CSE-557 Home Work No. 3

Anirudh Modi and Anupam Sharma
Graduate Students, Penn State University

We have developed a library which shall be very useful in writing a parallel code for solving problems with iterative methods. The subroutines in the library are pretty general and are not constrained by the number of processors used for working in parallel, grid size and relaxation technique. The functions in the library have been made as robust as possible so as to take care of any absurd data input from the user. The grid discretization (among the processors) and message passing is all handled by the library routines and the user has almost no way to meddle with it. Two groups of Communicator have been used, one each for row and column communication.

The library subroutines can perform the following without the user having to worry about parallelizing etc.

• New Communicator definition and initialization, which is then used to do all the message passing.
• Memory allocation in the processors given the total grid size and the total number of processors.
• Message passing to/from each neighbor. Any number of cells can be passed by specifying the number of ghost cells in each direction.