Fairy-Max 4.8_O released

[hgm]

I released a new version of Fairy-Max, which keeps track of the PV, and also supports multi-PV. (Perhaps not very interesting for Chess, as Fairy-Max is not a very strong engine, but there are not many engines that support multi-PV for Thai Chess, Shatranj or Gothic Chess!)

For the rest it has some minor tweaks and bugfixes: more general castling code, and support for promotion n the 6th rank.

http://home.hccnet.nl/h.g.muller/fairy.zip

[beachknight]

Thanks for the new release.

4.8O or 4.8.0?

Best,

[hgm]

Oh. The previous official release was 4.8 M, and fter Christmass I released an alpha version with multi-PV support as 4.8 N.

[beachknight]

Hmmm. I will use 4.8.0 then.

Best,

PS: Last version I have tested was 4.8, yes only 4.8

[hgm]

There should not really be any difference in playing strength between all the 4.8 versions so far. The bug fixes and enhancements mainly concern the interface, and often did not even touch the search and evaluation. E.g. in this version I changed the routine for reading the description of the variant from the ini file to be able to handle the case where the white and black Bishops (say) move differently, but you still want to indicate them both with the letter B (when you tell the engine where the Bishops are located when setting up a position). That is not something that would make it play any different in normal Chess.

In principle testing if 6th-rank promotion should be applied, on any Pawn move of the search, would slow it down a little, but this will not be more than a percent or so. That would cost about 1 Elo, but in fact different compiles that do exactly the same (e.g. where I deleted some code that was never executed anyway) might differ in speed by much more than 1%, so it might just as easily be faster because of it, and gain 1 or 2 Elo.

As for compiling, the binary at the downoad is a plain gcc -O2 compile. People that know how to do PGO compiles can make much faster versions. To the best of my knowledge the only version of Fairy-Max for which a PGO compile exist was 4.8 v (after 4.8 z I restarted with 4.8 A), made for me by Denis Mendoza. It was called fmax4_8w_v32, and is about 1.75 times as fast as my own compiles (but has a lot bigger .exe). For normal and Gothic / Capablanca Chess that is still the preferred version, playing-strength-wise, when it does not bother you that it plays an illegal move (a1a1) when it is stalemated.

From the changelog in the source tarball:

Code: Select all

/* w) piece indicators programable, multi-path support                     */
/* x) e.p. changed to support Berolina Pawns                               */
/* y) capture value of 6-7th-rank Pawn reduced in Shatranj                 */
/* z) bug in promotion input corrected                                     */
/* A) stalemate-detection bug in printResult fixed                         */
/* B) Invalidate hash on game-level promotion (might be under-promotion!)  */
/* C) King move evaluation based on negative piece value in stead of nr    */
/* D) WB memory command added, undo fixed                                  */
/* E) 15th piece read in                                                   */
/* F) accepts ini fileargument                                             */
/* G) bug in calculation ASCII promotion character fixed                   */
/* H) unified normal and shatranj source                                   */
/* J) rewrite under-promotion code, fixes persistent bug there             */
/* K) o[] and oo[] made int to make fairymax work on big-endian machines   */
/* L) added Resign option feature (using new WB protocol)                  */
/* M) char -> signed char for better portability                           */
/* N) add PV printing and multi-PV support                                 */
/* O) non-Rook castling, 6th-rank promotion, independent B/W piece naming  */
/***************************************************************************/

[beachknight]

Oooouch. An exposee is not that what I hoped

Just checked my RL. No previous tests of FairyMax seem included so far.
In few days FairyMax 4.8O will have a start rating in my RLs.

Best,

PS: I see no new M-max or joker compiles on the downloads section.

[hgm]

Indeed, I have not worked on micro-Max or Joker for about 2 years now. I did make a new engine last year (HaQiKi D), and the idea was that ultimately that should be able to play Chess. But so far it can only play Xiangqi...

[Jim Ablett]

I compiled some Win32 Intel compiler pgo executables for latest versions.

Download:
http://www.mediafire.com/?m3tygjmnjzk

Jim.

[beachknight]

Thank you, Jim.

I'll include your compiles in my tournaments cycle.

Best,

[MikeB]

I released a new version of Fairy-Max, which keeps track of the PV, and also supports multi-PV. (Perhaps not very interesting for Chess, as Fairy-Max is not a very strong engine, but there are not many engines that support multi-PV for Thai Chess, Shatranj or Gothic Chess!)

For the rest it has some minor tweaks and bugfixes: more general castling code, and support for promotion n the 6th rank.

http://home.hccnet.nl/h.g.muller/fairy.zip

Just got around to compiling FairyMax for the Mac. It does not seem to accept any castling moves - any ideas?