After increasing the number of threads, the histories were not cleared,
resulting in uninitialized memory usage.
This patch fixes this by clearing threads histories in Thread c'tor as
is the idomatic way.
This fixes issue 1227
No functional change.
We use Position::set() to set root position across
threads. But there are some StateInfo fields (previous,
pliesFromNull, capturedPiece) that cannot be deduced
from a fen string, so set() clears them and to not lose
the info we need to backup and later restore setupStates->back().
Note that setupStates is shared by threads but is accessed
in read-only mode.
This fixes regression introduced by df6cb446ea
Tested with 3 threads at STC:
LLR: 2.95 (-2.94,2.94) [-4.00,0.00]
Total: 14436 W: 2304 L: 2196 D: 9936
Bench: 5608839
Simplify out low level sync stuff (mutex
and friends) and avoid to use them directly
in many functions.
Also some renaming and better comment while
there.
No functional change.
In this rare case (e.g. go infinite on a stalemate),
just spin till ponderhit/stop comes.
The Thread::wait() is a renmant of the old YBWC
code, today with lazy SMP, threads don't need to
wait when outside of their idle loop.
No functional change.
But this time correctly set Threads.ponder
We avoid using 'limits' for passing pondering
flag because we don't want to have 2 ponder
variables in search scope: Search::Limits.ponder
and Threads.ponder. This would be confusing also
because limits.ponder is set at the beginning of
the search and never changes, instead Threads.ponder
can change value asynchronously during search.
No functional change.
This reverts commit 5410424e3d.
After the commit pondering is broken, so revert for now. I will
resubmit with a proper fix.
The issue is mine, Joost original code is correct.
No functional change.
Limits::ponder was used as a signal between uci and search threads,
but is not an atomic variable, leading to the following race as
flagged by a sanitized binary.
Expect input:
```
spawn ./stockfish
send "uci\n"
expect "uciok"
send "setoption name Ponder value true\n"
send "go wtime 4000 btime 4000\n"
expect "bestmove"
send "position startpos e2e4 d7d5\n"
send "go wtime 4000 btime 4000 ponder\n"
sleep 0.01
send "ponderhit\n"
expect "bestmove"
send "quit\n"
expect eof
```
Race:
```
WARNING: ThreadSanitizer: data race (pid=7191)
Read of size 4 at 0x0000005c2260 by thread T1:
Previous write of size 4 at 0x0000005c2260 by main thread:
Location is global 'Search::Limits' of size 88 at 0x0000005c2220 (stockfish+0x0000005c2260)
```
The reason of teh race is that ponder is not just set in UCI go()
assignment but also is signaled by an async ponderhit in uci.cpp:
else if (token == "ponderhit")
Search::Limits.ponder = 0; // Switch to normal search
The fix is to add an atomic bool to the threads structure to
signal the ponder status, letting Search::Limits to reflect just
what was passed to 'go'.
No functional change.
Instead of having Signals in the search namespace,
make the stop variables part of the Threads structure.
This moves more of the shared (atomic) variables towards
the thread-related structures, making their role more clear.
No functional change
Closes#1149
The main change of the patch is that now time check
is done only by main thread. In the past, before lazy
SMP, we needed all the threds to check for available
time because main thread could have been blocked on
a split point, now this is no more the case and main
thread can do the job alone, greatly simplifying the logic.
Verified for regression testing on STC with 7 threads:
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 11895 W: 1741 L: 1608 D: 8546
No functional change.
Closes#1152
the nodes, tbHits, rootDepth and lastInfoTime variables are read by multiple threads, but not declared atomic, leading to data races as found by -fsanitize=thread. This patch fixes this issue. It is based on top of the CI-threading branch (PR #1129), and should fix the corresponding CI error messages.
