No abstract

Parallelization of large-scale computational applications leads to the following challenges a programmer will face: the growing amount of modularity, dynamic application functionality and multilingual source code. In general program transformation is essential for program parallelization. However, manual program transformation is not straightforward in the case of a large amount of code. The growing size of code drastically complicates the main stages of the parallel programming: distribution of data on the processors and mapping of computations on the processors. Therefore, the incremental or step-by-step parallelization mode was introduced in SAPFOR system [1,2] to map sequential large-scale applications to parallel architectures with distributed memory. In addition, special attention was directed to the automatic finding the required transformations of the sequential program. SAPFOR also relies on the automatic execution of the found transformations. However, the three main challenges mentioned above also require their solution. This paper discusses possible ways to overcome these problems.

DVM-system is designed for the development of parallel programs of scientific and technical calculations in the C-DVMH and Fortran-DVMH languages. These languages use a single DVMH-model of parallel programming model and are an extension of the standard C and Fortran languages with parallelism specifications in the form of compiler directives. The DVMH model makes it possible to create efficient parallel programs for heterogeneous computing clusters, in the nodes of which accelerators, graphic processors or Intel Xeon Phi coprocessors can be used as computing devices along with universal multi-core processors. The article presents new features of DVM-system that have been developed recently.

Аннотация. DVM-система предназначена для разработки параллельных программ научно-технических расчетов на языках C-DVMH и Fortran-DVMH. Эти языки используют единую модель параллельного программирования (DVMH-модель) и являются расширением стандартных языков Си и Фортран спецификациями параллелизма, оформленными в виде директив компилятору. DVMH-модель позволяет создавать эффективные параллельные программы для гетерогенных вычислительных кластеров, в узлах которых в качестве вычислительных устройств наряду с универсальными многоядерными процессорами могут использоваться ускорители (графические процессоры или сопроцессоры Intel Xeon Phi). В статье представлены новые возможности DVMсистемы, которые были разработаны в последнее время.

Параллельный ввод-вывод и контрольные точки в DVM-системе Рекомендуемая форма библиографической ссылки