Author: Sune Fischer
Date: 14:28:19 08/18/03
Go up one level in this thread
On August 18, 2003 at 17:11:12, Russell Reagan wrote: >On August 18, 2003 at 16:24:09, Sune Fischer wrote: > >>If I cared about simplicity I would not be going parallel, I want performance! >>:) > >Well if you're going for "real" parallel, that's great. If you're only trying to >do "kind of" parallel, then I don't think the millisecond it costs to start up a >couple of threads before a search is going to hurt performance that much. > >I think we're talking about different things though. I'm only talking about >starting up threads for each root move, so maybe 35-40 per iteration, and never >more than 2 or 4 at a time. If your approach means that you would have to be >constantly starting threads for each node, then your approach is better of >course :) I think the first move consumes about 90% of the time, so there's only 10% left to run in parallel. That would be a lot of trouble for nothing, IMO, but of course it will be one of the first (primitive) tests :) >>Yes this is what I need, do you think I want to be stuck in windows forever? :) > >Well, if you only need those simple operations, then it would probably help >performance. That would mean non-portability via Windows functions or assembler. >It probably wouldn't be too bad to write a version for Windows (or just use the >Interlocked* functions) and then write a handful of assembler routines for >linux. Yep, specific OS optimizations can come later, first I want it working and portable. >Mutexes/etc. aren't terribly slow in my experience, and will be tuned for >efficiency, but there is some overhead that can't really be avoided. I imagine >there would be hardly any measurable overhead using the Interlocked* functions >or the assembly that they convert to (cmov or whatever). Right, hope that is true. I don't expect to be locking a lot though. -S.
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.