Meaning of prefetch_size

Hi, I have some problem with prefetch in memory instructions. I see the ptx document that I can use level::prefetch_size to prefetch some more data by this instruction, for example: cp.async.cg.shared.global.L2::128B.
But I don’t know what the prefetch_size means, is this mean that each thread will prefetch 128B data after current place, or each thread will jump 128B and prefetch the data at that place with the same size of cp.async. Thanks