Prospective multi-threaded parallel execution optimization method

Abstract

The invention discloses a prospective multi-threaded parallel execution optimization method which comprises the following steps that: (1) a serial program is divided into a plurality of threads; (2) whether the threads are prospectively executed is judged, and whether writing and reading data dependency exists between each thread and the direct precursor thread is analyzed for the prospectively executed threads; (3) whether the threads are prospectively executed with the set ratio of producer distance to consumer distance; and (4) the prospectively execution results are verified after prospective thread execution. The method can reduce the data dependency conflict and unnecessary delay waiting when in thread parallel execution; and the invention comprehensively considers a cycle part and a sub function part in the serial program which have higher parallel efficiency, can fully explore the writing and reading data dependency in the program, makes full use of the thread-level parallelism, selectively carries out prospective execution by quantizing writing and reading data dependency between the threads, and can improve the execution speed and the serial program on a multi-core processing unit and the utilization rate of the processing unit.

Description

technical field [0001] The invention relates to the computer field, in particular to a look-ahead multi-thread parallel execution optimization method. Background technique [0002] With the deepening of computer applications, the technology of single-chip multi-processors (Chip Multi-Processors, CMP) has attracted more and more attention. The key to using CMP technology to improve system performance is to give full play to the parallel advantages of multiple microprocessor cores. Although serial programs are easy to design and debug, it is difficult to take advantage of the parallel features of CMP technology; although parallel programs can benefit more from CMP technology, parallel programming requires higher programmer ability. For complex parallel programming languages, programmers prefer to use traditional serial programs, and hope to reuse the developed software as much as possible. In this case, parallelization methods for serial programs have received extensive atte...

AI Technical Summary

Problems solved by technology

[0005] Although thread forward execution can improve the efficiency of parallel execution to a certain extent, because of the forward execution, threads with read-after-write data dependencies will cause threads to restart, and the restart time and resource overhead are large, that is to say , for some threads, the efficiency of sequential execution will be better than the efficiency of parallel execution on multiple cores

Images