Avoid low granularity array views about gridap.jl

[fverdugo]

These two functions do the same:

function runview(a::Array{Float64,2})
    b = Array{Float64,1}(undef,(3,))
    for j in 1:size(a,2)
        v = view(a,:,j)
        b .= v 
    end
end

function runnoview(a::Array{Float64,2})
    b = Array{Float64,1}(undef,(3,))
    for j in 1:size(a,2)
        for i in 1:size(a,1)
            b[i] = a[i,j]
        end
    end
end

However, the first one seems to suffer from spurious memory allocation

a = zeros(3,100000)
@time runview(a)
@time runview(a)
  0.059263 seconds (136.37 k allocations: 6.273 MiB)
  0.002854 seconds (100.00 k allocations: 4.578 MiB)

but not the second one:

@time runnoview(a)
@time runnoview(a)
  0.037792 seconds (21.14 k allocations: 1.091 MiB)
  0.000632 seconds (5 allocations: 272 bytes)

[fverdugo]

I leave this comment here just for reference. Closing issue.