Make mapslices accept arbitrary number of inputs an allow broadcasting.

[andreasnoack]

To allow this, it was necessary to make the dims argument a keyword argument. Examples

julia> A = randn(3,2,2); B = randn(2,2,2);

julia> mapslices(svdvals, A, B, dims = (1,2))
2x1x2 Array{Float64,3}:
[:, :, 1] =
 1.35258
 3.10531

[:, :, 2] =
 7.87285 
 0.160165

julia> mapslices(svdvals, A, B[:,:,1], dims = (1,2))
2x1x2 Array{Float64,3}:
[:, :, 1] =
 1.35258
 3.10531

[:, :, 2] =
 1.28949 
 0.289479

julia> mapslices(norm, A, 1, dims = 1)
1x2x2 Array{Float64,3}:
[:, :, 1] =
 2.67363  2.65857

[:, :, 2] =
 1.16449  2.33811

This generalized version is slightly slower than the old version, but I guess this function is not used much in performance critical code.

cc: @alanedelman

[andreasnoack]

Closing, as this was not the desired solution.

[mschauer]

This seems to be difficult to get right. I think the point is that a multiarray map slices should behave differently. My unhappy try was #5140

[andreasnoack]

@mschauer Thanks for the reference. I didn't know about that one. I've prepared another solution, but the main focus is parallel execution so it might only fo into DistributedArrays.jl.