Supercomputer technologies in science, education and industry was held on May 30th, 2016 in Russia, Moscow, Lomonosov Moscow State University.
DVM system was developed in Keldysh Institute of Applied Mathematics, Russian Academy of Sciences, with the active participation of graduate students and students of Faculty of Computational Mathematics and Cybernetics of Lomonosov Moscow State University. It is designed to create parallel programs of scientific-technical calculations in C-DVMH and Fortran-DVMH languages. These languages use the same model of parallel programming (DVMH-model) and are the extensions of standard C and Fortran languages by parallelism specifications, implemented as compiler directives. The directives are invisible to standard compilers, so a programmer can have one program for sequential and for parallel execution on computers of different architectures.
DVMH model allows to create efficient performance parallel programs (DVMH-programs) for heterogeneous computational clusters, which nodes use as computing devices not only universal multi-core processors but also can use attached accelerators (GPUs or Intel Xeon Phi coprocessors). The computations mapped to the node can be automatically distributed between the computing devices of the node taking into account their performance.
The possibilities of DVM system implemented recently will be considered in the presentation:
- new version of C-DVMH compiler ;
- parallel input/output;
- new facilities for the functional debugging and analysis of parallel program performance.
Presented by V.A. Bakhtin.
Mathematics and Informational Technologies for Oil and Gas Industry (2016) was held on May 16th - 20th, 2016 in Russia, Surgut, Surgut state university.
Compositional multiphase simulation is used for more detailed flow modeling for reservoirs with light hydrocarbons (condensate, gas), for more accurate description of mass exchange between phases in enhanced oil recovery methods such as carbon dioxide or nitrogen injection, enriched gas injection or high-pressure gas injection, etc.
Presented by V.A. Bakhtin.
C-DVMH language description is available to study now. You can look the latest version always in the section "Documentation".
Now You can install DVM-system on your personal computer downloading the latest version from the SVN repository.
Parallel computational technologies (PCT) 2016 was held on March 28th - April 1st, 2016 in Russia, Arkhangelsk, Northern (Arctic) Federal University.
The result of applying a fixed, program-independent sequence of optimizations when compiling programs doesn’t reflect the features of their information structure. It affects the performance of parallel programs obtained by use of the non-adaptive automatically parallelizing compilers. The iterative process of program parallelizing, supported by SAPFOR system of automated parallelization, allows to select the needed transformations and apply them only when they are necessary to eliminate the problems preventing parallelization.
Presented by A.A. Smirnov, N.A. Kataev.
DVMH model is suitable first of all for writing of parallel programs on the regular rectangular grids, but it is possible to parallelize some types of programs on the irregular grids using available tools. There is offered a variant of DVMH model extension which, on the one hand, would organically be inscribed in the existing DVMH model, adding its constructions, and on the other hand would allow to remove the known problems and restrictions in the task parallelization on irregular grids, and not losing considerably in efficiency of parallel execution.
Presented by A.S. Kolganov, M.N. Pritula, V.A. Bakhtin.
Parallel implementation of the search algorithm for minimum spanning trees using central and graphic processors
The solution of the problem to search a minimum spanning trees is widespread in various fields of researches: recognition of different objects, computer vision, analysis and creation of networks (for example, telephone, electrical, computer, road, etc.), chemistry and biology and many others. There are at least three well-known algorithms that solve this problem: Boruvka’s, Kruskal’s and Prim’s. Large graph processing is a time-consuming task for Central processor unit (CPU) and is popular at present. The graphic accelerators (GPU) having greater computational power, than CPU, become more and more widespread for solution of general purpose tasks. But this problem, as well as many graph processing tasks, badly lays down on GPU architecture. In this article hybrid implementation of the algorithm will be considered.
Presented by A.S. Kolganov.