A Task Parallel Programming Framework Based on Heterogeneous Computing Platforms

Abstract

Along with the development of big data and artificial intelligence, high-performance heterogeneous parallel computing technology has received more and more attention from the industry. On the one hand, heterogeneous computing can significantly improve the computational efficiency. But on the other hand, it can also make the programming more difficult. Such bottlenecks make it harder to give full play to the advantages of heterogeneous hardware. There is currently no comprehensive solution to meet the efficient task scheduling requirements of heterogeneous computing systems. Therefore, this paper introduces a task parallel programming framework based on heterogeneous computing, including the design of programming model, adjusting task granularity, and task scheduling. A ST-HEFT-based static task scheduling method for heterogeneous computing system is proposed to improve computing efficiency. Simulation results show that the framework can obtain better average acceleration ratio. The difficulty of parallel programs for developers can be reduced, and the varying capabilities of heterogeneous components can be fully utilized.