ALMP: a shifting memory architecture for loop pipelining | IEEE Conference Publication | IEEE Xplore