Author: Gerd Isenberg
Date: 10:38:12 08/27/05
Go up one level in this thread
On August 27, 2005 at 06:44:14, Alessandro Damiani wrote:
>On August 27, 2005 at 06:40:43, Alessandro Damiani wrote:
>
>>On August 27, 2005 at 05:54:16, Gerd Isenberg wrote:
>>
>>>On August 27, 2005 at 05:33:17, Alessandro Damiani wrote:
>>>
>>>>On August 27, 2005 at 03:12:23, Gerd Isenberg wrote:
>>>>
>>>>>On August 26, 2005 at 16:11:53, Dan Honeycutt wrote:
>>>>>
>>>>>>Run the lowest bits down the a or h file and you can do pawn captures without
>>>>>>having to mask those files.
>>>>>>
>>>>>>Best
>>>>>>Dan H.
>>>>>
>>>>>Hi Dan,
>>>>>
>>>>>a very nice idea! Keeping the a-file in least significant byte and the h-file in
>>>>>the most significant byte (or vice versa) to avoid a-h-wraps for pawn attacks
>>>>>and further ands.
>>>>>
>>>>>eg. with a1 == 0:
>>>>>
>>>>>a1 a2 a3 a4 a5 a6 a7 a8
>>>>>b1 b2 b3 b4 b5 b6 b7 b8
>>>>>c1 c2 c3 c4 c5 c6 c7 c8
>>>>>d1 d2 d3 d4 d5 d6 d7 d8
>>>>>e1 e2 e3 e4 e5 e6 e7 e8
>>>>>f1 f2 f3 f4 f5 f6 f7 f8
>>>>>g1 g2 g3 g4 g5 g6 g7 g8
>>>>>h1 h2 h3 h4 h5 h6 h7 h8
>>>>>
>>>>>"left" attacks of white pawns ::= white pawns >> 7 ; eg. e4:d5
>>>>>"right" attacks of white pawns ::= white pawns << 9 ; eg. e4:f5
>>>>>"left" attacks of black pawns ::= black pawns >> 9 ; eg. e4:d3
>>>>>"right" attacks of black pawns ::= black pawns << 7 ; eg. e4:f3
>>>>>
>>>>>Gerd
>>>>
>>>>A nice idea indeed, but then vertical shifting needs masking, e.g. shiftUp(e8) =
>>>>f1. So, no real gain in the end, I guess.
>>>>
>>>>Or do I miss something?
>>>>
>>>>Alessandro
>>>
>>>Since pawns usually don't stay on back ranks, there is no need to worry about
>>>rank1/8 pawn push wraps. So for pawns there is a small gain for their attacks.
>>>For x86-32 it might be negligible since the double register shifts are much more
>>>expensive than two "ands" with immediate operands.
>>>
>>>For x86-64 there is only one reg64 shift. The "and reg64, mem64" reads the mask
>>>from memory - so the relative gain might be more significant here.
>>>
>>>Gerd
>>
>>Sorry, I mistakenly thought to apply this method to all piece bitboards. But,
>>right, if only applied to pawns then there is no drawback.
>>
>>Alessandro
>
>To determine the squares in front of the pawns we still need to mask. Something
>used in eval.
>
>Alessandro
Alessandro,
i fail to see why it is necessary to mask for pawn pushs:
a1 a2 a3 a4 a5 a6 a7 a8
b1 b2 b3 b4 b5 b6 b7 b8
c1 c2 c3 c4 c5 c6 c7 c8
d1 d2 d3 d4 d5 d6 d7 d8
e1 e2 e3 e4 e5 e6 e7 e8
f1 f2 f3 f4 f5 f6 f7 f8
g1 g2 g3 g4 g5 g6 g7 g8
h1 h2 h3 h4 h5 h6 h7 h8
let say white pawns on a7,b7, black pawns on g2,h2
a1 a2 a3 a4 a5 a6 wp a8
b1 b2 b3 b4 b5 b6 wp b8
c1 c2 c3 c4 c5 c6 c7 c8
d1 d2 d3 d4 d5 d6 d7 d8
e1 e2 e3 e4 e5 e6 e7 e8
f1 f2 f3 f4 f5 f6 f7 f8
g1 bp g3 g4 g5 g6 g7 g8
h1 bp h3 h4 h5 h6 h7 h8
square shift left one (bit shift right) for white pawn pushes,
square shift right (bit shift left) one for black pawn pushes,
to get the promote targets.
a1 a2 a3 a4 a5 a6 wp->>
b1 b2 b3 b4 b5 b6 wp->>
c1 c2 c3 c4 c5 c6 c7 c8
d1 d2 d3 d4 d5 d6 d7 d8
e1 e2 e3 e4 e5 e6 e7 e8
f1 f2 f3 f4 f5 f6 f7 f8
<<-bp g3 g4 g5 g6 g7 g8
<<-bp h3 h4 h5 h6 h7 h8
so result is
whitePawnPromotePushes{a8,b8}
blackPawnPromotePushes{g1,h1}
Since i like to combine all bitboards with the same mapping, i would use this
"vertical" mapping for all other pieces as well. Fill stuff works as well as
rotated of course (except that the none rotated maps consequtive squares on a
file rather than on a rank). While rooks left/right kogge stone need no
wrap-ands, but therefor up/down. Diagonal attacks needs wrap-ands anyway.
Or did i miss something?
Gerd
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.