Kashkovsky A.V.   Шершнёв А.А.   Кудрявцев А.Н.   Khotyanovsky D.  

An approach to development of multiplatform code for numerical simulation of compressible flows

Reporter: Kashkovsky A.V.

Many problems in computational fluid dynamics of supersonic flows require employment of high-resolution three-dimensional grids. One of the example of such problems is numerical simulation of laminar-turbulent transition. One of the ways to reduce wall-clock time of the computations is the employment of hybrid CPU/GPU (Central Processing Unit / Graphical Processing Unit) computational clusters. Usually GPUs are used either as co-processors, executing only some parts of the numerical algorithm, or as main computational devices. In latter case CPUs are only used as control devices calling computational kernels. However, the most efficient approach would be to use a hybrid computational cluster as a heterogeneous system, in which both CPUs and GPUs are used as computational devices.

The present paper overviews an approach to development HyCFS code, designed for numerical simulation of compressible high-speed flows on such computational systems on the basis of full unsteady Navier-Stokes equations. This approach allows one to use code on both GPUs and multicore CPUs. The main idea of the approach is to use a special wrapper allowing to compile two different executables for both GPU and CPU from the same source code. The developed numerical code was verified, its computational efficiency and parallel speed-up of the developed code were tested and compared on both GPU and CPU.

To reports list