Program auto parallelizer and vectorizer implemented on the basis of the universal translation library and LLVM technology

The paper is devoted to the integration of the compiler based on the LLVM library with the tools created using the Universal Translation Library (UTL)—automatic parallelizer and vectorizer. The intermediate representations used in the libraries to be integrated are analyzed and compared. Mechanisms which had to be implemented for integration are described. The most important UTL components are also presented. Finally we present comparative performance tests of the compilation system obtained as a result of this integration, and available compilers. These tests were run on multi-core systems based on ARM and x86 architectures using SPEC/CPU2006 and NAS Parallel Benchmarks packages.