On Wednesday, February 09, 2011 12:50:27 Sachin Gupta wrote:
Hi Hans,
Thanks for your inputs.We are part of Linaro organisation For more details on Linaro please refer to http://www.linaro.org . As part of our activities on Linaro we have been debating at whats the right solution for exposing camera support / features on a platform Openmax or v4l2.
It's simple, really. If you want to have support for the video hardware merged in the kernel, then V4L is the way to go. OpenMax drivers will never enter the kernel.
Also can you share some details/docs on how userside library/v4l2 partitioning is supposed to work.
It's really not yet known. The idea is as follows: the V4L driver gives access to the hardware. Our goal is that through this driver the full functionality of the hardware can be accessed.
There is also the library libv4l (also documented in the V4L spec: http://linuxtv.org/downloads/v4l-dvb-apis/, section 6). The main goal of this library is to do format conversion from custom formats to more common formats, software whitebalancing, image mirroring and similar tasks. The main use of this library is in generic webcam apps. You would not expect to see it used on a SoC with camera software written for a specific board. But work is in progress to extend this library to allow plug-ins as well.
This would make it possible to intercept V4L ioctls and call on some proprietary code to do e.g. whitebalancing or setting up scaling factors. Some patches have been posted already, but I don't know the current status.
In the case of complex video hardware I suspect that userspace libraries for specific video hardware might be needed. What I expect to be in there is code that will set up pipelines or specific use-cases in a simplified way. The library would know how to optimally configure the various blocks. A typical example would be scaling: there may be multiple scalers in the pipeline, and while the V4L2 API will give you access to those scalers it does not contain the knowledge how to achieve the best end result.
Now, it is too early to know whether or not this will actually be needed, and if it is needed, what form it will take. Perhaps it will be possible to write such libraries as a libv4l plugin, perhaps developers will just make their own solution.
Right now I would just forget about such libraries. It's way too early.
Regards,
Hans
Thanks Sachin
On Wed, Feb 9, 2011 at 1:13 PM, Hans Verkuil hverkuil@xs4all.nl wrote:
On Wednesday, February 09, 2011 07:34:09 Sachin Gupta wrote:
Looking at ppt from Robert , it seems v4l2 subdevices is the way to
support
different devices that may be involved in imaging processing chain, also from the ppt it seems a userside library for Media controller is needed particular to each platform which controls these subdevices.I have not
been
able to find detailed documentation on this but it seems we are talking about custom solution for every platform based on platform topology for image processing chain.
It is not clear yet whether custom libraries will be needed or not. For omap3 (the first driver to use the media controller) it doesn't seem to be needed (yet?).
However, the complexity of some of these video systems is such that I can't help thinking that some library will be required to simplify the use of such hardware.
In general it will not be possible to make a completely generic solution for video subsystems that will work everywhere. The various architectures simply are too varied for that. The media controller will go some way to solving this, but a 100% solution is in practice impossible.
If you go only for a subset (for example, setting up a standard simple pipeline for a camera-type system), you are probably able to make something generic, but if you want to get full control over such systems in order to get the best possible quality, then you will have to customize your code for that particular hardware.
It might help me if I could get a better idea of what you are working on and what the goal is. I came in in the middle of the discussion and I think I'm missing some of the pieces :-)
Regards,
HansOn Wed, Feb 9, 2011 at 11:53 AM, Subash Patel subash.rp@samsung.com
wrote:
In the reference architecture in ppt, we can directly wait for the RSZ interrupt, if we configure the hardware pipe. It was my
mis-understanding as
each of those hardware blocks can deliver interrupts too. In that way
ARM
needs to just work at finished frame, like forward it to the display or codec engine etc. V4L2 can be easily used for such hardware
architecture.
But if a ISP chooses to do the above work in a seperate (dsp)processor,
can
we still use V4L2? OMX seems better in such environment. Let me know if there is any other alternative.
Regards, Subash _______________________________________________ linaro-dev mailing list linaro-dev@lists.linaro.org http://lists.linaro.org/mailman/listinfo/linaro-dev
-- Hans Verkuil - video4linux developer - sponsored by Cisco