Author: Robert Hyatt
Date: 06:33:56 10/30/03
Go up one level in this thread
On October 30, 2003 at 08:19:05, Anthony Cozzie wrote: >I haven't parallelized Zappa yet, but its always in the back of my mind, and one >of the things I've been thinking about is processes v threads. The advantage of >processes is you can keep your globals, the advantage of threads is shared >tables. > >IIRC, Crafty used to use globals for the board, etc up till about 14-15. Would >you mind running the last version to use globals and the first version to use >the TREE struct on your Opteron? I'm curious what kind of speed improvement you >get on a machine with (at least a few) more registers than x86. > >anthony I don't have an opteron here to play with. The numbers I reported were produced by Eugene at Microsoft on a machine they have. I can tell you that going from globals to the current pointer approach was _way_ less than a 10% performance hit overall. I was hoping for 10% max. I actually got less than 5%. I don't remember the exact numbers, but might be able to reconstruct them by running the test, if it is interesting.
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.