利用控制流识别进行二进制翻译代码缓存压缩

Compress DBT Code Cache Using Control Flow Identification

为了扩展动态二进制翻译产生的本地翻译块,消除因分支目标重复产生的冗余本地代码和返回代码,降低代码缓存的内存占用,提出一种基于控制流识别的代码缓存压缩策略.通过分析投机与非投机2种扩展方式的利弊,结合嵌入式程序分支指令的特征,通过分支方向和偏移量识别对应的控制流;根据控制流选择适合本地翻译块的扩展方式,压缩动态二进制翻译的代码缓存.运行嵌入式基准测试程序时的实验数据表明,在实现该策略后,QEMU二进制模拟器代码缓存的内存占用降低了10％～40％.

Branches with same target produce redundant native code and exit code.A control flow identification based dynamic binary translation （DBT） code cache compression strategy is proposed to remove the redundant code and reduce DBT runtime memory space by extending dynamic generated translation blocks.Two extension methods,speculative and non-speculative,with branch characteristic are analyzed.The strategy identifies control flow with branch direction and offset,chooses extension method based on it,and compresses DBT code cache.The experimental results show that,the dynamic code cache size in QEMU based on this strategy could be reduced by 10％ ～ 40％.