Pipeline parallelization method for coarse-grained streaming application

Abstract

The invention discloses a pipeline parallelization method for coarse-grained streaming applications. The pipeline parallelization method includes performing classic data profiling and dependency analysis on serial C-codes to acquire a task dependence graph, performing dependence transformation on the task dependence graph to acquire a directed acyclic graph, building a system feature graph, performing task scheduling on the directed acyclic graph according to the system feature graph and judging whether a task scheduling result meets performance requirements or not, if not, then aggregating and splitting task of the directed acyclic graph to acquire a new directed acyclic graph, selecting and calculating the highest-cost task of the new directed acyclic graph to acquire a new calculated hot spot region, returning to performing the dependency analysis again, segmenting and modifying the serial C-codes according to the task scheduling result so as to obtain parallelized C-codes, encoding to generate parallel executable files through an encoder, and loading the parallel executable files to a target hardware platform to execute. The pipeline parallelization method is adaptable to multilayer nested loop structures and capable of extracting parallelism of the multilayer loop.

Description

technical field [0001] The invention relates to the field of computer applications, in particular to a method for realizing parallelization of flow for coarse-grained flow applications. Background technique [0002] In order to make full use of the resources of the multi-core system, the problem of parallel programming must be solved. Due to the popularity of the C language and the established serial programming thinking of programmers for a long time, a large number of C program codes are left behind, and these applications are often used as upper-level applications or software tools for multi-core systems. About 85% of embedded system developers are still using C / C++ to write programs. If programmers are asked to rewrite this type of application with a new parallel programming language, the development will be difficult and the development cycle will be long. Therefore, there is an urgent need to improve the parallel execution efficiency of serial C programs on multi-cor...

AI Technical Summary

Problems solved by technology

However, the above methods have the following defects: 1) They are mainly used to extract fine-grained thread-level parallelism, and they adopt an aggregation method for the processing of control dependencies, which is extremely easy to form too large a granularity for coarse-grained applications 2) The above methods are only suitable for the development of the innermost loop, and the effect is not ideal for the structure of multi-layer nested loops; 3) DOACROSS technology and DOALL technology are only suitable for regular storage access and simple control-dependent applications, similar to scientific computing, while most applications have irregular control flow and complex storage access, such as pointer access

[0005] There are also some research works to extract parallelism from the structural analysis of the source program, but this kind of work often focuses on modeling the program structure and collecting information, and lacks the transformation of the dependencies between various modules. Aspects rely on programmers to complete manually without systematic methods for guidance

Images