joergoster [Wed, 6 Aug 2014 10:36:04 +0000 (18:36 +0800)]
Remove insufficient material rule
The eval already returns zero in KK, KBK, KNK (see material.cpp). The difference is:
- we lose the "TB pruning" benefit of the draw rule (ie. search goes on even if eval is zero)
- we gain some speed by removing a useless test from the hot path
STC:
LLR: 0.05 (-2.94,2.94) [-3.00,1.00]
Total: 128000 W: 21357 L: 21560 D: 85083
LTC:
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 33023 W: 4613 L: 4509 D: 23901
bench
7461881
lucasart [Tue, 5 Aug 2014 23:04:27 +0000 (07:04 +0800)]
Remove useless code in Position::pretty()
First, remove some dead code (function never called with a Move argument).
Then, remove printing of legal moves, which does not belong here. Let's keep commands orthogonal and minimal:
- the "d" command should display the board, nothing more, or less.
- "perft 1" will display the list of legal moves.
No functional change.
lucasart [Tue, 5 Aug 2014 03:42:48 +0000 (11:42 +0800)]
Reduce minimum memory requirement by 16MB
Stockfish allocates the default hash (32MB) in main(), before entering UCI::loop(). If there is not enough
memory, the program will crash even before UCI::loop() is entered and the GUI is given a change to specify a
lower Hash value.
This defective design could be resolved by doing a lazy allocation upon "isready" command, as the UCI protocol
guarantees that "isready" will be sent at least once before any search. But it's a bit cumbersome when using
Stockfish "manually" to have to remember to type "isready" everytime.
So leave the current design, but reduce the default hash to 16MB instread of 32MB. In order to perform such
quick searches (depth=13), there is no reason to use so much Hash anyway. Another benefit is to introduce a
bit of hash pressure in bench, which increases chances to detect rare bugs related to TT replacement, for
example.
This is not a functional change, although it obviously changes the bench.
bench
7461879
lucasart [Tue, 5 Aug 2014 03:39:50 +0000 (11:39 +0800)]
Default Hash defined in a single place
Instead of defining it both in ucioption.cpp and benchmark.cpp. Obviously changing the default Hash will
change the bench as a result.
No functional change.
lucasart [Mon, 4 Aug 2014 05:54:09 +0000 (13:54 +0800)]
Retire divide command
The main purpose of perft is to help debugging. But without the breakdown in sum of perft(N-1), it is a
completely useless debugging tool.
So perft now displays the breakdown, and divide is therefore removed.
No functional change.
Marco Costalba [Sat, 2 Aug 2014 09:02:13 +0000 (11:02 +0200)]
Ensure printing UCI info in multi-pv case
After commit
94b1bbb68be6b0bc3aaf1cb804841a022bcc7007, in case available root moves are less than multiPV, we
could never reach condition:
PVIdx + 1 == multiPV
and as a consequence UCI output is not printed.
Fixed suggested by Joerg Oster.
No functional change.
Oskar Werkelin Ahlin [Fri, 1 Aug 2014 11:32:53 +0000 (19:32 +0800)]
Correct bench timing
No functional change.
lucasart [Mon, 28 Jul 2014 23:05:12 +0000 (07:05 +0800)]
Document Threat[] indices
From Marco's repo.
No functional change.
Marco Costalba [Sat, 26 Jul 2014 14:01:09 +0000 (16:01 +0200)]
Fix a warning with MSVC 2010
Warning C4267: 'argument' : conversion from 'size_t' to 'int', possible loss of data
No functional change.
David Zar [Sun, 27 Jul 2014 14:27:53 +0000 (17:27 +0300)]
Small code style reformatting
No functional change.
lucasart [Sun, 27 Jul 2014 01:05:28 +0000 (09:05 +0800)]
Aspiration: widen slower
STC:
LLR: -2.95 (-2.94,2.94) [0.00,4.00]
Total: 182323 W: 30664 L: 30234 D: 121425
LTC:
LLR: 3.51 (-2.94,2.94) [0.00,4.00]
Total: 59841 W: 8345 L: 8006 D: 43490
bench
7962536
David Zar [Tue, 22 Jul 2014 23:05:10 +0000 (07:05 +0800)]
Outpost tuning
double mg bonus and half eg bonus.
