*   TF Core:
    *   `tf.constant` always creates CPU tensors irrespective of the current
        device context.
    *   Eager `TensorHandles` maintain a list of mirrors for any copies to local
        or remote devices. This avoids any redundant copies due to op execution.
    *   For `tf.Tensor` & `tf.Variable`, `.experimental_ref()` is no longer
        experimental and is available as simply `.ref()`.

above, the changes are: // // 1. Negate [B] and [C] so they are positive. The invariant now involves a // subtraction. // 2. If step 2 (both [x] and [y] are even) runs, abort immediately. This // case needs to be handled by the caller. // 3. Subtract copies of [x] and [y] as needed in step 6 (both [u] and [v] // are odd) so coefficients stay in bounds. // 4. Replace the [u >= v] check with [u > v]. This changes the end // condition to [v = 0] rather than [u = 0]. This saves an extra // subtraction due...