Add support for 16-bit planar YUV422

[cwhite102]

Hello,
Our system processes data as 16-bit little-endian Planar YUV 422.
We are using MTL for both Tx and Rx 2110 in pipeline mode.

Currently we need to do an extra copy of the video data to/from MTL because 16-bit planar isn't supported directly by MTL.

What we wish to have is a frame format supported like:

/** YUV 422 planar 16bit little endian */
ST_FRAME_FMT_YUV422PLANAR16LE

Similar to the existing 10 & 12-bit formats:
ST_FRAME_FMT_YUV422PLANAR10LE
ST_FRAME_FMT_YUV422PLANAR12LE

With optimized (AVX-x) conversions where appropriate.
For example: st20_rfc4175_422be10_to_yuv422p16le

We'd be happy to help with contributing, but in order to do so we need to ask:

1. Is this an acceptable addition to MTL?
2. Could you provide some pointers/guidelines as to the necessary work?

On the other hand, if this feature is an easy addition by a main contributor who is familiar with the code, it would be greatly appreciated!

Thanks for your comments and feedback,
Craig

[Sakoram]

Hi,
We do not have the addition of ST_FRAME_FMT_YUV422PLANAR16LE on our roadmap, but I am more than happy to help add this format.

To add the format you would have to:

1. Add ST_FRAME_FMT_YUV422PLANAR16LE to the enum.
2. Extend converters list with the desired conversions.
3. Write actual conversion functions (in st_convert.c), like proposed st20_rfc4175_422be10_to_yuv422p16le. (I suggest using rfc4175_422be16 for the conversion to make it lossless)
4. Extend parsing in application/plugin. ( you can add it to the app/plugin you use, I will do it for the rest)
5. Add tests for conversions in cvt_test.cpp

If I didn't forget anything that should work. If you have any questions, please ask here.

[Sakoram]

https://github.com/OpenVisualCloud/Media-Transport-Library?tab=readme-ov-file#7-how-to-contribute

[cwhite102]

Thank you for all the info!
I will attempt to carve out the time to look into this.

[DianaEs2]

Hi Sakoram, some notes about support ST_FRAME_FMT_YUV422PLANAR16LE. Currently, we are interested in lossy conversions only st20_yuv422p16le_to_rfc4175_422be10 and st20_rfc4175_422be10_to_yuv422p16le. When yuv422p16le is used with 10-bit payload, the format layout looks like this (padding on the left, in least significant bits)

Only xxx_scalar options are implemented for now. Before we go ahead with AVX-512 optimization, can you please confirm that this approach is acceptable?

commit-695fbf5.patch

Thanks!

[awilczyns]

Hi cwhite102,

thank you for your notes regarding the support for ST_FRAME_FMT_YUV422PLANAR16LE. I can confirm that the proposed approach of focusing on lossy conversions between st20_yuv422p16le_to_rfc4175_422be10 and st20_rfc4175_422be10_to_yuv422p16le is acceptable.

I have reviewed the commit-695fbf5.patch, and it has been checked and is working fine. Please proceed to create a pull request with these changes so we can merge them into the main branch.
Thank you for your efforts and collaboration!

Best regards