Optimistic time management method based on GPU

Abstract

The invention discloses an optimistic time management method based on a GPU. The optimistic time management method includes the steps that S1, simulation initialization is carried out to generate a plurality of initialized simulation events; S2, whether simulation is finished or not is judged, if the simulation is finished, an algorithm is finished, and if the simulation is not finished, the step S3 continues to be executed; S3, a time window is built; S4, the events are extracted, wherein threads of the GPU inspect all the simulation events in parallel, and if timestamps of the simulation events are located inside the time window, the simulation events are added into target logic processes FEL; S5, forward event processing is carried out, wherein the threads of the GPU process the events in the logic processes FEL in parallel; S6, reverse event processing is carried out, wherein if cause-effect disordering occurs, repairing is carried out through reverse events; S7, event submitting is carried out, wherein all the processed events are submitted and affirmed, all logic processes PEL are set to be empty, and the step S2 is executed. The optimistic time management method has the advantages that the system parallelism degree can be increased, and the performance advantages of the GPU are efficiently exploited.

Description

technical field [0001] The invention mainly relates to the technical field of modeling and simulation, in particular to a GPU-based optimistic time management method. Background technique [0002] In recent years, due to various constraints such as system energy consumption, storage performance, and instruction set parallelism, the traditional method of obtaining higher-performance processors by increasing the main frequency is no longer feasible, and many-core technology has become a new trend in processor development. Among them, the GPU-based high-performance computing platform is becoming an important trend in the development of high-performance computers due to its powerful computing capabilities, low power consumption, and sustainable growth potential, and it also provides new opportunities for accelerating complex system simulation applications. Since the GPU is essentially an on-chip parallel processing cluster composed of many processing units, it must support paral...

AI Technical Summary

Problems solved by technology

[0004] However, since a GPU usually has hundreds of parallel processing units and needs to hide instruction execution delays through a large number of thread switching, its performance is highly dependent on the degree of parallelism

However, the synchronous conservative time management algorithm is too pessimistic, and its window size limit makes the algorithm only able to deal with the events that "certainly no causal disorder will occur", ignoring those events that "theoretically may cause causal disorder but not in practice". Events greatly limit the number of events that can be processed in parallel, making it difficult to effectively tap the computing potential of the GPU

Images