Ability to pad slices

[henrygouk]

The ability to pad slices could come in handy for image processing operations, among other things.

Something with similar behaviour to MATLAB's padarray function is what I am thinking of.

[henrygouk]

A solution to this would also solve #291

[9il]

Thanks!

[henrygouk]

It should be possible to implement this in a way that performs the padding lazily as elements are accessed, which would be nice in the sense that allocations wouldn't be needed. This would obviously increase the number of instructions required to access elements in the padded slice, but that problem can be somewhat mitigated by calling the slice function to allocate a new slice that has the padding eagerly evaluated.

[9il]

@henrygouk stack function was added . In combination with repeat from ndslice.topology it allows to create padding view and allocate new image with slice. Stack does not have [...] access primitives.

[9il]

It should be possible to implement this in a way that performs the padding lazily as elements are accessed, which would be nice in the sense that allocations wouldn't be needed. This would obviously increase the number of instructions required to access elements in the padded slice, but that problem can be somewhat mitigated by calling the slice function to allocate a new slice that has the padding eagerly evaluated.

Lazy ndslice view for stack can be achieved using mir.ndslice.topology.indexed and ndiota.
See libmir/mir-algorithm#29

I will reopen this Issue for mir-cv.

[9il]

constant padding was added libmir/mir-algorithm#31

[9il]

edge, symmetric and wrap paddings libmir/mir-algorithm#32