STC:
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 18142 W: 3094 L: 2948 D: 12100
LTC:
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 83561 W: 11706 L: 11329 D: 60526
bench
7831429
lucasart [Sun, 20 Jul 2014 00:50:54 +0000 (08:50 +0800)]
Revert Contempt = 20
Despite being neutral at STC, it turned out to be regressive at LTC:
40k games at LTC with Hash=8
ELO: -2.06 +-1.9 (95%) LOS: 1.4%
Total: 39720 W: 5740 L: 5976 D: 28004
40k games at LTC with Hash=128
ELO: -2.69 +-1.9 (95%) LOS: 0.2%
Total: 39149 W: 5702 L: 6005 D: 27442
bench
7477963
Marco Costalba [Mon, 14 Jul 2014 13:15:07 +0000 (15:15 +0200)]
Simplify evaluate_passed_pawns
From a suggestion by David Zar.
No functional change.
lucasart [Mon, 14 Jul 2014 23:03:37 +0000 (07:03 +0800)]
Contempt = 20
Also raise the admissible bounds to (-100,100), as there is no reason to prevent users from using high
values if they want to.
Does not regress in self play:
ELO: 0.10 +-2.0 (95%) LOS: 53.7%
Total: 40000 W: 7084 L: 7073 D: 25843
master vs SF 3
ELO: 182.86 +-2.7 (95%) LOS: 100.0%
Total: 40000 W: 21843 L: 2541 D: 15616
Contempt = 20 vs SF 3
ELO: 189.25 +-2.8 (95%) LOS: 100.0%
Total: 40000 W: 22721 L: 2859 D: 14420
Diff is therefore 6.4 +/- 3.9 elo against a 180-190 elo weaker engine, which is significantly positive,
as expected. This elo difference is likely understated, because of FishTest aggressive draw adjudication
though.
We could push Contempt further, but after 20cp, it would get in the way of FishTest draw adjudication
rule, and is likely to reduce the testing throughput as a result.
bench
8198667
David Zar [Mon, 14 Jul 2014 10:26:25 +0000 (18:26 +0800)]
Bonus passed pawn blocked by our pieces
passed STC:
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 63965 W: 10950 L: 10692 D: 42323
and LTC:
LLR: 2.95 (-2.94,2.94) [0.00,6.00]
Total: 19237 W: 2740 L: 2562 D: 13935
bench:
7477963
Marco Costalba [Thu, 10 Jul 2014 14:40:10 +0000 (16:40 +0200)]
Small reformat to Skill class
Steamline a bit the implementation of
skill levels. As a side effect we can
retire MultiPV global and use a local
variable instead.
No functional change.
lucasart [Thu, 10 Jul 2014 10:25:20 +0000 (18:25 +0800)]
Remove useless condition
Small simplification, suggested by Uri Blass.
passed STC:
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 25839 W: 4464 L: 4351 D: 17024
and LTC:
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 220039 W: 29981 L: 30131 D: 159927
No functional change.
Joona Kiiski [Mon, 7 Jul 2014 19:37:06 +0000 (20:37 +0100)]
Retire FakeSplit
- Currently broken
- Never been really useful
- Does not work well with new splitting model
Verified for no regression at STC with 3 threads:
LLR: 2.96 (-2.94,2.94) [-6.00,0.00]
Total: 81905 W: 12122 L: 12381 D: 57402
No functional change
Marco Costalba [Sat, 5 Jul 2014 06:15:57 +0000 (08:15 +0200)]
Avoid 'double assigments' tricks
Bitboard init code is already noteasy to follow,
so don't make it even harder using 'smart' code.
Also reindent a while loop in standard way.
No functional change.
joergoster [Thu, 3 Jul 2014 11:38:54 +0000 (19:38 +0800)]
Tune trapped rook penalty
Passed STC
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 15687 W: 3352 L: 3199 D: 9136
and LTC (parameter tweaks)
LLR: 2.95 (-2.94,2.94) [0.00,4.00]
Total: 27983 W: 5046 L: 4797 D: 18140
bench:
8330705
lucasart [Thu, 3 Jul 2014 10:23:56 +0000 (18:23 +0800)]
size_t cast in TranspositionTable::first_entry()
32-bit truncation would make this function bogus when clusterCount >= 2^33 (ie. Hash >= 256 GB).
No function change.
lucasart [Thu, 3 Jul 2014 10:22:53 +0000 (18:22 +0800)]
Use compiler intrinsic instead of assembly for popcnt
This time, do not break compatibility with some AMD machines that have SSE3 and popcnt, but do
not have SSE4.2.
