Codelet Scheduling Method Based on Genetic Algorithm

Abstract

The invention relates to a Codelet dispatching method based on a genetic algorithm. A Codelet data flow calculation model is used as a basis, a genetic algorithm dispatching method based on a multi-core processor parallel system and Codelet model task dispatching are combined, the encoding mode of an explicit two-dimensional array is adopted in the dispatching method when a task dispatching problem is solved, by means of hybridization, variation and other genetic operator operations, and the two arrays are fused to generate a new array serving as solution of the next generation. The dispatching method can effectively improve parallel task execution efficiency, reduce idle time of an execution unit, and increase the resource utilization rate of an on-chip multi-core system. The method can be extensively applied to data flow models of other types, and parallel task execution in the multi-core system is achieved.

Description

technical field [0001] The invention relates to the design of a task scheduling method for a multi-core data flow computer system, in particular to a Codele scheduling method based on genetic algorithm. Background technique [0002] For a long time, the mainstream computer system execution model is the Von Neumann computational model centered on the control flow. This model is characterized by placing a series of sequentially executed instructions into the same address space, controlled by a single program counter. With the emergence of distributed storage computer models and multi-core systems, parallel computing and parallel memory access have become the basic requirements of heterogeneous multi-core computer systems, and the application of big data algorithms has further promoted the development of large-scale data flow computing models. [0003] The data flow computing model is widely regarded as a potential alternative to the traditional parallel computing model center...

AI Technical Summary

Problems solved by technology

[0006] The present invention aims at the problems of poor adaptability of the original static scheduling and dynamic scheduling of the Codelet model and low concurrent execution of the system, and proposes a Codelet scheduling method based on a genetic algorithm. Based on the Codelet data flow calculation model, a method based on The codelet scheduling method of the genetic algorithm is applied to the actual codelet model. When solving the task scheduling problem, the scheduling method adopts the coding method of the explicit two-dimensional array, and through the operation of genetic operators such as hybridization and mutation, the two arrays " Fusion" produces new arrays as the next generation of solutions

Images