Method and device of pre-fetching data of compiler

Abstract

The invention provides a method and a device of pre-fetching data of a compiler so as to achieve efficient visitation and storage optimization. The method includes providing divided local storage spaces in a storage scratch pad memory (SPM) on a hardware instruction inquiry software management sheet, wherein the hardware instruction includes a main storage address of data; reading corresponding data of the main storage address if the corresponding data of the main storage address can be inquired in the SPM local storage spaces; and carrying out un-hitting processing according to the hardware instruction so as to pre-fetch the data to the SPM local storage spaces if the corresponding data of the main storage address can not be inquired in the SPM local storage spaces. According to the method of pre-fetching data of the compiler, high efficiency and software flexibility of the hardware are fully utilized, and the efficient visitation and storage optimization are achieved with little hardware expenditure. On the basis of realization of the efficient visitation and storage optimization, the method of pre-fetching data of the compiler is achieved, and the method combines the hardware support and a simplified pre-fetching algorithm, and is based on a software and hardware coordinated management mechanism. The method is superior to a pure software algorithm in performance, and flexibility of a software algorithm is guaranteed.

Description

technical field [0001] The invention relates to the field of data processing, in particular to a compiler data prefetch method and device. Background technique [0002] Due to the complex hardware logic of the cache (Cache), the Cache structure cannot be accommodated in some multi-core structures, and it is replaced by a software-managed on-chip memory (Scratch Pad Memory, SPM) with a simple hardware structure, small footprint, and lower power consumption. While SPM brings high performance and low power consumption, it also puts forward higher requirements and challenges for programming model and compilation optimization. Cache can automatically maintain data consistency by hardware, while SPM requires software to maintain consistency and manage data movement between storage systems at different levels. On the premise of ensuring program correctness, make full use of data localization To ensure the stability and local memory space, develop calculation and memory access in p...

AI Technical Summary

Problems solved by technology

[0004] However, the above-mentioned prior art lacks the synergistic support of both software and hardware, and the overhead of its software algorithm will have a great impact on the overall performance. In terms of hardware, SPM cannot complete some basic functions of Cache, causing a large burden on programmers.

Images