No functional change.
Gary Linscott [Tue, 1 Jul 2014 21:01:54 +0000 (17:01 -0400)]
Revert "Use compiler intrinsic instead of assembly for popcnt"
This reverts commit
a69f1d7c2060dc7b320a7b4801105df00b22a96a.
lucasart [Tue, 1 Jul 2014 12:46:16 +0000 (20:46 +0800)]
Use compiler intrinsic instead of assembly for popcnt
No functional change.
lucasart [Tue, 1 Jul 2014 10:13:20 +0000 (18:13 +0800)]
Raise max Hash to 1TB
And use size_t where appropriate, as suggested on FishCooking.
No functional change.
Ajith [Mon, 30 Jun 2014 14:55:10 +0000 (10:55 -0400)]
Add bonuses for Minors attacking enemy pieces(except pawns) even when they are protected by enemy pawns.
Patch passed STC
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 8206 W: 1426 L: 1304 D: 5476
and LTC
LLR: 2.97 (-2.94,2.94) [0.00,6.00]
Total: 19534 W: 2821 L: 2640 D: 14073
Bench:
9942172
joergoster [Sun, 29 Jun 2014 19:17:40 +0000 (20:17 +0100)]
Fix Singular extension condition to handle mate scores
With Eelco's patch "Don't special case for abs(beta) >= VALUE_MATE_IN_MAX_PLY" condition "abs(ttValue) < VALUE_KNOWN_WIN" has been removed from singular extension search, and condition "abs(beta) < VALUE_KNOWN_WIN" was added to the SingularExtensionNode definition.
This might lead to problems, especially in positions, where a mate is due.
For example, this position 5rk1/4K1pp/8/5PPP/8/8/8/1R6 w - - 12 1 triggers an assert.
stockfish: search.cpp:434: Value {anonymous}::search(Position&, Search::Stack*, Value, Value, Depth, bool) [with {anonymous}::NodeType NT = (<unnamed>::NodeType)2u; bool SpNode = false]: Assertion `-VALUE_INFINITE <= alpha && alpha < beta && beta <= VALUE_INFINITE' failed.
So let's re-insert the removed condition.
First spotted by Uri Blass, fix by me.
Bench:
8759675
Joseph R. Prostko [Sun, 29 Jun 2014 14:15:10 +0000 (15:15 +0100)]
Change the install prefix for Haiku
* /boot/common was removed from Haiku
* The equivalent path now that package management has been implemented is /boot/system/non-packaged
No functional change
Bench:
8759681
Ron Britvich [Sat, 28 Jun 2014 18:05:58 +0000 (14:05 -0400)]
Pack 3 TT entries in 32 bytes cluster
Idea from Ron Britvich
Code reworked by Marco Costalba and Joona Kiiski
Bench:
8095369
Resolves #3
Resolves #10
Gary Linscott [Thu, 26 Jun 2014 19:21:06 +0000 (12:21 -0700)]
Merge pull request #9 from glinscott/pawnspan
Scale down endgames with pawns on one or two adjacent files
Passed STC
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 16081 W: 2745 L: 2604 D: 10732
Passed LTC
LLR: 2.95 (-2.94,2.94) [1.00,6.00]
Total: 123832 W: 17292 L: 16584 D: 89956
128k games to measure ELO at 15+0.05:
ELO: 2.07 +-1.1 (95%) LOS: 100.0%
Total: 128000 W: 21632 L: 20869 D: 85499
New bench:
8028792
Gary Linscott [Thu, 26 Jun 2014 19:20:30 +0000 (15:20 -0400)]
A bit more cleanup
Gary Linscott [Thu, 26 Jun 2014 15:20:45 +0000 (08:20 -0700)]
Merge pull request #8 from glinscott/revert_ce1c260
Revert "Check for an available slave early on"
Gary Linscott [Thu, 26 Jun 2014 15:16:36 +0000 (11:16 -0400)]
Revert "Check for an available slave early on"
This reverts commit
ce1c260ea97c14eea996d0a2638f4876ccbc412b.
