Basic algorithm architecture supporting distributed parallel computing and computing method

Abstract

The invention discloses a basic algorithm architecture supporting distributed parallel computing and a computing method, and the basic algorithm architecture comprises a local algorithm library which is arranged in an application computer and comprises a plurality of programming language algorithm libraries; the CPU equipment node agent program is in communication connection with the CPU computing node of the application computer; the GPU equipment node agent program is in communication connection with the GPU computing node of the application computer; the FPGA equipment node agent program is in communication connection with the FPGA computing node of the application computer; the virtual computer resource management program is in communication connection with the local algorithm library, the GPU equipment node agent program, the CPU equipment node agent program and the FPGA equipment node agent program, and the virtual computer resource management program is used for selecting different equipment node agent programs for calculation according to the complexity of the calculation task of the application program. The system can support multiple programming languages, not only supports a CPU, but also can support GPU and FPGA computing platforms, and is high in computing efficiency.

Description

technical field [0001] The present invention relates to the technical field of basic software algorithm library, in particular to a basic algorithm architecture and computing method supporting distributed parallel computing. Background technique [0002] There are more and more standardized scientific computing libraries in the fields of machine learning, data mining, weather forecasting, image processing, signal processing, etc., such as: MKL algorithm library, FFTW algorithm library, OpenBLAS algorithm library, ArrayFire algorithm library, Eigen algorithm library , NumPy algorithm library, etc. However, there are more or less problems in the use of these algorithm libraries. For example, cross-node distributed parallel computing capabilities are weak, and usually cannot support multiple programming languages, and often only support CPU computing and GPU computing, and only support parallel computing of a single physical node, and the computing efficiency is low. Content...

AI Technical Summary

Problems solved by technology

[0003] The purpose of the present invention is to overcome the above-mentioned technical deficiencies, propose a basic algorithm architecture and calculation method supporting distributed parallel computing, solve the problem that the programming language supported by the scientific computing library in the prior art is single, and only supports parallel computing of a single physical node. Computationally inefficient technical issues

Images