Dynamic probability based instruction scheduling for low-power embedded system

Abstract

Power is a very important indicator in embedded system. In this paper, for decreasing the power of the embedded system, we use an efficiency method that does work in the code generate of compiler to solve the instruction scheduling and reordering problem. The instruction scheduling is a NP-Hard problem. For solving this NP-Hard problem, we design a proper power model to evaluate the power. And the multiple iterations of dynamic probability are used for the instruction scheduling selecting. The main contribution of our algorithm is that we make a dynamic probability based on the power of closed instructions to decide which is the better scheduling, and then through multiple iterations to select the best one for our final result.