Research On Energy Optimization Of Register Spilled Based On SPM

With the technology of integrated circuit design and manufacturing processes continuously improving, a leap-type development has been made in the performance of Embedded Systems with system-on-chip as the core .At the same time, however, energy consumption becomes an important bottleneck in the development of embedded systems. According to former researches, storage subsystem is main energy sources of system-on-chip, how to effectively reduce the energy consumption of storage subsystem has become the key to decreasing the whole system-on-chip energy, and the hierarchical architecture of storage subsystem with the SPM has emerged as the effective means of reducing energy consumption.With the effect on the storage subsystem's energy consumption based on SPM as the core, a SPM-based register allocation algorithm spill management strategy and a SPM-based -layout storage subsystem energy optimization strategy are introduced in this thesis.Different from variables to be spilled on off-chip memory, when no register will be available in Graph Coloring Register Allocation process, a SPM-based register allocation algorithm spill management strategy is introduced in this thesis. By analyzing the static single assignment (SSA) form generated in the phase of application compiler, before and after spilled variables SPM access instructions (sts/lfs) are inserted, SPM instead of off-chip memory become the target spill space, SPM access instructions assisted by SPM Manager, SPM Manager make spilled variables stored or loaded from SPM space. It reduces energy consumption of access by placing spilled variables in SPM space.Given the limitation and localization of the storage subsystem' energy consumption by the management of spilled variables in the SPM, a storage subsystem energy optimization strategy based on SPM layout optimization is presented in this paper. Generally considering the energy consumption of application program global variables, data, code, stack on the system, it will construct weight extended control flow graph (WECFG) reflecting the relationship of procedure calls and the number of visits, calculate the node energy density in the WECFG, and using 0-1 knapsack algorithm with energy density as priority to select energy consumption hot spots in program. This thesis divides SPM space into two layers: the first layer for spilled variables storage, the second for storage of hot energy consumption. It aims at reducing the energy consumption of storage subsystem through the rational layout of application program in the SPM space.The experimental results show that the above-mentioned strategy, variables spilled management in SPM and SPM spatial layout optimization, achieved about 4.24% and 34.8% energy consumption optimization in embedded systems respectively.