Author: Tom Kerrigan
Date: 16:05:36 08/16/00
Go up one level in this thread
On August 16, 2000 at 16:56:21, Andrew Dados wrote: >On August 15, 2000 at 19:24:25, Tom Kerrigan wrote: > >>On August 15, 2000 at 18:33:26, Larry Griffiths wrote: >> >>>On August 15, 2000 at 17:03:24, Tom Kerrigan wrote: >>> >>>>On August 15, 2000 at 13:43:55, Larry Griffiths wrote: >>>> >>>> >>>>No matter how you generate your moves, you should always be searching captures >>>>before killers. >>>> >>>>Also, at this early stage in your program's development, you should just >>>>generate all your moves at once and move on. Doing this "quick killer" stuff >>>>doesn't give you a very big performance gain and it makes your code much more >>>>complicated. >>>> >>>>-Tom >>> >>>I am using a lot of code concepts from my previous chess program. >>>This program is more object-oriented and uses bitboards heavily. >>>My "quick killer" code reduces the search time up to 50% and I consider >>>that a very good performance gain. >> >>That sounds odd to me. >> >>About 80-90% of your nodes should be in q-search, where you don't have to >>generate killers. So the most you can possibly hope for from "quick killer" >>stuff is 10-20% (and that obviously won't be achieved...). >> >>-Tom > >A well written qsearch shouldn't use more then 60% total nodes imo. > >Note a 'paradox' : if you can cut qsearch nodes by less then half from 90% down >to 50% your program overall speedup is up to 5 times (non qsearch nodes go up >from 10% to 50% of all). I think you're counting qsearch nodes differently than me. The way I count them, I believe it's impossible to be < 60%. -Tom
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.