The design of efficient algorithms for large-scale gas dynamics computations with hybrid (heterogeneous) computing systems whose high performance relies on massively parallel accelerators is addressed. A high-order accurate finite volume algorithm with polynomial reconstruction on unstructured hybrid meshes is used to compute compressible gas flows in domains of complex geometry. The basic operations of the algorithm are implemented in detail for massively parallel accelerators, including AMD and NVIDIA graphics processing units (GPUs). Major optimization approaches and a computation transfer technique are covered. The underlying programming tool is the Open Computing Language (OpenCL) standard, which performs on accelerators of various architectures, both existing and emerging.

Статья посвящена созданию эффективных алгоритмов для крупномасштабных расчетов задач газовой динамики на гибридных (гетерогенных) вычислительных системах, высокая производительность которых обусловлена применением массивно-параллельных ускорителей. Рассматривается конечно-объемный алгоритм для моделирования сжимаемых течений в областях сложной геометрии с использованием численной схемы повышенного порядка аппроксимации на основе полиномиальной реконструкции на неструктурированных гибридных сетках. Подробно описывается реализация базовых операций алгоритма для массивно-параллельных ускорителей, включающих в себя, в частности, графические процессоры GPU (Graphics processing units) производства AMD и NVIDIA. Описаны основные оптимизационные подходы и методика переноса вычислений. В качестве средства разработки используется открытый вычислительный стандарт OpenCL (Open computing language), позволяющий задействовать ускорители различных архитектур, как существующих, так и перспективных. Библ. 15. Фиг. 4. Табл. 4.

Considered in this article are various aspects associated with the implementation of large-scale supercomputer calculations of hydrodynamic and aeroacoustic problems; the problems of computational optimization are considered. Given are the means of achievement of high parallel efficiency for a large number of processors (up to several thousand). The implementation features of architecturally adapted hybrid paralleling, the optimization of memory access, and parallel processing of computational data and large grids are described. Given here are examples of direct numerical simulation applications for the simulation of both compressible and incompressible turbulent flows with the use of high-accuracy numerical algorithms on structured and unstructured grids.

This chapter discusses the big unstructured mesh processing on multiprocessor computer systems. The use of unstructured meshes with a large number of nodes makes it possible to approximate a space near the three-dimensional bodies of irregular shape with sufficient accuracy, but in this case, a great computational capability is required. Multiprocessor systems with distributed memory gives required capabilities and makes it possible to expand computational capability unrestrictedly for all intents and purposes. As a rule, multiprocessor simulation with the use of an explicit difference scheme and a simple iteration method proceeds in accordance to the following algorithms: reading the mesh topology from the disk and sharing computational load among the processors, reading the mesh description and mesh functions and determining the necessary parameters for computation of mesh geometry, initialization of data structures for data exchange, simulation, writing simulation results to the disk, and analysis or visualization of the results.