Re: [linux-sunxi] Cedrus driver

From: Maxime Ripard <maxime.ripard@free-electrons.com>
To: Nicolas Dufresne <nicolas@ndufresne.ca>
Cc: Giulio Benetti <giulio.benetti@micronovasrl.com>,
	Andreas Baierl <list@imkreisrum.de>,
	linux-sunxi@googlegroups.com, robh+dt@kernel.org,
	mark.rutland@arm.com, linux@armlinux.org.uk, wens@csie.org,
	devicetree@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org, thomas@vitsch.nl,
	linux-media@vger.kernel.org
Subject: Re: [linux-sunxi] Cedrus driver
Date: Fri, 17 Nov 2017 09:01:27 +0100	[thread overview]
Message-ID: <20171117080127.mtfkehxbofk56alv@flea> (raw)
In-Reply-To: <1510862395.8053.39.camel@ndufresne.ca>

[-- Attachment #1: Type: text/plain, Size: 1873 bytes --]

Hi Nicolas,

On Thu, Nov 16, 2017 at 02:59:55PM -0500, Nicolas Dufresne wrote:
> Le jeudi 16 novembre 2017 à 12:02 +0100, Maxime Ripard a écrit :
> > Assuming that the request API is in, we'd need to:
> >   - Finish the MPEG4 support
> >   - Work on more useful codecs (H264 comes to my mind)
> 
> For which we will have to review the tables and make sure they match
> the spec (the easy part). But as an example, that branch uses a table
> that merge Mpeg4 VOP and VOP Short Header. We need to make sure it does
> not pause problems or split it up. On top of that, ST and Rockchip
> teams should give some help and sync with these tables on their side.
> We also need to consider decoder like Tegra 2. In H264, they don't need
> frame parsing, but just the PPS/SPS data (might just be parsed in the
> driver, like CODA ?). There is other mode of operation, specially in
> H264/HEVC low latency, where the decoder will be similar, but will
> accept and process slices right away, without waiting for the full
> frame.

Sorry if it's a dumb question, but what branches and tables are you
talking about here?

> We also need some doc, to be able to tell the GStreamer and FFMPEG team
> how to detect and handle these decoder. I doubt the libv4l2 proposed
> approach will be used for these two projects since they already have
> their own parser and would like to not parse twice. As an example, we
> need to document that V4L2_PIX_FMT_MPEG2_FRAME implies using the
> Request API and specific CID. We should probably also ping the Chrome
> Devs, which probably have couple of pending branches around this.

We've had a prototype that wasn't based on libv4l but was based on the
VA-API, and it's been working great for us so far.

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 801 bytes --]