Re: [PATCH 0/2] V4L: Extended crop/compose API

[Sylwester Nawrocki <s.nawrocki@samsung.com>]

On 05/18/2011 02:31 PM, Hans Verkuil wrote:
> On Wednesday, May 18, 2011 14:06:21 Sylwester Nawrocki wrote:
>> On 05/16/2011 09:21 AM, Laurent Pinchart wrote:
>> > On Saturday 14 May 2011 12:50:32 Hans Verkuil wrote:
>> >> On Friday, May 13, 2011 14:43:08 Laurent Pinchart wrote:
>> >>> On Saturday 07 May 2011 13:52:25 Hans Verkuil wrote:
>> >>>> On Thursday, May 05, 2011 11:39:54 Tomasz Stanislawski wrote:
>> >
> 
>> > [snip]
>> ...
> 
>> >>>>> * resolution of an image combined with support for
> 
>> >>>>> VIDIOC_S_MULTISELECTION
> 
>> >>>>>
> 
>> >>>>> allows to pass a triple format/crop/compose sizes in a single
> 
>> >>>>> ioctl
> 
>> >>>>
> 
>> >>>> I don't believe S_MULTISELECTION will solve anything. Specific
> 
>> >>>> use-cases perhaps, but not the general problem of setting up a
> 
>> >>>> pipeline. I feel another brainstorm session coming to solve that. We
> 
>> >>>> never came to a solution for it in Warsaw.
> 
>> >>>
> 
>> >>> Pipelines are configured on subdev nodes, not on video nodes, so I'm also
> 
>> >>> unsure whether multiselection support would really be useful.
> 
>> >>>
> 
>> >>> If we decide to implement multiselection support, we might as well use
> 
>> >>> that only. We would need a multiselection target bitmask, so selection
> 
>> >>> targets should all be < 32.
> 
>> >>>
> 
>> >>> Thinking some more about it, does it make sense to set both crop and
> 
>> >>> compose on a single video device node (not talking about mem-to-mem,
> 
>> >>> where you use the type to multiplex input/output devices on the same
> 
>> >>> node) ? If so, what would the use cases be ?
> 
>>
> 
>> I can't think of any, one either use crop or compose.
> 
> 
> I can: you crop in the video receiver and compose it into a larger buffer.
> 
> Actually quite a desirable feature.

Yes, right. Don't know why I imagined something different.
And we need it in Samsung capture capture interfaces as well. The H/W
is capable of cropping and composing with camera interface as a data source
similarly as it is done with memory buffers.


Regards,
-- 
Sylwester Nawrocki
Samsung Poland R&D Center