That is the point: I can spend a couple months hand optimizing a hot spot, but for what CPU? The optimizations I'd do for an intel core i7 are different and intel core i9 - and these are generations of the same architecture try to support AMD ryzen and things are very different. Now add in an ARM CPU (which?)...

If I can find a better algorithm I can of course beat the compiler on all of the above. However does such an algorithm exist?