In recent years many computing clusters with accelerators in their nodes have been appeared. Most of them are graphics processors of NVIDIA company. The clusters with accelerators of other architecture – Xeon Phi by Intel Corporation – was appeared in 2012. The model of parallel programming for clusters with accelerators was developed at Keldysh Institute of Applied Mathematics, Russian Academy of Sciences in 2011. The model is extension of DVM model and is called DVMH (DVM for Heterogeneous systems). The Fortran-DVMH (FDVMH) language developed within the model simplifies a process of parallel program writing, and also allows to convert the program for cluster (DVM program) to the program for cluster with accelerators (DVMH program) with small changes. The FDVMH language is FORTRAN 95 language extended by parallelism specifications. These specifications are implemented as special comments called directives. The directives ”are invisible” to standard compilers. It allows to have single version of the program for sequential and parallel execution. The FDVMH language allows to write the program (DVMH program) which can be executed both as sequential and as parallel one for clusters with accelerators and without accelerators.
References in zbMATH (referenced in 1 article )
Showing result 1 of 1.
- Bondarev, A. E.; Galaktionov, V. A.; Kuvshinnikov, A. E.: Parallel solutions of parametric problems in gas dynamics using DVM/DVMH technology (2020)