Add unsigned fixed-point numbers

[timholy]

CC @StefanKarpinski, since this was your suggestion!

[timholy]

Fixed the error on 0.2. I was forced to test with a less stringent iteration syntax, due to issues with 0.2's range iteration.

[JeffBezanson]

Seems like this should be a method of reinterpret.

[timholy]

Seems reasonable. This will introduce a one-argument variant of reinterpret, but I agree that makes sense.

[JeffBezanson]

These data types are kind of cool!

[timholy]

The introduction of scaledual is a bit speculative, but it can essentially double the speed of certain operations. It has the following property:

bd, ad = scaledual(b, a)
b*a == bd*ad

but the RHS might be faster (particularly for floating-point b and an array a of fixed-point numbers).

The idea would be the following: say you're writing imfilter, then

function imfilter(img::AbstractArray, kern::AbstractArray)
    scalefac, imgd = scaledual(eltype(kern), img)
    if scalefac != 1
        kern = scalefac*kern
    end
    # Implement the filtering operation on imgd and kern
end

[timholy]

I added some documentation. If you like scaledual, I can document it, too.

[StefanKarpinski]

I assume this come from the thing I mentioned on the Color package about considering Uint8 as a way of representing floating-point values 0/255, 1/255, etc.?

[timholy]

Yes indeed! JuliaAttic/Color.jl#42.

[timholy]

Bump. Color is waiting on this, and Images is waiting on Color. (I probably won't merge the Images changes immediately, but I think the Color one is ready to go.)

[timholy]

Thanks!