With multiple inheritance and virtual functions, the derived class will contain multiple vptrs. In essence a derived class contains each of its base classes as subobjects, which in turn have different addresses within the derived class. So when used as one base class it has a different address than when used as a different base class. Code that uses the derived class as a base class just references the base class vtable as normal. To handle differences in the address of the base subobject vs. the natural address of the derived class, the vtable in addition to the pointer to member function also will contain a pointer offset, or a jump to a thunk that fixes up the this pointer when accessing the virtual function.

For more information you can read Design and Evolution of C++ by Bjarne Strostrup or the C++ Object Model by Stanley Lippman.

It may or may not need to change a pointer to a derived class when converting to a base class. In general, compilers generally have one base subobject be at the same address as the natural address of the derived class. When converting pointers between these two types no change in address needs to be done. However conversion between pointers of other base classes will need to have a pointer update done. This is one reason why reinterpret_casts can be so dangerous, since a reinterpret_cast may cause a valid pointer to a derived class to be converted to a invalid pointer to a base class even if a static_cast would have done the right thing.