Numerical calculation software acceleration method based on CPU-GPU (Central Processing Unit-Graphics Processing Unit) collaboration

Abstract

The invention discloses a numerical calculation software acceleration method based on CPU-GPU collaboration, and the method comprises the steps: dividing a program into basic blocks, and estimating the operation information of the basic blocks through a code analysis module, including whether the basic blocks are calculation-intensive tasks or not, whether the basic blocks are easy to parallel or not, and whether the memory copying time of the calculation tasks is less than the execution time of a CPU or not. And marking the basic blocks meeting the conditions as GPU modules according to an analysis result, and marking the rest as CPU modules. The GPU module is instantly compiled into GPU codes to be executed, if hot codes exist in the CPU module, the CPU module is instantly compiled into machine codes to be executed, and if not, the CPU module is interpreted to be executed. According to the method, a CPU-GPU collaborative instant compiling mode is adopted, CPU and GPU computing resources are fully utilized, the operation efficiency is higher than that of independent execution of the CPU or the GPU, meanwhile, the advantage of instant compiling of an interpreted language is exerted, and the performance of numerical calculation software is remarkably improved.

Description

technical field [0001] The invention relates to the technical field of heterogeneous computing, in particular to a numerical computing software acceleration method based on CPU-GPU cooperation. Background technique [0002] CPU-GPU collaboration is applied to parallel computing tasks. There are two types of collaboration. One is that the CPU is responsible for managing the GPU, providing calculation data and receiving calculation results for the GPU, and the GPU undertakes all calculation tasks; the other is that the CPU manages the GPU while being responsible for a part computing tasks. [0003] Numerical calculation software solves thread and nonlinear problems numerically, uses high-level programming language, simple coding, and runs in the way of interpretation and execution based on CPU. Taking Octave as an example, Octave has a grammar specification that is basically consistent with Matlab, and is weakly typed, but the interpretation and execution efficiency is low. ...

AI Technical Summary

Problems solved by technology

[0006] At present, the numerical calculation software execution method does not give full play to the advantages of hardware. If it is based on CPU-based just-in-time compilation, GPU computing resources will be wasted; if CPU-GPU collaborative parallel computing is used, the performance improvement brought by interpreted language just-in-time compilation will not be fully utilized.

Images