[Open-graphics] RGB/YUV converstion... importance?

[Timothy Miller]

On 9/13/06, Attila Kinali <attila at kinali.ch> wrote:
>
> 4:2:1 does that even exist? I've never seen it nor
> heard of it
>
> 4:2:0, the most important subsampling format is missing

Can you describe it?

>
> Important to note here is that the chroma samples are often
> half a pixel off the luma pixel positions. Thus every sample
> needs to be interpolated and noone can be just copied.
> But there are different standards even for that, depending
> on the codec.

This is a simple matter of how the texture coordinates relate to
screen coordinates.  No problem.

If we can send U and V at a different time from Y, I think we can
easily do any of these formats without any additional hardware
(besides the matrix multiply logic itself).

So, when you're ready to do the conversion, you just have to set up
the drawing engine right.  Using one or two textures, we can upload U
and V and then process Y on the fly as it's coming on.  If U and V are
in one texture, we could also put Y into another and combine them
after all image data has been uploaded.

The only difference between process-on-the-fly and
process-after-upload is latency.  Throughput would remain essentially
the same (if frames are software pipelined right), although memory
management for the latter is more complex, and the latter uses more
memory bandwidth.