taken for first QBUF on Output plane to reach driver with
> REQUEST API enabled (One way): 723us
> 3. Time taken for first QBUF on Output plane to reach driver without
> REQUEST API (One way) : 250us
> 4. Time taken by each IOCTL to complete ( round trip ) with REQUEST
> API enabled :
>     a. VIDIOC_S_EXT_CTRLS : 201us
>     b. VIDIOC_QBUF : 92us
>     c. MEDIA_REQUEST_IOC_QUEUE: 386us
> 
> Kindly share your feedback/comments on the design/call sequence.
> Also as we experimented and enabled the metadata on capture plane as
> well, please comment if any issue in
> allowing the metadata exchange on capture plane as well.
> 
> Reference for client side implementation can be found at [1].
> 
> Thanks,
> Dikshita
> 
> [1]
> https://git.linaro.org/people/stanimir.varbanov/v4l2-encode.git/log/?h=dikshita/request-api
> 
> Dikshita Agarwal (3):
>   Register for media device
>     - Initialize and register for media device
>     - define venus_m2m_media_ops
>     - Implement APIs to register/unregister media controller.
>   Enable Request API for output buffers
>     - Add dependency on MEDIA_CONTROLLER_REQUEST_API in Kconfig.
>     - Initialize vb2 ops buf_out_validate and buf_request_complete.
>     - Add support for custom Metadata control
> V4L2_CID_MPEG_VIDEO_VENUS_METADATA
>     - Implemeted/Integrated APIs for Request setup/complete.
>   Enable Request API for Capture Buffers
> 
>  drivers/media/common/videobuf2/videobuf2-v4l2.c |   4 +-
>  drivers/media/platform/Kconfig                  |   2 +-
>  drivers/media/platform/qcom/venus/core.h        |  36 ++++
>  drivers/media/platform/qcom/venus/helpers.c     | 247 
> +++++++++++++++++++++++-
>  drivers/media/platform/qcom/venus/helpers.h     |  15 ++
>  drivers/media/platform/qcom/venus/venc.c        |  63 +++++-
>  drivers/media/platform/qcom/venus/venc_ctrls.c  |  61 +++++-
>  drivers/media/v4l2-core/v4l2-ctrls.c            |  10 +
>  drivers/media/v4l2-core/v4l2-mem2mem.c          |  17 +-
>  include/media/v4l2-ctrls.h                      |   1 +
>  include/media/venus-ctrls.h                     |  22 +++
>  11 files changed, 465 insertions(+), 13 deletions(-)
>  create mode 100644 include/media/venus-ctrls.h

  parent reply	other threads:[~2020-05-26  6:31 UTC|newest]

Thread overview: 23+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-05-08  6:21 [RFC] METADATA design using V4l2 Request API Dikshita Agarwal
2020-05-08  6:21 ` [RFC PATCH 1/3] Register for media device Dikshita Agarwal
2020-05-08  6:21 ` [RFC PATCH 2/3] Enable Request API for output buffers Dikshita Agarwal
2020-05-08  6:21 ` [PATCH 3/3] Enable Request API for Capture Buffers Dikshita Agarwal
2020-05-26  6:31 ` dikshita [this message]
2020-05-26  7:03   ` [RFC PATCH 0/3] METADATA design using V4l2 Request API Hans Verkuil
2020-05-26 10:57 ` [RFC] " Hans Verkuil
2020-05-28 10:48   ` dikshita
2020-05-28 11:24     ` Hans Verkuil
2020-05-29  2:08       ` Nicolas Dufresne
2020-06-11 15:06         ` Nicolas Dufresne
2020-06-12 10:05           ` Hans Verkuil
2020-06-12 16:37             ` Nicolas Dufresne
2020-06-16 13:00               ` dikshita
2020-06-23 12:47                 ` dikshita
2020-07-02  6:33                   ` dikshita
2020-08-03 11:32                 ` Hans Verkuil
2020-09-14 10:21                   ` dikshita
2020-09-14 10:31                     ` Hans Verkuil
2020-05-29  2:18     ` Nicolas Dufresne
2020-05-29  7:31       ` Hans Verkuil
2020-06-05  7:02         ` dikshita
2020-06-05 17:43           ` Nicolas Dufresne

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=66208196d7668fe184f6c9d8c6c69c8b@codeaurora.org \
    --to=dikshita@codeaurora.org \
    --cc=hverkuil@xs4all.nl \
    --cc=jdas@codeaurora.org \
    --cc=linux-arm-msm@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-media-owner@vger.kernel.org \
    --cc=linux-media@vger.kernel.org \
    --cc=majja@codeaurora.org \
    --cc=sakari.ailus@linux.intel.com \
    --cc=stanimir.varbanov@linaro.org \
    --cc=vgarodia@codeaurora.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).