maximum(A, d), sum(A, d), ... should give an SMatrix for A::SMatrix

[stevengj]

Currently, this is using the generic reduction code:

julia> A = SMatrix{3,3}(rand(3,3))
3×3 StaticArrays.SMatrix{3,3,Float64,9}:
 0.0121247  0.371658  0.950669 
 0.450277   0.363499  0.0998775
 0.322925   0.158552  0.548777 

julia> maximum(A,2)
3×1 StaticArrays.MMatrix{3,1,Float64,3}:
 0.950669
 0.450277
 0.548777

julia> @which maximum(A,2)
maximum(A::AbstractArray, region) at reducedim.jl:305

[andyferris]

No... we shouldn't even do what is happening now (return MMatrix)! It's not type-stable, and I plan on fixing this (#22) soon.

To do this properly, we would need something like JuliaLang/julia#17880 (simultaneous inlining and inference) so that we have some way of constant-propagating the dimension you are selecting to reduce over inside another function.

Is this a function you find you use a lot? I almost never do, but maybe that is just me. (I would use things like sum(::Vector{SVector{3,Float64}}) for instance - reinterpret() is useful for this stuff). If someone absolutely needs this, we can add a Val version, e.g.

sum(rand(SMatrix{3,3}), Val{2})

[stevengj]

I'm using it in a package I'm working on, but it's not that performance-critical.

[wsshin]

Generated a PR addressing this: #86

[c42f]

Now that #86 is merged, maximum(A,Val{2}) works.

[andyferris]

I'm going to close this... I think maximum(A, Val{2}) is the best we can hope for at this stage.