ARM/NEON version and usage of floating point Mat

edit

Hi,

To satisfy different computations, like averaging of image from camera using weights I need to convert frame (1920x1200 px) from 8 or 16bit representation to floating point, This is pretty quick and works fine.

However conversion back from float to 8/16-bits integer is terribly slow.

Is there some missing NEON code supporting that conversion, maybe it is not advised to do so for some reason?