World Draughts Forum

And here the link towards the dropbox folder containing the latest Perft(), including sources.

Bert

https://www.dropbox.com/sh/2stq4b20k64wcxx/AAC1Asa5wxWFUhmofY0dJyUNa?dl=0

Ok... I will compile it and let you know the results somewhat later. I believe you that raymask can be faster as magics ...

Will have a look at the Damage GUI later, and will post a reply.
Bert

What I would like to know is what the various DXP buttons (Make, Wait, Run, Terminate) are doing.
I already saw that with 'Wait' it starts to wait for a connection and when I start an engine after this it does connect, but I ...

I also tried now the approach also used by Harm Jetten to detect first non-empty square in the main KingCapture detection routine.
Timing for the last Perft(9) position is now 7.40s, 7.44s, 7.43s and 7.43s.
See below code for the routine.
I guess some small optimizations are still possible.
As you ...

To determine king captures I basically do the following: ((magic_mask_for_current_direction & opposite_pieces & ~edge_squares) >> current_direction) & empty_squares;
This gives you the first empty square after the opposite piece you are allowed to capture.
It is possible to generate another set of ...

Joost, think further optimization does not make sense, as the Movegenerator is already very fast.
As I want to make some changes to the overall architecture of Damage, I want to spent still somewhat time on the Movegenerator.
I still believe that the King Capture (even without Magic) can be faster ...

I have removed the KingMove Mapping in the capture routine.
It seems that The Perft(9) position 2 is around 8% - 9% faster.
8.51s -> 7.69s

But hope Joost can confirm, and hope i approach now his results.

See the link as I have updated the Dropbox.

Bert

https://www.dropbox.com/sh/9yx1f7z3x1q6163 ...

BertTuyt wrote: Which timing function did you use?
I might need another one
Bert

I use QueryPerformanceFrequency() and QueryPerformanceCounter().

If you have your board class with bitboards for black, white and kings, then occupied == black ^ white. However, if you now define empty_squares == ~occupied, you have introduced 14 bits on the ghosts squares for empty_squares. This means that in patterns as opponent_pieces & (empty_squares << dir ...

I will post the Intel PGO results somewhat later, installing VS2015 update 3 broke something in the Intel VS2015 integration.
I tried with /QProf-gen and /QProf-use and that didn't work, I guess I have to repair or reinstall the Intel compiler.
Later...

Ok, here are the Intel results with PGO. It ...

These are the results for the Intel compiler without PGO.


Perft(1) N = 9 0.00 sec. KN/sec = 0
Perft(2) N = 81 0.00 sec. KN/sec = 0
Perft(3) N = 658 0.00 sec. KN/sec = 0
Perft(4) N = 4265 0.00 sec. KN/sec = 0
Perft(5) N = 27117 0.00 sec. KN/sec = 0
Perft(6) N = 167140 0.00 sec. KN/sec = 0
Perft(7 ...

Joost, thanks.
I would also like to see an as-is comparison.
So just run the executable on your (faster) computer, then I also better know , and can quantify what the effect of the Intel compiler is.
Bert

Bert,

These are the results for the MSVC compiler @ 3600MHz. I took the liberty to add 1 or ...

Why do you need the edge_squares masking? If you have the "ghost" squares at each edge, then the opposite_pieces & (empty_squares << current_direction) would be the cheapest way to isolate all opposite pieces with an empty piece behind them.

To be honest, I never looked at it in very great ...

Here the link to my Dropbox folder containing a Visual Studio 2015 Perft project (and Perft executable).
Including the Movegen source code.

Interesting what the Perft results are on a fast 4 Ghz machine and/or with better compiler optimization and/or Intel compiler.

Think it still could be faster ...

To determine king captures I basically do the following: ((magic_mask_for_current_direction & opposite_pieces & ~edge_squares) >> current_direction) & empty_squares;
This gives you the first empty square after the opposite piece you are allowed to capture.


Why do you need the edge_squares ...

Yesterday ( = Saturday) I tried to debug the Magic code which I had started the weeks before.
It also now works (as the Perft numbers are the same), but for an unknown reason it is not faster.
I only use Magic for the King Moves, so not the King Capture detection (or alike).


It is possible that ...