You have to think about things differntly in hardware. In hardware, memory is expensive and you typically build smaller arrays, and try to process video through line buffering. If you do need to story a large number of pixels, you would typically try to store them in external memory and then pull what you need to do processing.
I'm gussing that you would probably doing a kernal of say 8x8. In which case you only need to buffer up 8 lines at a time, while leaving the rest of the data in external memory.
Now to specically address your questions.
1. Yes you can save this as a 2D array, which could get converted to memory. You would want to look at the memory tempaltes in Project Navigator to see how to write this code so that it is synthesizable. But do keep in mind that comments about on using external memory if you can.
2. See above.
3. Floatingpoint is expensive in hardware, and there is limited support for synthesizable Floatoing Point operations in most syntheis tools. This is improving and will continue to improve. In the mean time you will have to deal with these as vectors if you want to use floating point values. You might want to consider converting to fixed point if possible. In most cases this is sufficient, though it does take some time to validate that you are expereincing any overflows.
Chris
