Learning by doing :: MPI -- Virtual Topologies In MPI, Part 8

This video is part of the new series called "Learning by doing :: MPI" which serves as an introductory course on MPI (Message Passing Interface). In this video we talked about virtual topologies in MPI! This topic corresponds to chapter 8 of the MPI standard v4.1. It is a lengthy chapter with a lot of information to digest. We followed the same pattern as for "Datatypes in MPI" by dividing a chapter into digestible ~25 mins long parts. In part 8 we continued our discussion by talking about the second code example of MPI_Neighbor_allgatherv. Then we continued with all to all pattern for neighbourhood collective communications. We started with MPI_Neighbor_alltoall with three examples, first one being with a distributed graph in a ring topology pattern, second one being a 3 dimensional Cartesian grid and the last one being the replication of example 8.2 in the MPI standard. In the end we had time to talk about MPI_Neighbor_alltoallv which is the vector version of MPI_Neighbor_alltoall. Note that the difference between MPI_Neighbor_allgather and MPI_Neighbor_alltoall is that you can send different data to different processes in the case of the latter, however you send always the SAME data to all other processes in the case of the former.

All the necessary info you need for MPI standard can be downloaded from here : https://www.mpi-forum.org/docs/mpi-4....

If you want to download MPICH, please visit here : https://www.mpich.org/downloads/
* The latest version at the time of this video is mpich-4.2.2

Hope you learned something from this video. If you have further questions don't hesitate to comment down below. Have fun!