compile errors, but solved !

[tissatussa]

at first i was unable to compile your v3.5.0 source code by 'make' .. [ using GCC version "Ubuntu 9.3.0-17ubuntu1~20.04" ]

so, i download your asset binary for linux and it works !

here's part of my terminal output when doing 'make' :

fortunately i was able to correct your code : the message says "did you forgot to '#include <cstddef>'", so i added that line to evaltable.h .. then compiling went OK without errors : i got a valid binary ! (size 342.5 KiB opposed to asset binary size 277.3 KiB) .. compiling only gave 1 warning, apparently corrected by GCC :

I'm happy i can run Loki in CuteChess !

I want to point out some other issues of your code (being minor though) :

• your README states : _"If compiling on GCC, 'usepopcount=no' should be added when running make" .. so i changed this value into 'no' in the makefile (you write that such variable can also be set / overruled by a command line argument .. i didn't know that !)
• i see this code part in makefile :

ifeq ($(debug), no) # Set debug mode CXXFLAGS += -DNDEBUG endif

This is unclear to me .. i'm a coder but not a cpp programmer .. the option '-DNDEBUG' should be added when debug is no ? I changed this into yes (but no apparent difference in terminal output).

• your linux binary asset is called 'Ubuntu[..]' but i guess it's a general linux binary ? (i DO use Ubuntu, so i have no clue)
• since several months i have a laptop which can do avx2 and bmi2 .. whatever those exactly mean .. it seems those architectures favor in speed, so i prefer to build such code .. i only see the option -march=native in makefile .. can your code be changed accordingly ? Btw. i find some (engine) programmers using some code to detect the optimal user 'profile' while compiling !?

[BimmerBass]

Thank you for opening this issue! I haven't been working on Loki for a while, but this will be fixed when I get back to it.

Since Loki has been developed on Windows 10, I haven't really tested its compatibility with Linux. I have been informed that version 3.5 works on Ubuntu since it needed to, to be able to run on an OpenBench server, but I don't know about the current development version. I think the problem you were experiencing could also be fixed by adding #include <cstdint> to defs.h, but I will try your solution as well. The last warning you received doesn't actually matter since it's only used in a debugging method.

Regarding your other issues:

• Yes, if you wish to build without popcount, you should be able to run the command make use_popcnt=no, so no need to change the makefile itself :)
• If NDEBUG is defined in the executable, the assert() method will be disabled, giving extra speed. Reversely, if it isn't defined, assert will run, which in some cases can cause a big slow down. For example, it would be unacceptable speed-wise to run assert(!pos->InCheck()) in a release build - NDEBUG prevents this.
• I haven't tested Loki 3.5's compatibility with other Linux platforms than Ubuntu, so that's why it is specified. I think it runs on most of them though.
• AVX2 and BMI2 speeds up execution of integer arithmetic (for example: with AVX2 you can multiply 32 8-bit integers with 32 other 8-bit integers in a single instruction -> 32x faster), but this doesn't really make sense to use for a traditional engine. Most authors only use it for NNUE or other neural network evaluations. march=native should therefore be enough machine-specific optimization for Loki.

[tissatussa]

could also be fixed by adding #include <cstdint> to defs.h

NO ! I tried it, but this does not solve the error .. however, adding #include <cstddef> to evaltable.h DOES help ! Thanks and good luck !

[Lecrapouille]

The fix is to add #include <cstddef> solves for me on Linux Debian :)