Parallel magnetohydrodynamics on the Cray T3D

The equations of magnetohydrodynamics (MHD) are discussed in the framework of parallel computing. Both linear and nonlinear MHD models are addressed. Special attention is given to the parallellisation of the kernels of the existing sequential MHD codes. These kernels involve matrix-vector multiplications and dot products for the linear MHD calculations, and Fast Fourier Transforms and tri-diagonal systems solvers for the nonlinear MHD simulations. The parallel results, such as scaling with the number of processors and the performances, are given for a 256 node Gray T3D. Impressive results are obtained by combining the implicit shared memory shmem_get routine of Gray with single node BLAS routines