A method and system for automatic migration and optimization of e-class supercomputing ocean models

Abstract

The invention discloses a method for automatically transplanting and optimizing an E-level supercomputing ocean model, comprising: determining the code data of the hotspot function by using an established code analysis tool based on the characteristics of the hotspot function; using a convolutional neural network to extract the code data of the hotspot function Get out hot function feature, and carry out vectorization processing to hot function feature, to obtain the feature vector of hot function; The feature vector of described hot function is input in the two-way LSTM neural network, and in described two-way LSTM neural network, introduce self- The attention mechanism is trained to determine the optimal two-way LSTM neural network model according to the first master-slave core code data of the output; according to the first master-slave core code data, the reinforcement learning method is used for iterative training to determine the optimal reinforcement learning Model; for the source code data obtained from the client, using the optimal two-way LSTM neural network model and the optimal reinforcement learning model to obtain the optimal master-slave core code scheme to realize automatic code transplantation.

Description

technical field [0001] The present invention relates to the technical field of Internet, and more specifically, relate to a kind of E-level supercomputing ocean mode automatic transplant optimization method and system. Background technique [0002] With the development of marine information technology and the national strategy of "Digital Ocean" and "Smart Ocean" centered on high-speed network, Internet of Things, cloud computing, artificial intelligence and other new-generation technologies in China, marine science and technology has entered a period of rapid development. Human beings have higher and higher requirements for information processing capabilities, and ordinary single-core or multi-core processors can no longer meet the needs of massive computing in the ocean model. The domestic many-core platform is a high-performance computing cluster independently developed by my country. It is based on a domestic many-core processor. Each processor contains four heterogeneou...

AI Technical Summary

Problems solved by technology

[0007] The present invention proposes a method and system for automatically transplanting and optimizing E-level supercomputing ocean models to solve the problem of how to automatically determine the master-slave core code scheme and realize automatic code transplantation

Images