Gary Linscott [Wed, 25 Jun 2014 20:45:14 +0000 (16:45 -0400)]
Original version of shane's patch
shane31 [Wed, 25 Jun 2014 20:01:00 +0000 (16:01 -0400)]
Scale down endgames with pawns on one or two adjacent files
Gary Linscott [Mon, 23 Jun 2014 16:14:03 +0000 (09:14 -0700)]
Merge pull request #5 from glinscott/authors
Add AUTHORS
Marco Costalba [Sat, 21 Jun 2014 12:32:11 +0000 (14:32 +0200)]
Some reformatting in pawns.cpp
No functional change.
Marco Costalba [Sat, 21 Jun 2014 11:07:29 +0000 (13:07 +0200)]
Ensure ttValue != VALUE_NONE in singular extension search
The assert:
assert(ttValue != VALUE_NONE);
Could fire for multiple reasons (although is very rare),
for instance after an IID we can have ttMove != MOVE_NONE
while ttValue is still set at VALUE_NONE.
But not only this, actually SMP is a source of corrupted
ttValue and anyhow we can detect the condition:
ttMove != MOVE_NONE && ttValue == VALUE_NONE
even north of IID.
Reported by Ronald de Man.
It is so rare that bench didn't change.
bench:
7710548
Gary Linscott [Sat, 21 Jun 2014 04:49:49 +0000 (00:49 -0400)]
Remove some duplicates
Gary Linscott [Sat, 21 Jun 2014 01:35:09 +0000 (21:35 -0400)]
Add AUTHORS
Marco Costalba [Fri, 20 Jun 2014 22:02:43 +0000 (00:02 +0200)]
Better value clipping in game_phase()
No functional change.
Marco Costalba [Fri, 20 Jun 2014 21:40:36 +0000 (23:40 +0200)]
Move game_phase() to Position
It seems a more natural to place this
function there.
No functional change.
Marco Costalba [Fri, 20 Jun 2014 21:15:31 +0000 (23:15 +0200)]
Small renaming in material weights
Also dropped some temporary variable: compiler
is more than able to push on stack temp values
by itself (verified).
No functional change.
Marco Costalba [Thu, 19 Jun 2014 13:44:42 +0000 (15:44 +0200)]
Simplify a condition in is_KXK()
No functional change.
Eelco de Groot [Mon, 16 Jun 2014 19:46:42 +0000 (21:46 +0200)]
Don't special case for abs(beta) >= VALUE_MATE_IN_MAX_PLY
Remove from the search this special case and apply
null search and razoring also in mate positions.
Tested in no-regression mode and passed both
STC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 65431 W: 10860 L: 10810 D: 43761
and LTC
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 34928 W: 4814 L: 4713 D: 25401
This patch kicks in only in mate positions and in
these cases it seems beneficial in finding mates
faster as Yery Spark measured on the Chest mate suite:
Total number of positions 6425
Fixed nodes 200K per position
master: 1049
new: 1154
And also the 5446 'hard' positions again with 2000K nodes
(those not found by both engines in 200K nodes):
master: 1069
new: 1395
bench:
7710548
Leonid Pechenik [Mon, 16 Jun 2014 10:32:11 +0000 (12:32 +0200)]
Simplify unstoppable pawns
Tested in no-regression mode and passed both
STC
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 17919 W: 3103 L: 2978 D: 11838
and LTC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 31460 W: 4414 L: 4308 D: 22738
bench:
7709279
Marco Costalba [Sun, 15 Jun 2014 08:29:53 +0000 (10:29 +0200)]
Remove -ansi flag for Clang
It seems this flag is only for gcc and
yields a warning under OSX Mavericks:
clang: warning: argument unused during compilation: '-ansi'
No functional change.
Marco Costalba [Sat, 14 Jun 2014 10:46:58 +0000 (12:46 +0200)]
Fix a warning with MSVC in 'analyze' mode
Here MSVC is worried that
StepAttacksBB[PAWN][psq]
could overflow, so change psq initialization
to clarify psq is always less than 64.
No functional change.
Marco Costalba [Sat, 14 Jun 2014 10:26:08 +0000 (12:26 +0200)]
Triviality in UCI::loop
Code style paranoid in action here :-)
No functional change.
Marco Costalba [Sat, 14 Jun 2014 10:16:22 +0000 (12:16 +0200)]
Factor out pawn attacks in 'lever'
Improves readibility and possibly speed.
No functional change.
