[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <2316c0cb-6a84-432e-8cd3-ca953123bd67@microchip.com>
Date: Wed, 15 Oct 2025 06:05:11 +0000
From: <Balamanikandan.Gunasundar@...rochip.com>
To: <kieran.bingham@...asonboard.com>, <eugen.hristev@...aro.org>
CC: <3chas3@...il.com>, <Nicolas.Ferre@...rochip.com>,
<alexandre.belloni@...tlin.com>, <claudiu.beznea@...on.dev>,
<Balakrishnan.S@...rochip.com>, <hverkuil@...nel.org>,
<ribalda@...omium.org>, <laurent.pinchart+renesas@...asonboard.com>,
<jacopo.mondi@...asonboard.com>, <dan.scally+renesas@...asonboard.com>,
<tomi.valkeinen@...asonboard.com>, <linux-kernel@...r.kernel.org>,
<linux-media@...r.kernel.org>, <linux-atm-general@...ts.sourceforge.net>,
<netdev@...r.kernel.org>, <linux-arm-kernel@...ts.infradead.org>,
<mchehab@...nel.org>
Subject: Re: [PATCH 00/18] media: microchip-isc: Color correction and
histogram stats
Hi Kieran,
On 10/10/25 1:51 pm, Kieran Bingham wrote:
>> Hi Bala,
>>
>> On 10/9/25 18:52, Balamanikandan Gunasundar wrote:
>>> Hi,
>>>
>>> This patch series has a set of enhancements to the Microchip Image Sensor
>>> Controller driver. The objective is to expand its image processing
>>> capabilities and to improve the colors.
>>>
>>> This series also introduces a new stats driver that exposes the histogram
>>> data to userspace via v4l2 controls. This allows applications such as
>>> libcamera to access real time image statistics for advanced image
>>> processing like automatic exposure, white balance adjustments etc.
> How much data do you anticipate to be passing through controls? (What
> can the hardware provide in total if we look at the bigger picture to
> support the full device?)
Thanks for the feedback. Currently the hardware provides histogram data
with 256 bins per color (R, G, B and Y) each 32 bit wide. So roughly
around 4 x 256 x 4 = 4K of data per frame. The intent to use it as v4l2
controls was to make this data easily accessible to user space without a
need for a new data interface initially. However I agree that we may
expand the statistics for future ISP blocks to include region based
stats (We are already in discussion with the design)
>
> For all other ISPs we've been working towards using structured parameter
> buffers to pass data - and we've been making that format extensible,
> which I think could also be a design that can apply to statistics.
That sounds great and we can certainly explore aligning with the
structured buffers for histogram data and any other future statistics.
Thanks,
Bala
>
> This would greatly reduce the overhead of managing 'one control per
> value' ... or things like passing large tables (like a lens shading
> table perhaps) through controls.
>
> --
> Kieran
Powered by blists - more mailing lists