Do upscaling and gradient detection directly on image colors

[luquinhas0021]

Commonly, this operators are done in each one of three components, apartly, and, then, the results are joined.
A RGB24 pixel has 16777216 possible full colors. They are valued from 1 to 16777216,
So can I do these types of image operator using full color information? Does it have bad results?

[feisty2]

Except the correlation between channels is practically useless to resampling

[luquinhas0021]

Maybe a hash table, inside it will be the R, G and B intensities of each full color value, can be useful.

Example: R = x, G = y, B = z a

a is taken from 1 to 16777216

[raffriff42]

MMX, SSE etc can operate on multiple integers in parallel. I assume that's what's used.
https://en.wikipedia.org/wiki/MMX_(instruction_set)

Yep. Search in the file below for 'MMX'
http://avisynth2.cvs.sourceforge.net/viewvc/avisynth2/avisynth/src/convert/convert_rgb.cpp?content-type=text%2Fplain