CMSC433 Spring 2021
Project 6 – Game of Life MPI Implementation Due: 5/10/2021 11:59:59 pm
Goal. In this project, you will be parallelizing Conway’s Game of Life1 in C using the OpenMPI library.
Setting up. For this project, you will need a machine that supports OpenMPI and the Make command. Follow the instructions below for your OS
Mac / Linux. Install MPI on your machine by calling: • run/install_mpi.sh
Typically, UNIX based machines will have make pre-installed. If this is not the case, you can install it by calling:
• sudo apt-get install build-essential
If you are on a Mac with Homebrew, you can install both by calling:
• brew install open-mpi
• brew install make
Windows. We recommend you enabled Windows Subsystem for Linux (WSL) for this project. Follow the instructions below:
• Go to “Control Panel” -> “Programs” -> “Turn Windows features on or off”
• Check “Windows Subsystem for Linux” and click “OK”
• Restart your device. Upon restart, open the Microsoft Store
• Search “Ubuntu” and install the “Ubuntu” app
o “Ubuntu20.04 LTS” and “Ubuntu18.04 LTS” should work if you already have them pre-installed
• Launch the Ubuntu app. If prompted, create an account
From WSL, you may access files on your Windows machine through /mnt/c/.
Navigate to the P6 directory by calling:
• cd /mnt/c/
Now let’s install make and OpenMPI. Run the following commands from the P6
directory within Ubuntu WSL:
• sudo apt update
• sudo apt install make
1 https://en.wikipedia.org/wiki/Conway%27s_Game_of_Life 1
CMSC433 Spring 2021
• run/install_mpi.sh
Getting started. We have provided you with many files to the program, which we will
describe below:
– A serial (sequential) version of the program called src/life_seq.c. Take a look at how the game is implemented. You may use code from this file in your MPI implementation. The serial version will also be used to generate the canonical output for the program. You should feed the input files into this program in order to see what the correct output is.
– The starter code for the program called src/life_mpi.c This is the only file you will be modifying in this project.
– A sample MPI program that shows how message passing works called
src/sample_mpi.c
– In order to compile all of the examples and the src/life_mpi.c programs, you should go to the run/ directory. You can either use the Makefile or the run_*.sh scripts to build and run your programs.
– Visualization and random input generation scripts are performed by
and respectively. There is an attached document with these two files
which explains how they function.
– For more information, please see the README file under the project directory.
How the game works. The Game of Life consists of an N by M matrix, where each cell can either be 0 (dead) or 1 (alive). The game will start with the cells corresponding to the (x, y) coordinates in the initial input file being set to 1 (alive). This is the “first generation” of the game. The game consists of constructing successive generations of the board based on certain rules which are described below:
1. Cellswith0,1,4,5,6,7,or8neighborsdie(0,1oflonelinessand4-8of overpopulation)
2. Cellswith2or3neighborssurvivetothenextgeneration
3. Anunoccupiedcellwith3neighborsbecomesoccupiedinthenextgeneration.
In life_seq.c, two copies of the board are defined: prev to store the previous generation and curr to compute the next generation. Boards are padded on all sides to allow for easier computation (there are less if-statements when computing neighbors of a cell on the board’s edge).
While you are not required to, we recommend you follow this paradigm when implementing Game of Life with MPI.
tools/life_visualizer.py
tools/input_generator.py
tools/TOOLS_README.txt
2
CMSC433 Spring 2021
An example graphic of one iteration of the game is found below:
For more information on the Game of Life, check out the link at the top of the pdf.
Implementing the game using MPI. We have provided some starter code, which includes all of the initialization of the MPI functionality for the program, as well as the code for the rank 0 process. You will need to implement the code for the rank 1 to n processes (worker processes).
The following is a high-level overview of this implementation:
1. Uponstart-up,therank0processwillpartitiontheboardintonon-overlapping sub-boards and assign them to each worker process.
a. Workerprocesseswillallocatememorytocreatetheirsub-board
b. Workerprocessesdonothavedirectaccesstoeachother’sboards
2. WorkerprocessesareresponsibleforcomputinguptotheN-thgenerationofthe
Game of Life for their sub-board. N is specified as a runtime parameter.
3
CMSC433 Spring 2021
a. Tocomputethenextgenerationofitstop/bottomrows,theworkerprocess needs to know the top/bottom rows of its neighboring processes
3. UsingMPI,workerprocesseswillcommunicatetheirtopandbottomrowstotheir respective neighbors. These rows will be stored in the sub-board’s padding
4. Withtheupdatedsub-board,eachworkerprocesswillcomputethenext generation of its sub-board. This logic should be the same as in life_seq.c
5. AfterNiterationsofthiscycle,allworkerprocesseswillsendtheirfinalsub-board state back to the rank 0 process.
6. Therank0processcollectsthesesub-boardstatesintoonelargeboard.Itthen prints out the overall N-th generation in the Game of Life.
The BOARD macro: Conceptually, you can think of boards as 2-D matrices with padding on all sides: they are of size (height+2)*(width+2). However, they are stored in C as a very long 1-D array of integers. We define the BOARD(board, x, y) macro to help you easily access indices of the 1-D array as if it were a 2-D matrix, centered around the original unpadded board.
With BOARD, the padded rows/columns are now at indices -1 and height/weight. The top left most colored cell can now be imagined as the 0th column, 0th row.
For example:
• BOARD(board, -1, -1) returns the value of the top-left in the padded board
• BOARD(board, 0, 1) returns the value at the 1st column, 2nd row in the padded
board (this is the same cell as the 0th column, 1st row in the unpadded board).
We highly recommend you use the BOARD macro to simplify computation in your implementation.
4
CMSC433 Spring 2021
Running the serial and MPI programs. First, go to the run/ directory. Then, you can use either the Makefile or the run_*.sh files to run the programs. Take careful note of the runtime parameters you must supply the run_*.sh files
Testing with the sample input/output. You can use the 4 sets of sample input/output files located in the run/ directory to test your program.
Testing with the input generator tool. First, you can create random inputs for the program using the input_generator.py script. Next, you should run these input files either through the visualizer or the serial version in order to get the correct output for the final generation that you specified. Finally, run the MPI program with the input file and get the output from that.
Comparing the output files. We recommend using the diff command (or https://www.diffchecker.com/ if you prefer a graphical display) to verify that the output from your MPI program is the same as the output from the serial version.
Submission. Submit the life_mpi.c file to the submit server. It will be graded offline. Grading. Grading will be performed offline. Since you can easily test your program
locally on random input, all tests on our side will be secret.
5