Author: Scott Gasch
Date: 22:42:54 11/10/04
Go up one level in this thread
On November 09, 2004 at 22:20:48, Russell Reagan wrote: >On November 09, 2004 at 21:43:11, Robert Hyatt wrote: > >>On November 09, 2004 at 16:21:56, Christophe Drieu wrote: >> >>>Hi, what is the basic idea to play chess on bi-processors ? Is it something >>>like: >>> >>>nb_moves=generate_all_moves(); >>> >>>Create 2 threads >>> The first one analyse the nb_moves/2 first moves >>> The second analyse the nb_moves/2 next moves >>> >>>Take the best move from the two analysis. >>> >>>Regards >> >>Or use dynamic thread scheduling and generate the list of moves first, then let >>each take a single move and search it, and when it completes the search, grab >>another and search it. This provides better load balancing... > >Request for clarification, please. When should this strategy be used? It doesn't >seem like it would be terribly effective at the root since the first move >usually takes the majority of the time, but your post could be interpretted as >recommending this strategy at the root. I assumed he was asking if this strategy >should be used at the root. Split the search when you are reasonably sure you have an all-node. I also do not allow the tree to be split at the root or with subtrees that are too shallow to make it worth the trouble.
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.