Per-thread singleton-like using thread_local storage

Is there any caveats of this usage of thread_local storage duration:

template <class T>
inline T &thread_local_get()
{
  thread_local T t;
  return t;
}

Then in different threads (for example)

thread_local_get<float>() += 1.f;

The doc at cppreference says this about thread local storage duration:

thread storage duration. The object is allocated when the thread begins and deallocated when the thread ends. Each thread has its own instance of the object. Only objects declared thread_local have this storage duration. thread_local can appear together with static or extern to adjust linkage.

Does this correctly allocate one thread_local instance for each T (during compilation) and each calling thread ? Is there any situation that can lead to e.g undefined behavior ?