Research On Optimization Of Conflict Detection Algorithm In Systemc Parallel Simulation

During the execution of SystemC parallel programs,race detection is an important means to detect potential conflicts between different threads to improve parallelism.Because multiple modules run concurrently in hardware design,modules share the same set of hardware resources(such as shared memory or shared bus),so there are race conditions.If these race conditions are not managed correctly,they can lead to errors in the design or failure of functionality.Race detection can not only find potential problems in the program but also reduce waiting time and resource waste in the system by detecting and avoiding race conditions,thereby improving the performance of the system.Although the early parallel discrete event simulation methods have improved the degree of parallelism,they have many limitations and cannot maximize parallelism.The out-of-order parallel discrete event simulation can break the limitation of the delta-cycle and further improve this effect,but it also requires a more stringent conflict detection algorithm to ensure the correctness of program execution,existing detection methods still have obvious deficiencies.On this basis,aiming at SystemC race detection algorithm based on out-of-order parallel discrete event simulation,this paper proposes an improved data race detection algorithm based on array element granularity.Based on the object granularity detection.The algorithm focuses on the optimization and improvement of the race detection of array structure access.To solve the problem that the conflict of data elements cannot be correctly determined,the algorithm combined with the built-in method of ROSE compiler to redesign the detection method in the segment graph construction stage,which can effectively eliminate some false positive conflicts in the fine-grained detection method.Make the test results more accurate.Based on the RISC project,the algorithm is improved and verified by experiments.The improvement works are applied to both static compile time and dynamic run time.In static conditions,the situation can be judged to be limited but can be comprehensively detected;in dynamic conditions,more information can be collected but the coverage is limited,and the two can be combined to obtain more accurate conflict results.The results show that the improved algorithm proposed in this paper can obtain more accurate race detection results with almost no extra time,ensure the correctness of the detection results and improve the parallel performance.