Author: Daniel Shawul
Date: 06:45:17 11/21/04
Go up one level in this thread
On November 21, 2004 at 08:49:52, Anthony Cozzie wrote: >Rather than have a big flame war here, (I'm saving my flamewar comments for the >arena thread :), I think I need to point out some things. > >First of all, if you are going to implement someone else's algorithm, I agree: Here we go again. If you are trying to imply, i borrowed algorithm from scott. I only implemented one thing scott told me that is splitting at first node and fourth node thing. That doesn't take a second to implement at all! (only 10 lines maximum).And you can be sure i won't use it in the release version. The other things i did it all by myself and help from CCC guys(including you,that 1kb copying thing i have figured it out). And if that's still so hard to belive i can post the relevant part of DanChess here. Nothing special everything of my own. If you are not implying what i thought above,sorry in advance. >it is mainly a matter of technique and debugging. It is still a challenging >debugging problem, but sure, I agree: it is a _programming_ problem rather than >a _computer science_ problem. yes my point is the splitting issue is the main science. I still don't use spin locks. I use critical sections which are quite enough for experiminting for the time being.I will postpone that until i understand them well. I was testing with ponder on against gerbil up to now(on single cpu), that disadvantaged danchess a lot and only manged to search depth 4 or 5. But when i changed to ponder off today,gerbil had no chance at all and it already has drawn fruit once. > >Secondly, this whole business gets much less complicated when you worry about >two processors only. All the fancy conditions occur when you have several >processors. If you design for two processors, you have a simple master/slave >relationship and everything works out. yes that's right i still don't lock/unlock the hashtable which is fine for a dual. > >Third, I strongly agree with Bob on the issue of design. With simple stuff you >can usually play it by ear, but I think its pretty important to do a reasonable >amount of thinking for the complicated stuff. I agree! >This doesn't mean I advocate lots >of UML and documentation, but that I think you should sit down a bit and >_think_. I will sit back and think to come up with my own splitting algorithm. >I know that I discarded 4 or 5 designs without writing a single line >of code, after I realized they were going to suck :) > >Anyway, I think we will see a parallel version of DanChess in the near future >(hopefully at CCT!). Good luck! Thanks! daniel > >anthony
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.