Design method aiming at binary program automatic parallelization of multi-core platform

Abstract

The present invention discloses a design method aiming at binary program automatic parallelization of a multi-core platform. The method comprises: translating the binary in the multi-core platform into LLVM IR, and then establishing a control flow diagram to find all the cycles in a program by firstly analyzing the LLVM IR; carrying out static analysis on the parallelism of cycles, and removing the cycles difficult in parallel; before regenerating a paralleled code for the cycles, setting simple criteria for the common memory access dependency, adding a corresponding memory address overlapping judging code, thereby reducing the performance loss introduced by the dynamic analysis on the memory access dependency; and then splitting the cycles into two cores to run, and rewriting LLVM IR to generate the paralleled code. According to the method, the parallelism of the binary can be analyzed and extracted under the condition that a binary program source code cannot be obtained, and the corresponding binary program capable of being subjected to multi-core paralleling can be automatically generated, so that the effect of accelerating the original program by using the multi-core platform is achieved.

Description

technical field [0001] The invention belongs to the technical field of binary automatic parallelization, and in particular relates to a design method for automatic parallelization of binary programs aimed at a multi-core platform. Background technique [0002] Since 2004, in order to further improve computing performance and alleviate energy consumption problems, the architecture of microprocessors has tended to be multi-core. In the field of embedded computers, ARM has also launched a series of multi-core microprocessors MPCore in recent years. Although the multi-core transformation of the microprocessor has further improved the performance of the product, the performance potential of the multi-core processor cannot be fully utilized because the traditional upper layer software does not support multi-core parallelization. In addition, due to the wide application of digital image processing technology in recent years, the number of images to be processed has increased and th...

AI Technical Summary

Problems solved by technology

[0009] The purpose of the present invention is to provide a design method for the automatic parallelization of binary programs for multi-core platforms, which can analyze and extract the parallelism in the binary when the source code of the binary program cannot be obtained, and automatically generate the corresponding multi-core Parallelize the binary program to achieve the effect of accelerating the original program by using the multi-core platform

Images