Author: Daniel Clausen
Date: 12:08:36 11/16/99
Go up one level in this thread
Hi On November 16, 1999 at 03:08:52, blass uri wrote: [snip] >I think that the result is not a good test because there can be different >definitions of nodes >It is easy by generating illegal moves to generate more nodes. > >I think it is more interesting to know how much time you need to generate all >the legal moves and not to repeat the same moves but to generate a tree of all >the possible legal games of 3 moves of both sides. I think this speed has a similar quality like NPS. But if you tune your own move generator you can compare it with an older version. So I think the test is at least a lil useful. (No pun intended :) Some points: -Generating legal moves: some engines only produce legal moves, others call Search() first and then notice the resulting position is illegal. So we have at least an additional MakeMove() call here. -The "output" of the MoveGenerator is not really defined. Ie my move-generator could do nothing, and the call GetNextMoveFromPool() could do all the work. Is my movegenerator now blindingly fast? :) Or with bitboards: Is the 'unroll' of a bitboard into ints (whereever the bit in the bitboard is '1') done in the move-generator or later? -Sorting: Maybe your move-generator automatically sorts the moves it generates and mine doesn't. (Or at least some kind of pre-sort is possible) -Some people (like me:) have move generation split into GenerateCapMoves() and GenerateNonCapMoves(). In this test this is a disadvantage whereas in practice it can speed up move generation since often you don't have to generate the non-cap moves. The list goes on and on, but I don't have more time now since the pizza is ready. [Hey, a mathematician became famous for a similar comment! =)] Kind regards, -sargon
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.