Crate tfe

A 2048 implementation that uses bit-shifting and a pre-computed move table this implementation is designed to provide low overhead when testing an algorithm on a large amount of games. On a mid-2015 MBP Retina (2.5GHz i7) 10,000,000 games take about 80 seconds to complete running on 8 threads (1,250,000 games per thread) by executing random moves, avg score ~2k.

The board itself is encoded as a u64. This means that each tile has 4 bits (64 / 16 = 4) to store its value. Since the maximum value of setting all four bits to 1 is 15 we cannot use it to display the value directly. Instead we use these 4 bits as the power value: 2 << 15 = 65536, 2 << 14 = 32768, 2 << 13 = 16384, 2 << 12 = 8192, etc...

A simple way to play the game automatically is to use tfe::Game::play:

extern crate tfe;
use tfe::{Game, Direction};

let game = Game::play(|board, failed| Direction::sample_without(failed));
println!("score: {:<6} board hex: {:016x}", Game::score(game.board), game.board);

The play method takes a closure that accepts a board: u64 and failed: &Vec<Direction> as parameters and returns a Direction to move in.

The game will terminate automatically if each distinct move has been attempted and failed without any successfull move in between the failed moves.

---

references:

• https://github.com/nneonneo/2048-ai/blob/master/2048.h
• https://github.com/nneonneo/2048-ai/blob/master/2048.cpp
• https://stackoverflow.com/questions/22342854/what-is-the-optimal-algorithm-for-the-game-2048

Structs

Game	Struct used to play a single game of 2048.

Enums

Direction

Enum that stores all available directions. This enum also provides some basic functions to allow a game to be using random moves.