Ajith [Wed, 11 Jun 2014 23:31:33 +0000 (08:31 +0900)]
Simplify pawn threats and merge into ThreatenedByPawn[]
Tested in no-regression mode,
passed STC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 14477 W: 2493 L: 2362 D: 9622
and LTC
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 51964 W: 7091 L: 7013 D: 37860
bench:
7875814
Reuven Peleg [Mon, 9 Jun 2014 10:45:56 +0000 (13:45 +0300)]
Tweak outpost name
This name is more accurate, since that function evaluates only one
outpost in every call.
No functional change.
Reuven Peleg [Wed, 11 Jun 2014 14:36:46 +0000 (23:36 +0900)]
Add bonus for advanced lever
Passed both STC
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 15413 W: 2670 L: 2530 D: 10213
And LTC:
LLR: 2.95 (-2.94,2.94) [1.00,6.00]
Total: 66908 W: 9398 L: 8960 D: 48550
Bench :
7859385
Marco Costalba [Sun, 8 Jun 2014 20:30:18 +0000 (05:30 +0900)]
Small renaming in Tracing
No functional change.
Marco Costalba [Sun, 8 Jun 2014 19:56:31 +0000 (04:56 +0900)]
Check for an available slave early on
Don't take the split lock if we don't have
available slaves (about 30-40% of times).
This new condition allows to retire the now
redundant one on number of threads.
No functional change.
Marco Costalba [Fri, 6 Jun 2014 23:16:31 +0000 (01:16 +0200)]
Use unsigned char as argument of std::isspace
Although signature allows an int:
int isspace( int ch );
The behavior is undefined if the value of ch is
not representable as unsigned char and is not
equal to EOF.
See
http://en.cppreference.com/w/cpp/string/byte/isspace
http://www.greenend.org.uk/rjk/tech/cfu.html
This is really a tricky corner case of C standard!
Spotted and reported by Ron Britvich.
No functional change.
Marco Costalba [Fri, 6 Jun 2014 09:12:05 +0000 (11:12 +0200)]
Avoid to use nullChild
Use instead:
(ss-1)->currentMove == MOVE_NULL
No functional change.
Marco Costalba [Fri, 6 Jun 2014 09:10:40 +0000 (11:10 +0200)]
Remove 'update gains' hack
Use (move != MOVE_NONE) condition to
filtering out updating gains at root.
bench:
8454456
Marco Costalba [Fri, 6 Jun 2014 09:07:12 +0000 (11:07 +0200)]
Temporary revert previous patch
Split previous patch in 2 steps: first remove
the MOVE_NULL hack, then retire nullChild.
The first step is a prerequisite
for second one and affects bench.
The second step (next patch) just removes nullChild
without affecting bench.
bench:
8205159
Marco Costalba [Fri, 6 Jun 2014 07:58:00 +0000 (09:58 +0200)]
Avoid to use nullChild
Should be a non functional change, but
for some reason bench is changed.
bench:
8454456
Marco Costalba [Fri, 6 Jun 2014 07:35:34 +0000 (09:35 +0200)]
Move Tempo to evaluation
No functional change.
Marco Costalba [Thu, 5 Jun 2014 21:55:18 +0000 (23:55 +0200)]
Revert "Score extractors"
Are broken for big-endian case and
I have verified with MSVC 2013 Premium
bench is correct and there is no
miscompilation, so the main reason
to change the original code drops.
No functional change.
Marco Costalba [Thu, 5 Jun 2014 21:46:58 +0000 (23:46 +0200)]
Fix a warning with MSVC Premium 2013
Reported by Ron Britvich.
No functional change.
pellanda [Wed, 4 Jun 2014 13:56:34 +0000 (10:56 -0300)]
Update Readme.md
Retire line about polyglot.ini and
book as they are not used anymore.
No functional change.
Lucas Braesch [Tue, 3 Jun 2014 12:48:43 +0000 (20:48 +0800)]
Symmetric King Safety: take 2
Another attempt at retiring current asymmetric
king evaluation and use a much simpler symmetric
one. As a good side effect we can avoid recalculating
eval after a null move.
Tested in no-regression mode and passed
STC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 21580 W: 3752 L: 3632 D: 14196
LTC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 18253 W: 2593 L: 2469 D: 13191
And a LTC regression test against SF DD to
verify we don't have regression against
weaker engines due to some kind of 'contempt'
effect:
ELO: 54.69 +-2.1 (95%) LOS: 100.0%
Total: 40000 W: 11072 L: 4827 D: 24101
bench:
8205159
Marco Costalba [Tue, 3 Jun 2014 10:38:58 +0000 (12:38 +0200)]
Use see() instead of see_sign() in previous patch
Before it was working by accident in case of
see_sign() and failing with see() due to how
castle moves are coded (king captures the rook).
