Changelog for cfish-13+git20210626.fb847b0d5781-1.50.x86_64.rpm :

* Wed Jan 12 2022 dsterbaAATTsuse.com- Update to version 13+git20210626.fb847b0d5781:
* Updated to \"Simplification for countermoves based pruning\".
* Updated to \"Remove coordination between searching threads\".
* Updated to \"Simplify LMR\".
* Updated to \"Simplify pawn moves generator\".
* Updated to \"Merge all move generators\".
* Updated to \"Use classical eval for Bishop vs Pawns\".
* Updated to \"Replace distanceFromPv with a better logic\".
* Updated to \"Change definition of between_bb()\".
* Updated to \"Removed advanced_pawn_push()\".
* Updated to \"Introduce ProbeCut for check evasions\".
* Cfish 13
* Updated to \"PV-Nodes likely to fail low\".
* Updated to \"Tune ordering of moves at internal nodes\".
* Updated to \"Simplify Chess 960 castling\".
* Updated to \"Improve play for closed positions\".
* Updated to \"Bad Outpost Pawn Scale\".
* Updated to \"Refine stat based reductions\".
* Updated to \"Remove Condition from Generate_Move Loop\".
* Updated to \"Reorder conditions in LRM and pruning\".
* Updated to \"WeakUnopposed penalty for backwards on file A or H\".
* Updated to \"Do more LMR for captures\".
* Updated to \"Correct Outflanking calculations in classical eval\".
* Updated to \"Simplify condition for assigning static-eval based bonus\".
* Updated to \"AVX512, AVX2 and SSSE3 speedups\".
* Updated to \"New Imbalance Tables Tweak\".
* Let Cfish compiled for AVX512 identify itself as Cfish AVX512. This commit also changes ARCH autodetection to use AVX512 when available.
* Updated to \"Introduce static history\".
* Updated to \"Update default net to nn-62ef826d1a6d.nnue\".
* Updated to \"Remove piece lists\".
* Updated to \"Search simplification\".
* Updated to \"use classical for certain endgames\".
* Updated to \"Calculate sum from first elements\".
* Updated to \"Assorted search and eval parameter tune\".
* Updated to \"Rook mobility tweak\".
* Updated to \"Increase reduction at root\".
* Updated to \"Qsearch pruning: follow-up\".
* Updated to \"Introducing King On File\".
* Small NNUE non-sparse optimization for AVX2 and AVX512. By rearranging the weights of the 32x32 layer, we can do the multiplication with AVX2 and AVX512 instructions without shuffling.
* Updated to \"Fix incorrect pruning in qsearch\".
* Updated to \"Update default net to nn-cb26f10b1fd9.nnue\".
* Sun Nov 01 2020 dsterbaAATTsuse.com- Update to version 12+git20201031.981e121e71bb:
* Non-sparse cleanup This also seems to give better code with gcc.
* Improve non-sparse multiplication. Completely based on
* AATTSopel97.
* Correctly display castling rights for Chess960. Black castling rights were capitalised, which is wrong.
* Switch to non-sparse multiplication for AVX2. This now turns out to be faster on my Intel laptop. To compile with sparse multiplication, add sparse=yes to the make command. To compile without sparse multiplication, add sparse=no. The default is sparse=yes, except for AVX2, BMI2 and AVX512.
* Updated to \"Do not skip non-recapture ttMove when in check\".
* NNUE: nn-2eb2e0707c2b.nnue
* Wed Oct 28 2020 dsterbaAATTsuse.com- Update to version 12+git20201028.b331733e4116:
* Update default net to nn-2eb2e0707c2b.nnue
* Use large pages for feature transformer biases and weights
* Do more reductions for late quiet moves in case of consecutive fail highs
* Scale factor tweak
* Sun Oct 11 2020 dsterbaAATTsuse.cz- initial import, v12