Bitboards ?: C# vs C++

[hgm]

To get this clear: your 'entire board' is an array of 64 bytes, indexed by square numbers that run from 0 to 63, right? So when you mask it, that is not a single machine operation. It might be 8 operations on 64-bit words, because I don't think the compiler would be smart enough to use SIMD instructions for this.

I still have no clue how you proceed after that. You talk about 'searching' (past a blocker), but searching is in general a much slower process than finding, where you first do many steps that do not find what you search.

[Chessnut1071]

To get this clear: your 'entire board' is an array of 64 bytes, indexed by square numbers that run from 0 to 63, right? So when you mask it, that is not a single machine operation. It might be 8 operations on 64-bit words, because I don't think the compiler would be smart enough to use SIMD instructions for this.

I still have no clue how you proceed after that. You talk about 'searching' (past a blocker), but searching is in general a much slower process than finding, where you first do many steps that do not find what you search.

It's more like 64 x 2 x 9 + 64 + 64. Actually, I don't mask the same way you do, I used that term, but, it's a simple array lookup with the KQRB[64,2,9] - all_pieces[ ] - and enemy_ check_mask[] arrays. The counter in front of each vector (ray) gives the legal steps terminated by the blocker. Two array lookups gives the move, direction, blocker and check. A third gives discovered check. Ent passant is a bit messy right now and I need to clean up pawn promotion. This approach is more than 50x faster than the one you described above.

I'd list the code, but, it's a GD mess and not cleaned up. In other words, it's not pretty right now, although it works.

[Chessnut1071]

To get this clear: your 'entire board' is an array of 64 bytes, indexed by square numbers that run from 0 to 63, right? So when you mask it, that is not a single machine operation. It might be 8 operations on 64-bit words, because I don't think the compiler would be smart enough to use SIMD instructions for this.

I still have no clue how you proceed after that. You talk about 'searching' (past a blocker), but searching is in general a much slower process than finding, where you first do many steps that do not find what you search.

It's more like 64 x 2 x 9 + 64 + 64. Actually, I don't mask the same way you do, I used that term, but, it's a simple array lookup with the KQRB[64,2,9] - all_pieces[ ] - and enemy_ check_mask[] arrays. The counter in front of each vector (ray) gives the legal steps terminated by the blocker. Two array lookups gives the move, direction, blocker and check. A third gives discovered check. Ent passant is a bit messy right now and I need to clean up pawn promotion. This approach is more than 50x faster than the one you described above.

I'd list the code, but, it's a GD mess and not cleaned up. In other words, it's not pretty right now, although it works.

corrigendum
woops, KQRB is [64 x 8 x 2 x 9] + 64 + 64

[hgm]

I have no idea what your all_pieces or enemy_check_mask arrays might contain, or for what purpose you might use those. Nor what KQRB might contain, and what it gets indexed with.

[lithander]

I just realized System.Numerics.BitOperations class "provides utility methods for intrinsic bit-twiddling operations. The methods use hardware intrinsics when available on the underlying platform; otherwise, they use optimized software fallbacks." See the doc. So no need for preprocessor directives, assuming Microsoft's software fallback code performs well.

I also wasn't aware of that. I guess I should get to work on a bitboard engine... it won't become any more convenient than that.

Sherlock - you found a 10 year old api.

System.Numberics.BitOperations was introduced with .Net Core 3.0 which released in 2019.

[Chessnut1071]

I have no idea what your all_pieces or enemy_check_mask arrays might contain, or for what purpose you might use those. Nor what KQRB might contain, and what it gets indexed with.

enemy_check_mask = Kd5
where 1 = 8; 2 = - 8; 3 = 1; 4 = -1; 5 = 9; 6 = -9; 7 = 7; 8 = -7
50010070
05010700
00517000
33304444
00826000
08020600
80020060
00020006

all_pieces

each square either empty or piece type: 1 - 6 & 7 - 12

hope it helps

[dangi12012]

I just realized System.Numerics.BitOperations class "provides utility methods for intrinsic bit-twiddling operations. The methods use hardware intrinsics when available on the underlying platform; otherwise, they use optimized software fallbacks." See the doc. So no need for preprocessor directives, assuming Microsoft's software fallback code performs well.

