Multithreaded Dynamic Binary Trans Lation Optimization Framework Based On Multi-core Platform

During dynamic software system's implementation, effective optimi-zation algorithms always introduce into some extra runtime overhead while improves the system performance [Massimiliano99, Sorav08]. In traditional static compiler's implementation, all optimizations are per-formed before the complied and linked execution file is created, so the overhead of performing these optimizations will not affect the generated file's execution time. But when it comes to DBT (dynamic binary transla-tion) system, all overhead spends on translate, optimize and execute will be counted into the final performance result. This fact will make our DBT system's optimization design must be effective and low-cost.In order to overcome this shortcoming, we use multithreaded optimi-zation mechanism to reorganize dynamic binary translation system's framework, the main thread is used to translate basic blocks, execute and link all the code blocks, while the help optimization thread is used to build, optimize and link the super blocks. All the threads are running on a multi-core processor to speed up. We innovatively proposed a well-suited arguments communication mechanism in dynamic binary translation sys-tem called ASLC(Assembly Language Communication), we also de- signed code management mechanism using multiple code caches com-bined hash function in our multithreaded framework. Compared with the original framework, multithreaded parallel execution make software rea-lized profile instruction more executed to achieve more accurate profile information[Jeffrey97], in the same time, the overhead of super block building[Vasanth00] has been hidden by the other modules'execution. According to our experiment, multithreaded dynamic binary translation system based on multi-core processor can improve almost 4 percent to 8 percent of system performance compared with the original implementa-tion.