Author: Murray
Date: 16:25:54 11/08/02
Go up one level in this thread
(snip) >>Just start 2 processes instead of 1, let one process do as if it is >>single cpu and play the game with that. the other process just fills the >>hashtable at the same time. >> Hmm so do you mean that the first process doesn't actually fill the hash table but tells the 2nd process what put add and which key to insert it? I can see that saving a bit of time, but I found with my program that the biggest hit was the probe from the hash table (not the update) as the probe was the one doing all the damage with most probes reading from main memory rather than a local CPU cache. the update however follows the probe and so there is more chance of hitting the CPU cache. (snip) > >Don't know where you have been, but the parallel search in Crafty took two weeks >to write. It may well _still_ have a bug or two that I haven't seen show up, of >course. >But the idea is to start off with something that has a chance of producing good >performance. Just sharing hash tables isn't going to offer much. What parallel algorithm would you recommend for targetting 2 CPU systems only? Which is the simplest to implement and works reasonably well for 2 CPU? PV split or some other?
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.