Author: Uri Blass
Date: 11:41:41 04/15/03
Go up one level in this thread
On April 15, 2003 at 13:43:50, Vincent Diepeveen wrote: >On April 15, 2003 at 12:48:06, Uri Blass wrote: > >>On April 15, 2003 at 10:13:07, Vincent Diepeveen wrote: >> >>>On April 14, 2003 at 04:47:44, Andreas Rueckert wrote: >>> >>>>Hi! >>>> >>>>I'm thinking about a incremental movegenerator for our little project. The idea >>>>is to store the moves for the pieces in a array and then check the last moved >>>>pieces, if they intersect with the potential moves of the piece on the square. >>>>If not, the moves that were computed before the piece was moved, could be >>>>reused. Has this been done before? (I guess so, since almost everything has been >>>>tried in this field, as it seems to me). Are there any numbers, how much >>>>performance gain is too expect from such a solution? >>> >>>for a beancounter that is psq only i assume is your question as in diep move >>>generation including generating all moves at every qsearch node is < 0.6% system >>>time. >>> >>>it is trivial that incremental generation is fastest. you just pick 1 piece and >>>generate all moves for it. if first move generates cutoff then you had a cheap >>>cutoff. >>> >>>Yet search efficiency is poor of course as perhaps some other move was better to >>>try first. >>> >>>Let's skip that. >>> >>>We all know fritz 5 and older than that must have used incremental move >>>generators. >>> >>>Perhaps you can get to 10 million nps speed at a fast K7 with incremental move >>>generator (of course pseudo legal, not legal). >> >>I think that the only right way to count is to count legal move as a node. > >No. Why no I think that people who count illegal moves do not count correctly(otherwise I can generate more nodes by having always 64^2 nodes and always discovering fast that a move from square a to square b is not legal(usually because a is empty). > >>If my target is to have more nps than I can get more than 10 million legal moves >>per second on p1000 by the following steps that I do when movei >>calculates perft. > >perft has nothing to do with this. Perft has clearly to do with it because my best perft generates more than 119060324 legal moves in less than 10 seconds (119060324 legal moves are only the moves in the last ply) Uri
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.