Scribd
Upload
8 views

Extra

Uploaded by

Aditi Allady
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
8 views

Extra

Uploaded by

Aditi Allady
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

COP5570 Extra Programming Assignment (7 points total)

OpenMP and MPI implementations of the Game of Life

OBJECTIVES

• Practice parallel programming with OpenMP and MPI

DESCRIPTION

This is an individual assignment. In the Game of Life, the world is modeled as a 2-


dimensional grid w[0..w X-1][0..w Y-1] with each entry w[i][j] representing a live or dead
cell, where w X and w Y specify the size of the world. The world can be represented as a C/C++
array w[w X][w Y]. Except for cells in the corners or borders, each cell w[i][j] has 8 neighbors:
w[i-1][j-1], w[i-1][j], w[i-1][j+1], w[i-1][j], w[i-1][j+1], w[i+1][j-1], w[i+1][j],
and w[i+1][j+1]. Border cells have less number of neighbors. For example, w[0][0] only
has three neighbors: w[0][1], w[1][0], and w[1][1]. Starting from an initial condition, the
program will simulate the world population change in each time step. The following are the
population change rules in each time step in the game:

1. Any cell with 0 or 1 living neighbor remains or becomes dead (dying out of loneliness).

2. Any cell with 2 living neighbors remains in the same state (live remains live, dead remains
dead).

3. Any cell with 3 living neighbors remains or becomes alive.

4. Any cell with 4 or more living neighbors remains or becomes dead (dying out of over-
population).

In this assignment, you are given a sequential program for this game. Your task is to parallelize
this particular sequential program and develop equivalent OpenMP and MPI programs for the
game. Your programs should work with any sized world allowed by the OS and any number
of threads/processes allowed by the OS.

DUE DATE AND MATERIALS TO BE HANDED IN

Due: April 22, 2024, 11:59pm (hard deadline).

• Tar all files including makefile, README file, and submit on canvas. When the sub-
mitted files are put in one directory, type ’make’ should make the OpenMP and/or MPI
executables. The README file should include instructions to demonstrate that your
programs are (1) correct and (2) efficient.

1
GRADING POLICY

Your program only needs to run on linprog. When the following conditions are met, a correct
and efficient OpenMP implementation earns 2 points in the final grade, and a correct and
efficient MPI implementation earns 5 points in the final grade.

1. Programs with compiler errors will get 0 point. Programs that do not simulate the game
of life will get 0 point. A submission whose README file does not include sufficient
instruction to demonstrate program correctness and efficiency will get at most 50% of
the points even if the programs are completely correct and efficient.

2. Your parallel implementations must have the same calculation, not just the results, as
the provided sequential code. A program will get 0 point if this requirement is not met.
This means that you must work with the provided code. If a random Game of Life
program from the Internet is submitted for grade, it will get 0 point.

3. Your OpenMP and MPI programs must correctly work with any reasonable number of
threads/processes and any world size. Failing the correctness test of any case will result
in 50% points deduction.

4. For the MPI implementation, each MPI process should only store a fraction of the world
(the array size should be roughly equal to w X × w Y /P or w X/P × w Y , where P is
the number of processes). An MPI program where an MPI process stores the equivalent
of the whole world (for any purpose) will receive 0 point.

5. Besides producing the same output as the sequential code, both your OpenMP and MPI
program must achieve a speedup of more than 2.1 for some configuration (world size,
number of threads, etc) on linprog to be considered correct, and a speedup of more
than 4.1 to be considered efficient. The given sequential program should be used as the
baseline.

MISCELLANEOUS

OpenMP implementation is simple (a few lines of code). MPI implementation can be quite
challenging if you have never programmed distributed memory systems.
All submitted programs will be checked by an automatic software plagiarism detection tool.

You might also like