Author: José Carlos
Date: 23:57:19 11/27/02
Go up one level in this thread
On November 28, 2002 at 01:50:49, Russell Reagan wrote: >On November 27, 2002 at 19:03:38, José Carlos wrote: > >> The program is still far from playing a whole game, and I'm not gonna have any >>spare time in the next few months, but I expect it to outplay Averno completely >>(though I'm having a horrible branching factor with MTD). > >Hello José, Hi Russel, >I too am writing a bitboard engine, and mine is also using non-rotated >bitboards. I'm almost done with move generation (still need promotion and en >passant). How fast is your non-rotated bitboards program compared to other >programs? I'm not focusing in speed at all. I chose to start easy and, eventually, I could rewrite critical parts for speed. However, I've found it's fast enough so far. My main source of comparison is Averno (0x88 + piece-lists). At this moment Anubis (my new program) does half the nps as Averno, being the move generator much slower. But I add all evaluation I want without hurting speed much, which is very good. Note also that I do SEE everywhere in Anubis (not only in quies but also in regular search to sort captures), but I don't use SEE at all in Averno. >I know that Gerd said IsiChess is a little faster when using his MMX >implementation of non-rotated bitboards, but I'm only using a C/C++ >implementation of those routines, and I'm afraid that once I get it >all finished that it's going to be fairly slow. What do you think? The only assembler I'm using is PopCount (from Athlon optimization manual) and FirstOne (from another manual I don't recall). I'm not using MMX at all. As for S.Wescott algorithms (Kogge-Stone etc.) I'm not using them. Only in the last days I did implement them (in C) to compare with mine, but I haven't had time to do a serious comparison. >I have a lot of functions I can inline also, so that might give me a boost. >Currently I don't have any functions inlined. Once all of the attack generation >and bit location lookup (which gets called a lot) gets inlined, that could be >significant, but I don't know how much inlining usually helps. I'm also passing >bitboards to those functions by value, so inlining can help there too. Does >anyone know how much inlining can help? A function that is called a lot is speed critical, so inlining must give a good speed improvement. How much? Well, it depends on the program, of course... >Russell José C.
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.