In mobile and embedded devices, the energy supply is strictly constrained with the battery capacity and energy saving ability. In these energy-constrained settings, the available energy budget is not sufficient to meet the optimal performance objective. This paper presents an energy-constrained software prefetching optimization approach, which can obtain the optimal performance under the limited energy resource. The approach is based on DVS-enabled CPU and memory. Through inserting frequency-scaling instructions, CPU and memory frequencies are simultaneously adjusted to meet two performance objectives (one is the time; the other is the processor gain) under a given energy constraint. A detailed analytical model is built and then the effectiveness of the approach is validated by a set of array-intensive applications. Experimental results show that the approach is effective for energy-constrained prefetching optimization problem.