Redundant Instruction Optimization Algorithm in Binary Translation

Binary translation is a main method to implement software migration. Dynamic binary translation is limited by dynamic execution and cannot be deeply optimized, resulting in low efficiency. Traditional static binary translation has difficulty to deal with indirect branch, and conventional optimization methods mostly affect in the intermediate code layer, paying less attention to a large number of redundant instructions that exist in the target code. According to this situation, this paper presents a static binary translation framework SQEMU and a target code optimization algorithm to delete redundant instructions based on the framework. The algorithm generates an instruction-specific data dependence graph(IDDG) based on the analysis of target codes, then combines liveness analysis with peephole optimization using IDDG, and effectively removes redundant instructions in target codes. Experimental results show that using the optimization algorithm for target codes, the execution efficiency is significantly increased, the maximal increase up to 42%, and the overall performance test shows that the optimized translation efficiency of nbench is increased by about 20% on average, and it is increased about 17% of SPEC CINT2006 on average.