Simplify time management code by removing hard stops for unchanging first root moves.
Search is now stopped earlier at the end iteration if it did not have fail-lows at root.
This simplification also fixes pondering bug. Ponder flag was true by default
and cutechess-cli doesn't change it to false even though no pondering is possible.
Fix the issue by setting the default value of 'Ponder' flag to false.
10+0.1:
ELO: 3.51 +-3.0 (95%) LOS: 99.0%
Total: 20000 W: 3898 L: 3696 D: 12406
40+0.4:
ELO: 1.39 +-2.7 (95%) LOS: 84.7%
Total: 20000 W: 3104 L: 3024 D: 13872
60+0.06:
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 37231 W: 5333 L: 5236 D: 26662
Stopped run at 100+1:
LLR: 1.09 (-2.94,2.94) [-3.00,1.00]
Total: 37253 W: 4862 L: 4856 D: 27535
Resolves#523Fixes#510
This is used by std::stable_sort() to sort moves from highest score to lowest score.
1) The comment is incorrect since highest to lowest means descending.
2) It's more natural to implement a less operator using another less operator rather than a greater operator.
No functional change.
Resolves#504
Unfortunately std::condition_variable::wait_for()
is not accurate in general case and the timer thread
can wake up also after tens or even hundreds of
millisecs after time has elapsded. CPU load, process
priorities, number of concurrent threads, even from
other processes, will have effect upon it.
Even official documentation says: "This function may
block for longer than timeout_duration due to scheduling
or resource contention delays."
So retire timer and use a polling scheme based on a
local thread counter that counts search() calls and
a small trick to keep polling frequency constant,
independently from the number of threads.
Tested for no regression at very fast TC 2+0.05 th 7:
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 32969 W: 6720 L: 6620 D: 19629
TC 2+0.05 th 1:
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 7765 W: 1917 L: 1765 D: 4083
And at STC TC, both single thread
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 15587 W: 3036 L: 2905 D: 9646
And with 7 threads
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 8149 W: 1367 L: 1227 D: 5555
bench: 8639247
The only interesting change is the moving of
stack[MAX_PLY+4] back to its original position
in id_loop (now renamed Thread::search).
No functional change.
Rely on well defined behaviour for message passing, instead of volatile. Three
versions have been tested, to make sure this wouldn't cause a slowdown on any
platform.
v1: Sequentially consistent atomics
No mesurable regression, despite the extra memory barriers on x86. Even with 15
threads and extreme time pressure, both acting as a magnifying glass:
threads=15, tc=2+0.02
ELO: 2.59 +-3.4 (95%) LOS: 93.3%
Total: 18132 W: 4113 L: 3978 D: 10041
threads=7, tc=2+0.02
ELO: -1.64 +-3.6 (95%) LOS: 18.8%
Total: 16914 W: 4053 L: 4133 D: 8728
v2: Acquire/Release semantics
This version generates no extra barriers for x86 (on the hot path). As expected,
no regression either, under the same conditions:
threads=15, tc=2+0.02
ELO: 2.85 +-3.3 (95%) LOS: 95.4%
Total: 19661 W: 4640 L: 4479 D: 10542
threads=7, tc=2+0.02
ELO: 0.23 +-3.5 (95%) LOS: 55.1%
Total: 18108 W: 4326 L: 4314 D: 9468
As suggested by Joona, another test at LTC:
threads=15, tc=20+0.05
ELO: 0.64 +-2.6 (95%) LOS: 68.3%
Total: 20000 W: 3053 L: 3016 D: 13931
v3: Final version: SeqCst/Relaxed
threads=15, tc=10+0.1
ELO: 0.87 +-3.9 (95%) LOS: 67.1%
Total: 9541 W: 1478 L: 1454 D: 6609
Resolves#474
Start all threads searching on root position and
use only the shared TT table as synching scheme.
It seems this scheme scales better than YBWC for
high number of threads.
