Compiler-directed Static Analysis And Code Transformation

With the rapid development of CPU production process, the core number of CPUdevelops from the original single-core, dual-core to the quad-core, eight-core or even more.In order to improve processing speed, multi-core runtime system emerges to effectivelyorganize existing physical core effectively. Multi-core runtime system can effectivelyorganize the multi-core computing resources, allocation and scheduling the computingtasks at run time and provide users with high-performance computing services. Staticanalysis technology is a key component of the compiler background, analysis and processthe program behavior without the execution process, generate the analysis result related tothe code. Lacking of effective analysis of the target code, most of existing schedulingpolicy of the runtime system increases the system overhead, thereby reducing theperformance of the system.By taking the result of the analysis system and user’s compiler-directed informationinto consideration, this paper presents an analysis based task management strategy,through making full use of the static program analysis and the code transformation,analysis based strategy want to reduce the overhead of task management. Consider thetask creation overhead and the computing power of core, most of the cores can maintain arelatively low idle rate within the specified time during task execution. In static programanalysis phase, the analysis system does the optimization work according to the degree ofparallelism and the execution plan. Through the code safety test to ensure the correctnessof the code transformation process. In run time phase, the choice of the task executionplan considering the depth information of the task node and the computational overhead ofthe compute cores, select the appropriate enforcement response calculation request. Thesystem dynamically manages tasks running status, reduce the overhead and ensure thecorrectness of the calculation.Through the simulation of the GOMP runtime system, the results show that theaverage running time of static analysis based task management strategy is18%lower thanthe GOMP built-in task management strategies, the CPU utilization is16%higher than thebuilt-in task management strategies in GOMP. The experiments further show that the useof static analysis and code transformation of great help to enhance the performance of theruntime system.