[ChrisRackauckas]

Do you have code to share and look at? All we can do is point to real-world code and benchmarks. For this specific case, see LoopVectorization.jl results (https://juliasimd.github.io/LoopVectorization.jl/latest/exam...), and the corresponding effects on stiff ODE solver benchmarks against C and Fortran packages (https://benchmarks.sciml.ai/html/StiffODE/Hires.html).