Verified for nor regression at STC 3 threads
LLR: -2.95 (-2.94,2.94) [-3.00,1.00]
Total: 40232 W: 6908 L: 7130 D: 26194
Verified for nor regression at LTC 3 threads
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 28186 W: 3908 L: 3798 D: 20480
Verified for nor regression at STC 7 threads
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 3607 W: 674 L: 526 D: 2407
Verified for nor regression at LTC 7 threads
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 4235 W: 671 L: 528 D: 3036
Tested with fixed games at LTC with 20 threads
ELO: 44.75 +-7.6 (95%) LOS: 100.0%
Total: 2069 W: 407 L: 142 D: 1520
Tested with fixed games at XLTC (120secs) with 20 threads
ELO: 28.01 +-6.7 (95%) LOS: 100.0%
Total: 2275 W: 349 L: 166 D: 1760
Original patch of mbootsector, with additional work
from Ivan Ivec (log formula), Joerg Oster (id loop
simplification) and Marco Costalba (assorted formatting
and rework).
Bench: 8116244
Introduce helper function Search::reset() which clears all kind of search
memory, in order to restore a deterministic search state.
Generalize TT.clear() into Search::reset() for the following use cases:
- bench: needed to guarantee deterministic bench (ie. if you call bench from
interactive command line twice in a row you get the same value).
- Clear Hash: restore clean search state, which is the purpose of this button.
- ucinewgame: ditto.
No functional change.
Resolves#346
When running more games in parallel, or simply when running a game
with a background process, due to how OS scheduling works, there is no
guarantee that the CPU resources allocated evenly between the two
players. This introduces noise in the result that leads to unreliable
result and in the worst cases can even invalidate the result. For
instance in SF test framework we avoid running from clouds virtual
machines because are a known source of very unstable CPU speed.
To overcome this issue, without requiring changes to the GUI, the idea
is to use searched nodes instead of time, and to convert time to
available nodes upfront, at the beginning of the game.
When nodestime UCI option is set at a given nodes per milliseconds
(npmsec), at the beginning of the game (and only once), the engine
reads the available time to think, sent by the GUI with 'go wtime x'
UCI command. Then it translates time in available nodes (nodes =
npmsec * x), then feeds available nodes instead of time to the time
management logic and starts the search. During the search the engine
checks the searched nodes against the available ones in such a way
that all the time management logic still fully applies, and the game
mimics a real one played on real time. When the search finishes,
before returning best move, the total available nodes are updated,
subtracting the real searched nodes. After the first move, the time
information sent by the GUI is ignored, and the engine fully relies on
the updated total available nodes to feed time management.
To avoid time losses, the speed of the engine (npms) must be set to a
value lower than real speed so that if the real TC is for instance 30
secs, and npms is half of the real speed, the game will last on
average 15 secs, so much less than the TC limit, providing for a
safety 'time buffer'.
There are 2 main limitations with this mode.
1. Engine speed should be the same for both players, and this limits
the approach to mainly parameter tuning patches.
2. Because npms is fixed while, in real engines, the speed increases
toward endgame, this introduces an artifact that is equivalent to an
altered time management. Namely it is like the time management gives
less available time than what should be in standard case.
May be the second limitation could be mitigated in a future with a
smarter 'dynamic npms' approach.
Tests shows that the standard deviation of the results with 'nodestime'
is lower than in standard TC, as is expected because now all the introduced
noise due the random speed variability of the engines during the game is
fully removed.
Original NIT idea by Michael Hoffman that shows how to play in NIT mode
without requiring changes to the GUI. This implementation goes a bit
further, the key difference is that we read TC from GUI only once upfront
instead of re-reading after every move as in Michael's implementation.
No functional change.
And reformat a bit time manager code.
Note that now we set starting search time in think() and
no more in ThreadPool::start_thinking(), the added delay
is less than 1 msec, so below timer resolution (5msec) and
should not affect time lossses ratio.
