Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: definition of clones: Danchess an Crafty (another note)

Author: Robert Hyatt

Date: 11:00:14 02/18/04

Go up one level in this thread


On February 18, 2004 at 10:34:01, Dan Honeycutt wrote:

>On February 17, 2004 at 20:58:28, Robert Hyatt wrote:
>
>>On February 17, 2004 at 17:38:13, Dan Honeycutt wrote:
>>
>>>I realize I'm adding where you are subtracting due to different numbering.  What
>>>I'm saying is that 3/4 of the time I still have to make "an adjustment" to the
>>>value bsf/bsr gives.  Do you not have to make an adjustment 4/4 of the time, or
>>>do you have to make 2 adjustments?
>>>
>>>Dan H.
>>
>>
>>I do it at the end, one time.  IE for the Opteron, it looks like this:
>>
>>movq    %rax,63
>>bsfq    %rbx, %rcx
>>subq    %rcx, %rax    ; done, result in rax (64 bit reg).
>>
>>for X86, I do it in three steps:
>>
>>first check one half, then add 32,
>>second check other half
>>subtract from 63
>>
>>I can post the x86 if you want...
>
>I'm with you.  on x86 I'm
>
>check one half, if there is a bit I'm done
>check other half
>add 32
>
>I get a slight advantage on 32 bit but more on 64 bit where I do nothing but you
>still have to subtract.
>
>Thanks for the help.
>Dan H.


I don't think this is a major issue, but the register needed for the subtract
certainly hurts on an X86 box.  AMD64 has 8 extra registers so it becomes less
of an issue 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.