used get_irn_op() instead of get_irn_opcode()
used get_irn_n(*, -1) instead of get_nodes_block(), so we
can use this optimization while nodes are floating

[r6132]