/laser-chess-engine

Laser, a UCI-compliant chess engine.

Primary LanguageC++GNU General Public License v3.0GPL-3.0

Laser

Laser is a UCI-compliant chess engine written in C++11 by Jeffrey An and Michael An.

The latest release and previous versions can be found here. Compiled binaries for 32 and 64-bit Windows are included.

Laser is a command-line engine only. To have a graphical interface, the executable can be used with any UCI chess GUI, such as Arena or Tarrasch.

Engine Strength

  • CCRL 40/4: 11th, 3281 ELO (4 CPU) as of October 21, 2018
  • CCRL 40/40: 11th, 3227 ELO (4 CPU) as of October 24, 2018

Makefile Notes

The code and Makefile support g++ on Linux and MinGW on Windows for popcnt processors only. For older or 32-bit systems with no popcnt instruction support, use the NOPOPCNT=true option. To compile, simply run make in the main directory. The USE_STATIC=true option creates a statically-linked build with all necessary libraries.

Thanks To:

  • The Chess Programming Wiki, which is a great resource for beginner chess programmers, and was consulted frequently for this project
  • The authors of Stockfish, Crafty, EXChess, Rebel, Texel, and all other open-source engines for providing inspiration and great ideas to move Laser forward
  • The engine testers and rating lists, for uncovering bugs, providing high quality games and ratings, and giving us motivation to improve
  • Cute Chess, the primary tool used for testing

Implementation Details

  • Lazy SMP up to 128 threads
  • Fancy magic bitboards for a 4.5 sec PERFT 6 @2.2 GHz (no bulk counting).
  • Evaluation
    • Tuned with reinforcement learning, coordinate descent, and a variation of Texel's Tuning Method
    • Piece square tables
    • King safety, pawns shields and storms
    • Isolated/doubled/passed/backwards pawns
    • Mobility
    • Outposts
    • Basic threat detection and pressure on weak pieces
  • A two-bucket transposition table with Zobrist hashing, 16 MB default size
  • An evaluation cache
  • Syzygy tablebase support
  • Fail-soft principal variation search
    • Adaptive null move pruning, late move reduction
    • Futility pruning, razoring, move count based pruning (late move pruning)
    • Check and singular extensions
  • Quiescence search with captures, queen promotions, and checks on the first two plies
  • Move ordering
    • Internal iterative deepening when a hash move is not available
    • Static exchange evaluation (SEE) and Most Valuable Victim / Least Valuable Attacker (MVV/LVA) to order captures
    • Killer and history heuristics to order quiet moves

To Dos

  • Enumerate or typedef basic values such as color, piece type, and scores
  • Chess960 support
  • Improved pruning rules
  • More efficient PERFT and eval