Heterogeneous many-core processor-oriented multi-task parallel scheduling method

Abstract

The invention discloses a heterogeneous many-core processor-oriented multi-task parallel scheduling method, which comprises the following steps of: S1, creating a task queue for storing tasks to be subjected to accelerated calculation; S2, when the task runs, enabling a main core to recover the completed task on a slave core and add a new task to the tail of the task queue; S3, grouping the slavecores according to proper granularity, and actively acquiring tasks from the head of the task queue by taking a group as a unit; S4, if a group of slave cores obtain a part of calculation of the current task, increasing the number of existing calculation resources of the current task; and S5, when the number of the computing resources required by the task is equal to the number of the existing computing resources, completing the task allocation. The task scheduling efficiency is improved, the fine management of the computing resources is realized, the problem of waste of the computing resources caused by many computing cores of a many-core processor, insufficient parallelism of practical application and unbalanced computing load is solved, and the utilization efficiency of the resources isimproved.

Description

technical field [0001] The invention relates to a multi-task parallel scheduling method for heterogeneous many-core processors, belonging to the technical field of computers. Background technique [0002] Heterogeneous many-core processors are generally divided into operation control components (hereinafter referred to as the main core) and accelerated computing components (hereinafter referred to as slave cores). The main core is responsible for task generation and unified scheduling, and loads the code and data of tasks to the Accelerate computing on the core. Before scheduling an accelerated task, the main core checks whether each slave core is ready to receive a new task; after the acceleration task is completed, the main core also checks whether the calculation of each slave core is completed. If there are multiple tasks that need to be accelerated, the main core will be scheduled in sequence according to a certain queuing order of the tasks. [0003] However, the exi...

AI Technical Summary

Problems solved by technology

[0004] The purpose of the present invention is to provide a multi-task parallel scheduling method for heterogeneous many-core processors, which improves the efficiency of task scheduling and realizes fine management of computing resources. It also solves the problem of waste of computing resources caused by many-core processors with many computing cores, insufficient parallelism in actual applications, and unbalanced computing load, and improves resource utilization efficiency

Images