Firenzina 2.2.2 xTreme, a clone of Fire 2.2 xTreme

[Gusev]

Matthew,
Given that I had set the default parameters in master to the so-called "Aggressive" option, see if you can
(a) keep a copy of your "Aggressive" compile,
(b) in fire.h, reset the default parameter values back to "Original" ones from my comments (keep the "Aggressive" values in comments),
(c) re-compile under the same compiler settings, and
(d) test to see which compile performs better against Stockfish 2.3.1 for Linux.
The test that I ran today ostensibly predicts that the original settings will fare better (that's why I am suggesting that you reset them in the code), but this is not certain.
Dmitri

[Gusev]

Master compiles fine, once LinuxLargePages is commented out of fire.h.

Matthew:out

I have updated master. I commented out LinuxLargePages and restored most of the default parameters back to the original values. I kept the new chess piece weights for the moment, because those weights are also different in the 2.2b GH compile from what they are in the 2.2 code. My weight set is based upon a simple additional statistical calculation that involved, among other contributions, weights from Kaissa 1.00 and Stockfish 2.2.2. I found that Stockfish is rather unusual here.

[ZirconiumX]

Matthew,
Given that I had set the default parameters in master to the so-called "Aggressive" option, see if you can
(a) keep a copy of your "Aggressive" compile,
(b) in fire.h, reset the default parameter values back to "Original" ones from my comments (keep the "Aggressive" values in comments),
(c) re-compile under the same compiler settings, and
(d) test to see which compile performs better against Stockfish 2.3.1 for Linux.
The test that I ran today ostensibly predicts that the original settings will fare better (that's why I am suggesting that you reset them in the code), but this is not certain.
Dmitri

I was thinking of trying "Orthogonal multi-testing" to increase efficiency of testing. It looks statistically sound, so it'd be interesting to try out.

http://talkchess.com/forum/viewtopic.php?t=27401

We'll also need to benchmark our machines, and try to keep testing conditions similar.

Matthew:out

[Marek Soszynski]

Master compiles fine, once LinuxLargePages is commented out of fire.h.

Matthew:out

I have updated master. I commented out LinuxLargePages...

Although Firenzena compiles okay on my 64-bit Linux system, in infinite analysis mode something goes wrong. In Scid it displays the move, but when that move is played, it no longer displays any analysis unless the entire system is rebooted.

There were no such problems with Jim Ablett's Fire code.

Incidentally, I don't see why the compile should be fire rather than Firenzina or likewise why the generated default configuration file is fire.cfg

[ZirconiumX]

Master compiles fine, once LinuxLargePages is commented out of fire.h.

Matthew:out

I have updated master. I commented out LinuxLargePages...

Although Firenzena compiles okay on my 64-bit Linux system, in infinite analysis mode something goes wrong. In Scid it displays the move, but when that move is played, it no longer displays any analysis unless the entire system is rebooted.

There were no such problems with Jim Ablett's Fire code.

Incidentally, I don't see why the compile should be fire rather than Firenzina or likewise why the generated default configuration file is fire.cfg

Confirmed. The starting position seems to be set at something other than the chess starting position, with black to move, so Scid goes WHAT?! and doesn't display any analysis.

EDIT: Even when fed the starting position by hand, it still thinks Black is to move.

Matthew:out

[ZirconiumX]

Master compiles fine, once LinuxLargePages is commented out of fire.h.

Matthew:out

Scratch that. Firenzina does NOT compile on my system, anymore.

Note to self: Make sure you hit Rebuild rather than Build.

Matthew:out

[ZirconiumX]

I have found the troublesome commit: 8a4bcae8d89f2e7174a9dc0ed6ebce291289febd

Or, if you can't read SHA sums, "Linux/Housekeeping/Bug fixes/Extend xTreme/Defs".

Now to find which particular patch caused this.

EDIT: hash.c is the culprit. It isn't needed anyway, a newer version of it is included in the source code, and overridden in hash.h.

Matthew:out

[velmarin]

I've compiled on Windows but not working.
No matter, it's what I have.

I repeat my advice that separate branches.

A couple of things:
how almost all Ippolit and Fire, Firenzina have this error:
in material_value.c

in line 145:
Flags |= BlackMinorOnly << 2;
if (!wN && !wB && !wR && !wQ && !bN && !bB && !bQ && !bQ&& wP + bP == 1)

Corret:
!bQ && !bR

In the same file can be added as IvanHoe 46 in White and BlackWeight

if (wQ == 1 && wPHASE == 4 && bPHASE >= 2 && bP >= 1)
wWEIGHT = 5;
}

In file Endgame can add cases enlargement by IvanHoe46,
Even remove kp_files (Jukebox Bouquet have eliminated).

Of course I would withdraw thousand configurable options, too many for a normal user, even for someone too advanced.
But it is a matter of personal taste.

I hope this little they will serve.

[ZirconiumX]

I've compiled on Windows but not working.
No matter, it's what I have.

I repeat my advice that separate branches.

A couple of things:
how almost all Ippolit and Fire, Firenzina have this error:
in material_value.c

in line 145:
Flags |= BlackMinorOnly << 2;
if (!wN && !wB && !wR && !wQ && !bN && !bB && !bQ && !bQ&& wP + bP == 1)

Corret:
!bQ && !bR

In the same file can be added as IvanHoe 46 in White and BlackWeight

if (wQ == 1 && wPHASE == 4 && bPHASE >= 2 && bP >= 1)
wWEIGHT = 5;
}

In file Endgame can add cases enlargement by IvanHoe46,
Even remove kp_files (Jukebox Bouquet have eliminated).

Of course I would withdraw thousand configurable options, too many for a normal user, even for someone too advanced.
But it is a matter of personal taste.

I hope this little they will serve.

Thank you for the bug report, José.

I was actually thinking of deprecating most of the UCI options, with a focus on fire.cfg.

Matthew:out

[Gusev]

I have found the troublesome commit: 8a4bcae8d89f2e7174a9dc0ed6ebce291289febd

Or, if you can't read SHA sums, "Linux/Housekeeping/Bug fixes/Extend xTreme/Defs".

Now to find which particular patch caused this.

EDIT: hash.c is the culprit. It isn't needed anyway, a newer version of it is included in the source code, and overridden in hash.h.

Matthew:out

After hash.c is excluded, does everything compile on your side? Also, is main.c needed? I believe that both came in as parts of the code from JA.