No functional change.
In case we stop the search during a fail-high
it is possible we return to GUI without a ponder
move. This patch try harder to find a ponder move
retrieving it from TT. This is important in games
played with 'ponder on'.
bench: 8080602
Resolves#221
Import C++11 branch from:
https://github.com/mcostalba/Stockfish/tree/c++11
The version imported is teh last one as of today:
6670e93e50
Branch is fully equivalent with master but syzygy
tablebases that are missing (but will be added with
next commit).
bench: 8080602
In some cases we want to go direcly to the moves loop
without checking for early return. The patch make this
logic more clear and consistent.
Tested for no regression, passed STC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 25282 W: 5136 L: 5022 D: 15124
and LTC
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 72007 W: 12133 L: 12095 D: 47779
bench: 9316798
Currently Search::RootMoves is accessed and even
modified by TB probing functions in a hidden
and sneaky way.
This is bad practice and makes the code tricky.
Instead explicily pass the vector as function
argument so to clarify that the vector is modified
inside the functions.
No functional change.
It is a non functional change, but because
we now skip copying of pv[] in SpNode, patch
has been tested for regression with 3 threads:
STC
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 54668 W: 9873 L: 9809 D: 34986
No functional change.
This is the first of a patch series to
rearrange and simplify accurate PV.
In this patch there is simple coding
style and reformatting stuff.
Verified with fishtest it does not crash
with MAX_PLY = 8
No functional change.
This gives SF accurate PVs, such that the evaluation of the leaf node in
the PV matches the score backed up to the root (99% of the time.
q-search will use the value stored in the hash table instead of the eval
value sometimes).
One drawback is that fail-high/low only get a minimal 2 move PV.
It doesn't add any additional overhead to the non-PV codepath except an
extra eight bytes to the SearchStack structure in multi-threaded
searches.
A core part of this is not pruning based on TT score in PV nodes. This
was measured as not being a regression at multiple TCs, except for one
exception, fast TC with huge hash, which is not realistic for longer
searches.
STC - 1 thread, 128 mb hash
ELO: 1.42 +-3.1 (95%) LOS: 81.9%
Total: 20000 W: 4078 L: 3996 D: 11926
STC - 3 thread, 128 mb hash
ELO: -3.60 +-2.9 (95%) LOS: 0.8%
Total: 20000 W: 3575 L: 3782 D: 12643
STC - 3 thread, 8 mb hash
ELO: 0.12 +-2.9 (95%) LOS: 53.3%
Total: 20000 W: 3654 L: 3647 D: 12699
LTC - 3 thread, 32mb hash
ELO: 2.29 +-2.0 (95%) LOS: 98.8%
Total: 35740 W: 5618 L: 5382 D: 24740
Bench: 6984058
Resolves#102
UCI specification is not clear on the size of
integers that are exchanged in the protocol, so
instead of a simple int, assume 'nodes' is a
int64_t because we need a bigger size to store
this value in many real cases, especialy with
very long searches.
No functional change.
Resolves#75
Unify various perft functions and move all the code
to search.cpp.
Avoid perft implementation to be splitted between
benchmark.cpp (where it has no reason to be) and
search.cpp
No functional and no speed change (tested).
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
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
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
Using memset on a std::vector is undefined behavior,
so manually init all the data memebers of LimitsType.
Bug intorduced in 41641e3b1e
No functional change.
When we have a fail-high of a quiet move, store it in
a Followupmoves table indexed by the previous move of
the same color (instead of immediate previous move as
is in countermoves case).
Then use this table for quiet moves ordering in the same
way we are already doing with countermoves.
These followup moves will be tried just after countermoves
and before remaining quiet moves.