Better to explicitly filter out castling moves
and use see() without any surprise/trick.
No functional case.
kinderchocolate [Tue, 3 Jun 2014 05:45:10 +0000 (15:45 +1000)]
Add some const qualifier
No functional change.
Reuven Peleg [Tue, 3 Jun 2014 09:32:45 +0000 (11:32 +0200)]
Decrease reduction for moves that escape a capture
Passed both STC
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 10094 W: 1833 L: 1704 D: 6557
and LTC
LLR: 2.95 (-2.94,2.94) [0.00,6.00]
Total: 27738 W: 4147 L: 3928 D: 19663
bench :
8599236
Marco Costalba [Sun, 1 Jun 2014 13:29:34 +0000 (15:29 +0200)]
Retire Polyglot Book management
Book handling belongs to GUI, we kept this code
for historical reasons, but nowdays there is
really no need of this old, (mostly) unused
and especially incorrect designed functionality.
It is up to the GUI to choose the book (far easier for
the user) and to select the book parameters. In no
place, including fishtest, TCEC, rating lists, etc.
the "own book" is used, moreover currently SF is
released without any book and even if in the future we
bundle a book in the release package, it will be the GUI
that will take care of it.
This corrects a wrong design decision that Galurung
and later Stockfish inherited from what was common
practice many yeas ago.
No functional change.
Marco Costalba [Sun, 1 Jun 2014 11:37:21 +0000 (13:37 +0200)]
Retire eval weights UCI options
There is really little that user can achieve (apart
from a weakened engine) tweaking these parameters
that are already tuned and have no immediate or visible
effect.
So better do not expose them to the user and avoid the
typical "What is the best setup for my machine?" kind of
question (by far the most common, by far the most useless).
No functional change.
Marco Costalba [Sat, 31 May 2014 21:37:20 +0000 (23:37 +0200)]
Retire polyglot.ini
All actively maintained GUI are able
to understand UCI protocol, so there
is no more need for this obsolete file.
No functional change.
Marco Costalba [Sat, 31 May 2014 21:34:36 +0000 (23:34 +0200)]
Restore development version
bench:
8732553
Marco Costalba [Sat, 31 May 2014 07:16:54 +0000 (09:16 +0200)]
Stockfish 5
Stockfish bench signature is:
8732553
Marco Costalba [Mon, 26 May 2014 19:39:48 +0000 (21:39 +0200)]
Revert "Symmetric king safety"
Regression test of 40K games at 60 secs shows
this commit to be a 2-3 ELO regression.
So revert to original king safety.
bench:
8732553
Marco Costalba [Sat, 24 May 2014 22:21:46 +0000 (00:21 +0200)]
Fix a warning with Intel compiler
warning #2259: non-pointer conversion from
"int" to "int16_t={short}" may lose significant
bits.
No functional change.
Marco Costalba [Sat, 24 May 2014 21:53:01 +0000 (23:53 +0200)]
Tidy up tt.h
Backport some non-functional changes
found working on 'dense TT' patch.
No functional change.
Marco Costalba [Sat, 24 May 2014 02:31:36 +0000 (04:31 +0200)]
Add perft 'divide' command
To show perft numbers for each move. Just
use 'divide' instead of 'perft', for instance:
position startpos moves e2e4 e7e5
divide 4
Inspired by Ronald de Man.
No functional change.
Marco Costalba [Mon, 19 May 2014 19:19:47 +0000 (21:19 +0200)]
Update polyglot.ini after last patch
No functional change.
Lucas Braesch [Mon, 19 May 2014 12:22:15 +0000 (14:22 +0200)]
Symmetric king safety
Retire current asymmetric king evaluation
and use a much simpler symmetric one.
As a side effect retire the infamous
'Aggressiveness' and 'Cowardice' UCI
options.
Tested in no-regression mode,
Passed both STC
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 33855 W: 5863 L: 5764 D: 22228
And LTC
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 40571 W: 5852 L: 5760 D: 28959
bench:
8321835
Marco Costalba [Sat, 17 May 2014 20:56:35 +0000 (22:56 +0200)]
Fix an off-by-one bug in extract_pv_from_tt
At root we start counting plies from 1,
instead pv[] array starts from 0. So
the variable 'ply' we use in extract_pv_from_tt
to index pv[] is misnamed, indeed it is
not the real ply, but ply-1.
