Hopcroft, Ullman Theorem 9.7 Reductions

Question

There exists a language L_d = {M | M doesn't belong to L(M)}. L_d is the collection of Turing machines (programs) M such that M does not halt and accept when given itself as input. It is known and proved that L_d is a non-Recursively Enumerable Language.

Now L_d can be reduced to the problem A_TM (Universal Language) as follows "To see if w ∈ L_d check if <w, w> ∈ A_TM."

The theorem 9.7 (Page 393) in Hopcroft Ullman states that if there exists a reduction from P1 to P2, then: if P1 is non RE then so is P2.

According to this theorem, since L_d can be reduced to A_TM, and L_d is non-RE, A_TM must be non-RE. But we know that A_TM  is Recursively Enumerable and undecidable. How is this possible?