A scheduling framework based on OpenCL kernel tasks

Abstract

The invention discloses a scheduling framework based on OpenCL kernel tasks. The framework includes the feature extraction and feature selection of OpenCL kernel tasks by LLVM compiler, and the compile-time static and run-time features of kernel code are obtained. In the process of feature extraction, the Greedy Feature Selection algorithm is used to select the most important features to avoid over-fitting. Then the selected features are used to predict the task partition ratio between CPU and GPU by using the static classifier in machine learning. At last, the scheduling algorithm is used todivide the tasks into proportions and the available equipment information of the platform, The invention completes the design details of the functions, realizes the algorithm and coding work, realizesthe automatic scheduling of maximizing the utilization of computing resources in the heterogeneous system through the combination of machine learning and scheduling algorithm, and improves the utilization rate of the heterogeneous system resources.

Description

technical field [0001] The invention belongs to the field of computer applications, and in particular relates to a task scheduling framework for realizing task division by using a machine learning method in a heterogeneous system. Background technique [0002] In the past ten years, major processor manufacturers have paid close attention to the field of high-performance computing and tried to develop the latest and most cutting-edge computing accelerator cards. Acceleration cards such as GPUs and FPGAs developed by Intel, Nvidia, and AMD are widely used in various fields. Heterogeneous computing technology has been produced in the 1980s and has the characteristics of high performance and good scalability. In addition, with the current hot accelerator card market and the development of artificial intelligence technology, heterogeneous computing has become one of the research hotspots in the field of parallel computing and distributed computing. [0003] For computing-intens...

AI Technical Summary

Problems solved by technology

However, in order to make full use of the computing resources of PE (Processing Elements) in the target device and write more efficient codes, programmers are required to be familiar with the underlying architecture of the target device, which increases the burden on programmers and requires experience.

Images