• CS 5030 Final Project
  • Compiling binary output to a video
  • Building
    • MPI
    • Cuda
    • OpenMP
    • Serial
  • Creating an initial starting grid

CS 5030 Final Project

Report (you may need to download it to view the links in the document).

A Video Example

There are multiple implementations in this project. Each directory contains a Makefile which will build that implementation. For most, a simple cd and make will do.

Every make will builds a gol binary. However, each implementation takes a different number of arguments, as documented below. The Cuda implementation needs to be run in a slightly different fashion.

Compiling binary output to a video

Every implementation produces file I/O exactly the same. When logging is turned on, each iteration in the output directory is labelled iteration-XXXXXXX.bin where iteration number is padded by 7 to make life easy.

There is a script in graphics that converts a raw binary dump into a .bmp (with some help from this Stack Overflow post). make-movie.sh converts a directory of ordered binary dumps to a video file with the arguments that are described in make-movie.sh (just provide none and a usage string will be ~echo~ed).

For instance, to make a movie of the outputs generated in cuda-global/output where each binary file is a grid of size 1920x1080 (at 8fps to a file named output-1920.mp4):

cd graphics

make

(On CHPC you will need to module load ffmpeg)

./make-movie.sh ../cuda-global/output 1920 1080 8 output-1920

Building

MPI

Firstly, module load gcc mpich. Then cd into mpi and make.

Then, you can run with

mpirun -np <cores> ./gol simulate <filename | random> <width> <height> <iterations> <log-each-step?1:0>

Cuda

Firstly, cd into cuda-global and make.

Then start an interactive gpu session on notchpeak:

salloc -n 1 -N 1 -t 0:10:00 -p notchpeak-shared-short -A notchpeak-shared-short --gres=gpu:k80:1

This implementation takes these arguments:

srun ./gol simulate <filename | random> <width> <height> <iterations> <log-each-step?1:0>

For example to do 1000 iterations at 1920x1080 with a random starting position (the last 1 will log each iteration into the output directory) with a block size of 32:

srun ./gol simulate random 1920 1080 1000 1 32

OpenMP

Firstly, cd into openmp and make.

This implementation takes these arguments:

./gol simulate <filename | random> <width> <height> <iterations> <log-each-step?1:0> <num_threads>

For example to do 100 iterations with 8 threads at 800x600 with a random starting position (and log each iteration into the output directory):

./gol simulate random 800 600 100 1 8

Serial

The most basic of the three implementations.

Firstly, cd into serial and make.

This implementation takes these arguments:

./gol simulate <filename | random> <width> <height> <iterations> <log-each-step?1:0>

For example to do 10 iterations with 8 threads at 400x400 with a random starting position (and log to output):

./gol simulate random 400 400 10 1

Creating an initial starting grid

Each gol binary also has a create-grid mode, mainly used for debugging:

./gol create-grid <width> <height> <filename>

You'll be prompted to enter in grid values (0/1) for each row, each seperated by a space.

For example to make a 10x10 grid and output it to output/testing.bin:

./gol create-grid 10 10 output/testing.bin

And then this file can be used in the filename argument when using simulate.