Author: Sune Fischer
Date: 10:04:05 01/28/02
Go up one level in this thread
On January 28, 2002 at 12:25:28, Vincent Diepeveen wrote:
>>Bah, you guys are no fun, live a little, risk it all!;)
>
>In fact there has been more experimented here than you thought :)
>
>But the hard facts are that if you would go for every thought of a
>18 year old kid who can't program, that you are busy for another 1000
>years checking out all the crap.
>
>Note that in this case it is not entire crap. The idea to fiddle with a
>bunch of MMX instructions in order to get something done in a bitboard
>concept is pretty interesting. Nevertheless it depends heavily on your
>evaluation.
I don't understand his MMX idea, I think it's just to increase speed in 64-bit
operations, I don't see why they should be needed here. We don't need to
re-reverse the bitboards, they are incrementally constructed.
I've made my little rook attack algorithm here on paper, it is about 20 64-bit
operations in total - no table lookups and no if's or while's!
Think of the potential speed increase when we get 64-bit chips, the rotated
bitboards will still be handicaped by the table lookups.
Anyway, I'll write down the algorithm now and post it here, then you can tell me
if it will stand a chance against rotated ;)
I hope there are no mistakes, I haven't tested it.
>IMHO bitboards are good for pawn structure but they suck further. Note
>there are another zillion solutions to pawn structures and 32 bits
>bitboards which are fast too.
>
>
>>If it works it should be faster, I'll think it through and if I can't find any
>>flaws I'll give it a try.
>
>>>Only the diagonals you need to take a look at at the design what is faster
>>>for them. if that problem isn't solved in reversed bitboards, then obviously
>>>rotated are going to be easier for you to take a look at.
>>
>>K, I will.
>
>the idea of rotated bitboards is pretty simple, it basically solves
>the problem of the sliding pieces. So when someone mentions reversed
>bitboards i don't care for knights for example, but *only* for
>bishops, rooks.
Yes we are talking about the sliding pieces only, the others are easy.
I know the idea of rotated bitboards is simple, but the implementation is not ;)
>I have missed the tricks how to do that with other solutions than rotated
>bitboards in an efficient way.
>
>>
>>>It seems there are however many methods at the K7 + intel processors to
>>>use with rotated bitboards. In isichess it seems smaller tables are
>>>getting used than in crafty for this.
>>>
>>>Others report they do it even different, but their speedresults are
>>>not impressive.
>>>
>>>>I found this link (it was actually dead, but google had it cached so I put it up
>>>>for a short while)
>>>>http://www.fys.ku.dk/~fischer/Temp/New%20Technology.htm
>>>>
>>>>It is about reversed bitboards, the author claims it is faster than rotated:
>>>>
>>>>"When taking account of memory latencies, calculating the piece attacks using
>>>>the forward and reverse bitboards can be done significantly faster due to total
>>>>independence on lookup tables and complex calculations. There are a few slight
>>>>snags with the diagonal calculations, but they are minor."
>>>>
>>>>
>>>>His description ("significantly faster") entrigues me (hehe) but I can't find
>>>>anything else on reversed bitboards and I've never heard of them before.
>>>>
>>>>What's the catch, is there something he is not telling?
>>>>
>>>>-S.
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.