Studies On Dynamic Optimization Technology And Its Aplication In SkyEye

Currently, dynamic binary translation has been paying more and more attention in the study field, for it could resolve the problem of software's cross-platform, which enables the binary programs, among different processors, to be easily transplanted to each other. Moreover, it expands the application scope between the hardware and software, and breaks the interacting situation between the processor and supporting software. The translator could translate the code at the run-time, based on information that collected at the run-time optimizing the frequently executed critical code blocks, it is also able to conduct dynamic optimization, so as to enhance the implementation performance for dynamic binary translation system.The hot paths-based dynamic optimization is an effective method for improving the operational efficiency of the software in dynamic binary translator. This study focuses on how to identify the hot paths by using the existing limited amount of previous operational information in basic blocks, raising the hit rate of the prediction, and on ensuring no increase in computational. In fact, there are few model-based methods with complicated implementation among existing hot paths predictive algorithms,therefore, based on Hidden Markov Model, an improved hot paths predictive algorithm is proposed in this paper. Firstly, Constructing the procedure diagram based on basic block. By using of process diagram, based on basic block, to represent the program's process flow, which is conducive to the research of execution path. Then, Extending procedure diagram through Hidden Markov Model. Enabling the procedure diagram meets the characteristics of Hidden Markov, in which the research on hidden Markov-based hot path predictive model derives. For the next, Constructing predictive. model through Hidden Markov Model, proposing an improved hot path recognition algorithm, which is easy to implement its unique sequence of state transition. At last, the efficiency of algorithm is verified by the experimental paltform SkyEye, results show its validity. This algorithm can improve the hit rate of hot path prediction and the performance of the dynamic binary translator without increasing the predictive delay at the same time.Constantly switching between the translation and implementation engine appears in the process of dynamic translation and implementation, thus resulting in a lot of schedule overhead. Instruction jump is the main cause to produce the switching. Therefore, the costs of translation and implementation could be significantly reduced provided we deal with the instruction jump efficiently.Direct jump Technique is applied in dynamic binary translation system (DBCT) of experimental platform-SkyEye, However, direct jump is only a small part of the jump instruction, so it can not efficiently reduce overhead caused by the jump in SkyEye. Only if the conditional branches jump instruction occupies a higher proportion in jump instruction is optimized, could it reduce the overhead efficiently generated by the jump instruction. For this reason, this paper deals with the conditional branches jump issue that appears in DBCT through applying that of similar method used in DynamoRIO system. The experimental result shows that the SkyEye's operating efficiency could be improved by the application of conditional branch jump optimization.