Avoiding allocations of small but non-trivial arrays (work array alternative?)

General Usage Performance
39

Looks like the multithreading is causing a type instability since order isn’t being constant-propagated across the task boundary. I’d rewrite it like this:

function run_v2_static_view(::Val{order}) where {order}
    A = LinRange(0,1,1_000_000)
    F = A.^4
    t = 0.3
    Threads.@threads for i ∈ 1:10000
        xnodes = SVector{order}(@view(A[i:i+order-1]))
        fvals = SVector{order}(@view(F[i:i+order-1]))
        u = v2(xnodes,fvals, t)
    end
end

julia> @btime run_v2_static_view(Val(6))
  3.371 ms (39 allocations: 7.63 MiB)
1 Like