🌹 Rose

Rose is an experimental classical chess and chess960 engine. Most notably, Rose was the first engine to utilise SIMD-accelerated attack tables.

Rose makes heavy use of SIMD; it relies on AVX512 for performance. Performance on other microarchitectures is not a focus of this project.

Rose is capable of communicating over both the Universal Chess Interface (UCI) protocol and the Chess Engine Communication Protocol (CECP/Xboard), although UCI is the preferred protocol.

The Rose project prides itself in only using original self-generated training data for its neural networks.

Releases

For official releases, please visit: https://github.com/87flowers/Rose/releases

We recommend using the AVX512 build for your operating system if you have a CPU that supports it, as this would be the fastest. You can get a speed comparison by running bench on each downloaded executable.

Building Rose

Building requires Make and a C++26 compiler. Clang is the recommended compiler, and GCC support is not guaranteed.

Run the command

make

in the root directory to build a ./rose executable. The current network will be automatically downloaded.

If you are building on Windows, using MSYS2 (UCRT64) is recommended. Rose is regularly tested to build with the mingw-w64-ucrt-x86_64-clang, mingw-w64-ucrt-x86_64-git, and mingw-w64-ucrt-x86_64-lld packages installed.

Non-standard UCI commands

• wait: Waits for the current search to complete before continuing.
• bench: Runs a search benchmark on a built-in list of positions, to provide a search fingerprint.
• perft <depth>: Counts all leaf nodes to the specified depth from the current position.
• moves [<move>]*: Make the specified list of moves on the current position.
• d: Print the current position as an ASCII board.
• getposition: Print the current position as a UCI command.
• dumpposition: Dumps internal position structure information.
• hashstack: Prints the current hash stack, which is used for repetition detection.
• xboard: Switch to CECP mode.

Non-standard Xboard commands

• wait: Waits for the current search to complete before continuing.
• getposition: Print the current position as a UCI command.
• dumpposition: Dumps internal position structure information.
• hashstack: Prints the current hash stack, which is used for repetition detection.
• uci: Switch to UCI mode.

Acknowledgements

• Members of the Stockfish, Engine Programming, and Alpha-Beta discords, for being a helpful and fun community to be a part of.
• The many public available open-source chess engines.
• Authors of Yukari, for ideas, hardware, and support.
• Authors of Reckless, pawnocchio, viridithas, Stormphrax, Soul, Clockwork, and many others.
• jw1912 for bullet, used to train the NNUE networks.
• The excellent {fmt} library, used in this project.

License

This project is licensed under the GNU Affero General Public License version 3. See the LICENSE file for more information.