We can convert all load/store instructions into register-based (no offset) and put the memory access in parallel with the ALU. What is the clock cycle time if this is done in the single-cycle and in the pipelined datapath?