Page 3 of 25

Re: A complete 2000 lines of code engine

Posted: Fri Oct 26, 2018 3:16 pm
by xr_a_y
I will upload binaries soon, last week it has evolved quite often so I was too lazy to make "real" releases ... just changed the version number.

Re: A complete 2000 lines of code engine

Posted: Fri Oct 26, 2018 7:00 pm
by tpoppins
Here are the bins for v0.6:

minic-06-tp.zip (Mediafire)
minic-06-tp.zip (Filedropper)

Static MinGW builds, 32- and 64-bit. I don't have a 32-bit OS to test on at the moment but the "32-old" build one should run on anything as old as a Pentium III.

I just realized that I'd forgotten to strip symbols on my previous releases, which resulted in the files being unnecessarily large. This has been been corrected for this release.

Re: A complete 2000 lines of code engine

Posted: Mon Oct 29, 2018 12:25 am
by Ronald
I saw you added some code to try the hash move first, some remarks to use this part probably even more:
- The hash entry is read with a depth, if the found entry has a lower depth then the current depth no entry is returned. You will get much more hash moves if you always return the hash entry, no matter the depth. You can not use the entry to return the hash score (so you have to check the depth for that) but you still can use the stored hash move and try that first.
- It will also help with IID 2 times, because first of all at this moment you will never get a hash entry / hash move after the IID search. IID is searched with depth/2 after which the result is read from hash with the check for "depth" :D And by always returning the entry you will also execute much less IID searches.
- I would try to integrate the hash move in the regular move loop(basic staged loop)
- Better not use a killer slot for the hash, the hash move already has it's own slot, the "hash move" slot and you might push out a valuable killer move with it

Re: A complete 2000 lines of code engine

Posted: Mon Oct 29, 2018 9:46 am
by xr_a_y
Thanks ! I'll try that. I also added SEE and LMP last night with a great elo boost.

Best regards

Re: A complete 2000 lines of code engine

Posted: Mon Oct 29, 2018 12:00 pm
by Guenther
tpoppins wrote: Fri Oct 26, 2018 7:00 pm Here are the bins for v0.6:

minic-06-tp.zip (Mediafire)
minic-06-tp.zip (Filedropper)

Static MinGW builds, 32- and 64-bit. I don't have a 32-bit OS to test on at the moment but the "32-old" build one should run on anything as old as a Pentium III.

I just realized that I'd forgotten to strip symbols on my previous releases, which resulted in the files being unnecessarily large. This has been been corrected for this release.
I miss a 64-old :)

Re: A complete 2000 lines of code engine

Posted: Mon Oct 29, 2018 4:14 pm
by tpoppins
I should have remembered. Graham also has a non-POPCNT box, although he's unlikely to run anything under 2000 Elo, which has lately been my domain.

minic-06-64-old-tp.zip

I expect that your h/w supports SSE2 at least, let me know if this won't run.

Re: A complete 2000 lines of code engine

Posted: Mon Oct 29, 2018 4:39 pm
by xr_a_y
Minic 0.8 is available with binaries for linux (64bits, sse4.2, stripped) and windows (x86_64-w64-mingw32-g++ and strip)
You'll find both inside "Dist" directory on the github repo.

Last additions : SEE, LMP, bug fixes thanks to Ronald

Now probably >2000elo

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   1 minic                         148      48     129   70.2%   39.5%
   2 fairymax                       41      51     128   55.9%   28.9%
   3 tscp                         -200      61     129   24.0%   20.2%
and still under 2000sloc :lol: .

Re: A complete 2000 lines of code engine

Posted: Mon Oct 29, 2018 4:50 pm
by CMCanavessi
Nice progress!

Re: A complete 2000 lines of code engine

Posted: Mon Oct 29, 2018 4:59 pm
by xr_a_y
Thanks ! great boost from SEE and using PST in move sorting.

Re: A complete 2000 lines of code engine

Posted: Mon Oct 29, 2018 6:27 pm
by tpoppins
Two 32-bit builds for v0.8:

minic-08-32-tp.zip

edit: removed broken 64-old build from archive.