10bit 422 to unsigned short RGB (3x16bit) ?

[iptrkpfn]

Hi Frank,

I wondered if/why there is no conversion in pixfc-sse for getting from 422 to a 48bit (or 64bit) RGB(A) pixel format (to keep the full bit depth of the source 10bit YCbCr buffers).

I browsed through the code of pixfc-sse, unfortunately without fully understanding it. But it seems you do an RGB packing step at the end of each YUV2RGB conversion. Could a packing be added that packs to 16bit RGB instead of 8it RGB? Or is the bit depth already lost at for that step?

Thanks for your help!

Best,

Patrick

[Pix FC]

Hi Patrick,
You are correct, at the output of yuv2rgb inlines each component is indeed 16-bit wide, but because the currently supported pixel formats are all 8-bit wide, all component values in the conversion pipeline will take up to 8-bit (in a 16-bit wide container). I started working on adding support for 10-bit formats and will be doing this in stages. I just finished adding conversions to/from 10bit YUV format (aka v210), and I am currently fixing a few conversion bugs here and there. If you wish to test it, checkout the SVN branch at http://pixfc-sse.googlecode.com/svn/branches/add-10bit . The next step is to add support for 10-bit RGB.

And yes, the macro "situation" makes it quite hard to follow what's going on inside pixfc. I m planning to write up a page detailing how to dig into the code, but essentially, you start with a conversion function in conversion_routines_from_xxx.c which is defined using a single macro, then go to the definition of that macro and find out which conversion recipe is used (called xxxx_TO_yyyy_RECIPE). The conversion recipes can be found in  yuv_conversion_recipes.h and rgb_conversion_recipes.h. They contain the core of the conversion code and calls the correct inline functions to do the job, usually broken up into the following blocks: unpacking -> up/downsampling -> conversion -> packing. Between each of these blocks, components are all 16-bit wide, but currently only use 8-bit as explained above.
Hope this helps.

Cheers,
Frank

[iptrkpfn]

Hi Frank,

thanks for clarifying! I don't know if a comprehensive documentation for the architecture of pixfc-sse is necessary, your comments on how to follow the macros down to the real implementation already helped, thanks! I will see if I can test the 8bit version of the v210 to RGB conversions next week.

Best,

Patrick