IL Optimization: Detecting and Eliminating Redundant Eflags by Flag Relevant Chain

In this paper, we proposed a systematic approach for automatically detecting and elimination redundant Eflags to optimize intermediate language (IL). We analyzed a broad spectrum of different IL and found that a number of IL expose all side effects explicitly by default and that not all the Eflags are relevant with subsequent analysis. Therefore, we proposed a unified approach, invertible analysis, to reduce the volume of IL. Our approach does not rely on any concrete IL, and thus can identify redundant Eflags in the IL. Moreover, we devised a method using flag relevant chain dependency analysis to remove redundant Eflags and shrink the IL. We developed a prototype, and conducted extensive experiments using representative samples from various categories. We demonstrated that our approach could diminish the volume of Vine IL obviously, and provide accurate representation about the assembly code.