I also wasn't aware of that. I guess I should get to work on a bitboard engine... it won't become any more convenient than that.

Sherlock - you found a 10 year old api.

System.Numberics.BitOperations was introduced with .Net Core 3.0 which released in 2019.

The question was about dotnet FRAMEWORK - and how to access bmi operations there. The answer is system.numerics nuget package.

[Chessnut1071]

I just realized System.Numerics.BitOperations class "provides utility methods for intrinsic bit-twiddling operations. The methods use hardware intrinsics when available on the underlying platform; otherwise, they use optimized software fallbacks." See the doc. So no need for preprocessor directives, assuming Microsoft's software fallback code performs well.

I also wasn't aware of that. I guess I should get to work on a bitboard engine... it won't become any more convenient than that.

Sherlock - you found a 10 year old api.

System.Numberics.BitOperations was introduced with .Net Core 3.0 which released in 2019.

The question was about dotnet FRAMEWORK - and how to access bmi operations there. The answer is system.numerics nuget package.

I tried Syetem. Numerics, where do you get the nuget package? I have 4.7.3. and badly need bfs. Microsoft has a new 6.0 for 2022 which should have it, but, I haven't downloaded it yet.

[hgm]

I have no idea what your all_pieces or enemy_check_mask arrays might contain, or for what purpose you might use those. Nor what KQRB might contain, and what it gets indexed with.

enemy_check_mask = Kd5
where 1 = 8; 2 = - 8; 3 = 1; 4 = -1; 5 = 9; 6 = -9; 7 = 7; 8 = -7
50010070
05010700
00517000
33304444
00826000
08020600
80020060
00020006

all_pieces

each square either empty or piece type: 1 - 6 & 7 - 12

hope it helps

OK, so all_pieces is the mailbox board, and enemy_check_mask a board size table containing the number of the direction you have to move in to hit a king on d5. (Do you have 64 such tables?) The latter does not seem useful for knowing where a slider gets blocked.

Now what is in KQRB? And what is it indexed by? I assume 64 is the square of origin, and the 2 is color? What do you use in the 9 dimension?

[Chessnut1071]

I have no idea what your all_pieces or enemy_check_mask arrays might contain, or for what purpose you might use those. Nor what KQRB might contain, and what it gets indexed with.

enemy_check_mask = Kd5
where 1 = 8; 2 = - 8; 3 = 1; 4 = -1; 5 = 9; 6 = -9; 7 = 7; 8 = -7
50010070
05010700
00517000
33304444
00826000
08020600
80020060
00020006

all_pieces

each square either empty or piece type: 1 - 6 & 7 - 12

hope it helps

OK, so all_pieces is the mailbox board, and enemy_check_mask a board size table containing the number of the direction you have to move in to hit a king on d5. (Do you have 64 such tables?) The latter does not seem useful for knowing where a slider gets blocked.

Now what is in KQRB? And what is it indexed by? I assume 64 is the square of origin, and the 2 is color? What do you use in the 9 dimension?

The KQRB board: below is the code for squares 30 and 31. In the example square 30, or b4, has 4 moves north +8, 38,46,54 & 62, 3 moves south, 1 move west, and 6 moves east, 1 move north west, 3 moves south east, 4 moves north east and 1 move south west.

if the square > o the vector is terminated, capture piece and direction are saved along with check from the check_mask. Discovered check combines piece board with the check mask, using the direction (1-8) with the piece type 3,4 or 5.

30,4,38,46,54,62,0,0,0,
30,3,22,14,6,0,0,0,0,
30,1,31,0,0,0,0,0,0,
30,6,29,28,27,26,25,24,0,
30,1,39,0,0,0,0,0,0,
30,3,21,12,3,0,0,0,0,
30,4,37,44,51,58,0,0,0,
30,1,23,0,0,0,0,0,0,

31,4,39,47,55,63,0,0,0,
31,3,23,15,7,0,0,0,0,
31,0,0,0,0,0,0,0,0,
31,7,30,29,28,27,26,25,24,
31,0,0,0,0,0,0,0,0,
31,3,22,13,4,0,0,0,0,
31,4,38,45,52,59,0,0,0,
31,0,0,0,0,0,0,0,0,

I'm not sure if this follows anybody's code, it just made logical sense to me. So, where am I wrong?