Identifying Compiler and Optimization Level in Binary Code From Multiple Architectures

While compiling a native application, different compiler flags or optimization levels can be configured. This choice depends on the requirements. For example, if application binary is intended for final release, and settings should set execution speed efficiency. Alternatively, to used debugging purposes, debug configured accordingly, usually involving minor no code optimization. However, this information cannot easily extracted from compiled binary. Nonetheless, ensuring same compilation particularly important when comparing files, avoid inaccurate unreliable analyses. Unfortunately, understand which optimizations have been used, deep knowledge of target architecture required. In study, we present two learning models detect both level in The study are O0, O1, O2, O3, Os x86_64, AArch64, RISC-V, SPARC, PowerPC, MIPS, ARM architectures. addition, x86_64 AArch64 architectures, also determine whether GCC Clang. We created dataset more than 76000 binaries it training. Our experiments showed over 99.99% accuracy detecting between 92% 98%, depending architecture, level. Furthermore, analyzed change amount data was extremely limited. shows that possible accurately flag with function-level granularity.