The fix is to leave ply name in extract_pv_from_tt
but assign it the correct start value and
consequentely change all the references to pv[].
Instead in insert_pv_in_tt it's simpler to rename
the misnamed 'ply' in 'idx'.
The off-by-one bug was unhidden when trying to use
'ply' for what it should have been, for instance in
this position:
position fen 8/6R1/8/3k4/8/8/8/2K5 w - - 0 1
at depth 24 mate line is erroneusly truncated due
to value_from_tt() using the wrong ply.
Spotted by Ronald de Man.
bench:
8732553
Marco Costalba [Sat, 17 May 2014 10:49:52 +0000 (12:49 +0200)]
Extract a reliable PV line
Truncate the extracted PV from the point where
the score stored in hash starts to deviate from
the root score.
Idea from Ronald de Man.
bench:
8732553
Michel Van den Bergh [Tue, 13 May 2014 20:33:41 +0000 (22:33 +0200)]
Drop to qsearch at low depth in razoring
If razoring conditions are satisfied and
depth is low, then directly drop in qsearch.
Passed both STC
LLR: 2.98 (-2.94,2.94) [-1.50,4.50]
Total: 12914 W: 2345 L: 2208 D: 8361
And LTC
LLR: 2.95 (-2.94,2.94) [0.00,6.00]
Total: 50600 W: 7548 L: 7230 D: 35822
bench:
8739659
Arjun Temurnikar [Tue, 13 May 2014 19:35:10 +0000 (21:35 +0200)]
Remove undefended minors
Tested in "no regression" mode.
Passed both STC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 68026 W: 12277 L: 12236 D: 43513
And LTC
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 85682 W: 12861 L: 12836 D: 59985
bench:
7311935
Marco Costalba [Sun, 11 May 2014 08:56:25 +0000 (10:56 +0200)]
Save stalemates in TT
When there aren't legal moves after
a search, instead of returning imediately,
save bestValue in TT as in the usual case.
There is really no reason to special case
this one.
With this patch is fully fixed (again) follwing
position:
7k/6p1/6B1/5K1P/8/8/8/8 w - - 0 1
Also in SMP case.
bench:
8802105
Marco Costalba [Sun, 11 May 2014 08:13:56 +0000 (10:13 +0200)]
Retire "Idle Threads Sleep" UCI option
After last Joona's patch there is no measurable
difference between the option set or unset.
Tested by Andreas Strangmüller with 16 threads
on his Dual Opteron 6376.
After 5000 games at 15+0.05 the result is:
1 Stockfish_14050822_T16_on : 3003 5000 (+849,=3396,-755), 50.9 %
2 Stockfish_14050822_T16_off : 2997 5000 (+755,=3396,-849), 49.1 %
bench: 880215
Arjun Temurnikar [Sat, 10 May 2014 06:18:09 +0000 (23:18 -0700)]
Make imbalance table more clear
No functional change.
Reuven Peleg [Thu, 8 May 2014 20:18:19 +0000 (23:18 +0300)]
Pass Position as const ref in update_stats()
No functional change.
Marco Costalba [Thu, 8 May 2014 07:09:35 +0000 (09:09 +0200)]
Clarify some comments in SMP code
Spotted by Joona.
No functional change.
Marco Costalba [Wed, 7 May 2014 06:47:18 +0000 (08:47 +0200)]
Reformat and simplify previous patch
No functional change.
Joona Kiiski [Sat, 3 May 2014 17:29:31 +0000 (18:29 +0100)]
Allow a slave to 'late join' another splitpoint
Instead of waiting to be allocated, actively search
for another split point to join when finishes its
search. Also modify split conditions.
This patch has been tested with 7 threads SMP and
passed both STC:
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 2885 W: 519 L: 410 D: 1956
And a reduced-LTC at 25+0.05
LLR: 2.95 (-2.94,2.94) [0.00,6.00]
Total: 4401 W: 684 L: 566 D: 3151
Was then retested against regression in 3 thread case
at standard LTC of 60+0.05:
LLR: 2.96 (-2.94,2.94) [-4.00,0.00]
Total: 40809 W: 5446 L: 5406 D: 29957
bench:
8802105
Ron Britvich [Mon, 5 May 2014 06:52:14 +0000 (08:52 +0200)]
Rewrite Score extractors
Less tricky and even a bit faster. With this
version Visual Studio Ultimate 2013 Update 2 RC
runs fine even in O2 optimization.
