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
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.
Commenst are obsolete now, an updated description
would be quite obscure, so better let the code
to talk and remove them all together.
No functional change.
On top of previous patch, rename time variables to
reflect the simplification of UCI parameters.
It is more correct to use as varibales directly the
corresponding UCI option, without intorducing redundant
intermediate variables.
This allows also to simplify the code.
No functional change.
When time remaining is less than Emergency Move Time,
we won't even complete one iteration and engine reports
a stale +M0 score.
To reproduce run "go wtime 10"
info depth 1 seldepth 1 score mate 0 upperbound nodes 2 nps 500 time 4 multipv 1 pv a2a3
info nodes 2 time 4
bestmove a2a3 ponder (none)
This patch fixes the issue.
Tested by Binky at very short TC: 0.05+0.05
ELO: 5.96 +-12.9 (95%) LOS: 81.7%
Total: 1458 W: 394 L: 369 D: 695
And at a bit longer TC:
ELO: 1.56 +-3.7 (95%) LOS: 79.8%
Total: 16511 W: 3983 L: 3909 D: 8619
bench: 7804908
In case ply is very high, function will round
to zero (although mathematically it is always
bigger than zero). On my system this happens at
movenumber 6661.
Although 6661 moves in a game is, of course,
probably impossible, for safety and to be locally
consistent makes sense to ensure returned value
is positive.
Non functional change.
Function move_importance() is already always
positive, so we don't need to add a constant
term to ensure it.
Becuase move_importance() is used to calculate
ratios of a linear combination (as explained in
previous patch), result is not affected. I have
also verified it directly.
No functional change.
Drop a useless parameter. This works because ratio1 and ratio2
are ratios of linear combinations of thisMoveImportance and
otherMovesImportance and so the yscale cancels out.
Therefore the values of ratio1 and ratio2 are independent
of yscale and yscale can be retired.
The same applies to yshift, but here we want to ensure
move_importance() > 0, so directly hard-code this safety
guard in function definition.
Actually there are some small differences due to rounding errors
and usually are at most few millisecond, that's means below 1% of
returned time, apart from very short intervals in which a difference
of just 1 msec can raise to 2-3% of total available time.
No functional change.
Use a skew-logistic function to replace the
MoveImportance[] array.
Verified it does not regress at fixed number
of games both at short TC:
LLR: -2.91 (-2.94,2.94) [-1.50,4.50]
Total: 39457 W: 7539 L: 7538 D: 24380
And long TC:
ELO: -0.49 +-1.9 (95%) LOS: 31.0%
Total: 39358 W: 6135 L: 6190 D: 27033
bench: 7335588
If the game got late enough that move_importance(currentPly) * slowMover / 100
rounds to 0, then we ended up dividing 0 by 0 when only looking 1 move ahead.
This apparently caused the search to almost immediately abort and Stockfish
would blunder in long games. So convert thisMoveImportance to a double.
No functional change.
These two changes go in opposite directions and it
seems that the combination is stronger than original.
Here are the positive tests at various TC:
15+0.05
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 24561 W: 4946 L: 4772 D: 14843
60+0.05
LLR: 2.96 (-2.94,2.94) [0.00,6.00]
Total: 15259 W: 2598 L: 2423 D: 10238
40/30
LLR: 2.96 (-2.94,2.94) [-3.00,3.00]
Total: 2570 W: 527 L: 422 D: 1621
Unfortunately there is also a bad result
with one sec time increment that needs
to be further investigated:
12+1
LLR: -2.97 (-2.94,2.94) [-3.00,3.00]
Total: 2694 W: 438 L: 543 D: 1713
bench: 8340585
Rationale:
- Speed of double and float is about the same (not on the hot path anyway)
- Double makes code prettier (no need to write 1.0f, just 1.0)
- Only practical advantage of float is to use less memory, but since we never
store large arrays of double, we don't care.
No functional change.
The ideal setting for super-blitz might be something like:
"Emergency Base Time" = 50
"Emergency Move Time" = 5
This would give a total emergency time buffer of:
50 + 40 * 5 = 250 ms
This setup replaces the previous half cooked hack
"Don't blunder under extreme time pressure".
Test results are very good at super blitz, but keep good even
at 60 secs.
At 5+0.05
ELO: 24.30 +-2.4 (95%) LOS: 100.0%
Total: 37802 W: 10060 L: 7420 D: 20322
At 15+0.05
ELO: 13.41 +-2.9 (95%) LOS: 100.0%
Total: 22271 W: 4853 L: 3994 D: 13424
At 60+0.05
ELO: 5.30 +-3.2 (95%) LOS: 99.9%
Total: 16000 W: 2897 L: 2653 D: 10450
No functional change.
Now that we use pre-increment on enums, it
make sense, for code style uniformity, to
swith to pre-increment also for native types,
although there is no speed difference.
No functional change.
We always attempt to keep at least this emergencyBaseTime
at clock. But if available time is very low it means that
we will force ourself to play immediately to satisfy the
emergencyBaseTime constrain and so leading to blunders.
Patch is good at short and very short TC (15secs and 5secs respectively)
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 26590 W: 5426 L: 5245 D: 15919
LLR: 2.96 (-2.94,2.94) [-1.50,4.50]
Total: 5767 W: 1397 L: 1268 D: 3102
Instead seems has no influence at longer TC (60 secs)
LLR: -2.96 (-2.94,2.94) [0.00,6.00]
Total: 79862 W: 13623 L: 13339 D: 52900
So it is committed to have a broader testing but is
to be consider still EXPERIMENTAL and can be reverted
easily.
No functional change.
Increase MaxRatio to use more time when in trouble.
After 16000 games at 60+0.05
ELO: 4.89 +-5.4 (95%) LOS: 99.9%
Total: 16000 W: 2700 L: 2475 D: 10825
No functional change.
There is no need to "invent" different names
from the original UCI parameters.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
With default value of 100 no change in regard of current
behaviour. Increasing the value makes SF to think a
longer time for each move. Decreasing the value makes SF
to move faster.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Greatly improves the usage. User defined conversions
are a novelity for SF, another amazing C++ facility
at work !
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Move global search-related variables under "Search" namespace.
As a side effect we can move uci_async_command() and
wait_for_stop_or_ponderhit() away from search.cpp
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
The timer will be fired asynchronously to handle
time management flags, while other threads are
searching.
This implementation uses a thread waiting on a
timed condition variable instead of real timers.
This approach allow to reduce platform dependant
code to a minimum and also is the most portable given
that timers libraries are very different among platforms
and also the best ones are not compatible with olds
Windows.
Also retire the now unused polling code.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
And pass correct currentPly to TimeManager::init().
This restores old behaviour, in particular now black has
a different timing than white becuase is no more:
currentPly = 2 * fullMoveNumber;
but becomes
2 * (fullMoves - 1) + int(sideToMove == BLACK)
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
And also tolerate a 0 value for full move number.
Revert BUG_41 patch, now we set initial King file only
if a castling is possible, so we don't need the fix
anymore in case of correct FEN.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
Mostly suggested by Justin (UncombedCoconut), the 0ULL -> 0 conversion
is mine.
No functional change.
Signed-off-by: Marco Costalba <mcostalba@gmail.com>
