Skip to content
Home » Wordfish 2.0 Falcon net

Wordfish 2.0 Falcon net

Summary

  • Fixed the build defaults so the engine now identifies as Wordfish 2.0 with a locked-in build date of 310825, ensuring consistent metadata across the Makefile and source headers
  • Standardized engine headers to embed the new name and date, keeping version information centralized in runtime utilities
  • Updated the UCI handshake to credit Stockfish developers, Jorge Ruiz Centelles, and ChatGPT as authors, and streamlined network verification to skip the optional Falcon net if unavailable or incompatible
  • Pointed the perft test harness at the built wordfish binary so automated regression tests run against the correct executable

Testing

  • ✅ cd src && sh ../scripts/net.sh
  • ✅ cd src && make -j2 build ARCH=x86-64-sse41-popcnt
  • ✅ cd tests && bash perft.sh

Download Wordfish 2.0

Download Wordfish 2.0 device

Changelog in Wordfish 2.0-dev 060925 avx

locate-tuned_bmc_window3-in-tests

Value  bestValue     = -VALUE_INFINITE;    Color  us            = rootPos.side_to_move();    double timeReduction = 1, totBestMoveChanges = 0;    int      delta, iterIdx                        = 0;    uint64_t previousBestMoveChanges               = 0;    // Allocate stack with extra size to allow access from (ss - 7) to (ss + 2):    // (ss - 7) is needed for update_continuation_histories(ss - 1) which accesses (ss - 6),@@ -435,7 +436,9 @@ void Search::Worker::iterative_deepening() {                // otherwise exit the loop.                if (bestValue <= alpha)                {                    beta =                      (alpha * 104 + beta * 4 + std::min(bestValue + delta, VALUE_INFINITE) * 9)                      / 117;                    alpha = std::max(bestValue - delta, -VALUE_INFINITE);                    failedHighCnt = 0;@@ -444,6 +447,15 @@ void Search::Worker::iterative_deepening() {                }                else if (bestValue >= beta)                {                    if (bestMoveChanges > previousBestMoveChanges)                        alpha = (alpha * 126 + beta * 3                                 + std::max(bestValue - delta, -VALUE_INFINITE) * 2)                              / 131;                    else                        alpha = (alpha * 109 + beta * 4                                 + std::max(bestValue - delta, -VALUE_INFINITE) * 21)                              / 134;                    beta = std::min(bestValue + delta, VALUE_INFINITE);                    ++failedHighCnt;                }@@ -510,6 +522,7 @@ void Search::Worker::iterative_deepening() {            skill.pick_best(rootMoves, multiPV);        // Use part of the gained time from a previous stable move for the current move        previousBestMoveChanges = bestMoveChanges;        for (auto&& th : threads)        {            totBestMoveChanges += th->worker->bestMoveChanges;

Root Experience Q-learning

Experience file working perfectly.

Download Wordfish 2.0-dev 060925