No functional change.
Marco Costalba [Sun, 4 May 2014 11:11:32 +0000 (13:11 +0200)]
Change search() signature
Pass SpNode as template parameter.
No functional change.
Marco Costalba [Sun, 4 May 2014 07:58:49 +0000 (09:58 +0200)]
Revert dynamic contempt
On a final fixed game number test it failed
to prove better than standard version.
STC 15+0.05
ELO: -0.86 +-1.7 (95%) LOS: 15.8%
Total: 57578 W: 10070 L: 10213 D: 37295
bench:
8802105
Marco Costalba [Sun, 4 May 2014 07:42:05 +0000 (09:42 +0200)]
Revert stalemate detection in evaluation
Unfortunatly we have a slow down that causes
a regression in STC with no-regression mode:
LLR: -2.96 (-2.94,2.94) [-3.00,1.00]
Total: 22454 W: 3836 L: 4029 D: 14589
bench:
8678654
Marco Costalba [Sun, 4 May 2014 07:31:25 +0000 (09:31 +0200)]
Revert back KBPsK to latest Gary's version
The bug was found to be elsewhere. This version
is correct and also is able to detect as draw
positions like:
8/8/5b2/8/8/4k1p1/6P1/5K2 b - - 6 133
bench:
8678654
Ronald de Man [Sat, 3 May 2014 19:37:24 +0000 (21:37 +0200)]
Fix KXK endgame
Position is win also if strong side has a bishop
and a knight (plus other material, otherwise
KBNK would be triggered instead of KXK).
This fixes a subtle bug where a search on position
k7/8/8/8/8/P7/PB6/K7 b - - 6 1
Instead of returning a draw score, suddendly returns
a big score. This happens because at one point in
search we reach this position:
8/Pk6/8/8/8/4B3/P7/K7 w - - 3 8
Where white can promote. In case of rook promotion (and also in case of
queen promotion) the resutling position gets a huge static eval that is
above VALUE_KNOWN_WIN (from the point of view of white). So for rook
promotion it is
&& futilityBase > -VALUE_KNOWN_WIN
that prevents futility pruning in qsearch. (Removing this condition indeed
lets the problem occur). Raising the static eval for K+B+N+X v K to a value
higher than VALUE_KNOWN_WIN fixes this particular problem without having to
introduce an extra futility pruning condition in qsearch.
I just checked and it seems K+R v K, K+2B v K and even K+B+N v K already get
a huge static eval. Why not K+B+N+P v K?
I think this fix corrects an oversight. There is special code for KBNK, but
KBNXK is handled by KXK, so the test for sufficient material should also test
for B+N.
bench:
8678654
Marco Costalba [Sat, 3 May 2014 10:09:56 +0000 (12:09 +0200)]
Fully correct stealmate detection
In the (rare) cases when the two conditions
are true, then fully check again with a slow
but correct MoveList<LEGAL>(pos).size().
This is able to detect false positives like
this one:
8/8/8/Q7/5k1p/5P2/4KP2/8 b - - 0 17
When we have a possible simple pawn push that
is not stored in attacks[] array. Because the
third condition triggers very rarely, even if
it is slow, it does not alters in a measurable
way the average speed of the engine.
bench:
8678654
Marco Costalba [Sat, 3 May 2014 09:45:58 +0000 (11:45 +0200)]
Add stealmate detection to evaluation
Currently a stealmate position is misevaluated
in a negative/positive score, this leads qsearch(),
that does not detects stealmates too, to return the
wrong score and this yields to some kind of endgames
to be completely misevaluated.
With this patch is fully fixed follwing position
7k/6p1/6B1/5K1P/8/8/8/8 w - - 0 1
Also in SMP case.
Correct root cause analysys by Ronald de Man.
bench:
8678654
Marco Costalba [Fri, 2 May 2014 08:27:13 +0000 (10:27 +0200)]
Revert to Galurung's KBPsK endgame
After reverting to the original Tord's
endgame, a search on position
7k/6p1/6B1/5K1P/8/8/8/8 w - - 0 1
Reports, correctly, a draw score instead of
an advantage for white.
Issue reported by Uri Blass.
bench:
8678654