Author: Gerd Isenberg
Date: 15:09:04 10/14/02
Go up one level in this thread
On October 14, 2002 at 16:01:22, Gian-Carlo Pascutto wrote: >On October 14, 2002 at 12:46:55, Gerd Isenberg wrote: > >>Hi all, >> >>as proposed some weeks ago, first conclusions of using Kogge-Stone and dumb7fill >>slideattack-generation with mmx-registers in IsiChess, which uses legal move >>generation with a kind of an infinite machine. >> >>IsiChessMMX is slightly but considerable faster on Athlon XP (but sucks on P4). > >Two questions Gerd: > Hi Gian-Carlo, >1) What was the slowdown after throwing out the rotated bitboards? About 3% on average in some tactical middlegame positions like BTtest. > >2) What is the gain after all changes? May be the same win after all changes, but in some positions more than 5%. One interesting thing - when i start both versions of my program, let them calculate on the same position, both heve the same CPU percentage of 50%. Surprisingly the MMX-version is four times faster than the rotated lookup version?! > >Looks fascinating, but I'll probably hold off until I get a 64 bit chip. Good idea, hammer will make it a lot easier. Eight additional 64-bit GP-registers, 16 128-bit XMM-registers and the eight MMX-registers. High time for Kogge-Stone and dumb7fill. See you in Leiden, Gerd > >-- >GCP
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.