S.M.A.R.T. Chess Engine / News: Recent posts

The quiescent search code was added to CVS and the two CVS trees (sourceforge + tofus') have been synchronized. NOTE: Quiescent search code significantly slows down the search because of the badly ordered moves that the movegen routines currently generate. These need to be fixed before quiescent search can be activated.

The Negascout/PVS algorithm that replaced the old alphabeta search has yet again been replaced itself; since SMART is currently using multi-level/single-target transposition caches, the search algorithm had to be changed to reflect this. Since negamax alphabeta is very simple to implement, i chose to use that instead of negascout. Once we get confirmation that zero-window negascout performs better with MTDf then plain negamax, i'll add a multi-level/single-target version of PVS/Negascout to replace the current Negamax implementation. Currently, however, it seems that Negascout produces somewhat more overhead in zero-window search then Negamax does. For the time being, we'll stick to plain negamax MTDf.

A lot of stuff has been changed. The most important change for now is that all low-level engine code will be released under the BSD license, while the development framework and some documentation will remain licensed under the GPL.

Although MTDf performance isn't really affected in theorie, since MTDf does zero-window searches on both Negascout and Alphabeta, i've changed the main tree-search engine with Negascout/PVS. Thus far, this seems to indeed caused a minor performance boost. This performance-win can also be caused by the new movegen-routines, which we're implemented almost synchronously with Negascout/PVS.

Implementation of xboard with some pseudocode. After the basic functionality the focus for frontend will be controlling the decision management and analyses.

The heart of the engine, namely the alphabeta and mtdf implementations, are done. There is still plenty of room for little improvements, but all the main components are ready. We can now focus more on the soul of the engine: the evaluation function.

Currently, all the low-level code concerning bitboard-manipulation and hash/transposition tables is done. Right now, the mtdf/alphabeta functions are being implemented.

I'm currently laying the last hand on the bitboard implementation; Programming bitboards is one of those weird 'black code-fu magics' from the ancient days, when CPU's had no power and no memory to remember that. SMART's bitboards occupy about 600Kb of memory...so it's a 'modern' implementation that shouldn't be portable to older C64's and ZX-Spectrums ;p

The development framework has been updated. Real development can start now.

The SMART Chess CVS tree has been created today. I have populated it with a standard framework, so it contains no running code yet. However, development can really start now.