Author: Frank Phillips
Date: 02:55:00 12/06/98
Go up one level in this thread
On December 05, 1998 at 21:36:46, Roberto Waldteufel wrote: > >On December 05, 1998 at 17:22:43, Frank Phillips wrote: > >>On a PII are programs based on bitboards inherently faster than those based on >>integer arrays? > >Hi Frank, > >On the Pentium, you have two very useful assembler instructions for bitboards, >the bsf and bsr instructions that find the first or last 1 bit in a 32-bit >number. These instructions are ideally suited to bitboard operations, but they >used to be terribly slow. Not any more, however: on the PII and PPro processors, >these instructions are very fast. As long as the testing is done on CPU >registers (not in memory), a bsf or bsr instruction executes in either 1 or 2 >CPU clocks (depending on pipelining efficiency). So on a PII or PPro I think >bitboards are a really big win, but if your program is going to be used on older >machines, you might think twice about using the bit scan instructions - they can >by more than 300 times slower in the worst case. > >Be careful how you code your bitboard routines if you try them. You will need to >code them in assembler, and it's quite fiddly to get right. But when you do, >bitboards fly on a PII. You might gain a good insight into bitboards by looking >at Earnst Heinz's Darkthought web site. If you are interested in implementing >rotated bitboards, Bob Hyatt is preparing (has prepared?) an article for ICCA >publication on the subject. He sent me a rough draft copy, and I found it >explains rotated bitboard implementation very well indeed. Whether or not >rotated bitboards are right for you depends on your program's data structures >and also on your implementation. My own experiments on a PII have shown that I >can generate the same information using ordinary bitboards faster than I can >using rotated bitboards, but the implementation I use relies very heavily on the >bit scan instructions, and as I said, these only became fast on the more recent >pentiums, and the rotated bitboards would undoubtedly be far superior on any >machine that did not offer this fast bit scan capability (eg pentium plain and >pentium MMX, also 486 has bit scans but they are slow) > >Hope this helps. Good luck! > >Roberto Roberto Thanks for the answer. I was wondering whether to make the leap to bitboards having just ‘completed’ the barebones of my first program based on the integer array approach in TSCP. I have no plans for the results of my rather mediocre efforts – just for personal interest. The problem as always is getting hold of information . Bob is your paper available other than by subscribing to ICCA? A while ago someone praised Bob Hyatt for all the help he gives to beginners and others alike. I whole heartedly endorse this and would also praise Tom Kerrigan for making TSCP readily available. As someone new to both programming and chess programming I found it a wonderful resource, without which I would probably never have got started. (Crafty is another well documented source, but way up the learning curve for a complete beginner). Frank
This page took 0.01 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.