Compilation time of generating a quadrature rule is very long

[fredrikekre]

Probably due to @nloops and friends? Mostly on v0.6, but pretty slow on v0.5 aswell:

v0.5:

julia> using JuAFEM

julia> @time qr = QuadratureRule{2, RefCube}(:legendre, 2);
  3.661755 seconds (5.95 M allocations: 239.781 MB, 2.25% gc time)

v0.6:

julia> using JuAFEM

julia> @time qr = QuadratureRule{2, RefCube}(:legendre, 2);
 10.056088 seconds (11.82 M allocations: 544.792 MiB, 1.60% gc time)

[fredrikekre]

function _points_weights(quad_type::Symbol, order::Int)
    if quad_type == :legendre
        p, w = gausslegendre(order)
    elseif quad_type == :lobatto
        p, w = gausslobatto(order)
    else
        throw(ArgumentError("unsupported quadrature rule"))
    end
    return p, w
end

gives:

julia> @time JuAFEM._points_weights(:legendre, 2);
  8.583793 seconds (11.41 M allocations: 529.193 MiB, 2.29% gc time)

but (in a new session):

julia> using FastGaussQuadrature

julia> @time p, w = gausslegendre(2);
  0.332452 seconds (231.99 k allocations: 12.070 MiB)

[fredrikekre]

Okay, it boils down to gausslobatto not beeing type stable

[fredrikekre]

JuliaApproximation/FastGaussQuadrature.jl#47

[fredrikekre]

#136