The patch passed an STC check for no regression:
http://tests.stockfishchess.org/tests/view/5925d5590ebc59035df34b9f
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 169597 W: 29938 L: 30066 D: 109593
Whereas rootDepth and lastInfoTime are not performance critical, nodes and tbHits are. Indeed, an earlier version using relaxed atomic updates on the latter two variables failed STC testing (http://tests.stockfishchess.org/tests/view/592001700ebc59035df34924), which can be shown to be due to x86-32 (http://tests.stockfishchess.org/tests/view/592330ac0ebc59035df34a89). Indeed, the latter have no instruction to atomically update a 64bit variable. The proposed solution thus uses a variable in Position that is accessed only by one thread, which is copied every few thousand nodes to the shared variable in Thread.
No functional change.
Closes#1130Closes#1129
If not explicitly initialized in a class constructor,
then all data members are default-initialized when
the corresponing struct/class is instanced.
For array and built-in types (int, char, etc..)
default-initialization is a no-op and we need to
explicitly zero them.
No functional change.
Under Windows it is not possible for a process to run on more than one
logical processor group. This usually means to be limited to use max 64
cores. To overcome this, some special platform specific API should be
called to set group affinity for each thread. Original code from Texel by
Peter sterlund.
Tested by Jean-Paul Vael on a Xeon E7-8890 v4 with 88 threads and confimed
speed up between 44 and 88 threads is about 30%, as expected.
No functional change.
Restore original behaviour to reset
the counter before a new move search.
Also fixed some warnings and added const
qualifier to a couple of functions, as
suggested by m_stembera.
Thanks to Werner Bergmans for reporting
the regression.
No functional change.
Use a per-thread counter to reduce contention
with many cores and endgame positions.
Measured around 1% speed-up on a 12 core and 8%
on 28 cores with 6-men, searching on:
7R/1p3k2/2p2P2/3nR1P1/8/3b1P2/7K/r7 b - - 3 38
Also retire the unused set_nodes_searched() and fix
a couple of return types and naming conventions.
No functional change.
Both Tablebases::filter_root_moves() and
extract_ponder_from_tt(9 were unable to handle
a mate/stalemate position.
Spotted and reported by Dann Corbit.
Added some mate/stalemate positions to bench so
to early catch this regression in the future.
No functional change.
Currently root moves are copied to all teh threads
but are DTZ filtered only in main thread at the
beginning of teh search.
This patch moves the TB filtering before the
copy of root moves fixing issue #679https://github.com/official-stockfish/Stockfish/issues/679
No bench change.
In this position we should have draw for repetition:
position fen rnbqkbnr/2pppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 moves g1f3 g8f6 f3g1
go infinite
But latest patch broke it.
Actually we had two(!) very subtle bugs, the first is that Position::set()
clears the passed state and in particular 'previous' member, so
that on passing setupStates, 'previous' pointer was reset.
Second bug is even more subtle: SetupStates was based on std::vector
as container, but when vector grows, std::vector copies all its contents
to a new location invalidating all references to its entries. Because
all StateInfo records are linked by 'previous' pointer, this made pointers
go stale upon adding more element to setupStates. So revert to use a
std::deque that ensures references are preserved when pushing back new
elements.
No functional change.
And passed in do_move(), this ensures maximum efficiency and
speed and at the same time unlimited move numbers.
The draw back is that to handle Position init we need to
reserve a StateInfo inside Position itself and use at
init time and when copying from another Position.
After lazy SMP we don't need anymore this gimmick and we can
get rid of this special case and always pass an external
StateInfo to Position object.
Also rewritten and simplified Position constructors.
Verified it does not regress with a 3 threads SMP test:
ELO: -0.00 +-12.7 (95%) LOS: 50.0%
Total: 1000 W: 173 L: 173 D: 654
No functional change.
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
Instead of creating a running std::thread and
returning, wait in Thread c'tor that the native
thread of execution goes to sleep in idle_loop().
In this way we can simplify how search is started,
because when main thread is idle we are sure also
all other threads will be idle, in any case, even
at thread creation and startup.
After lazy smp went in, we can simpify and rewrite
a lot of logic that is now no more needed. This is
hopefully the final big cleanup.
Tested for no regression at 5+0.1 with 3 threads:
LLR: 2.95 (-2.94,2.94) [-5.00,0.00]
Total: 17411 W: 3248 L: 3198 D: 10965
No functional change.
