Author: Vincent Diepeveen
Date: 20:21:56 08/07/03
Go up one level in this thread
On August 07, 2003 at 08:58:21, Gian-Carlo Pascutto wrote: It is a waste of time to actively PROGRAM in assembly for PC applications like this one. I won't say it's a waste for the NT kernel for example. Each new generation of cpu's requires an entire rewrite of the program. But well written C code keeps portable. If Frans each new version first has to rewrite in order to get it to work better at the new processor then some who read this for the first time now know what his daily work is. Working hard to get the current program ported. That he manages to make some modifications as well to it is really amazing. That some people like Sune and i know many other programmers who didn't post here but have loads of inline assembly too written. Half of them regrets it now. Of course taking a look to how assembly works and what is faster for a processor is a different thing. Study is never bad. Wasting time on getting routine X to work in your own written assembly just to get your program 0.001% faster is really a waste of time where you write at the same time that buying an opteron is not interesting yet as it is too expensive. Just buy that bit faster hardware and don't make it in assembly :) Saves time. Saves money. Just throw some hardware at it! Of course i won't say it is smart for everyone to throw 500 cpu's at it. I'll do that for them with DIEP :) Oh by the way in case i didn't mention it. I can compile diep at that machine because i use NO assembly at it. I find it very funny to read that most bitboarders first slow down their program in case of Sune to get 30% faster using inline assembly :) Same for Isenberg's Kogge Stone project. Move generation just in those registers? Hehehehehehehehehe. Waste of your time Gerd! First lookup how many clocks that is going to cost in advance then compare it with how many clocks i lose in nonbitboards generating at 73MLN nps a second at K7-2.127Ghz. I could get that faster even if i would write out black and white and each piece (see how crafty has written out all code there). But how many cycles do i actually lose a move? Now how many cycles is that going to be a move using the ambitious kogge stone? No need for assembly. All i need is a small calculator and an AMD manual to know in advance who is going to be faster generating at that opteron :) >On August 07, 2003 at 08:50:14, Vincent Diepeveen wrote: > >>But then ones program must be either incredible simple or you must be fulltime >>working at your chessprogram! > >Meh. Assembly languange programming is an art that's learned through >practise. I do not do it a lot, so I'm not very good at it. I think >it's hard, but I can beat the compiler, which is what matters. You >don't do it at all probably, so I guess for you it's even harder. > >Then there's people like Gerd Isenberg and Frans Morsch, and apparently >Sune that eat assembly for breakfast. I guess for them it's easy. > >I found an old snippet from Gerd Isenberg once that did the same as >a chunk of code I had written does. Gerd's version was ""a tad"" >smaller and faster. I'm not posting details because they're way too >embarassing. > >-- >GCP
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.