GPU-based game AI framework utilizing the Monte Carlo Tree Search algorithm

Due to the use of C++11 features, gcc 5+ and CUDA 8.0+ must be used. This requires setting the LD_LIBRARY_PATH environment variable so that the correct shared libraries can be dynamicly linked. On ECE GPU lab machines, gcc 5.2 can be found in /opt/rh/devtoolset-4/root/usr/bin/, which should be added to your PATH so that the correct version is used by make.
This can be done automaticly by adding the lines

set path=(/opt/rh/devtoolset-4/root/usr/bin $path)

if ($?LD_LIBRARY_PATH) then
 setenv LD_LIBRARY_PATH /usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH
else
 setenv LD_LIBRARY_PATH /usr/local/cuda-8.0/lib64
endif

to .cshrc or

export PATH=/opt/rh/devtoolset-4/root/usr/bin:$PATH

if [ -z $LD_LIBRARY_PATH ]; then
    export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH
else
    export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64
fi

to .bashrc

Simply run

make

or try

make nounicode=1

if your terminal doesn't print the Unicode checkers correctly.

bin/release/run_ai --help

Displays all runtime options

To play a game:

bin/release/run_ai -1 player1_type -2 player2_type

The default is human vs. mcts. Allowed player types are human, random, mcts, mcts_host, mcts_device_single, mcts_device_heuristic, mcts_device_multiple, mcts_device_coarse, and mcts_hybrid.

Running the human player will enter an interactive mode prompting you to enter a move. The move format consists of the simply the destination location, or if more than one piece can move to the same location, the source location followed by the destination, seperated by a space. For ambigous multiple jump moves, the intermediate locations can be entered between the source and destination seperated by spaces. Promotions are handled automaticly when a checker is moved to the oposite end of the board.
For example, to move a piece from location b6 to a5, you would type b6 a5 and press enter.

To run a test:

bin/release/run_ai --mode test -1 player1_type -2 player2_type -n num_tests

These were used to collect the benchmark results in our report. Allowed test types are host, device_single, device_heuristic, device_multiple, device_coarse, hybrid, and optimal.