ARC processor can use LP instruction to implement zero overlay loops.
The current inplementation doesn't handle the unlikely situation when
the loop iterator is located in memory. Refurbish the loop_end insn
pattern into a define_insn_and_split pattern.