Gus Correa wrote: > At each time step you exchange halo/ghost sections across > neighbor subdomains, using MPI_Send/MPI_Recv, > or MPI_SendRecv. > Even better if you use non-blocking calls > MPI_ISend/MPI_[I]Recv/MPI_Wait[all]. > Read about the advantages of non-blocking communication > in the "MPI The Complete Reference, Vol 1" book that I suggested > to you.
"Using MPI, 2nd Edition, by Gropp, et al, (the same people who wrote the above book, I think), also has a good discussion of this. > > You can do the bookkeeping of "which subdomain/process_rank is my > left neighbor?" etc, yourself, if you create domain neighbor > tables when the program initializes. > Alternatively, and more elegantly, you can use the MPI > Cartesian topology functions to take care of this for you. Also described in Using MPI, 2nd Ed. -- Prentice
