[IA64] optimize flush_tlb_range on large numa box