Comparing approaches ‐ simple timing comparison

Timings for running sin of a vector

CxxWraps approach:

module DACE
    using CxxWrap

    @wrapmodule(() -> "interfaces/cxx/libdace.so", :define_julia_module)
end

DACE.init(10, 2)

x = DACE.AlgebraicVector{DACE.DA}(1000)
for i in 1:1000
    x[i] = DACE.DA(1, 1.0)
end

function compute(x::DACE.AlgebraicVector{DACE.DA})
    sin(x)
end

@time begin
    for i in 1:1000
        compute(x)
    end
end

ccall approach:

using DACE

DACE.init(10, 2)

x = Vector{DACE.DA}(undef, 1000)
for i in 1:1000
    x[i] = DACE.DA(1, 1.0)
end

function compute(x::Vector{DACE.DA})
    sin(x)
end

@time begin
    for i in 1:1000
        compute(x)
    end
end

comparison

• ccall

  2.035978 seconds (4.86 M allocations: 97.188 MiB, 5.96% gc time, 0.23% compilation time)
  2.070674 seconds (4.86 M allocations: 97.188 MiB, 5.96% gc time, 0.23% compilation time)
  2.067645 seconds (4.86 M allocations: 97.188 MiB, 6.10% gc time, 0.24% compilation time)
  

• cxxwrap

  1.747713 seconds (6.60 k allocations: 392.842 KiB, 0.40% compilation time)
  1.754518 seconds (6.60 k allocations: 392.842 KiB, 0.39% compilation time)
  1.732835 seconds (6.60 k allocations: 392.842 KiB, 0.40% compilation time)
  

Note:

• keep in mind that no time has been spent optimising for performance at this stage
• many more allocations with ccall approach, this can probably be improved and would likely reduce run time