Slight optimisation on cache hit without match.

[Aracthor]

Actually, on cache hit, there is a new pointless dynamic_cast done if ptr is not an instance of required class.

It would be slightly faster to return directly nullptr if there is a cache hit but if this_vtable doesn't match with the stored one.

[tobspr]

I believe this does not work, imagine the following:

// Class hierarchy is A -> B -> C
A* a = new B;
fast_dynamic_cast<B*>(a);
// Cached offset is now A->B, Cached vtable is [A::vftable]

A* a2 = new C;
fast_dynamic_cast<B*>(a2);
// With your change, this returns zero, because
// if(src_vtable_ptr == this_vtable) fails and thus a nullptr is returned

I will let it run through my unit tests on monday though, to give you a definite answer.

[Aracthor]

You are right, I didn't consider that case. Forget about it.