Automatic Parallelization for Binary on Multi-core Platforms

While1 multi-core systems are becoming popular, many serial binaries exist. This paper proposed a method to exploit parallelism in serial binaries and automatically parallelize them on multi-core systems. We extended existing compiler optimization techniques to binaries and developed extra approaches to further improve the performance. Preliminary results on Polybench benchmark and our hand gesture recognition program showed an average speedup of 1.8 and 1.4 over serial performance respectively on an armv7 machine with 2 threads.