Author: Gerd Isenberg
Date: 09:17:36 06/02/03
Go up one level in this thread
On June 02, 2003 at 09:26:09, Peter Fendrich wrote: <snip> >Hi Gerd, I was up to something similar even if not directly Kogge-Stone. >This problem happens in 1 game out of 200 without being too comitted to >precisely this figure. I fear that it is quite costly to do this alg all the >time. One should need some kind of trigger to start it I think. Hi Peter, yes, may be if the cardinality of the attack or better move-target set is below some threshold... and if there are some unmovable pawns on squares of the color of the bishop in question. I don't use it at all in my current progam, but some "bad" bishop heurisics considering pawn rams, specially in the (extended) center - but i plan to try it in my future "hammer"-version ;-) Since AMD's Opteron/Athlon-64 optimization guide is available, i started to designing some SSE2 (white:black generators)/MMX (propagators) assembler routines in bottom-up manner... Unfortunately most SSE2 integer-, but also fp-instructions are so called "double" direct path instructions so far - but they "promised" wider 128-bit SSE2 ALUs in future hammer versions. "Movd reg64, (x)mm" or vice versa is still vector path and not "allowed". Same for bsf, bsr. Anyway, nice to have sixteen 128-Bit xmm registers. In the main routine for "normal" material distribution, i perform two opposite attack directions per time. For horicontal rays white:black rook1, white:black rook2, white:black queen and black:white king (as metaslider). For diagonals white:black bishops (they behave disjoint, one one dark and one on light squares), white:black queen and black:white king (as metaslider). A lot of register independent instructions. The swapped king colors makes it easier to determine pinned pieces in the same run... all branchless and intermixed with gp-register code for determing knight, pawn and king attacks and some other usefull tasks, to keep all the pipes busy ;-) All generated attacks are stored/aggregated in memory for later movegen and eval purpose. I intend to store disjoint attacks for each piece but also for each direction per piece type. > >Both rooks and bishops and sometimes knights are very vulnerable to this. Even >Rooks can be hindered but seldom by pawns only. For example a rook can be, if >not trapped, hindered by it's own king (king moves to f1 and g1 while the rook >is still on h1). In theory even Queens are subject to this as well but in >practice they can chose either the "bishop-way" or the "rook-way" out. Queens >are more vulnerable to be trapped behind the opponents pieces but that's another >story! > >Isn't "MvH" Swedish only?? >/Peter What about Norwegian and Danish? Mit freundlichen Grüssen, Gerd
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.