Parallel multigrid algorithms for some specialized computer systems
The paper deals with parallelization of multigrid algorithms for some specialized parallel computer systems. These algorithms are optimal because the number of operations that have to be performed linearly depends on the number of discrete unknowns which are to be evaluated. It is shown that for a SIMD-type computer the separate location of grid point values in orthogonal memory blocks and permutation network functions allow to implement the computational phases of the multigrid cycle in parallel. The effective mapping of partitioned grid onto the processors by the Gray code is described for hypercube multiprocessors. Finally, the VLSI design for a multigrid algorithm is presented where the best hitherto known upper bound for solving the model elliptic problem in the area-time complexity is achieved.
Unable to display preview. Download preview PDF.
- 1.1 Codenotti, B., Lotti, G., Romani, F.: VLSI implementation of fast solvers for band linear systems with constant coefficient matrix. Information Processing Letters 21, 1985, 159–163.Google Scholar
- 2.2 Gannon,D., Rosendahle,J.Van: Highly parallel multigrid solvers for elliptic PDEs: An experimental analysis. Rep. 82-36, ICASE, Hampton 1982.Google Scholar
- 3.3 Hackbusch, W., Trottenberg, U. (eds.): Multigrid methods. Lecture Notes in Mathematics 960, Springer-Verlag, Berlin, 1982.Google Scholar
- 4.4 Heath, M.T. (ed.): Hypercube Multiprocessors 1987. SIAM, Philadelphia 1987.Google Scholar
- 5.5 Mikloško, J., Vajteršic, M., Vrťo, I., Klette, R.: Fast Algorithms and their Implementation on Specialized Parallel Computers. North-Holland, Amsterdam 1989.Google Scholar
- 6.Vajteršic, M.: VLSI implementation of a fast Poisson solver. In: Proc. CONPAR 88 Conference / Stream "C"/, British Computer Society, Manchester 1988, 137–146.Google Scholar
- 7.7Volkert, J., Henning, W.: Multi-grid algorithms implemented on EGPA multiprocessor. In: Proc. 1986 Int. Conf. on Parallel Processing. IEEE Computer Society Press, Washington, D.C. 1986.Google Scholar