Changed search pattern for BrainLearn, CorChess, Crystal, Polyfish, ShashChess and Stockfish version number.
Apply -flto-partition=one for GCC. Live book can now be deactivated for BrainLearn and ShashChess.
CECSA is now also compatible with the community version of the GCC compiler.
Updated Berserk. The name of the neural network is now determined by the script itself.
Change -flto to -flto=thin and added -fexperimental-new-pass-manager for Clang and Android NDK.
Changed the makefiles from BrainLearn, CorChess, Crystal, Polyfish, ShashChess and Stockfish due to changes to the original makefile from Stockfish which introduces a download fallback for the network.
Added a new parameter CECSA_COMPILER_GCC_DISTRIBUTION_BUILD to easily switch between distribution builds (optimized for armv8-a based devices) and non distribution builds (optimized for the host CPU) when using the GCC compiler.
Replaced the makefiles from BrainLearn, CorChess, Crystal, Polyfish, ShashChess and Stockfish with my own makefiles in CECSA style. All supported chess engines now have their own makefiles.
Removed Demolito, Fire and Stockfish+. Big thanks goes to MichaelM for the hint, that PGO never worked with GCC in my script. Ignoring the warning messages was a big mistake. Without implementing the function "__gcov_flush()" PGO doesn't work with the GCC compiler.
CECSA now takes care about that and make the necessary changes to the source code (e.g. on Stockfish, changes are necessary in uci.cpp). Stockfish is now around 6 % faster than before.