XJCO3221 Coursework 2 1
School of Computing, University of Leeds
XJCO3221
Parallel computation
Coursework 2: Distributed memory parallelism with MPI
Deadline: 10am, Tuesday 19th April 2022
If you have any queries about this coursework please visit the MPI Discussion Forum on
Minerva (found in the Learning Resources folder for this module). If your query is not resolved
by previous answers, post a new message.
This piece of work is worth 20% of the final module grade.
Learning objectives
• Use collective communication to distribute the problem and accumulate the answer.
• Implement a binary tree using point-to-point communication.
• Perform timing runs for parallel scaling and interpret your findings.
Task
Your task is to implement an MPI-C program that performs matrix–vector multiplication in
parallel. That is, given the N × N matrix A and the N−vector x defined on rank 0, you need to
calculate b = Ax, i.e.
bj =
X
N
i=1
Aijxj
,
in parallel, and ensure the full N−vector b is on rank 0 at the end of the calculation.
For convenience, rather than store the matrix A as a two–dimensional float array, it is instead
stored as a one–dimensional array of size N2
. This ensures the rows of the matrix are stored
adjacent in memory, which makes the use of collective communication easier. With this choice,
the matrix element at row row and column col is A[row*N+col], and the serial code that performs
the multiplication is
i n t row , c o l ;
f o r ( row=0; row