Research On Techniques Of Parallel Program Transplantation Based On Binary Translation

Binary translation can translate an executable program of an instruction set architecture to another instruction set architecture without the source code,which is of great significance in solving the compatibility of old and new systems,isomeric CPU compatibility,system security and the transplant of the legacy code.Binary translation technology has already become one of the hot topics in the modern programming technology,and its analysis of binary code has important guiding significance for improving the performance of the program.However,in the current research of binary translation,most of them focuses on how to improve the performance of translation.Few studies have been made on the translation of parallel computing programs.And the lack of high-performance computing applications is a very important factor restricting the development of domestic processors.In this thesis,after analyzing the execution characteristics of parallel programs,a new feedback static binary translation framework FD-SQEMU is proposed,which improves the translation process of parallel programs,realizes the equivalent translation of parallel applications,and improves the execution efficiency of the software acquired through translation by improving the mode of register allocation.Focusing on how to solve the translation of parallel programs and how to improve the performance of binary translation target programs,the main contents and innovations of this thesis are as follows.1)A feedback static binary translation framework is proposed and implemented,with the aid of the two-level address mapping table,to solve the problem of code discovery and code location in static binary translation effectively.2)In view of the execution mode of parallel program,a translation method of parallel program is improved by combining localization technology of library function,to solve the translation problem of MPI and Pthreads parallel programs.3)A static and dynamic combined register allocation optimization method based on priority is proposed,with the idea of global register static mapping and local register dynamic allocation,which effectively reduces the unnecessary register overflows during the process of register allocation,reduces some redundant memory access instructions and improves the execution efficiency of the target program.Through the experiment,based on the feedback static binary translator FD-SQEMU designed and implemented in this thesis,the registers allocation optimization algorithm based on priority which combined static register mapping and dynamic register allocation have a good ability to solve the cross-architecture software translation problem of parallel programs.And the correctness and effectiveness of the framework and optimization methods were verified.