Author: Robert Hyatt
Date: 20:36:11 11/21/04
Go up one level in this thread
On November 21, 2004 at 19:14:35, Dan Honeycutt wrote: >On November 21, 2004 at 12:18:58, Robert Hyatt wrote: > >> >>If you plan on hacking and testing, repeated for a _long_ time, that is the way >>to go. The first parallel version of Crafty (version 15.0) was designed, then >>written, after the complete approach had been considered, designed, and >>carefully evaluated... only _then_ did the first line of code get written. >> > >* 15.0 this is the first version to support a parallel search using * >* multiple processors on a shared-memory machinel. this version * >* uses the "PVS" algorithm (Principal Variation Search) that is an * >* early form of "Young Brothers Wait". this is not the final * >* search algorithm that will be used, but it is relatively easy to * >* implement, so that the massive data structure changes can be de- * >* bugged before getting too fancy. the major performance problem * > >Sounds like what Daniel is doing, something relatively easy to implement so he >can test and debug before getting too fancy. > >Dan H. Didn't to me. The version above had the current parallel search implemented in total. All parts were not enabled initially to attempt to avoid introducing too many bugs at one time... but the program as it exists today was designed completely in 15.0, from the parallel search perspective. Only change after that point was adding the ability to split the tree at the root, something that is a bit more problematic than splitting at ply 2 or deeper. My reading Daniel's post sounds like he is just writing code left and right without having a clear idea of what the ultimate parallel search algorithm will look like. That's a recipe for lots of debugging and rewriting.
This page took 0.01 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.