Is this what you're looking for?
julia> using SIMD
julia> v = rand(2, 5)
2×5 Array{Float64,2}:
0.391832 0.785861 0.352291 0.874891 0.874593
0.697768 0.51637 0.711525 0.400077 0.266559
julia> w = reinterpret(Vec{2, Float64}, vec(v))
5-element Array{SIMD.Vec{2,Float64},1}:
Float64⟨0.391832,0.697768⟩
Float64⟨0.785861,0.51637⟩
Float64⟨0.352291,0.711525⟩
Float64⟨0.874891,0.400077⟩
Float64⟨0.874593,0.266559⟩
Den onsdag 16 november 2016 kl. 05:52:14 UTC+1 skrev Sheehan Olver:
>
>
> In some high performance code I want to convert an 2 x n `Matrix{Float64}`
> to a n long `Matrix{Vec{2,Float64}}`with no memory allocation (both types
> will store memory in the same order). The following works:
>
> *v=rand(2,100)*
> *w=unsafe_wrap(Vector{Vec{2,Float64}},*
> * reinterpret(Ptr{Vec{2,Float64}},pointer(v)),*
> * size(pts,1),true)*
>
> The variable *v* is a throw away variable, however, and as I understand
> it, the way its written the memory for *w* will be freed as soon as *v*
> is no longer referenced. Is there any way to tell the garbage collector to
> not collect *v*?
>
