std::move_backward within InlinedVector's ::Insert

[bdash]

I noticed that InlinedVector's Storage::Insert uses an explicit loop to move elements when making space for the inserted elements:

abseil-cpp/absl/container/internal/inlined_vector.h

Lines 799 to 808 in dc242cc

	for (Pointer<A>
	destination = move_assignment.data() + move_assignment.size(),
	last_destination = move_assignment.data(),
	source = move_assignment_values + move_assignment.size();
	;) {
	--destination;
	--source;
	if (destination < last_destination) break;
	*destination = std::move(*source);
	}

I'm curious if there is something about this code that requires the explicit loop rather than using std::move_backward. I can see that std::move_backward was used prior to ad1485c which was about improving the exception safety of insert. It's not clear to me whether the explicit loop is required for some subtle reason or if the use of std::move_backward just got lost during the exception safety refactoring.

The motivation behind this question is that both libc++ and libstdc++ have optimizations within std::move_backward that support lowering to memmove when it is appropriate for the types involved. Using it significantly speeds up insert on vectors of trivial types such as integers. It completely addresses the poor insert performance relative to other inline capacity vectors that others have observed.