Multistage memory access method and discrete memory access method based on heterogeneous multi-core processor

Abstract

The invention discloses a multistage memory access method and a discrete memory access method based on a heterogeneous multi-core processor. The multistage memory access method comprises that an auxiliary core reads in data needed by a first time calculation; besides the first time, the auxiliary core reads in data needed by a next time calculation while performing the data calculation; besides the last time, the auxiliary core writes back result data of the last calculation while performing the data calculation; and the auxiliary core writes back result data of the last time calculation. The discrete memory access method includes that storage orders of digit groups are adjusted on the auxiliary core, and discretely-stored digit groups are adjusted into continuously stored digit groups; and communication read-in data, calculation and communication write-back data are performed on the digit groups through the auxiliary core by the multistage memory access method. According to the multistage memory access method and the discrete memory access method based on the heterogeneous multi-core processor, maximum hides between expenditures and multistage storage resource access expenditures are effectively calculated, the multistage memory access efficiency is improved, the performance efficiency of the auxiliary core is fully developed, and accordingly, the calculation efficiency of a scientific calculation subject is enhanced.

Description

technical field [0001] The invention relates to the field of computers, in particular to a multi-level memory access method based on heterogeneous many-core processors and a discrete memory access method based on heterogeneous many-core processors. Background technique [0002] In recent years, high-performance computing software has developed rapidly, especially various scientific computing software. These scientific computing software are distributed in various disciplines such as astronomy, meteorology, earth simulation, aerospace, fluid mechanics, basic scientific research, etc. The problems studied are becoming more and more complex, and the requirements for high-performance computing resources are getting higher and higher. While scientific computing software has achieved great success, it also poses a huge challenge to today's high-performance computers. [0003] Because of its higher parallel execution performance and lower average power consumption, many-core proce...

AI Technical Summary

Problems solved by technology

[0009] The technical problem to be solved by the present invention is that the memory access efficiency is low when facing multi-level storage resources under the heterogeneous many-core architecture, and the computing resources are not fully utilized due to the low-efficiency memory access, and the performance of the coprocessor cannot be fully utilized. The problem

Images