Author: Vincent Diepeveen
Date: 16:16:16 09/03/02
Go up one level in this thread
here is your old email. Return-Path: <hyatt@cis.uab.edu> X-Authentication-Warning: crafty.cis.uab.edu: hyatt owned process doing -bs Date: Sun, 28 Jul 2002 11:39:49 -0500 (CDT) From: "Robert M. Hyatt" <hyatt@cis.uab.edu> X-X-Sender: <hyatt@crafty> To: Vincent Diepeveen <diep@xs4all.nl> cc: Gian-Carlo Pascutto <gcp@sjeng.org>, Thorsten Greiner <thorsten.greiner@web.de>, <brucemo@seanet.com>, FransMorsch <fmorsch@xs4all.nl>, <Rudolf.Huber@gmx.net>, <sgasch@hotmail.com>, <stefan@meyer-kahlen.de>, <tckerrigan@attbi.com>, <weilljc@club-internet.fr> Subject: Re: Some results with ABDADA On Sun, 28 Jul 2002, Vincent Diepeveen wrote: > Crafty doesn't copy even 1/10 of the data you are copying!! > How much is he copying? I copy about 44K bytes (the TREE structure) to split. With 4 processors, I copy it 4 times when I split from 1 to 4 in the tree... > At 11:01 AM 7/28/2002 +0200, Gian-Carlo Pascutto wrote: > > > > > >On Sat, 27 Jul 2002, Robert M. Hyatt wrote: > > > >> I disagree there. It is a fraction of a percent, overall, during any > >> search I have measured it on. I don't copy a lot of data, since I > >> can be selective... > > > >Does anyone have some concrete data on how evil it is to be copying > >over lots of tree-state data on the common current x86 SMP architectures? > > > >(Assuming a design similar to DTS or PVS) > > > >Vincent has been trying to convince me this is a Very Evil Thing, > >but I'm trying to assess what the impact is going from 4k to 1k or even to > >+- 150 bytes. > > > >I've already done the first, but this includes quite a bit of > >int->char->int conversions which I assume are also evil on current CPUs. > > > >The second would require a more through redesign. > > > >-- > >GCP > > > > > > > -- Robert Hyatt Computer and Information Sciences hyatt@cis.uab.edu University of Alabama at Birmingham (205) 934-2213 115A Campbell Hall, UAB Station (205) 934-5473 FAX Birmingham, AL 35294-1170 On September 03, 2002 at 16:33:12, Robert Hyatt wrote: >On September 03, 2002 at 16:25:23, Gian-Carlo Pascutto wrote: > >>On September 03, 2002 at 16:23:20, Eugene Nalimov wrote: >> >>>On September 03, 2002 at 16:10:33, Gian-Carlo Pascutto wrote: >>> >>>>On September 03, 2002 at 15:56:10, Eugene Nalimov wrote: >>>> >>>>>Wrong. Please re-read Vincent's message. "64k on Cray" *probably* was Bob's >>>>>number, but "44k in Crafty" not -- at least I cannot deduce that from Vincent's >>>>>message. >>>> >>>>I have it in personal email :) >>> >>>Does it says that all 44k are always copied? Because code definitely copies much >>>less. >> >>I think that's what Robert implied (sorry, email is on another machine). >> >>But I agree with you that's not what looks like the code does. >> >>I just wanted to point out Vincent never made up the 44k number, it was >>Robert that told Vincent that was his overhead. >> >>-- >>GCP > > >Again, "max overhead". Which might on rare occasions actually be hit. Perhaps >in fine 70 after several minutes, it might have to copy most of that stuff... > >This was when vincent was telling you your split overhead was too high, and >I pointed out mine _could_ be significantly higher with no ill effects since >fine 70 runs just "fine" on my box.. :)
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.