Passed both short TC
LLR: 2.95 (-2.94,2.94) [-1.50,4.50]
Total: 10350 W: 1998 L: 1866 D: 6486
And long TC
LLR: 2.95 (-2.94,2.94) [0.00,6.00]
Total: 14066 W: 2303 L: 2137 D: 9626
bench: 7205153
If we are still at first move, without a fail-low and
current iteration is taking too long to complete then
stop the search.
Passed short TC:
LLR: 2.97 (-2.94,2.94) [-1.50,4.50]
Total: 26030 W: 4959 L: 4785 D: 16286
Long TC:
LLR: 2.95 (-2.94,2.94) [0.00,6.00]
Total: 18019 W: 2936 L: 2752 D: 12331
And performed well at 40/30
ELO: 4.33 +-2.8 (95%) LOS: 99.9%
Total: 20000 W: 3480 L: 3231 D: 13289
bench: 8502826
1/ eval margin and gains removed:
16bit are now free on TT entries, due to the removal of eval margin. may be useful
in the future :) gains removed: use instead by Value(128). search() and qsearch()
are now consistent in this regard.
2/ futility_margin()
linear formula instead of complex (log(depth), movecount) formula.
3/ unify pre & post futility pruning
pre futility pruning used depth < 7 plies, while post futility pruning used
depth < 4 plies. Now it's always depth < 7.
Tested with fixed number of games both at short TC:
ELO: 0.82 +-2.1 (95%) LOS: 77.3%
Total: 40000 W: 7939 L: 7845 D: 24216
And long TC
ELO: 0.59 +-2.0 (95%) LOS: 71.9%
Total: 40000 W: 6876 L: 6808 D: 26316
bench 7243575
1/ eval margin and gains removed:
- gains removed by Value(128): search() and qsearch() now behave consistently!
2/ futility_margin()
- testing showed that there is no added value in this weird (log(depth), movecount)
formula, and a much simpler linear formula is just as good. In fact, it is most
likely better, as it is not yet optimally tuned.
- the new simplified formula also means we get rid of FutilityMargins[], its
initialization code, and more importantly ss->futilityMoveCount, and the hacky
code that updates it throughout the search().
- the current formula gives negative futility margins, and there is a hidden interaction
between the move coutn pruning formula and the futility margin one: what happens is
that MCP is supposed to be triggered before we use the non-sensical negative futility
margins.
3/ unify pre & post futility pruning
- pre futility pruning (what SF calls value based pruning) used depth < 7 plies,
while post futility pruning (what SF calls static null move pruning) used depth < 4 plies.
- also the condition depth < 7 in pre futility pruning was not obvious, and it seemd
to be depth < 16 (futility_margin() returns an infinite value when depth >= 7).
Tested with fixed number of games both at short TC:
ELO: 0.82 +-2.1 (95%) LOS: 77.3%
Total: 40000 W: 7939 L: 7845 D: 24216
And long TC
ELO: 0.59 +-2.0 (95%) LOS: 71.9%
Total: 40000 W: 6876 L: 6808 D: 26316
bench: 10206576
And #ifdef instead of #if defined
This is more standard form (see for example iostream file).
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Instead of a pointer. This should fix the issue of
remaining with a stale pointer when for instance calling
IID, but also null search verification, singular search
and razoring where we call search with the same ss
pointer. In this case ss->ei is overwritten in the
search() call and upon returning remains stale.
This patch could have a performance hit because Eval::Info
is big (176 bytes) and during splitting we copy 4 ss entries.
On the good side, this patch is a clean solution.
Proposed by Gary.
No functional change.
Allow to use EvalInfo struct, populated by
evaluation(), in search.
In particular we allocate Eval::Info on the stack
and pass a pointer to this to evaluate().
Also add to Search::Stack a pointer to Eval::Info,
this allows to reference eval info of previous/next
nodes.
WARNING: Eval::Info is NOT initialized and is populated
by evaluate(), only if the latter is called, and this
does not happen in all the code paths, so care should be
taken when accessing this struct.
No functional change.
