OpenMP and OpenACC programming technologies on supercomputers and modern Fortran in solid mechanics problems

The paper considers the features of parallelization of algorithms using OpenMP and OpenACC programming technologies. It covers the features that were manifested when implementing the iterative finite element algorithm for solving non-classical problems of mechanics of solids with singular points. A review of modern means of creation of high-performance mathematical software is provided. A comparative study of the parallelization of algorithms using OpenMP and OpenACC on INTEL and PGI compilers is conducted. Time characteristics of the compilers and parallelization technologies are given on the examples of multiplication operations for complete and sparse matrices. The influence of modern Fortran built-in functions on these characteristics and the representation of real numbers is also illustrated. The advantages of numerical methods for sparse matrices are shown. Practical recommendations for the application of Fortran algorithms as well as OpenMP and OpenACC technologies are given.