Now that we don't have anymore TimerThread, there is
no need of this long class hierarchy.
Also assorted reformatting while there.
To verify no regression, passed at STC with 7 threads:
LLR: 2.97 (-2.94,2.94) [-5.00,0.00]
Total: 30990 W: 4945 L: 4942 D: 21103
No functional change.
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
When changing 'search' and 'splitPointsSize' we have to
use thread locks, not split point ones, because can_join()
is called under the formers.
Verified succesfully with 24 hours toruture tests with 20
cores machine by Louis Zulli: it does not hangs.
Verifyed for no regressions with STC, 7 threads:
LLR: 2.94 (-2.94,2.94) [-3.00,1.00]
Total: 52804 W: 8159 L: 8087 D: 36558
No functional change.
Louis Zulli reported that Stockfish suffers from very occasional hangs with his 20 cores machine.
Careful SMP debugging revealed that this was caused by "a ghost split point slave", where thread
was marked as a split point slave, but wasn't actually working on it.
The only logical explanation for this was double booking, where due to SMP race, the same thread
is booked for two different split points simultaneously.
Due to very intermittent nature of the problem, we can't say exactly how this happens.
The current handling of Thread specific variables is risky though. Volatile variables are in some
cases changed without spinlock being hold. In this case standard doesn't give us any kind of
guarantees about how the updated values are propagated to other threads.
We resolve the situation by enforcing very strict locking rules:
- Values for key thread variables (splitPointsSize, activeSplitPoint, searching)
can only be changed when the thread specific spinlock is held.
- Structural changes for splitPoints[] are only allowed when the thread specific spinlock is held.
- Thread booking decisions (per split point) can only be done when the thread specific spinlock is held.
With these changes hangs didn't occur anymore during 2 days torture testing on Zulli's machine.
We probably have a slight performance penalty in SMP mode due to more locking.
STC (7 threads):
ELO: -1.00 +-2.2 (95%) LOS: 18.4%
Total: 30000 W: 4538 L: 4624 D: 20838
However stability is worth more than 1-2 ELO points in this case.
No functional change
Resolves#422
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.
Currently if we call it more than once, we crash.
This is not a real problem, because this function is
indeed called just once. Nevertheless with this small fix,
that gets rid of a hidden 'static' variable, we cleanly
resolve the issue.
While there, fix also ThreadPool::exit to return in a
consistent state. Now all the init() functions but
UCI::init() are reentrant and can be called multiple
times.
No functional change.
To sync UI with main thread it is enough a single
condition variable because here we have a single
producer / single consumer design pattern.
Two condition variables are strictly needed just for
many producers / many consumers case.
Note that this is possible because now we don't send to
sleep idle threads anymore while searching, so that now
only UI can wake up the main thread and we can use the
same ConditionVariable for both threads.
The natural consequence is to retire wait_for_think_finished()
and move all the logic under MainThread class, yielding the
rename of teh function to join()
No functional change.
Avoid redundant 'while' conditions. It is enough to
check them in the outer loop.
Quick tested for no regression 10K games at 4 threads
ELO: -1.32 +-3.9 (95%) LOS: 25.6%
Total: 10000 W: 1653 L: 1691 D: 6656
No functional change.
During the search, do not block on condition variable, but instead use std::this_thread::yield().
Clear gain with 16 threads. Again results vary highly depending on hardware, but on average it's a clear gain.
ELO: 12.17 +-4.3 (95%) LOS: 100.0%
Total: 7998 W: 1407 L: 1127 D: 5464
There is no functional change in single thread mode
Resolves#294
Idea and original implementation by Stephane Nicolet
7 threads 15+0.05
ELO: 3.54 +-2.9 (95%) LOS: 99.2%
Total: 17971 W: 2976 L: 2793 D: 12202
There is no functional change in single thread mode
Use Mutex instead.
This is in preparaation for merging with master branch,
where we stilll don't have spinlocks.
Eventually spinlocks will be readded in some future
patch, once c++11 has been merged.
No functional change.
