Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Intel Willamette and chess

Author: David Blackman

Date: 22:42:19 02/18/00

Go up one level in this thread


On February 18, 2000 at 21:11:58, Tom Kerrigan wrote:

>
>You just got the docs for the SIMD instructions? Where? Does it look like the
>chips have a FindFirstBit instruction, etc.?
>
>-Tom

I didn't see the docs, but read some of the discussion. The thing has a 20 stage
pipeline. They claim they use every trick there is for branch prediction, but
still, you should have a close look at your program and ask how you can make
your branches easier to predict (or remove them completely). Also, how you can
overlap various computations so you don't get 20 clock pipeline stalls. That
kind of stuff is probably going to be the most important optimisation for most
future chips, as pipelines get longer.

So, it's time to re-evaluate 12x10 vs 0x88 vs bitboards vs attack maps in the
light of long pipelines. I suspect bitboards or attack maps look better for eval
purposes (but they were always good there).



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.