浮点到定点的高效翻译策略研究

High-performance Floating-point to Fixed-point Translation Strategy

动态二进制翻译中,在目标平台没有浮点部件、不支持浮点运算的情况下,浮点指令只能通过模拟解释执行。浮点指令的解释执行造成翻译系统效率急剧下降。通过将浮点指令运算转化为定点运算,解决了浮点指令在目标平台的翻译,为浮点指令的翻译找到了新的途径。在动态二进制翻译系统中进行实验,验证了翻译方法的可行性。实验显示翻译系统的性能有明显提升,含有浮点指令的比例越高,算法能够获得的加速比越高,对含有25%浮点指令的程序,该算法能达到1.55的加速比。

When doing dynamic binary translation,if no floating-point component exists in the target platform so that it does not support floating-point operations,floating-point instructions can only be interpreted through simulation.Interpretation of floating-point instruction causes a sharp translation efficiency decrease.This article addresses a method that translating floating-point operations into fixed-point operations,which gives a new way to solve floating-point instruction translation in the no-floating target platform.Implementation of this method in a dynamic binary translation system proves the feasibility of the translation algorithm.The experiment also shows significantly the improvement of performance and the higher the proportion of floating-point instruction,the higher the speedup.And the speedup is up to 1.55 when the binary code contains 25% floating-point instructions.