New engine: Peacekeeper

[Sazgr]

I have been making a chess engine, named Peacekeeper, for a while. Finally it have gotten to the point where it seems to work decently, with no obvious bugs . Right now, it beats TSCP most of the time (about 80%). Some people might have seen it on GitHub already, but it is on https://github.com/Sazgr/peacekeeper. Most likely, I will use this thread for any further questions.

Right now the engine uses PESTO's evaluation, which I am planning to replace not far in the future.

[kranium]

I have been making a chess engine, named Peacekeeper, for a while. Finally it have gotten to the point where it seems to work decently, with no obvious bugs . Right now, it beats TSCP most of the time (about 80%). Some people might have seen it on GitHub already, but it is on https://github.com/Sazgr/peacekeeper. Most likely, I will use this thread for any further questions.

Right now the engine uses PESTO's evaluation, which I am planning to replace not far in the future.

Very nice...great work!

"I tried to support MSVC intrinsics, but since I do not have it, there might be errors."
These work in your project for a Windows compile:

Code: Select all

#ifdef _MSC_VER
#include <intrin.h>
#endif

#include "typedefs.h"

inline int popcount(u64 bits) {
#ifdef __GNUC__
    return __builtin_popcountll(bits);
#elif _MSC_VER
    return static_cast<int>(__popcnt64(bits));
#endif
}

inline int get_lsb(u64 bits) {
#ifdef __GNUC__
    return __builtin_ctzll(bits);
#elif _MSC_VER
    unsigned long idx;
    _BitScanForward64(&idx, bits);
    return static_cast<int>(idx);
#endif
}

inline int get_msb(u64 bits) {
#ifdef __GNUC__
    return __builtin_clzll(bits) ^ 63;
#elif _MSC_VER
    unsigned long idx;
    _BitScanReverse64(&idx, bits);
    return static_cast<int>(idx);
#endif
}

You may want to consider going w/ -std=c++20 and using the new intrinsics included in <bit>:
std::popcount
std::countr_zero

https://en.cppreference.com/w/cpp/header/bit

Good luck with your project!

[dangi12012]

If you have #pragma once just declare as static in the header and delete this file:
https://github.com/Sazgr/peacekeeper/bl ... magics.cpp

This is also very small:
https://github.com/Sazgr/peacekeeper/bl ... typedefs.h

But its more of a personal preference. I strongly recommend having as little code noise as possible + KISS principle.

[Sazgr]

If you have #pragma once just declare as static in the header and delete this file:
https://github.com/Sazgr/peacekeeper/bl ... magics.cpp

This is also very small:
https://github.com/Sazgr/peacekeeper/bl ... typedefs.h

But its more of a personal preference. I strongly recommend having as little code noise as possible + KISS principle.

Thank you!
I made the magics array static and did delete magics.cpp, however I feel the typedefs file might be expanded in the future.

[tcusr]

If you have #pragma once just declare as static in the header and delete this file:
https://github.com/Sazgr/peacekeeper/bl ... magics.cpp

This is also very small:
https://github.com/Sazgr/peacekeeper/bl ... typedefs.h

But its more of a personal preference. I strongly recommend having as little code noise as possible + KISS principle.

Thank you!
I made the magics array static and did delete magics.cpp, however I feel the typedefs file might be expanded in the future.

since C++17 you can use inline to declare global variables. this way "lookup_table" will only be initialized once.
i also agree to switch to C++20 so that your code will contain no platform/compiler specific code.

P.S.
i suggest you to use constexpr as much as possible, it will let compiler optimize better and you'll get a noticeable speed up.

[Sazgr]

If you have #pragma once just declare as static in the header and delete this file:
https://github.com/Sazgr/peacekeeper/bl ... magics.cpp

This is also very small:
https://github.com/Sazgr/peacekeeper/bl ... typedefs.h

But its more of a personal preference. I strongly recommend having as little code noise as possible + KISS principle.

Thank you!
I made the magics array static and did delete magics.cpp, however I feel the typedefs file might be expanded in the future.

since C++17 you can use inline to declare global variables. this way "lookup_table" will only be initialized once.
i also agree to switch to C++20 so that your code will contain no platform/compiler specific code.

P.S.
i suggest you to use constexpr as much as possible, it will let compiler optimize better and you'll get a noticeable speed up.

Sorry if I am being really dumb, but could you give an example of where to add cosntexpr? Thank you

[tcusr]

If you have #pragma once just declare as static in the header and delete this file:
https://github.com/Sazgr/peacekeeper/bl ... magics.cpp

This is also very small:
https://github.com/Sazgr/peacekeeper/bl ... typedefs.h

But its more of a personal preference. I strongly recommend having as little code noise as possible + KISS principle.

Thank you!
I made the magics array static and did delete magics.cpp, however I feel the typedefs file might be expanded in the future.

since C++17 you can use inline to declare global variables. this way "lookup_table" will only be initialized once.
i also agree to switch to C++20 so that your code will contain no platform/compiler specific code.

P.S.
i suggest you to use constexpr as much as possible, it will let compiler optimize better and you'll get a noticeable speed up.

Sorry if I am being really dumb, but could you give an example of where to add cosntexpr? Thank you

anywhere as long as the compiler doesn't complain.
this includes const variables/arrays or any functions where it makes sense (this excludes the Hashtable class because it modifies data allocated during runtime).

[Robert Pope]

This inspired me to test the PESTO eval in my engine. PESTO was more than twice as fast - really slick use of an array of PSTs compared to how I do it. However, I'm guessing my cut and paste has some bugs, since it performed 300 elo worse than my eval.

Did you implement a static eval command in Peacekeeper? It would be handy to have an independent version to check against.

[Sazgr]

This inspired me to test the PESTO eval in my engine. PESTO was more than twice as fast - really slick use of an array of PSTs compared to how I do it. However, I'm guessing my cut and paste has some bugs, since it performed 300 elo worse than my eval.

Did you implement a static eval command in Peacekeeper? It would be handy to have an independent version to check against.

I just modified my eval command to be more informative, the eval command should print the middlegame and endgame PST values, the phase and the weighted PST value now.

example:

Code: Select all

position startpos moves e2e4
eval

which outputs

Code: Select all

middlegame pst: -32
endgame pst: -20
phase: 24
weighted pst: -32

[StackFish5]

Congratulations on the new chess engine! Quite impressive.