Use LLVM for optimization of parallel execution of program code on the certain configuration

Automating of parallelization requires the analysis of program code from the point of view of semantic relations between operations. At the same time, up-to-date tools for automation of parallelization usually have a number of restrictions related to binding the parallelization system to the language on which the source code is written. The other restrictions appear from requirement to indicate the blocks in program code should be implemented in parallel or even make some “modifications” of the code in manual mode. The LLVM system provides manifold possibilities for automation of code analysis and optimization for its subsequent parallel execution. The article discusses the approach to parallelization of sequential program code by using LLVM for code analysis and its presentation in the form of semantic Petri-Markov nets.