|
Preprints of the Keldysh Institute of Applied Mathematics, 2013, 016, 20 pp.
(Mi ipmp16)
|
|
|
|
Plane electromagnetic wave diffraction: hybrid paralleling
A. V. Berezin, A. S. Vorontsov, M. B. Markov, D. N. Sadovnichii, A. V. Sysenko
Abstract:
The algorithm for solving of implicit mesh Maxwell equations is vectorized. Calculation module is developed in Cuda and OpenMP technologies with optimization of access to graphic processor memory. Vector operations are carrying out at graphic processors. Operations with random access to memory are carrying out at central processor in multithread regime. Graphic processors in supercomputer nodes accelerate calculations in comparison with traditional architecture. Calculation is accelerated in 11 times for problems with small RAM volume and in 2-3 times for rest problems.
Keywords:
mesh equation, vectorized calculation, central processor, graphic processor, random access memory, coalescent request.
Citation:
A. V. Berezin, A. S. Vorontsov, M. B. Markov, D. N. Sadovnichii, A. V. Sysenko, “Plane electromagnetic wave diffraction: hybrid paralleling”, Keldysh Institute preprints, 2013, 016, 20 pp.
Linking options:
https://www.mathnet.ru/eng/ipmp16 https://www.mathnet.ru/eng/ipmp/y2013/p16
|
|