Author: Matt Taylor
Date: 23:23:57 01/30/03
Go up one level in this thread
On January 30, 2003 at 08:42:07, Russell Reagan wrote: >On January 30, 2003 at 04:55:09, Bruce Moreland wrote: > >>This speed stuff is worth thinking about, but it's way more important to think >>about move ordering, pruning, extensions, and good eval. > >I agree. I had a little brain fart when thinking about it. I remembered Ed >saying "one instruction" and recalled it as saying "one cycle", and then I went >crazy thinking about the possibilities of generating moves at a rate of 1 move >per handful of cycles using a series of jump tables to jump tables. Then I >realized this must be too good to be true. But yes I agree with you, it's not >really important. Jump tables used to be free because the CPU had to fetch the next instruction anyway. Jump tables weren't free on the Pentium, and an interesting Pentium bug made the CPU mispredict things, and sometimes it would mispredict instructions that -can't- branch. However, it was still cheap. Pipelines have grown significantly since the (IIRC) 7-stage Pentium, and it is no longer even cheap. If you have 1,000 different cases, jump tables are obviously cheaper than 1,000 different compares. They are still useful...but in an eval where you can reduce the number of cases, there should be a better way. -Matt
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.