Link time optimization method for heterogeneous many cores

Abstract

The invention discloses a link time optimization method for heterogeneous many cores, which comprises the following steps: S11, a compiler processes an acceleration core code, and writes generated GIMPLE intermediate representation information into an ELF segment which takes'. Gnu.lto + function name 'as a segment name in an object file; s12, when the compiler links, the acceleration core object files are linked together, and thus reading GIMPLE intermediate representation information of all functions in the acceleration core object files; s13, the compiler takes the GIMPLE intermediate representation information of all the functions read in the step S12 as a compiling unit, link optimization is carried out, and an acceleration core assembly code is generated; s14, the compiler calls an acceleration core object file which is generated by the acceleration core assembler and can be directly linked by the linker; and S15, a heterogeneous fusion linker is called by the compiler, all control core object files are linked with the acceleration core object file obtained in the step S14, thus generating an executable code, and finishing linking. According to the invention, optimization is carried out when the acceleration core code is linked, the size of the executable code of the heterogeneous program can be reduced, and the overall running performance of the heterogeneous program is improved.

Description

technical field [0001] The invention relates to a link time optimization method for heterogeneous many-cores, and belongs to the technical field of computer compilation optimization. Background technique [0002] The compiler usually only performs compilation optimization in a source code file. Link-time optimization technology is that the compiler writes the GIMPLE intermediate representation information of the code into the object file when it processes the source code file to generate the object file, and the compiler reads it in the link phase. The function body information in multiple object files is obtained, and on this basis, the compiler can optimize in a larger code range, explore more optimization opportunities, and achieve the effect of reducing the executable code size of the program and improving the running performance of the program. [0003] Under the heterogeneous many-core architecture, the compiler compiles and processes the control core code and the acce...

AI Technical Summary

Problems solved by technology

The traditional link-time optimization technology can only achieve link-time optimization for the control core code, and does not support link-time optimization for the accelerated core code. The optimization performance of the accelerated core code is the key to determining the overall performance of heterogeneous programs. Therefore, the traditional The link-time optimization technology of the company cannot meet the link-time optimization requirements under the heterogeneous many-core architecture

Images