关键词: 窥孔优化, 循环计数值组合, 循环处理数据合并, 循环最优展开

Abstract:

Software developing in Embedded systems lacks the process of peephole optimization when using the GCC compiler, thus a poorer performance of the compiled code than that by the ARM compiler. This paper proposes a combination of cycle counter, consolidating data of cyclic process and cycle expansion optimally to optimize assembly code on the ARM9 processor, which are verified by running matrix multiplication, image merging and memory settings programs, respectively, on the ARM9 platform. Experimental data show that above mentioned methods reduce the number of register counts by 50% on average while nearly doubling the performance compared with the GCC compiled code.

Key words: peephole optimization, combination of cycle counter, consolidating data of cyclic process, cycle expansion optimally