Author: Uri Blass
Date: 12:24:36 08/15/04
Go up one level in this thread
On August 15, 2004 at 15:16:17, Mathieu Pagé wrote: >On August 15, 2004 at 12:13:59, Uri Blass wrote: > >>On August 15, 2004 at 12:01:56, Andrew Platt wrote: >> >>>Do not underestimate the cost of writing to Windows terminals! Particularly if >>>the output is unbuffered it can be quite intense. Try this: Open a file as a >>>test and send the output to there (default buffering of fopen is fine). See if >>>you see the same slowdown. >>> >>>Andy. >> >>It cannot be the cost of writing because the difference between times when there >>is no additional writing is also significantly bigger. >> >>I do not understand what is unbuffered output. >> >>I use fflush(stdout) after every printf but I know that I need to use >>fflush(stdout) after lines that I print (otherwise the program may work in text >>mode but not under winboard) > >Hi Uri, > >You are right, in order to work in Winboard you need to do fflush() after each >output if you want winboard to get it instantly. However there is some >informations that you send to winboard that did not need to be send so rapidely >(like PV informations). So I sugest you to try (i did not) to use fflush only >after critical informations like returning move, pong, etc. But not after pv. >I'm absolutely not sure if it will help, so you should test it. > >Mathieu P. If the problem is speed of fflush I could expect no slow down when there is no additional fflush but it is not the case. I already found a way to do the code more clearer and faster by dividing the file iterate.c to 2 files when one of them is for searching the root moves like crafty. Movei still has not enough files relative to crafty and I probably should divide the program to more files. Uri
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.