Date:   Tue, 12 Dec 2017 13:54:34 +0100
From:   Hans Verkuil <hverkuil@...all.nl>
To:     Dmitry Osipenko <digetx@...il.com>,
        Thierry Reding <thierry.reding@...il.com>,
        Jonathan Hunter <jonathanh@...dia.com>,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        Rob Herring <robh+dt@...nel.org>,
        Mauro Carvalho Chehab <mchehab@...nel.org>,
        Stephen Warren <swarren@...dotorg.org>,
        Vladimir Zapolskiy <vz@...ia.com>
Cc:     Dan Carpenter <dan.carpenter@...cle.com>,
        linux-media@...r.kernel.org, linux-tegra@...r.kernel.org,
        devel@...verdev.osuosl.org, devicetree@...r.kernel.org,
        linux-kernel@...r.kernel.org,
        Maxime Ripard <maxime.ripard@...e-electrons.com>,
        Giulio Benetti <giulio.benetti@...ronovasrl.com>
Subject: Re: [PATCH v4 3/5] staging: Introduce NVIDIA Tegra video decoder
 driver

On 10/12/17 19:56, Dmitry Osipenko wrote:
> On 05.12.2017 16:03, Hans Verkuil wrote:
>> On 12/05/17 13:17, Dmitry Osipenko wrote:
>>> Hi Hans,
>>>
>>> On 04.12.2017 17:04, Hans Verkuil wrote:
>>>> Hi Dmitry,
>>>>
>>>> As you already mention in the TODO, this should become a v4l2 codec driver.
>>>>
>>>> Good existing examples are the coda, qcom/venus and mtk-vcodec drivers.
>>>>
>>>> One thing that is not clear from this code is if the tegra hardware is a
>>>> stateful or stateless codec, i.e. does it keep track of the decoder state
>>>> in the hardware, or does the application have to keep track of the state and
>>>> provide the state information together with the video data?
>>>>
>>>> I ask because at the moment only stateful codecs are supported. Work is ongoing
>>>> to support stateless codecs, but we don't support that for now.
>>>>
>>>
>>> It is stateless. Is there anything ready to try out? If yes, could you please
>>> give a reference to that work?
>>
>> I rebased my two year old 'requests2' branch to the latest mainline version and
>> gave it the imaginative name 'requests3':
>>
>> https://git.linuxtv.org/hverkuil/media_tree.git/log/?h=requests3
>>
>> (Note: only compile tested!)
> 
> Thank you very much.
> 
>> This is what ChromeOS has been using (actually they use a slightly older version)
>> and the new version that is currently being developed will be similar, so any work
>> you do on top of this will carry over to the final version without too much effort.
>>
>> At least, that's the intention :-)
>>
>> I've CC-ed Maxime and Giulio as well: they are looking into adding support for
>> the stateless allwinner codec based on this code as well. There may well be
>> opportunities for you to work together, esp. on the userspace side. Note that
>> Rockchip has the same issue, they too have a stateless HW codec.
> 
> IIUC, we will have to define video decoder parameters in V4L API and then make a
> V4L driver / userspace prototype (ffmpeg for example) that will use the requests
> API for video decoding in order to upstream the requests API. Does it sound good?

Correct.

Hugues Fruchet made an example bit parser for mpeg:

https://www.spinics.net/lists/linux-media/msg115017.html

So something similar would work.

My recommendation would be to make a separate library that can be shared among
different implementations (i.e. in gstreamer, in vdpau, using a libv4l2 plugin, etc., etc.).

That can easily be hosted as part of v4l-utils to keep it in sync with the
kernel drivers. If it's independent of the various drivers, then it can be
hosted anywhere of course.

BTW, we as V4L2 core developers have no plans on working on such a library :-)

Regards,

	Hans

> 
>>>
>>>> Anyway, I'm OK with merging this in staging. Although I think it should go
>>>> to staging/media since we want to keep track of it.
>>>>
>>>
>>> Awesome, I'll move driver to staging/media in V5. Thanks!
>>
>> Nice, thanks!

Powered by blists - more mailing lists