Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: 0x88 and move generator speed

Author: Vincent Diepeveen

Date: 18:33:25 02/02/01

Go up one level in this thread


On February 02, 2001 at 06:59:03, Severi Salminen wrote:

>>But i don't see the speedup when compared to crafty when
>>talking about branches here. You use on average 2 JZ instructions
>>for each move. Let's do a very non-educated guess and estimate
>>that from each 2 JZ instructions at least 1 is mispredicted.
>>
>>That means that you lose in number of clocks (at least) on the
>>  P3 : 10
>>  P4 : 20
>>
>>So you waste 10 CLOCKS on a P3!!!!!!!
>>Just to get a single instruction out of a bitboard!
>>Added to that is a lot of extra overhead also, as you use
>>a lot of complex instructions. psllq i need to figure out what
>>it does but perhaps you can already post that!.
>
>Could you give me a basic rule of thumb for this branching stuff. The less we
>have jumps the faster we are? And what about when processor predicts right the
>destination?

You can freely download the specs of PIII processor and a lot of
info about it at at intel headquarters

www.intel.com then go to developers and look for the manuals for the PIII!

Basic rule of thumb is that something that looks nice in C
isn't always nice in assembly:

   a = solvemyproblem(X);

Doesn't mean that your implementation is theoretic perfect without
knowing what solvemyproblem is actually doing!

Ignoring the assembly output by the compiler is simply no good.

>Severi



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.