A high-performance heterogeneous multi-core shared cache buffer management method

Abstract

The invention discloses a heterogeneous multi-core sharing cache buffer management method for high performance. According to the method, first, a buffer with the same structure as a sharing final L2-level cache (LLC) is established at a GPU side, and a GPU message accesses the buffer first, so that the purpose of filtering a GPU flow request is achieved, and LLC space is emptied for a CPU application program. On the basis of adding the buffer, a reasonable replacement policy is adopted according to different characteristics of the CPU application program and a GPU application program to increase the cache hit rate. Finally, the size of the buffer is adjusted, the size of the buffer is changed before operation according to IPC division indicators, an optimal performance scheme is found, and therefore the purpose of improving system performance is achieved.

Description

Technical field [0001] The invention belongs to the field of computer system cache system structure, and specifically relates to a high-performance heterogeneous multi-core shared cache buffer management method. Background technique [0002] With the advancement of semiconductor technology and insurmountable obstacles such as physical limits and power consumption encountered by single processors, profound changes have taken place in architecture technology. After continuous research and development in recent years, the advanced architecture represented by multi-core processors has gradually replaced single-core processors as the main way to improve processor performance. A multi-core processor integrates multiple processor cores in a chip. These processor cores have the same or different functions and structures. They are integrated in the same chip in an effective way, and applications are distributed in an effective way. Parallel processing is performed on different microproce...

AI Technical Summary

Problems solved by technology

However, the existing cache management work is mainly used in the homogeneous multi-core system environment, which cannot adapt to the heterogeneous environment where the CPU and GPU are combined, let alone distinguish the requests from the CPU and the requests from the GPU, resulting in a shared last-level cache The unfairness of allocation seriously affects the performance and power consumption of the system

Images