Changes:

• Added lazy SMP. Supernova can now use more than one CPU when thinking!
• Fixed a Zobrist hashing bug regarding en passant.
• Fixed an issue in static exchange evaluation.
• Various optimizations.
• Changed data type in the 8x8 board.
• Encoded castling rights.
• Made quit command more stable.
• Compiled with GCC 11.2.

Testing Results:

• Supernova 2.4 (4 CPUs) vs Supernova 2.3:
  Time control: 10+0.1
  Games: 910
  Elo difference: 152.2 +/- 18.1

• Supernova 2.4 (1 CPU) vs Supernova 2.3:
  Time control: 10+0.1
  Games: 3000
  Elo difference: 17.2 +/- 9.1

Compared to version 2.2.1, Supernova has gained about 120 Elo in self-play.

Time control: 10+0.1
Elo difference: 123.7 +/- 7.2 (tested with 3-4-5 men Syzygy tablebases)

• Added Syzygy tablebases support using Fathom.
• Reworked late move reduction.
• Let hash table move to be tried first before move gen.
• Implemented counter move heuristic in move ordering.
• Added transposition table in quiescence search.
• Switched the main search to fail-soft.
• Let the history scores of previous quiet moves be reduced when cutoff.
• Changed the null move pruning formula.
• Changed IID so it won't activate when in check.

• Let LMR reduce more for moves with a bad history.
• Minor optimization through inline functions.
• 10-20 Elo gain compared to 2.2.

• Forward prunings and late move reduction have been improved.
• Added probcut, late move pruning, and IID.
• Evaluation tuning.
• Fixed "go" command handling.
• Improved check extension.
• Allowed hash table cutoff at PV nodes.
• Added functions to convert mate scores in the hash table.
• Merged Windows and Linux source code.
• Made it compatible with GCC 10.
• 166 Elo gain against 2.1.4 in self-play.

• Passed pawn scores have been increased.
• 169.6 Elo gain against Supernova 2.0 in self-play.

• After a search, "ponder" is no longer printed out if there is no ponder move.
• Max hash table size has been increased.
• More comments have been added.

• "bestmove" UCI command fix when pondering.

• Ponder fix.
• Trapped rook eval has been changed to work more precisely.

• Changed move representation from move string to move struct.
• Added pawn hash table.
• Added passed pawn extension.
• Added more midgame and endgame evaluations.
• Added FEN support.
• Changed UCI protocol handler so that it no longer depends on the "ucinewgame" command.
• Simplified mobility functions.
• Improved time management.
• Fixed bugs in SEE.

As a result, Supernova 2.1 has now reached 1400k+ nodes per second in search speed and it is a lot stronger than Supernova 2.0 after testing.

Supernova 2.0 is stronger than any previous version. There are some major changes comparing to Supernova 1.4. Its fundamental board representation was reworked to a structure containing all the board info as well as faster piece lists that are used in move generation and evaluation. Its move ordering now uses the faster selection sort instead of quicksort in the alpha-beta framework. These changes result in a tremendous speed boost that doubles its nodes per second to about 850k. Moreover, its evaluation was also improved since now it considers candidate passed pawn and more king safety evals like king attack and castling. SEE was changed to perform more correctly and it takes part in both the quiescence search and the move ordering. When playing against some 2160 to 2200 Elo engines in blitz time control, it scored almost the same points as its opponent.